547 34 11MB
English Pages 800 [1016] Year 2022
The Cambridge Handbook of Working Memory and Language Bringing together cutting-edge research, this Handbook is the first comprehensive text to examine the pivotal role of working memory in first and second language acquisition, processing, impairments, and training. Authored by a stellar cast of distinguished scholars from around the world, the Handbook provides authoritative insights on work from diverse, multidisciplinary perspectives, and introduces key models of working memory in relation to language. Following an introductory chapter by working memory pioneer Alan Baddeley, the collection is organized into thematic sections that discuss working memory in relation to: theoretical models and measures; linguistic theories and frameworks; first language processing; bilingual acquisition and processing; and language disorders, interventions, and instruction. The Handbook is sure to interest and benefit researchers, clinicians, speech therapists, and advanced undergraduate and postgraduate students in linguistics, psychology, education, speech therapy, cognitive science, and neuroscience, or anyone seeking to learn more about language, cognition, and the human mind. john w. schwieter is a Professor of Spanish and linguistics, Cross-Appointed in Psychology, and the director of Bilingualism Matters @ Laurier and the Language Acquisition, Multilingualism, and Cognition Laboratory at Wilfrid Laurier University. He is Executive Editor of the Bilingual Processing and Acquisition book series and Coeditor of Cambridge Elements in Second Language Acquisition. zhisheng (edward) wen is an Associate Professor at Macao Polytechnic University in Macau. He has extensive teaching and research experience in applied linguistics, second language acquisition, and cognitive science. He has authored and edited volumes on working memory and language aptitude published by Cambridge University Press, Routledge, Benjamins, and Multilingual Matters.
Published online by Cambridge University Press
cambridge handbooks in language and linguistics
Genuinely broad in scope, each handbook in this series provides a complete state-of-the-field overview of a major sub-discipline within language study and research. Grouped into broad thematic areas, the chapters in each volume encompass the most important issues and topics within each subject, offering a coherent picture of the latest theories and findings. Together, the volumes will build into an integrated overview of the discipline in its entirety.
Published titles The Cambridge Handbook of Phonology, edited by Paul de Lacy The Cambridge Handbook of Linguistic Code-Switching, edited by Barbara E. Bullock and Almeida Jacqueline Toribio The Cambridge Handbook of Child Language, Second Edition, edited by Edith L. Bavin and Letitia Naigles The Cambridge Handbook of Endangered Languages, edited by Peter K. Austin and Julia Sallabank The Cambridge Handbook of Sociolinguistics, edited by Rajend Mesthrie The Cambridge Handbook of Pragmatics, edited by Keith Allan and Kasia M. Jaszczolt The Cambridge Handbook of Language Policy, edited by Bernard Spolsky The Cambridge Handbook of Second Language Acquisition, edited by Julia Herschensohn and Martha Young-Scholten The Cambridge Handbook of Biolinguistics, edited by Cedric Boeckx and Kleanthes K. Grohmann The Cambridge Handbook of Generative Syntax, edited by Marcel den Dikken The Cambridge Handbook of Communication Disorders, edited by Louise Cummings The Cambridge Handbook of Stylistics, edited by Peter Stockwell and Sara Whiteley The Cambridge Handbook of Linguistic Anthropology, edited by N. J. Enfield, Paul Kockelman and Jack Sidnell The Cambridge Handbook of English Corpus Linguistics, edited by Douglas Biber and Randi Reppen The Cambridge Handbook of Bilingual Processing, edited by John W. Schwieter The Cambridge Handbook of Learner Corpus Research, edited by Sylviane Granger, Gaëtanelle Gilquin and Fanny Meunier The Cambridge Handbook of Linguistic Multicompetence, edited by Li Wei and Vivian Cook The Cambridge Handbook of English Historical Linguistics, edited by Merja Kytö and Päivi Pahta The Cambridge Handbook of Formal Semantics, edited by Maria Aloni and Paul Dekker The Cambridge Handbook of Morphology, edited by Andrew Hippisley and Greg Stump The Cambridge Handbook of Historical Syntax, edited by Adam Ledgeway and Ian Roberts The Cambridge Handbook of Linguistic Typology, edited by Alexandra Y. Aikhenvald and R. M. W. Dixon The Cambridge Handbook of Areal Linguistics, edited by Raymond Hickey The Cambridge Handbook of Cognitive Linguistics, edited by Barbara Dancygier The Cambridge Handbook of Japanese Linguistics, edited by Yoko Hasegawa
Published online by Cambridge University Press
The Cambridge Handbook of Spanish Linguistics, edited by Kimberly L. Geeslin The Cambridge Handbook of Bilingualism, edited by Annick De Houwer and Lourdes Ortega The Cambridge Handbook of Systemic Functional Linguistics, edited by Geoff Thompson, Wendy L. Bowcher, Lise Fontaine and David Schönthal The Cambridge Handbook of African Linguistics, edited by H. Ekkehard Wolff The Cambridge Handbook of Language Learning, edited by John W. Schwieter and Alessandro Benati The Cambridge Handbook of World Englishes, edited by Daniel Schreier, Marianne Hundt and Edgar W. Schneider The Cambridge Handbook of Intercultural Communication, edited by Guido Rings and Sebastian Rasinger The Cambridge Handbook of Germanic Linguistics, edited by Michael T. Putnam and B. Richard Page The Cambridge Handbook of Discourse Studies, edited by Anna De Fina and Alexandra Georgakopoulou The Cambridge Handbook of Language Standardization, edited by Wendy Ayres-Bennett and John Bellamy The Cambridge Handbook of Korean Linguistics, edited by Sungdai Cho and John Whitman The Cambridge Handbook of Phonetics, edited by Rachael-Anne Knight and Jane Setter The Cambridge Handbook of Corrective Feedback in Second Language Learning and Teaching, edited by Hossein Nassaji and Eva Kartchava The Cambridge Handbook of Experimental Syntax, edited by Grant Goodall The Cambridge Handbook of Heritage Languages and Linguistics, edited by Silvina Montrul and Maria Polinsky The Cambridge Handbook of Arabic Linguistics, edited by Karin Ryding and David Wilmsen The Cambridge Handbook of the Philosophy of Language, edited by Piotr Stalmaszczyk The Cambridge Handbook of Sociopragmatics, edited by Michael Haugh, Dániel Z. Kádár and Marina Terkourafi The Cambridge Handbook of Task-Based Language Teaching, edited by Mohammed Ahmadian and Michael Long The Cambridge Handbook of Language Contact: Population Movement and Language Change, Volume 1, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Language Contact: Multilingualism in Population Structure, Volume 2, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Romance Linguistics, edited by Adam Ledgeway and Martin Maiden The Cambridge Handbook of Translation, edited by Kirsten Malmkjaer
Published online by Cambridge University Press
Published online by Cambridge University Press
The Cambridge Handbook of Working Memory and Language Edited by John W. Schwieter Wilfrid Laurier University
Zhisheng (Edward) Wen Macao Polytechnic University
Published online by Cambridge University Press
University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108845342 DOI: 10.1017/9781108955638 © Cambridge University Press 2022 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2022 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Schwieter, John W., 1979- editor. | Wen, Zhisheng, editor. Title: The Cambridge handbook of working memory and language / edited by John W. Schwieter, Wilfrid Laurier University, Ontario ; Edward Z.S. Wen, Macao Polytechnic Institute. Description: Cambridge ; New York, NY : Cambridge University Press, 2022. | Series: Cambridge handbooks in language and linguistics | Includes bibliographical references and index. Identifiers: LCCN 2022001124 (print) | LCCN 2022001125 (ebook) | ISBN 9781108845342 (hardback) | ISBN 9781108958110 (paperback) | ISBN 9781108955638 (epub) Subjects: LCSH: Language acquisition–Psychological aspects. | Second language acquisition–Psychological aspects. | Language disorders in children. | Short-term memory. | Psycholinguistics. | BISAC: LANGUAGE ARTS & DISCIPLINES / Linguistics / General | LCGFT: Essays. Classification: LCC P118 .C365 2022 (print) | LCC P118 (ebook) | DDC 401/.93–dc23/ eng/20220113 LC record available at https://lccn.loc.gov/2022001124 LC ebook record available at https://lccn.loc.gov/2022001125 ISBN 978-1-108-84534-2 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.
Published online by Cambridge University Press
The Cambridge Handbook of Working Memory and Language Bringing together cutting-edge research, this Handbook is the first comprehensive text to examine the pivotal role of working memory in first and second language acquisition, processing, impairments, and training. Authored by a stellar cast of distinguished scholars from around the world, the Handbook provides authoritative insights on work from diverse, multidisciplinary perspectives, and introduces key models of working memory in relation to language. Following an introductory chapter by working memory pioneer Alan Baddeley, the collection is organized into thematic sections that discuss working memory in relation to: theoretical models and measures; linguistic theories and frameworks; first language processing; bilingual acquisition and processing; and language disorders, interventions, and instruction. The Handbook is sure to interest and benefit researchers, clinicians, speech therapists, and advanced undergraduate and postgraduate students in linguistics, psychology, education, speech therapy, cognitive science, and neuroscience, or anyone seeking to learn more about language, cognition, and the human mind. john w. schwieter is a Professor of Spanish and linguistics, Cross-Appointed in Psychology, and the director of Bilingualism Matters @ Laurier and the Language Acquisition, Multilingualism, and Cognition Laboratory at Wilfrid Laurier University. He is Executive Editor of the Bilingual Processing and Acquisition book series and Coeditor of Cambridge Elements in Second Language Acquisition. zhisheng (edward) wen is an Associate Professor at Macao Polytechnic University in Macau. He has extensive teaching and research experience in applied linguistics, second language acquisition, and cognitive science. He has authored and edited volumes on working memory and language aptitude published by Cambridge University Press, Routledge, Benjamins, and Multilingual Matters.
Published online by Cambridge University Press
cambridge handbooks in language and linguistics
Genuinely broad in scope, each handbook in this series provides a complete state-of-the-field overview of a major sub-discipline within language study and research. Grouped into broad thematic areas, the chapters in each volume encompass the most important issues and topics within each subject, offering a coherent picture of the latest theories and findings. Together, the volumes will build into an integrated overview of the discipline in its entirety.
Published titles The Cambridge Handbook of Phonology, edited by Paul de Lacy The Cambridge Handbook of Linguistic Code-Switching, edited by Barbara E. Bullock and Almeida Jacqueline Toribio The Cambridge Handbook of Child Language, Second Edition, edited by Edith L. Bavin and Letitia Naigles The Cambridge Handbook of Endangered Languages, edited by Peter K. Austin and Julia Sallabank The Cambridge Handbook of Sociolinguistics, edited by Rajend Mesthrie The Cambridge Handbook of Pragmatics, edited by Keith Allan and Kasia M. Jaszczolt The Cambridge Handbook of Language Policy, edited by Bernard Spolsky The Cambridge Handbook of Second Language Acquisition, edited by Julia Herschensohn and Martha Young-Scholten The Cambridge Handbook of Biolinguistics, edited by Cedric Boeckx and Kleanthes K. Grohmann The Cambridge Handbook of Generative Syntax, edited by Marcel den Dikken The Cambridge Handbook of Communication Disorders, edited by Louise Cummings The Cambridge Handbook of Stylistics, edited by Peter Stockwell and Sara Whiteley The Cambridge Handbook of Linguistic Anthropology, edited by N. J. Enfield, Paul Kockelman and Jack Sidnell The Cambridge Handbook of English Corpus Linguistics, edited by Douglas Biber and Randi Reppen The Cambridge Handbook of Bilingual Processing, edited by John W. Schwieter The Cambridge Handbook of Learner Corpus Research, edited by Sylviane Granger, Gaëtanelle Gilquin and Fanny Meunier The Cambridge Handbook of Linguistic Multicompetence, edited by Li Wei and Vivian Cook The Cambridge Handbook of English Historical Linguistics, edited by Merja Kytö and Päivi Pahta The Cambridge Handbook of Formal Semantics, edited by Maria Aloni and Paul Dekker The Cambridge Handbook of Morphology, edited by Andrew Hippisley and Greg Stump The Cambridge Handbook of Historical Syntax, edited by Adam Ledgeway and Ian Roberts The Cambridge Handbook of Linguistic Typology, edited by Alexandra Y. Aikhenvald and R. M. W. Dixon The Cambridge Handbook of Areal Linguistics, edited by Raymond Hickey The Cambridge Handbook of Cognitive Linguistics, edited by Barbara Dancygier The Cambridge Handbook of Japanese Linguistics, edited by Yoko Hasegawa
Published online by Cambridge University Press
The Cambridge Handbook of Spanish Linguistics, edited by Kimberly L. Geeslin The Cambridge Handbook of Bilingualism, edited by Annick De Houwer and Lourdes Ortega The Cambridge Handbook of Systemic Functional Linguistics, edited by Geoff Thompson, Wendy L. Bowcher, Lise Fontaine and David Schönthal The Cambridge Handbook of African Linguistics, edited by H. Ekkehard Wolff The Cambridge Handbook of Language Learning, edited by John W. Schwieter and Alessandro Benati The Cambridge Handbook of World Englishes, edited by Daniel Schreier, Marianne Hundt and Edgar W. Schneider The Cambridge Handbook of Intercultural Communication, edited by Guido Rings and Sebastian Rasinger The Cambridge Handbook of Germanic Linguistics, edited by Michael T. Putnam and B. Richard Page The Cambridge Handbook of Discourse Studies, edited by Anna De Fina and Alexandra Georgakopoulou The Cambridge Handbook of Language Standardization, edited by Wendy Ayres-Bennett and John Bellamy The Cambridge Handbook of Korean Linguistics, edited by Sungdai Cho and John Whitman The Cambridge Handbook of Phonetics, edited by Rachael-Anne Knight and Jane Setter The Cambridge Handbook of Corrective Feedback in Second Language Learning and Teaching, edited by Hossein Nassaji and Eva Kartchava The Cambridge Handbook of Experimental Syntax, edited by Grant Goodall The Cambridge Handbook of Heritage Languages and Linguistics, edited by Silvina Montrul and Maria Polinsky The Cambridge Handbook of Arabic Linguistics, edited by Karin Ryding and David Wilmsen The Cambridge Handbook of the Philosophy of Language, edited by Piotr Stalmaszczyk The Cambridge Handbook of Sociopragmatics, edited by Michael Haugh, Dániel Z. Kádár and Marina Terkourafi The Cambridge Handbook of Task-Based Language Teaching, edited by Mohammed Ahmadian and Michael Long The Cambridge Handbook of Language Contact: Population Movement and Language Change, Volume 1, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Language Contact: Multilingualism in Population Structure, Volume 2, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Romance Linguistics, edited by Adam Ledgeway and Martin Maiden The Cambridge Handbook of Translation, edited by Kirsten Malmkjaer
Published online by Cambridge University Press
Published online by Cambridge University Press
The Cambridge Handbook of Working Memory and Language Edited by John W. Schwieter Wilfrid Laurier University
Zhisheng (Edward) Wen Macao Polytechnic University
Published online by Cambridge University Press
University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108845342 DOI: 10.1017/9781108955638 © Cambridge University Press 2022 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2022 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Schwieter, John W., 1979- editor. | Wen, Zhisheng, editor. Title: The Cambridge handbook of working memory and language / edited by John W. Schwieter, Wilfrid Laurier University, Ontario ; Edward Z.S. Wen, Macao Polytechnic Institute. Description: Cambridge ; New York, NY : Cambridge University Press, 2022. | Series: Cambridge handbooks in language and linguistics | Includes bibliographical references and index. Identifiers: LCCN 2022001124 (print) | LCCN 2022001125 (ebook) | ISBN 9781108845342 (hardback) | ISBN 9781108958110 (paperback) | ISBN 9781108955638 (epub) Subjects: LCSH: Language acquisition–Psychological aspects. | Second language acquisition–Psychological aspects. | Language disorders in children. | Short-term memory. | Psycholinguistics. | BISAC: LANGUAGE ARTS & DISCIPLINES / Linguistics / General | LCGFT: Essays. Classification: LCC P118 .C365 2022 (print) | LCC P118 (ebook) | DDC 401/.93–dc23/ eng/20220113 LC record available at https://lccn.loc.gov/2022001124 LC ebook record available at https://lccn.loc.gov/2022001125 ISBN 978-1-108-84534-2 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.
Published online by Cambridge University Press
The Cambridge Handbook of Working Memory and Language Bringing together cutting-edge research, this Handbook is the first comprehensive text to examine the pivotal role of working memory in first and second language acquisition, processing, impairments, and training. Authored by a stellar cast of distinguished scholars from around the world, the Handbook provides authoritative insights on work from diverse, multidisciplinary perspectives, and introduces key models of working memory in relation to language. Following an introductory chapter by working memory pioneer Alan Baddeley, the collection is organized into thematic sections that discuss working memory in relation to: theoretical models and measures; linguistic theories and frameworks; first language processing; bilingual acquisition and processing; and language disorders, interventions, and instruction. The Handbook is sure to interest and benefit researchers, clinicians, speech therapists, and advanced undergraduate and postgraduate students in linguistics, psychology, education, speech therapy, cognitive science, and neuroscience, or anyone seeking to learn more about language, cognition, and the human mind. john w. schwieter is a Professor of Spanish and linguistics, Cross-Appointed in Psychology, and the director of Bilingualism Matters @ Laurier and the Language Acquisition, Multilingualism, and Cognition Laboratory at Wilfrid Laurier University. He is Executive Editor of the Bilingual Processing and Acquisition book series and Coeditor of Cambridge Elements in Second Language Acquisition. zhisheng (edward) wen is an Associate Professor at Macao Polytechnic University in Macau. He has extensive teaching and research experience in applied linguistics, second language acquisition, and cognitive science. He has authored and edited volumes on working memory and language aptitude published by Cambridge University Press, Routledge, Benjamins, and Multilingual Matters.
Published online by Cambridge University Press
cambridge handbooks in language and linguistics
Genuinely broad in scope, each handbook in this series provides a complete state-of-the-field overview of a major sub-discipline within language study and research. Grouped into broad thematic areas, the chapters in each volume encompass the most important issues and topics within each subject, offering a coherent picture of the latest theories and findings. Together, the volumes will build into an integrated overview of the discipline in its entirety.
Published titles The Cambridge Handbook of Phonology, edited by Paul de Lacy The Cambridge Handbook of Linguistic Code-Switching, edited by Barbara E. Bullock and Almeida Jacqueline Toribio The Cambridge Handbook of Child Language, Second Edition, edited by Edith L. Bavin and Letitia Naigles The Cambridge Handbook of Endangered Languages, edited by Peter K. Austin and Julia Sallabank The Cambridge Handbook of Sociolinguistics, edited by Rajend Mesthrie The Cambridge Handbook of Pragmatics, edited by Keith Allan and Kasia M. Jaszczolt The Cambridge Handbook of Language Policy, edited by Bernard Spolsky The Cambridge Handbook of Second Language Acquisition, edited by Julia Herschensohn and Martha Young-Scholten The Cambridge Handbook of Biolinguistics, edited by Cedric Boeckx and Kleanthes K. Grohmann The Cambridge Handbook of Generative Syntax, edited by Marcel den Dikken The Cambridge Handbook of Communication Disorders, edited by Louise Cummings The Cambridge Handbook of Stylistics, edited by Peter Stockwell and Sara Whiteley The Cambridge Handbook of Linguistic Anthropology, edited by N. J. Enfield, Paul Kockelman and Jack Sidnell The Cambridge Handbook of English Corpus Linguistics, edited by Douglas Biber and Randi Reppen The Cambridge Handbook of Bilingual Processing, edited by John W. Schwieter The Cambridge Handbook of Learner Corpus Research, edited by Sylviane Granger, Gaëtanelle Gilquin and Fanny Meunier The Cambridge Handbook of Linguistic Multicompetence, edited by Li Wei and Vivian Cook The Cambridge Handbook of English Historical Linguistics, edited by Merja Kytö and Päivi Pahta The Cambridge Handbook of Formal Semantics, edited by Maria Aloni and Paul Dekker The Cambridge Handbook of Morphology, edited by Andrew Hippisley and Greg Stump The Cambridge Handbook of Historical Syntax, edited by Adam Ledgeway and Ian Roberts The Cambridge Handbook of Linguistic Typology, edited by Alexandra Y. Aikhenvald and R. M. W. Dixon The Cambridge Handbook of Areal Linguistics, edited by Raymond Hickey The Cambridge Handbook of Cognitive Linguistics, edited by Barbara Dancygier The Cambridge Handbook of Japanese Linguistics, edited by Yoko Hasegawa
Published online by Cambridge University Press
The Cambridge Handbook of Spanish Linguistics, edited by Kimberly L. Geeslin The Cambridge Handbook of Bilingualism, edited by Annick De Houwer and Lourdes Ortega The Cambridge Handbook of Systemic Functional Linguistics, edited by Geoff Thompson, Wendy L. Bowcher, Lise Fontaine and David Schönthal The Cambridge Handbook of African Linguistics, edited by H. Ekkehard Wolff The Cambridge Handbook of Language Learning, edited by John W. Schwieter and Alessandro Benati The Cambridge Handbook of World Englishes, edited by Daniel Schreier, Marianne Hundt and Edgar W. Schneider The Cambridge Handbook of Intercultural Communication, edited by Guido Rings and Sebastian Rasinger The Cambridge Handbook of Germanic Linguistics, edited by Michael T. Putnam and B. Richard Page The Cambridge Handbook of Discourse Studies, edited by Anna De Fina and Alexandra Georgakopoulou The Cambridge Handbook of Language Standardization, edited by Wendy Ayres-Bennett and John Bellamy The Cambridge Handbook of Korean Linguistics, edited by Sungdai Cho and John Whitman The Cambridge Handbook of Phonetics, edited by Rachael-Anne Knight and Jane Setter The Cambridge Handbook of Corrective Feedback in Second Language Learning and Teaching, edited by Hossein Nassaji and Eva Kartchava The Cambridge Handbook of Experimental Syntax, edited by Grant Goodall The Cambridge Handbook of Heritage Languages and Linguistics, edited by Silvina Montrul and Maria Polinsky The Cambridge Handbook of Arabic Linguistics, edited by Karin Ryding and David Wilmsen The Cambridge Handbook of the Philosophy of Language, edited by Piotr Stalmaszczyk The Cambridge Handbook of Sociopragmatics, edited by Michael Haugh, Dániel Z. Kádár and Marina Terkourafi The Cambridge Handbook of Task-Based Language Teaching, edited by Mohammed Ahmadian and Michael Long The Cambridge Handbook of Language Contact: Population Movement and Language Change, Volume 1, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Language Contact: Multilingualism in Population Structure, Volume 2, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Romance Linguistics, edited by Adam Ledgeway and Martin Maiden The Cambridge Handbook of Translation, edited by Kirsten Malmkjaer
Published online by Cambridge University Press
Published online by Cambridge University Press
The Cambridge Handbook of Working Memory and Language Edited by John W. Schwieter Wilfrid Laurier University
Zhisheng (Edward) Wen Macao Polytechnic University
Published online by Cambridge University Press
University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108845342 DOI: 10.1017/9781108955638 © Cambridge University Press 2022 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2022 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Schwieter, John W., 1979- editor. | Wen, Zhisheng, editor. Title: The Cambridge handbook of working memory and language / edited by John W. Schwieter, Wilfrid Laurier University, Ontario ; Edward Z.S. Wen, Macao Polytechnic Institute. Description: Cambridge ; New York, NY : Cambridge University Press, 2022. | Series: Cambridge handbooks in language and linguistics | Includes bibliographical references and index. Identifiers: LCCN 2022001124 (print) | LCCN 2022001125 (ebook) | ISBN 9781108845342 (hardback) | ISBN 9781108958110 (paperback) | ISBN 9781108955638 (epub) Subjects: LCSH: Language acquisition–Psychological aspects. | Second language acquisition–Psychological aspects. | Language disorders in children. | Short-term memory. | Psycholinguistics. | BISAC: LANGUAGE ARTS & DISCIPLINES / Linguistics / General | LCGFT: Essays. Classification: LCC P118 .C365 2022 (print) | LCC P118 (ebook) | DDC 401/.93–dc23/ eng/20220113 LC record available at https://lccn.loc.gov/2022001124 LC ebook record available at https://lccn.loc.gov/2022001125 ISBN 978-1-108-84534-2 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.
Published online by Cambridge University Press
The Cambridge Handbook of Working Memory and Language Bringing together cutting-edge research, this Handbook is the first comprehensive text to examine the pivotal role of working memory in first and second language acquisition, processing, impairments, and training. Authored by a stellar cast of distinguished scholars from around the world, the Handbook provides authoritative insights on work from diverse, multidisciplinary perspectives, and introduces key models of working memory in relation to language. Following an introductory chapter by working memory pioneer Alan Baddeley, the collection is organized into thematic sections that discuss working memory in relation to: theoretical models and measures; linguistic theories and frameworks; first language processing; bilingual acquisition and processing; and language disorders, interventions, and instruction. The Handbook is sure to interest and benefit researchers, clinicians, speech therapists, and advanced undergraduate and postgraduate students in linguistics, psychology, education, speech therapy, cognitive science, and neuroscience, or anyone seeking to learn more about language, cognition, and the human mind. john w. schwieter is a Professor of Spanish and linguistics, Cross-Appointed in Psychology, and the director of Bilingualism Matters @ Laurier and the Language Acquisition, Multilingualism, and Cognition Laboratory at Wilfrid Laurier University. He is Executive Editor of the Bilingual Processing and Acquisition book series and Coeditor of Cambridge Elements in Second Language Acquisition. zhisheng (edward) wen is an Associate Professor at Macao Polytechnic University in Macau. He has extensive teaching and research experience in applied linguistics, second language acquisition, and cognitive science. He has authored and edited volumes on working memory and language aptitude published by Cambridge University Press, Routledge, Benjamins, and Multilingual Matters.
Published online by Cambridge University Press
cambridge handbooks in language and linguistics
Genuinely broad in scope, each handbook in this series provides a complete state-of-the-field overview of a major sub-discipline within language study and research. Grouped into broad thematic areas, the chapters in each volume encompass the most important issues and topics within each subject, offering a coherent picture of the latest theories and findings. Together, the volumes will build into an integrated overview of the discipline in its entirety.
Published titles The Cambridge Handbook of Phonology, edited by Paul de Lacy The Cambridge Handbook of Linguistic Code-Switching, edited by Barbara E. Bullock and Almeida Jacqueline Toribio The Cambridge Handbook of Child Language, Second Edition, edited by Edith L. Bavin and Letitia Naigles The Cambridge Handbook of Endangered Languages, edited by Peter K. Austin and Julia Sallabank The Cambridge Handbook of Sociolinguistics, edited by Rajend Mesthrie The Cambridge Handbook of Pragmatics, edited by Keith Allan and Kasia M. Jaszczolt The Cambridge Handbook of Language Policy, edited by Bernard Spolsky The Cambridge Handbook of Second Language Acquisition, edited by Julia Herschensohn and Martha Young-Scholten The Cambridge Handbook of Biolinguistics, edited by Cedric Boeckx and Kleanthes K. Grohmann The Cambridge Handbook of Generative Syntax, edited by Marcel den Dikken The Cambridge Handbook of Communication Disorders, edited by Louise Cummings The Cambridge Handbook of Stylistics, edited by Peter Stockwell and Sara Whiteley The Cambridge Handbook of Linguistic Anthropology, edited by N. J. Enfield, Paul Kockelman and Jack Sidnell The Cambridge Handbook of English Corpus Linguistics, edited by Douglas Biber and Randi Reppen The Cambridge Handbook of Bilingual Processing, edited by John W. Schwieter The Cambridge Handbook of Learner Corpus Research, edited by Sylviane Granger, Gaëtanelle Gilquin and Fanny Meunier The Cambridge Handbook of Linguistic Multicompetence, edited by Li Wei and Vivian Cook The Cambridge Handbook of English Historical Linguistics, edited by Merja Kytö and Päivi Pahta The Cambridge Handbook of Formal Semantics, edited by Maria Aloni and Paul Dekker The Cambridge Handbook of Morphology, edited by Andrew Hippisley and Greg Stump The Cambridge Handbook of Historical Syntax, edited by Adam Ledgeway and Ian Roberts The Cambridge Handbook of Linguistic Typology, edited by Alexandra Y. Aikhenvald and R. M. W. Dixon The Cambridge Handbook of Areal Linguistics, edited by Raymond Hickey The Cambridge Handbook of Cognitive Linguistics, edited by Barbara Dancygier The Cambridge Handbook of Japanese Linguistics, edited by Yoko Hasegawa
Published online by Cambridge University Press
The Cambridge Handbook of Spanish Linguistics, edited by Kimberly L. Geeslin The Cambridge Handbook of Bilingualism, edited by Annick De Houwer and Lourdes Ortega The Cambridge Handbook of Systemic Functional Linguistics, edited by Geoff Thompson, Wendy L. Bowcher, Lise Fontaine and David Schönthal The Cambridge Handbook of African Linguistics, edited by H. Ekkehard Wolff The Cambridge Handbook of Language Learning, edited by John W. Schwieter and Alessandro Benati The Cambridge Handbook of World Englishes, edited by Daniel Schreier, Marianne Hundt and Edgar W. Schneider The Cambridge Handbook of Intercultural Communication, edited by Guido Rings and Sebastian Rasinger The Cambridge Handbook of Germanic Linguistics, edited by Michael T. Putnam and B. Richard Page The Cambridge Handbook of Discourse Studies, edited by Anna De Fina and Alexandra Georgakopoulou The Cambridge Handbook of Language Standardization, edited by Wendy Ayres-Bennett and John Bellamy The Cambridge Handbook of Korean Linguistics, edited by Sungdai Cho and John Whitman The Cambridge Handbook of Phonetics, edited by Rachael-Anne Knight and Jane Setter The Cambridge Handbook of Corrective Feedback in Second Language Learning and Teaching, edited by Hossein Nassaji and Eva Kartchava The Cambridge Handbook of Experimental Syntax, edited by Grant Goodall The Cambridge Handbook of Heritage Languages and Linguistics, edited by Silvina Montrul and Maria Polinsky The Cambridge Handbook of Arabic Linguistics, edited by Karin Ryding and David Wilmsen The Cambridge Handbook of the Philosophy of Language, edited by Piotr Stalmaszczyk The Cambridge Handbook of Sociopragmatics, edited by Michael Haugh, Dániel Z. Kádár and Marina Terkourafi The Cambridge Handbook of Task-Based Language Teaching, edited by Mohammed Ahmadian and Michael Long The Cambridge Handbook of Language Contact: Population Movement and Language Change, Volume 1, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Language Contact: Multilingualism in Population Structure, Volume 2, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Romance Linguistics, edited by Adam Ledgeway and Martin Maiden The Cambridge Handbook of Translation, edited by Kirsten Malmkjaer
Published online by Cambridge University Press
Published online by Cambridge University Press
The Cambridge Handbook of Working Memory and Language Edited by John W. Schwieter Wilfrid Laurier University
Zhisheng (Edward) Wen Macao Polytechnic University
Published online by Cambridge University Press
University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108845342 DOI: 10.1017/9781108955638 © Cambridge University Press 2022 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2022 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Schwieter, John W., 1979- editor. | Wen, Zhisheng, editor. Title: The Cambridge handbook of working memory and language / edited by John W. Schwieter, Wilfrid Laurier University, Ontario ; Edward Z.S. Wen, Macao Polytechnic Institute. Description: Cambridge ; New York, NY : Cambridge University Press, 2022. | Series: Cambridge handbooks in language and linguistics | Includes bibliographical references and index. Identifiers: LCCN 2022001124 (print) | LCCN 2022001125 (ebook) | ISBN 9781108845342 (hardback) | ISBN 9781108958110 (paperback) | ISBN 9781108955638 (epub) Subjects: LCSH: Language acquisition–Psychological aspects. | Second language acquisition–Psychological aspects. | Language disorders in children. | Short-term memory. | Psycholinguistics. | BISAC: LANGUAGE ARTS & DISCIPLINES / Linguistics / General | LCGFT: Essays. Classification: LCC P118 .C365 2022 (print) | LCC P118 (ebook) | DDC 401/.93–dc23/ eng/20220113 LC record available at https://lccn.loc.gov/2022001124 LC ebook record available at https://lccn.loc.gov/2022001125 ISBN 978-1-108-84534-2 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.
Published online by Cambridge University Press
Contents
List of Figures List of Tables About the Editors About the Contributors Acknowledgments
page x xii xiv xvi xxxii
Overview of the Handbook 1 Working Memory and Language: An Overview of Key Topics
John W. Schwieter, Zhisheng (Edward) Wen, and Teresa Bennett Part I Introduction 2 Working Memory and the Challenge of Language
Alan Baddeley
3
19
Part II Models and Measures 3 The Evolution of Working Memory and Language 4 5 6 7 8 9 10
Frederick L. Coolidge and Thomas Wynn The Phonological Loop as a “Language Learning Device”: An Update Costanza Papagno The Embedded-Processes Model and Language Use Eryn J. Adams, Alicia Forsberg, and Nelson Cowan Long-Term Working Memory and Language Comprehension R. Lane Adams and Peter F. Delaney The Cognitive Neuroscience of Working Memory and Language Nina Purg, Anka Slana Ozimič, and Grega Repovš Computational Models of Working Memory for Language Graham J. Hitch, Mark J. Hurlstone, and Tom Hartley The Time-Based Resource Sharing Model of Working Memory for Language Valérie Camos and Pierre Barrouillet The Ease of Language Understanding Model Jerker Rönnberg, Emil Holmer, and Mary Rudner
Published online by Cambridge University Press
31 51 73 98 120 143 175 197
viii
CONTENTS
11 Assessing Children’s Working Memory Milton J. Dehn 12 Measuring Individual Differences in Working Memory Capacity and Attention Control and Their Contribution to Language Comprehension Alexander P. Burgoyne, Jessie D. Martin,
Cody A. Mashburn, Jason S. Tsukahara, Christopher Draheim, and Randall W. Engle Part III Linguistic Theories and Frameworks 13 Have Grammars Been Shaped by Working Memory and If So, How? John A. Hawkins 14 Branching and Working Memory: A Cross-Linguistic Approach Federica Amici, Alejandro Sanchez-Amaro,
and Trix Cacchione 15 Working Memory and Natural Syntax William O’Grady 16 The Role of Working Memory in Shaping Syntactic Dependency Structures Chunshan Xu and Haitao Liu 17 Working Memory in the Modular Cognition Framework
John Truscott and Michael Sharwood Smith 18 Short-Term and Working Memory Capacity and the Language Device: Chunking and Parsing Complexity Bingfu Lu and Zhisheng (Edward) Wen Part IV First Language Processing 19 Working Memory in Word Reading Sun-A Kim 20 The Role of Working Memory in Language Comprehension and Production: Evidence from Neuropsychology Rachel Zahn,
Autumn Horne, and Randi C. Martin 21 Working Memory and High-Level Text Comprehension Processes Ana I. Pérez Muñoz and M. Teresa Bajo 22 Working Memory and Speech Planning Benjamin Swets and Iva Ivanova 23 How Do Novice and Skilled Writers Engage Working Memory? Thierry Olive
219
247
275
304 322 343 368
393
421
435 459 482 504
Part V Bilingual Acquisition and Processing 24 How Measures of Working Memory Relate to L2 Vocabulary
Elisabet Service and Daphnée Simard
529
25 Working Memory and L2 Grammar Development in Children
Paul Leseman and Josje Verhagen 26 Working Memory and L2 Grammar Learning among Adults Timothy McCormick and Cristina Sanz 27 Working Memory and L2 Sentence Processing Ian Cunnings 28 Methodological Issues in Research on Working Memory and L2 Reading Comprehension Michael J. Leeser and Eric Herman
Published online by Cambridge University Press
550 573 593 613
Contents
29 Working Memory and Second Language Speaking Tasks
Peter Skehan
635
30 Working Memory in Second Language Interaction
Hyejin An
and Shaofeng Li 31 Working Memory and Interpreting Studies
656 Binghan Zheng
and Huolingxiao Kuang 32 A Methodological Synthesis of Working Memory Tasks in L2 Research Jihye Shin and Yuhang Hu
698 722
Part VI Language Disorders, Interventions, and Instruction 33 Specific Learning Disorders as a Working Memory Deficit
H. Lee Swanson
749
34 A New Perspective on the Connection between Memory and Sentence Comprehension in Children with Developmental Language Disorder James W. Montgomery, Ronald B. Gillam,
and Julia L. Evans 35 Working Memory and Childhood Deafness Gary Morgan 36 Working Memory Training in the Classroom
Tracy Packiam Alloway, Rachel K. Carpenter, Tyler Robinson, and Andrea N. Frankenstein 37 Working Memory and Classroom Learning Joni Holmes, Elizabeth M. Byrne, and Agnieszka J. Graham 38 Cognitive Load Theory and Instructional Design for Language Learning John Sweller, Stéphanie Roussel, and André Tricot 39 Working Memory Training Meta-Analyses and Clinical Implications Domenico Tullo and Susanne M. Jaeggi
776 801
820 835
859 881
Part VII Conclusion 40 Toward an Integrated Account of Working Memory and Language Zhisheng (Edward) Wen and
John W. Schwieter Index
Published online by Cambridge University Press
909 928
ix
Figures
2.1 The working memory model (Baddeley & Hitch, 1974) page 20 5.1 The embedded-processes model of working memory 75 7.1 Working memory and language-related brain regions and their structural connections 123 8.1 Illustration of simple (a) and compound (b) chaining models 151 8.2 Schematic of the architecture of context-free (a–c) and context-based (d–f ) CQ models and the steps involved in producing a three-item sequence 154 8.3 The central role of serial order in connecting detailed implementations of working memory in specific tasks with more abstract and general theories 159 8.4 Example of responses of syllabic phase model oscillators during processing of the sentence “Iguanas and alligators are tropical” 162 8.5 Phase and amplitude responses of a population of oscillators with different tunings (spaced between 0.1 Hz and 1.28 Hz) in the BUMP model of Hartley et al. (2016) 164 9.1 The cognitive architecture of working memory according to the TBRS model 180 10.1 The ELU-WM system 200 11.1 Working memory processes 220 12.1 The substantial and significant contribution of working memory capacity to fluid intelligence after accounting for short-term memory 253 12.2 A structural equation model depicting attention control fully mediating the relationship between working memory capacity and fluid intelligence 253 12.3 A structural equation model depicting task-unrelated thoughts partially mediating the relationship between attention control and reading comprehension 255
Published online by Cambridge University Press
List of Figures
12.4 Attention control supports information maintenance and disengagement in service of complex cognition 12.5 Reliability of a difference score (Y-axis) decreasing as the correlation between the component scores increases (X-axis) 14.1 Box plots representing the data distribution for the number of correctly recalled initial and final stimuli in the WM tasks 15.1 Direct mapping 15.2 Mediated mapping 16.1 Phrase structure and the dependency structure of a sentence 16.2 Long and short dependency distance in a sentence 18.1 Linguistic chunks 18.2 Procedure for obtaining DCs 18.3 Major-branch tree 18.4 The orbit-layering diagram 18.5 On-line chunking 18.6 Depth hypothesis 18.7 Comparison of MMCN, DMM, and W/IC ratio metrics 20.1 The domain-specific model of WM (adapted from Martin et al., 1999) 21.1 Reading times index (fourth sentence divided by the mean of the first three sentences or context, in milliseconds) 21.2 Electrophysiological activity (N400) for the disambiguating word, as a function of language, condition, and working memory (L2 divided by L1) 22.1 Model of language production (adapted from Levelt, 1989) 24.1 An example item in Word Associates Format 24.2 DKFVT item with the correct answers 30.1 Publication types included in the methodological synthesis 31.1 Mizuno’s (2005) model of interpreting based on Cowan’s (1988, 1995) Embedded-Processes Model of Memory 31.2 Dong and Li’s (2020) attentional control model of interpreting 31.3 Darò and Fabbro’s (1994) model of simultaneous interpreting 32.1 Change in the use of WM tasks over time 32.2 Change in the use of verbal and nonverbal WM tasks over time 32.3 Change in the use of simple and complex span tasks over time 34.1 Structural equation model of the direct relationships between cognitive processing and sentence comprehension 34.2 Structural equation model of the relationships between cognitive processing and sentence comprehension with working memory as a mediator
Published online by Cambridge University Press
257 259
313 324 324 344 347 397 399 399 400 401 405 405 436 471
472 484 532 532 664 700 701 703 728 728 729 788
789
xi
Tables
7.1 Event-related potentials (ERPs) related to language processes page 132 8.1 Benchmark findings of serial recall 150 11.1 Multibattery analysis completed worksheet example 237 12.1 Working memory tasks available for download from https://englelab.gatech.edu 263 12.2 Attention control tasks available for download from https://englelab.gatech.edu 266 13.1 Subject > Object and Object > Subject for full NPs in grammars (cf. Comrie, 2013) 293 13.2 Processing predictions for Subject > Object and Object > Subject Orders 295 18.1 First calculation of MMCN, DMM, and W/IC ratio metrics 407 18.2 Second calculation of MMCN, DMM, and W/IC ratio metrics 407 19.1 Summary of studies containing a WM variable in Chinese word reading 426 20.1 Examples of relative clause sentence types (adapted from Martin, 1987) 447 20.2 Examples of stimuli with syntactic and semantic interference (adapted from Tan & Martin, 2018) 452 21.1 Example of text used in the situation model revision task 465 22.1 Summary of evidence for the role of working memory in various stages of language production 485 23.1 Relations between writing processes and working memory components as proposed by Kellogg (1996) 508 23.2 An update on the relations between writing processes and working memory components in skilled writers 518 24.1 Correlations between versions of L1 pseudoword span and explicit memory for new L1 or L2 word forms (L1 was Finnish and L2 was Korean for young adults and eight-year-old children) 535
Published online by Cambridge University Press
List of Tables
24.2 Correlations between L2 word repetition and memory for intentionally or incidentally learned new word forms for young adults and eight-year-old children (L1 was Finnish and L2 Korean) 28.1 Comparison of L2 reading-WM studies that used written summary protocols 30.1 Predictor and criterion variables in the synthesis 30.2 Sample characteristics of included studies (K = 33) 30.3 Methodological features of included studies (K = 33) 32.1 Methodological features of the six most commonly used WM tasks 32.2 Scoring methods used in complex span tasks 32.3 Scoring methods used in simple span tasks 39.1 List of meta-analyses included and excluded from the current review 40.1 A unified understanding of working memory
Published online by Cambridge University Press
536 625 662 665 665 730 733 736 885 912
xiii
About the Editors
John W. Schwieter is a Professor of Spanish and Linguistics, CrossAppointed in Psychology, and the Director of Bilingualism Matters @ Laurier and the Language Acquisition, Multilingualism, and Cognition Laboratory at Wilfrid Laurier University in Canada. His research interests include psycholinguistic and neurolinguistic approaches to multilingualism and language acquisition; translation, interpreting, and cognition; and second language teaching and learning. He is Executive Editor of the Bilingual Processing and Acquisition series (Benjamins) and Coeditor of the Cambridge Elements in Second Language Acquisition series (Cambridge University Press, with Benati). His recent books include Introducing Linguistics: Theoretical and Applied Approaches (2021, Cambridge University Press), The Handbook of the Neuroscience of Multilingualism (2019, WileyBlackwell), and The Cambridge Handbook of Language Learning (2019, Cambridge University Press, with Benati). His forthcoming books include The Cognitive Neuroscience of Bilingualism (Cambridge University Press, with Festman), Introduction to Translation and Interpreting Studies (Wiley, with Ferreira), and The Routledge Handbook of Bilingualism, Translation, and Interpreting (Routledge, with Ferreira). Some of his research has appeared in journals such as Acta Psychologica, Bilingualism: Language and Cognition, Biological Psychology, Cerebral Cortex, Frontiers in Psychology, International Journal of Bilingualism, Language Learning, The Mental Lexicon, Psychophysiology, among others. Zhisheng (Edward) Wen is an Associate Professor in the School of Languages and Translation at Macao Polytechnic University. His research interests include second language acquisition, task-based language teaching and learning, psycholinguistics, and cognitive science, with a particular focus on the roles of working memory and language aptitude in language sciences and bilingualism research. His recent books include Working Memory in Second Language Acquisition and Processing (2015, Multilingual Matters, with Borges Mota & McNeill), Working Memory and Second
Published online by Cambridge University Press
About the Editors
Language Learning (2016, Multilingual Matters), Language Aptitude (2019, Routledge, with Skehan et al.), and Researching L2 Task Performance and Pedagogy (2019, Benjamins, with Ahmadian). His forthcoming books include Cognitive Individual Differences in Second Language Acquisition (de Gruyter, with Biedroń, Teng, & Sparks), Language Aptitude Theory and Practice (Cambridge University Press, with Skehan & Sparks), Memory in Science for Society (Oxford University Press, with Logie, Gathercole, Cowan, & Engle), and Working Memory in First and Second Language (Cambridge University Press, with Baddeley & Cowan). Some of his publications have appeared in Annual Review of Applied Linguistics, Language Teaching, International Journal of Applied Linguistics, and The Language Learning Journal, among others.
Published online by Cambridge University Press
xv
About the Contributors
The contributors to this handbook are international experts based at and/or affiliated with institutions and research centers in Australia, Canada, China (mainland), England, France, Germany, Greece, Hong Kong, Italy, Macao, Northern Ireland, Scotland, Slovenia, Spain, Sweden, Switzerland, Taiwan, The Netherlands, and the United States. Below are brief bios about these contributors: Eryn J. Adams is a PhD candidate and member of the Working Memory Lab at the University of Missouri in the United States. Her research interests include the development of working memory and language in young children. Some of her recent publications have appeared in the Journal of Cognition and Development and Language, Speech, and Hearing Services in Schools. R. Lane Adams is a graduate student at the University of North Carolina at Greensboro in the United States. His research interests include memory and expertise. Tracy Packiam Alloway is a Professorof Psychology and the Director of the Working Memory Lab at the University of North Florida in the United States. Her research interests include the role of working memory in mental health, decision-making, and education. Some of her significant publications have appeared in Child Development and the Journal of Experimental Child Psychology. Her recent book, Think like a girl (2021, Zondervan), explores the myths about the female brain. Federica Amici is a postdoctoral researcher at the Department of Human Behavior, Ecology and Culture at the Max Planck Institute for Evolutionary Anthropology and at the Leipzig University in Germany. Her main research interests lie in the evolutionary forces explaining behavioral and cognitive variation in humans and other species, including the role of language and culture in shaping human behavioral diversity. Her research has been published in several journals including
Published online by Cambridge University Press
About the Contributors
Current Biology, Proceedings of the National Academy of Sciences of the United States of America, and Nature Communications. Hyejin An is a PhD student in Foreign and Second Language Education at Florida State University in the United States. Her research interests include the role of learners’ cognitive and affective individual difference in language learning and the psycholinguistic aspects of foreign language written production. Alan Baddeley is an Emeritus Professor of Psychology at the University of York in England. He is best known for his development with Graham Hitch of the multicomponent model of working memory some 40 years ago. The model continues to be refined with an overview published in 2019 in Memory and Cognition and new developments outlined in his book Working Memories: Postmen, Divers, and the Cognitive Revolution (2019, Routledge). M. Teresa Bajo is a full professor at the Department of Experimental Psychology in the University of Granada. She is head of the Memory and Language research group. Her research interests include memory retrieval in young and older adults, and language processing and control in bilinguals and monolinguals. Some of her recent publications have appeared in Cognition, Neurobiology of Learning and Memory, Bilingualism: Language and Cognition, and the Journal of Experimental Psychology: Learning, Memory and Cognition. Pierre Barrouillet is an Honorary Professor of Cognitive Development and the Former Director of the Archives Jean Piaget at the University of Geneva in Switzerland. His research interests include working memory, reasoning, and mathematical cognition. He developed the Time-Based Resource Sharing model of working memory with Valérie Camos and coauthored with her several books and articles in leading scientific journals on working memory structure, functioning, and development. Teresa Bennett is a Research Assistant to coeditor Schwieter and a fourtime recipient of the Dean’s Honour Roll within the Faculty of Science at Wilfrid Laurier University in Canada. Throughout her time as an undergraduate student in the Health Sciences program, she has conducted research pertaining to both the social and physiological aspects of health. This includes studies on innate immune stimulants as cancer therapies, as well as how the social determinants of health, such as racism, discrimination, stigma, social exclusion, and the like, negatively impact the health and well-being of racialized/marginalized groups. Her interest in language and the mind primarily pertains to examining the cognitive benefits of bi- and multilingualism as she enjoys the field of cognitive and behavioral neuroscience and hopes to pursue a career in the field of healthcare in the future. Alexander P. Burgoyne is a Research Scientist II in the Attention and Working Memory Lab at the Georgia Institute of Technology in the United States. His research interests include individual differences in
Published online by Cambridge University Press
xvii
xviii
ABOUT THE CONTRIBUTORS
intelligence, experience, motivation, and personality, and their relationships to real-world outcomes such as academic achievement and job performance. Some of his recent publications have appeared in Current Directions in Psychological Science, Psychological Science, Cognition, and Psychological Bulletin. Elizabeth M. Byrne is a dual-affiliated Research Associate at the Centre for Play in Education, Development, and Learning at the University of Cambridge in England, and at the LEGO Foundation in Denmark. Her research interests include cognitive development, play-based learning, and educational interventions in early childhood. Some of her recent publications have appeared in Brain & Cognition and Child: Care, Health, & Development. She recently published a book chapter entitled “Cognitive plasticity and transcranial electrical stimulation” in Cognitive Training: An Overview of Features and Applications (2021, Springer, edited by Strobach & Karbach). Trix Cacchione is a Professor at the University of Applied Sciences and Arts Northwestern Switzerland. She is specialized in developmental psychology, with a focus on conceptual development. Her research interests span from psycholinguistics to comparative psychology and include crossspecific work aiming to trace the evolutionary origins of human cognition. Valérie Camos is a Full Professor in Cognitive Development and the Director of the Working Memory Development Lab at the University of Fribourg in Switzerland. Her research interests include the role of attention in working memory and the relationships between long-term and working memory in childhood and adulthood. Her work led to the development of the Time-Based Resource Sharing model with Pierre Barrouillet. In addition to several books, her significant publications have appeared in the Journal of Experimental Psychology: General, Journal of Memory and Language, Developmental Psychology, and Current Directions in Psychological Science. Rachel K. Carpenter is a PhD candidate in the Clinical Psychology Doctoral Program at East Tennessee State University in the United States. She completed her master’s degree in Psychological Science at the University of North Florida with a focus on working memory and augmented reality. She approaches the study of digital games and other media using a clinical approach, hoping to use advanced technology to reduce symptomology in a variety of clinical pathologies. Her published work has focused on working memory and testing modalities, augmented reality and mood, and the ecological considerations of intimate partner violence. Her current research explores working memory, intimate partner violence, suicide, and severe mental health considerations. Frederick L. Coolidge is a Professor and the Codirector of the University of Colorado, Colorado Springs Center for Cognitive Archaeology in the United States. He received his BA, MA, and PhD in Psychology and completed a two-year Postdoctoral Fellowship in Clinical
Published online by Cambridge University Press
About the Contributors
Neuropsychology at the University of Florida. He has received three Fulbright Fellowships, three outstanding teaching awards, the lifetime title University of Colorado Presidential Teaching Scholar, and two research excellence awards. Recently, he published his 10th book, Evolutionary Neuropsychology: An Introduction to the Evolution of the Structures and Functions of the Human Brain (2020, Oxford University Press). Nelson Cowan is Curators’ Distinguished Professor at the University of Missouri in the United States. His interests include working memory, its relation to selective attention, and its development. Based on his embedded-processes view of working memory, collaborations have explored working-memory factors in language impairment, dyslexia, autism, schizophrenia, amnesia, and Parkinson’s disease. Two recent publications in Perspectives on Psychological Science report on a theory of cognitive growth, and on the process of adversarial collaboration based on a working memory project. His coedited volume Working Memory: State of the Science (2020, Oxford University Press, with Logie & Camos) summarizes current working memory theories. Ian Cunnings is an Associate Professor of Psycholinguistics in the School of Psychology and Clinical Language Sciences at the University of Reading in England. His research interests are in language comprehension in different populations of speakers, with a focus on the working memory operations that subserve sentence processing. Much of his recent research has examined the similarities and differences between native and nonnative language comprehension. His recent publications have appeared in journals such as the Journal of Memory and Language, Applied Psycholinguistics, and Bilingualism: Language and Cognition. Milton Dehn was an Associate Professor and School Psychology Program Director at the University of Wisconsin–La Crosse in the United States until his retirement. Currently, he is the test development project director for Schoolhouse Educational Services. His interests include assessment of cognitive abilities, memory, dyslexia, and executive functions and using a pattern of strengths-and-weaknesses approach to the identification of specific learning disabilities. His publications on working memory include Working Memory and Academic Learning: Assessment and Intervention (2008, Wiley) and Essentials of Working Memory: Assessment and Intervention (2015, Wiley). Peter F. Delaney is a Professor of Psychology and the Director of the Cognition, Learning, and Memory Lab at the University of North Carolina at Greensboro in the United States. His research interests include expertise, intentional forgetting, and spacing effects in memory. His recent work has appeared in outlets such as Clinical Psychology Review, Memory, and the Journal of Expertise. Christopher Draheim is a Visiting Assistant Professor at Lawrence University in the United States. His research interests include the measurement and nature of individual differences in executive
Published online by Cambridge University Press
xix
xx
ABOUT THE CONTRIBUTORS
functioning, particularly working memory capacity and attention control. Some of his recent publications in this area have appeared in Psychological Bulletin and Journal of Experimental Psychology: General. Randall W. Engle is a Professor in the School of Psychology at the Georgia Institute of Technology in the United States. His research is on individual differences in attention control, working memory capacity, and fluid intelligence and the role those differences play in performing complex tasks. He is a member of the National Academy of Sciences. Julia L. Evans is a Professor in the Department of Speech, Language, and Hearing and Director of the Child Language and Cognitive Processes Lab at the University of Texas–Dallas in the United States. The focus of her research program is the neurobiology of language, cognitive processing, implicit learning, and memory in children with developmental language disorder. Her publications appear in a wide range of language, cognitive psychology, and neuroscience journals. Alicia Forsberg is a Postdoctoral Research Fellow at the University of Missouri in the United States, where she is part of Professor Nelson Cowan’s research group. She received her PhD in 2019 from the University of Edinburgh in Scotland. She is interested in the lifespan development of working memory, focusing on the causes and consequences of these lifespan changes. Some of her recent research on the relationship between working and long-term memory is published in Psychonomic Bulletin & Review. Andrea N. Frankenstein is a graduate student in Psychology at the University of Illinois at Chicago in the United States, where she specializes in cognitive psychology and statistics, methods, and measurement. Her broad research interests include cognitive and socialcognitive factors that influence learning outcomes. She is especially interested in applying this work in the context of higher education. Some of her recent publications have appeared in Cognition and Memory & Cognition. Ronald B. Gillam holds the Raymond and Eloise Lillywhite Endowed Chair in Speech-Language Pathology and is the Director of the Interdisciplinary Doctoral Program in Neuroscience at Utah State University in the United States. His research interests include the neurobiology of language and cognition as well as evidence-based practices for children with developmental language disorders. Some of his recent publications have appeared in Brain and Cognition, Human Brain Mapping, PLoS ONE, and the Journal of Speech, Language, and Hearing Research. Agnieszka J. Graham is a Lecturer in Applied Developmental Psychology at the Queen’s University of Belfast in Northern Ireland. She studies higher-level cognition and its development across the lifespan and is particularly interested in new and original approaches to tackling educational underachievement. The goal of her work is to understand how executive functions develop and how their development can be best
Published online by Cambridge University Press
About the Contributors
supported. She is currently working on a project exploring mind wandering in children, examining its nature, and considering the costs and benefits that mind wandering can bring in the context of educational attainment. Tom Hartley is a neuroscientist and Senior Lecturer in Psychology at the University of York in England. His research interests include brain mechanisms of perception, memory, and complex behavior, which he investigates with neuroimaging, computational modeling, neuropsychology, and experimental psychology. His PhD thesis at University College London developed a model of linguistic constraints on serial order in verbal short-term memory. After remaining at UCL for a series of postdoctoral positions focusing on the neuroscience of spatial memory, he moved to his current role in 2005, where his research continues on a broad range of topics, including studies of timing in auditory verbal short-term memory. Jack (John A.) Hawkins is a Distinguished Professor of Linguistics at the University of California, Davis, in the United States and Emeritus Professor of English and Applied Linguistics at Cambridge University in England. He has held previous positions at the University of Southern California, the Max Planck Institute for Psycholinguistics in Nijmegen, and the University of Essex. He has broad interests in the language sciences and has published widely on typology and universals, efficiency and complexity in grammars and usage, language processing and learning, the Germanic language family, and language change. Eric Herman is a PhD student at Florida State University in the United States. His research interests include second language reading comprehension and instructed second language learning. Graham J. Hitch is an Emeritus Professor of Psychology at the University of York in England. He completed his first degree in physics and switched to experimental psychology through an MSc at the University of Sussex. He went on to do a PhD on short-term memory at the Medical Research Council Applied Psychology Unit (now called the Medical Research Council Cognition and Brain Sciences Unit) at Cambridge University. This was followed by a postdoctoral fellowship with Alan Baddeley investigating working memory at the University of Sussex and University of Stirling. He has continued this life-long interest ever since, first back at the MRCAPU, then at the universities of Manchester and Lancaster and finally at the University of York, where he is presently an Emeritus Professor of Psychology. Emil Holmer is a Senior Lecturer of Disability Research and the Head of Division at the Disability Research Division at Linköping University in Sweden. His research interests include behavioral, cognitive, and neural aspects of language processing and development in populations with and without disabilities. Some of his recent publications have appeared in Frontiers in Psychology and Cerebral Cortex.
Published online by Cambridge University Press
xxi
xxii
ABOUT THE CONTRIBUTORS
Joni Holmes is a Senior Scientist and Head of the Centre for Attention Learning and Memory at the Medical Research Council’s Cognition & Brain Sciences Unit at the University of Cambridge in England. Her research interests center around understanding the cognitive mechanisms that give rise to developmental difficulties. Some of her significant publications have appeared in Current Biology, Developmental Science, Journal of Educational Psychology, Journal of Memory and Language, the Journal of the American Academy of Child and Adolescent Psychiatry. Autumn Horne is a PhD Student working under the guidance of Dr. Randi Martin in the Department of Psychological Sciences at Rice University in the United States. Her research interests include individual differences in working memory and their relationship with language comprehension, production, and prediction. Yuhang Hu is a PhD student in applied linguistics at Northern Arizona University. Her research interests include quantitative research methods in applied linguistics research, research synthesis and meta-analysis, and individual differences (particularly psychological variables) in bi/ multilingualism research. Mark J. Hurlstone is a Lecturer in the Department of Psychology at Lancaster University in England and an Honorary Research Fellow in the School of Psychological Science at the University of Western Australia. His research interests include working memory for order information and its relation to language processing, computational and mathematical modeling of cognitive processes, and the psychology of climate change, vaccination, and misinformation. His research on working memory has featured in top-tier psychology journals including Cognitive Psychology, Journal of Experimental Psychology: Learning, Memory, and Cognition, Journal of Experimental Psychology: Human Perception and Performance, and Psychological Bulletin. Iva Ivanova is an Assistant Professor of Psychology and the Director of the Language and Communication Lab at the University of Texas at El Paso in the United States. Her research interests include bilingualism, language production and dialogue, and their relationships with working memory and attention. A current project explores how bilinguals’ control of which language they speak when impacts the quality and fluency of spontaneous connected speech. Some of her recent publications have appeared in Cognition, Journal of Memory and Language, Journal of Experimental Psychology: Learning, Memory and Cognition, and Bilingualism: Language and Cognition. Susanne M. Jaeggi is a Professor in Education and Cognitive Science and Director of the Working Memory & Plasticity Lab at the University of California, Irvine, in the United States. She studies working memory and related cognitive functions across the lifespan, and within that domain, her major work focuses on the development of cognitive interventions and the investigation of whether, how, and for whom those interventions
Published online by Cambridge University Press
About the Contributors
generalize to nontrained cognitive domains. Her work has appeared in PNAS, Psychological Science, and Developmental Science. Sun-A Kim is an Associate Professor of Second Language Acquisition and Teaching Chinese as a Foreign Language in the Department of Chinese and Bilingual Studies at the Hong Kong Polytechnic University. Her research interests include individual differences in language processing and learning, psycholinguistic approaches to second language acquisition, and teaching Chinese and Korean as second languages. Huolingxiao Kuang is a PhD candidate at the School of Modern Languages and Cultures, Durham University in England. She holds an MA in Translation and Interpreting from Peking University. She has been working as a part-time interpreter, providing interpreting services for the Development Research Center of China State Council, the Ministry of Health in Zambia, Mercedes Benz, and Peking University. Her research interests include interpreting process research and note-taking in consecutive interpreting. Michael J. Leeser is an Associate Professor of Spanish and Linguistics and Director of Spanish Basic Language Instruction at Florida State University in the United States. His research interests include second language comprehension, input processing, and processing instruction. He is the coeditor of the recent volume Research on Second Language Processing and Processing Instruction (2021, Benjamins, with Keating & Wong). Paul Leseman is a Psychologist and Full Professor of Educational Sciences at Utrecht University in The Netherlands, where he leads a research unit focusing on early and middle childhood language, literacy, mathematics, and executive function development in relation to school achievement in multicultural and multilingual contexts. He is leader of national cohort studies on child development and coordinated European research projects among immigrant and ethnic minority communities on equity and inclusion. Recent work has appeared in journals such as Bilingualism: Language and Cognition, Child Development, Developmental Psychology, Learning and Instruction, and Review of Educational Research. Shaofeng Li is an Associate Professor of Foreign and Second Language Education at Florida State University in the United States. He received his PhD from Michigan State University. His main research interests include second language acquisition, working memory, language aptitude, and the interface between cognitive and affective learner differences and instruction type. His publications have appeared in Applied Linguistics, Applied Psycholinguistics, Language Learning, Language Teaching, Language Teaching Research, Modern Language Journal, and Studies in Second Language Acquisition, among others. Haitao Liu is a Qiushi Distinguished Professor of Linguistics and Applied Linguistics at Zhejiang University and a Yunshan Leading Visiting Professor at Guangdong University of Foreign Studies in China. He is a member of the Esperanto Academy. His research interests include
Published online by Cambridge University Press
xxiii
xxiv
ABOUT THE CONTRIBUTORS
computational cognitive science, quantitative linguistics, dependency grammar, and digital humanities. He is the author of more than 230 scientific publications with some recent research work appearing in Cognitive Linguistics, Language Sciences, Journal of Chinese Linguistics, Journal of Second Language Writing, Physics of Life Reviews, Frontiers in Psychology, Folklore, Digital Scholarship in the Humanities, Muttersprache, Français moderne, and Вопросы языкознания. Bingfu Lu is a Professor of Linguistics at Beijing Language and Culture University in China. His research interests include linguistic typology, syntax, and applied linguistics. Some of his publications in English have appeared in the Proceedings of the Third International Conference on Cognitive Science and Cahiers linguistique–Asie orientale and in books published by Hong Kong University Press and de Gruyter. Jessie D. Martin is a Senior Human Factors Psychologist with Battelle Memorial Institute. Her research interests include the measurement of attention control in applied contexts including linguistics and second language learning. Some of her recent publications in this area have appeared in the Journal of Experimental Psychology: General and the Journal of Applied Research in Memory and Cognition. Randi C. Martin is the Elma Schneider Professor of Psychological Sciences and Director of the T. L. L. Temple Foundation Neuroplasticity Lab at Rice University in the United States. Her research interests include the psychology and neuropsychology of language, with a focus on the role of working memory and executive function in language comprehension and production. A recent interest is whether the knowledge/access distinction for semantic memory is warranted. Her work has drawn on behavioral and neural findings from healthy and brain-damaged individuals. Significant recent publications have appeared in Brain, Cerebral Cortex Communications, Cortex, Current Directions in Psychological Science, and Psychonomic Bulletin & Review. Cody A. Mashburn is a PhD student in Cognition and Brain Science at the Georgia Institute of Technology in the United States. His research interests include individual differences in attention control, working memory capacity, processing speed, and intelligence, as well as applications of psychological testing. His research has appeared in journals such as Psychological Bulletin, Journal of Applied Research in Memory and Cognition, and Journal of Experimental Psychology: General. Timothy McCormick is a Language Data Researcher with Amazon’s Alexa. He completed his PhD in Spanish Linguistics at Georgetown University in Washington, DC, in the United States, where he focused on cognitive control and early and emergent bilingual sentence processing. His research interests include bilingualism and cognitive capacity, with publications in Studies in Second Language Acquisition and in volumes such as Research on Second Language Processing and Processing Instruction (2021,
Published online by Cambridge University Press
About the Contributors
Benjamins) and The Routledge Handbook of Study Abroad Research and Practice (2018, Routledge). James W. Montgomery is a Professor of Communication Sciences and Disorders and Director of the Developmental Psycholinguistics Lab at Ohio University in the United States. His research interests focus on understanding the intersection of working memory, long-term memory, and sentence comprehension in school-age children with Developmental Language Disorder. His publications appear in a variety of disciplinespecific journals as well as language and psychology journals. Gary Morgan is a Professor of Psychology at the Department of Language Communication Science at City University of London in England. He has researched a range of topics linked to childhood deafness including language development, theory of mind, and executive functions. His current research looks at the impact of early experiences on later language and cognitive outcomes in children. Some of his recent significant papers have appeared in Pediatrics, Child Development, and Infant Behaviour and Development. William O’Grady is a Professor in the Department of Linguistics at the University of Hawai‘i at Mānoa in the United States. He is well known for his writings on syntactic theory and language acquisition, as well as for his work on Korean and Jejueo. A major theme in his research is his commitment to emergentism, the idea that language is a complex system whose properties derive from the interaction for more basic factors and forces, especially processing pressures. He is author of numerous journal articles and several books including How Children Learn Language (2005, Cambridge University Press), Syntactic Carpentry, and Jejueo: The Language of Korea’s Jeju Island (2019, University of Hawai‘i Press, with Yang and Yang). He is the coeditor of The Handbook of Language Emergence (2014, WileyBlackwell, with MacWhinney) and the coeditor of a widely used textbook, Contemporary Linguistic Analysis (2021, Pearson, with Archibald), now in its ninth edition. Thierry Olive is a Senior Researcher at the Centre National de la Recherche Scientifique (CNRS) in France and Director of the Maison des Sciences de l’Homme et de la Société, a federative lab at the University of Poitiers. He is author of recent articles published in Reading and Writing, Frontiers in Psychology, Reading Research Quarterly, Written Communication, and Journal of Writing Research, and of the book Executive Functions in Writing (2021, Oxford University Press, with Teresa Limpo). He is also coeditor of the book series Studies in Writing (Brill). Anka Slana Ozimič is an Assistant Professor in Cognitive Science at the University of Ljubljana in Slovenia. She holds a PhD in Neuroscience and a master’s in Cognitive Science. She is active in the field of cognitive neuroscience and uses a combination of behavioral and fMRI experiments to study cognitive processes related to working memory and cognitive
Published online by Cambridge University Press
xxv
xxvi
ABOUT THE CONTRIBUTORS
control. Her recent publications focusing on systems limiting working memory capacity have appeared in the Journal of Memory and Language. Costanza Papagno is a Neurologist and Full Professor of Neuropsychology at the University of Milano-Bicocca and the Director of the Neurocognitive Rehabilitation Center at the University of Trento in Italy. Her research interests include verbal short-term memory and language, neuropsychological deficits in low-grade glioma patients, and the abstract/concrete dissociation. Some of her significant publications on verbal short-term memory have appeared in Cortex, Human Brain Mapping, Journal of Cognitive Neuroscience, Journal of Memory and Language, Memory, and Psychological Review. Ana I. Pérez Muñoz is a currently Principal Investigator of a MSCACOFUND Athenea3i project, working at the University of Granada in Spain. Her main research interest focuses on the study of high-level cognitive processes underlining text comprehension (such as inference making, comprehension monitoring, updating information, and information integration) in monolingual and bilingual children, young adults, and elderly people. Some of her significant publications have appeared in Memory and Cognition, Bilingualism: Language and Cognition, and Neuropsychologia. Nina Purg is a PhD Candidate in Neuroscience and a Teaching Assistant in Cognitive Science at the University of Ljubljana in Slovenia. She obtained a BSc degree in Biomedical Sciences and Synthetic Organic Chemistry and an MSc degree in Neuroscience from the University College London. Her research work focuses on studying human cognition, particularly executive functions, with a combination of behavioral and neuroimaging experimental approaches, such as fMRI and EEG. Grega Repovš is a Professor in General Psychology at the University of Ljubljana in Slovenia, where he leads the Mind and Brain Laboratory in the Department of Psychology. He holds a PhD and MSc in Psychology. His research focuses on the integration of behavioral, EEG, and fMRI methods in studying human cognition, with a specific focus on working memory, cognitive control, and integration of brain function in health and disease. His work includes the development of neuroimaging analytics with an emphasis on brain functional connectivity. His recent publications have appeared in the Journal of Memory and Language, Biological Psychiatry, Neuroimage, and Psychophysiology. Tyler Robinson is a PhD student with the Cognitive Neuroscience and Affective Psychopathologies (CNAPs) lab at Louisiana State University in the United States. His research interests are working memory, executive function, and cognition of aging using functional neuroimaging. His work in the CNAPs Lab has centered on cognitive and attentional emotion regulation. Following completion of his degree, he will begin a postdoctoral position at the University of Toronto in Canada to study white-matter changes in aging as they relate to cognitive performance.
Published online by Cambridge University Press
About the Contributors
Jerker Rönnberg is a Professor Emeritus of Psychology at Linköping University in Sweden. He has been the Director of the Swedish Institute for Disability Research (SIDR) for the last 20 years and continues to pursue his research at the Linnaeus Centre HEAD (HEaring And Deafness). With a background in memory research, he has recently published in Nature Communications, Frontiers in Systems Neuroscience, and Cerebral Cortex with a focus on hearing impairment and deafness, signed language, speech understanding in noise, signal processing in hearing instruments, and working memory capacity. He has contributed to the establishment of Cognitive Hearing Science as a new field of research. Stéphanie Roussel is an Associate Professor of applied linguistics and the Director of a Language Department at the University of Bordeaux in France. Her research interests include second language learning and teaching, second language acquisition, computer assisted language learning, and content and language integrated learning. Mary Rudner spent twenty years studying the nexus of cognition and communication, especially in relation to hearing loss and deafness as well as developmental processes. Her early work in collaboration with Rönnberg showed greater recruitment of superior parietal cortex when working memory tasks are performed in sign language rather than speech. More recent research with Rönnberg, Holmer, and others shows that regions of auditory cortex that process sound in hearing individuals are recruited during working memory tasks in profoundly deaf individuals. Rudner has retired from academic work and devotes her time to family, music, and art. Alejandro Sanchez-Amaro is a Postdoctoral Researcher at the Department of Comparative Cultural Psychology at the Max Planck Institute for Evolutionary Anthropology in Germany. His main research interests include the comparative study of humans and great apes with a special emphasis on how primates solve cooperative social dilemmas from a dyadic and a group-level perspective. He is also interested in the psychological mechanisms underlying the nature of primates’ decisionmaking biases and in the relationship between language and memory across cultures. Some of his recent publications have appeared in Proceedings of the Royal Society B: Biological Sciences and Evolution and Human Behaviour. Cristina Sanz is Full Professor of Spanish and Linguistics at Georgetown University in the United States, where she holds several administrative positions. She has published over 100 volumes, articles, and book chapters on the multiple interactions between learning context and individual differences, including working memory, in the development of multilingualism across the lifespan. Some of her recent publications have appeared in Neuropsychologia, Applied Psycholinguistics, Studies in Second Language Acquisition, Bilingualism: Language and Cognition, and Language Learning. Currently, she is coediting two volumes on methods in study
Published online by Cambridge University Press
xxvii
xxviii
ABOUT THE CONTRIBUTORS
abroad research (Benjamins) and on the teaching of Spanish (Wiley) and has recently coauthored Introducción a la lingüística hispánica (2020, Cambridge University Press). John W. Schwieter (see “About the editors”). Elisabet Service is a Professor and member of ARiEAL (Centre for Advanced Research in Experimental and Applied Linguistics) at McMaster University in Canada. She is Director of the Language, Memory, and Brain Lab, which employs behavioral and psychophysiological methods. Her research interests include working memory in language acquisition and processing, dyslexia, developmental language disorder, and cognitive load in bilingual task performance. She has published in the Journal of Memory and Language, Applied Psycholinguistics, Cerebral Cortex, Frontiers in Human Neuroscience, Frontiers in Psychology, among many others. Best known is her research of individual differences in secondlanguage acquisition. Michael Sharwood Smith is an Emeritus Professor at Heriot-Watt University in Scotland and a EUROSLA Distinguished Scholar. His current major research interest is in identifying the role of language representation, processing, and development (of any kind) within human cognition as a whole. He is currently working on a long-term project with John Truscott developing a cross-disciplinary framework called the Modular Cognition Framework. He is founding editor emeritus of Second Language Research and a former vice president of the European Second Language Association. His 13 books include the Internal Context of Bilingual Processing (2019, Benjamins, with Truscott), Introducing Language and Cognition: A Map of the Mind (2017, Cambridge University Press), The Multilingual Mind: A Modular Processing Perspective (2014, Cambridge University Press, with Truscott), and Second Language Learning: Theoretical Foundations (1994, Pearson). Jihye Shin is a Research Associate in the Systems Development & Improvement Center at the University of Cincinnati in the United States. She holds a PhD in Applied Linguistics from Northern Arizona University. Her research interests lie in second language reading, psycholinguistics, second language acquisition, TESOL, and research methods. Her recent publications on working memory and L2 reading have appeared in TESOL Quarterly, Applied Psycholinguistics, and in the edited volume Challenges in Language Testing around the World: Insights for Language Test Users (2021, Springer). Daphnée Simard is a Full Professor of second language acquisition in the Department of Linguistics at the Université du Québec à Montréal in Canada, where she was the director of the Institute of Cognitive Science from 2014 to 2017. She is currently one of the two editors-in-chief of the Canadian Modern Language Review. Her research interests are twofold. First, she investigates the role played by individual variables such as attentional capacity and memory in second language acquisition. She is
Published online by Cambridge University Press
About the Contributors
also interested in the relationship between metalinguistic behavior and second language acquisition. Her work has appeared in Bilingualism: Language and Cognition and Language Learning, among others. Peter Skehan is an Honorary Research Fellow in the Institute of Education at University College London in England. His interests are in second language acquisition, task-based learning and teaching, second language spoken performance, and foreign language aptitude. He has published in the Modern Language Journal, the Journal of Instructed Second Language Acquisition, the Annual Review of Applied Linguistics, System, as well as the book Second Language Task-Based Performance: Theory, Research, Assessment (2018, Routledge). H. Lee Swanson is a Research Professor in Educational Psychology at the University of New Mexico in the United States. He was previously Distinguished Professor and Peloy Endowed Chair of Educational Psychology at the University of California, Riverside. His primary research focuses on cognitive development in children at risk for learning disabilities. He has been the previous editor of the Journal of Learning Disabilities and Learning Disability Quarterly. He has published in several journals such as Developmental Psychology, Journal of Educational Psychology, Journal of Experimental Child Psychology, Intelligence, among others. John Sweller is an Emeritus Professor of Educational Psychology at the University of New South Wales in Australia. His research is associated with cognitive load theory. The theory is a contributor to both research and debate on issues associated with human cognition, its links to evolution by natural selection, and the instructional design consequences that follow. Based on many hundreds of randomized, controlled studies carried out by many investigators from around the globe, the theory has generated a large range of novel instructional designs from our knowledge of human cognitive architecture. Benjamin Swets is a Professor of Psychology at Grand Valley State University in the United States. His research interests include planning in language production, individual differences in working memory and language processing, and the effects of real-world pressures on language comprehension and production. Some of his recent publications have appeared in Frontiers in Psychology, Cognitive Science, and the Journal of Memory and Language. André Tricot is a Professor of Cognitive Psychology at Paul Valéry University of Montpellier in France and a researcher at the Epsylon lab. He obtained his PhD in cognitive psychology at Aix-Marseille University in 1995. His research investigates the relationships between natural and artificial memories in which he examines how the design of an artificial memory (a document) can help the natural memory instead of overloading it. Applications are in instructional design, human–computer interaction, ergonomics, and transport safety. He was the head of the group that designed the current curricula for grades 1, 2, and 3 in France.
Published online by Cambridge University Press
xxix
xxx
ABOUT THE CONTRIBUTORS
John Truscott is an Emeritus Professor at National Tsing Hua University in Taiwan. His primary research interest is in development of the Modular Cognition Framework (formerly known as MOGUL). He has published extensively on this and related topics in cognitive science, as well as in various areas of second language learning and teaching, notably including error correction. He is the author of Consciousness and Second Language Learning (2015, Multilingual Matters), Working Memory and Language in the Modular Mind (in press, Routledge) and coauthor of The Multilingual Mind: A Modular Processing Perspective (2014, Cambridge University Press, with Sharwood Smith) and The Internal Context of Bilingual Processing (2019, Benjamins, with Sharwood Smith). Jason S. Tsukahara is a PhD Student in Cognition and Brain Science at the Georgia Institute of Technology in the United States. His research interests include understanding the nature of individual differences in attention control, the role of the locus coeruleus in intelligence, and tracking the focus of attention using behavioral, physiological, and eyetracking measures. Some of his recent publications have appeared in Cognition and the Journal of Experimental Psychology: General. Domenico Tullo is a Postdoctoral Research Associate at the Centre Hospitalier Universitaire Sainte-Justine, affiliated with l’Université de Montréal in Canada. Domenico is currently examining the predictive validity of genetic and EEG biomarkers in response to an attention training intervention. More specifically, his area of research assesses effectiveness of training attention for children and adolescents with autism spectrum disorder and other neurodevelopmental conditions. Domenico’s doctoral research, which examined the feasibility and efficacy of training attention in children and adolescents with alternative learning profiles has been published in Developmental Science, Intelligence, and Journal of Vision. Josje Verhagen is an Associate Professor of Dutch Linguistics at the University of Amsterdam in the Netherlands. Her work focuses on language acquisition, with special reference to bilingualism in children. She is particularly interested in relationships between language and other aspects of cognition (memory and attention), statistical learning, and effects of the linguistic environment on acquisition. Some of her recent publications have addressed nonword repetition in monolingual and bilingual children, effects of input quantity and input quality on bilingual acquisition, and the role of sociopragmatic cues (e.g., eye gaze and pointing) in monolingual and bilingual children’s language learning. Zhisheng (Edward) Wen (see “About the editors”). Thomas Wynn is a Distinguished Professor of Anthropology at the University of Colorado, Colorado Springs, in the United States. In the 1970s his research opened a hitherto unexplored direction in Palaeolithic studies, the explicit use of psychological theory to interpret archaeological remains. He has published extensively (150+ books,
Published online by Cambridge University Press
About the Contributors
articles, and book chapters) with an emphasis on cognitive evolution. His books include The Evolution of Spatial Competence (1989, University of Illinois Press), The Rise of Homo Sapiens: The Evolution of Modern Thinking (2018, 2nd ed., with F. Coolidge), How to Think Like a Neandertal (2012, Oxford University Press, with F. Coolidge), and First Sculpture: Handaxe to Figure Stone (2018, Nasher Sculpture Center, with T. Berlant). Chunshan Xu is a Professor of the School of Foreign Studies at Anhui Jianzhu University in China. His research interests include dependency grammar and cognitive linguistics. Some of his recent publications have appeared in Poznań Studies in Contemporary Linguistics, Physics of Life Reviews, and Quantitative Analysis of Dependency Structures. Rachel Zahn is a PhD Student in the Department of Psychological Sciences at Rice University in the United States. She is working with Dr. Randi Martin in the T. L. L. Temple Foundation Neuroplasticity Lab. Her research interests include the role of working memory in language production, particularly longer utterances such as storytelling and sentence production. Her work has focused on neuropsychological populations, specifically individuals with aphasia poststroke. Binghan Zheng is an Associate Professor of Translation Studies and Director of the Centre for Intercultural Mediation at Durham University in England. His research interests include cognitive translation and interpreting studies, neuroscience of translation, and comparative translation and interpreting studies. His recent publications have appeared in journals such as Target, Across Languages & Cultures, Journal of Pragmatics, Brain & Cognition, Perspectives, LANS-TTS, Babel, Translation & Interpreting Studies, Foreign Language Teaching & Research, Translation Studies, and Journal of Foreign Languages. He is a guest editor of journals including Translation & Interpreting Studies and Foreign Language Teaching & Research.
Published online by Cambridge University Press
xxxi
Acknowledgments
First, we are very thankful to all of the individuals who kindly accepted our invitation to contribute to this massive project. Without their dedication, diligence, and, above all, their scholarly wisdom, this handbook simply would not exist. In particular, we wish to thank Alan Baddeley and John Hawkins for their additional advice, encouragement, and generous support to both the handbook and the working memory-language enterprise along the way. As always, the editorial team at Cambridge University Press was efficient, attentive, and supportive. A special thank you goes to Rebecca Taylor, Commissioning Editor, Linguistics, for initially discussing this handbook idea with us and shepherding it to production. We also want to thank Isabel Collins, Senior Editorial Assistant and Joshua Penney, Senior Content Manager, for their excellent work throughout the production of the handbook. We are grateful to our editorial assistant, Teresa Bennett, for her excellent work during the preparation of the manuscript. We gratefully acknowledge that financial support to hire her was provided by the Office of Research Services at Wilfrid Laurier University. Securing peer reviewers to volunteer their even-more-precious time in the wake of a global pandemic was challenging. But despite this, we are extremely grateful to the scholars, many of whom also contributed their own work, who dedicated their time and expertise to anonymously evaluate contributions to this handbook. It is without a doubt that their knowledge and expertise have strengthened the chapters and their implications for ongoing and future research. These individuals include the following: Federica Amici, University of Leipzig and Max Planck Institute for Evolutionary Anthropology, Germany Pierre Noël Barrouillet, Université de Genève, Switzerland Michael Bunting, University of Maryland, United States
Published online by Cambridge University Press
Acknowledgments
Rendong Cai, Guangdong University of Foreign Studies, China Valérie Camos, Université de Fribourg, Switzerland Frederick L. Coolidge, University of Colorado, Colorado Springs, United States Nelson Cowan, University of Missouri, United States Ian Cummings, University of Reading, England Peter Delaney, University of North Carolina at Greensboro, United States Denis Foucambert, Université du Québec à Montréal, Canada Ron Gillam, Utah State University, United States William Grabe, Northern Arizona University, United States Agnieszka J. Graham, Queens University Belfast, Northern Ireland John A. Hawkins, University of California, Davis, United States Graham Hitch, University of York, England Emil Holmer, Linköping University, Sweden Mark Hurlstone, Lancaster University, England Daniel O. Jackson, Kanda University of International Studies, Japan Susanne M. Jaeggi, University of California, Irvine, United States Sun-A Kim, Hong Kong Polytechnic University Huolingxiao Kuang, Durham University, England James Lee, Texas Tech University, United States Michael Leeser, Florida State University, United States Paul Leseman, Utrecht University, The Netherlands Shaofeng Li, Florida State University, United States Haitao Liu, Zhejiang University, China Bingfu Lu, Beijing Language and Culture University, China Jim Montgomery, Ohio University, United States Gary Morgan, City University London, England William O’Grady, University of Hawaiʻi at Mānoa, United States Thierry Olive, Centre national de la recherche scientifique and Université de Poitiers, France Anka Slana Ozimič, University of Ljubljana, Slovenia Dora Pan, Chinese University of Hong Kong Costanza Papagno, University of Trento and University of MilanoBicocca, Italy Ana I. Pérez Muñoz, University of Granada, Spain Nina Purg, University of Ljubljana, Slovenia Grega Repovš, University of Ljubljana, Slovenia Jerker Rönnberg, Linköping University, Sweden Nuria Sagarra, Rutgers, The State University of New Jersey, United States Elisabet Service, McMaster University, Canada Daphnée Simard, Université du Québec à Montréal, Canada Peter Skehan, University College London, England H. Lee Swanson, University of California, Riverside and University of New Mexico, United States John Sweller, University of New South Wales, Australia
Published online by Cambridge University Press
xxxiii
xxxiv
ACKNOWLEDGMENTS
Benjamin Swets, Grand Valley State University, United States André Tricot, University Paul Valéry Montpellier 3, France John Truscott, National Tsing Hua University, Taiwan Domenico Tullo, McGill University, Canada Josje Verhagen, University of Amsterdam, The Netherlands Thomas Wynn, University of Colorado, Colorado Springs, United States Janire Zalbidea, Temple University, United States Binghan Zheng, Durham University, England
Published online by Cambridge University Press
Overview of the Handbook
https://doi.org/10.1017/9781108955638.001 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.001 Published online by Cambridge University Press
1 Working Memory and Language An Overview of Key Topics John W. Schwieter, Zhisheng (Edward) Wen, and Teresa Bennett
1.1
Introduction
Working memory (WM) is our limited-capacity storage and processing (memory) system that permeates essential facets of our cognitive life such as arithmetic calculation, logical thinking, decision-making, prospective planning, language comprehension, and production. Since the first elaboration on WM in the early 1960s (Miller et al., 1960), its role in language acquisition and processing has been extensively investigated both empirically and theoretically by researchers from diverse fields of psychology and linguistics, accumulating an increasingly huge body of literature (e.g., see Gathercole & Baddeley, 1993; Baddeley, 2003 for reviews of early studies). Notwithstanding, the field still lacks a comprehensive and updated profile of conceptualizing and implementing working memory in the broad domains of native and second language acquisition, processing, and language impairments – a volume that is long overdue and expected by both WM theorists and language practitioners alike. Indeed, after a careful examination of recently published titles on working memory and language learning, we find it a bit surprising that, given the enormous interest in this important and intriguing topic by scholars from multidisciplinary fields of cognitive psychology and language sciences, there has not yet been a volume that provides a comprehensive account of these accumulative research findings and empirical evidence. This is particularly so when we realized that Gathercole and Baddeley’s seminal monograph, Working memory and language, had been published in 1993, almost thirty years ago! The succeeding three decades following its publication have recorded an exponential growth of both empirical and theoretical investigations in either working memory or language learning, either separately as two independent disciplines or occasionally jointly investigated as one phenomenon of human cognition.
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
4
JOHN W. SCHWIETER ET AL.
Separately, for example, we have seen, on the one hand, numerous empirical and theoretical papers addressing the nature, structure, and functions of WM per se. Plus, several edited volumes have offered comprehensive coverage of working memory theories and models in cognitive science and neuroscience. These have included, among others, Miyake and Shah’s (1999) Models of working memory, Conway et al.’s (2005) Variation in working memory, and most recently, an edited volume by Logie et al. (2021) entitled Working memory: State of science. At the other end of the pendulum, that is, language acquisition and processing, we have also seen numerous papers reporting both theoretical reviews and empirical studies in language learning per se; an apt example is the recent handbook compiled by the first editor, Schwieter and Benati (2019), though no such a volume has been set out to cover working memory and language issues specifically. Then, when working memory and language learning are considered as a joint venture, we have not found a comprehensive volume addressing the intricate relationship between working memory and the diverse and broad domains of first and second language learning, processing, and impairment. As far as we know, there are only three other recent titles that are addressing a similar topic, and they have greatly inspired and helped pave the way for the current handbook. To begin with, Altarriba and Isurin’s edited volume (2013), Memory, language, and bilingualism: Theoretical and applied approaches, has done an excellent job at showcasing some recent advances in studies on bilingual memory, language processing, and second language forgetting. They are also successful in integrating theoretical developments and real-world approaches to language learning from cognitive perspectives. Although this is very much appreciated in their title, we do feel that it was also a very concentrated and specialized book. It has served as a solid reference for bilingual processing and its interface with memory. That said, we still need a volume to speak more directly to the many other areas in which WM has been demonstrated to play a dynamic role in nuanced language research domains, such as first and second language acquisition, processing, bilingual development, not to mention the emerging domains of language impairment and cognitive training (e.g., Novick et al., 2019). Then, in terms of working memory and first or second language learning, we have failed to identify any recent volumes reviewing this topic comprehensively and thoroughly. In terms of second language learning though, the monograph authored by Wen (2016), “Working memory and second language learning: Towards an integrated approach and its sister volume edited with collaborators (i.e., Wen et al., 2015, Working memory in second language acquisition and processing) feature some succinct reviews of working memory theories, assessment procedures, and empirical studies, thus providing important insights into the intricate relationship between working memory and second language acquisition and processing. They also
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
An Overview of Key Topics
provided inculcate thoughts regarding integrating working memory and second language acquisition research. But again, these two volumes have not been able to address adequately the intricate relationship between working memory and first language acquisition and processing; neither did they manage to cover the well-established areas of WM and linguistic theories and models, as well as the expanding literature of WM and bilingualism. Given these obvious gaps in the research literature, we thus set out to compile this comprehensive handbook, with the goal in mind to fill up all these lacunae from previous research. Furthermore, we also aim for theoretical ingenuity and empirical robustness in our individual chapter reviews and devote independent sections to key areas of foundational theories, including working memory models and measures in cognitive psychology, as well as incorporating working memory within well-established linguistic theories and processing frameworks. As far as we know, many of these have not been done before. As such, we are hoping that the comprehensive coverage of key topics in all these essential areas in our handbook will benefit researchers and students not just from psychology and linguistics, but also readers from all other related fields of cognitive sciences to draw insights and inspirations from the chapters herein.
1.2
Organization of the Handbook
As briefly outlined above, we have thus compiled the handbook to cover all essential areas of the language sciences in which WM has been demonstrated to play a significant role. Specifically, the handbook is organized based upon the following seven parts (which somehow resonates with the very first impression of the “magical number seven” proposed by George Miller in 1956, another buzzword concept that has been associated with WM capacity that is widely known by almost everyone both inside and outside psychology). Part Part Part Part Part Part Part
I: Introduction II: Theoretical Models and Measures III: Linguistic Theories and Frameworks IV: First Language Processing V: Bilingual Acquisition and Processing VI: Language Disorders, Interventions, and Instruction VII: Conclusion
Following the present overview, Part I continues with a special introduction by Baddeley, arguably one of the most prominent scholars in WM. In the chapter, the author walks us through the history of various psycholinguistic models of language and memory and recounts how his classic WM model has evolved to address the arising challenges brought by
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
5
6
JOHN W. SCHWIETER ET AL.
language-related research. The chapter begins with a discussion of his seminal multicomponent model – which was developed during a time when Chomsky’s transformational grammar was at the forefront of psycholinguistics – and explores the link between short-term memory (STM) and long-term memory (LTM). After a brief discussion of the original tripartite WM model (Baddeley & Hitch, 1974), Baddeley summarizes related studies depicting the links between each component WM and language domains, the latest endeavors being focused on the relationship between the episodic buffer and binding. These reviews are also lending support to the distinction between verbal STM and WM. Toward the end, Baddeley acknowledges the different approaches between his own WM model and that of Popper’s view. Part II contains major chapters dedicated to discussing prominent WMlanguage theories, frameworks, and methods. Chapter 3 by Coolidge and Wynn opens the section with an exhaustive overview of the evolution of WM and language. The authors make mention of the fact that one of the pros of Baddeley and Hitch’s multicomponent model is the fact that it encompasses all forms and types of memory, whereas prior to its development, language research primarily focused on acoustic or written language (i.e., a two-component model of memory). Throughout the chapter, the varying components of the multicomponent model including the phonological loop, visuospatial sketchpad, and the development of the central executive are discussed in relation to their roles in language/speech production. In Chapter 4, Papagno presents an updated view of the utilization of the phonological loop as a “language learning device” (Baddeley et al., 1998), a key thrust of the WM-language endeavors in the early decades. The chapter begins with an overview of the role of the phonological loop as outlined in Baddeley and Hitch’s multicomponent model of WM. Furthermore, the author also looks at various studies that investigate the functional role of the phonological loop (i.e., the ability to retain sequences of verbal items for a short period of time). The elaboration of the phonological loop is provided throughout the chapter as various studies are examined on healthy people, children learning their mother tongue, children and young adults learning a second language (L2), and, crucially, on neuropsychological patients with a selective deficit of auditory-verbal shortterm memory. In all, the studies suggest that the primary function of the phonological loop is to hold new phonological representations in the memory long enough to build permanent representation. The author discusses how this explanation can be applied to what occurs during new language learning. Chapter 5 by Adams, Forsberg, and Cowan presents a critical overview of the Embedded-Processes Model of WM and its implications for language use. The Embedded-Processes Model was initially proposed in Cowan (1988) and was more formalized later in Cowan (1999). The model posits that WM is a collection of mechanisms that permit information to remain in an
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
An Overview of Key Topics
activated state. The authors first outline the basic concepts necessary to fully understand the model and then delve deeper into how each of its components is relevant to language use. They also examine the use of WM in both children and adults who are learning an L2. The premise of the chapter is to provide the foundational concepts necessary to understand the critical role that attention and LTM play in relation to successful language utilization at all levels. Chapter 6 by Adams and Delaney reviews long-term WM (LT-WM) and language comprehension. The LT-WM model (Ericsson & Kinstch, 1995) was developed to explain how individuals are able to effectively encode critical information into their LTM despite the somewhat limited capacity and constraints of WM. The authors go through the history of the LT-WM theory, starting with the rationale for its conception in the mid-1990s. Following this, there is an in-depth analysis of how the theory accounts for central phenomena in discourse comprehension and also how more recent work examining the theory proposes the involvement of syntactic processing. The chapter ends with a description of recent studies that review the relationship between neural activity and LT-WM in reasoning skills and language comprehension. Purg, Ozimič, and Repovš in Chapter 7 offer insight into the cognitive neuroscience of WM and language. There is a notable close relationship between WM and language as cognitive systems and as such, there is overlap and integration of brain systems and networks that support these processes. Through an examination of theoretical models and empirical evidence provided by a diverse range of study types and research methods, the chapter focuses on the brain substrate which plays a role in WM and language processing from the perspectives of their interconnectedness, synergy, and integration. In Chapter 8, Hitch, Hurlstone, and Hartley focus on computational models of WM for language processing. The authors focus on explaining how the limited capacity of the phonological loop utilized in verbal WM deals with information about serial order using computational models of the immediate recall of unstructured sequences of words, letters, or digits that provided baseline findings. The comparison of these computational models and the findings exemplify a serial ordering mechanism known as competitive queuing, a process that is evidenced to operate under various forms of linguistic constraint. In all, the chapter suggests that competitive queuing can potentially be a mechanism for unifying theories of WM and language processing. Chapter 9 by Camos and Barrouillet tracks the development of the TimeBased Resource Sharing (TBRS) model of WM for language (Barrouillet & Camos, 2015, 2020). The model investigates the function and development of WM specifically via the integration of executive attention and the timed components of cognitive processes. The authors take an in-depth look at the distinctions between domain-specific systems of maintenance that supervise verbal information and the domain-general system relying on attention. Throughout the chapter, the authors demonstrate how verbal
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
7
8
JOHN W. SCHWIETER ET AL.
information is maintained in WM from childhood to adulthood, how linguistic features impact the short-term memory capacity of verbal information, and how WM mechanisms in the model impact the creation of true and false verbal LTM traces. Rönnberg, Holmer, and Rudner transition to reviewing the Ease of Language Understanding model (Rönnberg, 2003) in Chapter 10. The premise of this model is that speed and accuracy of the signal matching in relation to existing multimodal language representations are determining factors of the ease of language comprehension. Furthermore, the authors examine the difference in matching pace and language understanding from the point of predictions and postdictions in WM. Ultimately, the interaction between WM and LTM is critical to language understanding and the importance of its efficiency becomes apparent in cases where its breakdown results in cognitive decline and dementia. Dehn’s Chapter 11 discusses assessing WM in youth. Given that language development, acquisition of academic skills, and performance of academic skills are heavily reliant on WM, assessing children’s WM is an essential consideration in cases in which it is recommended that children have neuropsychological and psychological evaluations. Children with a low WM capacity are prone to academic learning difficulties, specific learning disabilities (SLD), and deficits such as dyslexia (Alloway and Archibald, 2008; Pickering, 2006). The author also looks at the assessment measures that can be used to examine both storage and processing (e.g., the backward digit span test) and they also identify several assessments that are critically needed to measure WM function in children. In Chapter 12 by Burgoyne et al., the authors illustrate the importance of WM capacity and attention control measures for language researchers. The chapter investigates the origins of complex span measures of WM capacity to understand the underlying cognitive processes of language comprehension. Following this, there is a review of the evidence that supports an executive attention perspective to WM (Kane & Engle, 2002) – and a description of the relationship between WM capacity, attention control, and language comprehension. Specific attention is paid to how the functions supported by attention control contribute to performance across a variety of cognitive tasks. In closing, the authors provide recommendations and resources for researchers whose work involves measuring WM capacity and attention control. Part III presents important theoretical contributions from general linguistics. In Chapter 13, Hawkins provides insight on whether or not grammars are shaped by WM. The author examines variation patterns across the world’s grammars specifically in relation to psycholinguistic WM models. The following arguments are made and supported with compelling evidence: (1) constrained capacity WM exceeds some limit, (2) more versus less WM models are reflected in preferred vs dispreferred structures, and (3) integrated WM models that interact with other factors facilitate processing ease. Then, in Chapter 14, Amici, Sanchez-Amaro, and Cacchione discuss
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
An Overview of Key Topics
the link between branching and WM within a cross-linguistic approach. The authors first review the differences in processing information habits retained beyond the linguistic domain across languages as this affects how humans process stimuli other than words in a sentence. The chapter reflects on previously conducted studies that explore the association between word order and attention allocation and further investigate the effects of branching habits beyond the linguistic domain. The concluding sections promote a stronger cross-cultural approach to the study of branching and WM. In Chapter 15, O’Grady shifts focus toward WM and natural syntax. The chapter begins with a brief review of the birth of cognitive science with Miller’s (1956) paper on human memory and the storage of information. This seminal paper provided the foundation of the capacity limits of shortterm memory being established at 72 units of information. Subsequent work by Yngve (1960) builds on Miller’s work and suggested that these capacity limits apply to “immediate memory” in sentence production, a critical factor explaining the complexity of English syntax. The three lines of inquiry posited by Chomsky (1956) are also identified and reviewed. Finally, the author discusses the Performance-Grammar Correspondence Hypothesis (Hawkins, 2014) and speaks about how his differing approach is more rooted in emergentist thought. Chapter 16 by Xu and Liu offers a review of research on the role of WM in shaping syntactic dependency structures. The chapter looks at the relationship between dependency distance, the constraint of WM, and the least effort principle. More specifically, the authors indicate that the latter two variables are typically organized in a way that reduces dependency distance, which ultimately shapes the patterns of word order in human languages and may account for the linguistic universals seen in language typology. Finally, the chapter examines whether syntactic structures are the result of self-adaptation of language systems as shaped by external constraints and motivations including WM. Truscott and Sharwood Smith report on WM and the Modular Cognition Framework (Truscott & Sharwood Smith, 2004) in Chapter 17. The authors emphasize the involvement of the framework in language development and WM. WM and its capacity are considered integral parts of the cognitive system and as such, the authors apply this view to the study of language through their discussion of processing, learning, metalinguistic knowledge, bilingual control, codeswitching, the issue of selective versus nonselective access in bilingual processing, optionality in SLA, and translation and interpreting. In Lu and Wen’s Chapter 18, the authors explore the role of WM and the language acquisition device (LAD; Chomsky, 1965). To resolve the issue of the elusive nature of the distinction between STM and WM limitations as they relate to language processing, the authors propose an integrated memory and chunking-based metric of comprehension complexity. They posit that during language processing, STM limitations of 72
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
9
10
JOHN W. SCHWIETER ET AL.
chunks (Miller, 1956) are sensitive to the instant chunk number (i.e., the number of information chunks that the parser has kept active in mind when processing a new word) whereas WM limitations of 41 (Cowan, 2001) are sensitive to the mean dependency distance (i.e., the sum of all instant chunk numbers divided by the number of words in the construction). They provide examples of their metric calculations and discuss some of their limitations. Part IV reviews research on WM and first language (L1) processing and use. In Chapter 19 by Kim, the author reviews research on WM in word reading. This section provides a summary of the previous studies conducted on word reading from the perspective of Baddeley’s multicomponent WM model. The chapter also reviews studies on WM in Chinese word reading and offers clear directions for future research. Zahn, Horne, and Martin, in Chapter 20, examine the role of WM in language comprehension and production from neuropsychological perspectives. They address the role of verbal WM (specifically, the contributions of phonological and semantic WM buffers) in processes such as language production and comprehension, drawing on a mixture of data collected from brain-damaged and healthy individuals. Pérez Muñoz and Bajo discuss WM and high-level text comprehension processes in Chapter 21. The authors review a substantial body of research concerning how cognitive processes are supported by WM during online L1 and L2 text comprehension. Pérez and Bajo argue that WM is particularly critical when text comprehension requires updating of the situation model through the inhibition of nonrelevant information in the L1, but this relationship is still unclear in the L2. Chapter 22 by Swets and Ivanova addresses WM and speech planning. The authors examine how WM fine-tunes the flexibility of speech planning strategies and review the role of WM in individual levels of planning. Ultimately, they conclude that speech planning is determined by a complex system of compensatory factors that includes WM. In Chapter 23, Olive reviews how novice and skilled writers engage WM. The chapter utilizes specific perspectives that analyze how writing processes engage both the executive and nonexecutive components of WM. Part V synthesizes work on WM and L2 acquisition, processing, and use. Chapter 24 by Service and Simard reports on how measures of WM relate to L2 vocabulary. The authors aim to distinguish the differences between the intersecting effects of the components of WM, vocabulary knowledge, and methodological choices (i.e., measurement instruments) as they pertain to different aspects of L2 vocabulary knowledge. The authors critically examine previous research and bring to light their differing methodologies and implications of their results. The chapter concludes with a proposition of new perspectives on the interpretations of tasks typically used in WM and L2 vocabulary studies. In Chapter 25 by Leseman and Verhagen, the authors discuss WM and L2 grammar development in children. In reviewing previous work on statistical learning in first language acquisition and on cue
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
An Overview of Key Topics
competition in L2 learners, they suggest that the relationships between WM and L2 grammar are bidirectional and based on general mechanisms of perception, extraction, and integration of statistical information in the surrounding language. Specifically, they propose that children learn grammar in an L2 through chunking exemplars of constructions and that the resultant chunks in long-term memory, in turn, aid perception and attention, and enhance working memory capacity to facilitate further L2 learning. McCormick and Sanz’s Chapter 26 examines WM and L2 grammar development among adults. The authors review a plethora of studies that explain the differences in developmental rates such as the “more is better” hypothesis. McCormick and Sanz argue that future research may benefit from examining the role of WM in cognitively demanding tasks, which happen to change during L2 learning. In Chapter 27, Cunnings critically evaluates the various models of WM explaining L2 sentence comprehension. He also reviews existing studies that examine how WM influences L2 sentence processing and offers a concluding discussion on individual differences in WM and sentence processing with suggestions as to how to best measure it. In Chapter 28, Leeser and Herman report on methodological issues in research on WM and L2 reading comprehension. They argue that decisions about study designs can have significant effects on research outcomes and provide suggestions as to how to rectify these issues. This chapter also discusses the importance of understanding reading comprehension processes and their various components as these are vital for the promotion of L2 literacy skills and linguistic development. Finally, the authors discuss the fundamental issues that must be taken into consideration for more productive research to be done in this field. Skehan in Chapter 29 discusses WM and L2 speaking tasks. He examines the relationships between WM and performance on L2 speaking tasks. Based on the research discussed in the chapter, the author argues that WM plays a role in the formulation stage of speech production and proposes areas in which it would be the most helpful to research WM connections with L2 speaking tasks. In Chapter 30 by An and Li, the authors report on WM and L2 interaction. The chapter begins with an overview of WM as it pertains to L2 acquisition. Following this, a summary of the methods and findings from empirical research examining the role of WM in L2 interaction is noted. This synthesis focuses on three topics: (1) WM and interactional behaviors; (2) WM and task performance; and (3) WM and interaction-driven L2 gains. The chapter concludes with a discussion of the limitations, pedagogical implications, and future research directions. In Chapter 31, Zheng and Kuang analyze WM in interpreting studies by conducting a systematic review of work from the 1970s to the 2010s. Special attention is paid to simultaneous interpreting (SI) and consecutive interpreting (CI). The primary goal of the chapter is to fill the gap in SI and CI research by reviewing major WM models of interpreting, determining the similarities and differences between SI and CI, and examining relevant
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
11
12
JOHN W. SCHWIETER ET AL.
empirical evidence that either validates or invalidates the models. New research directions that can be used to investigate SI and CI are proposed. Chapter 32 by Shin and Hu begins by first discussing a methodological synthesis of WM tasks in L2 research. The authors review the last 20 years of WM and L2 learning research in order to describe WM task designs, scoring methods, and reporting practices. The chapter contributes to the ongoing dialogue promoting methodological rigor of WM measurement in L2 research (Wen et al., 2021). Part VI turns to engage readers with a dialogue on WM, language disorders, interventions, and instruction. Chapter 33 by Swanson presents a critical overview of specific learning disorders due to a WM deficit. The author draws on recent studies that suggest that growth in the executive component of WM is significantly related to growth in reading and/or math for children with specific learning disorders in these areas. Chapter 34 by Montgomery, Gillam, and Evans reviews a new perspective on the connection between memory and sentence comprehension in children with developmental language disorder (DLD). The authors present the first conceptually integrated and empirically validated model of sentence comprehension for school-age children with DLD which takes into account and describes the structural relationship between the aforementioned abilities. Morgan’s Chapter 35 examines WM and deafness while focusing on the development of WM in children who are born deaf. The author utilizes studies of deaf users of spoken and signed languages from both the medical and social models of deafness and reviews how variance in WM capacities has previously been attributed to deaf and hearing children. The chapter concludes that deafness disrupts early language exposure and reduces rehearsal abilities, ultimately impacting the WM’s operating system. Chapter 36 by Alloway, Carpenter, Robinson, and Frankenstein tracks the development of WM training in the classroom. The authors characterized WM training programs into two categories: those narrow or broad in scope. Furthermore, they also look at the efficacy of WM training as it pertains to transfer effects and report whether these effects are short- or long-term (maintenance effects). They also provide recommendations for implementing WM in the classroom. Holmes, Byrne, and Graham transition to WM and classroom learning in Chapter 37. The scholars discuss how minimal WM resources constrain classroom learning with a specific focus on the impacts of poor WM on the ability of children to follow learning-activity-relevant instructions and classroom management tasks. Finally, various methods that can be employed to assist these children are discussed, including an overview of current ideas about memory enhancement via brain stimulation and training and practical methods that can inform pedagogy. In Chapter 38 by Sweller, Roussel, and Tricot, the authors discuss cognitive load theory and instructional design for language learning which begins with employing evolutionary education psychology to determine which categories of knowledge are instructionally relevant. Chapter 39 by
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
An Overview of Key Topics
Tullo and Jaeggi conducts a meta-analysis of studies on WM training that explore the mechanistic properties and effectiveness of WM training. The authors propose alternative approaches that may fill gaps that traditional interventions have left open. Finally, in Part VII, the handbook concludes with Chapter 40, a summative piece written by the editors Wen and Schwieter. This first part of the chapter highlights and synthesizes multiple perspectives of WM from cognitive science as augmented by the major chapters of relevant sections in the handbook, which culminates in unifying characterizations of the construct. Then, the second part proposes an integrated account and taxonomy of WM (i.e., the phonological/executive model; Wen, 2016 & 2019) as a viable framework for language sciences and bilingualism research in the future.
1.3
Conclusion
In short, we hope that as collectively demonstrated throughout The Cambridge Handbook of Working Memory and Language, the broad fields of working memory and language sciences can now have a more interactive platform and thus can move closer not just to portray a much clearer picture of the emerging patterns of the interaction between WM and language processes, but also make a head start at more seamless integration that will benefit both disciplines and many others. So far, we have come to appreciate that the multiple components and functions of WM are playing critical albeit distinctive roles in essential aspects of language acquisition and processing and impairment in general (Wen, 2016). But beyond these, we also realize that there is still much to explore further, particularly toward deeper integration between the two independent disciplines of working memory in cognitive science on the one hand and language sciences on the other. Such an ambition has constituted not just the rationale and motivations of this handbook in the very first place, but also represents the vision and mission of this handbook. To allow this to happen, we thus call on more researchers and students from across the multiple fields to join us to embark on this rewarding journey to further study and learn more about WM and language, which proves to be the ultimate joy and reward of editing this enormous handbook.
References Alloway, T., & Archibald, L. (2008). Working memory and learning in children with developmental coordination disorder and specific language impairment. Journal of Learning Disabilities, 41(3), 251–262. Altarriba, J., & Isurin, L. (2013). Memory, language, and bilingualism: Theoretical and applied approaches. Cambridge University Press.
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
13
14
JOHN W. SCHWIETER ET AL.
Atkinson, R., & Shiffrin, R. (1968). Human memory: A proposed system and its control processes. In K. Spence & J. Spence (Eds.), The psychology of learning and motivation: Advances in research and theory (pp. 89–195). Academic Press. Baddeley, A. (2003). Working memory and language: An overview. Journal of Communication Disorders, 36(3), 189–208. Baddeley, A., Gathercole, S., & Papagno, C. (1998). The phonological loop as a language learning device. Psychological Review, 105, 158–173. Baddeley, A., & Hitch, G. (1974). Working memory. In G. Bower (Ed.), The psychology of learning and motivation (pp. 47–89). Academic Press. Barrouillet, P., & Camos, V. (2015). Working memory: Loss and reconstruction. Psychology Press. Barrouillet, P., & Camos, V. (2020). The time-based resource-sharing model of WM. In R. Logie, V. Camos, N. Cowan (Eds), Working memory: State of the science (pp. 85–115). Oxford University Press. Chomsky, N. (1956). Three models for the description of language. Institute of Radio Engineers Transactions on Information Theory, 2(3), 113–124. Conway, A., Jarrold, C., Kane, M., Miyake, A., & Towse, J. (Eds.) (2007). Variation in working memory. Oxford University Press. Cowan, N. (1988). Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human informationprocessing system. Psychological Bulletin, 104(2), 163. Cowan, N. (1999). An embedded-processes model of working memory. Models of working memory: Mechanisms of active maintenance and executive control, 20, 506. Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behaviour and Brain Sciences, 24, 87–185. Ericsson, K., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102, 211–245. Gathercole, S., & Baddeley, A. (1993). Working memory and language. Lawrence Erlbaum Associates. Hawkins, J. (2014). Cross-linguistic variation and efficiency. Oxford University Press. Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W., & Engle, R. W. (2004). The generality of working memory capacity: A latentvariable approach to verbal and visuospatial memory span and reasoning. Journal of Experimental Psychology: General, 133(2), 189–217. Logie, R., Camos, V., & Cowan, N. (Eds.). (2021). Working memory: State of the science. Oxford University Press. Novick, J., Bunting, M., Dougherty, M., & Engle, R. (2019). Cognitive and working memory training: Perspectives from psychology, neuroscience, and human development. Oxford University Press.
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
An Overview of Key Topics
Miller, G. (1956). The magical number of seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63 (3), 81–97. Miller, G., Galanter, E., & Pribram, K. H. (1960). Plans and the structure of behavior. Holt. Miyake, A. and Shah, P. (1999). Models of working memory: Mechanisms of active maintenance and executive control. Cambridge University Press. Pickering, S. (2006). WM in dyslexia. In T. Alloway & S. Gathercole (Eds.), Working memory and neurodevelopmental disorders (pp. 7–40). Psychology Press. Rönnberg, J. (2003). Cognition in the hearing impaired and deaf as a bridge between signal and dialogue: A framework and a model. International Journal of Audiology, 42 (Suppl 1), S68–S76. Schwieter, J. W., & Benati, A. (2019). the cambridge handbook of language learning. Cambridge University Press. Thiessen, E. D., & Erickson, L. C. (2013). Beyond word segmentation: A twoprocess account of statistical learning. Current Directions in Psychological Science, 22, 239–243. Truscott, J., & Sharwood Smith, M. (2004). Acquisition by processing: A modular approach to language development. Bilingualism: Language and Cognition, 7, 1–20. Wen, Z. (2016). Working memory and second language learning: Towards an integrated approach. Multilingual Matters. Wen, Z. (2019). Working memory as language aptitude: The Phonological/ Executive Model. In Z. Wen, P. Skehan, A. Biedron, S. Li, & R. Sparks (Eds.), Language aptitude: Advancing theory, testing, research and practice (pp. 189–214). New York, NY: Routledge. Wen, Z., Juffs, A., & Winke, P. (2021). Measuring working memory. In P. Winke & T. Brunfaut (Eds.), The Routledge handbook of second language acquisition and testing (pp. 167–176). Routledge. Wen, Z., M. Mota & McNeill, M. (2015). Working memory in second language acquisition and processing. Multilingual Matters. Yngve, Victor. (1960). A model and a hypothesis for language structure. Proceedings of the American Philosophical Society 104, 444–466.
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
15
https://doi.org/10.1017/9781108955638.002 Published online by Cambridge University Press
Part I
Introduction
https://doi.org/10.1017/9781108955638.003 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.003 Published online by Cambridge University Press
2 Working Memory and the Challenge of Language Alan Baddeley
I am pleased, if somewhat surprised, to find myself writing part of the introduction to a handbook comprising over three dozen chapters on language since I am afraid I have little psycholinguistic expertise. The multicomponent model developed during the period when psycholinguistics was dominated by Chomsky’s transformational grammar and its potential implications. These included the speculation by Miller (1962) that comprehension might involve the temporary storage of sentences while they are syntactically “unpacked,” a view that seemed initially to be promising (Savin & Perchonok, 1965), a promise that was however not fulfilled when subsequent studies uncovered a range of alternative interpretations of their results (e.g., Epstein, 1969; Glucksberg & Danks, 1969). However, such results did not rule out the possibility that temporary storage plays a role in language comprehension, a view that was broadly consistent with the modal model of memory proposed by Atkinson and Shiffrin (1968). This assumed a short-term store that functioned as a working memory and was involved in performing complex tasks, including those of a linguistic nature. By the early 1970s however the field of short-term memory (STM) was becoming increasingly diffuse with many different models of STM, typically based on limited experimental paradigms, often coupled with rather specific mathematical models, causing many to abandon the once popular but increasingly complex field for research on the newly developing area of semantic memory. The short-term storage component of the Atkinson and Shiffrin modal model was also being challenged on two fronts. The first was the assumption that it was merely necessary to hold information in the short-term store for it to be automatically transferred to long-term memory (LTM). Craik and Lockhart (1972) demonstrated that a more important variable than time in store was the operations performed on the material, resulting in their Levels of Processing hypothesis, which proposed that the more deeply the material was encoded, the more durable the memory trace
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
20
ALAN BADDELEY
Figure 2.1 The working memory model (Baddeley & Hitch, 1974)
(Craik & Lockhart, 1972; Craik & Tulving, 1975). A further challenge came from the discovery by Shallice and Warrington (1970) of patients who had a very substantially limited STM capacity with digit spans of two or less, and yet appeared to have normal LTM and normal language. If verbal STM acted as a working memory, both everyday cognitive functioning and language should have been greatly impaired. Graham Hitch and I entered the field at this point, with a grant focused on exploring the link between STM and LTM. We decided to begin by testing the hypothesis that verbal STM did indeed serve as a general working memory. In the absence of access to suitable patients, we did so using healthy subjects using a concurrent digit serial recall task to systematically load STM, at the same time as they performed tasks assumed to depend on working memory. Participants were required to retain and immediately repeat back digit sequences of varying lengths with the assumption that the longer the sequence, the heavier the load, the greater the demand on working memory, and the poorer the performance on any complex primary task. We tested this across verbal reasoning, language comprehension, and verbal long-term memory, in each case finding a clear effect for concurrent sequence lengths greater than three, but one that was by no means catastrophic. People were still able to perform a concurrent syntactic reasoning task with a constant error rate of around 5 percent even when holding sequences of span length, although reasoning time did increase systematically with sequence length. This suggested some involvement of short-term storage in reasoning but one that was no means a major factor. Broadly similar results were found using the concurrent digit load method to study the role of STM in both long-term learning and prose retention. This pattern of results prompted us to propose the three-component model shown in Figure 2.1. This assumed a limited capacity attentional control system, the central executive, aided by temporary verbal storage from the phonological loop together with visuospatial storage from the visuospatial sketchpad. Digit span was assumed to involve the loop with the load on the central executive
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
The Challenge of Language
increasing with the length of the digit sequence. We proposed that the phonological loop comprised a simple acoustic/phonological storage system where memory traces could be maintained by subvocal rehearsal (Baddeley & Hitch, 1974). Evidence that such rehearsal occurred in real time came from the word length effect (Baddeley et al., 1975) whereby accuracy of immediate memory for a sequence of five words was directly related to the spoken length of the words. More specifically, we argued that representations in the store faded over time but could be refreshed by subvocally articulating the words. Memory span was thus determined by a combination of the rate of decay of the short-term trace and the number of items that could be continually refreshed. We found that performance deteriorated once spoken length exceeded around two seconds, which appeared to be the maximum spoken length that could be maintained without error. However, while the word length effect is a very robust phenomenon, its interpretation whether in terms of trace decay, or some form of interference, remains controversial (Barrouillet & Camos, 2014; Lewandowsky & Oberauer, 2009). We initially termed the system an “articulatory loop,” later wishing to emphasize the store rather than the rehearsal process, we changed this to “phonological loop.” Further evidence for the role of subvocal rehearsal in STM performance came from the observation that blocking rehearsal through repeated concurrent articulation of a simple sound such as the word “the” would impair performance and remove the effect of word length on performance (Baddeley et al., 1975; Larsen & Baddeley, 2003). Evidence for the nature of the underlying memory trace was based on the observation initially made by Conrad (1962), that errors in STM for consonant sequences were similar in sound to the correct item (e.g., c for v), even with visual presentation. Furthermore, memory for verbal sequences was markedly reduced when they were similar in sound (Conrad & Hull, 1964). A further series of studies using monosyllabic words again showed a major effect of phonological similarity on STM while observing that similarity of meaning had a much more limited effect (Baddeley, 1966a). The opposite occurred for the learning of longer sequences of words, with similarity of meaning impairing performance, whereas similarity of sound had little impact (Baddeley, 1966b). Up to this point, the idea of the short-term verbal memory system using a phonological code and an articulatory rehearsal process seemed to account neatly for quite a large range of data. However, despite this promising series of studies, the simple proposal that STM was phonologically based while LTM was semantically based was encountering problems. One of these concerned the more precise nature of the temporary store and its link with articulation. Hence, while immediate memory for the order of the items was markedly impaired by concurrent articulation, suppression did not appear to disrupt the capacity to create a phonological representation of a visually presented item as reflected in either the speed or accuracy of making homophone judgments on visually presented words.
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
21
22
ALAN BADDELEY
Thus, participants were able to decide whether “wait” and “weight” were homophones, or whether “weight” and “wate” were homophonic. This implied that it was possible to set up and use some form of acoustic code during concurrent articulation, though this could apparently not be used for memory storage which, in contrast was substantially reduced by concurrent articulation (Baddeley & Lewis, 1981; Besner et al., 1981). Articulatory suppression did however interfere with rhyme judgments where the initial phoneme needs to be stripped off before a homophony judgment can be made on the rest of the word. (Baddeley & Lewis, 1981; Besner et al., 1981). We speculated that this might reflect two separate components of the phonological store, one being acoustic, which we termed the “inner ear” and the other articulatory, the “inner voice.” Both were assumed to be capable of setting up representations within the phonological loop, but only the inner voice was capable of maintaining the representation via active subvocal recycling of the input representation while manipulating the representation. Since that time, this distinction seems to have been comparatively neglected in the memory field until revived in the recent paper by Norris et al. (2018) and by our own interpretation of the results of a study into the role of the phonological loop in acquiring the accent of a second language (Mattys & Baddeley, 2019). This seems like a promising area for future work. A second series of puzzles arose concerning the role of the phonological loop in complex prose comprehension following a highly influential paper by Daneman and Carpenter (1980). They were interested in the role of working memory in comprehension and tackled this question using an individual difference-based approach, proposing that working memory involved a combination of short-term memory storage and concurrent processing, developing a working memory span test that required both of these. This involved requiring participants to read out a sequence of unrelated sentences and to then recall the final word of each sentence. Even with short sentences, this proves to be a demanding task with typical spans of three or four. Despite its apparent simplicity, it was found to correlate highly with performance on the comprehension component of the graduate record examination. This proved to be a robust finding that was subsequently extended very widely not only to language comprehension but to many other cognitive processing tasks ranging from the ability to resist perceptual distraction to performance on complex reasoning tasks of the type that feature in standard intelligence tests (Shipstead et al., 2015). A similar capacity to predict complex performance was shown for variants of working memory span, including, for example, one that substitutes a sequence of arithmetic operations (Turner & Engle, 1989) or basic transformations of simple items, provided the demand level is sufficient (Barrouillet & Camos, 2015). The sheer range of tasks that are correlated with variants of the working memory span task has led to the extensive exploration of its potential to provide a theoretical underpinning for the
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
The Challenge of Language
essentially statistical concept of general intelligence. However, while there is broad agreement, for example, that some form of inhibitory control is involved, there is less agreement on its precise nature and whether inhibition itself can be broken down into subcomponents (Shipstead et al., 2015). Returning to the multicomponent model, we find that the broad and robust evidence for the predictive capacity of working memory span, while emphasizing the generality and importance of working memory, raises a major challenge for the three-component model. Given that neither the loop nor the sketchpad appeared to have the capacity to hold several sentences, while the central executive was regarded as a purely attentional system, how could it explain performance on the working memory span test? Further problems had also arisen at about this time with the question of how the visuo-spatial component could interact with the phonological system given no joint code. Indeed, working memory needs to be able to combine inputs from perception and long-term memory as well as integrating information from its more specialized subsystems. Our model clearly required some sort of general multicode common space. As a response to this, the concept of an Episodic Buffer was proposed (Baddeley, 2000). The episodic buffer is assumed to be a multimodal temporary storage system of limited capacity that accepts and combines inputs from a range of different sources – visual, verbal, perceptual – combining them with information from various aspects of LTM. It was assumed to be accessible through conscious awareness and to be controlled by the central executive. To my relief, the concept was regarded as interesting and plausible enough to enter the mainstream literature. However, it raised the problem of whether the concept was not simply too powerful to be useful. We did not want it to act simply as a convenient “explanation” for awkward results. Our overall approach to theory did not exclude the possibility of adding a fourth component that was underspecified, provided it served a useful stage in allowing the model to grow by asking tractable and empirically answerable questions. We therefore set about attempting to use the concept of an episodic buffer productively. To do so, we incorporated the concept of “binding,” the process whereby potentially independent features such as color, shape, and spatial location are bound together into integrated objects or where sequences of words are bound together into meaningful phrases. We began with the assumption that the episodic buffer itself performs the act of binding, with the central executive playing a crucial role in controlling the process. We tested this by using dual task interference to disrupt the various components of working memory: the executive, the sketchpad, and the loop. We expected that the central executive would play a central role in the process of binding but interact with the sketchpad when binding color and shape into objects, and with the loop when binding words into meaningful phrases. We began with visual binding and studying the capacity for information on shape and color to be bound into memory for arrays of colored objects
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
23
24
ALAN BADDELEY
(Luck & Vogel, 1997). In the case of language, we looked at the effect of binding by comparing immediate memory for material in sentence form in which words are bound into meaningful chunks, comparing this with memory for the same words in random order. In both cases we got a clear answer to our question concerning the role of working memory. Both the visual and verbal studies yielded the same broad conclusion, that concurrent tasks impaired overall performance, but they did not prevent binding. In the case of memory for colored objects visual and executive disruption were more substantial than verbal, but the effect was no greater for bound than for individual features, suggesting that the binding of features into objects probably occurs at a perceptual processing level rather than in working memory (Allen et al., 2006). In the case of verbal material, executive and verbal concurrent tasks proved more disruptive than visual but the disruptive effects of the concurrent tasks were equivalent for the sentences and the random word sequences. In this case it appears that active chunking operates in long-term language processing systems, not in working memory, as we initially proposed (Baddeley et al., 2009). It is important to stress, however, that the concurrent tasks certainly did impact on short term storage in the expected way. What was not disrupted was the process of binding per se, which appears to occur outside working memory. This suggests that the episodic buffer is essentially a passive storage system that makes episodes bound elsewhere available to conscious awareness (See Baddeley, 2012, for further discussion). We then moved onto a more detailed analysis of the attentional control system, which we assume comprises an episodic buffer that provides temporary storage and a central executive, both drawing on a limited pool of attentional capacity. We assume that this can be used differentially either focused principally to control perceptual input with internally focused executive control minimized or alternatively with the emphasis being on executive control with consequent reduction of perceptually oriented attention (Hitch et al., 2020). We were happy to note that this is entirely consistent with approaches to the analysis of working memory control proposed by colleagues in visual attention (e.g., Chun et al., 2011). We regard our concept of the central executive as a control system focusing attention on an essentially passive episodic buffer, as entirely consistent with Nelson Cowan’s (2005) embedded processes model of working memory, which assumes the focusing of a limited attentional capacity on what he terms “activated long-term memory.” We view our own approach as offering a more detailed account of the interface between the storage and attentional components of working memory than that offered by Cowan’s more general concept of activated LTM. We do not, of course, deny the importance of long-term memory at many points within the working memory system. In the case of language processing for example, LTM clearly has a crucial influence at levels ranging from the role of spoken language skills in subvocal rehearsal to the impact of semantic
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
The Challenge of Language
memory and syntactic processing on sentence recall. Each of these of course depends on long-term knowledge of different kinds. We regard the concept of activated long-term memory as a convenient theoretical placeholder rather than a theoretically productive explanation, acknowledging the importance of factors beyond the limits of working memory while allowing Cowan to focus on the important attentional core of the system. I hope this brief account of the development of the multicomponent model has demonstrated the crucial importance to the multicomponent model of working memory that has been provided throughout its development by studies involving language, which have repeatedly resulted in challenges to earlier versions of the model. As described earlier, such challenges often led to hypotheses that were subsequently rejected. Should this therefore not have caused us to abandon the whole multicomponent model? According to one approach to science, it should. A still influential approach to the development of the scientific theory was proposed in the last century by Karl Popper (1959), who argued for the importance for falsification in theory development, suggesting that theories should lead to precise predictions, which if not fulfilled should lead to the abandonment of that theory. I adhered to this view early in my career until I realized that in my own case it did not seem to work. It seemed inappropriate for phenomena of the complexity of those I was investigating. It was certainly possible to make precise models, but in order to do so for an area as complex as working memory, one had to make a relatively large number of assumptions, which would either be unlikely to be correct in every case or the model would need to be highly flexible, allowing it to change depending on the precise details of the experimental paradigm, material, and conditions. This can be dealt with by allowing a range of parameters to vary. However, such an approach runs the danger of becoming little more than a complex curve-fitting exercise, a practice that in physics was criticized by von Neumann with the comment, “Give me four variables and I will fit an elephant, and with five I will make it wave its trunk!” That is not to deny the need for precise modeling in specific areas; the analysis of serial order in STM is an example (Hurlstone et al., 2014). I would however suggest that complex biological systems are likely to benefit from a less rigid approach than the simple demand for potential falsification proposed by Popper (1959). This is not of course a uniformly held position; there are efforts to provide detailed account of working memory in all its complexity currently being attempted by, for example, Oberauer (2021). We regard such attempts as complementary to our own, ultimately potentially leading to a deeper understanding of working memory, but currently less easy to apply beyond the laboratory than the much simpler multicomponent model. The multicomponent model was developed using a rather different approach from Popper’s emphasis on falsification, instead treating scientific theories as analogous to maps of a terrain that is gradually being
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
25
26
ALAN BADDELEY
explored (Toulmin, 1953). From this viewpoint, a theory is not seen as either correct or wrong, but rather as providing a more or less useful map, with usefulness based on the capacity of the theory to capture existing knowledge in a way that suggests potentially productive further investigation. Such investigation may well be driven by predictive hypotheses, but the value of the theory is not dependent on whether the predictions are fulfilled. Indeed, as I would suggest our own work shows, the greatest progress often occurs when reliable effects are found that do not fit. This then may indicate the need for further investigation that will itself enrich the theory by extending the scope and cohesiveness of the theoretical framework. The value of the overall model will thus depend on its capacity to give a relatively simple account of both the old and the new evidence. I would suggest further that one valuable source of evidence for the success of a model is its capacity to be extended fruitfully beyond the limits of the laboratory, which I suggest is a major feature of the multicomponent model of working memory (see Baddeley et al., 2019, 2021). I suggest that our attempts to use the multicomponent model to explore the interface between memory and language provides a good example of this approach. In conclusion, despite our lack of expert knowledge of psycholinguistics, we have gained enormously from the challenge offered by the attempt to apply our model to the rich field of language. Whether the study of language has gained from our efforts remains to be seen.
References Allen, R., Baddeley, A. D., & Hitch, G. J. (2006). Is the binding of visual features in working memory resource-demanding? Journal of Experimental Psychology: General, 135, 298–313. Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation: Advances in research and theory (Vol. 2, pp. 89–195). Academic Press. Baddeley, A. D. (1966a). Short-term memory for word sequences as a function of acoustic, semantic and formal similarity. Quarterly Journal of Experimental Psychology, 18, 362–365. Baddeley, A. D. (1966b). The influence of acoustic and semantic similarity on long-term memory for word sequences. Quarterly Journal of Experimental Psychology, 18, 302–309. Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4, 417–423. Baddeley, A. (2012). Working memory, theories models and controversy. The Annual Review of Psychology, 63, 12.11–12.29. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. A. Bower (Ed.), Recent advances in learning and motivation (Vol. 8, pp. 47-89). Academic Press
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
The Challenge of Language
Baddeley, A. D., Hitch, G. J., & Allen, R. J. (2009). Working memory and binding in sentence recall. Journal of Memory and Language, 61, 438–456. Baddeley, A. D., Hitch, G. J., & Allen, R. J. (2019). From short-term store to multicomponent working memory: The role of the modal model. Memory & Cognition, 47, 575–588. Baddeley, A. D., Hitch, G. J., & Allen, R. J. (2021). A multicomponent model of working memory. In R. H. Logie, V. Camos, & N. Cowan (Eds.), Working memory: State of the science (pp 10–43). Oxford University Press. Baddeley, A. D., & Lewis, V. J. (1981). Inner active processes in reading: The inner voice, the inner ear and the inner eye. In A. M. Lesgold & C. A. Perfetti (Eds.), Interactive Processes in Reading (pp. 107–129). Lawrence Erlbaum. Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of short-term memory. Journal of Verbal Learning and Verbal Behavior, 14, 575–589. Barrouillet, P., & Camos, V. (2015). Working memory: Loss and reconstruction Psychology Press. Besner, D., Davies, J., & Daniels, S. (1981). Phonological processes in reading: The effects of concurrent articulation. Quarterly Journal of Experimental Psychology, 33, 415–438. Chun, M. M., Golomb, J. D., & Turk-Browne, N. B. (2011). A taxonomy of external and internal attention. Annual Review of Psychology, 62, 73–101. Conrad, R. (1962). Practice, familiarity and reading rate for words and nonsense syllables. Quarterly Journal of Experimental Psychology, 14, 71–76. Conrad, R., & Hull, A. J. (1964). Information, acoustic confusion and memory span. British Journal of Psychology, 55, 429–432. Cowan, N. (2005). Working memory capacity. Psychology Press. Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing. A framework for memory research. Journal of Verbal Learning & Verbal Behavior, 11, 671–684. Craik, F. I. M., & Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General, 104 (3), 268–294. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behaviour, 19, 450–466. Epstein, G. F. (1969). Machiavelli and the devil’s advocate. Journal of Personality and Social Psychology, 11, 38–41. Glucksberg, S., & Danks, J. H. (1969). Grammatical structure and recall: A function of the space in immediate memory or of recall delay? Perception and Psychophysics, 6, 113–117. Hitch, G. J., Allen, R. J., & Baddeley, A. D. (2020). Attention and binding in visual working memory: Two forms of attention and two kinds of buffer. Attention, Perception & Psychophysics, 82, 280–293. Hurlstone, M. J., Hitch, G. J., & Baddeley, A. D. (2014). Memory for serial order across domains: An overview of the literature and directions for future research. Psychological Bulletin, 14, 339–373.
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
27
28
ALAN BADDELEY
Larsen, J., & Baddeley, A. D. (2003). Disruption of verbal STM by irrelevant speech, articulatory suppression and manual tapping: Do they have a common source? Quarterly Journal of Experimental Psychology A, 56, 1249–1268. Lewandowsky, S., & Oberauer, K. (2009). No evidence for temporal decay in working memory. Journal of Experimental Psychology: Learning, Memory and Cognition, 35, 1545–1551. Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory for features and conjunctions. Nature, 390, 279–281. Mattys, S., & Baddeley, A. D. (2019). Working memory and second language accent acquisition. Applied Cognitive Psychology, 33, 1113–1123. Miller, G. A. (1962). Psychology: The science of mental life. Harper & Row. Norris, D., Butterfield, S., Hall, J., & Page, M. P. A. (2018). Phonological recoding under articulatory suppression. Memory & Cognition, 46, 173–180. Oberauer, K. (2021). Towards a theory of working memory: From metaphors to mechanisms In R. H. Logie, V. Camos, & N. Cowan (Eds.), Working memory: State of the science (pp. 116–149). Oxford University Press. Popper, K. (1959). The logic of scientific discovery. Hutchison. Savin, H. B., & Perchonok, E. (1965). Grammatical structure and immediate recall of English sentences. Journal of Verbal Learning and Verbal Behavior, 4, 348–353. Shallice, T., & Warrington, E. K. (1970). Independent functioning of verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261–273. Shipstead, Z., Harrison, T. L., & Engle, R. W. (2015). Working memory capacity and the scope and control of attention. Attention, Perception & Psychophysics, 77, 1863–1860. Toulmin, S. (1953). The philosophy of science. Hutchison. Turner, M. L., & Engle, R. W. (1989). Is working memory capacity taskdependent? Journal of Memory and Language, 28, 127–154.
https://doi.org/10.1017/9781108955638.004 Published online by Cambridge University Press
Part II
Models and Measures
https://doi.org/10.1017/9781108955638.005 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.005 Published online by Cambridge University Press
3 The Evolution of Working Memory and Language Frederick L. Coolidge and Thomas Wynn
3.1
Introduction
The maintenance of sensory information long enough to make appropriate decisions (particularly movement) is a characteristic of all living things, including single cells, plants, and animals. This ability constitutes one of the two main functions of working memory, that is, short-term storage and processing, and thus, its evolution began with the very first forms of life, the prokaryotes about 3.9 billion years ago. Experimental psychologist Alan Baddeley in 1974 with his postdoctoral student Graham Hitch (Baddeley & Hitch presented a more highly detailed, multicomponent, working memory model). Baddeley’s (2001, 2007, 2012) working memory model has subsequently dominated memory research for nearly five decades. For example, a Google search since the year 2000 yields over 2 million hits. It is the purpose of this chapter to trace the evolutionary foundations for working memory and its relationships to the evolution of language. Baddeley’s multicomponent working memory model has at least three important attributes: (1) it subsumes nearly all current memory categories, including short- and long-term memory, explicit (declarative/semantic) memory and implicit (procedural) memory, and importantly, episodic memory and visuospatial memory; (2) it covers the generic definition of working memory, which is what can be actively held in attention despite interference; and (3) it provides a theoretical and empirical basis for the critically important executive functions of the frontal lobes (e.g., Coolidge & Wynn, 2001), which were first described by the Russian neuropsychologist Alexander Luria (1966/2012). Prior to Baddeley’s multicomponent working memory model, research in memory was dominated by verbal short- and long-term memory (see Atkinson & Shiffrin, 1968; Underwood, 1966). Much of this research focused almost exclusively on the study of an acoustic or written, limitedcapacity acquisition of words, nonsense syllables, or numbers. By the early
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
32
FREDERICK L. COOLIDGE AND THOMAS WYNN
1970s, psychologists like Baddeley had recognized the limitations in this simple two-component model of memory and its nearly sole focus on verbal material. Baddeley and Hitch’s (1974) initial work developed a more comprehensive cognitive theory that not only accounted the standard operations of both verbal and nonverbal, short- and long-term memory and but also attention, goal-directed behavior, decision-making, and inhibition. Their initial model included an attentional, multimodal controller or central executive, and two subsystems, the phonological loop and the visuospatial sketchpad. Baddeley (2001) expanded the central executive’s functions by adding a new component, the episodic buffer, which he viewed as integrating and temporarily storing information from the other two subsystems and serving as a temporary memory storage facility for the central executive. The phonological loop contains two elements: a short-term phonological store of speech and other sounds and an articulatory loop that maintained and rehearsed information either vocally or subvocally. Baddeley viewed the phonological store’s critical purpose to be the acquisition and comprehension of language. He hypothesized that the visuospatial sketchpad involved the maintenance and integration of visual (“what” information, like objects) information and spatial (“where” information, like location in space) elements and a means of refreshing them by rehearsal.
3.2
The Central Executive and Executive Functions of the Frontal Lobes
Baddeley (2001) viewed the central executive as having varying cognitive and behavioral functions including attention, active inhibition, decisionmaking, planning, sequencing, temporal tagging, and the updating, maintenance, and integration of information from the two subsystems. His concept of a central executive is synonymous with the characteristics of the executive functions of the frontal lobes from the clinical neuropsychology literature, which was first described by Russian neuropsychologist Alexander Luria (1966/2012). Luria’s patients were World War I and World War II military personnel and others who had impaired cognitive and psychological functioning because of war wounds, head injuries, strokes, and other means. It is important to remember that the central executive not only to focuses one’s attention to tasks and all the aforementioned executive functions of the frontal lobes, but it also serves as the critical liaison to the long-term memory systems, including language comprehension and production. Further, the amount of information that can be held in attention despite interference is referred to as working memory capacity. Research has also shown that working memory capacity is moderately or strongly predictive of a wide variety of higher cognitive measures, including reading and
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
Evolution of Working Memory and Language
listening ability, language comprehension, vocabulary, reasoning, writing, computer-language learning, and many others (e.g., Kane & Engle, 2002). Working memory capacity is also strongly correlated with intelligence, both crystalized intelligence (learned information) and fluid intelligence (novel problem-solving; Conway et al., 2005). Working memory capacity, whether measured through individual components like executive functions, phonological storage capacity, or visuospatial sketchpad capacity or by measuring a composite of the components, is also highly heritable and polygenic (i.e., involving multiple genes), as demonstrated by empirical research (see Wynn & Coolidge, 2010, for a review; also Friedman et al., 2008). It is also important to note that to date there is no measure of domain-free working memory capacity, as the cognitive functions associated with working memory must be measured in some domain, be it phonological or visuospatial.
3.3
The Evolution of the Central Executive
If the essence of the cognitive functions of the central executive is appropriate attention, active inhibition, decision-making, planning, and sequencing, then it is highly probable that their inchoate beginnings appeared in the simple flatworms of the Cambrian period, about 545 million years ago. Because modern versions of flatworms, like planaria and amphioxus, exhibit both nonassociative (habituation and sensitization) and associative learning (classical and operant conditioning), it can be assumed that the first simple flatworms exhibited many of the functions of Baddeley’s central executive: making decisions to move away from useless or toxic chemicals or organisms (habituation or active inhibition), deciding to move toward valuable chemicals or useful organisms (sensitization, planning, or sequencing), and the memorization of these movements and locations (procedural and visuospatial memories). Because these flatworms possessed photosensitive “eyes,” their inchoate central executive functions also incorporated visuospatial information into their movements. Thus, one can argue that the evolution of the central executive component and the visuospatial components in Baddeley’s working memory model began at least 545 million years ago. Two other evolutionary events ultimately influenced modern cognition: the divergence of mammals from reptiles about 200 million years ago and the divergence of primates from other mammals about 65 million years ago. The former included a vastly increased size of the cerebral cortex relative to the rest of the brain, such that it has been labeled the neocortex. It is thought that this cortical expansion gave mammals greater behavioral flexibility than the more behaviorally reflexive reptiles. The divergence of primates from other mammals placed greater emphasis on the expansion and importance of the phonological loop, as will be explicated shortly.
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
33
34
FREDERICK L. COOLIDGE AND THOMAS WYNN
3.4
The Phonological Loop and Its Evolution
Baddeley hypothesized that the phonological loop has two components: a brief, sound-based, phonological storage that fades within a few seconds and an articulatory processor. Spoken sounds can gain long-term memory storage by vocal or subvocal rehearsal by means of the articulatory processor. Auditory or spoken information has automatic and obligatory access to phonological storage by means of the strength of its emotional valence and the number of repetitions. A sound with a strong emotional valence does not require as many repetitions for later recall. A sound with a weaker emotional valence will require more repetitions (e.g., Coolidge, 2019, 2020). Baddeley proposed that the phonological loop evolved principally for the demands and acquisition of language. Baddeley et al. (1998) noted the critically important role that the phonological loop plays in the acquisition of a child’s native language and in adult second-language learning. He also wrote that “the phonological loop might reasonably be considered to form a major bottleneck in the process of spoken language comprehension” (Baddeley & Logie, 1999, p. 41). The process of vocal and subvocal articulation also appears to play a critical role in memorizing visual stimuli (Baddeley’s visuospatial sketchpad in conjunction with phonological storage), for example, thinking or saying, “Ah, a small green table!” Interestingly, Baddeley also proposed a fantastical characteristic of the result of the integration of contents of the phonological loop with the visuospatial sketchpad when he said how easy it is to imagine a blue, ice hockey–playing elephant. The functions of the phonological loop also help to explain why brain-damaged patients who cannot create a sound or speech through the phonological loop cannot memorize new material (e.g., Caplan & Waters, 1995). The latter most often occurs with damage to the temporal lobe area of the left hemisphere (also called Wernicke’s area or BA 40). The great emphasis of the phonological loop in modern humans (which includes inner speech and subvocal thinking and reasoning) can be traced to the emergence of primates about 65 million years ago when they diverged from the lineage of other mammals. To compete with other animals for food, particularly nutritious fruits, these small (fist-sized), arboreal, nocturnal, socially gregarious animals coordinated their foodgathering activities with intentionally directed vocalizations. Brains are an extraordinary calorie-requiring metabolic tissue, requiring over 10 times the number of calories than any other organ to function sufficiently. In addition, frugivorous (fruit-eating) primates have more convoluted and larger brains than folivorous (leaf-eating) primates, because of the greater caloric content in fruit. Chilean neuropsychologist Francisco Aboitiz and his colleagues emphasized that phonological storage’s carrying capacity was naturally selected for in these early primates because of their highly vocally mediated social activities (Aboitiz, 2017; Aboitiz et al., 2006, 2010).
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
Evolution of Working Memory and Language
The reasons for these vocalizations (i.e., pragmatics of speech) may have been limited: warnings of predators, alerts to conspecifics of food sources, and expressions of pleasurable or painful emotional states. This emphasis upon sounds and thus phonological storage capacity for communication among conspecifics and the coordination of their social activities was the inchoate foundation for modern human language according to Aboitiz and his colleagues. Further, they postulated that this selection for an extended phonological storage capacity allowed the processing of sounds and the conveyance of more elaborate meanings, which eventually led to syntactic processing (Aboitiz et al., 2006). They further hypothesized that this expanding memory system ultimately allowed for the processing and storage of more complex items, which may have allowed their subsequent combinatorial manipulation. It is also likely that this phonological expansion placed greater demands upon the central executive’s functions. The hypothesis that early primates may have had the ability to combine sounds is based on extant putty-nosed monkeys. They have at least two distinct vocal predator alarm calls, and they appear to elicit two different evasive actions. However, when those two sounds are combined, a group behavior is elicited, which appears to mean something like, “Let us move to another place for food” (Schlenker et al., 2016; Scott-Phillips & Blythe, 2013). Thus, a critical prerequisite of combinatorial sound communication is an organism’s sociocognitive capacities and intentionally outward expressions of sounds toward conspecifics and the recognition by conspecifics that these sounds represent intentions. It is important to note that the sociocognitive capacity of these organisms does not have to be substantial, and it does not have to be “symbolic.” A dog can learn to associate the sound of a bell to the advent of food, if the bell and food are paired in close time. This is called classical or Pavlovian conditioning. Even the simplest of animals (e.g., planaria) can be classically conditioned as long as the organism has a rudimentary brain and nervous system. It does not have to be assumed that the bell “symbolizes” food to the dog or a light symbolizes food to a planaria. The bell elicits a dog’s saliva because nervous systems can be classically conditioned. There is no need to invoke higher cognitive explanations if simpler ones suffice (Occam’s razor).
3.5
The Visuospatial Sketchpad and Its Evolution
As noted previously, the visuospatial sketchpad is a temporary store for the maintenance and manipulation of visual and spatial information (Baddeley 2001, 2012). Baddeley proposed that it played an important role in visuospatial navigation, in spatial orientation, and in solving visuospatial problems. It is thought that visuospatial sketchpad forms an interface between the two systems (Shah & Miyake, 2005), as empirical studies demonstrate
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
35
36
FREDERICK L. COOLIDGE AND THOMAS WYNN
that the visual and spatial processes actually constitute separate memory systems. Of course, visual information can be integrated with other senses, although the most common primary sense that visual information is integrated appears to be auditory information. Because the first simple flatworms that appeared about 545 million years ago had rudimentary photosensitive “eyes,” it might be surmised that the visual aspect of the Baddeley’s visuospatial sketchpad evolved well before the phonological component. In flatworms, the coevolution of the visual and spatial components appears to be most likely. However, the spatial component probably evolved much earlier than the visual component, as modern bacteria (which resemble the ancient prokaryotes), can orient spatially, even if it is not a true spatial navigation, but only moving toward or away from particular stimuli.
3.6
The Evolution of the Episodic Buffer
Nearly two and one-half decades after Baddeley first proposed his multicomponent theory of working memory, he recognized his concept of a central executive lacked a memory component. Thus, in 2001, he hypothesized the existence of an episodic buffer, which would serve as the temporary memory workspace for the central executive. Just as Luria’s metaphorical concept of the executive functions of the frontal lobes heralded Baddeley’s central executive, so did experimental psychologist Endel Tulving’s (1972) concept of episodic memory presage Baddeley’s episodic buffer. In Tulving’s now classic paper, he established a firm distinction between knowing something, like a fact or detail (i.e., declarative/semantic memory), and remembering a scene form one’s past that resembled a short movie clip (i.e., an episodic memory). As noted earlier, episodic memories are a subcategory of explicit memory, but they have distinctive characteristics from other explicit memories like declarative memory. For example, the recall of past experiences is typically visual, and if they have a high emotional valence (either positive or negative), they can persist for a very long time, as in the recall of traumatic events or joyful events. Most often episodic memories are recalled as short experiences (a few moments) and they are then reexperienced subjectively, usually with the original accompanying emotional valence (sad or happy). In the case of strong traumatic memories, such as in posttraumatic stress disorder, the episodic memory and its emotional attachment may persist indefinitely, although psychotherapy may reduce the negative feelings associated with the memory. Baddeley (2001, 2012) did elaborate beyond Tulving’s (2002) concept of episodic memory, as he hypothesized that the episodic buffer integrated, by means of a multimodal or panmodal code, the two subsystems, the phonological loop and the visuospatial sketchpad. It also integrated ideas and
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
Evolution of Working Memory and Language
meanings from long-term declarative/semantic memory. Because he imbued the episodic buffer with the ability to attend simultaneously to multiple sources of information, it gave the central executive the ability to create various models of the environment, which could be mentally manipulated to solve old and new problems, plan for future actions, and debate alternative courses of actions. Further, if an initial plan failed, another could be generated and executed. Baddeley also emphasized that the central executive’s use of the episodic buffer to solve problems did not simply consist of calling up old solutions with the best probability of success (which it is capable of doing), but that it genuinely generated new solutions to solve novel problems. Baddeley’s notion of a novel problem-solving central executive with help from the episodic buffer was also foreseen earlier by cognitive psychologist Roger Shepard (1997), who proposed the idea of “thought experiments.” He thought that every actual attempt to solve a problem was preceded by mental manipulations of that problem, which drew upon “countless” prior real experiences; however, these prior experiences could be recombined in one’s mind to produce a genuinely novel solution. This ability, he thought, was the critical evolutionary advantage of recollections. Further, he hypothesized that these internalized thought experiments would avoid trial and possibly fatal error. Shepard did not label his idea of thought experiments as types of episodic memory, but it is clearly evident that Shepard’s thought experiments and Baddeley’s episodic buffer are similar in theory, and they both understood their evolutionary advantages. Tulving (2002) had also proposed that the manipulative characteristic of episodic memories gave rise to the notion of a special form of consciousness, which he called autonoesis, where humans could become aware of the subjective nature of time. This awareness, he stated: allows autonoetic creatures to reflect on, worry about, and make plans for their own and their progeny’s future in a way that those without this capability possibly could not. Homo sapiens, taking full advantage of its awareness of its continued existence in time, has transformed the natural world into one of culture and civilization that our distant ancestors, let alone members of other species, possibly could not imagine. (p. 20) As further evidence of the importance of the episodic buffer in the evolution of modern thinking, and therefore, modern working memory, cognitive psychologist Daniel Schacter (2012) viewed memory systems as evidence of a “dynamic memory system that flexibly incorporates relevant new information” (p. 605). He proposed that the original function of episodic memory was not to recall an episode exactly or perfectly. He hypothesized that evolutionary value of an episodic memory was that relevant or
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
37
38
FREDERICK L. COOLIDGE AND THOMAS WYNN
salient parts of episodes could be recalled, reconstructed and reassembled, and subsequently used to simulate alternative future scenarios and outcomes. Baddeley (2012), as noted previously, proposed that greater working memory capacity would also benefit episodic memories by allowing for the formation of mental models that were more likely to be successful in the future. Individuals could actively choose a future action or create alternative actions, rather than simply choosing the highest path of probable success. One could still benefit from choosing alternatives based on probabilities of success from past attempts, but an active comparison of multiple simulated scenarios combined into novel actions would have a strong selection advantage. Could these models be created without an inner dialogue, that is, without language? Simply based on intuition, all organisms make decisions without language, including humans. People do not internally debate the myriad decisions they make daily. Of course, some decisions are internally debated via language (inner speech), but probably a majority of decisions are not: they are made through “gut feelings” and sometimes pure impulse. At this point, we shall formally address the surprisingly contentious issue of the nature of language and its evolution.
3.7
What Is Language?
The contentious nature of language begins with its very definition. Some linguists define language in a broad sense, that is, a communication system with both the required sensory-motor neural apparatus and a conceptualintentional system for the delivery of the communication mostly to conspecifics whether by intention or not (e.g., Hauser et al., 2002; Tallerman & Gibson, 2012). Hauser and colleagues labeled this communication system, shared by most animals (including humans), the faculty of language in the broad sense (FLB). They labeled the communication system shared only by humans, the faculty of language in the narrow sense (FLN), whose exclusive characteristic is recursion, that is, the theoretically infinite ability to nest phrases within phrases, which then often modifies or creates contingencies within sentences (e.g., if-then statements; also see Coolidge et al., 2010). They also called recursion, the “hallmark” of modern language. In their categorization, FLN is a subset of the larger FLB. Other linguists and anthropologists are not so gracious, and they reserve any use of the word language exclusively to humans. For example, Tallerman and Gibson (2012) wrote, “[L]anguage is an autapomorphy, i.e., a derived trait found only in our lineage, and not shared with our other branches of our monophyletic group (say, the group of primates, or the group of great apes). We also have no definitive evidence that any species other than Homo sapiens ever had language” (p. 2). Anderson (2013) similarly wrote, “[L]anguage in the human sense seems to be distinct to our species . . . just as other animals’ communication systems are part of their specific makeup” (p. 19). Corballis
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
Evolution of Working Memory and Language
(2017) stated, “[W]e seem to take language for granted, a gift bestowed on us for the privilege of being human. Of course, other animals do communicate, but their communications have nothing approaching the sheer vastness of human language” (p. ix).
3.8
Why Chomsky Is Wrong about the Evolution of Language
Let us begin with the strongest wordings possible about the evolution of language. Language evolved for the purposes of communication (e.g., Lieberman, 2005). Language was visible to natural selection (e.g., Barham & Everett, 2020) in part, due to the phenotypes of its polygenic basis (e.g., de Boer et al., 2020). Language evolved over millions of years, and it had a strong foundation with simpler forms of aurally directed communications in the earliest, highly social primates that first appeared about 65 million years ago (e.g., Aboitiz, 2017; Barham & Everett 2020). Language subsequently developed into a protoform (e.g., Botha, 2010; Botha & Everaert, 2013; Tallerman, 2007), called protolanguage, whose first users, very likely Homo erectus, may have possessed a substantial vocabulary, but likely simpler or absent syntax or grammar (e.g., Coolidge & Wynn, 2018). Finally, comparative studies of extant animals may shed some light on language’s evolution (e.g., Gibson, 2012; Lemasson et al., 2013). All of the these claims are contrary to Chomsky and his colleagues’ claims that they have made about language for nearly the past two decades (e.g., Berwick & Chomsky, 2017; Bolhuis et al., 2014; Fitch et al., 2005; Hauser et al., 2002). We shall not debate the merits (or lack of evidence) of their conjectures but focus upon ours and others’ previous claims. We shall also take a linguistic path much less traveled: We believe that language’s evolutionary trajectory was most highly influenced by its pragmatics, that is, the reasons for making sounds or speaking (e.g., Coolidge & Wynn, 2009). As noted earlier, putty-nosed monkeys have at least two distinct predator calls and a combination of predator sounds that affect group behavior and have nothing to do with predators. Further, modern vervet monkeys have at least three different sounds for three different predators that elicit distinctively different responses, and they teach these different sounds to their young (Price et al., 2015; Seyfarth et al., 1980). Homo erectus, which appeared about 2 million years ago, had brains at least 12 to 15 times larger than most extant monkey brains, which makes it highly unlikely that Homo erectus did not have at least the same or much greater abilities at attaching meanings or intentions to sounds as modern monkeys. Homo erectus also had a much more convoluted brain compared to all extant monkeys (providing an overall greater surface area for cognitive processing). Thus, it is an absurd contention to think that Homo erectus, and certainly Neandertals (with even 50 percent larger brains than Homo erectus),
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
39
40
FREDERICK L. COOLIDGE AND THOMAS WYNN
were mute, only gestured, only sang or whistled, or were alinguistic (e.g., Berwick et al., 2013; Mithen, 2006). It is critically important to remember that the simplest of organisms (like planaria), which have at least a rudimentary brain and nervous system, can be classically conditioned to respond to environmental stimuli. Again, the acquisition of meanings to sounds (or other stimuli) does not require symbolism or symbolic thinking (whatever those words might denote or connote) because meanings can be acquired through simple classical conditioning. By meanings, we mean that environmental stimuli can be reliably associated to particular behaviors, without resorting to any references whatsoever to symbols. This expanded use of meaningful sounds must have put evolutionary pressure on phonological storage and declarative memory, not only for keeping track of predators but also for remembering individuals in alliances, cheaters, alpha members, and so forth. (e.g., Aboitiz et al., 2006, 2010). As noted earlier, the pragmatics (reasons for speech) of these early primate vocalizations were likely limited: expressions of pleasurable or painful emotional states, warnings of predators, and alerts to conspecifics of food sources. Thus, the earliest pragmatic of language was likely an intrapersonal vocal expression of emotional states, which are called exclamatives. Interestingly, in patients with Broca’s aphasia (unable to speak but language comprehension is preserved), they are frequently able to cuss but they do so inadvertently, that is, short, overlearned expletives when a patient might drop something or be surprised. We believe that these intrapersonal expressions quickly evolved into interpersonal expressions, where an initial intrapersonal emotional expression is classically conditioned (Pavlovian conditioning) or operantly conditioned (through rewards or punishment for behaviors) into meanings for conspecifics or others. Thus, a sharp, sudden yelp of intrapersonal fear of a predator is learned (conditioned) by others to be associated with a warning. Of course, exclamative pragmatics were not restricted to the earliest primates (65 million years ago) or mammals (appearing about 200 million years ago), as the earliest reptiles (over 300 million years ago) had aurally directed sounds. The second pragmatic of speech may have developed subsequently or in conjunction with the first pragmatic of speech. This pragmatic of speech was the imperatives (commands), as extant primates, mammals, and reptiles display social hierarchies, often between dominant and submissive members, young and older members, or smaller and bigger members. While these sounds may not have been or be explicitly distinctive and consistent, as growls or yelps may vary, no doubt they were and are accompanied by posturing and facial gestures that makes their intent clear – move, get out of my way, give me that, and so forth. The third pragmatic of speech may have required a substantial cognitive leap, and thus a neurological leap as well. This evolutionary transition of primates to the first simians probably began about 35 million to 40 million years ago, as the foramen magnum shows the beginning of shift from a
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
Evolution of Working Memory and Language
more posterior position to a medial one. The latter change suggests a more upright posture, and most likely an increase in the brain-to-body ratio (encephalization quotient). If multiple kinds of modern monkeys have names for predators, which are taught to their young, it is highly likely so did these early simians, and naming things is a least a protoform of the declarative pragmatic of speech. The fourth pragmatic of speech, interrogatives (i.e., asking questions), might have developed well after the development of declaratives, and it most likely would also have required yet another substantial cognitive and neurological leap, perhaps, even beyond the appearance of the australopithecines and habilines at about 4 million years ago to 2.5 million years ago. Based on an average brain size of about 900 cc with a range of about 550 cc to 1,250 cc, Homo erectus might have been the first hominin to be able to ask or answer a question. As Homo erectus appeared about 2 million years ago, this would place the evolution of at least four pragmatics of speech in place by that time. It is well accepted that Homo erectus had made a full transition to terrestrial life (Coolidge & Wynn, 2006), unlike the largely arboreal lives of the australopithecines and habilines. Anthropologist Robin Dunbar (1998) has proposed that brains at the time of Homo erectus were challenged by the social demands of larger groups, as the number of australopithecine or habilines living in nests in trees would have had a limit before attracting predators rather than protecting from predators. There are suspicions that the Homo erectus group size would have doubled to about 100 or so members. Further, it is theorized that as the group size expanded, so did the territory necessary to sustain them. It is estimated that Homo erectus’s territory expanded 10-fold over that of the australopithecines and habilines, to over 100 square miles. Bramble and Lieberman (2004) and Wells and Stock (2007) hypothesized that Homo erectus had achieved fully metabolically efficient bipedalism and a body (tall and relatively thin) designed for long-distance running (Lieberman et al., 2009), which would have been an aid to territorial exploration. Yet not only would full terrestrial life placed demands for a larger vocabulary (and, of course, upon phonological storage capacity), but Homo erectus’s penchant for exploring new territory and its challenges would have selected for an expanded visuospatial sketchpad, greater behavioral plasticity, and enhanced fluid intelligence (novel problem-solving). The archaeological evidence for these four pragmatics of speech in Homo erectus is admittedly slim yet strong: the remarkable handaxes of Homo erectus (e.g., Overmann & Coolidge, 2019). These symmetrical, leaf-shaped stone tools were so qualitatively advanced over the australopithecine and habilines stone tools (sharp flakes, cores, and hammer stones), that their design persisted for nearly the next 2 million years. Cachel and Harris (1995) hypothesized that Homo erectus may be likened to a weed species, who were able to invade distant environments (leaving Africa for Europe, Asia, and southern Asia), disrupted environments (volcanoes, earthquakes, violent storms, fires, drought, floods, etc.), and varying
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
41
42
FREDERICK L. COOLIDGE AND THOMAS WYNN
environments (deserts, tropics, etc.). It appears that Homo erectus was not only able to live in these places but even to thrive in them. This capacity to adjust to fluctuating and unpredictable environments would also favor the behavioral plasticity associated with the basic language skills that the four pragmatics of speech we have described previously.
3.9
Why Chomsky May Have Gotten Something Right about Language’s Evolution
The evolution of language has now fallen into two camps: the gradualists like Darwin (1871), Pinker and Bloom (1990), Bickerton (2000), and many others. They believe that language’s origins began gradually, as we described previously, such as aurally directed primate calls and then protolanguages associated with Homo erectus. The other school of thought is saltationist. Like Chomsky and others, they believe that language’s onset was sudden, which they postulate took place about 100,000 years ago, and there was no language before that time (e.g., Bolhuis et al., 2014; Fitch et al., 2005; Hauser et al, 2002). They hypothesize that it was a single gene mutation in one person, and they offer archaeological evidence by Tattersall (2008) at this time as proof. There is no reasonable evidence whatsoever that a single gene caused language as a recent article attests (e.g., de Boer et al., 2020), and the archaeological evidence points to a much more recent dramatic change, that is, 50,000 years ago (Coolidge & Wynn, 2018; Klein, 2009; Mithen, 1996), rather than 100,000 years ago. Although Chomsky’s single gene hypothesis for language has no scientific basis, the notion that something happened to language about 100,000 years ago or more recently may have some basis. About 150,000–100,000 years ago, the overall brain size between Neandertals and Homo sapiens was not significantly different. Their brains were both comparatively large, about 1,550 cc, compared to modern Homo sapiens at about 1,350 cc. Archaeologically, their cultures were also more similar than different. They both intentionally buried their dead, but without grave goods, simple or elaborate. Neither of the two human types produced artistic figurines, painted on cave walls, or made personal ornaments. About 75,000 years ago, there were hints that both cultures were changing, but in Homo sapiens the change was more dramatic. Around this time, Neandertals and Homo sapiens both produced personal ornaments like shell beads and claw necklaces, and Neandertals may have worn feathers as ornamentation. They both used ochre for hafting points on spears (Wadley, 2010; Wynn, 2009). But about this time, Homo sapiens invented the bow and arrow technology (e.g., Coolidge et al., 2016), which Neandertals never did. Around 50,000 years ago, Homo sapiens began a migration into Europe, where Neandertals had dwelled successfully for over 200,000 years or more. Homo sapiens carried a culture into Europe that archaeologists term the Aurignacian.
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
Evolution of Working Memory and Language
This culture included elaborate and depictive cave paintings, which were very different from the simple geometric etchings or paintings of Neandertals (Rodríguez-Vidal et al., 2014). Homo sapiens’s burials began to include highly elaborate grave goods. Although there was some very limited interbreeding between Neandertals and Homo sapiens during this period, the two human groups remained largely genetically distinct. And within a relatively short period (about 10,000 years) of geographical overlap, Neandertals went extinct. Some refer to this dramatic change in culture in Homo sapiens as a Cultural Explosion (e.g., Klein, 2000; Mithen, 1996). However, some (Finlayson, 2019; Rodríguez-Vidal et al., 2014; Villa & Roebroeks, 2014.) refuse to acknowledge a qualitative distinction between a simple hashtag engraving by a Neandertal in Gibraltar about 41,000 years ago and the elaborate cave paintings in Chauvet made by modern humans only a few thousand years later. A majority of anthropologists and archaeologists do see a demonstrable qualitative difference between Homo sapiens’s Aurignacian culture and the final throes of Neandertal culture beginning at about 40,000 years ago. Some claim that these vast cultural differences could simple be attributed to differing cultural traditions or simple cultural drift, akin perhaps, to genetic drift. Evolutionary biologist E. O. Wilson (1978) insightfully pointed out that genes place a tight leash upon culture, and it is much more likely that genotypic differences between Neandertals and Homo sapiens yielded phenotypic differences that were visible to natural selection. Is it possible that a single gene mutation could make a significant difference in outcomes, that is, extinction versus survival? We would answer emphatically, yes! Evans et al. (2005) found evidence that a single gene (MCPH1), related to a group of genes that regulate brain growth (microcephalin) arose in Homo sapiens somewhere between 60,000 and 12,000 years ago. This variant MCPH1 dramatically changed Homo sapiens’s brains; however, the exact nature of the genotypic and phenotypic changes is still not yet understood. Paleoneurologist Emiliano Bruner (2004; Bruner & Iriki, 2016) was the first to demonstrate that modern Homo sapiens’s brains, although getting smaller in overall size compared to Neandertals’ over time, were subject to a parietal lobe expansion. Paleoneurologists Neubauer et al. (2018) have shown that this brain shape change in Homo sapiens occurred within the last 100,000 years, perhaps as recently as 30,000 years ago. This range of dates coincides nicely with the proposed cultural explosion beginning about 50,000 years ago and with the genetic influence on brains noted previously.
3.10
The Fifth and Final Pragmatic of Speech
We have written extensively about the cognitive nature of this confluence of changes (genotypic and phenotypic) in the evolution of modern Homo
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
43
44
FREDERICK L. COOLIDGE AND THOMAS WYNN
sapiens (e.g., Coolidge & Wynn, 2001, 2005, 2018; Wynn & Coolidge, 2009, 2010). We labeled the cognitive change enhanced working memory, and we have attributed these dramatic cultural changes in Homo sapiens within the last 100,000 years to a change in working memory capacity. Given the contentious nature of language discussions, it might be wise to sidestep language and recursion issues and their relationship to these dramatic cultural events. However, we believe they may be significantly linked. The fifth pragmatic of speech is said to be the subjunctive mood. Subjunctive statements express wishes, desires, things imagined, and hypotheses contrary to facts. Interestingly, at least in English, nearly all subjunctive expressions have two clauses, each containing a verb, one declarative (aka indicative mood) and one subjunctive. For example, in the sentence “I wish that I were stronger,” the first clause, “I wish,” is declarative, although it appears to be subjunctive. The second clause, “that I were stronger,” is subjunctive because it is a statement contrary to fact, that is, one would not actually wish one was stronger, if indeed, one was stronger already. Aside from the issue of body dysmorphia in that statement, it is clear that subjunctive sentences almost always contain two clauses with a verb each. The only exception might be (as TW reminds FLC), “Were I King.” However, as FLC reminds TW, implied in that statement is “I wish that. . .” and thus, the subjunctive is implied within the single clause.
3.11
Cognitive Requirements of the Subjunctive Pragmatic of Speech
One obvious yet often neglected characteristic of recursion is the phonological storage component, whether vocal or subvocal (inner speech). We shall not address the issue of thinking without language, but evidence for it is plentiful (e.g., Corballis, 2017; Pinker, 2007). However, if Chomsky’s (2015) notion is accepted of an I-language (internal thinking) preceding Elanguage (spoken or written), then adequate phonological storage capacity is likely a sine qua non for recursive thinking and recursive expression of those thoughts. As Corballis (2011) has noted, recursive thinking does depend upon working memory and executive processes, although he provided no greater detail. But what is the evolutionary value of recursive thinking and recursive expression? As noted previously, there is paleoneurological evidence that one of the most dramatic brain shape changes in Homo sapiens compared to Neandertals was an expansion in medial and superior regions of their parietal lobes and displacement of the inferior parietal regions into the posterior and superior regions of their temporal lobes (the home of inner speech and language comprehension). We believe that the modern cognitive functions of these regions might have played a critical role in Homo sapiens’s fully modern recursive thinking and language, particularly the
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
Evolution of Working Memory and Language
episodic buffer or episodic memory (Coolidge, 2014).We earlier noted the importance of Baddeley’s episodic buffer (2001), Tulving’s episodic memory (1995, 2002), and Shepard’s (1997) thought experiments in the solving of novel problems (i.e., fluid intelligence). It may be that the cultural explosion was at least in part due to enhanced episodic memory. Since the visuospatial sketchpad and phonological storage work in synergistic tandem, tagging imagined scenarios within linguistic tags may have aided novel problem-solving. However, the manipulation of episodic elements by the central executive may be the essential nature of recursive thinking. As noted earlier, Homo sapiens may have extracted salient aspects of previous episodic memories and mixed them in various ways (i.e., thought experiments), and thus, may have begun the subjunctive pragmatic of speech. As Corballis (2011) has noted, we were “intentionally sifting through different episodes in our lives to extract relevant information – perhaps in a fashion analogous to the insertion of phrases into sentences, or sentences into narratives (italics ours; p. 99).
3.12
Conclusion
We have presented the hypothesis that working memory and language did evolve in tandem. We have borrowed from the gradualists that language did evolve slowly from the aurally directed calls of early primates to the protolanguage forms of Homo erectus and that one of language’s first purposes was communication, although that position does not rule out Chomsky’s notion of inner language preceding external communication. However, we have presented the novel idea that the pragmatics of speech also evolved in tandem with the evolution of working memory. Further, the first four pragmatics of speech – exclamatives, imperatives, declaratives, and interrogatives – may well have been in place about 2 million years ago. However, we have also borrowed from the saltationists that something did indeed suddenly happen to language sometime more recent than 100,000 years ago. We are proposing that the fifth pragmatic of speech, the subjunctive mood, required fully modern working memory capacity and that a genetic mutation was responsible for a parietal lobe expansion, which is critically involved in episodic memory recall and simulation. The phenotypic result of this genotypic change meant that thought experiments could be conducted in a recursive manner. Thus, we agree with Chomsky and his colleagues that recursion may be considered the hallmark of modern language, but only because recursive thinking, based on the manipulation of episodic memories, allowed the emergence of the fifth and most powerful pragmatic of speech, the subjunctive. We think that the fruits of Homo sapiens’s cultural explosion – cave art, creative figurines, highly ritualized burials – were the direct result of the wishes, imaginings, and ideas contrary to fact that arise from subjunctive thinking.
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
45
46
FREDERICK L. COOLIDGE AND THOMAS WYNN
References Aboitiz, F. (2017). A brain for speech: A view from evolutionary neuroanatomy. Springer. Aboitiz, F., Aboitiz, S., & García, R. R. (2010). The phonological loop: A key innovation in human evolution. Current Anthropology, 51(S1), S55–S65. Aboitiz, F., García, R. R., Bosman, C., & Brunetti, E. (2006). Cortical memory mechanisms and language origins. Brain and Language, 98(1), 40–56. Anderson, S. R. (2013). What is special about the human language faculty and how did it get that way? In R. Botha & M. Everaert (Eds.), The evolutionary emergence of language: Evidence and inference (pp. 18–41). Oxford University Press. Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation: Advances in research and theory (Vol. 2, pp. 89–195). Academic Press. Baddeley, A. D. (2001). Is working memory still working? American Psychologist, 56, 851–864. Baddeley, A. D. (2007). Working memory, thought, and action. Oxford University Press. Baddeley, A. D. (2012). Working memory: Theories, models, and controversies. Annual Review of Psychology, 63, 1–29. Baddeley, A. D., & Hitch, G. (1974). Working memory. Psychology of Learning and Motivation, 8, 47–89. Baddeley, A. D., & Logie, R. H. (1999). Working memory: The multiplecomponent model. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (p. 28–61). Cambridge University Press. Baddeley, A., Gathercole, S., & Papagno, C. (1998). The phonological loop as a language learning device. Psychological Review, 105(1), 158–173. Barham, L., & Everett, D. (2020). Semiotics and the origin of language in the lower Palaeolithic. Journal of Archaeological Method and Theory, 1–45. Berwick, R. C., & Chomsky, N. (2017). Why only us: Recent questions and answers. Journal of Neurolinguistics, 43, 166–177. Berwick, R. C., Hauser, M., & Tattersall, I. (2013). Neanderthal language? Just-so stories take center stage. Frontiers in Psychology, 4, 671. Bickerton, D. (2000). How protolanguage became language. In C. Knight, M. Studdert-Kennedy, & J. Hurford, The evolutionary emergence of language: Social function and the origins of linguistic form (pp. 264–284). Cambridge University Press. Bolhuis, J. J., Tattersall, I., Chomsky, N., & Berwick, R. C. (2014). How could language have evolved? PLoS Biology, 12(8), e1001934. Botha, R. (2010). On the soundness of inferring modern language from symbolic behaviour. Cambridge Archaeological Journal, 20, 345–356.
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
Evolution of Working Memory and Language
Botha, R., & Everaert, M. (Eds.). (2013). The evolutionary emergence of language: Evidence and inference. Oxford University Press. Bramble, D. M., & Lieberman, D. E. (2004). Endurance running and the evolution of Homo. Nature, 432(7015), 345–352. Bruner, E. (2004). Geometric morphometrics and paleoneurology: Brain shape evolution in the genus. Homo: Journal of Human Evolution, 47, 279–303. Bruner, E., & Iriki, A. (2015). Extending mind, visuospatial integration, and the evolution of the parietal lobes in the human genus. Quaternary International, 369, 1–13. Cachel, S., & Harris, J. W. (1995). Ranging patterns, land-use and subsistence in Homo erectus from the perspective of evolutionary ecology. In J. R. Bower & S. Sartono (Eds), Human evolution in its ecological context: Palaeoanthropology: Evolution and ecology of Homo erectus (pp. 52–65). Pithecanthropus Centennial Foundation. Caplan, D., & Waters, G. S. (1995). Aphasic disorders of syntactic comprehension and working memory capacity. Cognitive Neuropsychology, 12(6), 637–649. Chomsky, N. (2015). Some core contested concepts. Journal of Psycholinguistic Research, 44, 91–104. Conway, A. R., Kane, M. J., Bunting, M. F., Hambrick, D. Z., Wilhelm, O., & Engle, R. W. (2005). Working memory span tasks: A methodological review and user’s guide. Psychonomic Bulletin & Review, 12(5), 769–786. Coolidge, F. L. (2014). The exaptation of the parietal lobes in Homo sapiens. Journal of Anthropological Sciences, 92, 295–298. Coolidge, F. L. (2019). The ultimate origins of learning and memory systems. Human Evolution, 34, 21–38. Coolidge, F. L. (2020). Evolutionary neuropsychology: An introduction to the evolution of the structures and functions of the human brain. Oxford University Press. Coolidge, F. L., Haidle, M. N., Lombard, M., & Wynn, T. (2016). Bridging theory and bow hunting: Human cognitive evolution and archaeology. Antiquity, 90, 219–228. Coolidge, F. L., Overmann, K. A., & Wynn, T. (2010). Recursion: What is it? Who has it? How did it evolve? WIRE Cognitive Science, 1, 1–8. Coolidge, F. L. & Wynn, T. (2001). Executive functions of the frontal lobes and the evolutionary ascendancy of Homo sapiens. Cambridge Archaeological Journal, 11, 255–260. Coolidge, F. L., & Wynn, T. (2005). Working memory, its executive functions, and the emergence of modern thinking. Cambridge Archaeological Journal, 15, 5–26. Coolidge, F. L., & Wynn, T. (2006). The effects of the tree-to-ground sleep transition in the evolution of cognition in early Homo. Before Farming: The Archaeology and Anthropology of Hunter-Gatherers, 4, 1–18.
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
47
48
FREDERICK L. COOLIDGE AND THOMAS WYNN
Coolidge, F. L., & Wynn, T. (2018). The rise of Homo sapiens: The evolution of modern thinking. Oxford University Press. Corballis, M. C. (2011). The recursive mind: The origins of human language, thought, and civilization. Princeton University Press. Corballis, M. C. (2017). The truth about language: What it is and where it came from. University of Chicago Press. Darwin, C. (1871). The descent of man, and selection in relation to sex. John Murray. de Boer, B., Thompson, B., Ravignani, A., & Boeckx, C. (2020). Evolutionary dynamics do not motivate a single-mutant theory of human language. Scientific Reports, 10(1), 1–9. Dunbar, R. I. M. (1998). The social brain hypothesis. Evolutionary Anthropology: Issues, News, and Reviews: Issues, News, and Reviews, 6(5), 178–190. Evans, P. D., Gilbert, S. L., Mekel-Bobrov, N., Vallender, E. J., Anderson, J. R., Vaez-Azizi, L. M.,. . .& Lahn, B. T. (2005). Microcephalin, a gene regulating brain size, continues to evolve adaptively in humans. Science, 309(5741), 1717–1720. Finlayson, C. (2019). The smart Neanderthal: Bird catching, cave art, and the cognitive revolution. Oxford University Press. Fitch, W. T., Hauser, M. D., & Chomsky, N. (2005). The evolution of the language faculty: Clarifications and implications. Cognition, 97(2), 179–210. Friedman, N. P., Miyake, A., Young, S. E., DeFries, J. C., Corley, R. P., & Hewitt, J. K. (2008). Individual differences in executive functions are almost entirely genetic in origin. Journal of Experimental Psychology: General, 137(2), 201–225. Gibson, K. R. (2012). Language or protolanguage? A review of the ape language literature. In M. Tallerman & K. R. Gibson, The Oxford handbook of language evolution (pp.46–58). Oxford University Press. Hauser, M. D., Chomsky, N., & Fitch, W. T. (2002). The faculty of language: What is it, who has it, and how did it evolve? Science, 298(5598), 1569–1579. Jackendoff, R., & Pinker, S. (2005). The nature of the language faculty and its implications for evolution of language (Reply to Fitch, Hauser, and Chomsky). Cognition, 97(2), 211–225. Kane, M. J., & Engle, R. W. (2002). The role of prefrontal cortex in workingmemory capacity, executive attention, and general fluid intelligence: An individual-differences perspective. Psychonomic Bulletin & Review, 9(4), 637–671. Klein, R. G. (2000). Archeology and the evolution of human behavior. Evolutionary Anthropology Issues News and Reviews, 9, 17–36. Klein, R. G. (2009). The human career: Human biological and cultural origins. University of Chicago Press. Lemasson, A., Ouattara, K., & Zuberbu, K. (2013). Exploring the gaps between primate calls and human language. In R. Botha & M. Everaert
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
Evolution of Working Memory and Language
(Eds.), The evolutionary emergence of language: Evidence and inference (pp. 181–203). Oxford University Press. Lieberman, D., Bramble, D., Rachlen, D., & Shea, J. (2009). Brains, brawn, and the evolution of human endurance running capabilities. In F. Grine, J. Fleagle & R. Leakey (Eds.), The first humans: Origin and early evolution of the genus Homo (pp. 77–92): Springer. Luria, A. R. (1966/2012). Higher cortical functions in man. Springer. Mithen, S. J. (1996). The prehistory of the mind: The cognitive origins of art, religion and science. Thames and Hudson. Mithen, S. (2006). The singing Neanderthals: The origins of music, language, mind and body. Harvard University Press. Neubauer, S., Hublin, J. J., & Gunz, P. (2018). The evolution of modern human brain shape. Science Advances, 4(1), eaao5961. Overmann, K. A., & Coolidge, F. L. (Eds.). (2019). Squeezing minds from stones: Cognitive archaeology and the evolution of the human mind. Oxford University Press. Parker, A. R. (2006). Evolving the narrow language faculty: Was recursion the pivotal step? In K. Smith (Ed.), Proceedings of the 6th International Conference on the Evolution of Language (pp. 239–246). World Scientific Publishing. Pinker, S. (2007). The stuff of thought. Penguin Books. Pinker, S., & Bloom, P. (1990). Natural language and natural selection. Behavioral and Brain Sciences, 13(4), 707–727. Pinker, S., & Jackendoff, R. (2005). The faculty of language: What’s special about it? Cognition, 95(2), 201–236. Price, T., Wadewitz, P., Cheney, D., Seyfarth, R., Hammerschmidt, K., & Fischer, J. (2015). Vervets revisited: A quantitative analysis of alarm call structure and context specificity. Scientific reports, 5(1), 1–11. Rodríguez-Vidal, J., d’Errico, F., Pacheco, F. G., Blasco, R., Rosell, J., Jennings, R. P., . . . & Finlayson, C. (2014). A rock engraving made by Neanderthals in Gibraltar. Proceedings of the National Academy of Sciences, 111(37), 13301–13306. Seyfarth, R. M., Cheney, D. L., & Marler, P. (1980). Monkey responses to three different alarm calls: Evidence of predator classification and semantic communication. Science, 210 (4471) 801–803. Schacter, D. L. (2012). Adaptive constructive processes and the future of memory. American Psychologist, 67(8), 603–613. https://doi.org/10.1037/ a0029869 Schlenker, P., Chemla, E., Arnold, K., & Zuberbühler, K. (2016). Pyow-hack revisited: Two analyses of putty-nosed monkey alarm calls. Lingua, 171, 1–23. Scott-Phillips, T. C., & Blythe, R. A. (2013). Why is combinatorial communication rare in the natural world, and why is language an exception to this trend? Journal of the Royal Society Interface, 10(88), 20130520. Shah, P., & Miyake, A. (Eds.). (2005). The Cambridge handbook of visuospatial thinking. Cambridge University Press.
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
49
50
FREDERICK L. COOLIDGE AND THOMAS WYNN
Shepard, R. N. (1997). The genetic basis of human scientific knowledge. In D. J. Chadwick & G. Cardew, Ciba Foundation Symposium (pp. 23–38). Wiley & Sons. Tallerman, M. (2007). Did our ancestors speak a holistic protolanguage? Lingua, 117, 579–604. Tallerman, M., & Gibson, K. R. (2012). Introduction: The evolution of language. In M. Tallerman & K. R. Gibson, The Oxford handbook of language evolution (pp. 1–35). Oxford University Press. Tattersall, I. (2008). An evolutionary framework for the acquisition of symbolic cognition by Homo sapiens. Comparative Cognition & Behavior Reviews, 3, 99–114. Tulving, E. (1972). Episodic and semantic memory. In E. Tulving & W. Donaldson (Eds.), Organization of memory (pp. 381–403). Academic Press. Tulving, E. (1995). Organization of memory: Quo vadis? In M. S. Gazzaniga (Ed.), The cognitive neurosciences (p. 839–853). The MIT Press. Tulving, E. (2002). Episodic memory: From mind to brain. Annual Review of Psychology, 53, 1–25. Underwood, B. J. (1966). Experimental psychology. Appleton-Century-Crofts. Villa, P., & Roebroeks, W. (2014). Neandertal demise: An archaeological analysis of the modern human superiority complex. PLoS ONE, 9(4), e96424. Wadley, L. (2010). Compound-adhesive manufacture as a behavioral proxy for complex cognition in the Middle Stone Age. Current Anthropology, 51, S111–S119. Wells, J. C., & Stock, J. T. (2007). The biology of the colonizing ape. Yearbook of Physical Anthropology, 50, 191–222. Wilson, E. O. (1978). What is sociobiology? Society, 15(6), 10–14. Wynn, T. (2009). Hafted spears and the archaeology of mind. Proceedings of the National Academy of Sciences, 106, 9544–9545. Wynn, T., & Coolidge, F. L. (2010). Beyond symbolism and language. Current Anthropology, 51, S5–S16. Wynn, T., & Coolidge, F. L. (2015). Technical cognition, working memory, and creativity. Pragmatics & Cognition, 22, 45–63.
https://doi.org/10.1017/9781108955638.006 Published online by Cambridge University Press
4 The Phonological Loop as a “Language Learning Device” An Update Costanza Papagno 4.1
Introduction
In the present chapter, which is an update of a paper that Baddeley, Gathercole, and I published in 1998, I will refer to the phonological loop, as presented in Baddeley and Hitch’s (1974, 2019 revisited) model. In the working memory model, the phonological loop corresponds to auditoryverbal short-term memory (STM) and therefore these terms will be used interchangeably, as well as the term phonological (P)STM. STM is the capacity to keep a small amount of information in mind for a brief period of time (several seconds to minutes), in the absence of sensory input. STM should be distinguished from working memory (Baddeley & Hitch 1974), which adds processes that support manipulation of the information, providing a basis for goal-directed behaviors (Baddeley 2003, 2012). Therefore, the difference between STM and working memory is in terms of operativity. Briefly, the concept of working memory includes the manipulation of information, while STM concerns only maintenance. The most popular psychological model of auditory-verbal STM distinguishes a component devoted to the storage of verbal information (the phonological short-term store or phonological input buffer), and a process, the articulatory rehearsal, which prevents the decay of the memory traces in the phonological store by recirculating them (see Shallice & Papagno 2019 for a review). After the acoustic analysis, auditory-verbal input has direct access to the phonological input buffer; instead, a written verbal information undergoes a visual analysis, then it is translated in a phonological form that accesses the phonological input buffer through rehearsal (see Papagno et al. 2008). Conrad (1964) and Baddeley (1966a, b) showed that phonological but not semantic similarity negatively affects short-term retention. This effect, namely the decrease of STM capacity in case of phonologically similar
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
52
CO STA N Z A PA PAG N O
material, is known as the phonological similarity effect: in immediate serial recall, level of performance is higher for lists of phonologically dissimilar stimuli than for similar strings (Conrad, 1964; Conrad and Hull, 1964). A second effect, the so-called word length effect, indicates that span is higher for short words than for long ones (Baddeley et al., 1975, 1984). The usual way of testing auditory-verbal STM is through digit span. One has to immediately repeat orally, in the same order, a sequence of digits that have been auditorily presented. Gathercole and Baddeley (1989), however, considered that nonword repetition is a more sensitive measure than digit span in assessing the capacity of the phonological loop, because it lacks lexical support, forcing the listener to rely on phonology only. Digits have already a lexical mental representation and are not, therefore, new phonological forms; indeed, nonword repetition is very much poorer than digit span if scored in terms of phonemes/syllables correct. A question about auditory-verbal STM concerns what this system is for. Around 1988, the relationship between the phonological loop and vocabulary learning started to become a major focus of investigation (e.g., Baddeley et al., 1988; Gathercole & Baddeley, 1989; Gathercole et al., 1992, 1999; Gupta et al., 2003; Martin et al., 1996; Martin & Saffran, 1997; Montgomery, 2002; for review, see Baddeley et al., 1998; Gathercole, 2006). Two major findings that prompted this discussion were the discovery that patients with STM deficits were unable to learn new words (Baddeley et al.,1988), and that novel word repetition ability correlated with vocabulary size in normally developing children (Gathercole & Baddeley, 1989) and in children with specific language impairment (SLI; Gathercole & Baddeley, 1990a). Since then, an impressive amount of evidence has been produced supporting the existence of a relationship between vocabulary size and nonword repetition or immediate serial recall (e.g., Gathercole & Baddeley, 1990b; Gathercole et al., 1992, 1997, 1999; Gupta, 2003; Gupta et al., 2003; Michas & Henry, 1994; Papagno et al.,1991; Papagno & Vallar, 1992; Service, 1992; Service & Kohonen, 1995). However, these associations were differently interpreted. Baddeley et al. (1998) suggested that phonological STM plays a causal role in vocabulary learning, and this would establish the existence of a connection between STM and long-term learning. An alternative perspective was that the association between nonword repetition and phonological vocabulary learning may be mediated by the number of phonological representations available (i.e., a greater vocabulary size producing an increased number of phonological representations and therefore a better repetition), which improves phonological knowledge itself, and consequently improves nonword repetition, challenging the need of the PSTM (e.g., Bowey, 1996; Snowling, 2006; Snowling, Chiat, & Hulme, 1991). Gupta and Teasdale (2009), by means of a computational model, demonstrated that the causal links posited by the two theoretical perspectives are both valid, indicating that an opposition between them is unnecessary.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
Several studies continue to be conducted to investigate the role of PSTM in learning new words. The relevance of the phonological loop in learning new words has been demonstrated in different populations, as previously cited: children learning their mother tongue (L1), healthy people, both children and adults, learning a second language (L2), children with specific language impairment (SLI), and brain-damaged people. The first observations concerned dyslexic children, whose common feature seemed to be a deficit in learning the order in which the months of the year occur (Miles & Ellis, 1981); they also seem to be poor at learning foreign languages, even if they have no intelligence deficits. Accordingly, Kail and Leonard (1986) reported that children with specific reading difficulties tend to have lower vocabulary level than expected from children of similar intelligence. This deficit is not due to a difficulty in reading (and consequently explained with a deficit in acquiring vocabulary through reading) because it appears quite early, before reading is achieved. It is somehow intuitive to explain the relation between vocabulary and auditory-verbal STM: if someone learns a new word, which is in some way a nonword when we hear it for the first time, the maintenance of the new phonological representation is required until a permanent trace is built. The difficulty increases along with the distance of the new phonological form from already existing phonological representations. In other words, one can take advantage of the similarity of the new word with a preexisting lexical representation. Cross-linguistic transfer can be observed at multiple levels, including phonology (Melby-Lervåg & Lervåg, 2011), and similarities in sound combinations can facilitate vocabulary acquisition (Bartolotti & Marian, 2017): for example, if I have to learn the Spanish word “garganta” meaning throat, learning can be enhanced by the fact that it sounds similar to “gargle”, while if I have to learn the Czech word “zmrzlina,” I probably have no similar phonological forms available. Both suggestions – namely, the need for a STM buffer allowing to build stable phonological representations and the possibility of bypassing this bottleneck, through similarities of phonological forms – have been confirmed by a number of studies in different populations. Successful encoding of a novel word depends to some extent on our ability to rehearse its phonological form, particularly during early stages of acquisition, and it is easier to rehearse words that resemble those of languages that we already know (Hayakawa et al., 2020).
4.2
Healthy Children Learning Their Mother Tongue
Childhood is the period in which people learn the greatest number of new words, and several studies have provided evidence that the capacity of auditory-verbal STM constrains new word learning. Children of the same age show large differences in STM capacity. Indeed, Gathercole and Adams
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
53
54
CO STA N Z A PA PAG N O
(1993) found that some children between 2 and 3 years could achieve a digit span of four, while others could not. In a personal experience, investigating Italian children learning Chinese in a nursery, some of those aged 3 could achieve a digit span of 4 and others only of 2. Similar variations are displayed in knowledge of the native vocabulary in the same groups of children. Crucially, it has been found that individual differences in STM performance are related to vocabulary knowledge: children with a better performance on STM also show better vocabulary knowledge (Gathercole et al., 1991, 1992, 1997; Gathercole & Adams, 1993, 1994; Gathercole & Baddeley, 1989; Michas & Henry, 1994). Children between ages 4 and 8 were tested in nonword repetition and general cognitive abilities, and vocabulary and nonword repetition were highly correlated at ages 4, 5, and 6, while the correlation decreased after that age. More specifically, a crosslagged correlational analysis showed that nonword repetition at age 4 was found to be significantly associated with vocabulary test scores one year later, whereas the vocabulary measure at age 4 was not a significant predictor of nonword repetition scores at age 5. Moreover, this correlation was stronger in the nonword to vocabulary direction only at an early age of 4, while it was stronger in the opposite direction at later timepoints, when an increased vocabulary induced a better nonword repetition in the following year. Cognitive abilities did not affect nonword repetition (Gathercole et al., 1994). This same relation between vocabulary and nonword repetition tested for English exists in different languages as Hayashi and Takahashi (2020) demonstrated, for example, in 90 5-year old Japanese-speaking children. As Gathercole (2006) pointed out, “every word we know was once unfamiliar to us and became part of our mental lexicon through several repetition attempts.” Therefore, this correlation is not surprising. As mentioned before, the correlation between nonword repetition and vocabulary acquisition decreases when the size of vocabulary increases, and this is explained with the fact that we use similarities among languages (see the example in Section 4.1), exploiting already existing phonological representations. Therefore, the greater the size of the lexicon, the more effective this strategy will be. With further exposure to vocabulary, everyone will be able to achieve a normal knowledge of new words. Indeed, in a longitudinal study on 5-year-old children with poor PSTM scores, Gathercole et al. (2005) found that at 8 years they showed normal native vocabulary knowledge even if memory span was still impaired. Gathercole and Baddeley (1990b) tested the abilities to learn new names of toy animals in two groups of 5-year-old children, one with a high and the other with a low nonword repetition score. The names were either familiar such as Peter and Michael or phonologically unfamiliar such as Pyemass and Meeton (obtained from the same phonological pool as the familiar names). The children with the low nonword repetition scores performed significantly worse at learning the unfamiliar names than the high repetition children, while the two groups did not differ in the rates at which they learned the familiar ones.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
These results support the hypothesis that new-word learning is linked to phonological memory skills, even when environmental exposure to new words is controlled across subjects. However, low-repetition children also had a lower vocabulary than high-repetition children, with a possible consequent lower availability of phonological forms.
4.3
Children with Specific Language Impairment (SLI) or Down Syndrome
Specific language impairment (SLI) is diagnosed when a child fails to develop spoken language at a normal rate and this deficit cannot be attributed to neurological damage, hearing impairment, or intellectual disability nor the impairment is caused by some emotional disorder (Leonard, 2014). Recently, the term Developmental Language Disorder (DLD) was endorsed for use when the language disorder was not associated with a known biomedical etiology (Bishop et al., 2017). Since the reported studies were run before the introduction of such terminology, I will maintain the term SLI that is, however, currently still in use. Phonological short-term memory problems have been reported in children with SLI (e.g., Bishop et al., 1996; Gathercole & Baddeley, 1990a), and it is plausible to suggest that they play a role in the language impairment given their severity. Nonword repetition is considered to be a reliable test for SLI in preschool children (see for example Dispaldro et al., 2013) but also for school students (Girbau & Schwartz, 2007). The role of PSTM in children with SLI has been related to reduced fast mapping in comparison to typically developing children (TD). This assumption refers to the mapping theory (Chiat, 2001) that can be applied to the stages of word learning, which involve input processing of phonological and semantic information and mapping between the two. Fast mapping is the first stage in the processing of laying down lexical representations: the child hears the word, stores its phonological form, and begins to map semantic features to that form (Chiat, 2001). Then, in slow mapping, the child hears and refines the phonological and semantic representations of each word as it recurs in different contexts. If the phonological form is not robust enough it may not be effectively recognized when it appears in different contexts. In a study performed on 23 children with SLI and 26 TD matched children, a nonword repetition test, as a measure of PSTM, and the Peabody Picture Vocabulary Test (PPVT), as a measure of receptive vocabulary, were administered (Jackson et al., 2016). Consistently, a poor performance on nonword repetition tasks was found in these children (see also Archibald & Gathercole, 2006a, and a meta-analysis across 23 nonword repetition studies, Graf Estes et al., 2007). As PSTM is considered a key process in fast mapping (Baddeley, 2003), by means of a regression analysis Jackson et al.,
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
55
56
CO STA N Z A PA PAG N O
(2016) found that nonword repetition was a significant predictor of fast mapping production, and children with limited PSTM performance showed more difficulties in creating stable phonological representations, and, consequently, limited lexical learning. That nonword repetition is a clinical marker of SLI has been demonstrated in different languages. For example, Girbau (2016) examined 40 native Spanishspeaking children of an age between 8 and 10, by means of a battery of psycholinguistic tests, including nonword repetition. Twenty were children with SLI and 20 were TD children. Children with SLI scored significantly worse than TD children on nonword repetition, with a length effect. Although a PSTM deficit alone cannot explain SLI, it seems to be a core one. Indeed, as Archibald and Gathercole (2006b) suggested, very large impairments are found in particular for nonword repetition. Therefore, nonword repetition proved to be more sensitive to the SLI deficit than the more traditional serial recall. A possibility is that nonword repetition is also affected by output phonological processes, in terms of output planning mechanisms (Snowling et al., 1991). A similar pattern of language impairment is found in children with Down syndrome. In a meta-analysis including 14 papers, children with Down syndrome performed significantly poorer on expressive vocabulary than nonverbal mental age-matched controls as well as in measures of verbal STM (Næss et al., 2011). However, these studies suffer from some limitations. First, even if poor memory in children with SLI or Down syndrome is consistent with the hypothesis that this is the cause of the language impairment, the opposite is also likely; poor verbal memory could result from weak language. Comparing language-impaired children with younger control children matched on language level could solve this problem, in that if their memory is poor even in relation to language-matched younger controls, then PSTM problems are not likely to be a consequence of language low performance (Bishop, 1992). Unfortunately, the literature is controversial on this point, as SLI children were not found to have impaired STM in comparison to younger language-matched control children in some studies (see, e.g., van der Lely & Howard, 1993). Second, all these studies are correlational in nature. The observations made in normal and language-disordered children suggest an important role of PSTM in the acquisition of vocabulary. If this is the case, subjects with a relatively normal language acquisition should have a preserved function of PSTM, even in the presence of relevant deficits in other areas of cognition. Giuseppe Vallar and I came across one of such cases, a young woman with Down syndrome who was fluent in Italian, English, and French, thanks to the fact that she spent her childhood in different countries due to her father’s job. She was severely impaired in tasks of general intelligence, visuospatial perception, and long-term memory. The spatial span was just above the cut-off score. Her vocabulary, as shown by the vocabulary subtest of the Wechsler Adult Intelligence Scale (WAIS), in the phonemic and semantic controlled association tasks, were within the normal range. Also, her digit
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
span and recency effect were normal, indicating normal PSTM (Vallar & Papagno, 1993). The alternative hypothesis that her preserved PSTM did not play any relevant role in her excellent vocabulary, and language acquisition is very unlikely, since the patient was unimpaired in a paired-associate learning task, in which the second item of the pair was a nonword, while, paradoxically, she was impaired when both words of the pair were Italian, therefore known, ones. The nonword learning task, a simple experimental analogue of vocabulary acquisition, involves a relevant contribution from PSTM, as previously reported.
4.4
Healthy Children and Adults Learning a Second Language (L2)
The link between PSTM and vocabulary acquisition also extends to older populations: nonword repetition ability in adults is highly associated with the rate of learning novel phonological forms that do not closely resemble familiar words (Atkins & Baddeley, 1998; Gupta, 2003) as when learning a foreign language. This association has been studied in children and adults, and within this population, seems to be stronger in older than young adults (Service & Craik, 1993).
4.4.1 Children Learning a Second Language Service (1992) studied the acquisition of English as a second language by Finnish children. They were tested before the beginning of the course and it worked out that their capacity to repeat English-sounding nonwords was the best predictor of their future achievement (at three years) in learning English. In a subsequent study, Service and Kohonen (1995) ran several different analyses to find out how foreign vocabulary learning is related to pseudoword repetition accuracy. The correlation was strong even when general academic achievement was partialed out. An additional recent study (White, 2020) explored 27 children in South Africa in the process of acquiring English as a second language (ELLs = English language learners) between the ages of 5 and 6 years old. Children were considered to be ELLs if they were sequential bilinguals and if both parents had any language other than English as their L1. The participants were tested three times throughout the year. The correlation between vocabulary and English nonword repetition was stable and remained significant throughout the year. However, longitudinal data showed that PSTM did not influence the developmental trajectory, in line with Gathercole et al.’s (1992) results showing that the causal relationship between vocabulary and nonword repetition decreases after the age of 5, although the correlation remains. Accordingly, Masoura and Gathercole (2005) showed that the children’s speed in a paired-associate (Greek-English) learning task was affected by
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
57
58
CO STA N Z A PA PAG N O
their current L2 vocabulary but independent of PSTM (namely, nonword repetition). The participants were Greek children who had been studying English for 3 years. There was, however, a strong relationship between the children’s PSTM and their present knowledge of English vocabulary. This pattern is in line with the hypothesis that PSTM plays a role in supporting the construction of stable long-term representations of the phonological structures of new words (Baddeley et al., 1998) but is no more crucial when the person has acquired sufficient long-term stable representations and can take advantage of known lexical structures. Verhagen and Leseman (2016) investigated whether the same relationships apply to children learning an L2 in a naturalistic setting and to monolingual children with a mean age of 5. Participants were 63 Turkish children who learned Dutch as an L2 and 45 Dutch monolingual children. Children completed, among others, a series of auditory-verbal STM tasks, a Dutch vocabulary task and a Dutch grammar task. Structural equation modeling showed that A-V STM, treated as a latent factor, significantly predicted vocabulary and grammar. No differences were found between the two groups. All these studies refer to a traditional way of learning L2 at school. An interesting point is to investigate to what extent L2 vocabulary acquired through the particular school context of early L2 immersion is linked to the same cognitive abilities. A total of 61 French-speaking 5-yearold children just enrolled in English immersion classes were administered a battery of tasks assessing three phonological processing abilities, namely PSTM (a French complex nonword repetition task), phonological awareness (a vowel phoneme detection task), and speech perception (minimal pair discrimination task), before starting the L2 immersion. Their English vocabulary knowledge was measured after one, two, and three years of school. Among the tested phonological processing abilities, PSTM and speech perception appeared to underlie L2 vocabulary acquisition in such a context of an early L2 immersion school program, at least during the first steps of acquisition, while phonological awareness seemed to have no influence (Nicolay & Poncelet, 2013). Indeed, performance on the nonword repetition task significantly correlated with performance on L2 vocabulary tests at T1, T2, and T3, in line with previous studies and studies on L1 learning. After controlling for possible effects of initial differences in age, nonverbal intelligence, and L1 vocabulary, nonword repetition performance still emerged as a predictor of both L2 productive vocabulary after 1, 2, and 3 years of immersion schooling and L2 receptive vocabulary after 1 and 2 years. It is worth mentioning a study that obtained contradicting results with respect to previous reported literature (Kormos & Safar, 2008). In a study on 121 secondary school students aged 15–16 who took the Cambridge First Certificate (the most important Cambridge exams of English),1 a nonword repetition test was used. The research was conducted in a HungarianEnglish bilingual school, and the participants who took part in an intensive
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
language program started from beginner level. In beginners PSTM did not correlate with English proficiency, while it was highly significant in the preintermediate stage. This contradiction is explained by the authors with the fact that for higher-level learners the ability to store verbal material in working memory plays an important role in acquiring a wide repertoire of words and expressions. It is not clear, however, why this does not apply to beginners. It is more likely that the contradiction is because, apparently, they never correlated PSTM with a pure vocabulary acquisition measure but with different composite measures of English proficiency, such as reading, listening, and speaking in general.
4.4.2 Adults Learning a Second Language All the above reported studies assessed L2 learning in a young, still-developing, population. Therefore, one could hypothesize that the role of auditory-verbal STM is apparent only in this particular period of life. In fact, a relation between vocabulary and PSTM has been found also in adults. Using a different paradigm, namely a narrated story in a foreign language, 55 French native speakers aged 19, relatively advanced L2 (English) learners, were exposed to new words in the context of a narrated story (Hummel, 2020). In this task, target words were explicitly presented, and participants were asked to try to remember them. In line with the previous studies, the hypothesis was that individuals with a strong PSTM capacity should perform better than those with poor PSTM capacity. This hypothesis also considered the auditory format of the material to be learned. Indeed, research on L1 has demonstrated that learning complex new material through an auditory format is more difficult compared to the visual format that allows one to revisit the material. Information presented in auditory format produces the so-called transient information effect (Leahy and Sweller, 2011). In Hummel’s (2020) study, a short story was presented that included 20 target words, one for each sentence. The context of the sentences contributed to facilitate the comprehension of the new words. PSTM capacity was assessed by means of a nonword repetition task. The performance on the PSTM tasks did not predict the performance on vocabulary acquisition if all participants were considered together. However, when the group was split in those with low PSTM and those with high PSTM, a significant association between nonword repetition and vocabulary acquisition was found for the low PSTM subgroup. In terms of vocabulary acquisition, no analyses were performed; however, the mean number of acquired words appears to be approximately the same for the two subgroups. A suggested explanation is that, as learning progresses, familiarity with the language system plays a larger role in performance and outweighs PSTM constraints, in line with L1 studies (see above), showing that the relation is stronger in 4-year-old children and decreases at 5, when the child already knows more words. Further L2 studies show that the role of PSTM decreases with more advanced learning stages (e.g., Serafini & Sanz, 2016).
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
59
60
CO STA N Z A PA PAG N O
The fact that the link between vocabulary scores and PSTM in Hummel’s study was evident only in the lower PSTM subgroup suggests that this capacity is critical for short-term vocabulary retention only when quite low, at least in a population of relatively advanced L2 learners. Also in a different study on bilingual (Catalan-Spanish) English learners, participants were divided in low PSTM and high PSTM, but in this case the relation between PSTM, as measured by means of a nonword recognition task, and accuracy scores were found in both populations, with better scores and greater perceptual accuracy (as measured through an identification and a discrimination task of vowels) in the high PSTM group (AliagaGarcia et al. 2010). Speciale et al. (2004) performed two experiments in order to verify whether individual cognitive differences affect the acquisition of L2 words. In the first experiment they assessed the ability to learn phonological sequences in a group of 40 undergraduate students. The stimuli were 12 target nonwords and 96 foils. Target items were presented eight times during the task and foils only once. On each trial a nonword was presented. The first 24 trials comprised 12 target items and 12 foil items, in a fixed quasirandom order. The following 168 trials were presented in random order: on half of these trials, one of the 12 target items was presented and on the other half, a foil item was presented. Therefore, each target was showed on eight trials, while foil items once. Participants indicated whether they had heard each stimulus previously. A formula derived from signal detection theory was used to measure the degree to which participants were able to identify target words. Speciale et al. (2004) also tested phonological short-term store capacity by means of nonword repetition, and learning of novel foreign (German) vocabulary. Phonological sequence learning predicted receptive vocabulary learning, while, together with phonological store capacity, they made independent additive contributions to productive vocabulary learning. In a second experiment, Speciale et al. performed a longitudinal study in a group of 44 undergraduate students learning Spanish as L2 during a 10week course. The initial ability to learn phonological sequences predicted the final level achieved in Spanish receptive language and their ability to repeat Spanish-sounding nonwords. Speciale et al.’s conclusion was that when beginning to learn a language, phonological store and sequence learning ability are separable. However, as exposure to the language increases, also does the degree to which a learner begins to recognize repeated phonological sequences (therefore learned) and to abstract their regularities. This determines how well long-term knowledge is reached. The greater the learning rate, the greater the resultant receptive vocabulary. Phonological sequence learning, in other words, enhances the acquisition of long-term phonological representations. Additional evidence comes from O’Brien et al. (2007), who found that PSTM correlated with vocabulary scores, both at the beginning and at the
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
end of a semester abroad of Spanish learning in a group of English adults. A correlation between PSTM and final vocabulary was found also using an artificial language (Martin & Ellis, 2012): nonword repetition correlated, and the correlation was moderately strong, with vocabulary production and comprehension. After the influences of age and foreign language study were accounted for, PSTM accounted for 14 percent of the variance in final vocabulary learning. Finally, a recent meta-analysis (Linck et al., 2014) indicated a robust, positive correlation between working memory and L2 outcomes. More specifically, the executive control component was more strongly related to L2 outcomes than is the storage component, which in any case showed still significantly positive effect sizes.
4.4.3 Disrupting the Phonological Loop Another source of evidence in adults comes from disruption of the phonological loop. This is the most powerful evidence of the PSTM role. Indeed, if the phonological loop is important for acquiring new vocabulary, its disruption would prevent this acquisition. A typical way of disrupting the phonological loop is through articulatory suppression. This consists in repeating an irrelevant utterance, such as a syllable, while the verbal material to be remembered is presented. Articulatory suppression interferes with the process of rehearsal that actively maintains the phonologically encoded material. Papagno et al. (1991) studied the effect of articulatory suppression in normal participants on the acquisition rate of pairs of familiar words (such as horse-book) and pairs of familiar words coupled with foreign ones, namely Russian words (such as rose-svieti). The participants in the initial experiments were native Italian speakers. Articulatory suppression was contrasted with tapping with the first producing a detrimental effect on the acquisition of Russian words but not on Italian native language. This decrease in vocabulary acquisition was evident with both the visual and the auditory presentation of stimuli. This result is in line with studies on children reported in section 4.2, in that PSTM constrains the acquisition of unfamiliar words but not learning of familiar names (Gathercole et al., 1997; Gathercole & Baddeley, 1990a). Once again, it appears that, when possible, people use existing language representations to mediate verbal learning, and only when phonological forms are new and therefore there is no preexisting representation to support learning, one can rely only on the phonological loop system. Further evidence of this possibility to use preexisting phonological representations to avoid the limitations due to the limited capacity of A-V STM came from the attempt to replicate the previous experiment with English participants. In fact, they found easy to form semantic associations to the selected Russian words. For example, a pair of words was “dictionary-slavar”. Participants reported that they imagined the utterance “slave to the
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
61
62
CO STA N Z A PA PAG N O
dictionary.” However, once the association values of the material were reduced by using nonsense CVC-CVC material and subsequently by using the more unfamiliar phonological structures provided by Finnish vocabulary, the results obtained with Italian participants were replicated (Papagno et al., 1991). As reported in the introduction of this chapter, our span is limited by phonological similarity and stimuli length. In contrast, long-term learning is not affected by these two variables but by semantic similarity (Baddeley, 1966a). However, if the unfamiliar new vocabulary items depend on the phonological loop for their initial acquisition, then one would expect to observe the same effects found for auditory-verbal STM. To investigate this aspect, which would be an indirect confirmation that the PSTM is involved in long-term acquisition of novel words, Papagno and Vallar (1992) manipulated the degree of phonological similarity among the items to be learned, predicting that learning known words with a paired associate paradigm would not be affected by phonological similarity, being the principal mode of coding lexical-semantic. On the contrary, paired associates with unfamiliar vocabulary involving the phonological loop, would be affected by phonological similarity and word length. The results proved the prediction to be true. In conclusion, while phonological similarity and word length had no influence on the participants’ acquisition of pairs of items in their native language, they had on the acquisition of unfamiliar Russian vocabulary.
4.4.4 Polyglots A final support to the role of PSTM in learning new words comes from a study on polyglots (Papagno & Vallar, 1995). When polyglot (people fluently speaking at least three different languages) and nonpolyglot (people who had studied only one language at school) university students were compared on a range of memory and long-term learning tasks, the two groups performed at the same level on tests of nonverbal ability, including visuospatial STM span, and visuospatial learning, and on general intellectual tests. However, the polyglots performed significantly better on auditory-verbal digit span and nonword repetition. In addition, the performance on the two PSTM measures significantly correlated with participants’ abilities to learn word-nonword pairs but were independent of word-word paired-associated learning. In sum, PSTM is crucial for the development of L2 vocabulary. Hence, a deficit in auditory-verbal STM should prevent learning of new words or words in a foreign language. The deficit would be extremely severe if the items to be learned were beyond the subject’s memory span. In that case, the subject would only be able to retain part of the stimulus item, showing a difficulty in integrating the different parts of the stimulus into a new phonological unit. However, even with an item within the memory span, a subject might need additional storage to build permanent phonological representations.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
4.5
Cognitive Deficits
The accepted view concerning PSTM is that it does not affect long-term learning based on semantic coding, namely the acquisition of meaningful material, such as a list of words or a story or any type of event. Indeed, in traditional tests of long-term memory (LTM) participants are usually required to learn arbitrary sequences of familiar words, not phonologically novel material as in the experiments reported so far. Hence, patients with PSTM deficits are not impaired in learning a short story, in a delayed recall of lists of words, and in learning paired associates in which both words are in a known meaningful language. However, they can repeat no more than 2–3 digits or words, in the same order of presentation in a typical span task. Also, they do not show the phonological (with visual presentation) and word length effect with visual and auditory presentation. Moreover, a very famous patient in the STM literature, PV (Basso et al., 1982), reported that she had hardly tried to learn French as a second language and, despite her unimpaired cognitive performance except in tests of auditory-verbal STM, her attempt failed. Her digit span was of 2–3 and she could not repeat more than two words. Baddeley et al. (1988) tested, therefore, her capacity to learn the vocabulary of Russian, which was an unfamiliar language. Stimuli were the same as those previously reported studying the effect of articulatory suppression on novel word learning. Her performance was compared with that of a group of 14 matched control participants in two tasks, one consisting in pairs of unrelated Italian words (the patient was an Italian native speaker), and one consisting in Italian-Russian equivalents. Both types of presentation were used, auditory and visual. Because the patient was unable to repeat back single quadrisyllabic Russian words, our stimuli were bi- or three-syllabic words. The patient proved to be absolutely comparable to controls in learning pairs of Italian (therefore known) words, while she could not learn any single words after eight attempts in the auditory modality, while the control participants had learned the whole list. With visual presentation her performance was slightly better, but it was still significantly worse than that of the control participants. In sum, her short-term phonological deficit was associated with a specific impairment in phonological long-term learning of unfamiliar material. A similar pattern of PSTM impairment and long-term learning deficit has been reported by Trojano et al. (1992) in a patient with a rarely reported linguistic syndrome: he could repeat words but not nonwords, while he could read both words and nonwords flawlessly. In other words, his difficulties were limited to auditory phonological coding: indeed, he had critical difficulties in discriminating minimal CV pairs, and in identifying single consonants with a binary choice procedure. This patient showed severe difficulties in learning lists of bisyllabic function words, while he was able to learn lists of bisyllabic concrete words. In a subsequent study,
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
63
64
CO STA N Z A PA PAG N O
Trojano and Grossi (1995) assessed the effects of phonological similarity and word length and nonword learning, using the same lists as in Papagno and Vallar (1992). The patient showed a normal learning of known words while he was specifically unable to learn nonwords regardless of presentation modality. Therefore, this is a case of secondary PTM deficit, since the final outcome of his defect of phonological processing was a reduced capacity of PSTM. We have discussed the pattern of learning in two patients with an impaired PSTM that was due to a selective impairment of the phonological input buffer in one case (Baddeley et al., 1988), and secondary to an auditory phonological coding defect in the second one (Trojano & Grossi, 1995). The same pattern was observed in a patient showing a deficit in the rehearsal component of the phonological loop, who showed, indeed, an impairment in learning Russian but not Italian words (Papagno et al., 2008). This patient had a normal recency effect - which represents the output of a phonological short-term buffer with no or minimal contribution of rehearsal - and an impairment on phonological judgments, an opposite pattern with respect to patients with a phonological input buffer deficit. The same stimuli used with the two previous patients were presented in auditory mode at a rate of 2 s per pair, and the patient was instructed that, following the presentation of all the word pairs, she would be asked to produce the corresponding Russian word for each stimulus, which consisted in an Italian word. The patient did not produce any response in the first, second, and ninth trial, while she recalled one single item for each of the remaining trials. When both words of the pair were Italian words, she was able to learn all stimuli after four trials. This pattern exactly reproduces what has been observed in the previous two patients and in healthy participants during articulatory suppression. All these experiments were carried out using the same material and with Italian patients. Thus, one could suggest that it is a deficit typical of a specific language. A confirmation that this is not the case comes from one English and three German patients. Friedmann and Martin (2001) assessed patients’ ability to learn Spanish word–English word translation pairs. They predicted that patients showing a phonological STM deficit would perform poorly on this task. One patient with a severe deficit in phonological STM but a better-preserved ability to retain semantic information showed better learning of new semantic information than new phonological information; indeed, she never averaged much more than one correct response per trial on the foreign word pairs. Her phonological STM deficit greatly reduced her ability to learn the novel phonology contained in the foreign words. Her high performance on phoneme discrimination task with no delay rules out a perceptual deficit as an explanation for her pattern of performance. Three other patients had a deficit in semantic short-term memory and showed lower retention of new semantic information, while the last patient was impaired in both.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
A German patient with a severe deficit of auditory-verbal STM showed, as Italian patients had, learning abilities within the range of controls when tested with real words, while in a test with one- and two-syllables pseudowords his performance was poor (Dittman & Abel, 2010). In a further experiment, Bormann et al. (2015) assessed learning of new words and of digit sequences in two patients with auditory-verbal STM impairment. They had to learn family names (that were in principle unfamiliar to them), phone numbers, ages, and professions. Therefore, compared to previous studies, this design was less artificial, mimicking an everyday situation such as meeting new people. The two STM patients were severely impaired in learning the family names of people, which corresponds to nonword learning, and in learning a sequence of digits. It is, however, to underline that family names could be also familiar, since a family name could have been already heard or known. Evidence that a long-term phonological learning deficit arises from impairments in the phonological loop has also been provided in a study of an individual who appears to have a developmental impairment of PSTM (Baddeley, 1993). A highly intelligent graduate student could not repeat sequences of more than four digits and performed very poorly on a task involving the immediate repetition of multisyllabic nonwords. When compared with a group of six fellow students on a wide range of short-term phonological memory tests, his performance was always poorer. In contrast, he performed very well on visual short- and long-term memory tasks. When this student was tested by using the same paradigm reported above, in which he had to learn pairs of meaningful English words and EnglishFinnish foreign language vocabulary, he showed normal learning of meaningful (English) paired associates, while his capacity to learn Finnish vocabulary was severely impaired with respect to control participants.
4.6
Conclusion
All the evidence provided so far is a clear demonstration of the role that the phonological loop has in vocabulary acquisition. The initial observation made on an Italian patient in 1988 (Baddeley et al., 1988) has been confirmed in other patients speaking different languages, such as German and English. Moreover, its role in acquiring vocabulary in children has been demonstrated, even though it is limited to the initial stages of learning. When sufficient phonological forms are available, the limitation imposed by the low capacity of STM is bypassed by sound similarities. It is a common observation, indeed, that language learners seek for similarities between the foreign and native language in order to make use of existing knowledge (Ringbom, 2007, cited in Hayakawa et al., 2020), namely, existing phonological representation to avoid the bottleneck imposed by PSTM.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
65
66
CO STA N Z A PA PAG N O
Before concluding this chapter, we need to mention that the contribution of the phonological loop in vocabulary acquisition has been discussed, but this is far from its only relevant role in language. Other research has explored the role of PSTM in acquiring L2 grammar, showing that PSTM can predict grammar achievement in L2 learning (French & O’Brien, 2008; Martin & Ellis, 2012; Verhagen & Leseman, 2016; White, 2020) and its role in syntactically complex sentences that load on memory, even though still debated, has been demonstrated in brain-damaged patients (for a review see Papagno & Cecchetto, 2019). However, these topics require an extensive discussion which is out of the scope of the present chapter.
References Aliaga-Garcìa, C., Mora, J. C., & Cervino-Provedano, E. (2010). Phonological short-term memory and L2 speech learning in adulthood. In M. Wrembel, M. Kul, & K. Dziubalska-Kołaczyk (Eds.), New sounds 2010: Proceedings of the 6th International Symposium on the Acquisition of Second Language Speech (pp. 19–24). Peter Lang. Archibald, L. M., & Gathercole, S. E. (2006a). Nonword repetition in Specific Language Impairment: More than a phonological short-term memory deficit. Psychonomic Bulletin & Review, 14, 919–924. Archibald, L. M., & Gathercole, S. E. (2006b). Short-term and working memory in specific language impairment. International Journal of Language & Communication Disorders, 41, 675–693. Atkins, P. W. B., & Baddeley, A. D. (1998). Working memory and distributed vocabulary learning. Applied Psycholinguistics, 19, 537–552. Baddeley, A. D. (1966a). Short-term memory for word sequences as a function of acoustic, semantic and formal similarity. Quarterly Journal of Experimental Psychology, 18, 362–365. Baddeley, A. D. (1966b). The influence of acoustic and semantic similarity on long-term memory for word sequences. Quarterly Journal of Experimental Psychology, 18, 302–309. Baddeley, A. D. (1993). Short-term phonological memory and longterm learning: A single case study. European Journal of Cognitive Psychology, 5, 129–148. Baddeley, A. D. (2003). Working memory: Looking back and looking forward. Nature, Reviews Neuroscience, 4, 829–839. Baddeley, A. D. (2012). Working memory: Theories, models and controversies. Annual Review of Psychology, 63, 1–30. Baddeley, A. D., Gathercole, S. E., & Papagno C. (1998) The phonological loop as a language learning device. Psychogical Review, 105, 158–173. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. A. Bower (Ed.), Recent advances in learning and motivation (Vol. 8, pp. 47–90). Academic Press.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
Baddeley, A. D., & Hitch, G. (2019). The phonological loop as a buffer store: An update. Cortex, 112, 91–106. Baddeley, A. D., Lewis, V., & Vallar, G. (1984) Exploring the articulatory loop. Quarterly Journal of Experimental Psychology, 36, 233–252. Baddeley, A. D., Papagno, C., & Vallar, G. (1988) When long-term learning depends on short-term storage. Journal of Memory and Language, 27, 586–595. Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of short-term memory. Journal of Verbal Learning and Verbal Behavior, 14, 575–589. Bartolotti, J., & Marian, V. (2017) Orthographic knowledge and lexical form influence vocabulary learning, Applied Psycholinguistics, 38, 427–456. Basso, A., Spinnleer, H. R., Vallar, G., & Zanobio, E. (1982) Left hemisphere damage and selective impairment of auditory-verbal short-term memory. Neuropsychologia, 20, 263–274. Bishop, D. V. M. (1992). The underlying nature of specific language impairment. Journal of Child Psychology and Child Psychiatry, 33, 1–64. Bishop, D. V. M., North, T., & Donlan, C. (1996). Nonword repetition as a behavioural marker for inherited language impairment: Evidence from a twin study. Journal of Child Psychology and Psychiatry, 37, 391–403. Bishop, D. M. V., Snowling, M. G., Thompson, P. A., Greenhalgh, T., & the Catalise-2 Consortium. (2017) Phase 2 of CATALISE: A multinational and multidisciplinary Delphi consensus study of problems with language development: Terminology. Journal of Child Psychology and Psychiatry, 58, 1068–1080. Bormann, T., Seyboth, M., Umarova, R., Weiller, C. (2015) “I know your name, but not your number”: Patients with short-term memory deficits are impaired in learning sequences of digits. Neuropsychologia, 72, 80–86. Bowey, J. A. (1996). On the association between phonological memory and receptive vocabulary in five-year-olds. Journal of Experimental Child Psychology, 63, 44–78. Chiat, S. (2001) Mapping theories of developmental language impairment: premises, predictions and evidence. Language and Cognitive Processes, 16, 113–142. Conrad, R. (1964). Acoustic confusions in immediate memory. British Journal of Psychology, 55(1), 75–84. Conrad, R., & Hull, A. J. (1964). Information, acoustic confusion and memory span. British Journal of Psychology, 55, 429–432. Dispaldro, M., Leonard, L. B., & Deevy, P. (2013) Real-word and nonword repetition in Italian-speaking children with specific language impairment: A study of diagnostic accuracy. Journal of Speech, Language, and Hearing Research, 56, 323–336. Dittmann, J., & Abel, St. (2010) Verbales Arbeitsgedachtnis und verbales Lernen: Wort- und Pseudowortlernen in einem Fall von pathologischer Arbeitsgedachtnisbeeintrachtigung. Sprache-Stimme-Gehor, 34, e1–e9.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
67
68
CO STA N Z A PA PAG N O
Freedman, M. I., & Martin, R. C. (2001) Dissociable components of shortterm memory and their relation to long-term learning. Cognitive Neuropsychology, 18 (3), 193–226. French, L. M., O’Brien, I (2008). Phonological memory and children’s second language grammar learning. Applied Psycholinguistics, 29 (3), 463–487. Gathercole, S. E. (2006) Nonword repetition and word learning: The nature of the relationship. Applied Psycholinguistics, 27, 513–543. Gathercole, S. E., & Adams, A. (1993). Phonological working memory in very young children. Developmental Psychology, 29, 770–778. Gathercole, S. E., & Adams, A. (1994). Children’s phonological working memory: Contributions of long-term knowledge and rehearsal. Journal of Memory and Language, 33, 672–688. Gathercole, S. E., & Baddeley A. D. (1989) Evaluation of the role of phonological STM in the development of vocabulary in children: A longitudinal study. Journal of Memory and Language, 28, 200–213. Gathercole, S., & Baddeley, A. (1990a). Phonological memory deficits in language disordered children: Is there a causal connection? Journal of Memory and Language, 29, 336–360. Gathercole, S. E., & Baddeley, A. D. (1990b) The role of phonological memory in vocabulary acquisition: A study of young children learning arbitrary names of toys. British Journal of Psychology, 81, 439–454. Gathercole, S. E., Frankish, C., Pickering, S. J., & Peaker, S. (1999). Phonotactic influences on short-term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25, 84–95. Gathercole S. E., Hitch G. J., Service, E.,& Martin, A. J. (1997). Short-term memory and new word learning in children. Developmental Psychology, 33, 966–979. Gathercole, S. E., Service, E., Hitch, G. J., Adams, A.-M., & Martin, A. J. (1999). Phonological short-term memory and vocabulary development: Further evidence on the nature of the relationship. Applied Cognitive Psychology, 13, 65–77. Gathercole, S. E., Tiffany, C., Briscoe, J., Thorn, A., & The ALSPAC Team. (2005). Developmental consequences of poor phonological short-term memory function in childhood: A longitudinal study. Journal of Child Psychology and Psychiatry, 46, 598–611. Gathercole, S. E., Willis, C., Emslie, H., & Baddeley, A. D. (1992). Phonological memory and vocabulary development during the early school years: A longitudinal study. Developmental Psychology, 28, 887–898. Girbau, D. (2016). The nonword repetition task as a clinical marker of specific language impairment in Spanish-speaking children. First Language, 36, 30–49. Girbau, D., & Schwartz, R. G. (2007). Non-word repetition in Spanishspeaking children with Specific Language Impairment (SLI). International Journal of Language & Communication Disorders, 42(1), 59–75.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
Graf Estes, K., Evans, J. L., & Else-Quest, N. M., (2007) Differences in the nonword repetition performance of children with and without specific language impairment: A meta-analysis. Journal of Speech, Language and Hearing Research, 50, 177–195. Gupta, P. (2003). Examining the relationship between word learning, nonword repetition, and immediate serial recall in adults. Quarterly Journal of Experimental Psychology, 56A, 1213–1236. Gupta, P., MacWhinney, B., Feldman, H. M., & Sacco, K. (2003). Phonological memory and vocabulary learning in children with focal lesions. Brain and Language, 87, 241–252. Gupta P., & Tisdale, J. (2009). Does phonological short-term memory causally determine vocabulary learning? Toward a computational resolution of the debate. Journal of Memory and Language, 61, 481–502. Hayashi, K., & Takahashi, N. (2020). The relationship between phonological short-term memory and vocabulary acquisition in Japanese young children. Open Journal of Modern Linguistics, 10, 132–160. Hayakawa, S., Bartolotti, J., & Marian, V. (2020). Native language similarity during foreign language learning: Effects of cognitive strategies and affective states. Applied Linguistics, 1–28. Hummel, K. M. (2020). Phonological memory and L2 vocabulary learning in a narrated story task. Journal of Psycholinguistic Research. Jackson, E., Leitao, S., & Claessen, M. (2016) The relationship between phonological short-term memory, receptive vocabulary, and fast mapping in children with specific language impairment. International Journal of Language Communication Disorders, 51, 61–73. Kail, R., & Leonard, L. B. (1986). Word-finding abilities in language-impaired children. (ASHA Monographs No. 25). American Speech-Language-Hearing Association. Kormos, J., & Sáfár, A. (2008). Phonological short-term memory, working memory and foreign language performance in intensive language learning. Bilingualism: Language and Cognition, 11, 261–271. Leahy, W., & Sweller, J. (2011). Cognitive load theory, modality of presentation and the transient information effect. Applied Cognitive Psychology, 25, 943–951. Leonard, L. B. (2014) Children with specific language impairment (p. 480). MIT Press. Linck, J. A., Osthus, P., Koeth, J. T., & Bunting, M. F. (2014). Working memory and second language comprehension and production: A metaanalysis. Psychonomic Bulletin Revue, 21, 861–883. Martin, K. I., & Ellis, N. C. (2012). The roles of phonological short-term memory and working memory in L2 grammar and vocabulary learning. Studies in Second Language Acquisition, 34, 379–413. Martin, N., & Saffran, E. (1997). Language and auditory-verbal short-term memory impairments: Evidence for common underlying processes. Cognitive Neuropsychology, 14, 641–682.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
69
70
CO STA N Z A PA PAG N O
Martin, N., Saffran, E. K., & Dell, G. (1996). Recovery in deep dysphasia: Evidence for a relation between auditory–verbal STM capacity and lexical errors in repetition. Brain and Language, 52, 83–113. Masoura, E. V., & Gathercole, S. E. (2005). Contrasting contributions of phonological short-term memory and long-term knowledge to vocabulary learning in a foreign language, Memory, 13:3–4, 422–429. Melby-Lervåg, M., & Lervåg, A. (2011). Cross-linguistic transfer of oral language decoding, phonological awareness and reading comprehension: A meta-analysis of the correlational evidence. Journal of Research in Reading, 34, 114–135. Michas, I. C., & Henry, L. A. (1994). The link between phonological memory and vocabulary acquisition. British Journal of Developmental Psychology, 12, 147–164. Miles, T. R., & Ellis, N. C. (1981). A lexical encoding difficulty II: Clinical observations. In G. Th. Pavlidis and T. R. Miles (Eds.), Dyslexia research and its applications to education. Wiley. Montgomery, J. (2000). Verbal working memory in sentence comprehension in children with specific language impairment. Journal of Speech, Language, and Hearing Research, 43, 293–308. Næss, K-A. B., Halaas Lyster S.-A., Hulme, C., & Melby-Lervag, M. (2011). Language and verbal short-term memory skills in children with Down syndrome: A meta-analytic review. Research in Developmental Disabilities, 32, 2225–2234. Nicolay, A.-C., & Poncelet, M. (2013). Cognitive abilities underlying secondlanguage vocabulary acquisition in an early second-language immersion education context: A longitudinal study. Journal of Experimental Child Psychology, 115(4), 655–671. O’Brien, I., Segalowitz, N., Collentine, J., & Freed, B. (2006). Phonological memory and lexical narrative, and grammatical skills in second language oral production. Applied Psycholinguistics, 27(3), 377–402. Papagno, C., & Cecchetto, C. (2019). Is STM involved in sentence comprehension? Cortex, 112, 80–90. Papagno, C., Cecchetto, C., Reati, F., & Bello, L. (2007). Processing of syntactically complex sentences relieson verbal short-term memory: Evidence from a STM patient. Cognitive Neuropsychology, 24(3), 292–311. Papagno, C., Lucchelli, F., & Vallar, G. (2008) Phonological recoding, visual short-term store and the effect of unattended speech. Cortex, 44, 312–324. Papagno, C., Valentine, T., & Baddeley, A. D. (1991). Phonological short-term memory and foreign-language learning. Journal of Memory and Language, 30, 331–347. Papagno, C., & Vallar, G. (1992). Phonological short-term memory and the learning of novel words: The effects of phonological similarity and item length. Quarterly Journal of Experimental Psychology, 44A, 47–67.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
The Phonological Loop
Papagno, C., & Vallar, G. (1995). Verbal short-term memory and vocabulary learning in polyglots. The Quarterly Journal of Experimental Psychology, 48, 98–107. Ringbom, H. (2007). Actual, perceived and assumed cross-linguistic similarities in foreign language learning. AFinLan Vuosikiria, 65, 183–196. Serafini, E., & Sanz, C. (2016). Evidence for the decreasing impact of cognitive ability on second language development as proficiency increases. Studies in Second Language Acquisition, 38, 607–646. Service, E. (1992). Phonology, working memory, and foreign-language learning. Quarterly Journal of Experimental Psychology, 45A, 21–50. Service, E., & Craik, F. I. M. (1993). Differences between young and older adults in learning a foreign vocabulary. Journal of Memory and Language, 32, 608–623. Service, E. K., & Kohonen, V. (1995). Is the relation between phonological memory and foreign language learning accounted for by vocabulary acquisition? Applied Psycholinguistics, 16, 155–172. Shallice, T., & Papagno, C. (2019). Impairments of auditory-verbal shortterm memory: Do selective deficits of the input phonological buffer exist? Cortex, 112, 107–121. Shallice, T., & Warrington, E. K. (1970). Independent functioning of verbal memory stores: A neuropsychological study. The Quarterly Journal of Experimental Psychology, 22(2), 261–273. Snowling, M. J. (2006). Nonword repetition and language learning disorders: A developmental contingency framework. Applied Psycholinguistics, 27, 588–591. Snowling, M., Chiat, S., & Hulme, C. (1991). Words, nonwords and phonological processes: Some comments on Gathercole, Willis, Emslie, & Baddeley. Applied Psycholinguistics, 12, 369–373. Speciale, G., Ellis, N. C., & Bywater, T. (2004). Phonological sequence learning and short-term store capacity determine second language vocabulary acquisition. Applied Psycholinguistics, 25, 293–321. Trojano, L., & Grossi, D. (1995). Phonological and lexical coding in verbal short-term memory and learning. Brain and Language, 51, 336–354. Trojano, L., Stanzione, M., & Grossi, L. (1992) Short-term memory and verbal learning with auditory phonological coding defect: A neuropsychological case study. Brain and Cognition, 18, 12–23. van der Lely, H. K. J., & Howard, D. (1993). Children with specific language impairment: Linguistic impairment or short-term memory deficit? Journal of Speech and Hearing Research, 36, 1193–1207. Verhagen, J., & Leseman, P. P. M. (2016). How do verbal short-term memory and working memory relate to the acquisition of vocabulary and grammar? A comparison between first and second language learners. Journal of Experimental Child Psychology, 141, 65–82.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
71
72
CO STA N Z A PA PAG N O
White, M. J. (2020). Phonological working memory and non-verbal complex working memory as predictors of future English outcomes in young ELLs. International Journal of Bilingualism.
Note 1 The written paper consists of reading and listening comprehension, composition, and Use of English test. The reading and listening comprehension sections contain three texts each, with multiple-choice items and questions requiring short answers. In the composition task students write in three different genres, which are evaluated on their content and accuracy. The oral part of the exam includes an interview, a picture description, task and a problem-solving task in pairs.
https://doi.org/10.1017/9781108955638.007 Published online by Cambridge University Press
5 The Embedded-Processes Model and Language Use Eryn J. Adams, Alicia Forsberg, and Nelson Cowan
5.1
Introduction
Perhaps one of the most complex and unique cognitive skills that humans possess, language has been studied for many centuries, guided by a desire to understand what allows us to communicate through unique sounds and symbols (Forbes, 1933). In the most recent decades, research in this field has been related to the study of what mechanism allows for the acquisition, maintenance, and everyday use of language skills. One of the mechanisms or resources that has emerged as a significant contribution to these processes is working memory. In this chapter, we explore the involvement of working memory in language use from one specific perspective: the embedded-processes model of working memory. We begin with a brief history and purpose of the model itself, followed by a breakdown of the different primary parts of the model. Then, we address some primary language processes and explain how the various parts of the model work together to achieve successful language processing. Finally, we discuss how working memory is useful for the acquisition of language, be it for children as they learn their own native language(s) or later learners, who seek to acquire an additional language. We hope to highlight the importance of an attention-based view of working memory for understanding language. There are limits in the scope of our chapter. The embedded-processes model is based on a foundation of activated long-term memory (including rapid new learning) and, clearly, past knowledge plays a crucial role in the ability to use working memory successfully. This certainly includes linguistic knowledge and language use during working memory tasks. However, we do not go into detail on the ways that linguistic knowledge can assist working memory, leaving that topic to others and concentrating instead on how working memory can assist language use.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
74
ERYN J. ADAMS ET AL.
5.2
The Embedded-Processes Model of Working Memory
Scientists have attempted to model the human working memory system for decades, with attempts as early as James (1890), who proposed primary memory, made fully of attention resources. Other theories, such as that of Atkinson and Shiffrin (1968), divided memory into slightly more distinct components, separating sensory, short-term, and long-term memory and highlighting mechanisms such as rehearsal and attention as being ways to move information between these stores. Baddeley and Hitch (1974) took this separation a step further by dividing the short-term memory part of the model into two more distinct stores, specialized for the type of material processed: the visuospatial store for nonverbal, visually based stimuli and the phonological loop for language (even if written) and other auditory stimuli. The episodic buffer would be added in a later iteration to account for information that did not fall distinctly into either store or contained both verbal and visual elements (Baddeley, 2000). There is also a proposed central executive in this model that performs actions in which information is moved between stores, such as moving information out of short-term memory and into long-term memory. This mechanism can manipulate information being held within the two primary stores and thus provides means for controlling attention (cf. Adams et al., 2018 for a review of various models and their relation to language; for recent attempts to deconstruct the central executive see Logie, 2016; Vandierendonck, 2016). The embedded-processes model of working memory, first proposed in Cowan (1988) and later given the name in Cowan (1999), is illustrated in Figure 5.1. It would endorse elements of each of these theories of working memory structure, with some modifications. Cowan proposed that working memory is not a single entity that can be fully extracted from other cognitive processes, but rather a collection of mechanisms that allow information to stay in an activated state. Thus, in the embedded-processes model, working memory is embedded within the long-term memory system. Part of longterm memory can be activated to a level higher than the rest of the long-term store (referred to as activated long-term memory). Within this activated portion lies the focus of attention, the most limited and most activated part of memory. This model also contains a central executive system similar to the one proposed by Baddeley, which allows for cognitive actions to be taken on information within the memory system. The model includes a set of principles but should not be taken as a precise mathematical model that yields very specific predictions; it provides a conceptual framework for any such models (e.g., Cowan et al., 2012). In the sections that follow, we visit each area of the embedded-processes model and explain them in more detail. We follow with an explanation of how information presumably moves throughout each section, to make the coherent memory system envisioned in the model.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
Figure 5.1 The embedded-processes model of working memory
5.2.1 Long-Term Memory We begin with the most inclusive repository of information in the model, long-term memory. Long-term memory is said comprise all memories that have been encoded substantially enough that they no longer need to be attended to in order to permit later retrieval. Given this characteristic, long-term memories can range from as old as years to as new as a couple minutes or less. Long-term memories are also distinct because of their relative permanence. Although they can become more difficult to retrieve as time goes on, long-term memories can be limitless in both duration and capacity. In the embedded-processes model, all aspects of memory are embedded within long-term memory instead of separated from it. This is an important feature because long-term memory can enhance or support information being held within working memory (e.g., Endress & Potter, 2014; Lewis-Peacock & Postle, 2008; Miller, 1956). Likewise, working memory is necessary for the formation of new long-term memories, particularly those that are declarative (personal or factual information, episodic memories, etc.; Atkinson & Shiffrin, 1968; Cowan, 2008, 2019). Working memory is not said to be heavily involved in long-term memories that are purely implicit, procedural (Yang et al., 2020), or semantic. We believe that for those kinds of memory, attention is still probably needed for adequate perception of the stimuli but not as much for retention. One function of the focus of attention is association between concurrently held items and their context, which is more important for declarative, episodic memory than for the other types. This interdependent relationship between working and long-term declarative memory serves as part of the evidence that helped to build the embeddedprocesses model, which seeks not to represent these mechanisms as fully separate, but as parts of the same system, functioning interactively.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
75
76
ERYN J. ADAMS ET AL.
Alternatively, some theories go as far as to propose that working (or short-term) memory and long-term memory are not at all separate from one another, but a part of the same unitary memory system. While theories that propose some degree of separation of these two types of memory highlight time- or capacity-based forgetting, unitary theories attribute most of forgetting to retroactive and proactive interference (Underwood, 1957). Contemporary theorists with similar views argue that memories that seem to be more available in the short term are only so because of certain features or cues about those stimuli; recent items are said to be more distinct from one another, and the encoding and retrieval contexts are similar. Effects that have been said to provide evidence for a distinct short-term store, such as the word length effect (i.e., better recall of words that can be rehearsed more quickly) can be eliminated by controlling other factors such as the frequency and lexicality of word stimuli (Nairne, 2002; Jalbert et al., 2011). Despite interesting findings like those, there remains a case to be made for a distinction between long- and short-term memory. Studies have reliably shown some sort of time-based (Ricker et al., 2014) or capacity-based (Awh et al., 2007; Cowan, 2001) limit for working memory, both of which are supported by the embedded-processes model.
5.2.2 Activated Long-Term Memory Within long-term memory, a subset can be activated to a heightened state. Activation can be defined as information that is more highly accessible to be attended to at any moment. Information can be activated without necessarily being attended to. One of the characteristics of information in activated longterm memory is that its activation is presumed to be limited by time. Items can only stay activated for so long before the activation decays, making the items inactive in long-term memory. (The next section indicates how reactivation using rehearsal or the focus of attention can stave off this decay.) Items in activated long-term memory can also be affected by interference from incoming information that possesses similar features to that of the activated information (e.g., visual, auditory, or tactile features). Whereas other theories of working memory divide working memory into distinguishable parts or stores that can process specific types of information, the embedded-processes model does not attempt to draw these boundaries between modalities within short-term memory. It is presumed that there can be an extensive number of different types of incoming information, which can be combined in many different ways. Thus, instead of focusing on creating these various silos for different material types, the model considers the role of interference between similar items (Cowan & Barron, 1987; Elliot et al., 1998). Either separate verbal and visual stores (as in the models built on Baddeley & Hitch, 1974) or separate verbal and visual features within a general store (as in the embedded-processes approach) can account for the finding that there is more interference between two activities having similar features.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
5.2.3 Focus of Attention Within activated long-term memory lies the information that is being attended to at the current moment. This is information that the individual is aware of and comprises the contents of consciousness. Given this awareness, the information is represented in greater detail because of higher levels of activation distributed to the neural representation. Though the representation itself is the same, the higher activation allows for more details to be discernible. Akin to shining a flashlight on an object in a dimly lit room, although the object is the same, the identity, shape, colors, and other fine details become more apparent with activation. In this state, the information is ready to be useful for carrying out complex tasks. This small amount of information can be said to be within the focus of attention, a central storage unit, housed within a larger storage unit, for this highly processed information in the form of up to several coherent, meaningful chunks. The focus of attention is not conceived as a separate copy of the information, but rather an index of relevant activated memory features from long-term memory, organizing them into a meaningful scene (for a review of relevant evidence see Cowan, 2019). The defining characteristic of the focus of attention is its capacity limit. Features are bound into coherent objects and scenes in the focus of attention, summing to no more than several separate ideas or chunks of information at once (Cowan, 2001). At a first glance at the literature, there may appear to be conflicting claims about the nature of this capacity. Some suggest that there is a time-based limit. For example, the multicomponent model of working memory suggests that people can remember only about 2 seconds’ worth of verbal information (Baddeley et al., 1975). While there is certainly evidence for such a time-based constraint, this sort of limit cannot fully explain the storage capacity of items currently being attended to. This becomes clear when working memory measures prevent individuals from relying on certain strategies such as rehearsal or chunking. One such measure is the running span, in which participants hear or see a string of stimuli and do not know when the list will end. Once the list ends, the individual is required to recall as many of the list-final items (e.g., digits) as possible. On average, people can consistently recall around 3–5 items (Cowan, 2001). Phenomena such as this inspired the proposal of a more precise average working memory capacity of about 4 1 items, updated from Miller’s (1956) 7 2 which is typically found when people can group or chunk information in order to expand recall. Cowan (2001) highlighted a series of studies pointing to an average capacity of 4 1 items when effective mnemonic activity seemed unlikely. For example, these small capacity limits can be found when individuals are asked to recall visual information presented briefly and simultaneously (Luck & Vogel, 1997) or when they must repeat a short word or syllable in order to prevent rehearsal (a manipulation known as articulatory suppression; Richardson & Baddeley, 1975; Soto & Humphreys, 2008). Thus, for the average human
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
77
78
ERYN J. ADAMS ET AL.
being, it is only possible to focus on a few items, thoughts, or grouped ideas at once, although these limited items can be from any modality (visual, verbal, tactile, etc.). At least one study shows that both a time limit and a capacity limit can operate, mechanisms of forgetting proposed within the activated portion of long-term memory and the focus of attention, respectively. Chen and Cowan (2009) presented participants with lists of paired words and lists of unpaired singletons. The arbitrary word pairs (e.g., cat-bog) were presented for paired associate recall until they were well-learned, and then they were used within lists to be recalled in order. Well-learned pairs would serve as learned chunks, whereas unpaired singletons (known words) would serve as shorter chunks. The results showed that when words recalled from the lists were counted correctly without regard for serial order and verbal rehearsal was prevented through articulatory suppression, there was a strict capacity limit: participants recalled about 3 chunks on average, regardless of whether these were pairs or singletons. However, when recalled words were counted correct only if they were recalled in the correct order, and rehearsal was possible, time came into play. Then participants could not recall as many word pairs as they could singletons, and they recalled about 2 seconds of speech in each case. This set of findings suggests that time and capacity operate together in the same task but that one or the other may limit recall, depending on whether serial rehearsal is involved.
5.2.4 Central Executive There is of course no use to being able to hold information in a highactivation state if it cannot be applied to a task. The central executive is defined as a collection of mechanisms or processes that can manipulate the information within the focus of attention, as well as within the activated portion of long-term memory. The latter occurs, however, by a search that brings the information back into focus. The central executive is especially used when the individual is intentionally attempting to perform an action on the information within working memory, whether that be guided by volition or external instruction. For example, one could be reading the words on this page at the moment, but with the central executive, choose to switch attention to another stimulus in their environment like a humming fan or the tick of a clock. The central executive is not one sole mechanism, but a collection of different processes.
5.2.5 The Cohesive Model One of the features of the embedded-processes model is the various number of ways information can move into and throughout the different subparts, through varying levels of activation. It is not always linear and at times can
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
depend on the nature of the information itself. We will describe how a simple unit of information, a word, may travel through the model (Figure 5.1) and be used in working memory. When an individual first hears a word, the information must travel through a brief sensory store, which can pick up on the shallow characteristics of the word (phonology, volume, etc.). If the individual is actively trying to attend to the word or if it is presented intensely (as in a loud voice), it enters into the focus of attention, especially if it has particular semantic relevance (e.g., one’s name or the presently attended topic). At the same time, features of the word will automatically activate parts of long-term memory that represent those features. They can include acoustic, phonological, and orthographic features, tone of voice or manner of writing, and, if there was some attention devoted to the word, its lexical identity and meaning. This activation supports the presence of the word in the focus of attention. If the person must keep the word in mind for some reason, they can recruit central executive processes to keep the word in this most accessible state by keeping the focus of attention on it. For example, the person may repeat the word silently in mind (i.e., subvocal rehearsal). If this word is repeated enough and associated with other information, it may gain a rich enough representation to become a new episodic memory that is available not only now, but also later. That kind of representation can free up the focus of attention by essentially off-loading information to long-term memory (Rhodes & Cowan, 2018). It may remain active for a short period of time, given its recent visit to the focus of attention, but without reactivation of the word, it will begin to fade from this activated form within seconds. Activated long-term memory of a recent episode always becomes a part of long-term memory according to this approach (see Cowan, 2019), but possibly not one that is strong enough to be retrieved given all of the interference in long-term memory and changes in retrieval contexts. The long-term representation can still be accessed in many cases, albeit not as quickly as if it were in a more activated state. Alternatively, if the word was not rehearsed or attended to for a sufficient period of time, it may decline for all practical purposes to an activated-only state. Without reactivation, it will decay to a point past retrieval (i.e., forgetting). However, if the word did sufficiently become a part of long-term memory, it is available to be activated when similar items or stimuli trigger the memory of the word. It can then enter back into the focus of attention to be used for a cognitive task such as production.
5.3
The Embedded-Processes Model in Active Language Use
In this section we consider the embedded-processes model in the context of language use. We consider the role of attention (5.3.1), the application to language comprehension (5.3.2), and the application to language production (5.3.3).
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
79
80
ERYN J. ADAMS ET AL.
5.3.1 Role of Attention Now that we have established a basic description of the different subparts of the embedded-processes model, we can begin to understand how language use may be handled by the model, as a whole. For typically developing individuals, much of language is processed automatically, and this processing requires little or no general attention and tends to be difficult to interrupt (Schneider & Shiffrin, 1977). A key issue is that there is a limit to the information that can be processed automatically, and we wish to know that limit. This issue was left as an unknown by Cowan (1988) but it was subsequently addressed, as we will see. In the embedded-processes model, it is assumed, a neural model of the current environment exists and allows adaptation to that environment. Repeated elements of the environment are not attended; there is habituation of the attentional orienting response. Changes in the environment that are detected and are discrepant from the neural model result in recruitment of attention to the change through an attentional orienting response (Sokolov, 1963). This response is viewed as dishabituation of orienting. In studies of what is known as the cocktail party effect (Broadbent, 1958; Cherry, 1953) effects of attention to speech can be studied. This effect is found in the laboratory setting using the selective listening task, in which participants receive two different messages, each to a different ear, simultaneously. They are typically asked to attend to one ear and repeat (shadow) its message while ignoring the other message. When this is done, participants generally can only remember basic characteristics about the stimuli in the unattended ear, such as whether the last few seconds was speech and what its pitch was, but not semantic information. If there is a dramatic change in pitch within the unattended channel it will be immediately noticed. Wood and Cowan (1995a) showed that it took a number of seconds to notice after the unattended channel changed from forward speech to backward speech, indicating that even phonological processing without attention was only rudimentary. However, attended semantic characteristics should be processed and should become part of the neural model, in which case a change in attended topic should be noticed (but not a change in ignored topic). Some subsequent studies have suggested that semantic information can be automatically activated. This can occur, for example, in the case of masked stimuli below the conscious threshold (Balota, 1983). There is not much evidence, however, that semantics can be automatically processed enough to produce new long-term memory traces. Eich (1984) concluded that semantic memories could persist without attention, after observing that presenting connected word pairs in the unattended channel (e.g., taxifare), altered the frequency with which participants who were then asked to spell the second word wrote fare instead of fair. However, Wood et al. (1997) found that Eich used too-slow stimuli to be shadowed, allowing attention to wander to the supposedly ignored channel; using a more usual shadowing rate abolished the effect.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
In another use of selective listening to examine the role of attention in semantic processing, Moray (1959) and Wood and Cowan (1995b) show that some people can remember the unattended information if it holds special relevance; about a third of people tested noticed and remembered their own name in the unattended channel. However, a limit in this evidence is the possibility that, as Wood et al. (1997) found, attention could have wandered to the supposedly ignored channel. To examine this possibility, Conway et al. (2001) explored how name-noticing varies by working memory capacity. In that study, participants completed the operation span task as a measure of working memory capacity. The results showed that while only 20 percent of individuals in the top quartile of working memory capacity noticed their name in the unattended channel, 65 percent of individuals with low capacities noticed their names. Participants with lower capacities also demonstrated more difficulty repeating the words in the attended channel. The authors proposed that a basis of individual differences in working memory is the ability to use attention to inhibit distracting information, leaving working memory free to concentrate on the task-relevant information (cf. Vogel et al., 2005). What is available in unattended speech is an auditory sensory memory that retains some of the useful information for a few seconds until attention can be turned to it. Cowan et al. (1990) examined memory for an ignored speech channel during reading. Occasionally, a change in the room lighting indicated that attention should be switched to remembering the last spoken consonant-vowel or vowel-consonant syllable. They found that if the last unattended syllable occurred 10 seconds prior to the attention switch, final vowels were remembered best; nonfinal phonemes, second best; and final consonants were forgotten the most. Balota and Duchek (1986) used a procedure in which a word list was followed by an interfering sound to show that immediately after the list, speech of any kind could interfere; after a 20-second, silent delay, the speech representation was gone, whereas an acoustic representation remained, so only a speech sound in the same voice as the list interfered with memory at that point. In sum, in the updated embedded-processes model (e.g., Cowan, 1999, 2019), it is assumed that little language processing is automatic except for the unprocessed acoustic representation and potentially some but not all speech-specific, phonological features. However, if enough attention is devoted to a channel of speech or writing to notice something meaningful or relevant, an orienting response is likely to occur, recruiting further attention to the stimulus to allow further analysis. In the following sections we outline two major language skills and describe how the embedded-processes model may account for how these tasks take place. We begin with the complex task of understanding language, then we move on to the equally complex task of producing language. Within each subsection, we discuss the theory as a whole, but highlight the parts of the model which may be especially important to the given task.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
81
82
ERYN J. ADAMS ET AL.
5.3.2 Comprehending Language From as early as 6 months of age, infants can begin distinguishing and mapping meaning to words spoken in their environment (Bergelson & Swingley, 2012; Saffran et al., 1996). As they develop, this ability expands and grows more efficient until comprehension of most native language structures becomes nearly effortless. Still, the act of understanding language in discourse or writing requires some level of working memory, especially when structures are less familiar or ambiguous (Kemtes & Kemper, 1997; Swets et al., 2007). Even in familiar language, one must use working memory to keep track of the themes, content, and the message of language being spoken or read in order to respond properly. 5.3.2.1 The Role of the Various Embedded Processes in Comprehension We can examine how the different parts of the embedded-processes model interact to contribute to active comprehension. Given that the focus of attention is nested within a larger long-term memory system, it is well equipped to deal with comprehending language, which most of the time depends on knowledge and recognition of previous structures. As people learn words and structures over time, they are added to the long-term memory store. In the case of words, this long-term memory can be described as the person’s lexicon. Broader and deeper lexicons should result in easier and faster comprehension (Ouellette, 2006; Perfetti, 2007; Qian, 1999). Syntax knowledge may work a bit differently given that there is some evidence that syntax relies more on procedural memory, a type of memory thought to be relatively separate from the explicit memory types such as working memory (also see work on statistical learning of syntax: Kidd, 2012; Thompson & Newport, 2007). For example, Ferreira et al. (2008) presented amnesic patients with a series of trials in which they would repeat a sentence that primed a specific syntax (e.g., the passive voice) and then were asked to describe a picture showing a transitive action. The syntax used to describe this image was examined to see if patients used the same syntax as what was previously heard, a phenomenon known as syntactic priming or syntactic persistence. After describing the target picture, they would hear the prime sentence again and be asked to recall if they heard it previously, to test simple recognition memory. Although patients did show impaired recognition for the prime sentence, they showed similar syntactic persistence as control subjects. These results suggest that at least some substantial portion of syntactic processing does not rely completely on declarative memory, the form of memory damaged in those with amnesia. Thus, people may be able to comprehend and remember syntactic structures without a complete reliance on working memory. However, if we still consider procedural memory to be a part of a broader long-term memory system, the focus of attention can still be embedded within and rely on these processes. So, in this model, working memory for language comprehension can rely heavily on previous knowledge stored within long-term memory. What
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
happens when we are actively trying to comprehend language being currently spoken or read? We can begin with the process of the activation of long-term memory, although other processes may be happening concurrently. When language is heard or read, certain linguistic structures and words will automatically activate portions of long-term memory that have overlapping characteristics. This activation is very similar to the process described by lexical activation hypotheses. In some lexical activation hypotheses, words are considered to be a part of connectionist-like models (Gwilliams et al., 2017). Words with overlapping features (e.g., phonemes) are more closely connected to one another. The presentation of one word can therefore activate morphemes that are similar or related, creating a sort of sensitivity to these similar stimuli (Pitt & Samuel, 2006). Words forms may also be associated with one another in long-term memory because of experience. A nurse may associate the words “heart” and “rate” together, while a psychologist may more quickly associate “rate” with “survey” or “proportion.” When words or ideas are activated in long-term memory, they become highly accessible for more detailed use, which allows for more rapid and efficient comprehension. If activated information is being actively attended to and is therefore in a listener or reader’s current awareness, it is said to be in the focus of attention. Information in the focus of attention is processed at the highest level of detail but is the most limited in the amount that can be processed at once, which calls for strategies to overcome these limits. For example, preexisting associations in long-term memory can be perceived within the stimuli to form chunks in working memory (Miller, 1956), e.g., making a 6letter list much easier to remember if it comprises the two known acronyms USA-FBI; and new perceived patterns can be rapidly memorized to form new chunks (Cowan, 2019; Gilchrist et al., 2008, 2009; Thalmann et al., 2019). It is possible that not every chunk of information being read or listened to is fully processed and represented. Some evidence of such impartial encoding is shown in work on ambiguous or garden path sentences. Ferreira et al. (2002) provided the common example sentence, “While Anna dressed the baby played in the crib” (the lack of a comma contributing to the ambiguity of the sentence). When participants answer comprehension questions, it often becomes clear that they come away with conclusions such as Anna is dressing the baby and the baby is playing. This provides some evidence that the sentence was most likely processed in parts, with the first part of the sentence already being somewhat pushed out of the focus of attention while the second part of the sentence is being processed. If a participant can reread the sentence, they might be able to come away with the correct meaning. This rereading may serve as sort of a rehearsal mechanism that allows the representation to become more strongly built than it was the first time it was being read. If a sentence can thus be processed as one chunk, it may be easier to come away with correct interpretations. For most language and discourse processing, partial representation may be adequate for successful comprehension.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
83
84
ERYN J. ADAMS ET AL.
5.3.2.2 Interaction Between Linguistic Levels An important issue beyond simple chunking is how a small amount of capacity-limited working memory can be used along with the activated portion of long-term memory to comprehend complex language. Notions to resolve that issue seem compatible with the embedded-processes approach. Sachs (1967) used a procedure in which a speech passage was followed by a recognition test for a sentence that had occurred 0, 80, or 160 syllables back in the passage. The test probe was a sentence identical to the one in the passage or else syntactically or semantically changed. All kinds of probe sentences were usually classified correctly when they were the most recent ones in the passage but after 80 syllables of interference, the semantic change was recognized well, but syntactic changes usually could not be distinguished from the correct sentence. This result suggested that listeners save the meaning and quickly lose the verbatim information, so it appears that what is saved by the working memory system by that time is only a unified conceptual scheme for the passage. Jarvella (1971) used cued recall to show that even with the amount of interfering material controlled, verbatim information about a clause was remembered better when it began the last sentence in a passage compared to when it ended the penultimate passage, suggesting that syntactic boundaries are used to close off the syntactic analysis and remember primarily the gist. This gist also includes the purpose (pragmatic value) of the speech. Jarvella and Collas (1974) found that repeated sentences were not recognized well when the intent of the sentence changed (e.g., “I’m always here” spoken as a complaint versus an invitation). In the theoretical treatments of discourse by Kintsch and van Dijk (1978) and Kintsch (1988), even the recognition of words does not come simply from the lexicon, but rather from the lexicon paired with the current situation, which can make the difference between, for example, the bank of a river and a bank to deposit money. This interactive, contextually based process would not be considered fully automatic in the embedded-processes model; attention to the situation would be needed to guide the interpretation of language and the summarizations of the key ideas held in the focus of attention. The central executive processes would direct the focus of attention so as to activate the relevant concepts so that automatic processes would be guided in the right direction to interpret the conversation most of the time.
5.3.2.3 Attention Filtering and Switching in Language Comprehension Dishabituation of the orienting response, mentioned earlier, is often used to understand how infants process various types of information, including elements of language. For example, Burns et al. (2007) studied monolingual and bilingual infants’ processing of phonemes in their native and nonnative language(s). In paradigms such as these, infants are habituated to repetitive stimuli, such as specific syllables pronounced by a voice. Then, when the stimulus changes, the infants’ looking times are measured as an indicator
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
of interest or awareness of a stimulus change. Burns et al. (2007) showed that at 6–8 months, both monolingual and bilingual infants are able to distinguish between similar sounding syllables present in English and French. However, by 14–20 months, only bilingual infants remain able to distinguish between these sounds. Here, the orienting response is able to tell us how infants use phoneme discrimination to learn their language(s). Orienting to new language sounds in the environment may be an important ability for learning new words and linguistic patterns. Though our attention can be drawn away to potentially important stimuli automatically as in the instances above, we can also select what information is to be stored and processed in the focus of attention using filtering and inhibition mechanisms. This ability can be especially important when trying to focus on listening to one speaker in a lecture hall full of distractions, maintaining attention on written words on a page, or keeping track of a list of directions in an unfamiliar city. This volitional recruitment of information into the focus of attention can follow a similar path as the information that did so automatically, beginning with surface-level processing through the brief sensory store, then directly into the focus of attention (portrayed by the long-dash line in Figure 5.1). These selective attention skills become especially important in highly complex language skills, such as interpreting (Cowan, 2000). Interpreters face the complex task of listening to one stream of language while maintaining the words and meaning while being ready to produce it in another language. This takes a substantial amount of attention filtering and selection in order to avoid distractions and miss part of an intended message.
5.3.3 Producing Language Along with the ability to understand language, we must be able to produce our own language in discourse and writing in order to carry out successful communication. The act of producing language certainly requires some level of working memory, although its involvement may change with development or mastery of language. Although there is an extensive amount of evidence for the role of working memory in language comprehension, much less work has been done on the relationship between working memory and active language production. Next, we will highlight ways in which the different parts of the embedded-processes model may combine to allow for successful language production. 5.3.3.1
The Role of the Various Embedded Processes in Language Production Much like comprehension, production relies on foreknowledge which can be permanently stored within long-term memory. Whereas comprehension relies heavily on recognition processes, language production relies on retrieval processes in long-term memory. We can follow the process of
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
85
86
ERYN J. ADAMS ET AL.
producing language using a model such as Bock and Levelt (1994). In their model, the speaker must begin with an intended message in mind. Next is the functional processing stage, in which they engage in lexical selection and role assignment. This lexical selection process, especially, requires a search through long-term memory for the correct word forms. The search can be carried out by central executive processes to choose the most relevant word for the intended message. While words are being selected, the focus of attention must be engaged in order to hold the intended message in mind and keep track of words that have already been produced. In this way, working memory can ensure that we do not lose the train of thought in the middle of a sentence. However, perhaps some of the most revealing aspects of production come from instances where the speaker does lose the train of thought or produces a speech error. These moments reveal aspects of language production that might be especially reliant on working memory, such as planning ahead. In work done on speech errors, Fromkin (1973) highlights how errors are often consistent and predictable across speakers. For example, people often inadvertently switch the nouns of a sentence (e.g., The dog is taking the babysitter for a walk) or switch the phonemes of two words in a sentence (e.g., I’ll nite it in my wrotebook). These slips of the tongue reveal that speakers are often thinking of the end of the sentence well before it is spoken, an ability that must require some planning and storage mechanism, which the focus of attention and activated long-term memory are well equipped for. Moreover, misplacement of higher-level units (e.g., a spoonerism such as I threw my window out the clock) typically occurs further apart than misplacement of lower-level units (e.g., instead of a bit nasty, saying a mit dasty). This phenomenon suggested to Fromkin that people plan ahead at a high level and cannot plan as far ahead at a lower level. As planned words or sentences are activated in long-term memory, similarities between units and spreading activation between different levels of units may cause the wrong ones to become more active than the correct ones (Dell, 1986; Dell et al., 2013). The central executive may then select incorrect phonemes, morphemes, or words from among the highly activated units. If the focus of attention and the central executive processes that manipulate the information within the focus of attention are taxed by other information, the incorrect syllable or lexical item may often be selected, leading to these slips of the tongue. Thus, the best way to ensure the intended message is produced is to have available attention to direct speech carefully and catch potential slips of the tongue before they are emitted. Activated long-term memory plays an important role in supporting the fluency of language. Like comprehension, speaking a word or thinking of an intended message may activate other related words in mind. This allows for a more rapid and efficient retrieval of related words and ideas. Not having some sort of activation mechanism should result in disjointed retrieval of
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
words and ideas and perhaps influent language. As we have discussed, items in activated long-term memory are susceptible to interference from information sharing similar features. Thus, words, phrases, or ideas similar to the message being spoken may interfere. This can manifest in a couple of different ways. One may choose a related but unintentional word while speaking. For example, a person may utter, “Are you going to the concert?” in place of the intended, “Are you going to the conference?” It could be that the word concert was activated in long-term memory at the thought of a conference because of the similarity in phonetic or morphemic structure (i.e., the prefix con-). Additionally, the thought of conferences may activate schemas of other large gatherings of people, hence making words such as concert more accessible. Another way this may happen is through a more external activation. You may have experienced a situation in which you were trying to speak or write while in the presence of another speaker (e.g., a television in the background or a nearby conversation). If attention briefly slips from your current speech to that of the other speaker, any words they say may be briefly activated in long-term memory, making them more accessible. By way of interference, you may even produce a word that was said by the other speaker, even if it is irrelevant to your current message. These sorts of slips give us a window into understanding how the mind processes words that are being heard while simultaneously trying to speak. The previously mentioned attention filtering and switching processes are especially important to avoid producing too many errors. The focus of attention, as well as central executive processes, can help keep track and monitor speech in order to ensure intended production. Very few studies have examined the role of working memory for active language production. Some studies do reveal reliable correlations between working memory and speech fluency (e.g., Eichorn et al., 2016). In general, the greater the working memory capacity, the better individuals perform on tasks that require language production skills. This relationship may not only be correlational. Hartsuiker and Barkhuysen (2006) conducted a study in which participants were to complete sentence fragments while holding a small amount of unrelated information in mind. They examined how this working memory load affected participants’ concurrent ability to correctly complete a sentence in which the subject and verb agreed. Results showed that working memory loads led to more subject-verb agreement errors, giving good evidence for the role of working memory in online sentence production. More attention is likely necessary for constructing more complex sentences or dealing with syntax that is more difficult. Central executive processes must be able to allocate the focus of attention to provide an adequate number of resources for performing the latter parts of the language production process (such as positional processing in the Bock & Level model). This may mean that smaller capacities could lead to more difficulty producing complex structures, which is evident in studies that examine verbal abilities in older adults, who typically score lower on measures of
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
87
88
ERYN J. ADAMS ET AL.
working memory than younger adults (Kemper et al., 2003). This may also be true during childhood, which we will address in the next section.
In sum, language production relies on successful retrieval processes from long-term memory in order to choose correct words, ideas, and syntax structures. The focus of attention and central executive processes work together to store and keep track of words that have already been said while planning for future intended words or messages. The activated portion of long-term memory is likely very important for allowing for fluent language as words that are related or similar to the currently produced words are made more highly accessible. Although interference may occur, attentional processes can be enacted in order to avoid errors. Altogether, the system works to allow for an incredible amount of production that can be successfully comprehended by a listener.
5.4
Using Working Memory to Acquire Language
Although language processes may seem effortless in typical cases, it is perhaps one of the most complex abilities that is gained within the first several years of life. People also remain capable of acquiring new language(s) later in life, although it may be more difficult once a native language has been established. The evidence for a childhood-specific sensitive period remains mixed (Bialystok, 1997; Norrman & Bylund, 2016). In either case, it seems that acquiring language requires cognitive resources such as working memory. In the following sections, we address how working memory might be used when infants and children are learning their language(s). Then, we discuss how working memory is useful during the later acquisition of another language.
5.4.1 Developmental Acquisition Children improve on working memory tasks with age (Gathercole et al., 2004), which is good evidence that working memory capacity is one of the resources that increases with maturation. Cowan (2016) reviewed the question of what allows for such cognitive growth, with several studies supporting a hypothesis for a maturational growth of working memory capacity itself. Although knowledge and experience can certainly enhance working memory performance (Halford et al., 2007; Hambrick & Engle, 2002), they cannot completely explain growth in capacity. Cowan et al. (2015) explored this question by presenting children in grades 1 through 7, as well as college students with arrays of English letters or unfamiliar characters. Given that people have more experience with English letters, theories that emphasize the importance of knowledge may expect better memory for letters than unfamiliar characters. They may also expect no gain in memory performance of the unfamiliar characters with age, as the unfamiliarity of
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
the characters controls for knowledge. Results did show that memory for letters was better than that of characters and that memory for letters improved with age. However, memory for the unfamiliar characters also improved with age. In fact, when the performance on both tasks were normalized via z-scores, the developmental trajectories were nearly identical. It seems that knowledge is not a fully sufficient explanation for working memory capacity growth in development. This is especially important when considering working memory for language. As children begin to learn words and sentence structures, they still must have adequate working memory resources in order to be able to successfully understand and produce these items. In the embedded-processes model, this capacity growth is largely attributed to a growth in the focus of attention. This growth can be due to structural changes within the first few years of life in relevant brain areas that allow for more storage and manipulation of items within storage, such as the intraparietal sulcus and more frontal regions (Cowan, 2011; Klingberg et al., 2002; Thomason et al., 2008). As this focus of attention grows, children are better able to store chunks of information, such as sentences during language processing. To study this specific ability, Gilchrist et al. (2009) asked children, 7 and 12 years of age, as well as adults to remember lists of spoken sentences of varying lengths and readability (some sentences were lists of random words intonated like sentences). With age, the number of sentence chunks that participants could access, defined as being able to remember at least one content word from the sentence, increased with age. However, the number of words they could remember from chunks they could access stayed relatively consistent with age at around 80 percent. This provided evidence for a growth of capacity with age, but also showed that it is useful to think of sentences as item chunks in memory. Once a chunk is formed, the vast majority of it can be remembered, likely enough for successful comprehension. As discussed, working memory appears to be especially important for planning ahead in speech and avoiding errors. This may be especially true for children, who are learning to master their native language(s). Adams and Cowan (2021) examined how working memory might be useful for young children while attempting to use an unfamiliar syntactic structure, the passive voice. Participants, 4 and 5 years of age, viewed transitive-action images and listened to an experimenter describe each one using a single passive voice sentence (e.g., The balloon was popped by the giraffe). Then, participants were presented with the same images and asked to recall what occurred while also attempting to use the passive voice like the experimenter. In two of the three experimental blocks, participants were also asked to hold a small amount of unrelated information (a few tokens dispersed in a 4 4 matrix or a few spoken digits) while producing the target sentence. The results unexpectedly showed that participants were more likely to transform the sentence to the active voice when there was no
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
89
90
ERYN J. ADAMS ET AL.
additional working memory load. Moreover, participants produced fewer errors when they did switch the sentence to the active voice. One of the takeaways was that young children might use working memory to transform sentences into more familiar forms, and more importantly, avoid errors in speech so that their conversational partner can understand the intended message. When there was a working memory load, children were more likely simply to repeat the passive-voice model that had been presented, but at the risk of making silly errors they would not otherwise make, such as saying The giraffe was popped by the balloon. Overall, working memory has reliably predicted children’s ability to understand and produce language (Adams, 1996; Blake et al., 1994). As the focus of attention grows, it is better able to add more information to long term memory. At least from the elementary school through adulthood, people appear to improve in ways to perceive patterns to allow information to be off-loaded from the focus of attention to preserve it for further processing (Cowan et al., 2018) and improve in the ability to be proactive, preserving what is needed for upcoming tasks instead of just the current task (Cowan et al., 2021). As central executive processes improve, more complex cognitive tasks can be performed on stored information. Thus, it seems critical to consider the role of working memory in most studies of language development and also to further clarify the connection between these two cognitive functions.
5.4.2 Acquiring a Second Language Although language acquisition is typically discussed in terms of natural acquisition during development, there are cases where one might want or need to learn another language. Thankfully, we maintain the ability to learn language later in the lifespan at fairly impressive speeds. Much like children learning language, second language learners must rely on working memory to help encode new words and structures. Most studies linking working memory to second language acquisition have been correlational, showing that higher working memory capacities lead to better fluency, processing, and proficiency (Darcy et al., 2015; Linck et al., 2014; van den Noort et al., 2006). Working memory may add to a number of individual differences predicting why some people have an easier time acquiring a second language than others (Hanulíková et al., 2012). There also appears to be some evidence for the in-the-moment use of working memory during the active acquisition of a language. For example, Ellis and Sinclair (1996) examined how rehearsing foreign language utterances in working memory is related to long-term knowledge of structures. They found that participants who were allowed to repeat foreign language sentences (Welsh for English speakers) were more likely to remember words and phrases as well as pronounce them correctly, which contributed to an overall more accurate rapid acquisition of the foreign structures. In the embedded-processes
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
model, we would suggest that being able to maintain attention along with the benefits of rehearsal allowed for a more successful consolidation of these forms into long-term memory. Once the words and structures are well established into long-term memory, they become available to be activated and used in the focus of attention again. While the evidence that working memory plays a role in second language acquisition is still being developed, the collection of studies thus far gives good reason to consider working memory and attention in language acquisition studies. It could help to understand why people may differ in efficiency and speed of acquisition and how we can best teach to the level of working memory so that all learners have a chance to master additional languages.
5.5
Conclusion
The embedded-processes model seeks to classify working memory as a system of interdependent processes that are not easily separable but do each possess some unique functions. In language use, each of these different processes can contribute to all aspects of communication, from understanding to producing language. Attention and long-term memory processes appear to play important roles in language use. With more consideration of these resources, we might then be better able to understand how such processes contribute to development and the mastery of language, as a whole by determining the manner and extent to which each kind of process relies on attention and capacity-limited working memory.
References Adams, A. M. (1996). Phonological working memory and spoken language development in young children. The Quarterly Journal of Experimental Psychology Section A, 49(1), 216–233. Adams, E. J., & Cowan, N. (2021). The girl was watered by the flower: Effects of working memory loads on syntactic production in young children. Journal of Cognition and Development, 22(1), 125–148. Adams, E. J., Nguyen, A. T., & Cowan, N. (2018). Theories of working memory: Differences in definition, degree of modularity, role of attention, and purpose. Language, Speech, and Hearing Services in Schools, 49(3), 340–355. Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. Psychology of learning and motivation, 2(4), 89–195. Awh, E., Barton, B., & Vogel, E. K. (2007). Visual working memory represents a fixed number of items regardless of complexity. Psychological Science, 18(7), 622–628.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
91
92
ERYN J. ADAMS ET AL.
Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417–423. Baddeley, A. D., & Hitch, G. (1974). Working memory. In Psychology of learning and motivation (Vol. 8, pp. 47–89). Academic Press. Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of short-term memory. Journal of Verbal Learning and Verbal Behavior, 14, 575–589. Balota, D. A., & Duchek, J. M. (1986). Voice-specific information and the 20-second delayed suffix effect. Journal of Experimental Psychology: Learning, Memory, and Cognition, 12, 509–516. Balota, D. A. (1983). Automatic semantic activation and episodic memory encoding. Journal of Verbal Learning and Verbal Behavior, 22, 88–104. Bergelson, E., & Swingley, D. (2012). At 6–9 months, human infants know the meanings of many common nouns. Proceedings of the National Academy of Sciences, 109(9), 3253–3258. Bialystok, E. (1997). The structure of age: In search of barriers to second language acquisition. Second Language Research, 13(2), 116–137. Blake, J., Austin, W., Cannon, M., Lisus, A., & Vaughan, A. (1994). The relationship between memory span and measures of imitative and spontaneous language complexity in preschool children. International Journal of Behavioral Development, 17(1), 91–107. Bock, K., & Levelt, W. J. (1994). Language production: Grammatical encoding (pp. 945–984). Academic Press. Broadbent, D. E. (1958). Perception and communication. Pergamon Press. Burns, T. C., Yoshida, K. A., Hill, K., & Werker, J. F. (2007). The development of phonetic representation in bilingual and monolingual infants. Applied Psycholinguistics, 28(3), 455–474. Chen, Z., & Cowan, N. (2005). Chunk limits and length limits in immediate recall: A reconciliation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31(6), 1235. Chen, Z., & Cowan, N. (2009). Core verbal working memory capacity: The limit in words retained without covert articulation. Quarterly Journal of Experimental Psychology, 62, 1420–1429. Cherry, E. C. (1953). Some experiments on the recognition of speech, with one and with two ears. The Journal of the Acoustical Society of America, 25(5), 975–979. Conway, A. R., Cowan, N., & Bunting, M. F. (2001). The cocktail party phenomenon revisited: The importance of working memory capacity. Psychonomic Bulletin & Review, 8(2), 331–335. Cowan, N. (1988). Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human informationprocessing system. Psychological Bulletin, 104(2), 163. Cowan, N. (1999). An embedded-processes model of working memory. Models of Working Memory: Mechanisms of Active Maintenance and Executive Control, 20, 506.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
Cowan, N. (2000). Processing limits of selective attention and working memory: Potential implications for interpreting. Interpreting, 5(2), 117–146. Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–114. Cowan, N. (2008). What are the differences between long-term, short-term, and working memory? Progress in Brain Research, 169, 323–338. Cowan, N. (2011). The focus of attention as observed in visual working memory tasks: Making sense of competing claims. Neuropsychologia, 49(6), 1401–1406. Cowan, N. (2019). Short-term memory based on activated long-term memory: A review in response to Norris (2017). Psychological Bulletin, 145(8), 822–847. Cowan, N., AuBuchon, A. M., Gilchrist, A. L., Blume, C. L., Boone, A. P., and Saults, J. S. (2021). Developmental change in the nature of attention allocation in a dual task. Developmental Psychology. 57(1), 33–46. Cowan, N., & Barron, A. (1987). Cross-modal, auditory-visual Stroop interference and possible implications for speech memory. Perception & Psychophysics, 41(5), 393–401. Cowan, N., Li, Y., Glass, B., & Saults, J. S. (2018). Development of the ability to combine visual and acoustic information in working memory. Developmental Science, 21, e12635, 1–14. Cowan, N., Lichty, W., & Grove, T. R. (1990). Properties of memory for unattended spoken syllables. Journal of Experimental Psychology: Learning, Memory, & Cognition, 16, 258–269. Cowan, N., Ricker, T. J., Clark, K. M., Hinrichs, G. A., & Glass, B. A. (2015). Knowledge cannot explain the developmental growth of working memory capacity. Developmental Science, 18(1), 132–145. Cowan, N., Rouder, J. N., Blume, C. L., & Saults, J. S. (2012). Models of verbal working memory capacity: What does it take to make them work? Psychological Review, 119, 480–499. Darcy, I., Park, H., & Yang, C. L. (2015). Individual differences in L2 acquisition of English phonology: The relation between cognitive abilities and phonological processing. Learning and Individual Differences, 40, 63–72. Dell, G. S. (1986). A spreading-activation theory of retrieval in sentence production. Psychological Review, 93(3), 283–321. Dell, G. S., Schwartz, M. F., Nozari, N., Faseyitan, O., & Coslett, H. B. (2013). Voxel-based lesion-parameter mapping: Identifying the neural correlates of a computational model of word production. Cognition, 128, 380–396. Eich, E. (1984). Memory for unattended events: Remembering with and without awareness. Memory & Cognition, 12(2), 105–111. Eichorn, N., Marton, K., Schwartz, R. G., Melara, R. D., & Pirutinsky, S. (2016). Does working memory enhance or interfere with speech fluency in adults who do and do not stutter? Evidence from a dual-task paradigm. Journal of Speech, Language, and Hearing Research, 59(3), 415–429.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
93
94
ERYN J. ADAMS ET AL.
Elliott, E. M., Cowan, N., & Valle-Inclan, F. (1998). The nature of cross-modal color-word interference effects. Perception & Psychophysics, 60(5), 761–767. Ellis, N. C., & Sinclair, S. G. (1996). Working memory in the acquisition of vocabulary and syntax: Putting language in good order. The Quarterly Journal of Experimental Psychology Section A, 49(1), 234–250. Endress, A. D., & Potter, M. C. (2014). Large capacity temporary visual memory. Journal of Experimental Psychology: General, 143(2), 548. Ferreira, F., Bailey, K. G., & Ferraro, V. (2002). Good-enough representations in language comprehension. Current Directions in Psychological Science, 11(1), 11–15. Ferreira, V. S., Bock, K., Wilson, M. P., & Cohen, N. J. (2008). Memory for syntax despite amnesia. Psychological Science, 19(9), 940–946. Forbes, P. B. R. (1933). Greek pioneers in philology and grammar. The Classical Review, 47(3), 105–112. Fromkin, V. A. (1973). Slips of the tongue. Scientific American, 229(6), 110–117. Gathercole, S. E., Pickering, S. J., Ambridge, B., & Wearing, H. (2004). The structure of working memory from 4 to 15 years of age. Developmental Psychology, 40(2), 177. Gilchrist, A. L., Cowan, N., & Naveh-Benjamin, M. (2009). Investigating the childhood development of working memory using sentences: New evidence for the growth of chunk capacity. Journal of Experimental Child Psychology, 104(2), 252–265. Gilchrist, A. L., Cowan, N., & Naveh-Benjamin, M. (2008). Working memory capacity for spoken sentences decreases with adult ageing: Recall of fewer, but not smaller chunks in older adults. Memory, 16, 773–787. Gwilliams, L., Poeppel, D., Marantz, A., & Linzen, T. (2017). Phonological (un)certainty weights lexical activation. arXiv preprint arXiv:1711.06729. Halford, G. S., Cowan, N., & Andrews, G. (2007). Separating cognitive capacity from knowledge: A new hypothesis. Trends in Cognitive Sciences, 11(6), 236–242. Hambrick, D. Z., & Engle, R. W. (2002). Effects of domain knowledge, working memory capacity, and age on cognitive performance: An investigation of the knowledge-is-power hypothesis. Cognitive Psychology, 44(4), 339–387. Hanulíková, A., Dediu, D., Fang, Z., Bašnaková, J., & Huettig, F. (2012). Individual differences in the acquisition of a complex L2 phonology: A training study. Language Learning, 62, 79–109. Hartsuiker, R. J., & Barkhuysen, P. N. (2006). Language production and working memory: The case of subject-verb agreement. Language and Cognitive Processes, 21(1–3), 181–204. Jalbert, A., Neath, I., & Surprenant, A. M. (2011). Does length or neighborhood size cause the word length effect? Memory & Cognition, 39(7), 1198–1210. James, W. (1890). The principles of psychology. Henry Holt.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
Jarvella, R. J. (1971). Syntactic processing of connected speech. Journal of Verbal Learning & Verbal Behavior, 10, 409–416. Jarvella, R. J. & Collas, J. G. (1974). Memory for the intentions of sentences. Memory & Cognition, 2, 185–188. Kemper, S., Herman, R. E., & Lian, C. H. (2003). The costs of doing two things at once for young and older adults: Talking while walking, finger tapping, and ignoring speech of noise. Psychology and Aging, 18(2), 181. Kemtes, K. A., & Kemper, S. (1997). Younger and older adults’ on-line processing of syntactically ambiguous sentences. Psychology and Aging, 12(2), 362. Kidd, E. (2012). Implicit statistical learning is directly associated with the acquisition of syntax. Developmental Psychology, 48(1), 171. Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction-integration model. Psychological Review, 93, 163–182. Kintsch, W., & van Dijk, T. A. (1978). Toward a model of text comprehension and production. Psychological Review, 85, 363–394. Klingberg, T., Forssberg, H., & Westerberg, H. (2002). Increased brain activity in frontal and parietal cortex underlies the development of visuospatial working memory capacity during childhood. Journal of Cognitive Neuroscience, 14(1), 1–10. Lewis-Peacock, J. A., & Postle, B. R. (2008). Temporary activation of longterm memory supports working memory. Journal of Neuroscience, 28(35), 8765–8771. Linck, J. A., Osthus, P., Koeth, J. T., & Bunting, M. F. (2014). Working memory and second language comprehension and production: A metaanalysis. Psychonomic Bulletin & Review, 21(4), 861–883. Logie, R. H. (2016). Retiring the central executive. Quarterly Journal of Experimental Psychology, 69, 2093–2109. Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory for features and conjunctions. Nature, 390(6657), 279–281. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63 (2), 81. Moray, N. (1959). Attention in dichotic listening: Affective cues and the influence of instructions. Quarterly Journal of Experimental Psychology, 11(1), 56–60. Nairne, J. S. (2002). Remembering over the short-term: The case against the standard model. Annual Review of Psychology, 53(1), 53–81. Norrman, G., & Bylund, E. (2016). The irreversibility of sensitive period effects in language development: Evidence from second language acquisition in international adoptees. Developmental Science, 19(3), 513–520. Ouellette, G. P. (2006). What’s meaning got to do with it: The role of vocabulary in word reading and reading comprehension. Journal of Educational Psychology, 98(3), 554.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
95
96
ERYN J. ADAMS ET AL.
Perfetti, C. (2007). Reading ability: Lexical quality to comprehension. Scientific Studies of Reading, 11(4), 357–383. Pitt, M. A., & Samuel, A. G. (2006). Word length and lexical activation: Longer is better. Journal of Experimental Psychology: Human Perception and Performance, 32(5), 1120. Qian, D. (1999). Assessing the roles of depth and breadth of vocabulary knowledge in reading comprehension. Canadian Modern Language Review, 56(2), 282–308. Rhodes, S., & Cowan, N. (2018). Attention in working memory: Attention is needed but it yearns to be free. Annals of the New York Academy of Science, 1424, 52–63. Richardson, J. T. E., & Baddeley, A. D. (1975). The effect of articulatory suppression in free recall. Journal of Verbal Learning and Verbal Behavior, 14(6), 623–629. Ricker, T. J., & Cowan, N. (2014). Differences between presentation methods in working memory procedures: A matter of working memory consolidation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 40(2), 417. Sachs, J. S. (1967). Recognition memory for syntactic and semantic aspects of connected discourse. Perception & Psychophysics, 2, 437–442. Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8month-old infants. Science, 274(5294), 1926–1928. Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84, 1–66. Sokolov, E. N. (1963). Perception and the conditioned reflex. Pergamon Press. Soto, D., & Humphreys, G. W. (2008). Stressing the mind: The effect of cognitive load and articulatory suppression on attentional guidance from working memory. Perception & Psychophysics, 70(5), 924–934. Swets, B., Desmet, T., Hambrick, D. Z., & Ferreira, F. (2007). The role of working memory in syntactic ambiguity resolution: A psychometric approach. Journal of Experimental Psychology: General, 136(1), 64. Thalmann, M., Souza, A. S., & Oberauer, K. (2019). How does chunking help working memory? Journal of Experimental Psychology: Learning, Memory, and Cognition, 45(1), 37. Thomason, M. E., Race, E., Burrows, B., Whitfield-Gabrieli, S., Glover, G. H., & Gabrieli, J. D. (2009). Development of spatial and verbal working memory capacity in the human brain. Journal of Cognitive Neuroscience, 21(2), 316–332. Thompson, S. P., & Newport, E. L. (2007). Statistical learning of syntax: The role of transitional probability. Language Learning and Development, 3(1), 1–42. Underwood, B. J. (1957). Interference and forgetting. Psychological Review, 64 (1), 49. Van den Noort, M. W., Bosch, P., & Hugdahl, K. (2006). Foreign language proficiency and working memory capacity. European Psychologist, 11(4), 289–296.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
The Embedded-Processes Model
Vandierendonck, A. (2016). A working memory system with distributed executive control. Perspectives on Psychological Science, 11, 74–100. Vogel, E. K., McCollough, A. W., & Machizawa, M. G. (2005). Neural measures reveal individual differences in controlling access to working memory. Nature, 438(7067), 500–503. Wood, N., & Cowan, N. (1995a). The cocktail party phenomenon revisited: Attention and memory in the classic selective listening procedure of Cherry (1953). Journal of Experimental Psychology: General, 124, 243–262. Wood, N., & Cowan, N. (1995b). The cocktail party phenomenon revisited: How frequent are attention shifts to one’s name in an irrelevant auditory channel? Journal of Experimental Psychology: Learning, Memory, & Cognition, 21, 255–260. Wood, N. L., Stadler, M. A., & Cowan, N. (1997). Is there implicit memory without attention? A re-examination of task demands in Eich’s (1984) procedure. Memory & Cognition, 25, 772–779. Yang, Y., Chen, M., He, W., & Merrill, E. C. (2020). The role of working memory in implicit memory: A developmental perspective. Cognitive Development, 55, 100929.
https://doi.org/10.1017/9781108955638.008 Published online by Cambridge University Press
97
6 Long-Term Working Memory and Language Comprehension R. Lane Adams and Peter F. Delaney 6.1
Introduction
Comprehending fluent speech or written discourse requires us to convert a perceptual signal into a meaningful representation in memory. Furthermore, retention of some types of information seems to last a very long time, such as information about who did what in a story, while other information like surface information often fades away quickly (e.g., Kintsch, 1988; Sachs, 1967). Few deny that transient working memory, which is a system that actively maintains information during ongoing processing, plays a role in some aspects of language comprehension. However, the remarkable retention of situation model information after reading suggests that special mechanisms for creating long-term memory (LTM) for connected discourse are also necessary for language comprehension. In the mid-1990s, the existing theories of working memory and LTM seemed insufficient to explain how situation models could be rapidly and reliably encoded into LTM. The notion that working memory might involve using attention to maintain information during processing was not yet fully developed. Instead, many researchers relied on older theories of memory that distinguished short-term and long-term memory, with short-term memory serving the role. Short-term memory (STM) was viewed as too short-lived and LTM as too unreliable to allow fluent reading (Ericsson & Kintsch, 1995). Moreover, experts in various domains (e.g., chess, medicine) showed superior LTM for materials in their area of expertise. Ericsson and Kintsch (1995) proposed that these two sets of phenomena could be explained by postulating the development of longterm working memory (LT-WM) through practice. LT-WM theory proposes that experts’ acquired memory abilities for meaningful information in their specialty are similar in kind to the acquired memory abilities found in language comprehension, and are mediated by acquired skills and knowledge (Ericsson & Delaney, 1998, 1999; Ericsson & Kintsch, 1995).
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
LT-WM enables experts to rapidly and reliably store information into LTM, provided they have enough practice. The LT-WM theory is now 25 years old. Publications in the field of language comprehension since that time have not only expanded upon Ericsson and Kintsch’s (1995) framework for LT-WM (e.g., Caplan & Waters, 2013), but have also identified some of the shortcomings of the theory (e.g., Foroughi et al., 2015). A competing theory for expertise that utilizes structures called templates exists, but it does not have a developed theory of language comprehension (Gobet & Simon, 1996b). In the present chapter, we will begin by reviewing the original LT-WM theory and the reasons it took the form it did. Then, we will review some of the recent advances that may lead to more contemporary theories of LT-WM and call for more extensive investigation of LT-WM in language.
6.2
Understanding Working Memory and Language Theories prior to and after LT-WM
In early cognitive psychology, STM was assessed by tasks like the digit span task, and was assumed to place limits on complex processing. For example, Newell and Simon (1972) showed that a limited STM was sufficient to support the memory processes necessary to solve simple puzzles. However, by the time Ericsson and Kintsch’s (1995) paper appeared, it was clear that simple span tasks that lack a processing component (e.g., digit span) tend to predict comprehension weakly at best (Perfetti & Lesgold, 1977), and consequently attention shifted from STM to working memory (for a brief review, see Neath et al., 2003). Working memory capacity (WMC) measures the ability to maintain information in the face of ongoing processing (e.g., Baddeley, 1986). The first complex span task was developed by Daneman and Carpenter (1980); their reading span task required people to read sentences interspersed with questions about the sentences, and then to recall the final word from each sentence. WMC was operationalized as the number of words correctly recalled. WMC differences were soon attributed to differences in controlled attention and the activity of the prefrontal cortex (e.g., Conway & Engle, 1994; Engle et al., 1999; Kane & Engle, 2003). Complex span strongly predicts individual differences in comprehension accuracy and question answering (e.g., Daneman & Carpenter, 1980, 1983; Daneman & Merikle, 1996; Engle et al., 1992). MacDonald et al. (1992) further found that WMC seemed to limit syntactic processing, too. Their argument was that people with low WMC have reduced resources available, limiting comprehension. The greater a sentence’s syntactic ambiguity, the more WMC was required to understand the sentence. Depending on the placement of the syntactic elements that constitute a sentence, a sentence can be made more difficult to understand. Changes in syntactic structural complexity were primarily measured using the time spent processing
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
99
100
R. LANE ADAMS AND PETER F. DELANEY
(King & Just, 1991), and syntactically ambiguous parts of sentences were argued to show longer reading times for people with low complex span scores than high complex span scores (MacDonald et al., 1992). In sum, the results in the mid-1990s seemed broadly consistent with a single, unitary WMC being involved in multiple aspects of language comprehension, including both syntax and discourse comprehension. Since that time, others have proposed theories postulating greater integration between WM and LTM (e.g., Baddeley, 2000; Cowan, 2008). This suggests that future work should revisit the relationship between LT-WM and WMC theories to see if they are truly at odds. In the next few sections, we will see that other challenges have emerged to the consensus view of the mid-1990s, and a competing view that suggested that long-term working memory is involved in sentence processing began to develop.
6.3
Development of the LT-WM Theory
The developments in working memory theory informed the development of LT-WM theory. Like those theories, it proposed that the ability to process effectively in experts, whether readers or chess masters, involved memory storage during concurrent processing (see Delaney, 2018). Unlike the theories concerned with WMC and attentional control, however, LT-WM proposed that experts develop skills to rapidly and reliably encode information into LTM (Ericsson & Kintsch, 1995). Before we describe the proposed mechanisms, we will trace how the theory evolved from earlier conceptions based on chunking and skilled memory expertise.
6.3.1 A Chunking Theory of Expanded Memory in Experts Early cognitive theories assumed that storage in LTM was slow and unreliable, but skilled chess players reconstructed briefly viewed chess positions too accurately and quickly for LTM, and recalled too many pieces for STM (e.g., Chase & Simon, 1973; DeGroot, 1965). Chase and Simon showed chess positions to three chess players with varying levels of expertise (beginner, class A, and master level). Players saw a chess position for 5 s, and then had to reproduce as much of it as possible. The process was repeated for several trials on each board, with the players adding new pieces each time. On chess positions that could have been produced in a real game, better players recalled more pieces. However, when presented with chess positions that could not be created in a real game (as they did not follow the rules of chess), skill differences evaporate. If anything, the master-level player performed the worst after several looks at the board, indicating superior chess memory was restricted to positions that followed the rules of chess (but for evidence of small advantages even on random positions, see Gobet & Simon, 1996a).
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
Chase and Simon surmised that chess masters use LTM to augment their STM capacity by capitalizing on their existing knowledge of chess, proposing a chunking mechanism whereby patterns that frequently appeared in chess games would be treated as a single perceptual unit called a “chunk.” The idea was first proposed by George Miller (1956) in his famous paper on the capacity of STM, which argued that STM holds about 72 items. He described both the process of grouping the input into familiar units called chunks, and the extensive efforts required to form chunks. Chunks were composed of information that is meaningful, so the items could be related together. The amount of information in each chunk could be expanded, as illustrated in their example about telegraph operators: “[A] man just beginning to learn radio telegraphic code hears each dit and dah as a separate chunk. Soon he is able to organize these sounds into letters and then he can deal with the letters as chunks. Then the letters organize themselves as words, which are still larger chunks, and he begins to hear whole phrases” (p. 93). Chase and Simon (1973) proposed that chess masters gain extensive experience with recurring patterns that form a familiar associative relationship, and hence learn chunks such as offensive or defensive positions or arrangements of pieces that could occur in famous games. Their theory explicitly assumed that for experts there is no direct storage of the memory trace in LTM; the board was coded into familiar perceptual chunks from LTM, which were stored in STM.
6.3.2 Skilled Memory Theory: Direct Storage in LTM Charness (1976) provided evidence against the chunking theory, showing that Class A players’ ability to recall briefly presented chess positions was still high in spite of a filled delay. Class A and C players briefly viewed a meaningful chess position and then had to reconstruct the board after a 30 s interruption that varied in processing difficulty and content. Although memory for the position was slightly reduced by a filled delay, had cues been primarily stored in STM, then the expected decrement in experts’ memory performance would have been much higher than the observed 10 percent. If people were storing chunks in STM, we would expect information that is stored in STM to be subject to catastrophic disruption from filled intervals of 30 s (cf. Brown 1958; Peterson & Peterson, 1959); indeed, this was true for trigrams. The results therefore suggested that skilled chess players store chess positions directly into LTM (see also Gobet & Simon, 1996b). Unlike chunking theory, skilled memory theory posited that we remember information not as perceptually available patterns, but rather as meaningful representations that are generated during processing of sentences (or boards). In support of this view, Chase and Ericsson (1982) described an unpublished study by Ericsson and Karat in which participants learned either a sentence or a list of the words from a sentence in a scrambled
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
101
102
R. LANE ADAMS AND PETER F. DELANEY
order. Performance was twice as good for the sentences (14 words) than for the scrambled text (6–7 words). Furthermore, they remembered many of the sentences in a postsession recall test, but almost none of the scrambled sentences. While measures suggesting greater language skill predicted higher recall of the sentences, language skill was unrelated to recall for scrambled sentences, consistent with the view that language skill is not different in kind from expertise in chess. In both instances, when a mental model cannot be produced, a person’s skills are no longer relevant and cannot aid later memory. Furthermore, the incorrect sentences recalled by participants were almost always semantically correct, suggesting that it is not just familiar perceptual chunks of words that enable people to remember the sentences. Instead, people must form some meaning-based structure to do the task, and later reconstruct the surface form from it. Such findings are problematic theories that suggest perceptual chunks encode language, rather than transforming the information into knowledge during comprehension. If people have extensive knowledge, they can generate associations that encode a pattern. For example, an avid runner might convert the number 4596 into a racing time of 4:59.6 and memorize it as Diane Leather, the first woman to run a sub-5-minute mile. These retrieval cues then can be regenerated later to distinguish groups of similar items (in this case, digits) from other items in memory. While even novices can recognize some patterns (e.g., 2000 is a year), to encode arbitrary patterns requires extensive practice to reliably encode a large percentage of the incoming information with a pattern and recall the anomalies where no pattern was available. Beyond simple retrieval cues, Chase and Ericsson argued that experts use retrieval structures to maintain the order of studied information. A retrieval structure is an LTM structure that organizes material in LTM using a set of retrieval cues that can be regenerated at the time of the test. Retrieval structures guarantee a cue to each item, so they need not rely on recency information alone to retrieve items using the retrieval structure; they have another, more reliable cue available. The classic example of retrieval structures came from experiments on expanding digit span with practice: over several hundred hours of practice, two participants (SF and DD) grew their digit spans from 7 to 84 and 104 (Chase & Ericsson, 1982; Ericsson & Staszewski, 1988). The results were surprising given that memory span appeared on IQ tests and was part of the evidence used by Miller (1956) to establish the size of STM. Verbal protocol data and interviews showed that they created retrieval structures that specified small groups of 3–4 digits that they would try to generate a pattern for. These digit groups were then assigned into an overarching “supergroup” until it grew too large to be reliably recalled, at which point a new “level” to the structure was added. By setting up the plan for encoding in advance, SF and DD could reliably regenerate the structure and use it to cue the groups of digits and their associated patterns.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
6.3.3 Transition to LT-WM and Its Mechanisms Skilled memory theory was primarily applied to explain amazing storage feats like memorizing long lists of digits. It was a skill-based account of how LTM is used to perform tasks that novices use STM and rehearsal for, like digit span. Just as working memory theory shifted attention from pure storage capacity to the ability to store information while engaged in processing, LT-WM shifted from an account of how pure storage capacity was expanded through skills to an account of how experts learned to use skills to store necessary information in LTM while using that information for processing. LT-WM is not a new storage system; it is just LTM, but used in a skilled way to support the tasks experts need to do. Like skilled memory theory, LT-WM proposes that a well-practiced, learned set of skills allows experts (including skilled readers) to use LTM for their WM needs, rather than having to rely exclusively on short-term working memory and controlled attention. At the time, most studies of LTM employed unrelated materials like word lists, which are difficult to store in LTM without extensive study and effort (e.g., Atkinson & Shiffrin, 1968). The material experts deal with is instead familiar, so they: (1) learn how to pick out what is important and reorganize it, (2) associate information together using their preexisting knowledge, (3) have practiced retrieval plans for getting the information back later, and (4) develop mechanisms for reducing interference between similar representations. 6.3.3.1 Picking Out the Important from the Unimportant Identifying the “important” in both reading and thinking happens automatically after extensive practice, but only when the material fits the structures developed naturally to support expert reading and thinking. For example, expert-level medical knowledge facilitates memory for text about symptoms, but only when the text can be processed using the skills typically used by doctors during diagnosis. Coughlin and Patel (1987) gave physicians and medical students a textual description of two patients’ symptoms in either the expected order or a randomly scrambled order. In one case (arteritis), the details in the text included laboratory findings that were highly associated with the correct diagnosis. For this case, physicians performed much better than students regardless of whether the text was scrambled. They recalled dramatically more of the relevant facts and did not alter the symptoms or draw unnecessary inferences as often as the students. Furthermore, when asked for a diagnosis, the physicians overwhelmingly gave the correct diagnosis, while none of the students did. For the other, much harder case (endocarditis tied to IV drug use), the diagnosis required reconstructing the order of the symptoms and inferring that a self-reported “cat bite” on the arm was in fact an injection point. The physicians were correct or partially correct almost all the time, while the students almost never were. However, randomly ordering the text reduced physician accuracy, since they were less able to associate the symptoms with the correct diagnosis when scrambled.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
103
104
R. LANE ADAMS AND PETER F. DELANEY
Their memory for the symptoms was no longer statistically higher than for students (though numerically they still recalled six more relevant facts). The students similarly showed poorer diagnostic accuracy with the scrambled version of the case, but overall, their performance was near floor: none was right or partially right in the scrambled case, and only two of them were even partially right in the typical order. Taken together, the results show that the physicians were much better than students at identifying the relevant information and recalling it, provided that temporal order was not necessary to make the diagnosis. Furthermore, the data support that correct memory for the symptoms was partially mediated by the supporting diagnosis, suggesting that the symptoms were recalled not just by recognizing patterns, but by organizing them in a way that supported generating the diagnosis. In other words, they anticipated what information would be needed later, and organized it using existing knowledge (but see Eva et al., 2002, for a failure to replicate using smaller expertise differences and cases deliberately containing features from multiple diagnoses).
6.3.3.2
Associating Information During Comprehension via Online Inferences Consider how readers must draw on their general world knowledge in order to understand these sentences (from Johnson-Laird 1983): (1) (2) (3)
Roland’s wife died in 1928. He married again in 1940. His wife now lives in Spain.
Unless we are reading a fantasy story, good readers will know that “his wife” in (3) must refer to Roland’s second wife, not the first, since dead wives generally do not live anywhere. The inference that connects these statements removes any inconsistency. There is no way in advance to recognize this information as a “chunk” or template; we must use world knowledge and reasoning to understand them.
6.3.3.3 Using Retrieval Structures Recall that experts often had practiced retrieval structures for getting information back from memory. In comprehension, there are natural retrieval structures involved in constructing situation models. For example, data supporting the event-indexing model (e.g., Pettijohn & Radvansky, 2016; Zwaan et al., 1995; Zwaan, Langston, & Graesser, 1995; Zwaan & Radvansky, 1998) suggest that readers attend to five kinds of event indices: time, place, protagonist/actor, intention/goal, and causality. Keeping track of these indices motivates inferences and guides the formation of a situation model that can be used later to answer important questions. Because we habitually are tested on these types of information, we tend to extract them when we read. That said, these goals are pragmatic, and readers are flexible and can
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
alter their strategies to focus more on other language factors when necessary. For example, Zwaan (1994) found that when people were told that the same text was a newspaper story or a narrative tale, they encoded more situation model details as a newspaper story and more details about the language and style as a narrative tale. Such results strongly suggest that expert readers anticipate the way they will be tested and alter their strategies for encoding accordingly.
6.3.3.4 Managing Interference Finally, Ericsson and Kintsch (1995) proposed that experts who must reuse similar retrieval structures have methods to manage proactive and retroactive interference in LT-WM. Meaningful associations can play this role. For example, when readers construct situation models, they contain associations that distinguish one situation from another. If one just memorized a list of times, places, or protagonist names without these connections, they would interfere with one another and cause dramatic forgetting. When integrated, however, the retrieval cues enable unique discrimination of the individual memories, reducing or even reversing interference. The same phenomenon has been observed in the fan effect, where unrelated sentences that share an element usually interfere with one another, but when the sentences can be meaningfully integrated, the effect is often smaller or absent (for a review, see Myers et al., 1984; but cf. Reder & Anderson, 1980 for evidence that plausibility is necessary for this to occur). Under the right circumstance, interference can even be replaced with facilitation. Consider a study by Jacoby and Walheim (2013) in which participants studied paired associates with competing targets. For example, participants studied a pair like knee-bone, and then later they had to learn knee-bend. The cue remained the same, but the target switched from bone to bend. Participants who were able to detect the change showed improved recall for both targets, while individuals who were unable to detect the change recalled them much less often. Thus, people can eliminate or even reverse interference by creating meaningful associations that enable them to distinguish individual responses that would otherwise compete. Furthermore, some associative encoding mechanisms emerge rather quickly with practice in a task that has considerable proactive interference (Wahlheim & Jacoby, 2013), suggesting that with experience, experts could become quite good at reducing interference using associations. Notably, though, just generating semantic mediators – words that semantically relate the cue and the target (cf. Crutcher & Ericsson, 2000) – is not sufficient to improve free recall of unrelated word pairs, and can even hurt cued recall (Lehman & Karpicke, 2016). It is not just having a mediator that helps memory, but rather generating mediators that can be regenerated later. Knowing how to select such mediators surely requires practice. Another mechanism that helps control interference is memory for temporal context, which enables people to retrieve the most recently
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
105
106
R. LANE ADAMS AND PETER F. DELANEY
encountered representation (cf. Klein et al., 2007; Sahakyan et al., 2013). If asked to reinstate earlier representations, a few cues are usually all that are necessary to access previously encoded representations from the recent past (e.g., Ericsson et al., 2004). For example, the digit memorist Rajan (Ericsson et al., 2004) is able to reproduce not only the most recent list of digits, but lists from earlier days when given the first few digits from the sequence.
6.3.3.5 Conclusions LT-WM does not involve a new kind of memory; it is LTM. What differs between LT-WM and ordinary LTM is the degree of skill with which experts (including readers) are able to anticipate future memory needs, form associations, and retrieve those associations later. Many experts are great at encoding familiar information and new information presented in a familiar way, but unusual or oddly structured material forces them to fall back on general methods that are used in unfamiliar tasks. The limitations of an expert’s coding ability are illustrated by the memorist Rajan’s memory limitations; his superior encoding ability was dependent on the material, such that learning a restricted set of numbers harmed his ability to encode patterns (see Ericsson et al., 2004).
6.4
LT-WM and Language: Memory in Discourse Comprehension
To apply the LT-WM theory to discourse comprehension, Ericsson and Kintsch (1995) extended the well-known construction-integration model (Kintsch 1988, 1992a, 1992b, 1994a, 1994b; Kintsch & Welsch, 1991) to include LT-WM. The model built on earlier ideas put forth by Bever (1970) to understand sentence comprehension. Bever suggested that well-learned heuristic strategies are developed through practice to solve common problems in sentence comprehension. In a similar fashion, LT-WM and the construction-integration model assume that strategies are applied to creating connected memory representations from coherent discourse. Comprehending text takes place in two phases – construction and integration – to allow for the construction of meaningful representations at three levels of representation: the surface structure, the text base, and the situation model. The earliest level is the surface structure (or the linguistic level), which consists of the lexical information explicitly presented in the text. Next, the reader constructs a textbase, which consists of a set of propositions that link nouns and adjectives together via a verb (e.g., IS [TALL, GEORGE]). These propositions are formed from the surface structure, and do not need to be coherent or even consistent. Since the construction phase is focused on using individual propositions to create a coherent representation in the text, it is assumed that the representations are full of redundancies and include irrelevant information.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
During the integration phase, the comprehender builds a network of interconnected propositions, integrating these new elements with their knowledge and experience to generate situation models. These situation models contain information about the situation described in the text. As each element is generated, it is integrated within the contextual information from the previous elements before being stored in LTM (the integration phase). Propositions that are no longer consistent with the emerging structure are dropped, eventually resulting in a coherent situation model. As a reader progresses through the text, new elements are connected to sentences and information that has already been stored in LTM. As we noted earlier, this integration process requires drawing inferences that go beyond what is given in the text. Inference processes during discourse comprehension are often inferred from reading times, priming of particular words, or the fact that people succeeded in understanding text (for a review, see Singer, 1990). However, several studies used think-aloud protocols to directly observe several types of complex inferences (Fletcher, 1986; Trabasso & Suh, 1993). Thinking aloud does not change the reading process, and verbalizations directly reflect the thoughts that are most accessible to the reader (for reviews, see Delaney et al., 2018; Ericsson & Simon, 1993). The situation model needs to be dynamic because new information can change the meaning. A reader not only needs to access the sentence they are currently reading but also needs to be able to alter representations in a text that have been viewed previously. This requires that a person can both detect a difference in information and be able to update the representation by integrating the new information into the previous understanding of the sentence. For example, when the identity of the killer is revealed in a murder mystery, all previous representations of the character have to be updated. To detect that something is inconsistent with what has been previously read, there must be a way to access the previous information. Ericsson and Kintsch believed that the updatable parts of the LTM structure serve as extended working memory and to integrate information, the reader would have to have access to distinctive cues within LT-WM. Thus, the construction-integration model proposes that people create multilevel representations that can later be drawn upon to recall information about the situations described in the text. Because the situation model contains meaningful associations between the parts of the discourse, people can use them to retrieve knowledge they need at a later point, or to integrate new information obtained from the text. Ericsson and Kintsch (1995) argued that the representations necessary for syntactic processing during reading usually resulted in a brief memory trace of the surface structure and a longer-lasting trace of the text base and situation model (e.g., Kintsch et al., 1990). Whereas the meaning of a text can usually be retained better than surface memory, retaining the surface form occurs in some cases. For example, the surface form is retained when something is pragmatically significant and relevant to the situation model. Ericsson and
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
107
108
R. LANE ADAMS AND PETER F. DELANEY
Kintsch suggested that the differences in retention, due to factors such as relevancy, of the (usually) short-lived surface and long-lasting deep structure representations are evidence of a multilevel representation of the text constructed in LTM. Discourse is often written in a way that surpasses the limitations of working memory. Readers therefore cannot hold every important piece of information in the focus of attention. This notion is similar to more recent work by McElree and Dyer (2013), who suggested that long-term memory processes could account for many of the apparent effects of WMC on language comprehension. Cues in the text must serve to reinstate previous information so that a passage is coherent. Languages are often structured so that the syntax provides clues to the reader about where bridging inferences must occur. For example, Givón (1995) identified elements of syntax that typically indicate how long ago a referent occurs. English usually uses a pronoun anaphor (e.g., he, she, or it) to indicate that the referent is still active, while it may use a more specific anaphor (e.g., “the sadistic killer”) for referents that occurred longer ago in the text. Readers that are unable to understand the syntax of a sentence (e.g., due to poor writing or sentence difficulty) spend more time trying to comprehend the sentence (Miller & Kintsch, 1980). Ericsson and Kintsch (1995) claimed this outcome is evidence that the reader must generate an appropriate retrieval cue to integrate information already in LT-WM, rather than knowing from the text. They divided processing during reading into the time when propositions were being constructed, which corresponds to decoding of syntax into the text base and situation model construction, and the time between propositions, when the information has already been extracted and encoded in LT-WM. They argued that when construction and integration were ongoing, interruptions would disrupt reading. However, interruptions between sentences – when such integration had already occurred – would have little effect on comprehension. Several studies demonstrated that comprehension was highly resistant to brief interruptions during reading that occurred between sentences (e.g., Fischer & Glanzer, 1986; Glanzer et al., 1981, 1984). Interruptions during reading still allowed people to continue reading, provided they could reinstate enough information to access their LT-WM structures. However, after an interruption task, the participants’ reading speed was slowed by approximately 350 ms, which coincided with the estimated retrieval time from LTM. Recently, Foroughi et al. (2015) questioned whether reading is as robust to interruptions as LT-WM theory claims. They found that long disruptions led to disrupted comprehension, suggesting that people were holding some information in short-term working memory. However, Delaney and Ericsson (2016) pointed out that there is rarely no forgetting from LTM after an interruption. Recall that Charness (1976) found a similar pattern in chess memory, where Class A players’ memory was slightly reduced by mentally taxing interruptions. Delaney and Ericsson (2018) argued that
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
what is surprising about Foroughi et al.’s (2015) data is not that there are some losses from interruptions, but rather how substantially intact comprehension remained despite the interruptions. Nonetheless, the data highlight the need for future investigations to clarify how much forgetting one should expect in LT-WM in order to resolve the competing understandings of how much forgetting should be predicted from lengthy interruptions.
6.5
Is There LT-WM for Sentence Structure to Support Syntactic Processing?
As noted previously, the framework for LT-WM was developed to explain discourse comprehension in language by using expertise frameworks, not sentence-level syntactic processing. Ericsson and Kintsch (1995) connected strategies that experts used to increase memory span such as the use of retrieval structures and generating associations that draw on world knowledge, and postulated similar structures played a role in comprehending discourse. They questioned whether conventional views of WMC were sufficient to explain the remarkably fast, durable storage in LTM of the text base and associated inferences generated during comprehension. Earlier work suggested that WMC underlies efficacy in processing syntactic structures. However, Caplan and Waters (1995a, 1995b; Waters et al., 1995) reevaluated the evidence for a role of general WMC in syntactic processing, and proposed instead that separate resources were available for online sentence interpretation (i.e., syntactic processing) and subsequent offline postinterpretive language processing efficacy (i.e., using the generated representations to answer questions). The original form of their separate resources theory holds that there are two separate resource pools used in encoding and postinterpretative processes. Caplan and Waters (1999) reviewed existing evidence that WMC predicted greater slowing when reading syntactically complex sentences at the point of difficulty. King and Just (1991, Exp. 1) compared reading times when reading object-relative clauses (e.g., The rat that the cat saw ran away.) and subject-relative clauses (e.g., The rat that saw the cat ran away.), and concluded there was a WMC by part of sentence interaction effect. However, they never presented statistics that supported their claim, and Caplan and Waters (1999) reviewed several studies that found that high-, medium-, and low-span participants all spent approximately equal listening time for object- versus subject-relative clauses (e.g., by Ferreira et al., 1996). Furthermore, MacDonald et al.’s (1992) critical finding that garden path sentences showed greater reading time slowing for low-span than high-span participants repeatedly failed to replicate. Recently, James et al. (2018) concluded that reliable individual differences in preinterpretive online syntactic processing are rarely observed, which would make it difficult for any individual differences measure to correlate with them. While total
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
109
110
R. LANE ADAMS AND PETER F. DELANEY
reading times for complex sentences are consistently longer than for less complex sentences, these patterns occur for everyone, and vary little as a function of who is doing the reading – consistent with Caplan and Waters’s argument that syntactic processing does not involve WMC during the online interpretation phase. However, once sentences are generated, to the degree that question answering required selective search of memory representations, WMC would be involved in question answering. Later work led Caplan and Waters (2013) to suggest that a form of LT-WM might be involved in syntactic processing. Ericsson and Kintsch’s (1995) original view of working memory relegated LT-WM to higher-order interpretation and other cognitively demanding operations. However, LT-WM need not be exclusive to higher-order cognition. Part of the original evidence that LT-WM existed in the first place was poor correlations between ST-WM tasks and domain-specific performance. Gordon et al. (2002) suggested that Ericsson and Kintsch’s (1995) characterization of LT-WM might also apply to sentence processing. Caplan and Waters similarly suggested that LT-WM could operate during the automatic processes of language comprehension, with acquired skills enabling readers to create knowledge-based associations and apply patterns and schemas (i.e., retrieval structures) to forge integrated memory representations. These representations are maintained in LTM, not in a transient working memory system. Caplan and Waters (2013) argued that key phenomena in online parsing mirror the LT-WM process. They proposed that parsing was a highly practiced skill, which explains why it is resistant to interruptions. Few studies examined whether interruption during parsing causes significant disruption of comprehension, but some evidence suggests it is surprisingly resistant (e.g., Wanner & Maratsos, 1978). In contrast to what happens during online parsing, once sentences are encoded in memory, unless they can be connected into discourse via the LT-WM mechanisms that are thought to produce situation models, they will be subject to the usual constraints of general WMC at retrieval. Consistent with this view, James et al. (2018) found that both general capacity and reading skill predicted comprehension of sentences, but not reading times during parsing. Individual differences in verbal working memory were recorded using three complex span measures (reading, listening, and operation span). Differences in language experience were assessed through several measures of vocabulary (Author Recognition Test, North American Adult Reading Test, etc.) By this view, more unusual sentence structures are harder to comprehend because we have less experience parsing them, but they are still processed using LT-WM. The mounting evidence from more recent publications suggests that uses for LT-WM exist outside of its original 1995 conceptualization. In addition to higher-order processes, LT-WM may also be implicated in reflexive/automatic functions such as syntactic processing. Future language research should continue exploring this possibility.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
6.6
The Neuroscience of LT-WM Development
Recent work investigated the neuroscience of the acquisition of LT-WM. Guida et al. (2012) examined data from 12 neuroimaging studies using trained novices and 8 using experts. As novices gained initial expertise, the brain regions employed changed little, although cerebral activity decreased with practice. Experts contrastingly showed greater activation in LTM-related regions of the brain, consistent with neural restructuring. Guida et al. theorized that acquiring expertise is a two-stage process; novices reduce cerebral activation with practice, eventually leading to direct storage in LTM once recognizable patterns are established. Activation within the fusiform gyrus may provide evidence that a person has developed LT-WM. The fusiform gyrus is thought to be involved in configural processing and recognition of faces, and is also implicated in other holistic, expert-level recognition, such as identifying cars (e.g., McGugin et al., 2012). Gauthier et al. (1998) found that extensive practice distinguishing artificially created objects produced signs of developing configural processing, suggesting fusiform gyrus involvement. Bilalicˊ (2018) reviewed evidence from their lab that examined brain activity in chess experts and novices. When tracking eye movement in a chess pattern recognition task, experts had fewer eye movements and required less time to complete the task in comparison to novices. Moreover, instead of using the same mental operations as a novice at the higher level, experts use a different level of brain activation. Both novices and experts activate the posterior middle temporal gyrus (pMTG) and supramarginal gyrus (SMG) when recognizing patterns but only experts have activation in both hemispheres. He termed the phenomenon of experts engaging areas in both hemispheres the double take of expertise (Bilalicˊ , 2017). These differences were drastically reduced when presented with a board that did not follow the familiar rules of chess. Further investigation into the causes of the double take of expertise may lead to greater understanding of how the brain adapts developing specialized knowledge or skills. Notably, Guida et al.’s (2012) experts each performed tasks where recognizable patterns could be easily established (chess, abacus calculation, and mental multiplication), and the studies reviewed above similarly dealt with perceptual and reasoning expertise. None of these tasks resembles expert reading. Examining the nascent cognitive neuroscience of discourse may provide new constraints on LT-WM theories of language. Recent studies suggest that words that are syntactically anomalous and words that violate world knowledge a sentence both elicit N400 responses, which are associated with semantic surprises, in about the same amount of time (e.g., Hagoort et al., 2004), as do words that are semantically sensible within a sentence but anomalous with respect to earlier-read information (e.g., van
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
111
112
R. LANE ADAMS AND PETER F. DELANEY
Berkum et al., 2003). The N400 is not driven by a single brain location, but has been associated with a broader system of regions tied to semantic memory (for a review of the N400, see Kutas & Federmeier, 2011). These data are consistent with the construction of the text base being sensitive to existing LT-WM information, semantic information, and world knowledge, and all at about the same speed. In sum, LT-WM access is sufficiently fast to act as efficiently as semantic knowledge when expectations are violated. Later updating of situation models sometimes involves more specific brain regions such as the precuneus and posterior cingulate in the medial parietal cortex. These regions are more active in fMRI studies of comprehension at the start of a narrative (when the situation model is being established) and at event boundaries such as changes of time or place, when the model needs to be updated (e.g., Speer et al., 2007, 2009; Whitney et al., 2009). These regions are often implicated in complex tasks that require integration across many brain areas and consciousness (e.g., Cavanna & Trimble, 2006), suggesting that LT-WM formation may also involve distributed brain processes. Furthermore, many of the studies observed bilateral activation consistent with a neural “double take” for situation model updating (e.g., Speer et al., 2007, 2009). A meta-analysis of the differences between coherent and incoherent text – equivalent to real and random positions in chess – revealed activation on both sides of the anterior temporal lobe and temporoparietal junction, suggesting that regions involved in integrating semantic information might be bilaterally activated during reading (Ferstl et al., 2008). Finally, regions involved in relevant types of perceptual processing were also activated during situation model updating. For example, Speer et al. (2009) found that regions involved in motor control were activated when characters interacted with objects, suggesting that LT-WM skills in reading may invoke mental perceptual simulation of the world (cf. Zwaan, 2004, for a related theory and review of perceptual theories of comprehension). In sum, more work is needed on the neuroscience of LT-WM in reading. The brain regions involved for perceptually oriented LT-WM (e.g., chess) and language-oriented LT-WM seem different. However, the skills involved in LT-WM likely integrate many brain areas to facilitate experts’ efficient processing both during reading and playing chess. Moreover, key brain regions may be recruited bilaterally to support LT-WM pattern generation and recognition.
6.7
Conclusion
This review highlights theoretical advancements in the fields of expertise and language that led to LT-WM theory. Originally developed to explain the superior WMC of experts and readers, the applications of LT-WM have greatly expanded since 1995. In language, LT-WM was originally designed
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
to account for people’s rapidly formed long-term memory during discourse comprehension, and proposed that people developed skills to meet these memory demands, including identifying relevant information, using retrieval structures, rapidly forming associations (e.g., via inferences), and capitalizing on associative memory to reduce interference. Newer evidence supports a possible role for LT-WM in syntactic parsing as well; specifically, WMC does not always limit syntactic processing, and parsing may be resistant to interruptions. Physical evidence for the usage of LT-WM has been shown in the broad recruitment of multiple brain regions by experts to process sensory and semantic information. Moving forward, we hope that new research will examine how both LT-WM and WMC support discourse and syntactic processing, both online and from memory.
References Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. Psychology of Learning and Motivation, 2(4), 89–195. Baddeley, A. D. (1986). Working Memory. Oxford University Press. Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417–423. Bever, T. G. (1970). The cognitive basis for linguistic structures. Cognition and the Development of Language, 279(362), 1–61. Bilalicˊ , M. (2017). The Neuroscience of Expertise. Cambridge University Press. Bilalicˊ , M. (2018). The double take of expertise: Neural expansion is associated with outstanding performance. Current Directions in Psychological Science, 27(6), 462–469. Brown, J. (1958). Some tests of the decay theory of immediate memory. Quarterly Journal of Experimental Psychology, 10(1), 12–21. Campitelli, G. (2015). Memory behavior requires knowledge structures, not memory stores. Frontiers in Psychology, 6, 1696. Caplan, D., & Waters, G. S. (1995a). Aphasic disorders of syntactic comprehension and working memory capacity. Cognitive Neuropsychology, 12(6), 637–649. Caplan, D., & Waters, G. S. (1995b). On the nature of the phonological output planning processes involved in verbal rehearsal: Evidence from aphasia. Brain and Language, 48(2), 191–220. Caplan, D., & Waters, G. S. (1999). Verbal working memory and sentence comprehension. Behavioral and Brain Sciences, 22(1), 77–94. Caplan, D., & Waters, G. (2013). Memory mechanisms supporting syntactic comprehension. Psychonomic Bulletin & Review, 20(2), 243–268. Cavanna, A. E., & Trimble, M. R. (2006). The precuneus: A review of its functional anatomy and behavioural correlates. Brain, 129(3), 564–583.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
113
114
R. LANE ADAMS AND PETER F. DELANEY
Charness, N. (1976). Memory for chess positions: Resistance to interference. Journal of Experimental Psychology: Human Learning and Memory, 2(6), 641–653. Chase, W. G., & Ericsson, K. A. (1982). Skill and working memory. Psychology of Learning and Motivation, 16, 1–58. Chase. W. G., & Simon. H. A. (1973). Perception in chess. Cognitive Psychology, 4(1), 55–81. Conway, A. R. A., & Engle, R. W. (1994). Working memory and retrieval: A resource-dependent inhibition model. Journal of Experimental Psychology: General, 123(4), 354–373. Coughlin, L. D., & Patel, V. L. (1987). Processing of critical information by physicians and medical students. Journal of Medical Education, 62(10), 818–828. Cowan, N. (2008). What are the differences between long-term, short-term, and working memory? Progress in Brain Research, 169, 323–338. Crutcher, R. J., & Ericsson, K. A. (2000). The role of mediators in memory retrieval as a function of practice: Controlled mediation to direct access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(5), 1297. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19(4), 450–466. Daneman, M., & Carpenter, P. A. (1983). Individual differences in integrating information between and within sentences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9(4), 561–584. Daneman, M., & Merikle, P. M. (1996). Working memory and language comprehension: A meta-analysis. Psychonomic Bulletin & Review, 3(4), 422–433. DeGroot. A. (1965). Thought and choice in chess. Mouton. Delaney, P. F. (2018). The role of long-term working memory and template theory in contemporary expertise research. Journal of Expertise, 1(3), 155–161. Delaney, P. F., & Ericsson, K. A. (2016). Long-term working memory and transient storage in reading comprehension: What is the evidence? Comment on Foroughi, Werner, Barragán, and Boehm-Davis (2015). Journal of Experimental Psychology: General, 145(10), 1406–1409. Delaney, P. F., Wallander, R., & Preheim, G. A. (2018). Protocol analysis. In D. S. Dunn (Ed.), Oxford bibliographies online: Psychology. Oxford University Press. Dixon, P., & Sharma, A. (2019). Distraction and temporal order in narrative situation models. Discourse Processes, 56(5–6), 402–414. Engle, R. W., Cantor, J., & Carullo, J. J. (1992). Individual differences in working memory and comprehension: A test of four hypotheses. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18(5), 972–992. Engle, R. W., Kane, M. J., & Tuholski, S. W. (1999). Individual differences in working memory capacity and what they tell us about controlled attention, general fluid intelligence and functions of the prefrontal cortex.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 102–134). Cambridge University Press. Ericsson, K. A., & Delaney, P. F. (1998). Working memory and expert performance. In R. Logie, & K. J. Gilhooly (Eds.), Working memory and thinking (pp. 93 114). Erlbaum. Ericsson, K. A., & Delaney, P. F. (1999). Long-term working memory as an alternative to capacity models of working memory in everyday skilled performance. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 257–297). Cambridge University Press. Ericsson, K. A., Delaney, P. F., Weaver, G. E., & Mahadevan, S. (2004). Uncovering the structure of a memorist’s superior “basic” memory capacity. Cognitive Psychology, 49(3), 191–237. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102(2), 211–245. Ericsson, K. A., & Simon, H. A. (1993). Protocol analysis: Verbal reports as data (Rev. ed.). MIT Press. Ericsson, K. A., & Staszewski, J. J. (1988). Skilled memory and expertise: Mechanisms of exceptional performance. In D. Klahr & K. J. Kotovsky (Eds.), Complex information processing: The impact of Herbert A. Simon (pp. 235–267). Erlbaum. Estevez, A., & Calvo, M. G. (2000). Working memory capacity and the time course of predictive inferences. Memory, 8(1), 51–61. Eva, K. W., Norman, G. R., Neville, A. J., Wood, T. J., Brooks, L. R. (2002). Expert-novice differences in memory: A reformulation. Teaching and Learning in Medicine, 14(4), 257–263. Ferreira, F., Henderson, J. M., Anes, M. D., Weeks, P. A., & McFarlane, D. K. (1996). Effects of lexical frequency and syntactic complexity in spokenlanguage comprehension: Evidence from the auditory moving-window technique. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22(2), 324–335. Ferstl, E. C., Neumann, J., Bogler, C., & von Cramon, D. Y. (2008). The extended language network: A meta-analysis of neuroimaging studies on text comprehension. Human Brain Mapping, 29(5), 581–593. Fischer, B., & Glanzer, M. (1986). Short-term storage and the processing of cohesion during reading. Quarterly Journal of Experimental Psychology, 38A (3), 431–460. Fletcher, C. R. (1981). Short-term memory processes in text comprehension. Journal of Verbal Learning and Verbal Behavior, 20(5), 264–274. Foroughi, C. K., Werner, N. E., Barragán, D., & Boehm-Davis, D. A. (2015). Interruptions disrupt reading comprehension. Journal of Experimental Psychology: General, 144(3), 704–709. Gauthier, I., Williams, P., Tarr, M. J., & Tanaka, J. (1998). Training “greeble” experts: A framework for studying expert object recognition processes. Vision Research, 38(15–16), 2401–2428.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
115
116
R. LANE ADAMS AND PETER F. DELANEY
Gernsbacher, M. A. (1990). Language comprehension as structure building. Erlbaum. Givón, T. (1995). Coherence in text vs. coherence in mind. In T. Givón & M. A. Gernsbacher (Eds.), Coherence in natural text (pp. 59–116). John Benjamins Publishing Company. Glanzer, M., Dorfman, D., & Kaplan, B. (1981). Short-term storage in the processing of text. Journal of Verbal Learning and Verbal Behavior, 20(6), 656–670. Glanzer, M., Fischer, B., & Dorfman, D. (1984). Short-term storage in reading. Journal of Verbal Learning and Verbal Behavior, 23(4), 467–486. Gobet, F., & Simon, H. A. (1996a). Recall of rapidly presented random chess positions as a function of skill. Psychonomic Bulletin & Review, 3(2), 159–163. Gobet, F., & Simon, H. A. (1996b). Templates in chess memory: A mechanism for recalling several boards. Cognitive Psychology, 31(1), 1–40. Gordon, P. C., Hendrick, R., & Levine, W. H. (2002). Memory-load interference in syntactic processing. Psychological Science, 13(5), 425–430. Guida, A., Gobet, F., Tardieu, H., & Nicolas, S. (2012). How chunks, longterm working memory and templates offer a cognitive explanation for neuroimaging data on expertise acquisition: A two-stage framework. Brain and Cognition, 79(3), 221–244. Guida, A., Gras, D., Noel, Y., Le Bohec, O., Quaireau, C., & Nicolas, S. (2013). The effect of long-term working memory through personalization applied to free recall: Uncurbing the primacy-effect enthusiasm. Memory & Cognition, 41(4), 571–587. Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004). Integration of word meaning and world knowledge in language comprehension. Science, 304(5669), 438–441. Jacoby, L. L., & Wahlheim, C. N. (2013). On the importance of looking back: The role of recursive remindings in recency judgments and cued recall. Memory & Cognition, 41(5), 625–637. James, A. N., Fraundorf, S. H., Lee, E. K., & Watson, D. G. (2018). Individual differences in syntactic processing: Is there evidence for reader-text interactions? Journal of Memory and Language, 102, 155–181. Johnson-Laird, P. N. (1983). Mental models. Erlbaum. Kane, M. J., & Engle, R. W. (2003). Working-memory capacity and the control of attention: The contributions of goal neglect, response competition, and task set to Stroop interference. Journal of Experimental Psychology: General, 132(1), 47–70. King, J., & Just, M. A. (1991). Individual differences in syntactic processing: The role of working memory. Journal of Memory and Language, 30(5), 580–602. Kintsch, W. (1988). The use of knowledge in discourse processing: A construction-integration model. Psychological Review, 95(2), 163–182
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction-integration model. Psychological Review, 95(2), 163–182. Kintsch, W. (1992a). A cognitive architecture for comprehension. In H. L. Pick, Jr., P. van den Broek, & D. Knill (Eds.), Cognition: Conceptual and methodological issues (pp. 143–164). American Psychological Association. Kintsch, W. (1992b). How readers construct situation models for stories: The role of syntactic cues and causal inferences. In A. F. Healy, S. M. Kosslyn, & R. M. Shiffrin (Eds.), From learning processes to cognitive processes: Essays in honor of William K. Estes (pp. 261–278). Erlbaum. Kintsch, W. (1994a). Discourse processes. In G. d’Ydewalle, P. Eelen, & P. Bertelson (Eds.), Current advances in psychological science: An international perspective (Vol. 2, pp. 135–155). Erlbaum. Kintsch, W. (1994b). Text comprehension, memory, and learning. American Psychologist, 49(4), 294–303. Kintsch, W., Patel, V. L., & Ericsson, K. A. (1999). The role of long-term working memory in text comprehension. Psychologia, 42(4), 186–198. Kintsch, W, & Welsch, D. M. (1991). The construction-integration model: A framework for studying memory for text. In W. E. Hockley & S. Lewandowsky (Eds.), Relating theory and data: Essays on human memory in honor of Bennett B. Murdoch (pp. 367–385). Erlbaum Kintsch, W., Welsch, D., Schmalhofer, F., & Zimny, S. (1990). Sentence memory: A theoretical analysis. Journal of Memory and Language, 29(2), 133–159. Klein, K. A., Shiffrin, R. M., & Criss, A. H. (2007). Putting context in context. In J. S. Nairne (Ed.), The foundations of remembering: Essays in honor of Henry L. Roediger III (p. 171–189). Psychology Press. Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: Finding meaning in the N400 component of the event related brain potential (ERP). Annual Review of Psychology, 62, 621–647. Lehman, M., & Karpicke, J. D. (2016). Elaborative retrieval: Do semantic mediators improve memory? Journal of Experimental Psychology: Learning, Memory, and Cognition, 42(10), 1573–1591. MacDonald, M. C., Just, M. A., & Carpenter, P. A. (1992). Working memory constraints on the processing of syntactic ambiguity. Cognitive Psychology, 24(1), 56–98. McElree, B., & Dyer, L. (2013). Beyond capacity: the role of memory processes in building linguistic structure in real time. In M. Sanz, I. Laka, and M. K, Tanenhaus (Eds.), Language down the garden path: The cognitive and biological basis for linguistic structures (pp. 229–240). Oxford University Press. McGugin, R. W., Gatenby, J. C., Gore, J. C., & Gauthier, I. (2012). Highresolution imaging of expertise reveals reliable object selectivity in the fusiform face area related to perceptual performance. Proceedings of the National Academy of Sciences, 109(42), 17063–17068.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
117
118
R. LANE ADAMS AND PETER F. DELANEY
McVay, J. C., & Kane, M. J. (2010). Adrift in the stream of thought: The effects of mind wandering on executive control and working memory capacity. In A. Gruszka, G. Matthews, & B. Szymura (Eds.), Handbook of individual differences in cognition (pp. 321–334). Springer. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81–97. Miller, J. R., & Kintsch, W. (1980). Readability and recall for short passages: A theoretical analysis. Journal of Experimental Psychology: Human Learning and Memory, 6(4), 335–354. Myers, J. L., O’Brien, E. J., Balota, D. A., & Toyofuku, M. L. (1984). Memory search without interference: The role of integration. Cognitive Psychology, 16(2), 217–242. Neath, I., Brown, G. D. A., Poirier, M., & Fortin, C. (2005). Short-term and working memory: Past, progress, and prospects. Memory, 13(3/4), 225–235. Newell, A., & Simon, H. A. (1972). Human problem solving. Prentice-Hall. Perfetti, C. A., & Lesgold, A. M. (1977). Discourse comprehension and sources of individual differences. In M. A. Just & P. A. Carpenter (Eds.), Cognitive processes in comprehension (pp. 1–65). Erlbaum. Peterson, L., & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58(3), 193–198. Pettijohn, K. A., & Radvansky, G. A. (2016). Narrative event boundaries, reading times, and expectation. Memory & Cognition, 44(7), 1064–1075. Reder, L. M., & Anderson, J. R. (1980). A partial resolution of the paradox of interference: The role of integrating knowledge. Cognitive Psychology, 12 (4), 447–472. Sachs, J. S. (1967). Recognition memory for syntactic and semantic aspects of connected discourse. Perception & Psychophysics, 2(9), 437–442. Sahakyan, L., Delaney, P. F., Foster, & Abushanab, B. (2013). List-method directed forgetting in cognitive and clinical research: A theoretical and methodological review. Psychology of Learning and Motivation, 59, 131–189. Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84(1), 1–66. Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory. Psychological Review, 84(2), 127–190. Singer, M. (1990). Psychology of language: An introduction to sentence and discourse processes. Erlbaum. Speer, N. K., Reynolds, J. R., Swallow, K. M., & Zacks, J. M. (2009). Reading stories activates neural representations of visual and motor experiences. Psychological Science, 20(8), 989–999. Speer, N. K., Reynolds, J. R., & Zacks, J. M. (2007). Human brain activity time-locked to narrative event boundaries. Psychological Science, 18(5), 449–455.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
Language Comprehension
Trabasso, T., & Suh, S. Y. (1993). Using talk-aloud protocols to reveal inferences during comprehension of text. Discourse Processes, 16(1), 3–34. Unsworth, N., Fukuda, K., Awh, E., & Vogel, E. K. (2014). Working memory and fluid intelligence: Capacity, attention control, and secondary memory retrieval. Cognitive Psychology, 71, 1–26. Unsworth, N., Spillers, G. J., & Brewer, G. A. (2012). Working memory capacity and retrieval limitations from long-term memory: An examination of differences in accessibility. Quarterly Journal of Experimental Psychology, 65(12), 2397–2410. van Dijk, T. A., & Kintsch, W. (1983). Strategies of discourse comprehension. Academic Press. Wahlheim, C. N., & Jacoby, L. (2013). Experience with proactive interference diminishes its effects: Mechanisms of change. Memory & Cognition, 39(2), 185–195. Wanner, H. E., & Maratsos, M. (1978). An ATN approach to comprehension. In M. A. Halle, J. Bresnan, & G. A. Miller (Eds.), Linguistic theory and psychological reality (pp. 119–161). MIT Press. Waters, G. S., & Caplan, D. (1995). What the study of patients with speech disorders and of normal speakers tells us about the nature of rehearsal. In R. Campbell and M. Conway (Eds.), Broken memories: Case studies in memory impairment (pp. 302–330). Blackwell. Whitney, C., Huber, W., Klann, J., Weis, S., Krach, S., & Kircher, T. (2009). Neural correlates of narrative shifts during auditory story comprehension. NeuroImage, 47(1), 360–366. van Berkum, J. J. A., Zwitserlood, P., Hagoort, P., & Brown, C. M. (2003). When and how do listeners relate a sentence to the wider discourse? Evidence from the N400 effect. Cognitive Brain Research, 17(3), 701–718. Zwaan, R. A. (1994). Effect of genre expectations on text comprehension. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(4), 920–933. Zwaan, R.A. (2004). The immersed experiencer: Toward an embodied theory of language comprehension. Psychology of Learning and Motivation, 44, 35–62. Zwaan, R. A., Langston, M. C., & Graesser, A. C. (1995). The construction of situation models in narrative comprehension: An event-indexing model. Psychological Science, 6(5), 292–297. Zwaan, R. A., Magliano, J. P., & Graesser, A. C. (1995). Dimensions of situation model construction in narrative comprehension. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21(2), 386–397. Zwaan, R. A., & Radvansky, G. A. (1998). Situation models in language comprehension and memory. Psychological Bulletin, 123(2), 162–185.
https://doi.org/10.1017/9781108955638.009 Published online by Cambridge University Press
119
7 The Cognitive Neuroscience of Working Memory and Language Nina Purg, Anka Slana Ozimic, ˇ and Grega Repovš 7.1
Introduction
A key feature of human cognition is the ability to actively represent, maintain, and manipulate complex information in the absence of immediate perceptual input. It allows us to hold on to a telephone number while typing it into our contact list, to keep in mind the location of the neighbor’s bicycle when backing a car into a parkway, to add, subtract and multiply numbers in our mind. It enables us to combine existing concepts into new ones and to maintain ordered plans of our actions to guide our cognition and support flexible goal-oriented behavior (D’Esposito & Postle, 2015). After the initial use of the term in the early 1960s (e.g., Miller et al., 1960), and its broad popularization by the seminal paper by Baddeley and Hitch (1974), we call this ability working memory. As a critical component of our cognitive system, working memory’s characteristics and limits profoundly shape our cognitive abilities, including language. Several chapters in this book provide evidence and discussion of how working memory shapes the structure of grammar (O’Grady, this volume) and syntax (Xu & Liu, this volume) and permeates “all essential linguistic domains” (Lu & Wen, this volume). However, language and working memory share and use an essential representational ability to form well-defined discrete representations. In language, words function both as codes that enable the exchange of information, as well as anchors that allow numerous bits of information from different modalities and continuous dimensions to be integrated into clearly defined concepts with which we can operate. In this way, discrete, categorical representations form the basis for symbolic logic, mathematics, and arguably rational thought. Easily separable and distinct representations are also crucial for working memory. As argued by O’Reilly and colleagues (1999), the representations used for active maintenance must be isolated to support the maintenance of
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
information in the absence of external input and to protect it from interference. In the case of overlapping or distributed representations, activation would spread rapidly, leading to dissipation and loss of precision. As suggested by the authors (O’Reilly et al., 1999), similarly to the role of words in language, the actively maintained representations function as anchors that bias the activity in relevant representational systems that encode specific content. This distribution of responsibilities between the activation and representational systems permeates cognitive neuroscience models of working memory (Slana Ozimiˇc & Repovš, 2020). The benefits of the discrete representations for working memory can be compounded when the target representation itself is a discrete, categorical representation. For the same reason as described above, such representations should be easier to maintain, more robust against interference and loss of precision, and easier to reconstruct in the case of partial loss of information. We suggest that this is one reason why the use of categorical representations and verbal recoding of information is a common strategy that participants employ in working memory tasks (e.g., Starc et al., 2017). The second reason for the use of verbal recoding is the possibility to reuse a system that enables robust encoding and maintenance of item identity and order. Hence, whereas in the cognitive psychology literature working memory is primarily regarded as enabling and shaping several language abilities and characteristics, it has to be acknowledged that language also provides representations and mechanisms that support and enhance working memory. While the ability to form discrete representations and operate with them enables us to conceptualize and reason, it can also lead us to draw boundaries where transitions are smooth, and systems overlap. We often think of complex cognitive abilities such as language and working memory as supported by separable independent modules. In reality, they could overlap and use the same mechanisms, representations, and processes. This insight is well illustrated by functional neuroimaging, which suggests that both working memory and language engage deeply intertwined, distributed networks of brain regions with both distinct and overlapping nodes. Instead of distinct, separable modules, working memory and language can be understood as emerging from and reusing a disparate set of representations, processes, and mechanisms evolved to solve specific computational challenges (see Postle, 2006). This view has also recently come to the fore in research and theories on working memory and language (Buchsbaum & D’Esposito, 2019; Schwering & MacDonald, 2020). The understanding of working memory and language has benefited considerably from the development of research paradigms and methods that provide insights into their neural basis. Single-cell recordings in animals have revealed neural mechanisms of encoding and maintenance of mnemonic representations at the cellular and circuit level
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
121
122
NINA PURG ET AL.
(e.g., Funahashi et al., 1989). Functional magnetic resonance imaging (fMRI) has enabled the transfer of animal research findings to the human brain. It has played a central role in identifying areas and large-scale systems in the brain that support working memory and language processes (e.g., Emch et al., 2019; Walenski et al., 2019). Electro- and magnetoencephalography (EEG, MEG) have enabled the observation of a precise temporal progression of neuronal activity underlying working memory and language processes (e.g., Schiller et al., 2003; Vogel & Machizawa, 2004), as well as the observation of long-range interactions between different neuronal populations and related brain systems (Herman et al., 2013). Finally, the link between conceptual ideas and empirical findings was established by computational modeling, which focused on the development of plausible step-by-step mechanisms of cognitive processes (e.g., O’Reilly et al., 1999; Hitch et al., this volume). In this chapter, we provide a conceptual outline of working memory and language and review the underlying brain systems. We then examine the interactions between the two systems in the context of language acquisition, comprehension, and production. Consider the sentence “The graduate student met with the professor and the research assistant with a curly hair showed the results of the experiment.” Can you figure out who showed the results? Were you able to understand the sentence on the first try or did you have to read it again? In sentences like this, we interpret the beginning of the sentence in a way that later turns out to be wrong. In the example sentence, we are initially led to believe that the verb “met” refers to both “the professor” and “the research assistant.” However, we are able to reevaluate our interpretation by maintaining the words in our working memory until we have gathered enough information to decode the correct meaning of the sentence. The seemingly simple task of reading a sentence goes through many different cognitive processes, such as identifying the written symbols, deriving the meaning behind the symbols, understanding the grammatical structure of the sentence, recoding the written information into sounds, and more. But how does our brain manage these complex processes? To what extent do language comprehension and production interact with working memory? Since working memory and language processes often overlap, does this mean that they also share some neural mechanisms? We will discuss these and similar questions in the following sections. To orient the reader to the relevant cortical anatomy, we have created an inflated cortical map and marked regions identified as engaged by working memory, language, or both (see Figure 7.1). Note that the regions shown are based on association test maps generated using an automated meta-analysis of the terms “working memory” and “language” provided by Neurosynth.org (Yarkoni et al., 2011). Due to the nature of the analysis, the results do not always match the findings of the studies reported in this chapter.
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
Figure 7.1 Working memory and language-related brain regions and their structural connections Note: The resulting maps are based on a Neurosynth term-based meta-analysis including 1,091 neuroimaging studies associated with the term “working memory” and 1,101 studies associated with the term “language”. Association test maps thresholded using whole-brain FDR corrected q < .01 were exported from Neurosynth, mapped to the surface, spatially smoothed, and thresholded again to create masks for regions associated with working memory and language. Finally, the overlap between the resulting working memory and language regions was calculated. MFG – middle frontal gyrus (sometimes referred to as dorsolateral prefrontal cortex, DLPFC), IFG – inferior frontal gyrus (sometimes referred to as ventrolateral prefrontal cortex, VLPFC), PMC – premotor cortex, SMA – supplementary motor area, pre-SMA – presupplementary motor area, ACC – anterior cingulate cortex, AI – anterior insula, PPC – posterior parietal cortex, STG – superior temporal gyrus, MTG – middle temporal gyrus. Numbers in the figure denote Brodmann areas corresponding approximately to pars opercularis (BA 44), pars triangularis (BA 45), pars orbitalis (BA 47), and Heschl’s gyrus (BA 41). Black dashed lines indicate dorsal and ventral white matter tracts between frontal and temporal language regions.
7.2
Working Memory
In the last 50 years, working memory has been an intensive field of research in both cognitive psychology and neuroscience. Research in the neurosciences has been stimulated by the observation of the persistent delay-related activity of neurons in the prefrontal cortex (e.g., Fuster & Alexander, 1971), while research in cognitive psychology has been inspired and shaped by the influential multicomponent model of working memory proposed by Baddeley and Hitch (1974). The model introduced the concept of working memory as a cognitive system that integrates components responsible for
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
123
124
NINA PURG ET AL.
representing and maintaining modality-specific information (the phonological loop and the visuospatial sketchpad) and an executive component with limited attentional capacity responsible for the control of storage components and manipulation of information in working memory (the central executive; for an overview, see Baddeley, this volume). The wealth of subsequent research has led to a further fractionation of storage components and to the addition of a new component, the episodic buffer, which is responsible for the storage of integrated multimodal information (Baddeley, this volume). It has also led to proposals of alternative models of working memory. Neuroimaging studies have shown an overlap between working memory processes and brain areas responsible for perception and long-term memory (e.g., Christophel et al., 2012). This has promoted the so-called state-based models (e.g., Adams et al., this volume), which argue that working memory can be described in terms of different activation states, which are determined by the allocation of attention to internal representations in sensory and motor systems or long-term memory stores (D’Esposito & Postle, 2015). On the other hand, work in the field of human expertise has led to an expansion of the concept of working memory to include strategies for efficient encoding and retrieval of information from long-term memory, the long-term working memory (see Adams & Delaney, this volume).
7.2.1 Neural Mechanisms Underlying Working Memory Processes The investigation of neural correlates of working memory began with electrophysiological recordings in the prefrontal cortex (PFC) of awake monkeys performing delayed-response tasks, such as making a saccade or a hand reach to the previously presented cue after a short delay period. These studies (e.g., Funahashi et al., 1989; Fuster & Alexander, 1971) showed that individual neurons exhibited persistent elevated activation throughout the delay period. After the development of noninvasive neuroimaging methods, such as positron emission tomography (PET) and fMRI, it was possible to observe persistent activity during the delay period in humans as well (e.g., Sweeney et al., 1996). The common assumption of delayed-response studies is that delay-period activity reflects the active maintenance of mnemonic representations necessary for task performance in working memory. Persistent delay-period activity has since been measured in various working memory tasks that required the storage and maintenance of information from different modalities. The role of delay-period activity in working memory has been supported by findings that its duration correlates with delay length (e.g., Funahashi et al., 1989), its amplitude correlates with working memory load (e.g., Vogel & Machizawa, 2004) and behavioural measures of memory precision (e.g., Curtis et al., 2004). Importantly, delayperiod activity is selective and stimulus-specific (e.g., Funahashi et al., 1989). Later studies identified persistent delay-period activity also in other brain
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
areas, including posterior parietal cortex (PPC), areas involved in visual processing, entorhinal cortex, and areas related to motor control, such as premotor cortex (PMC) and supplementary motor area (SMA), superior colliculus, and basal ganglia (for a review, see Constantinidis & Wang, 2004). The nature and necessity of the persistent delay-period activity has been extensively debated. Recent studies using multivoxel pattern analyses (MVPA) have shown that properties of stimuli that are maintained during the delay period can be decoded from the primary sensory cortices, even though these areas do not show increased activity during the delay period (e.g., Christophel et al., 2012). Conversely, several studies (e.g., Christophel et al., 2012) have failed to decode stimulus-specific information from frontal and parietal areas that otherwise show delay-related activity. Taken together, these findings suggest that increased above-threshold delay-period activity might support processes other than storage per se, such as maintenance and manipulation. Concurrently, stimulus-specific information might be encoded in the locally distributed patterns of activity across areas involved in earlier sensory processing. Using spatial smoothing and cluster-based significance tests in traditional analyses, we can ascertain that these regions would not show suprathreshold activity during the delay period (D’Esposito & Postle, 2015). The necessity of persistent delay-period activity for information maintenance has been further questioned by single-cell, fMRI, and EEG studies. Delayperiod activity measured in single neurons occurs in dynamic, sparse bursts that do not extend over the entire duration of the delay period (e.g., Shafi et al., 2007). Studies using MVPA showed that when a stimulus was indicated as irrelevant, the ability to decode it from the fMRI or EEG signal fell to a chance level, but the stimulus was successfully recalled and decoded when required by a subsequent postcue (e.g., Lewis-Peacock et al., 2012). These results suggested that delay-period activity may not be necessary to retain information in working memory. An alternative neural mechanism has been proposed by activity-silent computational models, which assume that information could be encoded and maintained by transient changes in synaptic plasticity of the circuit originally involved in stimulus processing (e.g., Stokes, 2015). The resulting temporary memory trace would then be recovered by subsequent reactivation of the circuit (D’Esposito & Postle, 2015). Research into the neural mechanisms of working memory has also provided insights into the basis of its limited capacity. EEG studies examining contralateral delay activity (CDA) during visual working memory tasks have shown that the amplitude of CDA in posterior cortical areas increases with working memory load and reaches a plateau when working memory capacity is reached or exceeded (Vogel & Machizawa, 2004). These results suggest an upper limit on the number of concurrent representations that can be robustly maintained in working memory. Whether this limit refers to a number of discrete representations (i.e., “slots”) that can be established in the representational system or to the availability of attentional resources for maintenance of the established representations is still a matter of
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
125
126
NINA PURG ET AL.
theoretical debate and empirical investigation (D’Esposito & Postle, 2015). Our own research indicates that constraints exist at both levels (Slana Ozimiˇc & Repovš, 2020) and that multiple strategies can be employed to retain relevant information (Starc et al., 2017).
7.2.2
The Working Memory Network and Functional Roles of Individual Brain Areas Whereas the persistent delay-period activity has been repeatedly replicated, the search for a unified working memory network has been more challenging. Various studies have identified a diverse range of brain areas engaged by working memory tasks, extending from prefrontal and parietal cortices to sensory cortical areas (for a review, see D’Esposito & Postle, 2015). Nevertheless, the meta-analysis of 189 fMRI studies (Rottschy et al., 2012) revealed a core network of most consistently activated cortical areas including Broca’s area, the anterior insula (AI), dorsal and ventral PMC, medial SMA, intraparietal sulcus (IPS), superior parietal lobule, lateral PFC, ventral visual cortex, the lobule VI of the cerebellum, and subcortical activation in bilateral regions of the thalamus and the left basal ganglia. The study also found areas selectively engaged by specific types of tasks. For example, Broca’s area was selectively active during verbal tasks, while ventral and dorsal PMC were most sensitive to object identity or location tasks. A more recent meta-analysis of verbal working memory studies (Emch et al., 2019) confirmed the engagement of lateral frontal cortices, including Broca’s region, SMA, operculum, parts of pre- and postcentral gyrus, parietal cortices, medial cingulate, subcortical nuclei, and cerebellum. Therefore, the common consensus is that working memory is supported by a network of spatially distributed brain areas with different functional roles. The observed brain areas are not specific or unique to working memory but are involved in other mental functions. Hence, working memory is believed to emerge from the interaction of areas that are typically active during mnemonic, attentional, sensory, and motor processes (D’Esposito & Postle, 2015). The functional roles of specific areas continue to be explored. An important research focus is the identification of regions that encode features of a stimulus stored in working memory by identifying neuronal activity that responds selectively to the relevant stimulus feature. For example, neurons in the PFC show selective responses to spatial locations, objects and natural images, colours, visual motion, and the like (for a review, see Christophel et al., 2017), but also to abstract information such as task rules, goals, and categories (D’Esposito & Postle, 2015). Therefore, representations in PFC are thought to encode high-dimensionality or high-order information, integrating different item-related and task-related information (Eriksson et al., 2015). PFC is also believed to play a central role in enabling attentional control, manipulation of information, and suppression of distractions (for a review, see Postle, 2006).
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
The PPC is assumed to provide both selective attentional control and store stimulus-specific information (Eriksson et al., 2015). Regional specificity has been observed in patients with lesions in the parietal cortex, where lesions in the right parietal cortex were found to affect spatial working memory (Koenigs et al., 2009). In contrast, lesions to the left parietal cortex resulted in impairments on verbal working memory (Vallar & Baddeley, 1984). Stimulus-specific information, such as complex visual patterns, have been identified in the PPC using MVPA applied to neuroimaging data (e.g., Christophel et al., 2012). MVPA studies have also revealed stimulus-specific activity in a wide range of sensory cortical areas. For example, visual features, such as orientation, color, motion, and complex patterns, have been decoded from early visual areas, whereas auditory stimuli were found to be stored in the primary auditory cortex, and tactile stimuli and vibrations in the somatosensory cortex (for a review, see Christophel et al., 2017). Taken together, these empirical findings suggest that representations in working memory are distributed across a wide range of low- and high-level cortical areas. The distributed nature of representations is thought to reflect different levels of abstraction, from low-level sensory areas that encode simple stimulus-specific information to high-level prefrontal areas that maintain more abstract representations, such as semantic, verbal, and response-related information, as well as different stages of information processing from early sensory input to the final memory-guided behavioural response (Christophel et al., 2017). Thus, different areas contribute to working memory depending on the nature of the stimulus-specific information and the context of the task.
7.3
Language
Language is a fundamental part of human cognition. It extends beyond “speech” or “communication,” which merely serve language externalization (Friederici et al., 2017), and should be understood as an internal mechanism for generating complex thought. It utilizes syntax, a set of rules and operations that allow lexical elements (e.g., words) to be combined into an unlimited number of structured expressions, such as phrases and sentences (Berwick et al., 2013). Similarly to working memory, language depends on the efficient integration of multiple cognitive and related brain systems engaged with storage and processing of information. Words can be considered as the basic building blocks of the language system. The set of all words that an individual has ever learned constitutes the lexicon or mental vocabulary. The exact nature of lexical representations is still debated; however, cognitive neuropsychology and neuroimaging studies (e.g., Howard & Nickels, 2005; Sanches et al., 2018) indicate that the lexicon can be fractionated into the representation of input and output word forms (e.g., a sequence of
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
127
128
NINA PURG ET AL.
phonemes or graphemes), syntactic (e.g., word categories such as noun or verb), and semantic (i.e., meaning of the word) information. Through the connections between lexicon and semantic memory, words are associated with specific concepts, and by combining multiple lexical elements, new and more complex concepts can emerge (Friederici, 2017). To use words as both external and internal means of communication, we can conceptually define two interface systems (Berwick et al., 2013). An external sensorymotor interface enables the perception and production of language, while a conceptual-intentional interface enables the association of words with concepts and intentions. Both interfaces are tightly integrated with and instantiated by related cognitive and underlying brain systems.
7.3.1 Processing of Language-Related Information The use of language involves many processes that jointly enable the perception, comprehension, and production of language. The specific processes involved depend on the input and output modality (e.g., spoken or written language) and the associated sensory (e.g., visual, auditory, tactile) and motor systems. The processes involved often consist of rapidly unfolding multiple stages, some sequential, others parallel. The perception and comprehension of spoken language begin with the identification of speech sounds (i.e., phonemes) and the acousticphonological analysis necessary to distinguish one word from another in a given language (e.g., /d/ and /t/ in “bad” and “bat”). If the perceived sequence of phonemes matches an entry in the input phonological form lexicon, information about the syntax and semantics of the word is accessed (Friederici, 2011). After identifying the lexical elements, a local syntactic structure such as a noun or verb phrases are formed based on the word category (e.g., Bornkessel & Schlesewsky, 2006). Then, to comprehend a sentence and understand “who is doing what to whom” (Bornkessel et al., 2005), one must also identify the syntactic, semantic, and thematic relations between lexical elements at the phrase or sentence level. Lastly, integration of all these different types of information, together with the processing of prosodic information (i.e., rhythmic and melodic variations in speech) that signal phrase boundaries and highlight relevant words in the sentence, leads to interpretation (Friederici, 2011). Although the initial processing steps described depend on the input sensory modality (e.g., auditory, visual, or tactile), the core linguistic phases remain the same across modalities. Language production is also a multistage process in which conceptual ideas are expressed through vocal articulation, gestures, or writing. Briefly, the initial stage consists of conceptual idea preparation and lexical selection (Levelt, 2001). Lexical selection involves accessing the word form and retrieving semantic and syntactic information, and includes competition with and suppression of other words with similar meaning (Price, 2010).
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
The following stages include grammatical and output (e.g., phonological) encoding (Levelt, 2001). During grammatical encoding, syntactic category and grammatical gender of words are retrieved (Levelt, 2001), words are assigned a thematic and grammatical role in the sentence, and function morphemes are added (e.g., plural, past tense) (Bock & Levelt, 1994). During output encoding, mental representations of words are translated into a sequence of actions in the order in which they will be produced (Schiller et al., 2003). Output encoding is followed by the initiation and coordination of motor activities (Price, 2010), which depends on the modality of language production. Finally, the produced language is monitored for online correction (Price, 2010). In the following sections, we introduce the core language brain network and discuss how the regions of this network support different languagerelated processes. Finally, we outline a time course of language processing indicated by relevant event-related potentials.
7.3.2
The Language Network and Functional Roles of Individual Brain Areas The first significant discoveries of the neural basis of language processing date back to the nineteenth century, when Paul Broca (Dronkers et al., 2007) and Carl Wernicke (Binder, 2017) discovered that parts of the left inferior frontal gyrus (IFG; Broca’s area, BA 44/45) and the left superior temporal gyrus (STG; Wernicke’s area, BA 42/22) are related to language production and comprehension, respectively. The advent of neuroimaging techniques and analytic approaches has allowed a more detailed investigation of the brain mechanisms underlying the language system in recent decades. The findings enabled a better understanding and fractionation of functions within Broca’s and Wernicke’s regions and revealed additional brain regions and their connections, which form a complex network supporting different aspects of language processing (for a recent review, see Friederici, 2017). Unfortunately, due to the language system’s complexity, a detailed analysis of the brain regions associated with language processes is beyond the scope of this chapter, so only a brief outline is provided here. For a recent meta-analysis of fMRI studies that identify brain networks associated with language comprehension and production, see Walenski et al. (2019). Language perception and comprehension are supported primarily by superior temporal regions extending to the superior temporal sulcus. See Friederici (2011) for a detailed review of the functional architecture of superior temporal regions. Briefly, early analysis of the acoustic signal is performed bilaterally by the primary auditory cortex (BA 41), located in the Heschl’s gyrus of both hemispheres, with the left responding preferentially to speech sounds and the right to tonal sounds. Areas posterior to BA 41 appear to be involved in identifying basic features of the acoustic signal,
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
129
130
NINA PURG ET AL.
relaying information to higher-order cortical regions, and supporting auditory-to-motor mapping. Areas anterior to BA 41 appear to discriminate speech from nonspeech sounds and to perceive phonemes. The processing continues with initial syntactic analysis in the anterior STG and frontal operculum (FO). The most anterior part of the STG is involved in syntactic and sentence-level semantic processing, while semantic-syntactic integration, which supports language interpretation, takes place in the posterior regions of the temporal lobe. In summary, the superior and medial temporal lobes consist of a patchwork of specialized brain regions that support both language comprehension and production (for a recent review, see also Binder, 2017). While superior temporal regions primarily support language comprehension, regions of the inferior frontal cortex (IFC) were initially associated with language production. The most prominent, Broca’s area, is usually (Rogalsky & Hickok, 2011) defined as the lateral posterior two-thirds of the left IFG, consisting of two subregions, the pars triangularis (BA 45) and the pars opercularis (BA 44), although FO, the PMC, and the SMA are sometimes also considered parts of Broca’s area. The initial hypotheses about the functional role of Broca’s area were based on findings of agrammatic speech with syntactically simple sentences lacking functional words and morphemes in patients with lesions in Broca’s area, suggesting its role in motor-based speech production (Dronkers et al., 2007) and syntactic processing (Rogalsky & Hickok, 2011). Neuroimaging and lesion studies, though, commonly found Broca’s area to be involved in verbal working memory tasks (for a review, see Chein et al., 2003), suggesting that Broca’s area primarily supports verbal working memory maintenance of information required for comprehension of syntactically complex sentences. Based on a comprehensive review, Rogalsky and Hickok (2011) concluded that no part of Broca’s area shows syntax-specific effects. Conversely, they suggested that the posterior parts (BA 44) of Broca’s area most likely support sentence comprehension through articulatory rehearsal or other processes related to phonological short-term memory. In contrast, anterior parts (BA 45) are more likely to be involved in cognitive control mechanisms that enable semantic integration or conflict resolution. In general, studies of IFC have resulted in an intricate pattern of sometimes contradictory findings, showing a considerable overlap between regions involved in language comprehension, production, verbal working memory, and integration. A more straightforward relationship with language production during initiation and execution of speech has been observed in bilateral motor areas, PMC and SMA (Chang et al., 2009). In addition, the anterior cingulate cortex (ACC) and bilateral caudate appear to be specifically involved in the production of speech sounds (Chang et al., 2009). As both regions have previously been associated with suppression of competing responses (e.g., Ali et al., 2010), activation of these regions suggests greater response selection demands for speech versus nonspeech sounds.
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
Finally, language production requires the monitoring of the output produced. For spoken language, this involves integrating auditory, articulatory, and somatosensory signals with motor output (Price, 2010). Somatosensory speech monitoring activates regions of the postcentral gyrus, while auditory monitoring activates bilateral superior temporal regions associated with speech perception (e.g., Peschke et al., 2009). When speech production becomes more difficult and requires more auditory monitoring, such as when masking sounds prevent participants from hearing the sound of their own speech, increased activation of the pedunculopontine tegmental and ventral supramarginal gyrus can be observed, suggesting that these regions are activated by the conflict between expected and actual auditory input (Zheng et al., 2010). Fiber tracking and interoperative deep stimulation research suggest that areas implicated in language processing are strongly interconnected by long-range pairs of dorsal and ventral connecting pathways between frontal and temporal language regions (for a detailed review, see Friederici, 2011; Friederici et al., 2017). The first dorsal pathway connects the STG to the PMC; it supports sound-to-motor mapping and is essential for language acquisition in infants via auditory-motor integration. The second dorsal pathway connects the STG to the BA 44 and supports higher-order semantic and syntactic functions. The dorsal pathways also connect to the inferior parietal cortex (IPC), supporting phonological working memory storage. The first ventral pathway connects the temporal cortex to BA 45 and presumably supports sound-to-meaning mapping. The second ventral pathway connects the anterior parts of the STG to the FO and supports combining adjacent sentence elements in the initial phase of structure building. The results of functional connectivity analyses support structural connectivity research findings. Resting-state-based analyses helped identify a persistent functional language network (Ji et al., 2019), while task-based studies showed that task demands modulate the network’s functional connectivity (e.g., Lohmann et al., 2010).
7.3.3 Neurotemporal Dynamics of Language Processing Due to their high temporal resolution, electrophysiological methods such as EEG and MEG have provided essential insights into temporal dynamics and revealed several event-related potentials (ERPs) related to language processes. Their exploration and use in research paradigms have helped identify the progression and timing of processing steps in language comprehension and production. They have become an essential tool in the cognitive neuroscience investigation of language in health and disease. Since a detailed exploration of ERPs is beyond the scope of this chapter, we provide a brief overview of some of the most studied language-related ERPs, the timing of their occurrence, and the functions and language features they refer to (see Table 7.1).
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
131
132
NINA PURG ET AL.
Table 7.1 Event-related potentials (ERPS) related to language processes ERP
Timing from stimulus onset
Related brain areas
Function
Early nonlanguage specific auditory analysis: sound onset, loudness, frequency, predictability Early non-language-specific auditory analysis: identification of deviant or missing stimuli, phoneme discrimination Initial syntactic structure building processes: identification of word categories and phrase structure The extraction of meaning and integration with information in short- and long-term semantic memory The identification of grammatical relationships Integration and interpretation of syntactic and semantic information
Language comprehension N100
Around 100 ms
The auditory cortex
MMN
Shortly after N100
The auditory cortex
ELAN
120–200 ms
The left frontal and left temporal cortex
N400
300–500 ms
The auditory and inferior frontal cortex
LAN P600
300–500 ms Around 600 ms
The left frontal cortex MTG, parts of the posterior temporal cortex, basal ganglia
Language production P2 –
Around 200 ms 180–230 ms
Parietal and occipital cortex Fronto-central regions
P300
Around 300 ms
Parietal cortex
N400
300–500 ms
–
275–400 ms
ERN
100 ms after an overt incorrect response
The auditory and inferior frontal cortex Left middle and posterior temporal cortex ACC and SMA
Lexical access (retrieval of words) Conceptual planning in sentences (linearization processes): the involvement of working memory to order the events according to the instructions Conceptual planning in sentences: complexity in linearization Morphological encoding Phonological encoding Error monitoring in speech production
Note: MMN–mismatch negativity, ELAN–early left anterior negativity, LAN–left anterior negativity, ERN–error related negativity. Please see Friederici (2017) and Ganushchak et al. (2011) for a review of the ERP components.
7.4
Working Memory and Language Acquisition
The first language is acquired in the first year of life through primary language faculties, such as speaking, signing, and language comprehension (Sakai, 2005). Secondary faculties, such as reading and writing, require more conscious effort and repetition and are usually acquired later in life through education (Sakai, 2005). The infant’s brain allows acquisition of any natural knowledge to which the infant is exposed, and it begins to respond to auditory speech very early; in newborns (Friederici et al., 2017). Dehaene-Lambertz et al. (2002) have shown that hearing spoken words activate a similar pattern of brain areas in infants at three months of age as in the adult brain. Brain responses were largely lateralized to the left hemisphere and distributed across multiple areas in the temporal cortex.
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
These results suggest that the infant brain already has some functional organization and lateralization for language processing. Nevertheless, the infant’s brain is still undergoing functional and structural changes in response to exposure to the native language, leading to maturation, strengthening, and refinement of the language processing pathways. First and second language acquisition is thought to be strongly dependent on working memory abilities. Several lines of evidence (see Papagno, this volume) show that the ability to hold new phonological forms in working memory is crucial for the formation of a new vocabulary. Patients with deficits in verbal working memory due to left hemispheric damage are unable to acquire the vocabulary of a new language even though their long-term linguistic memory remains normal (e.g., Baddeley et al., 1988). Neuroimaging studies have specifically identified the involvement of the IPC and supramarginal gyrus in learning new vocabulary (e.g., Lee et al., 2007). Conversely, however, verbal working memory also depends on longterm vocabulary knowledge. For example, children repeat nonwords more accurately if they are more similar to known words (e.g., Gathercole, 1995). Therefore, a strong reciprocal relationship between language acquisition, long-term vocabulary knowledge, and working memory is assumed (Baddeley, 2003). There are apparent differences between first and second language acquisition, both in language acquisition progression (Sakai, 2005) and in the mastery of a range of language-related abilities. The differences are observed primarily when the second language is acquired outside the “sensitive period” from infancy to early puberty and are attributed to the loss of flexibility for cerebral reorganization later in life (Sakai, 2005). Second language acquisition is also associated with increased activation of the left dorsal IFG in the early stages of language acquisition, which is attenuated when higher levels of language proficiency are reached (Sakai, 2005). Bilinguals appear to have enhanced abilities for cognitive control, attention, conflict monitoring, switching, and working memory, likely due to the consistent demands on executive functions during the bilingual language experience, which requires active monitoring, selecting, and switching between two competing languages (Grant et al., 2014). The impact of bilingualism is reflected in both brain structure and function. Bilinguals show higher white matter integrity in the corpus callosum, which projects to the bilateral superior longitudinal fasciculi, the right inferior fronto-occipital fasciculus, and the fasciculus uncinatus, and stronger long-range functional connectivity between the frontal cortex and posterior regions, including occipital and parietal cortices (Luk et al., 2011). Studies have also shown that bilinguals have increased gray matter volume in the IPC (Abutalebi et al., 2015), temporal pole, and orbitofrontal cortex (Abutalebi et al., 2014) compared to monolingual controls. The IPC has been specifically linked to the processes of lexical representations, semantic integration, and verbal working memory (Li et al., 2014). These results suggest that second language
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
133
134
NINA PURG ET AL.
acquisition can significantly enhance working memory and other cognitive processes across dispersed brain areas (Grant et al., 2014).
7.5
Working Memory and Language Comprehension
The literature on language comprehension has traditionally explored its reliance on both long-term and working memory processes. At the word level, working memory supports character decoding and word recognition (see Kim, this volume), while long-term memory enables access to semantic and syntactic features of individual words in the lexicon (Caplan & Waters, 2013). Identification of syntactic and semantic relations between words in sentence comprehension requires working memory to maintain related words that are not necessarily adjacent in a sentence (Van Dyke & Johns, 2012; Xu and Liu, this volume). Similarly, working memory is needed for comprehension beyond sentences, such as discourse and high-level text comprehension (Friederici, 2011; Pérez Muñoz & Bajo, this volume). Early studies of language comprehension attributed limitations in language comprehension primarily to limited working memory capacity. A meta-analysis by Daneman and Merikle (1996) found that complex measures of working memory that reflect both processing and storage capacity (e.g., reading span or operational span) are more predictive of comprehension ability than simple working memory measures of storage, such as word span or digit span. Many functional imaging and lesion studies explored the interaction of working memory and language comprehension in detail (e.g., Amici et al., 2007; Cooke et al., 2002). The studies consistently showed the involvement of a distributed network of temporal, parietal, and frontal regions that are individually associated with the retrieval of phonological forms, passive short-term storage supporting sentence comprehension, maintenance of long-distance syntactic dependencies, and other processes. An observation of functional connectivity between working memory and core language processing regions that increases with syntactic complexity of sentences (Makuuchi & Friederici, 2013) further illustrates the close integration of working memory in language comprehension processes. Further research has extended the research beyond investigating the role of working memory, and has shown that language comprehension also involves resources for planning (Novais-Santos et al., 2007) and semantic and syntactic interference resolution (Glaser et al., 2013). These results indicate that executive processes play an essential role in language comprehension. They also support the possibility that retrieval interference, rather than limitations in working memory capacity, contributes to comprehension difficulties (Van Dyke & Johns, 2012). The interaction between working memory and long-term memory in language comprehension is another topic of interest. Building on the long-term working memory model (Adams and Delaney, this volume) Caplan and Waters (2013) propose that short-term memory maintains the https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
retrieval cues for knowledge-based associations, patterns, and schemas that enable parsing and interpretation of sentences. The role of long-term memory in phonological short-term memory and sentence comprehension has also been highlighted in recent studies. Allen et al. (2018) reported that reliance on long-term knowledge benefits sentence recall, while Jones and Farrel (2018) showed superior memory for sequences that more closely follow English syntax. Together with the findings that span tasks for words, but not for nonwords, activate regions previously associated with the phonological word lexicon and lexico-semantic processes (Collette et al., 2001), they suggest a significant reciprocal relationship between language comprehension and working memory.
7.6
Working Memory and Language Production
In contrast to the extensive cognitive and neuroimaging research on the role of working memory in language comprehension, the literature examining the neural mechanisms of the role of working memory in language production is relatively sparse. Behavioral studies, though, indicate that working memory is engaged in speech planning (Swets and Ivanova, this volume) and that writing depends on, and places increased demands on all aspects of working memory (Olive, this volume). A comprehensive cognitive model of how working memory supports language production, particularly writing, was proposed by Kellogg (1996) and found to be strongly supported by empirical evidence in a review by Kellogg et al. (2013). However, recently, there has been a body of work that conversely examines the instrumental role of language production processes in verbal working memory. Neuroscience studies demonstrated that many complex abilities involve overlapping and closely integrated brain systems that may share common mechanisms and processes. Several studies further indicated that long-term phonotactic knowledge (Gathercole et al., 1999), sublexical features (Majerus et al., 2004), and lexical-semantic knowledge (Savill et al., 2019) influence performance on verbal working memory tasks. These and similar findings led to the consideration that strict divisions between verbal working memory, language comprehension, and production might be artifacts of a “divide-and-conquer approach to the study of human cognition” (Buchsbaum & D’Esposito, 2019, p. 134) that may be leading cognitive neuroscientists to “artificially … tear apart” (Ishkhanyan et al., 2019, p. 66) a complex system that integrates a distributed frontotemporal system evolved to support the perception and production of speech. The authors suggest that verbal working memory is supported by the recruitment of domain-specific mechanisms of the language system rather than by dedicated short-term storage mechanism. Specifically, comprehension processes enable encoding (Schwering & MacDonald, 2020) and production processes serial ordering (Acheson & MacDonald, 2009), maintenance, and recall (Schwering & MacDonald, 2020) of items in verbal working https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
135
136
NINA PURG ET AL.
memory. Phonological rehearsal, insightfully outlined in the description of the phonological loop (Baddeley & Hitch, 1974), could be realized through reciprocal connectivity between auditory and motor-speech systems of the brain (Buchsbaum & D’Esposito, 2019). This possibility is congruent with the observation of activity reverberating between the Sylvian-parietal-temporal and Broca’s areas in an MEG study (Herman et al., 2013), possibly reflecting a recurrent flow of information between phonological encoding and articulatory planning (Acheson et al., 2011). Tentative support for this hypothesis is indicated by impaired verbal working memory and language production during repetitive transcranial magnetic stimulation of the posterior STG (Acheson et al., 2011).
7.7
Conclusion
In recent decades, cognitive neuroscience has provided a wealth of information about how working memory and language arise from the integrated functioning of the nervous system. Along the way, studies have repeatedly shown that while we can attempt to “carve a system at its joints” (Kosslyn, 1994) in order to start mapping its complexity, the strong divisions between the identified systems—though helpful to study and understand them— can break down at the level of their implementation in the brain. Systems that we observe and conceptualize at the functional level as separable modules (e.g., phonological loop) may not be implemented as such in the brain. Instead, they reflect a complex and dynamic integration of circuits that have evolved to support a variety of processing needs and are often reused to enable other functionality. The permeable borders between cognitive abilities are evident in the close interdependence, shared representations, and mechanisms that underlie working memory and language. The challenge for future research and theory development is to integrate insights at different levels of observation into a coherent understanding of the cognitive machinery that supports complex thought, symbolic communication, and goal-oriented behavior, as exemplified by the research domains of working memory and language.
References Abutalebi, J., Canini, M., Della Rosa, P. A., Green, D. W., & Weekes, B. S. (2015). The neuroprotective effects of bilingualism upon the inferior parietal lobule: A structural neuroimaging study in aging Chinese bilinguals. Journal of Neurolinguistics, 33, 3–13. https://doi.org/10.1016/j .jneuroling.2014.09.008 Abutalebi, J., Canini, M., Della Rosa, P. A., Sheung, L. P., Green, D. W., & Weekes, B. S. (2014). Bilingualism protects anterior temporal lobe
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
integrity in aging. Neurobiology of Aging, 35(9), 2126–2133. https://doi.org/ 10.1016/j.neurobiolaging.2014.03.010 Acheson, D. J., Hamidi, M., Binder, J. R., & Postle, B. R. (2011). A common neural substrate for language production and verbal working memory. Journal of Cognitive Neuroscience, 23(6), 1358–1367. https://doi.org/10/bcv4c5 Acheson, D. J., & MacDonald, M. C. (2009). Verbal working memory and language production: Common approaches to the serial ordering of verbal information. Psychological Bulletin, 135(1), 50–68. https://doi.org/10/dxhnfj Ali, N., Green, D. W., Kherif, F., Devlin, J. T., & Price, C. J. (2010). The role of the left head of caudate in suppressing irrelevant words. Journal of Cognitive Neuroscience, 22(10), 2369–2386. https://doi.org/10/cfcdwr Allen, R. J., Hitch, G. J., & Baddeley, A. D. (2018). Exploring the sentence advantage in working memory: Insights from serial recall and recognition. Quarterly Journal of Experimental Psychology, 71(12), 2571–2585. https://doi.org/10.1177/1747021817746929 Amici, S., Brambati, S. M., Wilkins, D. P., Ogar, J., Dronkers, N. L., Miller, B. L., & Gorno-Tempini, M. L. (2007). Anatomical correlates of sentence comprehension and verbal working memory in neurodegenerative disease. Journal of Neuroscience, 27(23), 6282–6290. https://doi.org/10/db6h6v Baddeley, A. D. (2003). Working memory and language: An overview. Journal of Communication Disorders, 36(3), 189–208. https://doi.org/10.1016/ s0021-9924(03)00019-4 Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. A. Bower (Ed.), Recent advances in learning and motivation, Vol.8 (pp. 47–90). Academic Press. Baddeley, A. D., Papagno, C., & Vallar, G. (1988). When long-term learning depends on short-term storage. Journal of Memory and Language, 27(5), 586–595. https://doi.org/10.1016/0749-596X(88)90028-9 Berwick, R. C., Friederici, A. D., Chomsky, N., & Bolhuis, J. J. (2013). Evolution, brain, and the nature of language. Trends in Cognitive Sciences, 17(2), 89–98. https://doi.org/10/f4j39r Binder, J. R. (2017). Current controversies on Wernicke’s area and its role in language. Current Neurology and Neuroscience Reports, 17(8), 1–10. https://doi .org/10/gf26w5 Bock, J. K., & Levelt, W. J. M. (1994). Language production: Grammatical encoding. In M. A. Gernsbacher (Ed.), Handbook of psycholinguistics (pp. 945–984). Elsevier/Academic Press. Bornkessel, I., & Schlesewsky, M. (2006). The extended argument dependency model: A neurocognitive approach to sentence comprehension across languages. Psychological Review, 113(4), 787–821. https://doi.org/10/ ctj65q Bornkessel, I., Zysset, S., Friederici, A. D., von Cramon, D. Y., & Schlesewsky, M. (2005). Who did what to whom? The neural basis of argument hierarchies during language comprehension. NeuroImage, 26, 221–233. https://doi.org/10/fjrs3t
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
137
138
NINA PURG ET AL.
Buchsbaum, B. R., & D’Esposito, M. (2019). A sensorimotor view of verbal working memory. Cortex, 112, 134–148. https://doi.org/10/ghpwf3 Caplan, D., & Waters, G. (2013). Memory mechanisms supporting syntactic comprehension. Psychonomic Bulletin & Review, 20(2), 243–268. https://doi .org/10/f4qqws Chang, S.-E., Kenney, M. K., Loucks, T. M. J., Poletto, C. J., & Ludlow, C. L. (2009). Common neural substrates support speech and non-speech vocal tract gestures. NeuroImage, 47(1), 314–325. https://doi.org/10/d2jc3v Chein, J. M., Ravizza, S. M., & Fiez, J. A. (2003). Using neuroimaging to evaluate models of working memory and their implications for language processing. Journal of Neurolinguistics, 16(4–5), 315–339. https://doi.org/10/fw5shn Christophel, T. B., Hebart, M. N., & Haynes, J.-D. (2012). Decoding the contents of visual short-term memory from human visual and parietal cortex. Journal of Neuroscience, 32(38), 12983–12989. https://doi.org/10 .1523/JNEUROSCI.0184-12.2012 Christophel, T. B., Klink, P. C., Spitzer, B., Roelfsema, P. R., & Haynes, J.-D. (2017). The distributed nature of working memory. Trends in Cognitive Sciences, 21(2), 111–124. https://doi.org/10.1016/j.tics.2016.12.007 Collette, F., Majerus, S., Van der Linden, M., Dabe, P., Degueldre, C., Delfiore, G., Luxen, A., & Salmon, E. (2001). Contribution of lexicosemantic processes to verbal short-term memory tasks: A PET activation study. Memory, 9(4–6), 249–259. https://doi.org/10/bbzpf5 Constantinidis, C., & Wang, X.-J. (2004). A Neural Circuit Basis for Spatial Working Memory. The Neuroscientist, 10(6), 553–565. https://doi.org/10 .1177/1073858404268742 Cooke, A., Zurif, E. B., DeVita, C., Alsop, D., Koenig, P., Detre, J., Gee, J., Pinãngo, M., Balogh, J., & Grossman, M. (2002). Neural basis for sentence comprehension: Grammatical and short-term memory components. Human Brain Mapping, 15(2), 80–94. https://doi.org/10/dxk2x9 Curtis, C. E., Rao, V. Y., & D’Esposito, M. (2004). Maintenance of spatial and motor codes during oculomotor delayed response tasks. Journal of Neuroscience, 24(16), 3944–3952. https://doi.org/10.1523/JNEUROSCI.564003.2004 Daneman, M., & Merikle, P. M. (1996). Working memory and language comprehension: A meta-analysis. Psychonomic Bulletin & Review, 3(4), 422–433. https://doi.org/10/c659r8 Dehaene-Lambertz, G., Dehaene, S., & Hertz-Pannier, L. (2002). Functional neuroimaging of speech perception in infants. Science, 298(5600), 2013–2015. https://doi.org/10.1126/science.1077066 D’Esposito, M., & Postle, B. R. (2015). The cognitive neuroscience of working memory. Annual Review of Psychology, 66(1), 115–142. https://doi.org/10 .1146/annurev-psych-010814-015031 Dronkers, N. F., Plaisant, O., Iba-Zizen, M. T., & Cabanis, E. A. (2007). Paul Broca’s historic cases: High resolution MR imaging of the brains of Leborgne and Lelong. Brain, 130(5), 1432–1441. https://doi.org/10/ d5m4r2 https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
Emch, M., von Bastian, C. C., & Koch, K. (2019). Neural correlates of verbal working memory: An fMRI meta-analysis. Frontiers in Human Neuroscience, 13(180). https://doi.org/10/ggqqtf Eriksson, J., Vogel, E. K., Lansner, A., Bergström, F., & Nyberg, L. (2015). Neurocognitive architecture of working memory. Neuron, 88(1), 33–46. https://doi.org/10.1016/j.neuron.2015.09.020 Friederici, A. D. (2011). The brain basis of language processing: From structure to function. Physiological Reviews, 91(4), 1357–1392. https://doi.org/10/crdcmj Friederici, A. D. (2017). Language in our brain: The origins of a uniquely human capacity. MIT Press. Friederici, A. D., Chomsky, N., Berwick, R. C., Moro, A., & Bolhuis, J. J. (2017). Language, mind and brain. Nature Human Behaviour, 1(10), 713–722. https://doi.org/10.1038/s41562-017-0184-4 Funahashi, S., Bruce, C. J., & Goldman-Rakic, P. S. (1989). Mnemonic coding of visual space in the monkey’s dorsolateral prefrontal cortex. Journal of Neurophysiology, 61(2), 331–349. https://doi.org/10.1152/jn.1989.61.2.331 Fuster, J. M., & Alexander, G. E. (1971). Neuron activity related to short-term memory. Science, 173(3997), 652–654. https://doi.org/10.1126/science.173 .3997.652 Ganushchak, L. Y., Christoffels, I. K., & Schiller, N. O. (2011). The use of electroencephalography in language production research: A review. Frontiers in Psychology, 2, 1–6. https://doi.org/10/fn6k8p Gathercole, S. E. (1995). Is nonword repetition a test of phonological memory or long-term knowledge? It all depends on the nonwords. Memory and Cognition, 23(1), 83–94. https://doi.org/10.3758/bf03210559 Gathercole, S. E., Frankish, C. R., Pickering, S. J., & Peaker, S. (1999). Phonotactic influences on short-term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(1), 84–95. https://doi.org/ 10/cvchzn Glaser, Y. G., Martin, R. C., Van Dyke, J. A., Hamilton, A. C., & Tan, Y. (2013). Neural basis of semantic and syntactic interference in sentence comprehension. Brain and Language, 126(3), 314–326. https://doi.org/10/f5gdtz Grant, A., Dennis, N. A., & Li, P. (2014). Cognitive control, cognitive reserve, and memory in the aging bilingual brain. Frontiers in Psychology, 5(1401), 1–10. https://doi.org/10.3389/fpsyg.2014.01401 Herman, A. B., Houde, J. F., Vinogradov, S., & Nagarajan, S. S. (2013). Parsing the phonological loop: Activation timing in the dorsal speech stream determines accuracy in speech reproduction. Journal of Neuroscience, 33(13), 5439–5453. https://doi.org/10/f4svd5 Howard, D., & Nickels, L. (2005). Separating input and output phonology: Semantic, phonological, and orthographic effects in short-term memory impairment. Cognitive Neuropsychology, 22(1), 42–77. https://doi.org/10/fcjfkc Ishkhanyan, B., Boye, K., & Mogensen, J. (2019). The meeting point: Where language production and working memory share resources. Journal of Psycholinguistic Research, 48(1), 61–79. https://doi.org/10.1007/s10936-0189589-0 https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
139
140
NINA PURG ET AL.
Ji, J. L., Spronk, M., Kulkarni, K., Repovš, G., Anticevic, A., & Cole, M. W. (2019). Mapping the human brain’s cortical-subcortical functional network organization. NeuroImage, 185, 35–57. https://doi.org/10.1016/j .neuroimage.2018.10.006 Jones, T., & Farrell, S. (2018). Does syntax bias serial order reconstruction of verbal short-term memory? Journal of Memory and Language, 100, 98–122. https://doi.org/10.1016/j.jml.2018.02.001 Kellogg, R. T. (1996). A model of working memory in writing. In C. Levy & S. Ransdell (Eds.), The science of writing: Theories, methods, individual differences, and applications (pp. 57–71). Erlbaum. Kellogg, R. T., Whiteford, A., Turner, C., Cahill, M., & Mertens, A. (2013). Working memory in written composition: An evaluation of the 1966 model. Journal of Writing Research, 5(2), 159–190. https://doi.org/10/gfpqbv Koenigs, M., Barbey, A. K., Postle, B. R., & Grafman, J. (2009). Superior parietal cortex is critical for the manipulation of information in working memory. Journal of Neuroscience, 29(47), 14980–14986. https://doi.org/10 .1523/JNEUROSCI.3706-09.2009 Kosslyn, S. M. (1994). Image and brain: The resolution of the imagery debate. MIT Press. Lee, H., Devlin, J. T., Shakeshaft, C., Stewart, L. H., Brennan, A., Glensman, J., Pitcher, K., Crinion, J., Mechelli, A., Frackowiak, R. S. J., Green, D. W., & Price, C. J. (2007). Anatomical traces of vocabulary acquisition in the adolescent brain. Journal of Neuroscience, 27(5), 1184–1189. https://doi.org/ 10.1523/JNEUROSCI.4442-06.2007 Levelt, W. J. M. (2001). Spoken word production: A theory of lexical access. Proceedings of the National Academy of Sciences, 98(23), 13464–13471. https:// doi.org/10.1073/pnas.231459498 Lewis-Peacock, J. A., Drysdale, A. T., Oberauer, K., & Postle, B. R. (2012). Neural evidence for a distinction between short-term memory and the focus of attention. Journal of Cognitive Neuroscience, 24(1), 61–79. https://doi .org/10/dmmt42 Li, P., Legault, J., & Litcofsky, K. A. (2014). Neuroplasticity as a function of second language learning: Anatomical changes in the human brain. Cortex, 58, 301–324. https://doi.org/10.1016/j.cortex.2014.05.001 Lohmann, G., Hoehl, S., Brauer, J., Danielmeier, C., Bornkessel-Schlesewsky, I., Bahlmann, J., Turner, R., & Friederici, A. (2010). Setting the frame: The human brain activates a basic low-frequency network for language processing. Cerebral Cortex, 20(6), 1286–1292. https://doi.org/10.1093/cercor/bhp190 Luk, G., Bialystok, E., Craik, F. I. M., & Grady, C. L. (2011). Lifelong bilingualism maintains white matter integrity in older adults. Journal of Neuroscience, 31(46), 16808–16813. https://doi.org/10.1523/JNEUROSCI .4563-11.2011 Majerus, S., van der Linden, M., Mulder, L., Meulemans, T., & Peters, F. (2004). Verbal short-term memory reflects the sublexical organization of the phonological language network: Evidence from an incidental
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
The Cognitive Neuroscience
phonotactic learning paradigm. Journal of Memory and Language, 51(2), 297–306. https://doi.org/10/cjw86h Makuuchi, M., & Friederici, A. D. (2013). Hierarchical functional connectivity between the core language system and the working memory system. Cortex, 49(9), 2416–2423. https://doi.org/10/f5dq23 Miller, G. A., Galanter, E., & Pribram, K. H. (1960). Plans and the structure of behavior. Henry Holt. https://doi.org/10.1037/10039-000 Novais-Santos, S., Gee, J., Shah, M., Troiani, V., Work, M., & Grossman, M. (2007). Resolving sentence ambiguity with planning and working memory resources: Evidence from fMRI. NeuroImage, 37(1), 361–378. https://doi.org/10/bvn8g6 O’Reilly, R. C., Braver, T. S., & Cohen, J. D. (1999). A biologically based computational model of working memory. In A. Miyake & P. Shah (Eds.), Models of Working Memory (1st ed., pp. 375–411). Cambridge University Press. https://doi.org/10.1017/CBO9781139174909.014 Peschke, C., Ziegler, W., Kappes, J., & Baumgaertner, A. (2009). Auditory– motor integration during fast repetition: The neuronal correlates of shadowing. NeuroImage, 47(1), 392–402. https://doi.org/10.1016/j .neuroimage.2009.03.061 Postle, B. R. (2006). Working memory as an emergent property of the mind and brain. Neuroscience, 139(1), 23–38. https://doi.org/10.1016/j.neuroscience .2005.06.005 Price, C. J. (2010). The anatomy of language: A review of 100 fMRI studies published in 2009. Annals of the New York Academy of Sciences, 1191(1), 62–88. https://doi.org/10.1111/j.1749-6632.2010.05444.x Rogalsky, C., & Hickok, G. (2011). The role of Brocaʼs area in sentence comprehension. Journal of Cognitive Neuroscience, 23(7), 1664–1680. https://doi.org/10.1162/jocn.2010.21530 Rottschy, C., Langner, R., Dogan, I., Reetz, K., Laird, A. R., Schulz, J. B., Fox, P. T., & Eickhoff, S. B. (2012). Modelling neural correlates of working memory: A coordinate-based meta-analysis. NeuroImage, 60(1), 830–846. https://doi.org/10.1016/j.neuroimage.2011.11.050 Sakai, K. L. (2005). Language acquisition and brain development. Science, 310 (5749), 815–819. https://doi.org/10.1126/science.1113530 Sanches, C., Routier, A., Colliot, O., & Teichmann, M. (2018). The structure of the mental lexicon: What primary progressive aphasias reveal. Neuropsychologia, 109, 107–115. https://doi.org/10.1016/j .neuropsychologia.2017.12.018 Savill, N., Cornelissen, P., Whiteley, J., Woollams, A., & Jefferies, E. (2019). Individual differences in verbal short-term memory and reading aloud: Semantic compensation for weak phonological processing across tasks. Journal of Experimental Psychology: Learning, Memory, and Cognition, 45(10), 1815–1831. https://doi.org/10.1037/xlm0000675 Schiller, N. O., Bles, M., & Jansma, B. M. (2003). Tracking the time course of phonological encoding in speech production: An event-related brain
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
141
142
NINA PURG ET AL.
potential study. Cognitive Brain Research, 17(3), 819–831. https://doi.org/10 .1016/s0926–6410(03)00204-0 Schwering, S. C., & MacDonald, M. C. (2020). Verbal working memory as emergent from language comprehension and production. Frontiers in Human Neuroscience, 14, 68. https://doi.org/10.3389/fnhum.2020.00068 Shafi, M., Zhou, Y., Quintana, J., Chow, C., Fuster, J., & Bodner, M. (2007). Variability in neuronal activity in primate cortex during working memory tasks. Neuroscience, 146(3), 1082–1108. https://doi.org/10/bcz5k9 Slana Ozimiˇc, A., & Repovš, G. (2020). Visual working memory capacity is limited by two systems that change across lifespan. Journal of Memory and Language, 112, 104090. https://doi.org/10/ggjfhb Starc, M., Anticevic, A., & Repovš, G. (2017). Fine-grained versus categorical: Pupil size differentiates between strategies for spatial working memory performance. Psychophysiology, 54(5), 724–735. https://doi.org/10.1111/ psyp.12828 Stokes, M. G. (2015). “Activity-silent” working memory in prefrontal cortex: A dynamic coding framework. Trends in Cognitive Sciences, 19(7), 394–405. https://doi.org/10.1016/j.tics.2015.05.004 Sweeney, J. A., Mintun, M. A., Kwee, S., Wiseman, M. B., Brown, D. L., Rosenberg, D. R., & Carl, J. R. (1996). Positron emission tomography study of voluntary saccadic eye movements and spatial working memory. Journal of Neurophysiology, 75(1), 454–468. https://doi.org/10.1152/jn.1996 .75.1.454 Vallar, G., & Baddeley, A. D. (1984). Fractionation of working memory: Neuropsychological evidence for a phonological short-term store. Journal of Verbal Learning and Verbal Behavior, 23(2), 151–161. https://doi.org/10 .1016/S0022-5371(84)90104-X Van Dyke, J. A., & Johns, C. L. (2012). Memory interference as a determinant of language comprehension: Interference in comprehension. Language and Linguistics Compass, 6(4), 193–211. https://doi.org/10.1002/lnc3.330 Vogel, E. K., & Machizawa, M. G. (2004). Neural activity predicts individual differences in visual working memory capacity. Nature, 428(6984), 748–751. https://doi.org/10/dpb3j5 Walenski, M., Europa, E., Caplan, D., & Thompson, C. K. (2019). Neural networks for sentence comprehension and production: An ALE-based meta-analysis of neuroimaging studies. Human Brain Mapping, 40(8), 2275–2304. https://doi.org/10.1002/hbm.24523 Yarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C., & Wager, T. D. (2011). Large-scale automated synthesis of human functional neuroimaging data. Nature Methods, 8(8), 665–670. https://doi.org/10.1038/nmeth.1635 Zheng, Z. Z., Munhall, K. G., & Johnsrude, I. S. (2010). Functional overlap between regions involved in speech perception and in monitoring one’s own voice during speech production. Journal of Cognitive Neuroscience, 22(8), 1770–1781. https://doi.org/10.1162/jocn.2009.21324
https://doi.org/10.1017/9781108955638.010 Published online by Cambridge University Press
8 Computational Models of Working Memory for Language Graham J. Hitch, Mark J. Hurlstone, and Tom Hartley 8.1
Introduction
In this chapter, we begin by setting out briefly our view of working memory (WM) as a system, and how it stores verbal inputs. As a system, WM refers to the limited capacity cognitive resources we draw on in everyday activities such as mental arithmetic (Hitch, 1978) and following instructions (Yang et al., 2016) in which temporary information has to be both kept in mind and operated on. Theoretical accounts agree that WM involves an attentional resource that interacts with a set of transient memory representations in order to achieve some goal (Logie et al., 2021). However, beyond this broad consensus, there are many areas of disagreement and emphasis. For example, most approaches view WM as a general-purpose system whereas some see it as specialized for language (e.g., Schwering & MacDonald, 2020). However, this is probably best thought of as a difference in emphasis given evidence that verbal and nonverbal tasks draw at least in part on a common pool of limited resources (Kane et al., 2004; Morey, 2018). Another point of difference is between approaches in which transient information is assumed to correspond to currently activated information in long-term memory (LTM) (Cowan, 1995; Oberauer, 2002) and those in which it reflects the operation of short-term buffer stores (Baddeley, 2000; Baddeley & Hitch, 1974; Logie, 1995). We will argue that this difference is also best thought of as one of emphasis and that WM involves both activated LTM and the temporary storage of novel information. The crucial questions concern how these interact and combine, and we will go on to show that attempts to develop computational models allow us to test specific proposals.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
144
GRAHAM J. HITCH ET AL.
8.2
Overview of Working Memory as a System
In this section, we discuss the evidence for a component of WM capable of storing novel verbal input in phonological form over brief intervals as we regard this capacity as central to its role in language processing.
8.2.1 Transient Storage in Verbal WM The idea that recent perceptual inputs are briefly represented in a limited capacity buffer store was first developed by Broadbent (1958) and was an important feature of the Modal Model of memory, in which short-term memory (STM) was regarded as the gateway to LTM (Atkinson & Shiffrin, 1968). At that time, several strands of evidence were interpreted as supporting the idea of separate short-term and long-term memory stores. These included the limited span of immediate recall (Miller, 1956) coupled with the absence of a measurable limit on LTM; instances of selective neuropsychological impairment to STM in combination with preserved LTM (Shallice & Warrington, 1970) and the converse (Baddeley & Warrington, 1970); separate short-term and long-term components in free recall (Glanzer & Cunitz, 1966); and use of acoustic coding in STM and semantic coding in LTM (Baddeley, 1966a, 1966b). However, despite initial acceptance, certain assumptions in the Modal Model failed to stand up to scrutiny. For example, Shallice and Warrington’s (1970) STM patient showed normal long-term learning, challenging the idea of STM as the gateway to LTM. There was also evidence that identifying STM with acoustic coding and LTM with semantic coding was overly simplistic (Nelson & Rothbart, 1972; Shulman, 1972). Unsurprisingly, the concept of STM became unfashionable. Research interest switched to learning and LTM (Craik & Lockhart, 1972), and there were influential statements of STM’s demise (Crowder, 1982). To some, however, it was premature to abandon interest in STM given that Atkinson and Shiffrin (1968) saw it as serving the function of WM. Evidence that neuropsychological impairment of STM is not associated with general intellectual impairment (Shallice & Warrington, 1970) casts doubt on this suggestion. Nevertheless, Baddeley and Hitch (1974) considered it worth exploring further. They carried out a series of dual-task experiments to examine people’s ability to perform a range of cognitive tasks when STM is loaded with irrelevant verbal information. The cognitive tasks involved verbal reasoning, comprehending prose, and long-term learning in verbal free recall. The outcome was a consistent pattern across all these tasks in that small loads could be maintained with little interference and even when STM was loaded to capacity, the disruption was mild and far from catastrophic. This suggested that tasks loading STM tap into only one aspect of WM. To capture this Baddeley and Hitch (1974) proposed a multicomponent model of WM in which a limited capacity attentional resource directs
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
control processes over two modality-specific buffer stores, one corresponding broadly to verbal STM, the other to visuospatial STM. The underlying philosophy was to set out a broad framework capable of generating further questions that, if fruitful, would lead to progressive refinements to the model. Fortunately, this has proved to be the case in that the original framework has undergone a series of revisions in the light of further research (Baddeley, 1986, 2000; Baddeley et al., 2011) and is still widely used, despite having to compete with an ever increasing number of alternative models (Logie et al., 2021).
8.2.2 Phonological Loop In the multicomponent model of WM, the verbal buffer was initially regarded as a temporary store for speech-coded information. Information in this store was assumed to fade rapidly unless refreshed by vocal rehearsal that could be either overt or covert. The principal evidence for speech-coding was the observation that immediate serial recall is poorer when items are phonologically similar even when presented visually (Conrad, 1964). Evidence for the role of covert rehearsal came from the observation that suppressing articulation by repeating an irrelevant word impairs immediate serial recall and at the same time removes the phonemic similarity effect (Murray, 1968). Another line of evidence came from the word length effect, the observation that immediate serial recall declines as the time needed to articulate items is increased, an effect that is also removed by articulatory suppression (Baddeley et al., 1975). Taken together these observations could be explained by assuming information in the phonological loop decays in approximately 2 s unless refreshed by subvocal rehearsal. Subsequent findings have challenged the assumption that rapid forgetting is due to decay (Jalbert et al., 2011; Lewandowsky et al., 2009; Service, 1998). However, this does not undermine the idea that rapid forgetting happens, only the mechanism through which it occurs, and the basic concept remains influential. The operation of the phonological loop had to be spelled out in more detail to address effects of the modality used to present items for immediate recall. Thus, with visual presentation, articulatory suppression abolishes both the phonemic similarity and word length effects whereas with spoken presentation, suppression abolishes only the word length effect, leaving that of phonemic similarity intact (Baddeley et al., 1984). This is consistent with spoken items accessing the loop automatically, whereas visual items have first to undergo phonological coding, which depends on subvocalization. These effects of presentation modality emphasize a view of the phonological loop as a system specialized for speech input-output. Attempts to investigate recoding in more detail suggest the need to distinguish between different types of phonological code. Thus, whereas articulatory suppression disrupts the ability to decide whether printed
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
145
146
GRAHAM J. HITCH ET AL.
words such as blame and flame rhyme it has little effect on the ability to decide whether words such as ail and ale are homophones or whether a nonword such as pallis sounds like a real word (Besner et al., 1981; see also Baddeley et al., 1981). One interpretation of these differences is that a process of orthographic to phonological recoding feeds into, but is distinct from, the process of subvocal rehearsal. This “inner ear” is capable of supporting holistic judgments of homophony, whereas the “inner voice” of subvocalization is required to allow the segmentation operations involved in making rhyme judgments (Baddeley et al., 1981; see also Vallar & Papagno, 2002). Inner speech has been shown to activate neurophysiological motor-to-auditory mappings that together with auditory-tomotor mappings may form the neural substrate of the phonological loop (see, e.g., Ylinen et al., 2015).
8.2.3 Activated LTM Having briefly described the evidence for the phonological loop as a buffer for speech input-output, we find it instructive to consider the alternative view that these phenomena reflect the transient activation of representations in LTM. The principal drive for this type of account comes from evidence for substantial effects of prior long-term learning in STM tasks. To give some examples, immediate memory is better for words than nonwords (Hulme et al., 1991), for words with higher frequencies of occurrence in the language (Gregg et al., 1989), and for nonwords with higher phonotactic frequencies (Gathercole et al., 1999). It is possible to explain some of these effects in terms of redintegration processes at retrieval that use information in LTM to “clean up” degraded phonological representations (Hulme et al., 1991; Schweickert, 1993). However, more generally this seems an implausible way to account for phenomena such as the immediate recall of meaningful sentences, where word span is typically measured in double figures (Brener, 1940). A more appealing hypothesis is that performance in verbal STM tasks reflects the interactive activation of semantic, syntactic, lexical, and phonological information according to the combinatorial statistics of language knowledge (Martin & Saffran, 1997). Proponents of integrative accounts of WM and language push this approach to the limit by assuming no role for a phonological or indeed any other type of short-term buffer (see, e.g., MacDonald, 2016; Schwering & MacDonald, 2020). However, Norris (2017) pointed out the necessity of assuming some form of transient storage that goes beyond the mere reactivation of existing representations in LTM. This is needed to explain our ability to learn novel information, the ability to recall sequences that contain repeated items (where some form of temporary marker is needed to distinguish the repetitions), and selective neuropsychological impairments of STM and LTM. Norris accepts a role for activated representations in LTM in STM tasks and this is the position we take here. Indeed, the nature of language
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
processing forces us to consider the overlap. We assume that steps toward explaining the role of WM in language will involve understanding how transient memory for phonological sequences operates in the context of constraints that reflect the combinatorial statistics of linguistic knowledge (as we discuss in more detail later). In taking this approach, we have been encouraged by evidence implicating the phonological loop in various aspects of language processing.
8.2.4 Phonological Loop and Language Processing In this section, we discuss briefly some of the empirical evidence linking the phonological loop to the processes of learning new words, language production, and language comprehension. Several strands of research point to its role in vocabulary acquisition (Baddeley et al., 1988; see also the chapter by Papagno, this volume). Most striking is the case of a neuropsychological patient with an acquired selective impairment of auditory-verbal STM who was completely unable to learn word-nonword paired-associates in which the nonwords were foreign language translations, but totally unimpaired in learning word-word pairs (Baddeley et al., 1988). This suggests a key role for the loop in maintaining novel phonological sequences where, unlike familiar words, there is no support from lexical representations in LTM. Complementing the neuropsychological evidence, experimental manipulations of articulatory suppression, phonological similarity, and word length have been shown to affect healthy adults’ learning of word-nonword pairedassociates much more than word-word pairs (Papagno et al., 1991; Papagno & Vallar, 1992). In children, longitudinal studies of individual differences in development suggest that, initially, vocabulary acquisition is limited by phonological loop capacity, whereas at a later stage vocabulary boosts performance in STM tasks by providing lexical support (Gathercole et al., 1992). These findings are complemented by evidence for impaired nonword repetition in children diagnosed with Specific Language Impairment (Gathercole & Baddeley, 1989), Developmental Language Disorder (Graf Estes et al., 2004), and Dyslexia (Melby-Lervag & Lervag, 2012). Longitudinal studies of individual differences in older children indicate a specific association between the ability to repeat nonwords and the acquisition of foreign language vocabulary over a three-year period (Service, 1992). As might be expected, there is also strong evidence linking the phonological loop to speech production. Ellis (1980) demonstrated detailed correspondences between patterns of error in the immediate serial recall of consonant-vowel (CV) and vowel-consonant (VC) syllables and slips of the tongue in everyday speaking (MacKay, 1970; Nooteboom, 1973). Page et al. (2007) took this further by showing that characteristic patterns of error in recalling sequences of alternating phonemically similar and dissimilar letters also occur in speech errors when such sequences are repeatedly read aloud. Interestingly, however, detailed investigation of a neuropsychological case
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
147
148
GRAHAM J. HITCH ET AL.
with a marked selective impairment of auditory-verbal STM revealed that the patient’s speech was completely normal (Shallice & Butterworth, 1977). This would not be expected on a holistic account of the phonological loop as a speech input-output system and has been interpreted as suggesting two buffers, one for speech input, the other for spoken output (Vallar & Papagno, 2002), in line with an earlier proposal by Monsell (1987). Further evidence suggests the phonological loop is also involved in language comprehension, though not crucially. Thus, for normal adults, articulatory suppression disrupts oral comprehension only when sentences are syntactically complex (Rogalsky et al., 2008). This is complemented by evidence of almost normal oral language comprehension in patients with acquired impairment of auditory-verbal STM, with problems only when sentences are unusually long and syntactically complex (Vallar & Baddeley, 1987). Further research has shown the problem occurs when comprehension depends on reactivating phonological information over a distance (Friedmann & Gvion, 2001). We know also that performance in verbal STM tasks is a weak predictor of individual differences in reading comprehension, whereas performance in WM span tasks that measure the capacity to combine attention-demanding mental operations with storage in STM is a much more powerful predictor (Daneman & Carpenter, 1980). These observations suggest that the overall capacity of WM is more important for language comprehension than the phonological loop per se, in line with the earlier findings of Baddeley and Hitch (1974). It seems language inputs feed rapidly through to lexical and semantic streams of analysis, with the phonological loop becoming useful particularly when comprehension requires the reprocessing of verbatim information (Gvion & Friedmann, 2012).
8.2.5 Computational Modeling of the Phonological Loop So far, we have described the phonological loop as a component of WM and discussed briefly some of the evidence for its involvement in three broad aspects of language processing – vocabulary acquisition, speech production, and, to a more limited extent, comprehension. We have also highlighted a number of issues where the concept is in need of further development, the main ones being to incorporate effects of linguistic knowledge (LTM), the learning of new words, and differences between the “inner ear” and the “inner voice.” The initial concept was deliberately parsimonious, and the challenge is whether it can be elaborated to handle these and other additional effects. Most noteworthy among the latter are effects associated with maintaining the temporal order of a sequential input. The phonological loop was originally likened to a closed loop tape recorder, but this was offered as a metaphor rather than a mechanism and cannot account for the characteristic tendency to make transposition errors in immediate serial recall where a presented item is recalled in the wrong position in the
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
sequence. Furthermore, processing and retaining temporal order information is clearly central to language processing more generally, given that this involves a structured hierarchy in which information is serially ordered both within and between levels. Thus, for example, repeating back a spoken sentence involves parallel temporal sequencing at acoustic, phonological, lexical, semantic, and articulatory levels, each with their own characteristic time-scales. The need to process hierarchies of temporal order in language may be the most salient feature distinguishing verbal WM from visuospatial WM. As we will go on to show, we regard immediate serial recall as a tool for exploring verbal STM capacity for temporal order and as a steppingstone toward understanding the role of transient sequential information in language processing more generally.
8.3
Computational Models of Serial Order
As will be clear by now, immediate serial recall (henceforth, serial recall) is the dominant task used to study serial order in verbal STM. Participants are given a sequence of items (letters, digits, or words) that they must subsequently recall in the correct order. This task has generated a wealth of reproducible findings, yielding a rich set of constraints for theorizing (Lewandowsky & Farrell, 2008). The wealth and richness of findings makes serial recall an ideal task for computational modeling efforts. Accordingly, several computational models have been developed that provide a quantitative account of serial recall phenomena. Some of these models are cast within the phonological loop account (Burgess & Hitch, 1999; Page & Norris, 1998), essentially providing a computational instantiation of the verbal-conceptual model that includes an explicit mechanism for serial ordering, whereas others are cast outside the phonological loop concept. The models can be broadly divided into two categories, based on the serial ordering mechanism they employ, namely, chaining models and competitive queuing (CQ) models. In what follows, we discuss each of these classes of models in turn with respect to their ability to explain the list of benchmark findings shown in Table 8.1 (for a more comprehensive list of benchmarks, see Hurlstone, 2021; Hurlstone et al., 2014). We invite the reader to scrutinize the benchmarks, which include key terms referred to in the text, before advancing further.
8.3.1 Chaining Models Associative chaining is the oldest approach to serial order (Ebbinghaus, 1885/1964; Kahana, 2012) and the mechanism of serial recall in several computational models (Lewandowsky & Murdock, 1989; Murdock, 1993, 1995; Solway et al., 2012). In chaining models, serial order is encoded by forming associations between study items. Serial recall is accomplished by
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
149
150
GRAHAM J. HITCH ET AL.
Table 8.1 Benchmark findings of serial recall Study
Finding
Description
Drewnowski and Murdock (1980); Madigan (1971)
Serial-position curve
Crannell and Parrish (1957); Maybery et al. (2002) Henson (1996)
Sequence length effect Error types
Farrell and Lewandowsky (2004); Henson et al. (1996)
Transposition gradients
Farrell et al. (2013); Surprenant et al. (2005)
Fill-in
Hartley et al. (2016); Hitch et al. (1996); Ryan (1969b)
Temporal grouping effects
Baddeley (1968); Henson et al. (1996)
Phonological similarity effects
Baddeley et al. (1975) but see Jalbert et al. (2011); Lewandowsky et al. (2009); Service (1998)
Word length effect
Baddeley et al. (1975); Murray (1968)
Articulatory suppression effect
When recall accuracy is plotted as a function of the serial position of items, the resulting serial-position curve exhibits a large primacy effect (superior recall of early-sequence items) and a small recency effect (enhanced recall of end-ofsequence items). Serial recall performance decreases gradually as sequence length increases. Errors in serial recall can be transposition or item errors. Transpositions occur when items are recalled in the wrong serial positions. Item errors include intrusions (recall of priorsequence or extrasequence items), omissions (failure to recall an item in a position), and repetitions (items recalled on more than one occasion, despite occurring only once in the study sequence). Transpositions are typically more common than item errors. The probability of transpositions decreases with increasing ordinal distance from the target position. Thus, when an item is recalled in the wrong position it will tend to be close to its correct position. This tendency for transpositions to cluster around their target positions is known as the locality constraint. If an item i is recalled a position too soon, recall of item i – 1 is more likely at the next recall position than item i + 1. Thus, given the study sequence ABC, if B is recalled first, then a fill-in error, reflected by the subsequent recall of A, is more likely than an infill error, reflected by the subsequent recall of C. Inserting an extended temporal pause after every few items in a study sequence—known as temporal grouping— improves recall accuracy, causes mini-within-group primacy and recency effects, and a tendency for items to exchange groups but maintain their position within groups (a class of errors known as interpositions). Sequences of phonologically similar sounding items (e.g., B D G P T V) are recalled less accurately than sequences of phonologically dissimilar sounding items (e.g., F K L R X Y). This phonological similarity effect is also observed when sequences are constructed by alternating phonologically dissimilar and similar items (e.g., F B K G R T). Such mixed sequences exhibit a saw-toothed accuracy serial position curve, with peaks corresponding to the recall of dissimilar items and troughs corresponding to the recall of similar items. Serial recall declines as the time needed to articulate items is increased. Thus, ordered recall of sequences of words with long articulation times (e.g., coerce, harpoon, cyclone) is worse than for sequences of words with shorter articulation times (e.g., wicket, pectin, bishop). Suppressing articulation by repeating an irrelevant word (e.g., “the”, “the”, “the”. . .) impairs immediate serial recall and at the same time removes the phonological similarity effect (with visual item presentation) and word length effect.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
Figure 8.1 Illustration of simple (a) and compound (b) chaining models
traversing these associations, which serve as the retrieval cues for sequence production. For example, given the study sequence A, B, C, retrieval of A will cue retrieval of B, which will then cue retrieval of C. Chaining models can be divided into two classes: simple chaining and compound chaining. In simple chaining models (Figure 8.1a), such as the original Theory of Distributed Associative Memory (TODAM) model (Lewandowsky & Murdock, 1989), only forward associations between adjacent items are used to represent serial order. By contrast, in compound chaining models (Figure 8.1b; Solway et al., 2012), which includes later instantiations of TODAM (Murdock, 1993, 1995), serial order is represented by forward and backward associations between both adjacent and nonadjacent items, the strength of which decreases gradually as a function of the distance between items, with backward associations being weaker than forward associations. An additional assumption required in chaining models is that the first item in the sequence must be associated with a start of sequence marker to kickstart the chaining process at recall. Chaining models face several challenges. Although the chaining mechanism produces a primacy effect, it produces no recency effect. The primacy effect arises because successful recall of item i depends on correct recall of item i – 1, which in turn depends upon the correct recall of item i – 2. This dependency means that the chaining mechanism predicts recall performance will decrease monotonically across serial positions, with performance being worst at the final position. Furthermore, although the chaining mechanism generates a primacy effect, without ancillary mechanisms, the extent of primacy produced will tend to be weaker than observed empirically. To accurately model the primacy effect, Lewandowsky and Murdock (1989) had to augment TODAM with two mechanisms – a primacy gradient (see later) in the encoding strength of each successive association, and output interference during recall. Similarly, to explain the recency effect, two mechanisms were once again implemented – retroactive interference during the encoding of item and associative information, and response suppression (see later). While independent empirical evidence can be adduced for each of these mechanisms, the combination of all four to explain two of the most basic serial recall benchmarks is hardly parsimonious. Simple chaining models also encounter difficulties explaining the locality constraint on transposition errors – since a simple chaining mechanism only activates forthcoming items, it cannot readily explain how an earlier item can take the place of a later one as an error. Compound chaining models, by contrast, can capture the pattern of transpositions by
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
151
152
GRAHAM J. HITCH ET AL.
virtue of their use of bidirectional and graded associations between items (Murdock, 1995; Solway et al., 2012). Omission and intrusion errors are also problematic, since in both cases the cue for the next-to-be-recalled item will have been lost, meaning serial recall must terminate before the end of the sequence is reached. This prediction is at variance with the behavioral data, since participants frequently do recover from such errors. There are more serious objections to chaining. First, chaining accounts have difficulties explaining the pattern of findings associated with the recall of sequences in which phonologically similar and dissimilar items are intermixed (e.g., B K P R). Chaining accounts predict recall of the dissimilar items K and R should be impaired, because they possess similar (confusable) retrieval cues. However, this prediction is contrary to the data (Baddeley, 1968; Henson, et al. 1996), which show dissimilar items in mixed sequences are recalled as effectively as items in corresponding positions in pure dissimilar sequences. Second, chaining accounts predict more infill than fill-in errors, because an item recalled too soon will subsequently cue the item that followed it in the input sequence more strongly than any other by virtue of its direct associative link with that item. This prediction is at variance with the empirical data (Farrell et al., 2013; Surprenant et al., 2005). Due to these limitations, and others (see Hurlstone et al., 2014), theorists have largely discounted chaining as a viable account of serial recall (although see Logan, 2021 for a recent chaining model and Osth & Hurlstone, 2021 for a critique) and turned instead to an alternative class of models that we consider next.
8.3.2 Competitive Queuing (CQ) Models The current dominant class of serial recall models are CQ models (Grossberg, 1978a, 1978b; Houghton, 1990). Such models were motivated by the insight that slips in performance, such as transpositions in serial recall, imply that the representation of serial order is parallel (items are coactivated simultaneously), rather than serial (one item excites the next), as is the case in chaining models (Houghton & Hartley, 1995). CQ models are characterized by a two-stage parallel-sequence-planning and responseselection mechanism, comprising an activation layer and a selection layer. In the first stage, target items are activated in parallel according to a gradient of activation by an activating mechanism (Glasspool, 2005) that determines the relative output priority of items. These activated representations are projected to the second stage, wherein items compete for selection through mutual inhibition and self-excitation. The item with the strongest activation level is selected for recall, after which its corresponding representation in the first stage is inhibited – an assumption known as response suppression. This competition is rerun until all items have been selected for output in the second stage, and their corresponding representations in the first stage have been inhibited. By adding random
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
noise to item activation levels in the first or second stage, the CQ mechanism can simulate errors in sequence production. The main difference between CQ models relates to the activating mechanism used to generate the activation gradient. There are two main model variants, context-free and context-based.
8.3.2.1 Context-Free CQ models In context-free CQ models (Figure 8.2a–c), the activating mechanism generates a single activation gradient over items that is then held constant during sequence generation. This pattern of activation could originate from a short-term store, such as the phonological loop, or from LTM. The item activations are constrained by a primacy gradient, such that each item has an activation level that is weaker than its predecessor. Serial recall is accomplished via an iterative process of selecting the strongest item for recall before suppressing its activation, so the next strongest item can be selected. This is the functional mechanism for ordered recall in the models of Grossberg (1978a, 1978b), the primacy model of the phonological loop (Page & Norris, 1998), the Serial-Order-in-a-Box (SOB) model (Farrell & Lewandowsky, 2002), and the LIST PARSE model (Grossberg & Pearson, 2008). The postoutput suppression of items is a crucial ingredient in these context-free CQ models because without it they would perseverate on the first response, which would always be the most active. One limitation of context-free CQ models is that the order of repeated items in a study sequence cannot be represented using type representations with a single activation level. Repeated items must therefore be handled by incorporating multiple token representations of the same item. 8.3.2.2 Context-Based CQ Models In context-based CQ models (Figure 8.2d–f), each item in a sequence is associated with the current state of an internal context signal that changes gradually during encoding and represents the positions of items in the sequence. At recall, the context signal is reset and reevolves along its original path, with sequence items being activated according to the degree of similarity between the current state of the context signal and the state to which each item was originally associated. The context signal therefore serves as a dynamic activating mechanism that shifts the source of activation during retrieval. The dynamic nature of the activating mechanism enables these models to generate sequences containing repeated items using type rather than token representations. This is because the type representation of a repeated item will receive different sources of activation at different points during sequence generation. The postoutput suppression of items is therefore less crucial in these models because the context signal carries the primary burden for sequence generation. Context-based CQ models can be classified as event-based, time-based, or a hybrid of the two. In event-based models, the context signal changes when
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
153
154
GRAHAM J. HITCH ET AL.
Figure 8.2 Schematic of the architecture of context-free (a–c) and context-based (d–f ) CQ models and the steps involved in producing a three-item sequence. Note: Both classes of models (all panels) comprise an activation layer and a selection layer. Lines terminating with arrows represent excitatory connections, whereas lines terminating with circles represent inhibitory connections. Each node in the lower selection layer has an inhibitory connection to every other node in the same layer (for simplicity only adjacent-neighbor inhibitory connections are shown) and to its corresponding node in the activation layer (the pathway through which response suppression is implemented). Columns in the activation layer represent the activation levels of the nodes representing items in the to-be-recalled sequence. In context-free CQ models, a single activation gradient – known as a primacy gradient – is established over target items in the activation layer. The activations are projected to the selection layer, wherein items compete for selection via lateral inhibition, resulting in the production of the first item and the suppression of its corresponding representation in the activation layer (a). This allows the second item to become the most active and win the response competition (b), and its suppression, in turn, allows the third item to become the most active resulting in its production (c) and subsequent suppression (not shown). In context-based CQ models, the activation gradient established over target items in the activation layer varies in response to the activity of a reevolving context signal – originating from an additional layer, known as the context layer – to which items were temporarily bound during serial order encoding. Reinstatement of the context signal to its initial state in the context layer establishes an activation gradient over target items in the activation layer. Activations are projected to the selection layer, wherein items compete for selection, resulting in the production of the first item and the suppression of its corresponding representation in the activation layer (d). The context signal advances to the next state and a new activation gradient is established, allowing the second item to win the output competition (e), and its suppression, followed by the advancement of the context signal to its final state, generates a new activation gradient allowing the third item to become the most active, resulting in its production (f ) and subsequent suppression (not shown).
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
a new event (e.g., a new item) is experienced. In one class of event-based models, the context signal encodes absolute within-sequence position. Models falling into this class include C-SOB (Farrell, 2006; Lewandowsky & Farrell, 2008) and the original Burgess and Hitch (1992) network model of the phonological loop. For example, in the latter model, items are associated with a context signal implemented as a vector of inactive nodes containing a dynamic window of active nodes. The context vector changes gradually with the presentation of each item by sliding the moving window of activation from left to right by a constant one node per item. In another class of event-based models, the context signal encodes relative withinsequence position. For example, in the Start-End Model (SEM; Henson, 1998; see also Houghton, 1990) items are linked to the varying states of a context signal comprising two elements – a start marker that is strongest for the first position and decreases exponentially in strength across positions, and an end marker that is weakest for the first position and increases exponentially in strength across positions. Such a context signal represents approximate position relative to the start and end of the sequence. In time-based models, the context signal changes as a function of absolute time. Models in this class include more recent instantiations of the Burgess and Hitch (1999, 2006) model in which the same moving window context signal changes with time rather than events, and the OscillatorBased Associative Recall (OSCAR) model (Brown et al., 2000). In the OSCAR model, items are linked with the different states of a time-varying context signal driven by sets of temporal oscillators operating at different frequencies. At recall the context signal is reset to its initial state before being replayed, with sequence items being reactivated through their original associations with the timing signal. A similar, but more abstract, temporal coding scheme is utilized by the Scale-Invariant Memory, Perception, and LEarning (SIMPLE) model (Brown et al., 2007).
8.3.2.3 CQ Explanation of Benchmark Findings CQ models can account for the benchmark findings of serial recall. In contextfree CQ models, the primacy effect materializes because the activation levels of items near the beginning of the sequence are more distinctive, meaning these items encounter less competition during recall than items toward the end of the sequence. By contrast, the recency effect manifests because as successive items are recalled and suppressed, the number of response competitors is gradually reduced. In context-based CQ models, primacy and recency effects are partly, if not wholly, determined by “edge effects” – there are fewer opportunities for items near the beginning and end of a sequence to move around, compared to items at medial positions. In some models (e.g., Brown et al., 2007; Henson, 1998), an additional contributing factor is the greater distinctiveness of the context signal at terminal positions. In both context-free and context-based CQ models, sequence-length effects arise because the greater the number of items in the target sequence, the greater
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
155
156
GRAHAM J. HITCH ET AL.
the probability of an error being committed. In some context-based CQ models (e.g., Henson, 1998), an additional factor contributing to the sequence length effect is that the resolution of the context signal for longer sequences is weaker than for shorter sequences. It is generally easier for CQ models to produce transpositions than item errors. This is a natural consequence of the parallel sequence dynamics assumed by these models, which when perturbed by noise will alter the relative priority of items. Near-neighbor transpositions predominate because the representation of serial order via an activation gradient necessarily implies that the strongest competitors to the target item at each recall position will be items from adjacent, rather than remote, serial positions, thus accommodating the locality constraint. Omission errors are accommodated by incorporating an output threshold that the strongest item must exceed in order to be recalled, while intrusion errors are modeled by weakly activating extrasequence items to allow them to enter into the response competition. The scarcity of erroneous repetitions is accounted for in CQ models by the postoutput inhibition of items, which reduces the likelihood an item will be recalled more than once. Context-free CQ models can accommodate the finding that fill-in errors are more frequent than infill errors. This is because if an item i is recalled a position too soon and then suppressed, item i – 1 will be a stronger competitor at the next recall position than item i + 1, because the former item, by virtue of being presented earlier in the sequence, will have been encoded more strongly on the primacy gradient. Context-based CQ models can also accommodate this result (Burgess & Hitch, 1999; Henson, 1998), either by incorporating a primacy gradient as one component of the context signal (Henson, 1998), by incorporating a primacy gradient in the strength of the connections between items and the context signal (Brown et al., 2000; Lewandowsky & Farrell, 2008), or by incorporating a primacy gradient in conjunction with the context signal (Burgess & Hitch, 1999). A primacy gradient also appears to be necessary for context-based CQ models to provide an adequate account of the primacy effect, the distribution of omissions and intrusions across output positions, and the dynamics of transpositions (see Hurlstone, 2021 and Hurlstone et al., 2014 for discussion). To accommodate temporal grouping effects, context-based CQ models assume a hierarchical or multidimensional context signal, whereby one component of the signal represents the positions of items or groups within the sequence overall, and the second component represents the positions of items within groups (Brown et al., 2000; Burgess & Hitch, 1999; Hartley et al., 2016; Henson, 1998; Lewandowsky & Farrell, 2008). This provides a more distinctive two-dimensional representation of serial position, which accounts for the overall reduction in transposition errors in grouped sequences and the emergence of within-group primacy and recency effects. Interposition errors arise because items in the same positions in different groups are associated with the same state of the component of the context signal that represents within-group position, rendering them vulnerable to
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
confusion. Grouping effects are beyond the purview of context-free CQ models, which, because of their lack of positional representations, are unable to explain the pattern of interpositions. CQ models explain phonological confusions by assuming a third stage wherein an item chosen at the second stage undergoes a further competition in which it is vulnerable to confusion with other items based upon its degree of phonological similarity to those items (Burgess & Hitch, 1999; Henson, 1998; Page & Norris, 1998). The effect of this is to increase the likelihood that a similar item recalled from the second stage will be confused with another similar item in the third stage, thus accommodating the poorer ordered recall of phonologically similar compared to dissimilar items in pure and mixed sequences. Two CQ models of the phonological loop, namely the primacy model (Page & Norris, 1998) and the Burgess and Hitch (1999) model have been applied to the word length effect. Both models assume that forgetting in short-term memory is attributable to trace decay. Thus, in the primacy model the primacy gradient of activations held over items decays gradually with the passage of time but can be refreshed by a process of cumulative rehearsal. In the Burgess and Hitch (1999) model, such trace decay occurs in the connections between items and the context signal used to represent serial order, as well as connections between items and input and output phonemes that implement the phonological loop. Rehearsal involves an iterative process whereby items activate output phonemes corresponding to their pronunciation pattern. The output phonemes then send activation to input phonemes, which then send activation back to the items. On each cycle of rehearsal, the strength of connections between items and output phonemes and input phonemes and items is partially strengthened. In keeping with Baddeley’s verbal conceptual account, both models account for the word length effect because long words take longer to articulate then shorter words, meaning there is less opportunity to engage in rehearsal to offset the (item or weight) decay process. The model of Burgess and Hitch (1999) can also simulate the articulatory suppression effect. The effect is approximated by adding noise to the output phonemes corresponding to the spoken pronunciation of items. This noise propagates to the input phonemes and through to the items, producing interference and blocking the use of the phonological loop for rehearsal. This causes a reduction in serial recall performance, although the degree of disruption is much more pronounced than observed empirically. Through this mechanism, Burgess and Hitch (1999) are able to simulate the abolition of the word length effect and phonological similarity effect (with visual item presentation) under articulatory suppression. This section has focused on computational modeling of serial order at the lexical level but has ignored the constraints from language knowledge, new word representation, and learning we discussed at the outset. We turn next to computational approaches to serial order that take these linguistic constraints into account.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
157
158
GRAHAM J. HITCH ET AL.
8.4
Linguistically Constrained Computational Models
When considering the adequacy of computational models to account for phenomena in language and WM, it is necessary to go beyond the question of how arbitrary sequences of familiar items are transiently stored and recalled and begin to consider constraints that arise from the structure of language. In natural language, the sequence of phonemes in a syllable, morphemes in a word, or words in a sentence is each subject to constraints that make some sequences impossible, ungrammatical, or meaningless, while others vary hugely in their probability as determined by both perceptual and motor limitations and by long-term linguistic knowledge as reflected in phonetics, phonology, semantics, syntax, and so on. While computational models of serial order in STM provide a useful starting point, there is plainly a very long way to go in understanding how these influences contribute to memory for real-world language and the role of WM in comprehension, production, and communication. Working memory researchers, psycholinguists, and others with an interest in the mechanisms of verbal memory each approach these questions with different goals and priorities. Can the best insights of models of language, serial order, and WM be reconciled to arrive at a consensus theory, and if so how can such theoretical integration be achieved? In attempting to integrate different models it is important to consider the theoretical territory each model occupies. Where is a given model situated and how does it relate to other models, the empirical effects it explains, and those it does not address or with which it conflicts?
8.4.1 Central Role of Serial Order in Theoretical Integration We think about these questions in terms of a simplified and idealized “theoretical space” described by “dimensions” of scope and abstraction, which we can view from different perspectives within which certain empirical distinctions become more or less important. Scope (illustrated on the x-axis in Figure 8.3) refers to the range of empirical phenomena addressed by the model, the experimental results that it can simulate including existing observations and predictions. Models with broader scope typically explain different phenomena through more abstract mechanisms (for example, in language models mechanisms that are independent of modality or task), whereas models that have narrower scope can achieve greater granularity in their explanations of phenomena by incorporating specialized mechanisms (for example, specific to orthography, audition, or motor processing). To characterize this difference, we can think of an additional dimension abstraction (illustrated on the y-axis in Figure 8.3), which is perhaps loosely linked to Marr’s concept of levels of analysis (Marr, 1982).
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
Figure 8.3 The central role of serial order in connecting detailed implementations of working memory in specific tasks with more abstract and general theories. Note: (a) We can regard models (black ellipses) as occupying a position in a space defined by scope (empirical phenomena they address) and abstraction (the degree to which they characterize general principles). An abstract model (such as the concept of CQ) can provide general principles that span and connect theories of WM (right-hand side) and serial order (left-hand side). But more detailed and less abstract models are needed to address specific forms of serial memory (such as auditory-verbal STM). Models that overlap in terms of scope and abstraction can potentially form an explanatory chain, connecting abstract general principles to concrete specific empirical phenomena. (b) However, the theoretical space has many dimensions of scope. For example, language (top right) encompasses phenomena that extend well beyond WM, and phonological memory includes long-term lexicalphonological knowledge. (c) Viewed from a psycholinguistic perspective, a different chain of models may be needed to connect the general principles of language to those of WM, but the mechanisms of serial order are likely to play a central role. https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
159
160
GRAHAM J. HITCH ET AL.
Each implemented model tends not to cover the entire space, incorporating both fine-grain (e.g., modality-specific) mechanisms and general shared principles. This is because the explanatory value of general principles would often be undermined by more detailed specification, while the explanatory value of implementational detail is often necessarily limited to a specific domain. In trying to integrate models, we are looking to identify a coherent chain of models whose overlapping scope and abstraction can connect abstract general models to very specific and detailed predictions about empirical phenomena. In reality, however, scope is a multidimensional space. In the current context, we are considering computational models of WM for language, which leads us to focus on a specific range of relevant empirical phenomena as outlined above, but even within this area different scientists may take different views about the most important issues. For example, psycholinguists may favor models and explanatory chains that extend to and connect with linguistic phenomena beyond WM, while cognitive psychologists might prefer models that connect with more general WM phenomena (such as amodal attention) that are less immediately relevant to language. This means there are multiple valid perspectives on the same theoretical space. In the current context (as illustrated in Figure 8.3), mechanisms of serial order are central to both language and WM. Thus, they play a central role in connecting detailed implementations of verbal WM in specific tasks (such as learning novel phonological forms, digit span, or sentence production) with more abstract and general theories (such as Baddeley and Hitch’s (1994) multicomponent model of WM or Gupta and MacWhinney’s (1997) account of vocabulary acquisition). For this reason, some of the most promising avenues for theoretical integration lie in reconciling models of serial representation and processing, and in identifying common principles underpinning the serial representation and processing of linguistic content. Here we can identify a degree of consensus in the computational principles identified above. In particular, CQ models implicate context signals in the rapid acquisition of new unfamiliar sequences. Moreover, because all new words are unfamiliar the first time they are encountered, these principles are likely to play an important role in constraining the development of long-term lexical-phonological representations.
8.4.2 Context Signals Incorporating Linguistic Constraints In the context of language, these time-varying signals need to be sensitive to linguistic structure if they are to account for linguistic constraints on errors in immediate recall (i.e., in STM) and spontaneous speech (i.e., in the retrieval of long-term phonological representations). Errors in both situations are constrained by similar linguistic factors (see e.g., Boomer & Laver, 1968; Dell & Reich, 1981; Ellis, 1980; Shattuck-Huffnagel, 1979; Treiman & Danis, 1988; Vousden et al., 2000). For example, phonological errors tend to involve
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
paired or partial transpositions of phonemic units between corresponding parts of the same syllable. This contrasts with order errors in unstructured sequences (e.g., as used in digit-span tasks), which tend to involve transpositions between adjacent serial positions, but bear a resemblance to the interposition errors observed when such sequences are temporally grouped (see Table 8.1). In the CQ modeling framework, syllabic constraints on phonological errors might be explained by a context signal that is sensitive to syllable structure, so that phonemes associated with corresponding parts of nearby syllables are associated with similar states of the context signal. For example, in a spoonerism such as “barn door” ! “darn bore” the syllable initial /d/ and /b/ would be associated with similar states of the context signal, such that they compete with one another to be output at the initial position in each syllable. If the wrong phoneme is selected for output in the first syllable, it is suppressed, leaving the unselected initial consonant free to be selected at the initial position in the second syllable, but other phonemes (e.g., vowels, syllable-final consonants) do not compete because they are associated with distinct states of the context signal. Although the spoonerism example reflects constraints seen in spontaneous speech errors, as noted above a very similar pattern is seen in nonword repetition (e.g., Ellis, 1980; Treiman & Danis, 1988). In the context of this STM task, however, it becomes clear that any linguistic structure present in the context signal must derive from the unfamiliar sequence itself, because different syllable structures are subject to different patterns of error – each tending to allow transposition only to corresponding parts of different syllables. This implies, in the CQ framework, that some process must extract a representation of syllable structure online (as new words are encountered) to shape the context signal. An additional challenge – especially if we consider the learning of new phonological forms in the context of language development – is that we cannot assume that the process taps into a preexisting representation of syllable structure; rather, it must rely on cues present in the stimulus itself as it is first encountered. In other words, new word learning seems to demand a bottom-up representation of syllable structure, rather than one based on top-down application of preexisting linguistic knowledge. To resolve these problems, Hartley and Houghton (1996) proposed a CQ model for phonological sequencing in auditory verbal STM that incorporated a cyclical context signal sensitive to the sonority of successive phonemes. Sonority is a linguistic property of the phoneme corresponding to its relative loudness (for instance vowels are more sonorous than voiceless fricatives). Because of the sonority principle (Selkirk, 1984), phonemes in well-formed syllables typically conform to a pattern in which increasingly sonorous consonants build toward a peak (vowel) after which sonority declines over successive segments. This creates a wavelike pattern of peaks and troughs, and the phase of the cycle can be used to associate phonemes with the appropriate part of each syllable, accounting for the constraints seen in nonword repetition errors.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
161
162
GRAHAM J. HITCH ET AL.
Figure 8.4 Example of responses of syllabic phase model oscillators during processing of the sentence “Iguanas and alligators are tropical”. (a) shows the raw speech signal being processed. Phonetic and orthographic annotations show the locations of the phonemes and words. (b) and (c) show how the phase and amplitude responses, respectively, of oscillators varies over time. Source: TIMIT Database, NIST, 1990, Subject: mjmp0 Sentence:sx95. Reproduced under the Freedom of Information Act of 1966, https://www.nist.gov/foia
Although the original Hartley and Houghton model was based on the concept of sonority, it can be implemented using acoustic cues in the envelope of the speech signal (more sonorous phonemes coincide with peaks in the envelope for voiced frequencies, leading to a quasi-periodic amplitude modulation at the speech rate). Hartley (2002) constructed a syllable tracking context signal by combining the outputs of a number of oscillators processing and tuned to different frequencies of amplitude modulation spanning the range of typical speech rates. The syllabic phase model shows how auditory processing can track the speech rate, yielding – through a bottom-up process – a context signal that is sensitive to withinsyllable position. This is illustrated in Figure 8.4, which shows the responses of oscillators in the syllabic phase model during processing of the spoken sentence “Iguanas and alligators are tropical.” Panel (a) shows the raw speech signal, whereas panels (b) and (c), respectively, show the combined phase and amplitude responses of the syllabic phase oscillators over time. Syllable boundaries correspond with troughs in the envelope in panel (b) and it can be seen by comparison with panel (a) that these boundaries occur in phonologically plausible locations in the majority of cases. The syllabic phase model is consistent with parallel discoveries showing that activity in auditory cortex tracks the speech signal when participants hear natural connected speech (e.g., Ahissar et al., 2001, see Ding & Simon, 2014 for review).
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
8.4.3 Temporal Grouping in Working Memory: A Link to Prosody? It is natural to ask whether the same kind of mechanism might apply to larger scale serial structure in language, for example, governing the order of words in a sentence. However, natural sentences have a very complex hierarchical structure that potentially combines influences of syntax, semantics, and so on. Again, the serial recall literature from laboratory studies provides useful constraints. Even in the absence of top-down information used when recalling arbitrarily ordered sequences of words, as noted above, performance is influenced by the timing of the items. Regular grouping is advantageous for immediate recall and leads to characteristic changes in the distribution of order errors. The serial position curve shows multiple-bowing reflecting primacy and recency effects within groups, adjacent transposition errors are reduced, while transpositions between corresponding items in different groups increase. The latter effect is strikingly similar to the pattern seen in syllable-level transpositions between phonemes addressed in the Hartley and Houghton (1996) and Hartley (2002) models. Could similar mechanisms, reflecting prosodic and rhythmic structure, be at work in the sequencing of words? The regular grouping of sequences in serial recall experiments is indeed rather artificial, and it might be argued that grouping effects can be explained in terms of top-down expectations arising from the repeated and predictable pattern used in STM experiments. However, Ryan (1969a) had shown that when the grouping pattern of auditory sequences was varied, similar but necessarily more complex patterns emerged with local recency and primacy. Our own work (Hartley et al., 2016) showed that these patterns were reliable and reproducible even when the grouping pattern was varied on a trial-by-trial basis, so participants could have no foreknowledge of the sequence structure. We were able to explain these patterns using a CQ model with a context signal comprising a bottom-up multiscale population (BUMP, also described in Hartley et al., 2016) of oscillators similar to those used by Hartley (2002). Again, the oscillators are sensitive to local amplitude modulations in the envelope of incoming speech, with each oscillator possessing an intrinsic tuning – a tendency for its activity to oscillate at a specific rate. The frequency tunings of the oscillators are chosen to span the range of presentation rates encountered in spontaneous speech. Some oscillators are sensitive to slow modulations on a temporal scale corresponding to the length of the sequence (say 5 s), others are sensitive to more rapid fluctuations corresponding to groups of items (say 1–2 s), and yet others are sensitive to faster modulations corresponding to presentation of the individual items (0.75 s). In the BUMP model, when a sequence is presented, the rhythm and timing of items determines which oscillators become entrained to the auditory input. The context signal is thus similar to that suggested in OSCAR (Brown et al., 2000), but unlike that model, BUMP incorporates a bottom-up mechanism explaining how the context signal arises and evolves in response to irregular and unpredictable speech.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
163
164
GRAHAM J. HITCH ET AL.
Figure 8.5 Phase and amplitude responses of a population of oscillators with different tunings (spaced between 0.1 Hz and1.28 Hz) in the BUMP model of Hartley et al. (2016). For each axis, the upper trace shows the timing of triangular amplitude pulses representing each item in a nine-item sequence. The phase amplitude diagram represents the evolution of amplitude and phase of oscillators with different tunings (y-axis) over time (x-axis). Phase and amplitude are indicated by brightness (see Hartley et al. for a colored version of this diagram in which phase and amplitude are distinguishable). (a) ungrouped sequence, (b) sequence grouped into three groups of three (3-3-3), (c) 2-6-1 grouped sequence. Note that the same population of oscillators responds to all three sample sequences, but the phase and amplitude of the entrained responses is systematically affected by the sequence structure and in particular by local amplitude modulations on different scales (e.g., corresponding to sequence, group, item). Each item will be associated with the state of the oscillator population at the time it is presented. At retrieval items associated with similar states may be confused with one another, resulting in transposition errors.
To illustrate, in the case of an evenly timed, ungrouped sequence, oscillators with tunings close to the item presentation rate will respond strongly and in phase with the items (Figure 8.5a). These oscillators will go through one cycle per sequence item. Oscillators with tunings close to the sequence presentation rate respond to the larger-scale amplitude fluctuation associated with presentation of the entire sequence and are insensitive to the relatively rapid changes associated with individual items. These slower oscillators’ output will go through approximately half a cycle during presentation of the sequence. Oscillators with intermediate tunings respond
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
only weakly, and their responses are largely restricted to the beginning and end of the sequence. However, for a regularly grouped sequence, oscillators with tunings close to the group presentation rate are also recruited, and go through one cycle per group (Figure 8.5b). Oscillators with tunings close to the group presentation rate are also recruited for irregularly grouped sequences, but the inconsistency of group durations means oscillators in this range are less strongly activated than would be the case for regularly grouped sequences (Figure 8.5c). An important property of the model is that some temporal grouping patterns are more favorable than others. Regular grouping patterns powerfully activate oscillators tuned to the grouping rate, which enhances overall recall, albeit at the expense of interposition errors (as seen in the similarity of the outputs of filters tuned to the group presentation rate in Figure 8.5b). Irregular grouping patterns similarly favor intergroup transpositions although the correspondence between different positions is less clear-cut (see Figure 8.5c). In addition to explaining the pattern of errors based on the sequence structure, the BUMP model overcomes several theoretical barriers to the wider application of CQ in language processing. Specifically, it provides a mechanism that avoids two formerly unexplained problems that affected the capacity of earlier models to deal with more realistic verbal sequences seen in natural language: (1) how to anticipate the start and end of a sequence and (2) how to choose the appropriate rate of change of the context signal. BUMP is a hybrid model – the context signal changes smoothly over time, but these changes are driven by events.
8.4.4 Wider Considerations In our own work and in the models reviewed above, we have focused mainly on putatively hard-wired or bottom-up mechanisms that can rapidly learn and retrieve unfamiliar sequences in the absence of relevant longterm knowledge. We argue that these mechanisms must play a foundational role in language because STM for unfamiliar verbal sequences is a necessary precursor to long-term linguistic knowledge. However, it is clearly not the end of the story, and even in laboratory tasks, it is clear that long-term knowledge can influence memory for items and their serial order, as seen in semantic grouping effects in word-sequence recall (Kowialiewski et al., in press) and chunking in sentence recall (Baddeley et al., 2009). Thus, the bottom-up approach will only take us so far, even if we confine ourselves to the laboratory. CQ remains the most promising theoretical mechanism connecting and potentially unifying theories of WM and serial order (Figure 8.3a) with related mechanisms in language (Figures 8.3b and 8.3c). Some progress has been made toward understanding the mechanisms of sequencing at phonological and lexical levels, albeit largely focusing on laboratory tasks such as serial recall and nonword repetition. Further progress is likely to require
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
165
166
GRAHAM J. HITCH ET AL.
consideration of semantic, syntactic, pragmatic, and prosodic factors that potentially modulate these mechanisms, and which will likely require a much richer and more realistic account of linguistic representation. Recent developments in machine learning indicate that such models will demand a greater role for long-term statistical structure in the serial processes underpinning language production and comprehension – emergent top-down knowledge that can be acquired only with substantial experience (McClelland et al., 2010; Schwering & MacDonald, 2020). We earlier argued that “chaining” models could be ruled out in explaining serial order in working memory, because of the many problems they face in accounting for ordering errors in serial recall. It has also been argued (Houghton & Hartley, 1995) that chaining leads to unavoidable interference between sequences stored in LTM (for example, between the phonological sequences that comprise familiar words with overlapping subsequences). However, more recent work has demonstrated that, with extensive training, recurrent neural networks (Elman, 1990; Hochreiter & Schmidhuber, 1997; Jordan, 1986) – in some ways resembling compound chaining models – can overcome these limitations, developing rich representations that capture serial structure in their training material and can support STM for serial order (Botvinick & Plaut, 2006; intriguingly the trained STM model shows CQ-like dynamics). Building on these approaches, “large language models,” more recent recurrent architectures with billions of interconnections and trained on huge text corpora, are capable of generating remarkably realistic and meaningful linguistic output incorporating a wide variety of high-level linguistic constraints (Radford et al., 2019, GPT-2; Brown et al. 2020, GPT-3). These developments are beyond the scope of the current chapter, but they suggest that a full account of the role of WM in language may require a more detailed and realistic implementation of long-term linguistic knowledge, based on the statistical properties of language, and a big challenge for the future is to reconcile these mechanisms with the general mechanisms of verbal WM we have focused on here.
8.5
Conclusion
We began by reviewing evidence for a phonological loop in verbal WM involved in immediate serial recall and aspects of language processing that include lexical acquisition, speech production, and comprehension. Overall, our review of computational models of immediate serial recall indicates a central role for serial order in both WM and language. Verbal WM demands a mechanism that can encode novel sequences rapidly, and computational models of this process highlight the importance of CQ, which involves parallel sequence planning and competitive response selection. To account for linguistic constraints on nonword repetition errors and effects of rhythm in serial recall, we argue that this mechanism must also involve a
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
time-varying context signal that is sensitive to linguistic structure. We focused on low-level auditory-temporal structure that is most relevant to STM, but we acknowledge that a fuller account of linguistic structure will involve representations that incorporate the high-level statistical constraints that arise from syntax and semantics but that can only be learned over long-term experience. Language models and WM models have thus tended to occupy different levels of abstraction and empirical scope reflecting the distinct perspectives of psychologists, linguists, cognitive scientists, and, increasingly, AI developers. By targeting the remaining gaps in this space, such as the disjunction between mechanisms of sequencing in large language models and in models of STM, we believe modeling serial order and WM can play a vital role in understanding language processing.
References Ahissar, E., Nagarajan, S., Ahissar, M., Protopapas, A., Mahncke, H., & Merzenich, M. M. (2001). Speech comprehension is correlated with temporal response patterns recorded from auditory cortex. Proceedings of the National Academy of Sciences, 98(23), 13367–13372. Atkinson, R. C. & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 2, pp. 89–195). Academic Press. Baddeley, A. D. (1966a). Influence of acoustic and semantic similarity on long-term memory for word sequences. Quarterly Journal of Experimental Psychology, 18, 302–309. Baddeley, A. D. (1966b). Short-term memory for word sequences as a function of acoustic semantic and formal similarity. Quarterly Journal of Experimental Psychology, 18, 362–365. Baddeley, A. D. (1968). How does acoustic similarity influence short-term memory? The Quarterly Journal of Experimental Psychology, 20, 249–264. Baddeley, A. D. (1986). Working Memory. Oxford University Press. Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4, 417–423. Baddeley, A. D., Allen, R. J., & Hitch, G. J. (2011). Binding in visual working memory: The role of the episodic buffer. Neuropsychologia, 49, 1393–1400. Baddeley, A., Eldridge, M., & Lewis, V. (1981). The role of subvocalization in reading. Quarterly Journal of Experimental Psychology Section a: Human Experimental Psychology, 33, 439–454. Baddeley, A. D., Gathercole, S., & Papagno, C. (1998). The phonological loop as a language learning device. Psychological Review, 105, 158–173. Baddeley, A. D. & Hitch, G. J. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. VIII, pp. 47–90). Academic Press.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
167
168
GRAHAM J. HITCH ET AL.
Baddeley, A. D., Hitch, G. J., & Allen, R. J. (2009). Working memory and binding in sentence recall. Journal of Memory and Language, 61, 438–456. Baddeley, A. D., Lewis, V., & Vallar, G. (1984). Exploring the articulatory loop. Quarterly Journal of Experimental Psychology Section a: Human Experimental Psychology, 36, 233–252. Baddeley, A. D., Papagno, C., & Vallar, G. (1988). When long-term learning depends on short-term storage. Journal of Memory and Language, 27, 586–595. Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and structure of short-term-memory. Journal of Verbal Learning and Verbal Behavior, 14, 575–589. Baddeley, A. D., & Warrington, E. K. (1970). Amnesia and distinction between long-and short-term memory. Journal of Verbal Learning and Verbal Behavior, 9, 176–189. Besner, D., Davies, J., & Daniels, S. (1981). Reading for meaning: The effects of concurrent articulation. Quarterly Journal of Experimental Psychology Section a: Human Experimental Psychology, 33, 415–437. Boomer, D. S. & Laver, J. D. M. (1968). Slips of the tongue. British Journal of Disorders of Communication, 3, 2–12. Botvinick, M. M. & Plaut, D. C. (2006). Short-term memory for serial order: A recurrent neural network model. Psychological Review, 113(2), 201. Brener, R. (1940). An experimental investigation of memory span. Journal of Experimental Psychology, 26, 467–482. Broadbent, D. E. (1958). Perception and Communication. Pergamon Press. Brown, G. D. A., Neath, I., & Chater, N. (2007). A temporal ratio model of memory. Psychological Review, 114, 539–576. Brown, G. D. A., Preece, T., & Hulme, C. (2000). Oscillator-based memory for serial order. Psychological Review, 107, 127–181. Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P.,. . .& Amodei, D. (2020). Language models are few-shot learners. arXiv preprint arXiv:2005.14165. Burgess, N,. & Hitch, G. J. (1992). Towards a network model of the articulatory loop. Journal of Memory and Language, 31, 429–460. Burgess, N,. & Hitch, G. (1999). Memory for serial order: A network model of the phonological loop and its timing. Psychological Review, 106, 551–581. Burgess, N., & Hitch, G. J. (2006). A revised model of short-term memory and long-term learning of verbal sequences. Journal of Memory and Language 55, 627–652. Conrad, R. (1964). Acoustic confusions in immediate memory. British Journal of Psychology, 55, 75–84. Cowan, N. (1995). Attention and memory: An integrated framework. Oxford Psychology Series, no. 26. Oxford University Press. Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671–684.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
Crannell, C. W., & Parrish, J. M. (1957). A comparison of immediate memory span for digits, letters, and words. The Journal of Psychology, 44, 319–327. Crowder, R. G. (1982). The demise of short-term-memory. Acta Psychologica, 50, 291–323. Daneman, M., & Carpenter, P. A. (1980). Individual-differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. Dell, G. S., & Reich, P. A. (1981). Stages in sentence production: An analysis of speech error data. Journal of Verbal Learning and Verbal Behavior, 20(6), 611–629. Ding, N., & Simon, J. Z. (2014). Cortical entrainment to continuous speech: functional roles and interpretations. Frontiers in Human Neuroscience, 8, 311. Drewnowski, A., & Murdock, B. B., Jr. (1980). The role of auditory features in memory span for words. Journal of Experimental Psychology: Learning, Memory, and Cognition, 6, 319–332. Ebbinghaus, H. (1885/1964). Memory: A contribution to experimental psychology. Dover. Ellis, A. W. (1980). Errors in speech and short-term-memory: The effects of phonemic similarity and syllable position. Journal of Verbal Learning and Verbal Behavior, 19, 624–634. Elman, J. L. (1990). Finding structure in time. Cognitive Science, 14, 179–211. Farrell, S. (2006). Mixed-list phonological similarity effects in delayed serial recall. Journal of Memory and Language, 55, 587–600. Farrell, S., Hurlstone, M. J., & Lewandowsky, S. (2013). Sequential dependencies in recall of sequences: Filling in the blanks. Memory & Cognition, 41, 938–952. Farrell, S. & Lewandowsky, S. (2002). An endogenous distributed model of ordering in serial recall. Psychonomic Bulletin & Review, 9, 59–79. Farrell, S., & Lewandowsky, S. (2004). Modelling transposition latencies: Constraints for theories of serial order memory. Journal of Memory and Language, 51, 115–135. Friedmann, N., & Gvion, A. (2001). Sentence comprehension and working memory limitation in aphasia: A dissociation between semantic-syntactic and phonological reactivation. Brain and Language, 86(1), 23–39. Gathercole, S. E., & Baddeley, A. D. (1989). Evaluation of the role of phonological STM in the development of vocabulary in children: A longitudinal study. Journal of Memory and Language, 28, 200–213. Gathercole, S. E., Frankish, C. R., Pickering, S. J., & Peaker, S. (1999). Phonotactic influences on short-term memory. Journal of Experimental Psychology-Learning Memory and Cognition, 25, 84–95. Gathercole, S. E., Willis, C. S., Emslie, H., & Baddeley, A. D. (1992). Phonological memory and vocabulary development during the early school years: A longitudinal study. Developmental Psychology, 28, 887–898.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
169
170
GRAHAM J. HITCH ET AL.
Glanzer, M., & Cunitz, A. R. (1966). Two storage mechanisms in free recall. Journal of Verbal Learning and Verbal Behavior, 5, 351–360. Glasspool, D. W. (2005). Modelling serial order in behaviour: Evidence from performance slips. In G. Houghton (Ed.), Connectionist models in cognitive psychology (pp. 241–270). Psychology Press. Graf Estes, K., Evans, J. L., & Else-Quest, N. M. (2004). Differences in the nonword repetition performance of children with and without specific language impairment: A meta-analysis. Journal of Speech and Hearing Research, 50, 177–195. Gregg, V. H., Freedman, C. M., & Smith, D. K. (1989). Word-frequency, articulatory suppression and memory span. British Journal of Psychology, 80, 363–374. Grossberg, S. (1978a). A theory of human memory: Self-organization and performance of sensory-motor codes, maps, and plans. In R. Rosen & Snell (Eds.), Progress in theoretical biology (Vol. 5, pp. 233–374). New York: Academic Press. Grossberg, S. (1978b). Behavioral contrast in short-term memory: Serial binary memory models or parallel continuous memory models? Journal of Mathematical Psychology, 17, 199–219. Grossberg, S., & Pearson, L. R. (2008). Laminar cortical dynamics of cognitive and motor working memory, sequence learning and performance: Towards a unified theory of how the cerebral cortex works. Psychological Review, 115, 677–732. Gupta, P. & MacWhinney, B. (1997). Vocabulary acquisition and verbal short-term memory: Computational and neural bases. Brain and Language, 59(2), 267–333. Gvion, A. & Friedmann, N. (2012). Does phonological working memory impairment affect sentence comprehension? A study of conduction aphasia. Aphasiology, 26(3-4), 494–535. Hartley, T. (2002). Syllabic phase: A bottom-up representation of the temporal structure of speech. In J. Bullinaria & W. Lowe (Eds.), Connectionist models of cognition and perception (pp. 277–288). World Scientific Publishing Co. Hartley, T., & Houghton, G. (1996). A linguistically constrained model of short-term memory for nonwords. Journal of Memory and Language, 35, 1–31. Hartley, T., Hurlstone, M. J., & Hitch, G. J. (2016). Effects of rhythm on memory for spoken sequences: A model and tests of its stimulus-driven mechanism. Cognitive Psychology, 87, 135–178. Henson, R. N. A. (1996). Short-term memory for serial order (Doctoral dissertation, University of Cambridge). Henson R. N. A. (1998). Short-term memory for serial order: The start-end model. Cognitive Psychology, 36, 73–137. Henson, R. N. A., Norris, D. G., Page, M. P. A., & Baddeley, A. D. (1996). Unchained memory: Error patterns rule out chaining models of immediate serial recall. Quarterly Journal of Experimental Psychology, 49A, 80–115.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
Hitch, G. J. (1978). Role of short-term working memory in mental arithmetic. Cognitive Psychology, 10, 302–323. Hitch, G. J., Burgess, N., Towse, J. N., & Culpin, V. (1996). Temporal grouping effects in immediate recall: A working memory analysis. Quarterly Journal of Experimental Psychology, 49A, 116–139. Hochreiter, S. & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780. Houghton, G. (1990). The problem of serial order: A neural network model of sequence learning and recall. In R. Dale, C. Mellish, & M. Zock, (Eds.), Current research in natural language generation (pp. 287–319). Academic Press. Houghton, G., & Hartley, T. (1995). Parallel models of serial behavior: Lashley revisited. Psyche, 2(25), 1–25. Hulme, C., Maughan, S., & Brown, G. D. A. (1991). Memory for familiar and unfamiliar words: Evidence for a long-term-memory contribution to short-term-memory span. Journal of Memory and Language, 30, 685–701. Hurlstone, M. J. (2021). Serial recall. In M. J. Kahana & A. D. Wagner (Eds.), The Oxford handbook of human memory. Oxford University Press. Hurlstone, M. J., Hitch, G. J., & Baddeley, A. D. (2014). Memory for serial order across domains: An overview of the literature and directions for future research. Psychological Bulletin, 140, 339–373. Jalbert, A., Neath, I., Bireta, T. J., & Surprenant, A. M. (2011). When does length cause the word length effect? Journal of Experimental Psychology: Learning Memory and Cognition, 37, 338–353. Jordan, M. I. (1986). Serial order: A parallel distributed processing approach. Finding structure in time. Cognitive Science, 14, 179–211. Kahana, M. J. (2012). Foundations of human memory. Oxford University Press. Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W., & Engle, R. W. (2004). The generality of working memory capacity: A latentvariable approach to verbal and visuospatial memory span and reasoning. Journal of Experimental Psychology: General, 133, 189–217. Kowialiewski, B., Gorin, S., & Majerus, S. (2021). Semantic knowledge constrains the processing of serial order information in Working Memory. Journal of Experimental Psychology: Learning, Memory and Cognition, 12, 1958–1970. Lewandowsky, S., & Farrell, S. (2008). Short-term memory: New data and a model. The Psychology of Learning and Motivation, 49, 1–48. Lewandowsky, S., & Murdock, B. B. (1989). Memory for serial order. Psychological Review, 96, 25–57. Lewandowsky, S., Oberauer, K., & Brown, G. D. A. (2009). No temporal decay in verbal short-term memory. Trends in Cognitive Sciences, 13, 120–126. Lewis, R. L., Vasishth, S., & Van Dyke, J. A. (2006). Computational principles of working memory in sentence comprehension. Trends in Cognitive Sciences, 10, 447–454. Logan, G. D. (2021). Serial order in perception, memory, and action. Psychological Review, 128(1), 1–44.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
171
172
GRAHAM J. HITCH ET AL.
Logie, R. H. (1995). Visuo-spatial working memory. Erlbaum. Logie, R. H., Camos, V., & Cowan, N. (2021). Working memory: State of the science. Oxford University Press. MacDonald, M. E. (2016). Speak, act, remember: The language-production basis of serial order and maintenance in verbal memory. Current Directions in Psychological Science, 25(1), 47–53. MacKay, D. G. (1970). Spoonerisms: Structure of errors in serial order of speech. Neuropsychologia, 8, 323–350. Madigan, S. A. (1971). Modality and recall order interactions in short-term memory for serial order. Journal of Experimental Psychology, 87, 294–296. Marr, D. (1982). Vision: A computational approach. Freeman & Co. Martin, N., & Saffran, E. M. (1997). Language and auditory-verbal short-term memory impairments: Evidence for common underlying processes. Cognitive Neuropsychology, 14, 641–682. Martin, R. C., & Slevc, L. R. (2014). Language production and working memory. In M. Goldrick, V. Ferreira, & M. Miozzo (Eds.), The Oxford handbook of language production (pp. 437–450). Oxford University Press. Maybery, M. T., Parmentier, F. B. R., & Jones, D. M. (2002). Grouping of list items reflected in the timing of recall: Implications for models of serial verbal memory. Journal of Memory and Language, 47, 360–385. McClelland, J. L., Botvinick, M. M., Noelle, D. C., Plaut, D. C., Rogers, T. T., Seidenberg, M. S., & Smith, L. B. (2010). Letting structure emerge: Connectionist and dynamical systems approaches to cognition. Trends in Cognitive Sciences, 14(8), 348–356. Melby-Lervag, M. & Lervag, A. (2012). Oral language skills moderate nonword repetition skills in children with dyslexia: A meta-analysis of the role of nonword repetition skills in dyslexia. Scientific Studies of Reading, 16, 1–34. Miller, G. A. (1956). The magical number 7, plus or minus 2: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Monsell, S. (1987). On the relation between lexical input and output pathways for speech. In A. Allport, D. G. MacKay, W. Prinz, & E. Scheerer (Eds.), Cognitive science series. Language perception and production: Relationships between listening, speaking, reading and writing (pp. 273–311). Academic Press. Morey, C. C. (2018). The case against specialized visual-spatial short-term memory. Psychological Bulletin, 144, 849–883. Murdock, B. B. (1993). TODAM2: A model for the storage and retrieval of item, associative, and serial-order information. Psychological Review, 100, 183–203. Murdock, B. B. (1995). Developing TODAM: Three models for serial order information. Memory & Cognition, 23, 631–645. Murray, D. J. (1968). Articulation and acoustic confusability in short-term memory. Journal of Experimental Psychology, 78, 679–684.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
Computational Models
Nelson, T. O., & Rothbart, R. (1972). Acoustic savings for items forgotten from long-term memory. Journal of Experimental Psychology, 93(2), 357–360. Nooteboom, S. G. (1973). The tongue slips into patterns. In V. A. Fromkin (Ed.), Speech errors as linguistic evidence (pp. 144–156). Mouton. Norris, D. (2017). Short-term memory and long-term memory are still different. Psychological Bulletin, 143, 992–1009. Oberauer, K. (2002). Access to information in working memory: Exploring the focus of attention. Journal of Experimental Psychology: Learning Memory and Cognition, 28, 411–421. Osth, A. F., & Hurlstone, M. J. (2021). Do item-based context representations underlie serial order in cognition? Commentary on Logan (2021). Psychological Review. Page, M. P. A., Madge, A., Cumming, N., & Norris, D. G. (2007). Speech errors and the phonological similarity effect in short-term memory: Evidence suggesting a common locus. Journal of Memory and Language, 56, 49–64. Page, M. P. A., & Norris, D. (1998). The primacy model: A new model of immediate serial recall. Psychological Review, 105, 761–781. Papagno, C., Valentine, T., & Baddeley, A. (1991). Phonological short-term memory and foreign-language vocabulary learning. Journal of Memory and Language, 30, 331–347. Papagno, C., & Vallar, G. (1992). Phonological short-term-memory and the learning of novel words: The effect of phonological similarity and item length. Quarterly Journal of Experimental Psychology Section A: Human Experimental Psychology, 44, 47–67. Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I. (2019). Language models are unsupervised multitask learners. OpenAI blog, 1(8), 9. Rogalsky, C., Matchin, W., & Hickok, G. (2008). Broca’s area, sentence comprehension, and working memory: An fMRI study. Frontiers in Human Neuroscience, 2: 14. doi: 10.3389/neuro.09.014.2008 Ryan, J. (1969a). Grouping and short-term memory: Different means and patterns of grouping. Quarterly Journal of Experimental Psychology, 21, 137–147. Ryan, J. (1969b). Temporal grouping rehearsal and short-term memory. Quarterly Journal of Experimental Psychology, 21, 148–155. Schweickert, R. (1993). A multinomial processing tree model for degradation and redintegration in immediate recall. Memory & Cognition, 21, 168–175. Schwering, S. C., & MacDonald, M. C. (2020). Verbal working memory as emergent from language comprehension and production. Frontiers in Human Neuroscience, 14, 68. doi: 10.3389/fnhum.2020.00068 Selkirk, E. (1984). On the major class features and syllable theory. In M. Aronoff & R. T. Orehrle (Eds.), Language sound structure: Studies in phonology presented to Morris Halle by his teachers and students. MIT press.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
173
174
GRAHAM J. HITCH ET AL.
Service, E. (1992). Phonology, working memory, and foreign-language learning. Quarterly Journal of Experimental Psychology Section A: Human Experimental Psychology, 45, 21–50. Service, E. (1998). The effect of word length on immediate serial recall depends on phonological complexity, not articulatory duration. Quarterly Journal of Experimental Psychology Section A: Human Experimental Psychology, 51, 283–304. Shallice, T., & Butterworth, B. (1977). Short-term-memory impairment and spontaneous speech. Neuropsychologia, 15, 729–735. Shallice, T., & Warrington, E. K. (1970). Independent functioning of verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261–273. Shattuck-Huffnagel, S. (1979) Speech errors as evidence for a serial-ordering mechanism in sentence production. In W. E Cooper & E. C. T Walker, (Eds.), Sentence processing: Psycholinguistic studies presented to Merrill Garrett. Erlbaum. Shulman, H. G. (1972). Encoding and retention of semantic and phonemic information in short-term memory. Journal of Verbal Learning and Verbal Behavior, 9, 499–508. Solway, A., Murdock, B. B., & Kahana, M. J. (2012). Positional and temporal clustering in serial order memory. Memory & Cognition, 40, 177–190. Surprenant, A. M., Kelley, M. R., Farley, L. A., & Neath, I. (2005). Fill-in and infill errors in order memory. Memory, 13, 267–273. Treiman, R., & Danis, C. (1988). Short-term memory errors for spoken syllables are affected by the linguistic structure of the syllables. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(1), 145. Vallar, G., & Baddeley, A. (1987). Phonological short-term store and sentence processing. Cognitive Neuropsychology, 4, 417–438. Vallar, G., & Papagno, C. (2002). Neuropsychological Impairments of verbal short-term memory. In A. D. Baddeley, M. D. Kopelman, & B. A. Wilson (Eds.), Handbook of memory disorders (2nd. ed., pp. 249–270). Wiley. Vousden, J. I., Brown, G. D., & Harley, T. A. (2000). Serial control of phonology in speech production: A hierarchical model. Cognitive Psychology, 41 (2), 101–175. Yang, T. X., Allen, R. J., & Gathercole, S. E. (2016). Examining the role of working memory resources in following spoken instructions. Journal of Cognitive Psychology, 28, 186–198. Ylinen, S., Nora, A., Leminen, A., Hakala, T., Huotilainen, M., Shtyrov, Y., Mäkelä, J. P., & Service, E. (2015). Two distinct auditory-motor circuits for monitoring speech production as revealed by content-specific suppression of auditory cortex. Cerebral Cortex, 25(1), 1576–1586.
https://doi.org/10.1017/9781108955638.011 Published online by Cambridge University Press
9 The Time-Based Resource Sharing Model of Working Memory for Language Valérie Camos and Pierre Barrouillet 9.1
Introduction
In the same way as language has been studied since the first steps of scientific psychology in Wundt’s laboratory at the end of the nineteenth century, as Garnham et al. (2006) recalled, the first model of working memory (WM) proposed by Baddeley and Hitch (1974) in their seminal work was a model of verbal WM. Accordingly, the first task for measuring WM capacity, the reading span task designed by Daneman and Carpenter (1980), involved reading aloud unconnected sentences while remembering their last word. This does not come as a surprise when considering the importance of language for the human mind. Although the strongest version of linguistic determinism through the Sapir-Whorf hypothesis (Whorf, 1940) has been abandoned (Berlin & Kay, 1969), the relationships between thought and language have always been investigated by psychologists (e.g., Piaget, 1923; Vygotsky, 1934), and several theories have suggested linguistic and syntactic bases for the highest human cognitive functions such as reasoning and logic (Braine, 1990; Rips, 1994). Thus, due to the inherently sequential nature of language, a WM understood as a system dedicated to the temporary maintenance of verbal information in the purpose of its online processing should necessarily be a key function of the human mind. However, designing such a WM and its role in language acquisition, production and comprehension proved to be fraught with difficulties. The well-known solution proposed by Baddeley and Hitch (1974) was to replace the unitary short-term store that Atkinson and Shiffrin (1968) conceived of as a WM by a three-component mnemonic structure in which, along with a central executive responsible for the attentional control and a visuospatial sketchpad devoted to the maintenance and processing of visuospatial information, a system called the phonological loop was dedicated to the maintenance of speech-based information, and possibly purely acoustic information. Though the central executive was initially assumed
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
176
VALÉRIE CAMOS AND PIERRE BARROUILLET
by Baddeley and Hitch (1974) to involve some storage capacity, this hypothesis was subsequently abandoned and the storage of verbal information was exclusively devoted to the phonological loop (Baddeley, 1986). Because Baddeley’s model of WM was an important source of inspiration for our own model, the Time-Based Resource-Sharing model (TBRS;, Barrouillet & Camos, 2015, 2020) that includes a system akin to the phonological loop, we will first briefly present how Baddeley conceived verbal WM in order to make understandable in which way our conceptions differ from this inspirational proposal.
9.2
The Phonological Loop in Baddeley’s Model
When testing Atkinson and Shiffrin’s (1968) hypothesis that the short-term store of their model could be considered as a WM in charge of maintaining and processing information for ongoing cognition, Baddeley and Hitch (1974) observed that maintaining a verbal memory preload could have no measurable effect on concurrent cognitive activities (see also Baddeley and Lewis in a study reported by Baddeley, 1986). This meant that the system responsible for processing activities should be distinguished from the store involved in immediate serial recall. This latter system was conceived as a store in which verbal items are deposited in their phonological form and an articulatory rehearsal mechanism. The phonological traces that suffer from temporal decay would be recirculated into the store and hence reactivated through their overt or covert articulation. Because articulation is necessarily a sequential process, the amount of information that can be preserved from forgetting in the store is a function of the rates of both decay and articulation, short-term memory span corresponding to the number of items that can be articulated within the lifespan of a given item before its complete loss. As we will see in the following section, several of these ideas have been integrated in the TBRS model. Experimental and neuropsychological evidence was gathered by Baddeley and his collaborators to support the role of the phonological loop on language acquisition and comprehension as well as second language learning (Adams & Gathercole, 1996; Gathercole & Baddeley, 1989, 1990; Papagno et al., 1991; Vallar & Baddeley, 1987). Despite this experimental and neuropsychological evidence, Baddeley (2007) acknowledged that the decision to deprive the central executive of storage capacity led to a series of theoretical problems, what he pleasantly called “skeletons in the working memory cupboard” (Baddeley, 2007, p. 141). Among these skeletons, two were of particular interest for language. The first is that even when the use of the phonological loop is prevented by a concurrent articulation, individuals are still able to recall a substantial amount of verbal information, suggesting that there is another memory system able to store information in the short term. The second concerns memory for sentences. It is known
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
that individuals are typically able to recall about five unrelated words (Dempster, 1981), but this span increases to 15 when words form a sentence (Baddeley et al., 1987). This suggests that the system able to supplement the hypothesized phonological loop stores not only phonological, but also semantic features. The well-known solution adopted by Baddeley (2000) was to add the episodic buffer as a new component to the multicomponent model. This episodic buffer, conceived as a limited capacity store integrating multimodal information into coherent episodes, was assumed to bind information from perceptual inputs, the phonological loop, the visuospatial sketchpad, and long-term memory (LTM) under the control of the central executive. Although the hypothesis of the episodic buffer is appealing and solves several problems that the previous versions of the multicomponent model encountered, it also inherits some of the weaknesses of the different versions of this model and raises new questions about the relationships between the different stores. First, most of the questions investigated by Baddeley and his colleagues concerning the episodic buffer were about the binding process (e.g., Allen et al., 2006, 2009; Jefferies et al., 2004). However, important questions remain pending about the mechanisms by which information would be maintained and forgetting counteracted within the episodic buffer, about the relationships between these mechanisms and the central executive, and more generally about the relationships between the functions of storage and processing, which are yet distinctive of the concept of WM. Second, assuming that the episodic buffer integrates information from the slave systems, and more precisely from the phonological loop, poses the problem of the relationships between the two stores and more importantly of a possible multiplicity of simultaneous phonological representations for the same verbal item, one in the phonological loop, the other in the episodic buffer, potentially enriched by additional syntactic, orthographic, or semantic features. In the following section, after having presented the main tenets of the TBRS model, we give an overview of the answers that this model brings to these questions.
9.3
The Time-Based Resource Sharing Model
The first insight toward the TBRS model was provided by observations drawn from complex span tasks in which children had to maintain series of letters for further immediate serial recall while concurrently performing either a simple articulatory suppression (repeating the syllable “ba”) or the verification of arithmetic problems (6 + 3 + 8 = 18, true or false? Barrouillet & Camos, 2001). It appeared, as we expected, that the arithmetic task had a more detrimental effect on recall performance than a mere articulatory suppression, but this effect was far from dramatic. This finding suggested first that processing and storage share a common resource in WM, the
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
177
178
VALÉRIE CAMOS AND PIERRE BARROUILLET
resource allocated to processing being no longer available for storage, but that participants managed in some way or another to maintain memory traces while solving the equations, explaining why the effect was not so large as we expected. The solution was to imagine a WM in which the common resource is shared between processing and storage on a temporal basis in such a way that maintenance is admittedly disrupted by a concurrent activity but still possible to some extent. The TBRS model assumes that the common resource shared by processing and storage is attention. Working memory traces on which attention is focused receive activation, but as soon as attention is diverted by a concurrent activity, they suffer from a temporal decay. However, if the concurrent activity can be surreptitiously interrupted at least for short periods of time, attention could be redirected into the memory traces for their refreshing, hence avoiding their complete loss. Thus, in a complex span task in which each memory item (e.g., letters, digits, words) is followed by a concurrent activity (e.g., reading sentences, solving arithmetic problems), recall performance should be a function of the ratio between the time during which this concurrent activity occupies attention and memory traces suffer from temporal decay, and the time during which attention can be freed to refresh these traces and avoid their forgetting. In other words, the amount of information that can be held in WM should be a function of the proportion of time during which concurrent activities occupy attention, preventing the refreshing of memory traces, a proportion called cognitive load (CL). The prediction that, in complex span tasks, memory performance depends on the CL of the processing component was verified in several studies (e.g., Barrouillet et al., 2004, 2007; see Barrouillet & Camos, 2015, for review). It appeared that, still in line with the TBRS predictions, what matters for recall performance is not the nature of the concurrent activity, but its duration (Barrouillet et al., 2007, 2011). Thus, provided that temporal factors are controlled, a visuospatial task has the same detrimental effect on verbal memory as a numerical task (Barrouillet et al., 2007), in the same way as a verbal task has the same detrimental effect on the maintenance of visuospatial information as a visuospatial task (Vergauwe et al., 2010). These different studies confirmed the main tenet of the TBRS model: Time is the fundamental dimension of WM functioning. At the structural level, the model assumes that information is maintained in WM as multimodal representations within a buffer akin to the episodic buffer postulated by Baddeley (2000). These representations would integrate sensory inputs, information retrieved from LTM, and the product of cognitive operations on previously created WM representations. The capacity of this buffer would be limited to about four representations, with attention focusing at any time on only one of these representations. Following Anderson’s (1993) conception of the production systems, the content of the focused representation would be continuously read by a procedural system that selects, depending on the current goal, the most
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
appropriate production that runs through its set of actions. These actions can have for effect either to transform this representation, to select another representation, to protect this representation from degradation if it must be maintained, or to send appropriate commands for motor response. Thus, the episodic buffer along with the procedural system constitute an executive loop that is the keystone of the controlled cognitive activity. Because production firing is sequential, the cognitive operations oriented toward the preservation and maintenance of WM content, and those that aim to modify it for processing purpose compete for occupying the executive loop, cognitive functioning taking the form of a rapid switching between storage and processing. We have said that the representations held in the episodic buffer are multimodal. Indeed, several studies led us to the conclusion that there is no specific system like a visuospatial sketchpad dedicated to the maintenance of visual or spatial information (Vergauwe, Barrouillet, & Camos, 2009; Vergauwe, Camos, & Barrouillet, 2009). Vergauwe et al. (2010) demonstrated that, contrary to the conclusions drawn from the use of the selective interference paradigm in Baddeley and colleagues’ studies (see Baddeley, 2007, for review), verbal and visuospatial activities have the same detrimental effect on visuospatial memory when temporal aspects of the tasks are carefully controlled. In the same way, attention-demanding tasks, even visuospatial in nature, proved to have a detrimental effect on verbal maintenance, suggesting that attention plays a crucial role in verbal maintenance (Barrouillet et al., 2007, 2011). However, though the hypothesis of a specific system for visuospatial maintenance can be dismissed, this not the case for the verbal domain. Indeed, while Vergauwe et al. (2010) observed that verbal and visuospatial tasks had the same effect on the concurrent maintenance of visuospatial information, verbal memory proved more disrupted by verbal than visuospatial concurrent tasks. Accordingly, and contrary to what happens with visuospatial memory, we observed that individuals are able to maintain up to four letters without any attentional involvement (Vergauwe et al., 2014). Thus, it seems that along with the executive loop, there is another system for the maintenance of verbal information in WM.
9.4
Two Systems of Maintenance for Verbal Information
Besides the executive loop conceived by the TBRS model as a domaingeneral system in charge of the short-term maintenance of information, another loop would be specifically dedicated to the maintenance of verbal information, the phonological loop. Following Baddeley’s (1986) multicomponent model, we initially assumed that this phonological loop can maintain verbal information in its phonological format through an articulatory rehearsal mechanism that uses language-based processes to recirculate
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
179
180
VALÉRIE CAMOS AND PIERRE BARROUILLET
Figure 9.1 The cognitive architecture of working memory according to the TBRS model Note: Multimodal representations constructed from inputs provided by perception, longterm memory knowledge and current goals are held in an episodic buffer and continuously read and transformed by a production system in what is called an executive loop. Additionally, a phonological loop would be able to maintain some verbal items through articulatory rehearsal. See main text for a functional account of the system and the hypothesis that what was initially understood as a phonological loop might simply be an articulatory loop. Source: Barrouillet, P., & Camos, V. (2015). Working memory: Loss and reconstruction. Psychology Press. Reproduced with permission
verbal information either overtly or covertly through inner speech. The cognitive structure resulting from the addition of the phonological loop is depicted in Figure 9.1. The structure and functioning of WM as proposed by the TBRS model have particular consequences for language, and more specifically for the maintenance of verbal information. Because the phonological loop and the executive loop are hypothesized as two distinct systems, it should be possible to tease apart their respective impact on verbal maintenance and observe signs of their independence. In a first attempt to examine the involvement of these systems in verbal maintenance, Camos et al. (2009, Exp. 3) orthogonally manipulated their availability in complex span tasks. They had participants maintain series of letters, each letter being followed by a series of digits sequentially displayed on screen. To vary the availability of the executive loop, participants were asked to perform on these digits either a choice reaction time task (CRT, a
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
parity judgment) or a simple reaction time task (SRT, pressing a key as fast as possible when each digit appeared). The parity judgment task was assumed to capture attention and occupy the executive loop for a longer period of time than the SRT task. To vary the availability of the phonological loop, participants had to perform each of the concurrent tasks (the CRT and the SRT tasks) either silently through key presses or aloud, the latter condition preventing the use of articulatory rehearsal. As was expected, both increasing the attentional demand through a concurrent CRT task and introducing a concurrent articulation reduced recall performance. The novelty lay in the absence of interaction, the effect of the two manipulations being additive, supporting the hypothesis that two independent systems were involved in the maintenance of letters. Several experiments replicated these findings (Camos et al., 2009, Exp. 1, 2, & 4) that have been reported by other studies (Hudjertz & Oberauer, 2007). It should also be noted that some brain-imaging studies bring evidence in favor of the independence between a language-based and an attentionbased verbal maintenance mechanism (Gruber, 2001; Raye et al., 2007; Smith & Jonides, 1999). In the same vein, clinical examinations of patients with brain lesions led to descriptions of double-dissociations, bringing further support to the existence of two distinct systems (Trost & Gruber,2012). The independence of the two maintenance systems opens the possibility for individuals to make choice on the mechanism they favor for maintaining verbal information in WM. More specifically, it should be expected that depending on the constraints of the task, participants should make a different choice on which mechanism to use. Indeed, as previously said, articulatory rehearsal and attentional refreshing can be impaired by different factors, a concurrent articulation and a concurrent attentional demand, respectively. Hence, an adaptive choice should lead participants to favor the executive loop when they are under concurrent articulation and the phonological loop (and its poorly attention-demanding rehearsal) when a demanding concurrent task has to be achieved concurrently. This was exactly what Camos et al. (2011) observed in young adults performing complex span tasks in which they maintained series of words. It should be noted that the use of one or the other of the two systems can also be induced through verbal instructions to either repeat overtly or covertly the words for the phonological loop, or to think of the words for the executive loop. Each of these sets of instructions led to the same pattern of results as when experimental manipulations hinder the use of the other system (Camos et al., 2011).
9.5
The Impact of Linguistic Characteristics on Short-Term Maintenance
As mentioned previously, the two systems described by the TBRS model differ on the reactivation mechanism and the resource they are relying on. The phonological loop and its articulatory rehearsal involve language-based
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
181
182
VALÉRIE CAMOS AND PIERRE BARROUILLET
processes, whereas the executive loop and its attentional refreshing depend on attention. As a consequence, well-known phonological effects described for impacting recall from WM may depend on the involvement of the phonological loop, and should be immune to any variation of the implication of the executive loop. Two famous examples of such effects are the phonological similarity effect (PSE) and the word length effect (WLE), which lead to reduced recall performance for lists of phonologically similar words and of long words compared to dissimilar and short words, respectively. In parallel studies, Camos et al. (2013) and Mora and Camos (2013) tested the predictions that PSE and WLE were dependent on the phonological loop and independent from the executive loop. In complex span tasks, they asked participants to memorize series of words that differed either on their phonological similarity or length, each word being followed by a delay of some seconds. Moreover, the availability of each maintenance system was orthogonally manipulated. For varying the implication of the executive loop, the interword intervals were either filled with a location judgment task intended to distract attention from maintenance, or remained empty, hence freeing attention. The involvement of the phonological loop was made possible by asking participants to remain silent during the interword intervals or was impaired by asking them to concurrently and continuously utter the one-syllable word oui (“yes” in French). First, these studies replicated the absence of interaction between manipulations targeting each system. More importantly for the purpose of these studies, results showed that both PSE and WLE were not modulated by the presence or the absence of an attention-demanding concurrent task, but disappeared under concurrent articulation. This latter finding suggested that, as we predicted, these phonological effects are under the dependence of the use of the phonological loop. Blocking this loop with a concurrent articulation made these effects disappear, although recall was still possible (recall performance was not at zero), thanks to the maintenance of the words within the executive loop. Besides the phonological characteristics of the memoranda, other linguistics features of the words have been examined to assess their impact on short-term recall and to help understand WM functioning. Factors such as word frequency, familiarity, and lexicality are known for affecting the ease of retrieval from LTM. For example, high-frequency words are better and faster retrieved from LTM than low-frequency words (e.g., Morton, 1979). The impact of several of these factors has been examined in WM, and generally, those that favor retrieval from LTM also improve recall in WM tasks (e.g., Engle et al., 1990; Loaiza, Duperreaultm et al., 2015; Loaiza, Rhodes et al., 2015). Understanding these effects is still a challenge, and several of them have been used to test different accounts of attentional refreshing. Although it is now clear that the executive loop relies on attention for maintaining information in WM, different proposals have been put forward
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
about the precise nature and functioning of its maintenance mechanism, attentional refreshing (see Camos et al., 2018, for review). While some authors suggest that refreshing can be conceived as briefly thinking of one or more previously activated representations by bringing them into the focus of attention (Cowan, 1995, 1999; Johnson, 1992; Johnson et al., 2002; Raye et al., 2007), others have described refreshing as a covert retrieval from LTM (Loaiza & McCabe, 2012, 2013; McCabe, 2008), a reconstruction (Barrouillet & Camos, 2015), or an elaboration (Hudjetz & Oberauer, 2007; Loaiza & McCabe, 2012) of WM traces using knowledge stored in LTM. These different conceptions of refreshing can be categorized according to the role they attribute to LTM knowledge. Thus, examining how this knowledge and its availability impacts the functioning of attentional refreshing should help us decide between these different accounts. For this purpose, Camos et al. (2019) examined how word frequency and lexicality of memory items affect performance in complex span and Brown-Peterson tasks.1 In the former task, not surprisingly, varying the availability of refreshing as well as frequency or lexicality impacted recall performance. However, and contrary to what might be expected from accounts suggesting that refreshing relies in some sort on LTM, none of the refreshing manipulations interacted with the LTM effects, Bayesian analysis supporting the absence of such interactive effects (see Abadie & Camos, 2018, for similar finding). Further experiments confirmed this finding. Rosselet-Jordan et al. (in press) reported a similar absence of interaction between manipulation of refreshing and the level of relatedness of memory words (i.e., series of words that share a common gist, like rabbit, carrot, ear) on recall performance. These authors replicated on recall performance this finding and extended it to the examination of recall latency. To examine this issue through more fine-grained measure, Camos et al. (2019) proposed to use a Brown-Peterson paradigm that allows an estimate of the speed of refreshing by measuring the increase in refreshing duration with the number of memory items. Using this method, Camos et al. (2019) showed that refreshing speed was not impacted by the frequency or lexicality of memory items. On the contrary, it was similar for easy- and difficult-to-retrieve memory items (high- and low-frequency words, respectively) as well as for words and nonwords. In the same vein, Campoy et al. (2015) examined the impact of concreteness effect (i.e., concrete words are easier to recall than abstract words) in an immediate serial recall task in which they varied the availability of attentional resources by asking participants to concurrently perform either a simple tapping or a highly demanding random tapping task. Although the introduction of a higher concurrent attentional demand reduced recall performance, it did not modulate the concreteness effect. All these results disconfirm theoretical views that describe refreshing as a reconstruction or an elaboration based on LTM knowledge, and better fit with the idea that refreshing acts through an attentional reactivation of memory traces.
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
183
184
VALÉRIE CAMOS AND PIERRE BARROUILLET
However, some results rendered this picture slightly more complex. Indeed, while the time needed for refreshing memory items does not vary with their relative retrievability from LTM, things are rather different when examining items for which there is no previously stored knowledge in LTM. For example, Ricker et al. (2010) reported that impairing refreshing has an increasingly negative impact across longer retention intervals for English letters, but not for unfamiliar visual stimuli. Similarly, using a BrownPeterson task to assess refreshing speed, Vergauwe et al. (2014) showed that, contrary to what was observed with letters or words, the time needed to refresh a set of unfamiliar fonts did not vary with set size. To conclude, these findings are consistent with the hypothesis that the executive loop and its maintenance mechanism proceed through the rapid switching of attention from one representation to the other to enhance their level of activation. In such a view, so-called LTM effects (word frequency, lexicality, concreteness, or relatedness) on recall performance would originate from the encoding stage. Items for which information is easier to retrieve in LTM would be better encoded into WM, or as we proposed in the TBRS model, their mental representations would be easily constructed in the episodic buffer. By contrast, the representation of items for which there is no available LTM knowledge (unfamiliar visual stimuli or letter fonts) might be more difficult to construct, more volatile, and unfit to any form of refreshing.
9.6
The Impact of the Two Maintenance Systems on LTM Learning of Verbal Information
The existence of two distinct mechanisms for maintaining verbal information at short term has also implications for the long-term storage of this information, in other words, for learning. Using a design similar to the one implemented by Camos et al. (2019) in which the availability of the two systems was orthogonally manipulated in a complex span task, Camos and Portrat (2015) tested LTM learning with a delayed recall test performed after a distracting task (e.g., counting backward by threes starting from a 3-digit number) that erased all short-term memory traces. Varying the availability of the articulatory rehearsal had no impact on recall performance in this delayed test, whereas the impairment of attentional refreshing reduced it (see Loaiza & McCabe, 2013, for similar findings). Such a contrasting pattern of findings brings additional evidence on the separation between the two maintenance systems. This also echoes some early conceptions that distinguished Type I and Type II processing, also named maintenance rehearsal and coding rehearsal to account for the transfer of information from short- to long-term retention systems, well before Baddeley’s work and depiction of WM as a multicomponent system (Craik & Watkins, 1973; Glenberg et al., 1977;
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
Woodward et al., 1973). As articulatory rehearsal is described nowadays, maintenance rehearsal was considered as the rote repetition or recirculation of verbal information at a phonemic level. It was also assumed that it had only transitory effects on verbal maintenance (although some benefit was observed on recognition tests; Greene, 1987). By contrast, coding rehearsal was depicted as a deeper and more elaborative maintenance mechanism, which can support long-term retention. Such a conception is not far from some recent proposals about refreshing, which would be acting as an elaborative rehearsal (Hudjetz & Oberauer, 2007; Loaiza & McCabe, 2012). Beside the respective impact of the two maintenance systems on LTM learning, Abadie and Camos (2019) have shown that these systems are also involved in different ways in the emergence of false memory at short and long term. By integrating the Fuzzy Trace Theory (FTT; Brainerd & Reyna, 2002, 2005) and the TBRS model, Abadie and Camos (2019) propose a new theoretical account of the relationships between WM and LTM. One main tenet of this model is that articulatory rehearsal and attentional refreshing are manipulating qualitatively different types of representations. According to the studies we presented in the previous section, articulatory rehearsal is dealing with phonological representations. Hence, according to the FTT, these representations can be seen as verbatim memories, that is, as representations of the surface forms that cannot be maintained over the long term (e.g., Reyna & Brainerd, 1995). Moreover, because it has been shown that articulatory rehearsal emphasizes the phonological, but not the semantic, characteristics of the memory words (Loaiza & Camos, 2018; Rose et al., 2014, 2015), Abadie and Camos (2019) suggested that rehearsal was not appropriate for maintaining gist memories that represent the meaning of memory traces. By contrast, it was assumed that attentional refreshing can maintain both verbatim and gist representations. False memories are resulting from the retrieval of a gist representation that is not counteracted by the retrieval of a verbatim representation of the memory items. For example, the gist representation of the items forest, leaf, trunk, and lumberjack can lead to the false memory of having been presented with the item tree, except if the verbatim form of the presented words (their phonological or alphabetic images) is still available. Thus, Abadie and Camos’s (2019) model predicts that articulatory rehearsal should prevent the short-term semantic distortion of memory items by enhancing the retrieval of verbatim traces for these items. However, because articulatory rehearsal does not impact long-term learning, such an effect should not be efficient at long term. By contrast, the use of attentional refreshing for maintaining memory words should increase the risk of long-term false memories, because it favors the retrieval of gist memory. This is exactly what was observed in a series of experiments, showing that the different systems described by the TBRS model for maintaining verbal information at short term underlie also false memories at short and long term.
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
185
186
VALÉRIE CAMOS AND PIERRE BARROUILLET
9.7
The Development of Verbal Working Memory and Its Maintenance Systems
As described in the beginning of this chapter, the main tenet of the TBRS model and its originality compared with other WM models is to take into account the temporal dynamic of cognitive functioning. This is particularly clear in the functional description of the executive loop, and the role that time has in both the forgetting of short-term memory traces as well as the impact of processing duration in the availability of attentional refreshing. One central question is to understand when such a functioning appears in childhood. Few studies examined whether the relationship between recall performance and concurrent CL reported in adults is also observed in children (Corbin et al., 2012; Gavens & Barrouillet, 2004; Portrat et al., 2009; see Camos & Barrouillet, 2011, 2018, for reviews). Using a complex span task, Portrat et al. (2009) reported in 10-year-old children a reduction of recall performance of series of consonants when the concurrent task, which was to judge the location (up or down) of squares appearing successively on screen, was made more difficult by either diminishing the distance between the two possible locations or blurring the squares. Such findings, akin to what is observed in young adults, comforted Gavens and Barrouillet’s (2004) results. In this latter study, 8- and 10-year-old children had to maintain series of letters in a complex span task, while they read aloud series of digits appearing successively on screen. Digits in the concurrent reading digit task were presented either in a random or a canonical order (e.g., 5 3 2 4 1 vs. 1 2 3 4 5); the former requires more executive attention because the nature of the forthcoming digit cannot be anticipated, and children have to inhibit their natural tendency to utter digits in the canonical sequence. Moreover, the aloud reading of digits prevented the use of articulatory rehearsal, providing a better test to the functioning of attentional refreshing in children. The results revealed a more detrimental impact on recall performance of reading the digits in the random than the canonical order. Finally, it should be noted that the reduction of recall performance when the CL of concurrent task increases is not restricted to typically developing children. In the same task as the one used by Gavens and Barrouillet (2004), recall performance in atypically developing children with learning difficulties and lower WM capacities than their age-matched peers was also affected by variations in CL of the reading digit task (Corbin et al., 2012). Although these studies showed that time-based resource sharing accounts for the functioning of WM in children, this mechanism is not as efficient at any age. In a study involving four age groups of children from 8 to 14 years old performing a reading digit-span task, Barrouillet et al. (2009) varied the pace of presentation of the digits to be read from a rate of 0.4, 0.8, 1.2 to 2 digits per second, the latter inducing a stronger distraction of attention away from maintenance activity. The reduction of recall performance with the increased attentional demand of the reading digit task was stronger in older than
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
younger children. The authors suggested that this phenomenon evidences the poorer functioning of refreshing in the younger children, which are then less impaired by an increase in concurrent attentional demand. Importantly, it should be noted that the relationship between concurrent attentional demand and recall performance is not observed in very young children, aged 5 or 6. These children are not sensitive to depletion of attention, but to the retention interval between the presentation of the memory items and the recall phase (Barrouillet et al., 2009; Camos & Barrouillet, 2011). This latter finding is understood as an absence of use of attentional refreshing in children aged 5 or 6. Such an absence of refreshing could result on either the incapacity to implement it at this age or the neglect of implementation of maintenance strategies. Some authors suggested that the latter is a consequence of the goal neglect frequently observed in children of that age. However, when goal maintenance was scaffolded in WM span tasks by different cues, recall performance did not improve in children younger than 7 (Fitamen et al., 2019), suggesting that goal neglect does not cause the absence of (or difficulty in) setting up WM maintenance strategies in very young children. Beside the executive loop, we previously mentioned that verbal information can be maintained in a domain-specific system, the phonological loop, the two loops being independent in accounting for recall performance in adults. A similar pattern is also observed in children of different ages from 6 to 9 (Oftinger & Camos, 2016), suggesting that the two systems involved in maintenance have independent effects on the maintenance of verbal information from a relatively young age onward. Nevertheless, it seems that the preference of one or the other system varies with age, articulatory rehearsal being a preferred or default system to maintain verbal information in younger children. Indeed, at 7 years of age, recall performance varies with the concurrent attentional demand when articulatory rehearsal is impaired by a concurrent articulation, but not when memory items can be rehearsed. By contrast, whether articulatory rehearsal is available or not, the impact of a concurrent attentional load never affects 6-year-old children, but always 8-year-olds (Oftinger & Camos, 2018; see also 2017, for congruent findings). Finally, between 8 and 9 years of age, some children favor rehearsal while others prefer refreshing (Mora & Camos, 2015). This shows that, though adults have clearly the ability to adapt their maintenance mechanism to the task, children slowly acquire this ability, with many changes occurring around 7.
9.8
The Two Systems of Verbal WM, a Method for Their Optimal Use and How It Informs Us about the Nature and Functioning of the Articulatory Loop
The capacity of verbal short-term memory, primarily through the measurement of the digit span, has been regarded as an index of intellectual capacity and intellectual development from the very inception of psychometrics
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
187
188
VALÉRIE CAMOS AND PIERRE BARROUILLET
(Binet & Simon, 1904; Wechsler, 2014), and for modern psychology as an estimate of the capacity of the human mind understood as a channel of information (Miller, 1956). Considering the importance of language for human thinking and intelligence, being able to keep in mind a large amount of verbal information would be a prerequisite for an efficient mental functioning. The fact that WM capacity, often assessed through verbal tasks such as Daneman and Carpenter’s (1980) reading span task, proved to be the best predictor of fluid intelligence and of a series of intellectual achievements has reinforced this conception (Kane et al., 2004). These spans might receive an unequivocal interpretation as long as verbal short-term memory spans were considered as reflecting the capacity of a single store like the short-term store in Atkinson and Shiffrin’s (1968) model and the phonological loop in Baddeley’s (1986) model, or the efficiency of a single mechanism of shortterm maintenance for those who do not believe in a short-term memory separated from long-term memory (e.g., Brown et al., 2007; Crowder, 1993). However, if, as the TBRS assumes, verbal WM involves two distinct systems, what do short-term memory spans measure? Do they correspond to the sum of the capacity of the two systems? Barrouillet et al. (2021) noted that it does not seem to be the case. Whereas the capacity of both the phonological loop (Vergauwe et al., 2014) and the central attentional system (Cowan, 2005) has been estimated to be about four letters, adults’ letter span is notoriously lower than eight and only reaches 6.5 at best (Dempster, 1981). Barrouillet et al. (2020) reasoned that if the letter span does not correspond to the sum of the capacity of the two systems, it is probably because individuals, unaware of the structure of their verbal WM, misuse it. Because articulatory rehearsal appears at a first glance as the most natural and efficient way for maintaining verbal information, it might be that people try to rehearse too many items. This overload of the articulatory system would disorganize the schedule of articulation and lead to underuse the articulatory system. If this is the case, appropriate instructions based on the TBRS’s hypotheses about the nature, functioning, and capacity of the two systems should lead to an increase in individual’s letter span that should reach or at least approach the theoretical optimum of 8. For this purpose, adults were presented with series of letters for immediate serial recall, the length of which varied from 4 to 11 letters. In a control condition, they were just asked to recall these series without any specific instructions, as in a traditional simple span procedure. In the experimental condition, participants were asked to perform a cumulative rehearsal, but only of the first 3, 4, or 5 letters, no more, and to keep rehearsing these letters aloud until the end of the series and recall. The rationale of this method, which we called the maxispan procedure, was twofold. On the one hand, rehearsing a limited number of letters should not exceed the capacity of the phonological loop and allow their perfect maintenance. On the other, articulating this set of letters until the end of the series, and thus during the presentation of the following letters, should prevent these latter letters
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
from entering the phonological loop and force people to store them in the executive loop. In such a way, the two systems would hold different sets of letters without any transfer or interaction, optimizing their use. As we expected, whereas the simple span procedure led to spans about 6, as it is usually observed, the maxispan procedure dramatically increased this recall performance that was close to 8 (7.73) in the optimal condition in which the letters to be rehearsed were presented auditorily while the following letters were presented visually. Higher spans were also observed with the maxispan than the traditional simple span procedure when all the letters were presented visually or auditorily, the maxispan procedure resulting in a better recall of both the rehearsed and the nonrehearsed letters. In fact, in the maxispan procedure, the limited set of letters articulatorily rehearsed was virtually perfectly recalled, the following letters exhibiting a strong primacy effect with a very small or no recency effect. Interestingly, the recall of the rehearsed letters was not affected by the number of following letters, reinforcing the hypothesis of two distinct and separate systems. Finally, these findings led us to reconsider our hypotheses about the nature of what we called, following Baddeley (1986), the phonological loop. The phonological similarity effect already discussed led Baddeley to assume that this loop stores phonological representations. However, in a series of recent experiments, we observed that the maxispan procedure in which only four letters out of seven are rehearsed abolishes the phonological similarity effect. It is only when the capacity of the articulatory loop is exceeded (i.e., with the instruction of rehearsing six letters) that the phonological similarity effect reappears with the maxispan procedure. This might be due to the fact that trying to articulate too many letters leads to their encoding in the executive loop, probably in a phonological format resulting from their pronunciation, these phonological traces becoming confusable and more difficult to retrieve and recall when memory items are phonologically similar. The fact that the phonological similarity effect disappears under the maxispan procedure suggests that the items maintained through the articulatory process are not stored in a phonological format. Baddeley himself (Baddeley et al., 1975) envisioned this hypothesis in a first version of his model in which an executive WM system endowed with storage capacities was supplemented by an articulatory rehearsal loop with a capacity of about three items, and not seven as it was subsequently assumed (Baddeley, 1986). This smaller capacity was explained by the fact that when the access to the loop is prevented, memory items can be stored in the executive system. The other strong difference with the current phonological loop model was the nature of what is stored in the articulatory loop, which was considered as an output buffer holding the motor program necessary for the verbal production of the memory items. Interestingly, recent studies have shown that the mere perception of speech activates the motor areas involved in speech production (Wilson et al., 2004). It should also be noted that the cyclic articulatory process through which the loop
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
189
190
VALÉRIE CAMOS AND PIERRE BARROUILLET
maintains verbal items is one of the most basic sensorimotor processes, already present in the first months of life. Following Baldwin (1894), Piaget (1936) described primary circular reactions in which infants tend to repeat syllables pertaining to their own repertoire, the perception of these syllables triggering their production that causes their perception that in turn prompts their production in a cyclic behavior. In this account, the information held by the articulatory loop would not be representational in nature. More than a memory buffer, this loop would be a sensorimotor auxiliary system used to reinstate by its pronunciation the verbal material to be remembered. The tendency to try to rehearse too many items in the span tasks, rooted in one of the most basic and early acquired cognitive mechanisms, along with the fact that people are unaware of the structure and functioning of their memory, explains why short-term memory spans underestimate verbal WM capacity.
9.9
Conclusion
Although WM relies on two separate systems for the maintenance of verbal information, its capacity remains surprisingly limited, in part because as we have seen people underuse it, but also because the capacity of both systems is itself strongly limited. Surprisingly, such a small capacity seems to be sufficient for acquiring complex linguistic systems, lending support to the Less-Is-More hypothesis (Newport, 1990). This also suggests that, as Baddeley (2007) noted, most of language comprehension and production relies on automatic processes that do not require a strong WM involvement, except for understanding long and convoluted sentences. However, despite its limited capacity, the very structure of verbal WM as it is described by the TBRS model might be especially appropriate for language processing. Indeed, the variety of codes allowed by the dual-structure of verbal WM proposed by the TBRS model make it particularly well adapted to language processing, with an articulatory loop suitable for the maintenance of verbatim traces of the surface form of the words, whereas the executive loop and its capacity to form and maintain multimodal representations is also able to represent and maintain the gist of even complex sentences. This interplay between different levels of information within verbal WM is probably what makes it so efficient in language processing.
References Abadie, M., & Camos, V. (2018). Attentional refreshing moderates the word frequency effect in immediate and delayed recall tasks. Annals of the New York Academy of Sciences, 1424, 127–136.
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
Abadie, M., & Camos, V. (2019). False memory at short and long term. Journal of Experimental Psychology: General, 148(8), 1312–1334. Adams, A. M., & Gathercole, S. E. (1996). Phonological working memory and spoken language development in young children. Quarterly Journal of Experimental Psychology, 49A, 216–223. Allen, R. J., Baddeley, A. D., & Hitch, G. J. (2006). Is the binding of visual features in working memory resource-demanding? Journal of Experimental Psychology: General, 135, 298–313. Allen, R. J., Hitch, G. J., & Baddeley, A. D. (2009). Cross-modal binding and working memory. Visual Cognition, 17, 83–102. Anderson, J. R. (1993). Rules of the mind. Lawrence Erlbaum Associates. Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation: Advances in research and theory (Vol. 2, pp. 89-195). Academic Press. Baddeley, A. D. (1986). Working memory. Clarendon Press. Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4, 417–423. Baddeley, A. D. (2007). Working memory, thought, and action. Oxford University Press. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. A. Bower (Ed.), Recent advances in learning and motivation (Vol. 8, pp. 647–667): Academic Press. Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of short-term memory. Journal of Verbal Learning and Verbal Behavior, 14, 575–589. Baddeley, A. D., Vallar, G., & Wilson, B. A. (1987). Sentence comprehension and phonological memory: Some neuropsychological evidence. In M. Coltheart (Ed.), Attention and performance XII: The psychology of reading (pp. 509–529). Lawrence Erlbaum Associates. Barrouillet, P., Bernardin, S., & Camos, V. (2004). Time constraints and resource-sharing in adults’ working memory spans. Journal of Experimental Psychology: General, 133, 83–100. Barrouillet, P., Bernardin, S., Portrat, S., Vergauwe, E., & Camos, V. (2007). Time and cognitive load in working memory. Journal of Experimental Psychology: Learning, Memory and Cognition, 33(3), 570–585. Barrouillet, P., & Camos, V. (2001). Developmental increase in working memory span: Resource sharing or temporal decay? Journal of Memory and Language, 45, 1–20. Barrouillet, P., & Camos, V. (2015). Working memory: Loss and reconstruction. Psychology Press. Barrouillet, P., & Camos, V. (2020). The time-based resource-sharing model of working memory. In R. H. Logie, V. Camos, N. Cowan (Eds), Working memory: State of the science. Oxford University Press.
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
191
192
VALÉRIE CAMOS AND PIERRE BARROUILLET
Barrouillet, P. Gavens, N., Vergauwe, E., Gaillard, V., & Camos, V. (2009). Working memory span development: A Time-Based Resource-Sharing model account. Developmental Psychology, 45, 477–490. Barrouillet, P., Gorin, S., & Camos, V. (2021). Simple spans underestimate verbal working memory capacity. Journal of Experimental Psychology: General, 150(4), 633. Barrouillet, P., Portrat, S., & Camos, V. (2011). On the law relating processing and storage in working memory. Psychological Review, 118, 175–192. Barrouillet, P., Portrat, S., Vergauwe, E., Diependaele, K., & Camos, V. (2011). Further evidence for temporal decay in working memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 37, 1302–1317. Berlin, B., & Kay, P. (1969). Basic color terms. University of California Press. Binet, A., & Simon, T. (1904). Méthodes nouvelles pour le diagnostic du niveau intellectuel des anormaux [New methods for the diagnosis of the intellectual level of subnormals]. L’année psychologique, 11, 191–244. Braine, M. D. S. (1990). The “natural logic” approach to reasoning. In W. F. Overton (Ed.), Reasoning, necessity and logic: Developmental perspectives (pp. 135-158). Lawrence Erlbaum Associates. Brainerd, C. J., & Reyna, V. F. (2002). Fuzzy-trace theory and false memory. Current Directions in Psychological Science, 11, 164–169. Brainerd, C. J., & Reyna, V. F. (2005). The science of false memory. Oxford University Press. Brown, G. D., Neath, I., & Chater, N. (2007). A temporal ratio model of memory. Psychological Review, 114, 539–576. Camos, V., & Barrouillet, P. (2011). Factors of working memory development: The time-based resource-sharing approach. In P. Barrouillet & V. Gaillard (Eds.), Cognitive development and working memory: From neo-Piagetian to cognitive approaches (pp. 151–176). Psychology Press. Camos, V., & Barrouillet, P. (2018). Working memory in development. Routledge. Camos, V., Johnson M., Loaiza V., Portrat S., Souza A., & Vergauwe E. (2018). What is attentional refreshing in working memory? Annals of the New York Academy of Sciences, 1424(1), 19–32. Camos, V., Lagner, P., & Barrouillet, P. (2009). Two maintenance mechanisms of verbal information in working memory. Journal of Memory and Language, 61(3), 457–469. Camos, V., Mora, G., & Barrouillet, P. (2013). Phonological similarity effect in complex span task. Quarterly Journal of Experimental Psychology, 66(10), 1927–1950. Camos, V., Mora, G., & Oberauer, K. (2011). Adaptive choice between articulatory rehearsal and attentional refreshing in verbal working memory. Memory & Cognition, 39(2), 231–244. Camos, V., Mora, G., Oftinger, A-L., Mariz Elsig, S., Schneider, P., & Vergauwe, E., (2019). Does long-term memory affect refreshing in verbal
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
working memory? Journal of Experimental Psychology: Learning, Memory and Cognition, 45(9), 1664–1682. Camos, V., & Portrat, S. (2015). The impact of cognitive load on delayed recall. Psychonomic Bulletin and Review, 22, 1029–1034. Campoy, G., Castellà, J., Provencio, V., Hitch, G., & Baddeley, A. (2015). Automatic semantic encoding in verbal short-term memory: Evidence from the concreteness effect. Quarterly Journal of Experimental Psychology, 68(4), 759–778. Corbin, L., Moissenet, A., & Camos, V. (2012). Le fonctionnement de la mémoire de travail chez des enfants présentant des difficultés scolaires. Développements, 11, 5–12. Cowan, N. (1995). Attention and memory: An integrated framework. Oxford University Press. Cowan, N. (1999). An embedded-processes model of working memory. In P. Shah & A. Miyake, (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 62–101). Cambridge University Press. Craik, F. I. M., & Watkins, M. J. (1973). The role of rehearsal in short-term memory. Journal of Verbal Learning and Verbal Behavior, 12, 599–607. Crowder, R. G. (1993). Short-term memory: Where do we stand? Memory & Cognition, 21, 142–145. http://dx.doi.org/10.3758/BF03202725 Daneman, M. & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. Dempster, F. N. (1981). Memory span: Sources of individual and developmental differences. Psychological Bulletin, 89, 63–100. Engle, R. W., Nations, J. K., & Cantor, J. (1990). Is working memory capacity just another name for word knowledge? Journal of Educational Psychology, 82, 799–804. Fitamen, C., Blaye, A., & Camos, V. (2019). The role of goal cueing in kindergarteners’ working memory. Journal of Experimental Child Psychology, 187. Garnham, A., Garrod, S., & Sanford, A. (2006). Observation on the past and future of psycholinguistics. In M. J. Traxler, & M. A. Gernsbacher (Eds.), Handbook of psycholinguistics (pp. 1–18). Academic Press. Gathercole, S. E., & Baddeley, A. D. (1989). Evaluation of the role of phonological STM in the development of vocabulary in children: A longitudinal study. Journal of Memory and Language, 28, 200–213. Gathercole, S. E., & Baddeley, A. D. (1990). Phonological memory deficits in language-disordered children: Is there a causal connection? Journal of Memory and Language, 29, 336–360. Gavens, N., & Barrouillet, P. (2004). Delays of retention, processing efficiency, and attentional resources in working memory span development. Journal of Memory and Language, 51, 644–657.
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
193
194
VALÉRIE CAMOS AND PIERRE BARROUILLET
Glenberg, A. M., Smith, S. M., & Green, C. (1977). Type I rehearsal: Maintenance and more. Journal of Verbal Learning and Verbal Behavior, 16, 339–352. Greene, R. L. (1987). Effects of maintenance rehearsal on human memory. Psychological Bulletin, 102, 403–413. Gruber, O. (2001). Effects of domain-specific interference on brain activation associated with verbal working memory task performance. Cerebral Cortex, 11, 1047–1055. Hudjetz, A., & Oberauer, K. (2007). The effects of processing time and processing rate on forgetting in working memory: Testing four models of the complex span paradigm. Memory & Cognition, 35, 1675–1684. Jefferies, E., Lambon Ralph, M. A., & Baddeley, A. D. (2004). Automatic and controlled processing in sentence recall: The role of long-term and working memory. Journal of Memory and Language, 51, 623–643. Johnson, M. K. (1992). MEM: Mechanisms of recollection. Journal of Cognitive Neuroscience, 4(3), 268–280. Johnson, M. K., Reeder, J. A., Raye, C. L., & Mitchell, K. J. (2002). Second thoughts versus second looks: An age-related deficit in reflectively refreshing just-activated information. Psychological Science, 13, 64–67. Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W., & Engle, R. W. (2004). The generality of working memory capacity: A latentvariable approach to verbal and visuospatial memory span and reasoning. Journal of Experimental Psychology: General, 133, 189–217. Loaiza, V. M., & Camos, V. (2018). The role of semantic representations in verbal working memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 44, 863–881. Loaiza, V. M., Duperreault, K. A., Rhodes, M. G., & McCabe, D. P. (2015). Long-term semantic representations moderate the effect of attentional refreshing on episodic memory. Psychonomic Bulletin & Review, 22, 274–280. Loaiza, V. M., & McCabe, D. P. (2012). Temporal-contextual processing in working memory: Evidence from delayed cued recall and delayed free recall tests. Memory & Cognition, 40, 193–203. Loaiza, V. M., & McCabe, D. P. (2013). The influence of aging on attentional refreshing and articulatory rehearsal during working memory on later episodic memory performance. Aging, Neuropsychology, & Cognition, 20, 471–493. Loaiza, V. M., Rhodes, M. G., & Anglin, J. (2015). The influence of age-related differences in prior knowledge and attentional refreshing opportunities on episodic memory. Journals of Gerontology, Lists B: Psychological Sciences and Social Sciences, 70 (5), 729–736. McCabe, D. P. (2008). The role of covert retrieval in working memory span tasks: Evidence from delayed recall tests. Journal of Memory and Language, 58(2), 480–494.
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
Time-Based Resource Sharing Model
Miller, G. A. (1956). The magical number seven plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. http://dx.doi.org/10.1037/h0043158 Mora, G., & Camos, V. (2013). Two systems of maintenance in verbal working memory: Evidence from the word length effect. PLoS ONE, 8, e70026. Mora, G., & Camos, V. (2015). Dissociating rehearsal and refreshing in the maintenance of verbal information in 8-year-old children. Frontiers in Psychology (Developmental Psychology), 6(11). Morton, J. (1979). Word recognition. In J. Morton & J. C. Marshall (Eds.), Psycholinguistics, volume 2: Structures and processes (pp. 107–156). Paul Elek. Newport, E. L. (1990). Maturational constraints on language learning. Cognitive Science, 14, 11–28. https://doi.org/10.1207/s15516709cog1401_2 Oftinger, A-L., & Camos, V. (2016). Maintenance mechanisms in children’s verbal working memory. Journal of Educational and Developmental Psychology, 6(1), 16–28. Oftinger, A-L., & Camos, V. (2017). Phonological similarity effect in children’s working memory: Do maintenance mechanisms matter? Journal of Child Psychology, 1(1), 5–11. Oftinger, A-L., & Camos, V. (2018). Developmental improvement in strategies to maintain verbal information in children’s working memory. International Journal of Behavioral Development, 42(2), 182–191. Papagno, C., Valentine, T., & Baddeley, A. D. (1991). Phonological short-term memory and foreign-language vocabulary learning. Journal of Memory and Language, 30, 331–347. Piaget, J. (1923). Le langage et la pensée chez l’enfant. Delachaux et Niestlé. Portrat, S., Camos, V., & Barrouillet, P. (2009) Working memory in children: A time-related functioning similar to adults. Journal of Experimental Child Psychology, 102, 368–374. Raye, C. L., Johnson, M. K., Mitchell, K. J., Greene, E. J., & Johnson, M. R. (2007). Refreshing: A minimal executive function. Cortex, 43, 135–145. Reyna, V. F., & Brainerd, C. J. (1995). Fuzzy-trace theory: An interim synthesis. Learning and Individual Differences, 7, 1–75. Ricker, T., Cowan, N., & Morey, C. (2010). Visual working memory is disrupted by covert verbal retrieval. Psychonomic Bulletin & Review, 17, 516–521. Rips, L. J. (1994). The psychology of proof. MIT Press. Rose, N. S., Buchsbaum, B. R., & Craik, F. I. M. (2014). Short-term retention of a single word relies on retrieval from long-term memory when both rehearsal and refreshing are disrupted. Memory & Cognition, 42, 689–700. Rose, N. S., Craik, F. I., & Buchsbaum, B. R. (2015). Levels of processing in working memory: Differential involvement of frontotemporal networks. Journal of Cognitive Neuroscience, 27, 522–532. Rosselet-Jordan, F. L., Abadie, M., Mariz-Elsig, S., & Camos, V. (2022). Role of attention in the associative relatedness effect in verbal working memory:
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
195
196
VALÉRIE CAMOS AND PIERRE BARROUILLET
Behavioral and chronometric perspectives. Journal of Experimental Psychology: Learning, Memory, and Cognition. Advance online publication. https://doi.org/10.1037/xlm0001102 Rosselet-Jordan, F., Abadie, M., Mariz-Elsig, S. & Camos, V. (in press). Role of attention in the associative relatedness effect in verbal working memory: Behavioral and chronometric perspective. Smith, E. E., & Jonides, J. (1999). Storage and executive processes in the frontal lobes. Science, 283, 1657–1661. Trost, S., & Gruber, O. (2012). Evidence for a double dissociation of articulatory rehearsal and non-articulatory maintenance of phonological information in human verbal working memory. Neuropsychobiology, 65, 133– 140. Vallar, G., & Baddeley, A. D. (1987). Phonological short-term store and sentence processing. Cognitive Neuropsychology, 4, 417–438. Vergauwe, E., Barrouillet, P. & Camos, V. (2009). Visual and spatial working memory are not that dissociated after all: A time-based resource-sharing account. Journal of Experimental Psychology: Learning, Memory & Cognition, 35, 1012–1028. Vergauwe, E., Barrouillet, P. & Camos, V. (2010). Verbal and visuo-spatial working memory: A case for domain-general time-based resource sharing. Psychological Science, 21, 384–390. Vergauwe, E., Camos, V., & Barrouillet, P. (2014). The impact of storage on processing: Implications for structure and functioning of working memory. Journal of Experimental Psychology: Learning, Memory & Cognition, 40, 1072–1095. Vygotsky, L. S. (1934). Thought and language. MIT Press. Wechsler, D. (2014). Wechsler Intelligence Scale for Children: Fifth Edition technical and interpretive manual. NCS Pearson. Wilson, S. M., Saygin, A. P., Sereno, M. I., & Iacoboni, M. (2004). Listening to speech activates motor areas involved in speech production. Nature Neuroscience, 7, 701–702. http://dx.doi.org/10.1038/nn1263 Whorf, B. L. (1940). Science and linguistics. Technology Review, 42, 229–231, 247–248. Woodward, A. E., Bjork, R. A., & Jongeward, R. H. (1973). Recall and recognition as a function of primary rehearsal. Journal of Verbal Learning and Verbal Behavior, 12, 608–617.
Note 1 The difference between the two types of tasks is that in a complex span task, processing episodes are interspersed between memory items, whereas the Brown-Peterson task adopts a preload paradigm in which all the memory items are presented first and followed by an intervening task before recall.
https://doi.org/10.1017/9781108955638.012 Published online by Cambridge University Press
10 The Ease of Language Understanding Model Jerker Rönnberg, Emil Holmer, and Mary Rudner
10.1
Introduction
The Ease of Language Understanding (ELU) model describes how different memory systems are used to support language perception and understanding, as well as communicative discourse (Rönnberg et al., 2021). Initially, the ELU model was based on the observation that persons with acquired hearing impairment showed similar cognitive architectures underlying speech understanding (Arlinger et al., 2009; Rönnberg, 2003), irrespective of communication method. However, the relative weight of linguistic input and cognitive processing components depended on how well specified the signal was and how it was processed in a hearing aid. With very poorly specified input (auditory, visual, or tactile), cognitive functions were assumed and shown to come into play to a higher degree. With better specification of input language, processing could be more readily and rapidly executed. Moreover, hearing and cognitive functions can interface at different levels in the system, from subcortical, postcochlear levels, to modality-specific and amodal regions of cortex. In the ELU model (see Rönnberg et al., 2013, 2019, 2021), we have assumed that working memory (WM) is involved in both prediction and postdiction of phonological, semantic, and syntactic contents of communication. Prediction and postdiction are overarching processes tied to communicative content and intent. The interaction between prediction and postdiction depends on the degree and type of hearing loss, the difficulty of language understanding or communicative task (e.g., Mattys et al., 2012), as well as the particular WMC (Capacity, e.g., Arehart et al., 2013), lexical knowledge, and processing skills of the individual (Rönnberg et al., 2016). This research was supported by the Linnaeus Centre HEAD, financed by The Swedish Research Council Grant (3492007-8654), awarded to Jerker Rönnberg, and by a grant from the Swedish Research Council (2017-06092), awarded to Anders Fridberger.
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
198
JERKER RÖNNBERG ET AL.
In general, much of our work focuses on the cognitive demands and processes resulting from a mismatch between linguistic input delivered via the RAMBPHO buffer (Rapid Automatic Multimodal Binding of PHOnology) and Long-Term Memory (LTM) representations, causing postdiction. Postdiction (including inference-making and repair) feeds back into new predictions, creating a dynamic interacting system. Our notion of mismatch emphasizes the cognitive and communicative consequences of mismatch, and not just the physical parameters of the mismatch function per se. Mismatch arises when either the RAMBPHO information is poor or the phonological representations in LTM are poor. We propose a threshold above which a certain number of phonological attributes, embedded in lexico-semantic representations, must be part of the input to trigger lexical meaning. Subthreshold input may lead to different outcomes (for further details of mismatching conditions, see Rönnberg et al., 2013, pp. 10–11). If there are too many mismatches on a daily basis, and if hearing loss remains untreated for many years, there may be negative long-term consequences. One such consequence is a relative disuse of LTM systems due to fewer encodings, and hence fewer retrievals (Rönnberg et al., 2011, 2013, 2014, 2019), which is a potential disuse precursor of mild cognitive impairment (MCI, Farias et al., 2017; Fortunato et al., 2016), with a subsequent increased risk of dementia of the Alzheimer type (e.g., Lin, 2011; Lin et al., 2014; Livingstone et al., 2017; Rönnberg et al., 2021). Prediction represents fast preprocessing (primarily on the millisecond scale). It involves focusing attention on certain kinds of linguistic information and associations (Zekveld et al., 2011, 2013), resulting in priming. Priming in turn leads to better perception and greater perceived clarity of speech (Signoret et al., 2018; Signoret & Rudner, 2019; Wild et al., 2012). Our contribution in the ELU context is our assumption that predictions are kept and processed in WM, sometimes explicitly (e.g., Zekveld et al., 2013) but typically implicitly (Rönnberg et al., 2021). Prediction is fast and may help access the lexicon via syllabic, phonological information (see below about RAMBPHO under the WM-ELU system). The phonological/lexical composite representations are dependent on the episodic, semantic, and contextual conditions under which they are formed over time (Hickok & Poeppel, 2012; Holmer et al., 2016; Rönnberg et al., 2013). When perceived speech is fuzzy relative to LTM representations (e.g., due to hearing loss, suboptimal hearing aid settings), then a mismatch is likely to occur and will activate WM in another more explicit functional role, namely, postdiction. Postdiction operates on a different time scale than prediction (in sec) and is about WM – in interaction with Episodic and Semantic LTM (ELTM and SLTM). Postdiction is initiated when there is a mismatch. In contrast to prediction, it is typically explicit and elaborate. In postdiction, there may be several WMLTM storage and processing interactions before the listener has understood the gist of the conversation. In our view, WM plays a key role in balancing and coordinating prediction and postdiction to accomplish optimal ease of
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
language understanding. When there are unexpected or rapid fluctuations in the relationship between prediction and postdiction processes during discourse, WM may, depending on communicative intentions and communicative situation, give priority to one of the two functions. In everyday life, communication is always about a dynamic interplay between the two functions. Skill in rapid shifting between prediction and postdiction may be a cognitive key to successful communication. For example, a person with hearing loss may sometimes go completely wrong in their interpretation of a conversation, and hence, need to generate new hypotheses (predictions) via further reconstruction (postdiction) (cf., Poeppel et al., 2008).
10.2
The ELU-WM System
WM is a capacity limited pool of storage and processing resources (e.g., Baddeley, 2012; Daneman & Carpenter, 1980; Lunner et al., 2009; Rönnberg et al., 2013). Depending on task demands, these resources can be allocated flexibly to storage or processing of language. In our view, if processing of some kind consumes most of the resources, less will be available for storage and other kinds of processing, and vice versa (Lunner et al., 2009); if storage demands increase, then other processing activities will be dampened or inhibited (Sörqvist et al., 2012, 2016). The only component of the ELU-WM model that is relatively encapsulated from the task-dependent and dynamic storage and processing functions is RAMBPHO. The ELU model emphasizes rapid multimodal binding of sensory information rather than separate sensory stores (cf., the episodic buffer, Baddeley, 2000), it relies on multimodal representations in LTM (e.g., Cowan, 2005), and has a predictive attention component that is not necessarily modality specific but rather time-based (cf., the Time-Based Resource-Sharing model, Barrouillet & Camos, 2020). RAMBPHO is obligatory; it handles sensory information at any given moment (McGurk & MacDonald, 1976; Näätänen & Escera, 2000), with a high updating rate (cf. the “primal sketch”; Poeppel et al., 2008), and has the presumed aim of creating a sufficiently precise phonological code for a multimodal WM system. RAMBPHO is assumed to abstract and assemble relevant multimodal phonological (visual, auditory, tactile) information that can be used by WM and represented in LTM. It is crucial for lexical access, and consequently, for the flow of language processing and understanding (e.g., Rönnberg et al., 2008). It is primed and contextually framed by rapid less time-demanding operations (e.g., Signoret et al., 2018). RAMBPHO is an essential part of making predictions, but can be overruled if the context is so strong that hardly any sensory input is needed to understand what was intended (e.g., by using gating of final words in highly predictable sentences, Moradi et al., 2013, 2014; see Section 10.2.1, Section 10.2.2, and Figure 10.1). While RAMBPHO represents one modular function of WM, there are other storage and processing aspects of WM that are not typically captured
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
199
200
JERKER RÖNNBERG ET AL.
Figure 10.1 The ELU-WM system Note: Language input enters RAMBPHO (Rapid, Automatic, Multimodal Binding of Phonology) (a) where it is compared to multimodal representations accessed in semantic (SLTM) and episodic (ELTM) long-term memory (b). If there is a rapid match, the lexicon is accessed effortlessly (c) and gist is grasped immediately or full understanding achieved (d). Gist or understanding, combined with contextual information, generates predictions of upcoming input that feed back into RAMBPHO (e). If accurate, predictions facilitate RAMBPHO processing. In case of mismatch, postdiction commences (f ) involving explicit processing of language input fragments and further access to SLTM and ELTM (g) in order to unlock the lexicon. If postdiction is successful, lexical access will be achieved in the next RAMBPHO cycle (h). The match-mismatch interaction will affect ELTM and SLTM in the long run, not least in terms of relative disuse and adaptation/development.
by traditional memory tests. We have used complex tests of WMC to operationalize the explicit capacity that was assumed in the Rönnberg (2003) and Rönnberg et al. (2008) versions of the model. The model does not stipulate specific processing resources, apart from RAMBPHO. Nevertheless, the important feature of our WM operationalization in the context of speech understanding is that the particular test taps into both storage and processing in WM (Daneman & Carpenter, 1980). We have used, for example, visuospatial WM tests, where the task is to remember the sequence of cells in a matrix that are used for same-different comparisons of nonnamable circular objects. Performance on visual reading span (sets of sentences are presented to the participant who had to judge whether each sentence was appropriate or absurd, and at the end of the list recall either the first or last word in each sentence in their correct sentence-wise serial order) and the visually presented semantic word-pair span (apart from recalling either the first or the second word in the word-pair, the participant had to decide which word in the pair that was a living thing), correlated highly with performance on the visuospatial span test, suggesting a common denominator (Rönnberg et al., 2016). Additionally, all three of them (as a latent factor in a Structural Equation Model), or one at a time, predict recall and understanding of speech presented in a background of
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
speech noise to a larger extent than do simpler span tests (Rönnberg et al., 2016). In fact, we suggest that the dual demands of these WM tests tap into rapid abstraction from modality-specific to modality-free cognitive representations (Rönnberg et al., 2011, 2014). Moreover, the visuospatial test may tap into the same general executive attention system as all dual tasks do. This reduces the need to assume a separate visuospatial sketchpad to explain these data (cf., Camos & Barrouillet, this volume). We venture that WM storage and processing are mutually interdependent in all kinds of language perception, language understanding, or conversation tasks. In the ELU model we assume that fast predictions are nested under relatively slower postdictions. RAMBPHO has done its work (perhaps several timeover) before slower elaborate inference-making takes place (Rönnberg et al, 2013). One important question is whether the contents of both prediction and postdiction are abstracted to multimodal LTM representations in the brain, rapidly for prediction, and by implication, for postdiction purposes. These and related issues will be addressed in the current chapter and are visualized in Figure 10.1.
10.2.1 WM and Prediction We have conducted experimental studies that clearly show a “silent” involvement of WM during very early predictive processing (Sörqvist et al., 2012). Attending to a sound (counting deviants in an oddball paradigm) gives an increase in the amplitude of the brainstem response compared to listening only. However, the brainstem response decreases when attention is shifted from the auditory modality to a visually based, n-back WM task, including letter sequences. But the tones and deviants remain as stimuli, identical to the baseline conditions. And, when the cognitive load in the nback task is increased, the brain stem Wave 5 (i.e., 7 ms postcochlear response) is further dampened, especially for participants with a high WMC (Sörqvist et al., 2012; see also Anderson et al., 2013; Kraus & WhiteSchwoch, 2015). Thus, WM is implicated in the very early neural processing of stimuli. Recent conceptualizations of a top-down governed, subcortical, early auditory filter model are in line with these results and account for the WM data referred to above (Marsh & Campbell, 2016). When we later repeated the same experiment in an fMRI design, we observed that increasing the load in the visual n-back task also inhibits activity in superior temporal auditory cortical regions. The cortical activity seems to mirror the brain stem activity, namely, a cross-modality dampening of auditory processing when visual WM load is increased. Another way of putting it is: When WM is fully occupied, it is shielded by early attention mechanisms (Marsh & Campbell, 2016; Sörqvist et al., 2016). As a matter of fact, load also dampens amygdala activity – a finding that may imply that it takes a “cool” brain to focus on a demanding task. Thus, load contributes to reducing both internal and external distractions
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
201
202
JERKER RÖNNBERG ET AL.
(Sörqvist et al., 2016). This general feature of brain processing under pressure will help enhance predictions of upcoming events. Other “early” features of brain processing are the RAMBPHO conversion of stimulus-specific modality of presentation into multimodal or more abstract representational formats. Previously published data from our lab support evidence rapid abstraction into representational formats handled by WM. In one early study by Rönnberg et al. (1989), we looked at neural speed and amplitude in Visual Evoked Potentials (VEPs) and visual speechreading skill. We found a significant correlation with the P200 VEP amplitude and the visual reading span test performance. Thus, WMC (reading span) plays a role early on in visual information processing, just as it does for auditory processing (Sörqvist et al., 2012). In a recent magnetoencephalography (MEG) study of prediction of final words in spoken sentences (“Yes” for expected and “No” for not expected), we observed that WMC correlated with false alarms to meaning deviants, which rhymed with the expected final word. The participants had studied each sentence beforehand, so they had a strong expectation, not only grammatically and semantically, but also episodically (Signoret et al., 2020). Fewer false alarms were observed for individuals with higher WMC. This is the “bottleneck” we have described above: lexical access via phonological attributes. It should be noted that WM was stressed because the words preceding the final word were presented in noise at an individual intelligibility level of 50 percent, that is, the signal-to-noise ratio that for a given individual resulted in 50 percent correct recall. Higher WMC was also associated with lower N400 (peak in cortical response 400 ms after stimulus offset), and there was an association between WMC and the mean amplitude of the N400 effects only for the meaning deviants type (minus the expected condition, Signoret et al., 2020). Thus, again, WM is crucially involved at an early stage of processing, and the N400 result may indicate transfer to more integrative processing, suitable for possible SLTM and ELTM postdiction. Whether the cognitive representation at this particular point in time is still phonological and/or semantic is hard to decide, but it is possible that the high WMC participant has more rapidly and semantically accepted the deviance from the expected word (cf., our assumption about phonologically mediated lexical access speed, Rönnberg, 2003; Rönnberg et al., 2008), thus rendering a smaller N400 response. Also interesting in this context is a long series of experimental and crosssectional studies from our lab by Moradi and colleagues (Moradi et al., 2013, 2014, 2017) using the gating paradigm (Grosjean, 1980). In the gating paradigm, successive fragments of a given speech token (e.g., a consonant) are presented to participants, whose task is to guess the identity of that speech token as more fragments of the signal are successively presented. The major aim of the gating paradigm is to measure the isolation point (i.e., the shortest time required for correct identification of a speech stimulus, after successive additions of time slices of the stimulus).
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
Findings by Moradi et al. showed that people with greater WMC were able to identify consonants and words (unless embedded in highly predictable sentence contexts) faster than those with lower WMC (Moradi et al., 2013, 2014, 2017). Moradi and colleagues reasoned that WM plays a critical role in the rapid identification of a speech token when semantic context is lacking. With this paradigm, it takes around 100–150 msec of the speech token to identify a consonant. Hearing loss (even when compensated with hearing aids), age, and noisy signals all slow down the identification process, whereas a high WMCC counteracts these effects, presumably by reducing demands on the inference-making process in relation to SLTM, when supportive semantic context is missing. Thus, irrespective of sensory input modality, early abstraction into multimodal formats and other formats suitable for WM seems to be a common denominator in the ELU prediction ww process?. RAMBPHO is part of the prediction process but sometimes context overrides RAMBPHO (Rönnberg et al., 2013). One way of viewing the early predictive WM involvement exemplified above is that it “paves the way” to minimize mismatch and postdiction by setting multimodal parameters, already at very early stages of post-cochlear processing (cf., Sörqvist et al., 2012).
10.2.2 WM and Postdiction The original experimental evidence that actually convinced us about the assumption of cognitive consequences of mismatch (i.e., Rudner et al., 2008, 2009; Rönnberg, 2003) was obtained when we manipulated the signal processing in the participants’ hearing aids and observed higher WM dependence for mismatching conditions: The participants (on average 55–60 years, around 50 db loss, three-frequency average on the best ear) were introduced to a nonhabitual mode (other than that prescribed for the participant) of signal processing in their hearing aids (in this case either SLOW or FAST wide range dynamic compression) for a period of nine weeks. Nonhabitual modes of processing necessarily affect the efficiency of RAMBPHO and lexical retrieval. Pretesting and posttesting used Hagerman matrix sentences in amplitude modulated noise (Hagerman, 1982), and when mismatching PRE to POST conditions were used – that is, FAST to SLOW or SLOW to FAST conditions, also controlling for individually set 50 percent intelligibility levels – the dependence on WM increased significantly (Rudner et al., 2008, 2009). For matching conditions, that is, FAST-FAST and SLOW-SLOW, the correlations with WM vanished. The mismatch effect is not uniquely tied to this kind of signal-processing manipulation and could also be observed for speech recognition in noise with new compression release settings more generally, irrespective of whether they were FAST or SLOW (Foo et al., 2007; Rudner et al., 2009). For more extensive reviews of these and other kinds of signal processing data supporting a phonologically triggered mismatch mechanism, see Rönnberg et al. (2019), Souza and Sirow (2014), and Souza et al. (2015, 2019).
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
203
204
JERKER RÖNNBERG ET AL.
Furthermore, when we engage SLTM by means of speech maskers (Lunner, 2003; Lunner & Sundewall-Thorén, 2007; Sörqvist & Rönnberg, 2012; Kilman et al., 2014; Ng et al., 2013), especially in their native tongue (Kilman et al., 2015; Ng et al., 2015), a strong reliance on WM is also observed. This means that over and above phonological mismatch, there is a semantic component. SLTM harbors phonological, lexical, and conceptual representations that allow for the additional informational masking via phonological to lexical/ semantic representations (cf., the MEG study by Signoret et al., 2020). Mismatch in visual word-pair rhyming tasks generates behavior suggestive of similar WM compensation. We have tested experimentally whether high WMC would compensate for poorer phonological LTM representations in persons with long-term profound hearing loss (Classon et al., 2013). Especially for word pairs that are orthographically similar but non-rhyming, individuals with high WMC in the hearing-impaired group demonstrated extremely good rhyme judgments. The paradoxical consequence of this “superficial” phonological strategy was that later recognition of target words became worse. Persons with severe hearing impairment and lower WMC performed worse in the rhyming test, but perhaps due to their “deeper” semantic instead of phonological approach to the rhymes, they performed better in the final recognition test (e.g., Craik & Tulving, 1975). Thus, task demands and type of encoding operations relate to ELTM recognition. The above postdiction examples again show that WM is involved in early phonological stages of processing of a word or a sentence, and especially when SLTM is activated. Thus, the probability of mismatch increases when encoding/learning of phonological forms of processing requires transformation and recoding of stimuli at test. This implies that mismatch can occur at different levels of processing. But with a system that accomplishes rapid abstraction into multimodal representations, it will generally be less sensitive to the particular details of sensory transformations.
10.2.3 Extending the Mismatch Notion Recent studies of different speech distortions (Kennedy-Higgins et al., 2020) show that WM and vocabulary (i.e., SLTM) come out as the main predictors, irrespective of the type of distortion (time-compressed, noise-vocoded, and speech in noise). This tells us that the cognitive machinery underlying speech perception and speech understanding is rather invariant in its reliance on certain cognitive building blocks. But rapid abstraction into formats suitable for WM must take place for such a system to work. Another way of extending the notion of mismatch is explored in the study by Blomberg et al. (2019) of adults with Attention Deficit Hyperactivity Disorder (ADHD). She also used different kinds of speech distortions by orthogonally combining normal versus noise-vocoded speech (signal processed speech used to simulate cochlear implants) with type of background (clear speech, white noise, and speech babble). Age-appropriate
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
sentence materials were used (Swedish HINT, Hällgren et al., 2006). Results showed that compared to an age-matched control group, there was no interaction between group and type of masker or stimulus distortion (but expected main effects were observed), in line with the Kennedy-Higgins et al. (2020) findings. WMC was one significant predictor in all cases, which extends the generality and nonspecificity of the mismatch notion. Finally, a recent study elucidates that nonhabitual modes of processing may occur during you bimodally combined listening with a cochlear implant in one ear and a hearing aid in the other (Hua et al., 2017). These two types of signal reaching the brain will not necessarily be as rapidly integrated in RAMBPHO compared to regular sensory multimodal stimulation. The prediction was therefore that this kind of integration would demand WM resources to a high degree. As a matter of fact, the reading span correlation with performance in this bimodal condition was very high, suggesting still another kind of mismatch. Thus far, we have shown that habitual modes of speech understanding are less dependent on WM than nonhabitual modes, suggesting that the degree of mismatch is higher with nonhabitual modes.
10.2.4 Adaptation The ELU model predicts that when individuals become accustomed to the sounds transmitted by their hearing aids, they will automatize speech recognition and will be less dependent on WM to understand speech (Holmer & Rudner, 2020; Rönnberg et al., 2019). Ng et al. (2013) demonstrated that after a period of up to six months with new hearing aid settings, initial associations with WMC during speech recognition in noise became nonsignificant. However, the original ELU model did not appropriately cover such adaptive effects (see more under Section 10.4.1, “Development of Representations”). But, what was surprising to us was that in another recent sample of 200 hearing aid users (Rönnberg et al., 2016), dependence on WMC did not vanish for very experienced users (up to 10 years), especially for fourtalker babble backgrounds (Ng & Rönnberg, 2020). In Rönnberg et al. (2021), we ventured that given sufficiently dynamic and demanding communicative environments, we always need to explicitly activate WM, and that becoming accustomed to hearing aids does not solve all communication problems, especially in background talker noise, even for experienced hearing aid users. However, EEG-based attention-tracking techniques under development may improve perception of different talkerto-talker intentions during turn-taking (Alickovic et al., 2019).
10.3
ELU and Age-Related Hearing Loss
The ELU model assumes that different memory systems are differentially vulnerable to the long-term consequences of hearing loss. The reason, or
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
205
206
JERKER RÖNNBERG ET AL.
mechanism responsible, is assumed to be that of disuse. If we consider the everyday life of a hearing-impaired person, the ELU model posits that WM will always be actively engaged in trying to resolve ambiguities and misunderstandings during communication between individuals. At the same time, we can safely assume that not all misunderstandings and miscommunications are resolved by WM. This further implies that fewer phonologically based lexical accesses in SLTM will be encoded correctly into ELTM. That is, ELTM will fall into disuse: fewer encodings lead to fewer retrievals, and therefore, “the gymnastics” of ELTM will be less than optimal relative to WM. In the long-term, deterioration is expected. Another way to put it would be to say that the ELU model gives priority to the meaningful and matching SLTM activations, which then optimize ELTM encoding. Mismatching materials will not be readily encoded and retrieved from ELTM, or simply decayed and forgotten. It is also crucial for the development of SLTM that matching (or partially matching) are allowed to develop new representations (see below about the D-ELU model). While WM is engaged in postdiction, SLTM will be invoked for providing phonological constraints and lexical-semantic knowledge to narrow down the number of possible lexical candidates. This means that WM will use the fragments of information currently in mind, and then combine those with successively retrieved SLTM contributions. By this logic, WM has this inherent dual purpose of combining existing information (temporarily stored in WM) with inferences from information retrieved from SLTM (cf., storage and processing), while SLTM provides this semantic support only. ELTM encoding and retrieval will then be a consequence of the WM-SLTM interaction. So, on a use-disuse dimension, the general prediction is that the degree of deterioration of memory systems due to hearing loss is as follows, starting with the highest degree: ELTM > SLTM > WM. We also have empirical reason to believe that Age-Related Hearing Loss (ARHL) has consequences not only for encoding-storage in the auditory mode; it also drives memory system changes, memory systems that are multimodal in character. Two major studies demonstrate this: Rönnberg et al. (2011, 2014). In general, and in agreement with the disuse hypothesis, the negative hearing loss effect was manifest for the two LTM systems, but not for short-term memory, or WM. In Rönnberg et al. (2011) we tested the notion of multimodal representations that might be affected by hearing loss by using three different encoding tasks: oral recall of auditory presented word list (hearing aids on), oral recall of textually/auditory presented sentences, and thirdly, oral recall of motorically performed imperatives like “comb your hair” or “tie your shoelaces.” That is, all tasks had an encoding phase and a verbal, oral free recall phase after all items (12 for words and 16 for the enacted imperatives). All ELTM tests (free verbal recall but with different encoding instructions) were affected negatively by ARHL, and if anything, the highest simple correlation with hearing loss was the motorically encoded imperatives task.
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
The latent construct ELTM was significantly related to ARHL even when age was included in the structural equation models (Rönnberg et al., 2011). To accept these results we assume that representations in ELTM are multimodal and due to rapid binding of encoded information by RAMBPHO (cf., Baddeley, 2000). In the same vein, the study by Rönnberg et al. (2014), using a very large sample of participants from the UK Biobank Resource, showed that hearing loss was related to visual ELTM, yielding an effect size in the moderate range. This is a further argument with respect to the multimodality issue. Theoretically, the multimodality of LTM representations and selectivity of deterioration of multimodal memory systems is hard to reconcile with common cause accounts of cognitive aging and ARHL (e.g., Baltes & Lindenberger, 1997; Humes et al., 2013). These accounts typically assume some common neural degeneration that is responsible for a general cognitive decline, which would in our interpretation predict an equal deterioration across memory systems. The ELU model, on the other hand, assumes that hearing loss may cause cognitive decline in the disuse order discussed (Rönnberg et al., 2014, 2021). Data from other labs also show that gray matter volume is proportional to audiometric hearing loss, and is predictive of sentence comprehension (Lin et al., 2014; Peelle et al., 2011; Peelle & Wingfield, 2016). Lin (2011) and Lin et al. (2014) show that hearing loss may drive brain atrophy and cognitive impairment, which in turn will undermine cognitive integrity and ultimately lead to dementia (see review by Livingston et al., 2017). Combined, the above studies further testify to the possibility that hearing loss drives the successive, selective decline of memory systems. Indeed, we showed that even subclinical levels of poorer hearing in a middle-aged population are associated with smaller brain volumes in auditory and cognitive processing regions of the brain (Rudner et al., 2019). An even more recent study by Ayasse et al. (2019) shows that grammatical complexity is enough to tax the resources of participants with subclinical hearing losses (see also under Section 10.4, “Limitations of the ELU Model”). Furthermore, the data by Rönnberg et al. (2011, 2014) seem to rule out other explanations that suggest that the mechanism behind co-occurring auditory and cognitive decline is either information degradation (e.g., Schneider et al., 2002) or consumption of attention due to a degraded auditory stimulus (e.g., Verhaegen et al., 2014), because auditory only encoding did not suffer more than other encoding conditions (Rönnberg et al., 2011; see Hewitt, 2017; Livingstone et al., 2017; Roberts & Allen, 2016; Wayne & Johnsrude, 2015, for updated alternative accounts). Finally, the selectivity and multimodality effects cannot be explained by an aging SLTM only, because age alone did not alter the memory system selectivity (Rönnberg et al., 2011). SLTM structures like phonological neighborhoods, would be expected to deteriorate with age (Luce & Pisoni, 1998; Sommers, 1996), but, again, the selective hearing loss effect on memory
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
207
208
JERKER RÖNNBERG ET AL.
systems survived in the structural equation model (Rönnberg et al., 2011). As we have seen, even a minor hearing loss can lead to effects on memory and comprehension. This would fit with the overall argument that hearing loss drives problems with encoding (i.e., RAMBPHO delivered input) and subsequent retrievals from ELTM and SLTM systems, hence resulting in disused LTM systems (cf., Stamate et al., 2020). In short, ARHL is associated with multimodal deterioration of memory systems, presumably in the ELTM, SLTM, and WM order of sensitivity. The selectivity is dependent on use/disuse of a memory system, not on a particular sensory modality.
10.4
Limitations of the ELU Model
In this section, we will address some shortcomings of the ELU Model, specifically with respect to the development of representations, signed language, and hearing loss.
10.4.1 Development of Representations Holmer et al. (2016) proposed the D-ELU model to account for the way in which preexisting SLTM representations influence the development of new lexical representations. This model can also account for the adaptation of phonological representation driven by sustained hearing aid use. Furthermore, we proposed that novelty produces a mismatch, and that mismatch-induced postdictive processes push the system toward appropriate adjustment of SLTM, a process supported, and at the same time constrained, by existing representations in the lexicon (Holmer & Rudner, 2020; Rönnberg et al., 2019). The notion that WM dependence for hearing aid users becomes weaker over time (Ng et al., 2013), by reduced occurrence of mismatch (Rudner et al., 2008, 2009), is thus in line with the D-ELU model. The empirical base supporting the claims of the D-ELU model of how development of lexico-semantic representations in SLTM occurs is still limited. However, there are some experiments on novel word learning in hearing children reported by others (for an overview, see, e.g., Gray et al., 2020), and which we recently replicated in a Swedish study (Holmer & Witte, unpublished manuscript), with findings adhering to the principle that existing representations influence the probability of developing a lexico-semantic representation of a previously unfamiliar word. Even more specifically, there seems to be an interaction between the characteristics of the novel word and SLTM representations, which in turn is influenced by learning conditions. When word learning is challenging, such as for children (Hoover et al., 2010) or in background noise for adults (Han et al., 2016), optimal learning has been reported for novel words with a highly probable phonological pattern and a large number of phonological
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
neighbors in the lexicon, that is, novel words that in a sense have a high familiarity. However, learning rates at a similar level as for such “familiar” words have been reported for novel words with the exact opposite pattern of characteristics, that is, an unlikely phonological pattern and few phonological neighbors (Han et al., 2016; Hoover et al., 2010). This latter observation has been proposed to reflect that novel words that are very different from existing representations are easier to detect as novel, which leads to a higher probability of learning. Thus, the influence of SLTM representations might differ across different stages in the process of establishing new lexico-semantic representations; initially, they set the boundaries for novelty detection; in a later stage they support encoding of the novel item. In addition, from a D-ELU perspective, development of lexico-semantic representations is always constrained by WMC, and the general outline of developmental mechanisms of the model resembles ideas put forth by Gathercole and colleagues (e.g., Gathercole, 2006). Thus, investigating word learning, and the influence of SLTM on this process, when conditions are suboptimal due to contextual (e.g., noise) or individual (e.g., age, WMC, hearing loss) constraints is important for future refinement of the developmental version of the ELU model, D-ELU.
10.4.2 Signed Language Because early evidence suggested that sign language processing is organized in the brain in much the same way as speech processing (for reviews, see MacSweeney et al., 2008; Rönnberg, 2003), we initially believed that the ELU model could be applied with only minor adjustment to sign perception and communication. The ELU model is about language and should therefore be possible to generalize across language modality (Rönnberg, 2003; Rönnberg et al., 2008). However, we have successively noted some constraints on the ELU model when comparing WM for sign and speech. For example, in Rönnberg et al. (2004), WM for sign showed specific activations of superior parietal and temporo-occipital areas but communality with audiovisual speech for prefrontal/frontal areas of the cortex, typically associated with WM processing (Eriksson et al., 2015), was still true. These results were later replicated (e.g., Bavelier et al., 2008; Rudner et al., 2007). In a study on WM for sign compared to moving visual nonsense objects (displays of dots moving in a way that mimic human gestures/actions), Cardin et al. (2018) demonstrated another kind of language modality specificity: while frontal activations were similar in an n-back task for deaf native signers, compared to hearing native signers and hearing nonsigners, it turned out that auditory cortex was specifically active in the deaf native signing group for both kinds of visual stimuli (cf., Cardin et al., 2013). Further, resting-state connectivity analysis showed increased connectivity between frontal regions and auditory regions in deaf compared to hearing individuals (cf., Ding et al., 2016).
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
209
210
JERKER RÖNNBERG ET AL.
To investigate this supposition, Andin et al. (2021) parametrically manipulated load during a sign-based n-back WM task. Although there was greater activation in auditory cortex for deaf compared to hearing participants, activation decreased as load increased in both groups. Thus, it seems that auditory cortex represents a cognitive processing resource for deaf individuals, and its function seems inversely load-related. Generally speaking, WM for sign language shows two kinds of language modality specificity: (1) a larger degree of activity in parietal regions relative to spoken language and (2) plastic changes due to early sign language use regarding reliance on superior temporal regions. This limits the generality of the ELU model. However, WM load in the visual modality (signed or by using letters), dampens activity in the temporal cortex (Andin et al., 2021; Sörqvist et al., 2016), which actually reinforces similarities across language modalities.
10.4.3 Hearing Loss Limitations of the ELU model with respect to hearing status have also been addressed in one meta-analytic study by Füllgrabe and Rosen (2016), demonstrating that for hearing impaired and older participants, WM accounts for significant portions of variance in explaining speech-in-noise performance, whereas this was not the case for normal hearing young participants. However, later research has shown that even for subclinical/normal hearing individuals, small variations in hearing acuity is associated with atrophy of the brain in predicted auditory and cognitive cortical sites (Rudner et al., 2019). This result casts doubt on this putative limitation. The same message is true for a paper by Ayasse et al. (2019), where variations within the limits of normal hearing show that even minimal hearing loss has effects on listening to sentences in noise, when sentence grammar is more complex and WM-demanding. Thus, there are interesting findings on hearing status as a constraint on the ELU model such that it can be argued that the model also applies to so-called normal or subclinical variability, even though these latter studies did not involve WM capacity as such (see reference to WMC and context dependence in normal hearing participants, Rönnberg et al., 2019). The general message is that it does not take much hearing loss to affect an individual’s ability for information processing. Thus, this constraint is able to pick up interesting effects that need to be further explored using, for example, analyses of individuals with pure tone averages focusing on high frequency loss.
10.5
Implications
We have shown with our own and others´ experimental data that in the context of communication, especially in adverse conditions, a WMC
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
concept that emphasizes storage and processing functions plays important roles for both prediction and postdiction. Postdiction and prediction work on different time scales and interac with ELTM and SLTM. RAMBPHO encoding results in multimodal representations. ARHL results in selectivity in the deterioration of multimodal memory systems predicted by the ELU model, results that are not easily handled by other cognitive theories. We have also shown that the developing ELU framework, although a fruitful source of testable hypotheses, might not be straightforwardly applied to individuals with early profound deafness who are sign language users.
10.6
Conclusion and Future Directions
Future developments and experiments to test the ELU model must deal with the limitations mentioned: (1) a D-ELU model that takes into account under what conditions the putative learning mechanism from the postdiction-prediction interaction loop is possible; (2) language-modality constraints exemplified by signed languages; (3) a longitudinal ARHL study (is under way) that evaluates memory system selectivity in relation to hearing loss, cognitive components related to the ELU model (see e.g., Rönnberg et al., 2016), and eventual MCI and Alzheimer’s dementia; and (4) more intelligent hearing aids are needed that use the brain activity of an individual (e.g., top-down modulation of neural envelopes, Decrui et al., 2020). Here, the ELU model can play a decisive role.
References Andin, J., Holmer, E., Schönström, K., & Rudner, M. (2021). Working memory for signs with poor visual resolution: fMRI evidence of reorganization of auditory cortex in deaf signers. Cerebral Cortex. 31(7), 3165–3176. Alickovic, E., Lunner, T., Gustafsson, F., & Ljung, L. (2019). A tutorial on auditory attention identification methods. Frontiers in Neuroscience, 13, 153. Anderson, S., White-Schwoch, T., Parbery-Clark, A., & Kraus, N. (2013). A cognitive system supports speech-in-noise perception in older adults. Hearing Research, 300, 218–232. Arehart, K. H., Souza, P., Baca, R., & Kates, J. M. (2013). Working memory, age, and hearing loss: Susceptibility to hearing aid distortion. Ear and Hearing, 34(3), 251–260. Arlinger, S., Lunner, T., Lyxell, B., & Pichora-Fuller, M. (2009). The emergence of cognitive hearing science. Scandinavian Journal of Psychology, 50, 371–384
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
211
212
JERKER RÖNNBERG ET AL.
Ayasse, N., Penn, L., & Wingfield, A. (2019). Variations within normal hearing acuity and speech comprehension: An exploratory study. American Journal of Audiology, 28(2), 369–375. Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417–423. Baddeley, A. D. (2012). Working memory: Theories, models, and controversies. Annual Review of Psychology, 63, 1–29. Baltes, P. B., & Lindenberger, U. (1997). Emergence of a powerful connection between sensory and cognitive functions across the adult lifespan: A new window to the study of cognitive aging? Psychology and Aging, 12(1), 12–21. Barrouillet, P., & Camos, V. (2020). The time-based resource-sharing model of working memory. In R. H. Logie, V. Camos, N. Cowan (Eds), Working memory: State of the science. Oxford University Press. Bavelier, D., Newman, A. J., Mukherjee, M., Hauser, P., Kemeny, S., Braun, A., et al. (2008). Encoding, rehearsal, and recall in signers and speakers: Shared network but differential engagement. Cerebral Cortex, 18, 2263–2274. Blomberg, R., Danielsson, H., Rudner, M., Söderlund, G. B. W., & Rönnberg, J. (2019). Speech processing difficulties in Attention Deficit Hyperactivity Disorder. Frontiers in Psychology, 10, 1536. Cardin, V., Orfanidou, E., Rönnberg, J., Capek, C. M., Rudner, M. & Woll, B. (2013). Dissociating cognitive and sensory neural plasticity in human superior temporal cortex. Nature Communications, 4, 1473. Cardin, V., Rudner, M., De Oliveira, R. F., Andin, J., Su, M. T., Beese, L., Woll, B., & Rönnberg, J. (2018). The organization of working memory networks is shaped by early sensory experience. Cerebral Cortex, 28(10), 3540–3554. Classon, E., Rudner, M., & Rönnberg, J. (2013). Working memory compensates for hearing related phonological processing deficit. Journal of Communication Disorders, 46(1), 17–29. Cowan, N. (2005). Working memory capacity. Psychology Press. Craik, F., & Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General, 104 (3), 268–294. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19(4), 450–466. Decrui, L., Lesenfants, D., Vanthornhout, J., & Francart, T. (2020). Top-down modulation of neural envelope tracking: The interplay with behavioral, self-report and neural measures of listening effort. The European Journal of Neuroscience, 52(5), 3375–3393. Ding, H., Ming, D., Wan, B., Li, Q., Qin, W., & Yu, C. (2016). Enhanced spontaneous functional connectivity of the superior temporal gyrus in early deafness. Scientific Reports, 6, 23239.
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
Eriksson, J., Vogel, E. K., Lansner, A., Bergström, F., & Nyberg, L. (2015). Neurocognitive architecture of working memory. Neuron, 88(1), 33–46. Farias, S. T., Lau, K., Harvey, D. J., Denny, K. G., Barba C., & Mefford, A. N. (2017). Early functional limitations in cognitively normal older adults predict diagnostic conversion to mild cognitive impairment. Journal of the American Geriatrics Society, AA65 (6), 1152–1158. Foo, C., Rudner, M., Rönnberg, J., & Lunner, T. (2007). Recognition of speech in noise with new hearing instrument compression release settings requires explicit cognitive storage and processing capacity. Journal of the American Academy of Audiology, 18, 553–566. Fortunato, S., Forli, F., Guglielmi, V., De Corso, E., Paludetti, G., Berrettini, S., & Fetoni, A. R. (2016). A review of new insights on the association between hearing loss and cognitive decline in ageing. Acta Otorhinolaryngologica Italica, 36, 155–166. Füllgrabe, C., & Rosen, S. (2016). On the (un)importance of working memory in speech-in-noise processing for listeners with normal hearing thresholds. Frontiers in Psychology, 7, 1268. Gathercole, S. E. (2006). Nonword repetition and word learning: The nature of the relationship. Applied Psycholinguistics, 27(4), 513–543. Gray, S., Lancaster, H., Alt, M., Hogan, T. P., Green, S., Levy, R., & Cowan, N. (2020). The structure of word learning in young school-age children. Journal of Speech, Language, and Hearing Research, 63(5), 1446–1466. Grosjean, F. (1980). Spoken word recognition processes and gating paradigm. Perception & Psychophysics, 28, 267–283. Hagerman, B. (1982). Sentences for testing speech intelligibility in noise. Scandinavian Audiology, 11, 79–87. Hällgren, M., Larsby, B., & Arlinger, S. (2006). A Swedish version of the hearing in noise test (HINT) for measurement of speech recognition. International Journal of Audiology, 45, 227–237. Han, M. K., Storkel, H. L., Lee, J., & Cox, C. (2016). The effects of phonotactic probability and neighborhood density on adults’ word learning in noisy conditions. American Journal of Speech-Language Pathology, 25, 547–560. Hewitt, D. (2017). Age-related hearing loss and cognitive decline: You haven’t heard the half of it. Frontiers in Aging Neuroscience, 9, 112. Hickok, G., & Poeppel, D. (2007). The cortical organization of speech processing. Nature Reviews Neuroscience, 8, 393–402. Holmer, E., Heimann, M., & Rudner, M. (2016). Imitation, sign language skill and the developmental Ease of Language Understanding (D-ELU) Model. Frontiers in Psychology, 7, 107. Holmer, E., & Rudner, M. (2020). Developmental Ease of Language Understanding mode and literacy acquisition: Evidence from deaf and hard-of-hearing signing children. In Q. Y. Wang & J. F. Andrews (Eds.), Multiple paths to become literate: International perspective in deaf education. Gallaudet University Press.
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
213
214
JERKER RÖNNBERG ET AL.
Holmer, E., & Witte, E. (unpublished manuscript). Phonotactic probability and phonological neighborhood density interact in word learning in Swedish schoolchildren. Hoover, J. R., Storkel, H. L., & Hogan, T. P. (2010). A cross-sectional comparison of the effects of phonotactic probability and neighborhood density on word learning by preschool children. Journal of Memory and Language, 63(1), 100–116. Hua, H., Johansson, B., Lyxell, B., Magnusson, L., & Ellis, R. J. (2017). Speech recognition and cognitive skills in bimodal cochlear implant users. Journal of Speech, Language, and Hearing Research, 60(9), 1–12. Humes, L. E., Busey, T. A., Craig, J., & Kewley-Port, D. (2013). Are age-related changes in cognitive function driven by age-related changes in sensory processing? Attention, Perception, & Psychophysics, 75, 508–524. Kennedy-Higgins, D., Devlin, J. T., & Adank, P. (2020). Cognitive mechanisms underpinning successful perception of different speech distortions. Journal of the Acoustical Society of America, 147(4), 2728–2740. Kilman, L., Zekveld, A., Hällgren, M., & Rönnberg, J. (2014). The influence of non-native language proficiency on speech perception performance. Frontiers in Psychology, 5, 651. Kilman, L., Zekveld, A., Hällgren, M., & Rönnberg, J. (2015). Native and nonnative speech perception by hearing-impaired listeners in noise- and speech maskers. Trends in Hearing, 19. Kraus N., & White-Schwoch, T. (2015). Unraveling the biology of auditory learning: A cognitive sensorimotor- reward framework. Trends in Cognitive Science, 19, 642–654. Lin, F. (2011). Hearing loss and cognition among older adults in the United States. The Journals of Gerontology: Series A, 66A(10), 1131–1136. Lin, F. R., Ferrucci, L., An, Y., Goh, J. O., Doshi, J., Metter, E. J., (. . . ) Resnick, S. M. (2014). Association of hearing impairment with brain volume changes in older adults. NeuroImage, 90, 84–92. Lin, F. R., Metter, E. J., O’Brien, R. J., Resnick, S. M., Zonderman, A. B., & Ferrucci, L. (2011). Hearing loss and incident dementia. Archives of Neurology, 68(2), 214–220. Livingston, G., Sommerlad, A., Orgeta, V., Costafreda, S. G., Huntley, J., Ames, D.,. . .Mukadam, N. (2017). Dementia prevention, intervention, and care. Lancet, 390, 2673–734. Luce, P. A., & Pisoni. D. B. (1998). Recognizing spoken words: The neighborhood activation model. Ear and Hearing, 19(1), 1–36. Lunner, T. (2003). Cognitive function in relation to hearing aid use. International Journal of Audiology, 42(Suppl 1), 49–58. Lunner, T., Rudner, M., & Rönnberg, J. (2009). Cognition and hearing aids. Scandinavian Journal of Psychology, 50(5), 395–403. Lunner, T., & Sundewall-Thorén, E. (2007). Interactions between cognition, compression, and listening conditions: Effects on speech-in-noise
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
performance in a 2-channel hearing aid. Journal of the American Academy of Audiology, 18(7), 604–617. McGurk, H., & MacDonald, J. (1976). Hearing lips and seeing voices. Nature, 264(5588), 746–748. Marsh, J. E., & Campbell, T. A. (2016). Processing complex sounds passing through the rostral brain stem: The New Early Filter Model. Frontiers in Neuroscience, 10, 136. MacSweeney, M., Capek, C. M., Campbell, R., & Woll, B. (2008). The signing brain: The neurobiology of sign language. Trends in Cognitive Science, 12, 432–440. Mattys, S. L., Davis, M. H., Bradlow, A. R., & Scott, S. (2012). Speech recognition in adverse conditions: A review. Language and Cognitive Processes, 27 (7–8), 953–978. Moradi, S., Lidestam, B., Ng, E. H. N., Danielsson, H., & Rönnberg, J. (2017). Visual cues contribute differentially in audiovisual perception of consonants and vowels in improving recognition and reducing cognitive demands. Journal of Speech, Language, and Hearing Research, 60, 2687–2703. Moradi, S., Lidestam, B., & Rönnberg, J. (2013). Gated audiovisual speech identification in silence vs. noise: Effects on time and accuracy. Frontiers in Psychology, 4, 359. Moradi, S., Lidestam, B., Saremi, A., & Rönnberg, J. (2014). Gated auditory speech perception: Effects of listening conditions and cognitive capacity. Frontiers in Psychology, 5, 531. Näätänen, R., & Escera, C. (2000). Mismatch negativity: Clinical and other applications. Audiology and Neurootology, 5, 105–110. Ng, E. H. N., & Rönnberg, J. (2020). Hearing aid experience and background noise affect the robust relationship between working memory and speech recognition in noise. International Journal of Audiology, 59(3), 208–218. Ng, E. H. N., Rudner, M., Lunner, T., Pedersen, M. S., & Rönnberg, J. (2013). Effects of noise and working memory capacity on memory processing of speech for hearing-aid users. International Journal of Audiology, 52(7), 433–441. Ng, E. H. N., Rudner, M., Lunner, T., & Rönnberg, J. (2015). Noise reduction improves memory for target language speech in competing native but not foreign language speech. Ear and Hearing, 36(1), 82–91. Peelle, J. E., Troiani, V., Grossman, M., & Wingfield, A. (2011). Hearing loss in older adults affects neural systems supporting speech comprehension. Journal of Neuroscience, 31,12638–12643. Peelle J. E., & Wingfield, A. (2016). The neural consequences of age-related hearing loss. Trends in Neuroscience, 39(7), 486–497. Poeppel, D., Idsardi, W. J., & van Wassenhove,V. (2008). Speech perception at the interface of neurobiology and linguistics. Philosophical Transactions of the Royal Society B: Biological Sciences, 363, 1071–1086.
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
215
216
JERKER RÖNNBERG ET AL.
Roberts, K. L., & Allen, H. A. (2016). Perception and cognition in the ageing brain: A brief review of the short- and long-term links between perceptual and cognitive decline. Frontiers in Aging Neuroscience, 8, 39. Rudner, M., Foo, C., Rönnberg, J., & Lunner, T. (2009). Cognition and aided speech recognition in noise: Specific role for cognitive factors following nine-week experience with adjusted compression settings in hearing aids. Scandinavian Journal of Psychology, 50(5), 405–418. Rudner, M., Foo, C., Sundewall Thorén, E., Lunner, T., & Rönnberg, J. (2008). Phonological mismatch and explicit cognitive processing in a sample of 102 hearing aid users. International Journal of Audiology, 47 (Suppl. 2), S163–S170. Rudner, M., Fransson, P., Ingvar, M., Nyberg, L. & Rönnberg, J. (2007). Neural representation of binding lexical signs and words in the episodic buffer of working memory. Neuropsychologia, 45(10), 2258–2276. Rudner, M., Seeto, M., Keidser, G., Johnson, B., & Rönnberg, J. (2019). Poorer speech reception threshold in noise is associated with reduced brain volume in auditory and cognitive processing regions. Journal of Speech, Language, and Hearing Research, 62(4S), 1117–1130. Rönnberg, J. (2003). Cognition in the hearing impaired and deaf as a bridge between signal and dialogue: A framework and a model. International Journal of Audiology, 42 (Suppl. 1), S68–S76. Rönnberg, J. Arlinger, S., Lyxell, B., & Kinnefors, C. (1989). Visual evoked potentials: Relation to adult speechreading and cognitive function. Journal of Speech and Hearing Research, 32(4), 725–735. Rönnberg, J., Danielsson, H., Rudner, M., Arlinger, S., Sternäng, O., Wahlin, Å., & Nilsson, L-G. (2011). Hearing loss is negatively related to episodic and semantic long-term memory but not to short-term memory. Journal of Speech, Language, and Hearing Research, 54, 705–726. Rönnberg, J., Holmer, E., & Rudner, M. (2019). Cognitive hearing science and ease of language understanding. International Journal of Audiology, 58(5), 247–261. Rönnberg, J., Holmer, E., & Rudner, M. (2021). Cognitive hearing science: Three memory systems, two approaches, and the ease of language understanding model. Journal of Speech, Language, and Hearing Research, 64(2), 359–370. Rönnberg, J., Hygge, S., Keidser, G., & Rudner, M. (2014). The effect of functional hearing loss and age on long- and short-term visuospatial memory: Evidence from the UK Biobank Resource. Frontiers in Aging Neuroscience, 6, 326. Rönnberg, J., Lunner, T., Ng, E. H. N., Lidestam, B., Zekveld, A. A., Sörqvist, P., ( . . . ) Stenfelt, S. (2016). Hearing impairment, cognition and speech understanding: Exploratory factor analyses of a comprehensive test battery for a group of hearing aid users, the n200 Study. International Journal of Audiology, 55(11), 623–642.
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
Ease of Language Understanding Model
Rönnberg, J., Lunner, T., Zekveld, A. A., Sörqvist, P., Danielsson, H., Lyxell, B., (. . .) Rudner, M. (2013). The Ease of Language Understanding (ELU) model: Theoretical, empirical, and clinical advances. Frontiers in Systems Neuroscience, 7, 31. Rönnberg, J., Rudner, M., Foo, C., & Lunner, T. (2008). Cognition counts: A working memory system for Ease of Language Understanding (ELU). International Journal of Audiology, 47 (Suppl 2), S99–S105. Rönnberg, J., Rudner, M. & Ingvar, M. (2004). Neural correlates of working memory for sign language. Cognitive Brain Research, 20, 165–182. Schneider, B. A., Daneman, M., & Pichora-Fuller, M. K. (2002). Listening in aging adults: From discourse comprehension to psychoacoustics. Canadian Journal of Experimental Psychology, 56, 139–152. Signoret C., Andersen, L. M., Dahlström, Ö., Blomberg, R, Lundqvist, D., Rudner, M., & Rönnberg, J. (2020) The influence of form- and meaningbased predictions on cortical speech processing under challenging listening conditions: A MEG Study. Frontiers in Neuroscience, 14, 573254 Signoret, C., Johnsrude, I., Classon, E., & Rudner, M. (2018). Combined effects of form- and meaning-based predictability on perceived clarity of speech. Journal of Experimental Psychology: Human Performance and Perception, 44(2), 277–285. Signoret, C., & Rudner, M. (2019). Hearing impairment and perceived clarity of predictable speech. Ear and Hearing, 40 (5), 1140–1148. Sommers, M. S. (1996). The structural organization of the mental lexicon and its contribution to age-related declines in spoken-word recognition. Psychology and Aging, 11, 333–341. Souza, P., Arehart, K. H., Shen, J., Anderson, M., & Kates, J. M. (2015). Working memory and intelligibility of hearing-aid processed speech. Frontiers in Psychology, 6, 526. Souza, P., Arehart, K., Schoof, T., Anderson, M., Strori, D., & Balmert, L. (2019). Understanding variability in individual response to hearing aid signal processing in wearable hearing aids. Ear and Hearing, 40(6), 1280–1292. Souza, P., & Sirow, L. (2014). Relating working memory to compression parameters in clinically fit hearing aids. American Journal of Audiology, 23(4), 394–401. Stamate, A., Logie, R. H., Baddeley, A. D., & Sala, S. D. (2020). Forgetting in Alzheimer’s disease: Is it fast? Is it affected by repeated retrieval? Neuropsychologia, 138, 107351. Sörqvist, P., Dahlström, Ö., Karlsson, T., & Rönnberg, T. J. (2016). Concentration: The neural underpinnings of how cognitive load shields against distraction. Frontiers in Human Neuroscience, 10, 221. Sörqvist, P., & Rönnberg, J. (2012). Episodic long-term memory of spoken discourse masked by speech: What is the role for working memory capacity? Journal of Speech, Language, and Hearing Research, 55(1), 210–218.
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
217
218
JERKER RÖNNBERG ET AL.
Sörqvist, P., Stenfelt, S., & Rönnberg, J. (2012). Working memory capacity and visual-verbal cognitive load modulate auditory-sensory gating in the brainstem: Toward a unified view of attention. Journal of Cognitive Neuroscience, 24(11), 2147–2154. Verhaegen, C., Collette, F., & Majerus, S. (2014). The impact of aging and hearing status on verbal short-term memory: Neuropsychology, development, and cognition. B Aging Neuropsychology and Cognition, 21, 464–482. Wayne, R. V., & Johnsrude, I. S. (2015). A review of causal mechanisms underlying the link between age-related hearing loss and cognitive decline. Ageing Research Reviews, 23(Pt B), 154–166. Wild, C. J., Yusuf, A., Wilson, E., Peelle, J. P., Davis, M. H., & Johnsrude, I. S. (2012). Effortful Listening: The processing of degraded speech depends critically on attention. Journal of Neuroscience, 32(40), 14010–14021. Zekveld, A. A., Rudner, M., Johnsrude, I. S., Festen, J. M., van Beek, J. H. M., & Rönnberg, J. (2011). The influence of semantically related and unrelated text cues on the intelligibility of sentences in noise. Ear and Hearing, 32(6), e16–e25. Zekveld, A. A., Rudner, M., Johnsrude, I. S., & Rönnberg, J. (2013). The effects of working memory capacity and semantic cues on the intelligibility of speech in noise. Journal of the Acoustical Society of America, 134(3), 2225–2234.
https://doi.org/10.1017/9781108955638.013 Published online by Cambridge University Press
11 Assessing Children’s Working Memory Milton J. Dehn
11.1
Introduction
A comprehensive assessment of working memory should be conducted whenever a child is referred for a psychological or neuropsychological evaluation, especially when the referral concerns include language development or academic learning problems. Language development, acquisition of academic skills, and performance of academic skills all rely heavily on working memory. The strong relations between working memory and language development (Moser et al., 2007) and working memory and academic achievement (Swanson et al., 2009) are well documented. Correlations between working memory measures and academic achievement generally range from .3 to .6 (Swanson, 2006, 2011; Swanson et al., 2009; Swanson & Berninger, 1995, 1996; Swanson & Jerman, 2007). Students with low working memory capacity are at risk for learning problems. Gathercole and Alloway (2008) found that 80 percent of students who had working memory scores at the 10th percentile or lower experienced significant academic learning difficulties or specific learning disabilities (SLD). For example, the presence of working memory deficits in many students with dyslexia is well documented (Palmer, 2000; Pickering, 2006; Siegal & Ryan, 1989; Swanson et al., 2009; Wang & Gathercole, 2013). The predominant model of working memory was proposed by Baddeley and Hitch in 1974 and later expanded by Baddeley (1986, 2000, 2006). Baddeley’s hierarchical model is composed of four components: the phonological loop, the visuospatial sketchpad, the central executive, and the episodic buffer, with the central executive functioning as the manager of the other three components. In contrast, Kane and Engle’s (Engle, 2002) Executive Attention Model and Cowan’s (2005) Embedded-Process Model place less emphasis on a hierarchical arrangement while defining the essence of working memory as executive attention and the focus of attention, respectively.
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
220
M I LT O N J . D E H N
In 2008, Dehn proposed the Integrated Model of working memory, a fivecomponent (or five processes) model (see Figure 11.1) that is useful for assessing working memory in children, especially those referred for academic learning problems. These five working memory processes and their labels differ somewhat from Baddeley’s (2006) model. The main difference is that Baddeley does not divide the “central executive” into verbal and visual-spatial processing. Yet, verbal and visual-spatial working memory at the processing level can be differentiated (Alloway et al., 2006), and some assessment instruments make this distinction (e.g., Alloway, 2007; Roid, 2003). Furthermore, there is evidence that verbal and visual-spatial working memory can be differentiated at the neurological level (Bunge & Wright, 2007; Smith & Jonides, 1997). Similarly, the storage and processing aspects of working memory can also be separated (Alloway et al., 2006). The remainder of this chapter approaches working memory assessment from Dehn’s model. From Dehn’s assessment perspective, working memory might best be defined as the short-term retention of information while processing the same or other information. A measure that assesses both storage and processing is the classic backward digit span test. Retaining the presented digits is the storage, while reversing the sequence is assumed to tap the processing dimension of working memory. The neurological functioning of working memory is complex; thus, the construct warrants an in-depth evaluation that assesses each of its main constituents. The results of a simple span test, such as forward digit span, do not tell the whole story. The storage and processing dimensions need to be differentiated, and each of these needs to be examined from a verbal and visual-spatial perspective. Also, the role of working memory executive processes (i.e., the central executive), such as inhibition, needs consideration. Finally, a few highly related cognitive processes, such as processing speed, should be measured and compared with working memory functioning (Dehn, 2015).
Figure 11.1 Working memory processes
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
11.2
The Need to Assess Specific Working Memory Processes
In Dehn’s (2008, 2015) integrated assessment model there are five specific working memory components or processes (see Figure 11.1) that should be assessed during an evaluation: verbal short-term storage, visual-spatial short-term storage, verbal processing (also referred to herein as verbal working memory), visual-spatial processing (also referred to herein as visual-spatial working memory), and executive processes (also referred to herein as executive working memory). Further support for identifying five specific working memory components or processes for assessment purposes comes from ample neuroimaging evidence that each of these five components activates distinct brain regions. For example, specific areas of the parietal and temporal lobes are involved in verbal and visual-spatial short-term storage (Hedden & Yoon, 2006). The left prefrontal region (Broca’s area) is active when verbal content is being processed, whereas the right prefrontal region processes visual-spatial information. When both types are being processed simultaneously or when executive functions are needed, the dorsolateral prefrontal cortex is involved. Theoretically, these five working memory processes have a hierarchical arrangement whereby executive functions oversee the entire working memory system, and verbal and visual spatial processing manage their respective storage components (see Figure 11.1). Nonetheless, each process can function semiautonomously. For example, a short-term storage component may have normal retention even while the processing level functions poorly. Essentially, all five working memory processes should be tested so that a meaningful profile of strengths and weaknesses can be determined. Identification of the weak links in working memory functioning allows deeper understanding of the examinee’s impairments and enables focused interventions and accommodations. For instance, an individual may have weak storage capacity but stronger performance when there are processing demands. Alternately, an individual may have a weakness in executive working memory processes that impacts overall performance. For example, children with ADHD are likely to have normal short-term storage capacity but perform poorly when working memory processing is required (Martinussen & Tannock, 2006). It also been reported that visual-spatial working memory is weaker than verbal working memory in children with ADHD (Martinussen et al., 2005). Individuals with language delays and disorders are expected to frequently have weak verbal storage and processing, but omitting measurement of their visual-spatial storage and processing capabilities would result in an incomplete picture and potentially hamper designing an effective
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
221
222
M I LT O N J . D E H N
treatment. For example, teaching the examinee to use visual imagery to support weak verbal processing is more appropriate when visual-spatial processing is known to be a relative strength.
11.3
Selective, Multibattery Testing
Most contemporary cognitive and memory assessment batteries measure multiple aspects of working memory, but none produces scores for all five specific working memory processes. This reality will require examiners to draw from more than one battery in a selective-testing fashion. In multibattery testing, the examiner should begin by noting which working memory processes are covered by the primary battery. If a broad cognitive scale, such as the WISC-V, is administered, then only those working memory processes not assessed by the WISC-V need to be tested with other batteries. Supplemental batteries should be administered in a selectivetesting manner, whereby only the desired subtests are administered, and subtests that duplicate what has already been adequately tested are omitted. A combination of cognitive, neuropsychological, and memory scales may be used. To accomplish such multibattery testing efficiently, it is best to plan in advance, selecting scales and subtests before actual testing begins. Whenever feasible, each of the five processes should be tested with a minimum of two subtests (for broader sampling of the construct and to increase the reliability of the results). Whenever they are available, composite scores should be used to identify potential weaknesses and strengths. Composites are preferable because they have less measurement error (Sattler, 2020); they are more reliable than individual subtest scores. Unfortunately, not many standardized working memory composites can be recommended because many composites mix specific processes, such as including both verbal and visual-spatial subtests in the same composite. On occasion, one subtest score will suffice, especially if there is no reason to suspect a weakness in that specific process.
11.4
Identifying Subtests That Measure Specific Working Memory Processes
Choosing subtests that tap each of the five processes can be challenging. First, some subtest names are misleading, such as when a classic measure of verbal short-term storage is labeled as an “attention” subtest. Second, subtest names and tasks vary considerably across measurement scales. Third, even subtests designated as specific working memory subtests frequently measure more than one of the five processes, as well as measuring elements of other cognitive abilities. These kinds of subtests lack subtest
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
specificity. Consequently, it is frequently necessary to task analyze the content and demands of subtests to determine the primary working memory process(es) being measured. Selection of subtests should not be limited to memory batteries. Many cognitive, achievement, language, memory, and neuropsychological batteries contain subtests that are valid measures of specific working memory processes but are not identified as such. In such instances, a subtest might be classified and used as a specific working memory measure even when the instrument’s authors are using the subtest for another purpose. For example, in the Woodcock-Johnson IV Tests of Oral Language (Schrank et al., 2014) a subtest called Sentence Repetition is categorized by the authors as a measure of oral expression. However, it can also be used as a verbal working memory measure because the task primarily requires the verbatim repetition of sentences. Verbal repetition of orally presented information is the most common way that working memory’s verbal storage and processing are measured. The extent to which a subtest measures a specific process can also be determined through expert opinion, correlational studies, and factor analysis. An example of factor analysis reorienting how a subtest should be viewed is the Arithmetic subtest of the WISC-IV. The WISC-IV structure aligned Arithmetic with the Working Memory Index. However, a factoranalytic study revealed that Arithmetic was measuring reasoning more than working memory (Keith et al., 2006). Task analysis and other sources of validity data were used in Dehn’s Memory Processes Analyzer (2018) when creating lists of working memory measures divided into the five specific working memory processes. Clinicians themselves can employ task analysis to decide which specific working memory process a subtest is primarily measuring. With task analysis, input (directions and stimuli), processing, and output (responses) are all considered, but the processing requirements should be given the most weight. These variables should then be considered in view of how the specific working memory process is defined. For example, if the task primarily involves storage of phonological/auditory/verbal information without any significant need to process the information, it should be considered a measure of verbal short-term storage. Variables to consider when classifying subtests are presented in the next five subsections.
11.4.1 Verbal Storage Even though Baddeley’s (2006) model considers short-term storage to be part of working memory, testing short-term storage separately is important because storage and processing functions are different from each other and because they can be separated neurologically (Baldo & Dronkers, 2006; Cowan, 1995; Hedden & Yoon, 2006; Prabhakaran et al., 2000). When an assessment fails to include measures that isolate short-term storage from
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
223
224
M I LT O N J . D E H N
processing and executive dimensions, it can be difficult to determine where the specific working memory weaknesses lie. A subtest measures shortterm storage whenever storage but no additional processing, or only minimal processing, is required. Forward digit span is a good example of a verbal short-term storage task because the digits can immediately be repeated without any manipulation or processing. To be considered a relatively specific and valid verbal storage measure, the task should: (a) have orally presented stimuli, (b) not require the examinee to read the stimuli, (c) require an oral response, (d) require sequential recall of the stimuli, (e) not involve any significant manipulation or processing of information, (f ) not include an irrelevant, distracting task that creates interference, and (g) allow vocal or subvocal rehearsal. To reduce the influence of lexical knowledge from long-term memory, nonwords are ideal stimuli.
11.4.2 Verbal Processing When an assessment task requires immediate recall of verbal information and the examinee’s recall must be verbatim or nearly verbatim, the task can be used to measure verbal working memory. However, the task should involve more than just rote recall, which mainly requires just brief storage, such as recalling a list of unrelated words. There should also be some mental processing demands or some associations with prior knowledge to consider it more than a measure of brief storage. For example, a memory for sentences task (each item is an orally presented sentence that the examinee must immediately recall perfectly or with only one error) is classified as verbal working memory rather than short-term storage because sentences invoke meaning-based processing, which promotes grouping the words into meaningful phrases. The result is several more words being recalled than the span for a series of unconnected words. Another example of a challenging verbal task is remembering the last word in a series of sentences that are orally presented. Although most verbal working memory tasks involve sequential recall, nonsequential tasks are still tapping verbal working memory.
11.4.3 Visual-Spatial Storage Tasks that involve recalling visual-spatial stimuli or their location without a need to process the information further, deal with distractors, or manipulate images can be classified as visual-spatial short-term storage measures. Although standardized tasks of this nature often include pictures of common objects, it is best to use abstract stimuli that cannot be easily named. When the stimuli are named, performance on the task may be confounded by verbal processes, thereby prohibiting a relatively pure evaluation of visual-spatial storage. Visual-spatial storage tasks may or
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
may not require sequential recall. Classic block-tapping tasks require sequential recall, but much of real-world visual-spatial processing is not sequential in nature.
11.4.4 Visual-Spatial Processing A visual-spatial task graduates from storage to storage with processing when the mental manipulation of stimuli or internally generated images is required. Specific measures of visual-spatial working memory are uncommon. Examples of such tasks include maintaining information during rotation and reversing the sequence, such as in a block-tapping activity.
11.4.5 Executive Working Memory It can be difficult to differentiate between the verbal and visual-spatial processing levels and executive working memory. In general, executive tasks are more challenging than either verbal or visual-spatial tasks. There are four main ways that executive processing can be differentiated from the more modality-specific processing components: (a) the task requires integration of verbal and visual-spatial information, such as immediately recalling new symbols (rebuses) that represent words; (b) the task introduces distracting information or requires task-irrelevant processing; (c) the task requires extensive, ongoing inhibition, updating, or switching; (d) there is a conscious application of a strategy other than basic rehearsal (repetition), such as chunking (grouping) words or digits together or visualizing verbal information.
11.5
Assessing Related Cognitive and Executive Processes
Working memory is a core cognitive and executive faculty that is highly interrelated with most cognitive and executive processes. Some of these processes, such as fluid reasoning (Dehn, 2017) and oral language, depend heavily on adequate working memory functioning, some have reciprocal relations with working memory, and others strongly influence working memory. Those that influence working memory may serve to either enhance, support, or diminish working memory functioning and performance, depending on their strength. Thus, the ones with strong influences and reciprocal relations should be assessed and compared with working memory whenever working memory is evaluated. For instance, a weak influential cognitive process, such as processing speed, may impair otherwise normal working memory capability. The cognitive processes that influence working memory performance also directly impact academic learning and performance.
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
225
226
M I LT O N J . D E H N
11.5.1 Oral Language Several complex cognitive and linguistic processes are involved in oral language development and production, including phonological processing, word retrieval, morphology, and syntax. These linguistic processes and the development of first and second languages depend heavily on adequate verbal working memory processing and storage (Moser et al., 2007). For example, vocabulary development can be slowed by verbal storage limitations and oral expression can be limited by verbal processing.
11.5.2 Auditory Processing Auditory processing is the ability to perceive, analyze, synthesize, and discriminate speech and other auditory stimuli. Auditory processing subsumes such narrow abilities as phonetic coding, speech sound discrimination, resistance to auditory stimulus distortion, memory for sound patterns, maintaining and judging rhythm, musical discrimination and judgment, absolute pitch, and sound localization (Schneider & McGrew, 2012). Studies have reported moderate correlations between general auditory processing and working memory (Hitch et al., 2001). If some aspects of auditory processing are deficient, verbal storage and processing may be impacted.
11.5.3 Phonological Processing Phonological processing is an element of oral language and auditory processing that involves manipulation of the phonemes that make up words. The working memory component that is most closely related with phonological processing is verbal storage. However, verbal processing comes into play when phonemes are consciously manipulated, such as replacing one sound with another. Studies (reviewed by Wagner, 1996) have found measures of verbal storage, which according to Baddeley (2006) is primarily phonological in nature, to be highly correlated with phonological processing. The consensus among researchers is that the two processes have a reciprocal effect on each other (Cohen-Mimran & Sapir, 2007). Adequate short-term storage is necessary for proficient phonological processing; conversely, phonological processing ability affects short-term span (Hulme & Mackenzie, 1992).
11.5.4 Processing Speed Processing speed refers to how quickly the brain processes information and how efficiently simple cognitive tasks are executed over a sustained period of time. Broad processing speed can be divided into “simple” processing speed, which reflects the mental speed required to perform undemanding attentional tasks, and “complex” processing speed, which reflects the total
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
time to complete more demanding tasks, such as a task that involves decision-making. Simple processing speed is typically tested with tasks requiring the examinee to perform relatively easy, overlearned procedures that require little reasoning or higher-level complex processing. Cognitive processing speed is highly interrelated with all working memory components, and processing speed appears to have a strong influence on the growth of working memory capacity. Fry and Hale (1996) reported that 71 percent of age-related improvements in working memory capacity are related to developmental advances in processing speed. Processing speed has such a strong influence because working memory functioning is temporal. Short-term storage retains information for only a few seconds, at best. Faster processing speed allows better completion of cognitive tasks that are dependent on the rapidly decaying information in short-term storage. Slow processing speed allows loss of information from normal working memory storage before the cognitive task can be completed (Compton et al., 2012). Neurologically, both processing speed and working memory rely on extensive neurological connectivity (Klingberg, 2010; Linden, 2007) and good myelination (Briscoe et al., 2001).
11.5.5 Visual-Spatial Processing Visual-spatial processing is the ability to perceive, analyze, synthesize, manipulate, and transform visual patterns and images, including those generated internally. Visual-spatial processing is closely related with the visual-spatial dimensions of working memory (Hitch et al., 2001). Visualspatial working memory is involved in constantly refreshing the perceptual image of the visual field. This refreshing helps keep individuals oriented in space and aware of the current location of moving objects. Visual-spatial working memory is also involved when individuals consciously manipulate objects or mental images.
11.5.6 Executive Functions Executive functions include frontal lobe processes responsible for cuing, directing, and coordinating multiple aspects of perception, cognition, emotion, and behavior during purposeful, goal-directed, problem-solving behavior. Multiple executive functions, which are analogous to a board of directors, monitor and manage cognitive functions. Working memory is usually considered one of the executive functions (Miyake et al., 2000). However, the brief storage of information, which usually takes place outside the frontal lobes (Baldo & Dronkers, 2006), has weaker relations with general executive functions than the processing dimensions of working memory. It is important to distinguish between general executive functions and working memory executive functions. General executive functioning should not be viewed as the equivalent of executive working memory
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
227
228
M I LT O N J . D E H N
(St. Clair-Thompson, 2011). Certainly, working memory can vary independently of other executive functions (Bayliss et al., 2003). For example, working memory may have little to do with such executive functions as initiating an activity or the regulation of emotions. Also, verbal and visualspatial working memory may function adequately without general executive involvement. From this perspective, executive working memory might comprise a subset of executive functions on which working memory functioning is highly dependent. The three executive functions that seem to have the most influence on working memory are inhibition, shifting, and updating.
11.5.7 Inhibition Inhibition is an important executive function that includes the restraint of one’s impulses and one’s automatic, learned, and typical responses. It also includes the suppression of distractions and one’s own intruding thoughts. Good inhibitory control allows goal-directed persistence and the focusing of attention on the task at hand. Inhibition also performs a crucial function for working memory by suppressing and deactivating information that is not relevant to the immediate goal (Cowan, 2005). When inhibition fails, irrelevant information interferes with the retention and processing of information that should be the focus of attention, negatively impacting working memory and other cognitive performance. Children’s inhibition can be assessed with Stroop tasks and scales such as the NEPSY-2 (Korkman et al., 2007).
11.5.8 Attention Working memory functioning requires adequate attentional capacity and control. There are several different types of attention: focusing attention, sustaining attention, selective attention, and divided attention. Working memory and attention are highly interrelated (Kasper et al., 2012; Soto et al., 2005). Thus, working memory functions best when individuals have adequate abilities in all these different aspects of attention. These strong relations can make it difficult to differentiate between working memory and attention. For example, children with working memory deficits are often identified by their teachers as having short attention spans (Cornish et al., 2006). The control, focusing, maintenance, and dividing of attention are core functions of executive working memory (Soto et al., 2005). Focused attention is necessary during almost every aspect of working memory functioning (Engle et al., 1999). Executive working memory is also responsible for resisting interference and for shifting attention from one task to another, without losing relevant information. Individuals who score high on tests of executive working memory are better at focusing attention and inhibiting distracting information than
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
are low-span individuals (Conway et al., 2001). It has been proposed that the inability to inhibit interference may be the underlying trait that accounts for both weak attention and poor working memory performance (Cornish et al., 2006; Schecklmann et al., 2014). Concomitant deficits in attention and working memory are very common. Kasper et al. (2012) reported that 81 percent of children with ADHD have deficits in executive working memory. The assessment implication of the integral relation attention has with working memory is that youth with attention problems and ADHD diagnoses should have their working memory tested in depth.
11.6
Specific Academic Learning Difficulties
Documenting the specific academic learning difficulties displayed by struggling students should be part of a working memory evaluation. Gathercole’s research in the United Kingdom (e.g., Gathercole et al., 2006) discovered several learning behaviors highly associated with working memory deficits: abandoning activities before completion, not following instructions accurately, being reserved in group activities, poor monitoring of work quality, depending on teacher or peers for support, and having a short attention span. When the examiner is unable to observe the student in a learning environment, data may be collected through interviews and rating scales. Although any specific behavior may have multiple causes, when several academic learning behaviors known to be related to deficient working memory functioning are evident, the likelihood of a working memory impairment increases. This sort of data can be used to corroborate the results of performance-based testing. Knowing the empirical relations between working memory and specific academic skills may lead to a better understanding of the learner’s difficulties and a more focused academic intervention. At the very least, awareness of specific academic learning difficulties can lead to hypotheses about specific working memory weaknesses.
11.6.1 Basic Reading Skills Basic reading skills, also referred to as reading decoding skills, are primarily dependent on phonological processing – the ability to detect and manipulate the phonemes used to construct words (Gillon, 2004). Reading decoding involves more than simple storage of phonological sequences in short-term storage. Blending of phonemes into a word requires sequencing, thereby requiring processing in verbal working memory. There is strong evidence that updating of phonological information is essential for reading decoding (de Jong, 2006). Updating is an important executive process. Also, with beginning readers, visual-spatial short-term storage must maintain the grapheme long enough for the phonetic match to be found. In conclusion,
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
229
230
M I LT O N J . D E H N
all five working memory processes are involved in reading decoding (Swanson et al., 2009). Reading decoding behaviors that can be associated with working memory weaknesses include: • Difficulty learning letter sounds • Forgetting decoded phonemes in a word before they are blended • Losing his or her place when reading • Repeating words that were decoded correctly • Omitting words without realizing it • Replacing words with similar words • Forgetting words that were correctly pronounced seconds earlier • Inconsistent decoding
11.6.2 Reading Comprehension Although reading fluency, fluid reasoning, and prior knowledge have strong influences on reading comprehension, differences in working memory capacity also contribute to reading comprehension difficulties (De Beni & Palladino, 2000). To comprehend text a reader must store recently decoded words while complex processes construct meaning. Reading comprehension includes several skills and abilities that involve working memory: accessing word meanings from long-term memory, assembling word meanings into larger information units, integrating information across phrases and sentences, focusing attention on main ideas, creating visual images, forming new knowledge representations, drawing plausible inferences, monitoring the understanding of text as reading progresses, and associating and integrating information with prior knowledge stored in long-term memory. Most of these comprehension components make exceptionally high demands on both storage and processing. To add to the challenge, the executive functions of switching, updating, and inhibition are continually required. Switching occurs when the reader switches back and forth between pieces of information. Updating replaces old, no longer relevant content with newer, more relevant content. Inhibition prevents the intrusion of irrelevant information. The role each working memory process plays in reading comprehension differs from its role in reading decoding. There is a much stronger relation between verbal processing and reading comprehension than between verbal storage and reading comprehension (Engle et al., 1991). Verbal storage facilitates reading comprehension by holding words and sentences in consciousness until there is enough information to complete an idea. Executive working memory is another primary determinant of successful comprehension. The executive must coordinate many diverse processes, especially those that culminate in the integration of new information with an existing mental model. Reading comprehension behaviors that might be attributed to working memory weaknesses include:
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
• • • • • • • •
Frequently rereads sentences and paragraphs Cannot identify main ideas Struggles with inferential comprehension Poor comprehension in spite of adequate decoding skills Poor comprehension in spite of good prior knowledge Poor comprehension in spite of normal reading speed Answers to comprehension questions are entirely wrong Poor recall of what was just read
11.6.3 Mathematics To varying degrees, each working memory process is involved in each type of mathematics operation. Even simple mathematics calculation involves all the processes to varying degrees. However, the working memory components that best predict mathematics calculation are visual-spatial processing (Swanson, 2006) and executive control (Frisco-van den Bos et al., 2013; Gathercole & Pickering, 2000). Compared with calculation, mathematics problem-solving and reasoning place a greater load on the executive component (Swanson, 2006), When solving word problems, individuals must mentally construct an adequate problem representation – a process that depends heavily on executive working memory. Completing a story problem also requires executive working memory involvement in: (a) keeping track of incoming information; (b) integrating information; (c) retrieving mathematics facts and procedures from long-term storage; (d) matching the correct algorithm to the problem at hand; (e) updating the contents of working memory; (f ) making mental arithmetic calculations; (g) monitoring the computational process; and (h) evaluating the solution. The following behaviors are indicative of working memory weaknesses. • • • • • • • • •
Struggles more with story problems than calculation problems Does not know which story problem facts are needed to solve problem Struggles with mental arithmetic Has difficulty learning math facts Does notice the operation signs when doing calculations Confuses columns when doing calculations When young, loses track when counting objects Has poor estimating ability Does not know where to start in multistep solutions
11.6.4 Written Expression Written expression is a complex cognitive activity that requires the integration of several cognitive and memory processes. Once basic writing skills are automatized, executive and verbal working memory are the primary components involved in written expression. Verbal storage contributes by
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
231
232
M I LT O N J . D E H N
briefly storing the words or sentence under construction. The visual-spatial components are also involved, especially during planning, when the writer may be using imagery. However, executive working memory seems to be the most essential ability for translating ideas into writing. Executive working memory has been found to significantly predict planning, writing, and revision, as well as the majority of skills used in writing, such as punctuation, grammar, and vocabulary (Vanderberg & Swanson, 2007). The following writing behaviors may be associated with working memory weaknesses: • • • • • •
Expresses ideas orally but cannot get them on paper Struggles even though basic writing skills are adequate Writes short sentences for age Omits and repeats words Does not detect errors when editing Written sentences do not communicate what writer intends
11.6.5 Academic Performance Poor classroom productivity and academic performance, such as difficulties completing assignments and getting good grades, may be due, at least in part, to a working memory deficit, especially a deficit in executive working memory. This particularly applies to the underachiever who has adequate motivation and adequate academic skills in reading, math, and writing but gets poor grades. The following performance behaviors may be related to working memory deficits: • • • • • •
Has difficulty taking notes in class Slow to complete assignments and exams Does not complete assignments Forgets or misunderstands directions for assignments Frequently asks teachers or peers for assistance Is easily distracted when working on assignments
11.7
Home Environment and Daily Life Functioning Difficulties
The extent to which working memory weaknesses are apparent in the home environment and daily life depends on the working memory demands typically placed on the child. If the demands are minimal, weaknesses will not be very apparent. Examples of behaviors indicative of weaknesses include: • •
Not remembering whether an action was just completed Weak listening comprehension
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
• • • • • •
Forgetting some steps in multistep directions Forgetting where items were placed Slow to complete chores and activities Appearing to have a short attention span Poor planning Forgetting what he or she was doing
11.8
Informal Assessment Strategies
The consideration of informal and qualitative assessment data is good practice in any type of psychological assessment. Such data can be used to corroborate test scores and support assessment hypotheses.
11.8.1 Collecting Background Information Reviewing the client’s history should be the first step. Going all the way back to the individual’s birth is important because it increases the odds of uncovering a potential cause of working memory problems other than genetics. Interviewing, especially a parent interview, is a primary method of collecting this data. There should also be a review of available records – educational, medical, neurological, and psychological. A main objective in collecting background information is to confirm the presence or absence of risk factors. A short list of working memory deficit risk factors includes: • • • • • • • •
Traumatic brain injury Concussion Extreme prematurity Diabetes Dopamine deficiency Depression Anxiety Stress
However, any condition that is detrimental to the brain and cognitive functions is a general risk factor for working memory impairments. When a risk factor is identified, it is important to find out the client’s age at onset or when the incident occurred. Determining the number of episodes is also helpful information. If no risk factors are present, the odds are that the working memory deficiency is developmental. When the deficits are genetic and developmental, the first clues should emerge in early childhood. For example, a delay in speech and language development is a marker for a working memory delay. ADHD-like symptoms or delays in executive functions are other clues. Any indications of memory problems, even if they are attributed to long-term memory, may originate with
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
233
234
M I LT O N J . D E H N
working memory. The absence of any history related to working memory problems does not rule out the possibility of significant weaknesses. Sometimes, working memory deficits and the problems associated with them will not appear until the individual reaches an age (grade) where higher demands are placed on working memory functioning.
11.8.2 Interviews Discussing working memory during an interview can be challenging because most interviewees do not know what working memory is or how it is manifested. For example, teachers and parents often attribute behaviors resulting from working memory weaknesses to attention problems or lack of motivation. To add to the challenge, structured interview formats seldom include items specifically related to working memory. Consequently, the interviewer must make a special effort to generate questions that address working memory concerns. Examples of interview items are suggested in the sections that follow. 11.8.2.1 Teacher Interviews At the beginning of the interview, it may be best to ask about behaviors, learning problems, and cognitive processes that are related with working memory deficits, without explicitly mentioning working memory. For example, the interviewer might ask if the student appears to be having attention problems, processing speed problems, or academic learning problems. If the teacher’s responses give the impression of a working memory deficiency, the next round of questions should attempt to identify which working memory processes are implicated the most. Such interview items might pursue: (a) whether the observed problems are primarily verbal or visual-spatial; (b) how increasing demands in a learning situation impact the student’s performance; and (c) whether the student is using any memory strategies. Examples of interview items include: • • • • • •
Does it take the student exceptionally long to complete assignments? How well does the student remember directions and information? Does the student have any difficulties with listening comprehension? How well does the student stay focused on the task at hand? How well can the student do two things simultaneously, such as listen and take notes? How well does the student retain information during multistep procedures?
11.8.2.2 Parent Interviews The behaviors that should be addressed during a parent interview are similar to those in the teacher interview. The main difference is that the parent interview should contain home environment examples of working
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
memory problems. For example, the interviewer might ask how often the child forgets what was just said in a conversation. To reduce confusion, it might be best to refer to working memory as “short-term memory.”
11.8.2.3 Child Interviews Child interviews can be a valuable source of assessment data. It is important to elicit the child’s attributions regarding the cause of learning and performance problems. Some children lack self-awareness of their problems, and some simply deny they are having any problems. To avoid biasing their attributions, it is best to ask the individual about his or her attributions before there is any mention of memory problems. Then, children who are middle-school age or older should be directly questioned about behaviors related to working memory, and simple, age-appropriate items should also be attempted with elementary students. Some examples of child interview items include: • • • • •
How often do you forget what your teacher or parent just said? Do you ever forget what you were going to say? Is it hard for you to listen and take notes at the same time? Is it hard for you to do arithmetic in your head? What do you think might be causing your learning problems?
11.8.3 Observations The observer should be familiar with behaviors that are indicative of working memory limitations. Classroom and daily life difficulties that are manifestations of working memory weaknesses are detailed in previous sections in this chapter. These should be observed for or included in interviews when personal observations are not possible. Cautious interpretation of observed behaviors is recommended because any given behavior may be due to multiple causes. Observations during one-on-one standardized testing can provide valuable clinical information about the examinee’s working memory strengths and weaknesses. Many of the behaviors suggested for classroom observation can also be observed during testing. During testing, indications of working memory weaknesses should increase as the cognitive complexity of the task increases. For example, changing from digits forward to digits backward may produce more observable indicators of working memory weaknesses. Working memory weaknesses should also become more evident as the testing tasks place greater demands on executive functions such as planning. Examples of testing behaviors indicative of working memory deficits include: • • •
Asking for directions or items to be repeated Taking a long time to respond, especially on more difficult items Requesting supplemental materials, such as paper to write on
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
235
236
M I LT O N J . D E H N
• • • •
Increasing frustration as the complexity of the task increases Difficulty elaborating upon a response when requested to do so Difficulty staying focused on the task at hand Lacking confidence with or fearing immediate memory activities
11.9
Analyzing Multibattery Test Results
For most comprehensive assessments of working memory, examiners will use two or more scales in a multibattery, selective testing fashion. When an entire scale, such as the WISC-V, has been administered, the results for that scale should first be analyzed separately. This within-scale analysis will reveal how working memory compares with other cognitive abilities measured by the scale. However, it will not provide information about all of the examinee’s specific working memory processes. Consequently, practitioners should select the working memory and related scores from the primary scale and combine them with working memory and related scores obtained from other scales. This allows scores from various scales to be analyzed together. The purpose of including all of the relevant scores in one analysis is to identify the examinee’s strengths and weaknesses among the specific working memory processes, as well as strengths and weaknesses when working memory processes are compared with related cognitive abilities, long-term memory processes, and overall cognitive or intellectual ability. A structure for conducting a multibattery analysis is the Working Memory Analysis Worksheet (see a completed example in Table 11.1). The steps for completing the analysis and using the worksheet are detailed below.
Step 1. Decide Which Type of Analysis to Conduct If scores for one or more working memory processes have not been obtained, it is still acceptable to proceed with the analysis. There are three types of analyses that might be completed. For any given case, one, two, or all three types may be conducted. The first is a within working memory analysis, using only scores representing working memory processes. The second type is a within-memory analysis that includes long-term memory test scores along with working memory scores. The third type includes working memory scores along with related cognitive and long-term memory processes. The third option is the most efficient and provides the most comprehensive information. The third option has been used in Table 11.1.
Step 2. Enter the Scales, Composites, and Subtests Subtests from more than one scale may be used to measure a memory or cognitive process. In the completed example in Table 11.1, subtests from
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Table 11.1 Multibattery analysis completed worksheet example Composite and Subtest Scores
Process Score
Predictor
Intra individual S or W
Process
Scale
Composites and Subtests
Executive WM
WJ IV
81
81
97
16
W
W
WM Verbal Processing
WRAML2
97
14
W
W
WRAML2 WISC-V WJ IV WRAML2
95 92
97 97
2 5
– –
– –
93
97
4
–
–
CAS2 WJ IV
(7) 85 (6) 80 (9) 95 (6) 80 103 (8) 90 (9) 95 84 87
83
WM Vis-Spatial Proc. WM Verbal Storage
SHORT-TERM WORKING MEM. Sentence Mem. Story Memory Picture Mem. Digits Forward Mem for Words Design Mem. Finger Window ATTENTION AUDITORY PROCESSING
84 87
97 97
13 10
W W
W –
WISC-V WISC-V WRAML2 WJ IV WJ IV
FLUID REASONING PROCESSING SPEED Verbal Learn. Vis.-Aud. Learn Retrieval Fluency
111 89 8 (90) 86 98
111 89 88
97 97 97
+14 8 9
S W W
S – –
98
97
+1
–
–
WM Vis.-Spatial Stor. Attention Auditory Processing Fluid Reasoning Processing Speed LTM Encoding LTM Retrieval Fluency
Discrepancy
Normative S or W
238
M I LT O N J . D E H N
both the WISC-V and the WJ IV were used to measure verbal storage. Thus, both battery names are entered in the Scale column. Next, the names of the composites and subtests are entered in the third column. Composite scores are preferred whenever possible, as they are more reliable than subtest scores. When composite scores are used, the name of the composite should be listed in uppercase font, such as the WJ IV’s SHORT-TERM WORKING MEMORY that is in the Executive row. If a composite score is used, the names and scores of the individual subtests making up that composite should not be entered separately, unless subtest scores from another scale are also being used to measure the same process.
Step 3. Calculate and Enter Obtained Standard Scores Using each scale’s manual or scoring software, determine the norms-based, derived scores and enter them in the Composite and Subtest Scores column. Any derived score that is not a standard score with a mean of 100 and standard deviation of 15 will need to be transformed to that metric. For example, scaled scores and T-scores require this transformation. The scores from which the standard scores are derived are displayed within parentheses. In the Verbal Storage row of the completed example, Digits Forward has a scaled score of 6, which was transformed into a standard score of 80. Memory for Words comes as a standard score of 103, so no change is necessary.
Step 4. Calculate and Enter the Process Scores The score entered as the Process Score represents the functioning or ability level of that specific process. When only one obtained composite or subtest score is available, that score can be used directly. In the completed example, the only obtained score for visual-spatial processing is a standard score of 95. Thus, 95 becomes the score representing this process. When there are two or more subtest or composite scores available, the mean of the scores is used as the process score. This mean is referred to as a clinical composite. For verbal processing in the completed example, the scores of 85 and 80 average out to 82.5 and are rounded to 83. This averaged clinical process score is then used as the best estimate of verbal working memory.
Step 5. Determine Whether Each Process Score Is Unitary When the difference between the lowest and highest subtest scores used to calculate a process score is greater than 22 standard score points (1.5 standard deviations), the scores are not “unitary” (lack consistency). Scores that are nonunitary can be used in the analysis but should be interpreted cautiously. In the verbal storage row in Table 11.1 Digits Forward has a score of 80 while Memory for Words is 103. Thus, the process
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
score representing verbal short-term storage would be considered nonunitary. Ideally, examiners should be able to explain why the two scores involved are nonunitary. If not, further testing of the process involved is recommended.
Step 6. Enter the IQ or Mean as the Predictor The score entered in the Predictor column represents overall working memory, overall memory, or overall cognitive ability. In most cases, it is best to use an IQ or a similar cognitive composite, representing overall cognitive ability, as the predictor. IQ and cognitive composites are appropriate predictors because of the high correlations IQ has with different types of memory and cognitive processes. In the completed example, an IQ score of 97 is being used. The alternative is to use the mean of the scores in the Process Score column as the Predictor. Using the mean of the processes scores as the Predictor may result in fewer weaknesses being identified. This occurs when the majority of the processes tested have low scores. In such instance, it is better to use an IQ as the Predictor. However, at times, more of a “within memory” analysis is desired, in which case the mean of the scores involved can be used as the Predictor. In such cases, the mean is determined by summing the appropriate scores in the Process Score column and then dividing by the number of processes included in the analysis.
Step 7. Compute and Enter the Discrepancy The Discrepancy is computed by subtracting the predicted score (IQ or Mean) from the Process Score. The value is then entered along with a plus or minus symbol in front of it. In the completed example, the verbal processing score of 83 is 14 points lower than the predicted score of 97, resulting in an entry of 14.
Step 8. Determine Normative Strengths and Weaknesses In the Normative S (Strength) or W(Weakness) column, an “S” is entered when Process Score entries are above average, a “W” is entered for below average scores, and a dash (–) is entered when neither an S or W apply. “Normative” refers to a standard, normal distribution. Typically, the average range is considered to be 85–114. However, a narrower average range of 90–109 can be used, as it is in the Table 11.1 example. The worksheet scores used to identify normative S’s and W’s are the scores in the Process Score column. For example, in Table 11.1 a “W” is entered for the verbal processing score of 84 because it is below the average range, and an “S” is entered for the Fluid Reasoning score of 111 because it is above the average range.
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
239
240
M I LT O N J . D E H N
Step 9. Determine Intraindividual Strengths and Weaknesses From a pattern of strengths and weaknesses (PSW; Dehn, 2014, 2020) perspective, identification of intraindividual (within child) strengths and weaknesses is the primary benefit of using this analysis worksheet. In the Intraindividual S or W column, an “S” is entered for intraindividual strengths, a “W” for intraindividual weaknesses, and a dash (–) when there is neither. Given scores that have high reliability coefficients (.80 or higher), a critical value of 15 is very likely to be a significant difference at approximately the .05 probability level. However, a lower critical value of 12 points is recommended because requiring a 15-point discrepancy is likely to miss genuine, within-person weaknesses that are impacting functioning. Thus, a critical value of 12 points seems more appropriate, but will result in more false positives. The value in the Discrepancy column is used to make this determination. If the discrepancy is –12 or more points, then the process is identified as a weakness. If the Discrepancy is greater than +12 points, then it is labeled a strength. When a 12-point discrepancy is used as the critical value to identify intraindividual strengths and weaknesses, there should be corroborating evidence from other assessment data to support the identified weaknesses.
Step 10. Basic Interpretation Once the worksheet has been completed, the S’s and W’s can be used to interpret the examinee’s profile. For example, in Table 11.1 the examinee has normative weaknesses (below average) processing abilities in executive and verbal working memory processes, as well as in attention, auditory processing, processing speed, and long-term memory encoding. Therefore, it appears that related cognitive and memory processes may be contributing to the working memory weaknesses. In spite of some weak working memory weaknesses, the examinee displays a normative strength in fluid reasoning. The intraindividual (within child) weaknesses are fewer: executive and verbal working memory along with attention. The pairing of weak executive working memory with weak attention is a fairly typical profile. The processes that are both kinds of weaknesses (double weaknesses) are the greatest cause for concern.
11.10
Formal Statistical Analysis of Multibattery Test Results
There are two commercial software products available that can be used to identify strengths and weaknesses within working memory, within working and long-term memory, and within memory and related cognitive processes. Using multibattery test scores, both programs conduct a formal statistical analysis with unique critical values for each set of process scores
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
that determine the examinee’s strengths and weaknesses at .01 and .05 level of significance. This allows the examiner to more confidently and accurately identify the examinee’s profile of strengths and weaknesses. Each program can also compare memory and/or cognitive scores with related achievement test scores for consistency. For example, verbal working memory is compared with the reading comprehension. If both have low scores in the same range, there is evidence for the hypothesis that poor reading comprehension is related to weak verbal working memory. Drop-down menus for each process allow entering composites and subtest scores that measure each process. After all the test scores are entered, completing the calculations is accomplished with the click of button. An in-depth report with several tables and charts and a narrative with recommendations make it easy to interpret the examinee’s profile. The Memory Processes Analyzer (MPA; Dehn, 2018) identifies statistically significant strengths and weaknesses among 13 memory processes. Five of them are working memory processes and eight are long-term memory processes. Working memory test scores can be analyzed with or without long-term memory scores. The MPA program allows examiners to combine scores from different scales and to test discrepancies for statistical significance. The Psychological Processing Analyzer (PPA; Dehn 2020) identifies statistically significant strengths and weaknesses among 14 cognitive and memory processes and among 8 areas of achievement. The PPA divides both working and long-term memory only into verbal and visual-spatial components. Consequently, it does not allow the analysis of the five specific working memory processes. However, it does allow working memory to be compared with related cognitive processes; for example, verbal short-term storage will be compared with phonological processing for statistical consistency.
11.11
Using Assessment Results to Guide Treatment
The examinee’s profile of specific working memory strengths and weaknesses can be used to guide treatment and intervention decisions, and lead to more appropriate recommendations for academic instruction and accommodations. One standard treatment for all working memory deficits does not address the specific needs of many examinees. For example, executive working memory deficiencies might be addressed with promoting more self-awareness and more strategies that support working memory performance, as well as attempts to improve inhibition and better manage distractions. In contrast, a child with adequate executive development but poor verbal storage capacity would likely benefit from being taught rehearsal, chunking, organization, and visualization strategies. In general, evidence-based interventions for working memory deficits are lacking. Computerized working memory training, which recently has received the most attention, has produced inconsistent and controversial
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
241
242
M I LT O N J . D E H N
results (Melby-Lervåg & Hulme, 2013). Even if it were consistently effective, it may do little to improve the verbal dimension of working memory because the computerized training tasks are primarily visual-spatial. When verbal weaknesses are involved, teaching and applying strategies is probably the most effective. Dehn (2011, 2015) provides details on several strategies, such as teaching rehearsal, that do have a historical evidence-base. The child’s working memory profile should also be used to generate appropriate recommendations for classroom instruction and accommodations. Simply informing teachers and parents of a child’s working memory profile may change how they teach and interact with the child. Accommodations can be matched to the child’s specific weaknesses, such as providing shorter directions for those with verbal storage limitations. Regarding instruction and academic performance recommendations, Dehn (2015) draws from the literature on cognitive load (greater cognitive load increases demands on working memory) to make specific recommendations for classroom environments, instruction, and curriculum.
11.12
Conclusion
Dehn’s assessment model recommends a comprehensive assessment of working memory that measures the processing and storage components separately and differentiates between verbal, visual-spatial, and executive working memory. To tap each of these five working memory processes usually requires a selective, multibattery testing approach. To assist with selection of appropriate subtests, the chapter defines and gives assessment examples for each of the five working memory processes. Highly related cognitive processes, such as processing speed and phonological processing, should also be included in a working memory assessment in order to weigh reciprocal influences and account for cognitive and achievement performance weaknesses. Details are then provided on how specific working memory weaknesses are manifested in specific areas of academic achievement, as well as in academic performance and in the home environment. After recommending what to focus on during informal assessment procedures, step-by-step procedures for analyzing multibattery test results are provided with the goal of identifying the child’s profile of specific working memory strengths and weaknesses.
References Alloway, T. P. (2007). Automated working memory assessment. Harcourt Assessment. Alloway, T. P., Gathercole, S. E., & Pickering, S. J. (2006). Verbal and visuospatial short-term and working memory in children: Are they separable? Child Development, 77, 1698–1716.
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
Baddeley, A. D. (1986). Working memory. Oxford University Press. Baddeley, A. D. (2000). The episodic buffer: A new component in working memory? Trends in Cognitive Science, 4, 417–423. Baddeley, A. D. (2006). Working memory: An overview. In S. J. Pickering (Ed.), Working memory and education (pp. 1–31). Academic Press. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. A. Bower, (Ed.), Recent advances in learning and motivation (Vol. 8, pp. 47–89). Academic Press. Baldo, J. V., & Dronkers, N. F. (2006). The role of the inferior parietal and inferior frontal cortex in working memory. Neuropsychology, 20, 529–538. Bayliss, D. M., Jarrold, C., Baddeley, A. D., & Gunn, D. M. (2003). The complexities of complex span: Explaining individual differences in working memory in children and adults. Journal of Experimental Psychology: General, 132, 71–92. Briscoe, J., Gathercole, S. E., & Marlow, N. (2001). Everyday memory and cognitive ability in children born very prematurely. Journal of Child Psychology and Psychiatry, 42, 749–754. Bunge, S. A., & Wright, S. B. (2007). Neurodevelopmental changes in working memory and cognitive control. Current Opinion in Neurobiology, 17, 243–250. Cohen-Mimran, R., & Sapir, S. (2007). Deficits in working memory in young adults with reading disabilities. Journal of Communication Disorders, 40, 168–183. Compton, D. L., Fuchs, L. S., Fuchs, D., Lambert, W., & Hamlett, C. L. (2012). The cognitive and academic profiles of reading and mathematics learning disabilities. Journal of Learning Disabilities, 45, 79–95. Conway, A. R. A., Cowan, N., & Bunting, M. F. (2001). The cocktail party revisited: The importance of working memory capacity. Psychonomic Bulletin & Review, 8, 331–335. Cornish, K., Wilding, J., & Grant, C. (2006). Deconstructing working memory in developmental disorders of attention. In S. J. Pickering (Ed.), Working memory and education (pp. 157–188). Academic Press. Cowan, N. (1995). Attention and memory: An integrated framework. Oxford Psychology Series, 26.: Oxford University Press. Cowan, N. (2005). Working memory capacity. Erlbaum. De Beni, R., & Palladino, P. (2000). Intrusion errors in working memory tasks: Are they related to reading comprehension ability. Learning and Individual Differences, 12, 131–145. Dehn, M. J. (2008). Working memory and academic learning: Assessment and intervention. Wiley. Dehn, M. J. (2011). Helping students remember: Exercises and strategies to strengthen memory. Wiley. Dehn, M. J. (2014). Essentials of processing assessment (2nd Ed). Wiley. Dehn, M. J. (2015). Essentials of working memory assessment and intervention.: Wiley.
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
243
244
M I LT O N J . D E H N
Dehn, M. J. (2017). How working memory enables fluid reasoning. Applied Neuropsychology Child, 6(3), 245–247. doi: 10.1080/21622965.2017.1317490 Dehn, M. J. (2018). Memory processes analyzer 3.0.: Schoolhouse Educational Services. Dehn, M. J. (2020). Psychological processing analyzer 8.0. Schoolhouse Educational Services. de Jong, P. (2006). Understanding normal and impaired reading development: A working memory perspective. In S. Pickering (Ed.), Working memory and education (pp. 33–60). Academic Press. Engle, R. W. (2002). Working memory capacity as executive attention. Current Directions in Psychological Science, 11(1), 19–23. Engle, R. W., Carullo, J. J., & Collins, K. W. (1991). Individual differences in working memory for comprehension and following directions. Journal of Educational Research, 84, 253–262. Engle, R. W., Kane, M. J., & Tuholski, S.W. (1999). Individual differences in working memory capacity and what they tell us about controlled attention, general fluid intelligence and functions of the prefrontal cortex. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 102–134).Cambridge University Press. Frisco-van den Bos, I., van der Ven, S. H. G., Kroesbergen, E. H., & van Luit, J. E. H. (2013). Working memory and mathematics in primary school children: A meta-analysis. Educational Research Review, 10, 29–44. Fry, A. F., & Hale, S. (1996). Processing speed, working memory, and fluid intelligence: Evidence for a developmental cascade. Psychological Science, 7, 237–241. Gathercole, S. E., & Alloway, T. P. (2008). Working memory and learning: A practical guide for teachers. SAGE. Gathercole, S. E., Lamont, E., & Alloway, T. P. (2006). Working memory in the classroom. In S. J. Pickering (Ed.), Working memory and education (pp. 219–240). Academic Press. Gathercole, S. E., & Pickering, S. J. (2000). Assessment of working memory in six- and seven-year-old children. Journal of Educational Psychology, 92, 377–390. Gillon, G. T. (2004). Phonological awareness. Guilford Press. Hedden, T., & Yoon, C. (2006). Individual differences in executive processing predict susceptibility to interference in verbal working memory. Neuropsychology, 20, 511–528. Hitch, G. J., Towse, J. N., & Hutton, U. (2001). What limits children’s working memory capacity? Theoretical accounts and applications for scholastic development. Journal of Experimental Psychology: General, 130, 183–198. Hulme, C., & Mackenzie, S. (1992). Working memory and severe learning difficulties. Lawrence Erlbaum. Kasper, L. J., Alderson, R. M., & Hudec, K. L. (2012). Moderators of working memory deficits in children with attention-deficit/hyperactivity disorder (ADHD): A meta-analytic review. Clinical Psychology Review, 32, 605–617.
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
Assessing Children’s Working Memory
Keith, T. Z., Fine, J. G., Taub, G. E., Reynolds, M. R., & Kranzler, J. H. (2006). Higher order, multi-sample, confirmatory factor analysis of the Wechsler Intelligence Scale for Children—Fourth Edition: What does it measure? School Psychology Quarterly, 12, 89–107. Klingberg, T. (2010). Training and plasticity of working memory. Trends in Cognitive Sciences, 14, 317–324. Korkman, M., Kirk, U., & Kemp, S. (2007). NEPSY-II: A developmental neuropsychological assessment. Psychological Corporation. Linden, D. E. J. (2007). The working memory networks of the human brain. Neuroscientist, 13, 257–267. Martinussen, R., Hayden, J., Hogg-Johnson, S., & Tannock, R. (2005). A metaanalysis of working memory impairments in children with attentiondeficit/hyperactivity disorder. Journal of the American Academy of Child and Adolescent Psychiatry, 44, 377–384. Martinussen, R., & Tannock, R. (2006). Working memory impairments in children with attention-deficit hyperactivity disorder with and without comorbid language learning disorders. Journal of Clinical and Experimental Neuropsychology, 28, 1073–1094. Melby-Lervåg, M., & Hulme, C. (2013). Is working memory training effective? A meta-analytic review. Developmental Psychology, 49(2), 270–291. doi:10.1037/a0028228 Miyake, A., Friedman, N. P., Emerson, M. J., Witski, A. H., & Howerter, A. (2000). The unity and diversity of executive functions and their contributions to complex “frontal lobe” tasks: A latent variable analysis. Cognitive Psychology, 41, 49–100. Moser, D. D., Fridriksson, J., & Healy, E. W. (2007). Sentence comprehension and general working memory. Clinical Linguistics & Phonetics, 21, 147–156. Palmer, S. (2000). Phonological recoding deficit in working memory of dyslexic teenagers. Journal of Research in Reading, 23, 28–40. Pickering, S. J. (2006). Working memory in dyslexia. In T. P. Alloway & S. E. Gathercole (Eds.), Working memory and neurodevelopmental disorders (pp. 7–40). Psychology Press. Prabhakaran, V., Narayanan, K., Zhao, Z., & Gabrieli, J. D. E. (2000). Integration of diverse information in working memory within the frontal lobe. Nature Neuroscience, 3, 85–90. Roid, G. H. (2003). Stanford-Binet Intelligence Scales. 5th ed.: Riverside Publishing. Sattler, J. M. (2020). Assessment of children: Cognitive foundations and applications 6th ed.). Jerome M. Sattler. Schecklmann, M., Ehlis, A.-C., Plichta, M. M., Dresler, T., Heine, M., Boreatti-Hummer, A., . . . F allgatter, A. J. (2014). Working memory and response inhibition as one integral phenotype of adult ADHD? A behavioral and imaging correlational investigation. Journal of Attention Disorders, 17(6), 470–482.
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
245
246
M I LT O N J . D E H N
Schneider, J. W., & McGrew, K. S. (2012). The Cattell-Horn-Carroll (CHC) model of intelligence. In D. P. Flanagan, & P. L. Harrison (Eds.), Contemporary intellectual assessment (3rd ed., pp. 99–144). Guilford Press. Schrank, F. A., McGrew, K. S., & Mather, N. (2014). Woodcock-Johnson IV tests of cognitive abilities. Riverside. Siegel, L. S., & Ryan, E. B. (1989). The development of working memory in normally achieving and subtypes of learning disabilities. Child Development, 60, 973–980. Smith, E. E., & Jonides, J. (1997). Working memory: A view from neuroimaging. Cognitive Psychology, 33, 5–42. Soto, D., Heinke, D., Humphreys, G. W., & Blanco, M. J. (2005). Early involuntary top-down guidance of attention from working memory. Journal of Experimental Psychology: Human Perception and Performance, 31, 248–261. St. Clair-Thompson, H. (2011). Executive functions and working memory behaviours in children with a poor working memory. Learning and Individual Differences, 21, 409–414. Swanson, H. L. (2006). Cross-sectional and incremental changes in working memory and mathematical problem solving. Journal of Educational Psychology, 98, 265–281. Swanson, H. L. (2011). Working memory, attention, and mathematical problem solving: A longitudinal study of elementary school children. Journal of Educational Psychology, 103, 821–837. Swanson, H. L., & Berninger, V. W. (1995). The role of working memory in skilled and less skilled readers’ comprehension. Intelligence, 21, 83–108. Swanson, H. L., & Berninger, V. W. (1996). Individual differences in children’s working memory and writing skill. Journal of Experimental Child Psychology, 63, 358–385. Swanson, H. L., & Jerman, O. (2007). The influence of working memory on reading growth in subgroups of children with disabilities. Journal of Experimental Child Psychology, 96, 249–283. Swanson, H. L., Zheng, X., & Jerman, O. (2009). Working memory, shortterm memory, and learning disabilities: A selective meta-analysis of the literature. Journal of Learning Disabilities, 42, 260–287. Vanderberg, R., & Swanson, H. L. (2007). Which components of working memory are important in the writing process? Reading and Writing, 20, 721–752. Wagner, R. K. (1996). From simple structure to complex function: Major trends in the development of theories, models, and measurements of memory. In G. R. Lyon & N. A. Krasnegor (Eds.), Attention, memory, and executive function (pp. 139–156). Brookes. Wang, S., & Gathercole, S. E. (2013). Working memory deficits in children with reading difficulties: Memory span and dual task coordination. Journal of Experimental Child Psychology, 115, 188–197.
https://doi.org/10.1017/9781108955638.014 Published online by Cambridge University Press
12 Measuring Individual Differences in Working Memory Capacity and Attention Control and Their Contribution to Language Comprehension Alexander P. Burgoyne, Jessie D. Martin, Cody A. Mashburn, Jason S. Tsukahara, Christopher Draheim, and Randall W. Engle 12.1
Introduction
Imagine you are sitting at a coffee shop talking with a friend. The environment is replete with distractions, from the barista calling out orders to the espresso machine noisily letting off steam. To understand your friend, you must focus on what they are trying to say while (a) preventing your attention from being captured by these distractions and (b) maintaining the gist of what your friend has said in the midst of this sensory and cognitive maelstrom. Of course, this situation is not unique to the coffee shop. Everyday life is filled with distractions and interference, both from the external environment (e.g., receiving a text message) and from internal sources (e.g., thinking about lunch). A ubiquitous challenge, then, is keeping a running gist of the task you are performing while ignoring or suppressing task-relevant and irrelevant distractors. As it turns out, individual differences in these cognitive abilities play an important role in explaining individual differences in language comprehension. In this chapter, we discuss the measurement of working memory capacity and attention control. First, we examine the origins of complex span measures of working memory capacity, which were created to better This work was supported by grants from the Office of Naval Research (N00173–18-S-BA01; N00014–12-1-1011) to Randall W. Engle.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
248
A L E X A N D E R P. B U R G O Y N E E T A L .
understand the cognitive processes underpinning language comprehension. We then discuss the executive attention theory of working memory, which places attention control at the center of individual differences in working memory capacity and fluid intelligence. Next, we describe the relationship between working memory capacity, attention control, and language comprehension, and discuss how maintenance and disengagement – two functions supported by the control of attention – contribute to performance across a range of cognitive tasks. Afterward, we discuss challenges associated with measuring working memory capacity and attention control and identify factors that threaten the construct and criterion validity of these measures. We also detail the steps our laboratory has taken to refine the measurement of these cognitive constructs. We close by providing practical recommendations and resources to researchers who wish to use measures of working memory capacity and attention control in their work.
12.2
The Origins of Complex Span Measures of Working Memory Capacity
Researchers recognized early on that language comprehension requires the short-term storage of information (Kintsch & van Dijk, 1978; Perfetti & Lesgold, 1977). For example, in order to understand the referent of the pronoun “he” in the sentence “Although the doctor was playing golf, he still checked his phone often,” one must recall “the doctor” from the previous clause. The successful integration of information across words and sentences requires that information is not immediately forgotten, but rather is retained at the surface level for a short while and at the gist level for a considerably longer period. These memories must be brought to bear on subsequent comprehension processes to allow what Gernsbacher (1990) called “structure building” at the utterance and sentence level in the short term and at the paragraph and higher level in the longer term. Thus, the interdependence of semantic, syntactic, and contextual information during language processing led theorists to posit that short-term memory played an important role in comprehension. It followed that if short-term memory was indeed critical, individual differences in short-term memory should predict individual differences in language comprehension. And yet, time and time again, researchers found that short-term memory was largely unrelated to reading or listening comprehension (Hunt et al., 1973; Jackson & McClelland, 1979; Perfetti & Goldman, 1976). As recounted by Daneman and Merikle (1996), “simple span” measures of short-term memory rarely predicted reading comprehension, except when the sample consisted of young children or severely impaired readers (Farnham-Diggory & Gregg, 1975; Rizzo, 1939). This finding was so problematic that it led Crowder (1982) to petition psychologists to abandon the notion of short-term memory. It led Daneman and Carpenter (1980) to argue that
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
language comprehension required more than passive short-term storage; it also required the active maintenance and manipulation of information – in other words, an interplay between the storage of prior information and the processing of new information. This dual requirement of storage and processing was not adequately captured by simple span short-term memory tasks, they argued, because such tasks merely presented a series of items (e.g., digits) for participants to recall. To address this limitation, Daneman and Carpenter (1980) created the first complex span measures of working memory capacity, the reading span and listening span tests. In the reading span test (Daneman & Carpenter, 1980), participants read aloud a series of sentences, verified whether they made sense, and then recalled the final word of each sentence. The measure of performance was the number of sentences the participant could read while recalling the final words in the correct order. The listening span test was an auditory facsimile of the reading span test; participants listened to sentences instead of reading them aloud. Both complex span tasks interleaved the presentation of memoranda with a secondary task, requiring a trade-off between information storage (e.g., remembering the final words) and processing (e.g., interpreting the sentences). What’s more, both components of the tasks used verbal stimuli, which increased the likelihood that the processing and storage subtasks interfered with one another (Hale et al., 1996) and tapped the domain-specific demands of language comprehension. Daneman and Carpenter (1980) found that performance on the reading span and listening span tests predicted verbal SAT scores and two other reading comprehension measures, fact retrieval and pronominal reference. By contrast, they found that word span and digit span measures of shortterm memory did not predict reading comprehension. This pattern of results provided early evidence that the “active ingredient” in measures of working memory capacity was not simple short-term storage, but instead, the ability to control attention to successfully coordinate storage and processing subtasks. Sixteen years later, a meta-analysis of 77 studies (Daneman & Merikle, 1996) affirmed that complex span measures predict reading comprehension better than short-term storage measures do, and furthermore, that complex span tasks with verbal stimuli evoke the strongest relationships between working memory capacity and reading comprehension, likely due to the linguistic processing demands shared across predictor and criterion tasks. Nevertheless, even complex span tasks that do not use verbal stimuli for the processing subtask predict individual differences in reading comprehension (Turner & Engle, 1989), suggesting that these tasks tap a domain-general ability that is important for language processing. Daneman and Carpenter’s (1980) complex span tasks had an enormous impact on the field. Their article, which has been cited over 8,000 times according to Google Scholar, served as a guide for psychometricians to develop other complex span tasks, including operation span (Unsworth
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
249
250
A L E X A N D E R P. B U R G O Y N E E T A L .
et al., 2005), symmetry span (Unsworth et al., 2009), and rotation span (Shah & Miyake, 1996), which we discuss in detail later on. In turn, complex span measures of working memory capacity have been shown to predict individual differences in reading, writing, and speaking ability, as well as following directions (Bock & Miller, 1991; Daneman & Green, 1986; Engle et al., 1991; Gathercole & Baddeley, 1993). Finally, Daneman and Carpenter’s (1980) work gave rise to the capacity theory of comprehension (Just & Carpenter, 1992), which holds that working memory is essential for language comprehension because it facilitates the resolving of syntactic ambiguity via the maintenance of multiple interpretations, and supports syntactic modularity via the interaction of syntactic and pragmatic information. For this chapter on the measurement of individual differences, it is worth discussing how complex span tasks fit within the broader theoretical framework of working memory capacity and attention control. Below, we provide a description of what we mean by these terms, followed by a discussion of further evidence for the executive attention view, which places attention control at the center of individual differences in working memory capacity and fluid intelligence.
12.3
The Executive Attention View of Working Memory Capacity
Working memory refers to the cognitive system responsible for the temporary maintenance and manipulation of information in a highly accessible state (Baddeley, 1992). The working memory system comprises a controlled attention component and a short-term storage component, or components. Our view of the working memory system – the executive attention view (Kane & Engle, 2002) – emphasizes the role of attention control, which we define as the domain-general ability to maintain focus on task-relevant information while preventing attentional capture by task-irrelevant thoughts and events. In Baddeley and Hitch’s (1974) classic model of the working memory system (refer back to Figure 2.1), the controlled attention component is called the central executive. It is responsible for coordinating the flow of information between short-term storage components in a goal-driven manner (Baddeley, 1992). The visuospatial sketch pad, on the other hand, is responsible for the storage of visual information, such as mental imagery. Finally, the phonological loop is responsible for the storage of verbal and auditory information, such as speech. What is important for the present purposes is not the exact specifications of Baddeley and Hitch’s (1974) model, or subsequent models (Baddeley, 2002), but rather the idea that
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
working memory involves the interplay between (1) controlled attention and (2) short-term memory. Whereas working memory refers to the cognitive system responsible for the temporary maintenance and manipulation of information, working memory capacity refers to the measurement of an individual’s ability to use the working memory system. Working memory capacity is often measured using complex span tasks: dual tasks that require participants to simultaneously store and process information. Although working memory capacity is often indexed in terms of “number of items recalled,” short-term storage only tells part of the story. This is because people who are better able to flexibly allocate attention to the storage and processing subtasks perform better on working memory tasks than those who are more susceptible to distraction and interference. In fact, evidence suggests that the controlled attention component of working memory plays a large role in explaining its relationships with a range of outcomes and abilities (Kane & Engle, 2002), including language comprehension. Early evidence for the executive attention view was provided by studies that found differences between high and low working memory capacity participants on tasks that demanded controlled attention but placed little burden on short-term memory. For example, in the antisaccade task, participants must resist the urge to look at a flashing cue on one side of the screen, and instead rapidly look toward the opposite side of the screen. The task is challenging because it requires inhibiting a reflexive response: looking at a highly salient visual stimulus. Adding to the difficulty, participants cannot simply ignore the flashing cue, because it indicates the side of the screen they should look away from. In a sample of 203 participants, Kane et al. (2001) found that people with higher working memory capacity made fewer errors on the antisaccade task (i.e., looking at the cue instead of away from it), and were quicker to recover when they looked in the wrong direction. Because the antisaccade task does not burden participants with lists of items to remember, but does require controlled attention, Kane et al.’s (2001) results suggest that individuals’ working memory capacity is closely linked to the functioning of the central executive. As another example, in a dichotic listening task, participants are presented two auditory streams, one to each ear, and must repeat aloud the messages presented to one ear while ignoring the messages presented to the other. At some point, the participant’s name is surreptitiously presented to the unattended ear. Later on, the participant is asked whether they heard anything unusual. Around one-third of people report hearing their name in the unattended ear, an effect which has been termed the “cocktail party” phenomenon (Moray, 1959) Critically, low working memory capacity participants were significantly more likely to report hearing their name than high working memory capacity participants (Conway et al., 2001). This suggests that people with high working memory capacity are better able
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
251
252
A L E X A N D E R P. B U R G O Y N E E T A L .
to control their attention to ignore task-irrelevant distractors, and may explain why some people perform better than others in the coffee shop scenario described in the beginning of this chapter. Additional evidence for the executive attention view is provided by studies that use latent variables to model relationships between cognitive constructs. Latent variables are unobserved variables that capture variance common to a set of indicators (e.g., performance measures on different tasks). Latent variable analyses such as confirmatory factor analysis and structural equation modeling provide a number of advantages relative to other statistical approaches. For instance, latent variable analyses allow researchers to draw conclusions about cognitive constructs – the hypothesized source of the shared variance among a set of measures – as opposed to drawing conclusions about specific measures, which may only capture a slice of the cognitive construct of interest. Furthermore, latent variables are theoretically free of measurement error, which attenuates relationships (Kline, 2015). As we discuss in the “Practical Recommendations” section below, latent variable analyses require large samples (e.g., 250 or more participants) and multiple measures per construct (e.g., 3 or more measures). When the proper conditions are met, however, latent variable analyses can be a powerful tool for elucidating relationships between constructs. Using latent variable analyses, Engle et al. (1999) found that controlled attention drives working memory capacity’s relationship with fluid intelligence (i.e., reasoning ability). First, Engle et al. (1999) established that working memory capacity and short-term storage were dissociable at the latent level. The two constructs shared approximately 46 percent of their reliable variance – a substantial amount, but considerably less than 100 percent. More importantly, they found that working memory capacity contributed independently to fluid intelligence after accounting for shortterm storage. While the predictive path from short-term storage to fluid intelligence was near zero and nonsignificant, the path from working memory capacity to fluid intelligence was substantial and significant (Figure 12.1). This indicates that the controlled attention component of working memory contributes to fluid intelligence above and beyond short-term memory. Twenty-one years later, Draheim et al. (2021) corroborated and extended this finding by analyzing relationships between working memory capacity, attention control, and fluid intelligence. Whereas Engle et al. (1999) estimated the contribution of controlled attention by partialling out variance in working memory capacity attributable to short-term storage, Draheim et al. (2021) measured attention control directly using a battery of newand-improved attention tasks, described in greater detail in the “Measuring Attention Control” section. Using structural equation modeling, Draheim et al. (2021) found that attention control mediated the relationship between working memory capacity and fluid intelligence. That is, the once-significant relationship between working memory
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
Figure 12.1 The substantial and significant contribution of working memory capacity to fluid intelligence after accounting for short-term memory Note: latent factors are depicted as ovals and observed measures are depicted as rectangles. Working memory capacity and short-term memory were strongly correlated (.68). The contribution of working memory capacity to fluid intelligence (.59) was substantial and significant after accounting for short-term memory. By contrast, the contribution of short-term memory to fluid intelligence (.13, ns) was non-significant after accounting for working memory capacity. This pattern of results suggests that the controlled attention component of working memory contributes to fluid intelligence above and beyond short-term memory. Adapted from Engle et al. (1999).
Figure 12.2 A structural equation model depicting attention control fully mediating the relationship between working memory capacity and fluid intelligence. Working memory capacity was strongly related to attention control (.75), and attention control was strongly related to fluid intelligence (.65). Working memory capacity did not predict fluid intelligence (.21, ns) after accounting for attention control. The results suggest that attention control drives the relationship between working memory capacity and fluid intelligence. Adapted from Draheim et al. (2021).
capacity and fluid intelligence was no longer significant after accounting for attention control (Figure 12.2). This finding extends the work of Engle et al. (1999) by showing that individual differences in attention control can fully explain the relationship between working memory capacity and fluid intelligence.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
253
254
A L E X A N D E R P. B U R G O Y N E E T A L .
12.4
Working Memory Capacity, Attention Control, and Language Comprehension
Returning to the domain of language, evidence also suggests that the controlled attention component of working memory contributes to individual differences in comprehension. For example, Swanson and Ashbaker (2000) found that working memory capacity significantly predicted reading comprehension and word recognition performance in children with learning disabilities, even after accounting for short-term memory. Across a series of hierarchical regression analyses, the incremental validity of working memory capacity above and beyond short-term memory ranged from 5 percent to 27 percent. For comparison, the incremental validity of short-term memory above and beyond working memory capacity ranged from 1 percent to 7 percent. Because the unique variance in working memory capacity after accounting for short-term storage represents controlled attention, this pattern of results indicates that attention control drives the relationship between working memory capacity and language comprehension (see also Engel de Abreu et al., 2011). Further evidence for the role of attention control in supporting language comprehension is provided by experiments that burden the central executive with a distractor task and reveal concomitant decreases in language comprehension. For example, Waters et al. (1987) had participants maintain a random sequence of six digits while reading sentences of varying syntactic complexity. They found that burdening the central executive significantly impaired comprehension of syntactically complex sentences. As discussed by Caplan and Waters (1999), concurrent digit load primarily affects the comprehension of syntactically complex sentences when the presentation of one set of stimuli interrupts the presentation of the other. This suggests that attentional shifts induced by a secondary task may interfere with efforts to structure sentences syntactically or interpret their meaning (Caplan & Waters, 1999). Given that the relationship between working memory capacity and comprehension is partly attributable to controlled attention, some language researchers have attempted to measure attention control directly, rather than indirectly by partialling out variance in working memory capacity attributable to short-term storage or manipulating it by burdening the central executive. However, the measurement of individual differences in attention control poses its own challenges due to psychometric limitations, as we discuss in the section “Measuring Attention Control.” Nevertheless, researchers have found that measures of attention control predict individual differences in language abilities. For example, McVay and Kane (2012) found that latent variables representing attention control and reading comprehension ability were strongly correlated in a sample of over 200 participants. To explore the mechanism by which attention control contributed to comprehension, McVay and Kane
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
Figure 12.3 A structural equation model depicting task-unrelated thoughts partially mediating the relationship between attention control and reading comprehension Attention control was negatively related to the number of task-unrelated thoughts (.37), and task-unrelated thoughts were negatively related to reading comprehension (.46). The path from attention control to reading comprehension (.21) remained significant after accounting for task-unrelated thoughts, indicating that attention control contributed to reading comprehension even after accounting for mind wandering. Adapted from McVay and Kane (2012)
(2012) had participants report instances of “task-unrelated thoughts,” or mind wandering, during task performance. Attention control was negatively correlated with the frequency of task-unrelated thoughts, indicating that participants with greater attention control were better able to maintain task focus and were less susceptible to distractions. Furthermore, the relationship between attention control and reading comprehension was partially mediated by task-unrelated thoughts (Figure 12.3). In other words, mind wandering partly explained the relationship between attention control and reading comprehension. That said, the direct path from attention control to reading comprehension remained significant even after accounting for task-unrelated thoughts, suggesting that mind wandering only captured part of the covariance between attention control and comprehension. McVay and Kane (2012) speculated that this unexplained covariance between attention control and reading comprehension may represent effective competition resolution, which is measured by tests of attention control and required when readers encounter ambiguity in a passage of text. As another example, Blankenship et al. (2019) examined the development of infants’ attentional abilities and their relationship to reading achievement at age 6 in a longitudinal study of 157 children. Blankenship et al. (2019) measured the attention abilities of 5-month-old infants by showing them a 45 second video clip from Sesame Street. They counted the number of times the infants shifted their gaze and the longest duration they looked at the video. Blankenship et al. (2019) found that infants’ attentional abilities predicted their executive functioning five months later, as measured by the A-not-B task (i.e., an “updating” test in which infants are challenged to
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
255
256
A L E X A N D E R P. B U R G O Y N E E T A L .
find a toy hidden in a new location; to do so, they must avoid perseverating on a previously learned location). They also found that differences in executive functioning were reliable and displayed continuity with age, such that each subsequent measure (obtained at ages 3, 4, and 6 using span tasks and tests of attention control) was significantly related to the previous developmental measure. Critically, the relationship between attentional abilities in infancy and reading achievement at age 6 was mediated by executive functioning across development. This result held even after controlling for verbal intelligence, suggesting that individual differences in domain-general attentional abilities can be detected early on, and contribute to reading achievement above and beyond domain-specific language abilities.
12.5
Maintenance and Disengagement
Why does attention control contribute to performance on working memory, fluid intelligence, and language comprehension tasks? In our theoretical framework, attention control is necessary for performing two distinct but complementary cognitive functions that are important to a wide range of tasks. Those two functions are maintenance and disengagement (Burgoyne & Engle, 2020; Shipstead et al., 2016). Maintenance refers to the cognitive operations that support keeping track of information, particularly amid distraction and interference. For example, maintenance is required when building the gist of a complex story told by your friend in the coffee shop among lots of distractions and interruptions. Sources of interference can include taskirrelevant thoughts, as well as external events that threaten to capture attention. Disengagement, on the other hand, is responsible for removing no-longer-relevant information from active processing, and flagging it for nonretrieval. For example, one must disengage from irrelevant information that was processed during the interruptions in your friend’s story. We think most tasks require both information maintenance and disengagement, but the extent to which each is important depends on the cognitive demands of the task at hand (Figure 12.4). For example, in complex span working memory tests, maintenance plays a critical role because the performer must keep track of memoranda while completing secondary processing tasks. Disengagement seems less important than maintenance in complex span tasks; however, the performer must still disengage from memoranda from prior trials and the processing subtasks to perform well. By contrast, in fluid intelligence tests such as Raven’s matrices (Raven & Court, 1998) or number series (Thurstone, 1938), we think disengagement plays a larger role than maintenance. Many fluid intelligence tasks challenge participants to discover relationships or abstract rules among stimuli. As participants rule out disproven hypotheses, they must prevent these incorrect hypotheses from being reretrieved, reentering the focus of attention, and interfering with the discovery of novel
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
Figure 12.4 Attention control supports information maintenance and disengagement in service of complex cognition. Adapted from Burgoyne and Engle (2020)
solutions (Burgoyne et al., 2019b; Hambrick & Altmann, 2015). Maintenance, on the other hand, may help fluid intelligence test takers keep track of information used to generate novel hypotheses (see Burgoyne et al., 2019b). With respect to language comprehension, maintenance appears to make a substantial contribution because readers must keep track of previous information to contextualize new information, and must maintain multiple interpretations of ambiguous sentences until they are resolved in the service of structure building (Gernsbacher, 1990). Disengagement also appears to play a role; once ambiguity in a sentence has been resolved, such as after a garden-path sentence, the incorrect interpretation of that sentence should be removed from further consideration. In a recent large-scale study, Martin et al. (2020) estimated the contribution of information maintenance and disengagement to reading comprehension and second-language vocabulary learning. Martin et al. (2020) had 567 young adults (ages 18–35) complete tests of working memory capacity, memory updating, and fluid intelligence. Using structural equation modeling, Martin et al. (2020) partitioned variance in performance on these tasks into latent factors representing maintenance and disengagement. The models revealed that maintenance and disengagement were statistically separable at the latent level. Moreover, each made substantial and
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
257
258
A L E X A N D E R P. B U R G O Y N E E T A L .
significant contributions to reading comprehension and second-language vocabulary learning; together, they accounted for 58 percent of the variance in reading comprehension and 61 percent of the variance in secondlanguage vocabulary learning.
12.6
Measuring Attention Control
Although the preceding results suggest that attention control contributes to language comprehension, psychometric limitations have posed a challenge for researchers attempting to directly measure individual differences in attention control. These limitations were shown clearly at the latent level by Friedman and Miyake (2004), who had 220 undergraduates complete nine tasks designed to measure three attentional functions (inhibiting a prepotent response, resisting distraction, and resisting proactive interference). Friedman and Miyake (2004) found that most of the measures were unreliable, with an average internal consistency below .60. Because unreliability attenuates correlations, it is perhaps unsurprising that the measures correlated weakly with each other (only one was above r = .18), and that the average factor loading was below .40. In light of these results, Friedman and Miyake (2004) suggested that researchers develop new tests of attention control with greater reliability, process purity, and sensitivity to individual differences. Despite this suggestion, many of the attention tasks used by Friedman and Miyake (2004) are still used today. The use of psychometrically unsound tasks has led some to conclude that measures of attention reflect task-specific factors and not an underlying unitary ability (Kramer et al., 1994; Rey-Mermet et al., 2018). Others have argued that it is difficult to draw conclusions about the unity or diversity of attention control as a cognitive construct in the presence of psychometric limitations such as unreliability (Paap & Sawi, 2016), contamination by processing speed, strategy, semantic memory, and speed-accuracy trade-offs (Draheim et al., 2019; Hedge et al., 2020), and a small effect size to noise ratio (Rouder & Haaf, 2019). We have argued that problems affecting the measurement of attention control are largely due to the use of response time difference scores, which reduce reliability and induce contamination by processing speed and speedaccuracy trade-offs (Draheim et al., 2019; 2021). Difference scores use a subtraction methodology; an individual’s performance in one condition is subtracted from their performance in another condition. For example, in the Stroop task, participants must indicate the color a word is printed in, not the color it refers to (Stroop, 1935). Trials can be congruent (e.g., “RED” in red ink) or incongruent (e.g., “BLUE” in red ink). Performance on incongruent trials is hypothesized to require controlled attention, because participants must resolve conflict between the word and the color it is printed in. By contrast, performance on congruent trials requires largely nonattentional processes, given the lack of conflict resolution required and the automaticity of reading
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
Figure 12.5 Reliability of a difference score (Y-axis) decreasing as the correlation between the component scores increases (X-axis) Each line represents the reliability of the difference score when the reliabilities of the component scores are set to r(xx) = .60, .70, .80, or .90.
(see MacLeod, 1991, for a review). The difference between performance on incongruent and congruent trials is thought to reflect attention-specific variance, and for this reason many attention tasks use difference scores between performance on conditions requiring controlled attention and conditions thought to reflect largely automatic processes. The subtraction methodology appears to be a great tool for experimental researchers (see Chiou & Spreng, 1996), but the use of difference scores in individual differences research has been denounced by psychometricians for over half a century (Cronbach & Furby, 1970). Many researchers have noted that difference scores are poorly suited for correlational work because they are often unreliable and minimize between-subjects variance (Ackerman & Hambrick, 2020; Draheim et al., 2016; 2019; Friedman & Miyake, 2004; Hedge et al., 2018). Difference scores are less reliable than their component scores (i.e., the performance measures from each condition used to calculate the difference score) in all practical situations because subtraction removes the shared – and therefore reliable – variance of the component scores but preserves the error variance. As shown in Figure 12.5, the unreliability of a difference score depends on the reliability of its components and how strongly those components are correlated. For attention tasks, congruent and incongruent trials are typically highly reliable (e.g., .90) and strongly correlated (e.g., .80), leading to unreliable difference scores and subsequently poor validity (Draheim et al., 2021; Hedge et al., 2018; Paap & Sawi, 2016). With these issues in mind, we recently developed new-and-improved tasks to measure attention control (Draheim et al., 2021). We administered ten attention tasks to over 400 participants, including “classic” tasks (e.g., the antisaccade task), modified tasks (e.g., the Stroop task with an adaptive response deadline), and new tasks (e.g., the sustained-attention-to-cue task). Our new and modified tasks avoided the use of difference scores. Many of them used an adaptive procedure in which the tasks became easier or more difficult depending on how the participant performed. In these adaptive
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
259
260
A L E X A N D E R P. B U R G O Y N E E T A L .
tasks, we set the converged-upon accuracy rate to be constant across participants, and used the level of task difficulty at which the participant could perform at this accuracy rate as the dependent measure. The new accuracy-based attention tasks were markedly better than classic tasks that relied on difference scores or response times in terms of reliability, intercorrelations, loadings on an attention control factor, and associations with fluid intelligence. As we noted above, using these new tasks, we found that attention control fully mediated the relationship between working memory capacity and fluid intelligence at the latent level, a result that could not be attributed to processing speed. Furthermore, the results suggested that attention control is a unitary ability, when measured using psychometrically sound tasks.
12.7
Measuring Working Memory Capacity
The measurement of working memory capacity is considerably less contentious than that of attention control, with several psychometrically sound tasks available to researchers. The strong reliability and criterion validity of these measures is largely attributable to the tasks being designed for individual differences research and scored without using response times or difference scores. That said, there are at least three ongoing issues pertaining to the measurement of working memory capacity to consider, including whether tasks are interchangeable, whether they are appropriate for lowerand higher-ability samples, and whether administration time can be reduced without loss in reliability or criterion validity. Although researchers use a variety of tasks to draw conclusions about working memory capacity, these conclusions may differ depending on the tasks used to measure it. For example, while complex span tasks are popular, so are the n-back and running span tasks. In the n-back, participants are presented a continual series of stimuli (e.g., letters) and must respond when the current stimulus is identical to the stimulus presented n trials ago. In the running span task, participants are presented a series of stimuli and must recall the last x number of stimuli in the order they were presented. Despite ostensibly measuring the same construct, a meta-analysis by Redick and Lindsey (2013) revealed that performance on complex span tasks correlated weakly with performance on n-back tasks (r = .20) – the two measures shared only 4 percent of their variance. In another study, this time using latent variable analyses, Harrison (2017) found that complex span and n-back measures loaded onto separate factors that shared less than one-quarter of their reliable variance. Moreover, both factors accounted for unique variance in fluid intelligence. Harrison (2017) also found that n-back tasks with a larger stimulus pool loaded onto a separate factor than n-back with a smaller stimulus pool, which in turn affected their relationships with complex span performance and fluid intelligence.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
These results suggest that measures of working memory capacity based on complex span and n-back tasks may not reflect the same construct or source of variance, and that task-specific factors may play a role in explaining contradictory results across studies that use the n-back. On the other hand, Broadway and Engle (2010) found that running span and complex span performance was strongly correlated, and that both measures had nearly equivalent relationships with fluid intelligence. Furthermore, Broadway and Engle (2010) found that these relationships were largely invariant to task-specific factors in the running span task, such as the presentation rate and whether the participant knew the set size (i.e., the number of items to be remembered) in advance. Taken together, these studies indicate that working memory capacity measures are not always interchangeable. A robust assessment of working memory capacity should therefore include more than one type of task, as we discuss in the “Practical Recommendations” section. Another consideration is the match between the difficulty of the task and the ability level of the population of interest. To shed light on this issue, Draheim et al. (2018) used item response theory to analyze three complex span tasks: operation span, symmetry span, and rotation span. The analyses revealed that the standard operation span task was poor at differentiating between high- or even average-ability individuals, in part because there were ceiling effects (i.e., performance at or near 100 percent) on trials with lower set sizes. As a result, operation span performance and fluid intelligence were not significantly correlated among the top third of performers unless larger set sizes were added to make the task more difficult. For comparison, the standard rotation and symmetry span tasks were much better at distinguishing between average- and high-ability individuals, although they also benefited from adding larger set sizes. Because the smaller set sizes used in these three complex span tasks only tapped variance among the worst performers, Draheim et al. (2018) concluded that in many cases they could be removed to reduce administration time. A final consideration is that working memory capacity tasks are timeconsuming. A battery of three standard complex span tasks takes over an hour to administer. Recent efforts to shorten these tasks by reducing practice time, removing smaller set sizes, and reducing the number of trials have been relatively successful. For example, Foster et al. (2015) and Oswald et al. (2015) used different approaches to shorten complex span tasks but converged on a similar conclusion: although shortening the tasks reduced their internal consistency reliability, it decreased their administration time by 20–40 percent and left their criterion validity largely intact. Nevertheless, three shortened complex span tasks still require over 40 minutes to complete, compared to roughly 30 minutes for attention control and 25 minutes for fluid intelligence tasks (Draheim et al., 2021). As such, complex span tasks could benefit from further efforts to shorten their administration time, perhaps by making them adaptive in difficulty on a trial-by-trial basis.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
261
262
A L E X A N D E R P. B U R G O Y N E E T A L .
12.8
Practical Recommendations
In this section, we provide practical recommendations to researchers interested in conducting studies of individual differences in cognitive ability. Given that differential psychology (i.e., the study of individual differences) is rarely taught in undergraduate- and graduate-level methods courses, and that best practices for differential research are rarely discussed in scientific publications (see Burgoyne et al., 2020), these recommendations may not be obvious to the uninitiated differential researcher.
12.8.1
Carefully Consider Whether the Cognitive Tasks You Administer Will Reflect the Cognitive Construct You Intend to Measure Given Your Population of Interest It is all too easy to select a task described as a measure of a cognitive construct, administer it to a sample, and assume you are properly measuring that construct or ability. While this may sometimes be the case, researchers should consider whether the demographics of the sample the task was developed and validated for are comparable to the researcher’s sample of interest. For example, the same “working memory” task may reflect different abilities when administered to different age groups, such as young adults or children. Other scenarios might not be as obvious, for instance, administering a computerized task to a sample that is not proficient in using a computer, or ensuring that task instructions are fully understood by nonnative speakers or those with lower language proficiency.
12.8.2
Ensure Your Sample of Subjects Reflects a Broad Range of Abilities Measures of cognitive abilities are designed to identify individual differences, so it is critical to include individuals who differ in your sample if you are interested in the population at large. A homogenous sample (e.g., university students enrolled in an introductory psychology course) can result in severe restriction of range, leading to reduced between-subjects variance (i.e., reduced variability across participants) and therefore lower reliability and validity of the measures.
12.8.3
Individual Differences Studies Require Larger Samples Than Typical Experimental Studies For example, Schönbrodt and Perugini (2013) concluded that correlations for moderate-sized effects do not stabilize until the sample size approaches 250 participants. Determining the actual sample size required for stable correlations is more complex than conducting a power analysis; we suggest
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
263
Table 12.1 Working Memory Tasks Available for Download from https://englelab.gatech.edu Task
Versions
Description
Reading Span
Standard Shortened Translated Standard Advanced Shortened Translated Standard Advanced Shortened Translated Standard Advanced Shortened Translated
Participants read sentences and remember a word presented after each sentence. After a series of sentences, participants recall the words in the presented order. Participants solve simple three-term math problems such as “does (3 5) – 7 = 8?”, each of which is followed by the presentation of a letter. After a series of items, participants recall the letters in the presented order. Participants judge whether a pattern of black and white squares is symmetrical about the midline. Afterward, they are shown a 44 grid with one square depicted in red. After a series of items, participants recall the position of the red squares in the presented order. Participants mentally rotate uppercase consonants to the upright position and indicate whether the consonant is mirror-reversed. Afterward, they are presented with a big or small arrow pointing one of eight possible directions. After a series of items, participants recall the arrows in the presented order. Participants count the number of target shapes appearing in a display of targets (e.g., blue circles) and non-targets (e.g., green circles). After a series of displays, participants recall the number of targets in each array in the presented order. Participants are presented with a list of letters and must report the last x number of letters in the presented order.
Operation Span
Symmetry Span
Rotation Span
Counting Span
Standard Translated
Running Span
Standard Translated
Note: These tasks are available for download at https://englelab.gatech.edu/. Standard = standard administration; Advanced = includes larger set sizes for higher-ability participants; Shortened = reduced administration time; Translated = translated into a language besides English.
consulting Table 12.1 in Schönbrodt and Perugini (2013). Latent variable analyses are problematic when samples are too small to produce a robust and stable correlation matrix.
12.8.4 Avoid Using a Single Task to Measure a Cognitive Ability Although cognitive tasks are designed to measure processes associated with a particular cognitive ability, any single task will inevitably also reflect a wide variety of extraneous cognitive processes. This creates a problem when researchers attempt to equate scores on a single task to a cognitive ability. Instead, a latent variable approach should be used to capture variance common to a set of tasks. Extraneous processes will not be captured by the latent variable if they are unshared across tasks. As a rule of thumb, at least three tasks per latent construct is advised.
12.8.5
The Particular Set of Tasks Used to Measure a Cognitive Ability Matters Although a latent variable approach provides a more “process pure” and theoretically meaningful measure of cognitive ability relative to analyses of
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
264
A L E X A N D E R P. B U R G O Y N E E T A L .
observed variables, in practice, the reliability and validity of the latent variable will depend heavily on the set of tasks used. For instance, measuring working memory capacity by exclusively using complex span tasks, or n-back tasks, will result in somewhat different latent constructs. Even though both sets of tasks are considered working memory capacity tasks, there is evidence that they do not necessarily measure the same thing (Redick & Lindsey, 2013).
12.8.6
For Broad Cognitive Abilities, Use a Heterogeneous Set of Tasks That Reflect Different Domain-Specific Processes For instance, a set of tasks should tap both verbal and spatial abilities. This ensures that the common variance that is captured by the latent factor is domain-general and does not reflect more domain-specific abilities. Note that one potential limitation of current attention control tasks is that they often rely on visual-spatial processing, and there are not many that require verbal or auditory processing.
12.8.7
Do Not Use Difference Scores to Assess Individual Differences As we discussed in the “Measuring Attention Control” section, difference scores subtract an individual’s performance in one condition from their performance in another condition and should be avoided in correlational research. Difference scores are less reliable than their component scores, leading to a poor signal-to-noise ratio and attenuated validity. Furthermore, difference scores are not a necessary requirement to measure individual differences in attention control. In the “Task Downloads” section, we provide researchers with attention tasks that do not use difference scores.
12.8.8 Do Not Use an Extreme-Groups Design Extreme-groups design refers to comparing a subset of participants categorized into high- versus low-ability groups. Historically, extreme-groups comparisons have been used to circumvent the need for large samples because they do not require the full continuous range of an ability. However, from our own experience, extreme-groups designs have the potential to confound one ability with another. For instance, because working memory capacity and fluid intelligence are highly correlated, participants that are categorized as high working memory capacity will also have high fluid intelligence. Extreme-groups designs can be used for exploratory purposes or to justify a larger-scale study, but ideally, researchers should use a large sample representing a continuous range of an ability before making strong claims about cognitive mechanisms.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
12.8.9
Measure Cognitive Abilities That Are Highly Related to Predictor and Criterion Variables To draw conclusions about the mechanisms underlying performance in a domain, researchers should include measures that are highly related to the predictor and criterion variables. Simply showing that a cognitive ability is correlated with performance in a domain does not provide strong evidence that the cognitive ability underpins performance in that domain. For instance, if working memory capacity is moderately correlated with a criterion measure, it is likely that fluid intelligence and attention control will also correlate with the criterion measure. Without measuring these constructs, it cannot be determined which construct is most important. What is needed is incremental validity: Does working memory capacity predict the criterion measure above and beyond attention control and fluid intelligence? Granted, it is unfeasible to measure every variable that might be related to the predictor and criterion variables, but researchers should endeavor to rule out theoretically plausible third-variable explanations.
12.8.10
Report Detailed Demographic Information, Task Descriptions, Descriptive Statistics, Reliabilities, and Bivariate Correlations A detailed description of your sample (e.g., age, sex, level of education, race, ethnicity) helps researchers evaluate the validity and generalizability of your conclusions, and determine whether the tasks you administered were appropriate for the sample under investigation. The same is true for reporting descriptive statistics, reliabilities, and bivariate correlations between measures. Some cognitive tasks have standardized administration procedures, however, many attention control tasks do not. Laboratories vary widely in the properties of the stimuli that are presented, the proportion of trial types (e.g., congruent vs. incongruent), the total number of trials, and so on. Some of these differences will have a greater impact on the reliability and validity of the task than others. For instance, many attention control tasks may require more trials than are typically administered to attain adequate reliability (Rouder et al., 2019). However, tasks that are adaptive in difficulty may reduce administration time without significant loss in reliability and validity (Draheim et al., 2021). Also, the proportion of trial types interacts with individual differences in cognitive ability in both the Stroop and flanker tasks (Heitz & Engle, 2007; Kane & Engle, 2003). In general, a higher proportion of congruent trials to incongruent trials (e.g., 2:1) is optimal for capturing individual differences in attention control. While our research has shown that the order in which tasks are administered can actually change what the task is measuring (see, e.g., Kane et al., 2001), counterbalancing task order between participants may only add more noise if you do not include the counterbalance variable in the statistical model. Our recommendation is for researchers to think carefully
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
265
266
A L E X A N D E R P. B U R G O Y N E E T A L .
Table 12.2 Attention Control Tasks Available for Download from https://englelab.gatech.edu Task
Description
Antisaccade
Participants fixate centrally as a flashing cue appears to the side. Participants must rapidly look to the opposite direction to detect a briefly presented letter. Participants are shown color words appearing in different colors. On congruent trials, the word and its color match (“RED” in red color), whereas on incongruent trials, the word and its color do not match (“BLUE” in red color). Participants must report the color the word appears in. This task uses an adaptive response deadline; better performance results in a shorter response deadline, worse performance results in a longer response deadline. The dependent measure is the response deadline duration at the end of the task. Participants are presented five arrows and must indicate the direction of the central arrow. The flanking arrows can either be congruent ( ) or incongruent ( ! ) with the direction of the central arrow. This task uses an adaptive response deadline; better performance results in a shorter response deadline, worse performance results in a longer response deadline. The dependent measure is the response deadline duration at the end of the task. Participants are cued to pay attention to either the red or blue rectangles. Next, they are briefly shown a target array of red and blue rectangles in different orientations. After a visual mask, a test array appears that is either identical to the target array or differs by one item. A white dot appears on one of the items, and the participant must determine whether that item changed or remained the same across the two arrays. Participants are shown a circle cue that gradually narrows on a target location to be monitored for the presentation of a letter. After a variable wait interval, a centrally-presented asterisk flashes, and a letter surrounded by distractors is briefly presented at the target location. The participant must indicate the letter.
Color Stroop Adaptive Deadline
Arrow Flanker Adaptive Deadline
Selective Visual Arrays
Sustained Attention to Cue
Note: These tasks are available for download at https://englelab.gatech.edu/.
about the order of tasks presented to participants, with consideration of potential spill-over effects across tasks if counterbalancing is not used.
12.9
Task Downloads
Researchers can download working memory capacity and attention control tasks for free from our laboratory website: https://englelab.gatech.edu/. In Tables 12.1 and 12.2, we describe the tasks available for download. Many of the working memory capacity tasks available for download have been translated into languages besides English. We also provide versions of our tasks with shortened administration times, and advanced versions that contain larger set sizes for higher-ability samples. All tasks were programmed in E-Prime.
12.10
Conclusion
In this chapter, we discussed the measurement of working memory capacity and attention control, and how these measures have been used by language
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
researchers to better understand the cognitive processes underpinning comprehension. We also discussed challenges associated with measuring these cognitive abilities, and provided recommendations and resources to researchers interested in conducting studies of individual differences. Although we did not discuss what governs the top-down control of attention, one plausible explanation is that what we attend to is guided in part by the contents of long-term memory and their interaction with environmental cues (see Delaney, 2018, and also Adams & Delaney, this volume). At the very least, such attempts at explaining the top-down control of attention help circumvent an infinite regress of central executives controlling central executives. Research on attention control is evolving rapidly, and we are excited to see where the adoption of more sophisticated statistical techniques and measurement methods take the field.
References Ackerman, P. L., & Hambrick, D. Z. (2020). A primer on assessing intelligence in laboratory studies. Intelligence, 80, 101440. Baddeley, A. (1992). Working memory. Science, 255(5044), 556–559. Baddeley, A. (2002). Is working memory still working? European Psychologist, 7, 85–97. Baddeley, A., & Hitch, G. (1974). Working memory. In Psychology of learning and motivation (Vol. 8, pp. 47–89). Academic press. Blankenship, T. L., Slough, M. A., Calkins, S. D., Deater-Deckard, K., KimSpoon, J., & Bell, M. A. (2019). Attention and executive functioning in infancy: Links to childhood executive function and reading achievement. Developmental Science, 22, e12824. Bock, J. K., & Miller, C. A. (1991). Broken agreement. Cognitive Psychology, 23, 45–93. Broadway, J. M., & Engle, R. W. (2010). Validating running memory span: Measurement of working memory capacity and links with fluid intelligence. Behavior Research Methods, 42, 563–570. Burgoyne, A. P., & Engle, R. (2020). Attention control: A cornerstone of higherorder cognition. Current Directions in Psychological Science, 29(6), 624–630. Burgoyne, A. P., Hambrick, D. Z., & Altmann, E. M. (2019a). Is working memory capacity a causal factor in fluid intelligence? Psychonomic Bulletin & Review, 26, 1333–1339. Burgoyne, A. P., Hambrick, D. Z., & Altmann, E. M. (2019b). Placekeeping ability as a component of fluid intelligence: Not just working memory capacity. The American Journal of Psychology, 132, 439–449. Burgoyne, A. P., Tsukahara, J. S., Draheim, C., & Engle, R. W. (2020). Differential and experimental approaches to studying intelligence in humans and nonhuman animals. Learning & Motivation, 72.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
267
268
A L E X A N D E R P. B U R G O Y N E E T A L .
Caplan, D., & Waters, G. S. (1999). Verbal working memory and sentence comprehension. Behavioral and Brain Sciences, 22, 77–126. Chiou, J. S., & Spreng, R. A. (1996). The reliability of difference scores: A reexamination. Journal of Consumer Satisfaction Dissatisfaction and Complaining Behavior, 9, 158–167. Conway, A. R., Cowan, N., & Bunting, M. F. (2001). The cocktail party phenomenon revisited: The importance of working memory capacity. Psychonomic Bulletin & Review, 8, 331–335. Cronbach, L. J., & Furby, L. (1970). How should we measure “change” – or should we? Psychological Bulletin, 74, 68–80. Crowder, R. G. (1982). The demise of short-term memory. Acta Psychologica, 50, 291–323. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Memory and Language, 19, 450–466. Daneman, M., & Green, I. (1986). Individual differences in comprehending and producing words in context. Journal of Memory and Language, 25, 1–18. Daneman, M., & Merikle, P. M. (1996). Working memory and language comprehension: A meta-analysis. Psychonomic Bulletin & Review, 3, 422–433. Delaney, P. F. (2018). The role of long-term working memory and template theory in contemporary expertise research. Journal of Expertise, 3, 155–161. Draheim, C., Harrison, T. L., Embretson, S. E., & Engle, R. W. (2018). What item response theory can tell us about the complex span tasks. Psychological Assessment, 30, 116–129. Draheim, C., Hicks, K. L., & Engle, R. W. (2016). Combining reaction time and accuracy: The relationship between working memory capacity and task-switching as a case example. Perspectives on Psychological Science, 11, 133–155. Draheim, C., Mashburn, C. A., Martin, J. D., & Engle, R. W. (2019). Reaction time in differential and development research: A review and commentary on problems and alternatives. Psychological Bulletin, 145, 508–535. Draheim, C., Tsukahara, J. S., Martin, J. D., Mashburn, C. A., & Engle, R. W. (2021). A toolbox approach to improving the measurement of attention control. Journal of Experimental Psychology: General, 150(2), 242-275. Engle, R. W., Carullo, J. J., & Collins, K. W. (1991). Individual differences in working memory for comprehension and following directions. The Journal of Educational Research, 84, 253–262. Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. (1999). Working memory, short-term memory, and general fluid intelligence: A latent-variable approach. Journal of Experimental Psychology: General, 128, 309–331. Engel de Abreu, P. M. J. E., Gathercole, S. E., & Martin, R. (2011). Disentangling the relationship between working memory and language: The roles of short-term storage and cognitive control. Learning and Individual Differences, 21, 569–574.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
Farnham-Diggory, S., & Gregg, L. W. (1975). Short-term memory function in young readers. Journal of Experimental Child Psychology, 19, 279–298. Foster, J. L., Shipstead, Z., Harrison, T. L., Hicks, K. L., Redick, T. S., & Engle, R. W. (2015). Shortened complex span tasks can reliably measure working memory capacity. Memory & Cognition, 43, 226–236. Friedman, N. P., & Miyake, A. (2004). The relations among inhibition and interference control functions: A latent-variable analysis. Journal of Experimental Psychology: General, 133, 101–135. Gathercole, S. E., & Baddeley, A. D. (1993). Working memory and language. Erlbaum. Gernsbacher, M. A. (1990). Language comprehension as structure building. Psychology Press. Hale, S., Myerson, J., Rhee, S. H., Weiss, C. S., & Abrams, R. A. (1996). Selective interference with the maintenance of location information in working memory. Neuropsychology, 10, 228–240. Hambrick, D. Z., & Altmann, E. M. (2015). The role of placekeeping ability in fluid intelligence. Psychonomic Bulletin & Review, 22, 1104–1110. Harrison, T. L. (2017). N-back as a measure of working memory capacity. (Ph.D. dissertation, Georgia Institute of Technology). http://hdl.handle.net/1853/ 58762 Hedge, C., Powell, G., Bompas, A., & Sumner, P. (2020, February 1). Strategy and processing speed eclipse individual differences in control ability in conflict tasks. PsyArXiv. https://doi.org/10.31234/osf.io/vgpxq Hedge, C., Powell, G., & Sumner, P. (2018). The reliability paradox: Why robust cognitive tasks do not produce reliable individual differences. Behavior Research Methods, 50, 1166–1186. Heitz, R. P., & Engle, R. W. (2007). Focusing the spotlight: Individual differences in visual attention control. Journal of Experimental Psychology: General, 136, 217–240. Hunt E, Frost N, Lunneborg C. (1973). Individual differences in cognition: A new approach to intelligence. In G. H. Bower, (Ed.), The psychology of learning and motivation: Advances in research and theory (pp. 87–122). Academic Press Jackson, M. D., & McClelland, J. L. (1979). Processing determinants of reading speed. Journal of Experimental Psychology: General, 108, 151–181. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122–149. Kane, M. J., Bleckley, M. K., Conway, A. R., & Engle, R. W. (2001). A controlled-attention view of working-memory capacity. Journal of Experimental Psychology: General, 130, 169–183. Kane, M. J., & Engle, R. W. (2002). The role of prefrontal cortex in workingmemory capacity, executive attention, and general fluid intelligence: An individual-differences perspective. Psychonomic Bulletin & Review, 9, 637–671.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
269
270
A L E X A N D E R P. B U R G O Y N E E T A L .
Kane, M. J., & Engle, R. W. (2003). Working-memory capacity and the control of attention: The contributions of goal neglect, response competition, and task set to Stroop interference. Journal of Experimental Psychology: General, 132, 47–70. Kintsch, W., & Van Dijk, T. A. (1978). Toward a model of text comprehension and production. Psychological Review, 85, 363–394. Kline, R. B. (2015). Principles and practice of structural equation modeling. Guilford. Kramer, A. F., Humphrey, D. G., Larish, J. F., & Logan, G. D. (1994). Aging and inhibition: Beyond a unitary view of inhibitory processing in attention. Psychology and Aging, 9, 491–512. MacLeod, C. M. (1991). Half a century of research on the Stroop effect: An integrative review. Psychological Bulletin, 109, 163–203 Martin, J. D., Shipstead, Z., Harrison, T. L., Redick, T. S., Bunting, M., & Engle, R. W. (2020). The role of maintenance and disengagement in predicting reading comprehension and vocabulary learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 46, 140–154. McVay, J. C., & Kane, M. J. (2012). Why does working memory capacity predict variation in reading comprehension? On the influence of mind wandering and executive attention. Journal of Experimental Psychology: General, 141, 302–320. Moray, N. (1959). Attention in dichotic listening: Affective cues and the influence of instructions. Quarterly Journal of Experimental Psychology, 11, 56–60. Oswald, F. L., McAbee, S. T., Redick, T. S., & Hambrick, D. Z. (2015). The development of a short domain–general measure of working memory capacity. Behavior Research Methods, 47, 1343–1355. Paap, K. R., & Sawi, O. (2016). The role of test-retest reliability in measuring individual and group differences in executive functioning. Journal of Neuroscience Methods, 274, 81–93. Perfetti, C. A., & Goldman, S. R. (1976). Discourse memory and reading comprehension skill. Journal of Verbal Learning and Verbal Behavior, 15, 33–42. Perfetti, C. A., & Lesgold, A. M. (1977). Discourse comprehension and sources of individual differences. In M. A. Just and P. A. Carpenter (Eds.), Cognitive processes in comprehension. Lawrence Erlbaum Associates. Raven, J. C., & Court, J. H. (1998). Raven’s progressive matrices and vocabulary scales. Oxford Pyschologists Press. Redick, T. S., & Lindsey, D. R. (2013). Complex span and n-back measures of working memory: A meta-analysis. Psychonomic Bulletin & Review, 20, 1102–1113. Rey-Mermet, A., Gade, M., & Oberauer, K. (2018). Should we stop thinking about inhibition? Searching for individual and age differences in inhibition ability. Journal of Experimental Psychology: Learning, Memory, and Cognition, 44(4), 501.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Measuring Individual Differences
Rizzo, N. D. (1939). Studies in visual and auditory memory span with special reference to reading disability. The Journal of Experimental Education, 8, 208–244. Rouder, J. N., & Haaf, J. M. (2019). A psychometrics of individual differences in experimental tasks. Psychonomic Bulletin & Review, 26, 772–789. Rouder, J., Kumar, A., & Haaf, J. M. (2019, March 25). Why most studies of individual differences with inhibition tasks are bound to fail. https://doi .org/10.31234/osf.io/3cjr5 Schönbrodt, F. D., & Perugini, M. (2013). At what sample size do correlations stabilize? Journal of Research in Personality, 47, 609–612. Shah, P., & Miyake, A. (1996). The separability of working memory resources for spatial thinking and language processing: An individual differences approach. Journal of Experimental Psychology: General, 125, 4–27. Shipstead, Z., Harrison, T. L., & Engle, R. W. (2016). Working memory capacity and fluid intelligence: Maintenance and disengagement. Perspectives on Psychological Science, 11, 771–799. Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18, 643–662. Swanson, H. L., & Ashbaker, M. H. (2000). Working memory, short-term memory, speech rate, word recognition and reading comprehension in learning disabled readers: Does the executive system have a role? Intelligence, 28, 1–30. Thurstone, L. L. (1938). Primary mental abilities. Psychometric Monographs, 1, 270–275. Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task dependent? Journal of Memory and Language, 28, 127–154. Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R. W. (2005). An automated version of the operation span task. Behavior Research Methods, 37, 498–505. Unsworth, N., Redick, T. S., Heitz, R. P., Broadway, J. M., & Engle, R. W. (2009). Complex working memory span tasks and higher-order cognition: A latent-variable analysis of the relationship between processing and storage. Memory, 17, 635–654. Waters, G., Caplan, D. & Hildebrandt, N. (1987). Working memory and written sentence comprehension. In M. Coltheart (Ed.), Attention and performance XII: The psychology of reading (pp. 531–555). Erlbaum.
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
271
https://doi.org/10.1017/9781108955638.015 Published online by Cambridge University Press
Part III
Linguistic Theories and Frameworks
https://doi.org/10.1017/9781108955638.016 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.016 Published online by Cambridge University Press
13 Have Grammars Been Shaped by Working Memory and If So, How? John A. Hawkins 13.1
Introduction: Performance-Grammar Correspondences and Their Relevance for Psycholinguistic Models and Working Memory
During the last 20 years we have seen mounting evidence in the language sciences for a powerful correlation between performance preferences within languages and the conventions of grammars across languages. When language users have alternative structures to choose from (e.g., different word orders or relative clauses) the preferences found in corpora and online experiments look very much like the patterns and hierarchies seen in grammars that permit fewer structures of the relevant type. Many of these preferences and grammatical patterns have been attributed to working memory (WM) and its limitations as proposed in models of language processing. The purpose of this chapter is to review some of these WM explanations for grammars, and conversely to ask what grammars can tell us about the nature of WM and its role within a larger theory of language processing. The suggestion that grammars can tell us anything at all about WM would have seemed implausible not so very long ago. After all, earlier theorising in generative grammar (Chomsky, 1965) claimed that grammar was autonomous from performance and not shaped by it in any way, even though the competence grammar was an integral part of a model of performance.1 More recently this position has been modified (Chomsky, 2005; Trotzke et al., 2013) to allow for a more bidirectional relationship that also permits performance influences on grammar as “third factors” in language design (see O’Grady 2012, 2017 and Hawkins 2014, pp. 62–72, for critical discussion). Still, the fact that the conventions of grammars have been shaped by performance pressures has not yet
I would like to thank Elaine Francis, Tom Wasow, and an anonymous reviewer for giving me most helpful comments on an earlier version of this paper. They have resulted in numerous improvements and clarifications. The editors also provided useful feedback. I am grateful to them all.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
276
JOHN A. HAWKINS
been fully recognized, nor have its consequences been fully appreciated, either in linguistics or in psycholinguistics. And it is this influence of performance on grammars that enables us to consider both in relation to WM. Before discussing WM itself and the models and measures proposed for it, some more background is accordingly needed in the current state of the art on the correlations between performance data and grammars. They have been captured in the Performance-Grammar Correspondence Hypothesis of Hawkins (2004, p. 3): (1)
Performance-Grammar Correspondence Hypothesis (PGCH) Grammars have conventionalized syntactic structures in proportion to their degree of preference in performance, as evidenced by patterns of selection in corpora and by ease of processing in psycholinguistic experiments.
The significance of PGCH for linguistic theory and language typology is that grammatical conventions are not autonomous from performance: they cannot be explained independently of the principles that underlie language use and processing ease. For psycholinguists, grammars can now be seen as “frozen” or “fixed” conventionalizations of the very performance preferences that led to the formulation of one WM model over another.2 They provide an additional set of relevant data involving grammatical rules and constraints in the world’s languages, of the type that will be illustrated in this chapter. Such data go beyond corpora and experimental findings from individual languages, and these data are especially valuable when they involve structures and grammars that are typologically very different from English and Standard Average European (Haspelmath, 2001), on which most psycholinguistic research has focused hitherto (Jaeger & Norcliffe, 2009). Greenberg (1966) was the first to document systematic correspondences between grammars and performance, in his discussion of “markedness hierarchies” in morphology, like Singular > Plural > Dual > Trial/Paucal. Morphological inventories across languages provided evidence for the implicational patterns in these hierarchies – for example, if a language has a unique grammatical morpheme for Trial number it will have one for Dual, if it has one for Dual it will have one for Plural, and so on. Meanwhile declining frequencies of usage pointed to a clear correlation with performance data. Greenberg’s text counts for number inflections on nouns in Sanskrit were, for example: Singular = 70.3 percent, Plural = 25.1 percent, and Dual = 4.6 percent. In other words, if a lower frequency item on this hierarchy is conventionalized in the grammar of a language as a unique morpheme, then all higher frequency numbers will have unique morphemes as well. This led Greenberg to propose a causal role for performance frequencies in the evolution of grammatical regularities, with certain categories being morphologized before others (Greenberg, 1995,163–164); compare also Bybee and Hopper (2001) and Haspelmath (2008, 2021) for extensive development, testing, and support for this idea.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
In syntax, Keenan and Comrie (1977) proposed an Accessibility Hierarchy (AH) for relative clause formation (Subject > Direct Object > Indirect Object/ Oblique > Genitive; cf. Comrie, 1989). Grammatical cut-off points across languages were shown to follow the AH – that is, there are grammars that disallow relativization on one or more positions and these positions are structured by the AH.3 As an explanation, Keenan and Comrie proposed that the ease of processing relative clauses declined down AH, and they presented some preliminary evidence from English to support this. Extensive performance evidence has since come from numerous languages in which corpus frequencies decline while processing load and working memory demands increase, down AH under experimental conditions (cf., e.g., Diessel & Tomasello, 2006; Hawkins, 1999; Kwon et al., 2010). In another area of grammar, word order, the preferred orderings found in languages with considerable flexibility were argued in Hawkins (1994) to be those that are productively conventionalized as fixed and basic orders in languages with less flexibility. The preferred orders and preferred grammars minimize domains of phrase structure processing and so reduce online WM demands (see Section 13.3.1). In addition to the Keenan-Comrie AH, other “filler-gap” hierarchies for relativization and WH-movement have been shown to be structured by the increasing complexity of gaps in various embedded clauses (Hawkins, 1999, 2004, ch. 7, 2014, pp. 172–176; O’Grady, this volume). The grammatical cut-offs correspond to documented degrees of declining processing ease in languages with numerous gap-containing environments (including Subjacency-violating languages like Akan; cf. Saah & Goodluck, 1995) (see Section 13.3.1). (Nominative) subject before (Accusative) direct object ordering is highly preferred in the performance of languages in which both subject before object and object before subject are grammatical (Finnish, German, Japanese, Korean). It is also highly preferred as a basic order or as the only order across grammars (Gibson, 1998; Hawkins, 1994; Tomlin, 1986; Yamashita, 2002) (see Sections 13.3 and 13.4). Performance preferences in syntax like these have been linked explicitly to WM. In order to model how exactly they enter a grammar and become conventionalized, Haspelmath (1999) proposed a theory of diachrony couched in Optimality Theory (cf. Prince & Smolensky, 1993) in which usage preferences lead to new grammatical conventions over time. He argued that many of the basic constraints of Optimality Theory have clear functional motivations, and these include motivations ultimately related to processing ease. Stochastic Optimality Theory (Bresnan et al., 2001) defined two types of constraints: “soft” ones for the preferences of performance and “hard” ones for the grammatical conventions. It was observed (in accordance with the PGCH in [1]) that what is a soft constraint in one language, that is, a preference, can be a grammatical requirement or hard constraint in another, with ungrammaticality resulting from its violation.4 Meanwhile Hawkins (1994 pp. 19–24, 2014, pp. 78–85) showed how processing considerations can be incorporated into the traditional rule formulations of generative grammar,
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
277
278
JOHN A. HAWKINS
while Chomsky (2005) and Trotzke et al. (2013) now explicitly allow for their inclusion in grammars as third factors in language design, as we have seen. These proposals for integrating online processing considerations into grammars have been made as linguists have come to appreciate the correspondences between them. They have also led to some interesting debates and disagreements, especially for phenomena such as filler-gap dependencies (cf., e.g., the papers in Sprouse & Hornstein, 2013). Is there an actual grammatical convention distinguishing grammatical from ungrammatical sentences, or simply a principle of processing defining degrees of (un) acceptability? Phillips (2013a, b) argues against what he calls “reductionist” accounts (like Hofmeister & Sag, 2010; Kluender & Kutas, 1993) that dispense with a conventionalized Subjacency constraint in the grammar and that account for the relevant gradient judgments in terms of processing complexity alone. Berwick and Weinberg (1984) had proposed that there is a grammatical Subjacency convention and that the processing demands required for sentences of this type were the explanation for it (see Section 13.2). Similarly, Hawkins (1999, 2004) argued that patterns of variation in fillergap dependencies across languages, and their hierarchies, point to differences in grammatical conventions, explainable through processing. Hofmeister et al. (2013) discuss the general difficulties in “distinguishing the effects of grammar and processing on acceptability judgments,” to quote from their title (cf. further Sprouse et al., 2012a, b), while O’Grady (2005) argues in favor of eliminating grammars in the traditional sense altogether and subsuming their effects within the parameterized procedures and routines of an efficient processor for different languages (see also O’Grady, this volume). These problems in disentangling what is in the grammar from what is in the processor arise precisely because performance preferences and grammatical patterns are so highly intercorrelated (cf. the PGCH in [1]). But because they are intercorrelated, it follows that properties of grammars, and the variation we see across languages, are directly relevant for language processing theories, including theories of WM. In this chapter I review a number of attempts that have been made to link WM to grammars and thereby explain grammatical properties and constraints in terms of WM. In Section 13.2 some quite specific proposals are examined for what the capacity constraints of WM actually are. Capacity constraints have been argued to explain certain word order restrictions across languages involving head ordering. The Subjacency condition on filler-gap dependencies has also been explained in terms of capacity constraints. We will see that these explanations have not been successful empirically, however: grammatical structures that are predicted to be unattested do in fact occur, sometimes quite productively. These facts about grammars accordingly provide evidence against the relevant capacity constraints as proposed in the WM literature, typically on the basis of a limited range of languages. Section 13.3 reviews a different approach to linking WM to grammars, one which I will argue to have been more successful, in terms of “more
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
versus less” WM demands. The central claim here is that the greater or lesser demands on WM are reflected in dispreferred versus preferred structures across languages and in declining typological distributions. These facts about grammars are an interesting test case, therefore, for the degrees of processing difficulty defined by WM theories, and are in turn potentially explained by these degrees. Some of these theories still assume that capacity constraints exist in WM, but these are not exceeded in the examples we will discuss and do not form part of the explanation for the structural preferences. The strength of this more versus less approach lies in the interesting correlations it has uncovered between declining numbers of grammars around the globe and increasing WM demands. In Section 13.4, a third approach to linking WM to grammars is examined, one that is more in line with recent developments in language processing research and in which WM is integrated with other factors that facilitate processing, such as prediction and communicative efficiency. One consequence of this multifactor approach is that a structure that adds to WM load can sometimes be preferred across languages, because of its benefits for other aspects of processing, for example, prediction. A case study will be discussed involving the relative ordering of subjects and direct objects across and within languages. In this third approach grammars and cross-linguistic comparison provide evidence about how WM interacts with other processing considerations, and this interaction can in turn explain the cross-linguistic distributions and frequencies of certain grammatical properties. I conclude briefly in Section 13.5.
13.2
Constrained Capacity WM in Grammars
In 1956 George Miller proposed that there are limits on our capacity for processing information of 7 plus or minus 2 items. This soon led to the notion of a WM with a constrained capacity for processing language and other cognitive systems (cf. Cowan, 2005, for summary and review). In psycholinguistics it led to a lively and still ongoing debate about what the precise limits are, what the linguistic units are that need to be counted, and how WM relates to other memory systems (short-term and long-term), see Baddeley (this volume). Should the capacity constraints for language refer to words (Frazier & Fodor, 1978), to discourse referents (Gibson, 1998), or to chunks and phrases of various kinds (Lu & Liu, 2016; Lu & Wen, this volume), and what exactly are the quantities for these different units (see the chapters in Part 1, ”Models and Measures,” of this volume)? The present section considers how some of these capacity constraints in WM have been linked directly to properties of grammars as a potential explanation for them. It is important to stress at the outset that if we are going to look for evidence of processing constraints of any kind in grammars, it is not
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
279
280
JOHN A. HAWKINS
sufficient to just look at English or the handful of (generally European) languages on which most psycholinguistic research has traditionally been conducted. A constraint that is present in the grammars of such languages and that is potentially explainable through, for example, a proposed capacity limit in WM may be completely absent in other grammars, with the result that structures that exceed this limit can be produced by speakers of the language(s) in question, and so provide evidence against the WM capacity limit. It is now becoming increasingly recognized (Hawkins, 2007; Jaeger & Norcliffe, 2009; Norcliffe et al., 2015) that a broad range of languages must be considered when testing psycholinguistic claims about both usage and grammars, just as consideration of a broad genetic, geographical, and typological range of languages has become commonplace in language typology when testing purely linguistic claims about universal features of grammars (cf., e.g., Dryer, 1989, 1992; Haspelmath et al., 2005). There are not many psycholinguistic studies that have considered a broad range of grammars like this when discussing WM and capacity constraints. A couple that have done so will be considered here. Frazier (1979, 1985) examined the word order universals of Greenberg (1963) from the perspective of a two-stage parsing model that was developed in Frazier and Fodor (1978). The first stage of this parser was claimed to have a limited viewing window of 5–6 adjacent words. She proposed that certain syntactic phrases with nonadjacent heads X and Y like (2a) (with a postnominal relative clause in an NP co-occurring with a postposition within a PP) would regularly exceed this limit. Corresponding structures like (2b) with prepositions instead of postpositions, as in English (John went climbing) PP[on NP[mountains that he always wanted to explore]], permit parsing of the PP and recognition of its two immediate constituents, P and NP, in just two adjacent words X and Y. (2)
a.
b.
PP[NP[Noun
Rel Clause] Postposition] Y Modifier X of Y PP[Preposition NP[Noun Rel. Clause]] X Y Modifier of Y
Frazier’s justification for the 5–6 word limit was that syntactic and semantic processing could regularly not take place in (2a) (in sentences corresponding to (John went climbing) PP[NP[mountains that. . .] on]) because the number of words required to identify phrasal packages forming coherent semantic units in the parser’s first stage would regularly exceed the limit. A similar argument applies to (3) in which the higher head X is a verb and the postnominal relative in NP occurs in a verb-final VP in (3a), but in a verb-initial VP in (3b) (as in English (John) VP[climbed NP[mountains that he always wanted to explore]]).
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
(3)
a.
VP[NP[Noun
Y
b.
Rel Clause] Verb] Modifier X of Y
VP[Verb NP[Noun
X
Y
Rel Clause]] Modifier of Y
A second example of the application of WM constraints into grammars can be found in Berwick and Weinberg’s (1984) explanation for “Subjacency” referred to above. Subjacency prohibits rules from relating two constituents or structural positions that are separated by more than one “bounding node” (Chomsky, 1981). The bounding nodes of different languages can vary and are “parameterized” (Rizzi, 1982), but the limitation on rules of, for example, movement to just one such node is claimed to be universal. A classic example involves what was first proposed as the Complex NP Constraint by Ross (1967), illustrated in the WH-question and relative clause shown in (4ab), both of which are ungrammatical in English: (4)
a.*Which bonei [did you see NP[the dogj S[that Oj was biting Oi]]]? b.*. . . the bonei [whichi you saw NP[the dogj S[that Oj was biting Oi]]]
Berwick and Weinberg attribute their ungrammaticality to the bound on left context within their proposed parser, and hence to a capacity constraint in WM as they defined it. These explanations raise two fundamental questions, one psycholinguistic, the other linguistic. How good is the processing evidence for the WM capacity constraints proposed in these parsers? And are the grammatical generalizations they set out to explain actually supported beyond English and Standard Average European (Haspelmath, 2001)? Cross-linguistic data from grammars provide, in fact, counterevidence to both. Neither Frazier’s non-adjacent head orderings of (2a) and (3a) nor Subjacency in (4) are wellsupported universal constraints on grammars. There are certainly lots of individual grammars that do incorporate the adjacent heads and Subjacency, but there are also many that don’t, and those that don’t are exceptions both to the proposed grammatical constraints and to the WM limits that have been argued to explain them. The exceptions to the excluded co-occurrence of postpositions with postnominal relatives (i.e., Noun + Rel Clause) in (2a) are actually quite numerous. Hawkins (1983, p. 96) counted as many as 22/57 (39 percent) of grammars with postpositions that also have Noun + Rel Clause in their NPs, within his expanded Greenbergian sample. Prenominal relatives (i.e., Rel Clause + Noun) are preferred in postpositional languages (in 35/57 cases, 61 percent), but Postpositions & Noun + Rel Clause is also wellattested. Lakhota is an example (cf. Lehmann, 1984):
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
281
282
JOHN A. HAWKINS
(5)
PP[NP[xé
wã S[Tamalpais éciya-pi] wã] él] mountain INDEF Tamalpais they-called INDEF on ‘on a mountain that they called Tamalpais’
(Lakhota)
Interestingly, the opposite disharmonic head ordering of prepositions with prenominal relatives shown in (6) is exceedingly rare and found in only 1/106, or 0.9 percent of prepositional languages (this language being Mandarin Chinese) in Hawkins’s (1983) expanded sample: (6)
PP[Preposition
X
NP[Rel
Clause Noun]] Modifier Y of Y
Exceptions to the co-occurrence of Noun + Rel Clause with verb-finality are also well attested. There are as many as two thirds of SOV languages in the World Atlas of Language Structures (WALS) that do not have the Rel Clause + Noun structure of languages like Japanese (Kuno, 1973; cf. Hawkins, 2014, p. 266) and instead have alternative strategies that include mostly Noun + Rel Clause (Dryer, 2005a). German sentences that have not undergone Extraposition from NP like (7) are examples of (3a) (from Hawkins, 1983, p. 266): (7)
Wie soll ich also VP[NP[deinem Vater S[der wochenlang im Ausland war]] erklären] dass How shall I therefore to-your father who for-weeks abroad was explain that ‘How shall I therefore explain to your father who was abroad for weeks that. . .’
Again the opposite disharmonic ordering of Rel Clause + Noun in a verbinitial VP, as shown in (8), is exceedingly rare and found in only one grammar (Mandarin Chinese) in Hawkins (1983) (see Hawkins 2014, pp. 146–150 for discussion and a proposed explanation): (8)
VP[Verb
X
NP[Rel
Clause Noun]] Modifier Y of Y
Languages permitting violations of Ross’s Complex NP Constraint, and hence of Subjacency, include Swedish as in (9). Compare (9), which is grammatical (Allwood, 1982), with the English (4b) which is not: (9) ...NP[ett beni S[somi jag ser a bone which I see
NP[en
hundj S[somj Oj gnager på Oi]]]] a dog which is-gnawing on
Counterexamples to Subjacency are also found in Japanese as shown in (10) (Kuno, 1973, pp. 239–240):
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
(10)
NP[S[NP[S[Oi
Oj osiete-ita] seitoj ga] rakudaisita] senseii] teaching-was student NOM flunked teacher ‘the teacher who the students that (he) was teaching flunked’
Maxwell (1979) cites Korean and Tamil as further counterexamples in typologically similar SOV grammars, while Saah and Goodluck (1995) conducted acceptability experiments in SVO Akan (Niger-Congo: Kwa) and showed that Subjacency violations are not ungrammatical and unusable in this language. In all of these languages there is evidence that “Subjacency-violating” structures are indeed hard to process, and this supports the basic intuition behind Berwick and Weinberg’s (1984) processing explanation. However, the “cut-offs” in grammaticality clearly vary across grammars, so these structures are not unparsable, and they do not exceed capacity constraints in WM. Performance factors that influence degrees of acceptability in Swedish Subjacency structures have been discussed in the papers of Engdahl and Ejerhed (1982). For Japanese, Kuno (1973, pp. 239–240, 244–260), Haig (1996), Matsumoto (1997) and Comrie (1998) all document pragmatic and semantic factors that facilitate acceptability. And Saah and Goodluck (1995) have shown that gaps in the complex NPs of Akan are indeed less acceptable than in other simpler environments in this language, but more acceptable than the corresponding Subjacency violations of English, because, they argued, of the lack of any conventionalized Subjacency constraint in Akan versus its presence in English. In other words, Subjacency violations are indeed ungrammatical in English, but not in Akan. These grammatical differences between languages are problematic for any attempt to explain grammars in terms of capacity constraints in WM. But we should not throw the baby out with the bathwater. We know, both from the Keenan-Comrie Accessibility Hierarchy and from the Subjacency literature, that these structures excluded either by language-particular (Accessibility Hierarchy) or more general (Subjacency) grammatical conventions are those that pose more severe demands on processing. Frazier’s (1979, 1985) disharmonic head orderings are also more demanding than harmonic ones (see Section 13.3.1). We also know, from the Performance-Grammar Correspondence Hypothesis (1), that we can expect a proportional response in grammars: certain sentence types should be excluded as ungrammatical in proportion to their degrees of processing difficulty, with more difficult structures being permitted in fewer grammars. These degrees can be established based on experimental data and corpora from languages (like Swedish, Japanese, and Akan) in which the relevant structures are actually grammatical. This all suggests that we may be able to link WM to grammars more successfully by shifting the focus away from capacity constraints to a second aspect of WM on which there is more certainty and agreement.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
283
284
JOHN A. HAWKINS
This is simply the idea that processing becomes harder the more items are held and operated on simultaneously in WM when reaching any one parsing decision (e.g., What is the phrase structure for this portion of the sentence, or where is the gap for this filler?). I shall refer to these as “more versus less” approaches to WM. By the Performance-Grammar Correspondence Hypothesis (1) we can then predict that rules or constraints will be conventionalized in grammars with a productivity and frequency that declines as WM load increases when processing the relevant structures.
13.3
More versus Less WM in Grammars
A primary focus on more versus less WM demands can be seen in several of Gibson’s (1998) explanations for performance data. He observes, for example, that there is a preference for subjects to be positioned before direct objects in languages like Finnish and German in which both orderings are grammatical, as in the following German examples: (11) a.
Die kleine blonde Frau küsste den roten Bären. the small blond lady-NOM kissed the red bear-ACC
b.
Den roten Bären küsste die kleine blonde Frau. the red bear-ACC kissed the small blond lady-NOM ‘the small blond lady kissed the red bear’
Gibson (1998) appeals to the added WM cost of ordering the object before the subject in (11b): OVS orders are more complex at the initial nouns because it is necessary to retain the prediction of a subject noun at this location. SVO sentences are expected to be more frequent than OVS, because they require less memory to produce and comprehend. (p. 59) An object always co-occurs with a subject, so the clearly case-marked initial Accusative object in (11b) will activate the prediction that SU follows. Subject-first NPs do not activate the prediction of an object since many (in fact most) subjects occur in intransitive clauses that lack a direct object (cf. Ueno & Polinsky, 2009). Hence, according to Gibson, subject-first orders make fewer demands on WM in comprehension and production and should be easier to process and more frequent. Notice that this explanation does not refer to limits on WM capacity. It appeals instead to a gradient, more versus less WM load in the processing of subjects and objects in transitive clauses. Gibson’s (1998) general theory does assume the existence of certain capacity constraints in WM, and some
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
of the language-particular data he discusses from English and other languages (involving, e.g., center- and self-embeddings) are discussed in terms of these constraints. But the dispreferred OVS order in (11b) falls well within these capacity limits, and his explanation for the preference for SVO in (11a) relies simply on more versus fewer items in WM, and hence on more versus less processing load. Notice also that Gibson’s proposal can be extended to basic orderings across grammars, in accordance with the Performance-Grammar Correspondence Hypothesis in (1). It is well known since Greenberg (1963) that subject before object grammars (SOV, SVO, and VSO) are much more common than the reverse (VOS, OVS, and OSV). The figures from Tomlin’s (1986) sample of 402 languages examining this were 96 percent S > O and just 4% with O > S.5 We shall see in Section 13.4 that the cross-linguistic situation is actually more complex, and that a psycholinguistic explanation incorporating Gibson’s WM idea must also take account of numerous other factors, both grammatical (especially “ergative” morphology) and psycholinguistic. Lewis and Vasishth (2005) meanwhile developed an alternative WM model to Gibson’s in terms of activation, using psychological primitives taken from cognitive psychology and from mechanisms of neural activation (cf. Anderson, 1983). A key component of their model is the idea that small domains for processing grammatical properties and relations make processing easier because they contain fewer items that can interfere with current parsing decisions, for example, fewer potential gap sites for a filler, fewer NP arguments to be linked to a given verb, and so forth. Fewer items and less interference in WM mean less cognitive activity is required, and preferences in performance data are shown to follow from this. Once again, many of their predictions follow from a more versus less approach to WM. The time decay of neural activations in their model imposes limits, and ultimately capacity constraints, on certain processing operations, but many of their predictions, as in Gibson’s theory, follow simply from more versus fewer such demands. The research program of Hawkins (1994, 2004, 2014), to which we now turn, has examined typological variation from a processing perspective and is also based on a more versus less approach to WM. His theory is noncommittal about what the absolute limits on WM may be, and indeed about whether there are any capacity constraints at all, given the problems that we saw in 13.2 with explanations for proposed universals in these terms. Instead, he has focused on the greater “efficiency” of certain structures over others in explaining these patterns, where efficiency is defined, in part, in terms of the relative numbers of items that must be held in WM simultaneously (cf. 13.4 below and Hawkins, 2014, pp. 47–49, for a general summary). Hawkins’s theory of WM is fundamentally similar to that of Gibson (1998) in assuming that more or fewer items may occur in the integration domains for processing alternative word orders and relative clauses. The only difference is that he quantifies these items in terms of words rather than discourse referents.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
285
286
JOHN A. HAWKINS
13.3.1
Cross-Linguistic Patterns in Word Orders and Filler-Gap Dependencies Hawkins (1994) examined basic word order patterns across languages in the updated “Greenbergian” correlations of Dryer (1992) from a WM perspective. Greenberg (1963) had proposed correlations between verb position and prepositions or postpositions in phrases corresponding to “went to the beach” in (12): (12)
a. b. c. d.
VP[went PP[to
the beach]] beach to] went] VP[went PP[the beach to]] VP[PP[to the beach] went] VP[PP[the
(12a) is the English order, (12b) the Japanese order, and these two sequences are highly preferred (in 94 percent of grammars) over (12cd) with disharmonic head orderings in Dryer’s (1992) sample (measuring individual languages rather than groups of related languages, or “genera”; cf. Hawkins 1994, p. 257): (13)
a. b. c. d.
VP[V PP[P
NP]] VP[PP[NP P] V] VP[V PP[NP P]] VP[PP[P NP] V]
= 161 (41%) = 204 (52%) = 18 (5%) = 6 (2%)
(13a) + (13b) = 365/389 (94%) The adjacency of V and P in (13a) and (13b) guarantees the smallest possible string of words for the recognition and “construction” (Kimball, 1973) of VP and its two immediate constituents, namely V and PP. Hence, adjacency provides a minimal integration domain for the construction of VP and its daughters and makes the least possible demands on WM. The (c) and (d) structures make significantly more demands and are relatively rare by comparison (just 6 percent of grammars). We can capture Frazier’s (1985) intuition about the Greenbergian universals in an empirically more adequate way, therefore, by shifting the focus from capacity constraints to quantifiable differences in cross-linguistic distributions for different grammars. The phrasal orderings that make minimal demands on WM and whose processing is faster and more efficient are well attested and highly frequent across languages; those that make greater demands are quite rare (see the principle of Minimize Domains in Section 13.4). This reasoning was extended in Hawkins (1994) to typological patterns for which no grammatical principle had been proposed hitherto, or has been since. For example, there are certain hierarchies of center-embedded phrases that can now be explained from a WM perspective. In the environment PP[P NP[__ N]], with a head-initial preposition in PP and a final noun in NP, we have the distribution of basic NP-internal orderings shown in (14)
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
(for prepositional phrases corresponding to in yellow books vs. in books yellow {Adj, N}, in my father’s books vs. in books of my father {Possp, N}, and in I read which/that books vs. in books which/that I read {Rel, N}), cf. Hawkins (1983 p. 75, 2004 p. 128): (14)
Prepositional languages:
AdjN PosspN RelN
32% 12% 1%
NAdj NPossp NRel
68% 88% 99%
As the aggregate weight and complexity of nominal modifiers increases (relative clauses exceeding possessive phrases which in turn exceed singleword adjectives) so the distance between P (which constructs PP) and N (which constructs NP) increases in the prenominal order, adding to simultaneous WM demands in the processing of the PP and its immediate constituents. Even for single-word adjectives, prepositional languages prefer NAdj with an adjacent P and N over AdjN by 2–1 (68 percent to 32 percent). But as the distance between P and N grows larger in PosspN and RelN structures, and WM demands increase accordingly, the crosslinguistic preference for NPossp and NRel rises higher, to 88 percent and 99 percent respectively. For filler-gap dependencies I summarized the grammatical cut-off points for relative clauses on Keenan and Comrie’s Accessibility Hierarchy in footnote 3. A more detailed tabulation is given in Hawkins (2004, p. 189) where it is shown that the grammaticality of the gap strategy decreases precisely across grammars down the SU, DO, IO/OBL, and GEN positions on which relative clauses are formed (by 100 percent to 65 percent to 25 percent to 4 percent, respectively). Meanwhile the use of resumptive pronouns in place of gaps increases down the hierarchy (0 percent to 35 percent to 75 percent to 96 percent, respectively). It is argued (Hawkins, 2004, 177–190) that the Accessibility Hierarchy is a complexity ranking requiring more items and structural relations to be held and processed in WM down the hierarchy. The gradual shift in grammaticality from gaps, as in the professori that I met Oi for lunch, to pronouns, the professori that I met heri for lunch, makes processing easier by providing an explicit anaphor in the position relativized on and permitting a minimal integration domain for grammatical co-occurrence relations between the subcategorizing verb (met) and this position relativized on. This relieves simultaneous processing in WM. As these demands increase, so gaps are systematically replaced by pronouns down the Accessibility Hierarchy, in languages in which both are grammatical. Once again, more versus less WM load can be linked to fewer versus more grammars of the relevant types, and WM can provide an empirically well-supported explanation for the crosslinguistic distributions. The more complex gap environments of embedded clauses, on which the whole Subjacency literature has focused, can also be explained in a
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
287
288
JOHN A. HAWKINS
more empirically adequate way along these lines. Hawkins (1999, 2004 ch. 7) argued that Subjacency as a grammatical constraint is both too strong and too weak (even when attempts are made to “relativize” it to the different pairs of bounding nodes in Rizzi, 1982). It is too strong because several grammars (like Akan) do not have it at all. It is too weak because there are numerous grammaticality distinctions in environments permitted by Subjacency that it does not account for. For example, Russian disallows gaps in finite subordinate clauses like (15a) (from Comrie, 1973, p. 297), which are grammatical in English (see the translation), while gaps in the corresponding infinitival clauses of Russian are grammatical (15b). The surface structure of (15b) is argued in Hawkins (2004, pp. 193–197) to involve a smaller domain for filler-gap processing and fewer simultaneous demands in WM. More generally, Hawkins (2004) proposes a complexity hierarchy for embedded gaps, predicting cut-off points in grammars in proportion to their WM demands. Structures violating Subjacency are simply at the bottom of this hierarchy, and some languages do go all the way down and permit relative clauses in these hard-to-process environments, just as they do in the Accessibility Hierarchy.6 (15) a. *Vot NP[ogurcyi S[kotoryei ja obeščal here are cucumbers which I promised b. Vot obeščal NP[ogurcyi S[kotoryei ja here are cucumbers which I promised
S[čto
prinesu Oi]]] that I’d bring VP[prinesti
Oi]]]
to-bring
13.3.2 Issues for grammars and for processing Processing explanations for typological distributions like these raise similar issues for grammars to those we discussed in Section 13.1. Are the grammars actually different in some of these typological samples, and if so, how, or are we just seeing preferred and more frequent structures of the relevant types occurring more often in performance in the relevant languages, in accordance perhaps with the processing preferences of WM theories? For example, are the “basic” word orders of Greenberg’s (1963) and later typological samples actually conventionalized in grammatical rules of linear ordering (as they are in the fixed SVO of English), or are they simply the most frequent clausal orders for the relevant head-initial or head-final phrases (PP, NP, AdjP, etc)? Dryer (2005b) gives a pertinent discussion of this issue, while Hawkins (1994, 2004, 2015) points to quantitative differences in corpora that can reveal the presence of a grammaticalized convention, just as Saah and Goodluck (1995) saw evidence for conventionalized versus nonconventionalized Subjacency in their English versus Akan acceptability data. It is once again the Performance-Grammar Correspondence Hypothesis (1) that reveals why disentangling grammars from performance preferences is not straightforward. If a grammatical rule or constraint does exist, it will (by hypothesis) have responded to the preferred performance pattern. If it
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
does not, performance preferences will coexist with minority structures that are not ruled out by the grammar. Relevant data are required from more languages to sort out which is which (see in this connection the interesting recent paper of Futrell et al., 2020). When there is evidence for an actual convention in the grammar, of constituent ordering or filler-gap linking, we must then ask if this convention is best described by a traditional generative rule, or as an Optimality Theory constraint of some kind (Bresnan et al., 2001; Featherston, 2008; Prince & Smolensky, 1993), or in altogether different terms as in more usage-based theories (Bybee 2010), see Francis (2022) for a thorough review of the options (and footnote 2). For processing, meanwhile, the correspondences between more versus fewer items in WM and fewer versus more grammars raise general issues of a psycholinguistic nature. For even if WM load is increased in what are generally less frequent structures, does it necessarily follow that WM itself and any of its architectural properties is the actual cause of the increasing processing difficulty and of declining grammars? A growing consensus in the psycholinguistic literature is that WM per se is not necessarily the primary cause, and we have seen other considerations taking center stage, either in addition to or instead of WM. First, processing is known to be extremely fast. In the earliest set of data comparing performance preferences and grammars Hawkins (1994) argued that the correspondences were ultimately driven by the advantages of greater speed. This was reflected in the naming of his main principle, Early Immediate Constituents, which became Minimize Domains in Hawkins (2004, 2014). The emphasis shifted to more versus fewer items in integration domains and to WM load in this later work, but both ideas have been there all along, they are not incompatible, and both could be operative. Second, Lewis and Vasishth (2005) see reduced interference, and the avoidance of errors and garden paths, as one of the key benefits of smaller structural domains for processing. Third, many psycholinguists see frequency of occurrence as a, or the, major determinant of ease of processing (Bybee, 2010; Bybee & Hopper, 2001; MacDonald, 1999). There are strong correlations between greater frequency and the brevity of forms and smaller phrases (see Hawkins’ 2004 principle of Minimize Forms7, also Haspelmath, 2008, 2021; Zipf, 1949), and sometimes frequency seems to be the more fundamental and ultimate cause (see e.g. Diessel & Tomasello’s 2006 frequency-based account of Keenan & Comrie’s AH for relativization). Fourth, a growing body of work in psycholinguistics and computational linguistics now sees predictability and “surprisal” as the key determinant of processing ease and difficulty. A word or structure is easier to process in proportion to its predictability in context (Hale, 2001; Levy, 2008). An extension of this work (Dyer, 2017; Futrell, Gibson & Levy, 2020) now proposes that many of the word order universals summarized in Section 13.3.1 can be attributed to “information locality” and to the preference for keeping words and phrases together that predict one another.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
289
290
JOHN A. HAWKINS
Fifth, and finally in this context, we have recently seen the emergence of a general and integrative theory of “communicative efficiency” in psycholinguistics. This approach views languages as shaped by a trade-off between information transfer, ease of production and ease of comprehension under information processing constraints that are inherent to the human brain. A summary of its basic tenets with extensive supporting references can be found in Gibson et al. (2019). This approach provides a framework in which many traditional problems can be solved, including the relationship between the speaker’s needs and the hearer’s needs, between ease and complexity of processing and efficient information transfer, between prediction and integration in processing, and between performance and grammars. The kinds of efficiency considerations documented in the conventions of the world’s languages in Hawkins (2004, 2014) are directly relevant for this research program and have led to interesting experiments using multilingual corpora and treebanks (Futrell et al., 2015, Futrell, Levy & Gibson, 2020; Liu, 2008). The communicative efficiency model also provides a framework in which WM can be integrated with the other considerations that clearly impact ease or difficulty in processing. Their interaction can now be investigated more adequately, and the respective roles and strengths of each can be tested empirically. For example, Gibson’s (1998) explanation for the preference for Subject > Object over Object > Subject appealed to added WM load in the latter as a result of the object predicting the subject. But online predictions are supposed to be good for processing, according to Surprisal Theory (Levy, 2008), so shouldn’t this make the processing of these transitive clauses easier? And if not, why not? Performance data from German, Finnish, and also Japanese (see footnote 5) appear to show that the benefits of online prediction are outweighed by added WM load, as Gibson suggested. However, there are other relevant considerations, including the grammatical encoding and case marking of subjects and objects, their semantic roles as agents or patients, and their relative frequencies, and I am going to argue in Section 13.4 that cross-linguistic typology has an important role to play here. It can reveal patterns in the grammatical variation that are directly relevant for assessing the role of WM in a larger theory of efficiency and for teasing apart other co-operating and competing processing factors.
13.4
Integrated WM Models in Grammars
As mentioned above, the goal of Hawkins (1994, 2004, 2014) has been to examine grammars from an efficiency perspective and to test the Performance Grammar Correspondence Hypothesis (1). Three general
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
efficiency principles have been proposed, Minimize Domains, Minimize Forms, and Maximize Online Processing. The first and third of these are most relevant for WM and are defined in (16) and (17): (16) Minimize Domains (MiD) (Hawkins 2004, p. 31) The human processor prefers to minimize the connected sequences of linguistic forms and their conventionally associated syntactic and semantic properties in which relations of combination and/or dependency are processed. The degree of this preference is proportional to the number of relations whose domains can be minimized in competing sequences or structures, and to the extent of the minimization difference in each domain. MiD accounts for the word order patterns and filler-gap hierarchies summarized in Section 13.3.1. (17) Maximize Online Processing (MaOP) (Hawkins, 2004 p. 51) The human processor prefers to maximize the set of properties that are assignable to each item X as X is processed, thereby increasing O(nline) P(roperty) to U(ltimate) P(roperty) ratios. The maximization difference between competing orders and structures will be a function of the number of properties that are unassigned or misassigned to X in a structure/sequence S, compared with the number in an alternative. This principle accounts for several grammatical patterns that have not yet received the attention they deserve in psycholinguistics. They include the asymmetrical ordering preference of WH fillers before (rather than after) their gaps (see, e.g., [4a] above), head nouns before relative clauses even in otherwise head-final languages (see Section 13.2), topics before predications, wide scope quantifiers and operators before narrow scope items, and antecedent NPs before their anaphors (see Hawkins, 2004, ch. 8; Hawkins, 2014, ch. 2.3). These principles and the data they describe are ultimately explainable by the different causal factors that psycholinguistic models are now trying to disentangle: WM load (cf. Minimize Domains), speed, and earliness in the delivery of linguistic properties (cf. Minimize Domains and Maximize Online Processing), avoidance of online errors and garden paths (cf. Maximize Online Processing), and frequency and predictability (cf. Minimize Forms in footnote 7). The interaction between them, their respective strengths and weaknesses and their competition and cooperation, can actually be seen in quantitative data from grammars (Hawkins, 2014, ch. 9). I now present a case study showing how WM can be integrated with these other factors to predict degrees of preference (and dispreference) for grammars, and for corresponding performance data as well.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
291
292
JOHN A. HAWKINS
The two languages on which Gibson (1998) based his WM explanation for preferred SO ordering, German (cf. [11]) and Finnish, are both rich case marking languages (Tallerman, 1998) with Nominative and Accusative case morphology, of the kind found in the familiar Indo-European languages of Western Europe. A very different pattern of case marking is found in languages with Ergative-Absolutive marking, however, as seen in Avar (North-East Caucasian); compare Comrie (1978): (19)
a. Vas-as: jas j-ec:ula. boy-ERG girl-ABS SG.FEM.ABS-praise ‘The boy praises the girl.’ b. Jas j-ekerula. girl-ABS SG.FEM.ABS-run ‘The girl runs.’
The Agent subject receives distinctive (Ergative) coding (-as) in the transitive (19a) while the Patient object has (zero-marked) Absolutive case, like that of the intransitive subject in (19b). The significance of this coding in the present context is that it reverses the preference that Gibson’s WM explanation defines, because the Ergative is now the predicting category. Ergatives are unique to transitive clauses, and they predict the co-occurrence of an Absolutive. So when Ergative precedes Absolutive, as in (19a), it will add to WM load. The reverse Absolutive > Ergative should therefore be preferred, just as Nominative before a predicting Accusative is preferred in German and Finnish. Empirically, the majority of Ergative-Absolutive grammars go against this prediction, however. They prefer Ergative > Absolutive, as in Avar (Primus, 1999), and these languages fall within the 96 percent of grammars in Tomlin’s (1986) sample that he classifies as having subject before object. But interestingly for Gibson’s theory, a significant minority of Ergative-Absolutive languages do exemplify an Absolutive > Ergative preference, as seen in (20a) from the Australian language Dyirbal (Comrie, 1989, p.106; Dixon, 1972). (20)
a. Balan dyugumbil CLASSIF-ABS woman-ABS ‘The man hit the woman.’
baŋgul CLASSIF-ERG
b. Balan dyugumbil CLASSIF-ABS woman-ABS ‘The woman came here.’
baninyu came-here
yaraŋgu _ man-ERG
balgan hit
Grammars with Ergative-Absolutive morphology like Dyirbal comprise most of the languages that have been classified as Object > Subject hitherto in the typological literature (Tomlin’s 4 percent), according to Primus (1999).8
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
293
Table 13.1 Subject > Object and Object > Subject for full NPs in grammars (cf. Comrie, 2013) Order
Distribution
SNom > OAcc
44/50 (88%) (29 SOV, 3 VSO, 1 VOS, 9 SVO, 1 SVO & VSO, 1SVO & SOV) 2/50 (4%) (1 VOS, 1 OVS) 4/50 (8%) 2/52 18/27 (67%) (16 SOV, 1 VSO, 1 SVO) 2/27 (7%) (2 VOS) 7/27 (26%) 5/32
OAcc > SNom No dominant order (both SNom > OAcc and OAcc > SNom) No info given for transitive clauses with SNom and OAcc SErg > OAbs OAbs > SErg No dominant order (both SErg > OAbs and OAbs > SErg) No info given for transitive clauses with SErg and OAbs
We can use data from WALS (Comrie, 2013) in order to quantify the different morphological types here and link them to processing. Comrie provides case marking data for full noun phrases in 190 grammars, of which 52 have a Nominative-Accusative system like German and Finnish, and 32 have Ergative-Absolutive.9 In the former 52 there are 2 for which no information is given on the ordering of subject and object in transitive clauses, leaving 50. Of these a full 44 (88 percent) are classified as having basic Subject > Object order, that is, Nominative before Accusative, and just 2 (4 percent) as Object > Subject. For 4 (8 percent) there is no dominant order, that is, both SO and OS are productive in these languages. For the 32 Ergative-Absolutive grammars no information is given about the relative ordering of subject and object in 5. Among the remaining 27, the Ergative subject precedes the Absolutive object as the basic order in 18 (67 percent), 2 have object before subject (7 percent), and there is no dominant order in 7 (26 percent). In other words, a majority (67 percent) of these grammars also have a basic Agent subject before Patient object order, but a significantly smaller majority than the 88 percent of Nominative-Accusative languages. And whereas 33 percent (9/27) of Ergative-Absolutive languages have either basic OS or no dominant order (i.e., both SO/OS), just 12 percent of NominativeAccusative grammars have these two possibilities (6/50). These data are summarized in Table 13.1, together with further details about basic verb positions. Let us review the processing factors that are relevant for these orderings of subjects and objects. First, depending on the morphology, one order can be good for online prediction, which we can abbreviate as [+P], since the second argument of a transitive clause is predicted by the first, while the reverse order makes no such prediction from the first argument to the second, that is [ P]. Second, we then have (according to Gibson) reverse benefits and disadvantages for WM. If the first argument (e.g., Accusative) does predict the second (Nominative), this will increase the number of
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
294
JOHN A. HAWKINS
items in, WM, that is [ WM], whereas if the first (e.g., Nominative) does not predict the second (Accusative), this will be good for, and not add to WM, that is [+WM]. In other words, [+WM] co-occurs with the absence of a prediction, [ P], and [ WM] with its presence, [+P]. A third factor is frequency. Nominative is considerably more frequent than Accusative, and Absolutive more than Ergative, in their respective language types (Primus, 1999), since Nominative and Absolutive occur in both intransitive and transitive clauses, Accusative and Ergative only in transitive ones. More frequent items are more accessible in production (Levelt, 1989) and easier to comprehend (MacDonald, 1999). By the logic of De Smedt (1994), MacDonald (2013), and Wasow (2013) this should result in an “Easy First” ordering preference and earlier positioning, that is, [+F] for the more frequent argument in initial position in the transitive clause versus [ F] for the less frequent argument initially. Fourth, Agent is preferred before Patient [+AgPat], rather than the reverse Patient before Agent, that is, [ AgPat], because of the asymmetrical semantic dependency between them, which means that online access to the Agent is preferred first, as with other asymmetric dependencies (cf. Hawkins, 2004, ch. 8; Primus, 1999). Fifth, Animate entities are more accessible in production than inanimates, and in many (e.g. Bantu) grammars an Animate before Inanimate [+AnIn] order has been grammaticalized, as opposed to the reverse [ AnIn] (Comrie, 1989; Branigan et al., 2008; Tomlin, 1986). Transitive clauses generally have an animate subject, and an object that is regularly inanimate (or else a second animate argument), so the SO order is only rarely [ AnIn] and regularly [+AnIn]; OS orders, by contrast, are regularly [ AnIn]. Sixth, the structural configurations in which subjects (or VP-external NPs) precede objects (VP-internal NPs) have been calculated in Hawkins (2004, pp. 228–235) to involve smaller domains of phrase structure processing and/or earlier access to syntactic properties online, i.e. greater efficiency in syntactic processing for the structural configuration of a transitive clause, [+SC] versus [ SC]. In Table 13.2 I show how these six factors combine. The predictions we can make are complicated by the fact that (a) they have different relative strengths and (b) some of them are gradient and can apply to different degrees (see e.g., the relative weight effects of Hawkins, 1994 in relation to different degrees of WM load; cf. also Stallings & MacDonald, 2011). Nonetheless, enumerating these different factors and their plus or minus values can serve as a first approximation for the kind of multifactor integrated approach to processing and grammar that we now need. A comparison of Tables 13.1 and 13.2 shows a good match between the grammatical data and processing preferences. Five out of six processing factors predict a strong preference for Nominative subjects before Accusative objects, and this is reflected in the 88 percent of grammars that have basic SO, with just a small minority of 12 percent having dominant or
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
295
Table 13.2 Processing predictions for Subject > Object and Object > Subject Orders Order
WM
SNom > OAcc OAcc > SNom SErg > OAbs OAbs > SErg
+
P + +
+
F
AgPat
AnIn
SC
Prediction
+ -
+
+
+
+
+ +
5/6 1/6 4/6 2/6
+
Note: [WM] = good or bad for WM (i.e., more versus fewer items that need to be held simultaneously in WM when processing a transitive clause); [P] = good or bad for online prediction of the second argument in a transitive clause by the first; [F] = good or bad for the more frequent case preceding the less frequent one; [AgPat] = good or bad for the preferred Agent before Patient (per MaOP); [AnIn] = good or bad for the more animate argument preceding the less animate; [SC] = good or bad for the efficiency of phrase structure processing resulting from minimal phrasal combination domains (per MiD in Example [16]) and early online property assignments (per MaOP in Example (17) (cf. Hawkins, 2004, pp. 228–235).
nondominant OS. For Ergative-Absolutive grammars there are fewer, but still a majority of processing factors, 4/6, that predict Ergative subjects before Absolutive objects, and there are indeed now fewer grammars, but still a majority (67 percent), that have basic SO, and a larger minority (33 percent) that have dominant or nondominant OS. This multifactor approach exemplifies what I believe must now be done in future studies of cross-linguistic variation across grammars. A focus on WM alone is not sufficient, since WM interacts with, and is sometimes opposed by, other factors. A similar tension between WM and other factors can be seen in the cross-linguistic distribution of VO and OV languages, and more generally in head-initial and head-final grammars (Haspelmath et al., 2005; Hawkins, 1983). Verbs and heads are more predictive of their complements, that is, they are better for making predictions about their complements, than complements are of their heads (cf. Engelhardt et al., in prep. for experimental confirmation of the greater predictiveness of verbs for their complements rather than vice versa in a comparison of SVO English and SOV Japanese). VO languages are, therefore, better for online prediction of the object from the verb, [+P], than OV languages are for predicting the verb from the object, [ P]. By Gibson’s (1998) logic, we then have the reverse costs and benefits for online WM load, that is, added WM or [ WM] for VO languages and less WM load for OV languages, [+WM]. In terms of the additional processing factors that are relevant here, Hawkins (2014, ch. 7) points to numerous symmetrical dependencies between verbs and their objects that will lead to complementary advantages and disadvantages for online property assignments in each order, VO and OV (by Maximize Online Processing [17]). Overall it is argued in Hawkins (2014) and Engelhardt et al. (in press) that head-initial and head-final grammars will have roughly equal efficiencies and should be roughly equally distributed across the globe, which they are.10
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
296
JOHN A. HAWKINS
13.5
Conclusion
I have shown in this chapter that there are strong correlations between preferences in the variation data of performance in languages permitting variation and in the fixed conventions of grammars that allow fewer variants. These correlations make it plausible to consider processing explanations for grammars in terms of WM (Section 13.1). Conversely, grammars provide additional data of relevance for processing models. Explanations for grammatical universals in terms of proposed capacity constraints in WM have not worked well, however, since too many grammars provide counterexamples (Section 13.2). There have been interesting discussions throughout the psycholinguistic literature of particular languages, mainly Standard Average European ones (Haspelmath, 2001), in relation to various capacity constraints that have been proposed. But these are insufficient to establish whether grammars in general are similarly constrained, and the evidence of Section 13.2 is that many are not. A focus on more versus less WM load (in Section 13.3) has resulted in more successful predictions in the form of correlations with infrequent versus frequent grammars and declining typological distributions. Even greater empirical adequacy, however, in both performance and grammars, comes from integrating WM within a multifactor processing theory (Section 13.4) such as the communicative efficiency model of Gibson et al. (2019). We suggest that future work should be of this integrative nature. Current databases can be mined in order to shed light on the many processing factors that impact both performance preferences and grammars and so reveal their relative strengths, their competitions and their cooperation (cf. Hawkins, 2014, ch. 9), as argued in Section 13.4. An illustration of this methodology was provided for the relative ordering of subjects and objects. A further study was outlined involving a similar competition and cooperation between processing factors resulting in the roughly equal distribution of verb-object (VO) and objectverb (OV) languages across the globe.
References Allwood, J. (1982). The complex NP constraint in Swedish. In E. Engdahl & E. Ejerhed (Eds.) Readings on unbounded dependencies in Scandinavian languages (pp. 15–32). Almqvist & Wiksell. Anderson, J. R. (1983). The architecture of cognition. Harvard University Press. Ariel, M. (1999). Cognitive universals and linguistic conventions: The case of resumptive pronouns. Studies in Language, 23, 217–269. Baddely, A. (this volume). Working memory and the challenge of language. Berwick, R. C., & Weinberg, A. S. (1984). The grammatical basis of linguistic performance: Language use and acquisition. MIT Press.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
Branigan, H. P., Pickering, M. J., & Tanaka, M. (2008). Contributions of animacy to grammatical function assignment and word order during production. Lingua, 118, 172–189. Bresnan, J., Dingare, S., & Manning, C. D. (2001). Soft constraints mirror hard constraints: Voice and person in English and Lummi. In M. Butt & T. H. King (Eds.), Proceedings of the LFG 01 Conference (pp. 13–32). CSLI Publications. Bybee, J. L. (2010). Language, usage and cognition. Cambridge University Press. Bybee, J. L., & Hopper, P. (Eds.). 2001. Frequency and the emergence of linguistic structure. John Benjamins. Chomsky, N. (1965). Aspects of the theory of syntax. MIT Press. Chomsky, N. (1981). Lectures on government and binding. Foris. Chomsky, N. (2005). Three factors in language design. Linguistic Inquiry, 36, 1–22. Comrie, B. (1973). Clause structure and movement constraints in Russian. In C. Corum, T. C. Smith-Stark, & A. Weiser, (Eds.), You take the high node and I’ll take the low node (pp. 291–304). Chicago Linguistic Society, Comrie, B. (1978). Ergativity. In W. Lehmann (Ed.), Syntactic typology: Studies in the phenomenology of language. University of Texas Press. Comrie, B. (1989). Language universals and linguistic typology (2nd ed.). University of Chicago Press. Comrie, B. (1998). Rethinking the typology of relative clauses. Language Design, 1, 59–86. Comrie, B. (2013). Alignment of case marking of full noun phrases. In M. Dryer, & M. Haspelmath (Eds.), The world atlas of language structures (pp.366–369) online. Max Planck Institute for Evolutionary Anthropology. http://wals.info/chapter/98 Cowan, N. (2005). Working memory capacity. Psychology Press. De Smedt, K. J. M. J. (1994). Paralellism in incremental sentence generation. In G. Adriaens & U. Hahn (Eds.), Parallelism in natural language processing. Ablex. Diessel, H., & Tomasello, M. (2006). A new look at the acquisition of relative clauses. Language, 81, 882–906. Dixon, R. M. W. (1972). The Dyirbal language of North Queensland. Cambridge University Press. Dryer, M. S. (1989). Large linguistic areas and language sampling. Studies in Language 13, 257–292. Dryer, M. S. (1992). The Greenbergian word order correlations. Language, 68, 81–138. Dryer, M. S. (2005a). Order of relative clause and noun. In M. Haspelmath, M. Dryer, D. Gil, & B. Comrie (Eds.), The world atlas of structures (pp. 366–369). Oxford University Press. Dryer, M. S . (2005b). Relationship between the order of object and verb and the order of relative clause and noun. In M. Haspelmath, M. Dryer, D. Gil,
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
297
298
JOHN A. HAWKINS
& B. Comrie (Eds.), The world atlas of structures (pp. 390–393). Oxford: Oxford University Press. Dryer, M. S. (2013). Order of subject, object and verb. In M. Dryer, & M. Haspelmath (Eds.), The world atlas of language structures online. Max Planck Institute for Evolutionary Anthropology. http://wals.info/chapter/81. Dyer, W. E. (2017). Minimizing integration cost: A general theory of constituent order (Ph.D. dissertation, University of California Davis). Engdahl, E., & Ejerhed, E. (Eds.). (1982). Readings on unbounded dependencies in Scandinavian languages. Almqvist & Wiksell. Engelhardt, P., Filipovic, L., & Hawkins, J. A. (In prep.) Verbs and arguments as predictors in SVO and SOV languages: Processing and typological considerations (MS, University of East Anglia & University of California Davis). Featherston, S. (2008). The Decathlon model of empirical syntax. In S. Kepser & M. Reis (Eds.), Linguistic evidence: Empirical, theoretical and computational perspectives (pp. 187–208). de Gruyter. Francis, E. (2022). Gradient acceptability and linguistic theory. Oxford University Press. Francis, E., Lam, C., Zheng, C. C., Hitz, J., & Matthews, S. (2015). Resumptive pronouns, structural complexity, and the elusive distinction between grammar and performance: Evidence from Cantonese. Lingua, 162, 56–81. Frazier, L. (1979). Parsing and constraints on word order. University of Massachusetts Occasional Papers in Linguistics, 5, 177–198. Frazier, L. (1985). Syntactic complexity. In D. Dowty, L. Karttunen, & A. Zwicky (Eds.), Natural language parsing: Psychological, computational, and theoretical perspectives. Cambridge University Press. Frazier, L., & Fodor, J. D. (1978). The sausage machine: A new two-stage parsing model. Cognition, 6, 291–326. Futrell, R., Gibson, E., & Levy, R. P. (2020). Lossy-context surprisal: An information-theoretic model of memory effects in sentence processing. Cognitive Science, 44, 1–54. Futrell, R., Levy, R. P., & Gibson, E. (2020). Dependency locality as an explanatory principle for word order. Language, 96(2), 371–412. Futrell, R., Mahowald, K., & Gibson, E. (2015). Large-scale evidence of dependency length minimization in 37 languages. Proceedings of the National Academy of Sciences, 112(33), 10336–10341. Gibson, E. (1998). Linguistic complexity: Locality of syntactic dependencies. Cognition, 68, 1–76. Gibson, E., Futrell, R., Piantadosi, S. T., Dautriche, I., Mahowald, K., Bergen, L., & Levy, R. P. (2019). How efficiency shapes human language. Trends in Cognitive Sciences, 23(5), 389–407. Greenberg, J. H. (1963). Some universals of grammar with particular reference to the order of meaningful elements. In J. H. Greenberg (Ed.), Universals of language (pp. 73–113). MIT Press,. Greenberg, J. H. (1966). Language universals, with special reference to feature hierarchies. Mouton.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
Greenberg, J. H. (1995). The diachronic typological approach to language. In M. Shibatani & T. Bynon (Eds.), Approaches to language typology (pp. 143–166). Clarendon Press. Haig, J. H. (1996). Subjacency and Japanese grammar: A functional account. Studies in Language, 20, 53–92. Hale, J. (2001). A probabilistic early parser as a psycholinguistic model. In Proceedings of NAACL, 2001 (pp. 1–8). Association for Computational Linguistics. Haspelmath, M. (1999). Optimality and diachronic adaptation. Zeitschrift für Sprachwissenschaft, 18, 180–205. Haspelmath, M. (2001). The European linguistic area: Standard average European. In M. Haspelmath, E. König, W. Oesterreicher, & W. Raible (Eds.), Language typology and language universals: An international handbook (pp. 1492–1510). Walter de Gruyter, Haspelmath, M. (2008). Creating economical morphosyntactic patterns in language change. In J. Good (Ed.), Language universals and language change (pp. 185–214). Oxford University Press. Haspelmath, M. (2021). Explaining grammatical coding asymmetries: Formfrequency correspondences and predictability. Journal of Linguistics, 57(3), 605–633. Haspelmath, M., Dryer, M. S., Gil, D., & Comrie, B. (Eds.). (2005). The world atlas of language structures. Oxford University Press. Hawkins, J. A. (1983). Word order universals. Academic Press. Hawkins, J. A. (1994). A performance theory of order and constituency. Cambridge University Press. Hawkins, J. A. (1999). Processing complexity and filler-gap dependencies. Language, 75, 244–285. Hawkins, J. A. (2004). Efficiency and complexity in grammars. Oxford University Press. Hawkins, J. A. (2007). Processing typology and why psychologists need to know about it. New Ideas in Psychology, 25, 87–107. Hawkins, J. A. (2014). Cross-linguistic variation and efficiency. Oxford University Press. Hawkins, J. A. (2015). Typological variation and efficient processing. In B. MacWhinney & W. O’Grady (Eds.), The handbook of language emergence (pp. 215–236). Wiley-Blackwell,. Hofmeister, P., Jaeger, T. F., Arnon, I., Sag. I. A., & Snider, N. (2013). The source ambiguity problem: Distinguishing the effects of grammar and processing on acceptability judgments. Language and Cognitive Processes, 28 (1–2), 48–87. Hofmeister, P., & Sag, I. A. (2010). Cognitive constraints and island effects. Language, 86(2), 366–415. Jaeger, T. F., & Norcliffe, E. (2009). The cross-linguistic study of sentence production: State of the art and a call for action. Language and Linguistics Compass, 3(4), 866–887.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
299
300
JOHN A. HAWKINS
Keenan, E. L., & Comrie, B. (1977). Noun phrase accessibility and universal grammar. Linguistic Inquiry, 8, 63–99. Kimball, J. (1973). Seven principles of surface structure parsing in natural language. Cognition, 2, 15–47. Kluender, R., & Kutas, M. (1993). Subjacency as a processing phenomenon. Language and Cognitive Processes, 8, 573–633. Kuno, S. (1973). The structure of the Japanese language. MIT Press. Kwon, N., Gordon, P. C., Lee, Y., Kluender, R., & Polinsky, M. (2010). Cognitive and linguistic factors affecting subject/object asymmetry: An eye-tracking study of prenominal relative clauses in Korean. Language, 86, 546–582. Lehmann, C. (1984). Der Relativsatz. Gunter Narr Verlag. Levelt, W. J. M. (1989). Speaking: From intention to articulation. MIT Press. Levy, R. (2008). Expectation-based syntactic comprehension. Cognition, 106, 1126–1177. Lewis, R., & Vasishth, S. (2005). An activation-based model of sentence processing as skilled memory retrieval. Cognitive Science, 29, 375–419. Liu, H. (2008). Dependency distance as a metric of language comprehension difficulty. Journal of Cognitive Science, 9(2), 159–191. Lu, B., & Wen, Z. (this volume). Working memory and the language device. Lu, Q., Xu, C. & Liu, H. (2016). Can chunking reduce syntactic complexity of natural languages? Complexity, 21, 33–41. MacDonald, M. C. (1999). Distributional information in language comprehension, production and acquisition: Three puzzles and a moral. In B. MacWhinney (Ed.), The emergence of language. Lawrence Erlbaum Associates. MacDonald, M. C. (2013). How language production shapes language form and comprehension. Frontiers in Psychology, 4, 226. Matsumoto, Y. (1997). Noun-modifying constructions in Japanese: A frame semantic approach. Benjamins. Matthews, S., & Yip, V. (2003). Relative clauses in early bilingual development: Transfer and universals. In A. G. Ramat (Ed.), Typology and second language acquisition. Mouton de Gruyter. Maxwell, D. N. (1979). Strategies of relativation and NP accessibility. Language, 55, 352–371. Miller, G. (1956). The magical number seven plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Norcliffe, E., Harris, A. C., & Jaeger, T. F. (2015). Cross-linguistic psycholinguistics and its critical role in theory development: Early beginnings and recent advances. Language, Cognition and Neuroscience, 30(9), 1009–1032. O’Grady, W. (2005). Syntactic carpentry: An emergentist approach to syntax. Lawrence Erlbaum Associates.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
O’Grady, W. (2012). Three factors in the design and acquisition of language. Wiley Interdisciplinary Reviews: Cognitive Science, 3, 493–499. O’Grady, W. (2017). Working memory and language: From phonology to grammar. Applied Psycholinguistics, 38(6), 1340–1343. O’Grady, W. (this volume). Working memory and natural syntax. Phillips, C. (2013a). On the nature of island constraints I: Language processing and reductionist accounts. In J. Sprouse & N. Hornstein (Eds.) Experimental syntax and island effects (pp. 64–108). Cambridge University Press. Phillips, C. (2013b). Some arguments and nonarguments for reductionist accounts of syntactic phenomena. Language and Cognitive Processes, 28 (1– 2), 156–187. Primus, B. (1999). Cases and thematic roles: Ergative, accusative and active. Niemeyer. Prince, A., & Smolensky, P. (1993). Optimality theory: Constraint interaction in generative grammar. MIT Press. Rizzi, L. (1982). Issues in Italian syntax. Dordrecht. Rizzi, L. (1991). Relativized minimality. MIT Press. Ross, J.R. (1967). Constraints on variables in syntax (Ph.D. dissertation, Massachusetts Institute of Technology). Saah, K. K., & Goodluck, H. (1995). Island effects in parsing and grammar: Evidence from Akan. Linguistic Review, 12, 381–409. Sprouse, J., & Hornstein, N. (Eds.). (2013). Experimental syntax and island effects. Cambridge University Press. Sprouse, J., Wagers, M. & Phillips, C. (2012a). A test of the relation between working-memory capacity and syntactic island effects. Language, 88(1), 82–123. Sprouse, J., Wagers, M. & Phillips, C. (2012b). Working-memory capacity and island effects: A reminder of the issues and the facts. Language, 88(2), 401–407. Stallings, L. M., & MacDonald, M. C. (2011). It’s not just the “heavy NP”: Relative phrase length modulates the production of heavy-NP shift. Journal of Psycholinguistic Research, 40(3), 177–187. Tallerman, M. (1998). Understanding syntax. Arnold. Tomlin, R. S. (1986). Basic word order: Functional principles. Croom Helm. Trotzke, A., Bader, M. & Frazier, L. (2013). Third factors and the performance interface in language design. Biolinguistics, 7, 1–34. Ueno, M., & Polinsky, M. (2009). Does headedness affect processing? A new look at the VO-OV contrast. Journal of Linguistics, 45(3), 675–710. Wasow, T. (2002). Postverbal behavior. CSLI Publications. Wasow, T. (2013). The appeal of the PDC program. Frontiers in Psychology, 4, 236. Yamashita, H. (2002). Scrambled sentences in Japanese: Linguistic properties and motivation for production. Text, 22, 597–633. Zipf, G. K. (1949). Human behavior and the principle of least effort: An introduction to human ecology. Addison-Wesley.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
301
302
JOHN A. HAWKINS
Notes 1 Chomsky (1965) proposed that grammars are autonomous of, and cannot be formulated in terms of, factors that include WM limitations: “Acceptability. . .belongs to the study of performance. . .The unacceptable grammatical sentences often cannot be used for reasons having to do. . . with [memory limitations, intonational and stylistic factors,. . .and so on. . .[I]t would be quite impossible to characterize unacceptable sentences in grammatical terms. . .[W]e cannot formulate particular rules of the grammar in such a way as to exclude them” (pp. 11–12). 2 Exactly how the preferences of performance are claimed to have been conventionalized in grammars varies with different grammatical models. An excellent synopsis of the different approaches is given in Francis (2022, ch. 2), from some of the traditional generative proposals that I summarize in Section 13.2 to more usage-based theories (Bybee, 2010) in which frequencies and exemplars in mental representations play a central role. For a radical approach that dispenses with a grammar altogether that is separate from an efficient processor, see O’Grady (2005, this volume). 3 The cut-off points for relative clause formation down the Accessibility Hierarchy can be illustrated with the following languages from Keenan and Comrie (1977): SU only: Malagasy, Maori SU & DO only: Kinyarwanda, Indonesian SU & DO & IO/OBL only: Basque, North Frisian, Catalan SU & DO & IO/OBL & GEN: English, Hausa 4 Bresnan et al. (2001) illustrate the hard constraint/soft constraint distinction with a comparison between the Salish language Lummi and English with respect to the Person Hierarchy (1st, 2nd > 3rd). In English this hierarchy results in a (soft constraint) preference for the Passive “I was hit by the boy” over the Active “The boy hit me” when the direct object is a first person pronoun. In Lummi, sentences corresponding to this latter are ungrammatical and nonoccurring, on account of a hard constraint preventing subjects from containing a lower person on the hierarchy than their objects. 5 Interestingly, the performance frequency for OSV (versus SOV) in the Japanese corpus of Hawkins (1994, p. 145) was also 4–5% (see Yamashita, 2002, for similar statistics from Japanese). 6 The clausal embedding hierarchy for gaps given in Hawkins (2004, p. 194) is: infinitival (VP) complement > finite (S) complement > S within a complex NP 7 Hawkins’ (2004) principle of Minimize Forms (MiF) is defined as follows:
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
Have Grammars Been Shaped by Working Memory
The human processor prefers to minimize the formal complexity of each linguistic form F (its phoneme, morpheme, word, or phrasal units) and the number of forms with unique conventionalized property assignments, thereby assigning more properties to fewer forms. These minimizations apply in proportion to the ease with which a given property P can be assigned in processing to a given F.” (p. 38) MiF predicts reductions in both the surface complexity of forms and the number of unique form-property pairs, as in Greenberg’s (1966) markedness hierarchies, and greater ambiguity and polysemy, all in proportion to the ease of processing these minimal forms. 8 The Ergative-Absolutive languages discussed in Primus (1999) that have been classified hitherto as object before subject include the OSV languages Dyirbal, Hurrian, Siuslaw, Kabardian, Fasu, and the OVS Apalai, Arecuna, Bacairi, Macushi, Hianacoto, Hishkaryana, Panare, Wayana, Asurini, Oiampe, Teribe, Pari, Jur, Luo, and Mangarayi 9 A further 98 languages in Comrie’s (2013) sample have “neutral” marking on subject and object full noun phrases, as in English, which does not distinguish between them morphologically, while 8 have minor case marking types (tripartite and active/inactive). None of these types has an asymmetry between subjects and objects in the morphology relevant for online prediction and WM load, so they will not be considered here. 10 The proportions of basic word orders in the Expanded (Greenbergian) sample of Hawkins (1983), with 336 entries, were: 174 SOV (52%), 109 SVO (32%), 53 V1 (VSO or VOS), i.e. 52% OV to 48% VO. The most recent WALS sample of Dryer (2013), with 1,173 entries for languages with a dominant word order of one of these types, has: 565 SOV (48%), 488 SVO (42%), and 120 (10%) V1 (VSO or VOS), i.e. 48% OV to 52% VO.
https://doi.org/10.1017/9781108955638.017 Published online by Cambridge University Press
303
14 Branching and Working Memory A Cross-Linguistic Approach Federica Amici, Alejandro Sanchez-Amaro, and Trix Cacchione
14.1
Introduction
Experimental evidence suggests that different languages may foster specific habits of processing information, which are retained beyond the linguistic domain. Depending on the word order within sentences, for instance, realtime sentence comprehension may require a different allocation of attention, thus triggering processing habits that may also affect how humans process stimuli other than words. Below, we will first review previous studies on the link between word order, statistical learning habits, and attention allocation. Then, we will discuss experimental evidence supporting the hypothesis that branching habits affect working memory processes beyond the linguistic domain. Finally, we will conclude by fostering a stronger cross-linguistic approach to the study of branching and working memory, suggesting possible lines for future research.
14.2
Working Memory and the Focus of Attention
Working memory (WM) makes it possible to temporarily store information in memory in an easily accessible way, and is essential for a variety of cognitively complex activities, including language comprehension, production, and acquisition (Baddeley, 2003; Cowan, 2014). Several WM models emphasize the dynamic aspects of WM, describing it as a domain-general system controlling the focus of attention (Barrouillet & Camos, 2007; Engle &
We are sincerely grateful to John W. Schwieter and Zhizeng (Edward) Wen for having invited us to participate in this book, and to two anonymous reviewers, who made wonderful suggestions to improve a former version of the chapter. We would like to warmly thank Carla Sebastián-Enesco, Juan Salazar-Bonet, Matthias Allritz ,and Federico Rossano for their huge support during the work we coauthored in 2019, and to the Humboldt foundation, for awarding the first author a Research Fellowship for Postdoctoral Researchers (Humboldt ID number 1138999) during which data were collected.
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
Branching and Working Memory
Oransky, 1999). The embedded process model by Cowan, for example, conceives of WM as temporarily activated memory that is embedded in longterm memory, and includes the focus of attention (Cowan, 1995, 1999). The focus of attention is a limited resource when processing information, and can be controlled by both involuntary processes (like the automatic shift of attention produced by certain loud noises or lights) and voluntary executive control (Cowan, 1995, 1999). Similarly, the attention control model by Engle and Kane describes WM as a system of short-term storage, rehearsal processes to achieve and maintain memory activation, and executive attention, which keeps goal-relevant information activated in the face of interference and competition (Engle et al., 1999; Kane et al., 2007; Kane & Engle, 2003). In contrast to the short-term memory component of WM, which is defined as a nonstrategic passive store, the dynamic component of WM (i.e., executive attention) is thought to allow for the control of attention to maintain relevant information in an activated state (Engle, 2002). Despite important differences across theoretical models of WM, all of them highlight its essential role to encode, recall, and combine information from different sources, and to allow the incremental processing of sounds, words, and sentences, which are temporarily activated and maintained in memory until they have been processed (see Caplan & Waters, 2013; Just & Carpenter, 1992; Lewis et al., 2006).
14.3
Working Memory and Processing Demands: The Case of Branching
The computational capacity of WM is usually considered a limited resource (Cowan, 2001; Miller, 1956), and the amount of information that can be maintained in WM finite (e.g., Chen et al., 2005; Gibson, 1998; see Christiansen & Chater, 2016). Therefore, sentence production and comprehension might face computational resource constraints when WM limits are exceeded (e.g., Gibson, 1998; Gibson & Pearlmutter, 1998; Just & Carpenter, 1992; Pearlmutter & MacDonald, 1995). Traditionally, these limits have been interpreted as a general constraint to maintain more than a certain number of activated items in WM (e.g., Caplan & Waters, 1999; Carpenter & Just, 1988). In center-embedded sentences (such as the selfembedded sentence in example (1) by Frazier, 1985), the head of the subject phrase (i.e., “apartment”) is separated from the associated verb phrase (i.e., “was well decorated”) by a large number of verbal elements: (1)
The apartment [that the maid [who the service had sent over] was cleaning every week] was well decorated.
These sentences are usually considered very hard to process, because they imply that a large number of items should be maintained in WM in the face of decay and interference, until the final verbal phrase is processed and can be connected to the corresponding noun phrase (Frazier & Fodor, 1978;
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
305
306
FEDERICA AMICI ET AL.
Gibson, 1998; Lewis et al., 2006). Therefore, center embeddings would be generally hard to process by increasing the WM load and thus requiring higher computational resources (Gibson, 1998; Gibson & Thomas, 1999). Along the same line, researchers have traditionally considered specific word orders as entailing higher processing costs. Across languages, there is substantial variation in the position of the words within a sentence. Within languages, however, some pairs of elements tend to be consistently ordered in the same way, and there are correlations in how these pairs of elements are arranged in a sentence (i.e., “Greenbergian correlations”; see, e.g., Dryer, 1992; Greenberg, 1963; Hawkins, 2004). Verb-object (VO) languages, for instance, also tend to be prepositional, and the head of a phrase usually precedes its dependents (e.g., noun-genitive, noun-relative clause), creating parse trees that grow down and to the right (i.e., right-branching or RB languages). In contrast, OV languages are usually postpositional and headfinal, so that dependents may more likely precede the head (genitive-noun, relative clause–noun), creating parse trees that grow down and to the left (i.e., left-branching or LB languages). In typically RB languages, like English, speakers can therefore process information incrementally with a low risk of reanalysis, because heads are usually encountered before their dependents, and dependents seldom affect previous parsing decisions. In contrast, typical LB languages, like Japanese, have been traditionally considered harder to process: initial dependents might acquire an unambiguous meaning only when the final head has been processed, implying a higher risk of unassignments and misassignments, and/or the need to delay processing decisions until the head is parsed (Mazuka, 1998; cf. Hawkins 1994, 2004). To date, it is still debated how exactly LB and RB are processed, and how computational resources are allocated during this process. The traditional view that LB structures are harder to process, however, has largely been discredited. In natural conversation, all native speakers process their language quickly and effectively, regardless of their branching direction (e.g., Hawkins, 2004; Stivers et al., 2009). Indeed, both RB and LB structures are frequent across languages, and both VO and OV are highly productive. How is it possible that LB speakers can effectively process a final-head language that should impose high risks of unassignments and misassignments during processing? Researchers suggest that there are at least two nonmutually exclusive ways in which this might happen. First, the very existence of Greenbergian correlations would facilitate processing in both RB and LB languages. By consistently ordering words within a sentence, languages allow the human parser to stick to the same processing strategy and therefore reduce the processing difficulties associated with a mixture of RB and LB structures (e.g., Dryer, 1992, 2009; Greenberg, 1963). According to Hawkins (1994, 2004, 2014), moreover, Greenbergian correlations would decrease processing difficulties by minimizing the sequences of linguistic forms and properties that have to be processed in order to solve dependencies (i.e., elements that depend on each
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
Branching and Working Memory
other for the assignment of specific syntactic and semantic properties, like in head-dependent relationships). In particular, both RB and LB languages would enhance processing efficiency by increasing adjacency between the elements of these dependencies so that the ultimate syntactic and semantic representation can be accessed as early as possible. In prepositional languages like English, for instance, this may result in heads being consistently positioned to the left, and longer phrases moving to the end of the sentence, whereas in postpositional languages like Japanese, the opposite is the case: heads are consistently positioned to the right, and longer phrases might move to the beginning of the sentence (Hawkins, 2004, 2014). These processes might explain the very existence of Greenbergian correlations, and clarify why LB structures can also be highly efficient. However, despite minimizing processing costs through increased adjacency between elements of dependencies, LB languages are still linked to a higher risk of unassignments and misassignments (cf. Hawkins 2004, 2014). A second way to deal with this risk is therefore to rely on other strategies that efficiently resolve ambiguity during sentence processing. There are several strategies that can be used for this. Languages, for instance, can increase the occurrence of case-marking, so that the initial element of a dependency contains more syntactic information. This information can be used to partially reduce ambiguities before processing the final part of the dependency, thereby decreasing processing demands (see, e.g., Dryer 1992, 2002; Hawkins, 2004). In line with this, LB languages are much more likely to show case-marking than RB languages, and case-marking is also more likely to appear earlier in a clause, where it can be more quickly accessed during language processing (Hawkins, 2004). Moreover, it is possible to decrease the risk of unassignments and misassignments by using statistical information about the probabilities with which different syntactic structures and other linguistic material occur in the language (Garrod & Pickering, 2004; Hale, 2001; Jaeger, 2015; Levy, 2008; Nakatani & Gibson, 2010; Pickering & Garrod, 2007). Based on this “statistical knowledge,” it is possible to predict the structure of a sentence and reduce unassignments and misassignments, so that processing difficulties would only increase when the input fails to match expectations (Hale, 2001; Jaeger, 2015; Levy, 2008; Nakatani & Gibson, 2010; Pickering & Garrod, 2007). Crucially, reliance on statistical knowledge might be especially important in LB languages. A corpus analysis of a typically RB language (i.e., English) and a typically LB language (i.e., Korean), for instance, showed that word orders in English and Korean allowed forward probabilities, and thus more reliable predictions, only in the latter (Onnis & Thiessen, 2013). In their study, the authors used two corpora to measure how likely each word was to occur together with the preceding one (i.e., backward probability), or with the following one (i.e., forward probability). In English, for instance, two words like “in Sapporo” showed a higher backward probability, because “Sapporo” was more likely to be preceded by “in” than “in” was
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
307
308
FEDERICA AMICI ET AL.
to be followed by “Sapporo” (Onnis & Thiessen, 2013). In contrast, Korean followed the opposite pattern, as initial words could more reliably predict the following ones (i.e., forward probability), rather than the preceding ones (Onnis & Thiessen, 2013). Through increased forward probability, therefore, LB languages may use statistical information to create expectations about the following words and the syntactic structure of the sentence, significantly increasing processing efficiency. Experimental evidence from different research areas also converges in showing that RB and LB languages might deal with processing demands in different ways. In German, for example, increasing the distance between the verb and its dependents within a sentence does not necessarily increase processing costs. In some cases, sentences with a larger number of words before the verb (i.e., the head) may even be easier to process. In two selfpaced reading experiments, for instance, Konieczny (2000) provided participants with verb-final sentences such as those in (2). Despite the higher number of items between the main verb and its arguments, native German-speakers were quicker to read the verb in (2b), suggesting that they found the verb easier to process. (2a) Er hat den Abgeordneten begleitet, und. . . He has the delegate escorted, and. . . ‘‘He escorted the delegate, and. . .’’ (2b) Er hat den Abgeordneten an das große Rednerpult begleitet, und. . . He has the delegate to the big lectern escorted, and . . . ‘‘He escorted the delegate to the large lectern, and. . .’’ The author interpreted these results as evidence that in head-final sentences, heads can be largely predicted on the basis of stored information (and thus more quickly processed) through incremental integration of their arguments (Konieczny, 2000). This is in line with other studies showing that final verbs can be anticipated on the basis of the kinds and number of preceding arguments (Kamide et al., 2003; Konieczny, 1996; Lewis et al. 2006). In some cases, therefore, initial arguments can facilitate the anticipation of the final head, decreasing (rather than increasing) the cognitive load of sentence processing (Konieczny, 2000). Similar positive effects on processing times have also been reported for Hindi (Vasishth, 2003; Vasishth & Lewis, 2006). Verb-final structures are common in both languages, as the typical word order of Hindi sentences is OV, and this is also the most common word order of German subordinate sentences. Similar findings have been described in other cross-linguistic studies on center-embedded structures. In languages like English, or French, selfembedded sentences like that seen in (1) are so hard to process that most native speakers find them less acceptable or harder to process than their ungrammatical counterparts with no second verb phrase (see Christiansen & MacDonald, 2009; Frank et al., 2016; Frazier, 1985; Gibson & Thomas,
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
Branching and Working Memory
1999; Gimenes et al., 2009; Vasishth et al., 2010). Another example is shown in (4) by Gibson and Thomas (1999). (4)
The apartment [that the maid [who the service had sent over] was well decorated.
These findings have been traditionally interpreted as evidence for WM limits: as the WM system cannot simultaneously maintain three noun phrases active until the three verb phrases have been processed, the central verb-phrase prediction (which causes higher WM load) is lost, and its omission in the ungrammatical sentence would not only go unnoticed, but it would also decrease parsing difficulties (Gibson & Thomas, 1999). Crucially, however, this effect is not found when testing native German and Dutch speakers with analogous sentences (Frank et al., 2016; Vasishth et al., 2010). Indeed, both German and Dutch speakers are slower at processing the ungrammatical than the grammatical sentence, suggesting that there are no universal WM constraints when processing these sentences (Frank et al., 2016). Instead, cognitive adaptations to a specific word order have been argued to explain these results better. In particular, verb-final structures are more common in German and Dutch (Frank et al., 2016; Vasishth et al., 2010). Speakers of these languages are exposed to verb-final constructions more often, and they might therefore have a different perception about the “accessibility” of certain structures, as a result of their language-specific statistical patterns (Frank et al., 2016; see Hawkins, 2004). Thus, accessibility might be a function of habituation to a specific structure-schema. By learning to direct their attention toward specific structures, when they are more likely to happen, speakers might learn to differently allocate WM resources during language processing. This could increase the accessibility of these structures, in line with Hawkins’s (2004, 2014) perspective that word order across languages is linked to different processing demands.
14.4
Habits of Processing Information: Statistical Learning and the Allocation of Attention
Overall, cross-linguistic research suggests that humans are endowed with a powerful mechanism to extract the statistical properties of a language, and that they automatically and naturally use it during language acquisition. Experimental evidence, for instance, shows that language exposure is a crucial condition for infants, children, and even adults to acquire a language (e.g., Brent & Cartwright, 1996; Gomez & Gerken, 2000; Saffran, 2003; Saffran et al., 1996; Seidenberg, 1997). When exposed to a language, humans might not only acquire new words and grammatical rules, but they might also develop a strong sensitivity to certain statistical regularities (e.g., Brent & Cartwright, 1996; Gomez & Gerken, 2000; Saffran, 2003; Saffran et al., 1996; Seidenberg, 1997).
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
309
310
FEDERICA AMICI ET AL.
For some authors, humans have been argued to be especially sensitive to linguistic regularities in the way words are ordered. From early on, humans may develop a strong bias toward the branching direction of their native language (e.g., Mazuka, 1998; Mazuka & Lust, 1988), even generalizing it to other languages during second language acquisition (see Pienemann, 2005). Infants, for instance, can automatically detect statistical regularities in the language they listen to (e.g., segmenting fluent speech into words), even after short periods of exposure and even if receiving no rewards for that (Saffran et al., 1996). Moreover, previous knowledge of language-specific word order may affect the acquisition of subsequent information in a linguistic domain, both in infants and adults (Onnis & Thiessen, 2013; Thiessen et al., 2019). Onnis and Thiessen (2013), for instance, tested RB and LB adult speakers with a sequence learning task including linguistic and nonlinguistic stimuli. Depending on the branching of their language, participants developed different expectations about the novel linguistic stimuli, and different parsing preferences. These findings suggest that, through experience, humans may develop strong statistical learning habits that reflect the branching of their native language, and that these habits may in turn affect learning processes throughout their life, at least in the linguistic domain (Onnis & Thiessen, 2013). These statistical learning habits appear at the end of the first year of life, in response to extensive exposure to the predominant word order of the native language, and are likely adaptive to facilitate language acquisition (Thiessen et al., 2019). Such sensitivity to the typical word order of a language might therefore foster specific habits of language processing: humans would develop clear expectations about the position in which certain structures are more likely to occur, leading to a different allocation of WM resources.
14.5
Branching and Habits of Processing Information: Beyond the Linguistic Domain
Sensitivity to the branching of a language may be so pervasive as to also impact how humans process sequences of stimuli other than the words in a sentence. It is well known that extensive experience in specific domains (e.g. playing music, driving a taxi) may impact memory and attentional processes, and even create brain structural changes with effects across multiple domains (Draganski et al, 2006; Gaser & Schlaug, 2003; Maguire et al., 2000; Woollett & Maguire, 2011). Language experience should be no exception. In line with a weak interpretation of linguistic relativity, for instance, language may bias attention toward specific aspects of the world, fostering habits of processing information that might be generic and thus also extend to nonlinguistic domains (Boroditsky, 2001; Slobin, 1996). Clearly, there are different levels at which the interface between language and cognition can be detected (Hunt & Agnoli, 1991). For example, language
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
Branching and Working Memory
might have a semantic effect on thought, in that specific characteristics of languages may affect the way we conceptualize numbers (Casasanto, 2005; Pica et al., 2004; Spellke & Tsivkin, 2001), space (Haun et al., 2006; Levinson & Wilkins, 2006; Li & Gleitman, 2002), time (Boroditsky, 2001; Casasanto et al., 2004; Nuñez & Sweetser, 2006), color (Gilbert et al., 2006; Regier & Kay, 2009; Winawer et al., 2007) or even mental states (de Villiers, 2007; Pyers & Senghas, 2009). Moreover, syntax might affect the way in which we process events. Experimental evidence, for instance, has shown an effect of transitive/agentive and intransitive/nonagentive structures (e.g., “he broke the plate” versus “the plate broke by itself”) on the ability to recall the agent in accidental events (Fausey & Boroditsky, 2010, 2011). Beyond categorization and discrimination tasks, however, there is to our knowledge very little research on the effects that different processing habits across languages might have on cognitive processes in nonlinguistic domains. In order to address this issue, we recently tested whether specific branching directions trigger habits of processing information that might also be retained beyond the linguistic domain (Amici et al., 2019). In this study, we presented adult native speakers of four LB and four RB languages with a battery of three WM tasks (assessing the “dynamic” attention-allocation aspect) and three short-term memory tasks (assessing the passive storage aspect; Conway et al., 2005; Unsworth et al., 2005). We selected Khmer, Italian, Oshiwambo, and Northern Thai as strongly RB languages: they all follow a VO order, for instance, and have head nouns preceding relative clauses (Dryer & Haspelmath, 2013). As LB languages, we selected Japanese, Korean, Nama, and Sidaama, as they all follow an OV order, and have relative clauses preceding noun phrases (Dryer & Haspelmath, 2013). In our study, we tested participants in their native language, adapting the automated span tasks implemented by Unsworth and colleagues (2005, 2014). In short-term memory tasks, we presented participants with a series of stimuli (i.e., words, numbers, or spatial matrixes) that they had to sequentially recall at the end of the presentation. In the short-term memory task with word stimuli, for instance, participants were tested in 18 trials, in which they were shown 2–7 stimuli. The stimuli consisted of 600 px 800 px pictures depicting common animals and objects (e.g., a cat, a hen, a leaf, an ant, a cloth), which sequentially appeared in the middle of the screen for 2 seconds. Before being tested, participants were instructed to observe the pictures, name them aloud as they appeared on the screen, and then recall them at the end of the trial, in the same order as they had appeared. Traditionally, these tasks present participants with written words that have to be read aloud and later recalled, but it is common in cross-cultural research to use images instead of written words to control for differences in the degree of literacy across study groups. WM tasks were identical, but participants were also asked to perform a distracting task right after the stimuli presentation and before the recalling phase (see Unsworth et al., 2005, 2014), because keeping goal-relevant information activated in the face of interference and competition is exactly what
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
311
312
FEDERICA AMICI ET AL.
differentiates WM from short-term memory (Engle et al., 1999; Kane et al., 2007; Kane & Engle, 2003). In the WM task with word stimuli, for instance, participants were presented with 2–5 stimuli per trial that depicted common animals and objects, as for the short-term memory tasks. Before observing the following stimulus, however, participants had to solve a distracting task (i.e., subtracting quantities that also appeared on the screen). At the end of the trial, participants had to recall the stimuli in the same order as they had appeared, as in the short-term memory tasks. For each participant, we then coded the number of initial stimuli (i.e., the first half ) and the number of final stimuli (i.e., the second half ) that had been correctly recalled in each trial of each task. For instance, if subjects were presented with the stimuli ABCDE in a trial, and recalled ACFE, they would score 1 for the initial stimuli (as they correctly recalled the identity and position of the initial stimulus A, but not those of the following stimulus B), and 1 for the final stimuli (as they correctly recalled the identity and position of the final stimulus E, but not that of the preceding one, D). Finally, we ran two generalized linear mixed models (one with the dataset from the WM tasks, and one from the short-term memory task) to assess whether the number of correct stimuli identified in each trial was predicted by the three-way interaction of branching direction (RB or LB), kind of stimuli (numerical, spatial, and word) and stimuli position (initial or final), while controlling for a variety of potentially confounding variables (i.e., biographic and socioeconomic factors). In this way, we could, for instance, test whether LB speakers were more likely to correctly recall initial or final stimuli, as compared to RB speakers, and whether this was true for all the stimuli used (i.e., suggesting a predictive effect of branching in both the linguistic and nonlinguistic domains). Our study showed that, in the WM tasks, LB speakers were significantly better than RB speakers at recalling initial stimuli, but RB speakers were better than LB ones at recalling final ones (Figure 14.1). These findings were confirmed when separately analyzing all the eight languages of this study: in all RB languages, participants recalled final stimuli better than initial ones, whereas in all LB languages speakers were better at recalling initial than final stimuli, with the exception of Sidaama speakers (see Amici et al., 2019, for a possible explanation linked to the lower degree of LB in Sidaama, and/or the more secluded conditions in which Sidaama-speakers lived, which might explain the longer duration of WM trials for these participants and thus the fact that earlier stimuli might have become comparatively less accessible for them). Most importantly, these results were true for all kinds of stimuli used in the study (i.e., not only words, but also numbers and spatial matrixes). Therefore, our results strongly suggest that habits of processing sentences in specific directions might influence not only the way in which attention is allocated (and thus stimuli recalled) in the linguistic domain, but also in nonlinguistic domains (e.g., with numerical and spatial stimuli). Why should LB speakers be more proficient at recalling initial stimuli, and RB speakers more successful with final ones? One possible explanation
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
Branching and Working Memory
Figure 14.1 Box plots representing the data distribution for the number of correctly recalled initial and final stimuli in the WM tasks Data are reported separately for left-branching (LB) and right-branching (RB) speakers. The horizontal ends of the boxes represent the 25% and 75% quartiles, and the ends of the whiskers the 2.5% and the 97.5% quartiles. The dotted lines represent the model estimates. Source: Amici et al. 2019. The word order of languages predicts native speakers’ working memory. Scientific Reports, 9, 1124. Reproduced with permission under the Creative Commons Attribution v4.0.
is that all participants, regardless of their language, would tend to better recall final stimuli as compared to initial ones, because recency effects are usually stronger than primacy effects in these kinds of tasks (for a detailed discussion of these findings, and the possible methodological reason for this, see Morrison et al., 2014). However, extensive experience with the linguistic regularities of LB might lead LB speakers to preferentially focus their attention on the initial stimuli, which are more likely to entail casemarkings (Hawkins, 2004) and to provide the statistical information that LB speakers need to predict the structure of the sentence (Konieczny, 2000; Onnis & Thiessen, 2013; see Hawkins, 2004, 2014). In this way, LB speakers might reduce the risk of unassignments and misassignments during language processing and thus decrease processing costs. RB speakers, in contrast, can incrementally process information, have lower risks of unassignments and misassignments, and are less likely to reliably use forward probabilities and/or case-markings to predict the structure of the sentence, so that they are not expected to preferentially focus their attention on the initial stimuli. These processing habits would therefore mask
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
313
314
FEDERICA AMICI ET AL.
recency effects in LB speakers, but not in RB speakers, so that LB speakers would be better at recalling initial stimuli (on which they more strongly focus their attention), whereas RB speakers would find the final ones more accessible. Clearly, this pattern is only shown in WM tasks, but not in shortterm memory tasks, because linguistic habits of processing information should affect the way in which humans allocate the focus of attention (i.e., the dynamic component of the WM), but not their ability to passively store information (i.e., the short-term memory component; see Engle, 2002).
14.6
Conclusion and Directions for Future Research
Overall, cross-linguistic experimental evidence strongly suggests that the branching direction of their language may predict the way in which humans process and recall linguistic stimuli. Moreover, our previous research has shown that this might also be true for nonlinguistic stimuli (Amici et al., 2019). These findings are in line with other research emphasizing the significance of performance preferences and processing habits in shaping grammatical universals (see, e.g., Hawkins, 2004, 2014). Moreover, they confirm that humans are highly sensitive to picking up the statistical regularities of a language, and that they might efficiently use it to increase processing efficiency. Especially interesting, in our opinion, is the fact that linguistic practices might affect processing habits far beyond language comprehension and production. In particular, habits of language processing might not only predict the way in which we perceive and conceptualize the world surrounding us, but also how we process, store, and retrieve nonlinguistic information. Therefore, the interface between language and cognition would not only encompass syntactic structures, but also the sequential processing of information (Amici et al., 2019). This is especially important, considering that WM is fundamental for a variety of higher cognitive functions (e.g. reading, planning, problem-solving, decision-making; Baddeley, 1986; Conway et al., 2005; Finn et al., 2017; Melby-Lervåg & Hulme, 2013; Unsworth et al., 2005). In the future, it will be especially interesting to include data from more languages, with a larger variety of word order structures. In our study, for instance, we included participants from four RB languages (i.e., Italian, Thai, Khmer, and Ndonga) and four LB ones (i.e., Japanese, Korean, Sidaama, and Khoekhoe). To classify languages according to their branching, we mainly relied on the order of V and O, and the position of noun phrases and relative clauses (Dryer & Haspelmath, 2013). However, many other criteria can be used to assess the branching direction of a language, and not all of them might have the same importance. Sidaama, for instance, was the language characterized by fewer left-branching structures (Dryer & Haspelmath, 2013), in our study, and Sidaama speakers were also the only
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
Branching and Working Memory
LB speakers that failed to show the same clear WM pattern as speakers from the other languages (Amici et al., 2019). Is it possible that only branching that involves certain specific structures has a significant effect on processing habits? Moreover, it would be interesting to assess the performance of speakers of languages with mixed branching and free word order. In languages with free word order, for instance, speakers appear to consider sentences containing the same words, but ordered in a different way, as being equivalent (e.g., Hale, 1983). How would speakers of these languages allocate attention during processing of linguistic and nonlinguistic stimuli? Similarly, it would be important to assess how bilinguals speaking an RB and an LB language allocate attention across linguistic and nonlinguistic stimuli. Do they also develop specific patterns of attention allocation, and if so, how easily would they switch between branching-specific patterns? Furthermore, it would be interesting to investigate how sensitivity to branching direction emerges through development, and how patterns of attention allocation develop through age (see, e.g., Thiessen et al., 2019). Finally, more WM tasks should be used to assess whether RB speakers’ better performance with final stimuli was really an effect of more general recency effects, as we hypothesized above (cf. Morrison et al., 2014), or rather the result of other branching-specific processing habits. As with many other aspects of human behavior and cognition, crosscultural and cross-linguistic research appear to be incredibly powerful tools to further investigate the link between language and cognition, including the complex interplay between habits of processing information, attention allocation, and WM. Therefore, preserving linguistic diversity will not only be essential for maintaining different linguistic variants, but also to better test language universalities and the real limits of current linguistic theories (see Evans & Levinson, 2009).
References Amici, F., Sanchez-Amaro, A., Sebastian-Enesco, C., Allritz, M., Salazar, J., Cacchione, T., & Rossano, F. (2019). The word order of languages predicts native speakers’ working memory. Scientific Reports, 9, 1124. Baddeley, A. (2003). Working memory: Looking back and looking forward. Nature Reviews Neuroscience, 4, 829–839. Baddeley, A. D. (1986). Working memory. Oxford University Press. Barrouillet, P., & Camos, V. (2007). The time-based resource-sharing model of working memory. In N. Osaka, R. H. Logie, & M. D’Esposito (Eds.), The cognitive neuroscience of working memory. Oxford Scholarship Online. Boroditsky, L. (2001). Does language shape thought? English and Mandarin speakers’ conceptions of time. Cognitive Psychology, 43, 1–22. Brent, M. R., & Cartwright, T. A. (1996). Distributional regularity and phonotactic constraints are useful for segmentation. Cognition, 61, 93–125.
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
315
316
FEDERICA AMICI ET AL.
Caplan, D., & Waters, G. S. (1999). Verbal working memory and sentence comprehension. Behavioral and Brain Sciences, 22, 77–94. Caplan, D., & Waters, G. S. (2013). Memory mechanisms supporting syntactic comprehension. Psychonomic Bulletin & Review, 20, 243–268. Carpenter, P. A., & Just, M. A. (1988). The role of working memory in language comprehension. In D. Klahr & K. Kotovsky (Eds.), Complex information processing: The impact of Herbert A. Simon (pp. 31–68). Erlbaum. Casasanto, D. (2005). Crying “Whorf.” Science, 307, 1721–1722. Casasanto, D., Boroditsky, L., Phillips, W., Greene, J., Goswami, S., Bocanegra-Thiel, S., Santiago-Diaz, I., Fotokopoulu, O., Pita, R., & Gil, D. (2004). How deep are effects of language on thought? Time estimation in speakers of English, Indonesian, Greek, and Spanish. Proceedings of the Annual Meeting of the Cognitive Science Society, 26, 186–191. Chen, E., Gibson, E., & Wolf, F. (2005). Online syntactic storage costs in sentence comprehension. Journal of Memory and Language, 52, 144–169. Christiansen, M. H., & Chater, N. (2016). The now-or-never bottleneck: A fundamental constraint on language. Behavioral and Brain Sciences, 39, e62 Christiansen, M. H., & MacDonald, M. C. (2009). A usage-based approach to recursion in sentence processing. Language Learning, 59, 126–161. Conway, A. R., Kane, M., Bunting, M., Hambrick, D., Wilhelm, O., & Engle, R. (2005). Working memory span tasks: A methodological review and user’s guide. Psychonomic Bulletin & Review, 12, 769–786. Cowan, N. (1995). Attention and memory: An integrated framework. Oxford University Press. Cowan N. (1999). An embedded-processes model of working memory. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 62–101). Cambridge University Press. Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24, 87–185. Cowan, N. (2014). Working memory underpins cognitive development, learning, and education. Educational Psychology Review, 26, 197–223. Cowan, N., Elliott, E. M., Scott, S. J., Morey, C. C., Mattox, S., Hismjatullina, A., & Conway, A. R. (2005). On the capacity of attention: Its estimation and its role in working memory and cognitive aptitudes. Cognitive Psychology, 51, 42–100. de Villiers, J. G. (2007). The interface of language and theory of mind. Lingua, 117, 1858–1878. Draganski, B., Gaser, C., Kempermann, G., Kuhn, H. G., Winkler, J., Büchel, C., & May, A. (2006). Temporal and spatial dynamics of brain structure changes during extensive learning. Journal of Neuroscience, 26, 6314–6317. Dryer, M. S. (1992). The Greenbergian word order correlations. Language, 68, 81–138.
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
Branching and Working Memory
Dryer, M. S. (2002). Case distinctions, rich verb agreement, and word order type. Theoretical Linguistics, 28, 151–157. Dryer, M. S. (2009). The branching direction theory revisited. In S. Scalise, E. Magni, & A. Bisetto (Eds.), Universals of language today (pp. 185–207). Springer. Dryer, M. S., & Haspelmath, M. (Eds.). (2013). The world atlas of language structures online. Max Planck Institute for Evolutionary Anthropology. http://wals.info Engle, R. W. (2002). Working memory capacity as executive attention. Current Directions in Psychological Science, 11, 19–23. Engle, R. W., & Oransky, N. (1999). The evolution from short-term to working memory: Multi-store to dynamic models of temporary storage. In R. Sternberg (Ed.), The nature of cognition (pp. 514–555). MIT Press. Engle, R., Kane, M., & Tuholski, S. (1999). Individual differences in working memory capacity and what they tell us about controlled attention, general fluid intelligence, and functions of the prefrontal cortex. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 102–134). Cambridge University Press. Evans, N., & Levinson, S. 2009. The myth of language universals: Language diversity and its importance for cognitive science. Behavioral and Brain Sciences, 32, 429–448. Fausey, C. M., & Boroditsky, L. (2010). Subtle linguistic cues influence perceived blame and financial liability. Psychonomic Bulletin & Review, 17, 644–650. Fausey, C. M., & Boroditsky, L. (2011). Who dunnit? Cross-linguistic differences in eye-witness memory. Psychonomic Bulletin & Review, 18, 150–157. Finn, A. S., Minas, J. E., Leonard, J. A., Mackey, A. P., Salvatore, J., Goetz, C., West, M. R., Gabrieli, C. F. O., & Gabrieli, J. D. E. (2017). Functional brain organization of working memory in adolescents varies in relation to family income and academic achievement. Developmental Science, 20, e12450 Frank, S. L., Trompenaars, T., & Vasishth, S. (2016). Cross-linguistic differences in processing double-embedded relative clauses: Working-memory constraints or language statistics? Cognitive Science, 40, 554–578. Frazier, L. (1985). Syntactic complexity. In D. R. Dowty, L. Karttunnen, & A. M. Zwicky (Eds.), Natural language parsing: Psychological, computational, and theoretical perspectives (pp. 129–189). Cambridge University Press Frazier, L., & Fodor, J. A. (1978). The sausage machine: A new two-stage parsing model. Cognition, 6, 291–325. Garrod, S., & Pickering, M. J. (2004). Why is conversation so easy? Trends in Cognitive Sciences, 8, 8–11. Gaser, C., & Schlaug, G. (2003). Brain structures differ between musicians and non-musicians. The Journal of Neuroscience, 23, 9240–9245.
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
317
318
FEDERICA AMICI ET AL.
Gibson, E. A. (1998). Linguistic complexity: Locality of syntactic dependencies. Cognition, 68, 1–76. Gibson, E., & Pearlmutter, N. J. (1998). Constraints on sentence comprehension. Trends in Cognitive Sciences, 2, 262–268. Gibson, E., & Thomas, J. (1999). Memory limitations and structural forgetting: The perception of complex ungrammatical sentences as grammatical. Language and Cognitive Processes, 14, 225–248. Gilbert, A., Regier, T., Kay, P., & Ivry, R. (2006). Whorf hypothesis is supported in the right visual field but not the left. Proceedings of the National Academy of Sciences, 103, 489–494. Gimenes, M., Rigalleau, F., & Gaonac’h, D. (2009). When a missing verb makes a French sentence more acceptable. Language and Cognitive Processes, 24, 440–449. Gomez, R. L., & Gerken, L. A. (2000). Infant artificial language learning and language acquisition. Trends in Cognitive Sciences, 4, 178–186. Greenberg, J. H. (Ed.). (1963). Universals of language. MIT Press. Hale, J. (2001). A probabilistic Earley parser as a psycholinguistic model. Proceedings of NAACL, 2, 159–166. Hale, K. (1983). Warlpiri and the grammar of non-configurational languages. Natural Language & Linguistic Theory, 1, 5-47. Haun, D. B. M., Rapold, C., Call, J., Janzen, G., & Levinson, S. C. (2006). Cognitive cladistics and cultural override in Hominid spatial cognition. Proceedings of the National Academy of Sciences, 103, 17568–17573. Hawkins, J. A. (1994). A performance theory of order and constituency. Cambridge University Press. Hawkins, J. A. (2004). Efficiency and complexity in grammars. Oxford University Press. Hawkins, J. A. (2014). Cross-linguistic variation and efficiency. Oxford University Press. Hunt, E., & Agnoli, F. (1991). The Whorfian hypothesis: A cognitive psychology perspective. Psychological Review, 98, 377–389. Jaeger, L. A. (2015). Working memory and prediction in human sentence parsing (Doctoral dissertation, University of Potsdam). https://publishup.unipotsdam.de/opus4-ubp/frontdoor/deliver/index/docId/8251/file/jaeger_diss .pdf Just, M. A., & Carpenter, P. A. (1992), A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122–149. Kamide, Y., Altmann, G. T. M., & Haywood, S. L. (2003). The time-course of prediction in incremental sentence processing: Evidence from anticipatory eye movements. Journal of Memory and Language, 49, 133–156. Kane, M. J., Conway, A. R. A., Hambrick, D. Z., & Engle, R. W. (2007). Variation in working memory capacity as variation in executive attention and control. In A. R. A. Conway, C. Jarrold, M. J. Kane, A.
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
Branching and Working Memory
Miyake, & J. N. Towse (Eds.), Variation in working memory (pp. 21–48). Oxford University Press. Kane, M. J., & Engle, R. W. (2003). Working-memory capacity and the control of attention: The contributions of goal neglect, response competition, and task set to Stroop interference. Journal of Experimental Psychology: General, 132, 47–70. Konieczny, L. (1996). Human sentence processing: A semantics-oriented parsing approach (Doctoral dissertation, Universität Freiburg). https://www .researchgate.net/publication/36150321_Human_sentence_processing_a_ semantics-oriented_parsing_approach Konieczny, L. (2000). Locality and parsing complexity. Journal of Psychological Research, 29, 627–645. Levinson, S. C., & Wilkins, D. P. (Eds.). (2006). Grammars of space: Explorations in cognitive diversity. Cambridge University Press. Levy, R. (2008). Expectation-based syntactic comprehension. Cognition, 106, 1126–1177. Lewis, R. L., Vasishth, S., & Van Dyke, J. A. (2006). Computational principles of working memory in sentence comprehension. Trends in Cognitive Sciences, 10, 44–54. Li, P., & Gleitman, L. R. (2002). Turning the tables: Language and spatial reasoning. Cognition, 83, 265–294. Maguire, E., Gadian, D., Johnsrude, I, Good, D., Ashburner, J., Frackowiak, R., & Frith, C. (2000) Navigation-related structural changes in the hippocampi of taxi drivers. Proceedings of the National Academy of Sciences, 97, 4398–4403. Mazuka, R. (1998). The development of language processing strategies: A crosslinguistic study between Japanese and English. Psychology Press. Mazuka, R., & Lust, B. (1988). Why is Japanese not difficult to process? A proposal to integrate parameter setting in Universal Grammar and parsing. NELS, 18, 333–356. Melby-Lervåg, M., & Hulme, C. (2013). Is working memory training effective? A meta-analytic review. Developmental Psychology, 49, 270–291. Miller, G. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Morrison, A. B., Conway, A. R., & Chein, J. M. (2014). Primacy and recency effects as indices of the focus of attention. Frontiers in Human Neuroscience, 8, 6. Nakatani, K., & Gibson, E. (2010). An on-line study of Japanese nesting complexity. Cognitive Science, 34, 94–112. Núñez, R. E., & Sweetser, E. (2006). With the future behind them: Convergent evidence from Aymara language and gesture in the crosslinguistic comparison of spatial construals of time. Cognitive Science, 30, 401–450.
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
319
320
FEDERICA AMICI ET AL.
Onnis, L., & Thiessen, E. (2013). Language experience changes subsequent learning. Cognition, 126, 268–284. Pearlmutter, N. J., & MacDonald, M. C. (1995). Individual differences and probabilistic constraints in syntactic ambiguity resolution. Journal of Memory and Language, 34, 521–542. Pica, P., Lemer, C., Izard, V., & Dehaene, S. (2004). Exact and approximate arithmetic in an Amazonian indigene group. Science, 306, 499–503. Pickering, M. J., & Garrod, S. (2007). Do people use language production to make predictions during comprehension? Trends in Cognitive Sciences, 11, 105–110. Pienemann, M. (Ed.). (2005). Cross-linguistic aspects of processability theory. John Benjamins Publishing. Pyers, J. E., & Senghas, A. (2009). Language promotes false-belief understanding evidence from learners of a new sign language. Psychological Science, 20, 805–812. Regier, T., & Kay, P. (2009). Language, thought, and color: Whorf was half right. Trends in Cognitive Sciences, 13, 439–446. Saffran, J. R. (2003). Statistical language learning: Mechanisms and constraints. Current Directions in Psychological Science, 12, 110–114. Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8month-old infants. Science, 274, 1926–1928. Seidenberg, M. S. (1997). Language acquisition and use: Learning and applying probabilistic constraints. Science, 275, 1599–1603. Slobin, D. I. (1996). From “thought and language” to “thinking for speaking.” In J. J. Gumperz & S. C. Levinson (Eds.), Studies in the social and cultural foundations of language: Rethinking linguistic relativity (pp. 70–96). Cambridge University Press. Spelke, E. S., & Tsivkin, S. (2001). Language and number: A bilingual training study. Cognition, 78, 45–88. Stivers, T., Enfield, N. J., Brown, P., Englert, C., Hayashi, M., Heinemann, T., Hoymann, G., Rossano, F., de Ruiter, J., Yoon, K. E., & Levinson, S. C. (2009). Universals and cultural variation in turn-taking in conversation. Proceedings of the National Academy of Sciences, 106, 10587–10592. Thiessen, E. D., Onnis, L., Hong, S. J., & Lee, K. S. (2019). Early developing syntactic knowledge influences sequential statistical learning in infancy. Journal of Experimental Child Psychology, 177, 211–221. Unsworth, N., Fukuda, K., Awh, E. & Vogel, E. K. (2014). Working memory and fluid intelligence: Capacity, attention control, and secondary memory retrieval. Cognitive Psychology, 71, 1–26. Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R. W. (2005). An automated version of the operation span task. Behavior Research Methods, 37, 498–505. Vasishth, S. (2003). Working memory in sentence comprehension: Processing Hindi center embeddings. Garland Press.
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
Branching and Working Memory
Vasishth, S., & Lewis, R. L. (2006). Argument-head distance and processing complexity: Explaining both locality and anti-locality effects. Language, 82, 767–794. Vasishth, S., Suckow, K., Lewis, R. L., & Kern, S. (2010). Short-term forgetting in sentence comprehension: Cross-linguistic evidence from verbfinal structures. Language and Cognitive Processes, 25, 533–567. Winawer, J., Witthoft, N., Frank, M. C., Wu, L., Wade, A. R., & Boroditsky, L. (2007). Russian blues reveal effects of language on color discrimination. Proceedings of the National Academy of Sciences, 104, 7780–7785. Woollett, K., & Maguire, E. A. (2011). Acquiring “the knowledge” of London’s layout drives structural brain changes. Current Biology, 21, 2109–2114.
https://doi.org/10.1017/9781108955638.018 Published online by Cambridge University Press
321
15 Working Memory and Natural Syntax William O’Grady
15.1
Introduction
On September 11, 1956, the day that cognitive science was born according to some accounts (Gardner, 1985, p. 28), George Miller gave a talk entitled “Human memory and the storage of information” at the Symposium on Information Theory at MIT. Miller’s seminal paper drew attention to the importance of capacity limits in information processing, suggesting that the ceiling might be set at seven chunks of information, plus or minus two. The linguistic evidence for his hypothesis focused on the recall of monosyllabic words, but it was not long before Victor Yngve, a pioneer in the field of computational linguistics, suggested that Miller’s conjecture might also provide a limit on “immediate memory”1 in sentence production. Indeed, Yngve went so far as to claim that “much of the complexity of [English] syntax can be explained on the basis of [this] hypothesis” (1956, p. 464), as could the direction of language change and the difficulty of certain patterns for child language learners (1998, p. 634). Miller was not the only person at the 1956 symposium to put forward a seminal idea. Speaking on the same day, Noam Chomsky outlined the need for transformational operations in syntactic analysis (Chomsky, 1956). His ideas created a new paradigm for the study of language, proving a foundation for research on the role of working memory in syntactic computation. At least three separate lines of inquiry have been pursued in studies of how working memory interacts with syntax. The first involves explanations for why certain types of utterances are uninterpretable even though they are the product of well-established grammatical operations. Sentences with more than one level of center-embedding are a classic case (e.g., Chomsky & Miller, 1963, p. 286; Gibson, 1998, pp. 34ff, Miller & Chomsky, 1963, pp. 471ff; Yngve, 1956, p. 460). [S1 The rat [S2 that the cat [S3 that the dog worried] killed] ate the malt].
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
A second line of inquiry involves the role of processing in creating preferences for one acceptable pattern over another, as illustrated by the placement of the phrase to the girl in the following two sentences (e.g., Gibson, 1998, pp. 51ff ). Preferred – short phrase before long phrase: The young boy gave [PP to the girl] [NP the beautiful green pendant that had been in the jewelry store window for weeks]. Dispreferred – long phrase before short phrase: The young boy gave [NP the beautiful green pendant that had been in the jewelry store window for weeks] [PP to the girl]. A third set of questions, on which I focus in this chapter, involves a possible role for working memory in explaining various facts that are traditionally attributed to grammatical principles – such as the requirement that a reflexive pronoun have an antecedent in the same clause, or the unacceptability of the strange prohibition against the appearance of the complementizer that in certain types of question patterns. Mary insists that [Sally underestimates herself]. (herself = Sally, not Mary) *Who did you say that left early? (compare: Who did you say left early?) The possibility that grammatical principles might be derived from processing pressures – a highly disruptive notion within linguistics – has been put forward in different forms. One idea, pioneered by Hawkins (2004, 2014, this volume), proposes that a grammar’s rules and syntactic representations reflect the need to minimize processing cost, including the burden on working memory. Grammatical rules, Hawkins proposes, “have incorporated properties that reflect memory limitations and other forms of complexity and efficiency that we observe in performance” (2014, p. 6). Performance-Grammar Correspondence Hypothesis (abridged): Grammars have conventionalized syntactic structures in proportion to their degree of preference in performance. (2014, p. 3) A parallel approach, which I have been developing for some time (O’Grady, 2005, 2015, 2021), makes an even sharper break from the generative tradition by denying the existence of a grammar at all. Its starting point is the radical line of emergentist thought outlined in Sections 15.2 and 15.3, which lay the groundwork for the series of case studies presented in Sections 15.4, 15.5, and 15.6. Section 15.7 offers some concluding remarks.
15.2
Linguistic Emergentism
Emergentist approaches to language encompass a broad range of inquiry (MacWhinney, 2015, p. 9). Nonetheless, they tend to converge
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
323
324
WILLIAM O’GRADY
Figure 15.1 Direct mapping
Figure 15.2 Mediated mapping
on a commitment to the idea that the properties of complex systems, including language, arise from the interaction of simpler and more basic forces and factors. The version of linguistic emergentism that has been dubbed “Natural Syntax” (O’Grady, 2021) adopts three further assumptions.
15.2.1 Assumption 1: Direct Mapping The mapping between a sentence’s form and its meaning is direct, in the sense that it does not require reference to syntactic structure (see Figure 15.1). This idea contrasts with the more standard view that the relationship between form and meaning is mediated by syntactic representations (see Figure 15.2).
15.2.2 Assumption 2: Algorithmic Orientation The mapping between form and meaning is regulated by algorithms that operate in real time in the course of speaking and understanding. Here are two very simple examples that are part of the processing routine needed to manage intransitive sentences such as Harry left. The First-Argument Algorithm: Map the referent of the nominal onto the first argument position in the semantic representation.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
Harry
↦
PRED
The Predicate Algorithm: Map the event denoted by the verb onto the predicate position. Harry left.
↦
LEAVE
On this view, there are no grammatical rules per se, even for word order. English is a subject–verb language because of the order in which the two processing algorithms apply. If they were activated in the reverse order, English would manifest a verb–subject template like Irish, Hawaiian, and various other languages.
15.2.3 Assumption 3: Processing Determinism A further key claim of natural syntax is that the algorithms involved in mapping form onto meaning (and vice versa) are largely shaped by forces that contribute to processing efficiency (a view also adopted by Hawkins). [T]he [processor] should operate in the most efficient manner possible, promptly resolving dependencies so that they do not have to be held any longer than necessary. This is a standard assumption in work on processing, where it is universally recognized that sentences are built in real time under conditions that favor quickness. (O’Grady, 2005, p. 7) Speed in communicating the intended message from speaker to hearer and minimal processing effort in doing so are the two driving forces of efficiency. (Hawkins, 2014, p. 48; see also Hawkins, 2004, p. 9) Taken together, the three assumptions that I have just summarized ensure that the mapping between form and meaning – the essential activity of language – is shaped and constrained by processing considerations that have a natural and well-established role in cognition. In this, they differ from the formal laws of grammar on which most analytic work in contemporary syntax is based. I will focus in this chapter on the effects of processing determinism and on its close association with working memory.
15.3
Working Memory
A striking feature of research on working memory and language is the stark divide between studies of working memory as a mechanism and studies of language as a rule-driven phenomenon. Responsibility for the divide is shared by those on both sides.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
325
326
WILLIAM O’GRADY
There is a large and rich literature on working memory, dating back at least to the 1950s. As the other chapters in this volume illustrate, very important issues have yet to be resolved: What are the components of working memory? Are there specialized buffers, such as a phonological loop? Is memory for items distinct from memory for serial order? What is the link between long-term memory and verbal working memory? And so on. These are important questions, whose investigation is closely associated with linguistic phenomena ranging from the recall of nonce words to the difficulty of comprehending, say, direct object relative clauses compared to subject relative clauses. Crucially, however, the pursuit of these matters pays little attention to the issues that preoccupy linguists. One finds few references in the literature on working memory to, say, Archangeli and Pulleyblank’s (2015) theory of phonological contrasts or Kayne’s (1994) highly influential analysis of relative clauses. In contrast, work within mainstream linguistics tends to flow in the opposite direction, making quite minimal assumptions about working memory as it explores the intricacies of sentence processing. This approach equates working memory with “the set of cognitive processes involved in the temporary storage and manipulation of information” (Gathercole, 2008, p. 33), a commonly held view in the linguistic literature (e.g., Gibson, 1998, p. 2; Hawkins, 2004, p. 12; Jackendoff, 2007, p. 13; Lewis et al., 2006, p. 447; O’Grady, 2005, p. 6; p. 447; Schwering & MacDonald, 2020, p. 2). On this conception, which I will also adopt here, working memory capacity is closely aligned with a calculus of processing cost built on the following assumptions. • • •
The less information that has to be processed, the better. An operation that can apply immediately is less costly than an operation of the same type that has to be delayed. An operation that requires storage of material is more costly than an operation of the same type that does not require that storage.
Consistent with the conceptual divide in the field, my goal is not to offer new insights into working memory, any more than research on working memory is intended to offer new analyses of phonological contrasts or relative clauses. Rather, the point is to explore the possibility that even the most basic properties of working memory, however they are ultimately integrated into a theoretical model, can contribute to a deeper understanding of the workings of human language. As I will attempt to show in this chapter, this approach has considerable promise, yielding fundamental insights into a series of problems in the syntax of natural language which have traditionally been attributed to highly abstract formal principles of grammar, which now become unnecessary. This in turn opens the door for an emergentist theory of language built around a spare but uncontroversial conception of working memory.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
The way forward for this type of research involves a focus on phenomena whose morphosyntactic properties are both well understood and complex enough to strain the mind’s processing resources. Of the many phenomena that fit this description, I will focus here on just three – a restriction on the interpretation of reflexive pronouns, a curious prohibition on phonological contraction in a type of wh question, and a baffling constraint commonly known as the “that-trace effect.” In each case, I will outline the facts and their relevance as informally as possible, avoiding technical details where it is practical to do so. Readers who are interested in a more formal analysis may wish to consult the detailed treatment of these phenomena outlined in O’Grady (2021).
15.4
A Restriction on Pronoun Interpretation
It has long been understood that prototypical reflexive pronouns in languages such as English seek out a coargument antecedent (Jespersen 1933:111). Thus, the reflexive pronoun that serves as the second argument of cut in the example below can refer only to Richard, the first argument of the same verb. It cannot refer to Marvin, let alone to some unnamed party. Marvin just found out what happened. Richard cut himself while playing with scissors. Why should this be? An important clue comes from the following real-time scenario for interpreting the reflexive pronoun in the example above. a. The nominal Richard is encountered and is assigned a referent (represented here as the index r). Richard r b. The transitive verb cut is encountered and its two-place predicate-argument structure is projected, with Richard as the first argument. Richard cut CUT
c.
The reflexive pronoun is encountered and identified as the verb’s second argument (represented by the symbol x), thereby triggering the search for an antecedent. Richard cut himself. CUT
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
327
328
WILLIAM O’GRADY
d.
The processor interprets the reflexive pronoun immediately and locally with the help of the previously identified referent of the verb’s first argument, Richard. Richard cut himself. CUT
↳r
The first three steps map the words in the sentence directly onto a corresponding semantic representation, without the mediation of syntactic structure or grammatical rules. In the fourth and final step, the justencountered reflexive pronoun receives an interpretation that links it to the verbal predicate’s first argument. This way of proceeding has a quite obvious natural motivation: it has the fortuitous effect of resolving referential dependencies immediately and locally,2 in response to internal processing pressures. At the point where the reflexive pronoun is encountered and identified as the verb’s second argument, only the first argument is immediately available to serve as antecedent – the correct result, as predicted by the claim that processing pressures shape this particular part of language. Interestingly, the interpretive facts are mastered very quickly in the course of first language acquisition, despite a paucity of relevant data. (In the Brown corpus, for instance, there are just five third-person reflexive pronouns in caregiver speech directed to Sarah; see https://sla.talkbank.org/ TBB/childes/Eng-NA/Brown/Sarah). All of this makes perfect sense since, in a way, there is nothing for children to acquire; they have only to surrender to the natural impulse to minimize processing cost. The consequence of that impulse is the immediate resolution of the referential dependency by selecting the nearest possible antecedent – a prior coargument. In other words, “learning” in this instance simply consists of doing as little as possible.
15.5
A Prohibition on Contraction
A signature property of English and many other languages is the placement of wh words at the left edge of a question, creating a dependency with a sentence-internal lexical item – usually a verb, as in the following example. Who did Mary see _? " As illustrated here, the dependency is traditionally represented with the help of a gap (_). Although I will follow tradition in employing the term “filler-gap dependency” alongside “wh dependency,” I do not equate the gap with an actual structural position. Rather, I treat it simply as the point in the course of processing at which the dependency can be resolved – a view
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
that is also widely held in the psycholinguistic literature (e.g., Pickering, 2000; Traxler & Pickering, 1996; and see Hawkins, 2004, 171–172, for discussion). A good deal of research has converged on the finding that filler-gap dependencies place significant burden on the resources of working memory. There is good evidence that wh dependencies increase the difficulty of a sentence. Sentences with wh dependencies are rated as harder and less acceptable than sentences without wh dependencies. (Phillips, 2013, p. 90; see also Goodall, 2004, p. 102; and Hawkins, 2004, p. 173, among many others) This observation brings us to the question that lies at the heart of this chapter: how does processing cost shape the syntax of wh dependencies? A very striking example can be discerned in the effect that a sentenceinitial wh word can have on the pronunciation of want to in sentences such as the following. Yes-no question; contraction is natural: Do you want to stay? # wanna Wh question; contraction is unnatural: Who do you want to stay? # ?*wanna It is a remarkable fact about both language and linguistics that a single casually pronounced word might shed light on the workings of human cognition, but that does seem to be the case.3 As Getz notes in her review of the literature on the subject (2019, p. 120), the wanna phenomenon has played a “momentous role . . . in shaping modern linguistic theory.” Two different subtypes of want are in play here, above and beyond the type exemplified in patterns such as I want that bicycle. The first involves just an infinitival complement, as in the following example. I want to-stay. " The second involves both a nominal complement and an infinitival complement. I want Harry to-stay. " " It is the latter pattern that resists contraction when its second argument is a sentence-initial wh word – as in Who do you want to stay? The question is why.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
329
330
WILLIAM O’GRADY
The search for an explanation has yielded a number of spectacular proposals. Perhaps the most intriguing of these, put forward by Jaeggli (1980, p. 242), is that contraction is blocked by an invisible empty category – the case-marked “trace” (t) left by the operation that supposedly moves the wh word from its initial position between want and to to the left edge of the sentence (in theories of generative grammar). A case-marked trace blocks contraction. (Jaeggli, 1980, p. 242; see also Chomsky, 1980, p. 160.) Who do you want [NP tobj ] to stay? "–––––––––––––––| A different way to approach the contractibility asymmetry is to consider it from a processing perspective, with attention to two overlapping demands. A first demand involves articulatory processing: contraction is most natural in rapid speech. Natural Contraction Contraction of the string XY is most natural when X adjoins to Y without delay. (O’Grady, 2005, pp. 139ff ) It is for this reason that one can say I want to stay in a slow and deliberate manner with a pause after each word, including want, but that there is no natural pronunciation of I wanna stay in which there is a pause after wan. A second demand that shapes the syntax of contraction can be traced very directly to working memory. A long tradition of psycholinguistic research has established that the wh filler must be associated with an appropriate verb at the first opportunity. An early statement of this requirement was formulated by Clifton and Frazier (1989, pp. 292 & 297); see also Aoshima, et al. (2002, p. 2); Gibson (1998, p. 54); Hawkins (2004, p. 174); Wagers and Phillips (2009, pp. 396–397), among many others. The Active Filler Hypothesis (paraphrased) Associate the filler with a “gap” as quickly as possible. In sum, we have a situation defined by two demands: i. Phonological considerations call for the immediate adjunction of want and to if there is to be contraction. ii. Working memory pressures favor resolution of the burdensome fillergap dependency at the first opportunity, which arises upon encountering the verb want. (Recall that the gap represents the point in time at which the wh dependency is resolved, not a structural position.) Who do you want _ to stay? " The filler-gap dependency is resolved here. (cf. you want who to stay)
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
As can be seen here, the two requirements are mutually incompatible in the forbidden pattern: prompt resolution of the wh dependency at want undermines the opportunity for immediate adjunction with to – a prerequisite for contraction. There is striking independent evidence for this proposal: in patterns where contraction is impeded, want has a lengthier articulation and is followed by a prosodic break, suggesting activity at the verb consistent with resolution of the wh dependency (Warren et al., 2003). Who do you w a n t # to stay? " Lengthening and prosodic adjustment at the point where the filler-gap dependency is resolved Crucially, there is no such effect in the pattern that permits contraction, since there is no filler-gap dependency to resolve.4 Do you want to stay? # wanna In sum, the constraint on want to contraction is real, but it does not call for a syntactic explanation involving case-marked traces or any of the other devices in the traditional armory of generative grammar. Rather, it is an emergent property of working memory, which demands quick resolution of wh dependencies at the expense of opportunities for contraction.
15.6
A Baffling Constraint on Filler-Gap Dependencies
Let us turn now to an even more challenging case involving the syntax of wh dependencies. Because this phenomenon requires consideration of somewhat more complex syntactic patterns as well as data from languages other than English, I have deliberately left it for the final part of this chapter. Cross-clausal wh dependencies I take as my starting point the fact that some languages (e.g., English, but not Russian) allow filler-gap dependencies to extend across a clause boundary.5 English: Who did Mary say [that they saw _ ]? |–––––––––––––––––––––––––| Russian (Dyakonova 2009, p. 215): *Kogo Olga skazala [čto oni videli _ ]? |––––––––––––––––––––––––––––––––––––––––––––––| who.acc Olga.nom say.pst.fem that they.nom see.pst.pl
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
331
332
WILLIAM O’GRADY
It is not surprising that some languages reject cross-clausal dependencies. If the gap is located in an embedded clause, the filler must be transported across the embedded clause boundary.. . .Carrying a filler across a clause boundary results in additional processing cost. (Kluender, 1998, p. 253; see also Kluender & Kutas, 1993) Psycholinguistic research shows . . . that processing clause boundaries generally lowers acceptability ratings and causes an increase in processing time. (Hofmeister & Sag, 2010, p. 383; see also Alexopoulou & Keller 2007, pp. 133 & 136) There is no reason why all languages should allow cross-clausal dependencies, any more than they should all permit articulatorily difficult consonants – say, uvular ejectives or velar implosives. This is simply one of the points where the linguistic capacities of normal human beings permit variation. The difference between English and Russian is therefore not surprising. Moreover, from a processing perspective, the more interesting question is not why Russian prohibits cross-clausal wh dependencies; it is how English manages to permit them, given their additional processing cost. The key moment for the processor arises at the point where it has to make the transition to the second clause. In a sentence such as Who did Mary say they met?, that moment arises at the verb say. Who did Mary say. . . " At this point, the filler-gap dependency is still unresolved. However, the lexical properties of say permit a second clause and, therefore, a new set of possibilities – provided that the wh dependency can somehow be transferred into that clause. There is good reason to think that this is achieved by reactivating the wh word at the onset of the new clause, as illustrated below. (The horizontal line represents maintenance of the wh dependency in working memory, not movement.) Who did Mary say they. . . | –––––––––––––[who] " Reactivation of the filler at the onset of the second clause Evidence from a variety of sources supports the reactivation hypothesis.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
15.6.1 ERP Evidence Left Anterior Negativity, a sign that a wh dependency is being held in memory, is observed at the beginning of the second clause (Kluender & Kutas, 1993, pp. 608–610; see also Phillips et al., 2005, p. 423). What do you think [that the children bought _ ]? " LAN continues into this region.
15.6.2 Acquisition Evidence When young children produce wh questions containing a cross-clausal filler-gap dependency, they sometimes repeat the sentence-initial wh word at the beginning of the second clause. What do you think [what pigs eat]? Who did he say [who is in the box]? As observed by Lutken et al. (2020, p. 37), children appear to be reproducing the sentence-initial wh phrase in order to “strengthen their memory representation of the filler-gap dependency.” Similar suggestions have been put forward by Crain et al. (2006, p. 33), McDaniel et al. (1995), and Thornton (1990).
15.6.3 Typological Evidence Some languages place a copy of the wh word at the beginning of the second clause. Romany (McDaniel, 1989, 569n): Kas o Demìri mislinol [kas Arìfa whom does Demiri think whom Arifa “Who does Demiri think that Arifa saw _?”
dikhľa _ ]? saw
Passamaquoddy (Bruening, 2006, p. 28): Wen Mali wewitahamacil [wen kisiniskamuk _ ]? who Mary remember who I.dance.with “Who does Mary remember I danced with _?” So far, so good – but the syntax of wh dependencies raises many additional puzzles. A costly complementizer One of the great enduring mysteries of natural language involves the unacceptability of sentences like the one below, in which a filler-gap dependency extends into the subject position of an embedded clause that begins with the “complementizer” that.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
333
334
WILLIAM O’GRADY
Subject gap with a complementizer: *Who do you think [that _ is here]? complementizer
Strikingly, the sentence complementizer.
subject position
is
perfectly
acceptable
if
there
is
no
Subject gap with no complementizer: *Who do you think [_ is here]? subject position
I will call this contrast, first noted by Perlumutter (1968), the “that-effect.” (In generative grammar, it is generally referred as the “that-trace effect.”) The That-Effect The complementizer that is incompatible with a subject gap. Why should such an effect exist? The most common explanations make reference to one or another principle of Universal Grammar. (An early favorite was the Empty Category Principle, which required that empty positions be “properly governed” by meeting a series of intricate structural conditions; see Chomsky, 1986, pp. 10ff, for details.) At first glance, it seems unlikely that a phenomenon of this type could be reduced to a working memory effect. After all, that is just one small word, with a barely detectable semantics. Nonetheless, upon closer inspection, it is possible to discern a link to processing cost. A good place to start is with the question of why the absence of the complementizer has the ameliorating effect that it does. Given the general orientation of Natural Syntax, it makes sense to ask whether the complementizer-free pattern might offer a less costly mapping between form and meaning than its counterpart with that. To see how this might work, let’s pick up the acceptable sentence in midstream, right after the matrix verb think. Who do you think. . . | ––––––––––––––? In the next step, the processor encounters the verb is, the nucleus of the second clause. Who do you think is " The processor encounters the verb in the embedded clause. The availability of a subject-less verb at the very point where the new clause begins opens the door to the immediate resolution of the wh dependency – obviating the need for reactivation.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
Who do you think is |–––––––––––––" Resolution of the wh dependency by associating it directly with the verb is A key piece of evidence for this scenario comes from the fact that the verb in the lower clause is able to fuse with its counterpart in the higher clause. Who do you think’s here? (cf. Who do you think _ is here?) Contraction would not be possible here if reactivation had occurred between the two verbs. As noted in Section 15.5, phonological reduction can occur only if the second element adjoins to the first without delay. The acceptability of the sentence above is therefore strong evidence that the wh dependency was resolved immediately – before reactivation even became an option. Compare this scenario with what happens when the embedded clause is introduced by a complementizer. *Who do you think [that _ is here]? Under these circumstances, resolution of the filler-gap dependency is substantially more costly. To see this, let’s pick up the sentence just after the complementizer makes its appearance. Who do you think that . . . Upon encountering that, the processor finds itself committed to two operations that were not required in the case of the complementizerfree pattern. First, the complementizer must be assigned an interpretation. This essentially involves identifying its function, which is to signal the presence of a clausal argument. Projection of a second clause after encountering that: Who do you think that " The presence of that signals an upcoming embedded clause. Second, because the wh dependency cannot be resolved at this point, it must be reactivated if the search for a gap is to continue. Reactivation: Who do you think that | –––––––––––––[who] " reactivation of the wh filler as the processor prepares to engage with the embedded clause
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
335
336
WILLIAM O’GRADY
Upon subsequently encountering the verb is, the filler can be associated with the open subject position. Who do you think that is | –––––––––––––[who]–" resolution of the wh dependency by associating it with the verb is This way of proceeding is significantly more costly than the course of action allowed by the complementizer-free pattern, which offers an opportunity to resolve the wh dependency without the need for reactivation. Who do you think is . . . " The processor encounters an open subject position at the onset of the second clause. This contrast leads to a promising two-part hypothesis: • A cross-clausal wh dependency requires the least costly computational option. • Reactivation adds cost that should be avoided if possible. A precise prediction can now be made. The Prediction A that-effect will occur when the complementizer alone delays immediate access to the verb in the second clause, thereby triggering reactivation. Some further evidence This prediction appears to fare well in a variety of different circumstances and languages, offering a straightforward explanation for a series of important facts. I will mention just two such facts here; for more extensive discussion, see O’Grady (2021). Fact 1: There is no that-effect in direct object wh questions.
The presence of the complementizer should be irrelevant to the acceptability of embedded non-subject questions. Cross-clausal wh dependency involving a direct object gap: Who do you think (that) Mary met _? Even if that were suppressed here, the presence of the subject (Mary) would preclude immediate resolution of the wh dependency at the clause boundary, necessitating reactivation of the wh filler.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
Who do you think (that) Mary . . . | –––––––––––––[who] The processor encounters the preverbal subject in the second clause, precluding any opportunity for immediate resolution of the wh dependency and triggering reactivation of the wh filler. Since reactivation is therefore inevitable in any case, the complementizer cannot be held responsible for triggering an operation that could not have been avoided in any case. As predicted, there is therefore no that-effect. Fact 2: There is no that-effect in languages with postverbal subjects.
There should be no that-effect in languages such as Italian whose flexible word order delays the opportunity to resolve the wh dependency until midclause, where it can ascertained that the subject argument is missing. Chi hai detto [ che ha scritto _ questo libro]? (Rizzi 1982:NN) who has said that has written " that book “Who do you say that wrote that book?” Because resolution of the wh dependency at the onset of the second clause is therefore ruled out on independent grounds, the presence of the complementizer is benign and there is no that-effect. In sum, a single word with a sparse semantics can have a very significant impact on processing cost, creating a cascade of effects that include one of the most mysterious phenomena in syntax. The key to this puzzle lies in the realization that the complementizer is not so much the bearer of processing cost as it is the trigger. When that is solely responsible for triggering reactivation of the wh dependency, its presence is toxic. In contrast, its occurrence is benign in cases where reactivation is independently necessitated by other factors. No syntactic principle is in play, just the pressure to minimize processing cost where it is possible to do so.
15.7
Conclusion
In concluding, it is useful to return to Victor Yngve’s attempt to build a processing-based theory of syntax at the dawn of modern cognitive science. The idea was genuinely radical and could well have changed the history of linguistics, had it attained its goal of explaining “the utility of much of the observed syntactic complexity in English as an adaptation of grammar to the limited temporary memory available to speakers” (1998, p. 634). In fact, of course, the particular theory that Yngve put forward (often dubbed the
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
337
338
WILLIAM O’GRADY
“depth hypothesis”) fell short. Miller & Chomsky (1963, pp. 474ff ) noted one set of problems, and Yngve himself eventually rejected his idea for a different set of reasons (1998, p. 635).6 But dead ends and wrong turns were as commonplace in the early era of syntactic theory as they are today. In Chomsky’s (1956) presentation at the celebrated MIT symposium (and in Syntactic structures, which was published a year later), he proposed a distinction between two types of sentences: a “small possibly finite kernel” of simple, active declarative sentences derived by applying only obligatory transformations, and a set of sentences derived by applying optional transformations (p. 123). This idea was eventually abandoned, but not the underlying theory, which underwent numerous rounds of revision and reconceptualization – the Standard Theory, the Extended Standard Theory, Government and Binding theory, Principles and Parameters theory, Minimalism. . . By contrast, there were no second chances for early theories that sought to understand the intricacies of syntax in terms of limitations on working memory. Now, though, new discoveries and insights have revived interest in that line of inquiry, with somewhat different foundational assumptions but with the same general objective in mind – understanding syntax in terms of processing cost. Why do reflexive pronouns require a local antecedent? Is a case-marked trace really responsible for blocking want to contraction? Why can languages such as Russian and English differ in the admissibility of cross-clausal wh dependencies? How can a mere complementizer trigger the puzzling ripple of syntactic consequences that it does? Could these and other phenomena perhaps be nothing more than emergent properties of working memory? A further set of challenges stems from the need to bring together currently unaligned streams of research on working memory and syntactic computation. Working memory, in whatever form it exists, is obviously crucial for language-related cognition, but its relevance to actual theories of language use remains “scant,” as Schwering & MacDonald note (2020, p. 11). It is completely uncontroversial that language comprehension and production processes are constrained by what is commonly called ‘‘verbal working memory capacity’’. . . and yet the specific mechanisms posited in classic VWM models are, with only a few exceptions, absent from theorizing about how limited capacities shape language processes. (Schwering & MacDonald, 2020, p. 11; see also Caplan & Waters, 2013) This gap in our understanding of cognition must somehow be bridged, an undertaking that will no doubt shed light on both the mechanisms of memory and the computational operations that shape the properties of language. That is something to look forward to with considerable excitement.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
References Alexopoulou, T., & Keller, F. 2007. Locality, cyclicity and resumption: At the interface between the grammar and the human sentence processor. Language, 83, 110–160. Aoshima, S., Phillips, C., & Weinberg, A. 2002. Active filler effects and reanalysis: A study of Japanese wh-scrambling constructions. University of Maryland Working Papers in Linguistics, 12, 1–24. Archangeli, D., & Pulleyblank, D. 2015. Phonology without universal grammar. Frontiers in Psychology, 6, article 1229. Bresnan, J. 1977. Variables in the theory of transformations. In P. Culicover, T. Wasow & A. Akmajian (Eds.), Formal syntax (pp. 157–196). Academic Press. Bruening, B. 2006. Differences between the wh-scope-marking and wh-copy constructions in Passamaquoddy. Linguistic Inquiry, 37, 25–49. Caplan, D., & Waters, G. 2013. Memory mechanisms supporting syntactic computation. Psychonomic Bulletin & Review, 20, 243–268. Carpenter, P., Miake, A., & Just, M. 1994. Working memory constraints in comprehension: Evidence from individual differences, aphasia, and aging. In M. Gernsbacher (Ed.), Handbook of psycholinguistics (pp. 1075–1122). Academic Press. Chomsky, N. 1956. Three models for the description of language. Institute of Radio Engineers Transactions on Information Theory, 2(3), 113–124. Chomsky, N. 1980. Rules and representations. Columbia University Press. Chomsky, N. 1986. Barriers. MIT Press. Chomsky, N., & Miller, G. 1963. Introduction to the formal analysis of natural languages. In R. Luce, R. Bush, & E. Galanter (Eds.), Handbook of mathematical psychology (Vol. 2, pp. 269–321). Wiley. Cinque, G. 2020. The syntax of relative clauses: A unified analysis. Cambridge University Press. Clifton, C., & Frazier, L. 1989. Comprehending sentences with long-distance dependencies. In G. Carlson & M. Tanenhaus (Eds.), Linguistic structure in language processing (pp. 273–317). Kluwer. Cowan, N. 2015. George Miller’s magical number of immediate memory in retrospect: Observations on the faltering progress of science. Psychological Review, 122, 536–541. Crain, S., Goro, T., & Thornton, R. 2006. Language acquisition is language change. Journal of Psycholinguistic Research, 35, 31–49. Culicover, P. 1993. Evidence against ECP accounts of the that-t effect. Linguistic Inquiry, 24, 557–561. Dyakonova, M. 2009. A phase-based approach to Russian free word order. LOT. Gardner, H. 1985. The mind’s new science: A history of the cognitive revolution. Basic Books. Gathercole, S. 2008. Working memory. In H. Roediger (Ed.), Cognitive psychology of memory (pp. 33–52). Elsevier.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
339
340
WILLIAM O’GRADY
Getz, H. 2019. Acquiring wanna: Beyond universal grammar. Language Acquisition, 26, 119–143. Gibson, E. 1998. Linguistic complexity: Locality of syntactic dependencies. Cognition, 68, 1–76. Goodall, G. 2004. On the syntax and processing of wh-questions in Spanish. In V. Chand, A. Kelleher, A. Rodrígues, & B. Schmeiser (Eds.), Proceedings of the West Coast Conference on Formal Linguistics (pp. 101–114). Cascadilla Press. Hawkins, J. 2004. Efficiency and complexity in grammars. Oxford University Press. Hawkins, J. 2014. Cross-linguistic variation and efficiency. Oxford University Press. Hofmeister, P., & Sag, I. 2010. Cognitive constraints and island effects. Language, 86, 366–415. Jackendoff, R. 2007. A parallel architecture perspective on language processing. Brain Research, 1146, 2–22. Jaeggli, O. 1980. Remarks on to-contraction. Linguistic Inquiry, 11, 239–245. Jespersen, O. 1933. Essentials of English grammar. Allen and Unwin Kayne, R. 1994. The antisymmetry of syntax. MIT Press. Kluender, R. 1998. On the distinction between strong and weak islands: A processing perspective. In P. Culicover & L. McNally (Eds.), The limits of syntax (Syntax and Semantics 29), 241–279. Academic Press. Kluender, R., & Kutas, M. 1993. Subjacency as a processing phenomenon. Language and Cognitive Processes, 8, 573–633. Lewis, R., Vasishth, S., & Van Dyke, J. 2006. Computational principles of working memory in sentence comprehension. Trends in Cognitive Sciences, 10, 447–454. Lutken, C. J., Legendre, G., & Omaki, A. 2020. Syntactic creativity errors in children’s wh-questions. Cognitive Science, 44(7), e12849. MacWhinney, B. 2015. Introduction. In B. MacWhinney & W. O’Grady (Eds.), The handbook of language emergence (pp. 1–31). Wiley-Blackwell. McDaniel, D. 1989. Partial and multiple wh-movement. Natural Language and Linguistic Theory, 7, 565–604. McDaniel, D., Chiu, B., & Maxfield, T. 1995. Parameters for wh-movement types: Evidence from child English. Natural Language and Linguistic Theory, 13, 709–753. Miller, G. 1956. Human memory and the storage of information. Transactions on Information Theory, 2(3), 129–137. Miller, G., & Chomsky, N. 1963. Finitary models of language users. In R. Luce, R. Bush, & E. Galanter (Eds.), Handbook of mathematical psychology (Vol. 2, 419–491. Wiley. O’Grady, W. 2005. Syntactic carpentry: An emergentist approach to syntax. Erlbaum.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
Working Memory and Natural Syntax
O’Grady, W. 2015. Anaphora and the case for emergentism. In B. MacWhinney & W. O’Grady (Eds.), The handbook of language emergence, 100–122. Wiley-Blackwell. O’Grady, W. 2021. Natural syntax: An emergentist primer. http://ling.hawaii .edu/william-ogrady/ and researchgate.net Pearl, L., & Sprouse, J. 2013. Syntactic islands and learning biases: Combining experimental syntax and computational modeling to investigate the language acquisition problem. Language Acquisition, 20, 23–68. Perlmutter, D. 1968. Deep and surface constraints in syntax (Doctoral dissertation, Department of Linguistics, MIT). Phillips, C. 2013. On the nature of island constraints I: Language processing and reductionist accounts. In J. Sprouse & N. Hornstein (Eds.), Experimental syntax and island effects (pp. 64–108). Cambridge University Press. Phillips, C., Kazanina, N., & Abada, S. 2005. ERP effects of the processing of syntactic long-distance dependencies. Cognitive Brain Research, 22, 407–428. Pickering, M. 2000. No evidence for traces in sentence comprehension. Behavioral and Brain Sciences, 23, 47–48. Rizzi, L. 1982. Issues in Italian syntax. Foris. Schwering, S. & MacDonald, M. 2020. Verbal working memory as emergent from language comprehension and production. Frontiers in Psychology, 14, Article 68. doi: 10.3389/fnhum.2020.00068 Sprouse, J., Wagers, M., & Phillips, C. 2012. A test of the relation between working memory capacity and syntactic island effects. Language, 88, 82–123. Thornton, R. 1990. Adventures in long-distance moving: The acquisition of complex wh-questions (Doctoral dissertation, University of Connecticut). Traxler, M., & Pickering, M. 1996. Plausibility and the processing of unbounded dependencies: An eye-tracking study. Journal of Memory and Language, 35, 454–475. Wagers, M., & Phillips, C. 2009. Multiple dependencies and the role of the grammar in real-time comprehension. Journal of Linguistics, 45, 395–433. Warren, P., Speer, S., & Schafer, A. 2003. Wanna-contraction and prosodic disambiguation in US and NZ English. Wellington Working Papers in Linguistics, 15, 31–49. Yngve, V. 1960. A model and an hypothesis for language structure. Proceedings of the American Philosophical Society, 104, 444–466. Yngve, V. 1998. Clues from the Depth Hypothesis: A reply to Geoffrey Sampsons’ review. Computational Linguistics, 24, 633–640.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
341
342
WILLIAM O’GRADY
Notes 1 Yngve used this term interchangeably with “working memory” (e.g., his page 464); Miller’s paper uses only the term “immediate memory.” 2 The relevance of locality to processing cost requires careful consideration, especially if it is measured in terms of linear distance, whose relevance to processing cost has been called into question (Lewis et al., 2006, p. 451). This critique does not apply to locality that is defined in terms of argument structure, as in my proposal. 3 In fact, the phenomenon appears to extend well beyond the verb want. A similar contrast seems to occur in patterns such as the following, in which the contracted form expectənə seems fully natural only in the first sentence. Were you expecting to stay? Who were you expecting to stay? 4 At least one other computational operation is required here, namely the identification of you as the first argument of stay. Crucially, however, it takes place at a later point (at the verb stay) and therefore does not interfere with contraction. 5 In fact, technically, there are no clauses per se in the theory I adopt. The notion “clause” is really a proxy for a predicate-argument structure in a sentence’s semantic representation. PRED
6 In contrast, as noted by Cowan (2015), the conjecture by George Miller that inspired Yngve’s idea has aged quite well.
https://doi.org/10.1017/9781108955638.019 Published online by Cambridge University Press
16 The Role of Working Memory in Shaping Syntactic Dependency Structures Chunshan Xu and Haitao Liu 16.1
Introduction: Dependency Grammar, Dependency Parsing, and Dependency Distance
One fundamental property of human languages is their linearity (Saussure, 1959): linguistic symbols like words have to appear in succession, yielding a sequence extending in one dimension. Obviously, the words in this linear succession must be somehow related to one another; otherwise, the sequence will make no sense. But the relations cannot be merely between two neighboring words. If every word can only combine with the word immediately before or after it, structures thus formed would be too rigid and too inflexible to meet the vast communicative needs of human beings. Authentic sentences often have structures much more involved, which are not one-dimensional successions, but two-dimensional trees with complicated relationships among words. Grammar is mainly concerned with the analysis of these relations in sentences, and the description of the two dimensional structures underlying linear sentences (sentence structure), usually by the aid of word order, inflection, and functions words (Givón, 2009). As to grammar, there are two traditions or paradigms that are influential today. One is phrase structure grammar (Bresnan et al., 2015; Chomsky, 1957; Pollard & Sag, 1994) and the other is dependency grammar (Hudson, 2010; Jiang & Liu, 2018; Mel’čuk, 2003; Tesnière, 1959). Phrase structure grammar holds that phrases are the fundamental syntactic units of sentences, and the fundamental syntactic relations are the part-whole relations (den Dikken, 2013). Figure 16.1(a) is a phrase structure tree, where the basic units are phrases like S(sentence), NP
This work is partly supported by the National Fund of Social Scince of China (18BYY015).
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
344
CHUNSHAN XU AND HAITAO LIU
Figure 16.1 Phrase structure and the dependency structure of a sentence
(noun phrase), VP(verb phrase), and the part-whole relations among these units are often expressed as rules like S ! NP þ VP. Dependency grammar, instead, regards words, not phrases, as the fundamental units of sentences, and a sentence structure lies in the governor-dependent relations (dependency) between words (Hudson, 2010; Liu, 2009; Tesnière, 1959), not the partwhole relation of phrase structure (den Dikken, 2013). In terms of dependency grammar, the finite verb is the root of a sentence, with all other words relating to, or depending on, this verb, directly or indirectly. Figure 16.1(b) presents a dependency tree, where the root of the sentence is the finite predicate verb saw, which has two dependents boy and dogs, which in turn have their own dependents the and three, and thus a hierarchical dependency tree is formed. To note, though phrasal nodes like NP are absent in dependency structures, dependency grammar makes no denial of phrases or chunks, which in fact can be easily derived from a dependency tree. So, grammars provide the syntactic knowledge that is indispensable for describing and analyzing the syntactic structures of sentences. In phrase structure grammar, a very important part of that knowledge is the phrase structure rules, which, together with other lexical knowledge like subcategorization and the like govern the analysis of syntactic structures. However, in dependency grammar, there are no such autonomous rules, for, according to dependency grammar, words are not only the fundamental units of syntactic structures but also the source of syntactic knowledge for syntactic description and analysis. A typical word has certain properties, or knowledge, like meaning, syntactic valency (the number and type of words that are syntactically licensed by this word), and so forth, which will direct the analysis and the generation of syntactic structures (Hudson, 2007, 2010). For instance, the word destroy syntactically requires a nominal word to serve as its subject and another one as its object. This requirement is the syntactic valency of destroy. Of course, the subject and the object of the verb destroy may in turn have their valences or syntactic requirements. In languages with rather fixed word order, the relative position may also be https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
part of the syntactic knowledge (Hudson, 2010): the word destroy requires a nominal word in front of it to serve as its subject and another one behind it as its object. Furthermore, the probability distribution of different valencies of a word may also be one property of the word (Liu, 2006). In brief, it is on the basis of these properties of words such as valency, meaning, relative position, probability, and so forth, that we conduct a dependency analysis to find out how every word is related to another word in a sentence, and to finally establish a full dependency tree of the sentence. This process is dependency analysis or dependency parsing. To explain and to imitate the process of dependency parsing (analysis), various theories and models have been proposed. Some of them are engineering-oriented, committed more to computerized dependency parsing than to cognitive explanation and description of the dependency analysis. These engineering-oriented models generally fall into two types: graph-based dependency parsing and transition-based dependency parsing. In graph-based parsing, every possible dependency relation is built between every pair of words, and then the entire tree space is searched for the dependency tree with the highest probability or score (Li et al., 2009; McDonald et al., 2005). However, this is probably not the way human beings understand a sentence: in most cases, we build only one tree and give only one interpretation. Different from graph-based parsing models, transitionbased models (Nivre & Scholz, 2004; Yamada & Matsumoto, 2003) bear more similarity to the process in which human beings syntactically analyze a sentence. The fundamental parsing strategy of transition models is incremental processing. That is, the parser will start with the left-most word of a sentence, process the sentence word by word, and try linking each word to a previous word as its head or dependent (Covington, 2001). In this left-toright fashion, a dependency tree is incrementally built, with the words accepted one-by-one into the parser. One example is the parsing algorithm proposed by Covington (2003), which has two lists. One is the Headlist, storing words that have no heads yet, and the other is the Wordlist, containing all the words already processed. First, the algorithm accepts a word W and adds it to Wordlist: W + Wordlist. Then it searches the Headlist for the possible dependent(s) D of W. If successful, it links D to W as W’s dependent(s), and removes D from the Headlist. Then it searches the Wordlist for the possible governor (head) H of W. If successful, it links W as a dependent of H. If there is no head for W found, then it adds W to Headlist. This process will repeat until the entire sentence is processed, with all the words in the Wordlist, and only the finite verb in the Headlist. For example, before the algorithm parses the sentence I see a young boy, both the Wordlist and the Headlist are empty. Therefore, to process the first word I means to add it to both the Wordlist and the Headlist. Then the verb see is processed, which has a dependent I in the Headlist, but has no head in the Wordlist. So I is linked as a dependent of see, and then is removed from the Headlist. See is added to both the Wordlist [I see] and the Headlist [see]. https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
345
346
CHUNSHAN XU AND HAITAO LIU
The next word a can find neither a dependent in the Headlist nor a head in the Wordlist, and hence is added to the Wordlist [I see a] and the Headlist [see a]. The next word young also has neither a dependent in the Headlist nor a head in the Wordlist, and is hence added to the Wordlist [I see a young] and the Headlist [see a young]. The last word boy has two dependents, a and young, in the Headlist, which are linked to boy and removed from the Headlist. At the same time, this word boy has a head see in the Wordlist, so it is not added to the Headlist [see] but added to the Wordlist [I see a young boy]. The dependency parsing is thus finished, with all the words in the Wordlist a finite verb in the Headlist This incremental parsing strategy seems to be somewhat similar to how human beings comprehend sentences: we read and understand a sentence largely in a left-to-right fashion, one word after another. What is even more interesting is that there seems to be some similarity between human memory and Headlist: it may be assumed that the more words in the Headlist, the heavier memory burden they will make (Liu 2008). Take the above sentence as an example; before the last word boy is processed, the Headlist has to retain the most words during the parsing: the predicate verb see and two modifiers a and young. This may imply the heaviest burden during the parsing. A closer examination shows that two words a and young, which make for the burden, intervene between boy and its head see, giving rise to a nonadjacent dependency between them, that is, long distance between boy and see. In fact, cognitive models of dependency grammar, such as Word Grammar, do regard the linear distance between syntactically related words as correlating with syntactic difficulty (Hudson, 1995, 2007, 2010). Word Grammar (Hudson, 2007, 2010), grounded in cognitive sciences, is an important school of dependency grammar. This syntactic theory provides a cognition-based account of dependency parsing, an account that may not be strictly formalized, but is in line with existent theories of working memory. It is a basic tenet of cognitive linguistics that the understanding of a sentence is essentially a cognitive activity, based on and implemented through some general cognitive processes. Word Grammar is no exception, which describes the dependency analysis as a process implemented in working memory via some common cognitive mechanisms such as attention, network organization of knowledge, multiple inheritance, default inheritance, prototype effects, spreading activation, best global candidate, best fit principle, binding, and the like (Hudson, 2007, 2010). As has been mentioned, the processing of a sentence may well be incremental. During the dependency analysis, it may be assumed that a sentence will be accepted word by word, that is, one word is processed at a time. According to Word Grammar (Hudson 2007, 2010), seeing or hearing a word-token (such as sees) directs our attention to this word to activate it with the visual or audial stimulus; then a node will be built for it in the
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
Figure 16.2 Long and short dependency distance in a sentence
cognitive network, and a best match (in this case it is word-type see) for this new node will be searched in stored knowledge and bound to this new node through the mechanism of spreading activation. In this way, all the important properties of the word type see will be inherited by the new node sees, including its syntactic valency, which defines how this word should be related to other words, and will play vital roles in dependency analysis (Hudson, 2007, 2010). Word Grammar holds that, through the spreading activation (Anderson, 1983) and best fit principle, these activated properties will guide a search in words that have already been processed, the search for a best-fit dependent or governor of this word, and bind them together to build dependency relations between them (Hudson 2007, 2010). Then, after the present word is processed, the attention will be turned to the next word and this procedure will be repeated till the last word is processed. In terms of cognitive network, binding one node to another node means an operation in which the activation of the former, which may result from attended stimulus or other sources, spreads via its relations with other nodes (these relations are in fact its properties), and eventually activates another node or nodes (Hudson 2007, 2010). In this vein, the establishment of a dependency relation can be seen as an operation of binding the presently processed word, which is now at the focus of the attention, to a previously processed work, which has been removed from the focus of attention. In this operation of binding, the former is activated and spreads out its activation through properties that link valency to reactivate a word in working memory, which is linguistically termed as its head or dependent (Hudson 2007, 2010). When this process is over, the attention will be shifted to the next word, activating it to begin a new binding operation. In view of this processing model, we have an intuition that the longer a word is not attended to, the harder it will be to recall or to reactivate it in memory. To put it in a different way, the longer distance between two syntactically related words, the harder, it seems, to bind them (Hudson, 1995, 2010). This distance is termed in dependency grammar as dependency distance (see Figure 16.2). As can be seen in Figure 16.2, the predicate verb see is adjacent to its subject I, making for a very short dependency between them, while the object kids is separated from the predicate verb by three intervening words, leading to a much longer dependency. Dependency distance is held as closely related to syntactic difficulty (Gibson, 1998, 2000; Hudson, 1995 2010; Liu, 2008; Liu et al., 2017) The
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
347
348
CHUNSHAN XU AND HAITAO LIU
reason is that dependency distance between the dependent and the governor may have a significant effect on the binding operation, or, on the reactivation of a previously processed word, which can be explained in light of theories of working memory, especially those concerning the forgetting mechanism of working memory.
16.2
Working Memory Cost of Dependency Distance
The processing difficulty caused by dependency distance is closely related to working memory, which is usually understood as a cognitive system that maintains information no longer available for perception. Baddeley and Hitch (1974) proposed a multicomponent model of working memory, composed of three parts: the central executive, the phonological loop, and the visuospatial sketchpad. This model was later developed into a four-part one, with an additional episodic buffer (Baddeley, 2000a). The central executive directs attention, coordinates cognitive processes, and integrates information. The phonological loop bears on language processing, storing verbal information (the sound of language), and constantly rehearsing this information to prevent it from decaying; otherwise, the information retained in the articulatory loop will decay quickly and soon become unavailable (Baddeley, 1990, 2000b) This model treats working memory as a system independent of the longterm memory, while some other models conceive working memory as the activated part of long-term memory. One example is Cowan’s embedded processes model (Cowan, 1995, 1999) which holds “the information processing system as embedded processes: (1) the LTM (long-term memory) system, (2) within which, a subset is currently activated, that is, in a temporarily heightened state of activation, and (3) within which, a subset is currently in the focus of attention” (Vergauwe & Cowan, 2015, p. 904). Oberauer (2002) added to this model a third level of embedding, the focus of the focus of the attention, selecting and attending to only one item at a time. This model emphasizes that the operation of working memory hinges on activation (Cowan, 1995, 1999; Engle et al., 1992; Lovett et al., 1999), which is in the same line with the theory of Word Grammar discussed earlier: the establishment of a dependency relation can be seen as an operation of binding, or, of spreading activation of the present word through properties linked to reactivate a previous word in working memory. Both models of working memory believe that the decay of the information in working memory is one important reason for forgetting. According to multicomponent model, without repeated rehearsal, the information kept in the phonological loop will soon decay into disappearance (Baddeley, 1990, 2000b). Similarly, the embedded processes model claims that when attention is shifted away, the activation of an item will
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
decline with time (Cowan, 1995, 1999). This decay-based view of forgetting can also be found in time-based working memory resource-sharing model, according to which both processing and maintenance pool the same resource of attention (Camos & Barrouillet, 2011). To maintain items in working memory demands attention, and retrieval cues must be attended to for the reactivation of items in long-term memory (Anderson, 1993). Attention keeps information activated in working memory, whose diversion will leave the information in working memory in constant decay. The decay is a function of time, and therefore, the effort required by the recall (reactivation) is also a function of the time. These theories of working memory may explain the effect of dependency distance on language processing. It may be inferred, given these theories, that the word that has been processed and thus removed out of attention will suffer from constant decay of activation, which will reasonably cause trouble for later binding or reactivation (Gibson, 2000; Hudson, 1995, 2007, 2010). And it seems a safe bet that the degree of activation decay may correlate with the dependency distance, because longer distance implies more activation decay (Liu et al., 2017). Dependency analysis is largely incremental, which means that attention will shift from one word to the next. When the next word is processed, it will capture the attention, drive the preceding word out of attention, and leave it in continuing decay. According to Word Grammar (Hudson 2007, 2010), the binding will reactivate this out-of-attention word through spreading activation. Obviously, the more intervening words between the two syntactically related words, the more temporal interval there will be, which means more decay of the previous word and more difficulty in reactivating it when it is bound to the presently processed word. In other words, when two words are bound together to establish the dependency, longer distance between them means more burden on memory and more processing difficulty. Thus, decay of memory is probably one important reason for the processing difficulty invoked by dependency distance. However, decay is not accepted by some scholars as the sole reason for the forgetting of items in working memory. They believe that it is probably not time-based decay but representation-based interference that is responsible for forgetting (Oberauer & Lewandowsky, 2013; White, 2012). Linguistically, a word or an item is, from the perspective of Word Grammar, a bundle of properties that link to other nodes in the network of knowledge, and the activation of the word (item) node is nothing but the activation of these properties (Hudson 2007, 2010). This is compatible with the interference-based model of forgetting: an item in working memory is represented by a bundle of features, some of which, in a network of knowledge, may be part of the representations of other items, that is, features shared by other items, and the competition between items for shared features may result in degraded memory traces (Nairne, 1990; Neath, 2000). To put it in a different way, the activation may be degraded by similarity (Vasishth & Lewis, 2006), which is
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
349
350
CHUNSHAN XU AND HAITAO LIU
probably a result of the fan effect: activation of nodes depends on associative strengths, and strengths decrease with increasing associated nodes (Anderson & Reder, 1999). Therefore, it is believed by some scholars that the processing difficulty of long dependency distance may stem from the similarity-based interference, which is caused by the intervening words in a long dependency (VanDyke, 2007; VanDyke & Lewis, 2003). The establishment of a dependency relation between two words is a process of activation and reactivation. The attention will keep the presently processed word at a highly activated state, and the activation will spread to and reactivate a previously processed word through feature links, such as valency (specifications of its head or dependent), landmark (a feature usually merged with valency, specifying the relative position of the head or dependent of a word) (Hudson, 2010), and so forth. Then, the binding is finished with an established dependency relation. The problem is that, if there are two words that are similar in many ways (sharing the same valency and landmark feature), the spreading activation from the currently processed word may be shared by two or more words, which may lower the activation level the target word may reach, or even cause the reactivation of the wrong words. It may be expected that more intervening words (longer dependency distance) may mean more serious interference, or at least higher probability of interference. Of course, two adjacent words bring about the shortest dependency distance and no chance for interference, while longer dependency distance between two words means more intervening words and higher probability or higher degree of interference. In brief, both the decay-based and the interference-based models of forgetting are compatible with the cognitive process of dependency analysis: the distance between two syntactically related words may bring about both activation decay and similarity-based interference. In other words, dependency distance may be an index of the cognitive load of dependency analysis and an indicator of syntactic difficulty: longer dependency distance may call for more processing effort.
16.3
Empirical Evidence
The processing difficulty of long dependency distance has been attested by many psychological experiments, especially those experiments on relative clauses (Gibson, 1998; Gibson & Wu, 2013; Hsiao & Gibson, 2003; Levy et al., 2013). Many researches have reported more complexity and difficulty in English object-extracted relative clauses like The reporter [that the senator attacked] than in subject-extracted relative clauses like The reporter [that attacked the senator], as reflected by the different reading times of the verb of relative clauses (Fedorenko et al., 2012; Gibson, 1998; Grodner & Gibson, 2005). But in Chinese, more complexity is found in subject-extracted
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
relative clauses than in object-extracted relative clauses (Chen et al., 2008; Lin & Garnsey, 2010). This difference is probably due to the different word orders of English and Chinese. In English, relative clauses always follow their head nouns, which means that the object-extracted relative clause always has its subject intervening between the verb of relative clause and the head noun, as can be seen in The reporter [that the senator attacked]. As a result, English object-extracted clauses tend to involve longer dependency distance than subject-extracted relative clauses, which, as discussed in the preceding section, may cause more decay of, or more interference to, the previous word (such as reporter) in working memory . However, in Chinese, relative clauses always precede their heads. Hence subject-extracted relative clauses have longer dependency distance because the object of the relative clause will intervene between the verb of the relative clause and the head noun (Fedorenko et al., 2012; Hsiao & Gibson, 2003). In German and Russian, it is found that shorter dependencies often impose less processing difficulty in relative clauses (Levy et al., 2013; Levy & Keller, 2013). The effect of dependency distance on language processing is also found in other syntactic structures. For example, psychological experiments show that shorter dependency distance may reduce the processing difficulty of the main clause (Bartek et al., 2011; Husain et al., 2014; Lin, 2011), and increase the acceptability of wh-dependencies (Hofmeister et al., 2007; Hofmeister & Sag, 2010). In fact, the processing ease of short dependency plays an important role in guiding syntactic analysis, which has been captured by the principle of late closure (Frazier, 1979) – in a sentence, if there are two elements to which the presently processed word can be syntactically related, we tend to choose the element that is nearer to this word. In brief, psychological experiments suggest that longer dependency distance tends to bring about heavier processing load, increasing the difficulty of language comprehension. The phenomenon is probably a result of the mechanisms of working memory and forgetting, and may have significance influence on the syntactic dependency structures of human languages.
16.4
A General Tendency toward Dependency Distance Minimization
Psychological studies have shown that short dependency distance often contributes to the ease of language processing, and long dependency distance may demand considerable processing effort (Hudson, 2010; Gibson, 2000; VanDyke, 2007; VanDyke & Lewis, 2003). As a result, language users, who are subject to the universal principle of least effort (Zipf, 1949), may involuntarily prefer structures with short dependency distance so as to reduce the effort needed for processing. This preference may have tremendous influence on the syntactic structures of human beings, gradually giving rise to numerous syntactic patterns that are featured by short dependency distance.
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
351
352
CHUNSHAN XU AND HAITAO LIU
This view that working memory may have shaped dependency structures of human languages deviates from a long tradition that tries to model human languages, especially their syntax, as an autonomous formal system, a tradition traced back to Saussure (1959) and becoming dominant with the seminal work of Chomsky (1957). According to this tradition, a formal system of strict rules is the essence of human language, while the use of language is peripheral. However, there have recently been increasing different voices from cognitive linguistics (Lakoff & Johnson, 2003; Langacker, 2008), synergetic linguistics (Köhler, 1986, 2005), and usagebased grammar (Bybee, 2006), which believe that linguistic patterns are to a considerable degree grounded in the cognitive mechanisms of humans, and embodied in the actual use of language. In a sentence, language is not an autonomous system of rules, but a human-driven complex adaptive system (Liu, 2018; Liu et al., 2017). For example, synergetic linguistic, based on theories of complex adaptive system (Haken, 1983; Kauffman, 1993), argues that language is not an independent system, but restrained by many outside factors, among which the most important are probably the physiological and cognitive mechanisms of human being. These mechanisms underlie the verbal behaviors and the communicative needs of human beings. As a complex adaptive system, language is in constant self-organization, adapting to the outside constraints like the communicative needs and physiological-cognitive mechanisms. To recap, it is human beings who use languages, and languages in use will adapt to, through self-organization, the communicative needs and cognitive constraints of human beings. Linguistic patterns and laws are largely the results of this self-organization and self-adaptation in language use. Working memory is generally seen as one foundation of human cognition, supporting virtually all important human cognitive activities. Language is no exception, subject to the constraint of working memory. In this sense, it may be reasonable to assume that, under the effect of the principle of least effort, there might be a universal tendency in human languages to structure sentences in ways that minimize dependency distance, which, as discussed in the previous section, may lead to the decay or the interference in working memory. This is a reasonable assumption, but needs solid empirical evidences. Some early studies have found that human languages seem to exhibit rather short dependency distance. A statistical investigation into spoken texts of English and Japanese reported very short mean dependency distance in both languages: 0.386 word in English and 0.43 word in Japanese, which means that for two syntactically related words, there is an average of less than half a word intervening between them in these two languages (Hiranuma, 1999). Similarly, another study on the dependency distance of German and English also reports rather short mean dependency distance in both languages: 0.49 word in English, and 0.87 word in German (Eppler, 2004).
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
Subsequent comparative studies have yielded more convincing evidences that languages present a tendency toward dependency distance minimization (Ferrer-i-Cancho, 2004; Futrell et al., 2015, 2020; Gildea & Temperley, 2010; Liu, 2007, 2008; Temperley, 2007, 2008; Xiang et al., 2019). These studies generally generate baselines of random sentences by randomly rearranging the word order or the dependency structures of corresponding authentic sentences. Then the dependency distance of the random sentences is measured and compared with that of the corresponding authentic sentences to find out whether their dependency distance is significantly different from that of authentic sentences. Such comparative studies were first conducted into Romanian and Czech, with the findings that the mean dependency distance of random sentences with scrambled word order is much longer than that of authentic sentences, and that the average dependency distance increases drastically with sentence length in random sentences but only increases slightly with sentence length in authentic sentences (Ferrer-i-Cancho, 2004). These findings clearly point to a tendency toward dependency distance minimization in these two languages, which is absent in corresponding random languages (Ferrer-i-Cancho, 2004). Similar findings were reported in another study on Chinese, which scrambled the dependencies, not words, with the governors and dependents randomly assigned, and generated an additional type of random sentences by imposing, on the totally random sentences, a constraint of projectivity, which is generally found in natural languages. It is found that the mean dependency distance of authentic Chinese sentences is significantly shorter than those of the two randomly generated languages, and that, interestingly, the projective random language presents a much shorter mean dependency distance than the random language without this constraint (Liu, 2007). Some other scholars use a slightly different measure: mean dependency length, or the average per-sentence dependency length, but still reached similar findings: in both German and English, the average per-sentence dependency length of authentic sentences is significantly shorter than that of the corresponding random sentences (Gildea & Temperley, 2010; Temperley, 2007, 2008). These studies cover only a handful of languages. More languages are probably needed to verify the alleged universal preference for structures with short dependency distance. Liu (2008) conducted an investigation into 20 languages, and found that the mean dependency distances of all these 20 languages are much shorter than those of the corresponding random languages. Futrell et al. (2015) explored the dependency length of 37 languages of different language families, and arrived at similar findings: for all these languages, the average per-sentence dependency lengths are much shorter than those of the corresponding random languages. Two recent studies extended the range to more than 50 languages (Futrell et al., 2020; Xiang et al., 2019), and reached similar findings. All these studies seem to
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
353
354
CHUNSHAN XU AND HAITAO LIU
suggest that there is a universal tendency in dependency structures toward short dependency distance, which is motivated by the need to reduce the effort in retrieving items in working memory, abiding by the universal principle of least effort. These studies suggest that the dependency structures of human languages are shaped by the need to reduce the memory retrieval cost (dependency distance) during language processing. As a result, human languages may feature an overwhelming majority of short dependencies. which is confirmed by the studies on the distribution of dependency distance (Ferrer-i-Cancho, 2004; Jiang & Liu, 2015; Liu, 2007; Lu & Liu, 2015). These studies reported exponential or power-law distribution of dependency distance: adjacent dependencies prevail in dependency structures of authentic languages, and the increase of distance will sharply reduce the frequency of dependencies. Such distributions of dependency distance underlie the short mean dependency distance of human languages.
16.5
Dependency Structures Shaped by Dependency Distance Minimization
Previous corpus-based researches have repeatedly revealed a universal tendency of human languages toward dependency distance minimization. That tendency must be instantiated in concrete dependency structures or syntactic patterns. Dependency distance is closely related to the linearization of the dependency tree. As a result, the dependency structures with short dependency distance must result from some particular patterns in the dependency trees or in the linearization of these trees. Let’s start with the latter, for it affects dependency more explicitly than the former, whose influence, in contrast, is not readily seen at first glance. The linearization of the dependency tree involves largely the linear arrangement of word order, which makes a kernel part of syntax (Givon, 2009). In this sense, the short dependency distance bears closely on word order. Through manipulating the order of syntactic constituents, language users often produce sentence with dependency structures minimizing dependency distance. In many languages, word orders, despite their apparent diversity, seem to present various patterns that all contribute to short dependency distance. One of such patterns was first observed by Behaghel (1932): given two constituents, the shorter tends to precede the longer. This ordering of constituents may conduce to short dependency distance, and has been widely found at different syntactic levels in different languages. It is reported that the constituents in a noun phrase are often ordered according to their length in ways that are often optimal in reducing dependency distance (Crabbé et al., 2015). Another persistently-observed phenomenon is that short constituents after verbs often occur before long ones, and that the longest constituents usually come last (Behaghel, 1909; Bresnan et al., 2007; Quirk et al., 1972; Wang & Liu, 2014;
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
Wasow, 1997). In addition, this short-before-long tendency will lead to structures that are termed as “extraposition,” such as It surprises me that you came to the meeting, where a dummy it is extraposed and the heavy that-clause is displaced to the end of the sentence, which, though increasing the amount of dependencies, is one important syntactic arrangement to reduce dependency distance (Hudson, 1998, 2010). What is noteworthy is an opposite tendency that, in SOV (subject-objectverb) languages like Japanese, Korean, and Persian, the constituents before the verb generally are arranged in a long-before-short fashion, and people show a preference for such a pattern (Choi, 1999, 2007; Faghiri & Samvelian, 2014; Yamashita & Chang, 2001). Dependency distance minimization is probably the best explanation that account for both the short-before-long pattern in SVO (subject-verb-object) languages and the long-before-short pattern in SOV languages. These studies reveal that many seemingly diverse patterns of dependency structures are in fact driven by the universal constraint of working memory, and serve the same purpose of dependency distance minimization. Apart from these two patterns, it is also proposed that languages tend to be consistently head-first or head-last, forming a tendency of “samebranching” (Chomsky, 1988; Greenberg, 1963), which is believed to contribute to the minimization of the distances between heads and dependents, especially when each word has exactly one dependency (Frazier, 1985; Hawkins, 1994). Dryer (1992) has noted a more complicated scenario where the multiword phrases tend to branch consistently in a language, whereas one-word phrases generally do not. It is probably on the basis of these insights that the Dependency Length Minimization Rules (DLMRs) were put forward, which go beyond the superficial difference in syntax among different languages, extracting from the apparent diversity some common universal laws that govern the word order of dependency structures in various languages so as to reduce dependency distance (Gildea & Temperley, 2010; Temperley & Gildea, 2018). There are three Dependency Length Minimization Rules (DLMRs). DLMR 1 (same-branching rule) prefer structures that branch in the same direction (consistently left-branching or consistently right-branching). DLMR 2 (ordered nesting rule) describes the regularity that when a head has multiple dependents, the shorter dependent phrases should be located closer to the head. DLMR 3 (limited mixed-branching rule) points out that some opposite-branching of dependent phrases is desirable, when the dependent phrases are one word long and the parent head has multiple dependents (Gildea & Temperley, 2010; Temperley & Gildea, 2018). The role of these principles in minimizing dependency distance has been confirmed: artificial languages free of these rules have considerably longer dependency distance than those restrained by these rules (Temperley, 2007). Empirical studies on real languages also show that speakers generally prefer syntactic constructions conforming to these rules (Temperley, 2008).
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
355
356
CHUNSHAN XU AND HAITAO LIU
When word order is fixed, the tendency toward dependency distance minimization may show up in the length distribution of constituents in different positions of dependency structures. A recent study reveals that English objects tend to be much longer than English subjects, and such a difference persists even when their information status is controlled, which is probably due to the fact that long post-modifiers increase the dependency distance of subjects but have no influence on the dependency distance of English objects (Xu, 2018). Dependency distance is closely related to word order. But the relation between dependency distance minimization and word order may be more subtle and implicit than is shown in the apparent, direct, and explicit optimization of word order or constituent length. One example is the omission of arguments, as instantiated in frequent pro-drop (omission of subjects) and intransitive structures (Ueno & Polinsky, 2009). It has been found that arguments of verbs are more frequently omitted in Japanese than in English, which may be responsible for the fact that the dependency distance of Japanese is not statistically different from that of English, though theoretically, the former, as a SOV language with the object intervening between subjects, may well have longer dependency distance than English, an SVO language (Hiranuma, 1999). In fact, the pressure for short dependency distance is probably one reason why there is a tendency for human languages to shift from SOV order to SVO order. SOV order is often favored in simple structures (Langus & Nespor, 2010). However, language evolution seems to constantly augment the complexity of sentences (Givón, 2009), which may have motivated the shift to SVO order for the sake of dependency distance minimization. There are some empirical evidences that the preference for SOV order wanes in complex sequences (Langus & Nespor, 2010; Ferrer-i-Cancho, 2014). One diachronic study of English reveals that the length of the object is an important factor in determining the option between OV order and VO order (Tily, 2010). Even short transitive structures may shift from SOV order to SVO order when the intervening object may cause some trouble for communication (Gibson et al., 2013). The tendency toward dependency distance minimization have shaped many dependency structures and syntactic patterns that make use of work order, or the linearization of the dependency tree, to reduce dependency distance. This, however, does not mean dependency tree itself has no bearing on dependency distance, though this bearing is not as apparent as that of word order. In terms of dependency grammar, every sentence is underlain by a hierarchical dependency tree. The structure or graphic features of a dependency tree will exert much influence on the dependency distance of the sentence, which implies that the preference for short dependencies may somewhat shape the graphic structures of the dependency tree in a sentence. One instantiation of such influence is the rarity of crossing
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
dependencies in natural languages. That is, natural languages are largely projective. In fact, the constraint of projectivity will considerably reduce the dependency distance of random artificial sentences (Liu, 2007, 2008). What is more, computer experiments have also shown that in the computerized generation of dependency structures, manipulation of mean dependency distance will lead to variations in the number of crossing dependencies (Lu & Liu, 2016). These findings suggest that the efforts to reduce dependency distance may lead to the rarity of crossing dependencies (Ferrer-i-Cancho et al., 2018, Gómez-Rodríguez & Ferrer-i-Cancho, 2017; Liu, 2008). Another property of dependency trees that has to do with dependency distance is hubiness, that is, the degree of centralization of the tree (Ferreri-Cancho, 2013; Oya, 2013). High hubiness means that in the tree a few nodes have many links while the rest of the nodes in this tree have quite few links. It has been found that higher degree of centralization leads to longer minimum average dependency distance (Ferrer-i-Cancho, 2013). To reduce dependency distance, it seems that sentences should avoid structures with a high degree of hubiness. Linguistically, high hubiness means that in a sentence, a minority of words have many parallel dependents, whereas other words have very few dependents. In a linear sentence, if a word has three parallel dependents, at least one of them cannot be adjacent to the head word, and the more parallel dependents there are, the more long dependency relations there would be. Higher hubiness probably means that the tree is flatter, with fewer hierarchical levels. Therefore, to meet the need for dependency distance minimization, the dependency structures probably should not be too flat, especially in the cases of long sentences. Studies have found that shorter sentences often have higher hubiness (Oya, 2013), but longer sentences usually have more layers of hierarchical depth (Jiang & Liu, 2015). This is probably one property of dependency structures that prevents the mean dependency distance of sentences from drastically increasing with sentence length. In random sentences where such property is absent, the mean dependency distance increases as a linear function of sentence length. The hierarchy of a dependency tree bears closely on sentence chunking, and chunking may considerably reduce the number of long dependencies. Computational simulations have confirmed the role of chunking in curbing dependency distance, finding that, in terms of dependency distance minimization, the optimal chunk size is between 4 and 7 (Lu et al., 2015). The structural features of the dependency tree may have some influence on the dependency distance, which implies that investigation into dependency distance should be cautious about the dependency scheme used for annotating dependency treebanks, which determines to a considerable degree the dependency structures of sentences. For example, compared with treebanks annotated with dependency grammar, a dependency treebank that is automatically converted from a
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
357
358
CHUNSHAN XU AND HAITAO LIU
phrase structure treebank may exhibit significant difference in dependency distance (Liu et al., 2009). Another study by Yan and Liu (2019) has also reported significant difference in dependency distance between dependency treebanks annotated with Universal Dependencies (UD), which are semantically oriented, treating function words like adpositions, copula, and complementizers as dependents, and Surface-syntactic Universal Dependencies (SUD), which are syntactically oriented, treating these function words, instead, as governors. As a result, a treebank-based study concerning dependency distance should carefully weigh its research purpose so as to properly choose the annotation scheme. Interestingly, one work shows that the texts of different genres present significant differences in dependency distance (Wang & Liu, 2017), which are probably the result of the differences in syntactic structures favored in different text genres. In other words, dependency distance might be a valuable indicator of text genres or, rather, of the different preferences for syntactic structures in different genres.
16.6
Long-Distance Dependency Structures
Many researchers have found that in human languages various syntactic patterns and structures have been shaped by a cognitively driven tendency toward dependency distance minimization. However, it should be noted that in languages, in spite of the majority of short dependencies, there are invariably a minority of long dependencies, as revealed by the long tail in the distribution of dependency distance in various languages. As a matter of fact, the dependency distance of human languages, though much shorter than that of random languages, still has theoretic room for further minimization (Futrell et al., 2015). In other words, no language minimizes its dependency distance to the theoretical minimum, as evidenced by the sporadic long-distance dependencies in almost all languages (Futrell et al., 2015; Gildea & Temperley, 2010; Liu, 2008). There are several possible reasons for this phenomenon. One is simply the lack of the necessary implicit skills to organize dependency structures in such ways to minimize dependency distance. Another probable reason is that other needs may occasionally outweigh the need for dependency distance minimization, giving rise to long-distance dependencies. Syntax has emerged and evolved into increasing complexity to adapt to the communicative needs of human beings such as precision, reliability, expressiveness, or unambiguity (Givón, 2009; Hurford, 2012). These communicative needs may demand many specifications, details, and conditions, which have to be integrated into sentences, and possibly cause involved sentences with long dependencies, as might be found in legal documents, academic papers, and bureaucratic files. It has been reported that the average dependency distance of interlanguage increases significantly during the process of
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
second language acquisition, and seemingly correlates with second language proficiency (Jiang & Ouyang, 2018; Li & Yan, 2020), which probably suggests that language expressiveness and proficiency may partly lie in the ability to properly use some long and complicated dependency structures. The last possibility is, however, that, sometimes, long-distance dependencies do not invoke great processing difficulty. As has been discussed, the processing difficulty mainly lies in the difficulty in the retrieval, or the reactivation of one word, which is caused by the memory-decay and the memoryinterference from long dependency distance. Then it can be assumed that, in a long dependency, if the decay and the interference can be mitigated, this long dependency may not bring about much processing difficulty. Time-based decay can be lessened if the intervening words can be processed within a shorter period of time. In other words, there are probably some patterns or regularities in long-distance dependency structures that can save processing time by facilitating the processing of the intervening elements. Chunking via function words may be a solution. On one hand, the function words are easy and quick to process, and on the other, they may overtly mark the internal structures of the intervening words, contributing to the quick parsing of these words and thus improving the processing efficiency. It is found in both Chinese and English that function words like particles and complementizers are more frequently used in long dependencies than in short dependencies (Jaeger, 2010; Xu, 2015). In some cases, punctuation marks may serve similar ends. They are reported to appear much more frequently between English subjects and their postmodifiers than between English objects and their postmodifiers (Xu, 2018). The reason is probably that the postmodifiers often lead to a long dependency distance between a subject and its verb. To reduce the influence of dependency distance, it is desirable to quickly process the intervening elements and to direct attention to the key elements. The punctuation marks, which are often used in pairs following the subject and preceding the verb, may play the role of structure markers, signaling the subject, the verb, and postmodifier, facilitating the processing (Xu, 2018). Dependency distance makes difficulty in reactivating the previous word, which is a result of decay or interference. It then might be reasoned that a highly accessible word, which is rather easy to activate, may be somewhat insensitive to dependency distance. Cognitively, high frequency or high probability may render a word highly accessible (Ertel, 1977), and thus allow longer dependency distance. A quantitative probe into two Chinese treebanks reveals a correlation between Chinese subject dependency distance and the degree of contextual givenness – subjects more predictable from the context tend to have longer dependency distance, for a higher degree of contextual givenness means higher probability and more accessibility, which renders a word somewhat insensitive to decay (Xu & Liu, 2015). For those who believe that it is similarity-based interference that underlies the syntactic difficulty of long dependencies (Oberauer &
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
359
360
CHUNSHAN XU AND HAITAO LIU
Lewandowsky, 2013; VanDyke, 2007; VanDyke & Lewis, 2003), the solution is rather simple: a long-distance dependency should avoid intervening words that are similar to either the head word or the dependent word of this dependency. It is probably for this reason that case markers are apt to be used in long dependencies (Tily, 2010): it may contribute to the reduction of similarity. In short, long-distance dependencies are sometime unavoidable due to some communicative needs like precision and expressiveness. But language users may utilize some strategies to reduce the processing difficulty of these long-distance dependencies, giving rise some peculiar syntactic patterns in long-distance dependency structures.
16.8
Conclusion
Dependency grammar holds that the syntactic structure of a sentence lies in the dependency relations among words, and that the purpose of dependency analysis is to extract a hierarchical tree of dependencies from the linear word sequence. This analysis is an incremental process implemented in working memory and regulated by the operation mechanisms of working memory, especially the mechanism of forgetting. To reduce the memory burden during dependency analysis, the dependency structures of human languages generally present a tendency toward short dependency distance, which is a result of the universal principle of least effort. This tendency has profound influence on the dependency structures or syntactic structures of human languages, shaping diverse regularities in word order, constituent omission, syntactic evolution, and the graphic features of dependency trees. These regularities in dependency structures yield a rather short overall dependency distance in natural languages. However, the overall dependency distance of natural languages never reaches the theoretical minimum. Sometimes other communicative needs may override the need for short dependency distance, giving rise to a minority of long dependencies in human languages. The principle of least may also have motivated some syntactic patterns that make use of approaches such as chunking and the use of function words and punctuation marks to reduce the effect of dependency distance, and thus render it not necessarily difficult to process dependency structures with long distance. Language is not an autonomous system, but an adaptive system driven by human beings, by their communication needs and cognitive mechanisms, and working memory is probably one of such mechanisms. Therefore, languages have to adapt to the working memory of human beings, giving rise to various patterns in dependency structures to reduce dependency distance or the effect of dependency distance, which plays important roles in shaping dependency structures of human languages.
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
References Anderson, J. R. (1983). A spreading activation theory of memory. Journal of Verbal Learning and Verbal Behavior, 22(3), 261–295. Anderson, J. R. (1993). Rules of the mind. Erlbaum. Anderson, J. R., & Reder, L. M. (1999). The fan effect: New results and new theories. Journal of Experimental Psychology-General, 128, 186–197. Baddeley, A. D. (1990). Human memory, theory and practice. Erlbaum. Baddeley, A. D. (2000a). The episodic buffer: A new component of working memory. Trends in Cognitive Science, 4(11), 417–423. Baddeley, A. D. (2000b). Short-term and working memory. In E. Tulving, & F. I. M. Craik (Eds.), The Oxford handbook of memory (pp. 77–92). Oxford University Press. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. A. Bower (Ed.), Recent advances in learning and motivation (Vol. 8, pp. 647–667). Academic Press. Bartek, B., Lewis, R. L., Vasishth, S., & Smith, M. R. (2011). In search of online locality effects in sentence comprehension. Journal of Experimental Psychology: Learning. Memory and Cognition, 37(5), 178–198. Behaghel, O. (1909). Beziehungen zwischen Umfang und Reihenfolge von Satzgliedern. Indogermanische Forschungen, 25, 110–142. Behaghel, O. (1932). Deutsche Syntax: Eine geschichtliche Darstellung, Bd. 4: Wortstellung; Periodenbau. Winter. Bresnan, J., Asudeh, A., Toivonen, I., & Wechsler, S. (2015). Lexical functional syntax (2nd ed.). Wiley Blackwell. Bresnan, J., Cueni, A., Nikitina, T., & Baayen, H. (2007). Predicting the dative alternation. In G. Boume, I. Kraemer, & J. Zwarts (Eds.), Cognitive foundations of interpretation (pp. 69–94). Royal Netherlands Academy of Science. Bybee, J. (2006). From usage to grammar: The mind’s response to repetition. Language, 82(4), 711–733. Camos, V., & Barrouillet, P. (2011). Factors of working memory development: The time-based resource-sharing model approach. In P. Barrouillet, & V. Gaillard (Eds.), Cognitive development and working memory (pp. 151–176). Psychology Press. Chen, B., Ning, A., Bi, H., & Dunlap, S. (2008). Chinese subject-relative clauses are more difficult to process than the object-relative clauses. Acta Psychologica, 129, 616. Choi, H. W. (1999). Optimizing structure in context: Scrambling and information structure. CSLI Publications. Choi, H. W. (2007). Length and order: A corpus study of Korean dativeaccusative construction. Discourse and Cognition, 14(3), 207–227. Chomsky, N. (1957). Syntactic structures. Mouton. Chomsky, N. (1988). Language and problems of knowledge: The Managua lectures. MIT Press.
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
361
362
CHUNSHAN XU AND HAITAO LIU
Covington, M. A. (2001). A fundamental algorithm for dependency parsing. In J. A. Miller & J. Smith (Eds.), Proceedings of the 39th Annual ACM Southeast Conference, Athens, Georgia, 2001 (pp. 95–102). http://web.stanford.edu/ ~mjkay/covington.pdf Covington, M. A. (2003). Free-word-order dependency parser in prolog. University of Georgia: Artificial Intelligence Center. www .covingtoninnovations.com/mc/dparser/dparser.pdf Cowan, N. (1995). Attention and memory: An integrated framework. Oxford University Press. Cowan, N. (1999). An embedded processes model of working memory. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 62–101). Cambridge University Press. Crabbé, B., Gulordava, K., & Merlo. P. (2015). Dependency length minimisation effects in short spans: A large-scale analysis of adjective placement in complex noun phrases. In ACL Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Short Papers) (pp. 477–482). The Association for Computational Linguistics. den Dikken, M. (2013). Phrase structure grammar. Cambridge University Press Dryer, M. (1992). The Greenbergian word order correlations. Language, 68, 81–138. Engle, R. W., Cantor, J., & Carullo, J. J. (1992). Individual differences in working memory and comprehension: A test of four hypotheses. Journal of Experimental Psychology: Learning, Memory and Cognition, 5, 972–992. Eppler, E. (2004). The syntax of German-English code-switching. (Doctoral thesis, University College London). Ertel, S. (1977). Where do the subjects of the sentences come from? In S. Rosenberg (Eds.), Sentence production developments in research and theory (pp. 141–168). Lawrence Erlbaum. Faghiri. P., & Samvelian, P. (2014). Constituent ordering in Persian and the weight factor. In P. Christopher (Ed.), Empirical issues in syntax and semantics 10 (EISS10). www.cssp.cnrs.fr/eiss10/eiss10_faghiri-andsamvelian.pdf. Fedorenko, E., Piantadosi, S., & Gibson, E. (2012). Processing relative clauses in supportive contexts. Cognitive Science, 36, 471–497. Ferrer-i-Cancho, R. (2004). Euclidean distance between syntactically linked words. Physical Review A, 70, 056135. Ferrer-i-Cancho, R. (2013). Hubiness, length and crossings and their relationships in dependency trees. Glottometrics, 25, 1–21. Ferrer-i-Cancho, R. (2014). Random crossings in dependency trees. arXiv:1305.4561 [cs.CL]. https://arxiv.org/ftp/arxiv/papers/1305/1305.4561 .pdf Ferrer-i-Cancho, R., Gómez-Rodríguez, C., & Esteban, J. L. (2018). Are crossing dependencies really scarce? Physica A, 493, 311–329.
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
Frazier, L. (1979). On comprehending sentences: Syntactic parsing strategies. (Doctoral dissertation. University of Connecticut). Frazier, L. (1985). Syntactic complexity. In D. R. Dowty, L. Karttunen, & A. Zwicky (Eds.), Natural language parsing: Psychological, computational, and theoretical perspectives (pp. 129–89). Cambridge University Press. Futrell, R., Mahowald, K., & Gibson, E. (2015). Large-scale evidence for dependency length minimization in 37 languages. Proceedings of the National Academy of Sciences, 112(33), 10336–10341. Futrell, R., Levy, R. P., & Gibson, E. (2020). Dependency locality as an explanatory principle for word order. Language, 96(2), 371–412. Gibson, E. (1998). Linguistic complexity: Locality of syntactic dependencies. Cognition, 68, 1–76. Gibson, E. (2000). The dependency locality theory: A distance-based theory of linguistic complexity. In A. Marantz, Y. Miyashita, & W. O’Neil (Eds.), Image, language, brain (pp. 95–126). MIT Press. Gibson, E., Piantadosi, S., Brink, K., Bergen, L., Lim, E., & Saxe, R. (2013). A noisy-channel account of cross-linguistic word order variation. Psychological Science, 4(7), 1079–1088. Gibson, E., & Wu, I. (2013). Processing Chinese relative clauses in context. Language Cognition and Neuroscience, 28, 125–155. Gildea, D., & Temperley, D. (2010). Do grammars minimize dependency length? Cognitive Science, 34, 286–310. Givón, T. (2009). The genesis of syntactic complexity. Benjamins. Gómez-Rodríguez, C., & Ferrer-i-Cancho, R. (2017). Scarcity of crossing dependencies: A direct outcome of a specific constraint? Physical Review E, 96, 062304. Grodner, D., & Gibson, E. (2005). Some consequences of the serial nature of linguistic input. Cognitive Science, 29(2), 261–290. Greenberg, J. H. (1963). Some universals of grammar with particular reference to the order of meaningful elements. In J. H. Greenberg (Ed.), Universals of language (pp. 40–70). MIT Press. Haken, H. (1983). Synergetics, an introduction: Nonequilibrium phase transitions and self-organization in physics, chemistry, and biology. Springer-Verlag. Hawkins, J. A. (1994). A performance theory of order and constituency. Cambridge University Press. Hiranuma, S. (1999). Syntactic difficulty in English and Japanese: A textual study. UCL Working Papers in Linguistics, 11, 309–322. Hofmeister, P, Jaeger, T. F., Sag, I. A., Arnon, I., & Snider, N. (2007). Locality and accessibility in wh-questions. In S. Featherston & W. Sternefeld (Eds.), Roots: Linguistics in search of its evidential base. Mouton de Gruyter. http:// citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.141.5440&rep=rep1& type=pdf Hofmeister, P,. & Sag, I. A. (2010). Cognitive constraints and island effects. Language, 86, 366–415.
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
363
364
CHUNSHAN XU AND HAITAO LIU
Hsiao, F., & Gibson, E. (2003). Processing relative clauses in Chinese. Cognition, 90, 3–27. Hudson, R. (1995). Measuring Syntactic Difficulty. http://dickhudson.com/wpcontent/uploads/2013/07/Difficulty.pdf Hudson, R. (1998). English grammar. Routledge. Hudson, R. (2007). Language networks: A new word grammar. Cambridge University Press. Hudson, R. (2010). An introduction to word grammar. Cambridge University Press. Hurford, J. (2012). The origin of grammar. Oxford University Press. Husain, H. S., Vasishth, S., & Srinivasan, N. (2014). Strong expectations cancel locality effects: Evidence from Hindi. PLoS ONE, 9(7), e100986. Jaeger, T. F. (2010). Redundancy and reduction: Speakers manage syntactic information density. Cognitive Psychology, 61, 23–62. Jiang, J. Y., & Liu, H. T. (2015). The effects of sentence length on dependency distance, dependency direction and the implications. Language Sciences, 50, 93–104. Jiang, J. Y., & Liu, H. T. (Eds.). (2018). Quantitative analysis of dependency structures. De gruyter Mouton. Jiang, J. Y., & Ouyang, J. H. (2018). Minimization and probability distribution of dependency distance in the process of second language acquisition. In J. Y. Jiang & H. T. Liu (Eds.), Quantitative analysis of dependency structures (pp. 167–190). Walter de Gruyter. Kauffman, S. (1993). Origins of order: Self organisation and selection in evolution. Oxford University Press. Köhler, R. (1986). Zur linguistischen Synergetik. Struktur und Dynamik der Lexik. Brockmeyer. Köhler, R. (2005). Synergetic linguistics. In R. Köhler, G. Altmann, & R. G. Piotrowski (Eds.). Quantitative linguistics. An international handbook (pp. 760–775). Walter de Gruyter; 2005. Lakoff, G., & Johnson, M. (2003). Metaphors we live by. University of Chicago Press. Langacker, R. W. (2008). Cognitive grammar: A basic introduction. Oxford University Press. Langus, A., & Nespor, M. (2010). Cognitive systems struggling for word order. Cognitive Psychology, 60(4), 291–318. Levy, R., Fedorenko, E., & Gibson, E. (2013). The syntactic complexity of Russian relative clauses. Journal of Memory and Language, 69, 461–495. Levy, R., & Keller, F. (2013). Expectation and locality effects in German verbfinal structures. Journal of Memory and Language, 68, 199–222. Li, W. P., & Yan, J. W. (2020). Probability distribution of dependency distance based on a treebank of Japanese EFL learners’ interlanguage. Journal of Quantitative Linguistics, 28(2), 172–186. DOI: 10.1080/ 09296174.2020.1754611
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
Li, Z., Zhou, J., & Zhao, H. (2009). Cross language dependency parsing using a bilingual lexicon. Proceedings of the 47th Annual Meeting of the Association for Computational Linguistics and the 4th International Joint Conference on Natural Language Processing of the AFNLP (pp. 55–63). DOI: 10.3115/ 1687878.1687888 Lin, Y., & Garnsey, S. M. (2010). Animacy and the resolution of temporary ambiguity in relative clause comprehension in Mandarin. In H. Yamashita, Y. Hirose, & J. Packard (Eds.), Processing and producing headfifinal structures (pp. 241–275). Springer. Lin, Y. W. (2011). Locality versus anti-locality effects in Mandarin sentence comprehension. In Zhuo Jing-Schmidt (Ed.), Proceedings of the 23rd North American conference on Chinese linguistics (NACCL-23) (Vol. 1, pp. 200–214). University of Oregon. Liu, H. T. (2006). Syntactic parsing based on Dependency Relations. Grkg/ Humankybernetik, 47(3), 124–135. Liu, H. T. (2007). Probability distribution of dependency distance. Glottometrics, 15, 1–12. Liu H. T. (2008). Dependency distance as a metric of language comprehension difficulty. Journal of Cognitive Science, 9(2), 159–191. Liu, H. T. (2018). Language as a human-driven complex adaptive system. Physics of Life Reviews, 26–27, 149–151. Liu, H. T., Xu, C. S., & Liang, J. Y. (2017). Dependency distance: A new perspective on syntactic patterns in natural languages. Physics of Life Reviews, 21, 171–193. Liu, H. T., Zhao, Y. Y., & Li, W. W. (2009). Chinese syntactic and typological properties based on dependency syntactic treebanks. Poznań Studies in Contemporary Linguistics, 45(4), 509–523. Lovett, M. C., Reder, L. M., & Lebière, C. (1999). Modeling working memory in a unified architecture: An ACT-R perspective. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 135–182). Cambridge University Press. Lu, Q., & Liu, H. T. (2015). Does dependency distance distribute regularly? The Journal of Zhejiang University (Humanities and Social Science), 4, 63–76 (in Chinese). Lu, Q., & Liu, H. T. (2016). A quantitative study on the relationship between crossing and distance of human language. Journal of Shanxi University (Philosophy and social Sciences), 39(4), 49–56 (in Chinese). Lu, Q., Xu, C. S., & Liu, H. T. (2015). The influence of chunking on dependency crossing and distance. Complexity, 21, 33–41. McDonald, R., Crammer, K., & Pereira, F. (2005). Online large-margin training of dependency parsers. Proceedings of ACL 2005, 91–98. https:// www.aclweb.org/anthology/P05-1012.pdf Mel’čuk, I. (2003). Levels of dependency in linguistic description: Concepts and problems. In V. Agel, L. Eichinnger, H. W. Eroms, P. Hellwig,
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
365
366
CHUNSHAN XU AND HAITAO LIU
H. J. Herringer, & H. Lobin (Eds.), Dependency and valency. An international handbook of contemporary research (Vol. 1, pp. 189–229). De Gruyter. Nairne, J. (1990). A feature model of immediate memory. Memory & Cognition, 18, 251–269. Neath, I. (2000). Modeling the effects of irrelevant speech on memory. Psychonomic Bulletin & Review, 7, 403–423. Nivre, J., & Scholz, M. (2004). Deterministic dependency parsing of English text. Proceedings of the 20th International Conference on Computational Linguistics. www.aclweb.org/anthology/C04-1010.pdf Oberauer, K. (2002). Access to information in working memory: Exploring the focus of attention. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(3), 411–421. Oberauer, K., & Lewandowsky, S. (2013). Evidence against decay in verbal working memory. Journal of Experimental Psychology: General, 142(2), 380–411. Oya, M. (2013). Degree centralities, closeness centralities, and dependency distances of different genres of texts. Selected Papers from the 17th Conference of Pan-Pacific Applied Linguistics; 2013 (pp. 42–53). www .paaljapan.org/conference2012/pdf/006oya.pdf Pollard, C., & Sag, I. A. (1994). Head-driven phrase structure grammar. University of Chicago Press. Quirk, R., Greenbaum, S., Leech, G., & Svartvik, J. (1972). A grammar of contemporary English. Longman. Saussure, F. (1959). Course in general linguistics. Philosophical Library. Temperley, D. (2007). Minimization of dependency length in written English. Cognition, 105, 300–333. Temperley, D. (2008). Dependency length minimization in natural and artificial languages. Journal of Quantitative Linguistics, 15, 256–282. Temperley, D., & Gildea, D. (2018). Minimizing syntactic dependency lengths: Typological/cognitive universal? Annual Review of Linguistics, 4, 67–80. Tesnière, L. (1959). Eléments de la syntaxe structurale. Klincksieck. Tily, H. (2010). The role of processing complexity in word order variation and change (Doctoral dissertation, Stanford University). Ueno, M., & Polinksy, M. (2009). Does headedness affect processing? A new look at the VO–OV contrast. Journal of Linguistics, 45, 675–710. VanDyke, J. A. (2007). Interference effects from grammatically unavailable constituents during sentence processing. Journal of Experimental Psychology: Learning, Memory and Cognition, 33(2), 407–430. VanDyke, J. A., & Lewis, R. L. (2003). Distinguishing effects of structure and decay on attachment and repair: A retrieval interference theory of recovery from misanalyzed ambiguities. Journal of Memory and Language, 49(3), 285–316.
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
Shaping Syntactic Dependency Structures
Vasishth, S., & Lewis, R. L. (2006). Argument-head distance and processing complexity: Explaining both locality and anti-locality effects. Language, 82 (4), 767–794. Vergauwe, E., & Cowan, N. (2015). Theories of short-term memory. In J. D. Wright (Ed.), International encyclopedia of social & behavioral science (2nd ed., vol. 21, pp. 901–908). Elsevier. Wang, H., & Liu, H. T. (2014). The effect of length and complexity on constituent ordering in written English. Poznań Studies in Contemporary Linguistics, 50(4), 477–494. Wang, Y. Q., & Liu, H. T. (2017). The effects of genre on dependency distance and dependency direction. Language Sciences, 59, 135–147. Wasow, T. (1997). End-weight from the speaker’s perspective. Journal of Psycholinguistic Research, 26, 347–361. White, K. G. (2012). Dissociation of short-term forgetting from the passage of time. Journal of Experimental Psychology: Learning, Memory and Cognition, 38, 255–259. Xiang, Y., Agnieszka, F., & Jonas, K. (2019). Dependency length minimization vs. word order constraints: an empirical study on 55 treebanks. In X. Y. Chen & R. Ferrer-i-Cancho (Eds.), Proceedings of the First Workshop on Quantitative Syntax (Quasy, SyntaxFest 2019) (pp. 89–97). Association for Computational Linguistics. Xu, C. S. (2015). The use and the omission of Chinese conjunction “er.” Journal of Shanxi University(Philosophy and Social Sciences Edition), 38(2), 55–61. (in Chinese) Xu, C. S. (2018). Differences between English subject post-modifiers and object post-modifiers: From the perspective of dependency distance. In J. Y. Jiang, & H. T. Liu (Eds.), Quantitative analysis of dependency structures (pp. 261–76). Walter de Gruyter. Xu, C. S., & Liu, H. T. (2015). Can familiarity lessen the effect of locality? A case study of Mandarin Chinese subjects and the following adverbials. Poznań Studies in Contemporary Linguistics, 51(3), 463–486. Yamada, H., & Matsumoto, Y. (2003). Statistical dependency analysis with support vector machines. In Proceedings of the Eighth International Conference on Parsing Technologies (pp. 195–206). www.aclweb.org/anthology/ W03–3023.pdf Yamashita, H., & Chang, F. (2001). Long before short preference in the production of a head-final language. Cognition, 81, B45–B55. Yan, J. W., & Liu, H. T. (2019). Which annotation scheme is more expedient to measure syntactic difficulty and cognitive demand? In X. Y. Chen, & R. Ferrer-i-Cancho (Eds.), Proceedings of the First Workshop on Quantitative Syntax (Quasy, SyntaxFest 2019) (pp. 16–24). Association for Computational Linguistics. Zipf, G. (1949). Human behavior and the principle of least effort: An introduction to human ecology. Hafner.
https://doi.org/10.1017/9781108955638.020 Published online by Cambridge University Press
367
17 Working Memory in the Modular Cognition Framework John Truscott and Michael Sharwood Smith 17.1
Introduction
Working memory (WM) has generated an enormous amount of empirical research, including research in the development and use of language. A common limitation of such work is that it is not clearly connected to any coherent conception of the cognitive system as a whole. While there is no uncontroversial model of the mind, reasonably well-established frameworks do exist, making it possible to address the issues within this broad context. In this chapter we apply one such account, the Modular Cognition Framework (MCF), to the study of WM and its place in language use and language development. We begin with a brief introduction to the framework, pointing out the difference between a theoretical framework and a theory, then spell out the nature of WM and working memory capacity (WMC) within it. We then apply this view to the study of language development and language use. The discussion is broad ranging, and so some superficiality is unavoidable. For more in-depth discussion, readers are invited to consult the various MCF publications cited, especially Truscott and Sharwood Smith (2019) and Truscott (in press).
17.2
The Modular Cognition Framework (MCF)
The Modular Cognition Framework began development in 2000. Its name reflects the need to explain the role of language not only in its own terms but as an instantiation of principles that hold for the mind as a whole. As a theoretical framework it is designed to be relatively open-ended, allowing researchers in any area of cognitive science to elaborate it within their own field and in accordance with their own approach, provided it is compatible with the overall architecture. This architecture already incorporates insights and findings from research that can be integrated into its wider
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
perspective. The framework has thus far been elaborated, at least for the most part, in the areas of language acquisition, processing, and representation. Frameworks in this sense can vary considerably in their scope. For example, theoretical linguistic frameworks such as the Minimalist Program, itself a version of a wider theoretical (generative linguistic) approach, can accommodate sometimes sharply different approaches that still conform to its basic principles and assumptions (Chomsky, 1995). The scope of the MCF is, as just mentioned, much wider, covering cognition in general. In that MCF-applications thus far have concentrated on language, it might be compared to competing approaches such as the Three Graces, complex adaptive systems framework as set out by Ellis and associates and which itself draws on an even broader set of principles (Beckner et al., 2009). Nevertheless, this present discussion on working memory is part of the MCF as a framework for any area of cognition. As such it is comparable to ACT (Adaptive Control of Thought) as developed by Anderson (1983). Frameworks typically rest on many different sources of evidence. Although they themselves cannot be tested directly, they are indirectly vulnerable as they represent an integration of a range of hypotheses and empirical studies that are each open to challenges and testing. Not all broad frameworks that happen to be available are fully and explicitly incorporated into studies that target specific areas of cognition. We plan to show how actively exploiting MCF principles in order to elaborate on the notion of WM is a useful and promising way of advancing a general understanding of related phenomena, and here specifically with regard to language.
17.2.1 Architecture The MCF assumes a so-called modular architecture with the mind conceived as a set of functionally specialized systems that operate both independently and together, forming an interactive network. This reflects a view that is widely held in contemporary cognitive science (Barrett & Kurzban 2006; Bergeron, 2007; Carruthers, 2006) We shall be referring to them variously as “modules” or simply as “systems.” More specifically, although each of the modular systems in this network has a unique role in the mind as a whole, they all have a basic design and mode of operation in common. Each has a processor incorporating the unique principles that determine the composition of its representations, that is, the “code” in which they will be “written,” and each has a store where all of its particular type of representation reside. The store will be crucial in accounting for the nature and operation of working memory. The basic design of a processor and store is possible because this serves accounts of how the mind works. How they are instantiated in neural architecture is of course another matter, but it is important that research into the two areas, mind and brain, does not diverge too far and hinder meaningful relationships from being made between the two (Sharwood Smith, 2014). The way in which different types
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
369
370
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
of representation are distinguished, however, is also represented in the brain, although in this case each involves different physical locations and neural pathways. The precise number of systems is not absolutely fixed but certainly includes, apart from the two that handle linguistic structure (a) five (sensory) perceptual systems, the auditory and visual systems being the most relevant in this context, (b) the conceptual system, handling abstract meanings, (c) the affective system handling positive and negative value representations and those representing the basic emotions, and (d) the motor system, which is, for instance, required for explaining how language is articulated during production.
17.2.2 Representation Each store contains representations written in the specific code of the system (module) in question. Hence auditory representations would be written in auditory code as proposed by those working in auditory cognition. Linguistics already supplies a wealth of candidates for phonological and syntactic code, obvious uncontroversial syntactic examples being N (oun) and V(erb). Representations are described in the MCF as “structures” so that, for example, the abbreviation for an auditory representation is AS (auditory structure). Each store will have a set of biologically determined primitives, which are used to create the complex representations that accumulate during the individual’s lifetime according to that individual’s experience and the principles of the processor. Linguistic structures fall into two categories, phonological structures and syntactic structures. (Jackendoff, 1987, 1997). In the current setup, this takes care of morphology as well. In other words, what is sometimes called the language faculty is reflected in the MCF as the contribution of two independent systems and two types of representation for which we will reserve the term “linguistic,” namely, PS and SS. These systems that separately specialize in the processing of linguistic structure are in fact two of many that are involved in the processing of language in its broadest sense. Representations written in different codes may nevertheless become associated, and as will be described in the following section, coactivated in working memory. In this way, a generic sound representation (AS) may be associated with a phonological representation (PS), in other words, a speech structure.1 The creation of multiple networks of associated representations ranging across different stores is enabled via connecting interfaces. Representations thus associated have a shared index. This largely reflects the system developed by Jackendoff for language (Jackendoff, 1997), although there are some differences distinguishing the two perspectives. For example, we assign interfaces a smaller and somewhat different role, adopt a different view of consciousness (including recasting phonological awareness as auditory awareness), hypothesize unified perceptual output stores, offer a novel
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
view of learning, and, especially, treat working memory differently. In the MCF, representations are created via processing so a comprehensive description of their properties must also be in terms of processing principles where working memory plays a crucial role.
17.2.3 Processing Processing in the MCF has a local and a global dimension depending on whether the focus of interest is on what happens within particular stores or what happens between all the stores currently involved. As will be explained in more detail in following sections, the activation of structures in a given store puts them into a state called working memory (WM). Once in this WM state they will coactivate any representation with which they happen to be currently associated, that is, “coindexed.” When not being in WM, and thus defined, they are said to be at their resting level of activation. In the MCF, the resting level of a representation will change the more or the less frequently it is activated. An increase in resting level means that it will be more “accessible.” “Access” is one of a number of common metaphors to describe the relative likelihood of a representation participating in the execution of some processing operation. When candidate structures compete in WM, their resting level will influence the outcome, that is, which of them will ultimately turn out to be winners or losers. Since “level” itself is a metaphor suggesting relative height, one can describe resting levels of activation of representations in a store as being lower or higher so that when they enter into WM, the lower representations will normally be handicapped in any competition (e.g., see Sharwood Smith & Truscott, 2014, 75–78). Typically, newly created representations begin at a low level. The frequency, regularity, and degree of activation will determine how their resting level will increase or decline over time. In this way the accessibility of a representation simply depends on its activation history and does not depend on any notion of some mental entity literally “accessing” or “selecting” it (Truscott & Sharwood Smith, 2019, pp. 42–46). We return to this issue below. At the local level, within stores, activated representations will be combined in WM during online processing into larger representations according to the principles of the relevant processor. At a global level, associated representations will be coactivated, forming networks of representations across stores. Such networks, when frequently activated, are also called representational “schemas.” In other words, WM activity triggered within given stores – the phonological store, for example – will feature the coactivation of various individual phonological representations: the phonological processor will organize these into more complex configurations following its phonological principles. Parallel processing ensues: activated phonological representations will automatically coactivate associated
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
371
372
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
representations in other stores where similar activity takes place. In this way, whole schemas are activated in WM involving a number of stores with all representations in current WM conforming to the principles of their own processors. Processing is both parallel and incremental. Note that representations in the richly interconnected perceptual systems are characterized by their particularly high levels of activation when in WM. These extreme levels are manifested locally in the WM of their respective stores but also globally in the networks that link them with one another. This is something we will return to in our discussion of global working memory. In language processing, a distinction is made between what happens locally in the two linguistic systems and what happens globally as many other types of representation are coactivated as well. A simple example was given above as a generic sound representation (AS) is coactivated with a linguistic representation (PS). This PS will typically coactivate another type of linguistic representation (SS) giving a chain of structures activated in parallel as when the sound of the word “pin” coactivates an auditory structure (AS) and a PS /pin/ and an SS Nounsingular thus: AS , PS , SS The auditory structure may already be associated directly with a meaning, that is, a conceptual representation (the CS PIN), expanding the chain as follows: AS , PS , SS , CS The possibility of a direct AS(,)CS sound/meaning connection would convert what was a sequential chain of structures into a loop, thus already making what represents the word “pin” a small network without even considering a number of other types of representation that might be included. Nevertheless, the outcome of strictly linguistic processing that takes place in the working memories of the PS and SS systems can only be fully explained by also taking into account the nonlinguistic context taking place in working memories elsewhere. In the MCF, this nonlinguistic context is more precisely defined as the internal context of linguistic processing (Truscott & Sharwood Smith, 2019). That is to say, linguistic processing continually provides PS and SS associations that are internally consistent. However, which of these options survive in the end will depend on current activity in other, that is, nonlinguistic stores (“the internal context”). This will render the activated linguistic options more or less appropriate and more or less valued. In the end it is the overall fit across many WMs that matters. Internal context falls into two types. The outcome of linguistic (AS and SS) processing is particularly influenced by representations of goals in CS and value and emotion in affective structures. However, these representations are not directly involved in the first type of context, the mind’s own constructed version of the external environment: this type is termed the
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
“outside-in” internal context. But when coactivated in their respective WMs they belong to the other type, the “inherently” internal context (Truscott & Sharwood Smith, 2019, pp. 55–56, 81–99). These two types combine to form the total internal context influencing linguistic processing.
17.3
Working Memory in the Framework
Working memory, according to an increasingly accepted view, is not a location, a place where things are temporarily stored, but rather the set of currently active representations (e.g. Cantor & Engle, 1993; Cowan, 1995; D’Esposito & Postle, 2015; McElree, 1998; Oberauer, 2013). To say that a representation is “in working memory” is simply to say that it is active. In this section we wish to develop this view in the context of the Modular Cognition Framework. Subsequent sections will explore the implications for working memory in language learning and language use. Working memory, as it is generally understood, developed from the idea of short-term memory (STM), the temporary storage (or activation) of currently relevant items. But WM is more than storage/activation. Baddeley (e.g., 2007), for one, has always emphasized its active nature, that it is a set of items plus the use of those items. This is a rather murky notion, though. It could be taken to mean that everything that goes on in the system is WM, that a theory of WM is indistinguishable from a general theory of the cognitive system. So, clarification is needed. What exactly is the relation between activation of representations and their use in processing? Clarification begins with the concept of activation, which can serve as a bridge between working memory and processing, since the activation level of a representation is commonly defined as its availability for processing. More generally, though, this is a question about the cognitive system as a whole, so a serious answer requires an explicit framework specifying the architecture of the system and the nature of representation and processing within it – a framework like the MCF. One benefit of using such a framework is that it permits a more fine-grained analysis of what WM actually involves, including its precise relationship to the concepts of activation and consciousness. Discussion has traditionally assumed that an item either is or is not in working memory. But activation and availability are a matter of degree, so “in WM” cannot be a yes-or-no matter. The dichotomy is an abstraction that is useful for research but does not express the actual character of the system. What is usually taken as WM is the most active representations, associated particularly with conscious processing. This is a natural focus for research, as the ability to state a test item provides a convenient measure of memory. But this is the extreme case of availability and activation, not the definition. Most of what goes on in the system is not conscious but is processing nonetheless. It is simply more local and specialized and involves
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
373
374
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
lower activation levels, making it unconscious (see below). The idea of WM as making things available for processing thus logically implies that WM is not just those representations that move in and out of consciousness but is rather much more general. The biggest distinction in degree of activation is found in the contrast between local processing, in the individual expert systems, and global processing that constitutes the interaction among the various systems. We will consider these two types of processing in the two following sections.
17.3.1 Local Working Memory Processing takes place within individual expert systems – locally, in other words. Interactions among the various systems are also of great importance, and this is the topic of the following section. Here our concern is with WM as a local phenomenon, a feature of the individual systems. Local WM is simply active representations in the system in which the processing is taking place. Activation can come from a variety of sources, including coindexed representations in neighboring modules and representations within the local system that share features with one another, allowing activation to spread between them. At any given moment the local processor is using some of these active representations, typically the most active ones, to construct a new representation. Those that are active but not currently in use constitute the local context of the processing. This describes a constantly shifting scene. A representation typically is not a continuous part of processing for any length of time, but is quickly replaced in that role by other representations. Once it has ceased to be part of processing, it does not immediately leave WM, as the decline in activation level that accompanies nonuse is a gradual process. Thus it will continue for a time to influence processing from the sidelines and will be relatively available for further use by the processor. But if it receives no further stimulation, it will gradually disappear from WM. Other representations that do receive stimulation in that period will enter WM and then possibly participate for a time in processing. All of this is working memory – active representations and their use in processing. In language production, SS representations are activated by the activity of coindexed CS representations, which are active as part of the message to be expressed, or as the context of that message. Additional SS activity will result from spreading activation within the store. Perhaps the most important case involves the argument structure of a verb, which is a complex SS representation, consisting of V and its arguments (Sharwood Smith & Truscott, 2014). The SS of the verb tell, for example, is included in the representation [V NP NP], which is activated whenever the CS TELL is active. Activation of the frame can result from spreading activation from the verb, but also from additional CS activity. The concept TELL is associated with the concepts of someone doing the telling, something being told, and someone
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
receiving the information. Together they make up a conceptual frame corresponding to the syntactic frame, which is therefore coactivated with it. But CS activation also spreads within the store based on shared components. If the message includes the concept of verbally expressing an idea, for example, all the verbs that include this concept will be activated to varying degrees: tell, talk, say. . . Their frames in CS and SS will therefore also become active. All these conceptual (CS) representations constitute conceptual WM at that point in time. Thus, during language production, a great many SS representations are active to varying degrees, varying both among themselves and over time. At a given moment the syntax processor is using some of these active representations. The others constitute the local internal context in which the processing is occurring and are readily available for use by the processor. All of these active representations together constitute syntactic WM. The same principles apply to PS and in fact to all modules.
17.3.2 Global Working Memory The concept of local working memory is not a standard part of thinking in the literature. Researchers do however speak of central and peripheral components of WM (see Cowan et al., 2014), the former referring to executive processes, particularly attention, and the latter to domain-specific elements, which can be associated with local WM. A major difference from our proposal is that work in this area typically focuses on high-level processing, associated with conscious experience, while we also consider what is going on specifically at the local, peripheral level and is typically not conscious. Our approach thus recognizes that information availability is a much more general phenomenon than is typically recognized, making possible a deeper understanding and investigation of WM. This deeper view notably includes the factors underlying WM capacity, which cannot be understood without consideration of the local processing and representation. In this section we look at the central aspect, seeing it as a product of the local WMs. Working memory as it is commonly understood is, again, about activation. Executive factors activate peripheral elements and thereby make them the focus of processing. In the MCF approach, this is about a broad synchronization of processing brought about by the factors of internal context. Inherently internal context constitutes the executive element, but its influence must be understood within the context of the situation as it is internally represented. Consider what happens when we do mental arithmetic, such as this simple multiplication problem: 13 8
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
375
376
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
The processing is centered on CS, where concepts of numbers and rules for manipulating them are found, but is set up by construction of a perceptual representation of the problem, essentially the image shown above. The perceptual representation becomes the focus of the synchronization because of an active goal representation in CS, supported by active connections with value – if the person does not care about the problem, then the goal representation will have minimal activation and will not be able to guide processing. The multiplication process begins with the two immediately relevant numbers, 8 and 3. Each is a schema consisting of an abstract concept in CS (EIGHT and THREE) plus visual and auditory representations of the numbers in VS and AS, respectively, along with PS and SS representations of the words eight and three, and motor representations of the routines for writing them. Each of these representations is in its local WM. The global synchronization is focused on one of the perceptual representations (perhaps rapidly alternating between visual and auditory), giving it an extremely high activation level, which is to say it is in global WM. Substantial activation spreads to the other members of the schema. The CS processor then performs the multiplication using EIGHT and THREE, and so activates the representation TWENTYFOUR, which leads to activation of the entire 24 schema, with a focus of processing on the visual and/or the auditory representation. TWO and FOUR are then attached to the original representations of the problem, with the visual representation probably dominating. This is the experience of mentally writing a little 2 above the 1 and a 4 below the 8. This new representation becomes the focus of processing during its construction and retains considerable activation afterward, remaining available for processing in the next step. For this step, the focus shifts to 8 and 1. When this multiplication is complete, yielding EIGHT, the focus returns to the overall representation of the problem and the TWO that was attached to it in a previous step. The remainder of the task follows the same principles, so we will leave it at that. At each step, each of the relevant representations is in its local WM, either in use by the processor or serving as the local context and remaining available for processing. This continuous interplay of processing and its context is the essence of WM. Global WM has a crucial place in such interplay, giving especially high activation to the schema that is to be the current focus of processing.
17.3.3 Working Memory, Consciousness, and Attention In the multiplication example, the representations in global WM, at each point, are those that we are conscious of. This is not accidental. The MCF view of consciousness is expressed in the Activation Hypothesis (see Truscott, 2015a, 2015b):
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
A representation is the object of consciousness if and only if it has an extremely high current activation level. Setting aside affective awareness, these extreme levels of activation are found specifically in the perceptual stores. When activity in the various perceptual stores becomes synchronized, in harmony with activity in conceptual structures and affective structures, the focus of the synchronization becomes very highly active and therefore becomes the object of awareness. The association with global working memory should be clear. Global WM consists of the representations that are extremely active, as a result of the global synchronization. Thus conscious experience coincides with global WM, which corresponds to the standard idea of working memory. The multiplication example made no mention of attention as such, though it is commonly seen as the key to working memory and the gateway to consciousness (Baars, 1988; Dehaene & Naccache, 2001; Jackendoff, 1987; Posner, 1994). The problem with attention is that it is a loose notion, an umbrella term for a variety of processes related to the focus of processing, and so is best understood in terms of these various processes (e.g., Driver et al, 2001; Krauzlis et al., 2014; Nobre & Mesulam, 2014). In MCF, they are the factors of internal context, primarily those of inherently internal context. To say that we attend to something is to say that the combined influences of these factors has resulted in a focus of processing on that particular representation; in other words, that this representation is the current content of the global WM.
17.3.4 Some Thoughts on Neural Correlates We will conclude this section with some brief and somewhat speculative thoughts on neural mechanisms underlying WM as it is understood here. More detailed discussion can be found in Truscott and Sharwood Smith (2019) and Truscott (in press). Representations and schemas are neural circuits, which is to say groups of neurons that tend to fire together. The difference between them is that a representation, necessarily found within a given module, is likely to correspond to a local circuit, whereas a schema, consisting of several representations in different modules, will typically extend across the brain more widely. Goals are to be found in the prefrontal cortex (PFC), particularly its lateral areas (see especially Fuster, 2015). Orbitomedial PFC represents value and emotion in their role as controllers; in MCF terms, it is part of affective structures. It receives projections from lower parts of the brain where these more basic elements are rooted, especially via anterior cingulate cortex. These control areas, in conjunction with perceptual integration areas in the parietal cortex, selectively activate representations that are mostly stored in posterior regions (see Nobre & Mesulam, 2014). These
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
377
378
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
can be associated with the various memory stores. This activation is then working memory. The notion of activation can serve as a bridge between cognitive and neural levels, as it is a central notion for both. The function of a processor is to construct a single representation from representations that are currently active in its store – in its WM, in other words. In neural terms this means producing a local synchronization of activity, since the change from a number of independently active representations to a single active representation constructed from them represents a synchronization. Gamma oscillations, modulated by theta oscillations, are a hallmark of local processing (Womelsdorf et al., 2007) and might be taken as an expression of the local synchronization. An important feature of local synchronization is that it greatly enhances the influence of the local processing on activity elsewhere in the system (see Miller & Buschman, 2014). In other words, when a representation dominates a local WM it can strongly affect the WMs of other stores, particularly when the dominant representation is strongly connected to representations in the other stores; that is, when it is part of a schema. This interareal communication is associated with the gamma oscillations expressing the local synchronization (Palva et al., 2010) but more specifically with lower frequency oscillations (Bastos et al., 2015; Palva et al., 2010). Its ultimate form is a broad interareal synchronization. Cognitively, this is the expression of a schema becoming dominant, which is to say entering global WM. This discussion is sketchy and somewhat speculative. Its aim is to suggest possibilities and raise potentially interesting questions regarding the crucial issue of how cognitive and neural explanation can be reconciled. We believe the “neural friendliness” of the MCF is a strength of the framework.
17.4
Working Memory Capacity
Working memory capacity (WMC) has been extensively studied in experimental psychology (for review, see Oberauer et al., 2016). Based on that work, it has also become a prominent topic in second language acquisition (e.g., Juffs & Harrington, 2011; Szmalec et al., 2013; Wen, 2014; Wen et al., 2015; Williams, 2012).
17.4.1 What Is Working Memory Capacity? Working memory is made up of both local and global activity, so an understanding of working memory capacity must take both into account. We can thus speak of local capacity and global capacity, paralleling the more common distinction between peripheral and central capacity (Cowan et al., 2014). Observed limits on performance are then the combination of these factors.
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
Peripheral (local) capacity limits are commonly understood to be the product of interference, different processes needing to use the same representations in conflicting ways (Cowan et al., 2014; Oberauer et al., 2016). Explaining exactly how this occurs requires specific theories of the individual expert systems. There cannot be a simple principle that processing is limited to only one use of a given representation at a time, as shown by sentences like “People need people,” in which a representation of “people” must be used twice at each level – phonological, syntactic, and conceptual. The issue is how each processor is and is not able to bind particular kinds of representations to make larger representations. In any case, local limits on activity within a module must be taken into account in an analysis of WMC. Central capacity limits are commonly understood in terms of attention and its inherent limitations (Cowan et al., 2014; Oberauer, 2013). In the MCF, attention, as described above, is to be explained in terms of internal context, which selectively activates representations and maintains those elevated activation levels. Central (global) capacity limits exist because these influences work by producing a broad synchronization of activity around one particular representation. This is by nature a serial process. One representation is the focus and therefore has the highest activation level, while those that were recently the focus retain a substantial if lower degree of activation. Such a process cannot simultaneously raise an unlimited number of representations to extreme activation levels.
17.4.2 What Is Measured by Tests of Working Memory Capacity? Working memory tasks (see Daneman & Carpenter, 1980; Turner & Engle, 1989), like any other kind of task, involve a number of distinct, interacting systems. To understand and interpret measures of WMC, we must take all these factors into account. Thus, it is not possible to have a genuine understanding of WMC, or to know exactly what the results of a WMC test mean, without placing it within a specific account of the cognitive system, in other words, a framework like the MCF. A linguistic WM task necessarily makes use of AS (VS for written tasks) for processing of sensory input, and PS, SS, and CS for linguistic processing. Chains of representations must be kept active in these stores, long enough for the motor system to report them at the specified time. Thus the local WM of each of these modules must be considered. Interactions among the various WMs can affect the ability of each to keep a representation active. Activation levels within each module are also influenced, directly or indirectly, by the factors of internal context. Goals in CS, specifically the goals of remembering and reporting the test items, must be active. Affective activity, variably reinforcing the activation of the representations of both the target items and the goals, is also relevant. These factors also enter, crucially, by virtue of their role in global working memory.
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
379
380
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
The strictly linguistic systems process linguistic input in an automatic manner. They are also relatively isolated from outside influences and, partly for that reason, have relatively low activation levels in general. So the local WMs here should have little relevance to measures of WMC. More important are the local WMs of AS and CS. These themes will be pursued in more depth in the following sections.
17.5
Working Memory in Language Development
It is generally accepted that two distinct types of linguistic knowledge exist. In the framework the distinction is between (1) knowledge based in the linguistic modules, PS and SS and (2) metalinguistic knowledge, centered in CS. The initial character and function of SS and PS are specified by innate linguistic knowledge often referred to as universal grammar (UG), however that may be defined more specifically. The original basis for the UG assumption is the universal success of small children in mastering the extreme complexity of a language (or, often, languages) combined with the inevitable limitations in the input they receive and the lack of instruction. This success occurs in strikingly similar ways under a wide variety of conditions, making language development look more like learning to walk than like figuring out a challenging intellectual problem. This impression is reinforced by the general absence of a relation between intelligence and language development. Children with extremely low intelligence can acquire language with no particular difficulty, while children with no general intellectual weakness can encounter serious problems – problems that have a clear genetic basis. Further support for an innate linguistic endowment comes from the spontaneous development of creoles and sign languages, expressing a human “language instinct,” and the consistent failure of other species to acquire anything like a human language. For discussion and references, see Sharwood Smith and Truscott (2014) and Truscott (2015a, in press), among countless others. The idea behind innate knowledge in general is to greatly narrow the possible analyses of input, telling the system what things to look for and what to do with the things it finds. Without such guidance, the system is faced with an essentially infinite number of possible analyses for any given input, and learning becomes literally impossible. The notion of UG is often perceived as being an oddity in explaining cognition in general. On the contrary, in the MCF the idea is reflected in every system since each processor has its own set of principles for dealing with its particular input. In other words, the idea of UG is one manifestation of a biological norm. The implication is that the development of the specialist linguistic systems is the heart of language development and that the process is strongly guided by their in-built nature. But we also have to recognize that, in addition to such specialized learning, people have a general capacity to
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
acquire knowledge on essentially any topic, language being no exception. This nonspecialist learning cannot be expected to approach the success routinely achieved by the specialist system, but is nonetheless important. In this section we will address both types. In the MCF, learning of all types is based on the principle of Acquisition by Processing: Learning is the lingering effect of processing. Learning takes two forms. First, when a new representation is created, for the purpose of representing current input, it will linger in the store afterward. Second, after an existing representation is used in processing its resting activation level will be slightly higher as a result of this use. Note that both cases involve active representations. In other words, learning is about representations in WM. A representation’s presence in WM allows its use in the creation of new representations and assures it of a somewhat elevated resting level afterward. Each active representation is, by definition, in a local WM, but the extreme activation levels of global WM are important for learning, as will be seen below.
17.5.1 Development of the Linguistic Modules SS and PS are specialized for processing linguistic input, so they can typically deal with their input in an automatic manner. Given acquisition by processing theory, the implication is that the development of the SS and PS stores is also largely automatic. The processors simply follow specialized in-built routines to construct new representations from whatever is currently active in their store. This does not mean the resulting representations are the final, correct version of the grammar. Development requires very extensive input processing, inevitably resulting in continual changes in the contents of the stores. But each instance will be a quick, automatic process, with no demands on the global WM. The situation is more complex for the involvement of perceptual and conceptual modules because their scope is vastly greater and their processing is not nearly so constrained – they did not develop specifically to handle language. Their participation in development of the linguistic modules consists of providing input to those modules in useful forms, in terms of sound and meaning. In comprehension, PS automatically processes input from auditory structures (AS), based on its in-built principles, and SS automatically processes input from PS, based on its in-built principles, the ultimate result being a CS representation of the meaning of the input, as that input was represented in AS. In order for development to occur, the requirement is that the relevant perceptual representations must be present long enough and actively enough for PS to use them, that is, for appropriate PS representations to be activated, as part of local phonological WM. If this basic requirement is met, there is not much demand on WMC, as it is commonly measured.
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
381
382
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
If, on the other hand, the input representation is not sufficiently clear or sustained – due perhaps to an unfamiliar accent, unclear speech, background noise, or lack of attention – then additional processing is needed. This could mean reactivating the perceptual representation in order to give PS another chance. It could also mean conceptual analysis to reconstruct the representation in a way that is interpretable. In such efforts to construct an AS representation that PS can deal with, global WM is likely to be involved, and so WMC becomes relevant. Problems that result in further processing and a possible burden on WMC can also appear in the construction of the CS representation constituting the meaning of the input. When appropriate CS representations are active, based on context and existing linguistic knowledge, they can be automatically connected to SS representations and an adequate (if perhaps limited or flawed) representation of the meaning might result. In this case there is no involvement of global WM, and WMC is largely irrelevant. But the automatic processing might not yield an adequate CS representation, due to the limited current development of the linguistic modules and/or of SS connections to CS. In this case further processing may occur, with the goal of understanding the input. This amounts to thinking about the sentence, trying to make sense of it – a process that is likely to involve global WM and therefore to make WMC relevant. One aspect of this thinking is reconstruction of the input AS representation or perhaps some particularly challenging part of it. We should stress again that an automatic (or nonautomatic) construction of a representation or a connection between representations need not, and typically does not, produce a final, correct outcome. Development of representations and connections that are inconsistent with the target grammar is a routine part of learning, to be overcome by extensive additional processing experience.
17.5.2 Development of Metalinguistic Knowledge Metalinguistic knowledge, that is, the second type of linguistic knowledge, referred to above as a type of nonspecialist knowledge, is knowledge about language and is comparable to knowledge about any other topic that is acquired during the lifetime at school or more informally. Its acquisition requires a certain level of cognitive maturity so that preschool children already in control of one or more language systems nevertheless have limited knowledge of this kind. Multilingual children in particular may become metalinguistically aware at an early age, although their knowledge is not systematic and formalized. Metalinguistic knowledge is processed in the conceptual system and is therefore composed of CS representations, but these representations are typically part of a schema that includes perceptual representations. The knowledge that -s is added to English third-person singular present tense
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
verbs, for example, is conceptual but is connected to visual representations of the affix and its use, taking various possible forms, as well as auditory (AS) representations of the spoken form of the rule. If the conceptual representation attains a high enough activation level, through practice and/or affective support (the person cares about it), then the activation of the schema’s perceptual element(s) can cross the consciousness threshold and the schema as a whole can participate in global WM. We then have conscious language processing. Since such processing depends on global synchronization, which is by nature a serial process, strict limits exist on the use of metalinguistic knowledge to control ongoing language production in order to consciously monitor and correct utterances on the fly. PS and SS representations by contrast always operate efficiently at lower, nonconscious levels and, unlike the phonological and syntactic processors, the conceptual processor has no special ability to construct specialized linguistic representations locally, in conceptual WM. The conceptual system can only construct representations, for example, metagrammatical representations, that allow for the possibility of reflections about various aspects of language and more generally about all other objects of thought. Acquiring metalinguistic knowledge is largely a deliberate, controlled process engaging global WM. The growth that takes place in the conceptual system will have no obvious impact on PS and SS representations in the two linguistic stores. To take a grammatical example, while regularly hearing or reading the statement that “WH-questions in English are formed using the auxiliary do,” the resulting processing of that statement in phonological WM and syntactic WM will involve no exposure to interrogative constructions since this particular declarative sentence provides no relevant evidence: although metalinguistic knowledge of the grammar of English may be gained and boosted within the conceptual system as more examples of this statement are processed, no syntactic growth can result unless spoken and written utterances actually contain examples of the construction in question. Syntactic growth can only take place locally in the syntactic system and always at activation levels beneath those required for conscious awareness. This seems to suggest that, other things being equal, conceptually based metagrammatical knowledge may be acquired more rapidly than the equivalent syntactic knowledge due to the very high activation levels of the relevant conceptual representations required for global WM. One implication of this and the preceding section is that a strong relation exists between WMC and explicit processing, a point that was noted by Williams (2012). Processing in PS and SS is largely implicit and makes little use of WMC. Use and development of metalinguistic knowledge in CS is typically explicit and typically reliant on WMC, though automatization through extensive use can gradually make it more implicit. Another implication is that WMC is especially important for the processing of lower proficiency learners, especially formal classroom learners, because they are more dependent on metalinguistic knowledge due to the undeveloped
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
383
384
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
state of SS and PS. This prediction receives support from relevant research (French, 2003; Gathercole, 2006; Hummel, 2009; Masoura & Gathercole, 2005; Winke, 2005).
17.6
Working Memory in Language Use
The understanding of working memory and working memory capacity presented in the preceding sections has implications for a variety of topics in language use. The logic is quite similar to that just described for the development of the linguistic modules and metalinguistic knowledge. We will first describe processing for each type of case and then consider the implications for various topics in language use.
17.6.1 Processing in the Strictly Linguistic Systems: PS and SS As described above, processing in the two strictly linguistic systems is largely automatic. In most instances of phonological and syntactic processing, all the required information is directly, efficiently provided to the processors from their own stores and from the automatic interactions between them. This automaticity implies a minimal role for working memory in its commonly understood sense; thus, standard measures should show little or no role for WMC in this processing. Research supports this prediction, finding little relation between syntactic processing and WMC constraints (Juffs, 2015; Juffs & Harrington, 2011; Williams, 2012, 2015). The broad synchronization that underlies global WM is important for providing usable input, but it plays little or no role in activity within the linguistic system itself. WMC can become relevant, however, in certain circumstances. Additional processing could be needed in order to produce usable input if the speech is unclear or is in an unfamiliar accent or is accompanied by loud background noise. This additional processing could well involve global WM and thereby make WMC relevant. The same is true if the AS input is clear but difficult to interpret, as when a sentence is very long and complex or includes unknown words. In such cases the representation of the meaning cannot be constructed directly from the input that CS receives from SS, and so additional CS processing is needed, probably requiring reactivation/ reconstruction of the AS representation as a guide for the processing. Again, the global WM is likely to become involved in this processing, with the implication that WMC becomes relevant.
17.6.2 Processing That Involves Metalinguistic Knowledge Metalinguistic knowledge is, again, conceptual (CS) knowledge. CS, with its very broad scope, is not specialized to deal with linguistic input or to produce representations of language. So just as the development of
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
metalinguistic knowledge commonly involves global WM, so does its use. This can be seen in monitoring. An example is an English learner who knows that -s should be added to third person singular present tense nouns – who possesses this rule in the form of a CS representation – but omits the affix in spontaneous speech, which is to say the necessary PS-SSCS chain has not developed or is not yet strong enough. If the CS representation of the rule is sufficiently activated, it can in principle alter processing in AS and then PS, with the result that the affix does appear. Sufficient activation is likely to require involvement of the global WM, with accompanying awareness of the rule. The use of a metalinguistic representation can, however, become automatic through very extensive use, at least in principle. In such cases the resting activation levels of the representations and the connection between them become high enough to allow automatic use of the metalinguistic knowledge; the person can, perhaps, add -s in the appropriate places without involvement of the global WM and without awareness of doing it. But despite this theoretical possibility the use of metalinguistic knowledge in general requires more extensive processing than that in the strictly linguistic system, and this means that global WM routinely plays an important role and that WMC is relevant.
17.6.3 Bilingual Language “Selection” The view of working memory proposed here directly provides an answer to a familiar question in bilingual processing, that of “selective access” versus “nonselective access.” The question is whether processors have available to them both (all) of the bilingual’s languages or if those from the currently unwanted language are ruled out early on and “selection” of representations for processing is restricted to those of a single language. In the MCF account, there is no separation of representations of the different languages. All syntactic representations, regardless of language, are found in SS. All phonological representations, regardless of language, are found in PS. So representations from both languages are routinely “in WM” together, which is to say there is no preselection of representations from a particular language. The result is a constant competition for inclusion in processing. In other words, “access” is “nonselective.” This conclusion is consistent with the weight of the experimental evidence (e.g., de Groot & Starreveld, 2015; Kroll et al., 2015; Meuter, 2009; Schwartz, 2015). Which language is used at a given time is determined to a large extent by representations of the situation, as well as the value associated with each language in that situation and the person’s currently active goals. The effects become particularly strong when the influences of internal context produce a synchronization of activity centered on one representation, giving that representation an exceptionally high current activation level (i.e., when the representation is in global WM). An example is the identity
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
385
386
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
of the interlocutor. Talking to a person naturally involves an awareness of that person, which is to say that a representation of him/her is or recently was the focus of the global synchronization and now continues to be in global working memory. This in turn means that the representation is influencing speech production, possibly determining the language that is being used, as well as affecting various factors such as register and tone, not to mention the content. A question arises here as to why we do not see a constant free mixing of languages in speech. Three complementary answers can be offered. One is that the stores are to some extent self-organizing, in that representations from one language tend to become more strongly associated with others from the same language simply because they typically appear together in the input. The consequence is that activation of a representation in one language tends to activate other representations in that same language (see Shook & Marian, 2013). The second constraint on free mixing of languages is expressed in Grosjean’s (2010, p. 29) Complementarity Principle: “Bilinguals usually acquire and use their languages for different purposes, in different domains of life, with different people.” In MCF terms, representations of the external context are typically associated with one language or the other. Thus activity of particular outside-in representations tends to yield activation of representations from one language much more strongly than the other. The third factor involves deliberate choice of the language to be used. This is achieved through language representations in CS, which fall under the heading of metalinguistic knowledge. The representation ENGLISH, for example, which is simply the concept of English, is associated with various English-related representations and so, when activated by an active goal of using English, will enhance the activation of those representations.
17.6.4 Code-Switching Competent bilinguals can quickly and efficiently switch between their languages, inserting words or expressions from one into an utterance which is primarily in the other. This impressive performance is possible because representations from the different languages are found in the same stores and are simultaneously active, or “in” the local WMs, meaning that they are all readily available for use in processing. Spontaneous switching, we suggest, can be seen as a direct reflection of this simultaneous activation. For the case of a Spanish/English bilingual speaking in English, when a representation from Spanish has a higher current activation level than its English competitor, it appears in the utterance. In such cases there is no cost to the switch; in fact sticking with English would be costly (see de Bruin et al., 2018; Kleinman & Gollan, 2016). Deliberate switching, in contrast, can be seen as influence from outside the strictly linguistic system – other factors are interfering with the
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
spontaneous workings of the system. The extreme case is seen in many bilingual processing experiments, in which switching occurs in response to the experimenter’s instructions to switch (see Grosjean & Li, 2013; Jiang, 2015; van Hell et al., 2015). This sort of switching is likely to require involvement of the global WM and therefore to take additional time and perhaps effort.
17.6.5 Crosslinguistic Influence If representations from both languages are simultaneously present in each local WM, as we have suggested, and presence in WM means availability for processing, as is commonly assumed, it is no surprise that languages influence each other, both in their development and in their use. This influence should and does occur in both directions, though more so from a stronger language to a weaker one (Athanasopoulos, 2015; de Groot & Starreveld, 2015; Grosjean & Li, 2013; Jiang, 2015), simply because stronger implies higher resting activation levels. The much-studied phenomenon of optionality in second language acquisition is an example of this influence (e.g., Robertson & Sorace, 1999; White, 2003). Research has shown that learners commonly act as if a syntactic feature of the second language is optional, alternating between correct use of the L2 feature and an inappropriate form based on the L1. The explanation is straightforward. During production, structures from both languages are, again, part of syntactic WM. They can thus go back and forth between being used in processing and being the (unused) context of processing (but see Truscott, 2006, and Truscott & Sharwood Smith, 2019, for a more detailed MCF account).
17.6.6 Translation and Interpreting Although it is usually referred (in English) to tasks that do not need to be completed immediately, “translation” can be taken as a cover term for different types of conversion of written or spoken utterances from one language into another. In all cases processing in both global and local working memory is involved. Time pressure for the execution of a translation task in this general sense can vary considerably from what is required for translations of written text carried out at leisure, or under time pressure from an absent editor expecting the result after a brief lapse of time that could be hours up to a few days, to consecutive interpreting where a stretch of spoken or signed language needs to be translated immediately afterward, to simultaneous interpreting requiring rapid switches between languages and comprehension and production carried out at the same time. Even experienced simultaneous interpreters only manage to sustain this activity in stretches of twenty to thirty minutes at a time, showing that this type of activity is particularly effortful. This should not be surprising, as extensive
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
387
388
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
use of global WM is necessarily involved as a result of continual, rapid, and deliberate switching between two languages. Any type of translation is just another example of goal-driven behavior. Apart from the overall, shared goal of converting a source (spoken or written) text in one language into the terms of another, there will also be subsidiary goals depending on the precise nature, the degree of faithfulness to the original required, the time limit, the requirements appropriate to different kinds of interpreting, and so forth. Apart from that, there will be even more specific goals activated in the course of translation activity when problems arise such as working out a complex, badly articulated meaning in the source language or finding a less familiar word or construction that is stylistically appropriate in a given context. Translation tasks are therefore accomplished with varying degrees of conscious processing. In experienced translators, the basic goal of language conversion may cease to feature in global WM and be effortful, having become a standard routine and hence, in WM terms, may be carried out at lower levels of activation. However more specific goals will often continue to require global WM and therefore a greater use of processing resources and sense of effortfulness. To sum up, any translation process involves constant switching back and forth between languages, and this means sustained dynamic activity with both local and global working memory involved. All of this involves the use of different aspects of metalinguistic knowledge, that is, representations in conceptual WM.
17.7
Conclusion
In this chapter we have sought to show how working memory and an assortment of phenomena related to it can be handled within a broad cognitive framework, namely the Modular Cognition Framework. Working memory, as the activation of representations that makes them available for processing, is of two types. Each specialist system has its own local WM, consisting of its representations that are currently active and therefore available to its processor. A broad synchronization of activity across a number of local WMs constitutes global WM, closely associated with conscious experience. Working memory capacity is to be analyzed in terms of all the local and global activity involved in a given task. This conception supports the study of WM as part of the cognitive system, notably the study of the role that WM plays in the development and use of language. Large questions remain of course, notably including the measurement of individual WM capacities and exactly how the individual capacities together yield what is commonly understood as working memory capacity. The main conclusion we wish to draw here is that the study of working memory and language is best conducted in the context of what we would argue is a sorely needed general framework specifying the nature and operation of the cognitive system – a framework like the MCF.
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
References Anderson, J. R. (1983). The architecture of cognition. Harvard University Press. Athanasopoulos, P. (2015). Conceptual representation in bilinguals: The role of language specificity and conceptual change. In J. W. Schwieter (Ed.), The Cambridge handbook of bilingual processing (pp. 275–292). Cambridge University Press. Baars, B. (1988). A cognitive theory of consciousness. Cambridge University Press. Baddeley, A. (2007). Working memory, thought, and action. Oxford University Press. Barrett, H. C., & Kurzban, R. (2006). Modularity in cognition: Framing the debate. Psychological Review, 113, 628–647. Bastos, A. M., Vezoli, J., & Fries, P. (2015). Communication through coherence with inter-areal delays. Current Opinion in Neurobiology, 31, 173–180. Beckner, C., Blythe, R., Bybee, J., Christiansen, M., Croft, W., Ellis, N., Holland, J., Ke, J., Larsen-Freeman, D., & Schoenemann, T. (2009). Language is a complex adaptive system: Position paper. Language Learning, 59, 1–26. Bergeron, V. (2007). Anatomical and functional modularity in cognitive science: Shifting the focus. Philosophical Psychology, 20, 175–195. Cantor, J., & Engle, R. W. (1993). Working-memory capacity as long-term memory activation: An individual-differences approach. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 1101–1114. Carruthers, P. (2006). The architecture of the mind: Massive modularity and the flexibility of thought. Clarendon. Chomsky, N. (1995). The minimalist program. MIT Press. Cowan, N. (1995). Attention and memory: An integrated framework. Oxford University Press. Cowan, N., Saults, J. S., & Blume, C. L. (2014). Central and peripheral components of working memory storage. Journal of Experimental Psychology: General, 143, 1806–1836. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning & Verbal Behavior, 19, 450–466. de Bruin, A., Samuel, A. G., & Duñabeitia, J. A. (2018). Voluntary language switching: When and why do bilinguals switch between their languages? Journal of Memory and Language, 103, 28–43. de Groot, A. M. B., & Starreveld, P. A. (2015). Parallel language activation in bilinguals’ word production and its modulating factors: A review and computer simulations. In J. W. Schwieter (Ed.), The Cambridge handbook of bilingual processing. Cambridge University Press. Dehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of consciousness: Basic evidence and a workspace framework. Cognition, 79, 1–37.
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
389
390
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
D’Esposito, M., & Postle, B. R. (2015). The cognitive neuroscience of working memory. Annual Review of Psychology, 66, 115–142. Driver, J., Davis, G., Russell, C., Turatto, M., & Freeman, E. (2001). Segmentation, attention and phenomenal visual objects. Cognition, 80, 61–95. French, L. M. (2003). Phonological working memory and L2 acquisition: A developmental study of Quebec Francophone children learning English. (Doctoral dissertation, Université Laval, Quebec). Fuster, J. M. (2015). The prefrontal cortex (5th ed.). Elsevier. Gathercole, S. E. (2006). Nonword repetition and word learning: The nature of the relationship. Applied Psycholinguistics 27, 513–543. Grosjean, F. (2010). Bilingual: Life and reality. Harvard University Press. Grosjean, F., & Li, P. (2013). The psycholinguistics of bilingualism. WileyBlackwell. Hummel, K. M. (2009). Aptitude, phonological memory, and second language proficiency in nonnovice adult learners. Applied Psycholinguistics, 30, 225–249. Jackendoff, R. (1987). Consciousness and the computational mind. MIT Press. Jackendoff, R. (1997). The architecture of the language faculty. MIT Press. Jiang, N. (2015). Six decades of research on lexical representation and processing in bilinguals. In J. W. Schwieter (Ed.), The Cambridge handbook of bilingual processing. Cambridge University Press. Juffs, A. (2015). Working memory and sentence processing: A commentary. In Z. Wen, M. B. Mota, & A. McNeill (Eds.), Working memory in second language acquisition and processing (pp. 125–135). Multilingual Matters. Juffs, A., & Harrington, M. W. (2011). Aspects of working memory in L2 learning. Language Teaching, 44, 137–166. Kleinman, D., & Gollan, T. H. (2016). Speaking two languages for the price of one: Bypassing language control mechanisms via accessibility-driven switches. Psychological Science, 27, 700–714. Krauzlis, R. J., Bollimunta, A., Arcizet, F., & Wang, L. (2014). Attention as an effect not a cause. Trends in Cognitive Sciences, 18, 457–464. Kroll, J. F., Gullifer, J. W., McClain, R., Rossi, E., & Martín, M. C. (2015). Selection and control in bilingual comprehension and production. In J. W. Schwieter (Ed.), The Cambridge handbook of bilingual processing. Cambridge University Press. Masoura, E. V., & Gathercole, S. E. (2005). Contrasting contributions of phonological short-term memory and long-term knowledge to vocabulary learning in a foreign language. Memory, 13, 422–429. McElree, B. (1998). Attended and non-attended states in working memory: Accessing categorized structures. Journal of Memory and Language, 38, 225–252. Meuter, R. (2009). Neurolinguistic contributions to understanding the bilingual mental lexicon. In A. Pavlenko (Ed.), The bilingual mental lexicon: Interdisciplinary approaches (pp. 1–25). Multilingual Matters.
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
The Modular Cognition Framework
Miller, E. K., & Buschman, T. J. (2014). Neural mechanisms for the executive control of attention. In S. Kastner & A. C. Nobre (Eds.), The Oxford handbook of attention. Oxford University Press. Nobre, A. C., & Mesulam, M.-M. (2014). Large-scale networks for attentional biases. In S. Kastner & A. C. Nobre (Eds.), The Oxford handbook of attention. Oxford University Press. Oberauer, K. (2013). The focus of attention in working memory: From metaphors to mechanisms. Frontiers in Human Neuroscience, 7, 673. Oberauer, K., Farrell, S., Jarrold, C., & Lewandowsky, S. (2016). What limits working memory capacity? Psychological Bulletin, 142, 758–799. Palva, J. M., Monto, S., Kulashekhar, S., & Palva, S. (2010). Neuronal synchrony reveals working memory networks and predicts individual memory capacity. Proceedings of the National Academy of Sciences, USA, 107, 7580–7585. Posner, M. I. (1994). Attention: The mechanisms of consciousness. Proceedings of the National Academy of Sciences, USA, 91, 7398–7403. Robertson, D., & Sorace, A. (1999). Losing the V2 constraint. In E. C. Klein & G. Martohardjono (Eds.), The development of second language grammars: A generative approach (pp. 317–361). Benjamins. Sandler, W. (1989). Phonological representation of the sign: Linearity and nonlinearity in American Sign Language. Foris. Schwartz, A. (2015). Bilingual lexical access during written sentence comprehension. In J. W. Schwieter (Ed.), The Cambridge handbook of bilingual processing (pp. 327–348). Cambridge University Press. Sharwood Smith, M. A. (2014). In search of conceptual frameworks for relating brain activity to language function. Frontiers in Psychology, 5, 716. Sharwood Smith, M., & Truscott, J. (2014). The multilingual mind: A modular processing perspective. Cambridge University Press. Shook, A., & Marian, V. (2013). The Bilingual Language Interaction Network for Comprehension of Speech. Bilingualism: Language and Cognition, 16, 304–324. Szmalec, A., Brysbaert, M., & Duyck, W. (2013). Working memory and (second) language processing. In J. Altariba & L. Isurin (Eds.), Memory, language, and bilingualism: Theoretical and applied approaches (pp. 74–94). Cambridge University Press. Truscott, J. (2006). Optionality in second language acquisition: A generative, processing-oriented account. International Review of Applied Linguistics, 44, 311–330. Truscott, J. (2015a). Consciousness and second language learning. Multilingual Matters. Truscott, J. (2015b). Consciousness in SLA: A modular perspective. Second Language Research, 31, 413–434. Truscott, J. (in press). Working memory and language in the modular mind. New York: Routledge.
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
391
392
J O H N TR U SCOT T A N D M I C H A E L S H A R W OO D SM I TH
Truscott, J., & Sharwood Smith, M. (2019). The internal context of bilingual processing. Benjamins. Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task dependent? Journal of Memory and Language, 28, 127–154. van Hell, J. G., Litcofsky, K. A., & Ting, C. Y. (2015). Intra-sentential codeswitching: Cognitive and neural approaches. In J. W. Schwieter (Ed.), The Cambridge handbook of bilingual processing. Cambridge University Press. Wen, Z. (2014). Theorizing and measuring working memory in first and second language research. Language Teaching, 47, 174–190. Wen, Z., Mota, M. B., & McNeill, A. (Eds.). (2015). Working memory in second language acquisition and processing. Multilingual Matters. White, L. (2003). Second language acquisition and Universal Grammar. Cambridge University Press. Williams, J. N. (2012). Working memory and SLA. In S. M. Gass & A. Mackey (Eds.), The Routledge handbook of second language acquisition (pp. 427–441). Routledge. Williams, J. (2015). Working memory in SLA research: Challenges and prospects. In Z. Wen, M. B. Mota, & A. McNeill (Eds.), Working memory in second language acquisition and processing (pp. 301–307). Multilingual Matters. Winke, P. M. (2005). Individual differences in adult Chinese second language acquisition: The relationships among aptitude, memory and strategies for learning (Doctoral dissertation, Georgetown University). Womelsdorf, T., Schoffelen, J.-M., Oostenveld, R., Singer, W., Desimone, R., Engel, A. K., & Fries, P. (2007). Modulation of neuronal interactions through neuronal synchronization. Science, 316, 1609–1612.
Note 1 In sign language, it is visual, not auditory representations that connect up with PS (see, for example, Sandler, 1989). The principle is clear but how, in the MCF or elsewhere, sound and sign phonology exactly relate to one another is an open question.
https://doi.org/10.1017/9781108955638.021 Published online by Cambridge University Press
18 Short-Term and Working Memory Capacity and the Language Device Chunking and Parsing Complexity Bingfu Lu and Zhisheng (Edward) Wen
18.1
Introduction
The term working memory (WM) first appeared in Miller et al. (1960), and it generally refers to the ability to simultaneously store and process a small amount of task-relevant information in our heads during some cognitive activities such as mental calculation, logical reasoning, planning, and language comprehension (Cowan, 2005). Ever since the seminal model of WM proposed by the British cognitive psychologists Alan Baddeley and Graham Hitch (1974), research interest into the construct has been growing exponentially, particularly so after the publication of the well-cited volume on WM models by Miyake and Shah (1999; Wen, 2019). Throughout the various developmental stages of WM (Logie, 1996), both consensus and controversies have lingered over the construct’s (a) definition and nature (Cowan, 2017), (b) structure and functions, as well as (c) its inextricable relationship with long-term memory (Baddeley, 2012; Cowan, 2008). These multiple WM research camps can be generally described as pursuing two research paradigms, either the multicomponent model by Baddeley and colleagues or the executive control perspectives by cognitive psychologists mostly based in North America, such as Nelson Cowan and Randy Engle (Andrade, 2001). Inspired more or less by these two well-established traditions, a multitude of WM models and diverse perspectives have also blossomed and flourished in cognitive sciences, neurosciences, and many neighboring fields (Logie et al., 2021; Miyake & Shah, 1999). Amid these consensuses and controversies, most theoretical models would agree that WM has a limited capacity, We would like to thank John Hawkins, Ping Chen, Haitao Liu, and the anonymous reviewers for detailed and constructive comments on previous drafts of this chapter. This research is supported by the National Social Science Fund of China (18AYY023).
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
394
BINGFU LU AND ZHISHENG (EDWARD) WEN
though they are likely to disagree as to exactly what the underlying causes of this limitation are and on its exact quantification (Conway et al., 2007). Indeed, WM limitations have increasingly become a signature feature of this central construct of human cognition (Carruthers, 2013). Specifically, WM limitations are manifested in two ostensible ways: (a) the restricted units or chunks of information that can be held and worked upon (i.e., the “chunk capacity limit” as conceived by Cowan, 2008) and (b) the transient nature of this information being held (i.e., the “memory decay” conceived by Cowan, 2008). On the one hand, cognitive psychologists have disagreed on the exact sources of the transient or fleeting nature of information temporarily stored in WM for processing, which usually lasts for around 5–12 seconds (Waugh & Norman, 1965). Hypotheses related to explaining these phenomena include “memory decay” (Barrouillet & Camos, 2012, 2015; Cowan, 2008) and the “interference” view (Oberauer & Lewandowsky, 2013, 2014). Recently, Christiansen and Chater (2016) have proposed the “Now or Never” bottleneck effect. On the other hand, regarding the exact quantification of WM limitations, several candidate numbers have also been put forward, ranging from Miller’s (1956) famous magical number “7 2” to Cowan’s (2001) more realistic “4 1.” That said, it is fair to say that most WM theorists would agree to accept that the capacity of STM and WM can be set at between four chunks of unrelated items (Cowan’s view) and 7 units of information (Miller’s view). However, so far, there has been no exact distinction made between these two numbers and the potential implications they may have for research and practice. More relevantly in the field of language science and general linguistics, recent years have also witnessed several theoretical linguistic frameworks and hypotheses attempting to incorporate STM/WM limitations as a universal constraint on language design and processing (e.g., Gomez-Rodriguez et al., 2019; Jackendoff, 2007, 2011; O’Grady, 2015, 2017). For example, O’Grady (2017) in his recent commentary on the target article published by Pierce et al. (2017 in Applied Psycholinguistics), has cited examples from English, Russian, and Korean to argue that WM limitations not only constrain phonology in the language (as already discussed in extensive details by Pierce et al. 2017) but also constrain the character and acquisition of many grammatical phenomena, such as the typology of word order and the interpretation of pronouns during language processing. In light of this universal feature of WM, O’Grady (2015) postulates that the WM limitations should be a primary factor in language design and acquisition, as opposed to being relegated as the peripheral (“third”) factor by Chomsky (2005) that comes only after the first factor of universal grammar (UG) and the second factor of language experience. In a similar vein, Jackendoff in his “Parallel Architecture (2007 & 2011) has also proposed that linguistic WM should consist of three subdivisions or “departments” corresponding to the processing and construction of three grammatical structures (instead of just only focusing on the “phonological loop” in Baddeley’s classic model): (a) a phonological WM component for constructing and processing phonological structures, (b) a syntactic WM
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
component for processing syntactic structures, and (c) a semantic WM component for processing semantic structures. Interpreted in this way, WM functions as the sort of “workbench” or “blackboard” (Jackendoff, 2011, p. 13) where these grammatical structures are constructed and integrated online in a parallel manner. As demonstrated above, both O’Grady and Jackendoff have assigned a pivotal role to WM limitations. Unfortunately, WM limitations in their models and frameworks seem to be held only as a fixed trait construct, and it remains unclear how to operationalize it. In recent years, the field of linguistics has witnessed rapid developments with the advent of modern computational technology and the big data era. Both fields of language sciences and cognitive sciences have witnessed large-scale cross-linguistic corpus studies investigating the grammar dependency and locality phenomena across a large number of language databases (Gibson, 1998 & 2000; Gibson et al., 2019; also see Futrell, 2017; Liu, 2008). Most of these studies have the fundamental assumption and the starting point of incorporating the universal constraint of WM limitations (Ferrer-i-Cancho, 2017; Nicenboim et al., 2015). For example, Liu (2008) measured the average minimum dependency distance (MDD) in 20 languages and found that all fell under the WM capacity of 4 (with Chinese having the greatest MDD of 3.662). In a more recent study, Futrell et al. (2015) included 37 languages and found a similar pattern. Most recently, Gomez-Rodriguez et al. (2019) sampled about 16 billion syntactic dependency structures that differed in length and syntactic complexity and found that memory constraints, in the form of dependency distance minimization (DDM) have become inherent to formal linguistic grammars, leading the authors to conclude that all grammatical frameworks should incorporate memory constraints to capture our unbounded capacity for language productivity. Taken together, these recent large-scale studies are lending compelling evidence to the permeating principle of “the law of least efforts” (Zipf, 1949), or efficiency (Gibson et al., 2019; Hawkins, 2004), or compression and minimization (toward the tendency of MDD) in general language behavior, which are all presumably constrained by the limited capacity of STM/WM limitations (Ferrer-i-Cancho, 2017; Ferrer-i-Cancho et al., 2019; Liu, 2008). Based on this basic assumption, the current chapter aims to further demystify the distinction between STM and WM limitations as they relate to linguistic structures. To achieve this goal, we aim to posit our memory- and chunk-based metric of comprehension complexity vis-á-vis Liu’s MDD (Mean Dependency Distance) and Hawkins’s IC-to-Word ratio.
18.2
Chunking: The Basic Process of Linguistic Structure
In this section, we will first define what a chunk is and then discuss its operationalization in linguistics. We will then turn to distinguish STM-sensitive chunking versus WM-sensitive chunking as they relate to linguistic complexity.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
395
396
BINGFU LU AND ZHISHENG (EDWARD) WEN
18.2.1 What Is a Chunk in Linguistics? Though there has been some consensus that the basic form of the human information processing is chunking since the notion was posited by Miller (1956b), the constraint of the span of short-term memory and attention, that is, 7(2) (Miller, 1956a) has not been stipulated as a critical number in human language processing. This is not surprising, considering that Miller himself has not provided an answer as to what a chunk is in linguistic structure. Miller (1956b) states: We can usually repeat a 20-word sentence after hearing it once. How many items – 100 letters, 30 syllables, 20 words, 6 phrases, 2 clauses, or one sentence – does this sentence contain? We know that it contains about 120 bits of information because we have a definition of the bit that is independent of our subjective organization of sentences. But the essence of the chunk is that it is imposed by the person. For example, someone who knew nothing of English except the alphabet would have to treat the sentence as if it were 100 units long, while someone who knows English well might deal with it as if it were 6 units long. We cannot define the unit of organization independent of the hearer. (p. 131) Miller’s pessimism about defining an objective linguistic chunk independent of the individual language users, however, does not hold. The matter can be solved by assuming an English native user as an ideal parser, as Chomsky (1965) puts it: Linguistic theory is concerned primarily with an ideal speaker-listener, in a completely homogeneous speech community, who knows its language perfectly and is unaffected by such grammatically irrelevant conditions as memory limitations, distractions, shifts of attention and interest, and errors (random or characteristic) in applying his knowledge of the language in actual performance. (p. 3) Miller (1956b) himself conjectures intuitively and ingeniously that a native English user might deal with a fairly long sentence of 20 words, or 100 letters, as if it were 6 phrases long. The problem is how to identify these phrases. It should be noticed that the definition of a phrase must resort to a headword. According to a basic consensus in contemporary syntax, a phrase, that is, an Xmax is the projection of an X0 . Therefore, it can be claimed that a headword and all its dependents (arguments plus adjuncts), usually being phrases, are the basic chunks of the sentence, which inevitably and essentially includes a headword, but not only phrases (Lu, 2001). In other words, Miller’s 6-phrase sentence is actually composed of 7 chunks: six phrases plus a head verb, as fictionally demonstrated in Figure 18.1.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
Figure 18.1 Linguistic chunks
Our linguistic experience and intuition tell us that the overwhelming majority of sentences and noun phrases are composed of fewer than 7(2) dependent phrases of the headword. This quantitative constraint on linguistic structure might be taken as a quantitative universal of human language. Or in other words, the span of STM and attention has been incorporated into our language device. A headword and all its dependents can be called “Direct Constituents” (henceforth DC), which is distinguished from “Immediate Constituents” (IC) in the sense that the number of DCs per construction is sensitive to the span of STM, while IC is not. Thus, we may also refer to DCs as the chunks of linguistic construction. In other words, a DC can refer to both a Direct Constituent and a Direct Chunk. DC is similar to Dryer’s (1992, pp. 112–115) Major Constituents, which fits his Branching Direction Theory (BDT) much better than IC does. However, the notion of Major Constituents does not include the head verb. DC is furthermore similar to Hawkins’s IC in his EIC (Early Immediate Constituent), where the head verb is taken as an IC. In fact, DC is the very word-order unit in the practice of word order description in linguistic typology, such as V is a word while S and O are phrases in basic word orders of SVO, SOV, and VSO. In the description of the NP word order, such as DNum-Adj-N, or N-Adj-Num-D, a head noun is indispensable as well. Therefore, the term “chunk order” is preferable to the inaccurate “word order” and “phrase order,” or the vague “constituent/meaningful element/ unit order.” The term “chunk order” is at least more heuristic as it tells us that such a unit is sensitive to the span of STM and attention. That linguistic structure is hierarchical can be mainly shown in infinitive recursion. Any nonhead word in a construction can serve as a head to produce its own daughter dependents, and any nonhead words therein can further serve as a head word, producing its dependent. The shift of head hood, that is, motherhood, is the necessary condition for recursion. Hence, without a fixed headword, one just cannot clarify phrases at whichever level one is referring to. Only with a fixed headword, can one rule out the overlapping of phrases (Lu, 1983). According to the X-bar theory, a head word determines the structural pattern of a phrase. From the methodological perspective, it is obvious that one should analyze structural patterns one by one. The traditional phrase marker (tree diagram) conflates all phrases in different hierarchies within a
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
397
398
BINGFU LU AND ZHISHENG (EDWARD) WEN
matrix construction. The DCs in one construction do not overlap each other and cover all words in the construction exhaustively. Now, we turn to discuss how we can get DCs from IC analysis in the following section.
18.2.2 An Operational Definition of Linguistic Chunks The hierarchy of linguistic structure was first well described with IC analysis and later with the tree diagrams and phrase markers. However, such an analysis, in fact, conflates various structural patterns at different levels in one diagram, making syntactic patterns vague and fuzzy. To highlight syntactic patterns, one may simplify the binary phrase marker as below. Suppose that we cut a construction into its two ICs at first. As the next step in the analysis, if our cutting is indiscriminate, we should simultaneously cut each IC into two ICs of the lower layer and thus we get four units. Proceeding this way, we would maximally get 2n units at layer N, which may soon be well beyond the STM limit. Now suppose our cutting is selective in that we just choose the head IC of the two ICs as the target of the next segmentation,1 The head IC is defined as the one that contains the headword of the whole construction, that is., the root head, or global head. For example, in the construction carefully read a dictionary in the library, between the two ICs, carefully read a dictionary is the head IC while in the library is not, since the former contains the global head read. As this procedure goes on, one will quickly get only N + 1 units at layer N. Since the necessary condition of infinitive recursion is the shift of headhood, it can be inferred that the unidirectional segmentation toward a fixed head must be finite (Lu, 1983). In short, this analytic procedure cuts the sentence unidirectionally toward the global head, dispensing with all the internal analysis within all nonhead ICs that are irrelevant to “digging out” the global head, and hence irrelevant to the structural pattern of the matrix construction, the pattern of which is determined by its headword. The principle guiding this analysis can be termed as “One Head, One Pattern,” or Head-Pattern Corresponding. This procedure can then be taken as the operational definition of the linguistic chunk. As demonstrated in Figure 18.2, the bold line with the arrowhead figuratively stands for the tree trunk, while the middlethickness lines directly linked to the trunk are “major branches.” Hence it was once dubbed “Major Branch Analysis” (Lu, 1981). It results in a simplified tree diagram of phrase markers. The trunk plus all its major branches represent the structural pattern of the construction. The fine lines, standing for twigs and leaves, should be neglected in the analytic procedure of the matrix phrase. They represent embedded local structural patterns that are irrelevant to the matrix construction. Their presentation in the tree of the matrix phrase blurs the basic structure of the matrix phrase.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
Figure 18.2 Procedure for obtaining DCs Note: The superscript digits above the sentence stand for the order of cutting, the digits on the left side column for the order of levels.
Figure 18.3 Major-branch tree
When removing the twigs and leaves, the tree contains only DCs, as shown in Figure 18.3. Furthermore, the major-branch tree can easily be transferred into a headoriented orbit layering structure as shown in Figure 18.4.2 The structure of Orbital Layering can be horizontally expressed with a string of brace pairs: {{{{{{V}M}I}L}D}T}, where the layers of braces just imply the distance to the head Verb, but not directly the temporal linear ordering. Without the hierarchy of semantic distance, the permutation of these six units would be 720. With this hierarchy, the ordering pattern can be reduced to 24, though still much more than our attested 6 variants. The orbit-layering diagram reflects the cross-linguistic similarity of surface orderings, in terms of the head-oriented distance. This similarity is motivated by semantic iconicity (Lu, 1993). DCs can thus be taken as basic movable units both inner- and crosslinguistically of word order variations. Note that a “movable unit” is relative to its matrix construction. For example, an argument NP is a movable unit in the matrix clause it belongs to, and there are some movable units in turn within the NP itself as a matrix construction. In tree diagrams, the relationship between a mother node and its daughter nodes is easily misunderstood as if they were a whole-part one. However, it is
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
399
400
BINGFU LU AND ZHISHENG (EDWARD) WEN
Figure 18.4 The orbit-layering diagram
a dominant-dependent one instead. Such a misunderstanding will not happen in the orbit-layering diagram. In short, the characteristic of this analysis is that one headword determines one analytic procedure and one analytic procedure, in turn, highlights one structural pattern. The traditional way of doing IC analysis mixes up and lumps together all constituents at different levels and of different syntactic patterns, thus leading to Miller’s dilemma of being unable to quantify how many chunks there were in his 100-letter sentence.
18.2.3
From STM-Sensitive DC to WM-Sensitive MC (Momentary Chunk) However, it should be noted that DC is basically a static category in the sense that the DCs within a matrix chunk do not include transient chunks in the real-time parsing procedure, which are not DCs of the matrix chunk. To describe the online procedure, we need the notion “Momentary Chunk” (MC), referring to all discrete chunks that a parser has to deal with during the parsing procedure. In other words, MCs include each word and all the chunks that contain them, directly or indirectly. In nature, DCs are analytical and static, highly relevant to and determined by, the specific structural pattern; while MCs are synthetic, dynamic, transient, on-line, fleeting, and their patterns can be rather free. The simplified phrase marker composed only of DCs provides a syntactic pattern, which guides the parser on how and when to close the previous smaller momentary chunk(s) into a DC. For example, in Figure 18.5, carefully read does not count as a chunk, nor does he carefully read, he carefully read a dictionary, and so on. This is because they are not DCs of the matrix sentence. In other words, only words (X0 units) and the maximal projection of words (Xmax unit) count as chunks. This is consistent with the hypothesis that only X0 and Xmax units are movable units (Chomsky, 1986). For
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
Figure 18.5 On-line chunking This figure indicates the relations between DC and MC. MCNs is the number of MC at the moment the parser encounters the word above. The Mean of MCNs (hence MMCN) can be taken as the indicator of the perceptual complexity of the sentence.
example, since Carefully read is not an Xmax, it cannot count as an MC. The same treatment applies to he carefully read, and even he carefully read a dictionary. One of the motivations for this kind of treatment is that the material within a closed chunk is hard to be retrieved, as shown in (1). (1)
a. He knew the girl left exactly. b. He knew that the girl left exactly.
Hakes (1972), based on his experiment, finds that (1b) is harder to parse than (1a) is. This is because that tempts the parser to treat what follows as the last DC of the sentence, and thus combines all the preceding words into a finished sentence. When the parser unexpectedly meets exactly later on, he would feel slightly awkward about linking it to knew because the words inside the finished chunk are no longer as distinct as before, receding into the semantic background, hence harder to retrieve If a new word, say wordn+1, is unable to be chunked with the preceding word(s), its MCN will increase by 1. If a new word can be chunked with the preceding word(s), the MCN will remain unchanged or be reduced. Two words with the same MCN along with all the words in-between with higher MCNs constitute a super chunk (subscript digits in Figure 18.5 stand for MCNs): [wordn wordn+1 wordn+2. . ..wordn+m wordn] form a new, bigger chunk. For example, in Figure 18.5, [in5 the6 library5] makes a chunk. Supposing that the prepositional phrase expands to [in5 the6 new7 fancy8 library5]; these words also form a chunk as well. In our preliminary analysis of some Chinese data (Lu, 1993), MCN rarely exceeds 7(2), and MMCN, 4(1). This means that the chunking metric is
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
401
402
BINGFU LU AND ZHISHENG (EDWARD) WEN
sensitive to both STM and WM limitations. WM functions as a monitor; when an MCN surpasses 4, it prompts the parser to make a new chunk as soon as possible. Otherwise, the “late closure strategy” (Frazier, 1978) may be the better choice.
18.3
Mean Dependency Distance (MDD) as a Basic Metric of Structural Complexity
In Dependency Grammars (DG), especially in Word Grammar, researchers focus on word-to-word pairwise relations. Two syntactically related words form the dependency relation, with one word being the governor and the other the dependent. The distance between these two words is named “dependency distance” (DD, Hudson, 1995).3 Based on Hudson’s DD, Liu (2008; Xu & Liu, this volume) posits a metric of Mean Dependency Distance (MDD) as an indicator for structural complexity. MDD is the quotient of the sum of word-to-word distance divided by the number of all dependent words, as shown here:4 John
threw
John1threw threw
out the threw
1
old
out the2trash old1trash
thrash threw
4
thrsh
sitting 1 trash sitting
in the 1 sitting in
kitchen
the1kichen in2kitchen
Total dependency distance=1+1+2+1+4+1+1+1+2=14. MDD=14/9=1.56.
Above, each superscript digit of a word stands for its DD from its headword, which appears in the adjacent subscript. The divisor is 9 instead of 10 because threw is the roothead word, which is not dependent on any word within the construction. Based on his 20-language database, Liu (2008) summarizes that each language has its own overall MDD, usually fewer than 4. For example, Romania has the lowest MDD, 1.798; English is in the middle, 2.543; while Chinese has the highest, 3.662. It is thus claimed that MDD is sensitive to the psychologically important number 4, that is, the limited span of working memory (WM) (Cowan, 2001, 2005).5
18.4
Word-to-IC (W/IC) Ratio as a Basic Metric of Structural Complexity
Instead of focusing on complexity, Hawkins’s metric focuses on efficiency. Hawkins (2004) proposes his three principles of efficiency: Minimize Domains (MiD), Minimize Forms (MiF), and Maximize Online Processing (MaOP). The three principles can be realized in the metric of IC/W ratio (ICto-word ratio), being an efficiency index. Below, PCD stands for a Phrasal Combination Domain, defined as the string of terminal and nonterminal elements (or simply words) from the first IC of the phrase to the last, these ICs being constructed by Constructing
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
Categories (CCs). The final digits stand for the word number for recognizing the PCD. A CC is defined as being dominated by, or sister to, a phrase P and that can construct P on each occasion of use (Hawkins, 2014:121).6 (2) a.
[S
[that their time should not
PCD:S 1 PCD:VP b.
2
3
4
5
be
wasted]
6
7
[VP is important]] 1
8 2
IC/W ratio= 2/8 = 25% IC/W ratio= 2/2 = 100%
it [VP is important [that their time should not be wasted]]] [S 2 IC/W ratio=2/2 = 100% PCD:S 1 PCD:VP
1
2
3
IC/W ratio=3/3 = 100%
The metric of the IC-to-word ratio shows that the overall efficiency of (2b) is higher than that of (2a). Notice that Hawkins divides the overall efficiency into two parts in these sentences: one for PCD:S (the PCD for the higher S node) and the other for PCD: VP (the PCD for VP), This treatment is based on a classic Phrase Structure formalism, where there is a significant division of the clause (S) into a VP and a VP-external subject NP. However, in a number of languages, there is little evidence for a VP, and the subject is nothing more than an ordinary dependent to the head verb. Such is the case in Japanese, where the more important cut-off in a sentence is between the topic and the rest of the clause. Chinese is a little different in this respect. It is hard, perhaps as well as meaningless, to find a unique subject in a Chinese sentence like “Jintian Shanghai tianqi hen qinglang” (lit. Today Shanghai weather very sunny).7 The ordering of the three nouns is free, though the dominant one is the one shown here. Furthermore, in the languages where the basic word order is VSO, a VO-type VP does not exist at all, at least on the surface. We can therefore conflate PCD: S and PCD: VP into a single structure for processing as shown below. Therefore we conflate PCD: S and PCD: VP into one sentence as shown below. (3)
a. [S [that their time should not be wasted] is important] 2 3 4 5 6 7 8 9 IC/W ratio=3/9 = 33.3% PCD:S 1 it [VP is important [that their time should not be wasted]]] b. [S 2 3 4 IC/W ratio= 4/4 = 100% PCD:S 1
This change keeps the very essence of the IC/word ratio while conflating PCD:S and PCD: VP into one sentence, thus making the IC/word ratio metric simpler and applicable to more languages as well. And it makes the comparison with MMCN and MDD metrics much easier. Further, we change the efficiency index IC/word ratio to W/IC (Word-toIC) ratio, which can be taken as a complexity index like MMCN and MDD is. Thus the W/IC ratios of (3a) and (3b) would be 3 (9/3) and 1 (4/4), respectively.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
403
404
BINGFU LU AND ZHISHENG (EDWARD) WEN
18.5
Some Comparisons of the Three Metrics
This section devotes to the comparison of the three metrics in question. We will first discuss the comparison from algorithmic perspective with specific examples. Then, we shall demonstrate how to incorporate these different merits in various parsing models.
18.5.1 Algorithmic Comparison of Some Examples Yngve (1961) posits a Depth Hypothesis, which is encoder-oriented and taken as an indicator of complexity, as shown below. The digits under each word represent the “depth” at the moment the encoder is producing the word. When the encoders produce very, they activate in their plan 4 words (vividly, projected, pictures and appeared); thus, the depth for very is 4, and so on. This metric is clearly encoder-oriented and does not fit a parser, namely, a decoder.8 A preliminary comparison of the three metrics based on Yngve’s example are shown in Figure 18.7. The results seem to be roughly parallel in terms of complexity gradation, though the ratios of increase are different. Table 18.1 shows that Word/IC ratio is not as sensitive as the other two metrics to the wordorder change. If we add the to the NP, the new NP the very vividly projected pictures will contain two CCs (Constructing Categories), the and pictures. If we choose pictures as the major CC, the three numbers of W/IC ratio will remain the same. We thus choose the as the major CC. Below is the result. The calculations of the three sentences with the would be as below. (1’)
Here there are two constructing categories, “the” and “picture.” Which one should we take as the one used to calculate the ratio? If we take the head noun, the result will not change. If we take the definite article, as the first CC encountered by the parser that will construct the relevant mother phrase, the change is considerable, but the three sentences will have the same complexity, as shown in (2’) below.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
Figure 18.6 Depth hypothesis
Figure 18.7 Comparison of MMCN, DMM, and W/IC ratio metrics
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
405
406
BINGFU LU AND ZHISHENG (EDWARD) WEN
Figure 18.7 (cont.)
(2’)
(3’)
The calculations showing in Table 18.2 indicate the following differences. 1. From the perspective of the richness of the resulting digits along with the changes in target sentences, MDD is first, MMCN is the second, and Word/IC the third.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
Table 18.1 First calculation of MMCN, DMM, and W/IC ratio metrics
MMCN MDD W/IC Ratio
a
b
c
1.00 1.00 1.00
1.40 2.25 2.50
1.60 2.00 2.50
Note: The compared three sentences are: (a) Very vividly projected pictures appeared; (b) Pictures very vividly projected appeared; and (c) Pictures projected very vividly appeared.
Table 18.2 Second calculation of MMCN, DMM, and W/IC ratio metrics
MMCN MDD W/IC ratio
a’
b’
c’
1.33 1.60 3.00
1.33 2.00 3.00
1.50 2.20 3.00
Note: The compared three sentences are: (a’), The very vividly projected pictures appeared; (b’), The pictures very vividly projected appeared; and (c’), The pictures projected very vividly appeared.
That W/IC ratio is not as detailed and rich as the other two in reflecting the change of word order should be quite understandable from the point of view that Hawkins’s metric is mainly an index for efficiency, specifically for efficiency in the recognition of structural patterns (Hawkins, 2014; Section 2.4). The importance of pattern recognition has figured prominently in cognition literature. Neither MMCN nor MDD put this point into account. The importance of structural pattern recognition can be seen in the observation that when a constituent can be interpreted either as an argument or as an adjunct, the argument-interpretation makes it easier and faster to understand the whole sentence (Schutze, 1999; Speer & Clifton, 1998). This can be explained by saying that arguments are pattern-relevant constituents that facilitate pattern recognition. In a broad sense, both arguments and CCs are constructing categories, though being phrases and words respectively. 2. While MMCN and MDD do not emphasize pattern recognition, they both take every word that the parser encounters into account. Nevertheless, the difference between MMCN and MDD is also huge. Phrasehood of Xmax plays an important role in MMCN, but the calculation of MDD does not rely on phrasehood. As shown in Figure 18.5, since carefully read is not an Xmax, it cannot count as one momentary chunk. However, though carefully and read cannot form a momentary chunk, a semantic dependency relation does exist between them, which is helpful for parsing. The MDD metric counts this kind of semantic relation.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
407
408
BINGFU LU AND ZHISHENG (EDWARD) WEN
In addition to the above differences among the three metrics, there are many other differences. We just name a few here. MMCN is left-right asymmetric, that is, the results of left-right processing and of right-left are normally very different. MDD is symmetric with the processing direction. Left-right asymmetry is the essential characteristic of online dynamic processing. W/IC ratio is also asymmetric; since a phrase may have more than one CC, and the processing direction may affect the choice of the first CC, leading to different results. MMCN is online, dynamic. It is carried out one word at a time, without skipping any word. MDD does not miss any word as well, but it is static. W/IC ratio is basically dynamic, necessarily from left to right. Nevertheless, it does not count some non-CC words that are not in the scope of the PCD for the phrase in question (since the phrase in question and its ICs can typically be recognized on the basis of a subset of the words dominated by that phrase). The calculation of MCN never needs to look far forward or backward, while that of MDD does when one part of the dependency relation is located far away from the other. Namely, the former is always local, while the latter is not. Needless to say, many other factors, including context, and the like, affect perceptual complexity, sometimes even severely, such as ambiguity and the garden-path effect. However, they are casual, not omnipresent as are dependency relations and other indispensable resources of the three metrics. Therefore, the three metrics can be taken as the complexity base, which may be increased or reduced by other factors relevant for complexity.
18.5.2 Back to the Simplest Parsing Model Each of the three metrics probably has its own (dis)advantages respectively. A more sophisticated parsing model should incorporate the different merits of various parsing models. The problem is how. That the W/IC ratio distinguishes CCs and non-CCs enlightens us to consider distinguishing between content words and function words, in addition to the distinction between head words and nonhead words. Function words usually possess more constructing function than nonhead content words and they usually reduce the perceptual complexity. However, more distinctions make the parsing mechanism more complicated and harder to control. Let us start from a parsing model, which is as simple as possible, and then add more and more distinctions among different words. Suppose we face a string of 6 words ABCDEF, in which there is no difference of head versus dependent, lexical word versus functional one, and the string is unfinished, that is, its end does not appear yet. And suppose further that the 6-word string can be cut into three chunks:[A] [BC] [DEF], and each of them contains 1, 2, and 3 words, respectively. Furthermore, it does not have internal subchunks, that is, each word is a DC of the direct super chunk. Because the words are not categorically different,
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
only the MMCN of the tree metrics can be used for the parsing of the string. We will see that the result shows some important parsing phenomena. (4)
a.
[A] [BC] [D E F]. . .. . .So far, the structural complexity is MCN: 1 2 2 34 3 MMCN: (1+2+2+3+4+3) / 6 = 15 / 6 = 2.50
b.
[A] [D E F] [BC]. . .. . . MCN: 1 2 3 2 3 3 MMCN: (1+2+3+3+3) / 6 = 14 / 6 = 2.33
c.
[D E F] [A] [BC]. . .. . . MCN: 1 2 1 2 3 3 MMCN: (1+2+1+2+3+3) / 6 = 12 / 6 = 2.00
The calculation above shows first that a pure change of the chunk order affects the complexity. It shows secondly that as the largest DC, that is, [DEF], gradually moves forward, the complexity gradually decreases as well.9 This phenomenon indicates that the real-time parsing is left-right asymmetrical, more specifically, cumulative and incremental. Concretely, the momentary complexity of the word D in the strings of (4a), (4b), and (4c) is 3, 2, and 1 respectively. It is clear that everything else being equal, the later the same chunk appears in the string, the harder it is to be processed; because the later it is to be processed, the less the available the mental source of processing is, and the more the cumulative memory burden is. This seems at first glance to be contradictory to the widely observed parsing strategy named “save the hardest for the last,” which reduces the complexity of the overall construction (Bever, 1970). The explanation is that “the last position” is different from other “later positions.” When the parser notices that he enters the last position of the construction, he can use the law of closure principle in Gestalt psychology, which states that one perceives objects such as shapes, pictures, patterns, and the like, as being whole even part of a whole picture is missing. The difficult-reducing effect of this principle can be shown below. (5) MCN1: MCN2:
This is the cat that chased the rat that ate the 1 2 3 4 5 6 7. . .. . .. . .. . .. . . 1
2
3 4
5>1
2
3
4 5>1 2 3
malt
that
was in
4
5>1
2
the house. . .
3. . .. . .. . ..
Without the Gestalt Effect, the MCN would increase rapidly, as shown in the line of the MCN1. Our linguistic intuition disagrees with this calculation. The reason is that on each that, the decoder knows that the sentence enters the last DC of the direct mother node, they would therefore preclose
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
409
410
BINGFU LU AND ZHISHENG (EDWARD) WEN
the direct mother node as if it were a complete chunk because they recognize the structural pattern of the direct mother node already. For example, getting the first that, they know the following material is the modifier of cat, and thus closes the previous 4 words into one chunk and changes the MCN under cat from 5 to 1.10 The parsing model shown above might be the simplest because what it needs is only the notion of a chunk, subchunk, and chunking, the most indispensable element. The model would fit languages whose morphology is very simple and with very limited functional morphemes, such as some isolated languages. The first distinction of morphological category we need to add to the simplest model is the one between head and dependent, shared by all languages. Later on, more and more grammatical distinctions can be brought into consideration. The simpler the basic parsing model is, the more convenient it is to add various parsing-related factors, and to integrate different parsing factors.
18.6
Conclusion: An Introspection on the Language Device
Based on the above analysis, we can come to two general conclusions. First, there should be a clearer distinction between STM limitations versus WM constraints in relation to linguistic complexity. Second, contrary to some previous assumptions, performance may play a role in shaping linguistic structure and complexity.
18.6.1 STM versus WM Limitations Miller (1956b) mentioned a phenomenon called “subitizing,” that is, judging the number of objects directly and quickly without counting. When the number of objects does not exceed five or six, the number judgment is rarely wrong; when the number exceeds five or six, it is difficult to accurately judge, and can only be “estimated.” The number limit for subitizing is less than that of seven. This seems to show the difference to some degree between WM and STM. The WM limit reflects the number of chunks that people can quickly and easily process. MacGregor (1987) has confirmed that the “scan” efficiency is highest for sequences of four groups. This can explain why we have the highest efficiency in memorizing long strings of digits such as telephone numbers. In Cowan (2001), so-called WM is often replaced by “focus of attention.” In fact, “working memory” was only used 45 times, while “focus of attention” was used 83 times. This tells us that his working memory is mainly about “focus of attention,” which can be contrasted to “span of attention,” used by Miller (1956) as an alternative for STM. The difference between “focus” and “span” can be understood as follows: the number 72 is the
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
upper limit on real-time information processing, while 41 is the scope that can be focused on and processed almost instantly. Cowan does use “effective” frequently to describe WM phenomena. We may construe Cowan’s effectiveness based on the experience of daily life. For example, you can memorize a long string of digits by groups of five digits, as a result, you will soon be very tired and will not be able to remember many groups. However, it’s much easier to memorize digit strings by groups of four digits, and you can then remember more groups. As a result, you can remember more digits. For groups of two digits, although it is easier to process each group, the number of groups will increase rapidly. In groups of four, the number of total digits one can memorize is the highest. Therefore, “four” seems to be a sensitive point. Correspondingly, “seven” is the ultimate limitation. For example, a person’s normal appetite is to eat four pieces of bread. With four pieces, he can finish easily. He can eat more, but the eating will become harder and harder, and when he has finished seven pieces, he cannot eat more. This feature of WM may explain the fact that the number of arguments being constituting categories in the sentence cannot exceed three. In parsing, fast recognition of the patterns to be construed in a sentence is extremely important. Only when the number of pattern-determining constituents (the head verb plus its arguments) is no more than four can sentence pattern cognition be fast and easy.
18.6.2
Performance Shapes UG
According to Chomsky (1965), the language device should be unaffected by “such grammatically irrelevant conditions as memory limitations, distractions, shifts of attention and interest, and errors (random or characteristic) in applying his knowledge of the language in actual performance.” However, we argue that memory limitations are different from other performance factors in the sense that they play an essential role all the time in communication. They are not accidental, casual factors for communication but an omnipresent albeit hidden one (Gomez-Rodriguez et al., 2019). Such an indispensable factor should be an essential part of the language device (LD), or in other words, it should be incorporated into the language device. In addition, it is argued that such kinds of cognitive and even perceptual LD-incorporated factors may be much more than what we have expected before. Take one instance, it seems that size-words precede color-words in all head-final NPs, such as in English, we say “the big brown guard dog.” This is because firstly that human cognition is more sensitive to the property “size” than to the property of “color,” and secondly that the property of “size” is easier to be recognized and identified than the property “color” is
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
411
412
BINGFU LU AND ZHISHENG (EDWARD) WEN
in human perception.11 For example, in dim lighting, it is harder to identify the color of an object than its size. Indeed, accumulating evidence from multiple perspectives seems to converge on the instrumental role of these STM/WM limitations in the emergence of different types of language structures and on the way they constrain or shape how human languages are acquired, comprehended, and produced and how they have evolved. As argued in the first section, compelling evidence has now emerged from recent studies in language typology and big-data corpus studies (Futrell et al., 2015; Gibson et al., 2019; Gomez-Rodriguez et al., 2019; Liu, 2008). On the reverse side to the constrained STM/WM limits, the two manifestations of these limitations naturally give rise to two related mechanisms that can be deemed as adaptive and responsive strategies to overcoming the constraining effects of WM limitations through the evolution of language. One of these mechanisms, the articulatory rehearsal mechanism or more recently the procedure of “repetition” or “iteration” (Larsen-Freeman, 2012; cf. Majerus, 2013) is often associated with the phonological component of WM (in Baddeley’s view), as it allows information to be maintained much longer before fading away. The other compensatory mechanism offsetting WM limits is the chunking process, which is claimed to underpin and facilitate the consolidation process of information in (phonological) WM to form larger units or chunks (Ellis, 1996 & 2017). Chunking perceived in this way serves as an effective means to expand the limited capacity or span of WM, such as in the most famous case of expert chess playing (Gobet & Clarkson, 2004). Following these lines of development, in Wen (2016) as well as Wen and Li (2019), we have argued that these two mechanisms are part and parcel of the processes associated with phonological WM (PWM), which in turn carries significant implications for the acquisition and long-term development of key domains of language acquisition and processing such as vocabulary, formulaic sequences, and grammatical constructions in both the native and second language, thus rendering WM as part of the language acquisition device (Wen, 2019; cf. Baddeley et al., 1998; Chomsky, 1965). To conclude, the STM/WM limitations are now reconceptualized and distinguished from each other, consistent with exiting perspectives in cognitive psychology and with evidence from language typology data. Specifically, the STM capacity of 72 signals a limitation of MCN while the WM capacity of 41 implicates MMCN. Overall, it is argued that such a demarcation of the STM/WM distinction should have important implications for both theory construction (in terms of integrating with linguistic frameworks; Wen & Schwieter, this volume) and research methodology (in
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
terms of operationalization or calculations) when they are being implemented in the fields of the language sciences. With such finer-grained distinction among the cognitive constraints, we are now in the golden age of revisiting, revamping and revitalizing the core concept of the language device (LD).
References Andrade, J. (2001). Working memory in perspective. Taylor & Francis. Baddeley, A., Gathercole, S. E., & Papagno, C. (1998). The phonological loop as a language learning device. Psychological Review, 105, 158–173. Baddeley, A. D. (2012). Working memory: Theories, models and controversies. Annual Review of Psychology, 63, 1–30. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. A. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47–89). New York: Academic Press. Barrouillet, P., & Camos, V. (2012). As time goes by: Temporal constraints in working memory. Current Directions in Psychological Science, 21(6): 413–419. Barrouillet, P., & Camos, V. (2015). Working memory: Loss and reconstruction. Psychology Press. Bever, T. G. (1970). The cognitive basis for linguistic structure. Cognition and the development of language. ed. by John R. Hayes, 279–362. Wiley and Sons. Carruthers, P. (2013). The evolution of working memory. Proceedings of National Academy of Sciences, 110 (Suppl. 2), 10371–10378. Chen, B., Ning, A., Bi, H., & Dunlap, S. (2008). Chinese subject-relative clauses are more difficult to process than the object-relative clauses. Acta Psychologica, 129(1), 61–65. Chomsky, N. (1965). Aspects of the theory of syntax. MIT Press. Chomsky, N. (1986). Barrier. MIT Press. Christiansen, M. H., & Chater, N. (2016). Creating language: Integrating evolution, acquisition, and processing. MIT Press. Conway, A. R. A., Jarrold, C., Kane, M. J., Miyake, A., & Towse, J. N. (2007). Variation in working memory. Oxford University Press. Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24, 87–185. Cowan, N. (2005). Working memory capacity. Psychology Press. Cowan N. (2008). What are the differences between long-term, short-term, and working memory? Progress in Brain Research, 169, 323–338. Cowan, N. (2017). The many faces of working memory and short-term storage. Psychonomic Bulletin & Review, 24, 1158–1170. doi: 10.3758/ s13423-016-1191-6 Culicover, P., & Jackendoff, R. (2006). The simpler syntax hypothesis. Trends in Cognitive Sciences, 10, 413–418.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
413
414
BINGFU LU AND ZHISHENG (EDWARD) WEN
Dryer, M. (1992). The Greenbergian word order correlations. Language 68:81–139. Ellis, N. C. (1996). Sequencing in SLA: Phonological memory, chunking and points of order. Studies in Second Language Acquisition, 18, 91–126. Ellis, N. C. (2017). Chunking. In M. Hundt, S. Mollin, & S. Pfenninger (Eds.), The changing English language: Psycholinguistic perspectives (pp. 113–147). Cambridge University Press. Fan, J. (1984). Duoxiang NP ju [Multiple NP Sentences]. Zhongguo Yuwen, 1, 28–34. Ferrer-i-Cancho, R. (2017). Optimization models of natural communication. Journal of Quantitative Linguistics. http://arxiv.org/abs/1412.2486 Ferrer-i-Cancho, R., & Gómez-Rodríguez, C. (2019). Anti-dependency distance minimization in short sequences. A graph theoretic approach. Journal of Quantitative Linguistics, 28(1), 50–76. Frazier, Lyn, (1978). On comprehending sentences: Syntactic parsing strategies.) (Doctoral dissertation, University of Massachusetts–Amherst). Futrell, R. (2017). Memory and locality in natural language (Doctoral thesis, Massachusetts Institute of Technology). Futrell, R., Mahowald, K., & Gibson, E. (2015). Large-scale evidence of dependency length minimization in 37 languages. Proceedings of the National Academy of Sciences, 112(33), 10336–10341. Gibson, E. (1998). Linguistic complexity: Locality of syntactic dependencies. Cognition, 68, 1–76. Gibson, E. (2000). The dependency locality theory: A distance-based theory of linguistic complexity. Image, language, brain 2000, 95–126. Gibson, E., Futrell, R., Piantadosi, S., Dautriche, I., Mahowald, K., Bergen, L., & Levy, R. P. (2019). How efficiency shapes human language. Trends in Cognitive Sciences, 23(5), 389–407. Gobet, F., & Clarkson, G. (2004). Chunks in expert memory: Evidence for the magical number four. . .or is it two? Memory, 12, 732–747. http://dx.doi .org/10.1080/09658210344000530 Gomez-Rodriguez, C., Christiansen, M., & Ferrer-i-Cancho, R. (2019). Memory limitations are hidden in grammar. arXiv preprint arXiv:1908.06629. https:// doi.org/10.7910/DVN/XHRIYX Gómez-Rodríguez, C., Christiansen, M. H., & Ferrer-i-Cancho, R. (2020, April 14–17). Cognitive constraints built into formal grammars: Implications for language evolution. In A. Ravignani et al. (Eds.), The evolution of language: proceedings of the 13th international conference (EvoLang13). Brussels, Belgium. Hakes. D. T. (1972). Effects of reducing complement construction on sentence comprehension. Journal of Verbal Learning and Verbal Behavior. 11, 278–286 Hawkins, J. A. (2004). Efficiency and complexity in grammars. Oxford University Press. Hawkins, J. A. (2014). Cross-linguistic variation and efficiency. Oxford University Press
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
Hocket, C. (1961). Linguistic elements and their relation. Language, 37, 29–53. Hudson, R. (1995). Word meaning. Routledge. Jackendoff, R. (2007). A parallel architecture perspective on language processing. Brain Research, 1146, 2–22. Jackendoff, R. (2011). What is the human language faculty? Two views. Language, 87, 586–624. Kimball, J. (1973). Seven principles of surface structure parsing in natural language. Cognition, 2, 15–46. Larsen-Freeman, D. (2012). On the roles of repetition in language teaching and learning. Applied Linguistics Review, 3, 195–210. Liu, H. (2008). Dependency distance as a metric of language comprehension difficulty. Journal of Cognitive Science, 9(2), 159–191 Logie, R. H. (1996). The seven ages of working memory. In J. T. E. Richardson, R. W. Engle, L. Hasher, R. H. Logie, E. R. Stoltzfus, & R. T. Zacks (Eds.), Working memory and human cognition (pp. 31–65). Oxford University Press. Logie, R. H., Camos, V., & Cowan, N. (2021). Working memory: State of the science. Oxford University Press. Lu, Bingfu. (1983). Wuxian digui de tiaojian he youxian qiefen [The conditions of infinite recursion and finite segmentation]. Hanyu Xuexi [Chinese Learning], 1, 23–29. Lu, Bingfu. (1993). Hexin Tuidao Yufa [A head-oriented grammar]. Shanghai Education Press. Lu, Bingfu. (2001). What is the chunk in linguistic construction? Proceedings of the Third International Conference on Cognitive Science: 452–457. University of Science and Technology of China Press. Lu, Bingfu. (2009). The parallelism between NPs and clauses in terms of pragmatic effects on word order, Cahiers Linguistique–Asie Orientale 38(2), 177–219. Lu, Q., Xu, C., & Liu, H. (2016). Can chunking reduce syntactic complexity of natural languages? Complexity, 21, 33–41 Majerus, S. (2013). Language repetition and short-term memory: An integrative framework. Frontiers in Human Neuroscience 7, 357. doi: 10.3389/ fnhum.2013.00357 Miller, G. (1956a). The magical number seven plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Miller, G. (1956b). Human memory and the storage of information. IRE Transaction on Information Theory, 2(3), 129–137. Miller, G., Galanter, E., & Pribram, K. H. (1960). Plans and the structure of behavior. Holt. Miyake, A., & Shah, P. (1999). Models of working memory: Mechanisms of active maintenance and executive control. Cambridge University Press.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
415
416
BINGFU LU AND ZHISHENG (EDWARD) WEN
Nicenboim, B., Vasishth, S., Gattei, C., Sigman, M., & Kliegl, R. (2015). Working memory differences in long-distance dependency resolution. Frontiers in Psychology (6), 312. Oberauer, K., & Lewandowsky, S. (2013). Evidence against decay in verbal working memory. Journal of Experimental Psychology: General, 142, 380–411. Oberauer, K., & Lewandowsky, S. (2014). Further evidence against decay in working memory. Journal of Memory and Language, 73, 15–30. O’Grady, W. (2012). Three factors in the design and acquisition of language. Wiley Interdisciplinary Reviews: Cognitive Science, 3, 493–499. O’Grady, W. (2015). Processing determinism. Language Learning, 65(1), 6–32. O’Grady, W. (2017). Working memory and language: From phonology to grammar. Applied Psycholinguistics, 38(06), 1340–1343. Pierce, L. J., Genesee, F., Delcenserie, A., & Morgan, G. (2017). Variations in phonological working memory: Linking early language experiences and language learning outcomes. Applied Psycholinguistics, 38, 1265–1302. Schütze, C. (1999). English expletive constructions are not inflected. Linguistic Inquiry, 30(3), 467–484. Speer, S. R., & Clifton C., Jr. (1998). Plausibility and argument structure in sentence comprehension. Memory and Cognition, 26(5), 965–978. Waugh, N. C., & Norman, D. A. (1965). Primary memory. Psychological Review, 72, 89–104. Wells, R. S. (1947). Immediate constituent, Language, 23, 81–118. Wen, Z. (2016). Working memory and second language learning: Towards an integrated approach. Multilingual Matters. Wen, Z. (2019). Working memory as language aptitude: The Phonological/ Executive Model. In Z. Wen, P. Skehan, A. Biedron, S. Li, & R. Sparks (Eds.), Language aptitude: Advancing theory, testing, research and practice (pp. 187–214). Routledge. Wen, Z., & Li, S. (2019). Working memory in L2 learning and processing. In J. Schwieter & A. Benati (Eds.), The Cambridge handbook of language learning (pp. 365–389). Cambridge University Press. Yngve, V. H. (1961). Depth hypothesis. Proceedings of Symposia in Applied Mathematics, 11, 130–138 Zipf, G. (1949). Human behavior and the principle of least effort: An introduction to human ecology. Hafner.
Notes 1 The notion of the head has been extensively discussed in linguistics, which is beyond the limits of the current chapter. For the current purpose, suffice it to say that only “lexical heads” count, excluding “functional heads” in formal linguistics. 2 In contrast, Dryer (1992, 112–115) further treats the Major Constituent Trees as if they were flat structures, where the distance difference of the
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Chunking and Parsing Complexity
3
4 5
6
7
8
9
10 11
dependents to the head cannot be displayed clearly, especially when the dependents are on the two sides of the head. For simplicity, the dependents in the fictional sentence contain only adjuncts, but not arguments. The orders in Figure 18.3 are canonical. They are often overridden by superficial pragmatic motivations. For example, heavy chunks tend to appear in peripheral positions. Instead of zero as the minimal distance between two adjacent words (Hudson, 1995), Liu (2008) adopts one as the minimal distance, which is easy to express dependency direction with positive or negative digits in more sophisticated computation (Liu, 2010). Meanwhile, zero can be used to represent the DD of the root governor that is not dependent on any word. Here we use super- and subscripts instead of the arrowed curves. More recently, Futrell et al. (2015), based on their 37-language corpus, posit the principle of Dependency Length Minimization, sharing the essence of the findings by Liu (2008). Most CCs are MNCCs (within the kind of traditional Aspects-style generative syntax assumed in Hawkins 1994, and the Simpler Syntax assumed more recently in Culicover & Jackendoff 2006). Cf. Mother Node Construction (Hawkins, 1994, p. 62; 2014, p. 91), and The Axiom of Constructability (1994, pp. 379–380, 422). Fan (1984) finds that the number of NPs that can appear at the beginning of a “multiple NP sentence” cannot exceed 4. This might be also relevant to WM. Decoder-oriented processing is much simpler and more basic than an encoder-oriented one. An encoder must be a decoder concurrently because he must monitor his own encoding product, while a decoder does not need to be an encoder simultaneously. In fact, an ideally neutral decoder should not perform any encoding job, such as prediction, planning of future words. Synchronizing with Yngve (1961), Hockett (1961) posits a decoder-oriented parser. The difference between his approach and the current one is that his approach is mainly prospective in the sense that the parser’s major job is to predict the coming possible syntactic categories, while the current approach is completely retrospective. If the 6-word string is finished as a whole chunk, the last digit would reduce to 1, and the corresponding MMCN would correspondently reduce. The Gestalt Effect also shows the importance of early pattern recognition. Lu (2009) proposes a word order rule, which states that the easier the referent of a linguistic unit is to identify, the stronger is its tendency to precede. This rule can be seen as one realization of the information packaging from old/definite to new/indefinite.
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
417
https://doi.org/10.1017/9781108955638.022 Published online by Cambridge University Press
Part IV
First Language Processing
https://doi.org/10.1017/9781108955638.023 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.023 Published online by Cambridge University Press
19 Working Memory in Word Reading Sun-A Kim
19.1
Introduction
Working memory (WM) is one of the most significant cognitive factors affecting global reading and word reading. Closely linked to long-term memory (LTM), WM is characterized by its limited capacity available for maintaining and processing information. According to a multicomponent WM model, WM consists of four elements: the phonological loop, the visuospatial sketchpad, the episodic buffer, and the central executive (Baddeley, 2012). The phonological loop and the visuospatial sketchpad maintain and store phonological/verbal information and visuospatial input, respectively. The episodic buffer holds multidimensional chunks/episodes, binds information from various sources, and links WM to LTM. The central executive plays a role in focusing, dividing, and switching attention between targets from different modalities (Baddeley, 2000, 2012). Reading is a cognitive process of transferring written symbols to spoken language for comprehension of a given text, which involves bottom-up and top-down processing simultaneously. Through bottom-up processing, a reader identifies words and constructs meaning from them. At the same time, in the course of reading, the reader’s own prior knowledge related to the text is naturally activated, and hypothesis making and testing about the text takes place in the reader’s mind, which is called top-down processing. Bottom-up and top-down processing are interactive and inevitable in reading, and successful bottom-up processing, especially word reading, is an essential foundation of skilled reading. Due to the finite storage and processing feature of WM, if a reader consumes too much WM in bottom-up processing (e.g., decoding or understanding words), the person has less capacity available for top-down processing and then fails to comprehend the given text. Previous studies found that WM is one of the best predictors of reading comprehension (e.g., Daneman & Carpenter, 1980), and particularly the phonological loop has been demonstrated to be crucial in word
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
422
SUN-A KIM
reading and vocabulary learning by children and adults (see Baddeley et al., 1998 for a review). A large number of diverse WM models have been proposed (see Cowan, 2017 for a review). In the late 1990s, researchers determined that these seemingly disparate models actually converge on fundamental issues about WM but differ in their emphases and definitions. For example, the focus of Baddeley and colleagues’ research is on the phonological loop and the multicomponent features of WM, while Engle and colleagues concentrate on the roles of the central executive and controlled attention (see Wen, 2016 for a summary). Another view is that WM is not completely unitary/ domain-general nor entirely domain-specific (Miyake and Shah, 1999), and the hierarchical view of WM proposed by Engle, Kane et al. (1999) is an example. The hierarchical view of WM captures the dual nature of WM (Miyake, 2001) and deems WM a “hierarchical structure with a general domain-free factor overarching several subordinate domain-specific factors” (Engle, Kane et al., 1999, p. 125). This chapter views WM from the multicomponent and hierarchical perspective that WM has a limited storage element for different sensory domains (i.e., domain-specific short-term memory) plus finite attention (i.e., the domain-general central executive) (Engle, Tuholski et al., 1999; Kim et al., 2015). Therefore, in this chapter, verbal WM refers to WM which processes phonological/verbal information and maintains it in the phonological/verbal domain, whereas visuospatial WM refers to WM which processes and retains visuospatial information in the visuospatial domain. In contrast, verbal STM or visuospatial STM both denote a simple, passive storage unit without a processing role in each domain. WM is assessed by complex WM tasks taxing both processing and holding information (e.g., a backward recall task or a reading/listening/operation span task), while STM is measured by simple memory tasks mainly requiring memorization without simultaneous processing duties (e.g., a forward recall task). This clear distinction for WM is necessary and important because contradictory results related to WM and word reading in the literature seem to be mainly attributed to indistinct differentiation between WM and STM and to inconsistent or irrelevant task choices for measuring WM. In this chapter, research on word reading is first summarized from the perspectives of two distinctive modalities in the Baddeley’s WM model (i.e., phonological and visuospatial memory and processing). Then, issues found in recent studies are addressed on WM in Chinese word reading. Finally, areas that are worthwhile exploring in the future are suggested.
19.2
Verbal Memory and Processing in Word Reading
Verbal STM is essential to word reading, which can be assessed in various ways, including recalling a list of orally given words. Good readers tend to
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
Working Memory in Word Reading
show better memory than poor readers, and short-term verbal memory correlates with rudimentary reading (e.g., Wagner et al., 1997). Phonological awareness and rapid automated naming (RAN) have been documented as significant elements of word reading (see McBride, 2016, for a summary). Phonological awareness means the awareness of the sound system of a given language or sensitivity to its spoken language structure, such as phonemes, onsets, rimes, syllables, or tones for tonal languages. Awareness of each phonological unit in a language is measured by requiring participants to manipulate the target unit. For example, “How does cat sound without /k/?” is a question of an onset-awareness task (specifically, an onset-deletion task), and “Say rainbow without rain” is a question within a syllable-awareness task (specifically, a syllable-deletion task). A RAN task asks participants to name familiar objects, words, or digits as rapidly as possible and then measures the duration of naming. According to a review of the literature by McBride (2016), children’s phonological awareness and RAN are found to be more robust predictors of reading than verbal STM, and it is rare to find unique variance for verbal STM in reading when phonological awareness, RAN, and verbal STM are all considered – even though there is a correlation between verbal STM and word reading. This rarity probably stems from the fact that verbal STM is already incorporated in phonological awareness and RAN, such that without verbal STM, manipulating phonemes in a phonological awareness task or rapid naming of words or digits in a RAN task is impossible. Verbal STM, as a more essential and foundational skill for word reading, might develop before children’s phonological awareness or RAN abilities grow (McBride, 2016). The significance of verbal WM in reading has been demonstrated mostly in studies of English reading. Alloway and Alloway (2010) explored the unique contribution of verbal WM in literacy and numeracy with a longitudinal study in which each child was tested at two points in time. Participants were first tested at the age of 5 (Time 1) and then after six years when they were 11 years old (Time 2). To measure participants’ verbal WM, complex WM tasks involving both storage and processing (a backward digit span task and a listening recall test) were used. The backward digit span task required children to listen to a series of digits and then recall the digits in reverse order. In the listening recall test, participants had to judge the sentence they had just heard as true or false and simultaneously remember the last word in the sentence, followed by a recall task in which they recalled the final words of the previous sentence in correct order. The verbal WM measures were administered at both Time 1 and Time 2, while participants’ literacy and numeracy abilities were only tested at Time 2. Regression analysis suggested that verbal WM at Time 1 was a strong predictor of literacy in Time 2. Leather and Henry (1994) investigated the relationship between verbal WM, STM, and English reading ability among 7-year-old English-speaking children. Children’s verbal WM was measured by a listening span task and a counting span task. A regression analysis
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
423
424
SUN-A KIM
indicated that the listening span task made an independent and significant contribution to reading even after statistically controlling for the effects of the counting span test, the STM test, and the phonological awareness test. Comparisons between children with reading difficulties and normal reading children also revealed the importance of verbal WM in word reading. de Jong (1998) employed various measures for verbal STM, verbal WM, and processing speed and found that 10-year-old Dutch children whose single-word decoding abilities were very low performed worse on all measures than their normal reading peers. The results suggested that children with reading disabilities seem to have deficiency in the central executive which manages synchronous processing and storage of verbal information. Wang and Gathercole (2013) compared the verbal and visuospatial STM and WM of two groups of native English-speaking children aged 8 to 10 years. One group of children had some difficulties in reading single words, while the other group were typical readers whose age and cognitive abilities were matched with the first group. Their verbal STM and WM were measured by forward digit recall and backward digit recall, respectively. Visuospatial STM was assessed by a dot matrix task in which participants were asked to remember the position of a red dot in a series of 4 4 grids and then to recall the positions in the correct order, and visuospatial WM by a spatial recall task in which children were required to judge whether a series of a pair of shapes (one with a red dot and the other without a red dot) are the same or not and simultaneously remember the location of each red dot in the correct serial order. In all four memory measures, the scores of children with reading difficulties were significantly lower than those of typical readers. More importantly, group differences in verbal and visuospatial WM remained significant even after adjusting relevant STM, suggesting that the central executive, rather than the memory component of verbal or visuospatial domain, plays a core role in children’s word reading.
19.3
Visuospatial Memory and Processing in Word Reading
In this section, the focus will be on visuospatial memory and processing in Chinese word reading because visuospatial WM is more applicable to visually complex Chinese character/word reading than to alphabetic word reading where phonological processing plays a more important role than visuospatial analysis. A Chinese character consisting of a number of strokes is not only compressed into a square regardless of its visual complexity, but also sometimes has visually similar counterparts which differ only slightly in length or the slant of a stroke (e.g., 天 vs. 夫; 田 vs. 由; 干 vs. 千). More than 4,500 characters are required for fluent reading in Chinese (McBride, 2016), of which only around 2,600 frequently used characters are taught in elementary school in China (Shu et al., 2003). Despite the vast number of characters necessary for skilled reading, spoken Mandarin has about 1,200 syllables, including tones (Norman, 1988), which results in numerous https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
Working Memory in Word Reading
homophones. Thus, it is likely that reading Chinese characters requires not only visuospatial skills to correctly identify target characters and retrieve relevant information about them, but also attention-controlling skills to suppress visually similar characters or homophones which are irrelevant but automatically activated in the course of processing the given characters. Research on the role of visual skills in Chinese reading, especially among the normal population without reading difficulties, has been less explored, with the focus mostly on general visual skills , not on visuospatial WM. The results of numerous studies on general visual skills in Chinese reading development over two decades have not converged on a certain agreement (McBride, 2016; Yang et al., 2013). Some studies have found a significant relationship between various aspects of general visual ability and character reading (e.g., Huang & Hanley, 1995; McBride-Chang & Chang, 1995; Siok & Fletcher, 2001), while others failed to prove association of visual skills with character recognition (e.g., Ho, 1997; Huang & Hanley, 1997; McBride-Chang & Ho, 2000). It seems plausible that visuospatial WM approaches can shed new light on visuospatial processing in Chinese word reading. Some of the previous studies on Chinese children’s general visual skills included a visual STM test and showed that good readers have better visual STM than poor readers (e.g., Huang & Hanley, 1995; So & Siegel, 1997). Recently, Xu et al. (2020) adopted visuospatial WM approaches by using STM and WM measures (forward and backward Corsi block tasks) and unconventional analyses (treating STM and WM scores as separate variables), and found that Chinese children’s visuospatial WM, not visuospatial STM, accounted for the unique variance in character reading. It is likely that a clearer picture would emerge about visuospatial processing in Chinese word reading if researchers employ visuospatial WM measures and analysis methods that are proven to be valid and reliable in the literature.
19.4
Research on WM in Chinese Word Reading
WM has been actively investigated in Chinese word reading by children with reading difficulties or in native speakers’ Chinese reading comprehension, but the number of studies on WM in typically developing children’s character/word reading is quite limited (Yang et al., 2019). It appears that WM approaches to analyses of normal Chinese children’s word reading started not long ago. In this section, five recent studies on WM in normal children’s Chinese word reading will be focused upon, particularly those that address issues related to the conceptualization of WM and measures. Table 19.1 summarizes the five studies. Although there are at least nine major definitions or models of WM in the field (Cowan, 2017), the five studies on WM and Chinese word reading all adopted Baddeley’s WM model, which assumes that WM comprises four components of limited storage or processing capacity (the phonological loop, the visuospatial sketchpad, the episodic buffer, and the central https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
425
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
Table 19.1 Summary of studies containing a WM variable in Chinese word reading Measures Study
Participants
Mean age
Independent Variables
Dependent Variable
Analysis
Results
Chung & McBrideChang (2011)
85 Hong Kong kindergarteners (L1 Cantonese)
• Time 1: 4.99 yrs • Time 2: 5.92 yrs
Executive function (EF) [inhibitory control & WM]
Word reading (Reading aloud)
Hierarchical regression
• EF explained 14–16% of the variance in word reading in Time 1 & Time 2. • EF contributed unique variance in word reading in Time 2 with word recognition at Time 1 controlled.
Chung et al. (2018)
Yang et al. (2019)
369 Hong Kong kindergarteners (L1 Cantonese)
4.83 yrs
189 Chinese kindergarteners (L1 Mandarin)
5.73 yrs
• Visuomotor integration (VMI) • Executive function (EF) [WM, inhibitory control & cognitive measure]
Word reading & writing (Reading aloud & word writing)
Hierarchical regression
• General cognitive skills [WM, inhibition, & reasoning] • Metalinguistic awareness [phonological awareness, morphological awareness, & orthographic knowledge]
Character reading (Reading aloud)
Hierarchical regression
[Controlled age, vocabulary knowledge, phonological awareness & morphological awareness] VMI & EF were uniquely linked to word reading & writing. [Controlled age, gender, RAN & maternal education] General cognitive processing & metalinguistic awareness explained 16% and 20% of the variance in character reading. [Controlled age, gender, & receptive vocabulary]
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
Xu et al. (2020)
Wang et al. (2017)
154 Chinese primary school students (L1 Mandarin)
7.17 yrs (Grade 1)
• 8.31 yrs (Grade 2 • 66 Taiwanese [G2]; 31 students) primary school • 9.29 yrs (Grade 3 students (L1 [G3]; 35 students) Mandarin) • 28 college students • 21.32 yrs (young adults [YA])
• Visual WM • Visual temporal order judgment (VTOT)
• Binding memory (as an episodic buffer task) • Auditory-verbal memory • Visual memory
Character reading (Reading aloud)
Character reading (Writing pronunciation in Zhuyin Fuhao, the transliteration system used in Taiwan)
Moderation effects analysis
Hierarchical regression
• Visual WM & VTOT predicted character reading. • Moderation effect of VTOT on visual WM-character reading link [Controlled age, non-verbal intelligence, visual shortterm memory, morphological awareness, & orthographic awareness] Binding memory accounted for unique variance in character recognition. [Controlled age, auditoryverbal memory, visual memory, phonological awareness, & RAN]
3 3 ANOVA
3 3 ANCOVA
[age group x memory condition] • Binding memory: YA > G3 > G2 • Auditory-verbal memory: YA > G3 ~ G2 • Visual memory: YA > G3 ~ G2 [visual memory & auditoryverbal memory
as covariates] Binding memory: YA > G3 > G2
428
SUN-A KIM
executive). Nevertheless, inconsistency between the five studies is present in the details of viewing WM and choosing WM tasks, which may be responsible for the unclear and complex conclusions about WM and Chinese word reading in the studies. Chung and McBride-Chang (2011) and Chung et al. (2018) included WM as a component of a broader cognitive function called executive functioning (EF). Their definition of EF is the ability to regulate one’s attention, emotions, and behaviors, and in which WM, inhibitory control, and cognitive flexibility play major roles. Even though the two studies’ definitions of EF and WM were identical, the two groups of authors used very different measures for these constructs involving subjects with similar age ranges, native language, and backgrounds. Chung and McBride-Chang (2011) employed forward and backward digit span tests for WM, while Chung et al. (2018) used the hearts-and-flowers task consisting of three parts: the congruent block (used for measuring WM), the incongruent block (for inhibitory control), and the mixed block (for cognitive flexibility). The forward digit span task and the backward digit span task in Chung and McBride-Chang (2011) are known to measure verbal STM and the central executive, respectively, while the congruent block of the hearts-and-flowers task could be closer to assessing verbal STM (a passive storage mechanism) than verbal WM (a limited memory for both storing and processing). Yang et al. (2019) viewed WM as one of the three most important general cognitive skills in reading development, along with inhibition and reasoning. Yang and colleagues treated these three skills as separate, distinctive cognitive constructs. Inhibition was defined as “the ability to focus on attention and deliberately suppress dominant or automatic responses when necessary,” and reasoning as “the ability to use logic in new situations to identify patterns and solve novel problems.” The components of WM, including the phonological loop, the central executive, and the visuospatial sketchpad, were measured by a digit recall forward task, a digit recall backward task, and a simple visuospatial memory task that is similar to a forward Corsi test. The combined scores of the three tasks were used for analysis as the WM variable. The focus of Xu et al. (2020) was visual WM. A backward Corsi task was used for visual WM and a forward Corsi task for visual STM. The scores of visual WM and visual STM were included separately for analysis as independent variables. The approaches to measurements and analyses of Xu et al. (2020) are closer to common methods in the broader WM literature. Xu and colleagues found that visual WM involving both storage and processing explained unique variance in character reading, but not visual STM, which suggests that character recognition involves not only visuospatial memory but also processing in the visuospatial domain. The results are congruent with previous studies suggesting that complex WM measures (regardless of domain) involving both storage and processing have stronger predictive power than simple STM measures.
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
Working Memory in Word Reading
To investigate the role of the episodic buffer in character reading, Wang et al. (2017) developed the “binding memory task” as a measurement for the episodic buffer as well as two simple memory tasks for the auditory/ phonological and the visual domains (i.e., the auditory-verbal and the visual memory task). As an episodic buffer task, the binding memory task aims to measure the ability of binding information from two modalities, which were operationalized in this study as an associative learning task employing eight spoken syllables and eight abstract and nonnamable shapes. The auditory-verbal memory task and the visual memory task challenged subjects to recall the sequence of auditorily-given and visually-given items, respectively. Wang et al. (2017) is the first attempt to explore the episodic buffer in Chinese character/word reading. Within the five studies, in some cases there are grounds for criticism that the concept of WM might have been too general or abstract. In Chung and McBride-Chang (2011), Chung et al. (2018), and Yang et al. (2019), the construct of WM was viewed as a component of EF or general cognitive skill, but it was not clearly explained how the central executive of WM differs from EF or general cognitive skill. Also, even within the same WM framework, which assumes a multicomponent cognitive system involving both the maintenance and processing of information, various WM measures were used without specifying their purposes (e.g., whether a task was meant to measure passive STM or complex WM, which taxes both storage and processing in the phonological or the visuospatial domain). This indistinction hinders the revelation of how WM works in the process of reading Chinese words. There is a general consensus in the literature regarding WM measurement. A task of counting backward (e.g., backward digit span or backward Corsi block tasks) is an attention-demanding task requiring both retention of information and division of attention. Backward-focused tasks tax the central executive more than counting forward tasks because counting backwards requires the central executive to function by dividing and controlling attention in contrast to simply retaining information as a forward digit span demands (Baddeley, 2012). However, this distinction does not mean that backward digit span tasks are necessarily verbal WM tasks or that forward digit span tasks are necessarily verbal STM tasks, because the involvement of the central executive can vary depending on the subjects and the features of the items recalled. For example, Engle, Tuholski et al. (1999) categorized a backward word span in a STM assessment for their participants who were American college students, but acknowledged that the same task can function as a WM task for children because children place much higher demand on the central executive than adults do in order to carry out the same task. In sum, WM tasks should be selected carefully based on the study’s purposes and the characteristics of the subjects, and followed by clear explanation of their rationales for use. Then, the subsequent results of the experiments can be interpreted to unveil the mechanism of WM in Chinese word reading.
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
429
430
SUN-A KIM
19.5
Conclusion and Future Directions
In examining working memory in word reading, there are several areas of research that merit further exploration. In this section, some of these issued will be discussed.
19.5.1 Impact of Learning to Read on WM Development To date, research has been conducted mostly in one direction: the impact of WM on word reading. However, a reverse direction of investigation is not only possible but also promising. It appears that reading experience with visually complex Chinese characters aids in the development of native speakers’ visuospatial skills or processing. Cross-cultural studies have shown that Chinese children’s visual skills exceeded those of young alphabetic readers of Greek (Demetriou et al., 2005), Spanish, and Hebrew (McBride-Change et al., 2011). Demoulin and Kolinsky (2016) provides an extensive review on the proposal that learning to read shapes the development of verbal WM. In the future, this shifted perspective needs to be extended to more languages, diverse age groups, and research questions from various perspectives. Also, more significant results could be found by comparing the verbal or visuospatial WM of readers who use distant or contrastive writing systems.
19.5.2
Episodic Buffer and Visuomotor Aspect in Character Reading and Writing The episodic buffer is “a limited-capacity temporary storage system that is capable of integrating information from a variety of sources” (Baddeley, 2000) and thus plays a role in “binding of information in different codes from different sources” including visual, auditory, tactile, episodic, and semantic information (Rudner & Rönnberg, 2008). One example of episodic buffer involvement is writing Chinese characters/words. Visuomotor skills (e.g., copying or writing characters/words) have been found to be very important in Chinese word reading. Chinese children’s skills of copying unfamiliar characters predicted character reading (Tan et al., 2005), and children who participated in a character-copying intervention program for eight weeks outperformed those who did not do copying in word reading (Wang & McBride, 2017). Since copying and writing characters involves visual, auditory (in case of familiar character/words), and tactile codes, all information can be combined and integrated in the episodic buffer. Knowing more about the role of the episodic buffer in Chinese character/ word writing could enlighten us to the mechanisms of Chinese character/ word reading and the episodic buffer.
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
Working Memory in Word Reading
19.5.3 WM in Nonnative Speakers’ Chinese Word Learning Research is lacking on WM of Chinese word reading for second language readers of Chinese. Understanding the influence of reading experience in one’s native language versus reading in Chinese as a second language would shed light on WM in word reading, Chinese reading, and second language reading. Kim (2010) and Kim et al. (2015) conducted the first study on the role of visuospatial and verbal WM in second language speakers’ learning to read Chinese characters, and found that verbal WM, not visuospatial WM, was the most significant predictor of second language readers’ character learning when the readers’ native languages had alphabetic writing systems such as English, Korean, or Thai. These results suggest that second language readers might follow a similar pattern of using WM to read in their native language when learning to read in Chinese as a second language. The authors surmised that the results may have been attributable to the short period of time that the second language learners had been studying Chinese, from six weeks to fewer than three semesters. Further studies need to explore whether verbal WM continues to be a significant predictor for character learning or whether visuospatial WM would emerge as an influential variable due to the extensive reading experience in Chinese among very fluent, skilled second language readers of Chinese whose native scripts are alphabetic. Such studies might answer the question of whether substantial reading experience in visually complex and complicated Chinese could shape the use of cognitive resources among not only native speakers, but also second language learning populations. Diverse interpretations of the role of WM in nonnative speakers’ Chinese word learning (e.g., Opitz et al., 2014; Wang et al., 2017) will bring fresh perspectives to the field.
References Alloway, T. P., & Alloway, R. G. (2010). Investigating the predictive roles of working memory and IQ in academic attainment. Journal of Experimental Child Psychology, 106(1), 20–29. Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417–423. Baddeley, A. (2012). Working memory: Theories, models, and controversies. Annual Review of Psychology, 63, 1–29. Baddeley, A., Gathercole, S., & Papagno, C. (1998). The phonological loop as a language learning device. Psychological Review, 105(1), 158. Cowan, N. (2017). The many faces of working memory and short-term storage. Psychonomic Bulletin and Review, 24(4), 1158–1170.
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
431
432
SUN-A KIM
Chung, K. K. H., Lam, C. B., & Cheung, K. C. (2018). Visuomotor integration and executive functioning are uniquely linked to Chinese word reading and writing in kindergarten children. Reading and Writing, 31(1), 155–171. Chung, K. K., & McBride-Chang, C. (2011). Executive functioning skills uniquely predict Chinese word reading. Journal of Educational Psychology, 103(4), 909–921. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. de Jong, P. F. (1998). Working memory deficits of reading disabled children. Journal of Experimental Child Psychology, 70(2), 75–96. Demetriou, A., Kui, Z. X., Spanoudis, G., Christou, C., Kyriakides, L., & Platsidou, M. (2005). The architecture, dynamics, and development of mental processing: Greek, Chinese, or Universal? Intelligence, 33(2), 109–141. Demoulin, C., & Kolinsky, R. (2016). Does learning to read shape verbal working memory? Psychonomic Bulletin and Review, 23(3), 703–722. Engle, R. W., Kane, M. J., & Tuholski, S. W. (1999). Individual differences in working memory capacity and what they tell us about controlled attention, general fluid intelligence, and functions of the prefrontal cortex. In A. Miyake, & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 102–134). Cambridge University Press. Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. (1999). Working memory, short-term memory, and general fluid intelligence: A latent-variable approach. Journal of Experimental Psychology: General, 128 (3), 309. Ho, C. S. H. (1997). The importance of phonological awareness and verbal short-term memory to children’s success in learning to read Chinese. Psychologia: An International Journal of Psychology in the Orient, 40(4), 211–219. Huang, H. S., & Hanley, J. R. (1995). Phonological awareness and visual skills in learning to read Chinese and English. Cognition, 54(1), 73–98. Huang, H. S., & Hanley, J. R. (1997). A longitudinal study of phonological awareness, visual skills, and Chinese reading acquisition among firstgraders in Taiwan. International Journal of Behavioral Development, 20(2), 249–268. Kim, S. (2010). Developmental stages in reading Chinese as a second language. (Doctoral dissertation, University of Illinois at Urbana-Champaign). Kim, S., Christianson, K., & Packard, J. (2015). Working memory in L2 character processing: The case of learning to read Chinese. In Z. Wen, M. B. Mota, & A. McNeill (Eds.), Working memory in second language acquisition and processing (pp. 85–104). Multilingual Matters. Leather, C. V., & Henry, L. A. (1994). Working memory span and phonological awareness tasks as predictors of early reading ability. Journal of Experimental Child Psychology, 58(1), 88–111.
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
Working Memory in Word Reading
McBride, C. (2016). Children’s literacy development: A cross-cultural perspective on learning to read and write (2nd ed). Routledge. McBride-Chang, C., & Chang, L. (1995). Memory, print exposure, and metacognition: Components of reading in Chinese children. International Journal of Psychology, 30(5), 607–616. McBride-Chang, C., & Ho, C. S. H. (2000). Developmental issues in Chinese children’s character acquisition. Journal of Educational Psychology, 92(1), 50. McBride-Chang, C., Zhou, Y., Cho, J. R., Aram, D., Levin, I., & Tolchinsky, L. (2011). Visual spatial skill: A consequence of learning to read? Journal of Experimental Child Psychology, 109(2), 256–262. Miyake, A. (2001). Individual differences in working memory: Introduction to the special section. Journal of Experimental Psychology: General, 130(2), 163. Miyake, A., & Shah, P. (Eds.). (1999). Models of working memory: Mechanisms of active maintenance and executive control. Cambridge University Press. Norman, J. (1988). Chinese. Cambridge University Press. Opitz, B., Schneiders, J. A., Krick, C. M., & Mecklinger, A. (2014). Selective transfer of visual working memory training on Chinese character learning. Neuropsychologia, 53, 1–11. Rudner, M., & Rönnberg, J. (2008). The role of the episodic buffer in working memory for language processing. Cognitive Processing, 9(1), 19–28. Shu, H., Chen, X., Anderson, R. C., Wu, N., & Xuan, Y. (2003). Properties of school Chinese: Implications for learning to read. Child Development, 74(1), 27–47. Siok, W. T., & Fletcher, P. (2001). The role of phonological awareness and visual-orthographic skills in Chinese reading acquisition. Developmental Psychology, 37(6), 886. So, D., & Siegel, L. S. (1997). Learning to read Chinese: Semantic, syntactic, phonological and working memory skills in normally achieving and poor Chinese readers. Reading and Writing, 9(1), 1–21. Tan, L. H., Spinks, J. A., Eden, G. F., Perfetti, C. A., & Siok, W. T. (2005). Reading depends on writing, in Chinese. Proceedings of the National Academy of Sciences, 102(24), 8781–8785. Wagner, R. K., Torgesen, J. K., Rashotte, C. A., Hecht, S. A., Barker, T. A., Burgess, S. R., . . . & Garon, T. (1997). Changing relations between phonological processing abilities and word-level reading as children develop from beginning to skilled readers: A 5-year longitudinal study. Developmental Psychology, 33(3), 468. Wang, S., Allen, R. J., Fang, S. Y., & Li, P. (2017). Cross-modal working memory binding and L1–L2 word learning. Memory and Cognition, 45(8), 1371–1383. Wang, S., & Gathercole, S. E. (2013). Working memory deficits in children with reading difficulties: Memory span and dual task coordination. Journal of Experimental Child Psychology, 115(1), 188–197.
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
433
434
SUN-A KIM
Wang, Y., & McBride, C. (2017). Beyond copying: A comparison of multicomponent interventions on Chinese early literacy skills. International Journal of Behavioral Development, 41(3), 380–389. Wen, Z. E. (2016). Working memory and second language learning: Towards an integrated approach. Multilingual Matters. Xu, Z., Wang, L. C., Liu, D., Chen, Y., & Tao, L. (2020). The moderation effect of processing efficiency on the relationship between visual working memory and Chinese character recognition. Frontiers in Psychology, 11. (Article 1899). Yang, L. Y., Guo, J. P., Richman, L. C., Schmidt, F. L., Gerken, K. C., & Ding, Y. (2013). Visual skills and Chinese reading acquisition: A meta-analysis of correlation evidence. Educational Psychology Review, 25(1), 115–143. Yang, X., Peng, P., & Meng, X. (2019). How do metalinguistic awareness, working memory, reasoning, and inhibition contribute to Chinese character reading of kindergarten children? Infant and Child Development, 28(3), e2122.
https://doi.org/10.1017/9781108955638.024 Published online by Cambridge University Press
20 The Role of Working Memory in Language Comprehension and Production Evidence from Neuropsychology Rachel Zahn, Autumn Horne, and Randi C. Martin
20.1
Introduction
Successful language comprehension and production require interactions between language processing systems and other cognitive systems, such as working memory (WM). Well-developed theories of language production and comprehension should take into account the role that individual differences in WM play in language processing. In this chapter, we review the evidence for the claim that WM plays a role in language production and comprehension. We focus primarily on evidence from individuals with brain damage, while also reviewing complementary findings from healthy individuals. Our perspective on WM is the domain-specific model that includes WM buffers that are specific to phonological and semantic information and separate from long-term knowledge in these domains (Martin et al., 2021). Thus, the focus is on the separable contributions of phonological and semantic WM buffers to language processes.
20.2
Domain-Specific Model of Working Memory
The domain-specific model of working memory (WM) includes WM buffers that are specific to phonological, semantic, and graphemic information and separate from long-term knowledge in these domains (Martin et al., 2021). Domain-specific WM is distinguished from other influential models of WM including embedded processes models and standard buffer models.
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
436
RACHEL ZAHN ET AL,
Figure 20.1 The domain-specific model of WM (adapted from Martin et al., 1999)
Embedded process models assume that WM is the activated portion of longterm memory (LTM; e.g., Cowan et al., 2021). Information in LTM is kept activated by a domain-general attentional process, with limited capacity. In contrast, domain-specific WM assumes that the information contained in WM buffers is separate from long-term knowledge, although the two are closely linked. Although attention may be directed to the different buffers in domain-specific WM, information in the buffers can persist outside the scope of attention. Domain-specific WM (Martin et al., 1999, 2021) bears some similarities to standard buffer models (e.g., Baddeley et al., 2021), but there are two key distinctions. Standard buffer models typically include only one store and a rehearsal process that support verbal WM. Furthermore, standard buffer models emphasize the role of phonological/articulatory representations in this single verbal WM store. In contrast, the domain-specific WM model includes buffers for semantic, phonological, and graphemic representations. Prior research indicates that the semantic and phonological buffers play different roles in sentence processing in both language production and comprehension, and these findings are the focus of this overview.1 The domain-specific model of verbal WM from Martin and colleagues (Martin et al., 1999, 2021) is depicted in Figure 20.1. On the left side of the model is long-term knowledge of lexical representations. On the right, separate from long-term knowledge, are specialized buffers for maintaining semantic and phonological representations with bidirectional connectivity between lexical representations in LTM and WM. As you are hearing or preparing to produce language, representations in LTM are activated and transferred into the buffer. The content of these representations may be copied into the buffer (Yue et al., 2019) or there may be a placeholder in the buffer that points to the representation in LTM (Norris, 2017). This model of WM is considered domain-specific because the semantic and phonological
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
buffers are specific to processing in their respective domains. Within phonological processing, there is some evidence that a distinction should be made between buffers for input and output phonology, with the input buffer used in speech recognition and the output buffer in language production. The implications of a distinction between these buffers are discussed in the section on WM and language production. Semantic and phonological representations may be held in their respective buffers simultaneously, a feature of WM that is critical for successful language processing. Consider the case of language production. In production, WM is critical for storing verbal representations that are planned in advance of articulation. An intended message is formulated, and corresponding semantic representations are activated. Often, multiple words are planned ahead of articulation, with their semantic representations stored in the semantic buffer. These semantic representations activate corresponding phonological representations which guide articulation. During articulation, the next group of semantic representations can be planned. Thus, both semantic and phonological storage are drawn on simultaneously.
20.2.1 Neuropsychological Evidence for Domain-Specific WM The development of the domain-specific model of WM has been heavily influenced by neuropsychological work on brain damaged individuals with striking deficits in WM capacity (e.g., Martin et al., 1999). Individuals have been reported who show distinct patterns of performance that support their classification as having impairments in either semantic or phonological WM. Although patients with both types of WM deficits may perform below control subjects on tasks directed at assessing both semantic and phonological WM, their deficit is more prominent for one type of WM capacity than the other. The striking double dissociations observed between semantic and phonological WM capacities cannot be accounted for by either embedded processes models or standard buffer models of WM. Embedded processes models claim that WM is the activated portion of LTM. If the focus of attention brings LTM representations into WM, then you might expect the same attentional mechanism would be involved in both phonological and semantic WM. It’s unclear how a single attentional mechanism accounts for dissociations observed between semantic and phonological WM. Traditional buffer models also cannot account for dissociations between semantic and phonological WM because they do not propose any verbal buffer beyond the phonological loop (Baddeley et al., 2021). Patients with a semantic WM deficit have difficulty on tasks that require the maintenance of semantic representations (Martin & Romani, 1994; Martin, 2021). These patients show standard phonological effects, such as word length and phonological similarity effects. However, they fail to show an advantage of words over nonwords, in contrast to control subjects,
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
437
438
RACHEL ZAHN ET AL,
presumably because they do not benefit from the semantic information associated with words to support recall. Patients’ semantic WM capacity is often measured with the category probe task developed by Martin et al. (1994). In the category probe task, patients hear a list of words followed by a probe word and indicate whether the probe word was in the same category as any of the list words. For example, for the list “DESK TREE HAT,” followed by the probe word “CHAIR,” the correct response would be “Yes” because “DESK” is in the same category as “CHAIR.” This task contrasts with the rhyme probe task, also developed by Martin et al. (1994), where patients are asked to indicate if the probe item rhymes with any of the list words. For example, for the list “DRESS CABLE STORE,” and the probe word “FLOOR,” the correct response would be “Yes” because “STORE” rhymes with “FLOOR.” Semantic WM deficit patients perform worse on the category probe task compared to the rhyme probe task. Patients with a phonological WM deficit, on the other hand, have difficulty on tasks that require the storage of phonological representations (Martin, 1987; Martin et al., 1994). They do not show standard phonological effects, such as the word length or phonological similarity effects, at least with visually presented words. Their performance is very impaired on tasks using nonwords and digits, which have little associated semantic information; however, they do better on word than nonword span, where semantic information can be used to aid memory performance. Phonological WM patients typically perform better on memory tasks with visual than auditory presentation. This is in contrast to control subjects and semantic WM patients who typically perform better on memory tasks with auditory presentation, arguably because auditory input provides more direct access to phonological codes. Additionally, phonological WM patients perform better on the category probe task than the rhyme probe task described above, as the rhyme probe task depends primarily on the storage of phonological representations of the list words. For both types of WM deficits, it is important to confirm that the deficits are not the result of more general semantic or phonological processing deficits. Careful testing of speech perception, semantic knowledge, and single word comprehension and production is carried out in studies of patients with semantic and phonological WM deficits (e.g., Martin et al., 1994; Martin & Breedin,1992; Martin & He, 2004). In general, patients with both types of WM deficit perform within the control range on minimal pairs speech discrimination and auditory lexical decision. In the minimal pairs speech discrimination task, participants hear pairs of syllables that differ on a single feature of one phoneme (e.g., ba, pa, which differ in voicing) and must indicate if they are the same or different. In the auditory lexical decision task, participants must indicate if an item is a word or a nonword. Nonwords differed from words by one feature of one phoneme (e.g., pickle, bickle). Patients’ strong performance on both tasks indicates intact speech perception abilities. Additionally, they can perform within
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
the control range for single word comprehension and production tasks, such as picture-word matching and picture naming. Furthermore, Martin and Breedin (1992) demonstrated that even though an individual with a severe phonological WM deficit scored slightly below the control range on speech perception tests, her mild speech perception deficit could not account for her phonological WM deficit. Other patients with comparable perceptual deficits showed standard phonological effects on span and better phonological WM performance, with one scoring above the mean for controls on auditorily presented lists. These findings demonstrate that semantic and phonological WM patients’ deficits cannot be accounted for by general language processing deficits.
20.2.2 The Neural Basis of Phonological and Semantic WM While there is considerable behavioral evidence for domain-specific semantic and phonological WM buffers, the neural instantiations of the buffers are less well documented. Findings from early lesion overlap and fMRI studies suggested that the left inferior parietal region, particularly the left supramarginal gyrus (SMG), supports phonological WM (Paulesu et al., 1993). More recently, work by Yue and colleagues (2019) investigated the neural basis of phonological WM in healthy young adults using univariate and multivariate fMRI. They observed both sustained activation during the maintenance period of a phonological WM task and a memory load effect in the SMG. In addition, using multivariate pattern analysis, they showed that speech could be decoded during the delay period in the SMG and individual differences in such decoding related to behavioral performance on the WM task. Furthermore, Yue and Martin (2021) used representational similarity analysis (RSA) to examine the relationship between observed patterns of neural activity and a theoretical phonological similarity matrix. Yue observed RSA evidence of phonological storage in the SMG. Thus, findings from both univariate and multivariate fMRI approaches were consistent with the SMG being the location of a phonological WM buffer. A lesion overlap study of a large sample of patients by Pisoni et al. (2019) also found that damage to the SMG was related to phonological WM deficits. Beyond the inferior parietal region, frontal regions have been implicated in phonological WM and are thought to support the rehearsal processes required to maintain phonological forms (Paulesu et al., 1993). Furthermore, evidence from direct current stimulation of the SMG and IFG during awake surgery has implicated both regions in phonological maintenance (Papagno et al., 2017). Less work has been done to investigate the neural basis of semantic WM, but the limited fMRI research done with healthy young adults implicates the left middle frontal (MFG) and inferior frontal (IFG) regions (Hamilton et al., 2009; Martin et al., 2003; Shivde & Thompson-Schill, 2004). A recent multivariate lesion-symptom mapping study investigated the neural basis of both semantic and phonological WM in 94 patients at the
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
439
440
RACHEL ZAHN ET AL,
acute stage of left hemisphere stroke (Martin et al., 2021). By testing patients acutely (most within one week of stroke), the neural basis of the WM buffers could be investigated before reorganization of function could occur. Reduced phonological WM was related to SMG damage, consistent with past findings arguing that this region supports phonological storage (Paulesu et al., 1993; Yue et al., 2019). Phonological WM was also related to both subcortical (basal ganglia and thalamus) and cortical (postcentral gyrus and inferior frontal junction) speech-motor processing areas, likely due to their involvement in supporting subvocal rehearsal. Interestingly, reduced phonological WM was also related to damage to somatosensory regions for the lips and larynx as well as primary auditory areas, which may support somatosensory and auditory targets involved in the motor control underlying rehearsal (Hickok & Poeppel, 2007). Semantic WM was related to the IFG, as previously reported, but the region was more posterior than in previous studies (Hamilton et al., 2009; Martin et al., 2003). The posterior region of the IFG has been suggested by some studies to be involved in the selection of semantic representations among competing alternatives – for instance, when producing a verb semantically associated with a noun when there are two or more strongly associated responses (e.g., for door, selecting between open and close; Martin & Chao, 2001; Snyder & Munakata, 2008). Martin et al. (2021) also found that semantic WM was related to the posterior superior temporal sulcus, previously reported to be involved in mapping between phonological and semantic representations (Okada & Hickok, 2006), and the angular gyrus, a region thought to support integration of semantic representations (e.g., adjective-noun combinations such as “plaid jacket”; Bemis & Pylkkanen, 2013).
20.3
Domain-Specific WM and Language Production
WM supports speech production by storing representations that are planned in advance before the onset of articulation. Speech production proceeds across various steps (Bock & Levelt, 1994; Dell, 1986). Speech production begins with an intended message and its corresponding nonlinguistic conceptual representations. The conceptual representations are then encoded as lexical-semantic representations, sometimes called lemmas (Levelt, 1989), and fit within a syntactic structure (Allum & Wheeldon, 2007, 2009). Lexical-semantic representations then activate phonological representations, which are used to guide articulation. The process of speech production is incremental, meaning that part of the utterance is planned before articulation begins, and further planning of later segments can occur during articulation of earlier parts. When planning the utterance at the lexical-semantic and phonological levels, WM is critical because it must be relied on to keep the representations active until articulation takes place. Thus, the further ahead you are planning, the
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
more important WM becomes. How far in advance you are planning – the scope of planning – depends on the stage of the production process. While it is argued that the intended message is planned out in full at a clausal level, planning at the lexical-semantic and phonological levels is agreed to be more restricted (Garrett, 1980).
20.3.1 Scope of Planning Early investigations used speech error analysis to make inferences about the scope of planning at the lexical-semantic and phonological levels (Fromkin, 1971). Whole word exchange errors (e.g., “my chair seems empty without my room”) are assumed to arise from planning at the lexicalsemantic level and are frequently observed across multiple phrases but within the same clause. In contrast, sound exchange errors (e.g., “light knife” for “night life”), which are assumed to occur during planning at the phonological level, most frequently occur between adjacent words within the same phrase. Because of the difference in distance across which the two error types occurred, it was inferred that planning at the lexicalsemantic level had a larger scope than planning at the phonological level. However, these early findings were observational in nature, with attendant methodological limitations, such as perceptual biases (Alderete & Davies, 2019). More recent experimental work with healthy populations provides a more nuanced understanding of the scope of planning at different stages of the production process. In these studies, pictures or video displays are typically used to elicit utterances with specific grammatical constructions (e.g., a video eliciting “the ball moves above the block and the faucet” or a picture eliciting “the turtle was pushed by the frog”). Results of these studies generally indicate that the scope of planning differs for lexical-semantic and phonological planning. Claims regarding the scope of lexical-semantic planning have ranged from the single word (Griffin & Bock, 2000) to the phrase (Allum & Wheeldon, 2007, 2009; Smith & Wheeldon, 1999) to the whole clause (Meyer, 1996). Typically, studies that suggest the single word scope of lexical-semantic planning use an eyetracking paradigm in which subjects’ eye movements are monitored as they describe pictures of simple actions (Griffin & Bock, 2000). However, in these studies, target utterances often had identical structures, potentially resulting in participants adopting a word-by-word planning strategy because the syntactic structure was highly predictable (Martin et al., 2010). Other results have suggested that the scope of lexical-semantic planning is much larger, perhaps as large as the whole clause. Evidence for an extended scope of lexical-semantic planning comes from a pictureword interference study carried out by Meyer (1996). Meyer found that when producing a sentence to describe a picture such as “the arrow is next to the bag,” an auditorily presented distractor-affected speech-onset latencies when semantically related to either the first or second noun,
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
441
442
RACHEL ZAHN ET AL,
whereas a phonologically related distractor word had an effect only when related to the first noun, consistent with a clausal scope at the lexicalsemantic level and a single word scope at the phonological level. However, it is likely that the scope of lexical-semantic planning lies somewhere between the single word and the whole clause. A large body of evidence from Wheeldon and colleagues suggests that the minimal scope of lexical-semantic planning is the phrase, though healthy participants have the option of adopting a larger scope (Smith & Wheeldon, 1999). In the Smith and Wheeldon study, participants produced sentences to describe moving picture displays of three objects. The two target constructions were: (a) simple-complex, in which sentences began with a simple subject noun phrase followed by the verb and then a prepositional phrase that contained a complex noun phrase (e.g., “the tie moved above the candle and the foot”), (b) complex-simple, in which sentences began with a complex subject noun phrase that was followed by the verb and a prepositional phrase that contained a simple noun phrase (e.g., “the tie and the candle moved above the foot”). The sentence conditions were matched for overall length and content words (nouns and verbs). If planning occurs for the initial noun only or the whole sentence, then there should be no difference in the onset latencies for the two conditions. However, Smith and Wheeldon (1999) observed longer onset latencies for sentences beginning with a complex noun phrase, suggesting that participants were planning the initial noun phrase in its entirety before they began articulation. Later work confirmed that this finding could not be explained by visual grouping factors or by differences in retrieval fluency for the first versus the second content word in a phrase (Martin et al., 2010). Follow-up work by Allum and Wheeldon (2009) offered evidence corroborating that the initial phrase is planned prior to utterance onset at the lexical-semantic level, irrespective of whether the initial phrase was a subject, for English, which is a head-initial language, or a prepositional phrase, for Japanese, which is a head-final language. That is, the number of content words in the initial phrase impacted speech latencies for both languages (Allum & Wheeldon, 2009). It was also found that priming subjects with a picture of a noun that was not the initial noun, but in the initial phrase, speeded production in English as well as in Japanese, but did not do so when the noun pictured was not in the initial phrase. In summary, these three studies (Allum & Wheeldon, 2009; Martin et al., 2010; Smith & Wheeldon, 1999) provided strong evidence in English as well as Japanese that the default scope of lexical access is the grammatical phrase and this planning occurs at a lexical-semantic level. This evidence is corroborated by evidence from speakers with brain damage, which will be discussed later in this section. Much of the evidence from healthy speakers suggests a more restricted scope of planning at the phonological level. As mentioned above in Meyer’s (1996) study, phonological facilitation was observed when the auditorily
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
presented word was phonologically related to the first but not the second noun, suggesting that the phonology of only the first noun was selected before speech onset. Some researchers suggest that the phonological scope of planning is very small, perhaps as small as what is termed a phonological word in the linguistics literature, which consists of a content word and the unstressed function words associated with it (e.g., “the book”) (Wheeldon and Lahiri, 2002). Still, some others claim that the phonological scope of planning extends beyond the phonological word (Damian & Dumay, 2007; Schnur et al., 2006). Damian and Dumay (2007) demonstrated that the production of two words was faster when the two words began with the same phoneme (e.g., “blue bells”) than when they did not (e.g., “green bells”), suggesting that phonological planning for the second phonological word in the phrase was initiated before speech onset. However, even findings that the default phonological scope of planning extends across multiple words still generally suggest that this scope is smaller than that of lexical-semantic planning.
20.3.2
Consequences of Phonological and Semantic WM Deficits for Speech Production The differences in scope at the lexical-semantic and phonological levels would lead one to expect that planning at the lexical-semantic level would place greater demands on semantic WM than phonological planning places on phonological WM. Consistent with this logic, contrasting patterns have been observed in language production for speakers who have either semantic or phonological WM deficits. Martin and Freedman (2001) compared the phrase and sentence production abilities of two individuals with impaired semantic WM, and one with impaired phonological WM. The individual with a phonological WM deficit performed within control range, whereas the participants with semantic WM deficits had considerable difficulty in producing phrases that contained either one or two adjectives preceding the noun (e.g., “red book,” “small green leaf”), even though they could produce the nouns and adjectives in isolation. However, these individuals performed much better at producing the same content when it was split across multiple phrases in a sentence (e.g., “the book is red,” “the leaf is small and green”), consistent with a notion of a phrasal scope of planning at the lexical-semantic level. Martin et al. (2004) further investigated the relation between WM and speech production, using moving picture displays comparable to those in Smith and Wheeldon (1999). The participant with a semantic WM deficit had a substantially greater difference in onset latencies when describing pictures that required producing two nouns versus one in the initial phrase than was observed for controls (1027 ms vs. 66 ms). In contrast, the participant with a phonological WM deficit showed a difference between the two conditions that was comparable to controls (58 ms).
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
443
444
RACHEL ZAHN ET AL,
In both Martin and Freedman (2001) and Martin et al. (2004), speakers appear to be planning the lexical-semantic content of the head of the phrase and everything that comes before it (the lexical head principle) before beginning articulation. The phrase contains multiple words, so speakers must simultaneously hold all of the lexical-semantic representations for a phrase in WM while proceeding with the phonological encoding for the phrase. Patients with semantic WM deficits have patterns of speech output that indicate their difficulty holding multiple lexical-semantic representations in WM simultaneously, making it difficult to produce phrases containing many content words. On the other hand, patients with phonological WM deficits did not have the same difficulty with speech production because they were able to plan at a normal scope at the lexical-semantic level. At the phonological level, planning occurs only one or two words ahead, not far enough to place demands on their diminished phonological WM capacity (Wheeldon & Lahiri, 2002). Supporting these claims are findings showing that patients with phonological WM deficits also show fluent narrative speech (Shallice & Butterworth, 1977). In sum, work examining the relation between WM and speech production in participants with chronic WM deficits supports the claim that successful spontaneous speech production draws on semantic WM, but not phonological WM. This claim was further tested by Martin and Schnur (2019) in 37 individuals at the acute stage of left hemisphere stroke (i.e., within one week of stroke onset). A case series approach was used to relate their phonological and semantic WM to their language production abilities. In the case series approach, continuous variation in these WM capacities is measured and related to their performance on outcomes of interest (Schwartz & Dell, 2011). Testing at the acute stage allows for examination of performance before reorganization of function or the development of compensatory strategies. Language production was assessed with a Cinderella storytelling task, scored using Quantitative Production Analysis (QPA) (Saffran et al., 1989). Semantic WM was measured with the category probe task and phonological WM was measured using digit matching span (Martin et al., 1994). In the digit matching span task, participants heard two lists of digits and indicated whether the two lists matched. In non-matching trials the position of two digits had been transposed (e.g., “1752” vs “1572”). In order to control for single word processing abilities, performance was assessed on a single word–single picture matching task (Martin et al., 1999). Based on the previous research supporting a role for semantic WM in planning content words for speech production, it was predicted that impaired semantic WM would be related to reduced sentence elaboration (i.e., fewer content words in each noun and verb phrase) and shorter mean utterance length in the storytelling task, where distinct utterances were defined mainly by pauses and could be sentences, phrases, or single words. It was also predicted that a decreased semantic WM capacity would be
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
related to less fluent speech (measured as words per minute). In contrast phonological WM was not predicted to relate to these narrative measures. Using a multiple regression approach, Martin and Schnur (2019) found that semantic WM, but not phonological WM, had an independent contribution in predicting sentence elaboration and mean utterance length, as predicted. However, unexpectedly, they found that phonological WM had an independent contribution in predicting speech rate in words per minute, while semantic WM did not. The sentence elaboration results provide additional support from a large group of participants that semantic WM is critical to planning content words for language production. The unexpected relationship between phonological WM and speech rate requires further investigation. Two potential explanations seem possible: (1) variation in single word phonological retrieval or (2) the existence of separable input and output phonological buffers. According to the first possibility, both words per minute and phonological WM capacity are related to the efficiency of single word phonological retrieval, and variation in this third factor underlies the correlation between the two. The relationship of the efficiency of phonological retrieval to words per minute is straightforward – faster phonological retrieval at the single word level should contribute to faster production of words in a narrative. The relationship to phonological WM depends on assumption of a role for subvocal rehearsal in the phonological WM task. Although patients did not have to produce the lists in the digit matching task used by Martin and Schnur (2019), the digits were presented at a slow rate of one per second, allowing time for subvocal rehearsal. More efficient phonological retrieval may support rehearsal, as previous work in healthy individuals suggests a relationship between speech rate and the efficiency of subvocal rehearsal (Baddeley et al., 1975). In the case studies discussed earlier, patients with reduced phonological WM capacity were selected to have preserved single word production abilities (Martin et al. 2004, Martin & Freedman, 2001). In the acute study, however, all patients with left hemisphere stroke were included and produced varying proportions of phonological errors on tasks such as picture naming or repetition, implying difficulty with phonological retrieval for some. Thus, variation in phonological retrieval may underlie the unexpected relation of speech rate to phonological WM (Zahn et al., 2019). According to the second possibility, there may be a distinction between input and output phonological WM capacities. Some models consider both input and output phonological buffers to be part of the WM system (e.g., Martin et al.,1999, Romani, 1992; Shallice et al., 2000; Vallar, 2006), whereas others do not (e.g., Burgess & Hitch, 2006; Martin & Saffran, 2002). For those models including both buffers, the input phonological buffer is thought to be involved in the maintenance of input phonological codes derived from perception, while the output phonological buffer is used for the maintenance of output phonological codes that support articulation
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
445
446
RACHEL ZAHN ET AL,
(see Figure 20.1). To account for the unexpected speech rate results from Martin and Schnur (2019), one might argue that the individuals in the prior case studies had specific input phonological WM deficits, accounting for their preserved production, whereas those in the acute sample had a mix of deficits in input and output phonological WM capacities, leading to reductions in speech rate for those with the output buffer deficits. Evidence for the separation of input and output buffers comes largely from a small number of case studies of neuropsychological patients (Allport, 1984; Gvion & Friedmann, 2012b; Martin et al., 1999; Romani, 1992; Shallice et al., 2000). Most evidence involves a contrast in performance on phonological WM tasks that require speech output, such as serial list recall, versus those that do not, such as probe recognition or matching tasks. Patients with phonological output buffer deficits perform well on the tasks involving recognition or matching, but poorly on those requiring speech output. Patients with phonological input buffer deficits perform poorly on both, as the phonological input buffer is needed for recognition tasks, and also for holding perceived information for a transfer between the input and output buffers to support speech output. In language production, the phonological output buffer has been argued to be critical for retaining the phonological representations for multiple words prior to articulation (Romani, 1992; Shallice et al., 2000). For patients with a phonological output buffer deficit, sentence production and conversational speech show phonological errors such as substitutions, transposition, or deletion of phonemes (Gvion & Friedmann, 2012b; Romani, 1992). Martin et al. (1999) provided a different type of evidence for a phonological output buffer from a severely anomic patient. His performance on input phonological WM tasks was at a very high level: (1) 6-item digit matching trials: 95% correct (control mean 93%) (2) rhyme probe span: 6.4 items (control mean 7.08 items; SD 1.44), (3) 6-item nonword recognition probe task: 88% correct (control mean 86%). On output WM tasks, his performance was impaired only for lists of words that he had difficulty retrieving in picture naming or spontaneous speech, though he showed excellent comprehension for these words. Thus, he did not have an output buffer deficit per se, but had difficulty accessing output phonological forms to deposit into this buffer. Future work is needed to investigate whether additional patients show evidence supporting the proposed distinction between input and output buffers and, if so, clarifying the role of these buffers in supporting fluent speech.
20.4
Domain-Specific WM and Language Comprehension
Language comprehension involves maintaining, processing, and integrating linguistic information in an active and continuous manner. It is often assumed that the maintenance of information during comprehension must
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
rely on some WM capacity. Evidence from neuropsychological studies suggests that semantic WM is more critical than phonological WM to the comprehension process.
20.4.1 Phonological WM and Sentence Comprehension In an early study, Martin (1987) examined sentence comprehension abilities in patients with WM deficits, attributed to either phonological storage or rehearsal deficits. (As a semantic WM component had not yet been postulated, semantic WM was not assessed.) Patients were classified into three groups: (1) nonfluent agrammatic speakers, (2) nonfluent nonagrammatic speakers, and (3) fluent speakers. In addition, a fluent patient EA, previously argued to have a phonological storage deficit was included. Agrammatism in speech was assessed because agrammatism had been linked to a syntactic deficit, potentially compromising sentence comprehension (Berndt & Caramazza, 1980). WM was tested using word and letter list recall tasks in which they responded by pointing to corresponding pictures or letters in the order they were presented. Sentence comprehension was assessed using an active-passive (e.g., “The girl pushed the boy,” “The girl was pushed by the boy”) and relative clause (see Table 20.1) comprehension tasks. In both, participants heard a sentence and chose which of two pictures matched the sentence. Both groups of nonfluent patients showed similarly impaired patterns of WM performance, suggesting a rehearsal deficit, whereas the fluent patients performed substantially better. On sentence comprehension, however, nonfluent agrammatic speakers, with poor WM performance, showed impaired performance while both those classified as nonfluent nonagrammatic, with similarly impaired WM, and those classified as fluent, with higher WM performance, had intact comprehension. Thus, the results argued against a role for WM, at least with respect to rehearsal, in understanding complex sentences. However, the fluent patient EA was impaired at comprehension of sentences with embedded clauses and noncanonical word order in which the head noun played the role of the object rather than the agent (sentence types IV and V in Table 20.1). One might argue that these nonstandard embedded clauses slow down processing, leading Table 20.1 Examples of relative clause sentence types Sentence type
Example
I II III IV V
The The The The The
boy boy boy boy boy
that that that that that
had red hair carried the girl. carried the girl had red hair. had red hair was carried by the girl. was carried by the girl had red hair. the girl carried had red hair.
Source: Adapted from Martin, 1987
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
447
448
RACHEL ZAHN ET AL,
healthy subjects to rely on phonological storage of the sentence until semantic and syntactic processing can catch up. EA, unable to use the phonological store, would either have to stop processing the embedded clause to be able to attend to the following input or continue processing the embedded clause but miss the rest of the sentence. Other studies, however, have reported neuropsychological patients with phonological WM deficits who show comprehension within normal limits. Patient RE had a developmental disorder affecting storage of phonological information but showed good comprehension on all sentence types tested (Butterworth et al., 1986). Both RE and EA had a fluent speech rate and problems with verbatim sentence repetition. However, RE did not show the impaired comprehension of sentences with nonstandard embedded clauses that EA did. It may be the case that RE’s developmental deficit led to adaptive strategies for comprehension that differ from typical comprehension processes, though no evidence is available to support that claim. Gvion and Friedmann (2012a) reported two patients argued to have output phonological buffer deficits who had preserved comprehension even for complex sentences such as garden path sentences. Garden path sentences are sentences in which the comprehender is led to interpret the beginning of the sentence in the most likely fashion, which later proves to be incorrect. For example, in the sentence, “The complex houses married and single soldiers and their families,” the most likely initial interpretation is that “complex” is an adjective, not a noun and “houses” is a noun, not a verb. This turns out to be incorrect, requiring the reader to reevaluate the input to arrive at the correct interpretation. Because output buffer deficit patients are able to correctly comprehend garden path sentences, it is unlikely that output phonological code is necessary to comprehend even these complex sentences. Additionally, Friedmann and Gvion (2007) reported that patients with input phonological buffer deficits showed intact comprehension of complex sentences such as relative clause sentences. It has been proposed that the reactivation required for comprehension of relative clause sentences is semantic, not phonological reactivation (Gvion & Friedmann, 2012a). In contrast, a follow up study by Gvion and Friedmann (2012a) found that in garden path sentences with ambiguous words that have two potential meanings (such as “The toast that the elderly couple had every breakfast was always for happy life and for love”) patients with input buffer deficits performed worse on paraphrasing and judging the plausibility of the sentences as the length between the ambiguous word and the clarifying phrase increased. These findings suggest that input phonological WM is only critical for comprehension when phonological reactivation is needed to support the reinterpretation of an ambiguous word. Another patient, BO, was reported to have poor phonological WM, good comprehension, and slow, effortful speech (Waters et al., 1991). With auditory presentation, BO performed well on the embedded relative clause sentences like those that were difficult for EA (e.g., scoring 92% correct for
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
sentences similar to Type V in Table 20.1). With visual presentation of the sentences, particularly with limited viewing time, she struggled with object relative structures with three nouns (e.g., “The bear that the donkey kissed patted the goat”) (75% for 15 s viewing time vs. 20% for 5 s). It is difficult to attribute these difficulties, however, to her phonological WM deficit, given her much better performance with auditory presentation, which one might have thought would have put the greatest demand on phonological WM. It seems likely that BO’s problems with written presentation were due to reading difficulties per se, as she was less accurate and much slower than controls on single written word processing. These reading difficulties could have interacted with the inherent difficulty in processing more complex structures, as healthy individuals show greater effects of syntactic complexity when sentences are presented more rapidly (Miyake et al., 1994). BO also had difficulties when it was necessary to maintain multiple proper nouns or prepositions. From this pattern of performance, the claim was made that phonological WM was not necessary for online comprehension, but it did play a role in “postinterpretive” processes such as operating on propositional meanings or searching through noun phrases with multiple proper nouns. BO’s difficulty with proper nouns may have arisen because proper nouns have little meaning and therefore rely on phonological WM for storage and reactivation. Varkanitsa and Caplan (2018) conducted a meta-analysis of 26 studies relating verbal WM capacity to comprehension in people with brain damage. They included studies that measured verbal WM with a variety of recall span tasks for letters, digits, words and nonwords, complex span tasks such as reading and listening span (Daneman & Carpenter, 1980), and n-back tasks. When the studies that reported correlational data between WM performance and sentence comprehension abilities were pooled in a group-level meta-analysis a positive association between WM capacity and sentence comprehension was found. However, there are two limitations to this finding. The first is that it is difficult to determine from this association whether WM plays a role in online sentence comprehension or if it supports postinterpretive processes as suggested by Waters et al. (1991). The second limitation regards the types of verbal WM tasks used in the 26 studies included in the meta-analysis. The wide variety of tasks included do not differentiate between semantic and phonological WM, which may play distinct roles in language comprehension.
20.4.2 Role of Semantic WM in Sentence Comprehension While the neuropsychological evidence does not suggest a major role for phonological WM in language comprehension,2 it does seem that semantic WM is critical for maintaining semantic representations while integrating the meanings of words across a sentence. Work using rapid visual presentation of words suggests that semantic representations are activated
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
449
450
RACHEL ZAHN ET AL,
extremely rapidly, with Potter et al. (1986) demonstrating that readers can comprehend sentences that are presented word-by-word as quickly as 10 words per second. When synonyms for words in the sentence were presented either before or after the sentence, the synonym was frequently inserted into the recall of the sentence, suggesting that the sentence is reconstructed from the activated meaning representations rather than from the corresponding phonological representations (Potter & Lombardi, 1998). The activation and maintenance of meaning representations seems critical to language comprehension, and thus patients with semantic WM deficits should be impaired at sentence comprehension. Evidence indicates that semantic WM deficit patients have difficulty with comprehension as the number of word meanings to be maintained increases (Martin & Romani, 1994). For example, these patients performed poorly when answering attribute questions containing one adjective and two nouns (“Which is loud, a concert or a library?”) but had excellent performance on questions with only one adjective and one noun (“Is a library loud?”) (Martin et al., 1994). Subsequent research demonstrated that comprehension difficulties for those with semantic WM deficits depended on whether word meanings could be integrated immediately or if they must be maintained in an unintegrated state (Martin & He, 2004; Martin & Romani, 1994). These patients scored near chance at detecting the anomaly in sentences where they had to hold on to several semantic representations before integration of word meanings could be carried out (e.g., holding on to all of the nouns in the initial phrase before integration with the verb for a sentence such as “Rugs, vases, and mirrors cracked during the move.”) However, they performed well on sentences where the nouns directly followed the verb and the relation of each noun to the verb could be processed immediately as the noun was heard (e.g., “The movers cracked the mirrors, vases, and rugs.”). In contrast, a patient with a phonological WM deficit showed a pattern of effects in the delayed versus immediate integration sentences that was similar to controls but had very low accuracy on verbatim sentence repetition, often paraphrasing or providing a semantic approximation for complex sentences (Martin et al., 1994; Martin & He, 2004). Interestingly, neither semantic WM nor phonological WM deficit patients were affected by the distance between parts of a sentence that signaled a grammatical error such as a number mismatch between a determiner and noun (e.g., “these. . .girl”). This suggests that the retention of syntactic information does not rely on either semantic or phonological WM capacities. It may be that syntactic information is held in its own separate syntactic store. This possibility is difficult to explore because it is unclear how capacity for syntactic information would be tested that would be separate from an assessment of sentence processing. More recent research has focused on the role of phonological and semantic WM in resolving interference in sentence processing (Tan et al., 2017; Tan & Martin, 2018). This work has been done from the perspective of cue-based
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
parsing, in which sentence comprehension is thought to involve a series of cue-based memory retrievals. As words are encountered, associated syntactic and semantic features are encoded. When two words must be integrated, a retrieval cue is generated. Words are integrated based on matches between the retrieval cue and the features associated with earlier items. For example, in the sentence “Flowers from Maria arrived today,” “flowers” would be encoded with the grammatical features “plural,” “noun,” and “subject.” It would also be encoded with the semantic features “definite,” “inanimate,” and “used for decoration.” Later, when the verb “arrived” is encountered, retrieval cues are generated to locate the correct subject noun to integrate with the verb. Retrieval cues might include grammatical features like “noun” and “subject” as well as semantic features compatible with arriving. In this example, “flowers” would be the best candidate for integration with the verb “arrived,” but “Maria” would be a partial match, creating interference that makes sentence comprehension more difficult. In the cue-based parsing literature, it is often observed that comprehension relates more to the degree of interference than the distance between the items that need to be integrated. Many studies have shown longer sentence reading times and greater error rates on comprehension questions for sentences with semantic and syntactic interference (e.g., Van Dyke & Lewis, 2003; Van Dyke & McElree, 2011). Furthermore, cue-based parsing studies have demonstrated that the richness of semantic and syntactic knowledge is more predictive of the ability to resolve interference than individual differences in WM (Van Dyke et al., 2014). Thus, proponents of cue-based parsing often espouse embedded processes models of WM, which assume a limited scope of attention, making a relation between WM capacity and language processing unlikely. For this reason, few studies have examined the role that individual differences in WM play in resolving interference during sentence processing. To address this issue, Tan and Martin (2018) tested semantic WM, phonological WM, executive functioning, and sentence comprehension in nine individuals with left hemisphere stroke. Sentence comprehension was assessed by having patients answer comprehension questions about sentences that varied in their levels of semantic and syntactic interference. As shown in Table 20.2, in the high semantic interference condition, the noun in the embedded clause is semantically plausible as a subject of the main verb. This partial match between the retrieval cue for the main verb and the semantic features of the embedded noun creates a high level of semantic interference when attempting to integrate the main verb with its subject. In the low semantic interference condition, the noun in the embedded clause is semantically implausible as the subject of the main verb, creating relatively less semantic interference when integrating the main verb with its subject. In the high syntactic interference condition, the noun in the embedded clause was a grammatical subject. The partial match between the main verb’s retrieval cue and the grammatical features of the embedded noun creates a high level of syntactic interference. In the low syntactic interference condition, the
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
451
452
RACHEL ZAHN ET AL,
Table 20.2 Examples of stimuli with syntactic and semantic interference Sentence type
Example
LoSyn/LoSem
The jockey who had challenged the unbeatable record yesterday will win. The jockey who had challenged the unbeatable champion yesterday will win. The jockey who claimed that the record was unbeatable yesterday will win. The jockey who claimed that the champion was unbeatable yesterday will win. Will the jockey win?
LoSyn/HiSem HiSyn/LoSem HiSyn/HiSem Question
Source: Adapted from Tan and Martin (2018). Note. “Lo-” and “Hi-” refer to low and high interference conditions, and “-Syn” and “-Sem” refer to syntactic interference and semantic interference conditions.
embedded noun was a direct object. In this condition, there are fewer grammatical features shared by the main verb’s retrieval cue and the embedded noun, resulting in less syntactic interference. Participants’ semantic WM capacity was tested with category probe and synonymy judgment task. Phonological WM was tested with rhyme probe, digit matching, and digit span tasks. These tasks were combined to give composite scores for each WM capacity providing a more reliable assessment of WM abilities. This study also examined the relationship between executive functioning and resolving interference. Executive functioning was tested using the Stroop task and a picture-word interference task. Both tasks involve the inhibition of an incorrect response. When semantic knowledge was controlled for, semantic WM, and not phonological WM, predicted semantic interference in both sentence reading reaction times and accuracy on comprehension questions. Neither semantic WM nor phonological WM were related to effects of syntactic interference. However, syntactic interference effects in comprehension question accuracies were significantly predicted by executive function abilities. Tan et al. (2017) tested 120 healthy subjects on a similar set of tasks. When all measures were included in a mixed-effects model, semantic WM significantly predicted semantic, but not syntactic, interference effects in both sentence reading times and comprehension question accuracy. The phonological WM task did not predict either type of interference effect. Contrasting the patient results, the executive function measure did not significantly predict syntactic interference effects. However, a composite measure made up of two complex span tasks (reading span and operation span) did relate to syntactic interference. It has been suggested that executive function abilities play a role in performance on such complex span tasks (Engle, 2010). Both the Tan and Martin (2018) and Tan et al. (2017) studies support a role for semantic WM, but not phonological WM, in resolving semantic interference in sentence comprehension.
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
20.5
Conclusion
This chapter has summarized the domain-specific WM model and its relationship to language production and comprehension. We have focused on neuropsychological evidence that suggests that there are separable semantic and phonological WM buffers which make independent contributions to language processing (Martin & He, 2004; Tan & Martin, 2018). The semantic buffer has been implicated in processes requiring the short-term maintenance of meaning information (Martin et al., 1994; Martin & He, 2004). Neuroanatomical and neuroimaging evidence suggests that semantic WM may be localized in the left IFG, left angular gyrus, and left posterior superior temporal sulcus (Martin et al., 2021). Conversely, the phonological buffer has been implicated in processes requiring the short-term maintenance of speech sound information, and neuroanatomical and neuroimaging evidence suggests that it is located in the left SMG (Martin et al., 2021; Yue & Martin, 2021). The evidence for a semantic buffer goes against the assumptions of standard buffer models that assume only a phonological buffer in the verbal domain (e.g., Baddeley et al., 2021), and the evidence that both semantic and phonological buffers are distinct from LTM in these domains goes against the assumptions of embedded processes approaches (Cowan et al., 2021). In general, semantic WM plays a more extensive role in language processing. In production, semantic WM supports speech planning when multiple lexical-semantic representations must be maintained in advance of articulation (Martin & Freedman, 2001; Martin & Schnur, 2019). In comprehension, semantic WM supports the integration of conceptual representations when there are long-distance dependencies or conditions creating semantic interference (Martin & Romani, 1994; Tan & Martin, 2018). The role for phonological WM is more limited, but it may be important for supporting fluent speech rate in production (Martin & Schnur, 2019) and postinterpretive processes when comprehension is slow or effortful (Waters et al., 1991). However, the picture regarding phonological WM is complicated by findings suggesting a distinction between input and output phonological buffers (Martin et al., 1999). It may be that, while the input phonological buffer does not play a large role in speech production, the output phonological buffer is critical (Romani, 1992).
References Alderete, J., & Davies, M. (2019). Investigating perceptual biases, data reliability, and data discovery in a methodology for collecting speech errors from audio recordings. Language and Speech, 62, 281–317. Allport, D. A. (1984). Speech production and comprehension: One lexicon or two? In Cognition and motor processes (pp. 209–228). Springer.
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
453
454
RACHEL ZAHN ET AL,
Allum, P. H., & Wheeldon, L. R. (2007). Planning scope in spoken sentence production: The role of grammatical units. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33, 791. Allum, P. H., & Wheeldon, L. (2009). Scope of lexical access in spoken sentence production: Implications for the conceptual–syntactic interface. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35, 1240. Baddeley, A. D., Hitch, G. J., & Allen, R. J. (2021). A multicomponent model of working memory. In Working memory: The state of the science. Oxford University Press. Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of short-term memory. Journal of Verbal Learning and Verbal Behavior, 14, 575–589. Bemis, D. K., & Pylkkänen, L. (2013). Basic linguistic composition recruits the left anterior temporal lobe and left angular gyrus during both listening and reading. Cerebral Cortex, 23, 1859–1873. Berndt, R. S., & Caramazza, A. (1980). A redefinition of the syndrome of Broca’s aphasia: Implications for a neuropsychological model of language. Applied Psycholinguistics, 1, 225–278. Bock, K., & Levelt, W. (1994). Grammatical encoding. In M. A. Gernsbacher (Ed.), Handbook of psycholinguistics (pp. 945–984). Academic Press. Burgess, N., & Hitch, G. J. (2006). A revised model of short-term memory and long-term learning of verbal sequences. Journal of Memory and Language, 55, 627–652. Butterworth, B., Campbell, R., & Howard, D. (1986). The uses of short-term memory: A case study. The Quarterly Journal of Experimental Psychology, 38, 705–737. Cecchetto, C., & Papagno, C. (2011). Bridging the gap between brain and syntax: A case for a role of the phonological loop. In A. M. Di Sciullo & C. Boeckx (Eds.), Biolinguistic approaches to language evolution and variation. Oxford University Press. Cowan, N., Morey, C. C., & Naveh-Benjamin, M. (2021). An embeddedprocesses approach to working memory. In Working memory: The state of the science (p. 44). Oxford University Press. Damian, M. F., & Dumay, N. (2007). Time pressure and phonological advance planning in spoken production. Journal of Memory and Language, 57, 195–209. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Memory and Language, 19, 450. Dell, G. S. (1986). A spreading-activation theory of retrieval in sentence production. Psychological Review, 93, 283. Engle, R. W. (2010). Role of working-memory capacity in cognitive control. Current Anthropology, 51, S17–S26. Friedmann, N., & Gvion, A. (2007). As far as individuals with conduction aphasia understood these sentences were ungrammatical: Garden path in conduction aphasia. Aphasiology, 21, 570–586.
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
Fromkin, V. A. (1971). The non-anomalous nature of anomalous utterances. Language, 27–52. Garrett, M. (1980). Levels of processing in sentence production. In Language production, Vol. 1: Speech and talk (pp. 177–220). Academic Press. Griffin, Z. M., & Bock, K. (2000). What the eyes say about speaking. Psychological Science, 11, 274–279. Gvion, A., & Friedmann, N. (2012a). Does phonological working memory impairment affect sentence comprehension? A study of conduction aphasia. Aphasiology, 26, 494–535. Gvion, A., & Friedmann, N. (2012b). Phonological short-term memory in conduction aphasia. Aphasiology, 26, 579–614. Hamilton, A. C., Martin, R. C., & Burton, P. C. (2009). Converging functional magnetic resonance imaging evidence for a role of the left inferior frontal lobe in semantic retention during language comprehension. Cognitive Neuropsychology, 26, 685–704. Hickok, G., & Poeppel, D. (2007). The cortical organization of speech processing. Nature Reviews Neuroscience, 8, 393–402. Levelt, W. J. (1989). Speaking: From intention to articulation. MIT Press Series in Natural-Language Processing. MIT Press. Martin, A., & Chao, L. L. (2001). Semantic memory and the brain: Structure and processes. Current Opinion in Neurobiology, 11, 194–201. Martin, N., & Saffran, E. M. (2002). The relationship of input and output phonological processing: An evaluation of models and evidence to support them. Aphasiology, 16(1–2), 107–150. https://doi.org/10.1080/ 02687040143000447 Martin, R. C. (2021). The critical role of semantic working memory in language processing. Current Directions in Psychological Science. Martin, R. C. (1987). Articulatory and phonological deficits in short-term memory and their relation to syntactic processing. Brain and Language, 32, 159–192. Martin, R. C., & Breedin, S. D. (1992). Dissociations between speech perception and phonological short-term memory deficits. Cognitive Neuropsychology, 9, 509–534. Martin, R. C., Crowther, J. E., Knight, M., Tamborello II, F. P., & Yang, C.-L. (2010). Planning in sentence production: Evidence for the phrase as a default planning scope. Cognition, 116, 177–192. Martin, R. C., Ding, J., Hamilton, A. C., & Schnur, T. T. (2021). Working memory capacities neurally dissociate. Cerebral Cortex Communications. Martin, R. C., & Freedman, M. L. (2001). Short-term retention of lexicalsemantic representations: Implications for speech production. Memory, 9, 261–280. Martin, R. C., & He, T. (2004). Semantic short-term memory and its role in sentence processing: A replication. Brain and Language, 89, 76–82.
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
455
456
RACHEL ZAHN ET AL,
Martin, R. C., Lesch, M. F., & Bartha, M. C. (1999). Independence of input and output phonology in word processing and short-term memory. Journal of Memory and Language, 41, 3–29. Martin, R. C., Miller, M., & Vu, H. (2004). Lexical-semantic retention and speech production: Further evidence from normal and brain-damaged participants for a phrasal scope of planning. Cognitive Neuropsychology, 21, 625–644. Martin, R. C., Rapp, B., & Purcell, J. (2021). Domain-specific working memory. In Working memory: The state of the science. Oxford University Press. Martin, R. C., & Romani, C. (1994). Verbal working memory and sentence comprehension: A multiple-components view. Neuropsychology, 8, 506–523. Martin, R. C., & Schnur, T. T. (2019). Independent contributions of semantic and phonological working memory to spontaneous speech in acute stroke. Cortex, 112, 58–68. Martin, R. C., Shelton, J. R., & Yaffee, L. S. (1994). Language processing and working memory: Neuropsychological evidence for separate phonological and semantic capacities. Journal of Memory and Language, 33, 83–111. Martin, R. C., Wu, D., Freedman, M., Jackson, E. F., & Lesch, M. (2003). An event-related fMRI investigation of phonological versus semantic shortterm memory. Journal of Neurolinguistics, 16, 341–360. Meyer, A. S. (1996). Lexical access in phrase and sentence production: Results from picture–word interference experiments. Journal of Memory and Language, 35, 477–496. Miyake, A., Just, M., & Carpenter, P. (1994). A capacity approach to syntactic comprehension disorders: making normal adults perform like aphasic patients. Cognitive Neuropsychology, 11, 671–717. Norris, D. (2017). Short-term memory and long-term memory are still different. Psychological Bulletin, 143, 992. Okada, K., & Hickok, G. (2006). Identification of lexical-phonological networks in the superior temporal sulcus using functional magnetic resonance imaging. Neuroreport, 17, 1293–1296. Papagno, C., Comi, A., Riva, M., Bizzi, A., Vernice, M., Casarotti, A., Fava, E., & Bello, L. (2017). Mapping the brain network of the phonological loop. Human Brain Mapping, 38, 3011–3024. Paulesu, E., Frith, C. D., & Frackowiak, R. S. (1993). The neural correlates of the verbal component of working memory. Nature, 362, 342–345. Pisoni, A., Mattavelli, G., Casarotti, A., Comi, A., Riva, M., Bello, L., & Papagno, C. (2019). The neural correlates of auditory-verbal short-term memory: A voxel-based lesion-symptom mapping study on 103 patients after glioma removal. Brain Structure and Function, 224, 2199–2211. Potter, M. C., Kroll, J. F., Yachzel, B., Carpenter, E., & Sherman, J. (1986). Pictures in sentences: Understanding without words. Journal of Experimental Psychology: General, 115, 281.
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
Language Comprehension and Production
Potter, M. C., & Lombardi, L. (1998). Syntactic priming in immediate recall of sentences. Journal of Memory and Language, 38, 265–282. Rapp, B., & Fischer-Baum, S. (2014). Representation of orthographic knowledge. In M. Goldrick, V. S. Ferreira, & M. Miozzo (Eds.), The Oxford handbook of language production. Oxford University Press. Romani, C. (1992). Are there distinct input and output buffers? Evidence from an aphasic patient with an impaired output buffer. Language and Cognitive Processes, 7, 131–162. Saffran, E. M., Berndt, R. S., & Schwartz, M. F. (1989). The quantitative analysis of agrammatic production: Procedure and data. Brain and Language, 37, 440–479. Schnur, T. T., Costa, A., & Caramazza, A. (2006). Planning at the phonological level during sentence production. Journal of Psycholinguistic Research, 35, 189–213. Schwartz, M. F., & Dell, G. S. (2010). Case series investigations in cognitive neuropsychology. Cognitive Neuropsychology, 27, 477–494. Shallice, T., & Butterworth, B. (1977). Short-term memory impairment and spontaneous speech. Neuropsychologia, 15, 729–735. Shallice, T., Rumiati, R. I., & Zadini, A. (2000). The selective impairment of the phonological output buffer. Cognitive Neuropsychology, 17, 517–546. Shivde, G., & Thompson-Schill, S. L. (2004). Dissociating semantic and phonological maintenance using fMRI. Cognitive, Affective, & Behavioral Neuroscience, 4, 10–19. Smith, M., & Wheeldon, L. (1999). High level processing scope in spoken sentence production. Cognition, 73, 205–246. Snyder, H. R., & Munakata, Y. (2008). So many options, so little time. Psychonomic Bulletin & Review, 15, 1083–1088. Tan, Y., & Martin, R. C. (2018). Verbal short-term memory capacities and executive function in semantic and syntactic interference resolution during sentence comprehension: Evidence from aphasia. Neuropsychologia, 113, 111–125. Tan, Y., Martin, R. C., & Van Dyke, J. A. (2017). Semantic and syntactic interference in sentence comprehension: A comparison of working memory models. Frontiers in Psychology, 8, 198. Vallar, G. (2006). Memory systems: The case of the phonological store. A Festschrift for Cognitive Neuropsychology. Cognitive Neuropsychology, 23, 135-155. Van Dyke, J. A., Johns, C. L., & Kukona, A. (2014). Low working memory capacity is only spuriously related to poor reading comprehension. Cognition, 131, 373–403. Van Dyke, J. A., & Lewis, R. L. (2003). Distinguishing effects of structure and decay on attachment and repair: A cue-based parsing account of recovery from misanalyzed ambiguities. Journal of Memory and Language, 49, 285–316. Van Dyke, J. A., & McElree, B. (2011). Cue-dependent interference in comprehension. Journal of Memory and Language, 65, 247–263.
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
457
458
RACHEL ZAHN ET AL,
Varkanitsa, M., & Caplan, D. (2018). On the association between memory capacity and sentence comprehension: Insights from a systematic review and meta-analysis of the aphasia literature. Journal of Neurolinguistics, 48, 4–25. Waters, G., Caplan, D., & Hildebrandt, N. (1991). On the structure of verbal short-term memory and its functional role in sentence comprehension: Evidence from neuropsychology. Cognitive Neuropsychology, 8, 81–126. Wheeldon, L. R., & Lahiri, A. (2002). The minimal unit of phonological encoding: Prosodic or lexical word. Cognition, 85, B31–B41. Yue, Q., & Martin, R. C. (2021). Maintaining verbal short-term memory representations in non-perceptual parietal regions. Cortex, 138, 72–89. Yue, Q., Martin, R. C., Hamilton, A. C., & Rose, N. S. (2019). Non-perceptual regions in the left inferior parietal lobe support phonological short-term memory: Evidence for a buffer account? Cerebral Cortex, 29, 1398–1413. Zahn, R., Schnur, T. T., & Martin, R. C. (2019, October). Phonological retrieval mediates the relation of phonological short-term memory and narrative production. Academy of Aphasia Annual Meeting.
Notes 1 The research on the graphemic buffer has focused on single word processing (Rapp & Fischer-Baum, 2014). 2 For an argument for a larger role of phonological WM in comprehension see Cecchetto & Papagno (2011). We would note, however, that some of the evidence they cite is drawn from BO (Waters et al., 1991), which, as argued earlier is difficult to interpret as supporting a role for phonological WM in complex sentence comprehension.
https://doi.org/10.1017/9781108955638.025 Published online by Cambridge University Press
21 Working Memory and HighLevel Text Comprehension Processes Ana I. Pérez Muñoz and M. Teresa Bajo 21.1
Introduction
Imagine that you have decided to spend a Sunday afternoon reading for pleasure. From the bookshelf at home you take the book of poems Revolting Rhymes by Roald Dahl. On opening the cover, the story of Little Red Riding Hood and the Wolf draws your attention and you begin to read: As soon as Wolf began to feel that he would like a decent meal, he went and knocked on Grandma’s door.
At this point, you would probably expect that the wolf enters the grandmother’s house and devours her before putting on her clothes to fool Little Red Riding Hood and eat her next. This is a good prediction, and it is indeed what it happens. However, if you also predict that Little Red Riding Hood is going to be disturbed by the wolf (as we generally know from the classic story), you will be surprised by reading the following stanza: The small girl smiles. One eyelid flickers. She whips a pistol from her knickers. She aims it at the creature’s head and bang, bang, bang, she shoots him dead.
This example illustrates that when we are trying to understand linguistic information, a constant process of hypothesis generation based on our prior knowledge on the topic is carried out to predict what is going to happen in that specific context. In turn, when new information contradicts any of these hypotheses, we must be flexible to accommodate the new ideas and reject no longer valid expectations. These abilities require the involvement of working memory, as it is the capacity to actively represent, maintain, and manipulate information in mind, as well as to retrieve knowledge from long-term memory. In fact, most influential working memory models assign it functions of storage/activation and executive control
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
460
ANA I. PÉREZ MUÑOZ AND M. TERESA BAJO
(Baddeley, 2000; Cowan, 1999). Accordingly, working memory is essential for comprehension in general (i.e., any situation requiring understanding information) and text comprehension in particular (i.e., understanding information from a passage, compared to lexical or sentence comprehension). The present chapter will deal with the relationship between these complex cognitive processes underlining text comprehension and individual differences in working memory.
21.2
Working Memory and Reading Comprehension
By the 1980s, several studies demonstrated that working memory was at the base of individual differences in reading performance (e.g., Kintsch & van Dijk, 1978; Perfetti & Goldman, 1976). For example, Daneman and Carpenter (1980) developed the reading span task to assess both storage and information processing, and they found a positive correlation between this task and measures of fact retrieval and pronominal reference, suggesting that working memory underlies reading comprehension. Since then, extensive research has corroborated the influence of working memory in distinct comprehension abilities such as word-problem solving (e.g., Fuchs et al., 2015); the resolution of apparent inconsistencies in garden-path sentences (e.g., Farmer et al., 2017); as well as general achievement in reading comprehension (e.g., Schroeder, 2014). Generally speaking, these results demonstrate that low working memory readers (assessed by distinct span measures) show difficulties when comprehending a text, whereas high working memory readers manifest better general reading performance. Furthermore, some meta-analyses have also established the importance of working memory in second language (L2) reading comprehension. For instance, Linck et al. (2014) found a positive correlation between several working memory span tasks and general L2 processing, with larger effect sizes for tasks requiring executive function (vs. storage) and the verbal (vs. visuospatial) domain of working memory. Likewise, Shin (2020) analyzed a total of 25 bilingual studies using the reading span task and found a similar positive relationship between working memory and L2 reading. Thus, the recruitment of working memory during L2 comprehension seems to be as important (if not more) as the one carried out during native (L1) reading. Prior findings can be easily explained by conceiving working memory as a limited resource capacity. In 1992, Just and Carpenter already theorized that language comprehension is constrained by working memory capacity, understanding capacity as the “maximum amount of activation available” in the system to perform both storage and information processing at any linguistic level (e.g., semantic and/or pragmatic). In this way, if the total amount of activation is lower than the one necessary to carry out a specific comprehension process, then the activation supporting a previous task demanding maintenance and/or computation will be reduced, causing the
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
High-Level Text Comprehension Processes
forgetting of prior information and/or a problem processing new information. For instance, if inconsistent information is presented during the comprehension of a text, readers with less activation available might experience difficulties in detecting that inconsistency due to difficulties maintaining previous parts of the text in working memory. Because reading comprehension demands to be constantly activating, maintaining, processing, and deactivating information (sometimes in a simultaneous fashion), the cognitive resources available in working memory during online reading strongly predict comprehension’s ability. Therefore, individual differences in working memory capacity give rise to individual differences in comprehension. An influential approach in the field of individual differences and reading comprehension has been the Structure-Building model (Gernsbacher, 1990, 1997). This model proposes that when readers find information that is associated with their current representation, they enhance its activation and include this information in their situation model. On the contrary, when readers detect information that is incompatible with their memory representation, they attempt to suppress the no longer relevant information. Nevertheless, if readers have not sufficient memory (i.e., working memory) capacity to suppress the irrelevant information, they may form new substructures constructed out of the main mental representation, which reduces the accessibility of information in memory (e.g., Gernsbacher & Faust, 1991). Accordingly, high-capacity working memory readers are more effective not only activating, storing, and processing information, but also inhibiting larger or more complex pieces of information than low-capacity working memory readers who have reduced capacity to retain and process extra demands of information, and to achieve semantic integration of complex information. Moreover, prior knowledge may also modulate this relationship by allowing readers to retrieve appropriate information from long-term memory that it is subsequently integrated with current information, improving language comprehension (Was & Woltz, 2007). Taking into account the extensive literature demonstrating a relationship between working memory and reading comprehension, the aim of the present chapter is to understand whether the same is true for highly complex processes underlying text comprehension. To unravel this matter, we first explain what we refer by high-level text comprehension processes (section 21.3), then discuss the current empirical evidence supporting their connection to working memory in the native language (section 21.4) and the second language (section 21.5), and finish the chapter with some general conclusions (section 21.6).
21.3
High-Level Text Comprehension Processes
Different from lexical or sentence processing, successful text comprehension entails the construction of a coherent, integrated, and accurate mental
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
461
462
ANA I. PÉREZ MUÑOZ AND M. TERESA BAJO
representation of the state of affairs described by the text, which has been commonly called the situation model (Kintsch & van Dijk, 1978). That is, readers need to go beyond the surface characteristics of a text to generate inferences by bringing background knowledge from long-term memory. Moreover, the situation model must be constantly updated as the text unfolds, as each new word, sentence, or paragraph forces the modification of the current model to allow an integrated and coherent mental representation (e.g., McNamara & Magliano, 2009). Therefore, text comprehension demands high-level cognitive processes that are relevant in the construction of a situation model. A crucial process for language comprehension is inference making, which is the ability to generate information that has not been explicitly mentioned in the text (Cain & Oakhill, 1999). Inferences have been classified in many different ways such as online versus offline inferences, or local versus global (see e.g., Pérez, et al., 2014). An important distinction has been made between text-based inferences, which require the integration of different pieces of information from the text, and knowledge-based inferences, which entail the combination of text information and the reader’s prior knowledge (Kintsch, 1998). In addition, as was illustrated in the example of Little Red Riding Hood, an important type of knowledge-based inference is prediction. Predictive inferences help to anticipate upcoming events or future outcomes in a story, and tend to be automatically generated when (a) text information is highly constrained, making the activation of information from long-term memory quick and easy, and/or (b) they are necessary to provide text coherence (McKoon & Ratcliff, 1980). The study of the relation between working memory and inference making during reading is crucial, as inferential processing is inherent to language comprehension. Furthermore, the generation of predictions ensures that readers engage in reading for understanding (proactive comprehension) during text comprehension, which provides an ideal situation to explore the involvement of working memory. A second high-level language process is comprehension monitoring. This is a metacognitive ability considered an important executive function for efficient reading because it allows allocating cognitive resources to make sense of incoming information (Wagoner, 1983). It is commonly assessed by tasks requiring the detection of inconsistencies or information that conflict with readers’ prior knowledge. According to Baker (1985; see also Baker et al., 2014), comprehension monitoring encompasses an initial phase known as evaluation, which refers to the mere fact of being able to detect contradictory information in the text, and a subsequent phase named regulation, which is related to the repair processes that are carried out to solve the conflict after contradictory information has been encountered. Taking this distinction into account, the term “comprehension monitoring” will refer hereafter to the evaluation phase (see discussion of “updating information” for a deeper explanation of the regulation phase). Evaluation is conceived as
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
High-Level Text Comprehension Processes
a more routine, passive, and nonstrategic process that depends on the activation of new information and its integration with previous parts of the text and/or background knowledge (Kendeou, 2014). Therefore, the evaluation phase of comprehension monitoring seems to be a less cognitively demanding process, and it is not clear whether working memory capacity is involved here. Finally, updating information covers a broad range of processes going from the activation of new representations to the manipulation of the contents of working memory by replacing irrelevant information with new (or more appropriate) one (Miyake et al., 2000). As it occurs with comprehension monitoring, updating during reading can be measured by tasks entailing conflict detection. However, in order to construct a coherent situation model when incorporating new, inconsistent information, readers are forced to discard the information that it is no longer relevant or outdated (see the Structure-Building model [Gernsbacher, 1990, 1997] for a deeper understanding on the rationale). This particularly demanding type of updating is termed revision (Rapp & Kendeou, 2007). Because this is clearly an essential repair process, the term “updating information” will deal hereafter with the previously mentioned regulation phase of comprehension monitoring (Baker, 1985). Importantly, there is an apparent and direct link between updating and working memory capacity. However, differently from comprehension monitoring, the relation between updating and reading comprehension has traditionally involved working memory updating tasks with a list of words, and studies that have used comprehension tasks requiring updating information in relation to working memory are scarce. Thus, whether readers require working memory to update a situation model during text comprehension remains a fundamental question.
21.4
Working Memory and High-Level Comprehension in the Native Language
Previous literature has clearly demonstrated a positive relationship between working memory and inference-making during reading comprehension in both children (e.g., Potocki et al., 2017) and adults (e.g., BohnGettler & Kendeou, 2014). Moreover, when comprehending linguistic information, the ability to predict subsequent information is reduced in readers with lower executive resources (e.g., working memory) such as children or older adults, compared to young adults (see Ryskin et al., 2020, for a review). Thus, working memory seems to constrain the mechanisms underlying our ability to generate expectations during language comprehension. On the other hand, a considerable amount of scientific research has reported a relationship between working memory and comprehension monitoring in children (e.g., van der Schoot et al., 2012), where low-span children manifest poor comprehension due to difficulties with detecting
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
463
464
ANA I. PÉREZ MUÑOZ AND M. TERESA BAJO
both external and internal inconsistencies. However, although some studies have shown that low working memory young adults may fail to apply comprehension monitoring under certain circumstances (e.g., Linderholm et al., 2008), the relation between comprehension monitoring and working memory in young adults has been less conclusive (see e.g., De Beni et al., 2007). This suggests this population might not recruit working memory (or to a lesser extent) when detecting conflicting information during text comprehension. Furthermore, regarding updating of linguistic information, reading performance has been clearly associated with the ability to update working memory information in both children (e.g., GarcíaMadruga et al., 2014, but see Muijselaar & de Jong, 2015, for opposite results) and adults (e.g., Palladino et al., 2001). A typical measure in this context is the updating word span task, where readers are required to recall the last three-to-four/five smallest objects from a list of words (e.g., from the list “meeting, sense, woodpecker, passion, law, cow, happiness, amount, caterpillar, lamb, feast, and frog,” the three smallest words to be remembered are “woodpecker,” “caterpillar,” and “frog”). This task assesses updating information as it entails maintaining words in working memory to compare their physical size and select the smallest items, and then discarding any previously activated word that no longer meets the “smallest” criteria (i.e., inhibiting no longer relevant information). Studies using the updating word span task have demonstrated that poor comprehenders are also poor at updating the contents of working memory (García-Madruga et al., 2014; Palladino et al., 2001). These difficulties of poor comprehenders have been interpreted as due to problems inhibiting the no-longer-relevant text information, which in turn interferes with the activation of relevant information in working memory (Butterfuss & Kendeou, 2018). Nonetheless, notice that most of these studies have investigated the connection between updating and comprehension by measuring lexical or sentence-level updating instead of assessing readers’ ability to update information during text comprehension. Then, it is not clear whether working memory affects general comprehension when the text forces updating at the situation model level. In a study specifically designed to study high-level comprehension processes during text comprehension, we created the situation model revision task (Pérez et al., 2015), which critically assess inferential processing, comprehension monitoring and updating during online comprehension. In this task, short narratives are presented to participants one by one. The first three sentences of each narrative are used as a context that induces the generation of a predictive inference (e.g., “guitar”; see Table 21.1). Subsequently, participants encounter one of three possible conditions: (a) neutral, which does not refer to the prediction primed by the context (“The concert was taking place at the prestigious national concert hall”); (b) nonupdate, where the sentence is consistent with the initial prediction (“His instrument
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
High-Level Text Comprehension Processes
465
Table 21.1 Example of text used in the situation model revision task
Context Dan was a gypsy who had played flamenco since childhood. He had become a popular musician who played all over the world. Today, he was giving a recital of his favorite works. RT sentence The concert was taking place at the prestigious national concert hall. His instrument was made of maple wood, with a beautiful curved body. His instrument was made of maple wood with a matching bow. Disambiguating word The public was delighted to hear Dan playing the violin.
(Coming from. . . . . .“concert hall”) . . .“curved body”) . . .“matching bow”)
Comprehension sentence In the recital, Dan played his favorite works.
Condition
Process
(bias guitar)
Inference generation
Neutral Nonupdate Update
Comprehension monitoring (Reading times)
Uncertain Unexpected Expected
Updating information (ERPs)
(Yes/No question)
Global comprehension (Accuracy)
was made of maple wood, with a beautiful curved body”); and (c) update, where the information is inconsistent with the initial prediction and supports the generation of a new prediction (“His instrument was made of maple wood with a matching bow”). Because the latter condition prompts readers to detect the inconsistency, reading times in this condition compared to the other two conditions are taken as an index of comprehension monitoring. Finally, the last sentence of the narrative brings a disambiguating word (“The public was delighted to hear Dan playing the violin”), which creates three new conditions depending on the previous sentence: (a) uncertain, when coming from the neutral condition (“concert hall” ! “violin”); (b) unexpected, when coming from the nonupdate condition (“curved body” related to the idea of guitar ! “violin”); and (c) expected, when coming from the update condition (“matching bow” ! “violin”). Event-related potentials recorded in the disambiguating word are taken as indexes of updating information. Importantly, in this paradigm, both comprehension monitoring and updating information occur at the inferential level, which makes the task more cognitively demanding. Finally, general comprehension is assessed by means of a “Yes/No” comprehension sentence placed at the end (“In the recital, Dan played his favourite works”; response: “Yes”). In the original study (Pérez et al., 2015), young English adults were first assessed in working memory (measured by the reading span task and an updating word span task), and then divided into a high or low span group to be tested in the situation model revision task. Our results revealed different effects. First, we observed all readers took longer in the update condition
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
466
ANA I. PÉREZ MUÑOZ AND M. TERESA BAJO
(“matching bow” after predicting “guitar”) compared to the neutral (“concert hall”) and the nonupdate (“curved body”) conditions, with no difference between the last two. These findings signaled that readers were able to generate the initial prediction and then evaluated the information that was inferentially inconsistent with their prior interpretation (inferential comprehension monitoring). The reading times of the two working memory groups did not differ at this sentence, so we concluded that comprehension monitoring was not dependent on working memory capacity. In contrast, the electrophysiological results of the disambiguating word (“violin”) manifested working memory differences. Concretely, high span readers showed a more efficient reduced activation (i.e., parietal P300 and N400) in the expected condition compared to both the uncertain and unexpected conditions than readers with low span capacity. These results were interpreted as a better ability of high working memory readers to update an initial incorrect interpretation and to integrate new semantic information into their situation model. On the contrary, low working memory readers had difficulties updating their situation model probably because they fail to inhibit the initial (now outdated) prediction. In sum, our first study assessing young adults’ L1 text comprehension indicated that whereas comprehension monitoring did not require working memory, updating information at the inferential level involved working memory to be efficiently implemented. Interestingly, in a review, Butterfuss and Kendeou (2018) theorized that updating during text comprehension might be influenced by both domaingeneral and domain-specific factors. On the one hand, they state that effective updating during reading entails domain-general working memory processes related to storage and information processing. We believe this is in line with our previous results, as we showed that updating a situation model requires the activation and maintenance of incoming information (storage) such as the generation of the initial prediction or the alternative interpretation, as well as the detection of a possible conflict and the replacement of no longer relevant information with the new interpretation (processing). On the other hand, Butterfuss and Kendeou (2018) also hypothesized the involvement of domain-specific factors such as the verbal (instead of the visuospatial) domain of working memory to implement inhibitory control over irrelevant linguistic information. However, once again, this hypothesis was based on studies using an updating word (vs. number) span task similar to the one described above (e.g., Pelegrina et al., 2015) or a proactive interference task in which participants were presented with a list of words (vs. faces) and, after performing a distracting task, they were asked to recall one of the words in response to a category cue (e.g., type of fruit; Pimperton & Nation, 2010), rather than tasks measuring updating at the text level. In a different study, we explored working memory domain specificity of updating during text comprehension by means of eye movements (Pérez
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
High-Level Text Comprehension Processes
et al., 2016). Similar to Pérez et al.’s (2015) research, we tested young English adults in a mismatch detection task, where short narratives were presented (e.g., “It was already the 25th of December and Sophie was back home. As a special treat, her father was making her traditional Christmas dinner. The turkey was cooking, and it needed another hour in the ____ before it was done”), containing either an expected (“oven”) or unexpected (“grill”) concept. Subsequently, just below each narrative, readers encountered a comprehension sentence that could be either congruent or incongruent depending on the previous concept (e.g., “The turkey needed to be roasted/barbecued for one more hour”). Eye movements were recorded in the text and sentence target words. In addition, accuracy to the comprehension sentence assessed general reading comprehension. Moreover, this time working memory was measured in both the verbal domain (composite score of the listening recall and backward digits recall tasks) and visuospatial domain (odd one out and spatial recall tasks), to understand whether the differences previously observed in high compared to low span readers in the updating information process were specifically related to the verbal domain. Consistent with previous findings, readers manifested longer fixation durations (i.e., gopast time and total time) in the unexpected compared to the expected concept in the text, indicating successful comprehension monitoring. However, once more, this effect did not interact with either the verbal or visuospatial domain, signaling comprehension monitoring was not supported by working memory capacity. On the contrary, once readers encountered the unexpected concept, lower compared to higher verbal span readers spent more time rereading previous information (longer go-past time) after encountering a comprehension sentence that brought congruent information (“grill” ➔ “barbecued”), suggesting that lower verbal working memory readers experienced difficulties performing online updating. This relationship was not found with the visuospatial domain, indicating that the ability to inhibit no longer relevant information during text comprehension specifically recruits verbal working memory. Finally, in a recent study we explored whether the relation between updating at the situation model level and individual differences in working memory is essentially mediated by inhibitory mechanisms (Pérez et al., 2020). To this aim, we evaluated young English adults in a (multimodal) mismatch detection task, together with a working memory (backward digit span) task and an inhibitory control (flanker) task. Similar to previous findings, our results indicated that both comprehension monitoring and updating information were successfully implemented during the construction of the situation model. However, updating (but not comprehension monitoring) was specifically explained by individual differences in inhibitory control, where higher resistance to distractor interference (higher inhibitory skill) was translated into a better ability to suppress no longer relevant pictorial information and a more efficient ability to update the
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
467
468
ANA I. PÉREZ MUÑOZ AND M. TERESA BAJO
situation model across modalities. The general measure of (verbal) working memory did not explain any effect after inhibitory control was included in the statistical model, confirming updating of the situation model is essentially supported by inhibitory control mechanisms. Overall, studies on high-level text comprehension processes during the native language suggest comprehension monitoring is not explained by individual differences in working memory, even if it occurs at the inferential level. In contrast, working memory is consistently related to updating information: Low span readers experience problems updating the situation model because they fail to discard an initial erroneous prediction, whereas high span readers are faster and more efficient updating information that becomes outdated during online comprehension. This relationship seems to be specifically supported by the verbal (vs. visuospatial) domain of working memory, and it is mainly based on inhibitory mechanisms of cognitive control.
21.5
Working Memory and High-Level Comprehension in the Second Language
In line with studies on native language processing, individual differences in working memory also underlie inferential comprehension in L2. For example, Alptekin and Erçetin (2011) evaluated young Turkish adults, who were highly proficient in English, their L2. They were assessed in both literal and inferential L2 comprehension skills (a short autobiographical story), as well as in L2 working memory (reading span task). They observed that whereas all readers performed similarly in L2 comprehension when literal information was required, high compared to low span readers were significantly better in L2 inferential comprehension. In addition, several studies have demonstrated the connection between working memory and the ability to use prior knowledge to facilitate L2 reading (Joh & Plakans, 2017; Shin et al., 2019). Shin et al. (2019) evaluated young Korean-English bilingual adults in L2 text comprehension (different passages from the TOEFL Practice Test), L2 working memory (reading span task), and L2 background knowledge (vocabulary size test and C-test). Higher working memory span readers reached a better understanding of the text in their second language than lower span readers when they had prior knowledge than when they did not. This suggests L2 readers need to have some knowledge on the specific topic to be able to efficiently use their working memory, which is connected with the idea that working memory is essential to generate knowledge-based inferences (e.g., predictions), especially during L2 reading. In fact, similar to what it has been suggested for children and older adults (Ryskin et al., 2020), L2 comprehenders have a reduced ability to predict incoming information compared to native
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
High-Level Text Comprehension Processes
comprehension, which has been interpreted as due to limited availability of working memory resources (e.g., Hopp, 2013). Interestingly, in recent years, the topic of prediction has been gaining relevance in the bilingual literature (e.g., Foucart et al., 2014; Kaan et al., 2016; Martin et al., 2013; Zirnstein et al., 2018). Similar to the mismatch detection task, where narrative texts biasing a prediction are presented with either an expected or unexpected concept, these studies use highly constrained sentences biasing a lexical prediction (e.g., “It was raining so he grabbed his. . .”), followed by either an expected word (“umbrella”) or an unexpected but still plausible concept (“coat”). The typical effect is an N400 in the unexpected compared to the expected word, which is interpreted as signaling the degree of activation and integration of lexico-semantic information of the predicted concept. However, notice the N400 also reflects inferential comprehension monitoring at the sentence level, that is, the ability of comprehenders to detect that the unexpected word mismatches with the prediction biased by the sentential context. Moreover, some of the studies reporting N400 effects have also reported a late frontally-distributed postN400 positivity (PNP; see Van Petten & Luka, 2012 for a review) when the lexical prediction is replaced by the unexpected concept. The PNP has been related to several revision functions such as difficulty integrating information when constructing, reorganizing, or updating an utterance interpretation (Brouwer et al., 2012); the need to update information when incoming words are not followed from readers’ predictions (Kuperberg, 2016); disconfirmed lexical predictions (frontally located) or an attempt to check or reanalyze problematic information (parietally located, Van Petten & Luka, 2012); and an increase of cognitive resources to implement revision, updating, or conflict-monitoring/resolution processes (Boudewyn et al., 2015). Consistently, in the lexical prediction literature, the PNP has been hypothesized to indicate the cost that comprehenders experience when the predicted word is no longer valid at the sentence level, which has been also demonstrated at the text level (e.g., Brothers et al., 2015). Interestingly, all this literature suggests a connection between the PNP component and the updating information process. Despite the fact that the scientific literature on lexical prediction has yielded mixed results since the N400 and PNP components are not always found under the same L2 comprehension conditions (e.g., Martin et al., 2013; Zirnstein et al., 2018), several studies have shown that their occurrence during L2 comprehension depends on cognitive control. For instance, Zirnstein et al. (2018) presented high constraint sentences (e.g., “After their meal, they forgot to leave a ____ for the waitress”) including an expected (“tip”) or unexpected (“ten”) word, and found that both N400 and PNP effects depended on bilinguals’ L2 proficiency and cognitive control (AX-CPT task): Lower L2 proficient bilinguals showed worse lexical prediction (larger N400) than higher L2 proficient bilinguals, and those with lower inhibitory control incurred more costs (larger PNP) than bilinguals with larger
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
469
470
ANA I. PÉREZ MUÑOZ AND M. TERESA BAJO
inhibitory control ability, in the unexpected word1. These findings indicate that bilingual young adults are, at least in part, able to carry out inferential monitoring (observed in the N400 effect) and updating information (PNP effect) during L2 sentence comprehension, but this is dependent on several linguistic and cognitive factors, such as whether bilinguals are highly proficient in their second language and/or whether they have good cognitive control, among other factors. Importantly, in relation to the latter, L2 processing demands more working memory resources than L1 processing (e.g., Dussias & Piñar, 2010; Kaan, et al., 2016), and as more cognitive resources need to be allocated to lower-level linguistic processes (such as words or sentences), fewer resources are available for higher-level semanticdiscourse processes, where less proficient L2 comprehenders often manifest difficulties (e.g., Horiba & Fukaya, 2015). Thus, a straightforward hypothesis here is that bilinguals’ L2 working memory capacity strongly determines their ability to implement successful comprehension monitoring and updating information during L2 text comprehension. As far as we know, there is only one study that has investigated both inferential comprehension monitoring and inferential updating information during L2 text comprehension (Pérez et al., 2019). In this study, we evaluated native English adults highly proficient in Spanish. They were assessed in the situation model revision task (Pérez et al., 2015), cognitive control (AX-CPT task), L2 proficiency (linguistic background, vocabulary, and verbal fluency), as well as L1 and L2 working memory (operational span task). More concretely, we investigated high-level text cognitive processes during L1 and L2 comprehension by focusing on individual differences in cognitive control and L2 proficiency. Our results showed that bilinguals were able to monitor their inferential comprehension in both languages, but this was less efficient in the L2. In addition, regarding cognitive control, comprehension monitoring did not depend on this factor. In contrast, individual differences in cognitive control explained inferential updating (N400 effect) during the construction of the situation model. Concretely, more efficient inferential updating in the L1 was associated with a balance between proactive and reactive control, whereas a more nativelike L2 updating was related to a stronger proactive control. According to the dual mechanisms of control framework (Braver, 2012), proactive control is a preemptively implemented control mode, based on sustained goal maintenance and anticipatory monitoring during task performance, whereas reactive control involves a momentary and transient activation of the task goal in the light of conflict or interference. Taking into account this distinction, Pérez et al. (2019) interpreted their L1 findings as the ability of native comprehenders to anticipate information by actively generating the prediction suggested by the context (proactive control), and subsequently disengaging from that prediction to accommodate a new inference (reactive control). In contrast, efficient inferential updating in L2 comprehension seemed to involve only a more active control mode (proactive control).
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
High-Level Text Comprehension Processes
Figure 21.1 Reading times index (fourth sentence divided by the mean of the first three sentences or context, in milliseconds) It constitutes a “pure processing time” measure) for the fourth sentence, as a function of language, condition and working memory (L2 divided by L1; less than 1 means better working memory in the L1 compared to the L2, and more than 1 means the opposite). The three-way interaction was significant, F (2, 1738) = 6.72, p < .01, ηp2 = .008, which divided by language showed that the interaction of condition and working memory was marginally significant in the L1, F (2, 882) = 2.60, p =.07, and significant in the L2, F (2, 827) = 3.56, p < .05.
Beyond the complexity of this pattern of results, our findings shed light on the hypothesis that bilinguals’ working memory might underlie their ability to carry out efficient inferential updating during text comprehension. In the previous study, working memory was assessed as a control measure, but we did not analyze for it. Because the aim of the present chapter is to understand how working memory underlies high-level text comprehension processes, we reanalyzed Pérez et al.’s (2019) data to more directly explore whether individual differences in working memory predicted L2 and L1 comprehension processes. In line with prior results (Pérez et al., 2019), inferential comprehension monitoring occurred in both languages. This time, however, comprehension monitoring was determined by working memory. Specifically, in the native language, higher L1 span readers performed marginally better at inferential monitoring (longer reading times in the update condition) than higher L2 span readers, whereas the opposite was statistically significant in the second language, where higher L2 span readers were better than higher L1 span readers (see Figure 21.1). These findings suggest that bilinguals’ working memory capacity constrained the ability to inferentially monitor both L1 and L2 comprehension during the construction of a situation model. At first sight, this pattern seems to contradict our findings with monolingual participants showing that inferential monitoring was not related to working memory in young adults (Pérez et al., 2015, 2016). However, this relation between working memory and inferential monitoring in bilinguals provides support to the idea that bilingualism changes the way some linguistic processes are performed. Many studies have shown that coactivation of the two languages in
https://doi.org/10.1017/9781108955638.026 Published online by Cambridge University Press
471
472
ANA I. PÉREZ MUÑOZ AND M. TERESA BAJO
Figure 21.2 Electrophysiological activity (N400) for the disambiguating word, as a function of language, condition and working memory (L2 divided by L1) The three-way interaction was marginally significant, F (2, 10112) = 2.86, p = .06, ηp2 = .001, which divided by language demonstrated that the interaction of condition and working memory was significant in the L1, F (2, 5338) = 5.81, p .90) were reported by French (2006) for 11-year-old francophone children enrolled in a five-month
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
533
534
ELISABET SERVICE AND DAPHNÉE SIMARD
intensive English immersion program in a northern region of the Frenchspeaking Canadian province of Quebec. In this study, the items that had to be repeated were either English sounding pseudowords from the CNRep test (Gathercole & Baddeley, 1996) or Arabic real words, scored by two native speakers. English vocabulary learning was measured using a task probing productive BOV, provided by the provincial educational authority. Surprisingly, the correlations between the word translation score and English pseudoword repetition, on the one hand, and Arabic word repetition, on the other, were very similar, r = .78 and r = .80, respectively, suggesting that phonological similarity between the repeated nonwords and the words being learned did not make a difference. In this study, foreign word and pseudoword repetition predicted new word learning during the intensive learning period for the less L2-proficient half of the sample but not significantly for the more proficient half. The studies by Service and French speak to a debate in the L1 literature concerning the direction of causation, as they suggest that whatever pseudoword repetition reflects also predicts vocabulary learning in an academic setting over time rather than being itself predicted by already existing language skills. Furthermore, French’s results are also in line with other findings (Masoura & Gathercole, 2005) suggesting that phonological STM as measured by pseudoword repetition tasks only plays a role at the beginning stages of L2 vocabulary acquisition. Despite the very clear older findings of the role of phonological STM in L2 vocabulary acquisition, later research, often with older learners, has reported lower and sometimes nonsignificant correlations. Difficulty in controlling the potent effects of L2 exposure to the same extent as in the early studies with homogenous L2-naïve samples may explain some of the null results. However, aspects of the nonword repetition task itself may limit its sensitivity in adult learners. In an attempt to develop a phonological STM task suited for university students but not depending on articulatory skill, carried out a series of experiments comparing performance in young adults and 8-year-old children. The dependent variables were implicit and explicit aspects of memory for novel word forms adhering to L1 (Finnish) phonology and phonotactics. Implicit memory in this context was defined as aspects of learning that were not intentional or available to conscious reflection, such as improvement in phonotactic accuracy and response speed over task trials. Explicit memory refers here to the ability to explicitly recall from memory taskrelated language material, such as word stimuli, when asked to do so. Performance was predicted from variants of pseudoword span with task difficulty calibrated to the respective age groups. Previous research had shown that greater nonword length (Gathercole et al., 1994) inclusion of consonant clusters (Archibald & Gathercole, 2006) as well as foreign sounds and illegal or rarely encountered sound sequences (Kovács & Racsmány, 2008) add difficulty to nonwords in repetition tasks.
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
L2 Vocabulary
Table 24.1 Correlations between versions of L1 pseudoword span and explicit memory for new L1 or L2 word forms
Study Children Exp 1 N = 28 Adults Exp 1 N = 31 Adults Exp 2 N = 27 Adults Exp 3 N = 32
L1 pseudoword span & L1 wordpseudoword pairs
L1 pseudoword span & incidental L1 pseudoword learning
L1 pseudoword span & foreign word recognition (d’) after incidental learning
0.67***
0.70***
0.35a
0.26
0.08
0.16
0.23
0.09
‒0.04
0.27
Note: L1 was Finnish and L2 was Korean for young adults and 8-year-old children. *** p < .001 a p = .0679
In our experiments, the number and complexity of syllables in the pseudowords were manipulated to achieve more demanding variants of the pseudoword span task for our adult participants. In addition to pseudoword span, we used a foreign (Korean) word/collocation repetition task. The ability of the different phonological STM tasks to predict learning of new word forms in laboratory tasks can be seen in Table 24.1. Although the correlations were consistently positive (except for one very close to zero), robust effect sizes were, to our disappointment, found only for the children. A further explorative study correlating foreign-word repetition (see Table 24.2) with aspects of word learning also produced low correlations. Pseudoword span when the pseudowords had familiar phonology and phonotactics appeared only weakly predictive of memory for new wordforms in adults. Only foreignword repetition showed a moderate prediction effect, but performance would likely be highly dependent on the language pairs involved in the repetition and learning tasks. This led us to abandon the effort to develop a more sensitive phonological STM task for adults, for a decade. Over the last 10 years, a number of results have been published suggesting that fundamental abilities to represent the temporal dynamics of stimuli (e.g., Goswami, 2018) may underlie language-based skills such as phonological awareness (i.e., the ability to manipulate and reflect upon the sounds of a given language; Nesdale et al., 1984). This could be seen in, for instance, sensitivity to the shape of the speech envelope of spoken words. Another possible primitive skill is being able to tell the order in time of, for instance, animals arriving at the goal in a race (e.g., Majerus & Boukebza, 2013). This has led us to a new hypothesis based on the idea that the nonword repetition task has predictive validity because it captures the ability to form mental representations of patterns in time. Based on this idea, a student team: Maria de los Angeles Lopez Ricote, Erin DeBorba, and
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
535
536
ELISABET SERVICE AND DAPHNÉE SIMARD
Table 24.2 Correlations between L2 word repetition and memory for intentionally or incidentally learned new word forms for young adults and 8year-old children
Study Children Exp 1 N = 23 Adults Exp 1 N = 25
Foreign word repetition & L1 wordpseudoword pairs
Foreign word repetition & incidental L1 pseudoword learning
Foreign word repetition & foreign word recognition (d’) after incidental learning
0.30
0.35a
0.44*
0.38b
0.22
0.25
Note: L1 was Finnish and L2 Korean. Performance level on the repetition tasks was similar for both age groups. To avoid floor and ceiling effects, word-pseudoword learning was scored over the first two learning trials for the adults and over all four learning trials for the children. * p < .05 a
p = .0984
b
p = .0621
Meliha Horzum designed nonsentence repetition tasks based on English, Spanish, and Turkish (DeBorba, 2020; Lopez Ricote, 2020). The idea was that pseudowords embedded in prosodically rich contexts would be even more sensitive to memory for patterns in time. We found for native English speakers that the scores in the English and Turkish task variants were highly correlated with each other (r(70) = 0.59 for word-level scoring; r(70 = 0.61 for syllable scoring) and were also significantly correlated with learning the syllables of new Turkish three-syllable words. The correlation between a joint score of nonsentence repetition and complex foreign-word learning, r(70) = 0.47, was substantial, and may have been restricted mainly by the reliability of the difficult learning task. The study had been designed to explore whether a domain-general capacity for representing temporal patterns in STM forms part of the more specific ability to present linguistic structure. For this purpose, we also investigated the potential correlation of a rhythmic tapping task with the nonsentence repetition task score. In this task, participants heard a sequence of long and short “beeps.” Their task was to reproduce this Morse-code-like sequence by tapping on a key on a computer keyboard. The proportion of correct short and long taps was scored. We found a significant correlation [r(70) = 0.42] with nonsentence repetition, providing support for the hypothesis that temporal pattern frames might be used to represent the order of phonemes, syllable segments, and whole syllables in phonological STM when preexisting order chunks are not readily available to provide LTM support for novel bindings in STM. Our next question concerned a possible direct role for WM capacity for temporal patterns to support learning of the phonological structure of new
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
L2 Vocabulary
words in a language not closely related to L1. To answer this question, we investigated the paths between our rhythm measure and our word learning measure. It turned out that although performance in the tapping task significantly predicted wordform learning, the explained variance was not unique. A mediation analysis suggested that while there was a significant direct link between nonsentence repetition and word form learning, the linguistic repetition tasks mediated the association between rhythmical tapping and learning of syllables in L2 words. The nonsentence results add support to the hypothesis that the ability to immediately repeat aloud pseudowords or foreign words embedded in natural-language prosodic contexts is related to the ability to pick up the phonological structure of novel L2 words. The finding that STM for rhythmic patterns, tested with the tapping task, was related to nonsentence repetition but not directly to wordform learning, is compatible with the possibility that this type of memory belongs to the family supported by the phonological loop, as conceptualized in recent formulations of the Baddeley and Hitch WM framework (Baddeley, 2012). The results also fit well with recent findings (Gilbert et al., 2017) suggesting that covert rehearsal in auditory list recall tasks is temporally structured, mimicking the rhythm of stimulus presentation. Although our latest data are promising, it remains a puzzle why designing a task for adults that can detect a relationship between phonological STM and phonological word-form learning is so difficult. The accumulation of L1 phonological and phonotactic knowledge in the adult neural networks is usually thought to hinder rather than help L2 phonological learning. One possible answer emerged from a study where we compared 8year-olds and young adults in a foreign-word repetition task (Service et al., 2014). There were no significant performance differences between the two age groups in either repetition accuracy or improvement over repeated trials without feedback. However, when presented with a surprise recognition test where the participants had to discriminate foreign word stimuli that they had heard during the repetition task (old) from novel stimuli (new), the adults performed substantially better. The reason for this better episodic memory performance (i.e., ability to explicitly remember the stimuli that had formed part of the repetition task) in adults may have been their more readily available semantic rather than L1 phonological knowledge. With highly efficient mental search machines built over many years of language exposure, even foreign words were likely to trigger some activation in semantic memory, allowing the set-up of cues for episodic retrieval. Consequently, ability to recognize word forms or recall them when cued with L1 words would receive more semantic support for adults than it does for children who have less efficient semantic networks. At the moment, this is a speculative answer, but it is compatible with reports of patients whose semantic networks have been progressively disintegrating. These patients perform as poorly on memory tasks with real words as they do on tasks including pseudowords (Hoffman et al., 2009).
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
537
538
ELISABET SERVICE AND DAPHNÉE SIMARD
To summarize, phonological STM and early L2 word learning have repeatedly been found to be correlated. However, none of the phonological STM tasks can be considered a pure measure of the underlying construct for the reasons outlined above. Furthermore, these tasks come to reflect accumulating LTM knowledge of different kinds over the life span. Attempts to improve their sensitivity to STM in adult participants were only moderately successful. New repetition tasks involving pseudowords embedded in prosodic sentence structures in different languages showed promise in predicting the learning of syllables in multisyllabic foreign words with an unfamiliar prosodic structure. Moreover, these repetition tasks were correlated with a tapping task gauging STM for temporal patterns. If representations of rhythm provide the skeleton that phonological units are attached to for forming STM representations of novel word forms, pedagogic attention needs to be directed to such rhythm patterns and their manifestation in phrasal stress, intonation, and durational patterns. Once such patterns are established in a new language, learning of new word forms can rely on analogy with already acquired word forms. Familiar prosodic skeletons may also provide keys to the well-organised semantic networks of adult learners.
24.4
The Central Executive and L2 Vocabulary Knowledge
Many tasks have been designed to assess the executive aspect of WM, also called executive or general WM. The most widely studied one in L2 research is the reading span task, developed by Daneman and Carpenter (1980). In this task, participants are asked to read sentences, indicate whether they are plausible or not, and then memorize the final word of each for later recall. The number of sentences increases as the task proceeds. In L2 studies, the reading span task has been administered in both L1 and L2. In general, no significant correlations have been observed between results obtained in reading span tasks administered in either L1 or L2 and L2 vocabulary knowledge. These null effects do not depend on whether vocabulary was measured receptively using a breadth of vocabulary test, such as the Vocabulary Size Test or the Vocabulary Level Test (e.g., Coxhead, et al., 2014; Nation & Beglar, 2007), or using a receptive DOV task (e.g., Qian & Schedl, 2004). Variations on the reading span test include listening span, which requires participants to judge sentences they hear and remember the last word of each sentence. In a recent study conducted among adults (Masrai, 2020), correlations were observed between a listening span test administered in the participants’ L2 and both written and aural L2 breadth-of-vocabulary tests (based on the Yes/No Vocabulary Size Test, Meara, 1992). Also listening span administered in the participants’ L1 (Greek) has been found to predict both receptive and productive L2 breadth of vocabulary knowledge among children (Efstathiadi, 2016). However, when the listening span task was administered to children in their L2, no correlation could be observed (Zaretsky, 2020).
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
L2 Vocabulary
Speaking span involves the recall of sets of words and the production of sentences using the words in each set. There are substantially fewer studies that used the speaking span test to investigate the relationship between executive WM and vocabulary knowledge. In the few studies that used the speaking span among adults, correlations were observed between WM measured in the participants’ L2 and L2 receptive and productive breadth of vocabulary knowledge tasks (e.g., Fortkamp & Verçosa, 2019). Two different factors surface from this brief review as seeming to influence the relationship between WM and vocabulary knowledge: the modality and the language in which the WM task is administered. First, in the studies reviewed, the written test, that is, the reading span, revealed no association between executive WM and vocabulary knowledge (breadth or depth), while the aural and oral tasks were more successful. This difference, in particular between the reading and listening spans, is interesting, as written and aural comprehension have long been known to be correlated, especially among children (e.g., Carr et al., 1990; Oakhill & Cain, 2004). Some researchers go even as far as proposing a unitary construct (Hogan et al., 2014). However, although, both written and aural comprehension require the activation of representations in LTM during meaning construction when processing text (Kintsch et al., 1999), we propose that not only is access to information in LTM different in the two modalities, but the knowledge involved in each modality is of a different kind. Some researchers have even proposed that written and spoken language involve different language processing systems (e.g., Howard & Franklin, 1988). For instance, according to the Dual Route Cascaded model put forth by Coltheart et al. (2001), there are two access routes acting in parallel or in competition with each other when reading words: the lexical route, that is, the direct association between word meaning and its form, and the phonological route, which assumes an association between sublexical sounds and whole words. In visual word recognition using this second route, words are first constructed from their phonetic parts, that is, the existing associations between graphemes and phonemes. Therefore, the length of the path to meaning construction in the written modality could be affected by route dominance in reading span compared with the more direct listening and speaking versions. Individual variability in access to the less taxing lexical route could create noise in studies using reading span as the WM measure. Secondly, the differences observed in the results can certainly be explained at least in part by the language in which the WM spans were administered (see Simard et al., 2020). On the one hand, when either the reading or listening WM span is administered to children in their L2, the task may simply be too demanding to allow observation of an association with vocabulary knowledge, as their language knowledge system might not sufficiently support the execution of the WM task. On the other hand, in adult studies, when the WM task and the language tasks are administered in the same language, there is overlap between the content manipulated and memorized
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
539
540
ELISABET SERVICE AND DAPHNÉE SIMARD
(e.g., Turner & Engle, 1989). Therefore, some researchers have proposed using numerical tasks in studies investigating the relationship between general WM and language processing, with the assumption that WM is not specific to a particular domain, but rather represents a general capacity. In numerical WM task variants, participants are asked to either repeat sequences of numbers or perform mathematical calculations and to memorize words or letters presented along with the mathematical problems (Operation Span Test; Engle et al., 1992; see also Foster et al., 2015). In another type of numerical task, participants are asked to say out loud the highest number among a group of three numbers, and to memorize these highest numbers (Highest Number Task; Oakhill et al., 2011). Also, tasks in which participants are asked to repeat a sequence of numbers (digit span tasks) have been used as general WM measures. The most commonly studied ones are the forward or the backward digit spans. They are considered to tap into different constructs (Rosenthal et al., 2006). Some researchers consider only the backward digit span to measure the executive aspect of WM (see Juffs & Harrington, 2011); and that only the forward version should be employed for a STM assessment (e.g., St Clair-Thompson, 2010). Here, the characteristics of the participant group may be critical. Although backward span factors correlate with other executive WM measures in children (e.g., Alloway et al., 2006), this is not so in young adults, at least not in university students (Engle et al., 1999; St Clair-Thompson, 2010). As a common observation, the backward digit span administered to children in L1 appears to correlate with L2 receptive DOV (Blom et al., 2014; Bosma et al., 2017; Łockiewicz & Jaskulska, 2015). This result is corroborated by Efstathiadi (2016), who also found a significant association between the backward digit span (administered in L1 Greek) and both receptive and productive L2 BOV knowledge among children. Similar results have also been observed among adults between backward digit span in L1 and a productive BOV knowledge measure (e.g., Kormos & Sáfár, 2008). However, when the backward digit span was administered in the participants’ L2, no correlation with either receptive (recognition task) or productive (providing definition) L2 BOV knowledge tasks was observed (e.g., Varol & Erçetin, 2016). A variation of the usual forward and backward digit span is the Random version in which participants must produce sequences of random digits. It is considered to exert a heavy load on the central executive (Baddeley et al.,1998; Repovš & Baddeley, 2006). However, no association for either more proficient and less proficient participants could be observed between the random digit span task administered in either the participants’ L1 or L2 and an English L2 BOV knowledge task (Joyce, 2019). When operation span tasks were used to measure the executive aspect of WM in either L1 or L2, no significant correlation between the L1 operation span and either kind of L2 BOV knowledge (receptive and productive) was found. However, a weak association (r = 0.19, p = 0.05) was found between the L2 operation span and a productive vocabulary measure in which the
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
L2 Vocabulary
participants had to complete a sentence by providing the right word given two to five letters (resembling a c-test) (see Lu, 2015). Matielo et al. (2018) administered an operation span test in their participants’ L1, Brazilian Portuguese. No correlation was found between WM and the scores obtained on a receptive L2 BOV recognition task in which the participants had to provide definitions or translations. These results again confirm that the language in which the WM task is administered affects its relation with vocabulary knowledge. In a recent study (Simard & Molokopeeva, 2019), we looked at the potential role of WM as a cofactor in the relationship between reading comprehension and DOV, as it has been identified to predict both L2 reading comprehension (e.g., Alptekin & Erçetin, 2009, 2010) and DOV knowledge. We administered the Highest Number Task (Oakhill et al., 2011) to advanced adult learners of French in their L1 (Russian) and a DOV knowledge task (the DKFVT; Greidanus et al., 2005) in L2. We measured depth of L2 reading comprehension using a task based on Kintsch’s (1998) text representation levels, with separate items targeting the surface level, that is, explicit information contained in the text; the textbase level, that is, implicit information found in the text; and the situation model, that is, items integrating both implicit and background knowledge. As expected, a moderate correlation, r(30) = 0.414) was observed between WM and reading comprehension. However, no association between either WM or reading comprehension and DOV knowledge could be detected. Interestingly, when we partialed out the effect of WM, a moderate correlation, r(30) = 0.314, between DOV knowledge and reading comprehension was revealed. In other words, only when variation in WM was controlled could the relationship between DOV and depth of text representation, which taps into semantic representations in LTM, be observed. We speculate that the effect was carried specifically by the items targeting the textbase and situation models, as they both involve inferencing. On the one hand, although the role of WM in inferencing has been demonstrated numerous times, the specific way it supports the generation of inferences, the process used to interpret utterances or connect utterances together (Brown & Yull, 1983, p. 33) is still open for debate (Yeari, 2017). On the other hand, the centrality of depth of vocabulary in inferencing has be demonstrated before (e.g., Nassaji, 2004). Therefore, we propose that WM support for inferences during L2 reading comprehension is in the form of better semantic information processing and access to LTM when DOV knowledge is limited. All these executive WM tasks, as they are described above, involve temporary storage of verbal material (numbers, words) by means of the phonological loop component of WM. Thus, only the processing component of the task may be less verbal (numerical processing). Therefore, when numerical complex WM tasks are used in L2 studies, they should be administered in the participants’ L1. This would be done in order to reduce the potential language knowledge bias – the effect that the degree of proficiency in the language in which the task is administered has on the results obtained
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
541
542
ELISABET SERVICE AND DAPHNÉE SIMARD
(with better language knowledge being associated with better results) – in the storage component measure of the task (see French, 2006, for more detail). Studies that have compared verbal and numerical complex WM tasks in L1 have found significant positive correlations between the two types of measures (e.g., Turner & Engle, 1989), leading to the conclusion that WM is a general-domain capacity. However, in L2 research, WM has been more strongly correlated with L2 measures when the memory tasks were verbal (see Linck et al., 2014, for a metanalysis of studies investigating WM and L2 use and acquisition). For instance, WM tasks, complex or simple, that include only meaningful verbal material, such as the reading span task mentioned above, are more strongly correlated with L2 measures such as vocabulary or reading comprehension tasks, than WM tasks involving numerical material, such as complex WM tasks in which the participants have to perform mathematical operations. As the authors explain: this verbal/nonverbal difference could simply be due to the overlap in the content being manipulated (i.e., common method bias), despite the fact that the WM system per se is not specialized for or constrained to a specific content domain. (Linck et al., 2014, pp. 872–873) Finally, it should be mentioned that recently some researchers have preferred to avoid verbal processing altogether during the measurement of WM and used visuospatial tasks. For instance, Vafaee and Suzuki (2020) found that two visual complex WM tasks, namely, the blockspan and shapebuilder tasks, were positively correlated with receptive BOV and DOV knowledge tasks among L2 adult learners. In addition to offering a new outlook on the relationship between WM and vocabulary knowledge, these results support the position that WM is domain-general.
24.5
Conclusion
We have reviewed studies correlating measures of phonological STM and executive WM with L2 BOV and DOV. We point out a number of methodological biases in the measures in common use which can account for the variability in the observed results. However, we find that recent data reveal exciting new ways to understand the reported relationships between aspects of WM and L2 vocabulary. Firstly, we propose that the toolkit of phonological STM measures can be improved by adding nonsentence repetition tasks in different languages to catch memory for richer natural prosodic patterns. Moreover, a receptive version of the task should be devised and compared with the productive one. These tools should better reveal individual differences in beginning vocabulary learning that depends on ability to process temporal patterns. Secondly, we urge L2 researchers to pay more attention to the aspects of learner vocabulary they are testing. BOV has been the most common target
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
L2 Vocabulary
so far. However, DOV knowledge informs about the developing directaccess connections to the rich semantic LTM networks of adult learners. The need for limited executive WM resources in both L2 processing and learning can be hypothesized to depend on the number of steps that are needed to activate meaning from form.
References Abdallah, F. (2010). The role of phonological memory in L2 acquisition in adults at different L2 proficiency levels. (Thèse de doctorat inédite, Université Laval, Québec, Québec, Canada). Alloway, T. P., Gathercole, S. E., & Pickering, S. J. (2006). Verbal and visuospatial short-term and working memory in children: Are they separable? Child Development, 77(6), 1698–1716. Alptekin, C., & Erçetin, G. (2009). Assessing the relationship of working memory to L2 reading: Does the nature of comprehension process and reading span task make a difference? System, 37, 627–639. Alptekin, C., & Erçetin, G. (2010). The role of L1 and L2 working memory in literal and inferential comprehension in L2 reading. Journal of Research in Reading, 33, 206–219. Anderson, R. C., & Freebody, P. (1981). Vocabulary knowledge. In J. T. Guthrie (Eds.), Comprehension and teaching: Research reviews (pp. 77–117). International Reading Association. Archibald, L. M. D., & Gathercole, S. E. (2006). Nonword repetition: A comparison of tests. Journal of Speech, Language, and Hearing Research, 49, 970–983. Baddeley, A. D. (1986). Working memory. Clarendon Press. Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Science, 4, 417–423. Baddeley, A. D. (2010). Working memory. Current Biology, 20, 136–140. Baddeley, A. D. (2012). Working memory: Theories, models, and controversies. Annual Review of Psychology, 63, 1–29. doi:10.1146/annurev-psych120710-100422 Baddeley, A. D. (2015). Working memory in second language learning. In Z. E. Wen, M. B. Mota, & A. McNeill (Eds.), Working memory in second language acquisition and processing (pp. 17–28). Multilingual Matters. Baddeley, A. D., Allen, R. J., & Hitch, G. (2010). Investigating the episodic buffer. Psychologica Belgica, 50, 223–243. Baddeley, A. D., Gathercole, S., & Papagno, C. (1998). The phonological loop as a language learning device. Psychological Review, 105, 158–173. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (pp. 47–89). Academic Press.
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
543
544
ELISABET SERVICE AND DAPHNÉE SIMARD
Baddeley, A. D., & Logie, R. H. (1999). Working memory: The multiplecomponent model. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 28–62). Cambridge University Press. Baddeley, A. D., Papagno, C., & Vallar, G. (1988). When long-term learning depends on short-term storage. Journal of Memory and Language, 27, 586–595. Batista, R., & Horst, M. (2016). A new receptive vocabulary size test for French. Canadian Modern Language Review, 72, 211–233. Blom, E., Küntay, A. C., Messer, M., Verhagen, J., & Leseman, P. (2014). The benefits of being bilingual: Working memory in bilingual Turkish-Dutch children. Journal of Experimental Child Psychology, 128, 105–119. Bosma, E., Heeringa, W., Hoekstra, E., Versloot, A., & Blom, E. (2017). Verbal working memory is related to the acquisition of cross-linguistic phonological regularities. Frontiers in Psychology, 8, 1487. Brown, G., & Yull, G. (1983). Discourse analysis. Cambridge University Press. Carr, T. H., Brown, T. L., Vavrus, L. G., & Evans, M. A. (1990). Cognitive skill maps and cognitive skill profiles: Componential analysis of individual differences in children’s reading efficiency. In T. H. Carr & B. A. Levy (Eds.), Reading and its development: Component skills approaches (p. 1–55). Academic Press. Coltheart, M., Rastle, K., Perry, C., Langdon, R., & Ziegler, J. (2001). DRC: A dual route cascaded model of visual word recognition and reading aloud. Psychological Review, 108, 204-256. doi: 10.1037//0033-295X.108.1.204 Coxhead, A., Nation, P., & Sim, D. (2014). Creating and trialling six versions of the Vocabulary Size Test. TESOLANZ Journal, 22, 13–27. Daneman, M., & Green, I. (1986). Individual differences in comprehending and producing words in context. Journal of Memory and Language, 25, 1–18. DeBorba, E. (2020). The investigation of temporal order in language learning using behavioural tasks and MMN (Master’s thesis, McMaster University). http:// hdl.handle.net/11375/25801 Dunn, L., & Dunn, D. (2007). Peabody picture vocabulary test (4th ed.) (PPVT-4 scale). American Guidance Services. Dunn, L. M., Theriault-Whalen, C. M., & Dunn, L. M. (1993). Echelle de vocabulaire en images Peabody [French Peabody Picture Vocabulary Test]. ATM. Efstathiadi, L. (2016). Vocabulary acquisition by young Greek learners of L2 English: The predictive role of complex working memory in early foreign language learning. Selected Papers on Theoretical and Applied Linguistics, 21, 527–547. Engle, R. W., Cantor, J., & Carullo, J. J. (1992). Individual differences in working memory and comprehension: A test of four hypotheses. Journal of Experimental Psychology: Learning, Memory, and Cognition, IS, 972–992. Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. A. (1999). Working memory, short-term memory and general fluid intelligence:
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
L2 Vocabulary
A latent variable approach. Journal of Experimental Psychology: General, 128, 309–331 Fortkamp, M. B. M., & Verçosa, D. M. (2019). Working memory capacity and the retention of L2 vocabulary. Leitura, 1, 55–74. Foster, J. L., Shipstead, Z., Harrison, T. L., Hicks, K. L., Redick, T. S., & Engle, R. W. (2015). Shortened complex span tasks can reliably measure working memory capacity. Memory and Cognition, 43, 226–236. French, L. M. (2006). Phonological working memory and L2 acquisition: A developmental study of Quebec Francophone children learning English. Edwin Mellen Press. Gathercole, S. E. (1995). Is nonword repetition a test of phonological memory or long-term knowledge? It all depends on the nonwords. Memory & Cognition, 23, 83–94. Gathercole, S. E. (2006). Nonword repetition and word learning: The nature of the relationship. Applied Psycholinguistics, 27, 513–543. Gathercole, S. E., & Baddeley, A. D. (1989). Evaluation of the role of phonological STM in the development of vocabulary in children: A longitudinal study. Journal of Memory and Language, 28, 200–213. Gathercole, S. E., & Baddeley, A. D. (1996). The Children’s Test of Nonword Repetition. Psychological Corporation. Gathercole, S. E, Hitch, G. J., Service, E., & Martin, A. J. (1997). Phonological short-term memory and new word learning in children. Developmental Psychology, 33, 966–979. Gathercole, S. E., & Pickering, S. J. (2000). Assessment of working memory in six- and seven-year-old children. Journal of Educational Psychology, 92, 377–390. Gathercole, S. E., Service, E., Hitch, G. J., Adams, A., & Martin, A. J. (1999). Phonological short-term memory and vocabulary development: Further evidence on the nature of the relationship. Applied Cognitive Psychology, 13, 65–77. Gathercole, S. E., Willis, C., Baddeley, A., & Emslie, H. (1994). The Children’s Test of Nonword Repetition: A test of phonological working memory. Memory, 2, 103–127. Gilbert, R. A., Hitch, G. J., & Hartley, T. (2017). Temporal precision and the capacity of auditory-verbal short-term memory. Quarterly Journal of Experimental Psychology, 70(12), 2403–2418. Goswami, U. (2018). A neural basis for phonological awareness? An oscillatory temporal-sampling perspective. Current Directions in Psychological Science, 27(1), 56–63. Greidanus, T., Beks, B., & Wakely, R. (2005). Testing the development of French word knowledge by advanced Dutch-and English-speaking learners and native speakers. Canadian Modern Language Review, 62(4), 509–532. Hoffman, P., Jefferies, E., Ehsan, S., Jones, R. W., & Lambon Ralph, M. A. (2009). Semantic memory is key to binding phonology: Converging
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
545
546
ELISABET SERVICE AND DAPHNÉE SIMARD
evidence from immediate serial recall in semantic dementia and healthy participants. Neuropsychologia, 47(3), 747–760. Hogan, T. P., Adlof, S. M., & Alonzo, C. N. (2014). On the importance of listening comprehension. International Journal of Speech-Language Pathology, 16, 199–207. Howard, D., & Franklin, S. (1988). Missing the meaning? A cognitive neuropsychological study of the processing of words by an aphasic patient. MIT Press. Jones, G., & Macken, B. (2015). Questioning short-term memory and its measurement: Why digit span measures long-term associative learning. Cognition, 144, 1-13. doi:10.1016/j.cognition.2015.07.009 Joyce, P. (2019). The relationship between L2 listening proficiency and L2 aural language processing. PASAA: Journal of Language Teaching and Learning in Thailand, 57, 9–32. Juffs, A., & Harrington, M. W. (2011). Aspects of working memory in L2 learning. Language Teaching: Reviews and Studies, 42, 137–166. Kintsch, W. (1998). Comprehension: A paradigm for cognition. Cambridge University Press. Kintsch, W., Patel, V. L., & Ericsson, K. A. (1999). The role of long-term working memory in text comprehension. Psychologia, 42, 186–198. Kormos, J., & Sáfár, A. (2008). Phonological short term-memory, working memory and foreign language performance in intensive language learning. Bilingualism: Language and Cognition, 11, 261–271. Kovács, G., & Racsmány, M. (2008). Handling L2 input in phonological STM: The effect of non-L1 phonetic segments and non-L1 phonotactic sequences on nonword repetition. Language Learning, 58, 597–624. Laufer, B., & Aviad-Levitzky, T. A. M. I. (2017). What type of vocabulary knowledge predicts reading comprehension: Word meaning recall or word meaning recognition? The Modern Language Journal, 101, 729–741. Laufer, B., & Goldstein, Z. (2004). Testing vocabulary knowledge, size, strength, and computer adaptiveness. Language Learning, 54, 399–436. Linck, J. A., Osthus, P., Koeth, J. T. & Bunting, M. F. (2014). Working memory and second language comprehension and production: A meta-analysis. Psychonomic Bulletin & Review 861–883. doi: 10.3758/s13423-013-0565Łockiewicz, M., & Jaskulska, M. (2015). Mental lexicon, working memory and L2 (English) vocabulary in Polish students with and without dyslexia. Center for Educational Policy Studies Journal, 5, 71–89. Lopez Ricote, M. d. l. A. (2020). Serial order in language learning in bilinguals. (Master’s thesis, McMaster University). http://hdl.handle.net/11375/25799 Lu, Y. (2015). Working memory, cognitive resources and L2 writing performance. In Z. E. Wen, M. Mota, & A. McNeill (Eds.), Working memory in second language acquisition and processing (pp.175–188). Multilingual Matters. Macken, B., Taylor, J., & Jones, D. (2015). Limitless capacity: A dynamic objectoriented approach to short-term memory. Frontiers of Psychology, 6, 293. Majerus, S., & Boukebza, C. (2013). Short-term memory for serial order supports vocabulary development: New evidence from a novel word learning paradigm. Journal of Experimental Child Psychology, 116(4), 811–828. https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
L2 Vocabulary
Masoura, E. V., & Gathercole, S. E. (2005). Contrasting contributions of phonological short-term memory and long-term knowledge to vocabulary learning in a foreign language. Memory, 13(3-4), 422-429. Masrai, A. (2020). Exploring the impact of individual differences in aural vocabulary knowledge, written vocabulary knowledge and working memory capacity on explaining L2 learners’ listening comprehension. Applied Linguistics Review. 11, 423-447. Matielo, R., Oliveira, R. P. D., & Baretta, L. (2018). Subtitling, working memory, and L2 learning: A correlational study. Revista brasileira de linguística aplicada, 18, 665–696. Meara P. (1992). Network structures and vocabulary acquisition in a foreign language. In P. J. L. Arnaud & H. Béjoint (Eds.), Vocabulary and applied linguistics (pp. 62–70). Palgrave Macmillan. Meara, P., & Jones, G. (1990). The Eurocentres Vocabulary Size Tests. 10KA. Eurocentres. Meara, P., & Milton, J. (2003). X_Lex, The Swansea Levels Test, Express. Melby-Lervåg, M., Lervåg, A., Lyster, S. A., Klem, M., Hagtvet, B., & Hulme, C. (2012). Nonword-repetition ability does not appear to be a causal influence on children’s vocabulary development. Psychological Science, 23, 1092–1098. Milton, J. 2009. Measuring second language vocabulary acquisition. Multilingual Matters. Nader, M., Simard, D., Fortier, V., & Molokopeeva, T. (2017). Étude de la contribution de la mémoire de travail et de la mémoire phonologique dans la réalisation d’une tâche métasyntaxique chez des enfants de langue d’origine. Revue canadienne de linguistique appliquée/Canadian Journal of Applied Linguistics, 20, 55–75. Nassaji, H. (2004). The relationship between depth of vocabulary knowledge and L2 learners’ lexical inferencing strategy use and success. The Canadian Modern Language Review, 61, 107–134. Nation, I. S. P. (1983). Testing and teaching vocabulary. Guidelines, 5, 12–25. Nation, I. S. P. (1990). Teaching and learning vocabulary. Newbury House. Nation, I. S. P. (2001). Learning vocabulary in another language. Cambridge University Press. Nation, I. S. P., & Beglar, D. (2007). A vocabulary size test. The Language Teacher, 31, 9–13. Nesdale, A. R., Herriman, M. L., & Tunmer, W. E. (1984). Phonological awareness in children. In W. E. Tunmer, C. Pratt, & M. L. Herriman (Eds.), Metalinguistic awareness in children (pp. 56–72). Springer. Oakhill, J. V., & Cain, K. (2004). The development of comprehension skills. In T. Nunes & P. Bryant (Eds.), Handbook of children’s literacy (p. 155–180). Kluwer Academic Publishers. Oakhill, J., Yuill, N., & Garnham, A. (2011). The differential relations between verbal, numerical and spatial working memory abilities and children’s reading comprehension. International Electronic Journal of Elementary Education, 4, 83–106. https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
547
548
ELISABET SERVICE AND DAPHNÉE SIMARD
Ottem, E. J., Lian, A., & Karlsen, P. J. (2007). Reasons for the growth of traditional memory span across age. European Journal of Cognitive Psychology, 19(2), 233–270. doi:10.1080/09541440600684653 Qian, D. D., & Schedl, M. (2004). Evaluating an in-depth vocabulary knowledge measure for assessing reading performance. Language Testing, 21, 28–52. Read, J. (1993). The development of a new measure of L2 vocabulary knowledge. Language Testing, 10, 355–371. Repovš, G., & Baddeley, A. (2006). The multi-component model of working memory: Explorations in experimental cognitive psychology. Neuroscience, 139(1): 5–22. Rosenthal, E. N., Riccio, C. A., Gsanger, K. M., & Jarratt, K. P. (2006). Digit span components as predictors of attention problems and executive functioning in children. Archives of Clinical Neuropsychology: The Official Journal of The National Academy of Neuropsychologists, 21, 131–139. Schmitt, N. (2014). Size and depth of vocabulary knowledge: What the research shows. Language Learning, 64, 913–951. Schmitt, N., Ng, J., & Garras, J. (2011). The word associates format validation evidence. Language Testing, 28, 105–126. Schmitt, N., Schmitt, D., & Clapham, C. (2001). Developing and exploring the behaviour of two new versions of the vocabulary levels test. Language Testing 18, 55–88. Service, E. (1989). Phonological coding in working memory and foreign-language learning (Vol. B 9). University of Helsinki, General Psychology. Service, E. (1992). Phonology, working memory and foreign-language learning. Quarterly Journal of Experimental Psychology, 45A, 21–50. Service, E., & Kohonen, V. (1995). Is the relation between phonological memory and foreign-language learning accounted for by vocabulary acquisition? Applied Psycholinguistics, 16, 155–172. Service, E., Yli-Kaitala, H., Maury, S., & Kim, J.-Y. (2014). Adults’ and 8-Yearolds’ learning in a foreign word repetition task: Similar and different. Language Learning, 64(2), 215–246. doi:10.1111/lang.12051 Simard, D., & Molokopeeva, T. (2019, September). The mediating role of working memory in the relationship between reading comprehension and depth of vocabulary in L2 French. Communication présentée lors du Second Language Research Forum, Lansing, Michigan. Simard, D., Molokopeeva, T., & Zhang, Q. Y. (2020). Production d’autoreformulations autoamorcées par des apprenants adultes du français et capacité de mémoire de travail. Revue canadienne de linguistique appliquée/ Canadian Journal of Applied Linguistics. Snowling, M., Chiat, S., & Hulme, C. (1991). Words, nonwords, and phonological processes: Some comments on Gathercole, Willis, Emslie, and Baddeley. Applied Psycholinguistics, 12, 369–373.
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
L2 Vocabulary
St Clair-Thompson, H. L. (2010). Backwards digit recall: A measure of shortterm memory or working memory? European Journal of Cognitive Psychology, 22, 286–296 Tseng, W.-T., & Schmitt, N. (2008). Toward a model of motivated vocabulary learning: A structural equation modelling approach. Language Learning, 58, 357–400. Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task dependent? Journal of Memory and Language, 28, 127–154. Vafaee, P., & Suzuki, Y. (2020). The relative significance of syntactic knowledge and vocabulary knowledge in second language listening ability. Studies in Second Language Acquisition, 42, 383–410. Varol, B., & Erçetin, G. (2016). Effects of working memory and gloss type on L2 text comprehension and incidental vocabulary learning in computerbased reading. Procedia-Social and Behavioral Sciences, 232, 759–768. Vermeer, A. (2001). Breadth and depth of vocabulary in relation to L1/L2 acquisition and frequency of input. Applied Psycholinguistics, 22, 217–234. Wen, Z. E. (2016). Phonological and executive working memory in L2 taskbased speech planning and performance. The Language Learning Journal, 44, 418-435. Wen, Z. E., Borges Mota, M., & McNeill, A. (2015). Working memory in second language acquisition and processing. Multilingual Matters. Yamashita, J. (2013). Word recognition subcomponents and passage level reading in a foreign language. Reading in a Foreign Language, 25, 51–70. Yeari, M. (2017). The role of working memory in inference generation during reading comprehension: Retention, (re)activation, or suppression of verbal information? Learning and Individual Differences, 56, 1–12. Zaretsky, E. (2020). Verbal working memory and early literacy acquisition: Do ELLs allocate resources similar to their typical monolingual peers or monolingual children with SLI? International Journal of Bilingual Education and Bilingualism, 23, 1051–1070.
Note 1 Technically speaking, “nonword” and “pseudoword” should not be used interchangeably, as nonwords correspond to “unpronounceable letter strings consisting of orthographically implausible letter combinations,” whereas pseudowords are pronounceable letter strings consisting of orthographically plausible letter combinations (Yamashita, 2013, p. 60). In other words, only pseudowords follow the phonotactic rules of an existing natural language. This distinction is particularly important in the present discussion as only items created with pseudowords are likely to benefit from a semantic advantage derived from LTM representations.
https://doi.org/10.1017/9781108955638.030 Published online by Cambridge University Press
549
25 Working Memory and L2 Grammar Development in Children Paul Leseman and Josje Verhagen 25.1
Introduction
Previous research has shown that individual differences in children’s working memory abilities are correlated with their second language (L2) learning outcomes, for both vocabulary and grammar (Cheung 1996; Engel de Abreu & Gathercole, 2012; Kormos & Sáfár, 2008; Service & Kohonen, 1995; Verhagen & Leseman, 2016). While most earlier work has looked into the factors underlying the relationships between working memory and L2 vocabulary, especially regarding the involvement of long-term L2 knowledge (Cheung, 1996; Speciale et al., 2004), far less research has tried to explain the relationships between working memory and L2 grammar. In this chapter, we address this issue. Specifically, we ask why children with better developed working memory abilities generally acquire morphological and syntactic rules in an L2 more readily than children with lesswell-developed working memory abilities. In the attempt to answer this question, we explore to what degree several statistical learning frameworks that have been formulated within the field of first language (L1) acquisition can account for the role of working memory in L2 grammatical acquisition. Statistical learning frameworks assign a pivotal role to processes associated with working memory, such as chunking and storing of patterns, to explain the acquisition of linguistic structures by children learning novel or first language forms and structures, but have, to the best of our knowledge, not yet been widely applied to child L2 acquisition. The organization of this chapter is as follows. First, we describe current concepts of working memory, with particular reference to how working memory is assumed to relate to long-term memory knowledge. We then provide an overview of earlier research on relationships between working memory and L2 grammar. Subsequently, we discuss statistical learning accounts that have been proposed in the field of L1 acquisition, as well as earlier accounts of how working memory processes may be involved in
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
statistical learning. In the core part of our chapter, we address the extent to which these statistical learning accounts can be extended from L1 acquisition to L2 grammar learning, drawing from earlier working memory–based accounts of L2 learning. Finally, in our concluding remarks, we discuss to what extent working memory–based statistical learning accounts can account for the acquisition of L2 grammar, as well as any further steps to be taken to extend their scope from L1 to L2 acquisition.
25.2
Models of Working Memory
Working memory is generally defined as the capacity to hold a small amount of information in a temporary heightened state of activation to make this information available for further processing (Oberauer et al., 2018). One of the most-well-known models of working memory is the tripartite componential working memory model proposed by Baddeley and Hitch (1974), which contains modality-specific short-term memory systems for verbal and visuospatial (and kinesthetic) information, an episodic buffer for temporary multimodal integration and a central executive, which monitors and controls the information flows between the subsystems. After several revisions in the past decades (e.g., Baddeley, 2010), the current version of the model contains a central executive that has no storage capacity itself, but uses the limited capacity short-term memory systems for storage and the episodic buffer for constructing temporary integrated episodic representations. Long-term memory is not considered part of working memory in this model, but the central executive controls the storage and retrieval of processed information in and from long-term memory, and uses knowledge in long term memory for supporting working memory. Specifically, it refreshes or repairs the information in short-term memory to prevent forgetting, and it supports the chunking of information so as to increase the capacity of working memory. Other working memory models that have been proposed in the literature consider the actual processing of information as part of working memory, with limited capacity temporary storage systems being placed outside of working memory, while working memory proper is equated with limited capacity executive attention. In these models, mainly advocated by Engle and colleagues (Engle, 2002; Unsworth & Engle, 2007; Unsworth & Spillers, 2010), executive attention regulates processing, which is defined broadly, and includes reasoning and pattern matching. A larger role for long term memory has been proposed in models that define activated parts of long term memory as “long-term working memory” (Ericsson & Kintsch, 1995). In this chapter, we build on the generic model of working memory, as proposed by Cowan (2017). In this model, working memory is seen as a process embedded in long-term memory representations. Most of these representations are in a state of intermediate activation, and a much
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
551
552
PA U L L E S E M A N A N D J O S J E V E R H AG E N
smaller number is in a state of high activation, usually three or four separate information items, which may also be chunks (Cowan, 2010). In this generic model of working memory, the distinctions between shortterm memory, working memory, and long-term memory are not as strict as in the componential model of Baddeley and Hitch, but depend on how the memory system is involved in particular tasks. A limited capacity attention function plays a key executive role in allocating activation, that is, in activating, deactivating, and suppressing memory representations, to make them selectively available for processing (cf. Unsworth & Spillers, 2010). The model is specifically designed to account for the phenomenon that recently activated, but forgotten or suppressed, information is easily reactivated if demanded by the task. This is because this information is assumed to remain in an intermediate state of activation, and thus needs fewer attentional resources to be reactivated. As such, the model can account for the involvement of working memory in complex tasks such as language comprehension and reading, where understanding of a stretch of connected discourse or text requires relating the (propositional interpretation of the) last heard or read sentence to the (propositional interpretations of the) preceding sentences (Ericsson & Kintsch, 1995; Zwaan & Radvansky, 1998).
25.3
Working Memory and L2 Grammar Acquisition
In recent years, accumulating empirical evidence has shown that individual differences in working memory ability across children are correlated with differences in L2 acquisition, for both vocabulary (Cheung, 1996; Masoura & Gathercole, 2005; Messer et al., 2010) and grammar (French & O’Brien, 2008; Kormos & Sáfár, 2008; Service, 1992; Service & Kohonen, 1995; Verhagen et al., 2015; Verhagen & Leseman, 2016). In some of the studies on L2 grammar, relationships between working memory and L2 grammar disappeared once differences in L2 vocabulary were controlled. Service (1992) and Service and Kohonen (1995), for example, investigated the relationship between phonological storage and scores on several grammar tests in Finnish school-aged learners of English. In both studies, phonological storage, as assessed with nonword repetition, predicted children’s grammar test scores several years later. However, if vocabulary was included in the analysis, significant relationships were no longer found. Similarly, French (2006) found that effects of phonological storage on English grammar in French child learners of English were mediated by differences in L2 vocabulary knowledge. Not all studies found that relationships between working memory and L2 grammar were accounted for by L2 vocabulary, however. In a study by French and O’Brien (2008), for example, phonological storage, as assessed with nonword repetition tasks, was a significant predictor of 11-year-old
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
Francophone learners’ grammar skills in English, above and beyond English vocabulary. Verhagen et al. (2015) found that verbal memory, assessed through serial word and nonword recall, predicted the production of a number of grammatical structures in Dutch narratives, including subject-verb agreement and verb placement, in Turkish 4-year-old learners of Dutch. The relationship between children’s memory scores and grammatical production accuracy remained if differences in Dutch vocabulary were controlled. In another study, Verhagen and Leseman (2016) examined how the working memory components short-term storage and processing, as assessed with a set of different tasks, related to the knowledge of L2 vocabulary and grammar in Turkish child learners of Dutch and a monolingual Dutch comparison group. For grammar, both morphology and syntax were studied. The results showed that verbal short-term storage was significantly associated with vocabulary, while both short-term storage and processing were associated with grammar, for both syntax and morphology alike. There were no differences in the strengths of these relationships between the L2 and L1 learners, suggesting that the same working memory mechanisms are employed for learning vocabulary and grammar in L2 and L1 children. These results indicated, furthermore, that the processing component of working memory is uniquely needed for L2 grammar learning, but not vocabulary, if both vocabulary and grammar are considered simultaneously. Two points are noteworthy about these earlier studies. First, the children in both Verhagen et al. (2015) and Verhagen and Leseman (2016) were naturalistic L2 learners, who learned the L2 in an immersion setting with little to no explicit instruction. The fact that, in these studies with naturalistic learners, relationships between working memory and grammar remained even if vocabulary was controlled, whereas in studies looking at L2 classroom learners, vocabulary mediated the effects (French, 2006; Service, 1992; Service & Kohonen, 1995), suggests that relationships between working memory and L2 grammar may be specific to children who learn the L2 in uninstructed settings. However, in a study by Engel de Abreu and Gathercole (2012), who looked at trilingual Luxembourgian children learning both the L2 (German) and their L3 (French) at school, significant associations between measures of phonological storage and processing were found. This may indicate that, also for instructed L2 learners, working memory is implicated in the acquisition of L2 grammar. Second, an important note about the earlier studies on working memory and grammar in child L2 learners pertains to the design of these studies. Without exception, studies were correlational. Therefore, they leave unclear whether the relationships are causal. That is, these studies do not provide insight into the direction of the relationship between working memory and L2 grammar. However, evidence from a number of other studies on L2 children has indicated that the relationships between working memory and language learning may well be reversed such that increased
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
553
554
PA U L L E S E M A N A N D J O S J E V E R H AG E N
language knowledge supports working memory capacity. Evidence for this idea comes from frequently reported effects of word-likeness and phonotactic probability on tasks in which nonwords are repeated, showing that recall is superior for nonwords that are higher in wordlikeness and phonotactic probability than for nonwords lower in wordlikeness and phonotactic probability in L2 and L1 learners alike (Coady & Evans, 2008; Gathercole, 1995; Messer et al., 2010). For L2 learners, moreover, studies have shown that relationships between measures of working memory, in particular nonword repetition, and vocabulary are typically stronger within than across languages (Lee et al., 2013; Thorn & Gathercole, 1999). Finally, evidence for effects of long-term knowledge on working memory capacity comes from the finding that L2 children typically obtain the highest scores in working memory tasks based on the language they know best (Masoura & Gathercole, 1999; Messer et al., 2010). A study showing effects of long-term knowledge on working memory in child L2 learners was conducted by Messer and colleagues (2010). These authors investigated serial nonword recall, typically considered a measure of verbal storage, in monolingual Dutch children and sequentially bilingual Turkish 4-year-old learners of Dutch. They administered tasks with nonwords that were composed of phoneme combinations that were high-frequent (i.e., high-probability nonwords) or low-frequent (i.e., lowprobability nonwords) in either Dutch or Turkish. The results showed that the Dutch monolingual children obtained the highest scores for recall of Dutch-based nonwords of high-probability, whereas the Turkish-Dutch children obtained the highest scores for recall of Turkish-based high-probability nonwords. Recall of nonwords of low-probability in both Dutch and Turkish was equally low in the Dutch and Turkish-Dutch children. These findings demonstrate that existing language knowledge, in this case knowledge of the phonotactics of both languages, aided children’s ability to store the nonwords in working memory. The study also examined how the different types of recall were associated with the monolingual children’s Dutch vocabulary and the Turkish-Dutch children’s Dutch and Turkish vocabulary. The results revealed unique and significant contributions of both types of recall to the variance in vocabulary in both languages and in both groups. Thus, working memory, as assessed with tasks involving lowprobability nonwords, was significantly correlated with children’s vocabulary, even when differences in working memory, as assessed with highprobability nonwords, were controlled. In a study by Messer et al. (2015) the idea that acquired language knowledge facilitates working memory capacity was investigated further longitudinally. Specifically, this study investigated to what extent growth in children’s working memory that can be observed as they become older can be explained by growth in language knowledge. To test this, the findings from the 4-year-old children in Messer and colleagues (2010) were extended with assessments from the same children when they were 5 and
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
6 years old. Thus, children’s recall of (Dutch-based) high-probability nonwords and recall of (Dutch-based) low-probability nonwords was investigated through ages 4 to 6. Crucially, the authors hypothesized that if improvement in working memory with increasing age was due to growing language knowledge, there should be improvement of high-probability nonword recall (as language knowledge supports the recall of such nonwords and language knowledge typically increases during this period), but no or less improvement of low-probability nonword recall (since increasing language knowledge cannot be used to support the recall of these nonwords). This is exactly what was found: Growth modeling analysis of data from 72 monolingual Dutch children and 69 Turkish child learners of Dutch showed growth in children’s recall of high-probability nonwords from the ages 4 to 6, but no growth in recall of low-probability nonwords. These findings held for both groups. Thus, the results of this study indicate that verbal short-term memory growth can be explained by increases in long-term language knowledge (in this case phonotactic knowledge), due to increased exposure to language with age, indicating that not only is there an effect of working memory on language acquisition, as assumed in earlier work, but also vice versa, such that increases in acquired language knowledge result in better-developed working memory abilities.
25.4
Statistical Learning in (Monolingual First) Language Acquisition
Whereas earlier studies into the relationship between working memory and language learning in L2 children have mainly argued that children must be able to hold (phonological) information temporarily active in short-term memory to learn the words and rules in language, accounts formulated within the field of monolingual language acquisition have envisioned a more detailed picture of the complex interplay between perceiving, extracting, and integrating speech information to learn forms, structures, and meaning from language (e.g., Thiessen et al., 2013). Specifically, these studies have tried to uncover how young children learn language from the ambient input by picking up statistical regularities in this input, usually through experimental paradigms using novel language input or artificial languages. The ultimate aim of these studies is to understand the processes underlying the bidirectional relationships between working memory and language learning. Typically, studies have looked at very young children in the earliest phases of language learning, between zero and two years. In children this young, multiple distributional and frequency cues have been found to support (young) children’s detection of language structure (Thiessen et al., 2013), a process that has been referred to as statistical learning. Statistical learning thus refers to learning co-occurrence patterns
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
555
556
PA U L L E S E M A N A N D J O S J E V E R H AG E N
or distributions based on frequency information in the input. In a seminal study with young infants, Saffran and colleagues (1996) found that infants who were presented with a continuous speech stream of syllables with varying transitional probabilities picked up on these probabilities rather quickly, and were able to accurately distinguish between syllables with high versus low transitional probabilities in a subsequent test. These findings were taken as evidence that children can learn boundaries between combinations of syllables (i.e., “words”) from mere frequency information. Other infant studies demonstrated that young children are able to learn cooccurrence relations between elements, even if these are separated by intervening elements, in so-called nonadjacent dependency learning studies. Gómez and Maye (2005) showed, for example, that 15-month-old infants tracked the co-occurrence relationship between structures of the type a-X-b in which a and b were held constant, and X varied. Similar results have been obtained for older children and adults (de Bree et al., 2017; Misyak & Christiansen, 2012; Verhagen & de Bree, 2020). A central debate about the nature of these mechanisms underlying statistical learning centers around the following question: Does the brain compute probabilities or is statistical learning and, more generally, pattern detection, an emergent property of the interplay between perception and memory? The empirical and computational evidence to date favors the view according to which working memory processes of chunking, attentional biasing and prediction are involved in statistical learning (Dale et al., 2012; Hamrick, 2014; Isbilen et al., 2020; Sherman et al., 2020). According to the Extraction and Integration Framework formulated by Thiessen and colleagues (Erickson & Thiessen, 2015; Thiessen et al., 2013; Thiessen, 2017), for example, ‘category’ learning requires a memory system that (1) extracts, encodes and stores exemplars (e.g., particular temporal sequences of rising and falling pitch) from input based on conditional statistics, resulting in representations of exemplars which are initially episodic, idiosyncratic and noisy (i.e., also containing features that are less relevant), (2) matches new inputs to existing memory representations and activates the set of those representations that are most similar to the input (a mirroring mechanism) and suppresses or lets decay representations that match the input less well, (3) integrates the new input into the set of best matching representations while changing the internal association parameters of the set to reflect the features that are common across exemplars, after many inputs resulting in a prototype, and, (4) biases subsequent extraction processes to extract units that match the emerging category, thereby enhancing learning efficiency. This memory-based Extraction and Integration framework can explain frequency effects in statistical learning tasks, serial order effects in memory tasks, and non-adjacent dependency learning based on distributional characteristics of the input, that is, the learning of higher order distributional patterns that underlie syntactic knowledge (Erickson & Thiessen, 2015; Thiessen, 2017).
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
A basic assumption in the L1-based statistical learning studies discussed above is that statistical learning of language at different levels of complexity, from phonemes to syllables and words, from words to morphology and syntax, and to discourse, would require a huge amount of time and exposure if it were not biased by parallel processes to reduce uncertainty, involving attention and memory systems, and additional nonlinguistic information (Romberg & Safran, 2010; Thiessen et al., 2013). Therefore, a further assumption is that language learning in naturalistic situations receives additional support from the non-linguistic environment, in particular from the frequency distribution of co-occurrences of words (or phrases and sentences) and particular scenes with potential referents – referred to as cross-situational statistical learning (Romberg & Safran, 2010; Smith et al., 2014). These co-occurrences of words and scenes are relatively unambiguous for the young language learner, often carefully orchestrated and frequently repeated as routines in daily family life (Tamis-LeMonda et al., 2019; Weizman & Snow, 2001), and supported by prosody (Romberg & Safran, 2010) and pragmatic principles such as joint attention (Morales et al. 2000; Mundy & Gomes, 1998; Tomasello, 1988). As such, they form rather salient frequency-based chunks in the non-linguistic visuo-spatial array of objects, actions and events that regulate children’s attention to discard irrelevant information, activate relevant memory representations, and become rather consistently (in a statistical sense) associated with linguistic structures.
25.5
Relationships between Statistical Learning and Working Memory
Individual differences in statistical learning ability, the stability of these individual differences, the predictive value of statical learning for language learning, and the relationships between statistical learning and working memory or other domain-general cognitive functions (intelligence, executive function) have received relatively little attention until recently (Siegelman et al., 2017). Based on the extant research on adults, statistical learning is likely not a single domain-general capacity, but rather a set of statistical learning mechanisms that differ by modality (auditive vs. visual), statistical regularity (transitional vs. distributional), contingency (adjacent vs. non-adjacent), and material (verbal vs. non-verbal), that in the complex process of language acquisition may work in parallel (cf. Kidd, 2012). This suggests a multi-componential capacity (Arciuli, 2017; Siegelman et al., 2017) – a conclusion that is supported by the fact that, although separate measures of statistical learning have good test-retest stability, correlations between different measures are generally modest (McCauley et al., 2017; Siegelman & Frost, 2015). Thus, a likely explanation for these modest correlations across statistical learning tasks is that statistical learning
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
557
558
PA U L L E S E M A N A N D J O S J E V E R H AG E N
involves modality-specific perception, regularity-specific extraction, and modality-specific memory storage, updating and integration (Thiessen et al., 2013). All these subprocesses may require attentional resources, in line with the generic working memory model (Palmer & Mattys, 2016), thus sharing variance caused by individual differences in executive attention ability (Unsworth & Spillers, 2010). However, if the four subprocesses of statistical learning are indeed multiplicatively related, as proposed by Siegelman et al. (2017), it is not surprising that intercorrelations between different statistical learning measures that draw upon different modalities, statistical regularities, contingencies and materials, as well as the associations between statistical learning measures and domain-general working memory measures are moderate at best. Regarding the relationships between statistical learning and working memory, previous findings have been inconclusive. Siegelman and Frost (2015) examined lower level statistical learning (e.g., word and pattern segmentation) in adults in different modalities with different types of statistical regularities and contingencies. The authors found that different measures of statistical learning were only weakly intercorrelated (despite fair to good test-retest stability), and not related to working memory and other general cognitive functions. Misyak and Christiansen (2012) examined how verbal and visual adjacent and non-adjacent statistical learning were related to verbal short-term memory (forward digit span), verbal working memory (reading span) and several language comprehension measures. Their results showed that verbal working memory correlated moderately strongly (r’s in the .40 to .53 range) with both adjacent and nonadjacent statistical learning, while verbal short-term memory significantly correlated with adjacent statistical learning only. Thus, the relationships between measures of statistical learning of language and working memory assessments may differ between low and high level statistical learning, and only the latter may be associated with working memory, possibly due to higher information load and the need to control attention (Palmer & Mattys, 2016). Interestingly, moreover, when regressing the memory and statistical learning scores on tasks of language knowledge and language processing, the authors found unique effects of both adjacent and nonadjacent statistical learning but no direct effects of the verbal short-term and working memory tasks. For children, only very few studies have looked at the interrelationships between measures of working memory and statistical learning. Typically, these studies examined the correlations between nonword repetition as a measure of verbal short-term memory and non-adjacent dependency learning in artificial language learning experiments. The results revealed, just as for adults (Misyak & Christiansen, 2012), moderate correlations between verbal short-term memory and (higher level) non-adjacent dependency learning (de Bree et al., 2017; Verhagen & de Bree, 2020).
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
25.6
Chunking through Language Learning
Working memory capacity is limited to three or four items, according to the generic working memory model (Cowan, 2017; Oberauer et al., 2018), but can be substantially expanded by creating larger chunks that function as single units in working memory. Chunks can be defined as groups of items with stronger intragroup than intergroup associations (Brady et al., 2009; Cowan, 2010). Language knowledge in the form of chunked information in long-term memory supports short term memory and the creation of new, larger chunks. For example, recalling a string like USAEUNATO is relatively easy compared to a random sequence of the same letters, because long term knowledge helps to recognize the chunks USA, EU, and NATO. In language acquisition, statistical learning from input and the chunking of information in working memory based on distributional cues (which supports subsequent recognizing of more complex patterns in the input) work in parallel and interactively. Evidence for this comes from a study by Isbilen and colleagues (2020), who trained subjects in an artificial language. In this language, consonant-vowel syllables formed three-syllable nonwords with high within-word and low between-word transitional probabilities, similar to the stimuli used by Saffran et al. (1996). In a serial nonword recall task with more complex six-syllable items, half consisting of combinations of trained three-syllable nonwords with high internal transition probability and the other half of six-syllable control items with the syllables in varying, pseudorandomized order, recall was superior for the nonwords constructed out of syllables with high internal transitional probabilities compared to the nonwords constructed out of equally trained syllables but with low internal transitional probabilities. Similar results were found when natural language (English) syllable statistics were used to create input strings and recall items, revealing an advantage in serial recall for items made of high-probability words based on the frequency of the constituting syllables in natural language, as in the studies on L2-learning children by Messer and colleagues (2010, 2015) described above. These results indicate that statistical learning based on transitional probabilities between phonemes or syllables results in long-term memory representations of chunks (biphone and triphone units, syllables, words, phrases) that, in turn, influence working memory capacity in serial recall of larger units. Because the items were based on natural language, the memory tasks tapped into the subjects’ sensitivity to statistical cues in the natural language input they received, thus reflecting individuals’ statistical learning abilities in naturalistic language learning situations. In a study by McCauley and colleagues (2017), participants’ abilities to chunk the input were related to the processing of complex sentences. Two types of chunking ability were investigated: (a) phonological chunking ability (defined as the difference in repetition accuracy of nonwords
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
559
560
PA U L L E S E M A N A N D J O S J E V E R H AG E N
constructed of high vs. low chunklike syllables, as based on natural language statistics) and (b) multiword chunking ability (defined as the difference in recall accuracy of twelve-word strings composed of four subsets of three high vs. low chunklike words, as based on natural language statistics). Both nonword repetition and multiword recall were superior for high chunklike items compared to low chunklike control items, however, showing considerable individual differences. Furthermore, both types of chunking ability were related to the reading of (a) sentences with embedded object-relative clauses and distractions based on phonological similarity of the subject and object and of the two verbs (e.g., “The cook that the crook consoles controls the policemen”), and (b) sentences with long-distance subject-verb number agreement with inserted distracting number-marked nouns (e.g., “The key to the cabinets was rusty from many years of disuse”), but in different ways: subjects with higher phonological chunking ability showed less difficulty processing sentences of type (a), while subjects with higher multiword chunking ability showed less difficulty processing sentences of type (b). Interestingly, the two types of chunking ability based on statistical regularities in natural language at different levels were not correlated.
25.7
Construction Grammar and L2 Learning
Chunking in, and through, language learning is a key process according to the usage-based account of grammar learning, also referred to as the “construction grammar” approach (e.g., Ellis, 2002; Kidd, 2012; Tomasello, 2003). This account of language acquisition focuses on the learning and gradual abstraction of “constructions,” starting with the smallest meaning-carrying constructions in languages, such as morphemes and simple words, to more complex phrases and abstract syntactic frames (Tomasello, 2003; Wulff & Ellis, 2018). Constructions are regarded as formmeaning mappings, where meaning can be referential, but also functional (as in the passive, which shifts attentional focus to the recipient of an action), and concrete or abstract. The acquisition process is frequencydriven and assumed to involve general mechanisms of cross-situational statistical learning to map forms to meanings (Kidd, 2012). The most frequent constructions in language use are acquired early and become most entrenched, while very infrequent constructions may never be acquired by some language learners, as also suggested by studies on L2 learners who, despite apparent nativelike proficiency, do not show full nativelike mastery upon close scrutiny (Abrahamsson & Hyltenstam, 2009; Ioup et al., 1994). The idea of chunking in, and through, language has been applied to L2 learning as well. Specifically, researchers have proposed that L2 learners initially store chunked stretches of speech in memory that are analyzed only later on, leading to increasing L2 proficiency (Martin & Ellis, 2012;
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
Speidel, 1993; Speidel & Herreshoff, 1989). Speidel (1993), for example, argued that L2 learners store grammatical constructions in verbal shortterm memory, in the same way they store words, that is, as lexical items. In so doing, they build a “storehouse” of constructions in long-term memory from which they can gradually extract patterns to support their spontaneous L2 speech. Similarly, Ellis (1996) argued that most of L2 learning is in fact item learning at different levels of the language system, including grammar (which he termed “sequence learning”). Ellis proposed that L2 learners initially store sequences in a “database” that they later on use to abstract regularities (i.e., grammatical knowledge) from. When storing sequences, chunking plays a major role and the way in which chunks are formed is dependent on long-term knowledge. Specifically, Ellis assumed that the L2 learner first accumulates a sufficient mass of L2 phrases (“sequences”) and then uses the same statistical processes of abstraction that the L1 learner uses to discover or “construct” grammatical rules from this collection of sequences (cf. Tomasello, 2003; Ullman, 2001). As an example, he refers to the well-known acquisitional stages that L2 learners of diverse L1 backgrounds go through when acquiring negation in English: no/not + X (“no happy”), before no/not/don’t + V (“they not work”; “he don’t go”), before analyzed don’t (“she doesn’t sleep”) (e.g., Schumann, 1978). The first two sequences can be seen as co-occurrence patterns that are prevalent in the input and that L2 learners would store as ‘sequences’ (“I am not happy,” “There is no water anymore,” ”He does not work,” et cetera). Ellis did not go into detail as to which factors underlie the storage and chunking of L2 sequences. Mere frequency cannot be the only source of information that learners use: some constructions in languages are highly frequent, such as articles, verbal suffixes, the placement of negation and adverbs – yet many L2 learners are known to struggle with these constructions also after extensive exposure. Other factors that may facilitate storage and chunking are phonological salience, the extent to which a form carries a clear meaning, prototypicality, and redundancy (Ellis & Collins, 2009). Below we discuss a few of these factors in more detail, pointing to the combined effects of competition of L1 and L2 knowledge, cue salience, and reliability in the input, and the cumulative amount of input.
25.8
Second Language Learning and Working Memory
L2 acquisition (here broadly defined to include successive bilingualism and foreign language learning) differs from L1 acquisition in a number of ways (MacWhinney, 2005, 2012). First, while L1 learning children learn language and learn about the world at the same time, L2 learners usually already have knowledge about the world at the onset of L2 acquisition. Second, L2 learners already have acquired an L1 at least to some degree, which influences the way they process, comprehend, and produce L2 words and
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
561
562
PA U L L E S E M A N A N D J O S J E V E R H AG E N
sentences. Thus, L1 learners are learning with a still malleable brain that has not yet become specialized and modularized, whereas L2 learners, depending on their age, approach the language learning task with a brain much more dedicated to specific types of input and processes. Finally, a difference between L1 and L2 learners is that L1 learners are usually strongly supported by their caregivers in the family context, whereas L2 learners often have to pick up the L2 from less frequent, more distributed, and less well-organized language input (Jia & Aaronson, 2003; Leseman et al., 2019; Paradis, 2007; Place & Hoff, 2011; Scheele et al., 2010; Unsworth, 2014). Despite such differences, observations of child and adult L2 learning suggest that L1 and L2 are tightly interwoven in L2 learning, and that many mechanisms are similar (MacWhinney, 2012; Verhagen & Leseman, 2016). Both L1 and L2 learners need to segment speech into syllables, words, and phrases based on low-level statistical learning from input, learn word and sentence meanings by connecting form and meaning in cross-situational statistical learning, and figure out the patterns that govern word combinations in syntactic constructions based on high-level statistical learning, as proposed by the construction grammar approach. Moreover, within L2 learners, there is transfer and interference from L1 to L2, and vice versa. All this calls for a unified theory. Transfer and interference point to competition between L1 and L2 on different levels (e.g., phonology, morphosyntax, lexicon, syntax) and the outcome of this competition may be, at least in part, determined by statistical cue strength in the input (e.g., frequencies of particular word orders, subject-verb agreement cues, case marking, the role of agency; MacWhinney, 2012). This competition can be understood in terms of competing “‘resonances” in working memory of chunked L1 and L2 representations of both low and high-level constructions, as we will detail below. The Unified Competition Model developed by MacWhinney and colleagues (MacWhinney, 2005, 2012; Li & MacWhinney, 2013) assumes a central role for working memory in L2 learning. Specifically, in order to understand form-meaning mappings and L2-L1 mappings, items from both languages need to be in a temporarily activated state. Given that the capacity of working memory is limited, this requires chunking of constructions (Ellis, 2002; Wulf & Ellis, 2018). In online interactive situations such as in naturalistic L2 settings, attentional load may be heavy, as is evidenced by studies showing that even fully competent bilinguals tend to process sentences more slowly than monolinguals. Attentional load is also dependent on the structural characteristics of the L2. For example, in languages with a predominant subject-object-verb sentence structure (German), the processing load increases substantially when several elements are inserted between the subject and the verb. Long-term knowledge of syntactic frames (which can be L1 knowledge; see below), based on chunked exemplars encountered in language use, is needed to support working memory and
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
alleviate attentional load (Ellis, 2002). Chunking explains growing fluency in L2 (Hulstijn, 2002) and is particularly important for grammar learning, for instance, when acquiring complex inflectional morphology (Ellis, 2003). Chunking is also important at the sublexical level in beginning L2 learners, and helps them to identify the phonological composition of words and phrases, which sets the stage for learning more complex constructions (Gupta & MacWhinney, 1997; Hulstijn, 2002). Working memory – or, more specifically, the interplay between chunked information in long-term memory, attentional mechanisms, and temporarily activated items – is also involved in cross-language influences from the often stronger entrenched L1 on the still largely to be learned L2. Initially, L2 learners rely strongly on their L1 knowledge to process L2 input at the phonological, lexical, semantic, and syntactic level. The working memory mechanism is that of resonance, which is a bidirectional mechanism: cues in the input trigger activation (or “resonance”) of the best matching and most entrenched chunks in long term memory at all these levels of language processing, which initially are most likely L1 chunks in L2 learners. This, in turn, biases the attention to and perception of the input cues in a predictive way, in line with the general extraction-integration model described above (Sherman et al., 2020; Thiessen & Erickson, 2013; Thiessen et al., 2013). For example, if the L1 is well-entrenched, as in late L2 learners, the phonological form of an L2 word is likely to be perceived and represented as if it was an L1 word. Conceptual knowledge in L1 is also transferred to L2 to understand word and sentence meanings in L1, and L2 input is initially understood via L1 (called “parasitic use of L2”; MacWhinney, 2012). This is effective to the extent that words in L1 and L2 map to concepts in a highly similar way, but interferes when particular concepts in L1 and L2 do not match. Only when sufficient proficiency in L2 has been built up, thus after sufficient exposure, is L2 connected directly to the conceptual base. With respect to grammar, long-term knowledge of the L1 initially strongly biases the perception of and attention to cues for sentence processing such as word order, inflections, grammatical morphemes, definite articles, and noun animacy, yielding what has been called a “syntactic accent.” This bias can cause difficulties in language comprehension and production, and thus in learning from input. The influence of L1 only gradually disappears but may never be fully absent even in fully competent bilinguals. Strong entrenchment of L1, in this regard, presents the greatest “risk factor” for L2 learners (MacWhinney, 2012) – in addition, of course, to less exposure to L2 as a consequence of the often later age at which L2 learning starts and the inevitable need to distribute exposure time over two or more languages in multilingual situations, as was discussed above. There is evidence from L2 acquisition research supporting these claims. In L2 acquisition, L2 learners have been observed to prefer adverbial tense marking over verb inflection (Klein & Perdue, 1997; Verhagen, 2009). In
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
563
564
PA U L L E S E M A N A N D J O S J E V E R H AG E N
general, L2 learners show more difficulty with acquiring morphemes and closed-class grammatical constructions than with open-class constructions (vocabulary, lexicalized phrases). A number of principles can account for this, pointing to perceptual and statistical learning mechanisms involving working memory: cue availability, saliency, and reliability, and (in L1) learned attentional biases based on well-entrenched chunked constructions. Cue availability is determined by the frequency of the cue in language use. Also, the perceptual salience, or detectability, of cues in the language input matters. Less frequent or perceptually less salient linguistic cues are more difficult to pick-up by statistical learning mechanisms. Grammatical morphemes, although occurring frequently, are usually not stressed and often do not match with L1 constructions, so that there is no long-term knowledge that can support perception and chunking of these cues. Thus, in the latter case there is hardly positive transfer from L1 to L2, although this may differ between pairs of languages. In addition, there can be interference if a similar cue has a different function in the two languages (e.g., the determiner-number cues in English vs. Spanish; Li & MacWhinney, 2013). Cue reliability, or the degree of form-meaning contingency, represents the proportion of times a particular cue gives the correct interpretation of all uses of this cue. Cue reliability can be high for some constructions in a particular language, but low for other constructions. Cues can be frequent in a language, but not reliable because of a weak contingency between the cue and its meaning. For example, the highly frequent -s in English denotes the plural form but can also denote third person present tense and the possessive relation (Ellis, 2006). Learning casemarking in Russian is easier than in German, despite the fact that the Russian system is more complex, because Russian case-marking cues are more reliable for sentence interpretation (Kempe & MacWhinney, 1998). Less frequent, less salient, and less reliable cues, and cues that do not match L1 constructions or even interfere with L1 constructions, require substantially more exposure to be learned statistically, which may not be feasible in multilingual situations.
25.9
Conclusion
Earlier L2 research has shown that working memory is implicated in the acquisition of L2 grammar. However, within these studies on the relationships between working memory and L2 grammar learning, no detailed attempts have yet been made to explain how working memory may be involved in L2 grammar acquisition. In this chapter, we reviewed working memory-based statistical learning accounts from the language acquisition literature. We also reviewed a number of theoretically related accounts in the L2 literature that assume a pivotal role for processes associated with working memory (e.g., chunking) in L2 grammar learning.
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
Based on our discussion of these two strands of research, we propose that the relationships between working memory and L2 grammar learning are complex and most likely bidirectional, just like the relationships between working memory and L1 learning (Messer et al., 2010, 2015; Verhagen et al., 2019). Specifically, we propose that working memory enables children to learn constructions at increasing levels of complexity from the surrounding input involving the creation of chunks in long-term memory, while increases in this long-term language knowledge, in turn, enhance working memory capacity, bias attention, and perception, and thereby accelerate further language learning. Departing from the assumptions made in the generic model of working memory (Cowan, 2017), our chapter has attempted to describe how dynamic and interactive working memory processes involved in statistical language learning that have been assumed for monolingual language acquisition may provide a window on the processes involved in L2 grammar learning, at least in naturalistic situations. Importantly, L2 learners face a number of challenges that L1 learners do not face. Specifically, they are not only likely to encounter difficulties due to the competition between L1 and L2, but also they typically receive less input and less clear language input than L1 learners. These differences likely make language learning for L2 learners more challenging at all levels of the language system, starting with the ability to perceive and chunk phonemes and phoneme clusters in order to segment the speech stream as a prerequisite for word, phrase, construction, and syntax learning. Working memory enables learners to allocate their attentional resources to perceiving and chunking of the input and to memory representations of L1 and L2, and, as such, modulates the competition between L1 and L2, for instance by deactivating or suppressing L1 activation. A speculative idea to be explored in future research is that individual differences in these attentional resources explain individual differences in L2 learning that are still found even if all other factors that are at stake in L2 learning are kept equal. The ideas we reviewed and proposed in this chapter were specific to L2 learning in naturalistic situations. However, L2 learning may also take place in classroom settings, where it is supported through modified interactions and instructional materials in several ways. Specifically, interfering biases stemming from L1 may be deliberately addressed as part of metalinguistic instruction. Cues in the input that are not salient may be made salient by explicit instruction and corrective feedback, to focus the learners’ attention on these cues (and foster “noticing”; cf. Schmidt, 1990). Cues that are infrequent can be deliberately repeated to support the learning process. Negative evidence for certain constructions, though infrequent in natural language, may be brought to the attention of the L2 learner, too (TreffersDaller & Calude, 2015). In so doing, learning environments can be optimized to foster working memory-based processes at the level of perceiving,
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
565
566
PA U L L E S E M A N A N D J O S J E V E R H AG E N
extracting, and integrating information, as well as the interplay between these processes, which are needed to learn forms and structures of the L2 and overcome L1 interference. Our approach in this chapter is compatible with an influential model of language learning as based on two types of domain-general memory and learning mechanisms with distinct neurobiological substrates (Ullman, 2001): declarative and procedural learning (hence, the DP model), the first associated with lexical knowledge and the second with grammatical knowledge. Declarative learning in the DP model relates to what we have called item-based or exemplar learning, which characterizes L1 acquisition at different levels, from phonology to syntax, in young children acquiring their L1 and initial L2 learning until, by chunking, a sufficiently large storehouse is built up for processes of abstraction of categories and rules. Procedural learning in the DP model relates to these processes of abstraction and rule generation. A recent meta-analysis by Hamrick et al. (2018) provides support for the DP model, revealing associations between nonverbal behavioral and brain measures of declarative learning with lexical knowledge and also grammatical knowledge in both 5- to 10-year-old L1 learners and in beginning adult L2 learners, and between nonverbal measures of procedural learning with grammatical knowledge in experienced adult L2 learners (Hamrick et al., 2018). The double association of declarative learning with both lexical and grammatical knowledge in relatively inexperienced language learners (children, beginning adult L2 learners) may reflect, according to the authors, that declarative lexical knowledge gradually, through chunking mechanisms, feeds into procedural learning, as proposed in the construction grammar account (see also Hamrick, 2014). In line with this, we detailed in this chapter how working memory and statistical structures of the language input interact – through extraction of exemplars, chunking, and integration over chunked exemplars in statistical learning – to generate grammatical knowledge, both in L1 and L2. (Note that no studies were found for the meta-analysis of Hamrick et al. [2018] with older L1 learners to test whether, in older children, grammatical knowledge would also be uniquely associated with procedural learning as in experienced adult L2 learners.) Human working memory as a domain-general limited capacity resource, characteristics of grammars, and how language use in social contexts is governed by these grammars can be seen as constituting together a complex system in which basic properties of each of the three subsystems codetermine the (phylogenetic, cultural-historic, and ontogenetic) development and operation of the other subsystems. Indeed, as is discussed in the chapters by Hawkins and O’Grady in this volume, the grammars of many languages, if not all, reflect the capacity constraints of human working memory and generate in actual language use different types of statistical regularities to be picked up by general-purpose working memory, while prosodic and pragmatic features of language use in social contexts (e.g., the
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
use of stress, pauses, pointing, gesturing, repeated language routines, crosssituational form-meaning mapping) can be seen as devices to support language learning through working memory-based extraction and integration of statistical information in the input (Thiessen et al., 2013). Indeed, the stimulus is not poor at all and well-attuned to the human language learner.
References Abrahamsson, N., & Hyltenstam, K. (2009). Age of onset and nativelikeness in a second language: Listener perception versus linguistic scrutiny. Language Learning, 59, 249–306. Arciuli, J. (2017). The multi-component nature of statistical learning. Philosophical Transactions of the Royal Society B, 372 (20160058). Baddeley, A. (2010). Working memory. Current Biology, 20, R136–R140. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (pp. 47–89). Academic Press. Brady, T. F., Konkle, T., & Alvarez, G. A. (2009). Compression in visual working memory: Using statistical regularities to form more efficient memory representations. Journal of Experimental Psychology General, 138, 487–502. Cheung, H. (1996). Nonword span as a unique predictor of second-language vocabulary learning. Developmental Psychology, 32, 867–873. Coady, J. A., & Evans, J. L. (2008). Uses and interpretations of non-word repetition tasks in children with and without specific language impairment (SLI). International Journal of Communication Disorders, 43, 1–40. Cowan, N. (1988). Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human information processing system. Psychological Bulletin, 104, 163–191. Cowan, N. (2010). The magical mystery four: How is working memory capacity limited, and why? Current Directions in Psychological Science, 19, 51–57. Cowan, N. (2017). The many faces of working memory and short-term storage. Psychological Bulletin Review, 24, 1158–1170. Dale, R., Duran, N. D., & Morehead, J. R. (2012). Prediction during statistical learning, and implications for the implicit/explicit divide. Advances in Cognitive Psychology, 8, 196–209. De Bree, E. H., Verhagen, J., Kerkhoff, A. O., Doedens, W. J., & Unsworth, S. (2017). Language learning from inconsistent input: Bilingual and monolingual toddlers compared. Infant and Child Development, 26, e1996. Ellis, N. C. (1996). Sequencing in SLA: Phonological memory, chunking, and points of order. Studies in Second Language Acquisition, 18, 91–126 Ellis, N. C. (2002). Frequency effects in language processing and acquisition: A review with implications for theories of implicit and explicit language acquisition. Studies in Second Language Acquisition, 24, 143–188.
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
567
568
PA U L L E S E M A N A N D J O S J E V E R H AG E N
Ellis, N. C. (2003). Constructions, chunking, and connectionism: The emergence of second language structure. In C. J. Doughty & M. H. Long (Eds.), Handbook of second language acquisition. Blackwell. Ellis, N. C. (2006). Language acquisition as rational contingency learning. Applied Linguistics, 27, 1–24. Ellis, N. C., & Collins, L. (2009). Input and second language acquisition: The roles of frequency, form, and function: Introduction to special issue. The Modern Language Journal, 93, 329–335. Engel de Abreu, P. M. J., & Gathercole, S. E. (2012). Executive and phonological processes in second-language acquisition. Journal of Educational Psychology, 104, 974–986. Engle, R. W. (2002). Working memory capacity as executive attention. Current Directions in Psychological Science, 11, 19–23. Erickson, L. C., & Thiessen, E. D. (2015). Statistical learning of language: Theory, validity, and predictions of a statistical learning account of language acquisition. Developmental Review, 37, 66–108. Ericsson, K. A., & Kintsch, W. (1995). Long term working memory. Psychological Review, 102, 211–245. French, L. M. (2006). Phonological working memory and second language acquisition: A developmental study of Francophone children learning English in Quebec. Edwin Mellen Press. French, L. M., & O’Brien, I. (2008). Phonological memory and children’s second language grammar learning. Applied Psycholinguistics, 29, 463–487. Gathercole, S. E. (1995). Is nonword repetition a test of phonological memory or long-term knowledge? It all depends on the nonwords. Memory and Cognition, 23, 83–94. Gómez, R., & Maye, J. (2005). The developmental trajectory of nonadjacent dependency learning. Infancy, 7, 183–206. Gupta, P., & MacWhinney, B. (1997). Vocabulary acquisition and verbal short-term memory: Computational and neural bases. Brain and Language, 59, 267–333. Hamrick, P. (2014). A role for chunk formation in statistical learning of second language syntax. Language Learning, 64, 247–278. Hamrick, P., Lum, J. G., & Ullman, M. T. (2018). Child first language and adult second language are both tied to general-purpose learning systems. Proceedings of the National Academy of Sciences of the United States of America, 115(7), 1487–1492. Hulstijn, J. H.. (2002). Towards a unified account of the representation, processing and acquisition of second language knowledge. Second Language Research, 18, 193–223. Ioup, G., Boustagui, E., El Tigi, M., & Moselle, M. (1994). A case study of successful adult SLA in a naturalistic environment. Studies in Second Language Acquisition, 16, 73–98.
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
Isbilen, E. S., McCauley, S. M., Kidd, E., & Christiansen, M. H. (2020). Statistically induced chunking recall: A memory-based approach to statistical learning. Cognitive Science, 44, e12848. Jia, G., & Aaronson, D. (2003). A longitudinal study of Chinese children and adolescents learning English in the United States. Applied Psycholinguistics, 24, 131–161. Kempe, V., & MacWhinney, B. (1998). The acquisition of case-marking by adult learners of Russian and German. Studies in Second Language. Acquisition, 20, 543–587. Kidd, E. (2012). Implicit statistical learning is directly associated with the acquisition of syntax. Developmental Psychology, 48(1), 171–184. Klein, W., & Perdue, C. (1997). The basic variety (or: couldn’t natural languages be much simpler?). Second Language Research, 13, 301–347. Kormos, J., & Sáfár, A. (2008). Phonological short-term memory and foreign language performance in intensive language learning. Bilingualism: Language and Cognition, 11, 261–271. Lee, H. J., Kin, Y. T., & Yim, D. (2013). Non-word repetition in KoreanEnglish children. International Journal of Speech-Language Pathology, 15(4), 375–382. Leseman, P. P. M., Henrichs, L. F., Blom, E. & Verhagen, J. (2019). Young mono- and bilingual children’s exposure to academic language as related to language development and school achievement. In V. Grøver, P. Ucelli, M. Rowe & E. Lieven (Eds.), Learning through language (pp. 205–217). Cambridge University Press. Li, P., & MacWhinney, B. (2013). Competition Model. In C. A. Chapelle (Ed.), The encyclopedia of applied linguistics (pp. 3–5). Blackwell Publishing. MacWhinney, B. (2005). A unified model of language acquisition. In J. F. Kroll & A. M. de Groot (Eds.), Handbook of bilingualism: Psycholinguistic approaches (pp. 49–67). Oxford University Press. MacWhinney, B. (2012). The logic of the unified model. In S. Gass & A. Mackey (Eds.), The Routledge handbook of second language acquisition (pp. 211–227). Routledge. Martin, K. I., & Ellis, N. C. (2012). The roles of phonological short-term memory and working memory in L2 grammar and vocabulary learning. Studies in Second Language Acquisition, 34, 379–413. Masoura, E. V., & Gathercole, S. E. (1999). Phonological short-term memory and foreign language learning. International Journal of Psychology, 34, 383–388. Masoura, E. V., & Gathercole, S. E. (2005). Phonological short-term memory skills and new word learning in young Greek children. Memory, 13, 422–429. McCauley, S. M., Isbilen, E. S., & Christiansen, M. H. (2017). Chunking ability shapes sentence processing at multiple levels of abstraction. Proceedings of the 39th Annual Conference of the Cognitive Science Society (pp. 2681–2686).
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
569
570
PA U L L E S E M A N A N D J O S J E V E R H AG E N
Messer, M. H., Leseman, P. P. M., Boom, J., & Mayo, A. Y. (2010). Phonotactic probability effect in nonword recall and its relationship with vocabulary in monolingual and bilingual preschoolers. Journal of Experimental Child Psychology, 105, 306–323 Messer, M. H., Verhagen, J., Boom, J., Mayo, A. Y., & Leseman, P. P. M. (2015). Growth of verbal short-term memory of nonwords varying in phonotactic probability: A longitudinal study with monolingual and bilingual children. Journal of Memory and Language, 84, 24–36. Misyak, J. B., & Christiansen, M. H. (2012). Statistical learning and language: An individual differences study. Language Learning, 62, 302–331. Morales, M., Mundy, P., Delgado, C. E. F., Yale, M., Messinger, D., Neal, R., & Schwartz, H. K. (2000). Responding to joint attention across the 6- to 24month age period and early language acquisition. Journal of Applied Developmental Psychology, 21, 283–298. Mundy, P., & Gomes, A. (1998). Individual differences in joint attention skill development in the second year. Infant Behavior and Development, 21, 469–482. Oberauer, K., Lewandowsky, S., Awh, E., Brown, G. D. A., Conway, A., Cowan, N., Donkin, C., Farrell, S. A., Hitch, G. J., Hurlstone, M. J., Ma, W. J., Morey, C. C., Nee, D. E., Schweppe, J., Vergauwe, E., & Ward, G. (2018). Benchmarks for models of short-term and working memory. Psychological Bulletin, 144), 885–958. Palmer, S. D., & Mattys, S. L. (2016). Speech segmentation by statistical learning is supported by domain-general processes within working memory. The Quarterly Journal of Experimental Psychology, 69, 2390–2401. Paradis, J. (2007). Second language acquisition in childhood. In E. Hoff & M. Shatz (Eds.), Blackwell handbooks of developmental psychology: Blackwell handbook of language development (pp. 387–405). Blackwell Publishing. Place, S., & Hoff, E. (2011). Properties of dual language exposure that influence 2-year-olds’ bilingual proficiency. Child Development, 82(6), 1834–1849. Romberg, A. S. R., & Saffran, J. R. (2010). Statistical learning and language acquisition. WIREs Cognitive Science, 1, 906–914. Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8month-old infants. Science, 274, 1926–1928 Scheele, A. F., Leseman, P. P. M., & Mayo, A. Y. (2010). The home language environment of mono- and bilingual children and their language proficiency. Applied Psycholinguistics, 31, 117–140. Schmidt, R. (1990). The role of consciousness in second language learning. Applied Linguistics, 11, 129–158. Schumann, J. H. (1978). The acculturation model for second-language acquisition. In R.C. Gingras (Ed.), Second language acquisition and foreign language teaching. Center for Applied Linguistics. Scott, R. M., & Fisher, C. (2012). 2.5-year-olds use cross-situational consistency to learn verbs under referential uncertainty. Cognition, 122, 163–180.
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
L2 Grammar Development in Children
Service, E. (1992). Phonology, working memory, and foreign language learning. Quarterly Journal of Experimental Psychology, 45A, 21–50. Service, E., & Kohonen, V. (1995). Is the relation between phonological memory and foreign language learning accounted for by vocabulary acquisition? Applied Psycholinguistics, 16, 155–172. Sherman, B. E., Graves, K. N., & Turke-Brown, N. B. (2020). The prevalence and importance of statistical learning in human cognition and behavior. Current Opinion in Behavioral Sciences, 32, 15–20. Siegelman, N., & Frost, R. (2015). Statistical learning as an ability: Theoretical perspectives and empirical evidence. Journal of Memory and Language, 81, 105–120. Siegelman, N., Bogaerts, L., Christiansen, M. H., & Frost, R. (2017). Towards a theory of individual differences in statistical learning. Philosophical Transactions of the Royal Society B, 372(20160059). Smith, L. B., Suanda, S. H., & Yu, C. (2014). The unrealized promise of infant statistical word–referent learning. Trends in Cognitive Sciences, 18( 5), 251–258. Speciale, G., Ellis, N., & Bywater, T. (2004). Phonological sequence learning and short term store capacity determine second language vocabulary acquisition. Applied Psycholinguistics, 25, 293–320. Speidel, G. E. (1993). Phonological short-term memory and individual differences in learning to speak: A bilingual case study. First Language, 13, 69–91. Speidel, G. E., & Herreshoff, M. J. (1989). Imitation and the construction of long utterances. In G. E. Speidel & K. E. Nelson (Eds.), The many faces of imitation in language learning (pp. 181–197). Springer-Verlag. Tamis-LeMonda, C. S., Custode, S., Kuchirko, Y., Escobar, K., Lo, T. (2019). Routine language: Speech directed to infants during home activities. Child Development, 90(6), 2135–2152. Thiessen, E. D. (2017). What’s statistical about learning? Insights from modelling statistical learning as a set of memory processes. Philosophical transactions of the Royal Society B, 372, 1711. Thiessen, E. D., & Erickson, L. C. (2013). Beyond word segmentation: A twoprocess account of statistical learning. Current Directions in Psychological Science, 22, 239–243. Thiessen, E. D., Kronstein, A. T., & Hufnagle, D. G. (2013). The extraction and integration framework: A two-process account of statistical learning. Psychological Bulletin, 139(4), 792–814. Thorn, A. S. C., & Gathercole, S. E. (1999). Language-specific knowledge and short-term memory in bilingual and non-bilingual children. The Quarterly Journal of Experimental Psychology Section A, 52(2), 303–324. Tomasello, M. (1988). The role of joint attentional processes in early language development. Language Sciences, 10, 69–88. Tomasello, M. (2003). Constructing a language: A usage-based theory of language acquisition. Harvard University Press.
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
571
572
PA U L L E S E M A N A N D J O S J E V E R H AG E N
Treffers-Daller, J., & Calude, A. S. (2015). The role of statistical learning in the acquisition of motion event construal in a second language. International Journal of Bilingual Education and Bilingualism 18, 602–623. Ullman, M. T. (2001). The neural basis of lexicon and grammar in first and second language: The declarative/procedural model. Bilingualism: Language and Cognition, 4(1), 105–122. Unsworth, S. (2005). Child L2, Adult L2, Child L1: Differences and similarities. A study on the acquisition of direct object scrambling in Dutch. (Doctoral dissertation, Utrecht University). Unsworth, S. (2014). Comparing the role of input in bilingual acquisition across domains. In T. Grüter & J. Paradis (Eds.), Input and experience in bilingual development (pp. 181–201). John Benjamins. Unsworth, N., & Engle, R. W. (2007). The nature of individual differences in working memory capacity: Active maintenance in primary memory and controlled search from secondary memory. Psychological Review, 114, 104–132. Unsworth, N., & Spillers, G. J. (2010). Working memory capacity: Attention control, secondary memory, or both? A direct test of the dual-component model. Journal of Memory and Language, 62, 392–406. Verhagen, J. (2009). Finiteness in Dutch as second language (Doctoral dissertation Free University of Amsterdam). Verhagen, J., Messer, M. H., & Leseman, P. P. M. (2015). Phonological memory and the acquisition of grammar in child L2 learners. Language Learning, 65, 417–448. Verhagen, J., Boom, J., Mulder, H., de Bree, E. H., & Leseman, P. P. M. (2019). Reciprocal relationships between nonword repetition and vocabulary during the preschool years. Developmental Psychology, 55, 1125–1137. Verhagen, J., & de Bree, E. H. (2020). Effects of bilingualism on statistical learning in preschoolers. Linguistic Approaches to Bilingualism, 11(5), 611639. Verhagen, J. & Leseman, P. P. M. (2016). How do verbal short-term memory and working memory relate to the acquisition of vocabulary and grammar? A comparison between first and second language learners. Journal of Experimental Child Psychology, 141, 65–82. Weizman, Z. O., & Snow, C. E. (2001). Lexical output as related to children’s vocabulary acquisition: Effects of sophisticated exposure and support for meaning. Developmental Psychology, 37, 265–279. Wulff, S., & Ellis, N. C. (2018). Usage-based approaches to SLA. In D. Miller, F. Bayram, J. Rothman, & L. Serratrice (Eds.), Bilingual cognition and language: The state of the science across its subfields. (pp. 37–56). John Benjamins. Zwaan, R. A., & Radvansky, G. A. (1998). Situation models in language comprehension and memory. Psychological Bulletin, 123, 162–185.
https://doi.org/10.1017/9781108955638.031 Published online by Cambridge University Press
26 Working Memory and L2 Grammar Learning among Adults Timothy McCormick and Cristina Sanz 26.1
Introduction
Cognitive abilities in the field of psychology – “the skills involved in performing the tasks associated with perception, learning, memory, understanding, awareness, reasoning, judgment, intuition, and language” (American Psychological Association, 2020) – share considerable overlap with theories of both native and nonnative language acquisition. For example, in addition to noticing itself, Schmidt’s Noticing Hypothesis (1990), a seminal work in contemporary research on second language (L2) acquisition (SLA), gives a crucial role to “awareness.” Therefore, it is no surprise that cognitive capacity – a measure of the limits of these abilities – including working memory (WM), has played a major role in SLA research. A key question among linguists is whether unexploited features of human language that are not fully present in an adult’s first language (L1), such as grammatical gender in English, can still be learned in adulthood (e.g., the Failed Functional Features Hypothesis; Hawkins & Chan, 1997). Among applied linguists, this question is often framed in terms of processing demands, with WM emerging as (one of ) the stand-in(s) for human processing limitations. Of course, not all features pose the same complexities: Some features in L2 grammar may require more processing resources than others, therefore making their acquisition more difficult. For example, some models, such as the Primacy of Meaning Model (VanPatten, 1990), predict a hierarchy of processes that broadly account for the order of L2 grammar development. In the model, learners’ cognitive resources are charged first and foremost with deciphering meaning, the most fundamental of the processes in the hierarchy. Once comprehension becomes more automatized, recruiting fewer cognitive resources, the model predicts that learners’ resources move toward attending to more meaningful grammatical features (e.g., subject-verb agreement) before less meaningful features (e.g., nounadjective gender agreement). As these processes too become automatic,
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
574
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
cognitive resources are freed to attend to higher-order processes (see Sanz & McCormick, 2021, for a recent overview and study). Given this convergence of interests, research on WM in adult SLA is an area where specialists converse from different fields, including psycholinguistics and cognition, theoretical and applied language acquisition, laboratory and instructed SLA. Most often, this means a more holistic view of the issue, but it can also create disjointed terminology or fundamentals, so it is important to bear that in mind when comparing studies or findings within this research.
26.2
WM & SLA: Models, Operationalizations, and Predictions
Although there have been many competing models of WM in psychology research, most L2 researchers rely on Baddeley and Hitch’s (1974) model, one of the first to combine the storage role of what had previously been called short-term memory with a processing role, as well as the subsequent updates to this model (e.g., Baddeley, 2000, 2017). Their original model included three components: the phonological loop, responding to aural and verbal input; the visuospatial sketchpad, responding to visual stimuli; and the central executive, which coordinated the other two units, dubbed its “slave systems” (Baddeley & Hitch, 1974). The model was later updated to include the episodic buffer, a third slave system that also depends heavily on the central executive (Baddeley, 2000). The episodic buffer, linked through the central executive to the other slave systems, is responsible for the sequential ordering of stimuli that fall in the visuospatial or phonological domain; the episodic buffer also links WM with long-term memory. It was also proposed to serve as a “mental modelling space, allowing one to set up representations that might guide future actions” (Baddeley & Wilson, 2002, p. 1738). There are two main positions that researchers have taken concerning the role of WM in SLA. The more common is the “more is better” approach (Miyake & Friedman, 1998), whereby more WM resources correspond to greater L2 outcomes (e.g., Coughlin & Tremblay, 2013; Sagarra, 2007). This view was supported by a meta-analysis investigating the effects of WM on L2 outcome measures, which found a moderate, positive relationship (Linck et al., 2014). The second prediction is an extension of Newport’s (1990) “less is more” hypothesis regarding child versus adult acquisition. In this hypothesis, children’s limited cognitive resources restrict them to focusing on the elementary building blocks of their L1, while adults learning an L2 muddle the process by attending to more complex structures. For the same reason, the extension of this hypothesis predicts that adults with fewer available resources will also have an advantage in the L2 over those with more available resources (e.g., Cochran et al., 1999; Radulescu et al., 2019),
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
L2 Grammar Learning among Adults
though this research generally modulates cognitive load as a stand-in for cognitive differences. Unfortunately, the matter is more complex than either of these predictions can account for, which has led to findings that are interpreted as “mixed,” “contradictory,” or “inconclusive” (e.g., Indrarathne & Kormos, 2018, p. 357). In this chapter, we argue that WM’s role is (1) dynamic throughout the L2 developmental process and (2) contingent on task characteristics that challenge learners’ storage and processing capacities. Of the four components of WM, the key variables of interest in L2 research have been the phonological loop and the central executive. Given its role in verbal information processing, the phonological loop was originally proposed to be an important player in language acquisition (Baddeley, 2000; Baddeley et al., 1998), as it has proven to be since its inception (e.g., Hummel, 2009; Williams & Lovatt, 2003). Meanwhile, the central executive’s main functions include both updating the information stored within WM as new input is received and delegating resources selectively to incoming stimuli (Baddeley & Hitch, 1974), so researchers have often asked how individual differences (IDs) in this component manifest in L2 learning outcomes. To measure WM broadly or the central executive, researchers often use complex span tasks (e.g., listening, reading), which ask participants to process series of sentences and store some element from each (e.g., the last word, letters or digits that follow each sentence). The storage and processing required for these tasks is generally associated with tasks requiring both storage and processing (e.g., long distance dependencies or temporary ambiguities, see Juffs & Harrington, 2011; Papagno, Chapter 4, this volume). Given the paucity of research on the visuospatial sketchpad and the episodic buffer in adult SLA, this chapter focuses on the central executive and phonological loop. Nonetheless, Baddeley (2003, 2017) does propose that the visuospatial sketchpad may have some language functions, such as reading or learning new orthographies.1 One recent study investigated visuospatial WM in L2 morphosyntax learning. In their study on task difficulty, morphosyntactic salience and WM, Zalbidea and Sanz (2020) found that visuospatial WM positively correlated to L2 oral and written morphosyntax development. Importantly, they used the Corsi blocktapping task, which taps into processing visual data and storing its sequential order. The authors suggest that L2 learners (L2Ls) with higher visuospatial WM may be more able to reproduce the correct sequential order of the respective morphosyntactic elements. This would also implicate a role for the episodic buffer, given its role coordinating with the visuospatial sketchpad to process and store sequences in the visuospatial domains. However, quite little is known about the episodic buffer’s role in SLA, likely because it is the newest component in Baddeley and colleagues’ model. That said, its role in SLA is a promising avenue for future research given its importance in processing and storing sequences, and given the recent results of Zalbidea and Sanz (2020).
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
575
576
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
A considerable amount of adult L2 research looks at phonological shortterm memory (PSTM). PSTM itself is not a subsystem in Baddeley’s model. Rather, it is an operationalization of phenomena that are largely attributed to the phonological loop (Baddeley et al., 1998; Gathercole & Thorn, 1998; Harrington & Sawyer, 1992; Papagno et al., 1991). PSTM reflects the shortterm maintenance of aural information (Hummel, 2009). Span tasks were previously the main measurement of PSTM. Simple span tasks (e.g., digit, letter, or word span) ask participants to recall and repeat back series of items in the same order as presented. However, the nature of these items allows participants to use lexical knowledge to store them (see Gathercole et al., 1991). To disentangle PSTM from general WM phenomena (such as access to long-term memory), nonword repetition tasks (NWRT) have generally replaced span tasks as the principle means of operationalizing PSTM. This is in part because they task the participant with repeating a series of nonwords that obey the phonotactics of the target or native language without the potential for applying mnemonics, thereby isolating phonological aspects of WM. Nonetheless, in L2 research, given the phonological loop’s dominance as the slave system of interest, PSTM and WM are sometimes conflated. For example, digit spans have been used to measure WM (Kapa and Colombo, 2014, Study 1; Kormos & Sáfár, 2008; McDonough & Trofimovich, 2016) as well as PSTM as distinct from the central executive (Atkins & Baddeley, 1998; Serafini & Sanz, 2016). Therefore, it is important to understand how different tasks tap into different components of memory capacity. To facilitate this understanding, throughout this chapter, when we report a construct of interest in a study (e.g., WM), we also report its operationalization in parenthesis. As this field advances, it will also be important that researchers move toward a shared frame of reference wherein the concept and its operationalization are common among L2 researchers (see Shin & Hu, Chapter 2, this volume).
26.3
Major Research Strands
There are two dominant lines of research on WM and L2 learning that we highlight in this chapter. The first and more sizable strand we highlight is WM and the learning of discrete L2 morphosyntax, followed by WM’s role in more holistic measures of grammar learning such as oral proficiency and standardized tests. However, before addressing these two strands, we believe mention of WM as a component of L2 aptitude is merited. For over two decades, WM has been a variable of particular interest in research on adult L2 learning aptitude, both as a correlate to traditional aptitude as measured by Carroll and Sapon’s (1959) Modern Language Aptitude Test (MLAT, e.g., Hummel, 2009; Kormos & Sáfár, 2008; Robinson, 2005a) and as a component of aptitude in multicomponential perspectives (e.g., Doughty et al., 2010; Miyake & Friedman, 1998, Wen
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
L2 Grammar Learning among Adults
et al., 2017). In fact, Miyake and Friedman (1998) explore WM as “one (if not the) central component of this language aptitude” (p. 339). This shift to include broader-scope cognitive abilities in L2 learning aptitude allows researchers to account for general information processing, and has led to the development of new test batteries that measure WM (or subcomponents thereof ) as part of aptitude, such as the High-Level Language Aptitude Battery (Hi-LAB; Doughty et al., 2010; Linck et al., 2013). Given aptitude’s importance in L2 proficiency alongside the storage-processing processes that make WM a component in new perspectives of aptitude (see Li’s 2016 meta-analysis for further discussion of both of these ideas), it is important to bear aptitude in mind when considering WM’s role in L2 learning. Given the scope of this chapter, we do not explore aptitude, but we recommend Wen and colleagues’ (2017) article for more on WM in SLA aptitude research. Below, we address the major strands in which WM is investigated as a predictor of adult L2 learning.
26.3.1 WM in Grammar Development The role of WM in the acquisition of L2 grammar by adults accounts for two of the main strands mentioned above: first, the acquisition of discrete grammatical structures at specific moments of language development, measured through explicit knowledge or sensitivity to violations. The second strand includes grammar in broad terms, such as standardized tests or oral proficiency measures with a holistic selection of features. We will present findings from the first strand, discrete grammar development in adults, by separating novice learning of artificial and semiartificial languages from natural language development at varying proficiencies. 26.3.1.1 Discrete L2 Grammar Learning: Artificial Languages Both global WM and PSTM have been associated with novice language learning in studies using artificial or semiartificial grammars. The storage and processing functions of WM seem to play a role in developing early bases for grammar learning, more than certain executive functions (EFs) or more general cognitive abilities. PSTM has also been shown to play a role independent of and parallel to global WM. Aiming to elucidate how specific EFs play a role in novice grammar learning alongside the more global WM, and independent from verbal ability, Kapa and Colombo (2014, Study 1) trained participants in an artificial grammar during two 15-minute sessions. They calculated a composite language score that included vocabulary, sentence formation, and a grammaticality judgment task (GJT). They also measured WM (digit span) and several EFs (Attentional Network Task [ANT], Simon Task, Wisconsin Card Sorting Task). They found that WM had a large effect2 on artificial grammar learning, but of the EF measures, only inhibitory control – faster responses in ANT inhibitory trials – predicted greater learning outcomes. Misyak and
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
577
578
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
Christiansen (2012) also found a medium to large effect of verbal WM (reading span) on learning through exposure to adjacent and nonadjacent dependency patterns in an AG. Their study also included measures of shortterm memory, vocabulary, reading experience, cognitive motivation, and fluid intelligence, but the statistical learning of the patterns only correlated with natural language comprehension and with verbal WM. Baddeley and colleagues initially proposed the phonological loop as a language acquisition device because of its posited role in vocabulary learning. However, Williams and Lovatt (2003) argued that learning sequences of sounds and their arbitrary meaning (i.e., vocabulary) is similar to learning sequences of morphemes and how they map to meaning (i.e., grammar), even if the latter process is more global or may demand resources from other dimensions of WM beyond PSTM. Looking at PSTM’s role in semiartificial grammar learning, they found that PSTM (immediate serial recall task) had a large effect on participants’ ability to generalize gender based on the determiners that accompanied nouns, as well as on vocabulary learning and early association of determiner-noun pairs. They argued that PSTM, in its relation to word learning, is thus related to learning patterns between words, and therefore to grammar. Martin and Ellis (2012), in a similar paradigm, looked at both WM and PSTM’s role(s) in learning vocabulary, word order, and plural markers. They found WM (listening span) and PSTM (a NWRT and a nonword recognition task) had medium effects on vocabulary learning after one session and after two sessions with novel words. More critically for this chapter, they also found that WM and PSTM (only the NWRT) accounted for variance in grammar learning. In a series of hierarchical multiple regressions, both WM and PSTM independently accounted for variance in grammar production and grammar comprehension, with small to medium effect sizes. The authors claimed this confirmed their hypothesis that the processes involved in learning novel vocabulary and novel grammar rely on some common resources, despite earlier researchers’ claim that PSTM and WM had separate roles in SLA (lexical vs. sentential/discourse level). Nonetheless, there are studies that have not found effects for WM in the artificial grammar paradigm research. For example, Grey et al. (2015) found that success at acquiring word order and morphological case in an artificial language (Japanese order, English words) did not correlate to PSTM (two NWRTs). McDonough and Trofimovich (2016, Study 1), who included participants exposed to the transitive in Esperanto, also did not find any correlation between performance and WM (although there was a correlation between performance and statistical learning scores). In sum, these findings generally suggest that WM and PSTM play some role in learning linguistic (or linguistic-like) patterns during first exposure (s) to an artificial or semiartificial grammar in adulthood. Of course, as we will discuss below, artificial languages are not perfect stand-ins for natural languages, so it is important to consider these findings as informative to
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
L2 Grammar Learning among Adults
our understanding of natural language learning, and not necessarily as independent accounts for L2 learning per se. Before discussing WM in natural L2 learning, we first present a subset of studies using (semi)artificial languages to assess WM’s role in implicit and explicit learning conditions.
26.3.1.2 Discrete L2 Grammar Learning: Implicit and Explicit Instruction Although it has received considerable attention, still elusive to researchers is the role of WM in adult grammar learning under explicit (i.e., with a focus on form through rule presentation or instructions to find a rule) or implicit learning conditions (i.e., enabling learners to infer rules without awareness through meaning-oriented tasks, Ellis, 2009, p. 16). Early research on WM and instructional conditions suggested that WM played a role in L2 grammar learning under explicit conditions (e.g., Reber et al., 1991), but not implicit (or incidental) learning, which was thought to rely on other cognitive processes (Unsworth & Engle, 2005). However, some research shows that WM does have a role on grammar learning under implicit/incidental conditions (Hassin et al., 2009; Soto & Silvanto, 2014). Particularly, under implicit conditions, higher working memory capacity (WMC) enables keeping more information active to detect hidden patterns (see Martini et al., 2015). Interpreting this literature is complex, and the takeaway is not entirely clear. For example, Tagarelli et al. (2015) investigated how WM (Operation Span, letter-number ordering task) affected semiartificial learning in different instructional conditions (incidental, intentional). Tagarelli et al. (2016) also looked at the role of WM (reading span task) on learning in different instructional conditions (incidental, instructed + rule). In one study, there was only a positive relationship between WM and grammar learning in the intentional condition (Tagarelli et al., 2015). In the other, there was only a positive correlation between WM and performance on complex structures in the incidental group (Tagarelli et al., 2016). Clearly, WM’s role is tenuous even for researchers using similar procedures. Some scholars have also investigated grammar acquisition under different levels of instructional explicitness. For example, in the Latin Project, Sanz and colleagues investigated how WM and learning conditions contribute to English/Spanish bilinguals third language (L3) learning. Working with a reduced form of Latin, Sanz et al. (2016) provided reactive metalinguistic feedback and found that WM correlated to the comprehension of input only in the condition without prepractice grammar. The authors concluded that higher WMC is only advantageous when metalinguistic information is limited to reactive feedback, not preemptive instruction. Lado (2017), also using this paradigm, investigated WM and PSTM’s role in L3 learning, as well as linguistic analytic ability and rote memory. Lado manipulated the presence of metalinguistic feedback with no grammar lesson. Like Sanz and colleagues (2016), she found that higher WMC was advantageous in the presence of metalinguistic feedback. While linguistic
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
579
580
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
ability correlated to processing for meaning in this condition, only WM correlated to acquiring a sensitivity to the morphosyntax itself. We recommend the many publications in the Latin Project (e.g., Lado et al., 2014, 2017), as well as the Brocanto2 Project (e.g., Grey et al., 2017; Morgan-Short et al., 2010), that include WM and other cognitive variables, as well as neural correlates, if the reader is interested in WM’s interaction with learning conditions and degree of bilingualism. On top of the implicit versus explicit instruction divide, there is another complexity given researchers’ interest in the type of knowledge acquired – explicit or implicit. L2 learners’ WM has been correlated to increases in explicit knowledge under explicit conditions (Robinson, 2005b; Santamaria & Sunderman, 2015) and to increases in explicit knowledge under implicit conditions (Ellis & Sinclair, 1996; Martin & Ellis, 2012; Williams & Lovatt, 2003). However, while correlations between WM and implicit knowledge in implicit conditions were found by Ellis and Sinclair (1996), others have not reproduced similar results (Grey, Williams et al., 2015; Robinson, 2005b; Tagarelli et al., 2015). Finally, again adding to the complexity, there is the question of how artificial and natural language learning relate to each other. For example, Robinson (2010) included both an artificial and natural (Samoan) grammar in his study and only found a positive correlation between WM and learning Samoan’s patterns, not between WM and artificial grammar learning via explicit or implicit exposure. Recently, Indrarathne and Kormos (2018) asked how the strand’s findings translate to a natural language already known to the participants. For the L2Ls in their study, rated B1 or low B2 on the Common European Framework of Reference, they found that higher WMC facilitated the acquisition of an unknown grammatical structure in both explicit and implicit conditions; that WM played more of a role in explicit conditions than implicit conditions for explicit knowledge development (also see Issa, 2019); and that implicit knowledge development was not affected by condition, only WM differences. While these complex interactions are difficult to synthesize, it is a very promising avenue for future research, particularly given Indrarathne and Kormos’s (2018) recent extension of the question of WM’s role in implicit versus explicit learning to natural grammar acquisition at increased proficiency, which opens a host of possibilities to researchers.
26.3.1.3 Discrete L2 Grammar Learning: Natural Languages Discrete measures of morphosyntactic knowledge, especially agreement, are a primary focus for research on WM in adult SLA. This includes subject-verb agreement, noun-adjective, article-noun agreement, or, to a lesser extent, word order and morphological case markings. Most findings support the “more is better” hypothesis, but it is important to consider the cognitive load that a task poses for the relevant proficiency group in order to contextualize this literature. Specifically, the research suggests that the
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
L2 Grammar Learning among Adults
benefits of higher WM shift from lower-level morphosyntactic processes, like learning words’ inflectional morphology (e.g., Kempe & Brooks, 2008), to higher-level morphosyntactic processes, like sensitivity to violations in online processing (e.g., Coughlin and Tremblay, 2013). Using broad measures of proficiency, Linck and Weiss (2015) investigated WM’s role on longitudinal L2 learning in first- and third-semester Spanish and German L2Ls by comparing scores on a standardized grammar and vocabulary test at the beginning and end of one semester. They found that WM had a medium effect on end-of-term proficiency and a (marginally significant) medium effect on gains made from pretest to posttest. Sagarra (2017) also investigated second-semester Spanish L2Ls grammar development longitudinally, using a multiple-choice grammar test to assess knowledge of subject-verb, gender, and number agreement, among other grammatical constructs. Sagarra repeated the process with two cohorts of learners completing similar protocols: In Experiment 1, learners were assessed after one year of learning and completed the Daneman and Carpenter (1980) reading span test; in Experiment 2, learners were assessed after one semester and completed the Waters and Caplan’s (1996) reading span test, which adds processing demands to the earlier test. She found that gains from pretest to posttest only correlated with the Waters and Caplan test, the test whose dual processing-and-storage demands align more clearly with contemporaneous models of WM. Researchers looking at discrete grammar features have found similar results among low-proficiency learners. For example, among true beginners exposed to Russian, greater WM and PSTM provided a benefit to learn vocabulary and make connections between words and inflectional morphology like gender and case (Kempe et al., 2010; Kempe & Brooks, 2008), though not to generalize those associations to new tokens. Sagarra (2007) found that novice L2Ls of Spanish as a group were insensitive to gender violations in self-paced reading (SPR), but those with high WMC did reveal the sensitivity. Sagarra and Herschensohn (2010) performed a crosssectional study on number and gender violations in L2 Spanish. The study’s beginners lacked variability in WM and performed at floor for both an offline grammaticality judgment task and online SPR, with no sensitivity to either type of violation. Intermediate learners in the study did reveal sensitivity to violations in both tasks, and higher WMC among intermediate learners was associated with greater sensitivity to gender – but not number – violations. A native control group performed at ceiling for both the GJT and SPR, with no effects for WM. This tendency of positive effects for WM on inflectional morphology learning among low-proficiency learners seems to change for advanced learners. For example, Grey, Cox et al. (2015) investigated morphosyntactic development during a short-term study abroad by looking at changes in sensitivity to gender and number agreement and to word order. Participants were at least at the advanced level, and with the exception of
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
581
582
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
gender agreement, the speed and accuracy of their grammaticality judgments improved after a period of immersion abroad. However, neither WM (sentence span) nor PSTM (two NWRTs) explained their differential rates of success. The authors hypothesized that the richness of the input abroad may level the playing field for learners of different aptitudes, and that cognitive capacity may play a lesser role in moderating L2 learning among advanced learners. As we argue in this chapter, if the (timed) GJT did not push learners to their cognitive limits, WM would likely not emerge as a player. Indeed, in Coughlin and Tremblay’s (2013) study of intermediate and advanced learners, we see that adding a layer of complexity does reveal WM differences. In their study on WM L2Ls of French exposed to number agreement violations, learners (and native controls) completed an acceptability judgment, SPR, and WM tasks in French and English. There were no differences according to proficiency or WM in the offline acceptability judgments. However, only the advanced learners and native speakers revealed sensitivity to violations in the online SPR task, which varied as a function of French WMC. In other words, among these intermediate – and advanced-level learners, offline number agreement did not draw out WMC differences: both levels performed at ceiling. However, processing number agreement online increases task complexity and WM differences emerge (see Cunnings, Chapter 27, this volume, for more on WM and processing). This changing role of WM is especially visible in the longitudinal and cross-sectional study conducted by Serafini and Sanz (2016). These authors investigated one-semester gains in the automaticity of ten morphosyntactic structures for Spanish L2Ls at three proficiency levels. They measured EF (operation span) and PSTM (digit span), and used an online elicited imitation task (EIT) and an untimed GJT to measure linguistic knowledge. PSTM correlated to the online EIT and the untimed GJT scores for basic and intermediate learners. EF was also found to correlate to performance on the GJT for basic and intermediate learner groups, in addition to predicting variance in gains on the EIT for beginners. Unlike basic and intermediate learners, advanced learners obtained no robust relationships between these cognitive variables and morphological knowledge of Spanish, suggesting a decreased reliance on EF and PSTM at increasing proficiency levels. It is important to note that the same linguistic structures were assessed for all three groups, which therefore needed to be basic enough for early learners in the sample. Thus, the findings seem to indicate decreased recruitment of WM at higher L2 proficiencies as grammatical automaticity increases. Summarizing the literature, learners at lower proficiency levels who have a WM advantage also – more often than not – have a higher performance outcome in morphosyntactic agreement, but this advantage appears less consistent for learners with high WMC at high proficiency levels. This reflects what Serafini and Sanz (2016) found in their longitudinal, crosssectional study. As they point out, “studies tend to more consistently report positive effects for lower proficiency learners whereas findings for
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
L2 Grammar Learning among Adults
advanced learners with more exposure to and practice in the target language are mixed” (Serafini & Sanz, 2016, p. 614). However, this is not to say that WM plays less of a role in general at higher levels; rather, it plays less of a role in morphosyntactic features that are commonly studied. For example, among university students beyond the “advanced level,” comparable to participants in Foote (2011) and Grey, Cox et al. (2015), WM has been positively correlated to nativelike use of semantic plausibility (Dussias & Piñar, 2010) and to syntactic comprehension and native-like cue use (Miyake & Friedman, 1998), three grammatical features that are arguably more complex (i.e., more advanced) than subject-verb or noun-adjective agreement (NB: although we recognize that near-nativelike agreement is a late-acquired ability, the first signs of sensitivity to agreement violations, for example, are not; see Morgan-Short et al., 2010). Once learners develop a sensitivity to more basic morphosyntactic features (and violations thereof ) like agreement, their WM resources become available to intercede in higher-order processes, meaning that the advantage of learners with higher WMC changes across proficiency levels, but does not disappear. This also means that the tasks posed to L2Ls must be increasingly challenging in order to see a differential role of WM at higher proficiency levels.
26.3.1.4 Global Measures of L2 Grammar Learning: Production Using the Michigan Test of English Language Proficiency (MTELP), Hummel (2009) studied adult French L2L of English and found that their PSTM (NWRT) correlated to scores on English vocabulary, grammar, and reading comprehension. With this exception, studies on WM and L2 learning that use standardized tests have mostly evaluated adolescent L2 development (e.g., Harrington & Sawyer, 1992; Kormos & Sáfár, 2008; see Leseman and Verhagen, Chapter 25, this volume). Instead, research looking at WM and learning among adults often exploits the wide array of global proficiency measures available in oral or written production. As L2Ls advance, the tools and methodology at researchers’ disposal in order to answer their questions about the contribution of WM to grammar development change. One method is to look at increasingly complex discrete grammar structures using finer-grained measurements (e.g., Havik et al., 2009, who included a task that challenged even native speakers, and as a result, found a differential role for WM, even for late-stage language learners). Another method is to use global measures that depend on advanced grammar acquisition, such as oral or written production. Oral production is a common target among adult L2 research and WM. It is a complex aspect of L2 development that depends on the successful union of grammar, vocabulary, and other linguistic domains, so naturally, WM is a variable of considerable interest among researchers. Although hesitation phenomena do not fall within a typical understanding of grammar and different speakers use these devices to different degrees in the L1 (see Clark & Fox Tree, 2002), the frequency of their use in the L2 has also been linked
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
583
584
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
to level of proficiency (Lennon, 1990). Fehringer and Fry (2007) studied the presence of hesitation phenomena in oral speech among highly advanced L2 speakers of English and German. They found that, at high proficiency levels, higher WMC (aural WM span and the story retelling subtest of the Adult Memory and Information Processing Battery; Coughlan & Hollows, 1985) correlated negatively with the use of hesitation phenomena in oral speech. Their findings demonstrate that high-WMC speakers have an advantage over their low-WMC peers in overcoming the difficulties presented by a L2 in development (Fehringer & Fry, 2007). Weissheimer and Borges Mota (2009) investigated how WM (speaking span) related to learners’ oral proficiency development over two months. The level of these learners is not clear in the study because the focus is on development of both WMC and oral proficiency, but all learners were majoring in English at a Brazilian university. They found that WMC related to both fluency and complexity development, but not to oral accuracy development. While accuracy is one measure of grammatical control (how correct is the grammar?), grammar development is also necessary for development of both complexity (how advanced is the grammar?) and fluency (how accessible are the lexicon and grammar in real time?). Borges Mota (2003) found that WM (speaking span) correlated positively with advanced learners’ oral fluency, accuracy, and complexity (and negatively with lexical density), although this study only included 13 learners. Cho (2018), on the other hand, did not find a correlation between WM (reading and operation spans) and complexity, accuracy, or fluency in the written or oral modes. In this study on task complexity and proficiency among upperintermediate Korean learners of English, WM did not correlate to any L2 performance metrics, although the author does suggest that this may reflect tasks that “[were not] demanding enough compared to those used in the previous research” (Cho, 2018, p. 94). O’Brien and colleagues (2007) showed similar findings in their study on novice and intermediate learners of Spanish in both at-home and studyabroad contexts. They measured general oral ability and fluency through several measures (e.g., mean run without hesitation phenomena). They found that PSTM (NWRT) had a large effect on development of L2 oral fluency over time, namely, to the amount of speech produced, length of longest turn and of longest fluent run, speech rate, and amount of speech between filled pauses. The same authors also investigated the role of PSTM on gains made at home or abroad by learners at the intermediate level or above, divided in two groups by proficiency (O’Brien et al., 2006). They found that PSTM (NWRT) had a medium effect on gains in narrative abilities among the lower-proficiency group. They also found that PSTM had a very large effect on gains in accuracy of function word use among the more proficient learners, extending previous findings “by demonstrating that
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
L2 Grammar Learning among Adults
PSTM is implicated in the language development of adult L2 learners, at stages when language learning becomes more demanding (effortful)” (p. 398). They argue that as more elementary processes become simpler for the learner (e.g., lexical access to content words), WM, and specifically PSTM, is “redeployed for learning more complex grammar” (p. 399), in other words, fulfilling a dynamic role across L2 grammar development. Despite some research on discrete grammar acquisition that indicates a decreased role for WM at higher proficiency levels (e.g., Grey et al., 2015; Serafini & Sanz, 2016), upon considering advanced grammar (e.g., Havik et al., 2009) or global measures of the L2 that reflect advanced control of many aspects of grammar (e.g., O’Brien et al., 2006, 2007), WM has a clear role in managing and coordinating learners’ L2 knowledge during these more demanding tasks.
26.4
Conclusion
This chapter has summarized the literature on WM and L2 grammatical learning. The interest entered SLA research from cognitive psychology in the 1990s at a time when models of L2 development had just incorporated the role of attention and awareness in processing for acquisition. This established WM, the construct where incoming linguistic information is processed and temporarily stored, as a potentially key variable in explaining differential rates of success across L2Ls. The research has mostly focused on learning morphology and word order, either in the laboratory with novice learners of artificial or minigrammars or with beginner and intermediate L2Ls in cross-sectional designs. However, examples of longitudinal studies, those that include advanced learners in their samples, and those that manipulate learning conditions have proven to be most informative. As we have shown, most research in this strand supports the idea that greater WM resources are beneficial to adult L2 learning. Greater storage and processing capabilities facilitate the multifaceted mental processes involved in L2 learning, such as pattern learning among novice learners of artificial languages (e.g., Martin & Ellis, 2012). However, as we have argued, there is much to learn about how artificial grammar learning reflects natural L2 learning. Indrarathne and Kormos (2018) took a first step to answer this question by investigating an existing, albeit rare, L2 structure unknown to participants and found that WM played a considerable role. Future research might follow this approach to test the validity of findings from the artificial grammar paradigm beyond the laboratory. Within natural languages, as learners advance, the tasks that yield enough cognitive load to observe differences change (see Serafini & Sanz, 2016). Thus, to understand the “mixed,” “contradictory,” or “inconclusive”
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
585
586
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
findings (Indrarathne & Kormos, 2018, p. 357), researchers must consider how studies differ in the tasks posed to L2Ls: to what extent were the WM resources engaged given the specific proficiency level(s) under investigation? WM is only one of many variables leading to differences in ultimate L2 attainment, and its role reveals itself empirically only when the task pushes the learners to their cognitive limits. When learners undertake a task that is mundane or banal for their proficiency level, high- and lowWMC individuals will perform similarly (or, at least, differently to an unobservable degree). However, when these learners are challenged with a task that (1) requires storage and processing and (2) is still outside the threshold of automaticity, differences in WMC amplify the differences in L2Ls’ performance to reveal WM’s importance. We should expect such tasks to change across proficiency levels as we ask when and where WM is most involved in L2 learning. Although cross-sectional, longitudinal designs are burdensome, the findings obtained from studies such as Serafini and Sanz (2016), which showed that WM differences correlated to outcomes across ten morphosyntactic structures among beginner and intermediate, but not advanced, L2 Spanish learners, are incredibly valuable to the field because they can pinpoint not only what correlates to WM but when. Building on this, studies with both multiple proficiency levels and multiple tasks of increasing complexity may prove useful to further probe WMC advantages and the dynamic nature of WM. Throughout the past 30 years, our understanding of WM has remained ephemeral as we continually learn more about its role in cognition. It simultaneously correlates to broader constructs like general fluid intelligence and to narrower, more specific control tasks like Stroop (see Engle & Kane’s executive-attention theory of WMC variation, 2004). Moving forward, it will be important to consider how specific tasks and conditions – in both linguistic and WM measures – call upon specific executive control processes to better elucidate the contributions of specific control mechanisms to L2 processing and proficiency development. In this chapter, we have focused on the former – linguistic tasks at varying proficiencies – because we believe understanding its dynamicity is critical to understanding its role. However, as Linck et al. (2014) motivate, SLA researchers should also consider contemporary views of WM that treat EFs as independent. This may mean including both complex WM and specific EF measures (e.g., Kapa & Colombo, 2014), or asking what domain-general functions are reflected within a linguistic task, like inhibitory control in ambiguity resolution (e.g., McCormick, 2020). In this same vein, further exploration of the roles of the episodic buffer and visuospatial sketchpad in adult L2 grammar is warranted, as motivated by Baddeley (2003, 2017) and particularly after recent findings from Zalbidea and Sanz (2020).
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
L2 Grammar Learning among Adults
References American Psychological Association (2020). APA Dictionary of Psychology. https://dictionary.apa.org/cognitive-ability Atkins, P. W. B., & Baddeley, A. D. (1998). Working memory and distributed vocabulary learning. Applied Psycholinguistics, 19(4), 537–552. Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417–423. Baddeley, A. D. (2003). Working memory: Looking back and looking forward. Nature Reviews Neuroscience, 4, 829–839. Baddeley, A. D. (2017). Modularity, working memory and language acquisition. Second Language Research, 33(3), 299–311. Baddeley, A. D., Gathercole, S. E., & Papagno, C. (1998). The phonological loop as a language learning device. Psychological Review, 105, 158–173. Baddeley, A. D. & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. VIII, pp. 47–90). Academic Press. Baddeley, A. D., & Wilson, B. A. (2002). Prose recall and amnesia: Implications for the structure of working memory. Neuropsychologia, 40, 1737–1743. Borges Mota, M. (2003). Working memory capacity and fluency, accuracy, complexity, and lexical density in L2 speech production. Fragmentos, (24), 69–104. Carroll, J., & Sapon, S. (1959). Modern language aptitude test. The Psychological Corporation/Harcourt Brace Jovanovich. Cho, M. (2018). Task complexity, modality, and working memory in L2 task performance. System, 72, 85–98. Clark, H. H. & Fox Tree, J. E. (2002). Using “uh” and “um” in spontaneous speaking. Cognition, 84, 73–111. Cochran, B., McDonald, J. & Parault, S. (1999). Too smart for their own good: The disadvantage of superior processing capacity for adult language learners. Journal of Memory and Language, 41, 30–58. Coughlan, A. K. & Hollows, S. E. (1985). The Adult Memory and Information Processing Battery (AMIPB): Test manual. Psychology Department, St James’s University Hospital. Coughlin, C. E., & Tremblay, A. (2013). Proficiency and working memory based explanations of non-native speakers’ sensitivity to agreement in sentence processing. Applied Psycholinguistics, 34, 615–646. Daneman, M. & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. Doughty, C., Campbell, S., Mislevy, M., Bunting, M., Bowles, A., & Koeth, J. (2010). Predicting near-native ability: The factor structure and reliability of Hi-LAB. In M. Prior, Y. Watanabe, & S. Lee (Eds.), Selected proceedings of
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
587
588
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
the 2008 Second Language Research Forum (pp. 10–31). Cascadilla Proceedings Project. Dussias, P. E., & Piñar, P. (2010). Effects of reading span and plausibility in the reanalysis of wh-gaps by Chinese-English L2 speakers. Second Language Research, 26(4), 443–472. Ellis, N. C. & Sinclair, S. G. (1996). Working memory in the acquisition of vocabulary and syntax. Quarterly Journal of Experimental Psychology, 49, 234–250. Ellis, R. (2009). Measuring implicit and explicit knowledge of a second language. In R. Ellis, S. Loewen, C. Elder, R. Erlam, J. Philp & H. Reinders (Eds.), Implicit and explicit knowledge in second language learning, testing and teaching (pp. 3–25). Multilingual Matters. Engle, R. W. & Kane, M. J. (2004). Executive attention, working memory capacity, and a two-factor theory of cognitive control. In B. Ross (Ed.), The psychology of learning and motivation (pp. 145–199). Academic Press. Fehringer, C., & Fry, C. (2007). Hesitation phenomena in the language production of bilingual speakers: The role of working memory. Folia Linguistica: Acta Societatis Linguisticae Europaeae, 41(1–2), 37–72. Foote, R. (2011). Integrated knowledge of agreement in early and late English-Spanish bilinguals. Applied Psycholinguistics, 32, 187–220. Frost, R., Siegelman, N., Narkiss, A., & Afek, L. (2013). What predicts successful literacy acquisition in a second language? Psychological Science, 24, 1243–1252. Gathercole, S. E., & Thorn, A. (1998). Phonological short-term memory and foreign language learning. In A. F. Healy & L. E. Bourne (Eds.), Foreign language learning: Psycholinguistic studies on training and retention (pp. 141–158). Erlbaum. Gathercole, S. E., Willis, C., Emslie, H., & Baddeley, A. D. (1991). Nonword repetition, phonological memory, and vocabulary: A reply to Snowling, Chiat, and Hulme. Applied Psycholinguistics, 12, 375–379. Grey, S., Cox, J., Serafini, E., & Sanz, C. (2015). The role of individual differences in the study abroad context: Cognitive capacity and language development during short-term intensive language exposure. Modern Language Journal, 99(1), 137–157. Grey, S., Sanz, C., Morgan-Short, K., & Ullman, M. (2017). Bilingual and monolingual adults learning an additional language: ERPs reveal differences in syntactic processing. Bilingualism: Language and Cognition, 1–25. Grey, S., Williams, J. N., & Rebuschat, P. (2015). Individual differences in incidental language learning: Phonological working memory, learning styles, and personality. Learning and Individual Differences, 38, 44–53. Harrington, M., & Sawyer, M. (1992). L2 working memory capacity and L2 reading skill. Studies in Second Language Acquisition, 14, 25–38. Hassin, R. R., Bargh, J. A., Engell, A. D., & McCulloch, K. C. (2009). Implicit working memory. Consciousness and Cognition: An International Journal, 18(3), 665–678.
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
L2 Grammar Learning among Adults
Havik, E., Roberts, L., Van Hout, R., Schreuder, R., & Haverkort, M. (2009). Processing subject–object ambiguities in the L2: A self-paced reading study with German L2 learners of Dutch. Language Learning, 59, 73–112. Hawkins, R., & Chan, C. Y. (1997). The partial availability of Universal Grammar in second language acquisition: The “failed functional features hypothesis.” Second Language Research, 13(3), 187–226. Hummel, K. M. (2009). Aptitude, phonological memory, and second language proficiency in non-novice adult learners. Applied Psycholinguistics, 30, 225–249. Indrarathne, B., & Kormos, J. (2018). The role of working memory in processing L2 input: Insights from eye-tracking. Bilingualism: Language and Cognition, 21(2), 355–374. Issa, B. I. (2019). Examining the relationships between attentional allocation, working memory and second language development: An eye-tracking study. In R. P. Leow (Ed.), The Routledge handbook of second language research in classroom learning (pp. 464–479). Routledge. Juffs, A., & Harrington, M. W. (2011). Aspects of working memory in L2 Learning. Language Teaching: Reviews and Studies, 42, 137–166. Kapa L. L., & Colombo, J. (2014). Executive function predicts artificial language learning. Journal of Memory and Language, 7, 237–252. Kempe, V., & Brooks, P. J. (2008). Second language learning of complex inflectional systems. Language Learning, 58, 703–746. Kempe, V., Brooks, P. J., & Kharkhurin, A. (2010). Cognitive predictors of generalization of Russian grammatical gender categories. Language Learning, 60, 127–153. Kormos, J., & Sáfár, A. (2008). Phonological short-term memory, working memory and foreign language performance in intensive language learning. Bilingualism: Language and Cognition, 11, 261–271. Lado, B. (2017). Aptitude and pedagogical conditions in the early development of a nonprimary language. Applied Psycholinguistics, 38(3), 679–701. Lado, B., Bowden, H., Stafford, C. & Sanz, C. (2014). A fine-grained analysis of the effects of negative evidence with and without metalinguistic information in language development. Language Teaching Research, 18(3), 320–344. Lado, B., Bowden, H. W. & Stafford, C. & Sanz, C. (2017). Two birds, one stone, or how learning a foreign language makes you a better language learner. Hispania, 100(3), 361–378. Lennon, P. (1990). Investigating fluency in EFL: A quantitative approach. Language Learning, 40, 387–417. Li, S. (2016). The construct validity of language aptitude: A meta-analysis. Studies in Second Language Acquisition, 38(4), 801–842. Linck, J. A., Hughes, M., Campbell, S., Silbert, N., Tare, M., Jackson, S., Smith, B., Bunting, M., & Doughty, C. (2013). Hi-LAB: A new measure of aptitude for high-level language proficiency. Language Learning, 63(3), 530–566.
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
589
590
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
Linck, J.A., Osthus, P., Koeth, J. T., & Bunting, M. (2014). Working memory and second language comprehension and production: A meta-analysis. Psychonomic Bulletin & Review, 21, 861–883. Linck, J. A., & Weiss, D. J. (2015). Can working memory and inhibitory control predict second language learning in the classroom? SAGE Open, 5(4), 1–11. Martin, K. I., & Ellis, N. C. (2012). The roles of phonological short-term memory and working memory in l2 grammar and vocabulary learning. Studies in Second Language Acquisition 34, 379–413. Martini, M., Sachse, P., Furtner, M. R., & Gaschler, R. (2015). Why should working memory be related to incidentally learned sequence structures? Cortex, 64, 407–410. McCormick, T. (2020). Early and emergent bilingualism: The role of cognitive control in processing linguistic conflict (Doctoral dissertation, Georgetown University). McDonough, K., & Trofimovich, P. (2016). The role of statistical learning and working memory in L2 speakers’ pattern learning. The Modern Language Journal, 100, 428–445. Misyak, J. B., & Christiansen, M. H. (2012). Statistical learning and language: An individual differences study. Language Learning, 62, 302–331. Miyake, A., & Friedman, N. P. (1998). Individual differences in second language proficiency: Working memory as language aptitude. In A. F. Healy & L. E. Bourne, Jr. (Eds.), Foreign language learning: Psycholinguistic studies on training and retention (pp. 339–364). Lawrence Erlbaum. Morgan-Short, K., Sanz, C., Steinhauer, K., & Ullman, M.T. (2010). Second language acquisition of gender agreement in explicit and implicit training conditions: An event-related potential study. Language Learning, 60(1),154–193. Newport, E. (1990). Maturational constraints on language learning. Cognitive Science,14(1), 11–28. O’Brien, I., Segalowitz, N., Collentine, J., & Freed, B. (2006). Phonological memory and lexical narrative, and grammatical skills in second language oral production by adult learners. Applied Psycholinguistics, 27, 377–402. O’Brien, I., Segalowitz, N., Freed, B., & Collentine, J. (2007). Phonological memory predicts second language oral fluency gains in adults. Studies in Second Language Acquisition, 29, 557–582. Papagno, C., Valentine, T., & Baddeley, A. D. (1991). Phonological short-term memory and foreign-language vocabulary learning. Journal of Memory and Language, 30, 331–347. Plonsky, L., & Oswald, F. L. (2014). How big is “big”? Interpreting effect sizes in L2 research. Language Learning, 64, 878–912. Radulescu, S., Wijnen, F., & Avrutin, S. (2020). Patterns bit by bit: An entropy model for rule induction. Language Learning and Development, 16(2), 109–140.
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
L2 Grammar Learning among Adults
Reber, A. S., Walkenfeld, F. F., & Hernstadt, R. (1991). Implicit and explicit learning: Individual differences and IQ. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 888–896. Robinson, P. (2005a). Aptitude and second language acquisition. Annual Review of Applied Linguistics, 25, 45–73. Robinson, P. (2005b). Cognitive abilities, chunk-strength, and frequency effects in implicit artificial grammar and incidental L2 learning: Replications of Reber, Walkenfeld, and Hernstadt (1991) and Knowlton and Squire (1996) and their relevance for SLA. Studies in Second Language Acquisition, 27(2), 235–268. Robinson, P. (2010) Implicit artificial grammar and incidental natural second language learning: How comparable are they? Language Learning, 60(2), 245–263. Sagarra, N. (2007). Working memory and L2 processing of redundant grammatical forms. In Z. Han & E. S. Park (Eds.), Understanding second language process. Multilingual Matters, 133–147. Sagarra, N. (2017). Longitudinal effects of working memory on L2 grammar and reading abilities. Second Language Research, 33(3), 341–363 Sagarra, N., & Herschensohn, J. (2010). The role of proficiency and working memory in gender and number agreement processing in L1 and L2 Spanish. Lingua, 120, 2022–2039. Santamaria, K., & Sunderman, G. (2015). Working memory in processing instruction: The acquisition of French clitics. In Z. Wen, M. Borges Mota, & A. McNeill, (Eds.), Working memory in second language acquisition and processing (pp. 205–223). Multilingual Matters. Sanz, C., Lin., H., Lado, B., Bowden, H. & Stafford, C. (2016). One size fits all? Learning conditions and working memory capacity in ab initio language development. Applied Linguistics, 37(5), 669–692. Sanz, C., & McCormick, T. (2021). VanPatten 1990’s long and winding story and the nature of replication studies. In M. Leeser, W. Wong, & G. Keating, (Eds.), Research on second language processing and processing instruction: Studies in honor of Bill VanPatten (pp. 153–181). John Benjamins Publishing Company. Schmidt, R. W. (1990). The role of consciousness in second language learning. Applied Linguistics, 11, 129–158. Serafini, E. J., & Sanz, C. (2016). Evidence for the decreasing impact of cognitive ability on second language development as proficiency increases. Studies in Second Language Acquisition, 38, 607–646. Soto, D., & Silvanto, J. (2014). Reappraising the relationship between working memory and conscious awareness. Trends in Cognitive Sciences, 18, 520–525. Tagarelli, K. M., Borges Mota, M. & Rebuschat, P. (2015). Working memory, learning context, and the acquisition of L2 Syntax. In Z. Wen, M. Borges Mota, & A. McNeill, (Eds.), Working memory in second language acquisition
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
591
592
TI MOTHY MCCOR M ICK AN D CR I STI NA SANZ
and processing: Theory, research and commentary (pp. 224–247). Multilingual Matters. Tagarelli, K. M., Ruiz, S., Moreno, J. L. & Rebuschat, P. (2016). Variability in second language learning: The roles of individual differences, learning conditions, and linguistic complexity. Studies in Second Language Acquisition (Special Issue): Cognitive Perspectives on Difficulty and Complexity in SLA, 38(2), 293–316. Unsworth, N., & Engle, R. W. (2005). Working memory capacity and fluid abilities: Examining the correlation between operation span and raven. Intelligence, 33, 67–81. VanPatten, B. (1990). Attending to form and content in the input. Studies in Second Language Acquisition, 12, 287–301. Waters, G. S. & Caplan, D. (1996). The measurement of verbal working memory capacity and its relation to reading comprehension. The Quarterly Journal of Experimental Psychology, 49(A), 51–79. Weissheimer, J., & Borges Mota, M. (2009). Individual Differences in Working Memory Capacity and the Development of L2 Speech Production. Issues in Applied Linguistics, 17, 34–52. Wen, Z., Biedroń, A., & Skehan, P. (2017). Foreign language aptitude theory: Yesterday, today, and tomorrow. Language Teaching, 50, 1–31. Williams, J. N. & Lovatt, P. P. (2003). Phonological memory and rule learning. Language Learning, 53, 67–121. Zalbidea, J. & Sanz, C. (2020). Does learner cognition count on modality? Working memory effects on early L2 morphosyntactic attainment across oral and written tasks. Applied Psycholinguistics, 41(5), 1171–1196.
Notes 1 Frost et al. (2013) found that verbal WM was not related to L2 reading development for L1-English L2-Hebrew learners, but visual statistical learning ability was. The authors did not include any visuospatial WM measures, but their findings motivate further research in this area (also see Leeser & Herman, this volume, on reading comprehension and WM). 2 The studies highlighted in this chapter followed different statistical reporting procedures, so we have followed a reviewer’s recommendation and interpreted the effect sizes on behalf of the reader: For Pearson’s r: .25 as small, .40 medium, and .60 large (Plonsky& Oswald, 2014), and extending their guidelines (p. 17), for R2, .063 as small, .16 as medium, and .36 as large.
https://doi.org/10.1017/9781108955638.032 Published online by Cambridge University Press
27 Working Memory and L2 Sentence Processing Ian Cunnings
27.1
Introduction
Understanding either a native (L1) or nonnative (L2) language in real-time relies on the ability to keep track of who did what to whom during sentence and discourse comprehension. That real-time language processing relies on working memory, broadly defined as the ability to manipulate a limited amount of task-relevant information at one time (Baddeley, 2007; Cowan, 2017), is not contested, and it is clear that storing and accessing information from memory is crucial for successful comprehension (Jäger et al., 2017; Lewis et al., 2006; Vasishth et al., 2019). As such, it is not surprising that L2 researchers are interested in the role that working memory plays in determining outcomes in L2 acquisition and processing (e.g., Linck et al., 2014; Shin, 2020). However, different conceptualizations of working memory make different predictions about how working memory should influence language comprehension. In this chapter, I discuss how these different conceptualizations make different predictions about how working memory may influence L2 processing. I begin below by outlining different approaches to working memory during sentence processing, before discussing existing research that has examined how working memory may influence L2 processing. I end by discussing the importance of considering how task-based differences may influence conclusions that can be drawn about working memory and individual differences in L2 processing.
27.2
Capacity-Based and Interference-Based Approaches to Working Memory during Sentence Processing
A key distinction in different accounts of working memory is the extent to which there is hypothesized to be a distinct component dedicated to working memory. According to multicomponent theories (e.g., Baddeley,
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
594
IAN CUNNINGS
2007), there is a capacity-limited working memory component distinct from long-term memory. Other accounts assume there is no dedicated working memory, and instead describe working memory as the activated part of long-term memory (e.g., Cowan, 2017). Other accounts describe working memory in terms of the ability to allocate attention to taskrelevant information (e.g., Engle, 2002). These different approaches make different predictions about the role of working memory during L2 sentence processing. Below I discuss two different approaches in turn.
27.2.1
Capacity-Based Approaches to Working Memory during L2 Processing A dominant approach to working memory during L2 processing is the capacity-based view, originally proposed in the L1 processing literature (Daneman & Carpenter, 1980) and extended to L2 acquisition by Harrington and Sawyer (1992). Such accounts, which are broadly compatible with multicomponent views, assume that there is a dedicated working component that has a capacity that differs between individuals, which in turn leads to individual differences in language comprehension. Applied to the L2 context, capacity-based approaches predict that L2 comprehension is slower and more effortful than L1 comprehension, taxing the capacitylimited working memory component to a greater extent in L2 as compared to L1 processing. Such accounts would predict that L2 sentence processing difficulty may be neutralized for L2 learners with high enough L2 memory capacity, or alternatively that L1 processing may become error-prone if taxed in such a way as to mimic the capacity-limitations of L2 processing (McDonald, 2006). A contested issue in capacity-based approaches to sentence processing is how capacity is defined. Capacity could be described in terms of the number of words and/or phrases that an individual can maintain in working memory at one time or, alternatively, especially in the case of syntactic ambiguity resolution, capacity limitations may constrain how many interpretations of an ambiguous input a reader can consider in parallel at one time, or how many different sources of information a reader may be able to consult at one time to guide sentence processing (Daneman & Carpenter, 1980; Just et al., 1996; Just & Carpenter, 1992). Just and Carpenter (1992) claimed that individual differences in memory capacity influence sentence processing such that high-capacity readers are able to consult discourse and pragmatic information to guide syntactic ambiguity resolution and consider multiple parses of an ambiguous input, while low-capacity readers may focus on the syntactically simplest parse and have difficulty utilising discourse-based information to guide ambiguity resolution. Applied to L2 processing, such accounts would predict that capacitylimited L2 learners may maintain fewer words/phrases in working memory at one time than L1 speakers, which presumably may impact on L2
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
L2 Sentence Processing
learners’ ability to process sentences with nonadjacent constituents, as in the case of different linguistic dependencies. If the ability to take different sources of information, such as discourse context, into consideration during processing is capacity limited (Just & Carpenter, 1992), capacitybased approaches would also predict that L2 learners may have difficulty utilizing discourse context to guide sentence processing. However, L2 learners have been shown to be sensitive to how discourse-level information guides syntactic ambiguity resolution (Pan et al., 2015; Pan & Felser, 2011; Pozzan & Trueswell, 2016). Indeed, the work of Pan and colleagues has suggested that for at least certain types of syntactic ambiguity, L2 learners may be more sensitive to discourse-related biases than L1 speakers. For example, Pan et al. (2011) examined temporary syntactic ambiguities in which a prepositional phrase modified either a preceding verb (e.g., Bill glanced at the customer with strong suspicion) or a preceding noun (e.g., Bill glanced at the customer with ripped jeans). The prior discourse created interpretive bias for either verb or noun modification. L2 learners’ reading times of the critical sentences were faster for verb-modifying sentences in verb-biasing contexts, and faster for noun-modifying sentences in noun-biasing contexts. Although L1 readers demonstrated sensitivity to this discourse bias in an offline comprehension task (as did the L2 group), they did not show sensitivity to it during online reading. These findings would be unexpected under a view in which purportedly capacity-limited L2 readers have difficulty taking multiple sources of information into account during real-time sentence processing.
27.2.2
Interference-Based Approaches to Working Memory during L2 Processing In contrast to capacity-based models, other accounts do not posit a separate working memory component, but instead distinguish between items in memory and those in the focus of attention. Working memory, in this case, is characterized as the ability to allocate attention to bring task-relevant information in and out of the focus of attention (e.g., Engle, 2002). Such accounts focus on how memory representations are encoded, stored, and retrieved during the completion of cognitive tasks. Cue-based parsing (Lewis et al., 2006; Lewis & Vasishth, 2005; McElree et al., 2003; Van Dyke & Johns, 2012) provides a framework for how memory representations are encoded and retrieved from memory during sentence processing. To illustrate how memory encoding and retrieval influences language processing, consider anaphora resolution from the perspective of cue-based parsing. In (1), representations of each word and phrase in the sentence will be encoded as each are encountered during incremental processing. Reaching the pronoun “him” will initiate a memory-retrieval operation to find an antecedent for the pronoun. In cue-based parsing, memory retrieval is achieved by matching a set of retrieval cues against all items in memory
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
595
596
IAN CUNNINGS
in parallel, with the item that provides the best match being retrieved. A pronoun such as “him,” as in (1a), will initiate retrieval using a set of cues including, but not limited to, a cue for the current discourse topic (e.g., [+TOPIC]), and a masculine antecedent (e.g., [+MASC]), along with cues related to syntactic constraints on anaphora resolution. As memory retrieval involves matching cues against all items in memory in parallel, items that partially match a set of retrieval cues may be accidently retrieved causing interference. In (1a), the retrieval cues as described above uniquely identify “the boy” as the antecedent to be retrieved, but in (1b) “the man” provides a partial match as it is [+MASC]. Cue-based parsing predicts that partially matching items in memory may sometimes be retrieved, causing interference during processing. As interference is based on the similarity between a set of retrieval cues and the items in memory, this is called similarity-based interference. Cue-based parsing thus predicts that how information is retrieved from memory, especially in terms of similaritybased interference, is a key determinant of successful language comprehension (e.g. Lewis et al., 2006; Van Dyke & Johns, 2012; Vasishth et al., 2019). Note that although I focus here on memory encoding and retrieval operations during the processing of different linguistic dependencies, such as in anaphora resolution, in typical instantiations of cue-based parsing (e.g. Lewis & Vasishth, 2005), memory encoding and retrieval operations occur at each word in the sentence. (1a) The boy said that the lady spoke to him last night. (1b) The boy said that the man spoke to him last night. From this perspective of working memory during sentence processing, L1/ L2 differences can be explained in terms of how L2 learners encode and retrieve information from memory. Cunnings (2017) proposed that L2 learners may be more susceptible to similarity-based interference as a result of how they weight memory retrieval cues. Specifically, L2 learners may weight discourse-based cues more heavily than L1 speakers. Note that there are different instantiations of cue-based parsing (e.g., Parker et al., 2017; Vasishth et al., 2019). Simplifying somewhat, some recent debate has centered around whether particular findings index differences in the speed versus accuracy of a retrieval operation during sentence processing as a result of interference. It is beyond the scope of this chapter to discuss these issues here. Suffice to say that a key difference between capacity-based and interference-based approaches is that capacity-based views predict that a key determinant of successful sentence processing is the quantity or number of types of information in working memory at one time, while the interference-based account focuses instead on the content of information in memory. Another possibility is that both capacity and interference influence processing, although one of the two factors may still play a larger role in determining outcomes in L2 acquisition and processing.
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
L2 Sentence Processing
27.2.3
Teasing Apart Capacity-Based and InterferenceBased Approaches It can sometimes be difficult to tease apart capacity-based and interferencebased accounts of sentence processing because capacity and interference are often confounded. Consider a so-called filler-gap dependency as exemplified in (2), where the displaced noun “the book” needs to be interpreted as the direct object of the verb “read” for comprehension to be successful. Both capacity-based and interference-based approaches would predict that the short dependency in (2a) should be easier than the longer dependency in (2b). The capacity-based view would explain this in terms of the amount of information that needs to be maintained at one time being larger in (2b) than (2a). However, as dependencies become longer, the number of constituents that could potentially lead to interference will also naturally increase, in line with interference-based accounts. For example, the inclusion of another “readable” noun (“the newspaper”) in (2b) should lead to increased interference in comparison to (2a). Such interference would not be predicted in (2c), where the dependency is equally as long as (2b) in terms of the number of words or phrases, but importantly the additional noun (“the coffee”) is not plausibly “readable.” How the capacity-based view would predict differences between sentences like (2b) and (2c) is less clear, as the amount of information that needs to be maintained in both cases is similar, while it naturally follows that (2b) may cause difficulty in comparison to (2c) in the interference-based view, as the content of the sentences differ. (2a) John saw the book that the man read happily. (2b) John saw the book that the man with the newspaper read happily. (2c) John saw the book that the man with the coffee read happily. The capacity-based and interference-based approaches make different predictions about how individual differences should influence working memory during sentence processing. According to the capacity-based view, capacity-based limitations in L2 sentence processing should restrict the amount or types of information that an L2 reader can consult at one time. A key issue from this perspective is how memory capacity is assessed. This is typically assessed through administration of complex span tasks, where participants must remember a list of items while completing a secondary task (see, e.g., Conway et al., 2005; Mathy et al., 2018). In L2 sentence processing, this is most often operationalized in terms of performance on variants of the reading-span task, in which participants read a series of sentences and must remember the final word of each sentence for later recall (e.g., Daneman & Carpenter, 1980; Harrington & Sawyer, 1992). The number of sentences, and thus the number of words to recall, usually varies from two to five, and participants often also have to engage in a secondary task, such as grammaticality or plausibility judgments of the presented
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
597
598
IAN CUNNINGS
sentences. The number of words recalled is typically taken as the dependent measure, with a higher recall rate being taken as evidence of a larger working memory capacity. Note that the task is administered in different variants, for example it is sometimes tested in the L1 or L2 (or both), and is also not always scored in the same way, which may impact potential correlations between reading-span scores and other variables (e.g. Shin, 2020). I return to this issue in Section 27.4. Alternatively, according to the interference-based account, L2 sentence processing is predicted to be dependent on how L2 learners encode and retrieve information from memory during processing. One way in which individual differences may be explained in this case is in terms of individual differences in the quality of representations in memory. Van Dyke et al. (2014), for example, who studied individual differences in L1 sentence processing, argued that individuals with larger receptive vocabulary knowledge encode higher quality representations in memory that are less susceptible to interference. This proposal is in ways similar to some claims in the L2 processing literature, where Hopp’s (2018) Lexical Bottleneck Hypothesis predicts that successful L2 sentence processing is reliant on the quality of L2 lexical representations. Another source of individual variation from the perspective of cue-based parsing comes from how individuals weight different cues during memory retrieval (Vasishth et al., 2019). As noted above, Cunnings (2017) argued L2 learners may weight discourse-based retrieval cues more heavily during processing than L1 speakers. This hypothesis would predict that a primary determinant of L2 processing ability is the amount and type of linguistic experience an L2 learner has in the L2. Both robust lexical representations and the appropriate weighting of different memory retrieval cues require adequate linguistic experience of the relevant L2. This importance of linguistic experience in determining individual differences in L2 sentence processing parallels debate in the L2 acquisition literature regarding the roles of language proficiency and L1 transfer in L2 acquisition. L2 acquisition, from the perspective of cue-based parsing, involves learning to weight retrieval cues in an appropriate manner, while transfer can be described in terms of how an L2 learner’s L1 may influence how different L2 learner groups weight memory retrieval cues during L2 processing. More generally, accounting for individual differences in L2 sentence processing in this way can help provide greater integration between research on real-time L2 sentence processing and L2 acquisition.
27.3
Memory Capacity and Interference in L2 Sentence Processing
I now turn to existing research that bears on how memory capacity and interference may influence L2 sentence processing. For memory capacity,
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
L2 Sentence Processing
I focus on existing research that has examined the interaction between scores on the reading-span task and different aspects of sentence processing, before discussing research on anaphora resolution as a case study on how interference influences L2 sentence processing.
27.3.1
Reading Span and Relative Clause Attachment in L2 Sentence Processing Studies examining relative clauses in L2 processing have tested sentences like (3). (3a) is ambiguous, as the relative clause (“who injured himself”) can modify either the first noun phrase (NP; “the brother”), called high attachment, or the second (“the man”), called low attachment. A large literature has examined attachment preferences, and while results have been mixed, L1 English speakers typically prefer low attachment, while preferences in other languages differ (e.g. Carreiras & Clifton, 1993; Cuetos & Mitchell, 1988; Hemforth et al., 2015). Whether L2 speakers demonstrate L1-like attachment preferences has been widely debated (e.g. Felser et al., 2003; Hopp, 2014; Papadopoulou & Clahsen, 2003; Witzel et al., 2012). (3a) James saw the brother yesterday afternoon. (3b) James saw the sister yesterday afternoon. (3c) James saw the brother yesterday afternoon.
of of of
the the the
man
who
hurt
himself
man
who
hurt
himself
lady
who
hurt
himself
Individual differences in L2 attachment resolution have been examined in both offline comprehension tasks and online reading time experiments. In an offline task testing ambiguous sentences, Hopp (2014) reported a negative correlation between reading span, tested in the L2, and high attachment in L2 learners of English, with higher L2 span readers preferring low attachment. This finding has recently been replicated by Cheng et al. (2021), who found this pattern in both L1 and L2 English readers (with the reading-span task again being administered in English). Similar findings have also been reported in some previous L1 studies (e.g., Swets et al., 2007). However, the finding that lower-span readers chose high attachment more frequently than higher span readers is perhaps unexpected from a capacitybased view, which most obviously would predict that low-span readers should prefer the linearly closer NP. This would predict a low attachment preference in low-span readers, that is, the opposite pattern to what has been observed. Swets et al. argued that these results may reflect how lowerand higher-span readers chunk sentences into prosodic units, with different chunking strategies leading to different NPs in the relative clause becoming more salient. Whatever the source of this effect, the fact that
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
599
600
IAN CUNNINGS
similar findings have been observed in both L1 and L2 comprehension suggests a similar underlying cause in both L1 and L2 processing. Less consistent results have been observed in online studies. Kim and Christianson (2017) tested ambiguous relative clauses in Korean L2 learners of English in a self-paced reading experiment. Participants were tested on relative clauses in both their L1 (Korean) and L2 (English), and completed a reading-span task in their L1. Kim and Christian found longer reading times for individuals with higher L1 reading spans, a pattern that was found in both the readers’ L1 and L2. They interpreted this as indicating that highspan readers are more likely to consider both possible attachment sites in parallel, leading to competition. This would be consistent with claims that individuals with higher memory capacity are able to consider multiple interpretations of an ambiguous sentence at one time (Just & Carpenter, 1992). Note that as Kim and Christianson only tested ambiguous sentences in their study, and did not compare reading times of ambiguous sentences to those disambiguated to either low or high attachment, as in (3b) and (3c), respectively, this interpretation of their results is difficult to distinguish from one in which participants with higher L1 reading-span scores were simply slower, more careful readers overall. Cheng et al. (2021) did not report any significant effects of L2 reading span in an eye-tracking study that tested both ambiguous sentences like (3a), and sentences disambiguated to either low or high attachment as in (3b/c), respectively. Hopp (2014) also did not report any significant effects of L2 reading span in an eyetracking study of L2 readers testing sentences disambiguated to either low or high attachment. In sum, while reading span appears to be correlated with attachment preferences in offline tasks, whether reading span influences online processing of relative clauses during L2 comprehension is less clear. Note that cross-study comparisons are complicated in this case, as while Cheng et al. (2021) and Hopp (2014) tested reading-span scores in the L2, Kim and Christianson (2017) tested L1 reading spans.
27.3.2
Reading Span and Filler-Gap Dependencies in L2 Sentence Processing Individual differences in filler-gap dependency resolution have also been investigated. Dallas et al. (2013) examined filler-gap dependencies as in (4), which manipulated whether a nonadjacent noun was a plausible (“which player”) or implausible (“which football”) direct object of the verb “threatened”, using event-related brain potentials (ERPs). Implausible sentences were expected to yield the N400 ERP component (Kutas & Hillyard, 1980). L1 English speakers displayed the expected N400 effect, as did higher proficiency but not lower proficiency Chinese L2 learners of English. Participants completed three memory span tasks (alphabet span, subtract
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
L2 Sentence Processing
two span, and English reading span), but a composite memory span score did not predict ERP effects in the L2 group. (4)
The umpire asked which player/football the coach threatened before the game.
Other studies examining filler-gap dependencies using various different tasks have also not found significant effects of reading span. Felser and Roberts (2007) did not find any significant effects of L2 reading span in a cross-modal priming paradigm that investigated filler-gap dependency resolution during L2 listening comprehension, while Miller (2014) did not find any significant effects of L2 reading span in a reading comprehension task that also involved picture verification. Juffs (2005) examined filler-gap dependencies in a self-paced reading experiment but did not find any significant effects of either (L1 or L2) reading span or word span. In contrast to these null effects, Dussias and Piñar (2010) found that only L2 learners with high L2 reading spans behaved like L1 speakers in their self-paced reading experiment. However, Dussias and Piñar tested sentences like (5), where the displaced filler “who” must be interpreted as the subject of “killed” for successful comprehension, but it may initially be interpreted as the direct object of “know” during incremental processing. As such, these results might be related to reanalysis following temporary syntactic ambiguity, rather than effects related to filler-gap dependency resolution per se. (5)
Who did the policeman know killed the pedestrian?
While these results might indicate a specific effect of L2 reading span during reanalysis, given uncertain results in other aspects of ambiguity resolution, as discussed in Section 27.3.1, further research is required here to assess the replicability of this finding. In sum, although Dussias and Piñar reported an interaction between L2 reading span and (a particular aspect of ) filler-gap dependency resolution, other studies have not consistently found correlations between span scores and different sentence processing measures indexing L2 processing of filler-gap dependencies.
27.3.3
Reading Span and Morphosyntactic Agreement in L2 Sentence Processing A large literature has examined L2 processing of morphosyntactic agreement (for review, see Cunnings, 2017). It is beyond the scope of this chapter to provide a comprehensive overview, and instead I focus on studies that have examined how the length of agreement dependencies interacts with reading-span scores. Studies have examined how length influences the processing of L2 agreement based on the hypothesis that longer dependencies presumably impose greater demands on working memory than shorter ones.
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
601
602
IAN CUNNINGS
In an eye-tracking during reading study, Keating (2010) examined nounadjective gender agreement in L1 Spanish speakers and English L2 Spanish learners. He compared reading times for sentences such as (6), in which gender marking on an adjective (“abierta”/”abierto” and “rosado”/”rosada”) must agree with the sentence subject (“la tienda” and “el vestido”). Sentences contained either grammatical or ungrammatical gender marking on the adjective, and length was also manipulated by including both short (6a) and long (6b) conditions. The long condition contained an additional noun (“la muchacha”) intervening between the adjective and sentence subject. (6a) La tienda está abierta/∗abierto los sábados y domingos por la tarde. “The store-FEM is open-FEM/∗open-MASC Saturdays and Sundays in the afternoon.” (6b) El vestido de la muchacha es rosado/∗rosada y tiene lunares blancos. The dress-MASC of the girl-FEM is pink-MASC/∗pink-FEM and has white polka dots.” L1 readers had longer reading times for ungrammatical sentences irrespective of length, while L2 readers as a group showed this grammaticality effect for short dependencies only. Reading-span scores, assessed in the participants’ L1, however, positively correlated with the grammaticality effect in L2 readers only, with higher-span learners showing larger grammaticality effects in both short and long conditions. Two other studies are also relevant here. Coughlin and Tremblay (2013) used self-paced reading to examine sensitivity to number agreement violations during the processing of French clitics in L1 and L2 readers. L2 participants completed a reading-span task in both their L1 (English) and L2 (French). Using a length manipulation in a similar vein to Keating (2010), Coughlin and Tremblay found that high-proficiency, but not intermediateproficiency, L2 learners had longer reading times for ungrammatical than grammatical clitics in short and long conditions. This grammaticality effect was numerically larger in the short condition, however, and there was a tendency for L2, but not L1, reading-span scores to interact with grammaticality effects, though the relevant correlations were only marginally significant. Foote (2011) also reported a self-paced reading experiment that tested noun-adjective gender agreement and subject-verb number agreement in L1 and L2 Spanish speakers, again comparing short and long dependencies. Both groups showed grammaticality effects in short and long conditions, with differences between grammatical and ungrammatical sentences being smaller in the long conditions for both gender and number agreement. L2 reading-span scores, however, did not significantly correlate with the size of these grammaticality effects. In sum, there have been mixed results in terms of whether reading-span scores predict L2 sensitivity to agreement violations. L2 learners do,
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
L2 Sentence Processing
however, more consistently show grammaticality effects in shorter rather than longer conditions across the studies discussed above. This might be taken as evidence of a capacity-based disadvantage in L2 readers. However, in the studies above, the long conditions always include an additional intervening element that may interfere in the agreement dependency. In (6b), for example, the intervening noun (“la muchacha”) matches the gender of the adjective in the ungrammatical condition. As such, it is difficult to tease apart a capacity-based explanation from an interferencebased one based on these results. One way to tease these issues apart would be to systematically manipulate the agreement properties of the intervening element in the long dependencies, as cue-based parsing would predict that the properties of this constituent should influence the extent to which readers are susceptible to interference (for discussion, see Cunnings, 2017). Further research is required here to tease apart a capacity-based and interference-based account of these L1/L2 differences.
27.3.4 Interference in L2 Sentence Processing The results in the preceding sections highlight inconsistencies in whether reading-span scores interact with L2 sentence processing, and some results are confounded between capacity-based and interference-based accounts. I now turn to further discussion of how interference can provide an explanation for some findings in the L2 processing literature. For reasons of space, I discuss here briefly two relevant studies on the processing of reflexives. For in-depth review, see Cunnings (2017). Felser et al. (2009) examined the processing of reflexives in an eyetracking during reading study with L1 English speakers and L2 learners with L1 Japanese. They tested sentences as in (7). In (7), the only grammatical antecedent for the reflexive “himself” is “Richard,” while the gender of an ungrammatical potential antecedent has also been manipulated. In (7a), this ungrammatical antecedent (“John”) matches the gender of the reflexive, while in (7b), it mismatches (“Jane”). (7a) John noticed that Richard had cut himself with a very sharp knife. (7b) Jane noticed that Richard had cut himself with a very sharp knife. Although L2 learners demonstrated nativelike understanding of reflexives in an offline task, Felser et al. found longer reading times at the reflexive for L2, but not L1, readers when the ungrammatical antecedent matched the gender of the reflexive. This finding might be unexpected from a capacitybased perspective, given the ungrammatical antecedent is linearly more distant from the reflexive than the grammatical antecedent, but it follows naturally from an interference-based account. From this perspective, reflexive resolution involves retrieving an antecedent from memory. Retrieval is
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
603
604
IAN CUNNINGS
achieved by matching a set of retrieval cues against all items in memory. Retrieval cues will include structural cues that guide retrieval to the grammatical antecedent, but importantly also morphosyntactic agreement features, in this case [+MASC]. The longer reading times for L2 learners in (7a) may thus index interference from the discourse-prominent, gender-matching sentence subject. Felser and Cunnings (2012) reported L2 difficulty of a slightly different nature in their eye-tracking study of reflexives in L1 English speakers and German L2 English learners. They tested sentences similar to (7) but manipulated gender congruency between the reflexive and both the grammatical and ungrammatical antecedent. When readers first encountered the reflexive, L1 readers had longer reading times when the grammatical antecedent mismatched the gender of the reflexive, while L2 readers showed longer reading times when the ungrammatical antecedent mismatched the reflexive’s gender. L2 learners demonstrated nativelike understanding of reflexives in an offline task however, and their reading times were influenced by the gender of the grammatical antecedent, like L1 readers, in reading times for regions of text after the reflexive (i.e., in spillover processing). In sum, although the precise pattern of results in Felser and Cunnings (2012) and Felser et al. (2009) differ, they both suggest L2 difficulty during processing as a result of L2 learners’ temporary consideration of discourse-prominent, but ungrammatical, antecedents during reflexive resolution. Note that whether anaphora resolution in L1 processing is impervious to interference is debated (Dillon et al., 2013; Jäger et al., 2020), and interference clearly influences L1 processing more generally (for review, see Jäger et al., 2017). These L2 results are compatible, however, with the claim that a source of L1/L2 differences during processing is how the different populations weight different cues to memory retrieval (Cunnings, 2017). This discussion is not intended to be an exhaustive overview of how interference may influence L2 anaphora resolution (see Cunnings, 2017). However, it illustrates how interference may influence L2 processing, and highlights how consideration of the content of information in memory, in this case potential antecedents for an anaphor, rather than merely the amount of information in memory, can help inform our understanding of L2 processing. That L2 processing may be influenced by the content, rather than mere amount, of information in memory is predicted by interferencebased accounts that focus on memory encoding and retrieval, rather than capacity, during sentence processing. Although the studies discussed above may suggest a role of interference during L2 processing, further research is clearly required here to disentangle potential roles of memory capacity and interference in L2 anaphora resolution more broadly. Whether capacity or interference can explain L1/L2 differences in other linguistic dependencies also requires further systematic examination.
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
L2 Sentence Processing
27.4
Quantifying Individual Differences in L2 Sentence Processing
As discussed above, there have been inconsistent results in terms of individual differences in reading span and L2 sentence processing. At least some of this inconsistency may be related to the sample sizes tested in existing research, as these have varied across studies. Individual differences research requires large samples, and further research with large L2 samples is required here. Another difficulty in assessing the (in)consistency of these results is that the reading-span task has not been administered and scored in a systematic way across studies. Additionally, the extent to which there is indeed consistent individual variation in the psycholinguistic tasks used to examine L2 sentence processing has not been systematically examined. Below, I discuss the importance of considering these two issues when assessing the influence of working memory in L2 sentence processing.
27.4.1
Measuring and Characterizing Individual Differences in L2 Reading Span Inconsistency in the administration and scoring of reading-span tasks makes it difficult to systematically compare results across L2 studies, as the different findings may at least in part be due to different ways in which the reading-span task is administered and scored. Shin (2020) conducted a meta-analysis of the effects of reading span on L2 reading comprehension, and although she did not examine L2 sentence processing, her results are illustrative of this issue. Shin found that differences in how the task was administered, for example, in which language (the L1 or L2) the task was tested, in whether words needed to be recalled in the order in which they were remembered, in whether there was a secondary task, in the length of the sentences used, and also in the scoring method used all influenced the size of any potential correlation between reading span and L2 reading comprehension scores. Shin’s results indicate the need for standardization in the administration and scoring of the reading-span test when assessing the role that reading span may play in L2 comprehension. It is less clear how such issues may influence L2 sentence processing, as they have not been systematically examined in L2 processing research, but it is likely that at least some of the inconsistences in the literature may be due to different ways in which the reading-span task has been administered and scored across studies. For example, the language in which the reading-span task was administered was not consistent in the studies discussed in Sections 27.3.1–27.3.3. Some studies tested L1 reading span, some L2 reading span, and some both. The studies also did not use consistent scoring methods. These differences make it difficult to draw conclusions about how reading span may influence L2 sentence processing across studies.
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
605
606
IAN CUNNINGS
Additionally, how to interpret interactions between reading-span scores and sentence processing is contested. Although reading-span scores are often interpreted as indexing memory capacity, whether the reading-span task measures individual differences in capacity rather than individual differences in language experience has been debated (Kidd et al., 2018; MacDonald & Christiansen, 2002). MacDonald and Christiansen (2002), for example, argued that the reading-span task is a test of language processing ability with its own set of task demands, rather than an index of processing capacity, and Farmer et al. (2017) recently found that linguistic experience influenced reading-span scores in L1 speakers. The extent to which L2 linguistic experience influences L2 scores in the reading-span task has also been debated (Juffs & Harrington, 2011). Juffs and Harrington suggested that one way to alleviate this issue is to create a composite reading-span score based on L2 learners’ performance on reading-span tasks administered in both their L1 and L2. Additionally, a number of researchers (e.g., Waters & Caplan, 2003) have advocated calculation of composite span scores based on performance across several complex span tasks, including those that do not rely so heavily on language such as operation span (Turner & Engle, 1989), rather than assessing memory capacity using a single task. The rationale is that a composite score will provide a better estimate of an individual’s memory capacity that is not as confounded with performance in a single task. Further research is required here to tease apart the extent that these issues can explain the mixed results related to correlations between reading span and L2 sentence processing, and whether purported differences in L2 memory capacity can be explained in terms of individual differences in L2 linguistic experience. It is also important to note that differences in how psycholinguistic tasks are administered may also influence results. Recall that Kim and Christianson (2017) found an interaction between reading span and L2 processing of ambiguous relative clauses, whereas Cheng et al. (2021) did not. In addition to the fact that Kim and Christianson tested L1 reading span while Cheng et al. tested L2 reading span, another difference between the two studies is that Kim and Christianson used self-paced reading, whereas Cheng et al. adopted eye-tracking during reading. Selfpaced reading, which typically does not allow rereading of earlier parts of a sentence, may place different demands on working memory than eyetracking during reading, where rereading is possible. Note also that while both experiments required participants to answer comprehension questions about the sentences they read, only Kim and Christianson’s comprehension task probed interpretation of the relative clause ambiguity itself. These task-related differences make it difficult to draw direct comparisons between the two studies, and further research is required to examine how such task demands may influence individual differences in L2 sentence processing.
https://doi.org/10.1017/9781108955638.033 Published online by Cambridge University Press
L2 Sentence Processing
27.4.2 Measurement Reliability and Individual Differences A final issue to consider when assessing individual differences, be it in reading span or some other measure, is whether widely used cognitive tasks systematically measure individual differences to begin with (Hedge et al., 2018; Parsons et al., 2019). For an experimental task to measure individual differences, there must be systematic individual variation in the task to begin with, and to meaningful correlate two measures, such as reading-span scores and a measure of sentence processing, there must be systematic individual variation in both tasks. The systematicity with which a task measures individual differences is assessed by measurement reliability statistics, such as split-half reliability (for discussion, see Parsons et al., 2019). In split-half reliability, each participant’s data from a task is randomly split in half, and the two halves correlated. A correlation of .7 or above is considered consistent enough for a task to be used as a measure of individual differences (Nunnally, 1978). While this is not a strict cut-off, low measurement reliability is detrimental to statistical inference (Parsons et al., 2019). In Shin’s (2020) meta-analysis of reading span and L2 reading comprehension, measurement reliability was only reported in a minority of the studies surveyed. Although it is difficult to quantify how often reliability measures are reported for studies examining the role of reading span or other individual differences in L2 sentence processing, as the relevant meta-analysis has not been conducted, reliability measures are not routinely reported in L2 sentence processing research (Cunnings & Fujita, 2020). Cunnings and Fujita examined individual differences in a self-paced reading study that tested temporarily ambiguous and unambiguous sentences as in (8a) and (8b), respectively. Longer reading times are expected at “played” in (8a), as “the cat” may initially be interpreted as being conjoined with “the dog” as the direct object of “washed” when it is in fact the subject of “played.” (8b) is an unambiguous control in which the temporary ambiguity does not arise due to use of the conjunction “while” instead of “and.” (8a) Ken washed the dog and the cat in the garden played with a ball. (8b) Ken washed the dog while the cat in the garden played with a ball. Although whole group analyses indicated the expected differences in reading times between (8a) and (8b) in L1 and L2 readers, and split-half reliabilities for measures of overall participant reading speed were high (> .9), split-half reliabilities for the crucial ambiguity effect (the ambiguous/unambiguous difference) were low ( L2 French). Although L1 and L2 WM measures may be strongly correlated, they do not always equally predict or correlate with L2 reading comprehension. Shin’s meta-analysis revealed a stronger WM-L2 reading comprehension relationship when the RST was administered in the L2 (r = .35) than in the L1 (r = .17). On the one hand, this finding makes sense given that an L2 RST and L2 text comprehension both involve L2 reading. However, the other differences among studies complicate further interpretation. In one noteworthy study that examined the effects of design features of the RST on the relationship between WM and L2 reading, Alptekin et al. (2014) created four versions of the RST, manipulating task language (L1 Turkish, L2 English) and type of sentence judgment (semantic plausibility, morphosyntactic plausibility). Each version of the RST yielded two scores – recall of sentence final words and accuracy of (morphosyntactic or semantic) judgments. Comprehension was measured via a score on the Nelson-Denny Reading Test (Brown et al., 1993). Correlation analyses were performed between reading comprehension scores and the storage (recall) or processing component (sentence judgments) of the four RST versions.
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
621
622
MICHAEL J. LEESER AND ERIC HERMAN
The following RST components significantly correlated with L2 reading (.20 < r < .28): recall scores on L1 and L2 RSTs (semantic judgments), processing score on L1 RST (semantic judgments), and processing score on L2 RST (morphosyntactic judgments). Although these relationships between WM components and L2 reading comprehension are relatively weak, exploratory factor analysis on RSTs revealed interesting findings regarding the structure and underlying relationships among the measures from the various RSTs. Three meaningful factors were identified and accounted for 73 percent of the total variance of RST scores. The recall scores from all four RSTs loaded on the same factor and accounted for 40.27 percent of the variance, suggesting that storage operates independent of both language and task orientation. The processing scores from three tasks (L1 and L2 semantic judgments, L2 morphosyntactic judgments) loaded onto the same factor and accounted for 22.4 percent of the variance. Finally, processing scores from L1 morphosyntactic judgments loaded onto a separate factor, accounting for 10.41 percent of the variance. These findings suggest that the storage and processing functions of WM are language-independent factors when the processing component of the RST involves making semantic judgments; however, if the task requires grammaticality judgments, the processing function of WM depends on the task language.
28.5
Methodological Considerations for Comprehension Assessment
Acknowledging the multidimensional nature of reading comprehension processes and products, the question becomes what to measure and how. Research investigating the role of WM in L2 reading has most often used offline measures and focused on the product of comprehension. The two most widely used measures are recall protocols and multiple-choice (MC) tests. Shin’s (2020) meta-analysis revealed a stronger overall WM-L2 reading comprehension relationship when comprehension was assessed via free recall (r = .35) than via a multiple-choice test (r = .23). The reason for this difference can partly be understood by differences in the nature of these two assessment tasks. Nonetheless, the relationship between WM and L2 reading was not constant in all of the studies that assessed comprehension with a similar response format. Just as we suggested of measures of working memory, this inconsistency is, in part, a by-product of the numerous ways in which any one type of comprehension measure is designed and scored.
28.5.1 Multiple-Choice Tests MC tests offer a number of methodological advantages. These include ease of scoring and the targeting of specific information, such as main ideas,
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
Methodological Issues in Research
details, and various types of inferences. In addition, because MC tests are recognition tasks, the questions and options provide cues from which text representations can be retrieved. Therefore, they may be more sensitive to products of comprehension that cannot be retrieved without such cues. However, MC test scores do not exclusively reflect reading comprehension. Selecting from among options requires a degree of problem-solving and potentially involves the application of test-taking strategies or simply guessing. Perhaps the most serious challenge to test designers is to come up with distractors for each item, which, based on their relative plausibility, impact test difficulty (Drum et al.,1981). In a similar vein, the way in which readers answer initial questions may influence how they answer later questions in order to logically fit a story line or argument (see, e.g., Alderson, 2000, for other issues of test design). Although a handful of studies using MC tests have provided information about the different types of items, in most cases scores to the various question types are aggregated and a single, composite comprehension score is reported. Thus, there is no way of knowing the performance at different levels and types of comprehension. This is particularly the case with standardized tests. The Nelson-Denny and the TOEFL3 tests have been popular choices among the studies examining WM and L2 English comprehension. Yet, it has been repeatedly observed that standardized tests are “developed with a psychometric rationale and do not reflect our understanding of comprehension processes” (Kintsch & Kintsch, 2005, p. 86). For instance, analyses of questions used in standardized reading comprehension tests (e.g., Gates-MacGinitie and Nelson-Denny) revealed that standardized tests are not measuring the reader’s ability to construct a coherent textbase and situation model (Magliano et al., 2007; Miller et al., 2006). Furthermore, developing question items based on item analysis statistics can lead to creating questions that focus on peripheral and trivial information (Johnston, 1984). An additional issue with standardized tests, which is generally true of MC tests, is that question-answering occurs with access to the texts. One argument for text access is that taking away the texts makes it a memory test and underestimates what has been comprehended. Of course, it is expected that the sooner the test follows the reading, the less the effect of memory (Alderson, 2000). With text access, readers can search and reread. As McNamara et al. (2007) note, “The reader’s memory resources should be at least partially relieved if the text is made available when the reader answers the inference questions” (p. 238). However, this may have unexpected consequences for central questions. Johnston (1984) found that when the text was available during testing, readers did worse on central questions and on questions that required use of background knowledge. The proposed explanation was that when readers used search strategies with text access, they found some overlap between text content and the distractors, which made selecting the correct answer more difficult.
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
623
624
MICHAEL J. LEESER AND ERIC HERMAN
Ultimately, the decision to allow access to texts when assessing comprehension depends on the goals of a study and the kind of reading and tasks that researchers want to relate to individual differences in WM. For example, MC tests that instruct readers to select the most accurate statement from among partially correct alternatives while still having access to a text may be more appropriate for investigating an aspect of “deeper comprehension,” such as a reader’s ability to take a critical stance on the accuracy of information (Graesser, 2015). Among the few studies that compare the relationship of WM to different aspects of comprehension on MC tests (e.g., Alptekin & Erçetin, 2009, 2010, 2011; Choi, 2013; Park et al., 2016), the distinction typically made is between literal and inferential questions. The results of this research are conflicting. Studies have found everything from no differences in how WM relates to different question types (Park et al., 2016) to only finding a moderate relationship with literal questions (Choi, 2013) to the opposite finding in which only inferential questions have a small to moderate relationship with WM (Alptekin & Erçetin, 2009, 2010, 2011). Methodological differences are one likely source of the inconsistent relationships. The length and number of texts, the time given for reading, the types of inferences, and whether the text was available during testing are just some of the ways in which these studies varied.
28.5.2 Recall Protocols When it comes to recall protocols, there are various decisions about test design and scoring to keep in mind. First, instructions to participants can impact what and how much participants recall. More main ideas are recalled when the instructions are to write a summary than when instructions are to write everything that can be recalled (Riley & Lee, 1996). In terms of the various aspects of comprehension, summary writing primarily tests participants’ ability to recall and organize central information from the text, whereas writing anything that can be recalled includes more peripheral elements. In both cases, the text can be the primary source, which makes recalls “more text-base dependent” (McNamara et al., 1996, p. 19). Second, the language of the recall (L1 vs. L2) can also affect performance. Beginning and intermediate learners produce less information on recalls written in the L2 than in the L1 (Lee, 1986; Wolf, 1993). Lastly, there are different ways of scoring recalls. As an example, both Leeser (2007) and Fontanini and Tomitch (2009) counted the number of core and modifier propositions, but only Leeser gave half a point for partially correct propositions. However, there are additional scoring methods for L2 researchers to consider, including counting complex propositions (Kintsch, 1998), weighting the different idea units (Bernhardt, 1991), and counting pausal units (Johnson, 1970), whereby someone reads the text aloud and the natural pauses that are made mark the boundaries between units.
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
Methodological Issues in Research
Table 28.1 Comparison of L2 reading-WM studies that used written summary protocols Study
# Texts Read (# words) Text Type
Recall Language
Instructions
Correlation (r)
NR
L1 Portuguese: .50 L1 Chinese: .59
L1
Write as much as possible
Familiar topic: .43b Unfamiliar topic: .22
L1 or L2
Fill blanks with Loweras many words intermediate: .49 as wished Upperintermediate: ns
Fontanini & Tomitch (2009)
2 (911, 952) Expository L2
Leeser (2007)
2 (185-189)
Walter (2004)
8 in L1, Narrative 8 in L2 (124-150)
Narrative
a
Note. NR = not reported. ns = not significant. a Language not reported, but L2 assumed, because the participants were of two different L1s. b The effect sizes (Cohen’s d) of the differences between the high and low WM groups were transformed to r (Borenstein et al.,2009).
Because free recalls are completed in the absence of cues and distractors, as well as without text access, they have been considered the purest means of measuring comprehension (Bernhardt, 1991). Despite this advantage, recall/ summary protocols are an underused assessment tool in WM-L2 reading research. In fact, only three published studies from Shin’s (2020) meta-analysis used some kind of written summary task: free written recall (Fontanini & Tomitch, 2009; Leeser, 2007) or summary completion task (Walter, 2004). Yet, as shown in Table 28.1, differences among these studies exist on a number of dimensions, including text length, text type, recall instructions, and recall language. These factors, among others, undoubtedly contribute to the disparities in the relationship between WM and L2 reading, which ranged from weak to strong, according to Plonsky and Oswald’s (2014) guidelines for interpreting correlations. Furthermore, the findings of Leeser (2007) and Walter (2004) suggest two additional methodological features are of critical importance – topic familiarity and L2 proficiency of the participants. Summary tasks are not without limitations. Two drawbacks are that they provide less evidence of inferencing and give participants the chance to mask or leave out what they have not fully comprehended (Hammadou Sullivan, 2002). Thus, if an objective of research is to determine the relationship between WM and comprehension of specific inferences, then a viable option is to use a cued recall task (i.e., open-ended questions or summary completion tasks), as long as researchers verify, via piloting, that questions cannot be answered without reading the text(s) (e.g., Shibasaki et al.,2015). Although the problem-solving associated with MC tests is mitigated because cued recalls do not contain
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
625
626
MICHAEL J. LEESER AND ERIC HERMAN
distractor options, there still exists the possibility that readers will reason and guess an answer based on what they have comprehended (Magliano et al., 2007). That said, we believe cued recalls overcome many of the pitfalls associated with MC tests and free recalls, and they offer a promising avenue of exploring the role of individual differences in WM for specific types of comprehension.
28.6
Additional Considerations
Of the studies included in Shin’s (2020) meta-analysis, none directly measured the readers’ lexical coverage of the texts. The ability to select the context-specific meanings of words is fundamental to proposition encoding, and lexical knowledge is an important source for making connections between propositions and updating the situation model (Perfetti & Stafura, 2014). Having to infer the meanings of words during reading (texts and WM tasks) places an additional burden on WM. Given that word recognition is fundamental to comprehension and high levels of comprehension require knowledge of 98 percent or more of the words in texts (e.g., Herman & Leeser, in press; Schmitt et al., 2011), controlling for L2 readers’ lexical knowledge or employing a measure of lexical knowledge should be part of future research designs. To date, no WM-L2 reading studies have considered the ways in which WM may play a different role in reading for different purposes. Yet, Linderholm and van den Broek (2002) found that RST recall was more strongly related to recall performance of L1 readers when reading to study than when reading for entertainment. When reading for study purposes, low WM readers not only recalled less information than high WM readers, but think-aloud protocol analyses revealed that low WM readers also repeated the text more often and produced fewer comments about what they did and did not understand from the text. Another area into which research needs to expand is the role WM plays in the different levels and dimensions of a reader’s memory for texts. McNamara et al. (1996) was the first study that distinguished L1 comprehension tasks that primarily tapped the textbase or situation model, and Kintsch (2012) stated that reading assessment needs to, at the least, differentiate these two levels of comprehension. Although some L2 reading–WM research has distinguished between literal and inferential comprehension questions, this distinction does not correspond to comprehension at the textbase and situation model levels because inferences may pertain to either level (e.g., Schmalhofer et al., 2002). Moreover, there is good reason to maintain separate measures of different inference types, because of their differential demands on WM. Inferences vary in the extent to which they involve rapid and effortless processing or engage more strategic problemsolving processes, and determining the circumstances under which readers will put in more cognitive effort to build a mental representation of a text is
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
Methodological Issues in Research
a perennial issue in the field of discourse comprehension (e.g., Kintsch, 1993; McKoon & Ratcliff, 1992). An additional observation is that L2 studies have much to gain from incorporating online measures of reading comprehension. Product-oriented measures indicate what text representation readers have constructed (or can reconstruct), but it is more difficult to ascertain comprehension processes. Although mental activity still has to be inferred from online measures, time-sensitive measures provide a more direct account of processing. Online measures also have the potential to elucidate what would otherwise go unnoticed by offline measures. For instance, from eye-tracking and think-aloud data, Rapp et al. (2007) found that struggling L1 readers may obtain similar comprehension scores on product-oriented measures for different reasons. Recent efforts in L1 discourse comprehension using eyetracking have found that WM, as measured by the Swanson sentence span task, did not influence early processing (e.g., first-pass reading times), but did have an effect on later processing (e.g., fewer and shorter regressions) (van Moort et al., 2020). Furthermore, reading times of text-based and knowledge-based inconsistencies were not affected by WM. Clearly, the use of concurrent, online methods to explore the role of WM during L2 comprehension is a fruitful avenue of research. Finally, the vast majority of WM-L2 reading studies, and all of the studies included in Shin’s meta-analysis, utilized the RST. Although there are good reasons to continue using the RST as a processing and storage measure of verbal WM, future research should consider, and motivate appropriately, the use of other WM tasks. For example, in order to gain a broader perspective on the link between a general WM capacity (as opposed to a readingspecific WM) and L2 reading comprehension, future research should employ tasks that do not involve the reading of sentences, such as operation span.
28.7
Practical Recommendations
Burgoyne et al. (this volume) offer a number of recommendations for those interested in conducting individual differences research using cognitive tasks. Their recommendations regarding samples, use of multiple tasks, appropriate data and statistical reporting, and others all apply to research investigating the role of individual differences in WM in L2 comprehension (see Burgoyne et al. for further discussion). In this section we provide five recommendations based on the issues discussed in the previous three sections.
28.7.1
Clarify the Role(s) of WM in the Reading Process(es) and Products Under Investigation Fundamental to any L2 reading-WM study is clarifying the constructs of comprehension and WM. Therefore, we recommend that researchers
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
627
628
MICHAEL J. LEESER AND ERIC HERMAN
explicitly state the framework of comprehension guiding a study and the proposed roles of WM for the targeted level(s) of comprehension. In addition, because readers may have different goals (e.g., to study, to be entertained, etc.) and varying degrees of background knowledge when reading selected texts, we recommend addressing how readers’ goals and familiarity with text content and structure might mediate any role of WM. These issues are crucial for selecting the comprehension and scoring method(s) best suited for the targeted reading process and/or products.
28.7.2
Comprehension Measures Should Reflect the Reading Purpose and Levels of Comprehension We previously discussed the limitations of both multiple-choice and recall tests and noted that a cued-recall task can overcome many of these limitations by narrowing the focus of comprehension and minimizing the issues associated with selected response tasks. However, we also emphasized that the choice of assessment should ultimately be informed by the purpose of reading and the goals of a study. Whichever assessment method is selected, however, we recommend that researchers specify the comprehension level or kind of information captured in comprehension questions and scores. For example, questions that target a specific type of inference are preferable to testing an array of different inference types, unless separate scores for each of the inference types are reported and related to WM. Crucially, because comprehension is not a unitary process, reporting a single “reading comprehension” score is inadequate for understanding of the role of WM in the various component and subcomponent processes involved in reading comprehension.
28.7.3
Control for and Report L2 Lexical Knowledge of Texts and Tasks Readers require a high level of lexical coverage for the sufficient comprehension of texts. In addition, not knowing the meaning of a single word in an RST sentence can (minimally) affect sentence plausibility judgment accuracy and judgment reaction times. Thus, we encourage adequately controlling for lexical knowledge in L2 WM tasks and appropriate data gathering of participants’ lexical knowledge for L2 texts in order to mitigate the intrusion of a confounding variable. Furthermore, researchers can examine whether text coverage mediates the relationship between WM components and comprehension levels of particular texts.
28.7.4 Appropriately Motivate the Language (L1 or L2) of Span Tasks As with most experimental task decisions, the choice of an L1 or L2 version of the RST depends on researchers’ theoretical assumptions and
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
Methodological Issues in Research
experimental hypotheses. That is, if assuming a domain-specific L2 WM (e.g., Alptekin & Erçetin, 2010), then an L2 RST is a logical choice, provided that appropriate control for lexical knowledge is considered. However, if the only assumption is that there is a specific linguistic WM (e.g., Just & Carpenter, 1992), then the language of the RST is less important from a theoretical standpoint. If the researcher assumes WM is domain general and/or wants to test hypotheses regarding the domain specificity of WM, we advocate the use of an L1 and/or L2 RST together with another complex span task, such as operation span. In terms of practical issues, L1- or L2-specific RSTs may not exist or they may be researcher-created without providing any kind of reliability measures. Furthermore, if major syntactic differences exist between the L1 and L2, these differences may make either the processing or recall components of the two versions difficult to compare. For example, Erçetin (2015) notes that sentence final words in the English RST are nouns, but in Turkish they are verbs, thereby creating a comparison difficulty. Are words in one lexical category (i.e., noun, verb, adjective) easier to recall than those in another? Once again, we recommend the use of more than one complex span measure along with reporting the reliability for each.
28.7.5
Consider all Components of Complex Span Tasks and Check for Strategizing Given Leeser and Sunderman’s (2016) finding that the scoring method of a complex span task can alter a study’s outcomes, our recommendation is that researchers report basic descriptive statistics for each component of the span measure (judgment accuracy, judgment speed, recall) and conduct analyses to determine if participants are strategizing (e.g., trading off speed for accuracy). In addition, we recommend examining the relationship between comprehension and complex span components separately and together in order to shed light on theoretical debates surrounding which component or mechanism of WM accounts for greater variability of specific aspects of L2 comprehension.
28.8
Conclusion
In this chapter, we have pointed out that L2 reading comprehension may be more demanding of WM resources than L1 comprehension, yet this has not been reflected in individual differences research on the relationship between WM and L2 reading. We believe there is still much work to be done in this area, and we have identified important methodological issues surrounding both the assessment of WM and reading comprehension in order to advance the research agenda in this domain. Most importantly, we recommend that researchers clearly define the aspect of comprehension
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
629
630
MICHAEL J. LEESER AND ERIC HERMAN
they seek to relate to working memory and select assessment measures for both constructs accordingly. We also reiterate what many researchers before us have recommended (e.g., Alderson, 2000; Wolf, 1993). That is, that researchers seek to triangulate methods, because one assessment task can confirm and complement the findings from another task.
References Adams, R., & Shahnazari-Dorcheh, M. (2014). The relationship between working memory and L2 reading comprehension. Applied Research on English Language, 3, 19–34. Alderson, J. C. (2000). Assessing reading. Cambridge University Press. Alptekin, C., & Erçetin, G. (2009). Assessing the relationship of working memory to L2 reading: Does the nature of comprehension process and RST make a difference? System, 37, 627–639. Alptekin, C., & Erçetin, G. (2010). The role of L1 and L2 working memory in literal and inferential comprehension in L2 reading. Journal of Research in Reading, 33, 206–219. Alptekin, C., & Erçetin, G. (2011). Effects of working memory capacity and content familiarity on literal and inferential comprehension in L2 reading. TESOL Quarterly, 45, 235–266. Alptekin, C., Erçetin, G., & Özemir, O. (2014). Effects of variations in RST design on the relationship between working memory capacity and second language reading. Modern Language Journal, 98, 536–552. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47-89). Academic Press. Bernhardt, E. B. (1991). Reading development in a second language: Theoretical, research and classroom perspectives. Ablex. Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2009). Converting among effect sizes. In M. Borenstein, L. V. Hedges, J. P. T. Higgins, & H. R. Rothstein (Eds.), Introduction to meta-analysis (pp. 45–49). John Wiley & Sons. Brown, J. I., Fishco, V. V., & Hanna, G. S. (1993). Nelson-Denny Reading Test Form G. Riverside Publishing. Choi, S. (2013). Working memory capacity, vocabulary knowledge, and reading comprehension of EFL learners. English Education Research, 25, 25–42. Cowan, N. (2017). The many faces of working memory and short-term storage. Psychonomic Bulletin Review, 24, 1158–1170. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. Daneman, M., & Merikle, P. M. (1996). Working memory and language comprehension: A meta-analysis. Psychonomic Bulletin & Review, 3, 422–433.
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
Methodological Issues in Research
Drum, P. A., Calfee, R. C., & Cook, L. K. (1981). The effects of surface structure variables on performance in reading comprehension tests. Reading Research Quarterly, 16, 486–514. Erçetin, G. (2015). Working memory and L2 reading: Theoretical and methodological issues. ELT Research Journal, 4, 101–110. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102, 211–245. Ericsson, K. A., & Simon, H. A. (1993). Protocol analysis: Verbal reports as data. MIT Press. Fontanini, I., & Tomitch, L. M. B. (2009). Working memory capacity and L2 university students’ comprehension of linear texts and hypertexts. International Journal of English Studies, 9, 1–19. Friedman, N. P., & Miyake, A. (2005). Comparison of four scoring methods for the reading span test. Behavior Research Methods, 37, 581–590. Glanzer, M., Dorfman, D., & Kaplan, B. (1981). Short-term storage in the processing of text. Journal of Verbal Learning and Verbal Behavior, 20, 656–670. Grabe, W. (2009). Reading in a second language: Moving from theory to practice. Cambridge University Press. Graesser, A. C. (2015). Deeper learning with advances in discourse science and technology. Policy Insights from the Behavioral and Brain Sciences, 2, 42–50. Hammadou Sullivan, J. A. (2002). Advanced foreign language readers’ inferencing. In J. A. Hammadou Sullivan (Ed.), Literacy and the second language learner (pp. 217–238). Information Age Publishing. Herman, E., & Leeser, M. J. (in press). The relationship between lexical coverage and type of reading comprehension in beginning L2 Spanish learners. Johnson, R. (1970). Recall of prose as a function of the structural importance of linguistic units. Journal of Verbal Learning and Linguistic Behavior, 9, 12–20. Johnston, P. (1984). Prior knowledge and reading comprehension test bias. Reading Research Quarterly,19, 219–239. Just, M., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122–149. Kendeou, P., & O’Brien, E. J. (2018). Reading comprehension theories: A view from the top down. In M. F. Schober, D. N. Rapp, & M. A. Britt (Eds.), The Routledge handbook of discourse processes (2nd ed., pp. 7–21). Routledge. Kintsch, W. (1993). Information accretion and reduction in text processing: Inferences. Discourse Processes,16, 193–202. Kintsch, W. (1998). Comprehension: A paradigm for cognition. Cambridge University Press. Kintsch, W. (2012). Psychological models of reading comprehension and their implications for assessment. In J. P. Sabatini, E. R. Albro, &
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
631
632
MICHAEL J. LEESER AND ERIC HERMAN
T. O’Reilly (Eds.), Measuring up: Advances in how to assess reading abilities (pp. 21–38). Rowman & Littlefield Education. Kintsch, W., & Kintsch, E. (2005). Comprehension. In S. G. Paris & S. A. Stahl (Eds.), Children’s reading comprehension and assessment (pp. 71–92). Erlbaum. Koda, K. (2005). Insights into second language reading: A cross-linguistic approach. Cambridge University Press. Krashen, S. (2004). The power of reading: Insights from the research (2nd ed.). Heinemann. Lee, J. F. (1986). On the use of the recall task to measure L2 reading comprehension. Studies in Second Language Acquisition, 8, 201–212. Leeser, M. J. (2007). Learner-based factors in L2 reading comprehension and processing grammatical form: Topic familiarity and working memory. Language Learning, 57, 229–270. Leeser, M. J., & Sunderman, G. (2016). Methodological implications of working memory tasks for L2 processing research. In G. Granena, D. O. Jackson, & Y. Yilmaz (Eds.), Cognitive individual differences in second language processing and acquisition (pp. 89-104). John Benjamins. Linck, J. A., Osthus, P., Koeth, J. T., & Bunting, M. F. (2014). Working memory and second language comprehension and production: A metaanalysis. Psychonomic Bulletin & Review, 21, 861–883. Linderholm, T., & van den Broek, P. (2002). The effects of reading purpose and working memory capacity on the processing of expository text. Journal of Educational Psychology, 94, 778–784. Lorch, R. F. (2015). What about expository text? In E. J. O’Brien, A. E. Cook, & R. F. Lorch (Eds.), Inferences during reading (pp. 348-361). Cambridge University Press. Magliano, J., Radvansky, G. A., & Copeland, D. E. (2007). Beyond language comprehension: Situation models as a form of autobiographical memory. In F. Schmalhofer & C. Perfetti (Eds.), Higher level language processes in the brain: Inference and comprehension processes (pp. 379–391). Erlbaum. McKoon, G., & Ratcliff, R. (1992). Inference during reading. Psychological Review, 99, 440–466. McNamara, D. S., & Magliano, J. (2009). Toward a comprehensive model of comprehension. Psychology of Learning and Motivation, 51, 297–384. McNamara, D. S., de Vega, M., & O’Reilly, T. (2007). Comprehension skill, inference making, and the role of knowledge. In F. Schmalhofer & C. A. Perfetti (Eds.), Higher level language processes in the brain: Inference and comprehension processes (pp. 233–251). Lawrence Erlbaum Associates. McNamara, D. S., Kintsch, E., Songer, N. B., & Kintsch, W. (1996). Are good texts always better? Interactions of text coherence, background knowledge, and levels of understanding in learning from text. Cognition and Instruction, 14, 1–43. Millis, K., Magliano, J., & Todaro, S. (2006). Measuring discourse-level processes with verbal protocols and latent semantic analysis. Scientific Studies of Reading, 10, 225–240.
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
Methodological Issues in Research
Osaka, M., & Osaka, N. (1992). Language-independent working memory as measured by Japanese and English reading span tests. Bulletin of the Psychonomic Society, 30, 287–289. Osaka, M., Osaka, N., & Groner, R. (1993). Language-independent working memory: Evidence from German and French reading span tests. Bulletin of the Psychonomic Society, 31, 117–118. Park, H., Nam, K., & Lee, Y. S. (2016). The role of reading span in factual and inferential comprehension and retention in L2 reading. Linguistic Research, 33, 81–106. Perfetti, C., & Stafura, J. (2014). Word knowledge in a theory of reading comprehension. Scientific Studies of Reading, 18, 22–37. Plonsky, L., & Oswald, F. L. (2014). How big is “big”? Interpreting effect sizes in L2 research. Language Learning, 64, 878–912. Rapp, D., & van den Broek, P., McMaster, K., Kendeou, P., & Espin, C. (2007). Higher-order comprehension processes in struggling readers: A perspective for research and intervention. Scientific Studies of Reading, 11, 289–312. Riley, G. L., & Lee, J. F. (1996). A comparison of recall and summary protocols as measures of second language reading comprehension. Language Testing, 13, 173–189. Schmalhofer, F., McDaniel, M. A., & Keefe, D. (2002). A unified model for predictive and bridging inferences. Discourse Processes, 33, 105–132. Schmitt, N., Jiang, X., & Grabe, W. (2011). The percentage of words known in a text and reading comprehension. The Modern Language Journal, 95, 26–43. Shibasaki, H., Tokimoto, S., Ono, Y., Inoue, T., & Tamaoka, K. (2015). English reading comprehension by Japanese high school students: Structural equation modeling including working memory and L1 literacy. Open Journal of Modern Linguistics, 5, 443–458. Shin, J. (2020). A meta-analysis of the relationship between working memory and second language reading comprehension: Does task type matter? Applied Psycholinguistics, 41, 873–900. Unsworth, N., & Engle, R. W. (2007). On the division of short-term and working memory: An examination of simple and complex span and their relation to higher order abilities. Psychological Bulletin, 133, 1038–1066. Unsworth, N., & Spillers, G. J. (2010). Working memory capacity: Attention control, secondary memory, or both? A direct test of the dual-component model. Journal of Memory and Language, 62, 392–406. Van Heuven, W. J. B., Dijkstra, T., & Grainger, J. (1998). Orthographic neighborhood effects in bilingual word recognition. Journal of Memory and Language, 39, 458–483. van Moort, M. L., Koornneef, A., & van den Broek, P. W. (2020). Differentiating text-based and knowledge-based validation processes during reading: Evidence from eye movements. Discourse Processes, 1–20. VanPatten, B. (2020). Input processing in adult L2 acquisition. In B. VanPatten, G. D. Keating, & S. Wulff (Eds.), Theories in second language acquisition: An introduction (3rd ed., pp. 105–127). Routledge.
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
633
634
MICHAEL J. LEESER AND ERIC HERMAN
Walter, C. (2004). Transfer of reading comprehension skills to L2 is linked to mental representations of text and to L2 working memory. Applied Linguistics, 25, 315–339. Waters, G., & Caplan, D. (1996). The measurement of verbal working memory capacity and its relation to reading comprehension. Quarterly Journal of Experimental Psychology, 49, 51–79. Wolf, D. (1993). A comparison of assessment tasks used to measure FL reading comprehension. The Modern Language Journal, 77, 473–489. Zwaan, R. A. (2004). The immersed experiencer: Toward an embodied theory of language comprehension. In: B. H. Ross (Ed.), The psychology of learning and motivation (Vol. 44, pp. 35–62). Academic Press. Zwaan, R. A., & Radvansky, G. A. (1998). Situation models in language comprehension and memory. Psychological Bulletin, 123, 162–185.
Notes 1 See Cowan (2017) for a review of various conceptualizations and definitions of WM. 2 For an additional view on the role of long-term memory retrieval in complex span tasks, see Unsworth and Engle (2007) and Unsworth and Spillers (2010). 3 An anonymous reviewer pointed out that some L2 reading research studies utilize TOEFL practice tests from test preparation books rather than versions of the TOEFL developed by Educational Testing Services. We agree with this reviewer that researchers need to be explicit about the source of the test used.
https://doi.org/10.1017/9781108955638.034 Published online by Cambridge University Press
29 Working Memory and Second Language Speaking Tasks Peter Skehan 29.1
Introduction
The main focus of the present chapter is the relationship between working memory and performance on second language tasks. Prior to addressing this major area, two sections provide scene-setting information. The first provides brief coverage of issues in first and second language speaking. The second outlines what second language tasks are like, how they contrast with tasks in mainstream working memory research, and the role such tasks play in second language research.
29.2
First and Second Language Speaking
From a number of models of speaking which exist, Levelt’s (1989, 1999) account is chosen here, because it has been the model of first language speaking that has had the greatest influence on the second language domain. The basic model posits three general stages in speech production: Conceptualization, Formulation, and Articulation. The first, Conceptualization, is concerned with the underlying ideas that motivate speech, and implicates general knowledge, knowledge of the current situation and discourse flow, notions of stance and focus, and, importantly, the propositions the speaker wishes to express. It outputs a preverbal message to the Formulator, which, in turn, gets on with lexical/lemma selection, syntax building, and early articulatory preparation. Finally, the Articulator stage wrestles with the problem of encoding in sound the lexical and syntactic elements that have been activated. In addition to these three broad stages, there is potential for Monitoring of performance but only at particular stages within the speech production process. An important aspect of the first language model is that the different stages are encapsulated, and function in parallel. The output of one stage is passed on to the
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
636
PETER SKEHAN
next, which then works on that input while the “supplying” stage gets on with the next part of the message. The result is that there is a flow of operations that can sustain continuous speech. In first language speaking, each stage, generally, makes reasonable demands on the next. Most aspects of speech production are produced without undue pressure on attentional resources, and below the level of consciousness. This follows from the previous analysis – working within reasonable constraints means that a flow of speech is possible, and the mechanisms that underlie it do not intrude into consciousness (Skehan, 2018). One aspect of this account is worth highlighting. The Conceptualizer, as we have seen, outputs the preverbal message, and the Formulator then engages in processes of lemma retrieval, from the mental lexicon, and syntax building. The first language mental lexicon is well equipped to handle the demands that are made upon it, through the richness of the information held in lemmas (meaning, phonological form, morphosyntactic implications, association with other lemmas, etc.) and, importantly, speed of operation. The mental lexicon is able to handle these demands, in real time, and without any undue need to create a burden for working memory resources. Lemma-based information is placed in an assembly buffer, and then can be operated upon for syntactic, articulatory, and discourse implications. As a result, parallel communication proceeds (Kormos, 2006). The key issue we need to consider next is to see how second language speech production is different to first language speech production, and how limitations in working memory size and operation might have a different role to play. It has been argued (Skehan, 2014b) that overall, the same stages and processes apply in the second language case, and so Conceptualization, Formulation, and Articulation function broadly similarly, and that monitoring, as a process, is still important. Modularity, cycles of communication, encapsulation, and parallel processing are the ideal, in the same way as the first language case. Perhaps such processing goals are closest to reality with the Conceptualization stage. Bearing in mind that the output of this stage is the preverbal message, what leads to this output need not implicate linguistic processes extensively. The second language speaker can assess situations, organize ideas, make judgments about stance, and, potentially, do this as effectively as a native speaker. The problems are likely to occur at later stages. We saw that the assumption in first language speech production is that the mental lexicon is adequate for the demands that the Conceptualizer routinely makes of it. This situation is radically different in the second language case (although, of course, as proficiency increases, the disparity decreases). The second language mental lexicon (SLML) will be generally smaller, and in addition, lemmas may be much less rich in the information that they have stored. It is also likely that there will be much less formulaic language available, of
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
the sort that eases processing pressures during speaking. The entire system is also likely to be slower (Skehan, 2018). Kormos (2006) proposes that the second language speaker has to rely more on a declarative knowledge system. The processing implications are considerable, as are the consequences for working memory operation. The fundamental assumption, discussed above, that stages make “reasonable” demands on succeeding stages may not hold. So, while the Conceptualizer is getting on with the next cycle of communication, in the expectation that resources will be available to formulate it when needed, the Formulator may be forced to siphon off more attention and take more time to handle lexical and syntactic demands. More time may also be needed to handle deficiencies in that lexicon, as a message needs to be reformulated. The result will be that that attention is not freely available for concurrent Conceptualization (or Articulation) and indeed, signals may be received by the Conceptualizer that a reworked message is required, with consequent compensation, circumlocution, and diversion. First language speech production, as we have seen, is largely a parallel process. The limitations of the SLML, particularly at lower levels of proficiency, mean that second language speech production is often a serial process (Kormos, 2006). Indeed, the speaker may actually need to move in and out of parallel processing, recovering from problems but then being derailed again. Rhythm in speech production is disrupted and in more extreme cases, messages may simply need to be abandoned or radically reshaped. An additional important point is that the nature of monitoring is different in the second language case. First language monitoring may focus on infelicities of style, dealing with superficial error, or modifying things like stance. These factors are important, too, in the second language case. But in addition, there is the problem of error, and the way the second language speaker may be focused (or not) on detecting mistakes and correcting them. This may require more working memory resources than would be case with first language speech production.
29.3
The Role of Tasks in Second Language Performance
Second and foreign language instruction have undergone considerable changes in recent decades. An approach that emphasized control, habit formation, and copious quantities of drilling language patterns has been replaced by much more communicative approaches with a major emphasis on conveying meaning. One example of this is what is known as a taskbased approach (Crookes & Gass, 1993; Willis & Willis, 2007), more strongly influenced by language acquisition theories, and more likely to be linked to empirical studies (Long, 2015; Ellis et al. 2020). A task, in this context, is an activity that provides meaningful and probably unpredictable language use
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
637
638
PETER SKEHAN
(in contrast to a drill, which uses controlled, manipulated language, such as “change these verbs to the past tense”). Examples of tasks are discussing, in pairs, what sort of advice to give the writers of letters to a magazine Agony Aunt, or, also in pairs, comparing one another’s family trees, or (singly) devising a story based on a series of pictures. Such tasks are used for pedagogy, to help extend students’ repertoires in a language, and also for research, of the sort covered in this chapter. They contrast, in this respect, with “tasks” as used in much working memory research, where the term is used to refer to controlled elicitation devices useful for investigating specific hypotheses. Underlying the task-based approach are two propositions. The first is that acquisition and development are more likely if naturalistic learning processes are given scope to operate, as when learners are given tasks that push them to transact meanings and extend interlanguage (Long, 2015). It is assumed that tasks, by their interactive nature, will generate feedback, both implicit and explicit (An & Li, this volume). The second proposition is that it is not enough to learn about language and grammar, but that it is vital to develop skills in using language, in developing an ability for use, and a capacity to mobilize underlying resources for effective communication (Skehan, 2018). Second language task research has become a hotly researched area of second language acquisition, with many active lines of research. These include the effects on performance of different task characteristics, the effects of different task conditions, and of task processes, such as interaction. A focus on tasks within the second language domain has provoked a reevaluation of the relevance of working memory and its impact on task performance. We saw that for there to be change in an underlying language system, the learner must come up against problems that stretch that interlanguage system and possibly receive timely and relevant feedback that is relevant to restructuring and change in the system (Ellis et al., 2020). In such cases the learner has to notice that relevant feedback is being provided and to notice that there is a need for change. There is also a need for long-term memory to be accessed, and some connection made between the contents of current working memory and the longer-term language store. At a minimum, therefore, within a Baddeley-type working memory architecture (Baddeley, 2012), there will be important activity in the phonological buffer, in the central executive, and in episodic memory. An and Li (this volume) focus on interaction and working memory, so links in that area will not be developed further here. We also need to consider second language spoken task performance itself, and how it is characterized and assessed. The basic point here is that language performance has been shown to be multidimensional (Skehan & Foster 1997; Tavakoli & Skehan, 2005). Three broad areas have been contrasted: complexity (both structural and lexical), accuracy, and fluency. The three areas show statistical independence from one another, that is,
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
performance in these domains is not simply a reflection of general proficiency (Tavakoli & Skehan, 2005). In addition, an acquisitional link has been proposed, in that complexity is associated with the development of some new aspects of interlanguage, with risk-taking and a greater willingness to use more recently emerged language forms, and the beginning of system change (Skehan, 2013). Next, greater control is achieved over the emerging language, first of all so that error is reduced and accuracy increases (Anderson, 1995). There is still a need for attention at this stage. Then, as proceduralization continues, greater fluency is achieved, as performance becomes more rapid, error is less likely, and there is less need for conscious processing. Importantly, the different areas, complexity, accuracy, and fluency, can be associated with stages within the Levelt model (Skehan, 2018). Conceptualization links to complexity, both structural and lexical, in that developing more complex ideas, and relationships between ideas, is likely to push the learner to need more complex syntactic frames and more varied lexical elements. Formulation, as we have seen, has lexical (lemma-retrieval) and syntactic features. This is the stage when ideas are clothed in language through access to second language mental lexicon resources. The key issues become the availability of relevant lexical resources and then the ease with which lemma implications for syntax are feasible. It can be argued that the more effectively this is done, the more likely it is that error will be avoided and accuracy enhanced, and then that speed of access to SLML resources enables effective real-time communication. So, the Formulator is then associated with accuracy and fluency in performance. Turning to the details of measuring performance, structural complexity is typically measured in terms of subordination and/or an index of the length of clauses (Bui & Skehan, 2018). Lexical complexity is generally measured through a type-token ratio corrected for text length, termed lexical diversity, and a measure capturing the extent to which more difficult (usually defined in terms of low frequency) words are used, and this is termed lexical sophistication (Skehan, 2009). The most widely used measure of accuracy is the proportion of clauses which are error-free (Skehan, 2018). Fluency, in contrast, contains a number of subcomponents, comprising speed, breakdown (reflected in pausing), and repair (reflected in reformulations, false starts, and repetitions) (Tavakoli & Skehan, 2005; Tavakoli & Wright, 2021). Task research has grown enormously in scale over the last 30 years, and a number of themes have emerged that are relevant to working memory. A major distinction is between task characteristics and task conditions (Skehan, 2016). Research into task characteristics has demonstrated the importance of the type of information in a task (personal vs. impersonal; familiar vs. unfamiliar; concrete vs. abstract), its quantity (number of elements, density of information), organization (structured vs. unstructured), the operations carried out on that information (reasoning demands,
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
639
640
PETER SKEHAN
transformation of material, reorganization), and also the relationship of the participants to task demands (personal perspective). All of these have shown an impact on task performance in terms of complexity, accuracy, and fluency (Ellis et al. 2020). In addition, the conditions under which a task is completed have also been shown to be very important. A considerable literature now exists on the effect of preparedness for doing a task (Bui, 2014). Much of this literature is concerned with pretask planning. For example, giving second language speakers ten minutes to prepare for a task (such as a picture-based narrative retelling or an interactive task requiring decision-making) leads to greater language complexity and fluency, but not consistently greater accuracy. Other forms of readiness have also been explored, such as watching videos relevant to the topic of the task (Kim & McDonough, 2011), or dealing with personal areas of expertise (Bui, 2014). There have also been studies into what happens while a task is under way, principally exploring the impact of more relaxed time conditions (Ellis & Yuan, 2005). Then there has been work at the posttask stage. Repeating a task has been shown to have a clear impact in improving performance. Anticipation of the need to do a posttask activity, such as transcribing one’s own performance, has been shown to raise accuracy (Foster & Skehan, 2013, Li, 2014). Two broader accounts of second language task performance have emerged. Robinson (2011b, 2015) has proposed the Cognition Hypothesis and the SSARC Model (Stabilize, Simplify, Automatize, Restructure, Complexify) to account for such performance. This is also known as the Triadic Model, because it comprises Task Complexity factors, Task Conditions, and Task Difficulty. The first of these is most significant theoretically since task complexity is seen as an important driver for language performance and development. Task complexity is influenced by two categories of variable: resource-directing, which propose linkages between task features and linguistic elements, and resource-dispersing, which make links to processing resources. The former category includes characteristics such as time perspective, reasoning demands, and number of elements. The latter involves things like planning and structure. The assumption in this model is that attentional resources are not limited and draw upon multiple resource pools (Sanders, 1998). An alternative account, the Limited Attentional Capacity approach (Skehan, 2014b, 2018) does, contrastingly, assume attentional limitations and then relates these limitations to the way task characteristics and task conditions influence performance. The argument is that often the limitations mean that not all aspects of performance can be raised simultaneously, that there may be a trade-off between them, and that an interesting challenge is to try to overcome these limitations through artful task design and linkage of task features with appropriate task conditions. Working memory size and operations are clearly relevant for the way this is done.
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
29.4
Research into Working Memory and Second Language Speaking Task Performance
Perhaps the first thing to say here is that the database one can draw on relating working memory to second language speaking task performance is not extensive. But there is some research, and it is insightful already. In this section we will, first, cover working memory research linked to task characteristics and then, to task conditions. Next we will focus on the different aspects of performance (CALF), to explore what generalizations emerge about each of these areas separately. Finally, we will look at some methodological issues that need to be taken into account in future research.
29.4.1 Working Memory and Task Characteristics There are ten studies in this area. Five directly attempted to explore Robinson’s (2011b) Cognition Hypothesis. Cho (2018) explored the effect of varying the number of elements in a task, that is, a resource-directing variable. The participants had to defend a choice among five options, such as who should receive a scholarship or who should receive some financial support. In the simpler condition there were only two criteria which needed to be brought to bear, such as degree of financial hardship and level of grades, and in the more complex there were five such criteria. The study also contrasted oral and written modalities. There was mixed evidence regarding the Cognition Hypothesis itself (positive for complexity, negative for accuracy), but there was no evidence that working memory mediated any effects, with no significant correlations between Reading Span and Operation Span measures and CAF scores. Recio (2011) manipulated reasoning demands to be greater or smaller, following the CH prediction that working memory would be more relevant with more complex tasks. Phonological short-term memory (PSTM) correlated with performance in the simple and the complex conditions, contrary to prediction. Albarqi (2019) reports on a study linking working memory to self-monitoring, with a speaking task, at two levels of complexity – single task and dual task, and this at two levels of proficiency. There was limited evidence of WM involvement with the simpler task (weak negative correlations between WM and repetition and lexical repair), and a number of nonsignificant correlations with other aspects of fluency. There were no WM relationships with the more complex condition. Awwad & Tavakoli (2019) used the variable of intentional reasoning to modify complexity. They report positive results regarding the Cognition Hypothesis itself. In contrast there was no relationship between a working memory measure (backward digit span) and structural complexity or fluency, but some low correlations between working memory and accuracy and lexical complexity (0.32 in each case) for the greater intentional reasoning (more complex) condition. Finally, Zalbidea (2017) explored task complexity, operationalized through two levels of
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
641
642
PETER SKEHAN
number of elements/ reasoning demands, and these with a (monologic) argumentative task. In this study, modality (speaking vs. writing) had much more influence than task complexity. Working memory (an O-span task) showed 2 significant correlations (out of 32) with the more complex version of the task. Summarizing these studies, we see a very mixed picture. The CH would predict that WM would be relevant for the more complex condition in each study. This was true, although only to a limited extent, for Awwad and Tavakoli (2019) and Zalbidea (2017). The reverse held with Albarqi (2019), in that there was a WM-simple condition relationship. Recio (2011) reports a linkage at both levels of complexity, while Cho (2018) found no connection between WM and either complexity level. There seems, on this basis, to be little evidence linking the CH to working memory. Five other studies explored particular tasks in relation to working memory, although these were not motivated by any particular model, but rather simply investigated particular variables from the previous task literature. Three explored narrative tasks and two incorporated more interactive tasks. Gilabert & Munoz (2010) examined the relationship of WM with performance on a retelling of Modern Times, which was viewed twice before the retelling, with participants at two proficiency levels. There were no WM correlations with complexity, accuracy, or fluency for the overall sample. However, a correlation of .54 is reported between an L1 reading span measure and lexical diversity for the higher proficiency group. Kormos & Trebits (2011) also explored narrative retellings, with (a) a story told to a six-picture comic strip and (b) a task where six unrelated pictures had to be used to generate a story, so introducing a level of complexity into the research design (suggesting some CH relevance). Results were presented in tabular form, (based on a backward digit span test), with associated ANOVAs. There was no clear relationship between any of the CALF scores and working memory. Two significances (F scores) were generated, but these did not link monotonically with the WM scores. Duran-Karaoz (2020), in a broader, within-subjects study exploring L1-L2 consistencies in fluency based on oral narrative cartoon-retelling tasks, reported little impact for working memory measures. There were correlations between L1 and L2 fluency measures, but WM had no impact on these correlations. It would seem, then, that with narrative tasks, working memory seems to make, at best, a marginal contribution. Georgiadou and Roehr-Brackin (2017) researched the relationship between executive working memory (EWM: two measures, L1 and L2), PSTM and speech rate and hesitations. There were two proficiency levels, elementary and low intermediate, and with an oral interview task based on responses to card-based prompts. The L1-based EWM measure correlated negatively with the number of pauses in the low-intermediate group. There were no significant correlations, for any measure, EWM or PSTM, and speed or repair. The final study is Mota (2003), who used picture description and narrative tasks. The former required detailed description and some
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
analysis, and was allocated two minutes. The latter required an account of a film that had been seen, and from this performance the first two minutes were analyzed. There were strong correlations between working memory and structural complexity, accuracy, speed, and lexical diversity. However, there are some qualifications. There were only 13 participants, and the working memory test was an L2 speaking span, raising the possibility of a proficiency confound. One can say that across the range of studies, this last is somewhat out-of-step in findings. The others have generated at best low and not particularly systematic correlations, and so it seems safer to conclude that working memory does not link with task features, whether these features are derived from the Cognition Hypothesis, or from any previous literature on tasks.
29.4.2 Task Conditions The studies in this section fall into three groups. Some have investigated pretask (or strategic) planning, some have focused on online planning, and one study investigated repetition. Pretask planning is usually operationalized as giving participants time to prepare for a task after they have been given information about the task. On-line planning concerns the time conditions under which a task is done and is usually operationalized as the length of time for task completion, which is sufficient so that little time pressure is involved. Regarding pretask planning, Wen (2015) compared two groups who each had to narrate what happened in a seven-minute video. The control group watched the video and provided a narration immediately. The experimental group watched the video and then had ten minutes to prepare before they actually provided the narration. Wen (2015) used both L1- and L2-based nonword repetition measures for working memory. He reports no significant correlations between CALF performance and the working memory measures for either the planned or the unplanned condition. Guará-Tavares (2013, 2016) provides two related articles on a working memory study, again comparing planners (10 minutes to plan) and nonplanners, who each completed picture-based narratives. A speaking span working memory test was used in an extreme groups design – effectively top and bottom thirds on the WM measure. One study (Guara-Tavares, 2016) was qualitative in nature and focused on what participants said they did during the planning time, rather than their actual performance. Higher WM participants reported that they used the planning time with more focus on metacognitive strategies (rather than cognitive or social/affective strategies). Working with exactly the same participants, Guara-Tavares (2013) reported a more quantitative approach. For the planners, working memory correlated moderately with structural complexity and speed. For the control group, working memory did not correlate with these dimensions of performance, but did correlate highly with accuracy. Finally, Nielson (2013) explored the variables of working
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
643
644
PETER SKEHAN
memory and proficiency. She used picture-guided narratives and measured working memory through two spatial tests. Using a high-low group comparison design (not correlations), working memory did not have an impact on structural complexity or accuracy (although accuracy scores did show a trend). There was a significant effect of working memory on fluency, with the high WM grouping showing faster speech than the low WM group. The sub-literature comparing online with pretask planning has proved interesting. Ahmadian (2012) only investigated an online planning condition, using a 14-minute silent video retelling. With a listening span WM test, he reported significant correlations of working memory with accuracy and fluency performance scores, but no correlation with structural complexity measures. Li and Fu (2016), using a 6-minute video retelling, and measuring WM with an Operation Span test, compared the connection of working memory under pretask and online planning conditions. They reported that working memory correlates with performance for the online planners with accuracy and fluency (breakdown and speed), whereas there were no correlations under the pretask condition. The two studies in this section, then, present a reasonably consistent picture – working memory does not impact on performance after pretask planning, but does relate to performance when performance conditions are more relaxed as regards time pressure. In other words, working memory is not relevant in assisting how what is planned is remembered and then used in a subsequent performance. In contrast, where relaxed time pressure is involved, greater working memory is useful for the way the less pressured conditions are exploited, whether through more time for “on-the-fly” planning of language structure or more effective lemma retrieval and avoidance of error. There is only one study exploring task repetition. Ahmadian (2013), using a silent film retelling, explored the relationship of working memory with the original performance and also with an unanticipated repeated performance, two weeks later. Bui (2014) and Skehan (2014b, 2018) argue that the repeated performance can be regarded as having been prepared for by the original performance, and so has some connections with planning. In Ahmadian’s study there was, in fact, relatively little impact of the repetition condition on the CALF measures. However, what is interesting is that with the original performance, the correlations with working memory were very low. In contrast, in the repeated performance, while structural complexity did not correlate with working memory, accuracy and speed did, with correlations of 0.56 and 0.50, respectively. Intriguingly, therefore, the contribution of working memory is quite selective here.
29.4.3 Generalizations across the CALF Performance Areas The various studies have taken broadly similar approaches to measuring performance through CALF measures, and this allows some wider
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
generalizations to be made about the different performance areas. It is striking that working memory does not usually relate to structural complexity. Other than Mota (2003) and Guará-Tavares (2013) the studies are remarkably consistent and report quite low correlations. Having greater working memory seems not to have an impact on the amount of subordination in speech, or (though the evidence is less extensive) on any tendency to use longer clauses. In contrast, there is a stronger relationship between working memory and accuracy, typically measured by the percentage of clauses that are error-free, a relationship that is more widespread and, on occasion, quite high. Just under half of the studies show a relationship here, with the online planning studies making a major contribution. Fluency, as we saw earlier, is complex, and it has been proposed that we need to consider speed, breakdown fluency, and repair fluency separately. Speed seems to have a similar relation to working memory to that of accuracy – but perhaps a little lower. There is also a relationship with speed in the online planning studies. Equally interesting, the other subdimensions of fluency, pausing and repair, do not show much evidence of a link to working memory, with the slight exception of Li and Fu (2016), and online planning performance. Georgiadou and Roehr-Brackin (2017) and Albarqi (2019) also report slight evidence here – one group only in each case, and limited aspects of fluency. So higher working memory does seem to support slightly faster speech, but does not relate much to interruptions and modifications to that speech. The three performance areas considered so far have presented (contrasting but) clear pictures. That is not true when we consider influences on lexical complexity. Lexical diversity (i.e., length-corrected type-token ratios) is associated with positive results in three studies - Gilabert and Munoz (2010), with a higher proficiency group, Li and Fu (2016) and an online planning group, and Awwad and Tavakoli (2019) with the high Intentional Reasoning condition. In each of these cases, then, one condition within the research design shows an effect on lexical diversity, but the other does not. Then there are two studies where there are negative relationships (Mota 2003; Wen 2015), and two with no evidence of any relationship (DuranKaraoz, 2020, Georgiadou & Roehr-Brackin, 2017). So, we have an unclear picture, and considerable scope for more research questions to be posed – perhaps a tendency for a positive relationship with lexical diversity, but no more than a tendency. There is not enough information to discuss any relationship between WM and lexical sophistication, and so we will have to wait for more research on this point.
29.4.4 Methodological Issues The variety of research design decisions in the different studies renders comparability between studies hazardous. But perhaps there are some points that may be relevant to future work:
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
645
646
PETER SKEHAN
•
The different options in measuring working memory may well not be neutral. For example, backward digit span, and nonword repetition are not associated with strong WM-performance relationships (Awwad & Tavakoli, 2019; Kormos & Trebits, 2011; Wen, 2015). Possibly this is true with Operation Span studies also, but there are some positive examples here, such as Li and Fu (2016). It may be the case that stronger WMperformance relationships emerge when working memory is measured through an appropriate modality (oral-aural in this case) and with a clear language emphasis, bringing together phonological and central executive influences (Ahmadian, 2012; Guara-Tavares, 2013; Li & Fu, 2016). • It is clear that narrative second language speaking tasks predominate. This is understandable but does render generalizing from this narrow database more difficult. We need more studies with other, especially interactive, tasks. • Beyond the predominance of narratives, there are other issues with the input material in the studies. Some studies use picture-based narratives, while others use videos, an issue regarding processing demands and working memory. In addition, instructions can vary regarding the implicit emphasis they put on performance, such as highlighting details and amount of information, compared to producing a meaningful connected narration.
29.5
Theories of Speaking and the Role of Working Memory
We can now recapitulate the relatively few generalizations we now have from the review of the literature: •
working memory appears to have an impact on task performance when time pressure is reduced (cf. the online planning vs. the pretask planning comparison) • supportive task conditions are associated with a role for working memory (the above comparison plus the role of repetition) • task characteristics do not show a clear role for differences in working memory • structural complexity is not particularly influenced by working memory • accuracy and fluency, with supportive conditions, are more likely to show such an influence • the jury is still out on the relationship between working memory and lexical performance These more specific generalizations fit within a much broader claim. If we accept that second language speech, like first language speech, has broadly the three stages of Conceptualization, Formulation, and Articulation, then the major impact of working memory is with areas involving the
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
Formulator, since the SLML is likely to be more limited in nature. This follows from the online planning research, from the effects of task conditions such as repetition, and from the tendency for accuracy and fluency (Formulator-linked aspects of performance) to be influenced by working memory. It appears that working memory comes into play, in some circumstances, to ease and support Formulator operations. The evidence is not strong that working memory helps Conceptualizer operations, as reflected in the general lack of connection between working memory and task complexity variables. It is also interesting to pursue the circumstances working memory does have a positive role in Formulator operations. A key issue here is that variation in working memory size and operation appears to be limited. Miller (1956) famously estimated the magic number as 7, but equally importantly, plus or minus 2. In the first language case, that “minus 2” does not prevent people using language effectively. The narrow range is sufficient for most first language purposes. (Subsequently Miller’s characterisation of 7 units has been revised downward, as with Cowan’s [2001] proposal for a magical number of “4.” The point, however, is not the typical value, 7 or 4, but the lack of functional significance in much first language processing of the variation.) In second language speech, when SLML problems occur, and push the speaker into a serial mode of processing, the “plus 2” normally is neither here-nor-there. Working memory capacity has been exceeded, and there are likely to be serious difficulties as the problem is overcome and normal communication reestablished. In other words, speakers with a higher working memory are probably not at any particular advantage with most of the problems that occur. Essentially, the challenge is to understand the situations when there is scope for higher working memory to make a difference. We can return to the generalizations that do show an impact of working memory – online planning conditions and repetition – for illumination. Recall that the Formulator needs to take Conceptualizer preverbal input and then access the SLML in order to build language. This presupposes that the SLML is sufficiently large, rich, and fast for the task at hand. These qualities might be called into question in the second language case, leading to impairment in performance. This offers a perspective for understanding the role of online planning and repetition. On-line planning, with its more relaxed time conditions, gives a speaker more time for lemma access. If speed of access is the issue, or the problem is lack of resources requiring a replanning to be achieved, the additional time is exploitable more effectively by those with larger working memory. In other words, it appears to be the joint operation of more working memory and less time pressure that enables the SLML to be used more effectively. Repetition works slightly differently. Skehan (2018) argues that the first performance may semiactivate and prime some lemmas. Assuming that the SLML is less accessible than the first language mental lexicon, the issue may
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
647
648
PETER SKEHAN
be that lemmas are sometimes only partially retrieved in a first performance, and so a speaker has to make the most of the level of retrieval that has been achieved. Speaking is continuous, and the luxury of committing more time to achieve deeper and more complete retrieval is not available. But the first performance effectively primes the second performance. The incomplete lemma retrieval from the first performance then becomes a more effective starting point for the second retrieval, and the repeat performance then is likely to be better than the first. This is the generalization that emerges from the growing repetition literature (Bygate, 2001, 2018; Lambert et al., 2017). What is interesting is that Ahmadian (2013) reports a correlation with working memory for a repeated performance but not for the original. It seems that greater working memory does not have an impact under original conditions, but in the repeated performance, the interaction of greater working memory and semipriming leads to improvement. What we see from these results is that the typical amount of variation we see in working memory is not functionally relevant in many second language speaking situations, but when there is appropriate assistance, the advantage of having higher working memory does make a difference. So far, two sources of assistance that working memory can capitalize on have been identified: reduced time pressure in performance and repetition. Undoubtedly there are others, and future research may identify them. In so doing, we will gain a greater understanding of how to support speaking. To recapitulate the previous analysis, though, it appears to be the case that working memory confers advantages at the Formulator, rather than the Conceptualizer stage. Kormos (2006) argues that, with second language learners, the mental lexicon and a larger L2 morphosyntactic system are forms of declarative knowledge, whereas for first language speakers, they are implicit. This enables parallel, modular, encapsulated processing at the different stages of speech production. The second language speaker sometimes attains such a status, particularly as proficiency increases, but there is always the possibility that communication difficulties will make it serial, and not encapsulated, as difficulties cause the different stages to interfere with one another. Working memory, known to be more implicated in explicit processes (Williams, 2015), confers an advantage within this architecture only within certain narrow limits, and this applies principally to the Formulator stage. We turn next to the more theoretical accounts of second language speaking. We saw earlier that the Limited Attentional Capacity (LAC) approach (Skehan, 2018) and the Cognition Hypothesis (Robinson, 2015) each has an important role for working memory. One might have expected, therefore, to see extensive associated research literatures. What is most striking is how little working memory research has been done in relation to the two accounts, including by the authors of the two approaches. With the LAC approach, one might have thought that it would have been revealing to
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
explore whether limited attentional capacity is mitigated by greater working memory or strengthened by low working memory. But this has not been particularly explored in a direct manner. There has been research that impinges on the LAC approach if one draws on the planning literature. Here we have seen a contrast between an effect for working memory with online planning, and repetition, but little evidence linking working memory and pretask planning (Li & Fu, 2016; Nielson, 2013). Unfortunately, the task features highlighted by LAC approaches (information type, task operations, task structure) have yet to be researched in relation to working memory, so no conclusions can be drawn. For quantity of research, the CH has fared better, as we saw earlier. Awwad and Tavakoli (2019), Recio (2011), and Zalbidea (2017) with reasoning, Cho (2018), with number of elements, Albarqi (2019) with a single-dual task comparison, and Kormos and Trebits (2011), with a comparison between a simple picturebased narration and a need to make up a story, all explored task complexity from within a CH framework. Although there were some results which did provide limited support for the CH itself, there was very little systematicity in these results, with one or two studies giving supportive results, other studies showing no impact, and yet others being contrary to prediction. Disappointingly, then, we have yet to see any clear linkage between working memory and task complexity, the central component of the CH.
29.6
Conclusion and Future Research
The previous sections have brought out that in linking working memory to second language tasks, the database is not extensive. So, a first call in this section would simply be that significantly more research is done integrating working memory measures within general second language task studies. This would undoubtedly contribute to more robust generalizations in this area. But there are a number of areas where, even now, we can see there is a need for more focused studies: •
The problem of one-shot studies: There are a number of areas where conclusions rest on individual or just a pair of studies. For example, we saw interesting results reported by Ahmadian (2013) regarding repetition. We urgently need more repetition studies that incorporate WM measures. The same is true for proficiency (Bui, Skehan, & Wang, 2018), which could be important for charting the contribution of working memory. Even in the area of online planning we need more studies. Only Li and Fu (2016) actually compared pretask and online planning using working memory measures. Perhaps another underrepresented area is that of lexis. Under half of the studies covered included measures of lexis, and even within lexis the focus was generally on lexical
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
649
650
PETER SKEHAN
diversity, with little on lexical sophistication. In addition, the findings are mixed, with a wide variety of results reported. There is clearly scope for more research. • Working memory measurement is an important issue. Two broader issues are relevant here (and discussed elsewhere in this volume): the comparative contributions of different components of working memory (storage, attention control, manipulation), and whether working memory is domain-general or domain-specific (DeKeyser & Juffs, 2005). Interestingly, the measures that more often have generated correlations with performance tend to be reading or listening span measures. Less successful have been nonword repetition tests, spatial tests, and operation span tests. The database is small. It would be useful to have more studies that focus on this issue, incorporating different and contrasting measures within the same study. In this regard, second language task research has the potential to make a contribution to general working memory theorizing. In addition, there is scope to use research designs that do not rely solely on correlation (Duran-Karaoz, personal communication). Framing research questions in terms of more experimental designs (such as Kormos & Trebits, 2011) might address the complexity of the area more effectively. • The variable of task itself merits considerably more research, from at least three perspectives: ○ Within the existing database, there is a considerable emphasis on narrative monologic tasks. We need to have data on a more varied collection of tasks, certainly some involving interaction, with, possibly, problem-solving and information gap tasks to complement the range of monologic tasks that have been used. Narrative tasks have the advantage of facilitating greater experimental control, but they do lead to a particular set of influences on working memory operations, with few external pressures. Interactive tasks introduce unpredictability, a need to accommodate, but perhaps different time pressures when an interlocutor is speaking. There may be consequences for working memory from these differences. ○ From a LAC perspective, some task characteristics have been established as having an influence on performance, as mentioned earlier (the nature of the information, operations on information, task structure, etc.). It would be of considerable interest to see if these findings would be mediated by differences in working memory. ○ The Cognition Hypothesis (Robinson, 2015) proposes that greater task complexity is hypothesized to raise complexity and accuracy simultaneously. One wonders whether this prediction might be more appropriate with higher working memory participants, since greater attentional resources would be available to raise both. Research on this would be very useful. So, more broadly, would
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
research linking working memory to all of the resource-directing and resource-dispersing variables, the central variables in the Cognition Hypothesis (Robinson, 2011b). • It has been argued (Skehan, 2016) that the conditions under which tasks are done (planning, repetition, posttask conditions) generate more consistent results with task performance than do variables based on task characteristics (time perspective, number of elements, and so on). Skehan (2018) argues that task conditions link more naturally with the speaking stages outlined in the Levelt (1989) model. It would be interesting therefore to extend the research done with task conditions, and to broaden it in relation to other variables, such as proficiency, and other forms of pretask preparation and readiness. With proficiency, and assuming a larger and more effectively functioning second language mental lexicon, greater working memory might make more of a contribution, rather than processing resources being more easily swamped by more demanding task conditions. With preparation, it would be interesting to see if there is a different connection with working memory in pretask planning (where the challenge is to remember and use what has been planned; Pang & Skehan, 2014), compared to the sort of readiness from, say, previously established knowledge, or having spoken about a task on previous occasions. Other conditions, such as the use of posttasks (Foster & Skehan, 2013; Li, 2014) where it is hypothesized that foreknowledge of a posttask pushes participants to direct attention more to language form, particularly accuracy, would also be useful areas to build in working memory measures to research designs. The assumption is that attention direction is manipulable. Greater working memory might well be an important foundation for such attention direction. • Although already mentioned, it is worth discussing proficiency level further (Bui et al. 2018). Some studies, such as Gilabert and Munoz (2010) did incorporate this variable. But there is considerable scope for more research, perhaps over a greater range of proficiency. It has been argued in this chapter that an important factor in second language task performance is the Formulation stage and the way that the second language mental lexicon is drawn on. Exploring the relationship between working memory, proficiency level, and second language task performance may be means of exploring whether the narrow window for WM relevance is a consequence of level of proficiency, and that, as proficiency increases, the narrowness of this window widens, and working memory effects become more pervasive. Taken together, these various potential research strands will certainly lead to a greater understanding of second language task performance. It is to be hoped that research in this area may in turn make a contribution to our understanding of working memory more generally.
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
651
652
PETER SKEHAN
Acknowledgments The author would like to thank Benjamin Swets, Mohammad Ahmadian, Ghadah Ahmad Albarqi, Zeynep Duran-Karaoz, and Parvaneh Tavakoli for comments on previous drafts of this chapter, and for additional and very helpful material. The author would also like to thank the Editors of the volume.
References Ahmadian, M. (2012). The relationship between working memory capacity and L2 oral performance under task-based careful online planning conditions. TESOL Quarterly, 46, 1, 165–175. Ahmadian, M. (2013). Working memory and task repetition in second language oral production. Asian Journal of English Language Teaching. 23, 37–55. Albarqi, G. (2019). Self-monitoring behaviour of L2 learners: Proficiency level, dual task paradigm and working memory capacity (Doctoral dissertation, University of Reading). Anderson, J. (1995). Learning and memory. John Wiley. Awwad, A. & Tavakoli, P. (2019). Task complexity, language proficiency, and working memory: Interaction effects on second language speech performance. International Review of Applied Linguistics. https://doi.org/10 .1515/iral-2018-0378 Baddeley, A. (2012). Working memory: Theories, models, and controversies. Annual Review of Psychology, 63, 1–30. Bui, H. (2014). Task readiness: Theoretical framework and empirical evidence from topic familiarity, strategic planning, and proficiency levels. In P. Skehan (Ed.). Processing perspectives on task performance (pp. 63–94). John Benjamins. Bui, G. & Skehan, P. (2018). Complexity, accuracy, and fluency. In H. Nassaji (Ed.), TESOL encyclopedia of English language teaching. Wiley. Bui, G., Skehan, P., & Wang, Z. (2018). Task condition effects on advanced level foreign language performance. P. Malovrh & A. Benati (Eds.), Handbook of advanced proficiency in second language acquisition. (pp. 219–238). Wiley. Bygate, M. (2001). Effects of task repetition on the structure and control of oral language. In M. Bygate, P. Skehan, & M. Swain (Eds.). Researching pedagogic tasks: Second language learning, teaching and testing (pp. 23–48). Longman. Bygate, M. (2018). Learning language through task repetition. John Benjamins. Cho, M. (2018). Task complexity, modality, and working memory in L2 task performance. System, 72, 85–98.
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
Cowan, N. (2001). The magical number 4 in short term memory: A reconsideration of mental storage capacity. Behaviour and Brain Sciences, 24, 87–185. Crookes, G. & Gass, S. (1993). Tasks and language learning: Integrating theory and practice. Multilingual Matters. DeKeyser, R., & Juffs, A. (2005). Cognitive considerations in L2 learning. In E. Hinkel (Ed.), Handbook of research: Second language teaching and learning (pp. 437–454). Lawrence Erlbaum. Duran-Karaoz, Z. (2020). An exploratory study of second language oral performance: Fluence and lexical complexity in L1 Turkish and L2 English (Doctoral dissertation, University of Reading). Ellis, R., Skehan, P., Li, S., Shintani, N., & Lambert, C. (2020). Task-based language teaching. Cambridge University Press Ellis, R., & Yuan, F. (2005). The effect of careful within-task planning on oral and written task performance. In R. Ellis (Ed.), Planning and task performance in a second language. (pp. 167–192). John Benjamins. Foster, P. & Skehan, P. (2013) The effects of post-task activities on the accuracy of language during task performance. Canadian Modern Language Review, 69, 249–273. Georgiadou, E., & Roehr-Brackin, K. (2017). Investigating executive working memory and phonological short-term memory in relation to fluency and self-repair behavior in L2 speech. Journal of Psycholinguistic Research, 46(4), 877–895. Gilabert, R. & Munoz, C. (2010). Differences in attainment and performance in a foreign language: The role of working memory capacity. International Journal of English Studies, 10(1), 19–42. Guará-Tavares, M. G. (2013). Working memory capacity and L2 speech performance in planned and spontaneous conditions: A correlational analysis. Trabalhos Em Linguística Aplicada, 52(1), 9–29. Guará-Tavares, M. G. (2016). Learners’ processes during pre-task planning and working memory capacity. Ilha do Desterro, 69(1), 79–94. Kim, Y. & McDonough, K. (2011). Using pretask modelling to encourage collaborative learning opportunities. Language Teaching Research, 15(2), 183–199. Kormos, J. (2006). Speech production and second language acquisition. Lawrence Erlbaum. Kormos, J. & Trebits, A. (2011). Working memory capacity and narrative task performance. In Robinson, P. (Ed.). Second language task complexity (pp. 267–285). John Benjamins. Lambert, C., Kormos, J., & Minn, D. (2017). Task repetition and second language speech processing. Studies in Second Language Acquisition, 39(1), 1670150196. Levelt, W. J. (1989). Speaking: From intention to articulation. Cambridge University Press.
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
653
654
PETER SKEHAN
Levelt, W. J. (1999). Language production: A blueprint for the speaker. In C. Brown & P. Hagoort (Eds.), Neurocognition of language (pp. 83–122). Oxford University Press. Li, Q. (2014). Get it right in the end: The effects of post-task transcribing on learners’ oral performance. In P. Skehan (Ed.), Processing perspectives on task performance (pp. 129–154). John Benjamins. Li, S., & Fu, M. (2016). Strategic and unpressured within-task planning and their associations with working memory. Language Teaching Research, 20, 1–24. Long, M. (2015). Second language acquisition and task-based language teaching. Wiley. Miller, G. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Mota, M. (2003). Working memory capacity and fluency, accuracy, complexity, and lexical density in L2 speech production. Fragmentos, 24, 69–104. Nielson, K. (2013). Can planning time compensate for individual differences in working memory capacity? Language Teaching Research, 18(3), 272–293. Pang, F. & Skehan, P. (2014). Self-reported planning behaviour and second language performance in narrative retelling. In P. Skehan, Processing perspectives on task performance (pp. 95–128). John Benjamins. Recio, M. (2011). The effects of task complexity on L2 oral production as mediated by differences in working memory capacity (Master’s thesis, University of Barcelona). Robinson, P. (Ed.). (2011a). Second language task complexity: Researching the cognition hypothesis of language learning and performance. John Benjamins. Robinson, P. (2011b). Second language task complexity, the Cognition Hypothesis, language learning, and performance. In P. Robinson (Ed.), Second language task complexity (pp 3–38). John Benjamins Robinson, P. (2015). The Cognition Hypothesis, second language task demands, and the SSARC model of pedagogic task sequencing. In M. Bygate (Ed.). Domains and directions in the development of TBLT. (pp. 87–122). John Benjamins. Sanders, A. (1998). Elements of human performance. Lawrence Erlbaum Associates. Skehan P. (2009b). Lexical performance by native and non-native speakers on language-learning tasks. In B. Richards, H. Daller, D. D. Malvern, & P. Meara (Eds.). Vocabulary studies in first and second language acquisition: The interface between theory and application. (pp. 107–124). Palgrave Macmillan. Skehan, P. (2013). Nurturing noticing. In J. Bergsleithner, S. N. Frota, & J. K. Yoshioka (Eds.), Noticing and second language acquisition: Studies in honor of Richard Schmidt (pp. 169–180). National Foreign Language Center. Skehan, P. (2014a). Processing perspectives on task performance. John Benjamins.
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
Second Language Speaking Tasks
Skehan, P. (2014b). The context for researching a processing perspective on task performance. In Skehan, P. (Ed.), Processing perspectives on task performance (pp. 1–26). John Benjamins. Skehan, P. (2016). Tasks vs. conditions: Two perspectives on task research and its implications for pedagogy. In A. Mackey (Ed.), Annual Review of Applied Linguistics, 24, 34–49. Skehan P. (2018). Second language task-based performance: Theory, research, and assessment. Routledge. Skehan, P., & Foster, P. (1997). The influence of planning and post-task activities on accuracy and complexity in task based learning. Language Teaching Research, 1, 185–211. Tavakoli, P., & Skehan, P. (2005). Planning, task structure, and performance testing. In R. Ellis (Ed.), Planning and task performance in a second language (pp. 239–276). John Benjamins. Tavakoli, P. & Wright, C. (2020). Second language speech fluency: From research to practice. Cambridge University Press. Wen, Z. (2015). Working memory in second language acquisition and processing: The phonological/executive model. In Z. Wen, N. Mota, & A. McNeill (Eds.). Working memory in second language acquisition and processing. (pp. 41–62). Multilingual Matters. Williams, J. (2015). Working memory in SLA research: Challenges and prospects. In Z. Wen, N. Mota, & A. McNeill (Eds.), Working memory in second language acquisition and processing (pp. 301–308). Multilingual Matters. Willis, D. & Willis, J. (2007). Doing task-based teaching. Oxford University Press. Zalbidea, J. (2017). “One size fits all”? The roles of task complexity, modality, and working memory capacity in L2 performance. Modern Language Journal, 101(2), 335–352.
https://doi.org/10.1017/9781108955638.035 Published online by Cambridge University Press
655
30 Working Memory in Second Language Interaction Hyejin An and Shaofeng Li 30.1
Introduction
Individual difference variables have received much attention in second language acquisition (SLA). One cognitive variable that has been extensively investigated is working memory (WM), and researchers have paid particular attention to its role in L2 interaction. A general assumption is that WMC plays a significant role during L2 interaction due to the cognitive demands imposed on learners. During interaction, learners have to comprehend the interlocutor’s message in L2, relate it to what they know about the target language, plan the message they want to convey, and then decide what language they would use in the subsequent turn (Payne & Whitney, 2002). Although several meta-analyses and synthetic reviews (Li, 2017; Linck et al., 2014; Wen & Li, 2019) aggregated the results and tried to clarify the relationship between WM and L2 learning, none focused on the impact of WM on multiple aspects of interaction such as how learners behave during interaction, how different task conditions mediate the performance, and what they gain through the interaction. In this chapter, we conduct a narrative synthesis summarizing the methods and findings of the empirical research on the role of working memory in L2 interaction. The synthesis centers around three topics: (1) WM and interactional behaviors, (2) WM and task performance, and (3) WM and interaction-driven L2 gains. The chapter starts with an overview of working memory in SLA, followed by a synthesis of the methods and findings of primary research. A conclusion is offered at the end, focusing on limitations, pedagogical implications, and future directions.
30.2
Overview of Working Memory in SLA
Working memory is a multicomponent system involving “the temporary storage and manipulation of information that is assumed to be necessary
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
for a wide range of complex cognitive activities” (Baddeley, 2003, p. 189). According to Baddeley, WM is composed of four components: the visuospatial sketchpad, the phonological loop, the episodic buffer, and the central executive. The visuospatial sketchpad processes visual and spatial information; the phonological loop is responsible for handling phonological and verbal information; the episodic buffer serves to integrate information from the visuospatial sketchpad and the phonological loop and from longterm memory into a single multimodal episode; and the central executive is a device for attention control, which has three functions: “(1) shifting between tasks or mental sets, (2) updating and monitoring of working memory representations, and (3) inhibition of dominant responses when necessary” (Miyake et al., 2000, p. 54). Among the four components, the phonological loop and the central executive are considered to be particularly important in language acquisition (Juffs & Harrington, 2011), and a variety of measurements for these two components have been adopted in SLA research. As the phonological loop primarily functions for the “storage” of input, the memory processed via this component has been referred to as phonological short-term memory (PSTM). Example PSTM tests include nonword repetition/recognition, digit span, letter span, and word span, where learners are asked to memorize and recall strings of unrelated stimuli. The central executive in SLA literature has been generally measured together with the phonological loop by means of integrated tasks involving both the storage and processing functions. The tasks can be called complex tasks, and the measured construct is called executive working memory (EWM) by Wen and Li (2019). Complex tasks (such as listening span, reading span, and operation span) require participants to process and store information simultaneously. To exemplify, a listening span test asks learners to judge whether the sentences they listened to are semantically or grammatically acceptable and memorize the final word of each sentence simultaneously. Some researchers use independent tasks to measure the executive functions of WM (shifting, updating, and inhibition) separately from the storage components. For example, a Stroop test, a common tool to measure inhibitory control, is used to see how well participants are able to suppress irrelevant information (e.g., naming the word “red” colored in green). What has research shown about the role of working memory in SLA? Linck et al.’s (2014) meta-analysis revealed that the measures of EWM had larger effects than those of PSTM, suggesting a greater role of EWM in L2 learning. Furthermore, verbal measures that involve the processing of linguistic input such as words or sentences (as in reading span or listening span) were found to be more strongly associated with L2 outcomes than nonverbal measures consisting of nonlinguistic components such as digits or visuospatial images (as in operation span or a visual pattern test). Regarding the language used in WM measures, Linck et al. showed that WM had a stronger correlation with L2 learning when measured in the L2
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
657
658
HYEJIN AN AND SHAOFENG LI
than in the L1. This finding implies that L2 proficiency might influence the learners’ performance on WM measurements. While Linck et al. (2014) demonstrated a comparative predictive power of EWM and PSTM on L2 outcome, Wen and Li’s (2019) research synthesis revealed that PSTM and EWM play complementary roles in different SLA domains. PSTM is likely to facilitate vocabulary and grammar learning among beginning and intermediate learners and mediate L2 fluency development. EWM, on the other hand, has been revealed to predict the acquisition of L2 skills involving more cognitively demanding processes such as reading and speaking.
30.3
Working Memory and L2 Interaction
In this section, we discuss the theoretical perspectives on the role of working memory in SLA, followed by a brief summary of Li’s meta-analysis of the findings on working memory’s associations with the different dimensions of interaction. The theoretical discussion starts with the Interaction Hypothesis (Long, 1996, 2015), according to which interaction provides a forum where learners attend to linguistic forms in meaning negotiation. Major components of interaction that facilitate learning include corrective feedback and the two mechanisms derived from feedback: noticing and modified output. Corrective feedback refers to responses to learners’ erroneous utterances. Learners would have to utilize their WM resources to process the information contained in feedback in the ongoing communicative event. One function of feedback is to push learners to self-correct and modify their initial utterance, and this discourse move is called modified output in the literature. Because of the significance of modified output in L2 acquisition (Gass, 1997, 2003; McDonough, 2005; Swain, 1995), the role of WM in relation to the amount of modified output has been also investigated. In addition, WM has been researched as a factor that may help learners notice the corrective force of feedback (Schmidt, 2001). In order to notice the form in the feedback, the learners must devote their attentional resources to the target structure, and therefore individual differences in WM may mediate the ability to notice the form. Another behavior that learners may show during interaction is structural priming. It is defined as “a cognitive repetition phenomenon that refers to a speaker’s tendency to produce a sentence with a previously heard. . .syntactic structure rather than use an alternative structure” (McDonough et al., 2016, p. 112). Although the notion of structural priming is not explicitly involved in the Interaction Hypothesis, noticing, one of the essential elements in the Interaction Hypothesis, can explain the significance of priming behavior in L2 interaction. When L2 learners produce language influenced by structural priming, it implies that they attended to a particular structure in the interlocutor’s speech during meaning negotiation. That is, structural priming can be one of the apparent behaviors that denotes learners’
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
noticing of specific structures. Given that WM has also been researched in the priming studies as a cognitive individual difference factor, the effects of structural priming are discussed in the present synthesis as a criterion variable. Finally, Gass (1997) provides an overarching explanation for the role of working memory in facilitating learning in L2 interaction. Gass argues that L2 learners are often exposed to more input than they can actually process. Therefore, they are required to have a device that enables them to sort through the input and retain the information long enough to make a comparison between their utterance and targetlike forms, and working memory is such a device (Gass et al., 2013). Another theory that attaches great importance to working memory is Skehan’s Limited Attentional Capacity Hypothesis (1998), which draws on Levelt’s (1989) speech model. According to Levelt, speech production consists of three components: the conceptualizer, the formulator, and the articulator. The conceptualizer is where the content of the message is planned and generated, and the product of this stage is a preverbal message. The formulator finds linguistic forms for the planned message, that is, the syntactic structures and phonological representation for the content are encoded in the formulator. The output of the formulator is a phonetic plan, which is implemented through the articulator. Whereas the lexical access and articulation are likely automatic in L1 production, they need controlled processing when it comes to L2 production (Payne & Whitney, 2002). For learners at an early stage of L2 development, speech production in an L2 may require a greater amount of attention. As it is the individual’s WMC that determines the amount of attentional resources, the limited capacity in WM is assumed to have a direct influence on speech performance (Georgiadou & Roehr-Brackin, 2017). Based on Levelt’s model, Skehan suggested that tasks should be designed in such a way that assists conceptualization (i.e., message generation) and formulation (i.e., grammatical and phonological encoding) and eases the burden on learners’ WM. Furthermore, Skehan argued that learners may not be able to properly allocate attentional resources when they perform complex tasks due to the limited attentional capacities in WM. As a result, an increase in task complexity would force the learners to trade-off between complexity, accuracy, and fluency (CAF) (Skehan, 1989). This hypothesis, however, is challenged by another theoretical model, Cognition Hypothesis, according to which complex tasks enhance learners’ performance on multiple dimensions. According to the Cognition Hypothesis (Robinson, 2003b, 2007, 2011), L2 learners’ linguistic performance becomes more complex and accurate when tasks impose higher cognitive demands. Task complexity can be manipulated along the dimensions of resource-directing and resource-dispersing variables (Robinson, 2007). Resource-directing variables involve the manipulation of cognitive/conceptual demands (e.g., reasoning demands,
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
659
660
HYEJIN AN AND SHAOFENG LI
here and now, few elements, perspective taking) while resourcedispersing variables relate to the performative/procedural demands (e.g., planning time, prior knowledge, single task, few steps). He believes that increasing the complexity along the resource-dispersing dimension only depletes learners’ attention resources and thus makes learners struggle with focusing their attention on specific linguistic components (Robinson, 2003a). If task complexity is enhanced along both resourcedirecting and resource-dispersing aspects, the task can direct learners’ attentional resources to specific aspects of L2 resulting in more accurate and complex L2 performance. Given that complex tasks impose more attentional demands than simple tasks do, Robinson (2001) claims that WM may increasingly impact task-based performance as tasks increase in complexity along the above-mentioned variables. When tasks become more complex, learners are more likely to rely on their WM to process the increased cognitive demands. Therefore, those who have a greater capacity in WM may benefit more from complex tasks (Zalbidea, 2017). It should be noted, however, that Robinson has placed much emphasis on the effects of task complexity on L2 performance, leaving the question of how individual difference in WM interacts with task complexity unresolved. Regarding the empirical evidence on the role of working memory in L2 interaction, Li’s (2017) meta-analysis showed that WM has an overall weak predictive power on the process and product dimensions of L2 interaction. While the process aspect involves features that appear during interaction (e.g., noticing of feedback and producing modified output), the product dimension refers to the effectiveness of task treatments in interactionbased learning. The role of WM during the process of interaction was revealed to be inconsistent in previous research findings due to the diverse methodological designs (e.g., different measures of noticing). Moreover, the correlation between the learners’ WM and the immediate effects of interaction-based treatments was weak (r = .23, p < .01), and for delayed effects, the association was insignificant (r = .08, p = .21). When it comes to the role of WM in the efficacy of recasts – the most researched type of corrective feedback – the analysis showed a weak and nonsignificant correlation between WM and the effects of recasts (r = .22, p = .07). Given that the provision of recasts was highly controlled even in the classroom study setting, the weak predictive power of WM on the effectiveness of interactional treatments can be explained by the salience of the treatments (i.e., feedback) that attenuates the learners’ reliance on their WM. As quantitative analyses primarily draw on the numerical findings and may suffer from a lack of in-depth analysis and description of the primary research, the present narrative synthesis aims to provide a more elaborate synthesis of the methods and findings of the research, which may deepen our understanding of the relationship between WM and L2 interaction. Specifically, we provide a methodological synthesis of the retrieved studies as well as a narrative synthesis of the role of WM in multiple aspects of L2
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
interaction: interactional behaviors, the efficacy of interactional conditions, and interaction-driven learning gains. This synthesis is based on a thorough search of the literature, encompasses both the methods and findings of the research, and constitutes an update and complement to Li’s meta-analysis.
30.4
Methods
This section reports on the criteria employed to select relevant studies and the variables involved in the synthesis. It then introduces the coding scheme for synthesizing the included literature.
30.4.1 Search and Selection Criteria To search for the studies examining the relationship between WM and L2 interaction, we consulted the Linguistics and Language Behavior Abstracts (LLBA), Education Resources Information Center (ERIC), ProQuest Dissertation, Google Scholar, Web of Science, and PsycInfo. We also searched major journals in SLA and psychology. Example keywords used in literature search include working memory, central executive control, cognitive individual difference, phonological short-term memory, attention control, inhibition, shifting, switching, and updating for the predictor variable. Example key search words for the criterion variable include L2 interaction, language-related episodes (LREs), noticing, modified output, uptake, complexity, accuracy, fluency (CAF), communication, pair work, group work, corrective feedback, speech performance, and task-based learning. In order to select the studies for the synthesis, the following inclusion/ exclusion criteria were formulated. • •
•
• • •
Studies examining monological tasks were excluded. Studies examining either face-to-face or synchronous computermediated communication (SCMC) were included. CMC researchers argue that the real-time chat provides a similar interactional environment as face-to-face interaction. Studies that investigated computerized feedback are also included. Although in these studies the participants do not interact with a human interlocutor; they receive corrective feedback in response to their erroneous utterances. Furthermore, as the provision of computerized feedback tends to be more tightly controlled, it is expected to provide more valid evidence regarding the role of WM in the effectiveness of corrective feedback. The present synthesis includes studies published between 2002 to 2020. Published articles, book chapters, and unpublished dissertations were all included. Only studies published in English were included.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
661
662
HYEJIN AN AND SHAOFENG LI
Table 30.1 Predictor and criterion variables in the synthesis Predictor variables
Working memory
PSTM EWM Visuospatial memory Inhibitory control Attention control
Criterion variables
Interactional behavior
Noticing Modified output Structural priming Linguistic performance
Effectiveness of interactional treatments
Explicitness of feedback (implicit vs. explicit) Timing of feedback (pre- vs. within- vs. posttask) Modality of interaction (oral FTF vs. written CMC) Complexity of tasks (simple vs. complex)
Learning gains
Short-term treatments Long-term treatments
The preliminary search resulted in 46 studies. Among them, a total of 33 studies satisfied the inclusion criteria. The majority of the excluded studies adopted monologic tasks in treatment sessions (e.g., Gilabert & Munoz, 2010; Li & Fu, 2018; Wen, 2016). The decision to exclude was made because of monologic speech’s distinctive contrast with interactive speech, which may obscure our discussion regarding the role of WM during interactive conversations. While speakers have more freedom in controlling the topic or direction of discourse in monologic speech, they are less likely to control the flow of dialogic speech because they need to exchange conversational turns with the interlocutor. As L2 speakers are required to produce and comprehend L2 at the same time, the ability to control attentional resources in WM is assumed to be more significant in dialogic than in monologic interaction (Martin, 2018).
30.4.2 Variables in the Synthesis Table 30.1 shows the predictor and criterion variables that are included in our synthetic review. The predictor variables involve the components of WM such as phonological short-term memory, executive working memory, and visuospatial memory. Inhibitory control and attention control are also included because they represent the executive functions controlled by the central executive in WM. Yilmaz and Granena (2019) included the variable “attention control” specifically indicating the ability to shift attention among different cognitive tasks. The criterion variables are categorized into three aspects of L2 interaction-driven learning. The first is learners’ interactional behavior,
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
including noticing, modified output, structural priming, and CAF performance during L2 interaction. The second category includes studies investigating the associations between working memory and different interactional treatments. Four themes emerged from this category: (1) explicitness of feedback (implicit vs. explicit feedback), (2) timing of feedback (feedback in pre- vs. within- vs. post-tasks), (3) modality of feedback (oral face-to-face vs. written text-based feedback), and (4) task complexity (simple vs. complex tasks). The final category of studies examined the predictive power of WM in the L2 learning gain (via short-term and longitudinal treatments) itself through the interactive instruction without examining different types of interactional treatments.
30.4.3 Coding To examine the methodological designs of the included studies, we coded the literature in terms of publication type, participants, and methods. The studies were first coded by publication type, namely whether it was published in a book chapter, a journal, or as an unpublished dissertation. As for the sample, age group (children, adolescents, adults, the elderly), proficiency level, the first and target language, and the learning context were coded. If learners’ proficiency was not specified in the study, it was coded as “unspecified,” and if the participants in a study were reported to have different proficiency levels (e.g., beginner and intermediate learners in Georgiadou & Roehr-Brackin [2017]), it was coded as “mixed.” Methodological design was coded in terms of interaction type (dyadic or group), study setting (lab or classroom), direction of information flow (one-way or two-way), and the measurement of WM. Computerizedfeedback was coded as dyadic as the interaction takes place between the individual participant and the computer. For the studies that did not explicitly specify the direction of information, we examined how the meaning was exchanged during tasks. If the participants only received feedback from the interlocutor, it was coded as one-way interaction. If task participants exchanged information, it was coded as two-way interaction. Two studies conducted in L2 immersion settings did not have treatment tasks but involved two-way interactive tasks to measure L2 development (accuracy and fluency in Wright, 2013; general L2 learning gains in White, 2021); therefore, the direction of interaction was coded as two-way. Computerized-feedback studies, on the other hand, were coded as one-way interaction because feedback functions as a response to the learner’s error without providing further information.
30.5
Findings
This section starts with a methodological synthesis of the included studies, including sample characteristics and study designs. It proceeds to synthesize the research findings – pertaining to the role of working memory in (1)
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
663
664
HYEJIN AN AND SHAOFENG LI
Figure 30.1 Publication types included in the methodological synthesis
learner behavior during L2 interaction, (2) the efficacy of different L2 interactional treatments, and (3) in L2 interaction-based learning gains without instructional manipulation.
30.5.1 Methodological Synthesis The methodological information of the retrieved studies is displayed in Figure 30.1 and Tables 30.2 and 30.3. As seen in Figure 30.1, the literature was mostly published in journals (k = 19), and the remaining studies were book chapters (k = 8) or dissertations (k = 6). Table 30.2 describes the participants’ information in the studies. Twenty nine out of the 33 studies involved adult learners; one study was conducted with kindergarteners (White, 2021); one involved the elderly whose mean age was 72 (Mackey & Sachs, 2012); two studies examined adolescent learners (middle school learners in Li et al., 2019 and high school students in Revesz, 2012). Regarding the learners’ proficiency, 16 out of 33 studies included intermediate-level speakers, six studies involved beginners, and only one study investigated advanced learners (Wright, 2013). Furthermore, 6 examined learners with varied proficiency levels and 7 did not report the participants’ L2 proficiency. As for the L1 and the L2 of the participants, English was the most frequently examined language as both L1 and L2 (k = 14 as L1 and k = 18 as L2). Chinese and Korean were reported as L1 in 5 and 3 studies, respectively. 6 involved learners with various L1 backgrounds. Spanish and Chinese were examined as L2 in 9 and 2 studies, respectively. The languages introduced as either L1 or L2 one time each include Arabic, Farsi, Hungarian, Japanese, and Spanish as L1 and Italian, Japanese, Latin, and Turkish as L2. With regard to the instructional context, 25 studies were conducted in foreign language contexts while 8 were carried out in second language contexts. Table 30.3 summarizes the methodological features of the synthesized studies. Dyadic interaction was adopted in 29 out of 33 studies, 3 of which involved computerized interaction. Only 4 studies examined group interaction. For the experimental studies, 27 were carried out in the laboratory,
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
Table 30.2 Sample characteristics of included studies (K = 33) Aspects
Subcategories
K
Age
Children Adolescents Adult learners The elderly
1 2 29 1
Proficiency
Beginner Intermediate Advanced Varied Unspecified
6 16 1 3 7
First language
English Chinese Korean Varied Arabic, Farsi, Hungarian, Japanese, Spanish
14 5 3 6 1 (each)
Target language
English Spanish Chinese Italian, Japanese, Latin, Turkish
18 9 2 1 (each)
Instructional context
Foreign language Second language
25 8
Table 30.3 Methodological Features of Included Studies (K = 33) Aspects
Subcategories
K
Interaction type
Dyad Group Laboratory Classroom One-way Two-way PSTM
29 4 27 6 18 15 8 1 1 1 10 9 7 3 3 2 1 1 1 1 1 1
Study setting Direction of information during interaction Measurements of WM*
EWM
Visuospatial memory Inhibitory control Attention control
Nonword repetition Serial word recall Digit span Letter span Listening span Reading span Operation span Aural running span Backward digit span Counting span Story recall Visual pattern test Odd-one-out Stroop test Eriksen Flanker test Task switching numbers
Note. *Some studies used more than one WM measurements. For example, Baralt (2010) used Reading span, Operation span, and Counting span to measure the participants’ EWM.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
665
666
HYEJIN AN AND SHAOFENG LI
and only 6 had their treatments in the classroom. As for the information flow during interaction, similar number of studies adopted one-way and two-way interaction tasks (k = 18 and k = 15, respectively). With regard to the measurements of WM, it is noticeable that more research was conducted to examine the role of EWM in L2 interaction than other WM constructs. Among a variety of measurements of EWM, listening span (LSPAN) was the most frequently adopted test (k = 10), followed by reading span (RSPAN) and operation span (OSPAN). Other span tests used in the research include aural running span, backward digit span, and counting span. One study (Wright, 2013) chose to use a story recall task that was part of a test battery for native speakers with aphasia and other linguistic impairments. Listening and reading span are categorized as verbal WM measures as they specifically tap into the verbal processing in WM. Listening span measures the test taker’s ability to comprehend a set of sentences by listening to and recall information (e.g., the last word in each sentence) simultaneously. The administration of a reading span test is similar to that of a listening span test, with the only difference being in the modality in which the stimuli are presented. Nonverbal tests such as operation span have also been used. During an operation span test, individuals are presented with simple arithmetic equations followed by letters or words. They are asked to either solve the operation or judge the correctness of the given answer while remembering the accompanying word or letter. The measurements for the learners’ PSTM, on the other hand, only assess the ability to store and recall information (e.g., nonwords, digits, letters) without requiring learners to process information. Nonword repetition was most frequently used, and digit span and letter span were each used once. Visuospatial memory, inhibitory control, and attention control have rarely been used in L2 interaction research. Two studies examined visuospatial memory measured by a visual pattern test and an odd-one-out test. Inhibitory control was investigated in two studies by using a Stroop test and a flanker test, both of which measure learners’ ability to suppress irrelevant information. Only one study examined the ability to shift between different tasks, measured by means of a “task switching number” test. More detailed information about the design of the included studies as well as their major findings can be found in Appendices A–C.
30.5.2 Synthesis of Research Findings We now synthesize the findings of the retrieved studies based on the following questions the primary studies sought to answer: 1. What is the role of WM in the learners’ behavior during L2 interaction? 2. What is the role of WM in the efficacy of different L2 interactional treatments?
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
3. What is the role of WM in L2 interaction-based learning gains without instructional manipulation?
30.6
What Is the Role of WM in Learners’ Behavior during L2 Interaction?
The interactional behavior is operationalized as the speakers’ behavior shown during the performance of L2 interaction. As shown in Table 30.3, four subcategories emerged as the interactional behavior: noticing, modified output, structural priming, and linguistic performance.
30.6.1 Noticing Based on the premise that noticing during L2 interaction necessitates the focal attention to the target structure (Skehan, 2002), studies were conducted to examine the predictive power of WM for the amount of noticing. Mackey et al. (2002) investigated how the WM of 30 low-intermediate ESL learners predicts the ability to notice the corrective force of recasts. Each participant engaged in communicative tasks of picture drawing, picture difference, and story completion with a native speaker. During the conversation, the learners’ erroneous production of English questions was corrected via recasts. Noticing of recasts was measured through a stimulated recall and an exit questionnaire. For the assessment of WM, listening span and nonword repetition were adopted to measure the learners’ EWM and PSTM, respectively. A composite WM capacity (WMC) score was computed for analysis. The result of the study showed that those who reported less noticing are likely to have low WMC, whereas those who noticed more tend to have higher WMC. It should be noted, however, the noticing data were collected in different ways for different numbers of participants. Only 11 participants engaged in the stimulated recall, while the data of another 19 students were gathered from the exit questionnaire. The coding of noticing was also different in these two measures. In stimulated recall, it was coded “more noticing” if the participants reported noticing 25 percent or more of recasts from the three treatment sessions. In the exit questionnaire, on the other hand, the responses were classified as “more noticing” if they provided two or more reports of noticing in the five items asking about noticing. Investigating the same target structure, English question formation, Kim et al. (2015) also found a significant role of WM in the noticing of feedback. Eighty-one ESL learners performed two-way collaborative information gap tasks, during which their erroneous question formation was addressed by using recasts. Each recast was followed by immediate cued recall designed to measure the participants’ noticing. EWM was measured by an aural running span test in which the learners heard a series of letters and were
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
667
668
HYEJIN AN AND SHAOFENG LI
asked to recall the last n items from the list. The study revealed that WM had a positive and significant relationship with the amount of noticing. The positive role of working memory in noticing found by Mackey et al. (2002) and Kim et al. (2015) has been challenged by some contradictory findings. Dai (2013) expanded Mackey et al.’s (2002) study by involving two target structures: English question formation and the English past tense. While Mackey et al. found a significant role of WM in ESL learners’ noticing of English question formation, Dai revealed that WM does not predict EFL learners’ noticing of the question forms, but it does when it comes to the past tense. The participants in Mackey et al.’s study were recruited from an ESL instructional setting in which students are exposed to abundant opportunities to communicate with others asking and answering questions in the L2. Thus, the saliency of English questions in the ESL context may have assisted those with higher WM to notice more of the English question form in the implicit feedback. By contrast, Dai’s study was conducted in a Chinese EFL context, where language classes are generally teacher-centered and the learners rarely engage in communicative tasks. It is, therefore, possible that the participants in Dai’s study may have had difficulty in noticing the question form in recasts. However, the learners with high WM found the English past tense form easier to notice because the form can be more explicit in feedback than with question formation. This research finding suggests that the predictive power of WM in noticing can depend on the instructional setting and the explicitness of the target structure. Chen (2013) also failed to find a significant relationship between WMC and the noticing of recasts. Instead of targeting specific structures, Chen examined all kinds of errors. During the treatment session, each participant interacted with a native speaker in unfocused one-way information gap tasks. Noticing was measured through stimulated recall, and WM was assessed by a reading span test. The researcher suggested that a possible reason for the nonsignificant role of WMC in noticing the feedback is learners’ high noticing rate (over 70 percent) for both morphosyntactic and lexical/phonological recasts, which in turn can be partly attributed to the laboratory research setting, where the feedback was salient and easily noticeable. Another study found that the predictive role of WM in noticing depends on the contingency of feedback. Lai et al. (2008) investigated whether individual differences in ESL learners’ (n = 17) WM mediates the amount of noticing in two different types of feedback conditions: contingent feedback that immediately follows non-target-like utterances and noncontingent feedback separated from the erroneous utterance by a comment or irrelevant information. It was found that WM had a stronger correlation with the noticing of noncontingent recasts. It is proposed that noncontingent feedback requires the learners to store and process the target structure in WM while contingent feedback alleviates the burden on working memory.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
30.6.2 Modified Output The second L2 interactional behavior we discuss in relation to learners’ WM is modified output. Mackey et al. (2010, p. 508) explained the mechanism of how learners rely on WM when producing modified output as below: 1. Prompts (e.g., “What?”) are likely to invoke both the processing and storage functions of WM. 2. L2 learners must be able to shift attention away from meaning-based task completion and toward the particular L2 structures being used to accomplish the task. 3. In order to respond with a linguistic modification of the original utterance, the learner has to recall what s/he said immediately prior to the feedback and compare the utterance with the L2 structures in their long-term memory and identify the mismatch. A limited number of studies have been conducted to investigate the relationship between WM and modified output. Mackey et al. (2010) examined whether individual differences in WM among 42 learners of Spanish predicts the amount of modification during L2 interaction. During communicative tasks (a map task, picture drawing, spot-the-difference, and a story completion), the native speaker provided prompts to encourage the learners to modify their own erroneous utterances. Modified output was operationalized as “turns where learners made partial or complete changes to their original utterances” (p. 510). WM was tested with a listening span test in which the scores of processing and storing the given inputs were separately recorded. Simple linear regression analysis revealed a positive and significant correlation between the learners’ WM scores and the amount of modified output. When analyzing the predictive power of the processing and storage components of WM test separately, it was found that those with higher processing scores are more likely to “change” original utterances in their modified output, whereas the learners with higher recall scores tended to “repeat” the feedback without modifying the original utterances. While Mackey et al. (2010) investigated the role of WM in learners’ modified output after prompts, Zhao (2015) examined how WM predicts the learners’ repair following recasts. Although the researcher did not explicitly use the term “modified output,” the operationalization of “uptake-with-repair” – the learners’ response attempting to repair the erroneous utterance by repeating or integrating the feedback – is similar to modified output in Mackey et al. (2010). The participants performed oneway information gap tasks in the classroom. During group activities, the teacher provided either corrective recasts (prompt + recast) or implicit recasts (without a prompt) when the learners made errors on the two target structures: third person -s and the formation of embedded questions. Under the corrective recast condition, the teacher explicitly pointed out the erroneous part by repeating it and then provided a reformulation. In the condition of implicit recasts, the teacher simply provided a reformulation
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
669
670
HYEJIN AN AND SHAOFENG LI
without drawing the learner’s attention to the error. The correlation analysis between WM, target structures, feedback types, and uptake-with-repairs mostly revealed nonsignificant results. One exception, however, was a significant correlation between PSTM measured by nonword repetition (NWR) and the ability to repair the third-person -s errors in response to implicit recasts. The researcher attributed the weak association between WM and the repaired uptake to the classroom research context, where implicit recasts are nonsalient. Therefore, those with higher PSTM were more likely to perceive the corrective function of the implicit feedback and they could modify their utterances. Under the corrective recast condition, which was more explicit, both high and low PSTM learners were able to repair their non-target-like utterances after the teacher’s feedback.
30.6.3 Structural Priming Structural priming is also a behavior that learners can show during L2 interaction. L1 researchers found that speakers are more likely to produce structural priming with structures that are less frequent in speech, which is called “inverse-preference effects” (Hartsuiker et al., 1999; Scheepers, 2003). Based on such findings, L2 studies have examined how structural priming activities can facilitate learners’ subsequent production of more complex and nonsalient linguistic forms such as relative clauses (McDonough et al., 2016), passives (McDonough & Kim, 2016), and stranded prepositions (Kim et al., 2019). McDonough et al. (2016) explored the relationship between L2 learners’ structural priming of English relative clauses and their WM. During the priming activity, each of the 50 ESL learners interacted with one of the researchers during dyadic interaction. The interlocutors took turns asking and answering questions about the given information in the tasks. To facilitate the participants’ production of structural priming, the researcher always delivered the prime sentences before the learner’s utterance. In the baseline and the postpriming activity, the participants worked on the same format of the interactional activity, but in the absence of prime sentences from the researcher. The study found that WM scores, measured by backward digit span, are not significantly related to the production of relative clauses during or after the priming activities. The researchers speculated that the learners’ high proficiency might have helped them to produce structural priming regardless of their individual difference in WM. The learners’ report on their awareness of the target structure in prime sentences was further analyzed, and it was revealed that 45 out of 50 participants did not notice the structure; only 5 of them reported noticing. Interestingly, these 5 students had higher WM scores than those who did not notice the form. The study, therefore, suggests that the predictive power of WM in L2 structural priming can be mitigated by the learners’ high proficiency but WM may still predict higher noticing rates (also see Kim et al., 2015; Mackey et al., 2002).
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
McDonough and Kim (2016) conducted another experiment with a different group of participants with the target structure of English passives. While McDonough et al. (2016) incorporated face-to-face interaction tasks, McDonough and Kim (2016) asked the participants to describe the picture after listening to computerized prime sentences. For example, after listening to a prime sentence The computer was kicked by the man with the passive form in it, the participants were instructed to describe another picture of a farmer feeding apples to horses. WM, measured by an aural running span, was found to have a positive significant correlation with the production of structural priming in the absence of intervening sentences between the prime sentence and the target picture description. When irrelevant sentences are inserted between them, however, WM did not mediate the production of structural priming. The results suggest that the role of WM in structure priming may depend on the adjacency of prime sentences and the learners’ production of the target structure. A recent study conducted by Kim et al. (2019) compared L2 structural priming activities in face-to-face (FTF) and synchronous computer mediated communication (SCMC). The target structure was English stranded preposition construction (e.g., This is something (which) you take cookies from) which is considered to be a complicated structure, especially for the participants whose L1 was Korean in which postposition stranding is not allowed. During interactive picture description tasks, the participants were asked to guess what the object is after the interlocutor’s prime sentence describing an object in the picture. After the confirmation of the guess, the participant described another object using the given verb. The analysis of interactions between aural running span scores and the amount of structural priming showed that the participants’ WM was not significantly related with the degree of structural priming in both FTF and SCMC setting. The insignificant role of WM was attributed to the short length of prime sentences that might have reduced the cognitive load during the priming activities.
30.6.4 Task Performance Since Crookes’s study (1989) that investigated how different planning conditions influence L2 learners’ spoken performance in terms of complexity, accuracy, and fluency (CAF), a good deal of SLA research has examined CAF as three dimensions that describe learner’ linguistic performance while engaging in tasks. Although several studies investigated WM as a predictor of CAF, most of them used monologic tasks to find out the role of WM in linguistic performance. Therefore, it still remains questionable how WM predicts CAF performance during interaction. Only two studies have examined working memory and L2 accuracy and fluency in interaction. Wright (2013) investigated how learners in immersion contexts improved the accuracy and fluency or their oral proficiency and how WM mediated their oral gains. Two-way information gap tasks measuring the participants’ use
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
671
672
HYEJIN AN AND SHAOFENG LI
of the L2 were administered at two different time points to determine the longitudinal change in the accuracy and fluency of their L2 performance: within 4 weeks of arrival in the immersion context and after 11 months. Accuracy was operationalized as the number of targetlike question forms and the ratio of the produced questions to the total utterances. Fluency was measured by the type-token ratio (i.e., the ratio of the total number of different words by the total number of words); a higher type-token ratio at Time 2 served as evidence of increased fluency. Given that the type-token ratio alone cannot be a reliable measure of fluency, as acknowledged by the researcher, another scale of fluency measure – the number of pauses and repairs – was also adopted. During the communicative task, the participants were instructed to ask questions to the interlocutor to find the differences between two pictures. The learners’ WM was measured by a listening span test and a story recall task. It was found that the learners’ WM is related to the improvement in the accurate production of L2 questions. For L2 fluency, however, no correlation was found. The results suggest that WM is more closely associated with accurate L2 performance during interaction than fluent language production. While Wright (2013) included self-repair as a component of fluency, Georgiadou & Roehr-Brackin (2017) separated learners’ self-repair behavior from the dimension of fluency/disfluency. Seventy-seven EFL college-level learners (42 beginners and 35 lower-intermediate learners) participated in oral interviews with one of the researchers. During the communication, in which the participants were asked to answer a question related to their best friend, PSTM was measured by a serial word recall test that asked the participants to memorize and recall a set of words while EWM was assessed by means of backward digit span and listening span tests. The participants’ oral recordings were analyzed for three dimensions: fluency (i.e., words per minute), disfluency (i.e., number of pauses and repetitions), and self-repair behavior (i.e., number of overt self-corrections). The correlation analysis between WM measures and the fluency/disfluency revealed that neither PSTM nor EWM was related to the fluency for the beginner-level students while EWM was significantly and negatively correlated with the number of pauses for the lower-intermediate learners. The negative correlation indicates that those with higher EWM paused less during interaction. It is explained that hesitations (e.g., pauses and repetitions) allow L2 speakers extra time to conceptualize and formulate speech, and those with greater EWM may not have needed this additional time. With regards to the role of WM in the self-repair behavior, no significant correlation was found. The following is a summary of the findings regarding the role of WM in L2 interactional behavior. 1. Learners with higher WM are more likely to notice the gap between their non-target-like utterances and the correct forms contained in corrective feedback. However, the predictive power of WM in noticing
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
is insignificant when the target structure is too complex and nonsalient to notice, when the learners are already familiar with the target form, and when the feedback immediately follows the erroneous utterance. 2. The higher WM L2 learners have, the larger amount of modified output they are likely to produce. However, the effects of working memory are restricted to the modified output after feedback that prompt learners to self-correct, as opposed to feedback that provides the correct form such as recasts. 3. In syntactic priming, WM overall does not mediate learners’ tendency to produce the syntactic structure following the interlocutor’s previous utterance. However, it can predict structural priming under the condition where the prime and the target are contiguous and when the learners are at a low level of L2 proficiency. Higher WM learners are also likely to notice the target form in the prime sentence. 4. For task performance, learners with higher WM are more likely to improve the accurate use of L2 during interaction over time. As far as fluency is concerned, WM predicts fewer pauses for low intermediate learners, not for the beginner learners. In addition, self-repair is not associated with WM.
30.7
What Is the Role of WM in the Efficacy of Different L2 Interactional Conditions?
The second research question addresses whether WM is associated with the effectiveness of different interactional conditions measured by pre- and posttests. The studies reviewed in this section examined the mediating role of WM in the interactional environments with different (1) explicitness of feedback (i.e., explicit vs. implicit), (2) timing of feedback (i.e., pre-, within-, vs. post-task), (3) modality of interaction (i.e., oral vs. written), and (4) complexity of tasks (simple vs. complex).
30.7.1 Explicitness of Feedback A number of studies examined whether WM has differential associations with the effectiveness of different types of corrective feedback. Goo (2012) examined how 54 EFL students’ WM is related to the effects of recasts and metalinguistic feedback during one-way information gap activities in the classroom. In the task performance, the participants were instructed to ask the interlocutor a question looking at a picture. The interlocutor provided either a recast or metalinguistic feedback on learners’ errors in producing the English that-trace filter. That-trace filter (e.g., *Who do you think that likes Mary? ! Who do you think likes Mary?) was chosen as the target structure because (1) “that” is communicatively redundant, (2) learners should selectively attend to noticing how and when “that” should be deleted, and (3) it is
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
673
674
HYEJIN AN AND SHAOFENG LI
one of the structures rarely used in naturalistic or L2 classroom contexts. A grammaticality judgment test (GJT) and a written production test were administered before and after the two sessions of the communicative tasks. The participants’ WM was measured by using a reading span test and an operation span test. The analysis of the relationship between the WM scores and the efficacy of recasts and metalinguistic feedback revealed that WM is significantly and positively related to the effects of recasts on both the GJT and the written production test. For metalinguistic feedback, the learning gains of the metalinguistic group were not mediated by WM. The researcher interpreted the results as suggesting that learners in the recast group had to invest more cognitive effort in noticing the form during the ongoing interaction. Metalinguistic feedback, on the other hand, may have eased the burden on learners’ WM by providing an explicit explanation of the structure. The researcher conducted a replication of the 2012 study (Goo, 2016), which showed a contradictory finding that WM did not play any role in the efficacy of both recasts and metalinguistic feedback. The inconsistency was explained by referring to the different research settings. While Goo’s experiment (2012) was conducted in the classroom, where learners had to rely on their WM to process the input in recasts, the participants in Goo’s second experiment (2016) were given recasts in a laboratory, where the feedback was more clearly delivered. Therefore, the learners in the lab setting could more easily attend to the corrective force of the feedback and process the input regardless of their difference in WMC. These findings suggest that the significant role of WM in the effects of recasts may depend on the research context. Ahmadian (2020) compared the effects of implicit and explicit feedback in teaching refusal strategies. The researcher reasoned that the speech act of refusal is one of the most challenging L2 features because it is “highly face-threatening” (p. 171) and requires learners to spontaneously respond to the interlocutor’s request or suggestions without much preparation. The interaction between the effectiveness of each feedback type and WM was also examined. Seventy-eight upper-intermediate EFL learners engaged in peer interaction via role-play for six weeks to learn the pragmatic feature of refusals. For the implicit instruction, the participants were given an inputenhanced text where the target structure was underlined. During the roleplay, the researcher-teacher and two teaching assistants provided recasts on the learners’ inappropriate use of refusal expressions. For the explicit instruction group, on the other hand, the learners were explicitly taught refusal strategies and given explicit corrective feedback. The tests used to measure the learning gains included a discourse completion test (DCT) and a comprehension questionnaire (CQ). The DCT was designed to assess productive knowledge as the participants were asked to fill out a blank with a refusal expression. The CQ, on the other hand, measured receptive knowledge by requiring learners to rate the appropriateness of the refusal strategies. WM was measured by an operation span test. The analysis of the
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
scores revealed that WMC had a strong positive correlation with the development in the immediate and delayed posttests for both productive and receptive knowledge only in the implicit group. The researcher attributed the significant role of WM in the implicit group to the learners’ heavier cognitive and attentional load under the implicit instruction. Revesz (2012) also found evidence that lends support to the positive relationship between WM and the effects of recasts by comparing the role of WM in recast and nonrecast conditions. In the treatment session, 90 EFL learners at a beginner proficiency level were shown 30 photos and the researcher asked them to describe what people were doing to elicit the use of English past progressive form. For the recast group, the errors related to the target structure were treated with recasts while no feedback was given to the nonrecast group. The effects of recasts were assessed by means of a grammaticality judgment test, a written picture description test, and an oral description test. Overall, WM was revealed to only mediate the gain scores of those who received recasts during interaction. When examining PSTM and EWM separately, the results showed a strong correlation between PSTM and the oral description test, which was designed to measure procedural knowledge, and between EWM and the written tests (i.e., GJT and written description tests) assessing declarative knowledge. It was proposed that those with high PSTM were able to retain the input provided in recasts in short-term memory, and consequently they were better able to access the recently processed structure when producing the oral speech in a real-time speaking condition. High EWM capacity, by contrast, helps the speakers to control attention among a variety of cognitive demands of the task. The superiority of high EWM may have helped the learners to develop metalinguistic and declarative knowledge of the target structure from recasts. While the above-mentioned studies support the significant role of WM in the effectiveness of recasts, they are challenged by several experimental studies showing that WM is more likely to mediate the effects of explicit corrective feedback. Li (2013) investigated the interaction between feedback type (i.e., recasts and metalinguistic feedback) and two aptitude components (i.e., WM and language analytic ability) in L2 interaction-driven learning. Seventy-eight learners of Chinese as a foreign language performed picture description and spot-the-difference tasks during which non-targetlike utterances were treated with either recasts or metalinguistic correction. The target structure was the Chinese classifiers, which are commonly taught at an early stage of L2 Chinese instruction, but which constitute a difficult structure for learners at all proficiency levels. The learning gains were tested using a grammaticality judgment test and an elicited imitation test, as measures of explicit and implicit knowledge, respectively. WM was assessed by means of a listening span test, and language analytic ability was gauged via the Words in Sentences subtest of Modern Language Aptitude Test (MLAT). Unlike other studies that calculated only recall scores, Li included reaction time as well as plausibility scores to better assess the
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
675
676
HYEJIN AN AND SHAOFENG LI
processing component of WM. The finding was that language analytic ability was a significant predictor of the learning gains resulting from recasts while WM mediated the effectiveness of metalinguistic correction. Li argued that learners who received recasts may have had to analyze the syntactic pattern of the target structure because the feedback itself does not offer any linguistic information about the structure. Metalinguistic feedback, however, provided rule explanation and what remained to be done was to store and process the given input. Thus, the explicit feedback in the form of metalinguistic information may have prompted learners to consciously memorize and process feedback simultaneously (Ellis, 2009). Li (2015) reported on a study with a more complex structure, the Chinese perfective marker “-le”. Compared to the Chinese classifiers, the acquisition of the aspect of perfective marker is considered to be more complicated as the form-meaning mapping of the structure is not as transparent as that of classifiers. In this study, the participants were asked to narrate a story after watching a short silent movie. They also had interview tasks where they were given 16 questions about their recent experiences. Other aspects of the research designs were the provision of two different feedback types (i.e., recasts vs. metalinguistic feedback), tests of learning gains, measurements of language analytic ability and WM (i.e., a subtest of MLAT and listening span), which followed those of Li (2013). Surprisingly, the data analysis showed that WM had a significant negative correlation with the improvement of the metalinguistic feedback group. To explain the finding, the researcher drew on the “less is more” hypothesis (Miyake & Friedman, 1998). This would suggest that those with a greater WMC might have processed an excessive amount of information that is involved in the highly complex, rule-based target structure, and consequently this resulted in poorer performance. The learners with a limited WMC, on the other hand, were probably better able to engage in more detailed and deeper processing of the fewer inputs stored in their WM. The contradictory findings of Li (2013) and Li (2015), therefore, give credence to the idea that the role of WM in the efficacy of feedback can be mediated by the complexity of the target structure. Unlike other studies where the metalinguistic feedback was delivered in the format of a verbal comment on the learners’ errors, Sachs (2011) investigated “visual” metalinguistic feedback via tree diagrams that illustrate the syntactic relationship of the target structure in sentences. To ascertain the correlation between WM and the effects of visual metalinguistic feedback, the researcher measured the participants’ visuospatial memory (VSM) using the Visual Patterns Test (VPT). It assessed the participants’ ability to store the visuospatial information using the checkerboardlike grids of blank and filled-in squares. While practicing the target structure of Japanese reflexive zibun, the learners received three types of instruction: (1) no feedback, (2) right/wrong feedback that only gives information about the correctness of the answer, and (3) visual metalinguistic
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
feedback that tells whether the answer is right or wrong and provides a tree diagram. The learning gains were measured by a truth-value judgment test that is similar to the treatment task. The study found that learners’ VSM only mediated the effectiveness of the tree-feedback group. The result suggests that VSM plays a significant role when the feedback involves explicit metalinguistic information (here, the tree diagram) about the target structure. The learners with higher VSM may have benefited from feedback that contained tree diagrams because the feedback may have imposed an onerous processing load on the learners who had to extract the complicated syntactic rules from the visual information. While Li (2013, 2015) and Sachs (2011) involved metalinguistic feedback in the treatment, three other studies adopted explicit correction to compare its effectiveness with recasts. Yilmaz (2013) examined the mediating role of WM in the efficacy of explicit correction and recasts. Forty-eight beginners of Turkish as a foreign language performed one-way information gap tasks, and their non-target-like utterances were treated by either explicit correction or recasts. The performance in each experimental group was tested by oral production, comprehension, and recognition tests, which were combined to calculate a composite score for each learner. The participants’ WM, measured by operation span, was revealed to have a positive and significant correlation with the test scores of the explicit correction group. Yilmaz and Granena (2019) examined whether attention control was predictive of the effects of explicit correction and recasts. Attention control was assessed via a Task Switching Numbers test, which tapped the learners’ ability to shift attention among different cognitive tasks. PSTM was measured by a letter span test. A total of 112 learners of Spanish were randomly assigned to three groups (i.e., explicit correction, recast, and control), and they interacted with a native speaker during spot-the-difference tasks. Improvement in the target structure of Spanish gender agreement was tested by means of an oral production task and a grammaticality judgment test (GJT). In the recast group, neither PSTM nor attention control mediated the learning gains in oral production and GJT. In the explicit correction group, however, it was revealed that learners’ PSTM predicted the improvement on the GJT. That is, the learners with high PSTM benefited from the explicit correction when it came to the development of declarative knowledge (i.e., explicit knowledge). Recalling that Revesz (2012) found a positive correlation between PSTM and the improvement on the test measuring procedural knowledge (i.e., implicit knowledge) under the recast condition, the findings of the two studies appear to be inconsistent. Yilmaz and Granena attributed the different results to the saliency of the target structures and the way the oral production test was designed. While the English past progressive (i.e., was/were -ing) in Révész (2012) is more physically salient in its form, Spanish gender agreement in the current study was less salient, making it difficult for those with high PSTM to notice the structure when given recasts. Furthermore, the oral production test in Yilmaz and
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
677
678
HYEJIN AN AND SHAOFENG LI
Granena’s study likely did not measure procedural knowledge; rather it assessed declarative knowledge, which was more clearly tested in the GJT. In their study, the participants were asked to spot the difference between two pictures, which constitutes a discrete item test that likely measured explicit knowledge. In Revesz (2012), learners were required to describe what 8 people were doing for 10 minutes, and this task is relatively spontaneous and is likely a better measure of procedural knowledge. Zhao (2015), which found a positive correlation between PSTM and modified outputs, also revealed that the learners’ WM was more likely to predict the learning gains under the explicit correction condition. When analyzing the separate roles PSTM and EWM measured by nonword repetition and listening span respectively, it was shown that PSTM was significantly correlated with learning gains in the delayed posttest while EWM had a significant association with development on the immediate posttest.
30.7.2 Timing of Feedback Some studies have investigated whether working memory is related to the timing of corrective feedback. For example, Li et al. (2019) investigated the interaction between WM and the effectiveness of form-focused instruction provided at different stages of a form-focused lesson. A total of 150 EFL learners in 8th grade were provided the instruction about the English past passive in five different conditions: (1) explicit pretask instruction + task, (2) within-task interactional feedback, (3) both pretask instruction and withintask feedback, (4) posttask feedback, and (5) task only (i.e., control group). Dictogloss tasks were used, where the teacher presented a narrative, followed by group work, where the learners practiced retelling the narrative before presenting the narrative to the rest of the class. The pretask instruction involved a 10-minute grammar lesson about the target structure. Corrective feedback consisted of a prompt followed by a recast. The analysis of the relationship between WM measured by operation span and the learning gains assessed via a GJT and an EIT revealed that learners’ WM significantly mediated the development of the groups that received feedback during task performance (“within-task feedback” group and “pretask instruction and within-task feedback” group). The within-task feedback, regardless of whether the learners received pretask grammar lesson or not, imposed heavy cognitive processing load on learners’ WM, and consequently those with higher WM learned more. In the absence of feedback, however, the learners did not have to employ their cognitive resources to process ongoing external instruction, hence the lack of significant effects. The significance of WM in processing within-task feedback can be corroborated by Sanz et al. (2016), who found WM’s predictive power on L2 development when learners are challenged to process feedback during task performance. In their study, 44 college students learning Spanish as a foreign language attended computerized L2 lessons about the Latin “who
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
did what to whom” structure during which computerized metalinguistic feedback was provided. Prior to this main task, one group of students was given a structured pretask grammar lesson while the other group were “immediately immersed” (p. 669) into the main practice. The study found significant effects for working memory for the group that did not receive pretask grammar but not for the group that received pretask grammar instruction. The authors speculated that the grammar lesson in this study was highly structured and explicit, and consequently leveled off the learners’ individual differences in WM for L2 development. Yilmaz and Sagdic (2019) examined the predictive role of inhibitory control, one of the functions of the central executive, in relation to the effects of “immediate” and “delayed” provision of feedback. During oneway information gap interaction, the learners’ errors in the immediate feedback group received recasts. In the delayed feedback group, on the other hand, the errors were ignored during the task but addressed after the task. The participants’ inhibitory control was assessed by using an Eriksen Flanker test, which tapped learners’ ability to suppress the irrelevant information to make a quick decision. Their learning gains, measured by an oral production test and a GJT, were found not to be mediated by inhibitory control. The finding was attributed to the task environment of text-based SCMC for both feedback conditions. The written interaction in SCMC allows for extra time to process the input in the feedback and enables the learners to revisit the feedback in the chat, which may have reduced their cognitive load and leveled out the role of inhibitory control.
30.7.3 Modality of L2 Interaction As discussed in Yilmaz and Sagdic (2019) above, it is generally assumed that the cognitive load on WM is reduced in the SCMC condition due to the slower speed of interactional exchanges and the permanent nature of written information that allows the learners to check the target input multiple times. This section synthesizes the studies examining this assumption by comparing SCMS and face-to-face interaction. Payne and Whitney (2002) investigated whether WM plays different roles in two learning conditions. The experimental group engaged in hybrid instruction consisting of both face-to-face and SCMC while the control group only participated in FTF instruction. Both experimental and control groups participated in a 15week Spanish course, which required a variety of group interactions in the L2 (e.g., topic discussions, drill-and-practice with feedback, collaborative research, and writing projects). The learners’ oral proficiency was tested by using oral production tasks before and after the course. Their PSTM and EWM were measured by a nonword repetition test and a reading span test, respectively. No correlation was found between EWM and the learning gains in either the experimental or the control groups (r = .09). PSTM was found to have a weak association with the oral proficiency
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
679
680
HYEJIN AN AND SHAOFENG LI
development (r = .30) in both conditions; however, it showed a higher correlation with the scores of the control group, who engaged in only FTF interaction, than with the group that participated in both FTF and SCMC interaction. Sagarra and Abbuhl (2013) also found evidence that WM has a positive association only with the effects of orally delivered feedback within an SCMC context. A total of 218 beginner-level students of Spanish participated in computerized grammar practices, and their incorrect responses were treated with seven different ways of automated feedback: (1) no feedback, (2) written utterance rejection, (3) oral utterance rejection, (4) written recasts, (5) oral recasts, (6) written (typographically) enhanced recasts, and (7) orally enhanced recasts with stress on the target structure. A reading span test was used to measure the participants’ WM, and their learning gains were tested by written tests and oral FTF interactional tests. The learners’ WM was found to mediate the L2 development in the oral recast group regardless of enhancement type. The results were interpreted as suggesting that processing written input is less cognitively demanding than oral input, and that it diminishes the learners’ reliance on WM. Martin (2018) examined the role of working memory in text-based, videobased, and voice-based SCMC. To prevent the learners from rereading the feedback in the written L2 interaction, the researcher designed the textbased communication in such a way that the participants had to close the chat after the metalinguistic feedback was given and open it again to resume the interaction. Sixty-five intermediate learners of Spanish engaged in an interactive dialogic story retell task, during which the learners’ errors related to the target structure (i.e., Spanish past subjunctive) were corrected via metalinguistic feedback. They were randomly assigned to video-, voice-, and text-based L2 interaction group. It was hypothesized that WM would mediate the development in video- or voice-based interaction due to the more taxing level of oral communication. However, the results revealed that all the participants successfully acquired the target structure under the three communication modes, regardless of their WM. WM did not play any role in any modes in SCMC probably because the metalinguistic feedback was explicit and the task was not difficult for the learners, even for those in the video- and voice-based communication. In the same way that the above-mentioned studies suggested an overall insignificant role of WM in SCMC interaction (especially for the chat-based communication), Baralt’s studies (2010, 2015) also did not find a significant relationship between WM and L2 learning via SCMC. The results were consistently attributed to the lower cognitive demands in written chatbased interaction. Baralt further examined the interface among WM, interaction modality, and task complexity (i.e., simple versus complex task conditions), and revealed that WM does not predict L2 development within SCMC context regardless of task complexity.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
30.7.4 Task Complexity According to the Cognition Hypothesis, complex tasks impose higher processing demands on learners’ cognitive resources, and therefore individual differences in WM may play a greater role under complex task conditions than simple task conditions. Baralt (2010, 2015) conducted experimental studies to provide empirical evidence regarding this speculation. Both studies used the same task type – two-way information gap tasks (i.e., story completion) manipulated by task complexity (intentional reasoning). While the first study, Baralt (2010), differentiated the interaction modality (FTF vs. SCMC) resulting in four experimental conditions (i.e., FTF-Simple; FTF-Complex; CMC-Simple; CMC-Complex), Baralt (2015) only included the two groups involving the SCMC environment (i.e., Simple vs. Complex in SCMC). The complex task required the participants to draw on intentional reasoning to complete the story by providing empty thought bubbles in the comic strip. The simple version, however, provided the characters’ thoughts and only asked the learners to retell the story without requiring learners to engage in intentional reasoning. Having different sets of picture cards, the participant and the researcher interacted to complete the story. The improvement in productive knowledge and receptive knowledge was measured by means of story retell and multiple-choice tests, respectively. WM was assessed by reading span, operation span, and counting span tests. The studies revealed that WM did not have a significant correlation with the improvement of productive knowledge in either the FTF or the SCMC modality. When it comes to the receptive knowledge, however, Baralt (2010) found that WM had a significant effect on the simple task condition in the face-to-face modality. Baralt (2015) reported that WM does not play any role in SCMC, regardless of task complexity. These findings do not support the Cognition Hypothesis that predicts a more significant role for WM as task complexity increases. The complex FTF task in Baralt (2010) may have been too challenging, even for those with higher WM, because it asked the learners not only to attend to the real-time conversational exchanges but also to come up with intentional reasoning connected to the characters’ thoughts. The level of task complexity went beyond their WMC to the extent that WM is not significant. In the simple face-to-face task, the overburdened cognitive processing load was alleviated, and those with a greater WMC were better able to attend and process the target structure during interaction. Similarly, the complex task in SCMC mode in Baralt (2015) may have been extremely demanding for all learners because they had to type out, read, and send the messages, and draw on intentional reasoning at the same time. The findings of Baralt’s studies, therefore, suggest that the hypothesis regarding a greater role for WM in complex tasks can be undermined when the task-design is too complex.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
681
682
HYEJIN AN AND SHAOFENG LI
Kim et al. (2015), on the other hand, found evidence that lends support to the Cognition Hypothesis. The complexity of collaborative tasks was manipulated by reasoning demands. In the simple task, the learners simply exchanged information with a native speaker regarding the topics of traveling, college life, and cell phones. In the complex task condition, the participants were asked not only to exchange information but also to make a decision on the best solution based on several criteria. Treatment effects were evaluated by using oral production tests, and WM was measured via a running span test. Although it was found that WM was the only significant predictor for L2 development, post hoc analyses showed that learners’ WM is more strongly associated with the L2 development of the group who performed complex tasks. It suggests that a greater WM can be beneficial when the interactional task is cognitively more demanding. The finding contradicts that of Baralt’s (2010, 2015) studies where WM did not mediate performance in the complex task condition. The different levels of complexity imposed by +/- intentional reasoning and +/- reasoning demands may explain a nuanced and differential involvement of WM in task complexity. The task with +reasoning demands may have imposed a proper level of complexity by only asking them to think about the criteria and choose the best option. The complex task with +intentional reasoning, however, required the learners to guess what the story character might have thought. The latter may have exceeded the cognitive limit of all the learners and therefore failed to favor those with higher WMC. The following is a summary of the evidence on the associations between WM and treatment effects. 1. Some studies showed that WM predicts the effects of implicit feedback while other studies found it important in explicit feedback. However, there seems to be a consensus in that WM significantly mediates the efficacy of corrective feedback when the feedback imposes a heavy processing load on the learner. 2. With regard to working memory’s interface with the timing of formfocused instruction, WM plays a significant role when the form-focused instruction is provided during, but not before or after, task performance. Inhibitory control is not predictive of learning gains of any timing condition. 3. WM does not mediate the learning gains resulting from chat-based L2 interaction (i.e., written SCMC). However, when the SCMC involves orally delivered recasts, WM may mediate L2 development. 4. According to the Cognition Hypothesis, learners’ WM would mediate L2 development when learners perform cognitively demanding tasks. The hypothesis holds true when task complexity is within learners’ processing capacity but not when the task goes beyond the limit of their cognitive capacity.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
30.8
What Is the Role of WM in the Interaction-Driven Learning Gains When Instruction Is Not Manipulated?
Some studies have examined the role of WMC in L2 development during interaction without manipulating instruction type. In this section, we synthesize the findings of these studies and we divided them into two groups based on whether the studies investigated short or long treatments.
30.8.1 Short Treatments Mackey and Sachs’s (2012) study examined whether older learners with higher WM benefit from L2 interaction. Nine ESL learners with a mean age of 72 engaged in three communicative tasks (i.e., spot the difference, picture drawing, and picture sequencing), in which each of them interacted with a native speaker and was given feedback on non-target-like question forms. Their L2 development with English question formation was tested by spot-the-difference and picture-description tasks. PSTM and EWM were measured by nonword repetition and listening span (LSPAN), respectively. The correlation analysis of the development and WM scores revealed that 4 out of 9 learners who had highest LSPAN scores significantly improved on the immediate posttest. Furthermore, 2 of the learners with high EWM showed sustained improvement on the delayed-posttest. The result is consistent with Mackey et al. (2002), who found the learners who have high WMC showed more learning gains on delayed posttests. Although the results of both studies are limited in generalizability due to the small sample size, they suggest that high WMC allows the learners to retain the acquired knowledge for a longer period of time. Gass et al. (2013) examined how individual differences in inhibitory control and EWM predict L2 development via interaction. The researchers hypothesized that the predictive power of inhibitory control would be stronger than that of EWM because it directly functions as a mechanism to control the learners’ attention during L2 learning. In the treatment session, 27 college students who were learning Italian as a foreign language performed an object placement task in which they received feedback on gender and number agreement. Their inhibitory control ability was measured by using a Stroop test. A reading span test was adopted to measure EWM, and picture description tasks were used as pre- and posttests. The results showed that only inhibitory control played a significant role in the development – those who have less inhibitory control had low gain scores. EWM was not found to play any role in the learning gains. The study suggests the importance of examining the role of inhibitory control in L2 interaction-driven learning.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
683
684
HYEJIN AN AND SHAOFENG LI
30.8.2 Long Treatments Two studies investigated whether WM mediates learning gains over a longer period of time (i.e., a semester or a year). Payne and Ross (2005) investigated the role of WM in L2 development in an interaction-based online course. Twenty-four college students enrolled in a Spanish course attended 20 chat sessions involving various types of group interaction (e.g., role-plays, discussions of personal themes, jigsaw reading activities). The oral proficiency development was measured by means of oral production tasks at the beginning and end of the semester. Nonword repetition and reading span tests were used to assess the participants’ PSTM and EWM respectively. The results showed that while EWM did not play a role in the development of oral proficiency, the learners who had higher PSTM showed significantly more learning gains than those with lower PSTM over the 15 weeks of the course. While Payne and Ross exclusively examined the development of oral proficiency, White (2021) explored whether WM predicts a variety of dimensions of L2 learning (i.e., syntax, pragmatics, semantics) in interaction-based instruction that lasted a year. The participants were 27 young ESL learners, aged between 5 and 6 years, in the first year of formal immersive education. Under the immersion setting, they were not only taught different subjects in the L2 but also interacted with the teacher and peers in the L2. Throughout the year, their L2 proficiency was tested at three time points: at the beginning (T1), in the middle (T2), and at the end (T3) of the year. The outcome variables included L2 development in vocabulary, syntactic structures, semantic interpretation (e.g., verb contrast, preposition contrast, and quantifiers), and pragmatics (e.g., turn-taking, short narratives, and question-asking). Two WM tests were administered, comprising nonword repetition for PSTM and odd-one-out for visuospatial memory. The odd-one-out task asked the children to remember and recall the position of a shape that is different from the others. The analysis of WM and longitudinal language outcomes revealed positive and significant correlations between all WM measures and all linguistic performance measures across the year. When it comes to the predictive power of WM measures at T1 and T2 on the performance at T3, PSTM at T1 predicted the development on semantics while PSTM at T2 was related to the improvement in pragmatics at T3. Visuospatial memory, on the other hand, did not predict any outcomes at T3. The researcher explained that the test of semantics asked the learners to recognize phonotactic patterns and use them by remembering the information; therefore, those with higher PSTM performed better on the semantics. For the pragmatics test, the children had to hold L2 phonological representations to formulate questions and build stories, and the task might have imposed high cognitive demands on their PSTM. A summary of the relationship between WM and L2 interaction-driven learning gains is provided in the following.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
1. Learners’ WM predicts not only the immediate but also sustained L2 development in interaction-based learning. Learners’ ability to suppress irrelevant information (i.e., inhibitory control) can also be a predictor of L2 learning through interaction. 2. L2 speakers’ PSTM plays a significant role in long-term L2 developments for both adult and young learners
30.9
Conclusion
The aim of the current synthetic review is twofold: to summarize the methodological features of the studies investigating the relationship between WM and L2 interaction and to provide a narrative synthesis of their findings. Due to the limited space, the results were discussed focusing on the major findings of the included studies, so they may not be exhaustive. Also, as the synthesis did not aggregate the quantitative evidence, it does not provide a definitive conclusion for each research question. Despite the limitations, the present chapter provides an in-depth discussion of the role of WM in multiple aspects of L2 interaction. First, WM is likely to predict a variety of interactional behaviors such as noticing, modified output, syntactic priming, and task performance, but its predictive power is constrained by the instructional context, complexity of the target structure, learners’ proficiency, and type of feedback. Second, WM is closely related to treatment effects when L2 learners have a heavy processing load, when feedback is manipulated along the dimensions of explicitness, timing, modality, and task complexity. Finally, when instruction is not experimentally manipulated, learners’ WM is likely to facilitate both immediate and sustained L2 development via short-term interactional treatments. Inhibitory control is also related to interaction-based learning gains. PSTM has been found to be a predictor of longitudinal L2 development. It should be noted that the results above are suggestive rather than conclusive, as the literature has been limited in both amount and scope. Several areas merit more research to enhance the generalizability of findings. First, although CAF are important process features that can describe learners’ L2 speech performance, only two studies were conducted to examine the effects of WM on CAF during interaction. Second, the topic of task complexity in L2 interaction received scant attention from WM researchers (only three studies) although it is hypothesized that learners are more likely to rely on their attentional resources in WM in complex tasks than in simple tasks. Third, more longitudinal studies are needed to verify the facilitative role of WM in the L2 learning trajectory found in the two existing studies. Last but not least, it is still questionable how WM predicts the development of implicit and explicit knowledge through L2 interaction. Although a few studies distinguished implicit and explicit knowledge (oral
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
685
686
HYEJIN AN AND SHAOFENG LI
production test and grammaticality judgment test for each), the validity of the measurements is controversial, especially the oral production tests designed to measure implicit knowledge. In addition, there are numerous gaps that need to be filled. One is the role of WM in the process aspects of L2 interaction, such as the amount of speech or the number of turns between interlocutors. The relationship between learners’ WM and their emotions and attitudes during L2 interaction (e.g., willingness to communicate, anxiety, and enjoyment) has not been examined in the research. Another research gap is the role of WM in aspects of speech performance other than CAF, such as collocations or certain grammatical forms (e.g., past tense form). Such efforts would make our current understanding regarding the role of WM in L2 interaction more accurate and enable practitioners to better accommodate learners’ individual differences in WM in curriculum development and instruction.
References (Studies with asterisks are included in the synthesis) Abrams, Z. I. (2003). The effect of synchronous and asynchronous CMC on oral performance in German. Modern Language Journal, 87(2), 157–167. *Ahmadian, M. J. (2020). Explicit and implicit instruction of refusal strategies: Does working memory capacity play a role? Language Teaching Research, 24(2), 163–188. Baddeley, A. D. (2003). Working memory: Looking back and looking forward. Neuroscience, 4, 829–839. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In S. Dornic (Ed.), Recent advances in learning and motivation (pp. 47–89 ). Academic Press. *Baralt, M. (2010). Task complexity, the cognition hypothesis, and interaction in CMC and FTF environments (Doctoral dissertation, Georgetown University). *Baralt, M. (2015). Working memory capacity, cognitive complexity and L2 recasts in online language teaching. In Z. Wen, M. B. Mota, & A. McNeill (Eds.), Working memory in second language acquisition and processing (pp. 248–269). Multilingual Matters. *Chen, X. (2013). Chinese EFL learners’ noticing of recasts: Its relation to target structures, uptake, and working memory capacity (Doctoral dissertation, Michigan State University). Crookes, G. (1989). Planning and interlanguage variation. Studies in Second Language Acquisition, 11(4), 367–383. *Dai, B. (2013). Individual differences in learners’ working memory, noticing of L2 forms in recasts and their L2 development in task-based interactions (Doctoral dissertation, The Chinese University of Hong Kong). Ellis, R. (2009). Implicit and explicit learning, knowledge and instruction. In R. Ellis, S. Loewen, C. Elder, R. Erlam, J. Philp, & H. Reinders (Eds.), Implicit
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
and explicit knowledge in second language learning, testing and teaching (pp. 3–25). Multilingual Matters. Gass, S. (1997). Input, interaction, and the second language learner. Lawrence Erlbaum. Gass, S. (2003). Input and interaction. In C. Doughty & M. H. Long (Eds.), The handbook of second language acquisition (pp. 224–255). Basil Blackwell. *Gass, S., Behney, J. N., & Uzum, B. (2013). Inhibitory control, working memory, and L2 interaction. In K. Drozdzial-Szelest, & M. Pawlak (Eds.), Psycholinguistic and sociolinguistic perspectives on second language learning and teaching (pp. 91–114). Springer. *Georgiadou, E., & Roehr-Brackin, K. (2017). Investigating executive working memory and phonological short-term memory in relation to fluency and self-repair behavior in L2 speech. Journal of Psycholinguistic Research, 46(4), 877–895. Gilabert, R., & Muñoz, C. (2010). Differences in attainment and performance in a foreign language: The role of working memory capacity. International Journal of English Studies, 10(1), 19–42. *Goo, J. (2012). Corrective feedback and working memory capacity in interaction-driven L2 learning. Studies in Second Language Acquisition, 34(3), 445–474. *Goo, J. (2016). Corrective feedback and working memory capacity. In G. Granena, D. O. Jackson, & Y. Yilmaz (Eds.), Cognitive individual differences in second language processing and acquisition (pp. 279–302). Benjamins. Hartsuiker, R. J., Kolk, H. J., & Huiskamp, P. (1999). Priming word order in sentence production. Cognition, 1, 129–147. Juffs, A., & Harrington, M. (2011). Aspects of working memory in L2 learning. Language Teaching, 44(2), 137–166. *Kim, Y. J., Jung, Y. J., & Skalicky, S. (2019). Linguistic alignment, learner characteristics, and the production of stranded prepositions in relative clauses: Comparing FTF and SCMC contexts. Studies in Second Language Acquisition, 41(5), 937–969. *Kim, Y. J., Payant, C., & Pearson, P. (2015). The intersection of task-based interaction, task complexity, and working memory. Studies in Second Language Acquisition, 37(3), 549–581. *Lai, C., Fei, F., & Roots, R. (2009). The contingency of recasts and noticing. CALICO Journal, 26(1), 70–90. Levelt, W. J. M. (1989). Speaking: From intention to articulation. MIT Press. *Li, S. (2013). The interactions between the effects of implicit and explicit feedback and individual differences in language analytic ability and working memory. Modern Language Journal, 97(3), 634–654. *Li, S. (2015). The differential roles of language analytic ability and working memory in mediating the effects of two types of feedback on the acquisition of an opaque linguistic structure. In C. Sanz, and B. Lado (Eds.), Individual differences, L2 development & language program administration: From theory to application (pp. 32–52). Cengage Learning.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
687
688
HYEJIN AN AND SHAOFENG LI
Li, S. (2017). The effects of cognitive aptitudes on the process and product of L2 interaction: A synthetic review. In L. Gurzynski (Ed.), Expanding individual difference research in the interaction approach: Investigating learners, instructors and researchers (pp. 41–70). Benjamins. *Li, S., Ellis, R., & Zhu, Y. (2019). The associations between cognitive ability and L2 development under five different instructional conditions. Applied Psycholinguistics, 40(3), 693–722. Li, S., & Fu, M. (2018). Strategic and unpressured within-task planning and their associations with working memory. Language Teaching Research, 22(2), 230–253. Linck, J. A., Osthus, P., Koeth, J. T., & Bunting, M. F. (2014). Working memory and second language comprehension and production: A metaanalysis. Psychonomic Bulletin and Review, 21(4), 861–883. Long, M. H. (1996). The role of the linguistic environment in second language acquisition. In W. Ritchie and T. Bhatia (Eds.), Handbook of language acquisition (Vol. 2): Second language acquisition (pp. 413–468). Academic Press. Long, M. H. (2015). Second language acquisition and task-based language teaching. Wiley. *Mackey, A., Adams, R., Stafford, C., & Winke, P. (2010). Exploring the relationship between modified output and working memory capacity. Language Learning, 60(3), 501–533. *Mackey, A., & Sachs, R. (2012). Older learners in SLA research: A first look at working memory, feedback, and L2 development. Language Learning, 62(3), 704–740. *Mackey, A., Philp, J., Egi, T., Fujii, A., & Tatsumi, T. (2002). Individual differences in working memory, noticing of interactional feedback and L2 development. In P. Robinson (Ed.), Individual differences and instructed language learning (pp.181–209). Benjamins. *Martin, A. (2018). How to synchronize? A study of video-based, voice-based, and text-based synchronous computer-mediated communication, working memory, and second language learning (Doctoral dissertation, Georgetown University). McDonough, K. (2005). Identifying the impact of negative feedback and learners’ responses on ESL question development. Studies in Second Language Acquisition, 27, 79–103. *McDonough, K., Kielstra, P., Crowther, D., & Smith, G. (2016). Structural priming in L2 speech production: Examining relationships among English L2 speakers’ production, cognitive abilities, and awareness. In A. Mackey & E. Marsden (Eds.), Advancing methodology and practice: The IRIS repository of instruments for research into second languages (pp. 112–131). Routledge. *McDonough, K., & Kim, Y. (2016). Working memory and L2 English speakers’ primed and subsequent production of passives. In G. Granena,
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Second Language Interaction
D. O. Jackson & Y. Yilmaz (Eds.), Cognitive individual differences in second language processing and acquisition (pp. 205–222). Benjamins. Miyake, A., & Friedman, N. (1998). Individual differences in second language proficiency: Working memory as language aptitude. In A. Healy & L. Bourne (Eds.), Foreign language learning: Psycholinguistic studies on training and retention (pp. 339–365). Erlbaum. Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D. (2000). The unity and diversity of executive functions and their contributions to complex “frontal lobe” tasks: A latent variable analysis. Cognitive Psychology, 41(1), 49–100. *Payne, J. S., & Ross, B. M. (2005). Synchronous CMC, working memory, and L2 oral proficiency development. Language Learning and Technology, 9(3), 35–54. *Payne, J. S., & Whitney, P. J. (2002). Synchronous CMC: Output, working memory, and interlanguage development. CALICO Journal, 20(1), 7–32. *Révész, A. (2012). Working memory and the observed effectiveness of recasts on different L2 outcome measures. Language Learning, 62(1), 93–132. Robinson, P. (2001). Task complexity, task difficulty, and task production: Exploring interactions in a componential framework. Applied Linguistics, 22, 27–57. Robinson, P. (2003a). Attention and memory in SLA. In C. Doughty & M. Long (Eds.), Handbook of second language acquisition (pp. 631–678). Blackwell. Robinson, P. (2003b). The cognition hypothesis, task design, and adult taskbased language learning. Second Language Studies, 21(2), 45–105. Robinson, P. (2007). Criteria for classifying and sequencing pedagogic tasks. In M. P. Garcia Mayo (Ed.), Investigating tasks in formal language learning (pp. 7–26). Multilingual Matters. Robinson, P. (2011). Second language task complexity, the Cognition Hypothesis, language learning, and performance. In P. Robinson (Ed.), Second language task complexity: Researching the cognition hypothesis of language learning and performance (pp. 3–37). Benjamins. *Sachs, R. R. (2010). Individual differences and the effectiveness of visual feedback on reflexive binding in L2 Japanese (Doctoral dissertation. Georgetown University). *Sagarra, N., & Abbuhl, R. (2013). Optimizing the noticing of recasts via computer-delivered feedback: Evidence that oral input enhancement and working memory help second language learning. Modern Language Journal, 97(1), 196–216. *Sanz, C., Lin, H. J., Lado, B., Stafford, C. A., & Bowden, H. W. (2016). One size fits all? Learning conditions and working memory capacity in ab initio language development. Applied Linguistics, 37(5), 669–692.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
689
690
HYEJIN AN AND SHAOFENG LI
Scheepers, C. (2003). Syntactic priming of relative clause attachments: Persistence of structural configuration in sentence production. Cognition, 89(3), 179–205. Schmidt, R. (2001). Attention. In P. Robinson (Ed.), Cognition and second language instruction (pp. 3– 32). Cambridge University Press. Skehan, P. (1989). Individual differences in second language learning. Arnold. Skehan, P. (1998). A cognitive approach to language learning. Oxford University Press. Skehan, P. (2002). Theorizing and updating aptitude. In P. Robinson (Ed.), Individual differences and instructed language learning (pp. 69–93). Benjamins. Swain, M. (1995). Three functions of output in second language learning. In G. Cook & B. Seidlhofer (Eds.), Principle and practice in applied linguistics: Studies in honour of H. G. Widdowson (pp. 125– 44 ). Oxford University Press. Wen, Z. (2016). Phonological and executive working memory in L2 taskbased speech planning and performance. Language Learning Journal, 44(4), 418–435. Wen, Z., & Li, S. (2019). Working Memory in L2 Learning and Processing. In J. Schwieter & A. Benati (Eds.), The Cambridge handbook of language learning (pp. 365–389). Cambridge University Press. *White, M. J. (2021). Phonological working memory and non-verbal complex working memory as predictors of future English outcomes in young ELLs. International Journal of Bilingualism, 25(1), 318–337. *Wright, C. (2013). An investigation of working memory effects on oral grammatical accuracy and fluency in producing questions in English. TESOL Quarterly, 47(2), 352–374. *Yilmaz, Y. (2013). Relative effects of explicit and implicit feedback: The role of working memory capacity and language analytic ability. Applied Linguistics, 34(3), 344–368. *Yilmaz, Y., & Granena, G. (2019). Cognitive individual differences as predictors of improvement and awareness under implicit and explicit feedback conditions. Modern Language Journal, 103(3), 686–702. *Yilmaz, Y., & Sağdıç, A. (2019). The interaction between inhibitory control and corrective feedback timing. International Journal of Applied Linguistics, 170(2), 204–227. Zalbidea, J. (2017). “One task fits all”? The roles of task complexity, modality, and working memory capacity in L2 performance. Modern Language Journal, 101(2), 335–352. *Zhao, Y. (2015). The effects of explicit and implicit recasts on the acquisition of two grammatical structures and the mediating role of working memory (Doctoral dissertation, The University of Auckland).
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Appendix A Studies on the Role of WM in L2 Interactional Behaviors Publication type
Study setting
Measurements of the behaviors
Measurements of WM
Mackey et al* (2002)
Chapter
lab
picture drawing; picture difference, story completion
-Noticing: stimulated recall, questionnaire -Development: interactional tasks
-EWM: LSPAN -PSTM: NWR
Kim et al* (2015)
Journal
lab
two-way collaborative information gap (simple vs. complex tasks)
-Noticing: immediate cued recall -Development: 3 oral production tests
-EWM: aural running span
Chen (2013)
Dissertation
lab
one-way information gap
-Noticing: stimulated recall in L1
-EWM: RSPAN
Dai (2013)
Dissertation
lab
Spot-the-difference picture drawing story telling
-EWM: LSPAN -PSTM: NWR
Lai et al. (2008)
Journal
lab
Spot-the-difference
-Noticing: immediate stimulated recall, traditional stimulated recall -Learning: Interactional tasks -Noticing: think-aloud protocols, stimulated recalls
Study WM & Noticing
Treatment tasks
-EWM: Reverse digit span test
Major Findings -High WMC predicted more noticing. -Learners with low WMC showed more development in immediate post-test while high WMC learners outperformed in delayed post-test. -WM had a significant positive association with noticing and L2 development regardless of task complexity. -No significant relationship was found between WM and noticing. -WM had a significant relationship with noticing of English past tense, but not with English question forms. -WM had a strong significant relationship with noticing of noncontingent recasts.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Publication type
Study setting
Mackey et al. (2010)
Journal
lab
Zhao* (2015)
Dissertation
classroom
McDonough et al. (2016)
Chapter
lab
two-way information exchange tasks
McDonough & Kim (2016)
Chapter
lab
Picture description
Kim et al. (2019)
Journal
lab
interactive object description tasks
Study WM& Modified output
WM & Structural priming
Treatment tasks a map task; picture drawing task; spotthe-difference task; a story completion two way one-way information gap tasks (focused task)
Measurements of the behaviors
Measurements of WM
-Response to feedback: Interactional tasks
-EWM: LSPAN
-Learners with higher WMC modified their output more.
-Uptake: recording of the treatment; exit questionnaire -Implicit knowledge: EIT and oral production tests; and explicit knowledge: the written production and untimed GJT -Priming pattern: information-exchange activities -Awareness: Questionnaire -Priming pattern: picture description
-EWM: LSPAN -PSTM: NWR
-The only significant association was found between PSTM and the repairs after implicit recasts.
-EWM: Backward digit span task
-No significant relationship was found between WM and learners’ priming behavior. -WM was significantly positively correlated with the production of structural priming in the absence of intervening sentences between the prime sentence and the target picture description. -No significant relationship was found between WM and structural priming.
-proficiency: cloze test -prior knowledge: GJT -learning effects: production tests (picture description)
-EWM: aural running span
-EWM: aural running span
Major Findings
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
WM & CAF
Georgiadou & RoehrBrackin (2017)
Journal
lab
one-way information gap
-self-repair behavior: oral interview
-EWM: backward digit span test; LSPAN-PSTM: serial word recall test
Wright (2013)
Journal
lab
immersive language learning
-fluency and accuracy: linguistic task (Q/A two-way gap-fill task)
-EWM: LSPAN, Story Recall
-EWM was found to be negatively correlated with the number of pauses (only for intermediate learners, not for beginners). -No significant relationship was found between WM and self-repair behavior. -WM was significantly correlated with the improvement of accuracy, but not with fluency.
Notes. EIT: Elicited Imitation Test., GJT: Grammaticality Judgment Test., EWM: Executive Working Memory., PSTM: Phonological Short-Term Memory., LSPAN: Listening Span Task., NWR: Nonword Repetition Task., RSPAN: Reading Span Task., WMC: Working Memory Capacity. * Studies addressing more than one question in the synthesis.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Appendix B Studies on the role of WM in the effects of interactional conditions
Explicitness of feedback
Study
Publication Type
Study setting
Treatment tasks
Experimental conditions
Révész (2012)
Journal
lab
Picture description
recast vs. no-recast
Goo (2012)
Journal
classroom one-way info gap activities
Goo (2016)
Chapter
lab
Li (2013)
Journal
lab
Li (2015)
Chapter
lab
Sachs (2011)
Dissertation
lab
Yilmaz (2013)
Journal
lab
Yilmaz & Granena (2019)
Journal
lab
1) recast 2) metalinguistic feedback 3) control one-way info gap 1) recast activities 2) metalinguistic feedback 3) control -no feedback picture description, 1) implicit (recasts) spot-the-difference 2) explicit (metalinguistic feedback) 3) control (no feedback) video narration; 1) implicit (recasts) interview tasks 2) explicit (metalinguistic feedback) 3) control (no feedback) Truth-value judgment 1) Right/Wrong tasks Feedback 2) Tree Diagram Feedback 3) No Feedback one-way info gap task 1) recast 2) explicit correction 3) control one-way communicative tasks
1) explicit 2) implicit 3) control
Measurements of the efficacy of the conditions
Measurements of WM
Major Findings
-Learning: GJT, written picture description, oral description -Learning: GJT & a written production test
-PSTM: DS, NWS -EWM: RSPAN
-WM was correlated with the effects of recasts.
-EWM: RSPAN, OSPAN
-WM significantly predicted the effects of recasts, not metalinguistic feedback.
-Learning: Oral production, GJT
-EWM: OSPAN
-WM did not mediate the effects of either type of feedback
-Learning: GJT, EIT
-EWM: LSPAN
-WM mediated the effects of metalinguistic feedback.
-Learning: GJT, EIT
-EWM: LSPAN
-WM had a significant negative correlation with the effects of the metalinguistic feedback.
-Learning: truth-value judgment tasks
-VSM: Visual patterns test
-VSM was significantly correlated with learning gains only for Treediagram feedback group. -EWM: OSPAN -WM had a positive -Learning: oral significant correlation production, with the effects of comprehension, and explicit correction. recognition tests -Learning: oral -PSTM: Letter Span test -PSTM predicted the improvement on the GJT, production task, GJT -Attention Control: Task only in the explicit Switching Numbers correction group. Test
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Timing of feedback
Modality of interaction
-EWM: OSPAN -pragmatics development: DCT (discourse completion test); CQ (comprehension questionnaire) -EWM: LSPAN -Learning: aural interpretation, written interpretation, GJT, production test -Learning: GJT, EIT -EWM: OSPAN
Ahmadian (2020)
Journal
classroom role-play tasks
1) implicit instruction with recasts 2) explicit instruction with explicit correction
Sanz et al. (2016)
Journal
lab
2 conditions (+/grammar lesson; metalinguistic information)
Li et al. (2019)
Journal
classroom Dictogloss tasks
Yilmaz & Sagdic (2019)
Chapter
lab
Payne & Whitney (2002)
Journal
classroom online class sessions for collaborative work and language practice
With and without online text-based synchronous interaction
-EWM: RSPAN -Proficiency development: Oral -PSTM: recognitionbased NWR Proficiency Interview (OPI)
Sagarra & Abbuhl (2013)
Journal
lab
1) No feedback; 2) written utterance rejection; 3) oral utterance rejection written recast; 4) oral recast; 5) typographically enhanced recast; 6) orally enhanced recast
-Learning: written -EWM: RSPAN production test, oral FTF interactive tasks -Modifying output pattern: one of the interactive tasks
Computerized grammar practice with feedback
one-way information gap picture description activity
computer-based grammar practice
-Five Conditions 1) Explicit instruction +Task 2) Interactional Feedback 3) Explicit instruction + Interactional Feedback 4) Posttask Feedback 5) Task Only 1) Immediate feedback 2) Delayed feedback
Inhibitory control: -Learning: OPT (oral Eriksen Flanker Task production test) and GJT
-WM had a strong positive relationship with learning gains in the implicit instruction group. -WM predicted learning gains only when metalinguistic feedback was not provided prior to main-task. -WM had a significant relationship with learning gains of the groups who received within-task feedback.
-No significant relationship was found between inhibitory control and learning for both groups. -A stronger correlation between WM and learning gains was found in FTF group than in FTF+CMC group. -WM was significantly correlated with the effects of oral recasts, both enhanced and unenhanced.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Task complexity
Study
Publication Type
Study setting
Martin (2018)
Dissertation
lab
interactive dialogic story retell
Baralt (2010)
Dissertation
lab
two-way information gap
Baralt (2015)
Chapter
lab
two-way interactive story retells
Treatment tasks
Experimental conditions 1) text-based SCMC 2) video-based SCMC 3) voice-based SCMC 1) FTF, complex 2) FTF, simple 3) CMC, complex 4) CMC, simple (+/- intentional reasoning) Simple vs. complex tasks, both in SCMC (+/- intentional reasoning),
Measurements of the efficacy of the conditions
Measurements of WM
Major Findings
-Learning: production assessment tests (oral, written) -Learning: oral production task, multiple-choice test
-EWM: OperationLetter Span Task
-WM did not predict learning gains in any modes. -WM had a significant effect on receptive learning for FTF, simple task group.
-Learning: productive tasks multiplechoice test
-EWM: OSPAN, CSPAN, RSPAN
-EWM: OSPAN, CSPAN, RSPAN
-WM did not play any role in SCMC regardless of task complexity.
Notes. VSM: Visuospatial Memory., DS: Digit Span Task., NWS: Nonword Span Task., OSPAN: Operation Span Task., CSPAN: Counting Span Task., FTF: Face-to-Face Communication., CMC: ComputerMediated Communication., SCMC: Synchronous Computer-Mediated Communication.
https://doi.org/10.1017/9781108955638.036 Published online by Cambridge University Press
Appendix C Studies on the role of WM in the L2 interaction-driven learning gains Publication type
Interaction Study type setting
Mackey & Sachs (2012)
Journal
dyad
lab
spot-the-difference, picture-drawing, and picture-sequencing tasks
Gass et al. (2013)
Chapter
dyad
lab
White (2021)
Journal
dyad
Payne & Ross (2005)
Journal
group
Study short-term development
longitudinal development
Treatment
Measurements of learning gains
Measurements of WM
Major Findings
-Learning: spot-the-difference and picture-sequencing tasks
-EWM: LSPAN -PST: NWR
object placement task with corrective feedback
-Learning: Picture description
-EWM: RSPAN -Inhibition: Stroop test
lab
immersive language learning
-Learning: 1) Vocal (receptive skills): picture vocabulary test 2) Comprehension, production skills: DELV-CR (Developmental Evaluation of Language VariationCriterion Referenced Edition)
-PSTM: NWR -VSM: Odd-one-out
SCMC (classroom)
blended instruction
-Learning: oral production task
-EWM: RSPAN -PSTM: NWR
- Older learners with higher WM benefited from interaction in L2 learning. -Only inhibitory control played a significant role in L2 development. -PSTM consistently predicted learning gains on syntax and pragmatics for 1 year. -PSTM predicted further outcomes on semantics and pragmatics in later time points of the year. -Only PSTM predicted learning gains over the 15 weeks of the course.
31 Working Memory and Interpreting Studies Binghan Zheng and Huolingxiao Kuang
The work was supported by the National Social Science Fund of China (No. 20BYY014). Huolingxiao Kuang’s Ph.D. is funded by China Scholarship Council (CSC).
31.1
Introduction
Interpreting can be conducted either simultaneously or consecutively. In the simultaneous mode, interpreters listen, comprehend and translate the source speech in real time, which requires a coordinated use of limited working memory (WM) resources. In the consecutive mode, interpreters store the source speech in their WM, and then recall the stored speech in the target language by refreshing their WM (Dong et al., 2018). Both interpreting modes place high demands on the storage and processing functions of WM (Baddeley & Hitch, 1974), and require an effective executive control in WM operation (Nour, Struys, Woumans et al., 2020). Overall, previous interpreting research on WM can be divided into two strands. The first strand tests interpreters’ and noninterpreter bilinguals’ information storage and/or processing ability by measuring their WM capacity. The measurement is made using WM span tasks where subjects are required to recall a specific number of presented stimuli. The second strand investigates interpreters’ operation of WM executive control, which includes three executive functions: inhibiting phonological and multitasking interferences (Inhibition), replacing old information with new information as a method of continuous input processing (Updating) and switching between languages and subtasks of interpreting from source speech comprehension to target speech delivery (Shifting) (Nour, Struys, Woumans et al., 2020). So far, contradictory findings have been reported regarding the benefits of interpreting training and work experience on WM capacity improvement and executive control enhancement. Most of the studies in both strands deal with simultaneous interpreting (SI). This is because SI is generally considered much more demanding on the WM than consecutive interpreting (CI) and other interpreting modes (Dong & Cai, 2015) because of its concurrent execution of several subtasks from source
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
speech processing to target speech delivery. Owing to the limited amount of research on CI, there is a lack of comparative discussions on the role of WM in SI and CI, the two most representative interpreting modes. This chapter attempts to fill this gap in the research by systematically reviewing previous research on WM in SI and CI, focusing on the common features and the differences between the two modes from theoretical and empirical perspectives.
31.2
Working Memory in Theoretical Interpreting Studies
This section centers on the theoretical aspects of WM that interpreting researchers have explored to compare SI and CI. Comparisons are made in regards to both the development of WM capacity and the use of WM resources. The first half of the section focuses on the similarities between the two modes, while the second half discusses the differences.
31.2.1 The Similarities between SI and CI Interpreting in general is a complex cognitive activity that places a heavy demand on interpreters’ WM for language control and processing control. It is therefore essential for both simultaneous and consecutive interpreters to expand their WM capacity and interpreting-specific knowledge schemas in long-term memory to meet such great demands of interpreting. Specifics about the demand on WM and the need for WM development in SI and CI are introduced below. 31.2.1.1 Demand on WM for Language Control and Processing Control Baddeley and Hitch’s (1974) Working Memory Model and Cowan’s (1988, 1995) Embedded-Process Model of Memory have been extensively discussed in interpreting studies. Baddeley and Hitch (1974) looked into the “inside” of WM by proposing two slave systems in WM: the phonological loop and the visuospatial sketchpad, which are responsible for audio and visual information processing, respectively. Since the capacity of WM is limited, the two slave systems are coordinated by the commander in WM: the central executive. Through executive control, interpreters can effectively distribute WM resources to the audio channel (such as listening and speaking) and the visual channel (such as reading and note-taking). In contrast, Cowan (1988, 1995) focuses on the “outside” of WM by associating it with memory at different levels. In his model, WM is simply a “set of activated memory elements” (Cowan, 1995, p.100) that belong to long-term memory (LTM). Only when a memory is brought to the focus of attention can it be processed by WM. For example, when interpreters are processing the information in their focus of attention, they activate the part of LTM they need: linguistic knowledge and interpreting skills, for example, to accomplish comprehension or interpretation. The activated contents change constantly
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
699
700
BINGHAN ZHENG AND HUOLINGXIAO KUANG
Figure 31.1 Mizuno’s (2005) model of interpreting based on Cowan’s (1988, 1995) Embedded-Processes Model of Memory Source: Mizuno, A. (2005). Process model for simultaneous interpreting and working memory. Meta, 50(2), 739–752. Reproduced with permission
as the focus of attention moves. Overall, the two models explain how information is processed in interpreting with the two slave systems inside WM and with LTM outside it. Interpreting is by nature a verbal task, and this should be reflected in WM models. With this in mind, Mizuno (2005) added a language comprehension system and a language production system before and after the memory section to Cowan’s (1988, 1995) Embedded-Processes Model (see Figure 31.1). This addition seems to provide a complete description of the working flow of WM in interpreting from language input to language output. However, it neglects one of the most distinctive features of interpreting compared to other language processing tasks: language transfer (Dong & Cai, 2015). In SI, interpreters listen to one language and translate it into another language at almost the same time; in CI, by contrast, interpreters switch to the target language whenever the speaker finishes an utterance, and shift their interpreting direction when the source language changes. This frequent and regular switch between two languages is a distinctive feature of interpreting (Dong & Li, 2020) that should not be excluded from interpreting models. Therefore, Dong and Li (2020) proposed an attentional control model of SI and CI, taking both language control and processing control into consideration (p.10) (Figure 31.2). Under language control, working memory, together with monitoring, target enhancement, target disengagement, and shifting, help interpreters focus their attention on inhibiting the source language and activating the target language. When a particular language input-and-output modality is established and transformed into a schema, interpreters can activate it whenever they need it. This is especially important for CI interpreters, who sometimes interpret for interlocutors from both sides. Whenever the source language changes,
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
Figure 31.2 Dong and Li’s (2020) attentional control model of interpreting Source: Dong, Y. & Li, P. (2020). Attentional control in interpreting: A model of language control and processing control. Bilingualism: Language and Cognition, 23, 716–728. Reproduced with permission
they must switch the inhibited language and the activated language accordingly. In this situation, WM is heavily relied on to achieve focused attention. Under processing control, WM and coordination enable interpreters effectively to divide their attention among the various tasks involved in interpreting, such as listening comprehension, note-taking, target speech production, and self-monitoring. This is vital in SI, where the degree of simultaneity of listening comprehension and target speech production decides the quality of the interpreting product. It is worth mentioning that compared with previous models of WM in interpreting, Dong and Li’s (2020) model not only illustrates how WM operates during the process of interpreting, but also shows how to build interpreting expertise. In this model, the establishment of a language-modality connection and the improvement of language processing efficiency respectively contribute to better language control and processing control. Specifically, during interpreting training, interpreters should learn to form language-modality schemas, enhance their language abilities, and acquire interpreting strategies that will give them better attentional control. Hence, this attentional control model can be used not only as a process model of SI and CI, but also as a developmental model for interpreting training.
31.2.1.2 The Role of WM in the Development of Interpreting Expertise Compared with other language processing tasks, interpreting is especially complex because of its cognitive demand on both information retention and the need for fast processing. This demand is decided by the speaker-paced feature of SI and CI. In SI, interpreters reformulate the information they have heard and articulate it in the target language while storing the continuous
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
701
702
BINGHAN ZHENG AND HUOLINGXIAO KUANG
new input for later processing (Christoffels et al., 2006). In CI, interpreters must remember the whole source speech accurately and search for translation if there is any spare WM during listening. If an interpreter fails to keep up with the speaker, they will miss information, which will affect their comprehension of the source speech. To avoid information loss, interpreters can either increase their WM span to store more information or release their WM space by automatizing some information-processing procedures. Just and Carpenter (1992) observed that college students with larger reading spans had better language comprehension ability than those with smaller spans. Therefore, they claimed that the number of language elements (such as phrases and grammatical structure) individuals can activate in their WM decides the depth of language comprehension they can achieve. For interpreters, listening comprehension consumes at least 80 percent of their cognitive effort during interpreting (Padilla, 1995). If we put the two findings together, we can propose an assumption that a bigger WM span contributes to better language comprehension, which could save interpreters’ some effort in comprehension and improve their output quality during interpreting. An alternative theory to explain Just and Carpenter’s (1992) finding is the theory of expertise (Ericsson & Charness, 1994), which argues that experts outperform novices because of their extensive knowledge in a given domain. For example, if interpreters can process information in meaning chunks rather than individual words, even though the actual size of their WM remains unchanged, the amount of information contained in each unit increases. As a result, interpreters can store and process more information in their WM at a time. From a developmental perspective, once an individual is sufficiently proficient at performing a particular task, they can automize this operation, store it as a schema in LTM and use it at any time with little cognitive effort. For instance, Liu et al. (2004) reported that professional interpreters are more capable of detecting important ideas in speeches than novice interpreters, even though the two groups have a similar WM capacity. This is because professionals have automized this information-detection procedure in their LTM. Therefore, they can implement this procedure in WM without bearing much cognitive load. This interaction between LTM and WM has repeatedly been proved to be essential in providing specific skills and knowledge in tasks with high demands on WM, such as in playing chess (Holding & Reynolds, 1982) and computer programming (Adelson, 1984). However, in theories and models of interpreting studies, LTM has not been well focused when compared to WM. So far, interpreting models that include LTM can be categorized into three kinds. Darò and Fabbro (1994) support an independent and sequential view of WM and LTM. They believe that language is first processed in WM and then in LTM. In WM, interpreters hold what they hear in their phonological loop and conduct subvocal rehearsal to keep the information highly activated. After that, they search in their LTM at all levels (episodic memory, semantic memory, and procedural memory) for comprehension and interpretation.
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
Figure 31.3 Darò and Fabbro’s (1994) model of simultaneous interpreting Source: Darò, V., & Fabbro, F. (1994). Verbal memory during simultaneous interpretation: Effects of phonological interference. Applied Linguistics, 15(4), 365–381. Reproduced with permission
Interpreting is completed as soon as the search is done (see Figure 31.3). Ericsson and Kintsch (1995) proposed an interconnected relationship between WM and LTM, proposing the concept of long-term working memory (LT-WM). Specifically, when an interpreter has well-developed interpreting skills or when interpreting materials are familiar to the interpreter, they can activate corresponding interpreting knowledge in LT-WM with little cognitive effort. By contrast, Cowan (1988, 1995) and Mizuno (2005) proposed a hierarchical view of memory where WM is embedded in LTM (see Figure 31.1). When processing information, interpreters activate a subset of relevant items stored in their LTM and then transport them to WM. In this end, only a small fraction of the items will come into the focus of attention for further processing. The above three models illustrate the interaction between WM and LTM during the process of interpreting but not during the development of interpreting expertise. It is widely accepted that LTM is formed after learning from repeated practice with short-term memory. However, the process of this formation is not yet clear. According to Cognitive Load Theory (Paas & van Merriënboer, 1994), the WM-to-LTM transformation occurs when acquired knowledge and skills are transformed into schemas. When this condition is fulfilled, LTM can be enlarged, thus making available more capacity for other WM operations. However, this transformation requires
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
703
704
BINGHAN ZHENG AND HUOLINGXIAO KUANG
interpreters to bear a particular type of cognitive load called germane load (Sweller, 1988), which is concerned with knowledge acquisition and procedure automation. To accomplish learning something, interpreters must be sure that their WM is not fully occupied by intrinsic load (decided by the task) and extraneous load (decided by the interpreter’s efficiency in WM use), which means that there is spare germane load in their WM.
31.2.2 The Differences between SI and CI Although SI and CI share a great demand of WM resources, they have different requirements to the use of these cognitive resources and rely on different functions of the executive control. The following sections introduce the specific needs of the two interpreting modes in terms of WM resources and executive functions. 31.2.2.1 Particular Requirements of Interpreting Efforts Gile’s (1997/2002) effort model has described how SI and CI differ in the types of effort required (for a detailed review, see Dong & Cai, 2015). Firstly, SI is a single-stage activity where listening (L), memorizing (M), and target speech production (P) happen simultaneously through coordination (C). By contrast, CI is a two-stage activity consisting of comprehension and reformulation. At the comprehension stage, interpreters listen to the source speech (L), memorize as much of the content as possible (M), produce notes for later memory retrieval (NP), and coordinate (C) these activities; at the reformulation stage, interpreters read their notes (NR), reconstruct the source speech from memory (SR), produce the target speech (P), and coordinate all of these activities (C). Therefore, CI is generally considered less demanding on WM than SI as there is more leeway with regard to task coordination and production time. There are only four studies that have directly tested this assumption by comparing cognitive load in SI and CI from a product-oriented perspective, and three of these rejected this assumption. Lambert (1988) found that interpreters had similar free recall performance after SI and CI. It is assumed that the more demanding the task is, the smaller the number of recalled items will be. Therefore, his result suggests a similar level of cognitive demand across SI and CI. Gile (2001) and Russell (2002), respectively, found higher accuracy in SI and CI renditions. Gile (2001) concluded that there was a WM overload in SI, while Russell (2002) came to the opposite conclusion – that there was a WM overload in CI. Liang et al. (2017) compared the dependency distance, i.e., the distance between two syntactically related words, in SI and CI products. Results showed a smaller distance in the CI rendition, indicating higher processing difficulty in the producing the CI output than the SI output. Secondly, although in both SI and CI, efforts are required in the aspects of listening, memory, production, and coordination, the two modes differ in
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
the amount and type of effort required for each aspect. In SI, interpreters finish listening analysis, memory retrieval, and output production within seconds, requiring fast processing in WM; in CI, interpreters comprehend and retrieve the source speech across the two stages of CI, respectively emphasising the processing and storage functions of WM. Although no research has directly confirmed this assumption, Liang et al. (2019) indirectly support it through corpus analysis. In SI, they found interpreters relying on “the most tangible point of reference” (p.10) in the source speech to process language quickly owing to extreme time pressure, resulting in a more form-based interpreting. In CI, on the other hand, interpreters produced a higher frequency of functional words (as opposed to content words) than simultaneous interpreters to remind themselves of the syntactic structure of the source speech, producing a more meaning-based interpretation. In interpreting studies, form-based interpreting is considered faster and less effortful because it only requires shallow language processing, that is, transcoding, while meaning-based interpreting is considered slower and more effortful for it involves deep semantic processing, that is, deverbalization (e.g., Darò & Fabbro, 1994). Combing the results in Liang et al. (2019), both SI and CI rely on the processing function of WM, but the former involves language processing at a superficial level and the latter at a deep level. Moreover, the finding about more functional words in CI than in SI demonstrates interpreters’ need in source speech memory retrieval, which greatly relies on the storage function of WM.
31.2.2.2 Demands on WM executive functions SI requires concurrent listening comprehension and translation articulation, two activities that compete for the same cognitive resources in the phonological loop of WM. According to Baddeley and Hitch (1974), the phonological loop comprises two components: a temporary phonological store in which to keep perceived sounds and a subvocal rehearsal system for keeping the sounds active. It has been found that overt articulation hinders subvocal rehearsal, and information in the phonological store that is not rehearsed decays (Daró & Fabbro, 1994; Christoffels et al., 2006). This articulatory suppression effect explains SI’s particular demand on WM to inhibit phonological interference. Specifically, while interpreters are articulating the translation of an earlier piece of information, they must inhibit the hampering effect this articulation poses on listening to newly coming-in information. Darò and Fabbro’s (1994) model reflects this phonological interference in SI (see Figure 31.3) by linking the auditory processing of the source text with the overt delivery of the target text (the arrow from TL [target language] points toward WM, as shown in Figure 31.3). This assumption has been confirmed by a series of studies, in which interpreters’ recall performance after SI has been compared with that after other tasks such as listening, shadowing, and listening with articulatory suppression (e.g., Isham, 2000; Lambert, 1988; Padilla et al., 2005). It was repeatedly
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
705
706
BINGHAN ZHENG AND HUOLINGXIAO KUANG
found that recall after SI is worse than after other tasks, indicating a detrimental effect on memory caused by the phonological interference in SI. As a matter of fact, these findings provide empirical evidence for introducing phonological interference into future SI models. In comparison, CI puts enormous pressure on the memory for complete and accurate recall of source speeches. So far, there is no model dedicated to WM operation in CI. Dong et al. (2018) interpreted Mizuno’s (2005) process model of WM and interpreting from a CI perspective. They argue that Updating plays an essential role in CI. In the comprehension phase of CI, interpreters constantly update information in their WM as the source speech is delivered continuously; then, in the reformulation phase, they reactivate “the stretch of input which has already passed FOA (focus of attention)” to recall the source speech (Dong et al., 2018, p. 3). In their empirical study, Dong and her colleagues also confirmed that Updating is a predictor of CI performance (for a detailed discussion, see Section 31.3.2). Another major difference between SI and CI lies in note-taking. When interpreting long sections, consecutive interpreters usually take notes to alleviate the pressure on their memory. To complete note-taking, interpreters “translate” the verbal input from the phonological loop into written notes with the help of a visuospatial sketchpad, while in note-reading, they decode the visual notes and re-express them in the target language. In either process, the phonological loop and the visuospatial sketchpad work closely together and compete for the limited WM capacity. Therefore, interpreters must develop coordinating skills to balance these two components of WM. In WM executive control, note-taking is also a complex cognitive activity (Piolat et al., 2005). Firstly, interpreters shift between source speech comprehension and note production. During note production, they make a further shift between language-based notes and symbols. Secondly, interpreters constantly update information in WM to follow the source speech. Thirdly, they consistently inhibit the activation of inappropriate note forms, and subvocal rehearsal during writing. In summary, note-taking in CI places a heavy demand on the audio and visual components of WM and the three executive functions of WM. Note-taking research in its own right has found that students with larger WM spans experience less cognitive load in notetaking and cover more information in their notes in the classroom (Piolat, 2007). However, whether consecutive interpreters’ WM capacity affects the quality of their note-taking and the amount of effort required remains underexplored.
31.3
Working Memory in Empirical Interpreting Studies
Empirical evidence is essential in (in)validating the theoretical discussions on WM in interpreting. This section is devoted to reviewing empirical interpreting studies that have compared SI and CI from a WM perspective.
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
31.3.1 The Similarities between SI and CI This part of the review focuses on what SI and CI share in WM growth. It specifically looks into how interpreter training experience and interpreter work experience interact with the change of WM size and the use of executive control. 31.3.1.1 Impact of Interpreting Training on WM Span Since the 1990s, interpreting researchers have been attempting to determine whether interpreters have an advantage over noninterpreters in WM span (Darò & Fabbro, 1994, Padilla et al., 1995). Tasks used to test WM span can be divided into two kinds, based on the two functions of WM: information storage and information processing (Baddeley & Hitch, 1974). The first kind only tests WM storage through simple recall tasks such as the digit span task. The other kind evaluates the integral use of the two functions using complex span tasks, such as recall after semantic judgment (Timarová et al., 2014). In both situations, researchers have hypothesized an interpreter advantage for two reasons: first, memory has long been considered to be a basic skill required for completing interpreting tasks (Gile, 1997/2002; Seleskovitch, 1968/1978); second, interpreters are consistently confronted with concurrent storage and processing demands during interpreting, from which they can develop WM skills that will enable them to tackle the issue (Timarová et al., 2014). Overall, in previous research, in simple and complex span tests where digits, words, and sentences are presented visually, interpreters have consistently been found to have an advantage over noninterpreters in terms of WM span, and professional interpreters have consistently been found to have an advantage over novices. However, it was also found that this advantage sometimes disappeared when the stimuli were auditory. This runs counter to our expectation that interpreters would have an advantage in the auditory modality, as they are consistently exposed to auditory processing. There are two possible explanations for this modality effect. First, “reading” words or sentences aloud in some reading span experiments actually gives subjects an opportunity to strengthen their visual memory using self-generated phonological codes (Penney, 1989). According to Penney’s (1989) model of short-term memory, auditory items are automatically encoded in acoustic and phonological codes, while visual items must be intentionally translated into verbal forms to generate phonological codes. If no rehearsal is involved, phonological codes can fade away in seconds. If visual-to-auditory translation is successful, people can trace their memory of the visual stimuli through codes in both visual and audio modalities. Hence, visual presentation with overt vocalization can be more traceable than auditory presentation alone, and therefore helps subjects to complete recall tasks. Secondly, more often than not, interpreters consult prepared speeches, slides, notes, and other visual information during interpreting, thus developing a superior ability in visual-to-auditory translation.
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
707
708
BINGHAN ZHENG AND HUOLINGXIAO KUANG
In comparison, noninterpreters practice this visual-to-auditory translation much less frequently. Therefore, they need more time to complete this translation procedure and create corresponding phonological codes. However, the longer they take, the more the memory decays. As a result, in the studies referred to here, noninterpreters did not outperform interpreters in visually presented WM span tasks. Several researchers examined this modality effect, and all of them rejected the idea that such an effect exists. Daneman and Carpenter (1980) found a high correlation between college students’ reading span and listening span. Köpke and Signorelli (2011) and Chmiel (2018) also reported similar reading and listening spans in interpreters. Thus, in previous studies, this modality effect might just have been a coincidence caused by nonstandard experimental settings. Wen and Dong (2019) synthesized the findings of 10 primary studies on interpreters’ WM and short-term memory. They found that whether interpreters had an advantage over noninterpreters in WM span depended on task type. An interpreter advantage was observed in verbal and numerical/letter span tasks but not in spatial tasks. It is worth mentioning that in almost all studies on interpreters’ WM, researchers selected professional and novice interpreters according to the total time they had spent on interpreting work and/or training. However, few researchers have described the proportions of SI and CI in the interpreters’ work/training experience. In Hiltunen et al. (2014), simultaneous and consecutive interpreters were compared to linguistic nonexperts and foreign language teachers in a free recall task. They found that only the simultaneous interpreter group outperformed the linguistic nonexpert group. This suggests that SI and CI affect the growth of WM size in different ways. Therefore, when recruiting interpreter subjects and describing their background, the proportions of SI and CI work/training experience should be considered as an important variable to be controlled for. On the other hand, an increase in WM span has also been found to be related to higher interpreting accuracy and fluency. Tzou et al. (2012) reported that student interpreters had significantly larger reading spans than untrained bilinguals. And their interpreting accuracy of selected sentences and the overall speech was positively correlated with their reading spans. Lin et al. (2018) found a significant correlation between students’ reading span and the number of interruptions and hesitations in their interpretation, concluding that WM is a powerful predicator of SI fluency. This is in line with Injoque-Ricle et al. (2015) and Macnamara and Conway (2016), who also reported a positive correlation between WM span and SI quality. With regard to CI, Dong et al. (2018) tested students’ L2 listening span before and after CI training and found that the pretest span predicted CI performance. Yenkimaleki and van Heuven (2017) compared the CI performance of a control group (with no memory skill training) and an experimental group (with memory skill training). A positive effect of
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
memory training on reducing omissions in interpreting was observed, which indicates the benefits of an increased WM span on CI fluency. However, all of the above-mentioned studies were conducted with interpreter trainees. Timarová et al. (2015) conducted a study with professional interpreters and reported only a marginally significant correlation between interpreters’ letter span and interpreting accuracy in figures. So far, little evidence is available of a relationship between WM capacity and interpreting quality among professional interpreters.
3.1.2 Novice Advantage in Simple and Complex WM Span Tasks Novice interpreters have repeatedly been found to produce similar or even better performance than professional interpreters in simple WM span tasks, and sometimes in complex WM span tasks. Based on a meta-analysis of 10 studies on this topic, Wen and Dong (2019) concluded that expert interpreters outperform beginner interpreters but not intermediate interpreters in WM span tasks. It seems that the increase in WM span stops at an unclear point during the development of interpreting expertise. Several longitudinal studies conducted with novice interpreters have demonstrated the effect of interpreting training on WM span (Babcock & Vallesi, 2017; Dong & Liu, 2016; Nour, Struys & Stenger, 2020). However, there are contradictory findings concerning the effect of interpreting experience on WM span, with around half of the studies acknowledging professional interpreters’ advantage (Nour, Struys & Stenger, 2020; Padilla et al., 1995; Tzou et al., 2011), and the other half denying such an advantage (Köpke & Nespoulous, 2006; Liu et al., 2004; Padilla, 1995). There are two potential reasons why a novice might have an advantage in simple WM tasks. First, professional interpreters are usually older than novice interpreters, leading to an aging effect on WM span. Signorelli et al. (2011) proved that younger interpreters performed better than professionals in nonword repetition and cued recall, suggesting a trade-off between age and WM span. Second, compared with novice interpreters, professional interpreters usually process information in larger units. In this way, they can alleviate the pressure on their memory and focus more on language processing (Wen & Dong, 2019). A novice advantage has also been observed in complex span tasks. Liu et al. (2004) used a CI-like WM task where subjects first processed semantic meaning and then recalled the last words of sentences. The results show that professionals and novices performed similarly. Köpke and Nespoulous (2006) tested subjects with a SI-like task of free recall with articulatory suppression and found that novices recalled more words than professionals. These findings challenged the conventional view that expertise in interpreting relies on a greater WM capacity. Köpke and Nespoulous (2006) speculated that professional interpreters have developed an optimal language processing route that releases them from the constraints of WM size.
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
709
710
BINGHAN ZHENG AND HUOLINGXIAO KUANG
Novice interpreters without such optimal processing routes can easily experience an overload of WM during interpreting. It is not yet clear how this optimization of language processing works. However, to some extent it implies that the development of interpreting expertise is centred on domain-specific abilities like language processing rather than on general cognitive abilities.
3.1.3 Impact of Interpreting Experience on WM Executive Control An interpreter advantage has not only been observed in WM span tasks that test interpreters’ information storage and/or processing ability but also in tasks targeting on their operation of WM executive control. Overall, interpreters have been found to have an advantage in using the executive functions of WM, although the reasons for this advantage have not yet been clarified. With regard to Inhibition, Dong and Zhong (2017) reported an effect of interpreting experience, finding that more experienced student interpreters outperformed both less experienced students and more balanced noninterpreter bilinguals in a Flanker task. By contrast, Woumans et al. (2015) found that interpreters outperformed an unbalanced group of bilinguals but not balanced bilinguals in a Simon task and an Attention Network test,1 which implies that the (un)balance of bilingualism plays a decisive role in Inhibition execution. According to Nour, Struys, Woumans et al.’s (2020) detailed review of eight studies on interpreters’ Response-Distractor Inhibition, interpreters only show an advantage when they are compared with unbalanced bilinguals and not when compared with balanced bilinguals. With regard to Updating, Dong and Liu (2016) tested students’ execution of this function through a n-back task2 before and after CI training, and observed a significant improvement in students’ task performance. They explain that in CI, interpreters repeatedly replace old information with new information during the input stage and that they refresh their memory of the source speech during the output stage. In this way, interpreters strengthen their Updating ability. This could explain why Morales et al. (2015) and Timarová et al. (2014) did not find an advantage on the part of simultaneous interpreters in this regard. In SI, interpreters usually release their memory immediately after finishing the interpretation of the current information. Nevertheless, simultaneous interpreters can acquire this Updating ability quickly. Morales et al. (2015) compared simultaneous interpreters’ performance across the two blocks of a n-back task and found improved accuracy in the second block. In contrast, noninterpreter bilinguals showed no improvement throughout the task. Taken together, it seems that CI and SI experience has benefited the development of Updating in different ways. In comparison, findings concerning the positive impact of interpreting training and/or work experience on Shifting have been consistent (Dong
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
& Liu, 2016; Macnamara & Conway, 2014; Yudes et al., 2011), the only exception being Timarová et al. (2014). This is owing to the high sensitivity of Shifting to environmental factors. It has been found that over 20% of its variability is attributable to nongenetic factors (Nour, Struys, Woumans et al., 2020). In other words, the constant shift between languages during interpreting enables interpreters to improve their Shifting ability. In comparison, noninterpreter bilinguals face fewer Shifting needs in language use, resulting in slower reactions to shifting-related tasks. Moreover, interpreting is conducted under extremely high time pressure, meaning that interpreters must make Shifting decisions quickly. Compared with translators, who also switch between languages but with much lower time pressure, interpreters have been found to perform better in Shifting-related tasks (Henrard & van Daele, 2017). All in all, an interpreter advantage has been found in the execution of all three executive functions of WM, although how bilingual competence and interpreting practice interactively contribute to this advantage remains underexplored.
31.3.2 The Differences between SI and CI The following review introduces how SI and CI differ from each other in two aspects: the use of executive functions of WM and the fluctuation of cognitive load during the process of interpreting. Empirical evidence about the attributes of SI and CI in these regards has been found in studies on respective interpreting modes. 31.3.2.1 Working Memory: Inhibition in SI and Updating in CI The three executive functions of WM affect the subprocesses of interpreting differently (Timarová et al., 2014). In SI, there is an overlap between listening to the source speech and articulating the target speech for about 70% of the time (Chernov, 1994). Interpreters must therefore minimize the phonological interference caused by concurrent listening and speaking through Inhibition (e.g., Chincotta & Underwood, 1998; Padilla et al., 1995; Yudes et al., 2012). Moreover, the concurrent execution of subtasks like listening comprehension, interpretation delivery, and self-monitoring requires interpreters to inhibit the interferences caused by multitasking. In CI, interpreters memorize the inputted source speech and retrieve it when producing the output. Their pressure mainly derives from having to refresh their memory of the source speech, ranging from single words to entire speeches (Dam, 2010), through Updating (Dong et al., 2018; Dong & Liu, 2016). In short, in SI the emphasis is on Inhibition, while in CI the emphasis is on Updating, but both require Shifting for language transfer. To test simultaneous interpreters’ Inhibition advantage in resisting phonological interference, researchers usually employ a free recall task with articulatory suppression. This task requires interpreters to articulate
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
711
712
BINGHAN ZHENG AND HUOLINGXIAO KUANG
irrelevant sounds while listening to words or sentences, and thus creates an SI-like scenario. Using this task, Köpke and Nespoulous (2006) identified an interpreter advantage over noninterpreter bilinguals and a novice advantage over professional interpreters. Similarly, Padilla’s team (1995, 2005) found that novice interpreters outperformed noninterpreter bilinguals whether they had high reading spans or not. However, it has been argued that the novice advantage in Inhibition might be misleading since the selected task does not resemble SI as assumed. The suppressed recall task did prevent subvocal rehearsal as SI does because it requires interpreters to listen to and voice irrelevant information, whereas SI requires interpreters to articulate a translation of whatever they have heard. Moreover, the words or sentences used in the task are usually unrelated to each other, while the input in SI is normally a logical and coherent text, which provides interpreters with contextual hints that facilitate better source speech comprehension and prediction. Therefore, when the adopted WM task does not resemble SI, more-experienced interpreters might not be able to show an advantage over less-experienced interpreters in Inhibition. An alternative explanation for no Inhibition advantage in more-experienced interpreters may be that Inhibition is essential in the early stage of interpreting training but not as essential in building interpreting expertise. Research on Inhibition and multi-tasking has yielded more contradictory results than research on Inhibition and phonological interference in SI. A few researchers have reported an interpreter advantage in this regard (Köpke & Nespoulous, 2006; Woumans et al., 2015), while most researchers have denied it (Babcock & Vallesi, 2017; Dong & Liu, 2016; Dong & Xie, 2014; Yudes et al., 2011). Many of the tasks adopted in these studies focus on nonverbal information processing (arrows, shapes, and colours). However, not all SI skills can be transferred to nonverbal or general cognitive tasks. Morales et al. (2015) found that interpreters did not present any advantage in conflict resolution, but they showed an advantage in alertness and orienting tasks, which involve multitasking as SI does. Similarly, Dong and Liu (2016) did not detect an interpreting training effect on interpreters’ ability to resist conflicts, since interpreters are not often required to resolve conflicts of this kind during interpreting. All the above indicate that an interpreter advantage in Inhibition depends on the transferability of interpreting expertise to the adopted tasks. In contrast, research on Updating and CI has yielded more consistent results. Dong’s team first reported a positive impact of CI training on interpreters’ Updating ability (Dong & Liu, 2016). Later, they found that students’ Updating ability successfully predicted their CI performance both before and after CI training (Dong et al., 2018). Finally, they did a metaanalysis of four studies on Updating and interpreting and found that an interpreter advantage in updating ability had been observed in both CI and SI (Wen & Dong, 2019). It is noteworthy that an improvement in updating ability was observed in subjects who had received CI training for the shortest time (a 32-hour in-class
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
training program) (Dong & Liu, 2016). However, no such improvement was witnessed in memory span (Dong et al., 2018), shifting, or inhibitory control (Dong & Liu, 2016). Based on these findings, Wen and Dong (2019) concluded that “updating ability is probably the first taxed and trained memory skill in interpreting training relative to memory spans (short-term memory and WM spans)” (p. 12). This might be caused by the speaker-paced feature of SI and CI. Without Updating, interpreters would not be able to make enough memory space available for new information. In comparison, Inhibition and Shifting are more necessary after new information has entered the WM and is competing with old information for WM resources. To verify this assumption, researchers should conduct longitudinal studies with noninterpreters who start to learn interpreting and novice interpreters who gradually gain more interpreting experience. In this way, researchers can find out how the three executive functions of WM develop along with the course of interpreting training and practice.
31.3.2.2 Local Cognitive Load in SI and CI Cognitive load is “the load that performing a particular task imposes on the learner’s cognitive system” (Paas et al., 1995, p. 64). When it exceeds an individual’s WM capacity, task performance drops or even crashes. For interpreters, there are “global” and “local” levels of cognitive load. At a “global” level, interpreting as a single task entails more cognitive load than shadowing and listening do (Christoffels & de Groot, 2004; Köpke & Nespoulous, 2006; Padilla et al, 2005). At a “local” level, cognitive load fluctuates throughout the whole process of interpreting. In SI, interpreters’ cognitive load fluctuates below sentence level because they follow speakers closely to produce target speeches. But in CI, interpreters’ cognitive load varies beyond sentence level because they can flexibly allocate cognitive resources to the two stages of CI, as long as the sum of the allocated resources does not exceed their WM limit. Most of the research on this subject has attempted to identify the factors that affect the local cognitive load of interpreting, while only few studies have directly measured the amount of cognitive load. SI researchers have mainly measured interpreters’ local cognitive load by analyzing their interpreting products. It is assumed that interpreting quality drops as cognitive load increases. On interpreting accuracy, Gile (2017) asked 10 professional interpreters to interpret the same materials twice. He found new errors and omissions at different places in the interpreters’ second rendering, which indicates a change in the cognitive load fluctuation pattern between the two interpretations. On interpreting fluency, Plevoets and Defrancq (2016) took the occurrence rate of uh(m) as an indicator of local cognitive load and found that it was predicted by the delivery rate of the source text. In other words, local cognitive load increases as the speakers’ delivery rate increases. Shao and Chai (2020) found that interpreters’ local cognitive load reached a peak when processing four information
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
713
714
BINGHAN ZHENG AND HUOLINGXIAO KUANG
chunks at one time. When they exceeded four chunks, interpreters’ performance dropped dramatically. They also measured a specific kind of local cognitive load during interpreting: the current load, which is caused by the accumulated information of the current sentence. They measured it through ear-voice span (EVS), the time span between input and output, and it was found to be significantly correlated with SI performance. In other studies, the length of EVS has been found to increase with an increase in syllable and sentence length (Lee, 2002) and with an increase in syntactic and semantic complexity (Timarová et al., 2014), and to decrease with an increase in preparation time (Díaz-Galaz et al., 2015) and interpreting experience (Timarová et al., 2014). Although those researchers did not interpret EVS from the perspective of local cognitive load, they provided empirical evidence that the online cognitive load of SI changes within a sentence. Seeber (2011) measured interpreters’ current load using pupillometry, and found that interpreters’ pupils dilate more toward the end of sentences while handling asymmetrical structures in SI. This might be attributed to the four interpreting strategies interpreters adopted in this SI task: waiting, stalling, chunking, and anticipating. Among the four strategies, the first three allow interpreters to wait in order to obtain more information. Therefore, interpreters’ local cognitive load accumulates toward the end of sentences. In contrast to SI, very few studies have measured local cognitive load in CI. Wu and Wang (2009) analyzed the case of an interpretation delivered at a press conference held by the Chinese government and found that interpreters can save cognitive efforts by simplifying sentence structures in source speeches. They identified three simplification methods: deleting the overlapping part of different sentences, transforming sentences into the same structures, and merging sentences with similar structures. Chen (2017) measured interpreters’ ear-pen span during note-taking and eye fixations during note-reading. The data suggest that notes that entail longer ear-pen span and cost more cognitive efforts during note-taking result in shorter fixations and less cognitive effort during note-reading. In other words, there is a trade-off in cognitive load across the two phases of CI. All in all, interpreters experience fluctuating local cognitive load during interpreting. The amount of load varies within sentences in SI and beyond sentences in CI. More process-oriented research is needed to unveil the complex fluctuation pattern of local cognitive load during the interpreting process.
31.4
Discussion
Based on a systematic review of interpreting research on WM capacity and executive control, we observed that in most cases interpreters showed an advantage over noninterpreters in these regards. However, more-experienced
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
interpreters and less-experienced interpreters were alternately found advantageous than each other concerning WM capacity and executive control. From theoretical and methodological perspectives, these findings could lay the ground for future empirical research in this field. First, the inconsistent findings about the benefits of interpreting experience accumulation on WM might be derived from the mismatch between participants’ interpreting experience and the WM task selected. On the participant side, CI demands information storage during input and Updating for recall during output, while SI requires fast information processing and Inhibition for resisting phonological interference. The two interpreting modes contribute to the growth of WM in different ways. On the task side, researchers can test interpreters’ information storage and processing ability with simple and complex memory span tasks in visual and audio modalities, or their executive control with tasks specifically targeted at the three WM executive functions. Thus, if consecutive interpreters were tested using information processing tasks or simultaneous interpreters were tested using information storage tasks, then more-experienced interpreters may not be able to show an advantage over less-experienced interpreters in the tested aspect of WM. Future research should strictly control the variables involved in experimental designs to ensure that interpreters’ CI and SI experience composition or the selected WM task are not confounding factors affecting the research outcome. Second, a great deal of research has been conducted on WM in SI, while little attention has been paid to CI in this regard. Among those SI studies that included WM capacity control in their participant selection, almost all use recall tasks containing an input stage of stimuli presentation and an output stage of stimuli recall, which resemble CI tasks more than SI tasks. In other words, researchers tested interpreters with the same ability in CI-like tasks and correlated this ability with interpreters’ SI performance. Therefore, WM tasks involving separate input and output are suggested for CI research, and those requiring simultaneous input and output are advised for SI research. Moreover, it would be interesting to compare WM demands and interpreters’ coping strategies across the two stages of CI. At present, students usually receive CI training before SI training, as CI ability is generally considered to be a prerequisite for completing SI tasks. To decide whether this order is reasonable in curriculum design, more empirical evidence from comparative research on WM in CI and SI would be necessary. Third, concerning the contradictory findings regarding WM and interpreting quality, more process-oriented research should be conducted to see how WM demand fluctuates during interpreting from beginning to end. The process-oriented methodology has been used in several pioneering SI and CI studies that recorded interpreters’ eye movements, galvanic skin responses, and other physiological responses during interpreting (e.g., Chen, 2017; Seeber, 2011). These studies have proved that interpreting is a dynamic process, during which interpreters experience cognitive adaptation
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
715
716
BINGHAN ZHENG AND HUOLINGXIAO KUANG
and emotional fluctuation. However, there is a challenge in measuring the three types of cognitive load (intrinsic load, extraneous load, and germane load) separately, and in differentiating positive emotions (such as excitement) from negative emotions (such as disappointment). If cognitive load can be measured separately, the amount of germane load can reveal how many WM resources are devoted to “learning” during interpreting. Researchers could measure student interpreters’ germane load and score their interpreting performance before and after interpreting training. By dividing students into high-score and low-score groups, researchers could compare how different the two groups are in terms of the amount of germane load they devoted to learning and the quality of the interpretation, further illustrating the formation of interpreting expertise. Similarly, if positive and negative emotions can be presented separately, researchers will be able to understand better how interpreters deal with the overwhelming stress during interpreting. For instance, if more excitement correlates with better interpretation, then motivation enhancement would be an essential aspect for interpreting trainers to consider during curriculum design. Looking ahead, research on WM and interpreting would benefit from investigating several underexplored issues. First, interpreting in general is beneficial to the growth of WM. However, it is not yet clear how SI and CI contribute to this growth and in turn benefit from the growth through enhanced interpreting quality. Second, little attention has been paid to CI, in which the two stages of input and output place different demands on WM, leading to a shortage of comparative discussions on the role of WM in CI and SI. Third, cognitive load fluctuates during the process of interpreting, giving rise to the question of how speaker-related variables such as delivery speed and interpreter variables such as interpreting strategies affect the pattern of cognitive load fluctuation during interpreting.
31.5
Conclusion
This chapter has reviewed the operation of WM in SI and CI from theoretical and empirical perspectives. Theoretically, both interpreting modes place high demands on WM for language control and processing control. SI centers more on fast information processing because of its requirement for an immediate input-to-output transformation, while CI emphasizes both information processing and storage because of the need to recall source speeches. Empirically, WM and interpreting have been found to be interdependent on each other in a complex way. Questions remain regarding the development path of WM during the course of interpreting training and its interaction with interpreting performance. Future research could combine process-oriented and product-oriented approaches to unveil the cognitive mechanism governing the interpreting process and modulating the interpreting quality.
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
References Adelson, B. (1984). When novices surpass experts: The difficulty of the task may increase with expertise. Journal of Experimental Psychology: Learning, Memory and Cognition, 10, 484–495. Babcock, L., & Vallesi, A. (2017). Are simultaneous interpreters expert bilinguals, unique bilinguals, or both? Bilingualism: Language and Cognition, 20(2), 403–417. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G.H, Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (pp. 47-89). Academic Press. Chen, S. (2017). Exploring the process of note-taking and consecutive interpreting: A pen-eye-voice approach towards cognitive load (Doctoral dissertation, Macquarie University, Sydney, Australia). Chernov, G. V. (1994). Message redundancy and message anticipation in simultaneous interpretation. In S. Lambert & B. Moser-Mercer (Eds.), Bridging the gap: Empirical research in simultaneous interpretation (pp. 139–153). John Benjamins. Chincotta, D., & Underwood, G. (1998). Simultaneous interpreters and the effect of concurrent articulation on immediate memory: A bilingual digit span study. Interpreting, 3(1), 1–20. Chmiel, A. (2018). In search of the working memory advantage in conference interpreting: Training, experience and task effects. International Journal of Bilingualism, 22(3), 371–384. Christoffels, I. K., de Groot, A. M., & Kroll, J. F. (2006). Memory and language skills in simultaneous interpreters: The role of expertise and language proficiency. Journal of Memory and Language, 54(3), 324–345. Cowan, N. (1988). Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human informationprocessing system. Psychological Bulletin, 104(2), 163–191. Cowan, N. (1995). Attention and memory: An integrated framework. Oxford University Press. Dam, H. V. (2010). Consecutive interpreting. In Y. Gambier & L. van Doorslaer (Eds.), Handbook of translation studies (Vol. 1, pp. 75–79). John Benjamins. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19(4), 450–466. Darò, V., & Fabbro, F. (1994). Verbal memory during simultaneous interpretation: Effects of phonological interference. Applied Linguistics, 15(4), 365–381. Díaz-Galaz, S., Padilla, P., & Bajo, M. T. (2015). The role of advance preparation in simultaneous interpreting: A comparison of professional interpreters and interpreting students. Interpreting, 17(1), 1–25.
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
717
718
BINGHAN ZHENG AND HUOLINGXIAO KUANG
Dong, Y., & Cai, R. (2015). Working memory in interpreting: A commentary on theoretical models. In Z. Wen, M. Mota, & A. McNeill (Eds.), Working memory in second language acquisition and processing (pp. 63–79). Multilingual Matters. Dong, Y. & Li, P. (2020). Attentional control in interpreting: A model of language control and processing control. Bilingualism: Language and Cognition, 23, 716–728. Dong, Y., & Liu, Y. (2016). Classes in translating and interpreting produce differential gains in switching and updating. Frontiers in Psychology, 7, 1297. Dong, Y., Liu, Y., & Cai, R. (2018). How does consecutive interpreting training influence working memory: A longitudinal study of potential links between the two. Frontiers in Psychology, 9, 875. Dong, Y., & Xie, Z. (2014). Contributions of second language proficiency and interpreting experience to cognitive control differences among young adult bilinguals. Journal of Cognitive Psychology, 26(5), 506–519. Dong, Y., & Zhong, F. (2017). Interpreting experience enhances early attentional processing, conflict monitoring and interference suppression along the time course of processing. Neuropsychologia, 95, 193–203. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102(2), 211–245. Ericsson, K. A. & Charness, N. (1994). Expert performance: Its structure and acquisition. American Psychologist, 49, 725–747. Gile, D. (1997/2002). Conference interpreting as a cognitive management problem. In F. Pöchhacker & M. Shlesinger (Eds.), The interpreting studies reader (pp. 162–176). London: Routledge. Gile, D. (2001). Consecutive vs. simultaneous: Which is more accurate? Interpretation Studies, 1(1), 8–20. Gile, D. (2017). Testing the Effort Models’ tightrope hypothesis in simultaneous interpreting: A contribution. HERMES, 12(23), 153–172. Henrard, S., & van Daele, A. (2017). Different bilingual experiences might modulate executive tasks advantages: Comparative analysis between monolinguals, translators, and interpreters. Frontiers in Psychology, 8, 1870. Hiltunen, S., Pääkkönen, R., Vik, G. V., & Krause, C. M. (2014). On interpreters’ working memory and executive control. International Journal of Bilingualism, 20(3), 297–314. Holding, D. H., & Reynolds, R. I. (1982). Recall or evaluation of chess positions as determinants of chess skill. Memory & Cognition, 10(3), 237–242. Injoque-Ricle, I., Barreyro, J. P., Formoso, J., & Jaichenco, V. I. (2015). Expertise, working memory and articulatory suppression effect: Their relation with simultaneous interpreting performance. Advances in Cognitive Psychology, 11(2), 56–63. doi:10.5709/acp-0171-1
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
Isham, W.P. (2000). Phonological interference in interpreters of spokenlanguages: An issue of storage or process? In B. E. Dimitrova & K. Hyltenstam (Eds.), Language processing and simultaneous interpreting (pp. 133–150). John Benjamins. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122–149. Köpke, B., & Nespoulous, J. L. (2006). Working memory performance in expert and novice interpreters. Interpreting, 8(1), 1–23. Köpke, B., & Signorelli, T. M. (2011). Methodological aspects of working memory assessment in simultaneous interpreters. International Journal of Bilingualism, 16(2), 183–197. Lambert, S. (1988). Information processing among conference interpreters: A test of the depth-of-processing hypothesis. Meta, 33(3), 377–387 Lee, T. H. (2002). Ear voice span in English into Korean simultaneous interpretation. Meta, 47(4), 596–606. Liang, J., Fang, Y., Lv, Q., & Liu, H. (2017). Dependency distance differences across interpreting types: Implications for cognitive demand. Frontiers in Psychology, 8, 1–10. doi: 10.3389/fpsyg.2017.02132 Lin, Y., Lv, Q., & Liang, J. (2018). Predicting fluency with language proficiency, working memory, and directionality in simultaneous interpreting. Frontiers in Psychology, 9, 1543. doi:10.3389/fpsyg.2018.01543 Liang, J., Lv, Q., & Liu, Y. (2019). Quantifying interpreting types: Language sequence mirrors cognitive load minimization in interpreting tasks. Frontiers in Psychology, 10, 1–12. doi: 10.3389/fpsyg.2019.00285 Liu, M., Schallert, D. L., & Carroll, P. J. (2004). Working memory and expertise in simultaneous interpreting. Interpreting, 6(1), 19–42. Macnamara, B. N., & Conway, A. R. (2016). Working memory capacity as a predictor of simultaneous language interpreting performance. Journal of Applied Research in Memory and Cognition, 5(4), 434–444. Macnamara, B. N., & Conway, A. R. A. (2014). Novel evidence in support of the bilingual advantage: Influences of task demands and experience on cognitive control and working memory. Psychonomic Bulletin & Review, 21 (2), 520–525. Mizuno, A. (2005). Process model for simultaneous interpreting and working memory. Meta, 50(2), 739–752. Morales, J., Padilla, F., Gómez-Ariza, C. J., & Bajo, M. T. (2015). Simultaneous interpretation selectively influences working memory and attentional networks. Acta Psychologica, 155, 82–91. Nour, S., Struys, E., & Stengers, H. (2020). Adaptive control in interpreters: Assessing the impact of training and experience on working memory. Bilingualism: Language and Cognition, 23, 772–779. Nour, S., Struys, E., Woumans, E., Hollebeke, I., & Stengers, H. (2020). An interpreter advantage in executive functions? A systematic review. Interpreting, 22(2), 163–186.
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
719
720
BINGHAN ZHENG AND HUOLINGXIAO KUANG
Paas, F. G., & Van Merriënboer, J. J. (1994). Instructional control of cognitive load in the training of complex cognitive tasks. Educational Psychology Review, 6(4), 351–371. Padilla, P. (1995). Procesos de memoria y atención en la interpretación de lenguas (Doctoral dissertation, Universidad de Granada. Granada, Spain). Padilla, P., Bajo, M. T., Canas, J. J., & Padilla, F. (1995). Cognitive processes of memory in simultaneous interpretation. In J. Tommola (Ed.), Topics in interpreting research (pp. 61–72). Centre for Translation and Interpreting, University of Turku. Padilla F., Bajo M. T., & Macizo P. (2005). Articulatory suppression in language interpretation: Working memory capacity, dual tasking and word knowledge. Bilingualism: Language and Cognition, 8. 207–219. Penney, C. G. (1989). Modality effects and the structure of short-term verbal memory. Memory & Cognition, 17(4), 398–422. Piolat, A. (2007). Effects of notetaking technique and working-memory span on cognitive effort and recall performance. In M. Torrance, L. Van Waes, & D. Galbraith (Eds.), Writing and cognition: Research and applications (pp. 109–124). Elsevier. Piolat, A., Olive, T., & Kellogg, R. T. (2005). Cognitive effort during note taking. Applied Cognitive Psychology, 19(3), 291–312. Plevoets, K., & Defrancq, B. (2016). The effect of informational load on disfluencies in interpreting: A corpus-based regression analysis. Translation and Interpreting Studies, 11(2), 202–224. Russell, D. (2002). Interpreting in legal contexts: Consecutive and simultaneous interpreting. Linstok Press. Seeber, K. G. (2011). Cognitive load in simultaneous interpreting: Existing theories – new models. Interpreting, 13(2), 176–204. Seleskovitch, D. (1968/1978). Interpreting for international conferences-Problems of language and communication. Pen and Booth. Shao, Z., & Chai, M. (2020). The effect of cognitive load on simultaneous interpreting performance: An empirical study at the local level. Perspectives, 1–17. doi: 10.1080/0907676X.2020.1770816 Signorelli, T. M., Haarmann, H. J., & Obler, L. K. (2011). Working memory in simultaneous interpreters: Effects of task and age. International Journal of Bilingualism, 16(2), 198–212. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. Timarová, Š., Čeňková, I., & Meylaerts, R. (2015). Simultaneous interpreting and working memory capacity. In Ferreira, A. & J. W. Schwieter (Eds.), Psycholinguistic and cognitive inquiries into translation and interpreting (pp. 101–126). Benjamins. Timarová, Š., Čeňková, I., Meylaerts, R., Hertog, E., Szmalec, A., & Duyck, W. (2014). Simultaneous interpreting and working memory executive control. Interpreting, 16(2), 139–168.
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
Interpreting Studies
Tzou, Y. Z., Eslami, Z. R., Chen, H. C., & Vaid, J. (2012). Effect of language proficiency and degree of formal training in simultaneous interpreting on working memory and interpreting performance: Evidence from Mandarin-English speakers. International Journal of Bilingualism, 16(2), 213–227. Wen, H., & Dong, Y. (2019). How does interpreting experience enhance working memory and short-term memory: A meta-analysis. Journal of Cognitive Psychology, 31(8), 769–784. Woumans, E., Ceuleers, E., van der Linden, L., Szmalec, A., & Duyck, W. (2015). Verbal and nonverbal cognitive control in bilinguals and interpreters. Journal of Experimental Psychology: Learning, Memory, and Cognition, 41(5), 1579–1586. Wu, G., & Wang, K. (2009). Consecutive interpretation: A discourse approach. Towards a revision of Gile’s effort model. Meta, 54(3), 401–416. Yenkimaleki, M., & van Heuven, V. J. (2017). The effect of memory training on consecutive interpreting performance by interpreter trainees: An experimental study. International Journal of Interpretation and Translation, 15(1), 157–172. Yudes, C., Macizo, P., & Bajo, T. (2011). The influence of expertise in simultaneous interpreting on non-verbal executive processes. Frontiers in Psychology, 2, 309. Yudes, C., Macizo, P., & Bajo, T. (2012). Coordinating comprehension and production in simultaneous interpreters: Evidence from the articulatory suppression effect. Bilingualism, 15(2), 329–339.
Notes 1 These tasks are complex versions of a Flanker task. 2 Subjects decide whether the current stimulus matches the stimulus presented n items earlier.
https://doi.org/10.1017/9781108955638.037 Published online by Cambridge University Press
721
31 Working Memory and Interpreting Studies Binghan Zheng and Huolingxiao Kuang
The work was supported by the National Social Science Fund of China (No. 20BYY014). Huolingxiao Kuang’s Ph.D. is funded by China Scholarship Council (CSC).
31.1
Introduction
Interpreting can be conducted either simultaneously or consecutively. In the simultaneous mode, interpreters listen, comprehend and translate the source speech in real time, which requires a coordinated use of limited working memory (WM) resources. In the consecutive mode, interpreters store the source speech in their WM, and then recall the stored speech in the target language by refreshing their WM (Dong et al., 2018). Both interpreting modes place high demands on the storage and processing functions of WM (Baddeley & Hitch, 1974), and require an effective executive control in WM operation (Nour, Struys, Woumans et al., 2020). Overall, previous interpreting research on WM can be divided into two strands. The first strand tests interpreters’ and noninterpreter bilinguals’ information storage and/or processing ability by measuring their WM capacity. The measurement is made using WM span tasks where subjects are required to recall a specific number of presented stimuli. The second strand investigates interpreters’ operation of WM executive control, which includes three executive functions: inhibiting phonological and multitasking interferences (Inhibition), replacing old information with new information as a method of continuous input processing (Updating) and switching between languages and subtasks of interpreting from source speech comprehension to target speech delivery (Shifting) (Nour, Struys, Woumans et al., 2020). So far, contradictory findings have been reported regarding the benefits of interpreting training and work experience on WM capacity improvement and executive control enhancement. Most of the studies in both strands deal with simultaneous interpreting (SI). This is because SI is generally considered much more demanding on the WM than consecutive interpreting (CI) and other interpreting modes (Dong & Cai, 2015) because of its concurrent execution of several subtasks from source
Published online by Cambridge University Press
Interpreting Studies
speech processing to target speech delivery. Owing to the limited amount of research on CI, there is a lack of comparative discussions on the role of WM in SI and CI, the two most representative interpreting modes. This chapter attempts to fill this gap in the research by systematically reviewing previous research on WM in SI and CI, focusing on the common features and the differences between the two modes from theoretical and empirical perspectives.
31.2
Working Memory in Theoretical Interpreting Studies
This section centers on the theoretical aspects of WM that interpreting researchers have explored to compare SI and CI. Comparisons are made in regards to both the development of WM capacity and the use of WM resources. The first half of the section focuses on the similarities between the two modes, while the second half discusses the differences.
31.2.1 The Similarities between SI and CI Interpreting in general is a complex cognitive activity that places a heavy demand on interpreters’ WM for language control and processing control. It is therefore essential for both simultaneous and consecutive interpreters to expand their WM capacity and interpreting-specific knowledge schemas in long-term memory to meet such great demands of interpreting. Specifics about the demand on WM and the need for WM development in SI and CI are introduced below. 31.2.1.1 Demand on WM for Language Control and Processing Control Baddeley and Hitch’s (1974) Working Memory Model and Cowan’s (1988, 1995) Embedded-Process Model of Memory have been extensively discussed in interpreting studies. Baddeley and Hitch (1974) looked into the “inside” of WM by proposing two slave systems in WM: the phonological loop and the visuospatial sketchpad, which are responsible for audio and visual information processing, respectively. Since the capacity of WM is limited, the two slave systems are coordinated by the commander in WM: the central executive. Through executive control, interpreters can effectively distribute WM resources to the audio channel (such as listening and speaking) and the visual channel (such as reading and note-taking). In contrast, Cowan (1988, 1995) focuses on the “outside” of WM by associating it with memory at different levels. In his model, WM is simply a “set of activated memory elements” (Cowan, 1995, p.100) that belong to long-term memory (LTM). Only when a memory is brought to the focus of attention can it be processed by WM. For example, when interpreters are processing the information in their focus of attention, they activate the part of LTM they need: linguistic knowledge and interpreting skills, for example, to accomplish comprehension or interpretation. The activated contents change constantly
Published online by Cambridge University Press
699
700
BINGHAN ZHENG AND HUOLINGXIAO KUANG
Figure 31.1 Mizuno’s (2005) model of interpreting based on Cowan’s (1988, 1995) Embedded-Processes Model of Memory Source: Mizuno, A. (2005). Process model for simultaneous interpreting and working memory. Meta, 50(2), 739–752. Reproduced with permission
as the focus of attention moves. Overall, the two models explain how information is processed in interpreting with the two slave systems inside WM and with LTM outside it. Interpreting is by nature a verbal task, and this should be reflected in WM models. With this in mind, Mizuno (2005) added a language comprehension system and a language production system before and after the memory section to Cowan’s (1988, 1995) Embedded-Processes Model (see Figure 31.1). This addition seems to provide a complete description of the working flow of WM in interpreting from language input to language output. However, it neglects one of the most distinctive features of interpreting compared to other language processing tasks: language transfer (Dong & Cai, 2015). In SI, interpreters listen to one language and translate it into another language at almost the same time; in CI, by contrast, interpreters switch to the target language whenever the speaker finishes an utterance, and shift their interpreting direction when the source language changes. This frequent and regular switch between two languages is a distinctive feature of interpreting (Dong & Li, 2020) that should not be excluded from interpreting models. Therefore, Dong and Li (2020) proposed an attentional control model of SI and CI, taking both language control and processing control into consideration (p.10) (Figure 31.2). Under language control, working memory, together with monitoring, target enhancement, target disengagement, and shifting, help interpreters focus their attention on inhibiting the source language and activating the target language. When a particular language input-and-output modality is established and transformed into a schema, interpreters can activate it whenever they need it. This is especially important for CI interpreters, who sometimes interpret for interlocutors from both sides. Whenever the source language changes,
Published online by Cambridge University Press
Interpreting Studies
Figure 31.2 Dong and Li’s (2020) attentional control model of interpreting Source: Dong, Y. & Li, P. (2020). Attentional control in interpreting: A model of language control and processing control. Bilingualism: Language and Cognition, 23, 716–728. Reproduced with permission
they must switch the inhibited language and the activated language accordingly. In this situation, WM is heavily relied on to achieve focused attention. Under processing control, WM and coordination enable interpreters effectively to divide their attention among the various tasks involved in interpreting, such as listening comprehension, note-taking, target speech production, and self-monitoring. This is vital in SI, where the degree of simultaneity of listening comprehension and target speech production decides the quality of the interpreting product. It is worth mentioning that compared with previous models of WM in interpreting, Dong and Li’s (2020) model not only illustrates how WM operates during the process of interpreting, but also shows how to build interpreting expertise. In this model, the establishment of a language-modality connection and the improvement of language processing efficiency respectively contribute to better language control and processing control. Specifically, during interpreting training, interpreters should learn to form language-modality schemas, enhance their language abilities, and acquire interpreting strategies that will give them better attentional control. Hence, this attentional control model can be used not only as a process model of SI and CI, but also as a developmental model for interpreting training.
31.2.1.2 The Role of WM in the Development of Interpreting Expertise Compared with other language processing tasks, interpreting is especially complex because of its cognitive demand on both information retention and the need for fast processing. This demand is decided by the speaker-paced feature of SI and CI. In SI, interpreters reformulate the information they have heard and articulate it in the target language while storing the continuous
Published online by Cambridge University Press
701
702
BINGHAN ZHENG AND HUOLINGXIAO KUANG
new input for later processing (Christoffels et al., 2006). In CI, interpreters must remember the whole source speech accurately and search for translation if there is any spare WM during listening. If an interpreter fails to keep up with the speaker, they will miss information, which will affect their comprehension of the source speech. To avoid information loss, interpreters can either increase their WM span to store more information or release their WM space by automatizing some information-processing procedures. Just and Carpenter (1992) observed that college students with larger reading spans had better language comprehension ability than those with smaller spans. Therefore, they claimed that the number of language elements (such as phrases and grammatical structure) individuals can activate in their WM decides the depth of language comprehension they can achieve. For interpreters, listening comprehension consumes at least 80 percent of their cognitive effort during interpreting (Padilla, 1995). If we put the two findings together, we can propose an assumption that a bigger WM span contributes to better language comprehension, which could save interpreters’ some effort in comprehension and improve their output quality during interpreting. An alternative theory to explain Just and Carpenter’s (1992) finding is the theory of expertise (Ericsson & Charness, 1994), which argues that experts outperform novices because of their extensive knowledge in a given domain. For example, if interpreters can process information in meaning chunks rather than individual words, even though the actual size of their WM remains unchanged, the amount of information contained in each unit increases. As a result, interpreters can store and process more information in their WM at a time. From a developmental perspective, once an individual is sufficiently proficient at performing a particular task, they can automize this operation, store it as a schema in LTM and use it at any time with little cognitive effort. For instance, Liu et al. (2004) reported that professional interpreters are more capable of detecting important ideas in speeches than novice interpreters, even though the two groups have a similar WM capacity. This is because professionals have automized this information-detection procedure in their LTM. Therefore, they can implement this procedure in WM without bearing much cognitive load. This interaction between LTM and WM has repeatedly been proved to be essential in providing specific skills and knowledge in tasks with high demands on WM, such as in playing chess (Holding & Reynolds, 1982) and computer programming (Adelson, 1984). However, in theories and models of interpreting studies, LTM has not been well focused when compared to WM. So far, interpreting models that include LTM can be categorized into three kinds. Darò and Fabbro (1994) support an independent and sequential view of WM and LTM. They believe that language is first processed in WM and then in LTM. In WM, interpreters hold what they hear in their phonological loop and conduct subvocal rehearsal to keep the information highly activated. After that, they search in their LTM at all levels (episodic memory, semantic memory, and procedural memory) for comprehension and interpretation.
Published online by Cambridge University Press
Interpreting Studies
Figure 31.3 Darò and Fabbro’s (1994) model of simultaneous interpreting Source: Darò, V., & Fabbro, F. (1994). Verbal memory during simultaneous interpretation: Effects of phonological interference. Applied Linguistics, 15(4), 365–381. Reproduced with permission
Interpreting is completed as soon as the search is done (see Figure 31.3). Ericsson and Kintsch (1995) proposed an interconnected relationship between WM and LTM, proposing the concept of long-term working memory (LT-WM). Specifically, when an interpreter has well-developed interpreting skills or when interpreting materials are familiar to the interpreter, they can activate corresponding interpreting knowledge in LT-WM with little cognitive effort. By contrast, Cowan (1988, 1995) and Mizuno (2005) proposed a hierarchical view of memory where WM is embedded in LTM (see Figure 31.1). When processing information, interpreters activate a subset of relevant items stored in their LTM and then transport them to WM. In this end, only a small fraction of the items will come into the focus of attention for further processing. The above three models illustrate the interaction between WM and LTM during the process of interpreting but not during the development of interpreting expertise. It is widely accepted that LTM is formed after learning from repeated practice with short-term memory. However, the process of this formation is not yet clear. According to Cognitive Load Theory (Paas & van Merriënboer, 1994), the WM-to-LTM transformation occurs when acquired knowledge and skills are transformed into schemas. When this condition is fulfilled, LTM can be enlarged, thus making available more capacity for other WM operations. However, this transformation requires
Published online by Cambridge University Press
703
704
BINGHAN ZHENG AND HUOLINGXIAO KUANG
interpreters to bear a particular type of cognitive load called germane load (Sweller, 1988), which is concerned with knowledge acquisition and procedure automation. To accomplish learning something, interpreters must be sure that their WM is not fully occupied by intrinsic load (decided by the task) and extraneous load (decided by the interpreter’s efficiency in WM use), which means that there is spare germane load in their WM.
31.2.2 The Differences between SI and CI Although SI and CI share a great demand of WM resources, they have different requirements to the use of these cognitive resources and rely on different functions of the executive control. The following sections introduce the specific needs of the two interpreting modes in terms of WM resources and executive functions. 31.2.2.1 Particular Requirements of Interpreting Efforts Gile’s (1997/2002) effort model has described how SI and CI differ in the types of effort required (for a detailed review, see Dong & Cai, 2015). Firstly, SI is a single-stage activity where listening (L), memorizing (M), and target speech production (P) happen simultaneously through coordination (C). By contrast, CI is a two-stage activity consisting of comprehension and reformulation. At the comprehension stage, interpreters listen to the source speech (L), memorize as much of the content as possible (M), produce notes for later memory retrieval (NP), and coordinate (C) these activities; at the reformulation stage, interpreters read their notes (NR), reconstruct the source speech from memory (SR), produce the target speech (P), and coordinate all of these activities (C). Therefore, CI is generally considered less demanding on WM than SI as there is more leeway with regard to task coordination and production time. There are only four studies that have directly tested this assumption by comparing cognitive load in SI and CI from a product-oriented perspective, and three of these rejected this assumption. Lambert (1988) found that interpreters had similar free recall performance after SI and CI. It is assumed that the more demanding the task is, the smaller the number of recalled items will be. Therefore, his result suggests a similar level of cognitive demand across SI and CI. Gile (2001) and Russell (2002), respectively, found higher accuracy in SI and CI renditions. Gile (2001) concluded that there was a WM overload in SI, while Russell (2002) came to the opposite conclusion – that there was a WM overload in CI. Liang et al. (2017) compared the dependency distance, i.e., the distance between two syntactically related words, in SI and CI products. Results showed a smaller distance in the CI rendition, indicating higher processing difficulty in the producing the CI output than the SI output. Secondly, although in both SI and CI, efforts are required in the aspects of listening, memory, production, and coordination, the two modes differ in
Published online by Cambridge University Press
Interpreting Studies
the amount and type of effort required for each aspect. In SI, interpreters finish listening analysis, memory retrieval, and output production within seconds, requiring fast processing in WM; in CI, interpreters comprehend and retrieve the source speech across the two stages of CI, respectively emphasising the processing and storage functions of WM. Although no research has directly confirmed this assumption, Liang et al. (2019) indirectly support it through corpus analysis. In SI, they found interpreters relying on “the most tangible point of reference” (p.10) in the source speech to process language quickly owing to extreme time pressure, resulting in a more form-based interpreting. In CI, on the other hand, interpreters produced a higher frequency of functional words (as opposed to content words) than simultaneous interpreters to remind themselves of the syntactic structure of the source speech, producing a more meaning-based interpretation. In interpreting studies, form-based interpreting is considered faster and less effortful because it only requires shallow language processing, that is, transcoding, while meaning-based interpreting is considered slower and more effortful for it involves deep semantic processing, that is, deverbalization (e.g., Darò & Fabbro, 1994). Combing the results in Liang et al. (2019), both SI and CI rely on the processing function of WM, but the former involves language processing at a superficial level and the latter at a deep level. Moreover, the finding about more functional words in CI than in SI demonstrates interpreters’ need in source speech memory retrieval, which greatly relies on the storage function of WM.
31.2.2.2 Demands on WM executive functions SI requires concurrent listening comprehension and translation articulation, two activities that compete for the same cognitive resources in the phonological loop of WM. According to Baddeley and Hitch (1974), the phonological loop comprises two components: a temporary phonological store in which to keep perceived sounds and a subvocal rehearsal system for keeping the sounds active. It has been found that overt articulation hinders subvocal rehearsal, and information in the phonological store that is not rehearsed decays (Daró & Fabbro, 1994; Christoffels et al., 2006). This articulatory suppression effect explains SI’s particular demand on WM to inhibit phonological interference. Specifically, while interpreters are articulating the translation of an earlier piece of information, they must inhibit the hampering effect this articulation poses on listening to newly coming-in information. Darò and Fabbro’s (1994) model reflects this phonological interference in SI (see Figure 31.3) by linking the auditory processing of the source text with the overt delivery of the target text (the arrow from TL [target language] points toward WM, as shown in Figure 31.3). This assumption has been confirmed by a series of studies, in which interpreters’ recall performance after SI has been compared with that after other tasks such as listening, shadowing, and listening with articulatory suppression (e.g., Isham, 2000; Lambert, 1988; Padilla et al., 2005). It was repeatedly
Published online by Cambridge University Press
705
706
BINGHAN ZHENG AND HUOLINGXIAO KUANG
found that recall after SI is worse than after other tasks, indicating a detrimental effect on memory caused by the phonological interference in SI. As a matter of fact, these findings provide empirical evidence for introducing phonological interference into future SI models. In comparison, CI puts enormous pressure on the memory for complete and accurate recall of source speeches. So far, there is no model dedicated to WM operation in CI. Dong et al. (2018) interpreted Mizuno’s (2005) process model of WM and interpreting from a CI perspective. They argue that Updating plays an essential role in CI. In the comprehension phase of CI, interpreters constantly update information in their WM as the source speech is delivered continuously; then, in the reformulation phase, they reactivate “the stretch of input which has already passed FOA (focus of attention)” to recall the source speech (Dong et al., 2018, p. 3). In their empirical study, Dong and her colleagues also confirmed that Updating is a predictor of CI performance (for a detailed discussion, see Section 31.3.2). Another major difference between SI and CI lies in note-taking. When interpreting long sections, consecutive interpreters usually take notes to alleviate the pressure on their memory. To complete note-taking, interpreters “translate” the verbal input from the phonological loop into written notes with the help of a visuospatial sketchpad, while in note-reading, they decode the visual notes and re-express them in the target language. In either process, the phonological loop and the visuospatial sketchpad work closely together and compete for the limited WM capacity. Therefore, interpreters must develop coordinating skills to balance these two components of WM. In WM executive control, note-taking is also a complex cognitive activity (Piolat et al., 2005). Firstly, interpreters shift between source speech comprehension and note production. During note production, they make a further shift between language-based notes and symbols. Secondly, interpreters constantly update information in WM to follow the source speech. Thirdly, they consistently inhibit the activation of inappropriate note forms, and subvocal rehearsal during writing. In summary, note-taking in CI places a heavy demand on the audio and visual components of WM and the three executive functions of WM. Note-taking research in its own right has found that students with larger WM spans experience less cognitive load in notetaking and cover more information in their notes in the classroom (Piolat, 2007). However, whether consecutive interpreters’ WM capacity affects the quality of their note-taking and the amount of effort required remains underexplored.
31.3
Working Memory in Empirical Interpreting Studies
Empirical evidence is essential in (in)validating the theoretical discussions on WM in interpreting. This section is devoted to reviewing empirical interpreting studies that have compared SI and CI from a WM perspective.
Published online by Cambridge University Press
Interpreting Studies
31.3.1 The Similarities between SI and CI This part of the review focuses on what SI and CI share in WM growth. It specifically looks into how interpreter training experience and interpreter work experience interact with the change of WM size and the use of executive control. 31.3.1.1 Impact of Interpreting Training on WM Span Since the 1990s, interpreting researchers have been attempting to determine whether interpreters have an advantage over noninterpreters in WM span (Darò & Fabbro, 1994, Padilla et al., 1995). Tasks used to test WM span can be divided into two kinds, based on the two functions of WM: information storage and information processing (Baddeley & Hitch, 1974). The first kind only tests WM storage through simple recall tasks such as the digit span task. The other kind evaluates the integral use of the two functions using complex span tasks, such as recall after semantic judgment (Timarová et al., 2014). In both situations, researchers have hypothesized an interpreter advantage for two reasons: first, memory has long been considered to be a basic skill required for completing interpreting tasks (Gile, 1997/2002; Seleskovitch, 1968/1978); second, interpreters are consistently confronted with concurrent storage and processing demands during interpreting, from which they can develop WM skills that will enable them to tackle the issue (Timarová et al., 2014). Overall, in previous research, in simple and complex span tests where digits, words, and sentences are presented visually, interpreters have consistently been found to have an advantage over noninterpreters in terms of WM span, and professional interpreters have consistently been found to have an advantage over novices. However, it was also found that this advantage sometimes disappeared when the stimuli were auditory. This runs counter to our expectation that interpreters would have an advantage in the auditory modality, as they are consistently exposed to auditory processing. There are two possible explanations for this modality effect. First, “reading” words or sentences aloud in some reading span experiments actually gives subjects an opportunity to strengthen their visual memory using self-generated phonological codes (Penney, 1989). According to Penney’s (1989) model of short-term memory, auditory items are automatically encoded in acoustic and phonological codes, while visual items must be intentionally translated into verbal forms to generate phonological codes. If no rehearsal is involved, phonological codes can fade away in seconds. If visual-to-auditory translation is successful, people can trace their memory of the visual stimuli through codes in both visual and audio modalities. Hence, visual presentation with overt vocalization can be more traceable than auditory presentation alone, and therefore helps subjects to complete recall tasks. Secondly, more often than not, interpreters consult prepared speeches, slides, notes, and other visual information during interpreting, thus developing a superior ability in visual-to-auditory translation.
Published online by Cambridge University Press
707
708
BINGHAN ZHENG AND HUOLINGXIAO KUANG
In comparison, noninterpreters practice this visual-to-auditory translation much less frequently. Therefore, they need more time to complete this translation procedure and create corresponding phonological codes. However, the longer they take, the more the memory decays. As a result, in the studies referred to here, noninterpreters did not outperform interpreters in visually presented WM span tasks. Several researchers examined this modality effect, and all of them rejected the idea that such an effect exists. Daneman and Carpenter (1980) found a high correlation between college students’ reading span and listening span. Köpke and Signorelli (2011) and Chmiel (2018) also reported similar reading and listening spans in interpreters. Thus, in previous studies, this modality effect might just have been a coincidence caused by nonstandard experimental settings. Wen and Dong (2019) synthesized the findings of 10 primary studies on interpreters’ WM and short-term memory. They found that whether interpreters had an advantage over noninterpreters in WM span depended on task type. An interpreter advantage was observed in verbal and numerical/letter span tasks but not in spatial tasks. It is worth mentioning that in almost all studies on interpreters’ WM, researchers selected professional and novice interpreters according to the total time they had spent on interpreting work and/or training. However, few researchers have described the proportions of SI and CI in the interpreters’ work/training experience. In Hiltunen et al. (2014), simultaneous and consecutive interpreters were compared to linguistic nonexperts and foreign language teachers in a free recall task. They found that only the simultaneous interpreter group outperformed the linguistic nonexpert group. This suggests that SI and CI affect the growth of WM size in different ways. Therefore, when recruiting interpreter subjects and describing their background, the proportions of SI and CI work/training experience should be considered as an important variable to be controlled for. On the other hand, an increase in WM span has also been found to be related to higher interpreting accuracy and fluency. Tzou et al. (2012) reported that student interpreters had significantly larger reading spans than untrained bilinguals. And their interpreting accuracy of selected sentences and the overall speech was positively correlated with their reading spans. Lin et al. (2018) found a significant correlation between students’ reading span and the number of interruptions and hesitations in their interpretation, concluding that WM is a powerful predicator of SI fluency. This is in line with Injoque-Ricle et al. (2015) and Macnamara and Conway (2016), who also reported a positive correlation between WM span and SI quality. With regard to CI, Dong et al. (2018) tested students’ L2 listening span before and after CI training and found that the pretest span predicted CI performance. Yenkimaleki and van Heuven (2017) compared the CI performance of a control group (with no memory skill training) and an experimental group (with memory skill training). A positive effect of
Published online by Cambridge University Press
Interpreting Studies
memory training on reducing omissions in interpreting was observed, which indicates the benefits of an increased WM span on CI fluency. However, all of the above-mentioned studies were conducted with interpreter trainees. Timarová et al. (2015) conducted a study with professional interpreters and reported only a marginally significant correlation between interpreters’ letter span and interpreting accuracy in figures. So far, little evidence is available of a relationship between WM capacity and interpreting quality among professional interpreters.
3.1.2 Novice Advantage in Simple and Complex WM Span Tasks Novice interpreters have repeatedly been found to produce similar or even better performance than professional interpreters in simple WM span tasks, and sometimes in complex WM span tasks. Based on a meta-analysis of 10 studies on this topic, Wen and Dong (2019) concluded that expert interpreters outperform beginner interpreters but not intermediate interpreters in WM span tasks. It seems that the increase in WM span stops at an unclear point during the development of interpreting expertise. Several longitudinal studies conducted with novice interpreters have demonstrated the effect of interpreting training on WM span (Babcock & Vallesi, 2017; Dong & Liu, 2016; Nour, Struys & Stenger, 2020). However, there are contradictory findings concerning the effect of interpreting experience on WM span, with around half of the studies acknowledging professional interpreters’ advantage (Nour, Struys & Stenger, 2020; Padilla et al., 1995; Tzou et al., 2011), and the other half denying such an advantage (Köpke & Nespoulous, 2006; Liu et al., 2004; Padilla, 1995). There are two potential reasons why a novice might have an advantage in simple WM tasks. First, professional interpreters are usually older than novice interpreters, leading to an aging effect on WM span. Signorelli et al. (2011) proved that younger interpreters performed better than professionals in nonword repetition and cued recall, suggesting a trade-off between age and WM span. Second, compared with novice interpreters, professional interpreters usually process information in larger units. In this way, they can alleviate the pressure on their memory and focus more on language processing (Wen & Dong, 2019). A novice advantage has also been observed in complex span tasks. Liu et al. (2004) used a CI-like WM task where subjects first processed semantic meaning and then recalled the last words of sentences. The results show that professionals and novices performed similarly. Köpke and Nespoulous (2006) tested subjects with a SI-like task of free recall with articulatory suppression and found that novices recalled more words than professionals. These findings challenged the conventional view that expertise in interpreting relies on a greater WM capacity. Köpke and Nespoulous (2006) speculated that professional interpreters have developed an optimal language processing route that releases them from the constraints of WM size.
Published online by Cambridge University Press
709
710
BINGHAN ZHENG AND HUOLINGXIAO KUANG
Novice interpreters without such optimal processing routes can easily experience an overload of WM during interpreting. It is not yet clear how this optimization of language processing works. However, to some extent it implies that the development of interpreting expertise is centred on domain-specific abilities like language processing rather than on general cognitive abilities.
3.1.3 Impact of Interpreting Experience on WM Executive Control An interpreter advantage has not only been observed in WM span tasks that test interpreters’ information storage and/or processing ability but also in tasks targeting on their operation of WM executive control. Overall, interpreters have been found to have an advantage in using the executive functions of WM, although the reasons for this advantage have not yet been clarified. With regard to Inhibition, Dong and Zhong (2017) reported an effect of interpreting experience, finding that more experienced student interpreters outperformed both less experienced students and more balanced noninterpreter bilinguals in a Flanker task. By contrast, Woumans et al. (2015) found that interpreters outperformed an unbalanced group of bilinguals but not balanced bilinguals in a Simon task and an Attention Network test,1 which implies that the (un)balance of bilingualism plays a decisive role in Inhibition execution. According to Nour, Struys, Woumans et al.’s (2020) detailed review of eight studies on interpreters’ Response-Distractor Inhibition, interpreters only show an advantage when they are compared with unbalanced bilinguals and not when compared with balanced bilinguals. With regard to Updating, Dong and Liu (2016) tested students’ execution of this function through a n-back task2 before and after CI training, and observed a significant improvement in students’ task performance. They explain that in CI, interpreters repeatedly replace old information with new information during the input stage and that they refresh their memory of the source speech during the output stage. In this way, interpreters strengthen their Updating ability. This could explain why Morales et al. (2015) and Timarová et al. (2014) did not find an advantage on the part of simultaneous interpreters in this regard. In SI, interpreters usually release their memory immediately after finishing the interpretation of the current information. Nevertheless, simultaneous interpreters can acquire this Updating ability quickly. Morales et al. (2015) compared simultaneous interpreters’ performance across the two blocks of a n-back task and found improved accuracy in the second block. In contrast, noninterpreter bilinguals showed no improvement throughout the task. Taken together, it seems that CI and SI experience has benefited the development of Updating in different ways. In comparison, findings concerning the positive impact of interpreting training and/or work experience on Shifting have been consistent (Dong
Published online by Cambridge University Press
Interpreting Studies
& Liu, 2016; Macnamara & Conway, 2014; Yudes et al., 2011), the only exception being Timarová et al. (2014). This is owing to the high sensitivity of Shifting to environmental factors. It has been found that over 20% of its variability is attributable to nongenetic factors (Nour, Struys, Woumans et al., 2020). In other words, the constant shift between languages during interpreting enables interpreters to improve their Shifting ability. In comparison, noninterpreter bilinguals face fewer Shifting needs in language use, resulting in slower reactions to shifting-related tasks. Moreover, interpreting is conducted under extremely high time pressure, meaning that interpreters must make Shifting decisions quickly. Compared with translators, who also switch between languages but with much lower time pressure, interpreters have been found to perform better in Shifting-related tasks (Henrard & van Daele, 2017). All in all, an interpreter advantage has been found in the execution of all three executive functions of WM, although how bilingual competence and interpreting practice interactively contribute to this advantage remains underexplored.
31.3.2 The Differences between SI and CI The following review introduces how SI and CI differ from each other in two aspects: the use of executive functions of WM and the fluctuation of cognitive load during the process of interpreting. Empirical evidence about the attributes of SI and CI in these regards has been found in studies on respective interpreting modes. 31.3.2.1 Working Memory: Inhibition in SI and Updating in CI The three executive functions of WM affect the subprocesses of interpreting differently (Timarová et al., 2014). In SI, there is an overlap between listening to the source speech and articulating the target speech for about 70% of the time (Chernov, 1994). Interpreters must therefore minimize the phonological interference caused by concurrent listening and speaking through Inhibition (e.g., Chincotta & Underwood, 1998; Padilla et al., 1995; Yudes et al., 2012). Moreover, the concurrent execution of subtasks like listening comprehension, interpretation delivery, and self-monitoring requires interpreters to inhibit the interferences caused by multitasking. In CI, interpreters memorize the inputted source speech and retrieve it when producing the output. Their pressure mainly derives from having to refresh their memory of the source speech, ranging from single words to entire speeches (Dam, 2010), through Updating (Dong et al., 2018; Dong & Liu, 2016). In short, in SI the emphasis is on Inhibition, while in CI the emphasis is on Updating, but both require Shifting for language transfer. To test simultaneous interpreters’ Inhibition advantage in resisting phonological interference, researchers usually employ a free recall task with articulatory suppression. This task requires interpreters to articulate
Published online by Cambridge University Press
711
712
BINGHAN ZHENG AND HUOLINGXIAO KUANG
irrelevant sounds while listening to words or sentences, and thus creates an SI-like scenario. Using this task, Köpke and Nespoulous (2006) identified an interpreter advantage over noninterpreter bilinguals and a novice advantage over professional interpreters. Similarly, Padilla’s team (1995, 2005) found that novice interpreters outperformed noninterpreter bilinguals whether they had high reading spans or not. However, it has been argued that the novice advantage in Inhibition might be misleading since the selected task does not resemble SI as assumed. The suppressed recall task did prevent subvocal rehearsal as SI does because it requires interpreters to listen to and voice irrelevant information, whereas SI requires interpreters to articulate a translation of whatever they have heard. Moreover, the words or sentences used in the task are usually unrelated to each other, while the input in SI is normally a logical and coherent text, which provides interpreters with contextual hints that facilitate better source speech comprehension and prediction. Therefore, when the adopted WM task does not resemble SI, more-experienced interpreters might not be able to show an advantage over less-experienced interpreters in Inhibition. An alternative explanation for no Inhibition advantage in more-experienced interpreters may be that Inhibition is essential in the early stage of interpreting training but not as essential in building interpreting expertise. Research on Inhibition and multi-tasking has yielded more contradictory results than research on Inhibition and phonological interference in SI. A few researchers have reported an interpreter advantage in this regard (Köpke & Nespoulous, 2006; Woumans et al., 2015), while most researchers have denied it (Babcock & Vallesi, 2017; Dong & Liu, 2016; Dong & Xie, 2014; Yudes et al., 2011). Many of the tasks adopted in these studies focus on nonverbal information processing (arrows, shapes, and colours). However, not all SI skills can be transferred to nonverbal or general cognitive tasks. Morales et al. (2015) found that interpreters did not present any advantage in conflict resolution, but they showed an advantage in alertness and orienting tasks, which involve multitasking as SI does. Similarly, Dong and Liu (2016) did not detect an interpreting training effect on interpreters’ ability to resist conflicts, since interpreters are not often required to resolve conflicts of this kind during interpreting. All the above indicate that an interpreter advantage in Inhibition depends on the transferability of interpreting expertise to the adopted tasks. In contrast, research on Updating and CI has yielded more consistent results. Dong’s team first reported a positive impact of CI training on interpreters’ Updating ability (Dong & Liu, 2016). Later, they found that students’ Updating ability successfully predicted their CI performance both before and after CI training (Dong et al., 2018). Finally, they did a metaanalysis of four studies on Updating and interpreting and found that an interpreter advantage in updating ability had been observed in both CI and SI (Wen & Dong, 2019). It is noteworthy that an improvement in updating ability was observed in subjects who had received CI training for the shortest time (a 32-hour in-class
Published online by Cambridge University Press
Interpreting Studies
training program) (Dong & Liu, 2016). However, no such improvement was witnessed in memory span (Dong et al., 2018), shifting, or inhibitory control (Dong & Liu, 2016). Based on these findings, Wen and Dong (2019) concluded that “updating ability is probably the first taxed and trained memory skill in interpreting training relative to memory spans (short-term memory and WM spans)” (p. 12). This might be caused by the speaker-paced feature of SI and CI. Without Updating, interpreters would not be able to make enough memory space available for new information. In comparison, Inhibition and Shifting are more necessary after new information has entered the WM and is competing with old information for WM resources. To verify this assumption, researchers should conduct longitudinal studies with noninterpreters who start to learn interpreting and novice interpreters who gradually gain more interpreting experience. In this way, researchers can find out how the three executive functions of WM develop along with the course of interpreting training and practice.
31.3.2.2 Local Cognitive Load in SI and CI Cognitive load is “the load that performing a particular task imposes on the learner’s cognitive system” (Paas et al., 1995, p. 64). When it exceeds an individual’s WM capacity, task performance drops or even crashes. For interpreters, there are “global” and “local” levels of cognitive load. At a “global” level, interpreting as a single task entails more cognitive load than shadowing and listening do (Christoffels & de Groot, 2004; Köpke & Nespoulous, 2006; Padilla et al, 2005). At a “local” level, cognitive load fluctuates throughout the whole process of interpreting. In SI, interpreters’ cognitive load fluctuates below sentence level because they follow speakers closely to produce target speeches. But in CI, interpreters’ cognitive load varies beyond sentence level because they can flexibly allocate cognitive resources to the two stages of CI, as long as the sum of the allocated resources does not exceed their WM limit. Most of the research on this subject has attempted to identify the factors that affect the local cognitive load of interpreting, while only few studies have directly measured the amount of cognitive load. SI researchers have mainly measured interpreters’ local cognitive load by analyzing their interpreting products. It is assumed that interpreting quality drops as cognitive load increases. On interpreting accuracy, Gile (2017) asked 10 professional interpreters to interpret the same materials twice. He found new errors and omissions at different places in the interpreters’ second rendering, which indicates a change in the cognitive load fluctuation pattern between the two interpretations. On interpreting fluency, Plevoets and Defrancq (2016) took the occurrence rate of uh(m) as an indicator of local cognitive load and found that it was predicted by the delivery rate of the source text. In other words, local cognitive load increases as the speakers’ delivery rate increases. Shao and Chai (2020) found that interpreters’ local cognitive load reached a peak when processing four information
Published online by Cambridge University Press
713
714
BINGHAN ZHENG AND HUOLINGXIAO KUANG
chunks at one time. When they exceeded four chunks, interpreters’ performance dropped dramatically. They also measured a specific kind of local cognitive load during interpreting: the current load, which is caused by the accumulated information of the current sentence. They measured it through ear-voice span (EVS), the time span between input and output, and it was found to be significantly correlated with SI performance. In other studies, the length of EVS has been found to increase with an increase in syllable and sentence length (Lee, 2002) and with an increase in syntactic and semantic complexity (Timarová et al., 2014), and to decrease with an increase in preparation time (Díaz-Galaz et al., 2015) and interpreting experience (Timarová et al., 2014). Although those researchers did not interpret EVS from the perspective of local cognitive load, they provided empirical evidence that the online cognitive load of SI changes within a sentence. Seeber (2011) measured interpreters’ current load using pupillometry, and found that interpreters’ pupils dilate more toward the end of sentences while handling asymmetrical structures in SI. This might be attributed to the four interpreting strategies interpreters adopted in this SI task: waiting, stalling, chunking, and anticipating. Among the four strategies, the first three allow interpreters to wait in order to obtain more information. Therefore, interpreters’ local cognitive load accumulates toward the end of sentences. In contrast to SI, very few studies have measured local cognitive load in CI. Wu and Wang (2009) analyzed the case of an interpretation delivered at a press conference held by the Chinese government and found that interpreters can save cognitive efforts by simplifying sentence structures in source speeches. They identified three simplification methods: deleting the overlapping part of different sentences, transforming sentences into the same structures, and merging sentences with similar structures. Chen (2017) measured interpreters’ ear-pen span during note-taking and eye fixations during note-reading. The data suggest that notes that entail longer ear-pen span and cost more cognitive efforts during note-taking result in shorter fixations and less cognitive effort during note-reading. In other words, there is a trade-off in cognitive load across the two phases of CI. All in all, interpreters experience fluctuating local cognitive load during interpreting. The amount of load varies within sentences in SI and beyond sentences in CI. More process-oriented research is needed to unveil the complex fluctuation pattern of local cognitive load during the interpreting process.
31.4
Discussion
Based on a systematic review of interpreting research on WM capacity and executive control, we observed that in most cases interpreters showed an advantage over noninterpreters in these regards. However, more-experienced
Published online by Cambridge University Press
Interpreting Studies
interpreters and less-experienced interpreters were alternately found advantageous than each other concerning WM capacity and executive control. From theoretical and methodological perspectives, these findings could lay the ground for future empirical research in this field. First, the inconsistent findings about the benefits of interpreting experience accumulation on WM might be derived from the mismatch between participants’ interpreting experience and the WM task selected. On the participant side, CI demands information storage during input and Updating for recall during output, while SI requires fast information processing and Inhibition for resisting phonological interference. The two interpreting modes contribute to the growth of WM in different ways. On the task side, researchers can test interpreters’ information storage and processing ability with simple and complex memory span tasks in visual and audio modalities, or their executive control with tasks specifically targeted at the three WM executive functions. Thus, if consecutive interpreters were tested using information processing tasks or simultaneous interpreters were tested using information storage tasks, then more-experienced interpreters may not be able to show an advantage over less-experienced interpreters in the tested aspect of WM. Future research should strictly control the variables involved in experimental designs to ensure that interpreters’ CI and SI experience composition or the selected WM task are not confounding factors affecting the research outcome. Second, a great deal of research has been conducted on WM in SI, while little attention has been paid to CI in this regard. Among those SI studies that included WM capacity control in their participant selection, almost all use recall tasks containing an input stage of stimuli presentation and an output stage of stimuli recall, which resemble CI tasks more than SI tasks. In other words, researchers tested interpreters with the same ability in CI-like tasks and correlated this ability with interpreters’ SI performance. Therefore, WM tasks involving separate input and output are suggested for CI research, and those requiring simultaneous input and output are advised for SI research. Moreover, it would be interesting to compare WM demands and interpreters’ coping strategies across the two stages of CI. At present, students usually receive CI training before SI training, as CI ability is generally considered to be a prerequisite for completing SI tasks. To decide whether this order is reasonable in curriculum design, more empirical evidence from comparative research on WM in CI and SI would be necessary. Third, concerning the contradictory findings regarding WM and interpreting quality, more process-oriented research should be conducted to see how WM demand fluctuates during interpreting from beginning to end. The process-oriented methodology has been used in several pioneering SI and CI studies that recorded interpreters’ eye movements, galvanic skin responses, and other physiological responses during interpreting (e.g., Chen, 2017; Seeber, 2011). These studies have proved that interpreting is a dynamic process, during which interpreters experience cognitive adaptation
Published online by Cambridge University Press
715
716
BINGHAN ZHENG AND HUOLINGXIAO KUANG
and emotional fluctuation. However, there is a challenge in measuring the three types of cognitive load (intrinsic load, extraneous load, and germane load) separately, and in differentiating positive emotions (such as excitement) from negative emotions (such as disappointment). If cognitive load can be measured separately, the amount of germane load can reveal how many WM resources are devoted to “learning” during interpreting. Researchers could measure student interpreters’ germane load and score their interpreting performance before and after interpreting training. By dividing students into high-score and low-score groups, researchers could compare how different the two groups are in terms of the amount of germane load they devoted to learning and the quality of the interpretation, further illustrating the formation of interpreting expertise. Similarly, if positive and negative emotions can be presented separately, researchers will be able to understand better how interpreters deal with the overwhelming stress during interpreting. For instance, if more excitement correlates with better interpretation, then motivation enhancement would be an essential aspect for interpreting trainers to consider during curriculum design. Looking ahead, research on WM and interpreting would benefit from investigating several underexplored issues. First, interpreting in general is beneficial to the growth of WM. However, it is not yet clear how SI and CI contribute to this growth and in turn benefit from the growth through enhanced interpreting quality. Second, little attention has been paid to CI, in which the two stages of input and output place different demands on WM, leading to a shortage of comparative discussions on the role of WM in CI and SI. Third, cognitive load fluctuates during the process of interpreting, giving rise to the question of how speaker-related variables such as delivery speed and interpreter variables such as interpreting strategies affect the pattern of cognitive load fluctuation during interpreting.
31.5
Conclusion
This chapter has reviewed the operation of WM in SI and CI from theoretical and empirical perspectives. Theoretically, both interpreting modes place high demands on WM for language control and processing control. SI centers more on fast information processing because of its requirement for an immediate input-to-output transformation, while CI emphasizes both information processing and storage because of the need to recall source speeches. Empirically, WM and interpreting have been found to be interdependent on each other in a complex way. Questions remain regarding the development path of WM during the course of interpreting training and its interaction with interpreting performance. Future research could combine process-oriented and product-oriented approaches to unveil the cognitive mechanism governing the interpreting process and modulating the interpreting quality.
Published online by Cambridge University Press
Interpreting Studies
References Adelson, B. (1984). When novices surpass experts: The difficulty of the task may increase with expertise. Journal of Experimental Psychology: Learning, Memory and Cognition, 10, 484–495. Babcock, L., & Vallesi, A. (2017). Are simultaneous interpreters expert bilinguals, unique bilinguals, or both? Bilingualism: Language and Cognition, 20(2), 403–417. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G.H, Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (pp. 47-89). Academic Press. Chen, S. (2017). Exploring the process of note-taking and consecutive interpreting: A pen-eye-voice approach towards cognitive load (Doctoral dissertation, Macquarie University, Sydney, Australia). Chernov, G. V. (1994). Message redundancy and message anticipation in simultaneous interpretation. In S. Lambert & B. Moser-Mercer (Eds.), Bridging the gap: Empirical research in simultaneous interpretation (pp. 139–153). John Benjamins. Chincotta, D., & Underwood, G. (1998). Simultaneous interpreters and the effect of concurrent articulation on immediate memory: A bilingual digit span study. Interpreting, 3(1), 1–20. Chmiel, A. (2018). In search of the working memory advantage in conference interpreting: Training, experience and task effects. International Journal of Bilingualism, 22(3), 371–384. Christoffels, I. K., de Groot, A. M., & Kroll, J. F. (2006). Memory and language skills in simultaneous interpreters: The role of expertise and language proficiency. Journal of Memory and Language, 54(3), 324–345. Cowan, N. (1988). Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human informationprocessing system. Psychological Bulletin, 104(2), 163–191. Cowan, N. (1995). Attention and memory: An integrated framework. Oxford University Press. Dam, H. V. (2010). Consecutive interpreting. In Y. Gambier & L. van Doorslaer (Eds.), Handbook of translation studies (Vol. 1, pp. 75–79). John Benjamins. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19(4), 450–466. Darò, V., & Fabbro, F. (1994). Verbal memory during simultaneous interpretation: Effects of phonological interference. Applied Linguistics, 15(4), 365–381. Díaz-Galaz, S., Padilla, P., & Bajo, M. T. (2015). The role of advance preparation in simultaneous interpreting: A comparison of professional interpreters and interpreting students. Interpreting, 17(1), 1–25.
Published online by Cambridge University Press
717
718
BINGHAN ZHENG AND HUOLINGXIAO KUANG
Dong, Y., & Cai, R. (2015). Working memory in interpreting: A commentary on theoretical models. In Z. Wen, M. Mota, & A. McNeill (Eds.), Working memory in second language acquisition and processing (pp. 63–79). Multilingual Matters. Dong, Y. & Li, P. (2020). Attentional control in interpreting: A model of language control and processing control. Bilingualism: Language and Cognition, 23, 716–728. Dong, Y., & Liu, Y. (2016). Classes in translating and interpreting produce differential gains in switching and updating. Frontiers in Psychology, 7, 1297. Dong, Y., Liu, Y., & Cai, R. (2018). How does consecutive interpreting training influence working memory: A longitudinal study of potential links between the two. Frontiers in Psychology, 9, 875. Dong, Y., & Xie, Z. (2014). Contributions of second language proficiency and interpreting experience to cognitive control differences among young adult bilinguals. Journal of Cognitive Psychology, 26(5), 506–519. Dong, Y., & Zhong, F. (2017). Interpreting experience enhances early attentional processing, conflict monitoring and interference suppression along the time course of processing. Neuropsychologia, 95, 193–203. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102(2), 211–245. Ericsson, K. A. & Charness, N. (1994). Expert performance: Its structure and acquisition. American Psychologist, 49, 725–747. Gile, D. (1997/2002). Conference interpreting as a cognitive management problem. In F. Pöchhacker & M. Shlesinger (Eds.), The interpreting studies reader (pp. 162–176). London: Routledge. Gile, D. (2001). Consecutive vs. simultaneous: Which is more accurate? Interpretation Studies, 1(1), 8–20. Gile, D. (2017). Testing the Effort Models’ tightrope hypothesis in simultaneous interpreting: A contribution. HERMES, 12(23), 153–172. Henrard, S., & van Daele, A. (2017). Different bilingual experiences might modulate executive tasks advantages: Comparative analysis between monolinguals, translators, and interpreters. Frontiers in Psychology, 8, 1870. Hiltunen, S., Pääkkönen, R., Vik, G. V., & Krause, C. M. (2014). On interpreters’ working memory and executive control. International Journal of Bilingualism, 20(3), 297–314. Holding, D. H., & Reynolds, R. I. (1982). Recall or evaluation of chess positions as determinants of chess skill. Memory & Cognition, 10(3), 237–242. Injoque-Ricle, I., Barreyro, J. P., Formoso, J., & Jaichenco, V. I. (2015). Expertise, working memory and articulatory suppression effect: Their relation with simultaneous interpreting performance. Advances in Cognitive Psychology, 11(2), 56–63. doi:10.5709/acp-0171-1
Published online by Cambridge University Press
Interpreting Studies
Isham, W.P. (2000). Phonological interference in interpreters of spokenlanguages: An issue of storage or process? In B. E. Dimitrova & K. Hyltenstam (Eds.), Language processing and simultaneous interpreting (pp. 133–150). John Benjamins. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122–149. Köpke, B., & Nespoulous, J. L. (2006). Working memory performance in expert and novice interpreters. Interpreting, 8(1), 1–23. Köpke, B., & Signorelli, T. M. (2011). Methodological aspects of working memory assessment in simultaneous interpreters. International Journal of Bilingualism, 16(2), 183–197. Lambert, S. (1988). Information processing among conference interpreters: A test of the depth-of-processing hypothesis. Meta, 33(3), 377–387 Lee, T. H. (2002). Ear voice span in English into Korean simultaneous interpretation. Meta, 47(4), 596–606. Liang, J., Fang, Y., Lv, Q., & Liu, H. (2017). Dependency distance differences across interpreting types: Implications for cognitive demand. Frontiers in Psychology, 8, 1–10. doi: 10.3389/fpsyg.2017.02132 Lin, Y., Lv, Q., & Liang, J. (2018). Predicting fluency with language proficiency, working memory, and directionality in simultaneous interpreting. Frontiers in Psychology, 9, 1543. doi:10.3389/fpsyg.2018.01543 Liang, J., Lv, Q., & Liu, Y. (2019). Quantifying interpreting types: Language sequence mirrors cognitive load minimization in interpreting tasks. Frontiers in Psychology, 10, 1–12. doi: 10.3389/fpsyg.2019.00285 Liu, M., Schallert, D. L., & Carroll, P. J. (2004). Working memory and expertise in simultaneous interpreting. Interpreting, 6(1), 19–42. Macnamara, B. N., & Conway, A. R. (2016). Working memory capacity as a predictor of simultaneous language interpreting performance. Journal of Applied Research in Memory and Cognition, 5(4), 434–444. Macnamara, B. N., & Conway, A. R. A. (2014). Novel evidence in support of the bilingual advantage: Influences of task demands and experience on cognitive control and working memory. Psychonomic Bulletin & Review, 21 (2), 520–525. Mizuno, A. (2005). Process model for simultaneous interpreting and working memory. Meta, 50(2), 739–752. Morales, J., Padilla, F., Gómez-Ariza, C. J., & Bajo, M. T. (2015). Simultaneous interpretation selectively influences working memory and attentional networks. Acta Psychologica, 155, 82–91. Nour, S., Struys, E., & Stengers, H. (2020). Adaptive control in interpreters: Assessing the impact of training and experience on working memory. Bilingualism: Language and Cognition, 23, 772–779. Nour, S., Struys, E., Woumans, E., Hollebeke, I., & Stengers, H. (2020). An interpreter advantage in executive functions? A systematic review. Interpreting, 22(2), 163–186.
Published online by Cambridge University Press
719
720
BINGHAN ZHENG AND HUOLINGXIAO KUANG
Paas, F. G., & Van Merriënboer, J. J. (1994). Instructional control of cognitive load in the training of complex cognitive tasks. Educational Psychology Review, 6(4), 351–371. Padilla, P. (1995). Procesos de memoria y atención en la interpretación de lenguas (Doctoral dissertation, Universidad de Granada. Granada, Spain). Padilla, P., Bajo, M. T., Canas, J. J., & Padilla, F. (1995). Cognitive processes of memory in simultaneous interpretation. In J. Tommola (Ed.), Topics in interpreting research (pp. 61–72). Centre for Translation and Interpreting, University of Turku. Padilla F., Bajo M. T., & Macizo P. (2005). Articulatory suppression in language interpretation: Working memory capacity, dual tasking and word knowledge. Bilingualism: Language and Cognition, 8. 207–219. Penney, C. G. (1989). Modality effects and the structure of short-term verbal memory. Memory & Cognition, 17(4), 398–422. Piolat, A. (2007). Effects of notetaking technique and working-memory span on cognitive effort and recall performance. In M. Torrance, L. Van Waes, & D. Galbraith (Eds.), Writing and cognition: Research and applications (pp. 109–124). Elsevier. Piolat, A., Olive, T., & Kellogg, R. T. (2005). Cognitive effort during note taking. Applied Cognitive Psychology, 19(3), 291–312. Plevoets, K., & Defrancq, B. (2016). The effect of informational load on disfluencies in interpreting: A corpus-based regression analysis. Translation and Interpreting Studies, 11(2), 202–224. Russell, D. (2002). Interpreting in legal contexts: Consecutive and simultaneous interpreting. Linstok Press. Seeber, K. G. (2011). Cognitive load in simultaneous interpreting: Existing theories – new models. Interpreting, 13(2), 176–204. Seleskovitch, D. (1968/1978). Interpreting for international conferences-Problems of language and communication. Pen and Booth. Shao, Z., & Chai, M. (2020). The effect of cognitive load on simultaneous interpreting performance: An empirical study at the local level. Perspectives, 1–17. doi: 10.1080/0907676X.2020.1770816 Signorelli, T. M., Haarmann, H. J., & Obler, L. K. (2011). Working memory in simultaneous interpreters: Effects of task and age. International Journal of Bilingualism, 16(2), 198–212. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. Timarová, Š., Čeňková, I., & Meylaerts, R. (2015). Simultaneous interpreting and working memory capacity. In Ferreira, A. & J. W. Schwieter (Eds.), Psycholinguistic and cognitive inquiries into translation and interpreting (pp. 101–126). Benjamins. Timarová, Š., Čeňková, I., Meylaerts, R., Hertog, E., Szmalec, A., & Duyck, W. (2014). Simultaneous interpreting and working memory executive control. Interpreting, 16(2), 139–168.
Published online by Cambridge University Press
Interpreting Studies
Tzou, Y. Z., Eslami, Z. R., Chen, H. C., & Vaid, J. (2012). Effect of language proficiency and degree of formal training in simultaneous interpreting on working memory and interpreting performance: Evidence from Mandarin-English speakers. International Journal of Bilingualism, 16(2), 213–227. Wen, H., & Dong, Y. (2019). How does interpreting experience enhance working memory and short-term memory: A meta-analysis. Journal of Cognitive Psychology, 31(8), 769–784. Woumans, E., Ceuleers, E., van der Linden, L., Szmalec, A., & Duyck, W. (2015). Verbal and nonverbal cognitive control in bilinguals and interpreters. Journal of Experimental Psychology: Learning, Memory, and Cognition, 41(5), 1579–1586. Wu, G., & Wang, K. (2009). Consecutive interpretation: A discourse approach. Towards a revision of Gile’s effort model. Meta, 54(3), 401–416. Yenkimaleki, M., & van Heuven, V. J. (2017). The effect of memory training on consecutive interpreting performance by interpreter trainees: An experimental study. International Journal of Interpretation and Translation, 15(1), 157–172. Yudes, C., Macizo, P., & Bajo, T. (2011). The influence of expertise in simultaneous interpreting on non-verbal executive processes. Frontiers in Psychology, 2, 309. Yudes, C., Macizo, P., & Bajo, T. (2012). Coordinating comprehension and production in simultaneous interpreters: Evidence from the articulatory suppression effect. Bilingualism, 15(2), 329–339.
Notes 1 These tasks are complex versions of a Flanker task. 2 Subjects decide whether the current stimulus matches the stimulus presented n items earlier.
Published online by Cambridge University Press
721
32 A Methodological Synthesis of Working Memory Tasks in L2 Research Jihye Shin and Yuhang Hu 32.1
Introduction
As a limited-capacity system that governs cognitive control, working memory (WM) is responsible for active maintenance and simultaneous processing of information in a range of tasks (Baddeley, 2007). Learning takes place as a learner forges new association in memory while interacting with long-term memory and juggling the storage and processing of information. As such, it has generally been established that WM serves as a central resource for successful second language (L2) learning (e.g., Wen, 2016). Findings from primary studies as well as meta-analyses have shown that L2 learning and performance vary as a function of individual differences in WM capacity (e.g., Linck et al., 2014). Investigations into the role of WM in adult learners’ L2 acquisition and processing are ever increasing and have informed numerous key areas in L2 research (Jackson, 2020). Concomitant with an increasing body of research on WM, methodological procedures for measuring WM have garnered interest as well. Previous discussion in the field of psychology regarding WM assessment (Conway et al., 2005; Friedman & Miyake, 2005) has been a source of inspiration for L2 researchers in the fields of applied linguistics and bilingualism to examine WM measurement practices in their own research contexts (Grundy & Timmer, 2017; Wen et al., 2020). There exists a daunting number of WM tasks, and the lack of standardized measurement and scoring procedures adds to the challenge of accurate assessment of WM capacity. It is therefore critical to inform L2 researchers about inconsistencies in WM measurement practices to help make informed methodological decisions and compare findings across studies. To this end, this chapter sets out to review the past 20 years of WM-L2 research in the field of applied linguistics to describe the various WM task design, scoring methods, and WM reporting practices. By doing so, we aim
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
to add to the ongoing discussion of WM measurement in the context of L2 research and to contribute to promoting methodological rigor. Given the scope of our synthesis, we draw from studies that employed WM tasks in their investigations of the role of WM within the context of L2 research, specifically in L2 acquisition and processing.
32.2
Working Memory Tasks and Measurement
WM capacity is typically measured via simple or complex span tasks. While complex span tasks aim to measure simultaneous storing and processing of information (see more details below), simple span tasks are known to measure short-term memory storage (i.e., temporarily holding a limited amount of information; see Baddeley, 2012, and Cowan, 2008, for details) by asking participants to recall target items. In the forward digit span task, for example, strings of numbers are presented in sets, which can include two or more trials containing two to nine digits. Individuals are then asked to repeat the numbers in order of presentation after each trial. In backward digit span, the procedure is the same except that digits are recalled in reverse order. In terms of scoring, the longest set for which all digits are correctly recalled is often recorded as the final score. If a participant makes a mistake in Set 6, for instance, their set size is recorded as 5. However, this method has been criticized for its all-or-nothing scoring, and partial-credit scoring has been recommended (Conway et al., 2005). Complex span tasks, unlike simple span tasks, are integrated tasks that implement a dual-task paradigm. By interleaving a storage component of the task (e.g., word recall) with a concurrent processing task (e.g., arithmetic operation), complex span tasks tap both storage and processing functions of WM. In complex span tasks, allocation and manipulation of attentional resources occur with the operation of the central executive system of WM (see Baddeley, 2012, for details on WM models). We discuss two complex span tasks that are widely adopted below. The reading span task (Rspan) (Daneman & Carpenter, 1980) has enjoyed its popularity as the most widely used WM task in L2 research, particularly among North American scholars, to the extent that it is “almost synonymous with the WM test” (Wen, 2016, p. 49). Its spoken version, listening span, has been commonly used as well. In the original version of Rspan, individuals (a) read aloud sentences presented while memorizing the final word of the sentences and (b) recall the final words after each set. In a more recent, modified version, a judgment task (e.g., semantic judgment) is included to measure processing capacity of WM more explicitly. This modification has also led to the evolving of the scoring method because reflecting both storage and processing in the final WM score has become possible (see Waters & Caplan, 1996). With the benefit of technology, there are also several studies that recorded participants’ reaction times in judgment tasks
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
723
724
JIHYE SHIN AND YUHANG HU
and incorporated the values into a composite WM score (e.g., Leeser, 2007; Li & Roshan, 2019). However, Rspan has received criticism particularly due to its being a verbal WM task (Juffs & Harrington, 2011). Because of its language-based stimuli, questions arose as to whether or not Rspan results are limited to language-specific WM or if they can effectively represent WM for all cognitive processes (see Wen, 2016). In addition, the language-heavy processing component of Rspan, in particular, has led to controversy in regard to what Rspan measures. Some researchers have adopted a view that Rspan is an effective measure for individual differences in the WM resources (e.g., Waters & Caplan, 1996), whereas others have argued that linguistic experience explains variability in Rspan scores (Farmer et al., 2016; MacDonald & Christiansen, 2002). Another controversial issue of verbal WM tasks in general relates to task language, that is, whether or not L2 sentences should be used as stimuli in the WM task considering that there is a potential confound with L2 proficiency (Linck et al., 2014). With the goal of creating a general-domain, nonverbal complex span task, Turner and Engle (1989) developed the operation span task (Ospan). In their original version, participants (a) engage in verifying equations (e.g., 9/3‒1 = 4) while memorizing a word that appears with each equation and (b) report the words after each set. Though Ospan is different from Rspan in terms of the type of stimuli involved, the fundamental premise remains the same in that it comprises a storage component and a concurrent processing component. However, the demands on language processing knowledge in Ospan are lower than in Rspan, and recent years have seen it become another popular WM task in L2 research.
32.3
Variations in Working Memory Measurement
There exists a myriad of WM tasks, and importantly, it is not uncommon to see variations in design and scoring procedures among WM tasks. Such inconsistencies in assessment of WM can create confusion among researchers and research consumers when it comes to making methodological decisions and interpreting findings. Moreover, the lack of standardized methodological procedures makes it problematic to compare findings across studies and can blur the boundaries of components that make up the construct of WM. Although measurement issues can lead to conceptual discussion and/or stem from conflicts on theoretical stances, we intend to focus the scope of this chapter to a systematic review pertaining to WM measurement practices from a methodological standpoint. Inconsistencies in WM measurement were not explicitly addressed until recently. Conway et al. (2005) were some of the first scholars to outline different ways of scoring WM in complex span tasks and to raise a concern
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
regarding inconsistencies and their consequential ramifications. Similarly, Leeser and Sunderman (2016) examined different scoring methods and analysis procedures used in Rspan and Ospan and showed that study outcomes can vary depending on the task and scoring methods. Further support can be found in Linck et al. (2014), who conducted a meta-analysis on WM and L2 comprehension and production wherein they identified potential covariates that modulated the relationship between WM and L2 outcomes. Their findings showed that WM task type (simple, complex) and content (verbal, nonverbal) can affect study outcomes. More recently, Shin’s (2020) meta-analysis examined the relationship between WM and L2 reading comprehension and revealed that variability in effect sizes was attributed to inconsistencies in methodological features of Rspan, including task language, recall order, presence (or absence) of a processing task, scoring, and reliability reporting (cf. Grundy & Timmer, 2017). However, Shin’s findings were based on samples made up of L2 reading studies that used only Rspan as a WM measure, leaving questions regarding variations of other WM tasks unanswered. In response to the increasing attention to WM measurement practices, we conducted a methodological synthesis to describe the use of common WM tasks in L2 research by coding systematically for various methodological features. Our intention is by no means to criticize previous research but to illustrate the current WM measurement practices in the context of L2 research with a view to promoting methodological rigor in terms of consistency in WM measurement and complete reporting of research practices. In doing so, we pivot toward providing a snapshot of commonly used WM tasks for researchers and consumers to easily differentiate one WM task from another while being aware of the choices and inconsistencies in task design and implementation. We also describe reporting practices in the WM-L2 domain specifically focusing on the reporting of reliability of WM tasks. Our endeavor was guided by the following research questions: RQ1: What are the most common WM tasks used in L2 research from 2001 to 2020, and how has the trend changed over time? RQ2: How do the commonly used WM tasks vary in their task designs and scoring? RQ3: How often is reliability of the WM tasks reported?
32.4
Methods
In this section of the chapter, the methods of the present methodological synthesis are described, including the study retrieval and the coding process.
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
725
726
JIHYE SHIN AND YUHANG HU
32.4.1 Study Retrieval In this methodological synthesis, we retrieved published studies that met our inclusion criteria: studies that used one or more WM tasks targeting adult second language, foreign language, and bilingual adult participants with no known language disabilities or impairments. We consulted a research librarian to find the best search strategy and used the following key words: “second language,” “foreign language,” “bilingual*,” “working memory task,” “working memory test,” and “working memory capacity” in addition to a number of names of WM tasks (e.g., reading span, operation span, n-back) per their recommendation. The first three key words were used as a base search string to which the other key words were appended. Several databases commonly used in L2 research syntheses and metaanalyses were searched (Plonsky & Brown, 2015): Linguistics and Language Behavior Abstracts (LLBA), ERIC, Modern Language Association (MLA), and Google Scholar. We searched studies in the selected databases published from 2001, and the search concluded in April 2020. After eliminating duplicates and studies that focused on children, learning disabilities, and sign languages, 246 studies reporting 431 samples were identified. As we intended to focus on six most common WM tasks, our final selection was narrowed down to 206 studies with 329 unique samples after identifying samples in which at least one of the six WM tasks were used.
32.4.2 Coding Our coding scheme underwent a multistep development process. The initial scheme development was informed by previous literature in two areas: WM (e.g., Leeser & Sunderman, 2016; Linck et al., 2014; Shin, 2020) and methodological synthesis relating to L2 research (e.g., Marsden et al., 2018; Plonsky et al., 2020). We also consulted an expert in research synthesis and meta-analysis to guide our subsequent revision process. The coding scheme was piloted with 15 studies, followed by further refinements. As our purpose was to describe WM measurement and reporting practices in L2 research, our final coding scheme mirrored these two overarching areas and included the following four main categories: (a) study characteristics, (b) sample characteristics, (c) methodological features of WM tasks, and (d) reliability reporting of WM tasks. Three coders each coded a third of the studies independently after a training session with the first author. Once three coders finished coding, the first author double-coded 50 percent of the sample. The intercoder agreement, calculated based on all categories except for study characteristics, was 97 percent between the first coder and the other coders. The points of disagreement were resolved through discussion.
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
32.5
Results
32.5.1
RQ1: What Are the Most Commonly Used WM Tasks in L2 Research from 2001 to 2020, and How Has the Trend Changed over Time? In the initial 431 samples collected from 2001 to 2020, 44 different WM tasks were discovered. As we decided to limit the scope of this chapter to most common WM tasks, we first examined the occurrences of each WM task. Among the 44 tasks, the six most commonly used WM tasks together accounted for 76 percent of the initial sample. They were reading span (Rspan) (n = 109, 25 percent), operation span (Ospan) (n = 62, 14 percent), forward digit span (FDspan) (n = 52, 12 percent), backward digit span (BDspan), (n = 40, 9 percent), listening span (Lspan) (n = 37, 9 percent), and forward nonword span (FNWspan) (n = 29, 7 percent). The results described henceforth are based on 329 unique samples comprising these six subsets. Figure 32.1 displays the change in the use of the six WM tasks from 2001 to 2020 divided into four quarters. Rspan has maintained its position as the most frequently used WM measure in L2 research throughout the two decades and was used more than all the other tasks’ occurrences combined from 2001 to 2010. From the third quarter (2011–2015), there is a more apparent variety of WM tasks in L2 research. All tasks except for Rspan exhibited a noticeable increase in their occurrences compared to the first 10 years; in particular, Ospan showed a substantial rise in numbers in the third quarter, as did FDspan. The overall increase shown in the third quarter (n = 118), which continued into the last quarter (n = 130), is also an indication that far more studies on WM have been produced since 2011. The notable and stable increase in the use of Ospan culminated in the last quarter, where its occurrences became comparable to those of Rpsan in L2 research. BDspan is another task that has increased regularly throughout the years, from zero in the first quarter to 22 occurrences in the last quarter. Uses of FDspan, Lspan, and FNWspan, on the other hand, showed a slight decline in usage in recent years. The six most common WM tasks can be categorized according to task content (verbal, nonverbal) and demands (simple, complex). In terms of task content, Rspan, Lspan, and FNWspan fall into the category of verbal WM tasks which involve maintaining verbal information. Ospan, FDspan, and BDspan, on the other hand, typically belong to the category of nonverbal WM tasks due to the content domain of the stimuli included in the processing component (see Daneman & Merikle, 1996; Linck et al., 2014 for a similar classification). With respect to task demands, Rspan, Ospan, and Lspan are complex span tasks, which involve storage and processing components in the task, whereas FDspan, BDspan, and FNWspan are viewed as simple span tasks, as they require retrieval of items without a processing task. Note that we categorized BDspan as “simple” here, as was done in
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
727
728
JIHYE SHIN AND YUHANG HU
Figure 32.1 Change in the use of WM tasks over time
Figure 32.2 Change in the use of verbal and nonverbal WM tasks over time
Linck et al. (2014), simply due to the absence of a separate processing task. However, BDspan has been considered qualitatively different from FDspan as it necessitates manipulation of attention by asking individuals to recall digits in reverse order (Holdnack, 2019; Juffs & Harrington, 2011). We examined how the use of WM tasks, differentiated on task content and demands, has changed over time. As illustrated in Figure 32.2, L2 research relied heavily on verbal WM tasks from 2001 to 2010. This result goes hand in hand with Rspan having been unquestionably the most widely used WM task during those 10 years (see Figure 32.1). However, nonverbal WM tasks have shown a dramatic increase since 2011, eventually outnumbering their counterparts in the past five years. This overturn may be partly attributable to increased use of Ospan in L2 research. In terms of WM tasks that differ on task demands, Figure 32.3 illustrates how complex span tasks have been preferred over simple span tasks throughout the past 20 years.
32.5.2
RQ2: How Do the Commonly Used WM Tasks Vary in Their Task Design and Scoring? Our primary interest concerns task design and scoring methods of the common WM tasks. To answer RQ2, we focused on five methodological features: task language, storage task, recall order (of items in the storage
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
Figure 32.3 Change in the use of simple and complex span tasks over time
task), processing task, and scoring system. Below, we present the results with attention to each methodological feature with a summary in Table 32.1.
32.5.2.1 Task language Task language is an important methodological feature particularly for verbal WM tasks and when working with a linguistically heterogeneous participant group(s). Although the difference in percentages between L1 and L2 for the three verbal WM tasks (Rspan, Lspan, FNWspan) is not substantial, the overall pattern was that L1 was used more as task language (Rspan: 54 percent, Lspan: 68 percent, FNWspan: 31 percent) than L2 (Rspan: 39 percent, Lspan: 27 percent, FNWspan: 24 percent) (Table 32.1). For Rspan and Lspan, this means that the sentences used as task stimuli (written and auditory) were constructed in L1 more often than L2. For FNWspan, where pseudowords are created following either L1 or L2 phonotactic rules, L1-rule based pseudowords were used more often than L2. After coding dichotomously for L1 and L2, we noticed that several pairs of L1 and L2 belonged to the same studies; that is, several studies administered a WM task in both L1 and L2. There were 19 L1–L2 pairs that were from the same studies in the Rspan subset (22 percent of the primary studies that used Rspan), six pairs in the Lspan subset (19 percent of the primary studies that used Lspan), and three pairs in the FNWspan subset (13 percent of the primary studies that used FNWspan). We observed that the correlations (r) between L1 and L2 WM task results ranged from .36–.74. When it comes to non-verbal WM tasks, task language has not been a source of concern as much as it has been for the verbal WM tasks. However, there are cases in which language is involved in them, though not to the same extent as the verbal tasks. In the Ospan task, word recall was often used to measure participants’ storage capacity, in which case researchers make the decision whether to use L1 or L2 words. The results for Ospan’s task language, therefore, are limited to only when word recall was used as the storage task (n = 26). As shown in Table 32.1, more occurrences of L1 (58 percent) than L2 (23 percent) were identified in the Ospan subset. In FDspan and BDspan, we observed two types of tasks, which differed in terms of task modality: visual and auditory. In the visual version,
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
729
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
Table 32.1 Methodological features of the six most commonly used WM tasks Occurrences Category
Value
Task language
L1 L2 Unknown Word recall Letter recall Digit recall Symbol recall Unknown Same order (forward) Any order Reverse Unknown Semantic judgment Sentence repeat (readaloud in Rspan, listenrepeat in Lspan) Grammaticality judgment Comprehension checks Comprehension encouraged Semantic judgment + readaloud Arithmetic operations Animacy judgment Unknown
Storage task
Recall order
Processing task
Rspan (n = 109)
Ospan (n = 62)
FDspan (n = 52)
BDspan (n = 40)
Lspan (n = 37)
FNWspan (n = 29)
59 (54.13%) 43 (39.45%) 7 (6.42%) 91 (83.49%) 14 (12.84%) 0 0 4 (3.67%) 39 (35.78%) 26 (23.85%) 3 (2.75%) 41 (37.61%) 38 (34.86%) 32 (29.36%)
15 (57.69%)* 6 (23.08%)* 5 (19.32%)* 26 (41.94%) 28 (45.16%) 1 (1.61%) 1 (1.61%) 6 (9.68%) 34 (54.84%) 5 (8.06%) 0 23 (37.10%) 0 0
16 (43.24%)** 10 (27.03%)** 11 (29.73%)** 0 0 52 (100%) 0 0 52 (100%) 0 0 0 0 0
13 (48.15%)*** 8 (29.63%)*** 6 (22.22%)*** 0 0 40 (100%) 0 0 0 0 40 (100%) 0 0 0
25 (67.57%) 10 (27.03%) 2 (5.41) 33 (89.19%) 3 (8.11%) 0 0 1 (2.70%) 16 (43.24%) 9 (24.32%) 0 12 (32.43%) 19 (51.35%) 1 (2.70%)
9 (31.03%) 7 (24.14%) 13 (44.83%) 29 (100%) 0 0 0 0 29 (100%) 0 0 0 0 0
11 (10.09%) 5 (4.59%) 6 (5.50%)
0 0 0
0 0 0
0 0 0
3 (8.11%) 11 (29.73%) 2 (5.41%)
0 0 0
2 (1.83%)
0
0
0
0
0
0 0 15 (13.76%)
56 (90.32%) 0 6 (9.68%)
0 0 0
0 0 0
0 0 1(2.70%)
0 4 (13.79%) 0
Note. * Task language was examined only within the Ospan subset, where word recall was used (n = 26). ** Task language was examined only within the FDspan subset where the auditory version was used (n = 37) *** Task language was examined only within the BDspan subset where the auditory version was used (n = 27)
A Methodological Synthesis
participants are instructed to look at the strings of digits and recall the digits in writing. In the auditory version, participants listen to strings of digits and orally recall them, hence involving language. Therefore, we examined the subsets of FDspan and BDspan that employed the auditory version (FDspan: n = 37; BDspan: n = 27) and found that L1 again was the preferred choice (FDspan L1: 43 percent, L2: 27 percent, BDspan L1: 48 percent, L2: 30 percent).
32.5.2.2 Storage Task We coded for the types of storage task used to examine the extent to which they may vary across the six common WM tasks and the type of stimuli used (e.g., verbal, nonverbal). A total of four types of storage task were found in the sample: word recall, letter recall, digit recall, and symbol recall. Over 80 percent of Rspan and Lspan utilized word recall (see Table 32.1). Most of the word recall tasks required participants to recall sentence final words. However, we found a few exceptions in Rspan tasks, which required recall of sentence-internal words (Hayashi et al., 2016; Juffs, 2005), nouns printed in capitals (Hopps, 2015), and underlined words (Taguchi, 2008). Letter recall was the only other type used in the two subsets (Rspan: 13 percent, Lspan: 8 percent). In Ospan, letter recall was used slightly more (45 percent) than word recall (42 percent). One occurrence of digit recall and symbol recall was also found. In FNWspan, nonword recall was the only type of storage task used. In the two-digit span tasks, digit recall was used 100 percent of the time. 32.5.2.3 Recall Order Participants are asked to recall targeted items in the storage task in the same order as they were presented, in any order, or in reverse order. However, recall order can affect cognitive demands during recall (e.g., Oberauer et al., 2000) and was found to moderate WM-L2 study outcomes (correlation between WM and L2 reading comprehension) (Shin, 2020). We coded for recall order used in Rspan, Ospan, and Lspan to examine any variability (recall order is already specified in FDspan, BDspan, FNWspan). Overall, recall order was usually between the same order (same as item presentation) or any order. In Rspan, participants were most commonly asked to recall targeted items in the same order (36 percent); reporting in any order was less commonly done (24 percent); and there were only a couple of occasions in which recall was done in reverse order (3 percent). Similar patterns were found in Ospan (same order: 55 percent, any order: 8 percent) and Lspan (same order: 43 percent, any order: 24 percent), but no reverse order was found in their respective subsets. Note, however, that almost 38 percent of the Rspan, 37 percent of the Ospan, and 32 percent of the Lspan samples did not provide relevant information (Table 32.1).
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
731
732
JIHYE SHIN AND YUHANG HU
32.5.2.4 Processing task The concurrent processing task, often presented in some form of a judgment test, prevents participants from focusing only on remembering items and adds more cognitive demands. It is the primary factor differentiating complex span tasks from simple span tasks. Various types of processing tasks were observed only in Rspan, Ospan, Lspan, and, surprisingly, FNWspan. In Rspan alone, six different types were identified, demonstrating the highest level of inconsistency: semantic judgment (35 percent), read-aloud (29 percent), grammaticality judgment (10 percent), comprehension with and without actual comprehension questions (5 percent, 5 percent), and semantic judgment and read-aloud used in tandem (2 percent). Lspan most often used semantic judgment (51 percent), followed by comprehension checks (30 percent) and grammaticality judgment (8 percent). Ospan used only arithmetic operations as the processing task. As for FNWspan, we found that 4 out of 29 samples (14 percent) included a processing task, all of which were animacy judgment. Because FNWspan is typically considered a simple span task that includes a storage component only, this modification of FNWspan is rather uncommon. In comparison, no occurrences of a processing task were discovered in FDspan and BDspan. 32.5.2.5 Scoring Methods A we move toward scoring methods, we see that not all scoring procedures employed in the complex span tasks were comparable to those used in the simple span tasks. Therefore, we discuss the scoring methods for these two types of spans separately below. Scoring methods used in complex span tasks. To examine and categorize the scoring methods in the sample pertaining to the complex span tasks, we adapted Leeser and Sunderman’s (2016) categorization scheme, which included five different scoring methods. We also added five more that were identified in our sample (see Table 32.2). We use the same descriptions from Leeser and Sunderman (2016), which where applicable to help readers compare and distinguish between Leeser and Sunderman’s and our categorization. A list of the scoring methods is provided below with descriptions and example studies. 1. 2.
3.
4.
Recall: the total number of items recalled (e.g., Martin et al., 2020) Set size: the highest set size number at which participants recalled all of the items correctly for at least two-thirds of the sets (e.g., Stavrakaki et al., 2012) Set credit: sum of predetermined credits given to correct recall of words in a set (e.g., one point for all items correctly recalled within a set) (e.g., Taguchi, 2008) Recall with 100 percent judgment accuracy: number of items recalled for sentences or math equations judged accurately (e.g., Prior et al., 2017)
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
Table 32.2 Scoring methods used in complex span tasks WM components Storage
Storage & processing accuracy
Storage, processing accuracy, & processing speed Storage & processing speed
Scoring
Rspan (n = 109)
Ospan (n = 62)
Lspan (n = 37)
FNWspan (n = 4)
1. Recall 2. Set size 3. Set credit 4. Recall + 100% judgment accuracy 5. Recall + 80% or 85% judgment accuracy criterion 6. Recall score & judgment score 7. Composite z-score for recall and judgment accuracy 8. Recall + Accuracy + Time 9. Composite z-score for recall, judgment accuracy, and reaction times 10. Recall + Time
49 (44.95%) 6 (5.50%) 3 (2.75%) 12 (11.01%)
23 (37.10%) 3 (4.84%) 0 5 (8.06%)
10 (27.03%) 7 (18.92%) 0 5 (13.51%)
4 (100%)* 0 0 0
5 (4.59%)
9 (14.52%)
2 (5.41%)
0
8 (7.34%)
3 (4.84%)
4 (10.81%)
0
8 (7.34%)
2 (3.23%)
0
0
4 (3.67%) 6 (5.50%)
1 (1.61%) 3 (4.84%)
0 2 (5.41%)
0 0
3 (2.75%)
0
0
0
Multiple scoring methods used Unknown
2 (1.83%)** 3 (2.75%)
2 (3.23%)*** 11 (17.74%)
0 7 (18.92%)
0 0
* Four out of 29 FNWspan added a judgment component into their task thereby transforming into a complex span task. ** Scoring #2 and 6 were used. *** Scoring #1 and 2 were used.
734
JIHYE SHIN AND YUHANG HU
5.
Recall with 80 percent or 85 percent judgment accuracy criterion: participants with a total score below the 80 percent or 85 percent judgment accuracy rate excluded (e.g., Kam et al., 2020) 6. Recall score & judgment score: the number of each correct recall and each correct judgment (often summed or averaged) (e.g., Goo, 2012) 7. Composite z-score for recall and accuracy: an average of the z-scores for the total number for items recalled and correct judgments (e.g., Cho, 2018) 8. Recall + Accuracy + Time: the number of items recalled for sentences or math equations judged correctly within set time limitations (e.g., Sagarra, 2017) 9. Composite z-score for recall, accuracy, and reaction times: an average of the z-scores for the total number for items recalled, correct judgments, and reaction times (e.g., Shin et al., 2019) 10. Recall + Time: the number of items recalled within set time limitations (e.g., Gilabert & Munoz, 2010) Using these categories, we identified 10 different scoring methods utilized in Rspan, eight in Ospan, and six in Lspan. In Table 32.2, WM components accounted for by the scoring methods are also indicated in the far-left column. Below, we elaborate on the results presented in Table 32.2, going from top to bottom. Storage-based scoring methods were adopted more or less 50 percent of the time in complex span tasks even though a processing component is involved in the tasks. As shown in Table 32.2, Scoring #1–3 combined accounted for 53 percent in the Rspan subset and 42 percent and 46 percent in the Ospan and Lspan subsets, respectively. A closer examination revealed that Recall (Scoring #1) was used far more (Rspan: 45 percent, Ospan: 37 percent, Lspan: 27 percent) than the set-oriented calculations (Scoring #2 & 3) (Rspan: 8 percent, Ospan: 5 percent, Lspan: 19 percent). In fact, assigning one point to each correct recall (i.e., partial-credit scoring, Scoring #1) has been recommended over absolute scoring, such as recording the longest length of the set perfectly recalled (Scoring #2) or awarding 1 point only for the perfectly recalled trial (see Conway et al., 2005). We noticed that Scoring #3 is a looser version of absolute scoring because half a point was given to less than perfect recall in a set as long as the performance was above a predetermined cut-off. However, the absolute scoring procedures have been warned against because they show lower reliability and produce a restricted range of scores, thereby blunting variability in scores and affecting the measure’s sensitivity to detect individual differences in WM (Conway et al., 2005; Juffs & Harrington, 2011). Scoring #4–7 are scoring methods that take into account storage and processing accuracy. Overall, about 30 percent of Rspan, Ospan, and Lspan opted for this scoring method to reflect storage and processing accuracy in the final WM score (Rspan: 30 percent, Ospan: 31 percent, Lspan: 30
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
percent). Scoring methods in this category can be further divided into (a) focusing on recall scores while ensuring complete or acceptable judgment accuracy (Scoring #4 & 5) and (b) counting recall and judgment scores separately (Scoring #6 & 7). No sizable difference in the occurrences between (a) and (b) was found for Rspan (16 percent, 15 percent) while (a) was preferred over (b) in Ospan (23 percent, 8 percent) and Lspan (19 percent, 11 percent). Scoring #8 and 9 make use of three sources of scores from computerized complex span tasks: recall, judgment accuracy, and reaction times. Only a few occurrences of this three-component method were observed (Rspan: 9 percent, Ospan: 6 percent, Lspan: 5 percent). The low occurrences here make sense because including reaction times as part of the investigation of a possible trade-off between WM components has recently gained more interest (e.g., see Leeser, 2007; Li et al., 2019). Scoring methods #8 and 9 differ from each other in terms of the ways in which reaction times were incorporated into the final WM score. That is, some samples used the reaction times to impose time limitations and to use an inclusion/exclusion criterion (Scoring #8), while others treated reaction times as a separate component along with recall and judgment accuracy to create a composite score (Scoring #9). Turning to Scoring #10, there was one instance where Rspan incorporated recall scores and reaction times into the final WM score. We also found four instances (2 in Rspan, 2 in Ospan) in which more than one scoring method was used to produce multiple scores. For example, Tolentino and Tokowicz (2014) used Recall (Scoring #1) and Set size (Scoring #2) to generate two WM scores for each participant. In terms of reporting practices, authors were judicious about reporting their scoring methods. Rspan demonstrated a high reporting rate of 97 percent. The reporting rates of Ospan and Lspan were 82 percent and 81 percent, respectively. Scoring methods used in simple span tasks. We identified four scoring methods used in the simple span tasks (FDspan, BDspan, FNWspan). Because simple span tasks have a storage component only, all four scoring methods presented here are storage-based (see Table 32.3). Note that although the FNWspan subset includes 29 samples in total, 25 of them were counted for this part of analysis (see Table 32.3) because 4 samples were treated as part of the complex-span subset (see Table 32.2). 1. Set size 1: Maximum set size with at least one correct trial (e.g., Hayashi, 2019) 2. Set size 2: Maximum set size with at least two correct trials (e.g., Brunfaut & Révész, 2015) 3. Set credit: Sum of predetermined credits given to correct recall of words in a set (e.g., Serafini & Sanz, 2016) 4. Recall: The number of items recalled correctly (e.g., Drozdova et al., 2017)
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
735
736
JIHYE SHIN AND YUHANG HU
Table 32.3 Scoring methods used in simple span tasks. WM component
Scoring method variations
FDspan (n = 52)
BDspan (n = 40)
FNWspan (n = 25)
Storage
1. Set size 1 2. Set size 2 3. Set credit 4. Recall Multiple scoring methods Unknown
21 (40.38%) 5 (9.62%) 10 (19.23%) 1 (1.92%) 2 (3.85%) 13 (25%)
15 (37.5%) 2 (5%) 9 (22.5%) 3 (7.5%) 2 (5%) 9 (22.5%)
4 (16%) 0 2 (8%) 17 (68%) 0 5 (20%)
Note. Set size 1 was used in digit span when there were two trials for each set. Set size 2 in digit span when there were more than two trials for each set
Overall, FDspan and BDspan showed similar trends in scoring WM performance. As shown in Table 32.3, both the FDspan and BDspan tasks tended to rely on the set-oriented calculations (Scoring #1–3) (FDspan: 69 percent, BDspan: 65 percent) rather than Recall (Scoring #4) (FDspan: 2 percent, BDspan: 8 percent). FNWspan, on the other hand, did not follow the digit spans tasks’ general patterns; 69 percent opted for Recall (Scoring #4) and 24 percent used set-oriented calculations (Scoring #1–3). Notice that Scoring #1–3 here corresponds to absolute scoring, which comes with various shortcomings as discussed above. Unlike the complex span tasks, we observed far more use of absolute scoring especially among the two digitspan tasks. We also found four instances in which multiple scoring systems were used; WM performance was evaluated based on (a) the total number of correct trials (same as #3), (b) the total number of digits recalled (same as #4), and (c) the number of correct syllables (Lopéz et al., 2016). Finally, there were fewer reports of scoring systems used in the simple span tasks compared to the complex span tasks. No relevant information was found in 25 percent of FDspan, 23 percent of BDspan, and 20 percent of FNWspan.
32.5.3 RQ3: How Often Is Reliability of the WM Tasks Reported? With our intention to evaluate reporting practices in the WM-L2 research domain, we examined whether and to what extent reliability of the six WM tasks were reported. For complex span tasks, the ideal practice would be calculating and reporting reliability of both recall and judgment tasks separately. For simple span tasks, on the other hand, reliability reporting only for the recall task is expected as it is the only task involved. Overall, only 15 percent of the sample reported reliability, mostly using Cronbach’s alpha (n = 28), followed by test–retest reliability (n = 6), splithalves (n = 4), and Kuder Richardson coefficient (n = 1). Twelve samples did not specify which reliability measures were used. Among the complex span tasks, 17 percent of Rspan, 19 percent of Ospan, and 10 percent of Lspan did not provide any reliability information at all. When examined closely,
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
8 percent of Rspan, 18 percent of Ospan, and 5 percent of Lspan reported reliability of the task without stating whether it was obtained from examining the recall or judgment task. Separate reliabilities of both tasks were reported in pairs in only 9 percent of Rspan, 2 percent of Ospan, and 5 percent of Lspan subsets. (Note that reliabilities of the storage and processing tasks came in pairs from the same samples and therefore counted only once.) Among the simple spans, only five or fewer occurrences of reliability reporting were identified. To gain a better understanding of levels of instrument reliability of WM tasks in our sample, we calculated the mean reliability values from each sample which did report reliability estimates and found that they ranged from .72 to .89, indicating high instrument reliability overall.
32.6
Discussion
This methodological synthesis surveyed the literature spanning 20 years of WM-L2 research to review WM measurement and reporting practices in L2 research. After discovering 44 different WM tasks in our initial sample, we opted to narrow the scope of investigation and focus on the six most-used WM tasks for a more in-depth methodological discussion. In our 329 samples collected from 2001 to 2020, we observed both substantial variability in WM measurement in L2 research and neglected areas in reporting practices. We highlight the main findings below and provide future directions to contribute to making informed decisions for measuring WM and interpreting findings critically. The growing interest in WM in L2 research was well represented in the increased frequencies of the six WM tasks from 2001 to 2020, as they are also a reflection of the expanding literature. In this fast-developing research domain, we have seen more variety of WM measurement introduced and explored since a decade ago along with guidance and influence from cognitive psychology. In this evolving process, however, a few concerning trends have surfaced. First, our findings indicate that the way WM tasks are constructed tends to diverge more than converge within and among the six common tasks. One example occurs in the most frequently used WM task in this synthesis, Rspan. We observed six different types of secondary processing tasks used in Rspan. By comparison, there was only one type of processing task observed in Ospan. The underlying assumption of the processing task is that it places considerable cognitive demands on the participants while they are trying to retrieve target items, thereby necessitating allocation and manipulation of attentional resources (i.e., executive WM). Our concern is regarding the varying degree to which the processing function of WM is taxed by these six processing tasks. For example, while one would have to be engaged in the meaning of a sentence to make a correct semantic
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
737
738
JIHYE SHIN AND YUHANG HU
judgment, reading-aloud may not demand enough cognitive efforts to evoke sufficient dual processing or manipulation of attention (Sagarra, 2017), which can potentially lead to limiting the measure’s sensitivity to detect individual differences in WM capacity. As the very goal of a complex span task is to evoke executive WM, researchers should take caution in designing such a task to serve its purpose effectively and should critically evaluate potential caveats of certain design features seen in previous literature. Looking at all tasks together, we found that the six WM tasks differed in terms of task demands (complex, simple), content (verbal, nonverbal), and modality (auditory, visual). Simple span tasks are designed to tap shortterm memory while complex span tasks tap dual-processing systems in WM. They also function differently in L2 processes and performance (Kormos & Sáfár, 2008). Some argue that FDspan is a measure of rote memory rather than WM because the demands on attention are minimal, especially on healthy adults (e.g., Cullum & Larrabee, 2010). Further support can be found in Linck et al.’s (2014) meta-analysis which showed that WM measured in complex span tasks showed higher correlations with L2 outcomes. However, as seen in our results, both types of WM tasks have been continuously observed in WM-L2 research (Figure 32.3), making it difficult to interpret study findings. Standardization of WM measurement will take time as it will and must involve multidisciplinary efforts and insights. Nonetheless, perhaps we can start small by making it a norm to specifically and explicitly state the operational construct of WM and components (e.g., storage, processing, executive WM) targeted by the WM task used to help research consumers and researchers differentiate between and compare across studies. Employing a verbal WM task in our L2 research domain comes with the possibility of language skills and experience contributing to the variance in WM (e.g., Farmer et al., 2016). Recognition of this risk and the potential threat to internal validity may have contributed to the increased use of nonverbal WM tasks and decreased use of verbal WM tasks in the past five years (Figure 32.2). Another methodological decision related to using verbal stimuli is whether to use L1 or L2 as task language, or both. Meta-analytic findings have in fact reported that task language in a WM task can influence WM-L2 study outcomes (Linck et al., 2014; Shin, 2020). Our results here showed that L1 is more commonly used than L2 in the WM tasks, replicating Linck et al.’s (2014) finding. While choosing to use L1 is likely an indication of the researchers’ cautious decision-making, using L2 does not mean the opposite but a matter of choice made based on research circumstances such as diversity of L1s and research focus (e.g., Malone, 2018). It was also revealed that verbal WM tasks are not the only type to which task language is of relevance. Depending on task modality, L2 knowledge can be called into play even in the tasks that are commonly known as nonverbal WM tasks. For instance, we observed dozens of cases of FDspan and BDspan in which digits were presented and reported in participants’ L2
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
in the auditory (as opposed to visual) version of the tasks (Table 32.1). In addition, we found that L2 words were presented and orally recalled in the word recall task in Ospan. Given the range of differences within and among WM tasks along with the accumulating meta-analytic findings, perhaps there are good grounds to speculate that inconsistent WM measurement contributes to inconclusive findings being observed in WM-L2 studies (Zalbidea, 2017). In addition to the wide variability in the design and implementation of the WM tasks, an equally crucial or even more consequential methodological issue concerns the scoring methods. As an anonymous reviewer pointed out, the variety of scoring methods observed in complex span tasks indicates an increase in the number of studies that include a processing component in their WM task and in scoring. Perhaps, this trend may be attributed to many L2 researchers subscribing to the notion that processing and storage compete for a limited resource, suggesting processing-storage trade-offs. It may have also been influenced by Waters and Caplan’s (1996) inspection of storage, processing accuracy, and reaction times, which resulted in informative findings of the underlying processes of WM. Though some studies have reported a positive correlation between processing accuracy and storage, arguing against the processing-storage trade-offs (e.g., Unsworth et al., 2009), others have observed a negative correlation, showing evidence for the trade-offs (e.g., Leeser, 2007). Thus, we speculate that the variety of scoring procedures may partly be a by-product of L2 researchers’ attempts to further explore the relationship between storage and processing while addressing the mixed findings. Undoubtedly, however, there is very little consistency in the way WM performance was scored within and among the six common WM tasks, which calls construct validity into question. Ten different scoring methods were found among the complex span tasks, with four different configurations of WM components represented in the scores (Table 32.2). Among the simple span tasks, the traditional absolute scoring prevailed. However, Conway et al. (2005) warned against absolute scoring because (a) it can threaten “span reliability across different tasks,” (b) “information on other trials is discarded,” and (c) a narrow range of scores is produced (p. 774). Moreover, consequences of inconsistent scoring methods in complex span tasks, such as Rspan and Ospan, have been empirically demonstrated (Leeser & Sunderman, 2016), further supported by Shin’s (2020) finding that the various scoring methods in Rspan influence the resulting relationship between WM and L2 reading. Our suggestion for future WM-L2 research regarding scoring procedures would be that, following Conway et al.’s (2005) recommendations, L2 researchers (a) move away from the traditional absolute scoring as it reduces the sensitivity of the WM task to subtle individual differences and affects construct validity; (b) when possible, employ more than one WM task to obtain an average of scores to estimate WM capacity; and
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
739
740
JIHYE SHIN AND YUHANG HU
(c) report all the information collected with the task (recall, processing accuracy, reaction times). With the increased attention to WM measurement issues, we encourage researchers to examine if and to what extent different administration and scoring procedures of WM tasks affect our ability to capture individual differences in WM. In addition, thorough reporting of task subcomponent scores (recall, processing accuracy, reaction times) obtained via complex WM tasks will enable future synthetic and meta-analytic studies to collect important and necessary information on methodological issues in the L2-WM research domain. At this juncture with mixed findings on processing-storage trade-offs, we believe that WM research, in general, will benefit from more studies that further explore the relationship between processing and storage. We hasten to note here that some aspects of variations in the scoring methods seem to go hand in hand with the evolving WM measurement. For instance, technology affords researchers to record reaction times, which have been used to explore the potential trade-off between WM components (e.g., Leeser, 2007). Nevertheless, the inconsistent measurement practices illustrated thus far in terms of scoring, in addition to task design and implementation, not only pose a threat to validity of WM study results but also seriously hinder comparisons of the findings across studies in the WML2 domain, which has blossomed in the past decade. Our findings, therefore, point to an urgent need for standardization of WM measurement. The final point of discussion regards reliability reporting practices in the domain. Only 15 percent of our sample reported instrument reliability, falling well below the average reporting rate of instrument reliability estimates in L2 research (21 percent) (see Plonsky, 2013). This finding indicates that the reliability of 87 percent of the WM tasks used in the past 20 years of L2 research remains unknown. Such oversight is alarming especially considering the variety of task design and scoring issues at hand as well as the fact that numerous WM tasks in our sample were researcherdeveloped tasks. Though we do not question that there is a link between WM and L2 learning and processing, the correlations reported in L2 studies do vary and can be moderated by the reliability of the WM task, among other factors (e.g., Shin, 2020). In addition, few studies reported separate reliability estimates of the subcomponents (storage, processing accuracy) of complex span tasks. We therefore encourage L2 researchers to assess their WM task (as well as other instruments used in the study) and report all relevant information for the following reasons. First, researchers will be able to have more confidence in the instrument quality and identify possible factors that bear influence on reliability of their chosen WM task. This will enable researchers to make justifiable and necessary adjustments to their instrument and/ or administration and scoring procedures, which can also lead to more accurate and insightful recommendations for future research, if combined with transparent and complete reporting of such procedures. Second,
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
researchers and consumers will be able to maximize the utility of the reliability-related findings seen across WM-L2 studies to interpret study outcomes and make more informed methodological decisions. Third, reported reliability coefficients will be useful and helpful in future metaanalyses as many meta-analyses use the reporting of reliability as one of the inclusion criteria in the sampling process. As pointed out by Plonsky and Derrick (2016), reliability coefficients can also provide valuable information to future meta-analyses when correcting for the attenuation based on measurement error. We hope that future WM-L2 studies will further address the issues discussed above to promote more consistent WM measurement as well as more complete reporting practice in the domain.
32.7
Conclusion
This chapter surveyed the literature spanning 20 years of WM-L2 research to provide a methodological synthesis of WM measurement in L2 research. There is no denying that previous research in the WM-L2 domain has provided a remarkable amount of information on the underlying cognitive resources in L2 learning and processing. Our goal here was to describe the methodological trends and possible gaps as we reassess the WM measurement practices in the context of L2 research. To achieve this goal, we addressed issues regarding various WM task design, scoring methods, and WM reporting practices, specifically instrument reliability. It is our hope that this methodological synthesis, along with others mentioned throughout this chapter, will serve as an impetus for future work on WM measurement that advances research in the WM-L2 domain and facilitate future endeavors to standardize WM measurement in the years to come. Further, we hope that concerted and interdisciplinary efforts can be made with our sister fields, such as psychology and education, to standardize WM measurement practices. A complete list of studies included in our synthesis/ analysis and our coding scheme will be available on IRIS upon publication.
References Baddeley, A. (2007). Working memory, thought, and action. Oxford University Press. Baddeley, A. (2012). Working memory: Theories, models, and controversies. Annual Review of Psychology, 63, 1–29. Brunfaut, T., & Révész, A. (2015). The role of task and listener characteristics in second language listening. TESOL Quarterly, 49, 141–168. Cho, M. (2018). Task complexity, modality, and working memory in L2 task performance. System, 72, 85–98.
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
741
742
JIHYE SHIN AND YUHANG HU
Conway, A. R. A., Kane, M. J., Bunting, M. F., Hambrick, D. Z., Wilhelm, O., & Engle, R. (2005). Working memory span tasks: A methodological review and user’s guide. Psychonomic Bulletin & Review, 12, 769–786. Cowan, N. (2008). What are the differences between long-term, short-term, and working memory? Progress in brain research, 169, 323–338. Cullum, C. M., & Larrabee, G. J. (2010). WAIS-IV use in neuropsychological assessment. In L. G. Weiss, D. H. Saklofske, D. Coalson, & S. E. Raiford (Eds.), WAIS-IV clinical use and interpretation (pp. 167–188). Academic Press. Daneman, M., & Merikle, P. M. (1996). Working memory and language comprehension: A meta-analysis. Psychonomic Bulletin & Review, 34, 422–433. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. Drozdova, P., Van Hout, R., & Scharenborg, O. (2017). L2 voice recognition: The role of speaker-, listener-, and stimulus-related factors. The Journal of the Acoustical Society of America, 142, 3058–3068. Farmer, T. A., Fine, A. B., Misyak, J. B., & Christiansen, M. H. (2016). Reading span task performance, linguistic experience, and the processing of unexpected syntactic events. Quarterly Journal of Experimental Psychology, 70, 413–433. Friedman, N. P., & Miyake, A. (2005). Comparison of four scoring methods for the reading span test. Behavior Research Methods, 37, 581–590. Gilabert, R., & Muñoz, C. (2010). Differences in attainment and performance in a foreign language: The role of working memory capacity. International Journal of English Studies, 10, 19–42. Goo, J. (2012). Corrective feedback and working memory capacity in interaction-driven L2 learning. Studies in Second Language Acquisition, 34, 445–474. Grundy, J. G., & Timmer, K. (2017). Bilingualism and working memory capacity: A comprehensive meta-analysis. Second Language Research, 33, 325–340. Hayashi, Y. (2019). Investigating effects of working memory training on foreign language development. The Modern Language Journal, 103, 665–685. Hayashi, Y., Kobayashi, T., & Toyoshige, T. (2016). Investigating the relative contributions of computerised working memory training and English language teaching to cognitive and foreign language development. Applied Cognitive Psychology, 30, 196–213. Holdnack, J. A. (2019). The development, expansion, and future of the WAIS-IV as a cornerstone in comprehensive cognitive assessments. In G. Growth-Marnat & A. J. Wright (Eds), Handbook of psychological assessment (pp. 103–139). Academic Press. Hopp, H. (2015). Individual differences in the second language processing of object-subject ambiguities. Applied Psycholinguistics, 36, 129–173.
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
Jackson, D. O. (2020). Working memory and second language development: A complex, dynamic future? Studies in Second Language Learning and Teaching, 10, 89–109. Juffs, A. (2005). The influence of first language on the processing of whmovement in English as a second language. Second Language Research, 21, 121–151. Juffs, A., & Harrington, M. (2011). Aspects of working memory in L2 learning. Language Teaching, 44, 137–166. Kam, E. F., Liu, Y. T., & Tseng, W. T. (2020). Effects of modality preference and working memory capacity on captioned videos in enhancing L2 listening outcomes. ReCALL, 32, 213–230. Kormos, J., & Sáfár, A. (2008). Phonological short-term memory, working memory and foreign language performance in intensive language learning. Bilingualism, 11, 261. Leeser, M. J. (2007). Learner-based factors in L2 reading comprehension and processing grammatical form: Topic familiarity and working memory. Language Learning, 57, 229–270. Leeser, M., & Sunderman, G. (2016). Methodological issues of working memory tasks for L2 processing research. In G. Granena, D. O. Jackson, & Y. Yilmaz (Eds.), Cognitive individual differences in second language processing and acquisition (pp. 89–104). John Benjamins. Li, S., & Roshan, S. (2019). The associations between working memory and the effects of four different types of written corrective feedback. Journal of Second Language Writing, 45, 1–15. Linck, J. A., Osthus, P., Koeth, J. T., & Bunting, M. F. (2014). Working memory and second language comprehension and production: A metaanalysis. Psychonomic Bulletin & Review, 21, 861–883. López, E., Steiner, A. J., Hardy, D. J., IsHak, W. W., & Anderson, W. B. (2016). Discrepancies between bilinguals’ performance on the Spanish and English versions of the WAIS Digit Span task: Cross-cultural implications. Applied Neuropsychology: Adult, 23, 343–352. MacDonald, M. C., & Christiansen, M. H. (2002). Reassessing working memory: Comment on Just and Carpenter (1992) and Waters and Caplan (1996). Psychological Review, 109, 35–54. Malone, J. (2018). Incidental vocabulary learning in SLA. Studies in Second Language Acquisition, 40, 651–675. Marsden, E., Mackey, A., & Plonsky, L. (2016). The IRIS repository: Advancing research practice and methodology. In A. Mackey & E. Marsden (Eds.), Advancing methodology and practice: The IRIS repository of instruments for research into second languages (pp. 1–21). Routledge. Marsden, E., Thompson, S., & Plonsky, L. (2018). A methodological synthesis of self-paced reading in second language research. Applied Psycholinguistics 39, 861–904. Martin, J. D., Shipstead, Z., Harrison, T. L., Redick, T. S., Bunting, M., & Engle, R. W. (2020). The role of maintenance and disengagement in
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
743
744
JIHYE SHIN AND YUHANG HU
predicting reading comprehension and vocabulary learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 46, 140. Oberauer, K., Süå, H.-M., Schulze, R., Wilhelm, O., & Wittmann, W. W. (2000). Working memory capacity: Facets of a cognitive ability construct. Personality & Individual Differences, 29, 1017–1045. Park, H., Nam, K., Lee, Y. (2016). The role of reading span in factual and inferential comprehension and retention in L2 reading. Linguistic Research, 33, 81–106. Plonsky, L. (2013). Study quality in SLA: An assessment of designs, analyses, and reporting practices in quantitative L2 research. Studies in Second Language Acquisition, 35, 655–687. Plonsky, L., & Brown, D. (2015). Domain definition and search techniques in meta-analyses of L2 research (Or why 18 meta-analyses of feedback have different results). Second Language Research, 31, 267–278. Plonsky, L., & Derrick, D. (2016). A meta-analysis of reliability coefficients in second language research. The Modern Language Journal, 100, 538–553. Plonsky, L., Marsden, E., Crowther, D., Gass, S. M., & Spinner, P. (2020). A methodological synthesis and meta-analysis of judgment tasks in second language research. Second Language Research, 36, 583–621. Prior, A., Degani, T., Awawdy, S., Yassin, R., & Korem, N. (2017). Is susceptibility to cross-language interference domain specific? Cognition, 165, 10–25. Sagarra, N. (2017). Longitudinal effects of working memory on L2 grammar and reading abilities. Second Language Research, 33, 341–363. Serafini, E. J., & Sanz, C. (2016). Evidence for the decreasing impact of cognitive ability on second language development as proficiency increases. Studies in Second Language Acquisition, 38, 607–646. Shin, J. (2020). A meta-analysis of the relationship between working memory and second language reading comprehension: Does task type matter? Applied Psycholinguistics, 41, 873–900. Shin, J., Dronjic, V., & Park, B. (2019). The interplay between working memory and background knowledge in L2 reading. TESOL Quarterly, 53, 320–347. Stavrakaki, S., Megari, K., Kosmidis, M. H., Apostolidou, M., & Takou, E. (2012). Working memory and verbal fluency in simultaneous interpreters. Journal of Clinical and Experimental Neuropsychology, 34, 624–633. Taguchi, N. (2008). The effect of working memory, semantic access, and listening abilities on the comprehension of conversational implicatures in L2 English. Pragmatics & Cognition, 16, 517–539. Tolentino, L. C., & Tokowicz, N. (2014). Cross-language similarity modulates effectiveness of second language grammar instruction. Language Learning, 64, 279–309. Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task dependent? Journal of Memory and Language, 28, 127–154.
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
A Methodological Synthesis
Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R. W. (2005). An automated version of the operation span task. Behavior Research Methods, 37, 498–505. Unsworth, N., Redick, T. S., Heitz, R. P., Broadway, J. M., & Engle, R. W. (2009). Complex working memory span tasks and higher-order cognition: A latent-variable analysis of the relationship between processing and storage. Memory, 17, 635–654. Waters, G., & Caplan, D. (1996). The measurement of verbal working memory capacity and its relation to reading comprehension. Quarterly Journal of Experimental Psychology, 49, 51–79. Wechsler, D. (1997). WAIS-III, Wechsler Adult Intelligence Scale: Administration and scoring manual. The Psychological Corporation. Wen, Z. E. (2016). Working memory and second language learning: Towards an integrated approach. Multilingual Matters. Wen, Z. E., Juffs, A., & Winke, P. (2020). Measuring working memory. In P. Winke & T. Brunfaut (Eds.), The Routledge handbook of second language acquisition and language testing (pp. 167–176). Routledge. Woodcock, R. W., McGrew, K. S., & Mather, N. (2007). Woodcock-Johnson III Test of Cognitive Abilities. Riverside Publishing. Zalbidea, J. (2017). “One task fits all?” The roles of task complexity, modality, and working memory capacity in L2 performance. The Modern Language Journal, 101, 335–352.
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
745
https://doi.org/10.1017/9781108955638.038 Published online by Cambridge University Press
Part VI
Language Disorders, Interventions, and Instruction
https://doi.org/10.1017/9781108955638.039 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.039 Published online by Cambridge University Press
33 Specific Learning Disorders as a Working Memory Deficit H. Lee Swanson 33.1
Introduction
The chapter provides a review as well as an update on a model outlined by Swanson and Siegel (2001a, b), which suggested that specific processes related to the phonological and executive system of working memory (WM) underlie specific learning disorders in reading and/or math. We find (e.g., Swanson, 1992, 1993c; Swanson et al., 2020; Swanson & Fung, 2016; Swanson & Jerman, 2007), as do others, that children with average intelligence but who suffer specific learning disorders in reading (referred to as reading disabilities, RD) and/or math (referred to as math disabilities, MD) experience considerable difficulties on WM tasks (e.g., Attout & Majerus, 2015; Brandenburg et al., 2015; De Weerdt et al., 2013; Passolunghi & Siegel, 2004; Peng et al., 2012; Wang & Gathercole, 2013). Within a multiple-component model of WM and when compared to average achievers, our model views children with RD and/or MD as manifesting deficits related to the phonological loop, a component of WM that specializes in the retention of speech-based information. However, in situations that place high demands on processing, such as in the areas of reading comprehension and math word problem solving, individuals with RD and/or MD have deficits related to the executive component of WM, primarily related to controlled attentional processes (e.g., updating and/or maintaining task-relevant information in the face of distraction or interference) that can operate independently of their problems in the phonological system. In this chapter, we review work that provides an empirical This chapter draws from previous discussions in the final report to U.S. Dept. of Education, Institute of Education Sciences (R305H020055; R324A09002), Swanson (2011, 2017, 2020), Swanson and Zheng (2013) and Swanson and Siegel (2001a, b) and the reader is referred to those sources for additional information. Furthermore, this chapter was written as the author served as a research professor at UNM and supported by a National Science Foundation grant (DRL award number 1660828). This chapter does not necessarily reflect the views of the National Science Foundation or the Institute of Education Sciences.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
750
H. LEE SWANSON
foundation for the view that specific learning disorders in reading and/or math reflect a fundamental deficit in WM. Before discussing the research linking learning disorders in reading and/ or math to WM, we review an operational definition of specific learning disorders and WM. We next review syntheses of the literature linking WM to RD and/or MD and then review work from our lab linking specific learning disorders in reading and/or math to specific components of WM. We then review common criticisms of the model to children with RD and/ or MD.
33.2
Definition of Terms
33.2.1 Specific Learning Disorders Our earlier work used the term “specific learning disabilities.” Because this chapter has an international audience, however, we will use the term “specific learning disorders.” This is because The American Psychiatric Association (2013) Diagnostic and Statistical Manual of Mental Disorders (DSM5) does not use the term learning disabilities or such terms as “reading disabilities,” “dyslexia,” “math disabilities,” or “dyscalculia.” The DSM-5 (2013) uses the term “specific learning disorder” rather than “learning disabilities.” A specific learning disorder in reading and/or math, as reflecting a neurodevelopmental disorder, is considered to have a biological origin. In general, the concept of specific learning disorders rests on two assumptions: (a) academic difficulties are not due to inadequate opportunity to learn, general intelligence, physical or emotional disorders, but to basic disorders in specific psychological processes and (b) these specific processing deficits are a reflection of neurological, constitutional, and/or biological factors. Thus, to assess specific learning disorders at the cognitive level, systematic efforts are made to detect: (a) normal psychometric intelligence, (b) consistently below-normal achievement in a specific academic skill across grade levels, (c) below-normal performance in specific cognitive processes (i.e., phonological awareness, WM), (d) that systematic opportunity to learn (documentation that optimal instruction has been presented but deficits in isolated processes remain) has occurred, and (e) that processing deficits are not directly caused by environmental factors (e.g., poor instruction) or contingencies (e.g., Social Economic Status, child-rearing). We operationally defined specific learning disorders as those children and adults who have general intelligence scores on standardized tests above 85 and who have reading scores and/or math scores below the 25th percentile (90 standard score) on a standardized norm-referenced reading/and or mathematics achievement measure. Our more recent classification studies that utilize latent class and latent transition analysis suggest that a discrete class of children with RD and MD emerges that is consistent with our cutoff points (e.g., Swanson et al., 2016, 2018).
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
We have also identified children with specific learning disorders in reading and math who are English language learners with Spanish as a first language (Swanson et al., 2016, 2019a, b, 2020). Children with specific learning disorders exhibit low reading and/or math achievement on normative tests in each language system (in our case normative tests in both English and Spanish). The conundrum, of course, in our studies, is that children yielded clear variations in English and Spanish language proficiency. The literature varies in methodology for operationalizing variations in bilingualism (e.g. Peña et al., 2016; Rosselli et al., 2016), with some studies suggesting a focus on proficiency within each language system that in turn yields a total score and/or focusing on conceptual proficiency (tests that allow for responses in either language that yield the highest score in the preferred language [see Peña et al., 2016]; for a comprehensive review). At present, conceptual scoring has been used in our studies because it combines children’s vocabulary knowledge across languages (Swanson et al., 2019a, b, 2020).
33.2.1.1 Working Memory Working memory consists of a limited capacity system of temporary stores and functions related to the preservation of information while simultaneously processing other information as well as maintaining attention control related to these functions (e.g., Baddeley, 2012; Cowan, 2014; Unsworth & Engle, 2007; see Swanson, 1999, 2017, for review). Cowan (2014) defines WM “as the small amount of information that can be held in the mind and used in the execution of cognitive tasks, in contrast with long-term memory, which is the vast amount of information saved in one’s life” (p. 197). Working memory or complex span tasks share the same processes (e.g., rehearsal, updating, controlled search) as short-term memory (STM) or simple span tasks. However, simple tasks (e.g., recalling words or digits in the order of presentation) have a greater reliance on phonological processes than WM or complex span tasks (e.g., recalling words or digits in the context of interference/distraction) (see Unsworth & Engle, 2007, pp. 1045–1046, for a review). Whereas WM tasks assess an individual’s ability to maintain task-relevant information in an active state and to regulate controlled processing. Individuals performing WM tasks must remember some task elements and ignore, or inhibit, other elements as they complete task-relevant operations (e.g., Miyake et al., 2000). In contrast, STM tasks typically involve situations that do not vary from initial encoding (e.g., Unsworth & Engle, 2007). In elaborating on the distinction between STM and WM, Cowan (1997) emphasized the role of attentional processes. WM is depicted as a subset of items of information stored in STM that are in turn submitted to limited attentional control processing (see also Engle et al., 1999). This assumes that when the contents of STM are separated from WM, what is left of WM is some form of controlled attention or processing related to the central executive system (also referred to as the
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
751
752
H. LEE SWANSON
central executive component of WM). As will be reviewed, several of our studies finding that children with RD and/or MD suffer difficulties on both types of tasks cross various age ranges (e.g. Swanson et al., 2006, 2008, 2019; Swanson & Ashbaker, 2000; Swanson & Fung, 2016). No doubt the distinction between STM and WM is less clear in younger children (e.g., Hutton & Towse, 2001), but becomes more distinct as children approach the upper- middle grades (Gathercole et al., 2004). Our regression modeling finds that both types of tasks (STM and WM) uniquely contribute to reading and math performance among children at risk for RD and/or MD (Swanson et al., 2008; Swanson & Fung, 2016). The independent contribution of these two memory systems (STM, WM) to reading and math will be further elaborated upon later in the chapter.
33.3
Theoretical Foundation
By far the most utilized framework of understanding the role of WM as it applies to specific learning disorders is Baddeley’s multiple-components model (Baddeley, 2012; Baddeley & Logie, 1999). In this model, WM consists of three components: visual-spatial sketchpad, phonological loop, and central executive. The visual-spatial sketchpad is for the temporary storage of visual and spatial information, and it is important for the manipulation of mental images, such as symbols and shapes. The phonological loop is for the temporary storage of verbal information, and it is important for the temporary storage of text and verbal information, such as the story within a text reading. The central executive coordinates activities between the two subsystems (i.e., visual-spatial sketchpad and phonological loop) and increases the amount of information that can be stored in the two subsystems. This model has been revised to include an episodic buffer (Baddeley, 2012), but support for the tripartite model has been found across various age groups of children (e.g., Alloway et al., 2004; Gathercole et al., 2004; Gray et al., 2017). The tripartite model has also been supported in studies that include samples of children with RD and/or MD (e.g., Swanson, 2008; Swanson et al., 2019; Swanson & Fung, 2016; Swanson & Jerman, 2007).
33.4
Research on Working Memory
33.4.1 Synthesis of Literature Before selectively reviewing some of our work, a quantitative overview of the literature on WM is necessary to provide a context for our findings. We briefly summarize our quantitative syntheses of the published literature on learning disorders as it applies to reading (RD), math(MD), and memory (e.g., Johnson et al., 2010; Kudo et al., 2015; Peng et al., 2018; Swanson &
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
Jerman, 2006). A common metric utilized in these meta-analyses is referred to as effect size (ES) and reported as a d-index. To make d’s more interpretable, statisticians have adopted Cohen’s (1988) system for classifying d’s in terms of their size (i.e., .00 to .19 is described as trivial; .20 to .49 , small; .50 to .79 , moderate; .80 or higher, large).
33.4.1.1 Reading Disorders Our recent synthesis has found a strong relationship between WM and reading in average achievers (Peng et al., 2018). This meta-analysis investigated the relationship between reading and WM among average readers Overall the relation between reading and WM was moderate (r = .29) and was significantly influenced by grade level. The importance of this finding was that WM made unique contributions to reading performance even when decoding and vocabulary were controlled for. Further meta-analyses by Peng and his colleagues have’ synthesized research on WM among children with reading difficulties (RD), children with mathematics difficulties (MD), and children with reading and mathematics difficulties (RD/MD). As expected, typically developing children outperformed all learning difficulty groups on WM tasks. MD children and RD children showed comparable verbal WM deficits, whereas children with combined RD/MD showed the most severe WM deficits. Although there was no clear demarcation between STM and WM tasks, the results are consistent with an earlier analysis (Swanson et al., 2009) that found difficulties on both STM and WM tasks for children with RD. This earlier meta-analysis provided a more differentiated analysis of performance on STM (simple span tasks) and WM (complex span) tasks and found that children with RD vary from their average reading counterparts on STM and WM tasks. Effect sizes (ESs) across a broad age, and reading and IQ range, yielded a mean ES (RD vs. average readers) across memory studies of –.89 (SD = 1.03) in favor of their chronologically matched average reading counterparts. The moderate range for STM measures consisted of 255 ESs (M = –.61, 95% confidence range of –.65 to –.58) and 320 ESs were in the moderate range for WM measures (M = –.67, 95% confidence range of –.68 to –.64). The synthesis indicated that children with RD were distinctively disadvantaged compared to average readers on (a) STM measures requiring the recall of phonemes and digit sequences and (b) WM measures requiring the simultaneous processing and storage of digits and words within sentence sequences and final words from unrelated sentences. The results of this synthesis also indicated, via a meta-regression analysis, that no significant moderating effects related to age, IQ, or reading level emerged when predicting memory effect sizes. In general, the findings indicated STM and WM differences between ability groups persisted across age, suggesting that both a phonological (STM) and components of the executive system (WM) underlie RD.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
753
754
H. LEE SWANSON
33.4.1.2 Math Disorders Of interest is whether MD and RD share the same and/or different WM problems. Peng and Fuchs’s (2016) meta-analysis indicated that both groups suffer WM difficulties compared to average achievers, but the RD group tends to have deficits primarily on verbal WM tasks, whereas the MD group experience deficits on verbal and numerical WM tasks. To address this question directly as it applies to STM and WM tasks, Swanson and Jerman (2006) performed an earlier quantitative synthesis of the literature comparing the cognitive functioning of children with MD with (1) average achieving children, (2) children with RD, and (3) children with comorbid disorders (RD + MD) on this issue. When focusing on children with MD, approximately 194 effect sizes compared children with MD with average achievers (M ES = –.52, SE = .01), 58 effect sizes compared MD and children with RD (M = –.10, SE = .03), and 102 effect sizes compared children with MD to children with MD + RD (M = .26, SE = .02) on various cognitive measures. Overall, average achievers outperformed children with MD-on measures of verbal problem solving (M = –.58), naming speed (M = –.70), STM (M = –.71), verbal WM (M = –.70), visual-spatial WM (M = –.63) and long-term memory (LTM, M = –.72). These moderate to high ESs were consistent across age and severity of the MD. The results indicated children with MD could only be differentiated on measures naming speed (M = –.23) and visual WM (M = –.30). More importantly, both children with MD and RD share similar difficulties on measures of STM (M = .16) and WM (M = .07) tasks. The results further indicated that children with MD outperformed children with comorbid disorders (MD + RD) on measures of literacy (M = .75), visual-spatial problem solving (M = .51), LTM (M = .44), STM for words (M = .71), and verbal WM (M = .30). Interestingly, as also found by also by Peng and Fuchs (2016), children with MD could only be differentiated (although weakly) from children with RD on measures of naming speed and visual-spatial WM. Both children with RD or MD, outperformed the comorbid group on STM and WM measures. More important, hierarchical linear modeling of the data showed that the magnitude of effect sizes in overall cognitive functioning (criterion measure) between MD and average achievers was primarily related to WM deficits when the effects of all other variables (e.g., age, IQ, reading level, other domain categories) were partialed out in the analysis.
33.4.2 Shared Memory Difficulties? The takeaway message from the above syntheses of the literature is that children with RD and/or MD share common WM and STM problems. Thus, the question emerges: “how is it that both groups can share STM and WM deficits, but manifest themselves as reading or math deficits”? Recall that a neurological inefficiency is assumed to underlie RD and/MD the new DSM-5
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
(2013), however finding the commonality is still under investigation. Because RD and/or MD as assumed to have a neurological base to learning problems (DSM-V), it is important to note there are correlates in the neuropsychological literature that complement the tripartite WM structure of Baddeley and Logie’s (1999) model, suggesting that some functional independence exists among the systems (e.g., Acheson et al., 2011; Bledowski et al., 2010; Chein, & Fiez, 2010; Nee et al., 2013; Smith & Jonides, 1997; 1999). Neuropsychological evidence suggests that children with RD and/or MD experience difficulties related to these structures (e.g., Beneventi et al., 2009, 2010; Rosenberg-Lee et al., 2009). Further, recent meta-analyses of functional magnetic resonance imaging (fMRI) study of children with RD and/or MD may provide some light on this issue. Neuroimaging studies have indicated that math and reading processes tap several common areas of the brain (e.g., Evans et al., 2016) and the manifestations of these deficits reflect hypoactivation (under activation) of these common areas of the brain. For example, Kaufmann et al.’s (2011) meta-analysis of fMRI studies directed at children with MD (dyscalculia) showed observable hypoactivation in number processing in the prefrontal and occipital cortex when compared to children without MD. Likewise, Richlan et al. (2011) meta-analysis of fMRI studies that included children with dyslexia (RD) also found a hypoactivation in the anterior portion of the left occipito-temporal cortex for dyslexic children. Richlan’s (2012) reanalysis of the data also found little support for the assumption that standard neural anatomical models of developmental dyslexia are localized to problems primarily related to phonological decoding deficits in the left temporo-parietal regions. Rather, Richlan found evidence that points to dysfunction in dyslexics in the left hemisphere reflecting a larger reading network that included the “ hypoactivation” of the occipital-temporal, inferior frontal and inferior parietal regions. In general, the sharing of common problems in STM and WM suggests that “hypoactivation” (underactivation) may occur in common areas of the brain for children with RD and/or MD. Given this overview of a multi-component view of WM, we will now briefly review some of our earlier findings linking RD and/or MD to specific components of WM. The links between various components of WM to various achievement measures have been reviewed previously (e.g., Swanson, 2020; Swanson & Alloway, 2012; Swanson & Zheng, 2013) and because of space limitations are not reviewed here. Instead, some selective studies are reviewed that have included children with specific learning disorders in reading and/or math under each WM component.
33.4.3 Executive Component of WM A crucial activity of the central executive of WM as it applies to children with RD and/or MD is controlled attention. The involvement of controlled attention difficulties in children with RD and/or MD is inferred from three
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
755
756
H. LEE SWANSON
outcomes: (1) depressed performance across both verbal and visual-spatial tasks that require concurrent storage and processing, (2) poor performance on complex divided attention tasks (attention to relevant and irrelevant information), and (3) poor concurrent monitoring of attention on high demand tasks.
33.4.3.1 Combined processing and storage demands The majority of our recent studies (Swanson, 2012; 2017; Swanson et al., 2008; Swanson & Jerman, 2007) on executive processing have included tasks that follow the format of Daneman and Carpenter’s Sentence Span measure, a task strongly related to achievement measures (see Swanson & Alloway, 2012, for a review). For example, three unrelated sentences are presented and the child is asked to recall the last word of each sentence. Before the recall of last words, however, the child is asked a process question. Our WM studies typically engage participants in at least two activities after initial encoding: (1) a response to a question or questions about the material or related material to be retrieved, and (2) a response to recall item information that increases in set size. These laboratory tasks have an analog to everyday learning (Uppal & Swanson, 2016). For example, holding a person’s address in mind while listening to instructions about how to get there, listening to a sequence of events in a story while trying to understand what the story means, locating a sequence of landmarks on a map while determining the correct route, listening to specific word features among several in one ear and suppressing the same features in the other ear, and so on. All these tasks have some aspects of interference (a competing memory trace that draws away from the targeted memory trace) and monitoring (decisions related to the allocation of attention to the stimulus that is under consideration together with the active consideration of several other stimuli whose current status is essential for completing the task). An example of the difficulties experienced by children and adults with RD on these complex span tasks can be found in an earlier cross-sectional study (Swanson, 2003). The study compared four age groups (7, 10, 13, 20) on phonological, semantic and visual-spatial WM measures administered under a variety of conditions (also see Swanson, 1992, 1996): initial (no probes or cues), gain (cues that bring performance to an asymptotic level), and maintenance conditions (asymptotic conditions without cues). The general findings of the Swanson (2003) study were that both participants with and with RD showed continuous growth in verbal and visual-spatial WM. The results suggested that initial constraints in WM can be modified under gain conditions. However, the study provided little evidence that participants with RD catch up in WM skills with average readers as they age, suggesting that a deficit model rather than a developmental lag model best captured such readers’ age-related performance.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
In terms of MD, a longitudinal study (Swanson et al., 2008) examined the influence of growth in WM on growth in mathematical problem solution accuracy in elementary school children with and without MD across three years. Their results showed that children identified with MD in wave 1 showed less growth rate and lower levels of performance on WM measures than children not at risk at wave 3 (three years later). More importantly, the results indicated that WM contributed unique variance to problem-solving beyond what phonological processes (e.g., phonological knowledge), fluid intelligence, reading skill, inhibition, and processing speed contributed. Thus, there was clear evidence that WM contributed important variance to math performance three years later beyond processes related to speed, phonological knowledge, fluid intelligence, and reading skill. Implications?
33.4.3.1.1 Complex divided attention tasks Our earlier research showed that children with RD can be distinguished from average achievers in how they handled attentional demands (Swanson, 1989). For example, Swanson (1989) compared central and incidental recall on WM tasks between children with RD, children with low verbal IQ, average achievers and intellectually gifted children. In general, the results suggested that children with low verbal IQ and children with RD recalled less information as a function of encoding conditions than higher ability groups on both central and incidental memory tasks. However, the results also indicated that lower ability groups differed from higher ability groups in executive processing (how they shared, discriminated, and selectively allocated resources between the central and secondary recall tasks). Children with low verbal IQ compensated for their executive processing deficiencies by maintaining a constant resource supply (recall was comparable between central and secondary tasks); whereas children with RD failed to effectively prioritize items (recall fluctuated between central and secondary tasks) despite having a higher verbal IQ. Take-home? Implications? 33.4.3.2 General Monitoring Difficulties In another one of our earlier experiments (Swanson, 1993b, Exp. 1), a concurrent memory task, adapted from Baddeley (Baddeley et al., 1984) was administered to RD and average readers. The task required subjects to remember digit strings (e.g., 9, 4, 1, 7, 5, 2) while they concurrently sorted blank cards, cards with pictures of nonverbal shapes, or cards with pictures of items that fit into semantic categories (e.g., vehicles–car, bus, truck; clothing–dress, socks, belt). Demands on the central executive capacity system were manipulated through the level of difficulty (3- vs. 6-digit strings) and type of sorting required (e.g., nonverbal shapes, semantic categories, blank cards). Sorting activities that placed demands on the
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
757
758
H. LEE SWANSON
verbal storage (phonological system) included the categorization of pictures into semantic categories, whereas sorting activities that made demands on the visual store (i.e., visual-spatial sketch pad) included discrimination among complex nonverbal shapes. Baddeley et al. (1984) found that in such activities the main task difficulty (sorting) interacted with concurrent memory load, but only with a memory load of 6 digits. Performance for the 6-digit memory load condition placed processing demands on the central executive, thereby interfering with the main task. Swanson’s (1993b) results indicated a clear effect on memory load. The results showed that children with RD can perform comparably to their chronological age (CA)matched counterparts on verbal and visual-spatial sorting conditions that included 3-digit strings (low demands) and that only when the coordination of tasks becomes more difficult (6-digit strings) do ability group differences emerge. More importantly, the results for the high memory load condition (6-digit strings) showed that children with RD were inferior to the CA-matched readers (and reading matched controls for ordered recall) in their ability to recall digits during both verbal and nonverbal sorting. Because recall performance for children with RD was not restricted to a particular storage system (i.e., verbal storage), compared with the performance of CA-matched average readers, one can infer that processes other than a language-specific system accounted for the results. Implications?
33.4.4 Section Summary We have selectively reviewed some of our studies finding that children with RD and/or MD manifest WM deficits that, depending on the task and information processing demand, reflect problems related to an executive processing system. Although our research suggests that difficulties related to the executive component of WM underlie RD and/or MD, we also want to emphasize that several activities that involve executive processing are very much intact for children with RD/MD. Depending on the task, of course, some of these intact executive processes relate to planning (Swanson et al., 2010). For example, although planning (such as mapping out a sequence of moves to solve a problem) is considered a component of the executive system (e.g., see Miyake et al., 2000, p. 91), we have not found overall solution differences between children with RD (and/or MD) and average achieving students on problem-solving tasks (see Swanson, 1988, 1993a). Thus, only a few activities related to executive processing underlie RD and/or MD.
33.5
Phonological Loop
The previous section highlighted the role of the executive system of WM in RD and /or MD children. However, previous regression analyses have found that both STM and WM contribute unique variance to academic performance
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
in children with RD and/or MD (e.g. Swanson et al., 2008; Swanson & Ashbaker, 2000). Thus, the role of STM, referred to as the phonological loop is reviewed. The phonological loop is specialized for the retention of verbal information over short periods of time. A cognitive process consistently implicated in RD is phonological awareness (e.g., Melby-Lervåg et al., 2012). Phonological awareness is “the ability to attend explicitly to the phonological structure of spoken words” (Scarborough, 1998, p. 95). This difficulty in forming and accessing phonological representations impairs their ability to retrieve verbal information from STM. A substantial number of studies support the notion that children with RD experience deficits in phonological processing (e.g., see the synthesis of Kudo et al., 2015), such as forming or accessing phonological representations of information. These findings build on research that has shown that the manifestations of this phonological deficit are poor word recognition, poor performance on phonological awareness tasks, computation, and slow naming speed and impaired verbal STM. When attributing the component of WM to mathematical performance, several studies have attributed arithmetic skill to the phonological loop (e.g., Logie et al., 1994) or to a combination of both the phonological and executive systems (e.g., Furst & Hitch, 2000). The research to date indicates that some children who are less proficient in math appear to rehearse less and perform more poorly on tasks requiring short-term retention of order information than children proficient in math, especially on measures of digits (Geary, 2011, 2013; Geary et al., 2012; McLean & Hitch, 1999), suggesting inefficient utilization of the phonological rehearsal process. Swanson and Kim (2007) found that both short-term memory (STM) and WM tasks each uniquely predicted individual differences in mathematical performance. WM was independent of the contribution of STM and naming speed in predicting children’s mathematical performance. The results were interpreted as support for the notion that both the central executive system (controlled attention) and storage system of WM (i.e., phonological loop) predict children’s math performance. A key finding in several of our studies is that the contribution of STM (phonological loop) and WM (storage + controlled attention) can contribute unique variance to achievement measures and therefore can operate independently of one another. Swanson and Ashbaker (2000) directly tested whether the operations related to STM and WM operated independently of one another. In this study, they compared children with and without RD and younger reading level-matched children on a battery of WM and STM tests to assess executive and phonological processing, respectively. The study found that reading-group differences were pervasive across verbal and visual-spatial WM tasks, even when the influence of verbal STM was statistically removed, suggesting that reading-group differences are domain-general. These results suggested that the executive component of WM predicted reading problems independent of problems related to the phonological component of WM.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
759
760
H. LEE SWANSON
33.5.1 Visual-Spatial Sketch Pad The visual-spatial sketchpad is specialized for the processing and storage of visual material, spatial material, or both, and for linguistic information that can be recoded into imaginable forms (see Baddeley, 2012, for a review). Our work linking RD to visual-spatial memory deficits is mixed. For example, when visual-spatial WM (combined storage and processing demands) performance is considered, some studies find that visual-spatial WM in students with RD is intact when compared with their same-age counterparts (e.g., Swanson et al., 1996, Exp. 1), whereas others suggest problems in various visual-spatial tasks (e.g., Mammarella et al., 2017, 2018; Swanson et al., 1996, Exp. 2). Most studies suggest, however, that depending on the type of academic disability, greater problems in performance are more likely to occur on verbal than visual-spatial WM tasks for children with RD and/or MD (Peng et al., 2018; Swanson & Jerman, 2007). Likewise, our meta-analysis synthesizing research on the cognitive studies of MD (Kudo et al., 2015; Swanson & Jerman, 2006) suggests that memory deficits are more apparent in the verbal than visual-spatial domain. In general, our studies suggest that that the visual-spatial system of RD children is generally intact, but when excessive demands are placed on the executive system, their visual-spatial performance is depressed compared with chronological age-matched average achievers (Swanson, 2000).
33.5.2 Section Summary When performance demands on various tasks directly tax the WM capacity of individuals with RD and/or MD, deficiencies related to the accessing of speech-based information and/or the monitoring of attentional processes emerge. Thus, these two areas of deficiencies are related to components of WM referred to in Baddeley’s model (Baddeley & Logie, 1999) as the phonological loop and the executive system. Individuals with RD and/or MD do not exhibit deficits in all aspects of the phonological loop (e.g., they have relatively normal abilities in producing spontaneous speech and have few difficulties in oral language comprehension) or the executive system (e.g., they have relatively normal abilities in planning). Those aspects of the phonological system that appear particularly problematic for individuals with RD and/or MD relate to the accurate and speedy access of speech codes in STM, and those aspects of the executive system that appear faulty are related to the concurrent monitoring of processing and storage demands and the suppression of conflicting (e.g., irrelevant) information. Taken together, we place our findings with a multiple component model of WM. Two components are viewed as critical. The model views children with RD and/or MD as manifesting deficits related to the phonological loop, a component of WM that specializes in the retention of speech-based information. However, in situations that place high demands on processing, individuals with RD and/or MD have deficits related to the executive
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
component of WM, primarily related to controlled attentional processes (e.g., updating and/or maintaining task-relevant information in the face of distraction or interference) that can operate independently of their problems in the phonological system.
33.6
Alternative Explanations
No doubt, the above theoretical model seems counterintuitive when applied to specific learning disorders because WM, especially the executive component, has been associated with several other domain-general processes, such as intelligence and attention monitoring. We will briefly consider four alternative explanations as to why executive processing deficits in WM may underlie RD and/or MD.
33.6.1 Problems in WM Are Merely a Function of Intelligence No doubt, it seems counterintuitive that children with average intelligence can experience deficits related to tasks that tap the executive component of WM. This is because current studies show that the executive component of WM (updating, suppression of competing traces) is strongly correlated with fluid intelligence (e.g., Ackerman et al., 2002; Engle et al., 1999; Kyllonen & Christal, 1990; Swanson, 2012). Thus, it is rather unexpected that children with average intelligence but with RD and/or MD will have difficulties on WM tasks. This paradox can be addressed in at least two ways. First, the relationship between the executive component of WM and fluid intelligence in children with RD and/or MD may be indirect. For example, Swanson and Alexander (1997) found that the magnitude of the correlations between the executive component of WM and fluid intelligence (Raven Colored Progressive Matrices Test) was significantly smaller in children with RD than average readers (Swanson & Alexander, 1997, table 4). This finding suggests evidence that fluid intelligence while related to the executive system of WM is not an exclusive manifestation of such a system. Second, children with RD and/or MD may use different routes or processes to problem solve, even though solution accuracy is comparable to chronological age (CA)-matched peers (Swanson, 1988, 1993a). For example, Swanson (1988, 1993a) found that students with RD yield comparable performance to average readers on fluid measures of intelligence, but relied on different cognitive routes than average readers in problem-solving (tasks were Tower of Hanoi task, as well as two Piagetian problem-solving tasks [Combinatorial, Pendulum]). On measures of fluid intelligence, problemsolving for children with RD was augmented by “emphasizing problem representation (defining the problem, identifying relevant information or facts given about the problem) rather than procedural knowledge or processes used to identify algorithms” (Swanson, 1993a, p. 864). Thus, there is
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
761
762
H. LEE SWANSON
evidence suggesting that performance by individuals with RD on fluid measures of intelligence may involve compensatory processing. This compensation can partially overcome problems in executive processing (e.g., attention allocation) that, in turn, may allow them to perform in the normal range.
33.6.2
Attributing RD and/MD to Executive Processing Is Counterintuitive to the Notion of Specific Disorders Another assumption is that executive processing deficits represent only domain-general deficits and not specific deficits. That is, difficulties in executive processing are viewed as general processing activities and not related to specific operations. However, executive processing is made up of several parts (i.e., updating, inhibition, task switching) that reflect several different mental activities – each which can represent a specific cognitive process and/or operation. For example, some of these operations can reflect a specific mental constraint in children with RD and/or MD related to (a) maintaining task-relevant information in the face of distraction or interference, (b) suppressing and inhibiting information irrelevant to the task if necessary, and (c) quickly accessing information from long-term memory, all which has been attributed to specific disorders in reading and/or math (e.g., see Kudo et al., 2015; Swanson & Jerman, 2006).
33.6.3
Problems in the Executive System Are Secondary to Deficits in the Phonological Storage System Some studies have argued that deficits in phonological processes for children at risk for RD and/or MD create a bottleneck in the processing of information within the executive system (e.g., Hecht et al., 2001; Peng et al., 2018). Basic structural deficiencies in the storage of phonological input are assumed to impair higher-level processing, such as executive processing. This bottom-up processing approach views lower-level linguistic and cognitive analysis as subserving or influencing higher-order processing in an “upstream” manner. Therefore, performance differences on executive processing measures between children with RD and/or MD when compared to average achievers would be eliminated once performance related to phonological processing was partialed out in the analysis. Several of our earlier regression studies have shown, however, that executive processing deficits within the WM system exist in children with RD and/or MD independent of their deficits in phonological processing (e.g., Swanson, 2004; Swanson & Ashbaker, 2000; Swanson & Berninger, 1995). Also, problems in WM have been found to persist in children with RD and MD even after partialing out the influence of verbal articulation speed (e.g., Swanson & Ashbaker, 2000; Swanson & Beebe-Frankenberger, 2004),
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
short-term memory (STM, e.g., Swanson et al., 1996), or IQ scores (e.g., Swanson & Sachse-Lee, 2001a, b) in various regression models.
33.6.4
Executive Processing Problems in Children with RD and/or MD Are Manifestations of Attention Disorders Because executive deficits are manifestations of monitoring attention, it is easy to attribute executive processing deficits that might arise in children with RD and/or MD as manifestations of Attention-Deficit/Hyperactivity Disorder (ADHD). This is because RD and ADHD are frequently comorbid in epidemiological studies. However, a distinction can be made between executive processing deficits related to the self-monitoring of attention versus constraints in attentional capacity. However, the symptoms commonly attributed to ADHD children’s poor attentional monitoring (impulsivity, distractibility, diminished persistence, diminished sensitivity to feedback, lack of planning and judgment) appear intact relative to ADHD children for children with RD and/or MD (see Kudo et al., 2015). In contrast to ADHD children, research with RD children has shown normal levels of planning and judgment on various problem-solving tasks (e.g., Tower of Hanoi; Swanson, 1993a), and signal detection measures (d’) on vigilance tasks show comparable persistence (although less attentional capacity) as average achievers in their use of attentional resources across time (Swanson, 1981, 1983). These findings do not imply that children with RD and/or MD do not experience monitoring and planning difficulties on high order tasks such as reading comprehension (e.g., Locascio et al., 2010) and/ or math problem-solving (e.g., Swanson et al., 2008, 2015), but rather our results suggest that on some tasks that require complex problem solving that such children’s procedural knowledge (e.g., steps to solve a problem) parallels average achievers.
33.7
Future Directions
We have two directions in our current research. One focuses on the need to separate fundamental deficits in WM from fundamental deficits related to language acquisition. Our current research focuses on English language learners (ELL) with RD and MD whose first language is Spanish. We feel this line of work is important because children with Spanish as a first language in the United States have been found to yield low reading and mathematics scores when compared to other ELL groups on national assessments across several years (e.g., National Assessment of Education Progress, 2017, 2020). Our recent studies have focused on the role of WM in predicting children at risk for RD (e.g., Swanson et al., 2015, 2016) and MD (Swanson et al., 2019a, b) within English language learning samples. Past and more recent studies (Lanfranchi & Swanson, 2005, 2015, 2016; Swanson et al., 2004, 2006, 2011)
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
763
764
H. LEE SWANSON
have attempted to determine (a) whether the memory processes implicated in reading and second language acquisition extend beyond the phonological domain and (b) whether the cognitive processes that underlie difficulties in second language acquisition are the same as those that underlie difficulties in reading for monolingual students. In one of our earlier studies of reading (Swanson et al., 2004), elementary school bilingual and nonbilingual children were administered a battery of cognitive measures (STM, WM, rapid naming, random letter and number generation, vocabulary, and reading [real word and pseudoword reading]) in both Spanish and English. The results showed that English word identification performance was best predicted by a general verbal WM latent factor (a factor that reflected loading from both English and Spanish WM measures) and a Spanish STM factor, whereas English pseudoword reading performance was best predicted by Spanish pseudoword reading and a WM factor. The results also showed that WM and STM performance differentiated among ELL children with and without RD (Swanson et al., 2004). In terms of math, our recent study (Swanson et al., 2019a, b) determined those components of WM that played a significant role in predicting math growth in children who are English language learners with MD. A battery of tests was administered in English and Spanish that assessed computation, reading, vocabulary, inhibition, and components of WM in grade 1 children with follow-up testing in grades 2 and 3. The results indicated that growth in the executive component of WM was significantly related to growth in math performance even when covariates (STM, vocabulary, reading, fluid intelligence) were entered into the latent growth models. In our attempts to isolate fundamental processing difficulties in reading and math, the question emerges, however, as the whether the tripartite structure fits with English learners. Swanson et al. (2019c) tested the three tripartite structure (e.g., executive, phonological, visual-spatial) on five age groups (6-, 7-, 8-, 9-, 10-year-olds) English language learners. The tasks were administered in Spanish and then in English on separate occasions. Confirmatory factor analysis showed a three-factor structure of WM emerged in both L1 and L2 administrations for each age group. Thus, there are three important implications for our work with ELL children: (1) the theoretical framework for WM in English and Spanish fit the data for children who are ELLs, (2) the distinction between the phonological loop and executive system in young children becomes stronger with increases in age, and (3) the cross-language structure of measures is stronger for the executive component of WM than for the phonological loop. Taken together, this cross-sectional study found that Baddeley’s model of WM is an adequate conceptualization for English language learners and that the construct of WM can be measured within both English and Spanish language systems. The second area of research focuses on intervention. Although our previous work found the WM can be modified, via dynamic testing conditions
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
(Swanson, 1992, 2003, 2011; Swanson et al., 1996), whether such modifications improve academic performance is unclear. That is, WM training studies as a means to improve academic performance have yielded mixed results (e.g., Melby-Lervåg & Hulme, 2013; Schwaighofer et al., 2015), and few studies have shown any direct influence of WM training on academic performance (e.g., see Melby-Lervåg & Hulme, 2013, Shipstead et al., 2010, for a review of this literature). Instead of trying to modify WM directly, we are taking a different approach in our studies by embedding WM demands within classroom curriculum materials. Our more recent intervention studies have been directed toward focusing on math word-problem-solving strategies and including WM demands within these lesson plans (e.g., Swanson, 2014, 2016; Swanson et al., 2014, 2015; Swanson, Moran et al., 2013). That is, we have been focusing on how strategies can enhance math problem solving as a function of variations of both the WM demands within the curriculum and the WM capacity of the individual child. For example, a study (see Swanson, 2016) that replicated an earlier study (Swanson et al., 2013), included randomized control/experimental conditions, which investigated the role of cognitive strategy instruction on problem-solving solution accuracy in elementary school children (grade 3) with MD (N = 162). Comparisons were made on three popular cognitive strategy conditions. These included verbal and/or visual-spatial-schema activities + controlled attention training and a controlled attention training condition (referred to as MOC, materials only condition,-in the article). The MOC conditions included the same materials, practice, small group setting, and teacher feedback as the other conditions. The materials included practice with solving word problems that systematically increased into a 5 sentence “load” conditions. For example, Load 1 included 7 lessons of four sentences and one irrelevant sentence. In contrast, Load 5 included lessons 18 through 20, which required solving problems of eight sentences that included five irrelevant sentences. We also found all these conditions effective relative to the control condition (e.g., Swanson et al., 2013). However, we also found that the controlled attention condition (MOC) yielded higher effect sizes relative to the other strategy conditions at posttest on measures of problem-solving and WM. Our rationale for the controlled attention condition (referred to in the article as MOC) was that some of our pilot work found that various activities (underlining key words, circling the question, placing numbers in the diagram, etc.) placed unnecessary cognitive demands on the processing of materials for children with MD. The challenge, of course, is developing intervention materials that can be used in L1 and/or L2 settings
33.7.1 Section Summary Although our preliminary work with ELL children has suggested that growth in reading and math is directly tied to the development of the executive WM
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
765
766
H. LEE SWANSON
system, we would not argue at this point that this system acts independently of cross-language skills across grades. Likewise, although our preliminary intervention work that has embedded WM training within the curriculum materials has yielded some positive outcomes, its application has been applied only to English monolingual children. Thus, additional research is necessary before substantive conclusions can be drawn in terms of its application to ELL children at risk for RD and/or MD.
33.8
Conclusion
We conclude that WM deficits are fundamental problems of children with a specific learning disorder in reading and/or math (i.e., RD and/or MD). Students with RD and/or MD exhibit STM deficits related to the phonological loop, a component of WM that specializes in the retention of speech-based information. However, this subsystem is not the only aspect of WM that underlies RD and/or MD. We find that in situations that place high demands on processing, which in turn place demands on controlled attentional processing (such as monitoring limited resources, suppressing conflicting information, updating information), children with RD and/or MD are at a clear disadvantage when compared with their chronological aged counterparts. Further, these deficits are sustained when articulation speed, phonological processing, fluid intelligence and verbal STM are partialed from the analysis. No doubt, there are gaping holes in our knowledge about how WM and learning problems are related to specific learning disorders in reading and/or math, and therefore additional research is needed. For example, our classroom research has not identified all factors of WM amenable to manipulations, such as extended practice, specific instructions, and strategy use. Additional research needs to be directed toward explaining why WM tasks are good predictors of academic performance. Although, for example, it makes sense that controlled attention ability (e.g., the ability to switch attention between processing and storage requirements) may be particularly good for reading comprehension but not necessarily for simple sight-word recognition, this has not been tested experimentally. No doubt the complexity of the task determines whether general or domain-specific factors come into play.
References Acheson, D. J., Hamidi, M., Binder, J. R., & Postle, B. R. (2011). A common neural substrate for language production and verbal working memory. Journal of Cognitive Neuroscience, 23(6), 1358–1367. doi:10.1162/jocn.2010.21519 Ackerman, P. L., Beier, M., E., & Boyle, M. O. (2002). Individual differences in working memory within a nomological network of cognitive and
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
perceptual speed abilities. Journal of Experimental Psychology: General, 131, 567–589. doi: 10.1037/0096-3445.131.4.567 Alloway, T. P., Gathercole, S. E., Willis, C., & Adams, A. (2004). A structural analysis of working memory and related cognitive skills in young children. Journal of Experimental Child Psychology, 87, 85–106. doi:10.1016/j. jecp.2003.10.002 American Psychiatric Association (2013). Diagnostic and statistical manual of mental disorders (5th ed). American Psychiatric Association. Attout, L., & Majerus, S. (2015). Working memory deficits in developmental dyscalculia: The importance of serial order. Child Neuropsychology, 21(4), 432–450. doi:10.1080/09297049.2014.922170 Baddeley, A. D. (2012). Working memory: Theories, models, and controversies. Annual Review of Psychology, 63, 1–29. doi:10.1146/annurev-psych120710-100422 Baddeley, A. D., Eldridge, M., Lewis, V., & Thomson, N. (1984). Attention and retrieval from long-term memory. Journal of Experimental Psychology: General, 113, 518–540. Baddeley, A., Gathercole, S., & Papagno, C. (1998). The phonological loop as a language-learning device. Psychological Review, 105, 158–173. doi:10.1037//0033-295X.105.1.158 Baddeley, A. D., & Logie, R. H. (1999). The multiple-component model. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 28–61). Cambridge University Press. doi:10.1017/CBO9781139174909.00 Beneventi, H., Tønnessen, F. E., & Ersland, L. (2009). Dyslexic children show short-term memory deficits in phonological storage and serial rehearsal: An fMRI study. International Journal of Neuroscience, 119(11), 2017–2043. doi:10.1080/00207450903139671 Beneventi, H., Tønnessen, F. E., Ersland, L., & Hugdahl, K. (2010). Executive working memory processes in dyslexia: Behavioral and fMRI evidence. Scandinavian Journal of Psychology, 51(3), 192–202. doi:10.1111/j.14679450.2010.00808.x Bledowski, C., Kaiser, J., & Rahm, B. (2010). Basic operations in working memory: Contributions from functional imaging studies. Behavioural Brain Research, 214(2), 172–179. doi:10.1016/j.bbr.2010.05.041 Brandenburg, J., Klesczewski, J., Fischbach, A., Schuchardt, K., Büttner, G., & Hasselhorn, M. (2015). Working memory in children with learning disabilities in reading versus spelling: Searching for overlapping and specific cognitive factors. Journal of Learning Disabilities, 48(6), 622–634. doi:10.1177/0022219414521665 Chein, J. M., & Fiez, J. A. (2010). Evaluating models of working memory through the effects of concurrent irrelevant information. Journal of Experimental Psychology: General, 139(1), 117–137. doi:10.1037/a0018200 Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). L. Erlbaum Associates.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
767
768
H. LEE SWANSON
Cowan, N. (1997). The development of working memory. In N. Cowan (Ed.), The development of working memory in childhood (pp. 163–199). Psychology Press. Cowan, N. (2014). Working memory underpins cognitive development, learning, and education. Educational Psychology Review, 26(2), 197–223. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19(4), 450–466. doi:10.1016/S0022-5371(80)90312-6 De Weerdt, F., Desoete, A., & Roeyers, H. (2013). Working memory in children with reading disabilities and/or mathematical disabilities. Journal of Learning Disabilities, 46(5), 461–472. Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. (1999). Working memory, short-term memory, and general fluid intelligence: A latent variable approach. Journal of Experimental Psychology: General, 128, 309–331. Evans, T. M., Flowers, D. L., Luetje, M. M., Napoliello, E., & Eden, G. F. (2016). Functional neuroanatomy of arithmetic and word reading and its relationship to age. NeuroImage, 143, 304–315. doi:10.1016/j.neuroimage.2016.08.048 Fürst, A. J., & Hitch, G. J. (2000). Separate roles for executive and phonological components of working memory in mental arithmetic. Memory & Cognition, 28(5), 774–782. Gathercole, S. E., Pickering, S. J., Ambridge, B., & Wearing, H. (2004). The structure of working memory from 4 to 15 years of age. Developmental Psychology, 40, 177–190. Geary, D. C. (2011). Cognitive predictors of achievement growth in mathematics: A 5-year longitudinal study. Developmental Psychology, 47(6), 1539–1552. Geary, D. C. (2013). Learning disabilities in mathematics: Recent advances. In H. L. Swanson, K. Harris, & S. Graham (Eds.), Handbook of learning disabilities (2nd ed., pp. 239–256). Guilford. Geary, D. C., Hoard, M. K., Nugent, L., & Bailey, D. H. (2012). Mathematical cognition deficits in children with learning disabilities and persistent low achievement: A five-year prospective study. Journal of Educational Psychology, 104(1), 206–223. Gray, S., Green, S., Alt, M., Hogan, T., Kuo, T., Brinkley, S., & Cowan, N. (2017). The structure of working memory in young children and its relation to intelligence. Journal of Memory and Language, 92, 183–201. Hecht, S. A., Torgesen, J. K., Wagner, R. K., & Rashotte, C. A. (2001). The relations between phonological processing abilities and emerging individual differences in mathematical computational skills: A longitudinal study from second to fifth grades. Journal of Experimental Child Psychology, 79(2), 192–227. Hutton, U. M. Z., & Towse, J. N. (2001). Short-term memory and working memory as indices of children’s cognitive skills. Memory, 9(4–6), 383–394.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
Kaufmann, L., Wood, G., Rubinsten, O., & Henik, A. (2011). Meta-analyses of developmental fMRI studies investigating typical and atypical trajectories of number processing and calculation. Developmental Neuropsychology, 36 (6), 763–787. Kudo, M. F., Lussier, C. M., & Swanson, H. L. (2015). Reading disabilities in children: A selective meta-analysis of the cognitive literature. Research in Developmental Disabilities, 40, 51–62. Kyllonen, P. C., & Christal, R. E. (1990). Reasoning ability is (little more than) working-memory capacity? Intelligence,14, 389–433. doi:10.1016/S01602896(05)80012-1 Johnson, E. S., Humphrey, M., Mellard, D. F., Woods, K., & Swanson, H. L. (2010). Cognitive processing deficits and students with specific learning disabilities: A selective meta-analysis of the literature. Learning Disability Quarterly, 33(1), 3–18. Lanfranchi, S., & Swanson, H. L. (2005). Short-term memory and working memory in children as a function of language-specific knowledge in English and Spanish. Learning and Individual Differences, 15(4), 299–319. Locascio, G., Mahone, E. M., Eason, S. H., & Cutting, L. E. (2010). Executive dysfunction among children with reading comprehension deficits. Journal of Learning Disabilities, 43(5), 441–454. Logie, R. H., Gilhooly, K. J., & Wynn, V. (1994). Counting on working memory in arithmetic problem solving. Memory & Cognition, 22(4), 395–410. Maguire, M. J., Schneider, J. M., Middleton, A. E., Ralph, Y., Lopez, M., Ackerman, R. A., & Abel, A. D. (2018). Vocabulary knowledge mediates the link between socioeconomic status and word learning in grade school. Journal of Experimental Child Psychology, 166, 679–695. Mammarella, I. C., Caviola, S., Giofrè, D., & Szűcs, D. (2018). The underlying structure of visuospatial working memory in children with mathematical learning disability. British Journal of Developmental Psychology, 36(2), 220–235. McLean, J. F., & Hitch, G. J. (1999). Working memory impairments in children with specific arithmetic learning difficulties. Journal of Experimental Child Psychology, 74(3), 240–260. Melby-Lervåg, M., & Hulme, C. (2013). Is working memory training effective? A meta-analytic review. Developmental Psychology, 49(2), 270–291. Melby-Lervåg, M., Lyster, S. H., & Hulme, C. (2012). Phonological skills and their role in learning to read: A meta-analytic review. Psychological Bulletin, 138(2), 322–352. Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., & Howerter, A. (2000). The unity and diversity of executive functions and their contributions to complex “frontal lobe” tasks: A latent variable analysis. Cognitive Psychology, 41, 49–100. National Assessment of Educational Progress (2017). The condition of education (update 2017). US Department of Education.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
769
770
H. LEE SWANSON
National Center for Education Statistics (2020) Common Core of Data (CCD): Local education agency universe survey. U.S. Department of Education. Nee, D. E., Brown, J. W., Askren, M. K., Berman, M. G., Demiralp, E., Krawitz, A., & Jonides, J. (2013). A meta-analysis of executive components of working memory. Cerebral Cortex, 23(2), 264–282. Passolunghi, M. C., & Siegel, L. S. (2004). Working memory and access to numerical information in children with disability in mathematics. Journal of Experimental Child Psychology, 88(4), 348–367. Peña, E. D., Bedore, L. M., & Kester, E. S. (2016). Assessment of language impairment in bilingual children using semantic tasks: Two languages classify better than one. International Journal of Language & Communication Disorders, 51(2), 192–202. Peng, P., Barnes, M., Wang, C., Wang, W., Li, S., Swanson, H. L., & Tao, S. (2018). A meta-analysis on the relation between reading and working memory. Psychological Bulletin, 144(1), 48–76. Peng, P., Congying, S., Beilei, L., & Sha, T. (2012). Phonological storage and executive function deficits in children with mathematics difficulties. Journal of Experimental Child Psychology, 112(4), 452–466. Peng, P., & Fuchs, D. (2016). A meta-analysis of working memory deficits in children with learning difficulties: Is there a difference between verbal domain and numerical domain? Journal of Learning Disabilities, 49(1), 3–20. doi:10.1177/0022219414521667 Richlan, F. (2012). Developmental dyslexia: Dysfunction of a left hemisphere reading network. Frontiers in Human Neuroscience, 6 doi:10.3389/ fnhum.2012.00120 Richlan, F., Kronbichler, M., & Wimmer, H. (2011). Meta-analyzing brain dysfunctions in dyslexic children and adults. NeuroImage, 56(3), 1735–1742. doi:10.1016/j.neuroimage.2011.02.040 Rosselli, M., Ardila, A., Lalwani, L. N., & Vélez-Uribe, I. (2016). The effect of language proficiency on executive functions in balanced and unbalanced Spanish–English bilinguals. Bilingualism: Language and Cognition, 19(3), 489–503. Rosenberg-Lee, M., Lovett, M. C., & Anderson, J. R. (2009). Neural correlates of arithmetic calculation strategies. Cognitive, Affective & Behavioral Neuroscience, 9(3), 270–285. Scarborough, H. S. (1998). Early identification of children at risk for reading disabilities: Phonological awareness and some other promising predictors. In B Shapiro, P. Accardo, & A. Capute (Eds.). Specific reading disability: A view of the spectrum (pp. 75–119). York Press. Schwaighofer, M., Fischer, F., Bühner, M. (2015). Does working memory training transfer? A meta-analysis including training conditions as moderators. Educational Psychologist, 50(2), 138–166 Shipstead, Z., Redick, T. S., and Engle, R. W. (2010). Does working memory training generalize? Psychologica Belgica, 50(3&4), 245–276
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
Siegel, L. S., & Ryan, E. B. (1989). The development of working memory in normally achieving and subtypes of learning disabled children. Child Development, 60(4), 973–980. Smith, E. E., & Jonides, J. (1997). Working memory: A view from neuroimaging. Cognitive Psychology, 33(1), 5–42. Smith, E. E., & Jonides, J. (1999). Storage and executive processes in the frontal lobes. Science, 283(5408), 1657–1661. Swanson, H. (1981). Vigilance deficit in learning disabled children: A signal detection analysis. Child Psychology & Psychiatry & Allied Disciplines, 22(4), 393–399. Swanson, H. (1983). A developmental study of vigilance in learning-disabled and nondisabled children. Journal of Abnormal Child Psychology, 11(3), 415–429. Swanson, H. (1988). Learning disabled children’s problem solving: Identifying mental processes underlying intelligent performance. Intelligence, 12(3), 261–278. Swanson, H. (1989). The effects of central processing strategies on learning disabled, mildly retarded, average, and gifted children’s elaborative encoding abilities. Journal of Experimental Child Psychology, 47(3), 370–397. Swanson, H. (1992). Generality and modifiability of working memory among skilled and less skilled readers. Journal of Educational Psychology, 84(4), 473-488. Swanson, H. (1993a). An information processing analysis of learningdisabled children’s problem solving. American Educational Research Journal, 30(4), 861–893. Swanson, H. (1993b). Executive processing in learning-disabled readers. Intelligence, 17(2), 117–149. Swanson, H. (1993c). Working memory in learning disability subgroups. Journal of Experimental Child Psychology, 56(1), 87–114. Swanson, H. (1995). S-Cognitive Processing Test (S-CPT): A dynamic assessment measure. PRO-ED. Swanson, H. (1996). Individual and age-related differences in children’s working memory. Memory & Cognition, 24, 70–82. Swanson, H. (1999). What develops in working memory? A life span perspective. Developmental Psychology, 35(4), 986–1000. Swanson, H. (2000). Are working memory deficits in readers with learning disabilities hard to change? Journal of Learning Disabilities, 33(6), 551–566. Swanson, H. (2003). Age-related differences in learning disabled and skilled readers’ working memory. Journal of Experimental Child Psychology, 85(1), 1–31. Swanson, H. (2004). Working memory and phonological processing as predictors of children’s mathematical problem solving at different ages. Memory & Cognition, 32(4), 648–661.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
771
772
H. LEE SWANSON
Swanson, H. (2008). Working memory and intelligence in children: What develops? Journal of Educational Psychology, 100(3), 581–602. Swanson, H. (2011). The influence of working memory growth on reading and math performance in children with math and/or reading disabilities. In P. Barrouillet & V. Gaillard (Eds.), Cognitive development and working memory: A dialogue between neo-Piagetian theories and cognitive approaches (pp. 203–231). Psychology Press. Swanson, H. (2012). Intellectual growth in children as a function of domain specific and domain general working memory subgroups. Intelligence, 39, 481–492. Swanson, H. (2014). Does cognitive strategy training on word problems compensate for working memory capacity in children with math difficulties? Journal of Educational Psychology, 106(3), 831–848. Swanson, H. (2016). Word problem solving, working memory and serious math difficulties: Do cognitive strategies really make a difference? Journal of Applied Research in Memory and Cognition, 5(4), 368–383. Swanson, H. (2017). Verbal and visual-spatial working memory: What develops over a life span? Developmental Psychology, 53(5), 971–995. Swanson, H (2020). Learning disabilities as a working memory deficit: A revised model. In R. Sperling, A. Martin, & K. Newton, Handbook of educational psychology & students with special needs (pp. 19–51). Routledge. Swanson, H., & Alexander, J. E. (1997). Cognitive processes as predictors of word recognition and reading comprehension in learning-disabled and skilled readers: Revisiting the specificity hypothesis. Journal of Educational Psychology, 89(1), 128–158. Swanson, H., & Alloway, T. P. (2012). Working memory, learning, and academic achievement. APA educational psychology handbook. Vol 1: Theories, constructs, and critical issues (pp. 327–366). American Psychological Association. Swanson, H., & Ashbaker, M. H. (2000). Working memory, short-term memory, speech rate, word recognition and reading comprehension in learning disabled readers: Does the executive system have a role? Intelligence, 28(1), 1–30. Swanson, H., Ashbaker, M. H., & Lee, C. (1996). Learning-disabled readers’ working memory as a function of processing demands. Journal of Experimental Child Psychology, 61(3), 242–275. Swanson, H., & Beebe-Frankenberger, M. (2004). The relationship between working memory and mathematical problem solving in children at risk and not a risk for serious math difficulties. Journal of Educational Psychology, 96, 471–491. Swanson, H., & Berninger, V. (1995). The role of working memory in skilled and less skilled readers’ comprehension. Intelligence, 21(1), 83–108. Swanson, H., & Fung, W. (2016). Working memory components and problem-solving accuracy: Are there multiple pathways? Journal of Educational Psychology, 108(8), 1153–1177.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
Swanson, H., Howard, C. B., & Sáez, L. (2006). Do different components of working memory underlie different subgroups of reading disabilities? Journal of Learning Disabilities, 39(3), 252–269. Swanson, H., & Jerman, O. (2006). Math disabilities: A selective metaanalysis of the literature. Review of Educational Research, 76(2), 249–274. Swanson, H., & Jerman, O. (2007). The influence of working memory on reading growth in subgroups of children with reading disabilities. Journal of Experimental Child Psychology, 96(4), 249–283. Swanson, H., Jerman, O., & Zheng, X. (2008). Growth in working memory and mathematical problem solving in children at risk and not at risk for serious math difficulties. Journal of Educational Psychology, 100(2), 343–379. Swanson, H., Jerman, O., & Zheng, X. (2009). Math disabilities and reading disabilities: Can they be separated? Journal of Psychoeducational Assessment, 27(3), 175–196. Swanson, H., Kehler, P., & Jerman, O. (2010). Working memory, strategy knowledge, and strategy instruction in children with reading disabilities. Journal of Learning Disabilities, 43(1), 24–47. Swanson, L., & Kim, K. (2007). Working memory, short-term memory, and naming speed as predictors of children’s mathematical performance. Intelligence, 35(2), 151–168. Swanson, H., Kong, J., & Petcu, S. (2018). Math difficulties and working memory growth in English language learner children: Does bilingual proficiency play a significant role? Language, Speech, and Hearing Services in Schools, 49(3), 379–394. Swanson, H., Kong, J. E., & Petcu, S. D. (2019a). Individual differences in math problem solving and executive processing among emerging bilingual children. Journal of Experimental Child Psychology, 187, 25. Swanson, H., Kong, J., & Petcu, S. D. (2019b) Growth in math computation among monolingual and English language learners: Does the executive system have a role? Developmental Neuropsychology, 44(8), 566–593. Swanson, H., Kong, J, & Petcu, S. (2020) Emerging bilingual children at risk and not at risk for math difficulties. Child Neuropsychology, 26(4), 489–517. Swanson, H., Kudo, M., & Guzman-Orth, D. (2016). Cognition and literacy in English language learners at risk for reading disabilities: A latent transition analysis. Journal of Educational Psychology, 108(6), 830–856. Swanson, H., Kudo, M. F., & Van Horn, M. L. (2019c). Does the structure of working memory in EL children vary across age and two language systems? Memory, 27(2), 174–191. Swanson, H., Lussier, C. M., & Orosco, M. J. (2015). Cognitive strategies, working memory, and growth in word problem solving in children with math difficulties. Journal of Learning Disabilities, 48(4), 339–358. Swanson, H., Lussier, C. M., & Orosco, M. J. (2015). Cognitive strategies, working memory, and growth in word problem solving in children with math difficulties. Journal of Learning Disabilities, 48(4), 339–358.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
773
774
H. LEE SWANSON
Swanson, H., & McMurran, M. (2018). The impact of working memory training on near and far transfer measures: Is it all about fluid intelligence? Child Neuropsychology, 24(3), 370–395. Swanson, H., Moran, A. S., Bocian, K., Lussier, C., & Zheng, X. (2013). Generative strategies, working memory, and word problem solving accuracy in children at risk for math disabilities. Learning Disability Quarterly, 36 (4), 203–214. Swanson, H., Moran, A., Lussier, C., & Fung, W. (2014). The effect of explicit and direct generative strategy training and working memory on word problem-solving accuracy in children at risk for math difficulties. Learning Disability Quarterly, 37(2), 111–122. Swanson, H., Olide, A. F., & Kong, J. E. (2018). Latent class analysis of children with math difficulties and/or math learning disabilities: Are there cognitive differences? Journal of Educational Psychology, 110(7), 931–951. Swanson, H., Orosco, M. J., & Kudo, M. (2017). Does growth in the executive system of working memory underlie growth in literacy for bilingual children with and without reading disabilities? Journal of Learning Disabilities, 50(4), 386–407. Swanson, H., Orosco, M. J., & Lussier, C. M. (2012). Cognition and literacy in English language learners at risk for reading disabilities. Journal of Educational Psychology, 104(2), 302–320. Swanson, H., Orosco, M. J., & Lussier, C. M. (2015). Growth in literacy, cognition, and working memory in English language learners. Journal of Experimental Child Psychology, 132, 155–188. Swanson, H., Orosco, M. J., Lussier, C. M., Gerber, M. M., & Guzman-Orth, D. (2011). The influence of working memory and phonological processing on English language learner children’s bilingual reading and language acquisition. Journal of Educational Psychology, 103(4), 838–856. Swanson, H., & Sachse-Lee, C. (2001a). A subgroup analysis of working memory in children with reading disabilities: Domain-general or domain-specific deficiency? Journal of Learning Disabilities, 34(3), 249–263. Swanson, H., & Sachse-Lee, C. (2001b). Mathematical problem solving and working memory in children with learning disabilities: Both executive and phonological processes are important. Journal of Experimental Child Psychology, 79(3), 294–321. Swanson, H., Sáez, L., Gerber, M., & Leafstedt, J. (2004). Literacy and cognitive functioning in bilingual and nonbilingual children at or not at risk for reading disabilities. Journal of Educational Psychology, 96(1), 3–18. Swanson, H., & Siegel, L. (2001a). Elaborating on working memory and learning disabilities: A reply to commentators. Issues in Education: Contributions from Educational Psychology, 7, 107–129. Swanson, H., & Siegel, L. (2001b). Learning disabilities as a working memory deficit. Issues in Education: Contributions from Educational Psychology, 7, 1–48.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
Specific Learning Disorders
Swanson, H., & Zheng, X. (2013). Memory difficulties in children and adults with learning disabilities. In H. L. Swanson, K. Harris, & S. Graham (Eds), Handbook of learning disabilities (2nd ed, pp. 214–238). Guilford Press. Swanson, H., Zheng, X., & Jerman, O. (2009). Working memory, short-term memory, and reading disabilities: A selective meta-analysis of the literature. Journal of Learning Disabilities, 42(3), 260–287. Uppal, H. K., & Swanson, H. L. (2016). Teachers’ ratings of working memory in English language learners: Do laboratory measures predict classroom analogues? Applied Cognitive Psychology, 30(6), 871–884. Unsworth, N., & Engle, R. W. (2007). On the division of short-term and working memory: An examination of simple and complex span and their relation to higher order abilities. Psychological Bulletin, 133(6), 1038–1066. Wang, S., & Gathercole, S. E. (2013). Working memory deficits in children with reading difficulties: Memory span and dual task coordination. Journal of Experimental Child Psychology, 115(1), 188–197.
https://doi.org/10.1017/9781108955638.040 Published online by Cambridge University Press
775
34 A New Perspective on the Connection between Memory and Sentence Comprehension in Children with Developmental Language Disorder James W. Montgomery, Ronald B. Gillam, and Julia L. Evans
34.1
Introduction
Children with developmental language disorder (DLD) demonstrate significant difficulty mastering spoken and written language yet exhibit broadly normal-range nonverbal intelligence, normal hearing sensitivity, articulation, and no known neurological impairment (Leonard, 2014). During the preschool years, the learning and use of finite verb morphology is a defining and discriminating feature of DLD (Leonard et al., 1997; Rice et al., 1998). Beyond age seven, however, deficits in grammatical morphology no longer discriminate children with DLD and typically developing (TD) children, even though children with DLD continue to make errors (Moyle et al., 2011). Instead, significant weaknesses in syntactic knowledge and sentence comprehension appear to become dominant features of the communicative and academic profile of DLD (Friedmann & Novogrodsky, 2007; Montgomery et al., 2017; Montgomery & Evans, 2009; van der Lely & Stollwerck, 1997). Weaknesses in syntactic knowledge become increasingly evident during the school-age years as the academic requirements of the language arts curriculum place greater demands on syntactic knowledge and use (Nippold, 2017; Nippold et al., 2009).
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
Syntactic and sentence comprehension deficits are not the only significant limitations in children with DLD. These individuals also exhibit a wide range of memory limitations, including reduced verbal short-term memory capacity (Archibald & Gathercole, 2007; Conti-Ramsden et al., 2001), reduced verbal working memory capacity (Archibald & Gathercole, 2007; Ellis Weismer et al., 1999; Montgomery et al., 2019), and weak long-term memory abilities (Conti-Ramsden et al., 2015; Evans et al., 2009; Lum et al., 2014). They also exhibit poor attentional control, which is a key component of working memory (Engle, 2018; Victorino & Schwartz, 2015). Though there is a general consensus that the memory and sentence comprehension deficits of these children relate, our historical understanding of the nature of the relationship has been poor. In this chapter, we describe a new, conceptually integrated and data-driven structural model of this relationship.
34.2
What Is Auditory Sentence Comprehension?
Spoken sentence comprehension represents a unique problem-solving space because listeners must construct structure and meaning from a fleeting acoustic signal while trying to manage two fundamental challenges. First, sentence structure and meaning must be developed in the moment. Listeners may manage these temporal challenges by initiating comprehension from sentence onset (Marslen-Wilson & Zwitserlood, 1989; Zwitserlood, 1989) and incrementally building structure and meaning from all available cues in the input (phonological, morphological, syntactic, semantic, real-world). From the developing sentence representation, listeners then may be able to predict upcoming language material, predictions that may occur at a general level (e.g., next word is a verb or noun) or lexical level (i.e., specific word) depending on the availability and strength of semantic/real-world cues (Altmann & Kamide, 1999; Ferretti et al., 2001). Such processing abilities enable listeners to comprehend what they hear with some immediacy. The second challenge is listeners must wrestle with a severely capacity-limited working memory system that needs to hold the products of earlier processing while interpreting newly arriving language material. Such challenges may be managed by listeners creating just a few, integrated linguistic chunks (e.g., phrases, clauses) out of the stream of words and then combining these chunks into a single cohesive representation (e.g., McCauley et al., 2017; McCauley & Christiansen, 2015). Take the complex sentence The criminal that the judge sentenced to 20 years was upset by the ruling. Even though this sentence is not a high-frequency canonical/typical noun-verb-noun (NVN)/subject-verb-object (SVO) structure (e.g., The judge sentenced the criminal to 20 years), adults know that noun phrase one (NP1), the criminal, is the patient/recipient of the action (sentenced) and the judge (NP2) is the agent of the action. Adults understand such sentences with relative immediacy because they can quickly activate from
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
777
778
JAMES W. MONTGOMERY ET AL.
long-term memory (LTM) relevant language knowledge to build appropriate structure and meaning. The relative ease of such a sentence also owes to the fact that listeners can rely on semantic/real-world knowledge about who does what to whom, that is, judges sentence criminals. In short, listeners manage their memory constraints by chunking the input stream into a few intermediate units like noun phrases, verb phrases, and clauses (dependent, independent), storing these units in working memory (WM), and then combining the units into a single cohesive representation. Chunking, of course, depends on the state of listeners’ language knowledge.
34.3
DLD Sentence Comprehension: Two Historical Views
School-age children and adolescents with DLD demonstrate marked sentence comprehension deficits relative to TD peers. For these individuals, passive structures (The lion [i] was bitten [ti] by the monkey) and object relative structures (The lion [i] that the monkey bit [ti] was brown) pose special problems (Dick et al., 2004; Friedmann & Novogrodsky, 2007; van der Lely & Stollwerck, 1997). These forms are hard because there is a mismatch between the typical mapping of subject to agent and object to patient. Noun phrase 1 functions as the grammatical subject but semantically as the patient (not the agent), and NP2 functions grammatically as the object but semantically as the agent (not the patient/recipient). Even though passives have an NVN surface form (with an additional by-phrase) and object relatives an obvious noncanonical NNV form, children must come to realize that in both structures NP1 is the patient and NP2 is the agent. Historically, two broad theoretical perspectives have been advanced to help explain the comprehension limitations of children with DLD: a linguistic account and a narrow, memory-based account (see Montgomery et al., 2016 for a review).
34.3.1 Linguistic View of DLD Sentence Comprehension Deficits The computational grammatical complexity hypothesis, which is the primary linguistic account, assumes children with DLD have trouble comprehending noncanonical sentences due to difficulties building hierarchical grammatical structures (Marinis & van der Lely, 2007; Marshall et al., 2007; Marshall & van der Lely, 2006). Such structures presumably require the building of a long-distance syntactic (filler–gap) dependency via syntactic movement in which there is movement/displacement of the logical object NP (e.g., the lion in the example sentences above) of the verb (bitten, bit) to the subject position. The canonical relationship of the verb to its object is maintained by the moved element leaving a trace ([ti]) behind in its original object position, and the trace shares a coreferential relationship with the moved element ([i]). Both structures also presumably entail reactivation of
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
the moved element to establish the filler–gap dependency in that NP1 (filler, marked as [i]) is reactivated after processing the verb (trace/gap site). The grammatical complexity hypothesis assumes that the primary problem in complex sentence comprehension is that the representation of syntactic movement is treated as optional by children with DLD, not as obligatory.
34.3.2
Memory-Based View of DLD Sentence Comprehension Deficits The memory-based account focuses on the relationship between sentence comprehension and verbal WM (e.g., Montgomery, 1995, 2000; Montgomery & Evans, 2009; Robertson & Joanisse, 2010). Importantly, however, some investigators have begun to examine the potential relationship between sentence comprehension and LTM, most notably procedural memory (e.g., Conti-Ramsden et al., 2015; Hamrick et al., 2018; Hedenius et al., 2011) but this line of work is scarce. 34.3.2.1 Working Memory Working memory is the ability to temporarily maintain in the moment a limited amount of information in an accessible state while performing some kind of mental activity (Baddeley, 2012; Cowan et al., 2012). It is well established that children with DLD demonstrate significantly reduced WM capacity relative to TD peers (Archibald & Gathercole, 2007; Conti-Ramsden et al., 2015; Ellis Weismer et al., 1999; Marton et al., 2014; Montgomery et al., 2019). The majority of DLD studies have taken indirect approaches to examining the association between verbal WM and sentence comprehension in these children to test the assumption that, relative to TD children, these children have difficulty remembering the products of earlier processing (e.g., phrases, clauses) or the details related to previously processed chunks as they process new, incoming material. Most previous studies have three methodological characteristics in common: (1) manipulation of the length of sentences and often the structure of sentences; (2) use of conventional picture pointing tasks to assess comprehension; and (3) inclusion of an independent measure to index memory capacity. The general pattern of findings implicates limited WM capacity in these children’s difficulty comprehending verbal be passives (Montgomery & Evans, 2009; Robertson & Joanisse, 2010) and even lengthy SVO structures (Leonard et al., 2013; Montgomery, 2000; Montgomery et al., 2009; Robertson & Joanisse, 2010). Such findings have been interpreted to suggest that the children have greater difficulty than TD peers in holding on to linguistic chunks that have already been processed and simultaneously processing new language material. Working memory comprises more than storage. Controlled attention is another critical component of WM. The ability to allocate, sustain, and switch attention are important to support information storage (Baddeley,
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
779
780
JAMES W. MONTGOMERY ET AL.
2012; Barrouillet et al., 2009; Engle et al., 1999). Limited research has examined the potential influence of controlled attention on the sentence comprehension of children with DLD. Results of the few studies focusing on the association between attention and the comprehension of elaborated SVO sentences suggest there is a positive relationship (Leclercq et al., 2013; Montgomery, 2008, Montgomery et al., 2009). The assumption behind such studies was that children with DLD have insufficient attentional resources and/or difficulty sustaining these resources over the course of processing a sentence to support comprehension. Correlation analyses and pattern analyses reveal that controlled attention is related to comprehension in children with DLD but not TD children (Leclercq et al., 2013; Montgomery, 2008; Montgomery et al., 2009). Such results have been taken to suggest that even simple grammar processing by children with DLD is not yet automatic. Compared with TD peers, children with DLD appear to exert significant mental effort to understand even simple grammar.
34.3.2.2 Long-Term Memory One especially relevant LTM framework that has the potential to help us understand the broad range of language-learning difficulties in DLD is the declarative/procedural memory model of Ullman and colleagues (Hamrick et al., 2018; Hedenius et al., 2011; Ullman, 2004). Ullman (2004, 2016) proposes that these LTM systems underpin language learning, particularly morphology and syntax (Ullman, 2004, 2016). Procedural memory relates to the unconscious (implicit) learning of sequential patterns (Squire et al., 1993; Ullman, 2004). Implicit learning involves learners taking advantage of their cognitive tendencies to track the distributional or statistical regularities in input. Regarding syntax, these regularities correspond to different word-order patterns like SVO/NVN, passive (NVN with a byphrase), and object relative (NNV) structures. TD children use input regularities to learn different syntactic patterns, including passives and object relatives (Gómez & Gerken, 1999; Kidd, 2012; Kidd & Arciuli, 2016; Savage et al., 2003). Additionally, performance on independent implicit learning tasks predicts TD children’s syntactic learning (Conti-Ramsden et al., 2015; Kidd, 2012; Kidd & Arciuli, 2016). The declarative memory system is critical to lexical learning but may also be important to syntactic learning, especially early in life, by supporting the learning of linguistic chunks (Hamrick et al., 2018; Ullman, 2016). Compared with TD children, children with DLD show poor implicit learning across the nonverbal and verbal domains (Evans et al., 2009; Hedenius et al., 2011; Lum et al., 2014). Researchers have taken these findings to suggest that children with DLD have an implicit learning deficit (Conti-Ramsden et al., 2015; Garraffa et al., 2018; Hsu & Bishop, 2010, 2014; Lum & Conti-Ramsden, 2013), a suggestion first proposed by Ullman and Pierpont (2005) in their procedural deficit hypothesis. Two sources of evidence support the procedural deficit hypothesis. First, performance on
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
independent implicit learning tasks fails to predict sentence comprehension in children with DLD (Conti-Ramsden et al., 2015). Second, children with DLD who have syntactic deficits demonstrate significant difficulty relative to TD children with long-term retention of newly learned novel visual sequences (Hedenius et al., 2011). By contrast, children with DLD tend to show better declarative memory abilities, indexed by lexical knowledge, than procedural abilities (Conti-Ramsden et al., 2015). Whether their declarative memory and syntactic learning abilities relate is unknown.
34.4
A New Perspective on the Memory-Comprehension Connection in DLD
Memory limitations are implicated in the spoken sentence comprehension deficits of those with DLD. However, a clear understanding of the relationship has been missing from the DLD literature (see Montgomery et al., 2016). To our minds, the lack of clarity owes to one major factor – the absence of an integrated model of sentence comprehension, one that specifies the structural relationship between the components of memory and comprehension. We believe a model specifying such a relationship would advance our understanding of the comprehension abilities of children and adolescents with DLD in important ways. Below, we describe the first such model.
34.4.1 Model Components and Motivation We refer to our model as the GEM (Gillam, Evans, and Montgomery) model. It includes four mechanisms: fluid reasoning; controlled attention; WM; and language LTM. The selection of the mechanisms was motivated by their relevance to comprehension based on empirical findings in the DLD, TD, and adult sentence comprehension literatures. The model aims to understand the structural relationship between cognitive processing, especially the role of memory, and syntactic sentence comprehension in young school-age children with and without DLD. 34.4.1.1 Fluid Reasoning The influence of fluid reasoning in the spoken sentence comprehension of children has received little attention. Findings in the adult sentence comprehension literature, however, have relevance to our thinking. Some investigators (Andrews et al., 2017) have argued that fluid reasoning and comprehension are similar to some extent because both involve recognizing and interpreting patterns in the input. Relative to adults with weak fluid reasoning, those with stronger abilities are more accurate at determining the agent-patient relationship in complex sentences containing weak or nonbiasing semantic/real-world cues (The woman that the man helped
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
781
782
JAMES W. MONTGOMERY ET AL.
sang well). In this object relative sentence, neither NP (the woman, the man) is semantically favored as the agent, rendering the sentence more difficult than a sentence like The criminal that the judge sentenced was unhappy, which contains strong semantic/real-world cues. Such findings imply that those adults with stronger general pattern recognition abilities are better able to determine the agent-patient relationship in the presence of weak semantic/ real-world cues. Fluid reasoning and syntactic comprehension have also been shown to significantly correlate in adults (Engelhardt et al., 2017). The same is true for school-age TD children (Garcia-Madruga et al., 2014; Motallebzadeh & Yazdi, 2016).
34.4.1.2 Controlled Attention Controlled attention includes both sustaining activation of attention and switching the focus of attention to something new. We might reason that sustaining attention for a sentence should allow children to attend to the incoming words of a sentence (Montgomery et al., 2009), thus promoting their chunking of the input into linguistic units. Attention switching might be important in that children need to toggle their attention between storing linguistic chunks in WM that have already been created (i.e., keep the chunks in an active state) and language LTM, which generates new chunks from incoming input (Finney et al., 2014). 34.4.1.3 Language LTM Language knowledge resides in LTM and includes phonological, morphological, lexical, semantic, and syntactic representations. Our model borrows heavily from the chunk-and-pass model of language processing developed by Christiansen and colleagues (Chater et al., 2016; McCauley et al., 2017 McCauley & Christiansen, 2014, 2015). This model assumes a connectionist (Christiansen & MacDonald, 1999; MacDonald & Christiansen, 2002; McCauley & Christiansen, 2015) and usage-based (Abbot et al., 2006; Lieven et al., 2009) perspective of language learning and processing. One of the key processing principles of the chunk-and-pass model is rapid input chunking and passing chunks created at lower levels (e.g., phonological, lexical) to higher levels (e.g., multiword units, syntactic). Immediately chunking input into multiword units allows listeners to create intermediate and more abstract structures (NPs, verb phrases, clauses). Chunking occurs iteratively throughout the input until all necessary structures are developed, at which point they are combined into a single, cohesive structure. As structures are built, listeners use available semantic-pragmatic cues to assign meaning to the components of the structures. Developmentally, syntactic structure develops as a natural by-product of repeated language processing experiences, with children creating memory traces (templates) corresponding to different syntactic structures like SVOs, subject relatives, passives, and object relatives. The ability to chunk improves with age as the language system gains more language processing
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
experience. These greater experiences lead to more robust and stable multiword traces, which, in turn, leads to further gains in language processing efficiency (Thiessen, 2017). Emerging evidence in the developmental literature suggests that multiword chunks are crucial building blocks for syntactic development and that such chunks are used for comprehension and production (Arnon & Clark, 2011; Arnon et al., 2017; Arnon & Snider, 2008; Bannard & Lieven, 2012; Bannard & Matthews, 2008; Cornish et al., 2017). Repeated language processing experiences provide children crucial information about the distributional patterns across words, allowing them to create multiword chunks and the opportunity to reuse the components (nouns, verbs) within these chunks to acquire new multiword chunks (Cornish et al., 2017; Theakston & Lieven, 2017). The importance of learning and using multiword chunks cannot be overstated. As already mentioned, listeners are faced with two major hurdles during comprehension. The first is that comprehension takes place in the moment, requiring listeners to make immediate sense of what the speaker is saying. Second, listeners must store in memory earlier parts of a sentence while they process downstream material. Reliance on multiword chunk templates can greatly minimize both challenges. Activation of multiword representations allows listeners to efficiently chunk a continuous stream of words into fewer, more cohesive structural units. Doing so also allows listeners to anticipate upcoming words, either in a general sense or specific lexical sense. Activation of multiword templates thus not only speeds up comprehension but also conserves memory space.
34.4.1.4 Working Memory Also noted earlier, WM involves holding information in an active state while some kind of cognitive activity takes place (Baddeley, 2012; Cowan et al., 2012, 2014; Unsworth & Engle, 2007). Sentence comprehension is all about concurrent verbal processing and storage. To understand a sentence, listeners must store the products of earlier processing while interpreting newly arriving information. Viewing the WM-comprehension link in this way leads us to adopt the embedded processes WM model of Cowan and associates (Adams et al., 2018; Cowan et al., 2012, 2014; Hardman & Cowan, 2015). We believe Cowan’s model holds special relevance to comprehension because of its parsimonious, integrated view of WM and LTM, and the chunking function of language LTM. A central tenet of Cowan’s model is that WM is nested within LTM, a view that is both theoretically and empirically supported by the work of others (Engle et al., 1999; Loaiza & Camos, 2018; Nee & Jonides, 2013; Öztekin & Cowan, 2015; Öztekin et al., 2010; Unsworth & Engle, 2007). Related to comprehension, the basic idea is that incoming words are activated in LTM, and these representations then become the momentary objects of WM to be processed by the language system. Working memory includes a central
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
783
784
JAMES W. MONTGOMERY ET AL.
storage component and a peripheral component. Central storage is limited to about one item (e.g., phrase, clause) and is analogous to the focus of attention. Peripheral storage contains the other activated items lying just outside focal attention. Total WM capacity, which is limited to about 4–5 chunks, is defined as the combination of items in central and peripheral storage. Findings from WM studies also suggest that it is central storage that holds the most recently activated item(s) and peripheral storage that holds earlier encountered items (Öztekin et al., 2010). Items in WM may be of variable size depending on whether they have undergone chunking or not. Chunking is a crucial functional feature of the model because it allows individuals to consolidate many encountered items into fewer, larger integrated units, thereby conserving memory space. This view of chunking corresponds nicely with the Chunk-and-Pass view of language comprehension and provides a strong motivation for adopting Cowan’s model as part of our integrated model of comprehension. At its essence, sentence comprehension involves continuous chunking of incoming material into fewer, more integrated units like clauses and ultimately deriving a cohesive sentence representation.
The Conduit or Mediator Function of WM We mentioned that WM should serve as a conduit or mediator for fluid reasoning, controlled attention, and language LTM to support children’s sentence comprehension. We reasoned that WM should be the mechanism through which fluid reasoning, controlled attention, and language LTM would operate to influence comprehension indirectly. This assumption seemed very reasonable if one assumes that of these four mechanisms it is WM that is most proximal to comprehension, given the primary function of WM is to coordinate concurrent verbal processing and storage. It is WM that should allow children to hold on to the products of prior processing while at the same time processing newly arriving language material. A second reason for thinking WM should be the conduit is that each of the other mechanisms relates positively to WM, which, in our view, should elevate WM to this functional status. For instance, in terms of fluid reasoning, novel problem-solving appears to involve both information processing and storage. Evidence shows a moderate-to-strong relationship between WM and fluid reasoning in adults (Burgess et al., 2011; Fukuda et al., 2010; Kane et al., 2004) as well as TD children (Engel de Abreu et al., 2010). The storage demands on fluid reasoning tasks appear to reflect participants’ ability to hold in memory stimulus items they have associatively “bound” together in some fashion (Wilhelm et al., 2013). The use of controlled attention during fluid reasoning may reflect the toggling of attention between the stimulus items and the processes of associatively binding the items together (Shipstead et al., 2016). The link between WM and controlled attention is inherent in processing, as controlled attention is part of the WM system (Baddeley, 2012; Barrouillet & Camos, 2001; Cowan
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
et al., 2012; Engle et al., 1999). Controlled attention enables participants to toggle their attention between processing incoming information and previously stored information. Finally, the link between WM and language LTM should play out in two ways. First, WM corresponds to the temporary “storage” of those items that have been activated in language LTM (e.g., words, phrases, or sentences). Another way to think about this relationship is that those items activated in LTM are the very same ones that become the temporary objects of WM. The second and related link between WM and LTM is that WM functions as the conduit for language LTM during comprehension. It is WM that stores the intermediate products of prior processing while language LTM continues to process new, incoming input. Recall that within Cowan’s model, total WM storage capacity is the combination of chunks residing in central storage and peripheral storage. It would seem reasonable then to assume that during comprehension, it is peripheral storage that holds the chunks that have already been created while central storage holds the most recently created chunk (Öztekin et al., 2010).
34.4.2 A Test of the New Perspective A large-scale project was conducted with 117 children with DLD and 117 TD children propensity matched for age (7–11 years), gender, mother’s education, and family income (see Montgomery et al., 2018 for details). The two overarching aims of the project were to (1) provide a detailed description of the syntactic comprehension abilities of children and (2) build a structural model describing the structural relationship between cognition (including memory) and sentence comprehension. Children’s syntactic comprehension was indexed by performance on a conventional picture-pointing task. All of the sentences expressed highly implausible events as they contained no semantic/real-world cues. The children were thus forced to rely on their structural (word order) knowledge to guide comprehension. Children listened to two kinds of canonical sentences: SVOs (The square had changed the bed under the very new dry key) and subject relatives (The watch that had hugged the truck behind the kite was bright). They also listened to two kinds of noncanonical sentences: verbal be passives (The watch was bumped by the wheel near the very bright clock) and object relatives (The chair that the bread had splashed under the square was new). The SVO and passive sentences contained a single clause, whereas the subject and object relative sentences included two clauses (relative, main). The internal consistency of the two canonical structures was .84 and the internal consistency for the noncanonical structures was .89 (Montgomery et al., 2017). Immediately after each sentence, the children saw three images on a screen: one of the agent, one of the patient, and one of the object named in the prepositional phrase. They were asked to touch the picture of the agent. Each of the cognitive mechanisms was indexed by two or three measures. Fluid reasoning was indexed by performance on the figure ground,
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
785
786
JAMES W. MONTGOMERY ET AL.
sequential order, and repeated patterns subtests of the Leiter International Performance Scale-Revised (Roid & Miller, 1997). Controlled attention was represented by performance on a sustained attention task and an attention switching task. Language LTM was represented by performance on the receptive and expressive portions of the Test of Narrative Language (Gillam & Pearson, 2004). Finally, WM was indexed by performance on two tasks involving concurrent processing and storage of words and digits or high and low tones.
34.4.2.1 Patterns of Sentence Comprehension The overall performance of the children with DLD was not surprising. The children with DLD had significantly greater difficulty comprehending both the canonical and noncanonical structures. The poorer performance of the children with DLD was evident on both canonical sentence types and noncanonical types. The more interesting findings had to do with group differences in developmental changes in comprehension abilities. Each group was divided into a younger cohort (AgeMean = 8 years, 1 month) and an older cohort (AgeMean = 10 years, 8 months) of approximately equal numbers. The TD children’s comprehension of both the canonical and noncanonical structures improved with age. The children with DLD, however, showed developmental improvement only for the canonical structures. Even more interesting, the oldest children with DLD were significantly outperformed even by the youngest TD children on each noncanonical structure. These findings are the clearest demonstration to date of just how big a limitation children with DLD have learning and using syntactic structure, especially noncanonical structure, to guide comprehension. The findings also, by extension, align very well with the view that these children have a significant procedural/implicit learning deficit (Conti-Ramsden et al., 2015; Hedenius et al., 2011; Hsu & Bishop, 2010; Ullman & Pierpont, 2005). Framing these findings within the broader connectionist (Christiansen & MacDonald, 1999; Joanisse & McClelland, 2015) and usage-based perspectives of language learning (Abbot-Smith & Tomasello, 2006; Lieven et al., 2009) also offers us a unique way to think about the comprehension difficulties of children with DLD. These findings suggest that these children have significant limitations (1) discovering and creating multiword templates for LTM storage and (2) chunking input into multiword units. These interpretations implicate the role of memory deficits in the syntactic learning and comprehension difficulties of children with DLD. 34.4.2.2 Cognitive Nature of Sentence Comprehension The second phase of this work was to describe the structural relationship between cognition and sentence comprehension, with a special focus on memory (Montgomery et al., 2018). The primary aims were to: (1) describe the relationship among WM, language LTM, and sentence comprehension;
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
(2) determine whether the relationship was similar or different for children with DLD and TD children; and (3) determine whether the relationship was similar or different for canonical and noncanonical sentence comprehension. To address these aims, we used a multistep modeling process. In the first step, each cognitive mechanism was represented as an aggregate construct that combined their respective tasks into a composite measure. Correlation analyses revealed moderate to strong correlations between the measures within each mechanism construct, and lower, but still significant, correlations across measures indexing the different mechanisms. Results of a confirmatory factor analysis also indicated a very good model fit for the measurement model comprising the constructs of fluid reasoning, controlled attention, language LTM, and complex WM, both when the group variance was combined and when it was analyzed separately. Next, structural equation modeling was conducted to evaluate the adequacy of five models describing the structural relationship among the cognitive mechanisms and sentence comprehension. A direct model was first evaluated to determine whether one or more of the cognitive mechanisms had a direct influence on sentence comprehension. This model (Figure 34.1) proved to be statistically inadequate, indicating that no single mechanism functioned to influence comprehension directly. Four indirect models were then evaluated to determine whether any of the cognitive mechanisms functioned as a mediator for the other mechanisms to indirectly influence sentence comprehension. Only the model with WM as the mediator proved to be a statistically good fit with the data (Figure 34.2). Not only was the model adequate for all the children combined, but it was also adequate for each group separately. This model confirmed our hunch that it is WM that functions as the conduit through which fluid reasoning, controlled attention, and language LTM operate to indirectly influence the sentence comprehension of both TD children and children with DLD. Importantly, however, even though the WM-mediator model applied to each group, there were some very important and interesting group differences in the magnitude of the indirect influences of fluid reasoning, controlled attention, and language LTM on sentence comprehension. For instance, the indirect influence of fluid reasoning was very small and nonsignificant on the comprehension of both canonical and noncanonical structures for the children with DLD. By contrast, for the TD children, the influence was significant for each sentence type. Regarding controlled attention, the pattern was reversed. For the children with DLD, controlled attention had a significant indirect influence on both canonical and noncanonical sentence comprehension. For the TD children, attention had no detectable indirect influence. The differential role played by language LTM in the groups was especially interesting. Language LTM had a significant indirect influence on canonical sentence comprehension of both the children with DLD and TD children. The relationship between LTM and comprehension as mediated by WM was about the same for each group.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
787
788
JAMES W. MONTGOMERY ET AL.
Figure 34.1 Structural equation model of the direct relationships between cognitive processing and sentence comprehension Note: the exogenous latent variables of controlled attention (CATT) are indexed by auditory sustained attention (SusAtt) and auditory attention switching (AttSW), fluid reasoning (FLD-R) indexed by Leiter figure ground (Leiter FG), Leiter sequential order (Leiter SO), Leiter repeated patterns (Leiter RP), complex working memory (cWM) indexed by verbal working memory (WJ-AWM) and auditory working memory for tones (Hi-Low), and long-term memory language knowledge (LTM-LK) indexed by narrative language comprehension (TNLREC) and narrative language expression (TNL-EXP); propensity score (PROP S); and canonical sentence comprehension accuracy (CANACC) and noncanonical sentence comprehension accuracy (NOCANACC). The model fit statistics are as follows: X2 = 107.40, df = 94, p = .94, RMSEA = .035 (90% CI = .00 ‒ .063), CFI = .980, SRMR = .060; NPD.
However, the groups parted company in a dramatic fashion with respect to the noncanonical sentences. LTM had a moderate and significant influence on the sentence comprehension of TD children but for the children with DLD it had no significant effect. Strikingly, the indirect effect of LTM was 191% greater for the TD group than the DLD group.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
Figure 34.2 Structural equation model of the relationships between cognitive processing and sentence comprehension with working memory as a mediator Note: complex working memory (cWM) mediating the relationships between fluid reasoning (FLD R), controlled attention (CATT, long-term memory language knowledge (LTM-LK); a propensity score (PROP S) control; and canonical sentence comprehension accuracy (CANACC) and noncanonical sentence comprehension accuracy (NOCANACC). Path Stdyx values for the TD and DLDgroups are in parentheses. The model fit statistics are as follows: X2 = 109.56, df = 100, p = .94, RMSEA = .029 (90% CI = .00 ‒ .058), CFI = .986, SRMR = .055
34.5
Putting Things Together: The Memory-Comprehension Connection in Children with DLD
While the overall finding that WM played a vital role in the comprehension of the children with DLD is consistent with previous literature (e.g. Delage & Frauenfelder, 2020; Montgomery & Evans, 2009; Robertson & Joanisse, 2010), these new findings paint a much more nuanced picture of the role of WM than previous research. The typical interpretation is that children with DLD have insufficient storage capacity to support comprehension. But this interpretation assumes a false binary choice between sufficient or insufficient capacity. The dichotomy ignores the role of language LTM and the chunking function of language LTM. Our findings suggest that the children with DLD, like their TD peers, have sufficient WM storage capacity to support the comprehension of at least two-clause canonical and noncanonical sentences. This conclusion comes from the fact that (1) our sentences included just one or two critical clauses and (2) the children with DLD had a 2-chunk WM span while the TD children had a 3-chunk span, spans that were sufficient to support comprehension. The new and important understanding of the comprehension of children with DLD from this study is that comprehension is the product of the structural relationship between language LTM knowledge, which either facilitates or hinders effective chunking of input into fewer integrated linguistic units, and the ability to temporarily store these units in WM.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
789
790
JAMES W. MONTGOMERY ET AL.
Even though the model was a good description of the comprehension of both children with DLD and TD children, two striking differences between the groups were evident. We focus on the role of language LTM and controlled attention. Whereas LTM indirectly influenced both canonical and noncanonical sentence comprehension in the TD children, its influence in the children with DLD was restricted to canonical structures. That LTM influenced the canonical sentence comprehension of the children with DLD is important because it suggests that these children, like TD children, (1) have multiword templates in LTM that correspond to SVO-like structures, (2) can use these templates to chunk input into relevant linguistic chunks (phrases, clauses), and (3) can combine chunks into a coherent canonical sentence structure. These findings and interpretation are in keeping with those of Borovsky et al. (2012), who showed that school-age children and adolescents with DLD use their SVO templates to facilitate real-time sentence processing. Regarding noncanonical sentence comprehension, the role of language LTM was altogether different for the DLD group than the TD group. For the DLD group, LTM exerted no significant influence on comprehension. We took these results to mean that the children possess severely weak, if not nonexistent, noncanonical word order templates (Montgomery et al., 2017, 2018). We further interpreted the findings to be consistent with the view that children with DLD have a significant deficit in implicitly learning noncanonical word order patterns (e.g. Conti-Ramsden et al., 2015; Garraffa et al., 2018; Montgomery et al., 2017). The second striking difference was that both canonical and noncanonical sentence comprehension was mentally effortful for the children with DLD, but not for the TD children. This interpretation is based on the finding that controlled attention significantly contributed to the sentence comprehension of children with DLD, but not the TD children. Because we argue that children with DLD appear to have NVN templates in LTM, it may seem counterintuitive to say that such structures take significant mental effort to comprehend, especially since they are of high frequency (Wells et al., 2009). However, if these children’s NVN templates are weak or unstable, it is not surprising that the children would need to expend significant mental energy to chunk and comprehend the sentences relative to TD children who appear to chunk and comprehend more automatically (Lum et al., 2017; Montgomery, 2000; Montgomery et al., 2009; Montgomery & Evans, 2009; Robertson & Joanisse, 2010). For complex structures, the situation is different. Recall that language LTM had no influence on the children’s comprehension, implying the children with DLD had very weak or even nonexistent word order knowledge of these structures to aid chunking and comprehension. Thus, the children had to rely on the only other mechanism available to them to support comprehension – controlled attention. In the absence of multiword knowledge of complex structures, the children would be forced to process the entirety of the input in a word-by-word manner. Such an inefficient processing approach would lead to severe difficulty managing the temporal and memory constraints of
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
comprehension. Children would likely be slowed down in processing incoming material, which, in turn, would lead to a complete swamping of WM storage.
34.6
Conclusion and Future Directions
The comprehension model we have described here is the first conceptually integrated and empirically validated model of the sentence comprehension abilities of school-age children with DLD and TD children. The model provides new and important insights into children’s comprehension. Language LTM, controlled attention, and fluid reasoning do not appear to influence comprehension independently of each other, for either children with DLD or TD children. Rather each of these mechanisms indirectly influences sentence comprehension by way of their relationship with WM. For TD children, deeper and stronger language knowledge in LTM together with better general pattern recognition abilities lead to relatively efficient/automatic chunking of input, which, in turn, leads to serviceable sentence comprehension. By stark contrast, sentence comprehension by children with DLD is effortful. General pattern recognition abilities play no appreciable role in comprehension. Language LTM appears to be indirectly influential, but only for canonical structures. For noncanonical structures, it provides no support, thus forcing the children to comprehend input in a very inefficient way – word by word. Future research directions are plentiful. Researchers may wish to explore the learning of different multiword patterns as a function of the availability of different semantic/real-world cues and cue combinations in children with DLD. This approach would enable us to evaluate our claim that the learning of multiword units by these children is dependent on the presence of strong semantic-pragmatic cues. If it can be shown that the availability of strong semantic-pragmatic cues in noncanonical structures leads to reliably better comprehension, such findings would yield new and important theoretical implications about the grammar learning abilities of these children and the conditions that promote stable learning/retention of noncanonical word order patterns. An important related issue would be to determine whether such learning translates to more automatic multiword chunking of input, and if so, whether greater automaticity leads to a different relationship between cognitive processing and comprehension. Future research may also wish to investigate the role of declarative memory in these children’s syntactic learning abilities (Hamrick et al., 2018). An obvious extension of our model is to determine whether the structural relationship we observed here holds across a wider age range or whether it may change with age. Adolescents with DLD, for example, having accrued more language processing experience, may acquire stronger representations of both noncanonical and canonical multiword representations. If so, we might expect language LTM to play an even stronger role in adolescent
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
791
792
JAMES W. MONTGOMERY ET AL.
comprehension than it does during the school-age years, that is, more automatic and accurate linguistic chunking of the input, for different structures containing a range of different linguistic cues. If WM is a true mediator of controlled attention and this mediated relationship supports sentence comprehension, it should be possible through neuroimaging techniques to identify those brain regions typically associated with verbal WM (e.g., left prefrontal cortex, anterior cingulate cortex, temporal language area) (Chein et al., 2011; Osaka et al., 2003) and controlled attention related to WM performance (dorsolateral prefrontal cortex, frontoparietal system) (Peterson & Posner, 2012). Researchers should also be able to assess the timing and direction of coactivation of these regions during language processing. Another approach could be to model the effects of training language LTM, controlled attention or both on children’s sentence comprehension abilities using pre-test measures of WM as the mediator. Neuroimaging studies could also be combined with experimental studies to look for changes in strength and timing of neural (co) activation following different kinds of training exposures.
Acknowledgements Preparation of this chapter was supported by grant R01 DC010883 from the National Institute on Deafness and Other Communicative Disorders.
References Abbot-Smith, K., & Tomasello, M. (2006). Exemplar-learning and schematization in a usage-based account of syntactic acquisition. The Linguistic Review, 23, 275–290. Adams, E., Nguyen, A., & Cowan, N. (2018). Theories of working memory: differences in definition, degree of modularity, role of attention and purpose. Language Speech and Hearing Services in Schools, 49, 340–355. Altmann, G., & Kamide, Y. (1999). Incremental interpretation at verbs: Restricting the domain of subsequent reference. Cognition, 73, 247–264. Andrews, G., Ogden, J., & Halford, G. (2017). Resolving conflicts between syntax and plausibility in sentence comprehension. Advances in Cognitive Psychology, 13, 11–27. Archibald, L., & Gathercole, S. (2007). The complexities of complex memory span: Storage and processing deficits in specific language impairment. Journal of Memory and Language, 57, 177–194. Arnon, I., & Christiansen, M. (2017). The role of multiword building blocks in explaining L1-L2 differences. Topics in Cognitive Science, 9, 621–636.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
Arnon, I., & Clark, E. (2011). Why brush your teeth is better than teeth: Children’s word production is facilitated in familiar sentence-frames. Language Learning and Development, 7, 107–129. Arnon, I., McCauley, S., & Christiansen, M. (2017). Digging up the building blocks of language: Age-of-acquisition effects for multiword phrases. Journal of Memory and Language, 92, 265–280. Arnon, I., & Snider, N. (2008). More than words: Frequency effects for multiword phrases. Journal of Memory and Language, 62, 67–82. Baddeley, A. (2012). Working memory: Theories, models, and controversies. Annual Review of Psychology, 63, 1–29. Bannard, C., & Lieven, E. (2012). Formulaic language in L1 acquisition. Annual Review of Applied Linguistics, 32, 3–16. Bannard, C., & Matthews, D. (2008). Stored word sequences in language learning. Psychological Science, 19, 241–248. Barrouillet, P., & Camos, V. (2001). Developmental increase in working memory span: Resource sharing or temporal decay? Journal of Memory and Language, 45, 1–20. Barrouillet, P., Gavens, N., Vergauwe, E., Gaillard, V., & Camos, V. (2009). Working memory span development: A time-based resource-sharing model account. Developmental Psychology, 45, 477–490. Borovsky, A., Elman, J., & Fernald, A. (2012). Knowing a lot for one’s age: Vocabulary skill and not age is associated with anticipatory incremental sentence interpretation in children and adults. Journal of Experimental Child Psychology, 112, 417–436. Burgess, G., Gray, J., Conway, A., & Braver, T. (2011). Neural mechanisms of interference control underlie the relationship between fluid intelligence and working memory span. Journal of Experimental Psychology: General, 140, 674–692. Chater, N., McCauley, S., & Christiansen, M. (2016). Language as skill: Intertwining comprehension and production. Journal of Memory and Language, 89, 244–254. Chein, J., Moore, A., & Conway, A. (2011). Domain-general mechanisms of complex working memory span. NeuroImage, 54, 550–559. Christiansen, M., & Arnon, I. (2017). More than words: The role of multiword sequences in language learning and use. Topics in Cognitive Science, 9, 542–551. Christiansen, M., & Chater, N. (2016). The now-or-never bottleneck: A fundamental constraint on language. Behavioral and Brain Sciences, 39, 1–19. Christiansen, M., & MacDonald, M. (1999). Fractionated working memory: Even in pebbles, it’s still a soup stone. Behavioral and Brain Sciences, 22, 97–98. Conti-Ramsden, G., & Faragher, N. (2001). Psycholinguistic markers for Specific Language Impairment. Journal of Child Psychology and Psychiatry, 6, 741–748.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
793
794
JAMES W. MONTGOMERY ET AL.
Conti-Ramsden, G., Ullman, M., & Lum, J. (2015). The relation between receptive grammar and procedural, declarative, and working memory in specific language impairment. Frontiers in Psychology, 6, 1–11. Cornish, H., Dale, R., Kirby, S., & Christiansen, M. (2017). Sequence memory constraints give rise to language-like structure through iterated learning. PLoS ONE, 12(1), Article e0168532. https://doi.org/10.1371/journal.pone .0168532. Cowan, N., Rouder, J. Blume, C., & Saults, J. (2012). Models of verbal working memory capacity: What does it take to make them work? Psychological Review, 119, 480–499. Cowan, N., Saults, J., & Blume, C. (2014). Central and peripheral components of working memory storage. Journal of Experimental Psychology: General, 143, 1806–1836. Delage, H., & Frauenfelder, U. (2020). Relationship between working memory and complex syntax in children with developmental language disorder. Journal of Child Language, 47, 600–632. Dick, F., Wulfeck, B., Krupa-Kwiatkowski, M., & Bates, L. (2004). The development of complex sentence interpretation in typically developing children compared with children with specific language impairment or early unilateral focal lesions. Developmental Science, 7, 360–377. Ellis Weismer, S., Evans, J., & Hesketh, L. (1999). An examination of verbal working memory capacity in children with specific language impairment. Journal of Speech, Language, and Hearing Research, 42, 1249–1260. Engel de Abreu, P., Conway, A., & Gathercole, S. (2010). Working memory and fluid intelligence in young children. Intelligence, 38, 552–561. Engelhardt, P., Nigg, J., & Ferreira, F. (2017). Executive function and intelligence in the resolution of temporary syntactic ambiguity: An individual differences investigation. The Quarterly Journal of Experimental Psychology, 70, 1263–1281. Engle, R. W. (2018). Working memory and executive attention: A revisit. Perspectives on Psychological Science, 13(2), 190–193. Engle, R., Tuholski, S., Laughlin, J., & Conway, A. (1999). Working memory, short-term memory, and general fluid intelligence: a latent-variable approach. Journal of experimental psychology: General, 128, 309–331. Evans, J., Saffran, J., & Robe-Torres, K. (2009). Statistical learning in children with specific language impairment. Journal of Speech, Language, and Hearing Research, 52, 321–335. Fernald, A., Perfors, A., & Marchman, V. (2006). Picking up speed in understanding: Speech processing efficiency and vocabulary growth across the 2nd year. Developmental Psychology, 42, 98–116. Ferreira, V., & Bock, K. (2006). The functions of structural priming. Language and Cognitive Processes, 21, 1011–1029. Ferretti, R., McRae, K., & Hatherell, A. (2001). Integrating verbs, situation schemas, and thematic role concepts. Journal of Memory and Language, 44, 516–547.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
Finney, M., Montgomery, J., Gillam, R., & Evans, J. (2014). Role of working memory storage and attention focus switching in children’s comprehension of spoken object relative sentences. Child Development Research, 54, 1–11. Friedmann, N., & Novogrodsky, R. (2007). Is the movement deficit in syntactic SLI related to traces or to thematic role transfer? Brain and Language, 101, 5063. Fukuda, K., Vogel, E., Mayr, U., & Awh, E. (2010). Quantity, not quality: The relationship between fluid intelligence and working memory capacity. Psychonomic Bulletin & Review, 17, 673–679. García-Madruga, J. A., Vila, J. O., Gómez-Veiga, I., Duque, G., & Elosúa, M. R. (2014). Executive processes, reading comprehension and academic achievement in 3rd grade primary students. Learning and Individual Differences, 35, 41–48. Garraffa, M., Coco, M., Branigan, H. (2018). Impaired implicit learning of syntactic structure in children with developmental language disorder: Evidence from syntactic priming. Autism & Developmental Language Impairments, 3, 1–15. Gillam, R., Montgomery, J., Evans, J., & Gillam, S. (2019). Cognitive predictors of sentence comprehension in children with and without developmental language disorder: Implications for assessment and treatment. International Journal of Speech-Language Pathology, 21, 240–251. Gillam, R. & Pearson, N. (2004). Test of narrative language. Pro-Ed. Gómez, R., & Gerken, L. (1999). Artificial grammar learning by 1-year-olds leads to specific and abstract knowledge. Cognition, 70, 109–135. Grunow, H., Spaulding, T., Gomez, R., & Plante E. (2006). The effects of variation on learning word order rules by adults with and without language-based learning disabilities. Journal of Communication Disorders, 39, 158–170. Hamrick, P., Lum, J., & Ullman, M. (2018). Child first language and adult second language are both tied to general-purpose learning systems. Proceedings of the National Academy of Sciences of the United States of America, 7, 1487–1492. Hardman, K., & Cowan, N. (2015). Reasoning and memory: People make varied use of the information available in working memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 42, 700–722 Hedenius, M., Persson, J., Tremblay, A., Adi-Japha, E., Veríssimo, J., Dye, C., et al. (2011). Grammar predicts procedural learning and consolidation deficits in children with specific language impairment. Research in Developmental Disabilities, 32, 2362–2375. Hsu, J., & Bishop, D. (2010). Grammatical difficulties in children with specific language impairment: Is learning deficient? Human Development, 53, 264–277. Joanisse, M., & McClelland, J. (2015). Connectionist perspectives on language learning, representation and processing. WIREs Cognitive Science, 6, 235–247.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
795
796
JAMES W. MONTGOMERY ET AL.
Kane, M., Hambrick, D., Tuholski, S., Wilhelm, O., Payne, T., & Engle, R. (2004). The generality of working memory capacity: A latent-variable approach to verbal and visuospatial memory span and reasoning. Journal of Experimental Psychology: General, 133, 189–217. Kidd, E. (2012). Implicit statistical learning is directly associated with the acquisition of syntax. Developmental Psychology, 48, 171–184. Kidd, E., & Arciuli, J. (2016). Individual differences in statistical learning predict children’s comprehension of syntax. Child Development, 87(1), 184–193. Leclercq, A., Majerus, S., Prigent, G., & Maillart, C. (2013). The impact of dual tasking on sentence comprehension in children with specific language impairment. Journal of Speech, Language and Hearing Research, 56, 265–280. Leonard, L. (2014). Children with specific language impairment (2nd ed.). MIT Press. Leonard, L., Deevy, P., Fey, M., & Bredin-Oja, S. (2013). Sentence comprehension in specific language impairment: A task designed to distinguish between cognitive capacity and syntactic complexity. Journal of Speech, Language and Hearing Research, 56, 577–589. Leonard, L., Eyer, J., Bedore, L., & Grela, B. (1997). Three accounts of the grammatical morpheme difficulties of English-speaking children with specific language impairment. Journal of Speech, Language, and Hearing Research, 40, 741–753. Lieven, E., Salomo, D., & Tomasello, M. (2009). Two-year-old children’s production of multiword utterances: A usage-based analysis. Cognitive Linguistics, 20, 481–507. Loaiza, V., & Camos, V. (2018). The role of semantic representations in verbal working memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 44, 863–881. Lum, J., & Conti-Ramsden, G. (2013). Long-term memory: A review and meta-analysis of studies of declarative and procedural memory in specific language impairment. Topics in Language Disorders, 33, 282–297. Lum, J., Conti-Ramsden, G., Morgan, A., & Ullman, M. (2014). Procedural learning deficits in specific language impairment (SLI): A meta-analysis of serial reaction time task performance. Cortex, 51, 1–10. Lum, J., Youssef, G., & Clark, G. (2017). Using pupillometry to investigate sentence comprehension in children with and without specific language impairment. Journal of Speech, Language and Hearing Research, 60, 1648–1660. MacDonald, M., & Christiansen, M. (2002). Reassessing working memory: Comment on Just and Carpenter (1992) and Waters and Caplan (1996). Psychological Review, 109, 35–54. Marshall, C., Marinis, T., & van der Lely, H. (2007). Passive verb morphology: The effect of phonotactics on passive comprehension in typically developing and Grammatical-SLI children. Lingua, 117, 302–320.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
Marshall, C., & van der Lely, H. (2006). A challenge to current models of past tense inflection: The impact of phonotactics. Cognition, 100, 302–320. Marslen-Wilson, W., & Zwitserlood, P. (1989). Accessing spoken words: The importance of word onsets. Journal of Experimental Psychology: Human Perception and Performance, 15, 576–585. Marton, K., Campanelli, L., Eichorn, N., Scheuer, J., & Yoon, J. (2014). Information processing and proactive interference in children with and without specific language impairment. Journal of Speech, Language, and Hearing Research, 57, 106–119. McCauley, S., & Christiansen, M. (2013). Toward a unified account of comprehension and production in language development. Behavioral and Brain Sciences, 36, 38–39. McCauley, S., & Christiansen, M. (2014). Acquiring formulaic language: A computational model. The Mental Lexicon, 9, 419–436. McCauley, S., & Christiansen, M. (2015). Individual differences in chunking ability predict on-line sentence processing. Proceedings from The 37th Annual Conference of the Cognitive Science Society, 1553–1558. McCauley, S., & Christiansen, M. (2017). Computational investigations of multiword chunks in language learning. Topics in Cognitive Science, 9, 637–652. McCauley, S., Isbilen, E., & Christiansen, M. (2017). Chunking ability shapes sentence processing at multiple levels of abstraction. In G. Gunzelmann, A. Howes, T. Tenbrink, & E. Davelaar (Eds.), Proceedings of the 39th Annual Conference of the Cognitive Science Society (pp. 2681–2686). Cognitive Science Society. Montgomery, J. (1995). Sentence comprehension in children with specific language impairment: The role of phonological working memory. Journal of Speech and Hearing Research, 38, 187–199. Montgomery, J. (2000). Relation of working memory to offline and real-time sentence processing in children with specific language impairment. Applied Psycholinguistics, 21, 117–148. Montgomery, J. (2008). Role of auditory attention in the real-time processing of simple grammar by children with specific language impairment: A preliminary investigation. International Journal of Language and Communication Disorders, 43, 499–527. Montgomery, J., & Evans, J. (2009). Complex sentence comprehension and working memory in children with specific language impairment. Journal of Speech, Language, and Hearing Research, 52, 269–288. Montgomery, J., Evans, J., Fargo, J., Schwartz, S., & Gillam, R. (2018). Structural relationship between cognitive processing and syntactic sentence comprehension in children with and without developmental language disorder. Journal of Speech, Language, and Hearing Research, 61, 2950–2976.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
797
798
JAMES W. MONTGOMERY ET AL.
Montgomery, J., Evans, J., & Gillam, R. (2009). Relation of auditory attention and complex sentence comprehension in children with specific language impairment: A preliminary study. Applied Psycholinguistics, 30, 123–151. Montgomery, J., Evans, J., Gillam, R., Sergeev, A., & Finney, M. (2016). “Whatdunit?” Developmental changes in children’s syntactically based sentence comprehension abilities and sensitivity to word order. Applied Psycholinguistics, 37, 1281–1309. Montgomery, J., Gillam, R., Evans, J., Fargo, J., & Schwartz, S. (2019). Comparison of the storage-only deficit and joint mechanism deficit hypotheses of the verbal working memory capacity limitation of children with developmental language disorder. Journal of Speech, Language, and Hearing Research, 62, 3808–3825. Montgomery, J., Gillam, R., Evans, J., & Sergeev, A. (2017). Whatdunit? Sentence comprehension abilities of children with SLI: Sensitivity to word order in canonical and noncanonical sentences. Journal of Speech and Language Hearing Research, 60, 2603–2618. Motallebzadeh, K., & Yazdi, M. (2016). The relationship between EFL learners’ reading comprehension ability and their fluid intelligence, crystallized intelligence, and processing speed. Cogent Education, 3, 1–8. Moyle, M., Karasinski, C., Ellis Weismer, S., & Gorman, B. (2011). Grammatical morphology in school-age children with and without language impairment: A discriminant function analysis. Language, Speech, and Hearing Services in Schools, 42, 550–560. Nee, D., & Jonides, J. (2013). Trisecting representational states in short-term memory. Frontiers in Human Neuroscience, 7, 1–15. Nippold, M. (2017). Reading comprehension deficits in adolescents: Addressing underlying language abilities. Language, Speech, and Hearing Services in Schools, 48, 125–131. Nippold, M., Hesketh, L., Duthie, J., & Mansfield, T. (2005). Conversational versus expository discourse: A study of syntactic development in children, adolescents, and adults. Journal of Speech, Language, and Hearing Research, 48, 1048–1064. Nippold, M., Mansfield, T., Billow, J., & Tomblin, B. (2008). Expository discourse in adolescents with language impairments: Examining syntactic development. American Journal of Speech-Language Pathology, 17, 356–366. Nippold, M., Mansfield, T., Billow, J., & Tomblin, B. (2009). Syntactic development in adolescents with a history of language impairments: A followup investigation. American Journal of Speech-Language Pathology, 18, 241–251. Osaka, M., Osaka, N., Kondo, H., Morishita, M., Fukuyama, H., Aso, T., & Shibasaki, H. (2003). Neural basis of individual differences in working memory: An fMRI study. NeuroImage, 18, 789–797. Öztekin, I., & Cowan, N. (2015). Editorial: Representational states in memory: Where do we stand? Frontiers in Human Neuroscience, 9, 1–3.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
Memory and Sentence Comprehension
Öztekin, I., Davachi, L., & McElree, B. (2010). Are representations in working memory distinct from representations in long-term memory? Neural evidence in support of a single store. Psychological Science, 21, 1123–1133. Peterson, S., & Posner, M. (2012). The attention system of the human brain: 20 years after. Annual Review of Neuroscience, 35, 73–89. Plante, E., Patterson, D., Sandoval, M., Vance, C., & Asbjørnsen, A. (2017). An fMRI study of implicit language learning in developmental language impairment. NeuroImage: Clinical, 14, 277–285. Rice, M., Wexler, K., & Hershberger, S. (1998). Tense over time: The longitudinal course of tense acquisition in children with specific language impairment. Journal of Speech, Language, and Hearing Research, 41, 1412–1431. Robertson, E., & Joanisse, M. (2010). Spoken sentence comprehension in children with dyslexia and language impairment: The roles of syntax and working memory. Applied Psycholinguistics, 31, 141–165. Roid, G., & Miller, L. (1997). Leiter International Performance Scale–Revised. Stoelting. Savage, C., Lieven, E., Theakston, A., & Tomasello, M. (2003). Testing the abstractness of children’s linguistic representations: Lexical and structural priming of syntactic constructions in young children. Developmental Science, 6, 557–567. Shipstead, Z., Harrison, T., & Engle, R. (2016). Working memory capacity and fluid intelligence: Maintenance and disengagement. Perspectives on Psychological Science, 11, 771–779. Squire, L., Knowlton, B., & Musen, G. (1993). The structure and organization of memory. Annual Review in Psychology, 44, 453–495. The state of learning disabilities: Facts, trends and emerging issues (2014). National Center for Learning Disabilities (3rd Ed.). Theakston, A., & Lieven, E. (2017). Multiunit sequences in first language acquisition. Topics in Cognitive Science, 9, 588–603. Thiessen, E. (2017). What’s statistical about learning? Insights from modelling statistical learning as a set of memory processes. Philosophical Transactions of the Royal Society B, 372, 20160056. Ullman M. (2016). The declarative/procedural model: A neurobiology model of language learning, knowledge, and use. In G. Hickok & S. Small (Eds.), Neurobiology of language (pp. 953–968). Elsevier. Ullman, M. (2004). Contributions of memory circuits to language: The declarative/procedural model. Cognition, 92, 231–270. Ullman, M., & Pierpont, E. (2005). Specific language impairment is not specific to language: The procedural deficit hypothesis. Cortex, 41, 399–433. Unsworth, N., & Engle, R. (2007). The nature of individual differences in working memory capacity: Active maintenance in primary memory and
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
799
800
JAMES W. MONTGOMERY ET AL.
controlled search from secondary memory. Psychological Review, 114, 104–132. van der Lely, H., & Stollwerck, L. (1997). Binding theory and grammatical specific language impairment in children. Cognition, 62, 245–290. Victorino, K., & Schwartz, R. (2015). Control of auditory attention in children with specific language impairment. Journal of Speech, Language, and Hearing Research, 58, 1245–1257. Weisleder, A., & Fernald, A. (2013). Talking to children matters: Early language experience strengthens processing and builds vocabulary. Psychological Science, 24, 2143–2152. Wells, J., Christiansen, M., Race, D., Acheson, D., & MacDonald, M. (2009). Experience and sentence processing: Statistical learning and relative clause comprehension. Cognitive Psychology, 58, 250–271. Wilhelm, O., Hildebrandt, A., & Oberauer, K. (2013). What is working memory capacity, and how can we measure it? Frontiers in Psychology, 4, 1–22. Zwitserlood, P. (1989). The locus of the effects of sentential-semantic context in spoken-word processing. Cognition, 32, 25–64.
https://doi.org/10.1017/9781108955638.041 Published online by Cambridge University Press
35 Working Memory and Childhood Deafness Gary Morgan
35.1
Introduction
Working memory (WM) involves the temporary storage and manipulation of incoming sensory information and is considered a crucial cognitive process in developmental psychology. For example, solving the sum 14 5, requires the child to hold the first calculation in mind (10 5) while carrying out the second multiplication (4 5) and then add the two numbers together. WM is strongly linked to language e.g. during vocabulary learning, spoken language comprehension, and reading (Cowan, 2010). Evidence also indicates that WM capacity is associated with children’s academic skills, such as mathematics (Alloway et al., 2005) and reading comprehension (Nouwens et al., 2017). Considering the importance of WM, there are serious repercussions for any WM gaps between deaf and hard of hearing (DHH) children and their typically hearing peers (Pisoni & Cleary, 2003; Kronenberger et al., 2013). To illustrate this, DHH children’s variable performance on WM measures predicts their spoken language processing (Cleary, Pisoni, & Kirk, 2000), reading achievement (Harris & Moreno, 2004), and mathematics skill (Gottardis et al., 2011). Therefore, it is important to understand how WM operates in DHH children and identify possible explanations for observed capacity differences. Most research on WM is performed on individuals who have full access to auditory input and who use spoken language. In terms of the focus of the current chapter, a common explanation for the differences in WM capacity found in DHH children is one based on reduced access to sound and/or spoken language (Pisoni et al., 2011). This is a natural explanation given that the vast majority of DHH children are exposed to spoken language, and their development of WM has been studied in this context. Deafness reduces the quality of the sound signal that DHH children receive via hearing aids or cochlear implants with ensuing difficulties in forming clear phonological representations. This leads to detrimental effects on word
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
802
GARY MORGAN
learning and WM. However, because some DHH children are exposed simultaneously to signed languages (e.g., British Sign Language: BSL) during their spoken language developmental period, it is also important to understand how this influences WM processes (Marshall et al., 2015). The first linguistic descriptions of American Sign Language (ASL) demonstrated a duality of patterning (i.e., an organizational level of phonology and grammar). In particular, linguists have identified that signed languages have a sublexical representation (i.e., signs can be separated into smaller units, a handshape, location, etc.) corresponding to the phonology of speech (Brentari, 2019). A small group of 5 percent of DHH children are native users of sign (deaf children of deaf parents) and acquire signed languages following well attested general milestones (Mitchell & Karchmer, 2004). In contrast, 95 percent of DHH children are born to hearing parents, and if they are exposed to signing, learn to sign in atypical circumstances from nonfluent models (Lu et al., 2016).
35.2
Deafness, Models of Disability, and Cognitive Development
Deafness is diagnosed shortly after birth through universal newborn hearing screening (Harlor & Bower, 2009) or acquired later in the lifespan. This chapter focuses on children who have a newborn diagnosis of deafness (for WM in acquired deafness, see Wingfield et al., 2005). In the UK, congenital deafness occurs in 1–2 infants for every 1,000 live births (National Institute for Health and Clinical Excellence, 2019), with hearing loss ranging from mild to profound. DHH children may use hearing-aids to amplify and access sound around them. The type of hearing aids given, e.g., digital, depends on characteristics of the child’s deafness. DHH children with hearing aids are not be able to hear in the same way as a hearing child and still may struggle to understand a speaker not facing them, too far away, or in high levels of background noise. Today, DHH children with severe to profound hearing losses often do not gain adequate benefit from hearing aids and receive cochlear implants (CIs). CIs are a type of implanted hearing device that converts sound into electrical signals. CIs comprise an internal receiver, which is implanted surgically behind the ear, and an external hearing aid, which contains a speech processor microphone (Pisoni et al., 2011). Many children with CIs achieve spoken language skills that are comparable to their hearing peers, but there continues to be considerable variation (Geers et al., 2016). Because WM deals with the encoding, storage, rehearsal, and retrieval of the phonological representations of spoken words, early studies suggested that the reason for variability in spoken language development of DHH children was due to WM capacity. However, the study of WM in DHH children is complicated by the large heterogeneity of the population. While the development of WM is potentially at risk in DHH
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
Working Memory and Childhood Deafness
children, a range of factors modulate this risk, including: degree of hearing loss and age of deafness onset (Stiles et al., 2012);, parental hearing status (deaf or hearing parents: Marshall et al., 2015), mode of communication (oral and signed communication: Fitzpatrick et al., 2016); additional disabilities (Lina-Granade et al., 2010); and benefit obtained from amplification (years of hearing aid use: Stiles et al., 2012, years of cochlear implant: Pisoni et al., 2011). To make the situation even more complex, these variables do not stand in isolation; rather, they interact to create a set of conditions that is unique for every individual. Baddeley (this volume) writes WM is a complex biological system in all humans. The heterogeneity of the DHH child population adds several more layers of complexity. Before continuing down the main avenue of this chapter - WM and childhood deafness following Keehner and Atkinson (2006), it is necessary to take a side road to mention models of deafness.
35.2.1 The Medical versus Social Views of Deafness Within research on the WM development of DHH children there exist differences in opinion as to what deafness means between the medical and social models (Power, 2005). Briefly stated, the medical model views deafness as a sensory deficit and investigates how subsequent medical interventions can remediate language and WM development delays. Interventions such as CIs attempt to restore functional hearing at a young enough age to reduce risk of delays in access to spoken language and the storage of phonological codes. This assumes that because there is a reciprocal relationship between language and WM, functional hearing will have positive consequences for WM development. Research in the medical model evaluates decreases in children’s age at implantation and increases in duration of usage of CI, as both are associated with gains in WM development (e.g., Soleymani et al., 2014). In the contrasting social model, deafness is viewed as an aspect of human diversity – rather than a hearing “loss,” the social model refers to the importance of signed language, culture and deaf “gain.” For example, the visuospatial WM capacity of DHH adults who sign has been found in some studies to be greater than that of hearing adults (Hall & Bavelier, 2010), although not in all visuospatial WM tasks. Lauro et al. (2014) evaluated deaf native Italian Sign Language (LIS) users on the Corsi Block test and the Visual Pattern Test (VPT). The signers significantly outperformed nonsigners on the Corsi but performed significantly worse on the VPT, which the authors attributed to the static nature of the VPT. Explanations for poor WM in DHH children offered by researchers in the social model are nonsigning parents who offer poor access to a signed language at a young enough age and WM tests being unfairly weighted toward spoken language skills. The main evidence in WM research within the social model comes from studies of DHH children with DHH parents. Between 5 and 10 percent
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
803
804
GARY MORGAN
of DHH infants have DHH parents and predominantly use a signed language to communicate in the home. As we shall see in the section on WM and signed language, these native signers have age-appropriate WM capacity when tested in a signed language. The other 90–95 percent of DHH infants are born to hearing parents with no experience of what deafness means for communication, and the majority use a spoken language with varying levels of success and with related wide-ranging delays in WM development (Mitchell & Karchmer, 2004). The cultural model argues that if the 90–95% of DHH children were to experience native acquisition of a signed language, this would ensure appropriate language and WM development (Hall et al., 2019). However, it is far from clear how this can be achieved and how much exposure to a signed language is needed to ensure not only typical language but also good WM development (López-Crespo et al., 2012). DHH children are thus influenced by factors originating from both the medical and social models. They are diagnosed deaf and receive early interventions focusing on their hearing via medical professionals, as well as come into contact with professionals, families, and other DHH adults and children who provide experiences of signed language and what deafness means from within a social and educational model (Hall et al., 2019). Thus, research in each model puts emphasis on different factors, that is, hearing loss leads to poor establishment of the WM system (medical model); poor access and development of signed language leads to cognitive delays (social model).
35.3
Models of Working Memory
As seen throughout the chapters in this volume, there exist differing accounts of the architecture and functioning of WM. The need to temporarily store information while working on it requires a basic WM architecture; however, beyond this basic function, different WM models posit a single- or dual-component system. Single-component models have storage and processing as one global cognitive capacity (e.g., Just & Carpenter, 1992). In line with single-component models, the assessment of an individual’s WM is often through a span test where a series of sentences are read or listened to. After presentation of each sentence, the individual makes a judgment as to whether the statement is true or false. After all the sentences are completed, they are then asked to recall as many of the first or last words from any of the sentences. All retention of information and processing of language happens in a single-space in these models. In contrast, proponents of dual-component models (e.g., Baddeley, 2000) propose that storage and processing are dealt with by different components: the phonological loop, the visuospatial sketchpad, a memory buffer, and a central executive, which does the processing of any stored information.
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
Working Memory and Childhood Deafness
For the organization of the current chapter, we use the dual-component WM model proposed by Baddeley and Hitch (1974) and Baddeley (2000): (1) the visuospatial sketchpad, a subsystem responsible for the storage and manipulation of visual information, and (2) the phonological loop that allows for the temporary storage and processing of language and/or acoustic information. The phonological loop is broken down further into the phonological store, which holds phonological content for a few seconds before it fades, and an articulatory rehearsal process where phonological information may be rehearsed in order to refresh the memory trace. Lastly, (3) is a central executive, which uses attentional mechanisms for controlling these subsystems. The advantages of the dual-component model (Baddeley, 2000) is that it has been used to explain DHH children’s WM previously (e.g., Nittrouer et al., 2013), and it is able to deal with the mixed modalities of sign/spoken language and the particular phonological-loop issues that the DHH child population presents. With regard to working memory capacity, measures are split into simple span and complex span tasks (Cowan, 2010; Gathercole, 1999). Simple span tasks measure reproduction of verbal or nonverbal stimuli, where tasks tapping storage and processing are called complex span tasks. Complex span tasks such as the backward digit span, listening span, counting span, operational span (only used with older children and adults), and (versions of ) the Stroop task (Cowan, 2010; Gathercole, 1999). Complex WM tasks crucially require rehearsal processes.
35.3.1 Relationship between WM and Language Development A range of interdependent perceptual, motoric, social, and cognitive skills come together to support children’s language development (D’Souza et al., 2017). In many studies, WM is positively associated with several aspects of language development: phonological skills (Alloway et al., 2005), vocabulary learning, (Baddeley, 2003), and grammatical development (Adams & Gathercole, 1995). One explanation being that children’s progress in language learning feeds back to support real-time performance during verbal WM tasks (e.g., Cowan et al., 2012; Gathercole & Adams, 1993). This linkage is seen across a wide age range; for example, a relationship between the capacity of the phonological loop and language acquisition is seen as early as 2 years of age, but articulatory rehearsal does not appear until around age 7 (Gathercole & Hitch, 1993).
35.3.2
Variation in WM Capacity across Deaf and Hearing Individuals Cognitive development including WM capacity is variable across the child population (Gathercole & Adams, 1993), with some accounts linking this to innateness (e.g., Baddeley et al., 1998) as well as variability of early
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
805
806
GARY MORGAN
language learning experience. For example, Kapa and Colombo (2013) report that children who develop bilingualism (by 3 years) have attentional monitoring advantages compared to later bilinguals (after age 3). The explanation offered for this advantage is the need to keep languages separate from an early age. Other individual differences have been linked to WM variability, such as temperament (Wolfe & Bell, 2004). Although WM abilities vary somewhat among typically developing children, variance goes beyond this in a diagnosis of Developmental Language Disorder (DLD). Much research reports WM deficits in children with DLD relative to typically developing peers in both verbal (Vugs et al., 2016, and nonverbal tasks [Henry et al., 2012]). Kapa and Erikson (2019) conclude that these studies emphasize the connection between WM and language abilities. Following this argument, the main focus of the current chapter is the very wide variability in WM development of DHH children and its connection to language development.
35.4
Development of WM in Deaf Children
The study of WM in individuals born deaf has a long history (Conrad, 1970). While studying groups of DHH children is challenging because of their heterogeneity, research consistently shows that as a group, DHH children have lower scores in WM tasks than hearing children of the same chronological age, although different studies show wide variance and different patterns of results dependent on the type of task used. For example, in an early study, DHH children had shorter WM spans than their hearing peers when they had to remember a series of easily named pictures but better recall on spatial locations and pictures that were not easily named (O’Connor & Hermelin, 1972). This suggests that DHH children have difficulty with ordered sequences of items that need to be encoded verbally in speech, performing better when items are presented simultaneously and free recall is allowed (Keehner & Atkinson, 2006). This last possibility suggests that although verbal-sequential representations are viewed as fundamental to the operation of WM in hearing people (Baddeley, 2003), DHH children’s experience means this may not be universal. Whereas hearing children rely primarily on speechbased phonological coding in WM, DHH children, in addition to speechbased phonology, may recruit additional strategies, such as visual codes, fingerspelling codes, and sign codes (Keehner & Atkinson, 2006). DHH children might exploit a mixture of codes depending on their experiences during development. If a DHH child is a fluent user of a signed language, they might be able to exploit sign based phonological codes in WM. An important challenge for research, therefore, is to better understand the factors that lead to different patterns of WM functioning among this population.
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
Working Memory and Childhood Deafness
35.4.1 The WM Components in DHH Children The main WM components outlined in section 35.2 are now discussed in terms of research studies with DHH children. To recapitulate, these were the storage and manipulation of visual information, the phonological loop, store and articulatory rehearsal process, and the central executive. Typically, WM is separated between processes that deal with visuospatial information – the visuospatial sketchpad – and verbal processes in the phonological loop/store of the sound-based phonological code. Baddeley and Hitch’s model applies equally well to deaf and hearing individuals because some flexibility in the definition of verbal processes is sanctioned by the inclusion of phonologically recodable visual items in WM studies, such as print (Baddeley, 2003) or lip-read stimuli (Campbell & Dodd, 1980). Both of these codes, however, are derivatives of spoken language, whereas signed languages are more autonomous.
35.4.2 Signed Languages and WM The WM model in Baddeley (2000) contains the phonological loop for language and visuospatial sketchpad for spatial information. Sign languages are both linguistic and visual-spatial – it is therefore not clear how they are processed in the WM system. A breakthrough for understanding WM in deafness came from studies of how signed languages were processed in DHH adults. Early studies revealed several WM effects during sign language processing: for example, phonological similarity effects during signers’ immediate serial recall of signs (Wilson & Emmorey, 1997) and manual articulatory suppression (MacSweeney et al., 1996). These early studies raised the possibility that there was a sign-based phonological loop (also called a sign loop). Later research investigated the architecture and capacity of such a sign loop. Hall and Bavilier (2010) argued WM was a part of the human genetic endowment and not a consequence of auditory experience. They suggested that DHH individuals did not differ substantially in their basic WM architecture and proposed universal properties of WM for signed and spoken language. It was suggested that a sign-based phonological loop entailed: a phonological store that retains information using sign-based phonological codes (e.g., handshapes) and a manual articulatory rehearsal mechanism that refreshes information in the phonological store. However, in terms of the capacity of a sign loop, some research (Bavelier et al, 2008) suggests sign WM capacity is more similar to visuospatial capacity (5 2) than speech-based verbal capacity (7 2). Bavelier et al. (2008) further explained that differences between WM for signed and spoken language were based on the different information processing capabilities of the visual and phonological modalities. For example, visuospatial coding takes up more WM capacity and phonological information decays at a slower rate than visual information, while rehearsal of visual information is faster than rehearsal of auditory information (Bavelier et al., 2008).
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
807
808
GARY MORGAN
It is important to note that signed language WM studies are typically based on native signers (DHH children/adults with DHH parents). However, the vast majority of DHH children learn signed language as late learners from non-native models, as they have nonsigning hearing parents. There are very few studies of DHH non-native signers’ WM. For example, Wilson et al. (1997) reported signed language digit span is significantly poorer in nonnative signers than in hearing peers’ spoken language digit span. In one of the few studies of this topic in DHH children, Marshall et al. (2015) investigated WM and its relation to language processing in DHH native users of BSL, DHH non-native BSL users, as well as in a control group of typically developing hearing children with no knowledge of sign language. All three groups performed two complex nonverbal WM tasks. The non-native signers performed more poorly than the hearing participants on both WM tasks, while there was no difference in performance between the native signers and the hearing participants. The authors concluded that it is not just exposure to a signed language but also age of acquisition that leads to variation in WM capacity. To explain an age-of-acquisition effect, we can look at research in spoken language development that has proposed a potential bidirectional influence of language and WM (Cowan et al., 2017). We do not know if early development of visual representational knowledge (sign language) in native DHH affords a similar WM advantage over non-native DHH participants (see further discussion of this question in Section 35.5).
35.4.3 Other Visual Language Codes: Speech-Reading DHH individuals, like their hearing counterparts in certain circumstances, have access to a spoken language phonological code through lip-reading (now termed speech-reading). Speech-reading allows DHH and typically hearing children to perceive the correspondence between lip patterns for similar sounds. DHH children who are visually exposed to speech and who learn to speech-read are significantly more likely to develop phonological codes, and studies suggest this code is articulatory, rather than auditory (Dodd et al., 1983), and better speech-reading is associated with longer digit spans and sensitivity to regularity in letter strings (Hanson, 1986). Findings such as these led to a reconceptualization of WM abilities in DHH children whereby researchers concluded span was not contingent on deafness per se but on the type of underlying memory code used on a given task (i.e., sound, visual speech, or sign). Another important issue which has not received sufficient attention to date is the prevalence of bilingual or mixed memory codes in DHH individuals (in adults: Emmorey et al., 2017) and DHH children with spoken and signed language exposure (Davidson et al., 2014). This situation is a pervasive one at home and in school, and these DHH children presumably use both sign and speech-based coding in WM tasks depending on the specific demands. Speech-based codes might permit more efficient serial span performance and signing codes lean toward spatial arrays.
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
Working Memory and Childhood Deafness
35.4.4
Studies of WM in DHH Children Who Predominantly Use Spoken Language While there have been breakthroughs in explaining how signed languages might function in the architecture and development of WM, for the vast majority of DHH children with hearing and speaking parents, spoken language remains the most common target language, although it is very common that hearing parents use lexical signs while they are speaking as a form of sign-supported speech (Lu et al., 2016). Within research that focuses only on a hearing impairment, these DHH children will experience an early reduction of spoken language exposure followed by a medical hearing intervention, that is, a CI. Despite increasing success with acquiring spoken language, there continues to be great variability and the wide range of outcomes are argued to originate in the reciprocal relationship between language and WM (Geers & Sedey, 2011). Young children with good development of language are able to optimally use their phonological skills in verbal WM tasks such as nonword repetition. Good phonological representations are more likely to be retained in the phonological loop and in articulatory rehearsal. In turn, a more efficient phonological loop also facilitates the central executive function (Baddeley, 2003). Conversely, some researchers appeal to the language-WM link via a phonological “bottleneck” hypothesis to explain WM difficulties in DHH children for complex speech-based tasks (Nittrouer et al., 2014). Children with poor language development are at risk of WM difficulties because of their poor language skills. For example, Dillon et al. (2004) reported that DHH children with CIs were only capable of repeating 5 percent of nonwords correctly. Researchers explain this difficulty as weaknesses in sequential processing and in the efficacy of verbal WM (e.g., AuBuchon et al., 2015; Marschark et al., 2007; Torkidsen et al., 2018). Pisoni and Cleary (2003) tested forward and backward digit span tasks in 8–9-year-old CI users and a group of typically hearing peers. The hearing children outperformed the DHH children, but crucially there was less difference in span between forward and backward recall in the DHH children. This suggests reduced use of a spoken phonological WM in the DHH children. Kronenberger et al. (2013) reported that long-term CI users performed more poorly than typically hearing peers on several verbal WM tasks but did not differ on visuospatial WM. Similar findings were reported in a longitudinal study by Harris et al. (2013). In both studies, stimuli were presented visually and required a manual response so that actually hearing the stimuli was not necessary, and speech production as a response was minimalized. Differences in performance were therefore attributed to the internal efficiency of the phonological loop. Executive Function (EF) is a multidimensional construct that includes a set of higher order, cognitive processes related to monitoring, reasoning, and control (Elliott, 2003). EFs include inhibition, cognitive flexibility, and WM. In a longitudinal study of EF development in DHH children, Jones
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
809
810
GARY MORGAN
et al. (2020) measured nonverbal visuospatial WM in DHH and hearing 6–11-year-olds at two time points (24 months apart). Analysis showed both DHH and hearing groups improved performance between the two testing time points, but the hearing group performed significantly better than the DHH group on tasks. The authors also reported that the one variable that predicted improvements in WM over time was vocabulary level (whether signed or spoken). One explanation for the role of language was via subvocal rehearsal. Cowan et al. (2012) proposed that hearing children’s progress in language learning supports real-time performance during verbal WM tasks. Indeed, Bebko and Metcalfe-Haggert (1997) reported the use of rehearsal strategies in adults with high WM spans, but the authors noted that DHH students showed less evidence of using rehearsal, either in oral or signed codes, than hearing students. Inconsistent rehearsal in WM tasks was also reported in British DHH children by MacSweeney et al. (1996). Apart from differences on speech-based WM tasks, researchers have identified other consequences of variability in WM for DHH children, especially for school performance, such as in mathematics and literacy (Nittrouer et al., 2012). As a group, DHH children have lower reading skills, on average, than their hearing peers (e.g., Buchanan-Worster et al., 2020; Conrad, 1970). This gap is still present despite recent advances in hearing technology such as CIs (Buchanan-Worster et al., 2020). Fagan et al. (2007) in a study of DHH children with CIs aged 6–11 years old reported significant correlations between scores on reading and WM digit span measures. Lastly, BuchananWorster et al. (2020) looking at differences in reading skills between DHH children and their hearing peers proposed that the processing of phonological information via speech reading is more effortful for DHH children, which extends the phonological bottleneck proposal to reading skills. A growing set of studies also report concomitant effects of deafness on other aspects of cognitive processing, such as reduced control of attention (Almomani et al., 2021). However, this work highlights a final and important point as to whether wider cognitive differences in DHH children are the outcome of reduced access to sound, that is, “the auditory scaffolding hypothesis” or a language development delay (Campbell et al., 2014).
35.4.5 Summary In many studies of DHH children there are differences in performance on a range of WM tasks whether tested using a spoken or signed language, compared with either hearing peers or native signers, respectively. There are several, not-mutually exclusive, explanations for this performance gap. Many studies report comparisons between DHH and hearing participants based on WM tasks that require the participant to encode and manipulate spoken verbal information before producing a response. In these tasks, DHH children might have increased problems in the processing of verbal information due to early auditory deprivation altering sequential
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
Working Memory and Childhood Deafness
information processing, a language development delay, difficulties with attention orientation, less efficient oral language articulation and finally less use of subvocal verbal rehearsal strategies. It may be the case that one explanation for WM differences may not be applicable to all DHH children. We conclude the chapter with further exploration of the proposal in Pierce et al. (2017): congenital deafness both disrupts the early set-up of the spoken language phonological loop, as well as the use of subvocal rehearsal processes, with consequences for the development and functioning of WM.
35.5
Early Language Experience Modulates WM Skills
35.5.1 Ontogeny The development of the phonological loop in Baddeley’s model of WM has been linked in hearing children to early language input to the infant (Cowan, 2010). Building on Kappa and Colombo (2013), Pierce et al. (2017) proposed that deafness leads to variation in the timing, quality, and/or quantity of early spoken language input, which affected the development of the phonological loop in WM. This disruption in language input due to congenital deafness occurs during a purported sensitive period for the representation and processing of phonology in WM. Note there are similar proposals for a less optimal WM system due to inborn cognitive variance (Gathercole et al. 2005), as well as positive effects of enriched language experience (Kappa & Colombo, 2013), which all leads to the idea of a spectrum of language influences on WM. In the case of DHH infants, even when CI onset is early, there is a period of several months before it becomes fully operational, and speech therapy impacts on the child’s growing linguistic ability. Thus, disruptions in the uptake and storage of spoken phonology, as well as impoverished signed language exposure from hearing parents, during this early sensitive period leads to delays in the setting up of the WM-language relationship. As mentioned previously, children’s development of language supports realtime performance during verbal WM tasks (e.g., Cowan et al., 2012; Gathercole & Adams, 1993). These linkages are seen in the capacity of the phonological loop and language ability in children as young as 2 years of age, as well as through articulatory rehearsal in older children (Gathercole & Hitch, 1993). Early exposure to specific spoken or signed languages allows young children to form phonological representations including visual ones, for example, specific handshapes in signed languages. These phonological representations link to the establishment and maturation of WM. Pierce et al. (2017) argued that many DHH children, despite advances in hearing technology, experience disruptions in the timing, quality, and/or quantity of early language input. A degraded input leads to children’s storage of less optimal phonological representations and increasing processing difficulties, leading to a possible phonological bottleneck. All of these disruptions will have consequences for WM capacities in DHH children.
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
811
812
GARY MORGAN
35.5.2 Do DHH Children Use Less Subvocal Rehearsal in WM Tasks? The second aspect of the language and WM link comes in older children and has been reviewed previously in terms of subvocal rehearsal (Cowan et al., 2012). Doebel and Zelazo (2016) maintained that good language development enables children to automatically process information in WM via integrated language representations, thus freeing up cognitive load to engage in meta-cognitive strategies. Rehearsal in WM tasks is important, especially in those requiring complex manipulation of information, for example, in mental arithmetic. A language-delayed DHH child is less likely to be able to exploit automatic language rehearsal in order to free up cognitive load (Bebko & Metcalfe-Haggert, 1997). The reasons for less developed language rehearsal could stem from delayed language development, as well as reduced learning through social-interaction. Vygotsky (1962) argued that inner speech grew from early social-interaction, and is then modeled by the child in private. In this light, deaf children with hearing parents experience poorer early social interaction (Desjardin, 2006), and ensuing difficulties with language development might reduce the development of rehearsal during WM tasks (MacSweeney et al., 1996). Further research is required to understand what subvocal rehearsal strategies DHH children use during complex WM tasks.
35.6
Conclusion
The chapter has reviewed some of what we understand about WM in DHH children and how language is implicated in this. Because WM influences so many important developmental milestones, it is vital researchers establish how deafness, in all its instantiations, impacts on WM development. DHH children are a very heterogeneous group with some children following typical language and cognitive development and others demonstrating large and long-lasting delays. The role of signed language is important for many situations where there is a barrier to spoken language development. Yet significant numbers of DHH children struggle to develop sufficient spoken or signed language skills in the sensitive period for the set-up and implementation of WM. Many questions remain; for example, little is known about how WM develops and functions in the many (probably the majority) of DHH children who use both signing and speaking, either separately or at the same time. Future directions for research should remove the artificial separation of signing or speaking DHH children. If DHH children use both to communicate at home or at school, WM studies should be able to measure development and efficiency of the integrated WM system. There have been some attempts to look at WM development in small groups of DHH children who use different combinations of signed and spoken language (López-Crespo et al., 2012). The literature is still scarce in studies comparing different communication modes, and studies
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
Working Memory and Childhood Deafness
conducted on the topic have found no evidence of different WM characteristics between DHH children with different communication backgrounds (Koo et al., 2008, Miller, 2001). Equally, as many DHH children reduce or stop signing all together once spoken language begins to get stronger, future research could examine how this shift in language use impacts on WM. In studies of internationally adopted hearing children who switch spoken languages in early childhood (Pierce et al., 2017) there are long-lasting costs for WM skills. Another area of future research concerns the claims that native exposure to a signed language ensures typical development of language and presumably positive consequences for WM development. This is largely unattested in DHH children with hearing non-signing parents. It is not clear how native-like exposure can be achieved and indeed how much exposure to a signed language is needed by a DHH infant from hearing nonfluent parents to ensure not only typical language but also WM development. Linked with the earlier point about signing and speaking DHH children, it is probably the case that when hearing parents begin to sign with a DHH infant they do so at the same time as they use their native spoken language. Is this natural situation where sign supports access to spoken language for a DHH child a positive one for typical WM development? Finally, we do not know if WM delays once detected in DHH children can be remediated. Future directions for research need to move beyond documenting delays in WM development in DHH children and move toward WM training studies. To conclude, the study of deafness has stimulated many alternative ways to think about how the WM system functions and has illuminated what language experiences are implicated in its optimal development.
References Adams, A.-M., & Gathercole, S. E. (1995). Phonological working memory and speech production in preschool children. Journal of Speech & Hearing Research, 38(2), 403–414. https://doi.org/10.1044/jshr.3802.403 Alloway, T. P., Gathercole, S. E., Adams, A. M., & Willis, C. (2005). Working memory abilities in children with special educational needs. Education and Child Psychology, 22. Alloway, T. P., Gathercole, S. E., Adams, A. M., Willis, C., Eaglen, R., & Lamont, E. (2005). Working memory and other cognitive skills as predictors of progress towards early learning goals at school entry. British Journal of Developmental Psychology, 23, 417–426. Almomani, F., Al-Momani, M. O., Garadat, S., Alqudah, S., Kassab, M., Hamadneh, S., Rauterkus, G., & Gans, R. (2021). Cognitive functioning in deaf children using cochlear implants. BMC Pediatrics, 21(1), 71. https:// doi.org/10.1186/s12887-021-02534-1
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
813
814
GARY MORGAN
AuBuchon, A. M., Pisoni, D. B., & Kronenberger, W. G. (2015). Verbal processing speed and executive functioning in long-term cochlear implant users. Journal of Speech, Language, and Hearing Research, 58(1), 151–162. https://doi.org/10.1044/2014_JSLHR-H-13-0259 Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4, 417–423. Baddeley, A. (2003). Working memory and language: An overview. Journal of Communication Disorders, 36, 189–208. Baddeley, A., Gathercole, S. E., & Papagno, C. (1998). The phonological loop as a language learning device. Psychological Review, 105,158–173. Baddeley, A., & Hitch, G. J. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 742–775). Academic Press. Bavelier, D., Newman, A. J., Mukherjee, M., Hauser, P., Kemeny, S., Braun, A., & Boutla, M. (2008). Encoding, rehearsal, and recall in signers and speakers: Shared network but differential engagement. Cerebral Cortex, 18 (10), 2263–2274. https://doi.org/10.1093/cercor/bhm248 Bebko, J., & Metcalfe-Haggert, A. (1997). Deafness, language skills, and rehearsal: A model for the development of a memory strategy. Journal of Deaf Studies and Deaf Education, 2(3), 131–139. https://doi.org/10.1093/ oxfordjournals.deafed.a014319 Brentari, D. 2019. Sign language phonology. Cambridge University Press. Buchanan-Worster, E; MacSweeney, M; Pimperton, H; Kyle, F; Harris, M; Beedie, I; Ralph-Lewis, A., & Hulme, C. (2020) Speechreading ability is related to phonological awareness and single-word reading in both deaf and hearing children. Journal of Speech, Language, and Hearing Research, 63, 3775–3785 Campbell, R., & Dodd, B. (1980). Hearing by eye. Quarterly Journal of Experimental Psychology, 32, 85–99. Campbell, R., MacSweeney, M., & Woll, B. (2014). Cochlear implantation (CI) for prelingual deafness: The relevance of studies of brain organization and the role of first language acquisition in considering outcome success. Frontiers in Human Neuroscience, 8, 834. Cleary, M., Pisoni, D. B., & Kirk, K. I. (2000). Working memory spans as predictors of spoken word recognition and receptive vocabulary in children with cochlear implants. The Volta Review, 102, 259–280. Conrad, R. (1970). Short-term memory processes in the deaf. British Journal of Psychology, 61(2), 179–195. Cowan, N. (2010). The magical mystery four: How is working memory capacity limited, and why? Current Directions in Psychological Science, 19(1), 51–57. Cowan, N., Li, Y., Glass, B., & Saults, J. S. (2017). Development of the ability to combine visual and acoustic information in working memory. Developmental Science, 21, e12635, 1–14.
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
Working Memory and Childhood Deafness
Cowan N., Rouder J. N., Blume C. L., & Saults, J. S. (2012). Models of verbal working memory capacity: What does it take to make them work? Psychological Review, 119, 480–499. D’Souza, D., D’Souza, H., Karmiloff-Smith, A. (2017). Precursors to language development in typically and atypically developing infants and toddlers: The importance of embracing complexity. Journal of Child Language, 44, 591–627. Davidson, K., Lillo-Martin, D., & Chen Pichler, D. (2014). Spoken English language development among native signing children with cochlear implants. Journal of Deaf Studies and Deaf Education, 19(2), 238–250. Desjardin, J. (2006). Family empowerment: Supporting language development in young children who are deaf or hard of hearing. Volta Review, 106, 275–298. Dillon, C. M., Cleary, M., Pisoni, D. B., & Carter, A. K. (2004). Imitation of nonwords by hearing-impaired children with cochlear implants: Segmental analyses. Clinical Linguistics & Phonetics, 18(1), 39–55. Dodd, B., Hobson, P., Brasher, J., & Campbell, R. (1983). Deaf children’s short-term memory for lip-read, graphic and signed stimuli. British Journal of Developmental Psychology, 1(4), 353–364. Doebel, S., & Zelazo, P. D. (2015). A meta-analysis of the Dimensional Change Card Sort: Implications for developmental theories and the measurement of executive function in children. Developmental Review, 38, 241–268. Elliott, R. (2003). Executive functions and their disorders. British Medical Bulletin, 65, 49–59. Emmorey, K., Giezen, M. R., Petrich, J. A. F., Spurgeon, E., & O’Grady Farnady, L. (2017). The relation between working memory and language comprehension in signers and speakers. Acta Psychologica, 177, 69–77. Fagan, M. K., Pisoni, D., Horn, D., & Dillon, C. (2007). Neuropsychological correlates of vocabulary, reading, and working memory in deaf children with cochlear implants. Journal of Deaf Studies and Deaf Education, 12, 461–471 Fitzpatrick, E., Hamel, C., Stevens, A., Pratt, M., Moher, D., Doucet, S., . . . Na, E. (2016). Sign language and spoken language for children with hearing loss: A systematic review. Pediatrics, 137, e20151974 Gathercole, S. E. (1999). Cognitive approaches to the development of shortterm memory. Trends in Cognitive Science, 3, 410–418. Gathercole, S. E., & Adams, A. M. (1993). Phonological working memory in very young children. Developmental Psychology, 29, 770–778. Gathercole, S. E., & Hitch, G. J. (1993). Developmental changes in short-term memory: A revised working memory perspective. In A. F. Collins, S. E. Gathercole, M. A. Conway, & P. E. Morris (Eds.), Theories of memory (p. 189–209). Lawrence Erlbaum Associates. Gathercole, S. E., Tiffany, C., Briscoe, J., Thorn, A. S. C., & the ALSPAC Team (2005). Developmental consequences of phonological loop deficits during
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
815
816
GARY MORGAN
early childhood: A longitudinal study. Journal of Child Psychology and Psychiatry, 46, 598–611. Geers, A. E., & Sedey, A. L. (2011). Language and verbal reasoning skills in adolescents with 10 or more years of cochlear implant experience. Ear and Hearing, 32(1 Suppl), 39S–48S. Geers, A. E., Nicholas, J. G., Tobey, E., & Davidson, L. (2016). Persistent language delay versus late language emergence in children with early cochlear implantation. Journal of Speech, Language and Hearing Research, 59, 155–170. Gottardis, L., Nunes, T., & Lunt, I. (2011). A synthesis of research on deaf and hearing children’s mathematical achievement. Deafness & Education International, 13, 131–150. Hall, M. L., & Bavelier, D. (2010). Working memory, deafness, and sign language. In M. Marshark & P. E. Spencer (Eds.), The Oxford handbook of deaf studies, language, and education (Vol. 2, pp. 458–472). Oxford University Press. Hall, M. L., Hall, W. C., & Caselli, N. K. (2019). Deaf children need language, not (just) speech. First Language, 39(4), 367–395. Hanson, V. L. (1986). Access to spoken language and the acquisition of orthographic structure: Evidence from deaf readers. Quarterly Journal of Experimental Psychology, 38A, 193–212. Harlor, A. D. B., & Bower, C. (2009). Hearing assessment in infants and children: Recommendations beyond neonatal screening. Pediatrics, 124, 1252–1263. Harris, M., & Moreno, C. (2004). Deaf children’s use of phonological coding: evidence from reading, spelling, and working memory. Journal of Deaf Studies & Deaf Education, 9(3), 253–268. Harris, M. S., Kronenberger, W. G., Gao, S., Hoen, H. M., Miyamoto, R. T., & Pisoni, D. B. (2013). Verbal short-term memory development and spoken language outcomes in deaf children with cochlear implants. Ear and Hearing, 34(2), 179–192. Henry, L. A, Messer, D. J., & Nash, G. (2012). Executive functioning in children with specific language impairment. Journal of Child Psychology and Psychiatry, 53, 37–45 Jones, A., Atkinson, J., Marshall, C. R., Botting, N., St Clair, M., & Morgan, G. (2020). Expressive vocabulary predicts non-verbal executive function: A 2year longitudinal study of deaf and hearing children. Child Development, 91, e400–14.63. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99(1), 122–149. Kapa, L. L., & Colombo, J. (2013). Attentional control in early and later bilingual children. Cognitive Development, 28(3), 233–246. Kapa, L. L., & Erikson, J. A. (2019). Variability of executive function performance in pre-schoolers with developmental language disorder. Seminars in Speech and Language, 40(4), 243–255.
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
Working Memory and Childhood Deafness
Keehner, M., & Atkinson, J. (2006). Working memory and deafness: Implications for cognitive development and functioning. In Susan J. Pickering (Ed.), Working memory and education. Academic Press. Koo, D., Crain, K., LaSasso, C., & Eden, G. (2008). Phonological awareness and short-term memory in hearing and deaf individuals of different communication backgrounds. Annals of the New York Academy of Sciences, 1145, pp. 83–99. Kronenberger, W. G., Pisoni, D. B., Henning, S. C., & Colson, B. G. (2013). Executive functioning skills in long-term users of cochlear implants: A case control study. Journal of Pediatric Psychology, 38, 902–914. Lauro, L. J. R., Crespi, M., Papagno, C., & Cecchetto, C. (2014). Making sense of an unexpected detrimental effect of sign language use in a visual task. Journal of Deaf Studies and Deaf Education, 19 (3), 358–365. Lina-Granade, G., Comte-Gervais, I., Gippon, L., Nappez, G., Morin E., & Truy, E. (2010). Correlation between cognitive abilities and language level in cochlear implanted children. Cochlear Implants International, 11(Suppl. 1), 328–331. López-Crespo, G., Daza, M. T., & Méndez-López, M. (2012). Visual working memory in deaf children with diverse communication modes: Improvement by differential outcomes. Research in Developmental Disabilities, 33(2), 362–368. Lu, J., Jones, A., & Morgan, G. (2016). The impact of input quality on early sign development in native and non-native language learners. Journal of Child Language, 43, 537–552 MacSweeney, M., Campbell, R., & Donlan, C. (1996). Varieties of short-term memory coding in deaf teenagers. Journal of Deaf Studies and Deaf Education, 1(4), 249–262. Marschark, M., Rhoten, C., & Fabich, M. (2007). Effects of cochlear implants on children’s reading and academic achievement. Journal of Deaf Studies and Deaf Education, 12, 269–282. Marshall, C., Jones, A., Denmark, T., Mason, K., Atkinson, J., Botting, N., & Morgan, G. (2015). Deaf children’s non-verbal working memory is impacted by their language experience. Frontiers in Psychology, 6, 527. Miller, P. (2001). Communication mode and the information processing capacity of Hebrew readers with prelingually acquired deafness. Journal of Developmental and Physical Disabilities, 13(2001), 83–96 Mitchell, R. E., & Karchmer, M. A. (2004). Chasing the mythical ten percent: parental hearing status of deaf and hard of hearing students in the United States. Sign Language Studies, 4(2), 138–163. National Institute for Health and Clinical Excellence (NICE). (2019). Cochlear implants for children and adults with severe to profound deafness. NICE technology appraisal guidance 566. [online]. https://www.nice .org.uk/guidance/ta566 Nittrouer, S., Sansom, E., Low, K., Rice, C., & Caldwell-Tarr, A. (2014). Language structures used by kindergartners with cochlear implants:
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
817
818
GARY MORGAN
Relationship to phonological awareness, lexical knowledge and hearing loss. Ear and Hearing, 35(5), 506–518. Nittrouer, S., Caldwell, A., Lowenstein, J.H., Tarr, E., & Holloman, C. (2012). Emergent literacy in kindergartners with cochlear implants, Ear Hear, 33 (6) 683–697. Nittrouer, S., Caldwell-Tarr, A., & Lowenstein, J. H. (2013). Working memory in children with cochlear implants: Problems are in storage, not processing. International Journal of Pediatric Otorhinolaryngology, 77(11), 1886–1898. Nouwens, S., Groen, M. A., & Verhoeven, L. (2017). How working memory relates to children’s reading comprehension: The importance of domainspecificity in storage and processing. Reading and Writing, 30(1), 105–120. O’Connor, N., & Hermelin, B. (1972). Seeing and hearing and time and space. Perception and Psychophysics, 11, 46–48. Pierce, L., Genesee, F., Delcenserie, A., & Morgan, G. (2017). Variations in phonological working memory: Linking early language experiences and language learning outcomes. Applied Psycholinguistics, 38, 1265–1300. Pisoni, D. B., & Cleary, M. (2003). Measures of working memory span and verbal rehearsal speed in deaf children after cochlear implantation. Ear and Hearing, 24(1 Suppl), 106S–120S. Pisoni, D. B., Kronenberger, W. G., Roman, A. S., & Geers, A. E. ( 2011). Measures of digit span and verbal rehearsal speed in deaf children after more than 10 years of cochlear implantation. Ear and Hearing, 32, 60S– 74S. Power, D. (2005). Models of deafness: Cochlear implants in the Australian daily press. The Journal of Deaf Studies and Deaf Education, 10, 451–459. Soleymani, Z., Amidfar, M., Dadgar, H., & Jalaei, S. (2014). Working memory in Farsi-speaking children with normal development and cochlear implant. International Journal of Pediatric Otorhinolaryngology, 78. Stiles, D. J., McGregor, K. K., & Bentler, R. A. (2012). Vocabulary and working memory in children fit with hearing aids. Journal of Speech, Language, and Hearing Research, 55, 154–167. Torkildsen, J. V., Arciuli, J., Haukedal, C. L., & Wie, O. B. (2018). Does a lack of auditory experience affect sequential learning? Cognition, 170, 123–129. Vugs, B., Knoors, H., Cuperus, J., Hendriks, M., & Verhoeven, L. (2017). Executive function training in children with SLI: A pilot study. Child Language Teaching and Therapy, 33(1), 47–66. Vygotsky, L. (1962). Studies in communication, thought and language. In E. Hanfmann & G. Vakar (Eds.), MIT Press. Wilson, M., & Emmorey, K. (1997). A visuospatial “phonological loop” in working memory: Evidence from American Sign Language. Memory & Cognition, 25(3), 313–320.
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
Working Memory and Childhood Deafness
Wingfield, A., Tun, P. A., & McCoy, S. L. (2005). Hearing loss in older adulthood: What it is and how it interacts with cognitive performance. Current Directions in Psycholological Science, 14, 144–148. Wolfe, C. D., & Bell, M. A. (2004). Working memory and inhibitory control in early childhood: Contributions from physiology, temperament, and language. Developmental Psychobiology, 44(1), 68–83.
https://doi.org/10.1017/9781108955638.042 Published online by Cambridge University Press
819
36 Working Memory Training in the Classroom Tracy Packiam Alloway, Rachel K. Carpenter, Tyler Robinson, and Andrea N. Frankenstein
36.1
Introduction
Working memory is the retention of a small amount of information that is readily accessible for manipulation. It is used for the execution of a wide variety of cognitive tasks including planning, comprehension, reasoning, and problem-solving (Baddeley, 2010). Working memory is critical for a variety of activities in the classroom, from complex subjects such as reading comprehension, mental arithmetic, and word problems to simple tasks like copying from the board and navigating around school (see Alloway & Copello, 2013, for a review). Working memory is also important from kindergarten to the tertiary level (Alloway & Gregory, 2013; Nguyen & Duncan, 2019), and is an excellent predictor of longitudinal academic success (Alloway & Alloway, 2010; Giofrè et al., 2017).
36.1.1 Working Memory and Reading A key foundational skill in reading is known as phonological awareness, where the child must dissect a word into its parts, such as rhyming words with the same initial sounds or the ability to name pictures rapidly. Ruan and colleagues’ (2018) recent meta-analysis indicated that phonological awareness skills predicted reading accuracy, fluency, and comprehension. One explanation for why WM is critical for reading is that we use a “Post-it Note (i.e., available WM capacity)” to keep all the relevant speech sounds in mind, match them up with the corresponding letters, and then combine them to read the words. Children with reading disabilities show a limited capacity for processing and storing information and indicate significant WM deficits (Chen et al., 2018; De Weerdt et al., 2013).
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
Working Memory Training in the Classroom
36.1.2 Working Memory and Mathematics Working memory also plays a central role in mathematical skills (Friso-Van den Bos et al., 2013, for a review). Low WM scores are closely related to poor performance on arithmetic word problems (Alloway & Passolunghi, 2011) and poor computational skills (Lin et al., 2021). The relationship between mathematical skills and WM is mediated by the age of the child, as well as the task. Verbal WM is a reliable indicator of mathematical disabilities in the first year of formal schooling (Träff et al., 2020, for a review). Verbal WM plays a crucial role for basic arithmetic (both to learn arithmetic facts and to retain relevant data such as carried digits) but as children grow older, other factors such as number knowledge and strategies, play a greater role in overall mathematical skill (Thevenot & Oakhill, 2005). Visuospatial memory is also closely linked with mathematical skills (Hawes & Ansari, 2020), which functions as a mental blackboard, supporting number representation, such as placing values and aligning them in columns, and in counting and arithmetic (D’Amico & Guarnera, 2005).
36.1.3 Working Memory versus IQ Both IQ and WM are related to learning, and educators can target and support the cognitive skills that underpin success.1To investigate how well IQ and WM predict reading, writing, and math skills, we evaluated a group of 5-yearolds (n = 194) as they began kindergarten and tracked their progress over a sixyear period. The findings at the first time point (age 5) indicated that WM was a significant predictor of academic success. Children with high WM did well in reading, writing, and math; while those with low WM struggled in these tasks (Alloway et al., 2005). The children were tested again when they were 11 years old to explore the best predictors of learning outcomes over time: WM or VIQ/PIQ. They were also tested on standardized tests of language and math. The results indicated that a student’s WM ability at 5 years of age was a significant predictor of language and math scores six years later (Alloway & Alloway, 2010). This finding is important as it indicates that while IQ is still viewed as a benchmark of success, other skills, such as WM, may provide more useful information on a student’s potential to learn. Working memory is distinct but correlated with IQ (Cain, Oakhill, & Bryant, 2004; Rey-Mermet et al., 2019) and may uniquely predict learning outcomes. For example, WM skills predict a child’s performance in language and math, even after a child’s IQ scores have been statistically accounted for (Gathercole et al., 2006; Swanson & Saez, 2003, for a review). This same pattern is evidenced at the university level, where WM is a better predictor of grades than entrance exams like SAT scores (Hannon, & McNaughton-Cassill, 2011), and WM performance has been predictive of high school retention rates (Fitzpatrick et al., 2015). While WM and IQ are closely related, they represent dissociable skills with unique links to learning outcomes (Alloway & Alloway, 2010).
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
821
822
T R AC Y PAC K I A M A L LOW AY ET A L .
Further, WM may be a better predictor of learning than IQ (Siquara et al., 2018). A common WM test is to remember a sequence of numbers in the reverse order that it was presented. If students struggle in this test, it is not because they cannot count, or understand number magnitude, but perhaps because their “Post-it Note” (or working memory capacity) is not large enough to remember three or four numbers. Working memory is an accurate predictor of learning from kindergarten to college because it measures students’ ability to learn, rather than what they have learned (Lee et al., 2009). In contrast, other measurements like school tests and IQ tests measure stored, accumulated knowledge. If a student understands vocabulary definitions such as “bicycle” or “police,” then they will likely get a high IQ score. However, if they do not know the definitions of these words or perhaps do not articulate them well, this will be reflected in a low IQ score. In this way, IQ tests are very different from WM tests because they measure how much students know and how well they can articulate this knowledge. For example, one research project involved two different schools: one was in an urban, developed area, while the other was in an underprivileged neighborhood (Alloway, Alloway, & Wootan, 2014). Students were tested on their IQ using a vocabulary test, and one of the vocabulary words—police— drew very different responses. Students from the urban school provided definitions relating to safety or uniforms, which corresponded to the examples in the manual. However, those from the underprivileged neighborhood responded with statements like “I don’t like police” or “They are bad because they took my dad away.” Although both responses were drawn directly from the children’s experiences, only one type of answer matched the IQ manual’s definitions. This example illustrates how performance on IQ tests is strongly driven by a child’s background and experiences.
36.2
WM Training Programs
Given the importance of WM in learning, research efforts to understand the impact and efficacy of training this skill, particularly in an educational context have expanded (Colmar et al., 2020; Kerns et al., 2017; Sala & Gobet, 2017 for a review). To begin, we characterize WM training programs into two categories: those that are narrow in scope and those that are broad in scope. We define narrow-scope WM training programs as those that are very similar to a WM test. For example, they require the student to remember numbers in backward order or the location of dots in backward order. In contrast, we define broad-scope WM training programs as those that train WM in the context of broader abilities, such as executive function, attention, or learning skills. A narrow-scope program typically targets one area such as backward digit span; in contrast, a broad-scope program has a wider application like improving overall cognitive function.
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
Working Memory Training in the Classroom
36.3
Transfer Effects
When considering the efficacy of WM training, it is important to investigate the nature of transfer effects. This refers to whether a training program improves anything other than the target skill. Practicing a skill will naturally enhance it, known as practice effects, but it is important to evaluate if the benefits of a WM training program transfer to real world activities. Transfer effects can be classified as near transfer or far transfer. Near transfer refers to improvements that are similar to the training program such as improved performance on WM tests involving verbal and visuospatial stimuli. Some caution should be exercised when considering near transfer effects, as the measures used to assess near transfer effects often share similarities with the activities used in the training programs (Linares et al., 2019). As a result, it can be difficult to ascertain whether improvements are the result of practice effects, or broader, more efficient cognitive functioning. Far transfer effects refer to improvements in skills related to the area of training. In the context of WM training programs, far transfer refers to other executive function skills such as inhibition, updating, planning, and attention. Another far transfer skill that is related to WM is fluid intelligence. Investigation of far transfer effects provide some indication of whether training WM yields improvements in tasks that do not directly mirror the training activities. Another key factor to consider is whether transfer effects are short-lived or long-lasting. In some cases, the improvements may be the result simply of the novelty of using the training program, but in other cases, there is evidence of sustained improvements for specific populations such as children diagnosed with ADHD (Bigorra et al., 2016, for a review). A metaanalysis evaluating the near and far transfer effects of various WM training programs indicated that for the majority of the studies analyzed, there were immediate improvements on intermediate transfer (verbal and visuospatial WM). For far transfer (verbal capabilities, word decoding, reading comprehension, mathematics) there was no evidence of sustained improvements when WM training was compared to a control condition (Melby-Lervåg et al., 2016).
36.4
WM Training Programs in an Educational Context
This section explores the efficacy of WM training programs regarding both near and far transfer effects and programs that are both narrow and broad in scope in the context of education.
36.4.1 Narrow-Scope WM Training Programs One of the most researched narrow-scope commercial programs is the Cogmed Working Memory Training program (CWMT; www.Cogmed.com).
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
823
824
T R AC Y PAC K I A M A L LOW AY ET A L .
CWMT is an adaptive, five-week training that mirrors WM tests. The games include backward digit recall, letter recall, and dot location, and much like a WM test, the user has to recall numbers in backward order, letters, and the locations of dots in a grid. The target age in the majority of the published research on the CWMT is between 8 and 12 years, and has typically been studied on students with ADHD, though a few studies have been conducted with students with learning difficulties, as well as typically developing students (see Kirk et al., 2014 for a review; also Schwaighofer et al., 2015). Research investigating transfer effects with CWMT shows consistent near transfer effects to tests of both verbal and visuospatial WM, in children aged 6–12 years (Aksayli et al., 2019; Kirk et al., 2014 for reviews). However, far transfer effects are harder to substantiate (Anderson et al., 2018 for a review). In skills related to WM such as attention, teacher ratings of attention are typically not different posttraining with CWMT (Mezzacappa & Buckner, 2010). The pattern of parent ratings of attention and ADHD symptoms are similarly mixed, where some have reported differences attributed to the training (Klingberg et al., 2005), while others have found nonsignificant results in this area (Dentz et al., 2020; Green et al., 2012; see Sala & Gobet, 2017; Wu, 2020 for reviews). To date, there has not been evidence that CWMT improves scores in either verbal or nonverbal IQ tests (Kirk et al., 2014 for a review). Similarly, there has not been strong support for far transfer effects of CWMT to learning domains, such as mathematical reasoning or word reading (Holmes et al., 2009), though one study did report improvements in reading comprehension in a special needs population (Dahlin et al., 2008). The persistence of training effects with the CWMT has not been widely investigated; however, Holmes et al. (2009) reported the maintenance of near transfer effects in both verbal and visuospatial WM tests when typically developing students were assessed six months later. It is important to note that Holmes (2009) did not originally compare these transfer effects to a control group, which may have inflated the results. Other narrow-scope programs include the Odd Yellow (Van der Molen et al., 2010), which mirrors the Odd-One-Out test. The user is shown three shapes, has to identify the odd-one-out, and then remember the location of the yellow shape on a grid. Students with intellectual disabilities (IQ: 55–85) trained three times a week, during a five-week period. In the adaptive version, no improvements were found between the training group and the control in either verbal or visuospatial WM tests (no near transfer effects) and no far transfer effects in tests of fluid intelligence, response inhibition, or scholastic abilities. Additional WM programs include computerized variations of the n-back task (BrainTwister, Lumosity), a WM paradigm frequently used in the cognitive literature (Redick & Lindsey, 2013). Hardy and colleagues (2015) examined over 4,000 participants placed into two groups. Some
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
Working Memory Training in the Classroom
participants received cognitive training through 49 games of the Luminosity online platform, while the remaining participants completed crossword puzzles (active control group). The Luminosity training group showed greater improvements in speed of processing, short-term memory, WM, problem solving, and fluid reasoning assessments, and greater improvements in concentration. Overall narrow-scope WM training programs seem to improve near-transfer effects, while the literature supporting far transfer effects is minimal.
36.4.2 Broad-Scope WM Training Programs There are several broad-scope training programs that train WM in the context of learning activities, such as reading or math. Broad-scope WM training programs are important in light of research linking WM to a range of scholastic skills. There are a number of WM training programs that focus on specific outcomes (see Titz & Karbach, 2014, for a review). For example, Cornoldi et al. (2015) administered typically developing 8-to10-year-olds a problem-solving training program that involved WM. They were presented with a numerical math problem, had to solve it, as well as use WM to recall the problem and words and digits in the task. They reported near transfer effects of improvements in a WM updating task; far transfer to arithmetic problem-solving task; and maintenance effects for both these improvements when the students were tested three months later. We have also published research on a broad-scope WM training program known as Jungle Memory, which trains WM in the context of reading, math, and letter recognition. In a pilot study with students with learning difficulties (ages 12–13), half of the students played Jungle Memory, while an active control group received targeted educational support (Alloway, 2012). There were near transfer effects to verbal and visuospatial WM tests and far transfer effects to verbal IQ and a standardized math test. Additionally, Alloway et al. (2013) conducted a larger study of almost 100 students with learning difficulties (ages 8–12). Students were allocated into one of three groups: Nonactive Control; Active Control, where they trained once a week (WMT–Low frequency); and the Training group, where they trained four times a week (WMT–High frequency). All three groups were tested on measures of WM, verbal and nonverbal IQ, and standardized measures of academic attainment before training, retested on the same measures after training, and eight months later. The data indicated near transfer effects in both verbal and visuospatial WM tests for the highfrequency Training group. There were also far transfer effects in verbal and nonverbal IQ tests, as well as spelling, in the high-frequency Training group. Maintenance effects were reported when students were tested eight months later.
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
825
826
T R AC Y PAC K I A M A L LOW AY ET A L .
36.4.3 Working Memory Training in the Classroom Several studies have investigated if WM training is effective in the classroom environment compared to the laboratory. Rode et al. (2014) used a math-based WM training program in mainstream schools. They tested almost 300 third graders and presented them with tasks similar to the operation span task, where they were presented with a math problem to solve (e.g., 2 + 3 = ?) and then a number to remember (e.g., 9). They found near transfer effects to verbal and visuospatial WM tasks, as well as far transfer effects to math, but not reading. They did not measure maintenance effects. Another study conducted within the classroom examined the academic performance of two age-matched groups over the course of two years (Söderqvist & Bergman Nutley, 2015). Students in one classroom (n = 20) completed CWMT, whereas children in the other classroom (n = 22) received education per usual, and their performance on math and reading standardized tests were the outcome measures. At grade 6, reading and math improved for the training group compared to the control group (medium effect size, Cohen’s d = 0.66, p = 0.045). Similarly, Roberts and colleagues explored if CWMT improved long-term academic outcomes of children 6–7 years of age with low WM capacity compared to usual classroom teaching. They found no improvements in math and reading but did indicate WM improvements, specifically visuospatial memory. Narrow- and broad-scope WM training programs have been investigated both within the classroom and within laboratory settings. Typical WM training programs such as CWMT are costly and time intensive, so we provide specific, feasible strategies educators can use to improve WM functioning for their students.
36.5
Supporting WM Through Strategies
Classroom teachers can make small tweaks in the daily routine of the students to support their learning (Alloway, 2010). The following case studies illustrate how these steps can be implemented in a classroom setting. 1. Detect working memory failures. Is the student struggling to keep up with their peers? Are they beginning to disengage from the activity? Are they acting out in frustration? Once you have identified these signs in a student, you can follow the next two recommendations. 2. Break down information. If an activity exceeds the WM capacity of a student, they will be unable to complete the task. 3. Build long-term knowledge. This process can foster automaticity of knowledge in the student, which can ease the likelihood of WM overload.
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
Working Memory Training in the Classroom
36.5.1 Case Study 1 Jimmy, 10 years, had difficulty recalling information as well as completing simple tasks. His writing skills were poor, and his reading comprehension was also much lower than the rest of his classmates. 1. Detecting working memory failures. When the researcher (EC) reviewed his lesson plans, he realized that the time that the class worked on certain assignments changed every day. Jimmy found it difficult to work within this varying schedule and was frequently frustrated. In order to support his learning and keep him apace with his classmates, he started by developing a structured schedule for him. For example, every day at the same time he would complete writing activities, regardless of what the rest of the class was doing. Now that his day was structured, he knew exactly what to anticipate and was mentally ready to tackle his next assignment. His behavior improved as a result. 2. Break down information. During writing assignments, the researcher would break down complex sentences and have Jimmy write one sentence at a time. After each paragraph, he would read it aloud to him and then ask him to read it to me. Eventually he was able to write multiple sentences at a time without prompting and read the paragraph aloud to the researcher before it was read to him. 3. Build long-term knowledge. Each day, the researcher would review the multisyllable words with Jimmy and reinforce meaning. The next day the researcher would ask him what the word meant. The word “because” perplexed him at first. One day they used it in a sentence and read over it together, and the researcher explained the meaning of the word, as well as the proper usage. The next day during Jimmy’s writing assignment, he had to use the word in one of his sentences. He was able to use the word effectively and continued to better its usage throughout the week. Developing a schedule, breaking down the writing assignments, and explaining the meaning of words allowed him to catch up with his classmates in writing and reading assignments.
36.5.2 Case Study 2 Janine, 11 years, was having difficulty with manipulating three- and fourdigit numbers. 1. Detecting working memory failures. When the researcher (EC) first assessed her ability, he realized she was proficient with two-digit multiplication, but the borrowing system with three digits confused her. The same issue was evident with her long division. 2. Break down information. The researcher began with asking her to add multidigit numbers together (e.g., 345 þ 678). After she successfully
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
827
828
T R AC Y PAC K I A M A L LOW AY ET A L .
completed this task, he asked her to multiply the same numbers together, guiding her through each step along the way and talking about each step. They spent a week doing this together. Her homework assignments had to be completed in a quiet room away from televisions and radios. If she had any difficulties with the math problems, she was instructed to make a note next to the problem and move onto the next one (to avoid her adopting incorrect techniques). The following day they would review her notes, and the next homework assignment contained problems that mimicked the ones she had difficulties with. Eventually she was able to confidently perform complex multiplication with ease. Our next challenge was long division. The multiplication sessions involved going over simple division (e.g., 16/4) and going through them step by step. She began only dividing numbers that were even, and eventually integrated simple numbers that would have decimals. After integrating decimals, she was moved on to even-number long division (e.g., 100/5). The researcher went through these step by step with her until she could do them with ease. She began to remember the steps in the process, as well as the rules of long division. Once she demonstrated signs of proficiency in these problems, we moved onto three- and fourdigit long division. After she showed progress with those, she was given integrated decimal division. Her homework assignments mirrored what she learned in class, as well as added a few complex problems. 3. Build long-term knowledge. After a month of learning complex multiplication and division, she was given an assessment. The assessment required her to write down each step of the process for the multiplication and division problems. She was then given math problems that steadily progressed in difficulty. She improved greatly from the start of our sessions. By starting at the basic level of each process, she was able to build the necessary rules to do more complex problems later on. She eventually was teaching herself with ease. By building proper study habits and learning habits she was able to learn and recall information that used to be difficult for her.
36.6
Discussion
Ultimately, the goal of WM training is to improve WM capacity and performance across the full range of tasks presented to students throughout their academic career, if not beyond. Specifically, narrow-scope WM training programs typically yield near transfer effects to verbal and visuospatial WM tasks (Loosli et al., 2012; Melby-Lervåg et al., 2016), while broadscope WM training programs show both near transfer effects and far transfer effects to learning outcomes (Swanson & McMurran, 2018). Of equal importance is some evidence that these improvements in WM
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
Working Memory Training in the Classroom
performance are still present upwards of eight months after the conclusion of the training (Alloway et al., 2013). To examine the feasibility of WM training in the classroom, Holmes and Gathercole (2014) examined teachers who administered WM training to their own pupils based on previous laboratory research. Two trials of WM training were administered to children aged 8–11 and were compared to a control condition. Following training, children in Trial 1 improved significantly in both trained and untrained WM tasks, with effect sizes comparable to those reported in laboratory studies. These findings indicated that teacher-administered training was relatively easy to implement, led to improvements in WM, and significantly enhanced academic performance. Similarly, a feasibility study examining children with neurodevelopmental disorders found that the Caribbean Quest (CQ), an interactive game that targets different aspects of attention and WM, was easily delivered within the school day, enjoyable, and children transferred metacognitive strategies learned in game play into the classroom (Kerns et al., 2017). Moreover, Phakey et al. (2021) developed and evaluated a WM intervention using a combination of mobile phone–based applications and an activity booklet, within a lower socioeconomic setting. The authors concluded that this approach was feasible and effective in improving WM deficits in a lowresource settings. As noted in the case studies, implementing small changes within the classroom environment may lead to significant academic improvements. While larger scale implementation of WM training programs such as CWMT are relatively expensive and time consuming (Colmar et al., 2020), implementing smaller WM programs such as Jungle Memory, Memory Booster, Memory Mates, Memory! and Monster Hunt are cost-effective options that show improvements in near transfer effects. Additionally, detecting where a student is struggling with WM, breaking down instruction into simple components, and building long-term knowledge takes few additional resources to implement.
36.7
Conclusion and Future Directions
Effectively applying WM training to an educational setting will require both targeted training for the base structures of WM, intentional avoidance of memory techniques that might otherwise inhibit the structural improvement of WM systems, and sufficient time dedicated to training the individual child that demonstrates the deficits. One of the first steps in implementing WM training would be to determine if a child is struggling with WM by using the Working Memory Rating Scale (Alloway et al., 2008). If deficits are detected, this may then be followed by the Automated Working Memory Assessment (AMWA) to measure the student’s verbal and visual-spatial memory. After these assessments, the educator should
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
829
830
T R AC Y PAC K I A M A L LOW AY ET A L .
have a better understanding of the student’s strengths and weaknesses to provide targeted support. In addition to the recommendations provided within the case studies and throughout this chapter, please see Alloway (2010) and Gathercole and Alloway (2008) for further information pertaining to specific WM interventions and teacher recommendations (including a table for specific concerns).
References Alloway, T. P. (2010). Improving working memory: Supporting students’ learning. Sage. Alloway, T. P. (2012). Can interactive working memory training improve learning? Journal of Interactive Learning Research, 23(3), 197–207. Alloway, T. P., & Alloway, R. G. (2010). Investigating the predictive roles of working memory and IQ in academic attainment. Journal of Experimental Child Psychology, 106(1), 20–29. Alloway, T. P., Alloway, R. G., & Wootan, S. (2014). Home sweet home: Does where you live matter to working memory and other cognitive skills? Journal of Experimental Child Psychology, 124, 124–131 Alloway, T. P., Bibile, V., & Lau, G. (2013). Computerized working memory training: Can it lead to gains in cognitive skills in students? Computers & Human Behavior, 29(3), 632–638. Alloway, T. P., & Copello, E. (2013). Working memory: The what, the why, and the how. The Australian Educational and Developmental Psychologist, 30 (2), 105–118. Alloway, T. P., Gathercole, S. E., Adams, A. M., Willis, C., Eaglen, R., & Lamont, E. (2005). Working memory and other cognitive skills as predictors of progress towards early learning goals at school entry. British Journal of Developmental Psychology, 23, 417–426. Alloway, T. P., Gathercole, S. E., & Kirkwood, H. (2008). Working memory rating scale. Pearson. Alloway, T. P., Gathercole, S. E., Kirkwood, H., & Elliott, J. (2009). The cognitive and behavioral characteristics of children with low working memory. Child Development, 80(2), 606–621. Alloway, T. P., & Gregory, D. (2013). The predictive ability of IQ and working memory scores in literacy in an adult population. International Journal of Educational Research, 57, 51–56 (2013). Alloway, T. P., & Passolunghi, M. C. (2011). The relationship between working memory, IQ, and mathematical skills in children. Learning and Individual Differences, 21(1), 133–137. Aksayli, N. D., Sala, G., & Gobet, F. (2019). The cognitive and academic benefits of Cogmed: A meta-analysis. Educational Research Review, 27, 229–243.
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
Working Memory Training in the Classroom
Anderson, P. J., Lee, K. J., Roberts, G., Spencer-Smith, M. M., Thompson, D. K., Seal, M. L., . . . & Pascoe, L. (2018). Long-term academic functioning following Cogmed working memory training for children born extremely preterm: A randomized controlled trial. The Journal of Pediatrics, 202, 92–97. Baddeley, A. (2010). Working memory. Current Biology, 20(4), R136–R140. Bigorra, A., Garolera, M., Guijarro, S., & Hervás, A. (2016). Long-term fartransfer effects of working memory training in children with ADHD: A randomized controlled trial. European Child & Adolescent Psychiatry, 25(8), 853–867. Cain, K., Oakhill, J., & Bryant, P. (2004). Children’s reading comprehension ability: concurrent prediction by working memory, verbal ability and component skills. Journal of Educational Psychology, 96, 31–42. Chen, X., Ye, M., Chang, L., Chen, W., & Zhou, R. (2018). Effect of working memory updating training on retrieving symptoms of children with learning disabilities. Journal of Learning Disabilities, 51(5), 507–519. Colmar, S., Double, K., Davis, N., Sheldon, L., Phillips, N., Cheng, M., & Briddon, S. (2020). Memory mates: An evaluation of a classroom-based, student-focused working memory intervention. Journal of Psychologists and Counsellors in Schools, 30(2), 159–171. Cornoldi, C., Carretti, B., Drusi S., & Tencati, C. (2015) Improving problem solving in primary school students: The effect of a training programme focusing on metacognition and working memory. British Journal of Educational Psychology, 85(3), 424–439. doi: 10.1111/bjep.12083 Dahlin, E., Nyberg, L., Bäckman, L., & Neely, A. S. (2008). Plasticity of executive functioning in young and older adults: Immediate training gains, transfer, and long-term maintenance. Psychology and Aging, 23(4), 720–730. doi: 10.1037/a0014296 D’Amico, A., & Guarnera, M. (2005). Exploring working memory in children with low arithmetical achievement. Learning and Individual Differences, 15, 189–202. De Weerdt, F., Desoete, A., & Roeyers, H. (2013). Working memory in children with reading disabilities and/or mathematical disabilities. Journal of Learning Disabilities, 46(5), 461–472. Dentz, A., Guay, M. C., Parent, V., & Romo, L. (2020). Working memory training for adults with ADHD. Journal of Attention Disorders, 24(6), 918–927. Dietrichson, J., Bøg, M., Filges, T., & Klint Jørgensen, A. M. (2017). Academic interventions for elementary and middle school students with low socioeconomic status: A systematic review and meta-analysis. Review of Educational Research, 87(2), 243–282. Fitzpatrick, C., Archambault, I., Janosz, M., & Pagani, L. S. (2015). Early childhood working memory forecasts high school dropout risk. Intelligence, 53, 160–165.
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
831
832
T R AC Y PAC K I A M A L LOW AY ET A L .
Friso-Van den Bos, I., Van der Ven, S. H., Kroesbergen, E. H., & Van Luit, J. E. (2013). Working memory and mathematics in primary school children: A meta-analysis. Educational Research Review, 10, 29–44. Gathercole, S., & Alloway, T. P. (2008). Working memory and learning: A practical guide for teachers. Sage. Gathercole, S. E., Alloway, T. P., Willis, C., & Adams, A. M. (2006). Working memory in children with reading disabilities. Journal of Experimental Child Psychology, 93, 265281. Green, C. T., Long, D. L., Green, D., Iosif, A. M., Dixon, J. F., Miller, M. R., et al. (2012). Will working memory training generalize to improve offtask behavior in children with attention-deficit/hyperactivity disorder? Neurotherapeutics, 9, 639–648. Giofrè, D., Borella, E., & Mammarella, I. C. (2017). The relationship between intelligence, working memory, academic self-esteem, and academic achievement. Journal of Cognitive Psychology, 29(6), 731–747. Hannon, B., & McNaughton-Cassill, M. (2011). SAT performance: Understanding the contributions of cognitive/learning and social/personality factors. Applied Cognitive Psychology, 25(4), 528–535. Hardy, J. L., Nelson, R. A., Thomason, M. E., Sternberg, D. A., Katovich, K., Farzin, F., & Scanlon, M. (2015). Enhancing cognitive abilities with comprehensive training: A large, online, randomized, active-controlled trial. PLoS ONE, 10(9), e0134467. Hawes, Z., & Ansari, D. (2020). What explains the relationship between spatial and mathematical skills? A review of evidence from brain and behavior. Psychonomic Bulletin & Review, 1–18. Holmes, J., & Gathercole, S. E. (2014). Taking working memory from the laboratory into schools. Educational Psychologist 34(4), 440–450. Holmes, J., Gathercole, S. E., & Dunning, D. L. (2009). Adaptive training leads to sustained enhancement of poor working memory in children. Developmental Science, 12(4), F9–F15 (2009). Kerns, K. A., Macoun, S., MacSween, J., Pei, J., & Hutchison, M. (2017). Attention and working memory training: A feasibility study in children with neurodevelopmental disorders. Applied Neuropsychology: Child, 6(2), 120–137. Kirk, H. E., Gray, K., Riby, D., & Cornish, K. M. (2015). Cognitive training as a resolution for early executive function difficulties in children with intellectual disabilities. Research in Developmental Disabilities, 38, 145–160. Klingberg, T., Fernell, E., Olesen, P. J., Johneson, M., Gustafsson, P., Dahlström, K., & Westerberg, H. (2005). Computerized training of working memory in children with ADHD: A randomized, controlled trial. Child and Adolescent Psychiatry, 44(2), 177–186. Lee, K., Lee, M. P., Ang, S. Y., & Stankov, L. (2009). Do measures of working memory predict academic proficiency better than measures of intelligence? Psychology Science.
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
Working Memory Training in the Classroom
Lin, X., Peng, P., & Luo, H. (2021). The deficit profile of elementary students with computational difficulties versus word problem-solving difficulties. Learning Disability Quarterly, 44(2), 110–122. Linares, R., Borella, E., Lechuga, M. T., Carretti, B., & Pelegrina, S. (2019). Nearest transfer effects of working memory training: A comparison of two programs focused on working memory updating. PLoS ONE, 14(2), e0211321. Loosli, S. V., Buschkuehl, M., Perrig, W. J., & Jaeggi, S. M. (2012). Workingmemory training improves reading processes in typically developing children. Child Neuropsychology, 18(1), 62–78. Melby-Lervåg, M., Redick, T. S., & Hulme, C. (2016). Working memory training does not improve performance on measures of intelligence or other measures of “far transfer” evidence from a meta-analytic review. Perspectives on Psychological Science, 11(4), 512–534. Mezzacappa, E., & Buckner J. C. (2010). Working memory training for children with attention problems or hyperactivity : A school-based pilot study. School Mental Health, 2(4), 202–208. Nguyen, T., & Duncan, G. J. (2019). Kindergarten components of executive function and third grade achievement: A national study. Early Childhood Research Quarterly, 46, 49–61. Phakey, N., Sharma, S., Garg, D., Mukherjee, S. B., Sapra, S., Wadhawan, A. N., & Shukla, G. (2021). Development and evaluation of a working memory intervention kit in children with epilepsy in low-resource settings. The Indian Journal of Pediatrics, 1–4. Redick, T. S., & Lindsey, D. R. (2013). Complex span and n-back measures of working memory: A meta-analysis. Psychonomic Bulletin & Review, 20(6), 1102–1113. Rey-Mermet, A., Gade, M., Souza, A. S., Von Bastian, C. C., & Oberauer, K. (2019). Is executive control related to working memory capacity and fluid intelligence? Journal of Experimental Psychology: General, 148(8), 1335. Rode, C., Robson, R., Purviance, A., Geary, D., & Mayr, U. (2014). Is working memory training effective? A study in a school setting. PLoS ONE, 9(8), 1–8. Roberts, G., Quach, J., Spencer-Smith, M., Anderson, P. J., Gathercole, S., Gold, L., . . . & Wake, M. (2016). Academic outcomes 2 years after working memory training for children with low working memory: A randomized clinical trial. JAMA Pediatrics, 170(5), e154568–e154568. Ruan, Y., Georgiou, G. K., Song, S., Li, Y., & Shu, H. (2018). Does writing system influence the associations between phonological awareness, morphological awareness, and reading? A meta-analysis. Journal of Educational Psychology, 110(2), 180. Sala, G., & Gobet, F. (2017). Working memory training in typically developing children: A meta-analysis of the available evidence. Developmental Psychology, 53(4), 671.
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
833
834
T R AC Y PAC K I A M A L LOW AY ET A L .
Schwaighofer, M., Fischer, F., & Buhner, M. (2015). Does working memory training transfer? A meta-analysis including training conditions as moderators. Educational Psychologist, 50(2), 138–166. Siquara, G. M., dos Santos Lima, C., & Abreu, N. (2018). Working memory and intelligence quotient: Which best predicts on school achievement? Psico, 49(4), 365–374. Söderqvist, S., & Bergman Nutley, S. (2015). Working memory training is associated with long term attainments in math and reading. Frontiers in Psychology, 6, 1711. Swanson, H. L., & McMurran, M. (2018). The impact of working memory training on near and far transfer measures: Is it all about fluid intelligence? Child Neuropsychology, 24(3), 370–395. Swanson, H. L., & Saez, L. (2003). Memory difficulties in children and adults with learning disabilities. In H. L. Swanson, S. Graham, & K. R. Harris (Eds.), Handbook of learning disabilities (pp. 182–198). Guildford Press. Thevenot, C., & Oakhill, J. (2005). The strategic use of alternative representations in arithmetic word problem solving. The Quarterly Journal of Experimental Psychology, 58A, 1311–1323. Titz, C., & Karbach, J. (2014). Working memory and executive functions: Effects of training on academic achievement. Psychological Research, 78(6), 852–868 (2014). Träff, U., Olsson, L., Östergren, R., & Skagerlund, K. (2020). Development of early domain-specific and domain-general cognitive precursors of high and low math achievers in grade 6. Child Neuropsychology, 26(8), 1065–1090. Van der Molen, M., Van Luit, J., Van der Molen, M., Klugkist, I., & Jongmans, M. (2010). Effectiveness of a computerised working memory training in adolescents with mild to borderline intellectual disabilities. Journal of Intellectual Disability Research, 54(5), 433–447. Wu, Y. (2020). Does Cogmed working memory training improve school-age ADHD children’s academic achievement? Cambridge Educational Research eJournal 7, 141–167.
https://doi.org/10.1017/9781108955638.043 Published online by Cambridge University Press
37 Working Memory and Classroom Learning Joni Holmes, Elizabeth M. Byrne, and Agnieszka J. Graham
37.1
Introduction
Several children in a typical classroom experience persistent learning difficulties that are likely to reflect weak cognitive skills in one or more areas that include phonological processing, short-term and working memory, and/or executive function (e.g., Bishop & Snowling, 2004; Holmes et al., 2020; Peng et al., 2018; Swanson & Jerman, 2006; Yeniad et al., 2013). There are multiple definitions of working memory (see Cowan, 2017, for a review). We do not subscribe to any particular theoretical framework in this chapter, instead preferring to view working memory as a higher-order cognitive system that combines short-term storage with the attentional capacity to update and process information (e.g., Allen et al., 2012; Baddeley, 2010; Cowan, 2017). We classify tasks measuring storage-only (e.g., digit span) as simple span tasks, and those that explicitly involve the simultaneous processing of information alongside storage as complex span tasks (e.g., listening span). Finally, different ways to help children with poor working memory are discussed, including an overview of current ideas about memory enhancement by training and brain stimulation (e.g., Byrne et al., 2020), as well as more practical ways for teachers to use action to improve children’s instruction-following.
37.2
Working Memory and Learning
In this section, we briefly review research that has examined links between working memory and children’s attainments in two key areas of academic learning – reading and mathematics – and present some of the contemporary debates in the field.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
836
JONI HOLMES ET AL.
37.2.1 Reading and Mathematics Impairments in working memory are common in poor readers (e.g., Carretti et al., 2009; Jeffries & Everatt, 2004; Kudo et al., 2015). These deficits are predominantly verbal in nature and extend across both simple and complex verbal span tasks (e.g., Cain et al., 2004; Catts et al., 2002; Gathercole et al., 2003; Kudo et al., 2015; Peng & Fuchs, 2016; Pimperton & Nation, 2010; Swanson & Ashbaker, 2000; Swanson et al., 2009; Wang & Gathercole, 2013). It has been proposed that working memory supports both the maintenance and assembly of phonological segments in the process of word identification (Preåler et al., 2014), and the integration of phonological representations and lexical-semantic knowledge from longterm memory for reading comprehension (Cantor & Engle, 1993; Swanson & Ashbaker, 2000). Mathematics outcomes are typically more closely associated with visuospatial than verbal simple and complex span tasks (e.g., Holmes et al., 2008; Li & Geary, 2013; McLean & Hitch, 1993; Meyer et al., 2010). This domainspecific profile is common among children with specific impairments in mathematics (Child et al., 2019; Cirino et al., 2018; Szucs et al., 2013; Willcutt et al., 2013). Domain-general deficits are more typical of children with comorbid reading and mathematics problems (e.g., Bull & Johnston, 1997; Hitch & McAuley, 1991; Passolunghi & Siegel, 2001; Swanson & BeebeFrankenberger, 2004; Van de Weijer-Bergsma et al., 2015). Working memory may contribute to mathematical development, enabling children to remember number bonds to commit to a network of arithmetic facts in long-term memory (Geary, 1993, 2003). It may also support the storage of problem information, and both the storage and retrieval of partial results for simple and complex mathematical tasks (De Smedt et al., 2009; Passolunghi et al., 2008; Peng et al., 2016).
37.2.2 Contemporary Debates There is debate concerning the extent to which associations between working memory and learning reflect a direct causal association. The notion that children’s learning problems originate within working memory has been challenged (e.g., Gathercole & Holmes, 2014). Contemporary views suggest that: (1) there is no one-to-one mapping between a specific deficit in a cognitive skill and a particular outcome (Astle & Fletcher-Watson, 2020), and (2) links between cognitive development and academic outcomes are ubiquitous (Kievit et al., 2019; Mareva et al., 2021; Peng & Kievit, 2019). It is possible, too, that working memory capacity limitations constrain only the learning of novel information (e.g., Chandler & Sweller, 1991, Paas et al., 2003, 2004). These points are considered below. Working memory is part of a broader system of higher-order cognitive control that receives input from perceptual systems that process phonological and visual material. The quality of the inputs from associated
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
perceptual systems affects the quality of their representations in working memory: poor perceptual processing leads to the storage of degraded information (Gathercole & Holmes, 2014). Consistent with this view, some have argued that there are no causal links between working memory and reading, and that their association is mediated by a common contribution of phonological processing skills (Melby-Lervåg, Lervåg, et al., 2012; MelbyLervåg, Lyster, et al., 2012). Working memory is one of many executive functions mediated by frontal networks in the brain (e.g., Duncan & Owen, 2000). Any disruption or weaknesses in these connections will disrupt multiple higher-order control functions including working memory (Gathercole & Holmes, 2014). As such, observing working memory impairments in children with reading and mathematics problems is not sufficient to conclude that the source of their difficulties arises within working memory. Supporting this, deficits in other executive functions, including inhibitory control and attention switching, are common in children with reading and mathematics difficulties (Booth et al., 2010; Bull & Scerif, 2001; Caretti et al., 2009; Follmer, 2018; Landerl & Kölle, 2009; Rotzer et al., 2009; Szucs et al., 2013; Yeniad et al., 2013). Working memory deficits observed among children with specific reading or mathematical problems provide support for the hypothesis that memory systems are critical for academic learning (Szucs et al., 2013; Wang & Gathercole, 2013). This aligns with core deficit theory: the assumption that a set of characteristics can be explained by a single mechanistic impairment. Core deficit theories are appealing due to both their simplicity and implications for intervention, but they are challenged by evidence for equifinality, the notion that specific learning problems can arise through multiple causes (see Astle & Fletcher-Watson, 2020, for a review). For example, some struggling readers have difficulties in applying letter-sound correspondences to decode words, the primary role ascribed to phonological processing in reading (Castles & Friedmann, 2014). However, other poor readers have no difficulties with decoding, but instead struggle with reading comprehension, an aspect of reading assumed to be supported by other skills such as working memory (Cain, Oakhill, & Lemmon, 2004; Nation, 1999). Emerging evidence suggests that the development of cognitive and academic skills might be reciprocal: the development of one might influence the development of the other (e.g., Peng & Kievit, 2020; Van Der Maas et al., 2006). As such, working memory deficits might be both the consequence and cause of learning problems. Consistent with this, Nation (1999) suggests that verbal complex span deficits are a consequence of, rather than a contributor to, the language processing problems of children with poor reading comprehension. Mareva et al. (2021) have also shown that links between academic outcomes and measures of working memory are mutually reinforcing in both typical and struggling learners. Finally, it has been suggested that working memory capacity limitations constrain the learning of novel information, but that knowledge stored in
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
837
838
JONI HOLMES ET AL.
long-term memory might provide a resource for working memory when performance becomes skilled (Chandler & Sweller, 1991; Paas et al., 2004). According to cognitive load theory, skilled performance develops through the construction of increasingly complex schemas stored in long-term memory, which are formed by combining elements of lower-level schemas into higher-level schemas. Sweller and colleagues (e.g., Paas et al., 2003) argued that knowledge organized in schemas allows learners to categorize multiple interacting elements of information as a single episode, thus freeing working memory capacity. After extensive practice, schemas become more automated, thereby allowing learners to further bypass working memory capacity limitations. In this way, as learning progresses, the impact of working memory on learning is diminished.
37.3
Working Memory Constraints on Classroom Learning
Notwithstanding the debates around the causal role of working memory in learning, the capacity to simultaneously process and store information is vital for many classroom activities. Seminal work by Gathercole demonstrated that children with working memory difficulties struggled in learning activities that placed heavy demands on working memory (Gathercole & Alloway, 2008; Gathercole et al., 2006, 2008). Common problems included forgetting classroom or learning-activity-related instructions, place-keeping errors such as missing out letters in words, and failing to cope with the storage and processing demands imposed in structured learning activities. These failures often resulted in the abandonment or noncompletion of learning activities in the classroom, which Gathercole and colleagues suggested may be the cause of slow rates of learning in children with working memory problems. In this section, we focus on the role of working memory in supporting instruction following, a learning-related skill that has received significant attention over the last decade.
37.3.1 Working Memory and Instruction-Following One of the most crucial aspects of classroom learning is following spoken instructions given by the teacher. Teacher instructions are often multistep, containing both vital information about the content of learning activities and sequences of actions to perform to achieve success in said activities (e.g., put away your science book and writing materials, take all the equipment you have used back to the store cupboard, wash your hands, and then join the line by the door). To perform these actions, children must be able to remember the different parts of the instruction while carrying out the various steps – an activity that is challenging and prone to error (e.g., Gathercole & Alloway, 2008; Osterberg & Blaschke, 2005; Wickens et al., 2008). One important constraint on successful instruction-following is the
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
capacity of working memory to retain critical information bridging the period while instructions are being received through to their successful completion (e.g., Engle et al., 1991; Gathercole, Durling, et al., 2008). When the time needed to execute an extended ongoing sequence exceeds the temporal duration of working memory (estimated at between 2 and 18 seconds; e.g., Baddeley & Scott, 1971; Cowan et al., 1997), the rememberer must actively maintain the instruction in working memory, either through rehearsal or through the reactivation of memory traces (Camos et al., 2009). The storage of information may also be supplemented by long-term memory, but because it is less accurate at retaining verbatim than gist information (e.g., Cowan, 2008; Pause et al., 2013; Tulving, 2002), it may not be as effective as working memory in preserving the literal content needed to follow relatively arbitrary instructions. Predictably, following multistep instructions is particularly difficult for children with small working memory capacities. When the instructional load supersedes a child’s working memory capacity, they typically either carry out the first command in a sequence, skip straight to the last step, or simply abandon the task altogether as they are unable to remember all the necessary parts (Gathercole & Alloway, 2008; Gathercole et al., 2006). Failure to act upon instructions can impede progress in individual learning activities, and over time delay the development of desired proficiencies. A growing body of research has established that working memory capacity is a crucial limiting factor in the ability to follow instructions (e.g., Brener, 1940; Engle et al., 1991; Gathercole et al., 2008; Jaroslawska et al., 2018; Jaroslawska, Gathercole, Logie, et al., 2016; Kaplan & White, 1980). The first evidence for a relationship between working memory and following instructions came from Engle et al. (1991). In this study, children aged 7–12-years old completed pencil-and-paper instruction tasks (e.g., Point to the picture at the top of page three and copy it twice) and action-oriented commands (e.g., Sit on the floor with your legs crossed) that resembled the kinds of instructions children typically complete over the course of a school day. They found that instruction following improved with age. Working memory, as measured by word- and sentence-span tasks, was closely associated with the ability to follow multistep instructions, and the strength of this association increased with age. Gathercole et al. (2008) later extended this work using a task in which children followed a sequence of instructions and either manipulated objects placed in front of them or verbally repeated the instructions back to the experimenter. The sequences used by Gathercole et al. (2008) varied in length (i.e., the number of to-be-performed or to-be-repeated steps) but not in linguistic complexity or grammatical construction (e.g., Touch the white bag, then pick up the yellow ruler, then put it in the blue folder). The objectaction pairs were arbitrary (objects were not paired with what might be considered their typical use or action), reducing the contribution of longterm memory or prior knowledge to performance. One of the key findings
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
839
840
JONI HOLMES ET AL.
was that children’s memory for instructions was better when they had to physically perform, rather than verbally repeat, the information at test; an effect described in more detail later in the chapter. The accuracy of performing instructions, but not verbally repeating them back, was also strongly associated with working memory capacity assessed using forward and backward digit recall tasks. These links were strongest between instruction-following and backward digit recall, indicating that the crucial constraint when following practical instructions is not simply the passive storage of information, but rather the processing of the representation of the required action sequence in working memory over the course of memory retrieval. As Gathercole et al. (2008) used simple language to describe to-be-performed actions, the associations between working memory and the ability to follow instructions is unlikely to reflect the mediating impact of working memory on language comprehension. To accommodate the more real-life aspects of following instructions in a classroom environment, in which teacher instructions are often more complex and less predictable, and carried out over longer time periods, Jaroslawska, Gathercole, Logie, et al. (2016) developed 2D virtual versions of Gathercole et al.’s (2008) following-instructions paradigm. These tasks required children to navigate through a virtual school environment to perform a sequence of spoken instructions in multiple rooms (e.g., Go to the IT Suite and touch the red ball, and then go to Mrs. Bolton’s room and pick up the green box). The additional navigation demands of these activities aligned the tasks more closely with the everyday practical demands imposed on children in their school life. Children’s performance on these tasks was closely related to their performance on both simple and complex verbal span tasks, underscoring a potential role for working memory in following instructions over extended periods of activity. Others have used the dual-task methodology to explore the potential role of different subcomponents of the tripartite working memory model (e.g., Baddeley, 1986, 2012; Baddeley & Hitch, 1974) in executing and verbally repeating spoken instructions in young adults. Across these studies, memory for instructions was impaired by concurrent activities taxing phonological short-term memory, spatial short-term memory, and executive control, demonstrating that the encoding and retention of verbal instructions depend on multiple aspects of working memory (e.g., Jaroslawska et al., 2018; Yang et al., 2014, 2016).
37.3.2
The Benefits of Enactment-Based Encoding and Action-Based Recall Having established a link between working memory and following instructions, researchers have more recently discovered that physical movement can enhance the recall of instructions. This work is broadly embedded in theories of embodied cognition: the idea that physical properties of the
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
human body, particularly the perceptual and motor systems, play an important role in cognition (Shapiro, 2007). In this section, we review two distinct mnemonic effects of physical movement on following instructions. One is the advantage associated with action-based retrieval, which is observed when sequences of practical instructions are physically performed, rather than verbally repeated, during recall (e.g., Allen & Waterman, 2015; Gathercole et al., 2008; Jaroslawska, Gathercole, Allen, et al., 2016; Koriat et al., 1990; Yang et al., 2014, 2017). The other is the advantage driven by enactment-based encoding, which occurs when the physical performance of to-be-recalled actions at the time of presentation improves their subsequent recall over short delays (e.g., Allen & Waterman, 2015; Charlesworth et al., 2014; Jaroslawska, Gathercole, Allen, et al., 2016; Waterman et al., 2017; Wojcik et al., 2011; Yang et al., 2017). Jaroslawska, Gathercole, Allen, et al. (2016) provided a nice illustration of both the action advantage and enactment effect with children, demonstrating that both effects may have a common source. They instructed 7–9-yearolds to recall sequences of spoken instructions under presentation and recall conditions that either did or did not involve their physical performance. Memory accuracy was enhanced by carrying out the instructions as they were presented and when they were performed at recall. Crucially, the benefits of action-based recall were reduced following enactment during presentation, suggesting that the positive effects of action at encoding and recall may have a common origin (see also Allen & Waterman, 2015). Jaroslawska, Gathercole, Allen et al. (2016) proposed that when performing physical actions during encoding, or planning for action recall, participants may actively construct action plans that incorporate spatial-motoric information and representations of intended movements (Koriat et al., 1990; Wolpert & Ghahramani, 2000), which are held in a specialized motor store in working memory (see also Jaroslawska et al., 2018; Smyth & Pendleton, 1989, 1990). By this account, planning to execute action sequences at recall did not enhance children’s performance in the enactment condition because the action sequence was already represented in the hypothesized motor store. In a series of three experiments, Waterman et al. (2017) explored how planning for action, self-enactment, and researcher-demonstration affected 6–10-year-old children’s abilities to follow instructions. They found that the benefits of self-enactment at encoding were contingent on sequence complexity as determined by the sequence length and number of possible actions and objects in the experimental set. More specifically, selfenactment benefits were only evident when sequences were shorter, with less variation in the possible actions or objects involved. Demonstration, on the other hand, enhanced recall across the board, regardless of the level of complexity. These differential benefits of self-enactment and demonstration were replicated in children with attention-deficit hyperactivity-disorder (ADHD) and age-matched typically developing controls (Yang et al.,
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
841
842
JONI HOLMES ET AL.
2017). Waterman et al. (2017) suggested that the difference in efficacy between the two encoding strategies might reflect the increased cognitive cost associated with self-enactment compared to demonstration. By this account, self-enactment is a more resource-demanding means of acquiring information than passive observation. Any benefits from self-enactment are, therefore, offset by the competing executive costs associated with self-generating the visuospatial-motoric information – a cost that does not emerge when participants simply observe actions. Together, these findings indicate that working memory supports the retention of information during instruction following, and that different forms of physical movement can enhance recall. In the following section, we review other ways researchers have tried to directly enhance working memory performance.
37.4
Enhancing Working Memory
In this section we provide an overview of two approaches that have been used to try to improve working memory, cognitive training and transcranial electrical brain stimulation.
37.4.1 Working Memory Training The past two decades have been populated by working memory training studies. These aimed to improve working memory capacity through repeated practice on computerized training activities, with a view to enhancing other outcomes that rely on working memory, including children’s learning outcomes. Early findings were promising: individual studies reported positive training effects with transfer to novel memory tasks (near transfer) and other skills that likely depend on working memory (far transfer), including reading and mathematics (e.g., Holmes et al., 2009; Jaeggi et al., 2008; Klingberg et al., 2005; Loosli et al., 2012). Despite the promise of these early findings, the majority of meta-analyses and systematic reviews have since revealed that gains are largely restricted to memory tasks that are highly similar to the trained activities (e.g., Cortese et al., 2015; Gathercole et al., 2019; Melby-Lervåg et al., 2016; Melby-Lervåg & Hulme, 2013; Rapport et al., 2013; Redick et al., 2015; Sala et al., 2019; Sala & Gobet, 2017; Schwaighofer et al., 2015; Shipstead, Redick, et al., 2012; Shipstead, Hicks, et al., 2012; Simons et al., 2016; Soveri et al., 2017). There are some exceptions, with positive transfer beyond the trained tasks reported by some (e.g., Au et al., 2015; Karbach & Verhaeghen, 2014; Peijnenborgh et al., 2016; Spencer-Smith & Klingberg, 2015). Differences in experimental methodologies likely account for the different outcomes across studies and reviews: the quality of the control groups impacts on reported training benefits, and comparisons made to no-contact control groups may inflate estimates of training gains (Morrison & Chein,
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
2011). A clear example of this comes from a meta-analysis evaluating the effects of working memory training on children’s cognitive and academic skills (Sala & Gobet, 2017). Small to moderate effects were reported, but the effect sizes were inversely related to the quality of the experimental designs (e.g., presence of an active control group). The following more specific examples highlight the same issue. Holmes and colleagues (2009) found that training improved working memory and mathematical skills in children with poor working memory six months after training was completed. However, when they replicated the study using a more robust double-blind randomized controlled trial (RCT) design (Dunning et al., 2013), training did not benefit children’s mathematical learning (or any other aspect of learning). Similarly, Jaeggi, et al. (2008) found evidence of transfer to a measure of fluid intelligence for a training group relative to a no-intervention control group. However, Redick et al. (2013) failed to replicate this effect when the benefits for the training group were compared to an active control group (Redick et al., 2013). A final example comes from a study examining the impact of training on children’s attentional skills. Klingberg et al. (2005) reported that training reduced parent-rated symptoms of inattention and hyperactivity. However, meta-analytic studies found little evidence for this effect once raters were blinded to intervention condition (Cortese et al., 2015; Rapport et al., 2013; Sonuga-Barke et al., 2013). Overall, these data show the importance of using rigorous methodologies to evaluate the effectiveness of working memory training, and highlight that the benefits of training for children’s learning are likely to be very limited when strict study designs are used. Due to the absence of reliable training benefits for everyday outcomes, experimental research has more recently focused on investigating the limits of training transfer within working memory. By systematically manipulating the degree of overlap in key task features between training and transfer measures, studies have attempted to establish the conditions under which transfer does and does not occur (e.g., Minear et al., 2016; von Bastian & Oberauer, 2013). The majority of this work has been conducted with adults, but is summarized briefly here to demonstrate that the benefits of training are very narrow. Several experimental studies have shown that training on one type of working memory task (e.g., n-back) does not benefit performance on other types of working memory task (e.g., complex span), indicating that paradigm is a boundary condition to transfer (Byrne et al., 2020; Dunning & Holmes, 2014; Gathercole et al., 2019; Holmes et al., 2019; Li et al., 2008; Minear et al., 2016; Redick et al., 2013; Sprenger et al., 2013; Thompson et al., 2013). The type of stimuli within a paradigm does not appear to constrain transfer, although the effects are smaller when the training and test materials are from different domains (e.g., verbal and visuospatial). For example, in a recent RCT study, Byrne et al. (2020) examined whether two task features – stimuli and paradigm – limited transfer. Adult participants
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
843
844
JONI HOLMES ET AL.
trained on a single backward digit recall task and transfer was tested on multiple within- (backward recall with letters and spatial locations) and cross-paradigm (n-back with digits and letters) measures. Transfer was observed to novel backward span but not n-back tasks. Transfer within the backward span tasks was greater for the novel verbal (letters) than visuospatial (spatial locations) tasks. These findings, along with those from others, indicate that memory items do not constrain transfer when the task structure is held constant between trained and untrained activities (Byrne et al., 2020; Holmes et al., 2019; Jaeggi et al., 2010; Küper & Karbach, 2016; Minear et al., 2016). Together, these findings suggest that computerized working memory training is not an effective tool for enhancing children’s learning outcomes, nor is it enhancing an underlying working memory construct. Sweller’s cognitive load theory might explain why working memory training is not effective, and why any gains on specific tasks do not transfer to improvements in learning (e.g., Sweller 2016; Sweller et al., 2011). According to cognitive load theory, working memory constrains the acquisition of domain-specific secondary skills and knowledge (e.g., topics taught at school that require formal instruction), but not domain-general primary knowledge (e.g., arguably biological or innate skills acquired through evolution that would not need formal instruction, such as speaking, listening, or recognizing faces). Sweller (2016) argues that secondary knowledge is acquired with the assistance of primary knowledge (Paas & Sweller, 2012), and that attempts to teach domain-general primary knowledge fail because this type of knowledge is acquired without tuition (Sweller, 2016). Thus, if working memory is considered a domain-general primary skill, then it will not be teachable or trainable. Moreover, without changes to working memory (primary knowledge), benefits to new school-based topics (secondary knowledge) will not arise. Other studies suggest that strategy-based memory training (e.g., Peng & Fuchs, 2015; St. Clair-Thompson & Holmes, 2008; Witt, 2011), or noncomputerized methods might be more effective for enhancing working memory (Cornoldi et al., 2015; Henry et al., 2014; Passolunghi & Costa, 2016; Rowe et al., 2019). However, these fields are relatively new, meaning it may be too early to tell whether, like many new interventions, they are marked by high levels of early positive results that will not be sustained over longer periods (e.g., Dwan et al., 2013).
37.4.2 Transcranial Electrical Stimulation Another promising tool for enhancing cognition is transcranial electrical stimulation (tES), a noninvasive form of brain stimulation that delivers a weak electrical current to the scalp to affect processing in the underlying cortex (Nitsche & Paulus, 2000). Many studies have shown that tES can enhance working memory performance in single sessions (for reviews, see
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
Berryhill et al., 2014; Coffman et al., 2014; Hill et al., 2016; Kuo & Nitsche, 2012; Tremblay et al., 2014). There is some evidence that tES can be used to enhance the benefits of cognitive training in different cognitive domains (for a review, see Elmasry et al., 2015). However, the literature reporting additive benefits of tES for working memory training is limited. Some studies report positive findings when real (active) stimulation is compared to control stimulation (e.g., sham stimulation or active stimulation applied over a control cortical location). For example, training gains are reported to be improved by tES during n-back training, with greater improvements also found on untrained versions of n-back (Au et al., 2016; Ruf et al., 2017). However, many studies report null effects (e.g., Byrne et al., 2020; Holmes et al., 2016; also see Byrne et al., 2021 for more details), and the outcomes of others that report positive effects are limited by various methodological shortcomings. For example, although some studies reported that tES boosted performance on transfer tasks (e.g., Martin et al., 2013; Richmond et al., 2014), these findings were only significant for active versus no-training control groups. Critically, no significant differences were found between groups who received working memory training with active stimulation compared to those who received working memory training with sham stimulation, meaning the benefits can be attributed to training alone. Furthermore, meta-analytic studies have concluded that there is limited-to-no evidence that tES alters working memory training performance relative to sham stimulation (Mancuso et al., 2016; Nilsson et al., 2017). The majority of investigations on the effects of training combined with tES have focused on adult populations. However, despite ethical concerns surrounding the application of noninvasive brain stimulation to developing brains (e.g., Cohen Kadosh et al., 2012; Davis et al., 2014), there is some limited evidence that tES enhances children’s learning when combined with training. Most of this work has been conducted with children who have difficulties in mathematics or reading, and it is not specific to working memory training, but we include a short summary here for completeness. Costanzo and colleagues found that children and adolescents with dyslexia showed greater improvements in reading skills following training on reading tasks with real versus sham tES (Costanzo et al., 2016, 2019). In another study, children with math difficulties who took part in a numerical training intervention with active stimulation had better accuracy and steeper rates of learning during training relative to a control training group who received sham stimulation (Looi et al., 2017). tES also modulated generalization to an untrained mathematics task. Working memory training in children does not consistently improve either reading or mathematical abilities (Simons et al., 2016), but targeted cognitive interventions combined with tES may yield more promising results. However, more research is needed to assess the feasibility and safety of administering transcranial brain stimulation with children.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
845
846
JONI HOLMES ET AL.
37.5
Classroom Support for Children with Working Memory Issues
Evidence that working memory can be improved directly is severely limited, as discussed in the preceding section. An alternative way to support children with small working memory capacities is to alter the environment in which they learn to minimize the adverse educational consequences of working memory overload (Elliott et al., 2010). Key failures associated with working memory overload in the classroom include failing to see multistep tasks through to completion, place-keeping difficulties, and distractibility. Reducing the working memory demands of learning activities can help increase children’s chance of success in individual learning activities and boost their learning outcomes (Elliott et al., 2010). This can be achieved by reducing the length or complexity of verbal information to be remembered, using external memory aids for the child (such as digital audio recorders), as well as practice in using mnemonic strategies in areas of strength (see Gathercole & Alloway, 2008, for further details). In this section, we focus specifically on the ways in which our knowledge about the role of working memory in instruction following can help children in the classroom. Impairments in the ability to follow verbal instructions given by the teacher may impair academic progress (Engle et al., 1991; Gathercole et al., 2006). The robust and specific advantage to children of performing spoken instructions has practical relevance for classroom practice. Demonstration and physical engagement when instructions are presented and at recall may boost accuracy when remembering instructions over short as well as longer periods. By recruiting a mnemonic benefit in the form of spatial-motoric representations, children may be able to minimize the adverse consequences of weak verbal memory skills. In this way, incorporating physical engagement within curricular activities may accelerate learning. The type of physical engagement at encoding is important. Children benefit from creating additional visuospatial and motoric codes when they anticipate performing the instructions physically at recall. However, their ability to use additional spatial-motoric representations during self-enactment at presentation depends on the cognitive load of the task, which can sometimes harm recall due to additional executive demands. When someone else demonstrates instructions at encoding, executive demands are reduced, and children can use the additional forms of spatial-motoric encoding. It is therefore plausible that action will benefit children’s recall of instructions if they expect to perform them at recall or if someone else demonstrates them.
37.6
Conclusion
In this chapter, we have synthesized evidence for the role of working memory in children’s learning and highlighted some key debates around
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
both links between working memory and academic performance, and methods for enhancing working memory abilities. To summarize, there is a wealth of evidence showing an association between working memory and children’s academic outcomes, but it is debated whether working memory is the key cognitive determinant of learning success. While the ability to simultaneously hold information in mind and use that information in the course of ongoing activities is clearly important for some classroom tasks, it is possible that learning activities also promote the development of working memory skills. In terms of interventions for children with poor working memory skills, there is limited evidence that direct training or training combined with brain stimulation will yield positive benefits for classroom learning. However, harnessing what we know about how to improve children’s abilities to follow instructions using demonstration and physical movement may hold promise.
References Allen, R. J., Hitch, G. J., Mate, J., & Baddeley, A. D. (2012). Feature binding and attention in working memory: A resolution of previous contradictory findings. The Quarterly Journal of Experimental Psychology, 65(12), 2369–2383. Allen, R. J., & Waterman, A. H. (2015). How does enactment affect the ability to follow instructions in working memory? Memory & Cognition, 43(3), 555–561. Astle, D. E., & Fletcher-Watson, S. (2020). Beyond the core-deficit hypothesis in developmental disorders. Current Directions in Psychological Science, 29(5), 431–437. Au, J., Katz, B., Buschkuehl, M., Bunarjo, K., Senger, T., Zabel, C., Jaeggi, S. M., & Jonides, J. (2016). Enhancing working memory training with transcranial direct current stimulation. Journal of Cognitive Neuroscience, 28(9), 1419–1432. Au, J., Sheehan, E., Tsai, N., Duncan, G. J., Buschkuehl, M., & Jaeggi, S. M. (2015). Improving fluid intelligence with training on working memory. Psychonomic Bulletin & Review, 22(2), 1–12. Baddeley, A. D. (1986). Working memory. Oxford University Press. Baddeley, A. (2010). Working memory. Current Biology, 20(4), R136–R140. Baddeley, A. (2012). Working memory: Theories, models, and controversies. Annual Review of Psychology, 63, 1–29. Baddeley, A. D., & Hitch, G. (1974). Working memory. In Psychology of learning and motivation (Vol. 8, pp. 47–89). Elsevier. Baddeley, A. D., & Scott, D. (1971). Short term forgetting in the absence of proactive interference. The Quarterly Journal of Experimental Psychology, 23(3), 275–283.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
847
848
JONI HOLMES ET AL.
Berryhill, M. E., Peterson, D. J., Jones, K. T., & Stephens, J. A. (2014). Hits and misses: Leveraging tDCS to advance cognitive research. Frontiers in Psychology, 5, 1–12. Bishop, D. V. M., & Snowling, M. J. (2004). Developmental dyslexia and specific language impairment: Same or different? Psychological Bulletin, 130(6), 858–886. Booth, J. N., Boyle, J. M. E., & Kelly, S. W. (2010). Do tasks make a difference? Accounting for heterogeneity of performance of children with reading difficulties on tasks of executive function: Findings from a meta-analysis. British Journal of Developmental Psychology, 28(1), 133–176. Brener, R. (1940). An experimental investigation of memory span. Journal of Experimental Psychology, 26(5), 467–482. Bull, R., & Johnston, R. S. (1997). Children’s arithmetical difficulties: Contributions from processing speed, item identification, and short-term memory. Journal of Experimental Child Psychology, 65(1), 1–24. Bull, R., & Scerif, G. (2001). Executive functioning as a predictor of children’s mathematics ability: Inhibition, switching, and working memory. Developmental Neuropsychology, 19(3), 273–293. Byrne, E. M., Ewbank, M. P., Gathercole, S. E., & Holmes, J. (2020). The effects of transcranial direct current stimulation on within- and crossparadigm transfer following multi-session backward recall training. Brain and Cognition, 141, 105552. Byrne, E. M., Nord, C. L., & Holmes, J. (2021). Cognitive plasticity and transcranial electrical stimulation. In Cognitive Training (pp. 85–105). Springer International Publishing. Cain, K., Oakhill, J., & Bryant, P. (2004). Children’s reading comprehension ability: Concurrent prediction by working memory, verbal ability, and component skills. Journal of Educational Psychology, 96(1), 31–42. Cain, K., Oakhill, J., & Lemmon, K. (2004). Individual differences in the inference of word meanings from context: The influence of reading comprehension, vocabulary knowledge, and memory capacity. Journal of Educational Psychology, 96(4), 671–681. Camos, V., Lagner, P., & Barrouillet, P. (2009). Two maintenance mechanisms of verbal information in working memory. Journal of Memory and Language, 61(11), 457–469. Cantor, J., & Engle, R. W. (1993). Working-memory capacity as long-term memory activation: An individual-differences approach. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19(5), 1101–1114. Carretti, B., Borella, E., Cornoldi, C., & De Beni, R. (2009). Role of working memory in explaining the performance of individuals with specific reading comprehension difficulties: A meta-analysis. Learning and Individual Differences, 19(2), 246–251. Castles, A., & Friedmann, N. (2014). Developmental dyslexia and the phonological deficit hypothesis. Mind and Language, 29(3), 270–285.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
Catts, H. W., Fey, M. E., Tomblin, J. B., & Zhang, X. (2002). A longitudinal investigation of reading outcomes in children with language impairments. Journal of Speech, Language, and Hearing Research, 45(6), 1142–1157. Chandler, P,. & Sweller, J. (1991). Cognitive Load Theory and the format of instruction. Cognition and Instruction, 8(4), 293–332. Charlesworth, L. A., Allen, R. J., Morson, S., Burn, W. K., & Souchay, C. (2014). Working memory and the enactment effect in early Alzheimer’s disease. ISRN Neurology, eCollection 2014. Child, A. E., Cirino, P. T., Fletcher, J. M., Willcutt, E. G., & Fuchs, L. S. (2019). A cognitive dimensional approach to understanding shared and unique contributions to reading, math, and attention skills. Journal of Learning Disabilities, 52(1), 15–30. Cirino, P. T., Child, A. E., & Macdonald, K. T. (2018). Longitudinal predictors of the overlap between reading and math skills. Contemporary Educational Psychology, 54, 99–111. Coffman, B. A., Clark, V. P., & Parasuraman, R. (2014). Battery powered thought: Enhancement of attention, learning, and memory in healthy adults using transcranial direct current stimulation. NeuroImage, 85, 895–908. Cohen Kadosh, R., Levy, N., O’Shea, J., Shea, N., & Savulescu, J. (2012). The neuroethics of non-invasive brain stimulation. Current Biology, 22(4), R108–R111. Cornoldi, C., Carretti, B., Drusi, S., & Tencati, C. (2015). Improving problem solving in primary school students: The effect of a training programme focusing on metacognition and working memory. The British Journal of Educational Psychology, 85(3), 424–439. Cortese, S., Ferrin, M., Brandeis, D., Buitelaar, J., Daley, D., Dittmann, R. W., Holtmann, M., Santosh, P., Stevenson, J., Stringaris, A., Zuddas, A., & Sonuga-Barke, E. J. S. (2015). Cognitive training for attention-deficit/ hyperactivity disorder: Meta-analysis of clinical and neuropsychological outcomes from randomized controlled trials. Journal of the American Academy of Child and Adolescent Psychiatry, 54(3), 164–174. Costanzo, F., Rossi, S., Varuzza, C., Varvara, P., Vicari, S., & Menghini, D. (2019). Long-lasting improvement following tDCS treatment combined with a training for reading in children and adolescents with dyslexia. Neuropsychologia, 130, 38–43. Costanzo, F., Varuzza, C., Rossi, S., Sdoia, S., Varvara, P., Oliveri, M., Giacomo, K., Vicari, S., & Menghini, D. (2016). Evidence for reading improvement following tDCS treatment in children and adolescents with dyslexia. Restorative Neurology and Neuroscience, 34(2), 215–226. Cowan, N. (2008). What are the differences between long-term, short-term, and working memory? Progress in Brain Research, 169, 323–338. Cowan, N. (2017). The many faces of working memory and short-term storage. Psychonomic Bulletin & Review, 24(4), 1158–1170.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
849
850
JONI HOLMES ET AL.
Cowan, N., Saults, J. S., & Nugent, L. D. (1997). The role of absolute and relative amounts of time in forgetting within immediate memory: The case of tone-pitch comparisons. Psychonomic Bulletin & Review, 4(3), 393–397. Dahlin, E., Neely, A. S., Larsson, A., Bäckman, L., & Nyberg, L. (2008). Transfer of learning after updating training mediated by the striatum. Science, 320(5882), 1510–1512 Davis, N. J., Oberman, L. M., & Earp, B. D. (2014). Transcranial stimulation of the developing brain: A plea for extreme caution. Frontiers in Human Neuroscience, 8, 600. De Smedt, B., Swillen, A., Verschaffel, L., & Ghesquière, P. (2009). Mathematical learning disabilities in children with 22q11.2 deletion syndrome: A review. Developmental Disabilities Research Reviews, 15(1), 4–10. Duncan, J., & Owen, A. M. (2000). Common regions of the human frontal lobe recruited by diverse cognitive demands. Trends in Neurosciences, 23 (10), 475–483. Dunning, D. L., & Holmes, J. (2014). Does working memory training promote the use of strategies on untrained working memory tasks? Memory & Cognition, 42(6), 854–862. Dunning, D. L., Holmes, J., & Gathercole, S. E. (2013). Does working memory training lead to generalized improvements in children with low working memory? A randomized controlled trial. Developmental Science, 16, 915–925. Dwan, K., Gamble, C., Williamson, P. R., & Kirkham, J. J. (2013). Systematic review of the empirical evidence of study publication bias and outcome reporting bias: An updated review. PLoS ONE, 8, e66844. Elliott, J. G., Gathercole, S. E., Alloway, T. P., Holmes, J., & Kirkwood, H. (2010). An evaluation of a classroom-based intervention to help overcome working memory difficulties and improve long-term academic achievement. Journal of Cognitive Education and Psychology, 9, 227–251. Elmasry, J., Loo, C., & Martin, D. M. (2015). A systematic review of transcranial electrical stimulation combined with cognitive training. Restorative Neurology and Neuroscience, 33(3), 263–278. Engle, R. W., Carullo, J. J., & Collins, K. W. (1991). Individual differences in working memory for comprehension and following directions. Journal of Educational Research, 84(5), 253–262. Follmer, D. J. (2018). Executive function and reading comprehension: A meta-analytic review. Educational Psychologist, 53(1), 42–60. Gathercole, S. E., & Alloway, T. P. (2008). Working memory and learning: A practical guide for teachers. Sage. Gathercole, S. E., Brown, L., & Pickering, S. J. (2003). Working memory assessments at school entry as longitudinal predictors of National Curriculum attainment levels. Educational and Child Psychology, 20(3), 109–122.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
Gathercole, S. E., Dunning, D. L., Holmes, J., & Norris, D. (2019). Working memory training involves learning new skills. Journal of Memory and Language, 105, 19–42. Gathercole, S. E., Durling, E., Evans, M., Jeffcock, S., & Stone, S. (2008). Working memory abilities and children’s performance in laboratory analogues of classroom activities. Applied Cognitive Psychology, 22(8), 1019–1037. Gathercole, S. E., & Holmes, J. (2014). Developmental impairments of working memory: Profiles and interventions. Perspectives on Language and Literacy, 40, 36–39. Gathercole, S. E., Lamont, E., & Alloway, T. P. (2006). Working memory in the classroom. In S. Pickering (Ed.), Working memory and education. Academic. Geary, D. C. (1993). Mathematical disabilities: Cognitive, neuropsychological, and genetic components. Psychological Bulletin, 114(2), 345–362. Geary, D. C. (2003). Learning disabilities in arithmetic: Problem-solving differences and cognitive deficits. In H. L. Swanson, K. R. Harris, & S. Graham (Eds.), Handbook of learning disabilities (pp. 199–212). The Guilford Press. Henry, L. A., Messer, D. J., & Nash, G. (2014). Testing for near and far transfer effects with a short, face-to-face adaptive working memory training intervention in typical children. Infant and Child Development, 23(1), 84–103. Hill, A. T., Fitzgerald, P. B., & Hoy, K. E. (2016). Effects of anodal transcranial direct current stimulation on working memory: A systematic review and meta-analysis of findings from healthy and neuropsychiatric populations. Brain Stimulation, 9(2), 197–208. Hitch, G. J., & McAuley, E. (1991). Working memory in children with specific arithmetical learning difficulties. British Journal of Psychology, 82(3), 375–386. Holmes, J., Adams, J. W., & Hamilton, C. J. (2008). The relationship between visuospatial sketchpad capacity and children’s mathematical skills. European Journal of Cognitive Psychology, 20(2), 272–289. Holmes, J., Byrne, E. M., Gathercole, S. E., & Ewbank, M. P. (2016). Transcranial random noise stimulation does not enhance the effects of working memory training. Journal of Cognitive Neuroscience, 28(10), 1–13. Holmes, J., Gathercole, S. E., & Dunning, D. L. (2009). Adaptive training leads to sustained enhancement of poor working memory in children. Developmental Science, 12(4), 1–7. Holmes, J., Guy, J., Kievit, R. A., Bryant, A., Mareva, S., Gathercole, S. E., & CALM Team. (2020). Cognitive dimensions of learning in children with problems in attention, learning, and memory. Journal of Educational Psychology. Advance online publication.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
851
852
JONI HOLMES ET AL.
Holmes, J., Woolgar, F., Hampshire, A., & Gathercole, S. E. (2019). Are working memory training effects paradigm-specific? Frontiers in Psychology, 10(May), 1103. Jaeggi, S. M., Buschkuehl, M., Jonides, J., & Perrig, W. J. (2008). Improving fluid intelligence with training on working memory. Proceedings of the National Academy of Sciences of the United States of America, 105(19), 6829–6833. Jaeggi, S. M., Studer-Luethi, B., Buschkuehl, M., Su, Y.-F., Jonides, J., & Perrig, W. J. (2010). The relationship between n-back performance and matrix reasoning: Implications for training and transfer. Intelligence, 38(6), 625–635. Jaroslawska, A. J., Gathercole, S. E., Allen, R. J., & Holmes, J. (2016). Following instructions from working memory: Why does action at encoding and recall help? Memory & Cognition, 44(8), 1183–1191. Jaroslawska, A. J., Gathercole, S. E., & Holmes, J. (2018). Following instructions in a dual-task paradigm: Evidence for a temporary motor store in working memory. Quarterly Journal of Experimental Psychology, 71(11), 2439–2449. Jaroslawska, A. J., Gathercole, S. E., Logie, M. R., & Holmes, J. (2016). Following instructions in a virtual school: Does working memory play a role? Memory & Cognition, 44(4), 580–589. Jeffries, S., & Everatt, J. (2004). Working memory: Its role in dyslexia and other specific learning difficulties. Dyslexia, 10(3), 196–214. https://doi .org/10.1002/dys.278 Kaplan, C. H., & White, M. A. (1980). Children’s direction-following behavior in grades K-S. The Journal of Educational Research, 74(1), 43–48. Karbach, J., & Verhaeghen, P. (2014). Making working memory work: A meta-analysis of executive-control and working memory training in older adults. Psychological Science, 25(11), 2027–2037. Kievit, R. A., Hofman, A. D., & Nation, K. (2019). Mutualistic coupling between vocabulary and reasoning in young children: A replication and extension of the study by Kievit et al. (2017). Psychological Science, 30(8), 1245–1252. Klingberg, T., Fernell, E., Olesen, P. J., Johnson, M., Gustafsson, P., Dahlström, K., Gillberg, C. G., Forssberg, H., & Westerberg, H. (2005). Computerized training of working memory in children with ADHD: A randomized, controlled trial. Journal of the American Academy of Child and Adolescent Psychiatry, 44(2), 177–186. Koriat, A., Ben-Zur, H., & Nussbaum, A. (1990). Encoding information for future action: Memory for to-be-performed tasks versus memory for tobe-recalled tasks. Memory & Cognition, 18(6), 568–578. Kudo, M. F., Lussier, C. M., & Swanson, H. L. (2015). Reading disabilities in children: A selective meta-analysis of the cognitive literature. Research in Developmental Disabilities, 40, 51–62.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
Kuo, M.-F., & Nitsche, M. A. (2012). Effects of transcranial electrical stimulation on cognition. Clinical EEG and Neuroscience, 43(3), 192–199. Küper, K., & Karbach, J. (2016). Increased training complexity reduces the effectiveness of brief working memory training: Evidence from shortterm single and dual n-back training interventions. Journal of Cognitive Psychology, 28(2), 199–208. Landerl, K., & Kölle, C. (2009). Typical and atypical development of basic numerical skills in elementary school. Journal of Experimental Child Psychology, 103(4), 546–565. Li, S.-C., Schmiedek, F., Huxhold, O., Röcke, C., Smith, J., & Lindenberger, U. (2008). Working memory plasticity in old age: Practice gain, transfer, and maintenance. Psychology and Aging, 23(4), 731–742. Li, Y., & Geary, D. C. (2013). Developmental gains in visuospatial memory predict gains in mathematics achievement. PLoS ONE, 8(7), e70160–e70160. Looi, C. Y., Lim, J., Sella, F., Lolliot, S., Duta, M., Avramenko, A. A., & Cohen Kadosh, R. (2017). Transcranial random noise stimulation and cognitive training to improve learning and cognition of the atypically developing brain: A pilot study. Scientific Reports, 7(1), 4633. Loosli, S. V., Buschkuehl, M., Perrig, W. J., & Jaeggi, S. M. (2012). Working memory training improves reading processes in typically developing children. Child Neuropsychology, 18(1), 62–78. Mancuso, L. E., Ilieva, I. P., Hamilton, R. H., & Farah, M. J. (2016). Does transcranial direct current stimulation improve healthy working memory? A meta-analytic review. Journal of Cognitive Neuroscience, 28(8), 1063–1089. Mareva, S., The CALM Team, & Holmes, J. (2021). Cognitive and academic skills in two developmental cohorts of different ability level: A mutualistic network perspective, Journal of Applied Research in Memory and Cognition. Martin, D. M., Liu, R., Alonzo, A., Green, M., Player, M. J., Sachdev, P., & Loo, C. K. (2013). Can transcranial direct current stimulation enhance outcomes from cognitive training? A randomized controlled trial in healthy participants. The International Journal of Neuropsychopharmacology, 16(9), 1927–1936. McLean, J. F., & Hitch, G. J. (1999). Working memory impairments in children with specific arithmetic learning difficulties. Journal of Experimental Child Psychology, 74(3), 240–260. Melby-Lervåg, M., & Hulme, C. (2013). Is working memory training effective? A meta-analytic review. Developmental Psychology, 49(2), 270–291. Melby-Lervåg, M., Lervåg, A., Lyster, S.-A. H., Klem, M., Hagtvet, B., & Hulme, C. (2012). Nonword-repetition ability does not appear to be a causal influence on children’s vocabulary development. Psychological Science, 23(10), 1092–1098.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
853
854
JONI HOLMES ET AL.
Melby-Lervåg, M., Lyster, S. A. H., & Hulme, C. (2012). Phonological skills and their role in learning to read: A meta-analytic review. Psychological Bulletin, 138(2), 322–352. Melby-Lervåg, M., Redick, T. S., & Hulme, C. (2016). Working memory training does not improve performance on measures of intelligence or other measures of “far transfer”: Evidence from a meta-analytic review. Perspectives on Psychological Science, 11(4), 512–534. Meyer, M. L., Salimpoor, V. N., Wu, S. S., Geary, D. C., & Menon, V. (2010). Differential contribution of specific working memory components to mathematics achievement in 2nd and 3rd graders. Learning and Individual Differences, 20(2), 101–109. Minear, M., Brasher, F., Guerrero, C. B., Brasher, M., Moore, A., & Sukeena, J. (2016). A simultaneous examination of two forms of working memory training: Evidence for near transfer only. Memory & Cognition, 44, 1010–1037. Morrison, A. B., & Chein, J. M. (2011). Does working memory training work? The promise and challenges of enhancing cognition by training working memory. Psychonomic Bulletin & Review, 18, 46–60. Nation, K. (1999). Reading skills in hyperlexia: A developmental perspective. Psychological Bulletin, 125(3), 338–355. Nilsson, J., Lebedev, A. V., Rydström, A., & Lövdén, M. (2017). Direct-current stimulation does little to improve the outcome of working memory training in older adults. Psychological Science, 28(7), 907–920. Nitsche, M. A., & Paulus, W. (2000). Excitability changes induced in the human motor cortex by weak transcranial direct current stimulation. The Journal of Physiology, 527(3), 633–639. Osterberg, L., & Blaschke, T. (2005). Adherence to medication. New England Journal of Medicine, 353, 487–497. Paas, F. G. W. C., Renkl, A., & Sweller, J. (2003). Cognitive load theory and instructional design: Recent developments. Educational Psychologist, 38(1), 1–4. Paas, F. G. W. C., Renkl, A., & Sweller, J. (2004). Cognitive Load Theory: Instructional implications of the interaction between information structures and cognitive architecture, Instructional Science, 32, 1–8. Paas, F., & Sweller, J. (2012). An evolutionary upgrade of cognitive load theory: Using the human motor system and collaboration to support the learning of complex cognitive tasks. Educational Psychology Review, 24 (1), 27–45. Passolunghi, M. C., & Costa, H. M. (2016). Working memory and early numeracy training in preschool children. Child Neuropsychology, 22(1), 81–98. Passolunghi, M. C., Mammarella, I. C., & Altoè, G. (2008). Cognitive abilities as precursors of the early acquisition of mathematical skills during first through second grades. Developmental Neuropsychology, 33(3), 229–250.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
Passolunghi, M. C., & Siegel, L. S. (2001). Short-term memory, working memory, and inhibitory control in children with difficulties in arithmetic problem solving. Journal of Experimental Child Psychology, 80(1), 44–57. Pause, B. M., Zlomuzica, A., Kinugawa, K., Mariani, J., Pietrowsky, R., & Dere, E. (2013). Perspectives on episodic-like and episodic memory. Frontiers in Behavioral Neuroscience, 7, 1–12. Peijnenborgh, J. C. A., Hurks, P. M., Aldenkamp, A. P., Vles, J. S. H., & Hendriksen, J. G. M. (2016). Efficacy of working memory training in children and adolescents with learning disabilities: A review study and meta-analysis. Neuropsychological Rehabilitation, 26(5–6), 645–672. Peng, P., & Fuchs, D. (2015). A randomized control trial of working memory training with and without strategy instruction: Effects on young children’s working memory and comprehension. Journal of Learning Disabilities, 50(1), 62–80. Peng, P., & Fuchs, D. (2016). A meta-analysis of working memory deficits in children with learning difficulties: Is there a difference between verbal domain and numerical domain? Journal of Learning Disabilities, 49(1), 3–20. Peng, P., & Kievit, R. A. (2020). The development of academic achievement and cognitive abilities: A bidirectional perspective. Child Development Perspectives, 14(1), 15–20. Peng, P., Namkung, J., Barnes, M., & Sun, C. (2016). A meta-analysis of mathematics and working memory: Moderating effects of working memory domain, type of mathematics skill, and sample characteristics. Journal of Educational Psychology, 108(4), 455–473. Peng, P., Wang, C., & Namkung, J. (2018). Understanding the cognition related to mathematics difficulties: A meta-analysis on the cognitive deficit profiles and the bottleneck theory. Review of Educational Research, 88(3), 434–476. Pimperton, H., & Nation, K. (2010). Suppressing irrelevant information from working memory: Evidence for domain-specific deficits in poor comprehenders. Journal of Memory and Language, 62(4), 380–391. Preåler, A.-L., Könen, T., Hasselhorn, M., & Krajewski, K. (2014). Cognitive preconditions of early reading and spelling: A latent-variable approach with longitudinal data. Reading and Writing, 27(2), 383–406. Rapport, M. D., Orban, S. A., Kofler, M. J., & Friedman, L. M. (2013). Do programs designed to train working memory, other executive functions, and attention benefit children with ADHD? A meta-analytic review of cognitive, academic, and behavioral outcomes. Clinical Psychology Review, 33(8), 1237–1252. Redick, T. S., Shipstead, Z., Harrison, T. L., Hicks, K. L., Fried, D. E., Hambrick, D. Z., Kane, M. J., & Engle, R. W. (2013). No evidence of intelligence improvement after working memory training: A randomized, placebocontrolled study. Journal of Experimental Psychology: General, 142(2), 359–379.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
855
856
JONI HOLMES ET AL.
Redick, T. S., Shipstead, Z., Wiemers, E. A., Melby-Lervåg, M., & Hulme, C. (2015). What’s working in working memory training? An Educational Perspective, 27(4), 617–633. Richmond, L. L., Wolk, D., Chein, J. M., & Olson, I. R. (2014). Transcranial direct current stimulation enhances verbal working memory training performance over time and near-transfer outcomes. Journal of Cognitive Neuroscience, 26(11), 2443–2454. Rotzer, S., Loenneker, T., Kucian, K., Martin, E., Klaver, P., & von Aster, M. (2009). Dysfunctional neural network of spatial working memory contributes to developmental dyscalculia. Neuropsychologia, 47(13), 2859–2865. Rowe, A., Titterington, J., Holmes, J., Henry, L., & Taggart, L. (2019). Interventions targeting working memory in 4–11 year olds within their everyday contexts: A systematic review. Developmental Review, 52, 1–23. Ruf, S. P., Fallgatter, A. J., & Plewnia, C. (2017). Augmentation of working memory training by transcranial direct current stimulation (tDCS). Scientific Reports, 7(1), 876. Sala, G., Deniz Aksayli, N., Semir Tatlidil, K., Tatsumi, T., Gondo, Y., & Gobet, F. (2019). Near and far transfer in cognitive training: A secondorder meta-analysis. Collabra: Psychology, 5(1), 18. Sala, G., & Gobet, F. (2017). Does far transfer exist? Negative evidence from chess, music, and working memory training. Current Directions in Psychological Science, 26(6), 515–520. Schwaighofer, M., Fischer, F., & Bühner, M. (2015). Does working memory training transfer? A meta-analysis including training conditions as moderators. Educational Psychologist, 50(2), 138–166. Shapiro, L. (2007). The embodied cognition research programme. Philosophy Compass, 2(2), 338–346. Shipstead, Z., Hicks, K. L., & Engle, R. W. (2012). Cogmed working memory training: Does the evidence support the claims? Journal of Applied Research in Memory and Cognition, 1(3), 185–193. Shipstead, Z., Redick, T. S., & Engle, R. W. (2012). Is working memory training effective? Psychological Bulletin, 138(4), 628–654. Simons, D. J., Boot, W. R., Charness, N., Gathercole, S. E., Chabris, C. F., Hambrick, D. Z., & Stine-Morrow, E. A. L. (2016). Do “brain-training” programs work? Psychological Science in the Public Interest, 17(3), 103–186. Smyth, M. M., & Pendleton, L. R. (1989). Working memory for movements. The Quarterly Journal of Experimental Psychology, 41(2), 235–250. Smyth, M. M., & Pendleton, L. R. (1990). Space and movement in working memory. The Quarterly Journal of Experimental Psychology, 42(2), 291–304. Sonuga-Barke, E. J. S., Brandeis, D., Cortese, S., Daley, D., Ferrin, M., Holtmann, M., Stevenson, J., Danckaerts, M., Van Der Oord, S., Döpfner, M., Dittmann, R. W., Simonoff, E., Zuddas, A., Banaschewski, T., Buitelaar, J., Coghill, D., Hollis, C., Konofal, E., Lecendreux, M.,. . . Zuddas, A. (2013). Nonpharmacological interventions for ADHD:
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
Working Memory and Classroom Learning
Systematic review and meta-analyses of randomized controlled trials of dietary and psychological treatments. American Journal of Psychiatry, 170(3), 275–289. Soveri, A., Antfolk, J., Karlsson, L., Salo, B., & Laine, M. (2017). Working memory training revisited: A multi-level meta-analysis of n-back training studies. Psychonomic Bulletin & Review, 24(4), 1077–1096. Spencer-Smith, M., & Klingberg, T. (2015). Benefits of a working memory training program for inattention in daily life: A systematic review and meta-analysis. PLoS ONE, 10(3), 1–18. Sprenger, A. M., Atkins, S. M., Bolger, D. J., Harbison, J. I., Novick, J. M., Chrabaszcz, J. S., Weems, S. A., Smith, V., Bobb, S., Bunting, M. F., & Dougherty, M. R. (2013). Training working memory: Limits of transfer. Intelligence, 41(5), 638–663. St. Clair-Thompson, H. L., & Holmes, J. (2008) Improving short term and working memory: Methods of memory training. In N.B. Johansen (Ed.), New research in short term memory (pp. 125–54). Novascience. Swanson, H. L., & Ashbaker, M. H. (2000). Working memory, short-term memory, speech rate, word recognition and reading comprehension in learning disabled readers: Does the executive system have a role? Intelligence, 28(1), 1–30. Swanson, H. L., & Beebe-Frankenberger, M. (2004). The relationship between working memory and mathematical problem solving in children at risk and not at risk for serious math difficulties. Journal of Educational Psychology, 96(3), 471–491. Swanson, H. L., & Jerman, O. (2006). Math disabilities: A selective metaanalysis of the literature. Review of Educational Research, 76(2), 249–274. Swanson, H. L., Zheng, X., & Jerman, O. (2009). Working memory, shortterm memory, and reading disabilities: A selective meta-analysis of the literature. Journal of Learning Disabilities, 42(3), 260–287. Sweller, J. (2011). Cognitive load theory. In Psychology of learning and motivation (Vol. 55, pp. 37–76). Academic Press. Sweller, J. (2016). Working memory, long-term memory, and instructional design. Journal of Applied Research in Memory and Cognition, 5(4), 360–367. Szucs, D., Devine, A., Soltesz, F., Nobes, A., & Gabriel, F. (2013). Developmental dyscalculia is related to visuo-spatial memory and inhibition impairment. Cortex, 49(10), 2674–2688. Thompson, T. W., Waskom, M. L., Garel, K. L. a, Cardenas-Iniguez, C., Reynolds, G. O., Winter, R., Chang, P., Pollard, K., Lala, N., Alvarez, G. a., & Gabrieli, J. D. E. (2013). Failure of working memory training to enhance cognition or intelligence. PLoS ONE, 8(5). Tremblay, S., Lepage, J.-F. F., Latulipe-Loiselle, A., Fregni, F., Pascual-Leone, A., & Théoret, H. (2014). The uncertain outcome of prefrontal tDCS. Brain Stimulation, 7(6), 773–783. Tulving, E. (2002). Episodic memory: From mind to brain. Annual Review of Psychology, 53, 1–25.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
857
858
JONI HOLMES ET AL.
Van Der Maas, H. L. J., Dolan, C. V., Grasman, R. P. P. P., Wicherts, J. M., Huizenga, H. M., & Raijmakers, M. E. J. (2006). A dynamical model of general intelligence: The positive manifold of intelligence by mutualism. Psychological Review, 113(4), 842–861. Van de Weijer-Bergsma, E., Kroesbergen, E. H., & Van Luit, J. E. (2015). Verbal and visual-spatial working memory and mathematical ability in different domains throughout primary school. Memory & Cognition, 43(3), 367–378. von Bastian, C. C., & Oberauer, K. (2013). Distinct transfer effects of training different facets of working memory capacity. Journal of Memory and Language, 69(1), 36–58. Wang, S., & Gathercole, S. E. (2013). Working memory deficits in children with reading difficulties: Memory span and dual task coordination. Journal of Experimental Child Psychology, 115(1), 188–197. Waterman, A. H., Atkinson, A. L., Aslam, S. S., Holmes, J., Jaroslawska, A. J., & Allen, R. J. (2017). Do actions speak louder than words? Examining children’s ability to follow instructions. Memory & Cognition, 45(6), 877–890. Wickens, C. M., Toplak, M. E., & Wiesenthal, D. L. (2008). Cognitive failures as predictors of driving errors, lapses, and violations. Accident Analysis & Prevention, 40(3), 1223–1233. Willcutt, E. G., Petrill, S. A., Wu, S., Boada, R., DeFries, J. C., Olson, R. K., & Pennington, B. F. (2013). Comorbidity between reading disability and math disability: Concurrent psychopathology, functional impairment, and neuropsychological functioning. Journal of Learning Disabilities, 46(6), 500–516. Witt, M. U. (2011). School based working memory training: Preliminary finding of improvement in children’s mathematical performance. Advances in Cognitive Psychology, 7, 7–15. Wojcik, D., Allen, R., Brown, C., & Souchay, C. (2011). Memory for actions in autism spectrum disorder. Memory, 19(6), 549–558. Wolpert, D. M., & Ghahramani, Z. (2000). Computational principles of movement neuroscience. Nature Neuroscience, 3(11), 1212–1217. Yang, T., Allen, R. J., & Gathercole, S. E. (2016). Examining the role of working memory resources in following spoken instructions. Journal of Cognitive Psychology, 28(2), 186–198. Yang, T., Allen, R. J., Holmes, J., & Chan, R. C. (2017). Impaired memory for instructions in children with attention-deficit hyperactivity disorder is improved by action at presentation and recall. Frontiers in Psychology, 8, 39. Yang, T., Gathercole, S. E., & Allen, R. J. (2014). Benefit of enactment over oral repetition of verbal instruction does not require additional working memory during encoding. Psychonomic Bulletin & Review, 21(1), 186–192. Yeniad, N., Malda, M., Mesman, J., van IJzendoorn, M. H., & Pieper, S. (2013). Shifting ability predicts math and reading performance in children: A meta-analytical study. Learning and Individual Differences, 23, 1–9.
https://doi.org/10.1017/9781108955638.044 Published online by Cambridge University Press
38 Cognitive Load Theory and Instructional Design for Language Learning John Sweller, Stéphanie Roussel, and André Tricot 38.1
Introduction
Cognitive load theory (Sweller et al., 1998, 2011, 2019; Tindall-Ford et al., 2020) has been applied to a wide variety of instructional areas including second language learning (Roussel et al., 2017). Initially, the theory was based primarily on our knowledge of human cognitive architecture, specifically the intricate relations between working memory and long-term memory. Subsequently, relations between working memory and long-term memory were integrated with evolutionary psychology. The incorporation of evolutionary principles is important for all curriculum areas but has specific currency when dealing with language learning. Working memory has a different role when learning a first language versus a second language, a difference that is ignored by many instructional recommendations. We will begin by using evolutionary educational psychology to indicate instructionally relevant categories of knowledge.
38.2
Evolutionary Educational Psychology and Categories of Knowledge
Knowledge can be categorized in many ways. From an instructional perspective, the only categories that matter are ones that have instructional consequences. A major, data-based categorization scheme with important instructional consequences was devised by David Geary (Geary, 2008, 2012; Geary & Berch, 2016). He divided knowledge into two evolutionarily determined categories: biologically primary and biologically secondary knowledge. Biologically primary knowledge is knowledge that we have specifically evolved to acquire. Examples are learning to listen to and speak a first language, using general problem-solving strategies, generalizing from specific instances, learning how to learn, establishing social
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
860
JOHN SWELLE ET AL.
relationships, or learning how to recognize faces. There are several important characteristics of biologically primary knowledge. Different types of primary knowledge are modular, having usually evolved for different purposes and probably during different evolutionary epochs. They are acquired easily, indeed unconsciously, without the need for instruction even if they are informationally dense. Commonly, though not universally, they consist of generic-cognitive skills (Sweller, 2015, 2016; Tricot & Sweller, 2014) that are central to the human condition. We have evolved to acquire these skills precisely because of their critical importance. Biologically secondary skills are important for cultural reasons and vary from culture to culture and even from species to species (Tomasello, 2009). We can acquire the knowledge underlying these skills but frequently require considerable conscious effort to do so. Because we have not evolved to automatically acquire biologically secondary knowledge, societies often need to establish societal procedures, structures, and institutions in order to ensure that the requisite knowledge is acquired. Schools and other education and training institutions were established in order to specifically impart biologically secondary knowledge that otherwise is unlikely to be acquired. Virtually everything that is taught in such institutions can be classified as biologically secondary, and so most curricula documents provide examples of biologically secondary knowledge. Unlike biologically primary knowledge which tends to be generic-cognitive in nature, biologically secondary knowledge tends to be domain-specific (see Fischer et al., 2018, for discussions of this issue). For example, while natural language is needed to teach mathematics or science, what is learned tends to be very specific to the branch of mathematics or science under consideration. Indeed, even the language associated with a particular topic is likely to be specific to that topic. This categorization of knowledge is critically important for language learning. We have evolved to acquire listening and speaking skills of our first language, which is learned when young. Because of that evolved skill, we are neither taught how to listen to or speak our first language nor do we consciously need to learn how to listen to or speak it. Our biologically primary listening and speaking skills develop automatically, easily, and without conscious effort. All other language skills are biologically secondary and so acquired in a very different manner. Thus, learning to read and write in our first language is biologically secondary. In addition, when learning a second language not only are reading and writing biologically secondary, so are listening and speaking. These distinctions have profound instructional implications. Frequently, there is an implicit, sometimes explicit, assumption that instructional procedures should be similar for all aspects of language learning. In the field of language learning, second language and first language learning have been widely compared. As a result, language teaching under social and political pressure as well as on the basis of theoretical justifications has
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
borrowed, is borrowing, and will no doubt continue to borrow learning procedures that are based on what is known about the way in which language learning can take place in the natural environment. Many second language learning approaches are grounded in first language acquisition research, in interaction-based learning theories, in sociocultural and/or social-constructivist learning theory, and in input-based theory (DaltonPuffer, 2007). Many of these theories such as constructivism (Bruner, 1978; Vygotsky, 1978), input theory (Krashen, 1985), the interactionist (Gass & Mackey, 2007) and naturalistic approaches (Krashen & Terrell, 2000), assume that learning a first and a second language are based on the same mechanisms, and that the second should imitate the first. More recently, these theories furnished a “convenient rationale” for the reliance on incidental second language learning while learning other academic content simultaneously (Nikula et al., 2013, p. 75). According to these theories, the best second language learning conditions should be similar to those of first language acquisition. But, if learning to listen to and speak a first language is biologically primary while all else is biologically secondary, then that assumption is invalid. If the cognitive processes underlying learning to listen to and speak a first language are very different to the cognitive processes underlying all other language learning, then learning to listen to and speak a first language cannot be used as a general language learning model. Simply placing learners in a language-oriented environment, which is all that is needed to learn to listen to and speak a first language, is likely to be insufficient. Other second language acquisition theories have however underlined that in contexts where academic content is taught in a second language, it might be necessary to guide the attention of the learners on linguistic forms (Long, 1991); otherwise, those linguistics features may go unnoticed, unprocessed, and unlearned (see also Robinson, 2003; Schmidt, 2001). We will argue that the cognitive load involved in processing this kind of academic context can be so heavy that different instructional procedures are required. The cognitive architecture associated with the acquisition of biologically secondary knowledge is discussed next.
38.3
Human Cognitive Architecture
Cognitive load theory is concerned with the acquisition of biologically secondary, domain-specific knowledge. Biologically primary, genericcognitive knowledge may assist in the acquisition of secondary knowledge (Paas & Sweller, 2012) but the major goal of instruction is the acquisition of biologically secondary, domain-specific knowledge. There is a well-defined cognitive architecture associated with the acquisition of secondary knowledge. That architecture can be described by five principles that also underpin the functioning of biological evolution itself (Sweller & Sweller, 2006).
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
861
862
JOHN SWELLE ET AL.
While each of the principles is concerned with acquiring, processing, storing, or using biologically secondary, domain-specific information, each principle itself provides an example of a biologically primary, genericcognitive skill. Accordingly, the five principles themselves do not need to be taught or consciously learned.
38.3.1 Acquisition of Novel Information There are two ways in which novel information may be acquired. Most commonly, it is acquired from other people via the borrowing and reorganizing principle. Less frequently, when relevant information cannot be obtained from others, it is obtained during problem-solving using the randomness as genesis principle. 38.3.1.1 The Borrowing and Reorganizing Principle We obtain the vast bulk of biologically secondary, domain-specific information from other people using the borrowing and reorganizing principle. Unlike most other species, we have evolved to both provide and obtain information from others (Thornton & Raihani, 2008; Tomasello, 2009) as a biologically primary skill. We imitate what other people do (Bandura, 1986), listen to what they say, and read what they write. Once new information is processed, it is reorganized to be congruent with previous information that has been stored (Bartlett, 1932; Piaget, 1928). New information is rarely stored like a recording device in the same form in which it is obtained. Rather, it is altered by what is already known and stored schematically (Bartlett, 1932). For example, when learning to read, we can read a variety of scripts (including handwriting) that approximate each other. That schematic storage of information can have negative as well as positive consequences when, for example, learners of Latin-based scripts confuse “p” and “q.” With respect to those aspects of language learning that are biologically secondary, explicit instruction and conscious effort are vital. Neither explicit instruction nor conscious effort is required when learning the biologically primary skills associated with a first language. Immersion in the native-speaking culture is all that is required. In the absence of the distinction between biologically primary and secondary knowledge, the tendency to assume that all language learning should mimic the procedure of immersion in the language, which is used when learning to listen to and speak a first language, often has been overwhelming (see Roussel et al., 2017). It is a tendency that needs to be resisted. Simple immersion is never effective when dealing with biologically secondary information, an issue that is discussed in more detail with respect to second language acquisition below. While as indicated above, all principles including the borrowing and reorganizing principle are biologically primary, when applied to secondary knowledge, the borrowing and reorganizing principle requires explicit
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
teaching and conscious effort (Kirschner et al., 2006; Sweller et al., 2007). As an instructional design theory, cognitive load theory is concerned primarily with how to organize the presentation of what is said, written, or drawn by others in order to facilitate learning – see the section on Instructional Design and Cognitive Load Effects below. Consequently, the borrowing and reorganizing principle is central to the theory.
38.3.1.2 The Randomness as Genesis Principle Sometimes, we require information that we are not able to obtain from other people either because we do not have access to relevant information from others or because the required information has not as yet been generated. In the absence of sufficient knowledge held by ourselves, or by others whose knowledge is accessible to us via the borrowing and reorganizing principle, we can attempt to generate the information ourselves during problem-solving. All problem-solving when dealing with novel problems includes the random generation of problem-solving moves followed by tests of effectiveness. Effective moves are retained and ineffective ones are discarded. When faced with a problem for which we do not have a known solution ourselves and in the absence of a solution provided by someone else via the borrowing and reorganizing principle, we can use this random generate-and-test procedure to provide a complete problem solution. The randomness as genesis principle can be used without restriction as a means of acquiring information. It does not require other people or sources of information like the borrowing and reorganizing principle. Nevertheless, if it can be used, the borrowing and reorganizing principle is a far superior way of obtaining needed information. It is vastly more efficient. Consider someone learning to write an essay in a second language. One way of learning is to practice composing essays. In the absence of knowledge, composition consists of random generating and testing using the randomness as genesis principle. Alternatively, we could provide learners with model essays to study as a substitute. Providing model essays makes use of the borrowing and reorganizing principle. Empirical evidence using randomized controlled trials indicated superior essay-writing performance by students presented model essays to study (Kyun et al., 2013).
38.3.2 Processing and Storing of Information When novel information is encountered, it first needs to be processed in order to determine whether it is usable and required for current and future use, and if required, it then needs to be stored for future use. Two structures are required. Information first is processed in working memory using the narrow limits of change principle and may be stored in long-term memory using the information store principle.
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
863
864
JOHN SWELLE ET AL.
38.3.2.1 The Narrow Limits of Change Principle By definition, learners most frequently are confronted with novel rather than familiar information irrespective of whether it is obtained via the borrowing and reorganizing principle or the randomness-as-genesis principle, because if it is not novel and already fully learned, learning is not required. To be usable, that novel information must be appropriately processed, and working memory is the cognitive structure required to process novel information. Working memory has two critical characteristics. It is very limited in capacity (Miller, 1956) and duration (Peterson & Peterson, 1959). (Both of these papers were published before the term “short-term memory” was largely replaced by “working memory.” Nevertheless, the finding applies equally to working memory.) With respect to capacity, at any given time, we are unable to hold more than about 7 elements of novel information. We are unable to process more than about 3–4 elements of novel information, where processing elements means organising, comparing, and contrasting or working with the elements in some manner (Cowan, 2001). With respect to duration, we cannot hold information in working memory for more than about 20 seconds without refreshing the information by rehearsal (Peterson & Peterson, 1959). Several points need to be noted concerning the narrow limits of change principle. It only applies to working memory when it is dealing with novel, unfamiliar information such as unfamiliar words in a foreign language. The next two principles, with completely different characteristics, are concerned with previously learned and so familiar information. The narrow limits of change principle is important to cognitive load theory, and indeed, the theory’s name derived from this principle. It should be important to all instructional theories because characteristically, when teaching and learning, we usually deal with novel, not familiar information. The principle specifically is important when we are acquiring biologically secondary, domain-specific information. The limitations of working memory may not apply to the acquisition of biologically primary skills that we have evolved to acquire. With respect to language learning, while the narrow limits of change principle does not apply to the acquisition of the listening and speaking skills of a first language because those skills are biologically primary, it is central to the biologically secondary reading and writing skills and to second language listening and speaking skills. Children will effortlessly, unconsciously, and rapidly learn the complex biologically primary speaking rules used by their peer group. Simply hearing a large number of phrases in normal interactions is sufficient to induce the relevant grammatical structures. The same rules, learned as part of a second language, become biologically secondary and so in that case, are much more difficult to acquire and despite considerable conscious effort, for some adults, may never be adequately acquired. When learned as biologically secondary rather than primary rules, the limitations of working memory become critical. When
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
acquiring primary rules that we have specifically evolved to acquire, many elements of information can be processed simultaneously. When acquiring the same information as secondary rules, the same elements of information may be difficult to process in a limited working memory. The inability to process large numbers of elements in a limited working memory is likely to contribute, at least in part, to the relative difficulty of acquiring a second language compared to a first language. The difficulty of acquiring biologically secondary knowledge with its limited working memory association provided the raison d’être of cognitive load theory.
38.3.2.2 The information store principle The narrow limits-of-change principle as exemplified by the limitations of working memory when dealing with novel information, if considered in isolation, is unable to explain the heights to which the human cognitive system can reach. A structure able to hold large amounts of information for long periods of time is required. Long-term memory provides such a structure. In combination with working memory, the information store principle, which is instantiated by long-term memory in the human cognitive system, provides one of the missing pieces of the jigsaw. In contrast to working memory, long-term memory has no known capacity or duration limits. Everything that is learned must be stored in long-term memory if it is to be of permanent use to us. If nothing has been stored in long-term memory, nothing has been learned. The realization of the ultimate purpose of longterm memory was not obvious and provided one of the most important findings of cognitive psychology. The seminal work was carried out by De Groot (De Groot, 1946/1965; De Groot & Gobet, 1996). De Groot was concerned with the factors that explained why chess masters always defeated amateur players. The only difference De Groot could find between amateur players and masters was in memory of chess board configurations taken from real games. Chase and Simon (1973) obtained similar results but in addition found no differences between chess masters and amateurs for random board configurations. The superiority of chess masters only occurs for board configurations taken from real games, not for any board configurations. Similar results in a variety of educationally relevant fields have been obtained (Egan & Schwartz, 1979; Jeffries et al., 1981; Spilich et al., 1979; Sweller & Cooper, 1985). Experts have a far superior memory of combinations of elements that they are likely to encounter in their area of expertise. These findings established that skilled performance in areas of expertise is due to the contents of long-term memory. This structure is not just a random collection of items of information but rather is central to human cognitive functioning, including the high-level functioning characteristic of expert problem-solving. An expert is someone who has accumulated very large amounts of information in long-term memory over extended periods
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
865
866
JOHN SWELLE ET AL.
of practice allowing experts to recognize most of the problem states that they encounter and to know the best problem-solving moves for each state (Ericsson & Charness, 1994; Ericsson et al., 1993). Language provides strong evidence for the importance of long-term memory in skilled performance. A personal experiment similar to the ones referred to above easily reveals that importance. If readers are presented the letters “t ma e h t n taos t a c e h t” for 5 seconds and asked to repeat them in correct order, most will fail to repeat more than a few of them. In contrast, if readers are presented the same letters in the order: “t h e c a t s a t o n t h e m a t,” almost everyone will be able to repeat all of the letters with 100% accuracy. The difference is due to the contents of long-term memory available to competent readers of English, contents that took many years to acquire.
38.3.3 Using Stored Information to Govern Action The purpose of our cognitive system is to increase the likelihood of our taking action that is appropriate to our environment. The final principle, the environmental organizing and linking principle has this function. It provides a justification for each of the preceding four principles. 38.3.3.1 The Environmental Organising and Linking Principle Once information has been stored in long-term memory, it can be transferred back to working memory to govern appropriate action. Information from the environment is used as a trigger to determine which information is transferred. There are immeasurably large amounts of information stored in long-term memory, and environmental signals are required to ensure that the information that is selected for transfer back to working memory is appropriate to the environment. Information transferred to working memory then generates the required action. As indicated above in the discussion of the narrow limits of change principle, when processing novel information, working memory is severely limited in both capacity and duration. In contrast, when previously processed and stored information is transferred from long-term to working memory, the capacity and duration limits of working memory no longer apply. Working memory has no known capacity or duration limits when dealing with organized information transferred from long-term memory (Ericsson & Kintsch, 1995). The ability of a university teacher to discourse for extended periods of time on very complex issues provides an example. The environmental organizing and linking principle explains the transformational capacity of education. Once usable information is stored in long-term memory, it can be transferred to working memory, allowing us to act in a manner that would otherwise be inconceivable. We can think and solve problems with a degree of effectiveness that would be impossible if we had only our working memory to determine functioning. On this reasoning, the purpose of education is to facilitate the storage of usable
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
information in long-term memory. An expert in an area is someone who has stored large amounts of domain-specific, biologically secondary information in long-term memory. Language acquisition beyond the biologically primary listening and speaking skills exemplifies this process. For example, the written word “language” has been stored in long-term memory along with its many associations. When we see the complex squiggles that constitute the word embedded in a sentence, by using the environmental organizing and linking principle, those squiggles trigger the entire context of the word held in long-term memory. Despite the complexity of the squiggles and their associations, all the appropriate information linked with them can be transferred to working memory to determine a small part of the understanding associated with the relevant sentence in which the word is embedded. The same processes occur with respect to all the other words in the sentence and to relations between words. In that manner, we can derive meaning from the entire set of complex squiggles. The more familiar we are with the combinations of squiggles because they are entrenched in long-term memory, the easier it is to read a sentence and derive meaning. The less familiar we are with the squiggles, the more we have to engage in problem-solving and the harder it is to derive meaning. Such unfamiliar information must be processed as novel information in working memory with its capacity and duration limits. The advantages of an unlimited long-term memory that is available when dealing with familiar information is lost when dealing with unfamiliar information because the environmental organizing and linking principle cannot fully function.
38.4
Cognitive Load Theory
Cognitive load theory uses this cognitive architecture to devise novel instructional procedures. There are two basic, additive sources of cognitive load: Intrinsic and Extraneous cognitive load (Sweller, 2010). Intrinsic cognitive load refers to the essential complexity of the information that needs to be processed in order to learn. Consider someone who is learning the vocabulary of a second language. The task is very difficult because of the large number of words. Nevertheless, many of the words can be learned individually without reference to any other words. For example, the translation of the English word “cat” into the French word “chat” can be learned without reference to the translation of “dog” into “chien.” The elements that need to be learned do not interact and so can be learned independently. Because element interactivity is low, working memory load is low despite the fact that with a large number of vocabulary items that need to be learned, the task is difficult. For this task, intrinsic cognitive load is low. Other tasks may have fewer elements but because they interact, element interactivity is high and so intrinsic cognitive load is high. For example,
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
867
868
JOHN SWELLE ET AL.
learners of English have to learn appropriate word orders. It is appropriate to say “when learning a language” but inappropriate to say “a language learning when.” In order to learn appropriate word orders in English, each of the words and their order must be learned, and so the individual words are elements that interact rather than being independent of each other. Element interactivity is relatively high, resulting in a high intrinsic cognitive load due to the amount of information that must be simultaneously processed in working memory. It needs to be emphasized that element interactivity and task difficulty are different. A task such as learning many of the nouns of a second language can be very difficult because of the many elements but have very low element interactivity, allowing the elements to be learned independently. Such tasks do not impose a heavy cognitive load. Other tasks can be difficult despite only requiring the learning of a few elements. If those elements interact, a heavy cognitive load is imposed, and it is that cognitive load that renders the task difficult. Cognitive load theory applies primarily to tasks that are difficult because of high element interactivity rather than because there are many elements per se (Sweller & Chandler, 1994). The cognitive architecture described above also is critical when determining element interactivity. The number of interacting elements is not just determined by the structure of the information. It also is determined by the knowledge held in long-term memory. The written word “language” is a single element for readers of this book. For someone beginning to learn to read English, it may consist of eight elements, one for each letter of the word. For someone learning to read the Latin alphabet, the number of elements is likely to be considerably more than eight. Element interactivity cannot be measured mechanically simply by considering the structure of the information. What is held in long-term memory also needs to be considered. Since we need to consider the knowledge of learners, element interactivity can be estimated only by counting the assumed number of elements for the population under consideration, rather than precisely determined. It follows that intrinsic cognitive load can be altered either by changing what learners are required to learn or by changing what they hold in long-term memory. Intrinsic cognitive load should be optimized so that working memory is fully used but not overloaded. Extraneous cognitive load also is determined by element interactivity (Sweller, 2010). Unlike intrinsic cognitive load, which is determined by the intrinsic properties of the area being learned and the knowledge of learners, extraneous cognitive load is determined by the manner in which the information is presented and the activities required of learners. In other words, extraneous cognitive load is determined by instructional design.
38.4.1 Instructional Design and Cognitive Load Effects While element interactivity associated with intrinsic cognitive load should be optimized, in the case of extraneous cognitive load, element interactivity
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
always should be decreased by altering instructional designs. Since intrinsic and extraneous cognitive load are additive, any increase in extraneous cognitive load will decrease learning due to a decrease in working memory resources available to be devoted to the element interactivity associated with intrinsic cognitive load. (Resources devoted to dealing with intrinsic rather than extraneous cognitive load are sometimes referred to as “germane resources” or germane load.) Instructional procedures can be designed using the above cognitive architecture and the concepts of element interactivity, along with intrinsic and extraneous cognitive load. Cognitive load theory consists of the above theoretical concepts as well as the instructional designs generated by the theory. Using randomized, controlled trials, each design is tested for effectiveness by comparing it with a more conventionally used alternative. When the new design proves superior to its alternative, a new cognitive load effect has been generated. A large number of cognitive load effects have been identified (Sweller et al., 2019). Each effect provides a prescription for instructional design. Several of the designs are directly relevant to language instruction and will be discussed next.
38.4.1.1 The Worked Example Effect This effect occurs when learners who are presented worked examples to study rather than problems to solve obtain statistically higher test scores. It assumes that competent problem solvers gain their competence from the acquisition of large numbers of domain-specific problem solutions for the various problems that they will encounter in an area. Learners can acquire those solutions either by using the randomness-as-genesis principle during problem-solving or by the borrowing and reorganizing principle when obtaining the same information from other people. The randomness-asgenesis principle increases extraneous cognitive load compared to the borrowing and reorganizing principle by unnecessarily increasing element interactivity. Searching for and constructing a novel solution requires many more elements to be considered than being presented a solution. This hypothesis was tested in a language-based area by Kyun et al. (2013). They tested Korean students being taught to write essays in English about children’s fairy tales. Composing an essay is a problem-solving task. One group of students were presented the essay topics and had to compose their essays themselves. Another group were given the same topics but in addition were presented with model answers to study. The model answers provided both content and structural information. Students for whom this topic was novel and consequently difficult performed at a higher level on subsequent test essay questions after studying the worked examples than after composing an essay themselves, thus providing evidence for the hypothesis. Similar results were not obtained testing more knowledgeable students due to the expertise reversal effect, discussed below.
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
869
870
JOHN SWELLE ET AL.
38.4.1.2 The Split-Attention Effect This effect occurs when learners are faced with multiple sources of information that must be mentally integrated in order to be understood. Physically integrating the information eliminates the need for mental integration and so reduces working memory resources needed for mental integration. If physically integrated information results in increased test performance compared to the same information presented in a split-source format, the split-attention effect is obtained. Some of the results obtained by Yeung et al. (1998) replicated the splitattention effect in language learning. In two experiments using both native English speakers and English as a second language speakers, Yeung et al. presented learners with passages containing vocabulary with which they were unlikely to have a high familiarity. For example, a passage that includes the sentence “It never thinks of roaming unless it fears there is a drought at hand” will not be intelligible if learners do not understand the meaning of the word “roaming.” The meaning of the word – “travelling without an aim” – can be provided separately in a vocabulary list after the passage or it can be presented immediately above the occurrence of the word in the passage. Using a vocabulary list provides an example of splitattention. In order to understand the sentence, learners who do not understand the word must split their attention between the passage and the vocabulary list. The alternative format physically integrates the word meaning with the passage. The results of these experiments indicated that for novice learners, the integrated format increased comprehension of the passage, demonstrating the split-attention effect. Nevertheless, additional results indicated that the split-attention format increased vocabulary knowledge due to the redundancy effect (see next) with a complete reversal of the results using more knowledgeable learners due to the expertise reversal effect (discussed below).
38.4.1.3 The Redundancy Effect The split-attention effect occurs when at least one source of information is unintelligible without considering other sources of information. In contrast, the redundancy effect occurs when all sources of information are intelligible in isolation. We frequently assume that presenting learners with additional, unnecessary information at worst will have no adverse consequences and may have positive effects. For example, presenting the same information simultaneously in written and spoken form might be beneficial by allowing learners to attend to both or just to the source that they prefer. In fact, redundancy is a major source of extraneous cognitive load. For example, if learners attend to identical versions of both written and spoken information, as often occurs during PowerPoint presentations, they must coordinate the two redundant sources of information. Requiring learners to unnecessarily
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
coordinate multiple sources of information increases element interactivity and imposes an extraneous cognitive load. Kalyuga et al. (2004) found that learners asked to both read and listen to text simultaneously learned less than learners who just listened to the text or who were presented the text in both modalities successively rather than simultaneously. Redundancy also should be taken into account when teaching second language students to listen and to read. They should not be taught to listen and read simultaneously. Diao et al. (2007) found that native Chinese speakers who were learning to listen to English could understand auditory text better if it was accompanied by written text, but they were poorer on a subsequent listening task. The written text was beneficial if the aim was to assist in comprehension but was redundant if the aim was learning to comprehend purely auditory input. With respect to learning to read, Diao and Sweller (2007), again teaching native Chinese speakers, found that reading comprehension increased when learners practiced reading without the presence of auditory text. When learning to read, the presence of auditory text is redundant and so should be omitted. When discussing the split-attention effect above (Yeung et al., 1998), it was pointed out that for novice learners, a separate vocabulary list interfered with passage comprehension compared to physically integrated information. Nevertheless, if the aim is vocabulary acquisition, Yeung et al. found the separate list superior. In the case of vocabulary acquisition, the presence of the passage is redundant, and the separate list better allows learners to ignore the redundant passage, resulting in superior performance on vocabulary acquisition because more working memory resources can be applied to vocabulary learning. These results were all obtained using novice learners. They were reversed using more knowledgeable learners (the expertise reversal effect) as discussed in the next section.
38.4.1.4 The Expertise Reversal Effect Most cognitive load effects apply to novice learners who are just beginning to learn a particular, high element interactivity concept or procedure. As indicated above, as levels of expertise increase, element interactivity decreases. In turn, decreases in element interactivity decrease the need to decrease working memory load because it is already low. An instructional design intended to reduce extraneous cognitive load has a reduced advantage and may even result in a reversal of the expected cognitive load effect. This expertise reversal effect is due to the redundancy effect. For example, as indicated by Kyun et al. (2013), when learning to write English essays, novices find it advantageous to study model essays rather than to compose essays themselves. This result was obtained by Kyun et al. using novice learners. With increases in learner expertise, the advantage of studying the model answers decreased. At some point, studying worked examples becomes redundant. More learning occurs by practicing writing essays oneself.
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
871
872
JOHN SWELLE ET AL.
Yeung et al. (1998) obtained a similar expertise reversal effect associated with the split-attention effect. Recall, they found that text comprehension improved by providing vocabulary information close to the difficult vocabulary item, due to reduced split-attention effects. This effect was obtained using less able learners. Testing more able learners, the effect reversed. For more able learners, additional vocabulary information was redundant. That information could be easily ignored if it was presented in a separate list. When physically integrated with the text, it was much harder to ignore. By unnecessarily processing the redundant vocabulary information, more working memory resources were required and so for these more expert learners, physical integration reduced rather than increased comprehension.
38.4.1.5 The Transient Information Effect Some categories of information impose a heavy cognitive load because the information is transient. Spoken information belongs to this category, and so transience is directly relevant to language learning. When people listen to speech, what is heard now will need to stay in working memory if it is relevant to what will be heard later, because what is heard later may otherwise be unintelligible without the current information. Written text was invented precisely in order to turn transient spoken information into something that is permanent, allowing this problem to be less likely to arise. In contrast, if we cannot hold essential written information in working memory, we can return to it repeatedly, a practice that is impossible with spoken information, unless it is recorded. The difficulty associated with transient spoken information has instructional implications when learning to listen in a second language. MoussaInaty et al. (2012) conducted a study testing native Arab speakers learning to listen to English. They found that these learners acquired greater listening skills by reading rather than listening. They attributed this improbable finding to the transience of spoken information. When reading, learners could return to or spend more time on difficult elements of information, which is impossible when actually listening. Having to hold transient information in working memory can impose an impossibly high working memory load. Jiang et al. (2018) replicated Moussa-Inaty et al.’s (2012) result testing Chinese students learning English. While the result is readily replicable and so we can conclude that the ability to listen can improve more when reading than actually listening, it is obvious that at some point listeners have to listen rather than just read. Jiang et al. found that when they tested less knowledgeable students who had had only limited exposure to the second language (Chinese students learning English in one experiment and Chinese students learning French in another experiment), the reverse result was obtained. Listening was superior to reading when learning to listen for these novice learners who had to have some experience listening.
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
These results provide another example of the expertise reversal effect. Optimal instructional designs are strongly determined by levels of expertise. The expertise reversal effect provides a natural limit to the effectiveness of all other cognitive load effects, including the transient information effect. Increases in expertise result in decreases in element interactivity, and any task that is low in element interactivity may not impose a sufficiently high cognitive load to generate a cognitive load effect (Chen et al., 2017).
38.4.1.6 The Isolated Elements Effect and the Variability Effect These two effects are concerned with alterations in intrinsic cognitive load. It may be recalled that unlike extraneous cognitive load that should always be reduced, intrinsic cognitive load should be optimized. If too much information is presented for working memory to process, it needs to be reduced for the same reason that extraneous cognitive load needs to be reduced, but if too little information is presented, intrinsic cognitive load needs to be increased in order to increase the amount that is learned. The use of isolated elements decreases intrinsic cognitive load while the use of variability increases intrinsic cognitive load. Lu et al. (2020) tested the isolated elements and variability effects when students were learning to write Chinese characters. The Integratedintegrated group of learners was presented complex Chinese characters that they had to learn by tracing the full, integrated character three times followed by a practice test. This procedure was followed twice. The Isolatedintegrated group had the same characters divided into three components. They traced one of the components three times followed by the same procedure for each of the remaining two components, following an AAABBBCCC pattern. They then were presented the fully integrated character in a manner identical to the Integrated-integrated group. Lastly, the Variable-integrated group followed the same procedure as the Isolatedintegrated group, except that rather than being presented with a single component to trace three times in succession, they were required to trace each of the three components in succession three times (an ABCABCABC pattern) followed by the fully integrated character. Test results indicated that the Isolated-integrated group learned more than the Integrated-integrated group, demonstrating the isolated elements effect. Learning the isolated elements first, reduced working memory load sufficiently to enhance learning of the full character compared to attempting to initially learn the full character. Furthermore, when comparisons were made with the variable-integrated group, it was clear that working memory load was sufficiently reduced by the isolated elements to allow learners to increase learning by simultaneously comparing and contrasting the three segments of the character. The Variable-integrated group was superior to both of the other two groups, demonstrating the variability effect. Based on these results, the best way to learn to write Chinese
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
873
874
JOHN SWELLE ET AL.
characters is to divide complex characters into their simpler components and present those simpler components sequentially prior to students learning the full character.
38.4.1.7 Simultaneous Learning of Content and a Second Language Learning subject content in a second language requires learners to deal with two domain-specific, biologically secondary fields simultaneously. We would expect the burden on limited working memory to be severe. Roussel et al. (2017) asked participants to read a text in (1) a first language only condition (no second language involved, only content learning involved); (2) in a second language only condition (second language learning and content learning involved); (3) and in a bilingual version (second language is presented and translated into the first language, with content learning also involved). Some critical words and phrases were underlined and highlighted in bold. In the third version, the sentences were alternated between the two languages, and underlined critical words and phrases were linked with corresponding translations by arrows. The attentional resources of the learners were consequently explicitly devoted to linguistics forms. In a set of three experiments with French-speaking students, the content to be learned was varied (European Court of Justice, with law students; Machine language, with computer science students) as well as varying the second language used (German or English). The results showed that the best performance was obtained in the third condition both for content and language learning. These experiments have been replicated with oral material (Roussel et al., in press; Roussel, 2019). French students of law and political science listened to an audio document about the European Court of Humans Rights under one of the four following experimental conditions: in their first language (French) twice; in a second language (German) twice; first in French, then in German; or first in German, then in French. After the listening task, the participants were tested about their understanding of the German language and of the academic content. The results indicated again that listening in the second language only was never the best condition for learning either content or the second language. On the contrary, listening to the content in French before listening to it in a second language was beneficial for both content and language learning. This conclusion is in line with research indicating the need for pedagogical adjustments in situations where content and language have to be learned simultaneously. These results all illustrate that attempting to simultaneously learn content and a second language, both biologically secondary tasks, can overwhelm working memory resources. They also indicate how research on working memory can be used to discuss curricula and education policies. Indeed, at all levels of education in Europe, for example, more and more courses in different academic disciplines are taught in languages other than the official language of the country (mostly in English). The aim of these
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
approaches is well summarized by the expression “killing two birds with one stone” (Dallinger et al., 2016) that is to say, teaching both academic content and a second language at the same time without a loss to either content or language learning. In higher education, this approach is called English as a Medium of Instruction (EMI), while at secondary school level it is called Content and language integrated learning (CLIL). As is often the case in education, the implementation of these approaches is paved with good political intentions such as globalization, development of students’ language skills to prepare them to work internationally, competition between universities, attractiveness of international programs, and participation in the construction of a nation capable of aligning itself with its Englishspeaking neighbors (Dearden, 2015). Research results on the impact of these pedagogical practices at the secondary school level, both on language learning and on the learning of disciplinary content remain very heterogeneous and controversial (Bruton, 2013; Hüttner & Smit, 2014; Cenoz et al., 2013; Dalton-Puffer et al., 2014). In higher educational contexts, Macaro et al. (2018) concluded “that the research evidence to date is insufficient to assert that EMI benefits language learning nor that it is clearly detrimental to content learning” (p. 36). On a methodological level, it seems also that a great deal of EMI and CLIL research compared the performance of learners who take academic courses in a second language with their peers who learn academic content in the first language with second language classes at the same time. As underlined by numerous researchers (Gablasova 2014a, 2014b; Bruton, 2013; Rumlich, 2016) both groups of students often do not have the same time exposure, the same practice of the second language, the same average general aptitude level nor the same motivation for learning languages. This kind of comparison can sometimes yield good results and support the political will to internationalize training, but they undoubtedly tend to overestimate the pedagogical effectiveness of these approaches.
38.5
Conclusion
Cognitive load theory, based on an evolutionary approach that distinguishes between biologically primary and secondary knowledge, assumes that second language learning is biologically secondary knowledge. In contrast, first language learning is biologically primary knowledge. This distinction has several consequences: (1) second language learning is demanding of working memory resources; (2) designing second language instruction should optimize the use of working memory resources; (3) several effects obtained in cognitive load theory research are relevant to language learning; (4) there is increasing evidence that learning a second language and content simultaneously may be too demanding of working memory resources to be effective. This view of working memory resources involved in second language learning is opposed to theories of language learning that assume that we should
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
875
876
JOHN SWELLE ET AL.
learn a second language implicitly by simple immersion and exposure (DaltonPuffer, 2007; Krashen, 1985; Krashen & Terrell, 2000; Nikula et al., 2013). According to these theories, learning academic content through a second language is seen as an excellent way of reproducing a “natural” learning context. It is assumed that by attending a course in a second language, students will learn the language implicitly, without explicit instruction in the same way they learn their first language. Other second language acquisition theories (Long, 1991; Robinson, 2003; Schmidt, 2001) have been used to argue that it is not possible to learn a second language implicitly while acquiring academic context. We strongly agree based on human cognitive architecture and especially on working memory characteristics. Second language learning is a domain where working memory research can inform instructional design, and more generally, education policies.
References Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Prentice Hall. Bartlett, F. C. (1932). Remembering: A study in experimental and social psychology. Macmillan. Bruner, J. (1978). The role of dialogue in language acquisition. In A. Sinclair, R. Jarvella, & W. Levelt (Eds.), The child’s conception of language (pp. 241–256). Springer-Verlag. Bruton, A. (2013). CLIL: Some of the reasons why. . .and why not. System, 41, 587–597. Cenoz, J., Genesee, F., & Gorter, D. (2013). Critical analysis of CLIL: Taking stock and looking forward. Applied Linguistics, 35, 243–262. https://doi.org/ 10.1093/applin/amt011 Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55–81. Chen, O., Kalyuga, S., & Sweller, J. (2017). The expertise reversal effect is a variant of the more general element interactivity effect. Educational Psychology Review, 29, 393–405. Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24, 87–114. Dallinger, S., Jonkmann, K., Hollm, J., & Fiege, C. (2016). The effect of content and language integrated learning on students’ English and history competences: Killing two birds with one stone? Learning and Instruction, 41, 23–31. Dalton-Puffer, C. (2007). Discourse in content and language integrated (CLIL) classrooms. John Benjamins. Dalton-Puffer, C., Llinares, A., Lorenzo, F., & Nikula, T. (2014). “You can stand under my umbrella”: Immersion, CLIL and bilingual education.
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
A response to Cenoz, Genesee, & Gorter (2013). Applied Linguistics, 35, 213–218. Dearden, J. (2015). English as a medium of instruction: A growing global phenomenon. British Council. De Groot, A. (1946/1965). Thought and choice in chess. Mouton. De Groot, A., & Gobet, F. (1996). Perception and memory in chess: Heuristics of the professional eye. Van Gorcum. Diao, Y., Chandler, P., & Sweller, J. (2007). The effect of written text on learning to comprehend spoken English as a foreign language. American Journal of Psychology, 120, 237–261. Diao, Y., & Sweller, J. (2007). Redundancy in foreign language reading instruction: Concurrent written and spoken presentations. Learning and Instruction, 17, 78–88. Egan, D. E., & Schwartz, B. J. (1979). Chunking in recall of symbolic drawings. Memory & Cognition, 7, 149–158. Ericsson, K. A., & Charness, N. (1994). Expert performance; Its structure and acquisition. American Psychologist, 49, 725–747. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102, 211–245. Ericsson, K. A., Krampe, R., & Tesch-Romer, C. (1993). The role of deliberate practice in the acquisition of expert performance. Psychological Review, 100, 363–406. Fischer, F., Chinn, C. A., Engelmann, K., & Osborne, J. (Eds.). (2018). Scientific reasoning and argumentation: The roles of domain-specific and domain-general knowledge. Routledge. Gablasova, D. (2014a). Issues in the assessment of bilingually educated students: Expressing subject knowledge through L1 and L2. The Language Learning Journal, 42(2), 151–164. Gablasova, D. (2014b). Learning and retaining specialized vocabulary from textbook reading: Comparison of learning outcomes through L1 and L2. The Modern Language Journal, 98(4), 976–991. Gass, S., & Mackey, A. (2007). Input, interaction, and output in second language acquisition. In B. VanPatten, & J. Williams (Eds.), Theories in second language acquisition (pp. 175–199). Erlbaum. Geary, D. (2008). An evolutionarily informed education science. Educational Psychologist, 43, 179–195. Geary, D. (2012). Evolutionary educational psychology. In K. Harris, S. Graham, & T. Urdan (Eds.), APA educational psychology handbook (Vol. 1, pp. 597–621). American Psychological Association. Geary, D., & Berch, D. (2016). Evolution and children’s cognitive and academic development. In D. Geary & D. Berch (Eds.), Evolutionary perspectives on child development and education (pp. 217–249). Springer. Hüttner, J., & Smit, U. (2014). CLIL (Content and Language Integrated Learning): The bigger picture. A response to: A. Bruton (2013). CLIL: Some of the reasons why. . .and why not. System, 41, 587–597. System, 44, 160–167.
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
877
878
JOHN SWELLE ET AL.
Jeffries, R., Turner, A., Polson, P., & Atwood, M. (1981). Processes involved in designing software. In J. R. Anderson (Ed.), Cognitive skills and their acquisition (pp. 255–283). Erlbaum. Jiang, D., Kalyuga, S., & Sweller, J. (2018). The curious case of improving foreign language listening skills by reading rather than listening: An expertise reversal effect. Educational Psychology Review, 30, 1139–1165. Kalyuga, S., Chandler, P., & Sweller, J. (2004). When redundant on-screen text in multimedia technical instruction can interfere with learning. Human Factors, 46, 567–581. Kirschner, P., Sweller, J., & Clark, R. (2006). Why minimal guidance during instruction does not work: An analysis of the failure of constructivist, discovery, problem-based, experiential and inquiry-based teaching. Educational Psychologist, 41, 75–86. Krashen, S. D. (1985). The input hypothesis: Issues and implications. Longman. Krashen, S. D., & Terrell, T. D. (2000). The natural approach: Language acquisition in the classroom (2nd ed.). Prentice Hall. Kyun, S., Kalyuga, S., & Sweller, J. (2013). The effect of worked examples when learning to write essays in English literature. Journal of Experimental Education, 81, 385–408. Long, M. (1991). Focus on form: A design feature in language teaching methodology. In K. de Bot, R. Ginsberg, & C. Kramsch (Eds.), Foreign language research in cross-cultural perspective (p. 39–52). John Benjamins. Lu, J., Kalyuga, S., & Sweller, J. (2020). Altering element interactivity and variability in example-practice sequences to enhance learning to write Chinese characters. Applied Cognitive Psychology, 34, 837–843. Macaro, E., Curle, S., Pun, J., An, J., & Dearden, J. (2018). A systematic review of English medium instruction in higher education. Language Teaching, 51, 36–76. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Moussa-Inaty, J., Ayres, P., & Sweller, J. (2012). Improving listening skills in English as a foreign language by reading rather than listening: A cognitive load perspective. Applied Cognitive Psychology, 26, 391–402. Nikula T., Dalton-Puffer, C., & Llinares, A. (2013). CLIL classroom discourse: Research from Europe, Journal of Immersion and Content-Based Language Education, 1, 70–100. Paas, F., & Sweller, J. (2012). An evolutionary upgrade of cognitive load theory: Using the human motor system and collaboration to support the learning of complex cognitive tasks. Educational Psychology Review, 24, 27–45. Peterson, L., & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58, 193–198. Piaget, J. (1928). Judgement and reasoning in the child. Harcourt. Robinson, P. (2003). Attention and memory during SLA. In C. Doughty, & M. Long (Eds.), Handbook of second language acquisition (pp. 631–662). Blackwell.
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
Cognitive Load Theory
Roussel, S., Joulia, D., Tricot, A., & Sweller, J. (2017). Learning subject content through a foreign language should not ignore human cognitive architecture: A cognitive load theory approach. Learning and Instruction, 52, 69–79. Roussel, S., Tricot, A., & Sweller, J. (in press). The advantages of listening to academic content in a second language may be outweighed by disadvantages: A cognitive load theory approach. British Journal of Educational Psychology. Roussel, S. (2019). Comprendre un contenu disciplinaire en allemand L2: quels ajustements pédagogiques? Études en didactique des langues (EDL), 32, 89–100 Rumlich, D. (2016). Evaluating bilingual education In Germany. Clil students’ general English proficiency, Efl self-concept and interest. Peter Lang. Schmidt, R. (2001). Attention. In P. Robinson (Ed.), Cognition and second language instruction (p. 3–32). Cambridge University Press. Spilich, G., Vesonder, G., Chiesi, H., & Voss, J. (1979). Text processing of domain-related information for individuals with high and low domain knowledge. Journal of Verbal Learning and Verbal Behavior, 18, 275–290. Sweller, J. (2010). Element interactivity and intrinsic, extraneous and germane cognitive load. Educational Psychology Review, 22, 123–138. Sweller, J. (2015). In academe, what is learned and how is it learned? Current Directions in Psychological Science, 24, 190–194. Sweller, J. (2016). Working memory, long-term memory and instructional design. Journal of Applied Research in Memory and Cognition, 5, 360–367. Sweller, J., Ayres, P., & Kalyuga, S. (2011). Cognitive load theory. Springer. Sweller, J., & Chandler, P. (1994). Why some material is difficult to learn. Cognition and Instruction, 12, 185–233. Sweller, J., & Cooper, G. (1985). The use of worked examples as a substitute for problem solving in learning algebra. Cognition & Instruction, 2, 59–89. Sweller, J., Kirschner, P., & Clark, R. E. (2007). Why minimally guided teaching techniques do not work: A reply to commentaries. Educational Psychologist, 42, 115–121. Sweller, J., & Sweller, S. (2006). Natural information processing systems. Evolutionary Psychology, 4, 434–458. Sweller, J., van Merriënboer, J., & Paas, F. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10, 251–296. Sweller, J., van Merriënboer, J., & Paas, F. (2019). Cognitive architecture and instructional design: 20 years later. Educational Psychology Review, 31, 261–292. Thornton, A., & Raihani, N. (2008). The evolution of teaching. Animal Behaviour, 75, 1823–1836. Tindall-Ford, S., Agostinho, S., & Sweller, J. (Eds.). (2020). Advances in cognitive load theory: Rethinking teaching. Routledge. Tomasello, M. (2009). Why we cooperate. MIT press.
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
879
880
JOHN SWELLE ET AL.
Tricot, A., & Sweller, J. (2014). Domain-specific knowledge and why teaching generic skills does not work. Educational Psychology Review, 26, 265–283. Vygotsky, L. S. (1978), Mind in society, The development of higher psychological processes. Harvard University Press. Yeung, A. S., Jin, P., & Sweller, J. (1998). Cognitive load and learner expertise: Split-attention and redundancy effects in reading with explanatory notes. Contemporary Educational Psychology, 23, 1–21.
https://doi.org/10.1017/9781108955638.045 Published online by Cambridge University Press
39 Working Memory Training Meta-Analyses and Clinical Implications Domenico Tullo and Susanne M. Jaeggi
39.1
Introduction
Over the past two decades, there has been increasing interest in cognitive training and the idea that human cognition can be remediated, improved, and even enhanced (Green et al., 2019; Simons et al., 2016). Cognitive training is based on the idea that repeated practice on a task that targets specific aspects of cognition can translate to improvement in other domains that are related to the trained target. While there are many approaches to cognitive training, interventions that target working memory (WM) and related executive functions (EF) are among the most frequently implemented and have demonstrated success in various populations (Au et al., 2014; Klingberg, 2010; Weicker et al., 2016), and thus, those interventions will be the focus of the current chapter. WM is the cognitive process involved in storing and manipulating stimuli, which allows individuals to complete commonplace tasks (e.g., mental math operations, recalling a list of digits, solving puzzles; Kane et al., 2007). In addition to storage and manipulation, WM typically involves other processes, such as attention, inhibition, and interference resolution, and thus, characteristic for EF, WM is far from process pure (Diamond, 2013; Engle et al., 1999; Kane et al., 2004). Therefore, many WM training paradigms attempt to account for the multidimensional nature of WM (Shah & Miyake, 1999) and incorporate several processes in order to maximize the intervention’s benefits. For example, a training paradigm might be targeting both storage and manipulation processes by using a variant of the well-known complex span task as a training paradigm, in which participants have to recall an increasing number of stimuli while making decisions about the items (e.g., their orientation; Loosli et al., 2012); whereas other paradigms also incorporate interference and inhibitory control processes in addition to updating, (e.g., via an n-back paradigm), in which participants have to keep track of and recall an increasing number of
https://doi.org/10.1017/9781108955638.046 Published online by Cambridge University Press
882
DOM EN ICO TULLO AND SUSANNE M. JAEGGI
items in order (Jaeggi et al., 2008). Typically, paradigms that are used for training are different from assessment versions of the tasks in that they are adaptive, that is, they become more difficult as participants get better at the task, or easier if participants struggle, in order to adjust to participants’ fluctuating WM resources and/or skills (Jaeggi et al., 2008, 2011; Stepankova et al., 2014). The complexity of training WM is even further highlighted in gamelike implementations of an intervention, given that in addition to the task demands themselves, participants have to process additional information, such as the game environment that might be distracting, feedback, and/or incentives (Deveau et al., 2015; Katz et al., 2014; Pergher et al., 2019). The drawback of using complex interventions is that it is hard to determine the exact underlying mechanisms that might drive any benefits (e.g., Gibson et al., 2012; Webb et al., 2018). As such, this limitation illustrates the discrepancy between theoretical and applied goals (e.g., Wass et al., 2012). The importance of targeting WM processes via training is highlighted by research that has demonstrated the extensive differences in how well individuals can store and manipulate information, and how well they are able to maintain attention, ignore distractors, and resolve interference (Kane et al., 2007; Tullo, Faubert, et al., 2018). Individual differences in WM are indicative of one’s ability to complete a variety of complex tasks including reasoning and problem-solving (Conway et al., 2003; Cowan et al., 2005), and they are predictive for scholastic achievement and job success (Diamond, 2013). Individuals with low WM capacities; that is, a smaller capacity of cognitive resources that can be allocated to satisfy the task’s demands, demonstrate a lower probability to successfully complete any given task compared to those with high WM capacities (Kane et al., 2007). WM capacity is also related to the presence of maladaptive behaviors (Fabiano et al., 2009; Jensen et al., 2001; Knight et al., 2008). For instance, disruptive classroom behavior is more prevalent in children with poor WM (Barriga et al., 2002). Taken together, the links between WM, higher-order cognition, and critical components of everyday functioning, coupled with the promising findings demonstrating benefits of WM training in typically developing populations (Au et al., 2014; Karbach & Verhaeghen, 2014), have justified the rationale for the use of this approach to improve cognition for those demonstrating deficits in this domain (Chacko et al., 2013; Cortese et al., 2015; Sonuga-Barke et al., 2013, 2014). Specifically, WM training aims to address these individual differences across the general population, and more specifically between typically and atypically developing populations, where the intervention can serve to mitigate these individual differences (e.g., Wang et al., 2019). One population frequently targeted for WM training are individuals diagnosed with Attention-Deficit/Hyperactivity Disorder (ADHD), who often struggle with allocating sufficient cognitive resources to task demands while suppressing the allocation of their resources to task-irrelevant and/or
https://doi.org/10.1017/9781108955638.046 Published online by Cambridge University Press
Meta-Analyses and Clinical Implications
distracting information (Forster et al., 2014; Passolunghi et al., 2005). Nonpharmacological and cost-effective interventions such as WM training have been suggested as another potential approach to treat ADHD, specifically as an alternative and/or supplemental form of treatment (Jones et al., 2020; Shah et al., 2012; Sonuga-Barke et al., 2014). The demand for alternative treatments has become more prevalent given ongoing discussions about the viability of traditional pharmacological interventions due to pervasive side effects (Gau et al., 2008), issues with persistence (Coghill, 2019), and loss of efficacy or effectiveness over time (Swanson, 2019), all of which have been shown to be contributing factors toward treatment cessation (Biederman et al., 2008; Molina et al., 2013). Likewise, psychotherapy (e.g., Cognitive Behavioral Therapy) offered to individuals diagnosed with ADHD is limited by efficacy (Sonuga-Barke et al., 2013), can be burdensome to deliver and receive, and is costly (Jensen et al., 2001; Saxena et al., 2007). Indeed, WM training for the ADHD population has shown some promise, as demonstrated the translation of benefits from training across an array of targeted domains (see Jones et al., 2020; Loosli et al., 2012; Salmi et al., 2020; Sonuga-Barke et al., 2014). However, the validity of WM training’s benefits remains a contentious issue. The divide over the validity of the industry derives from the unsubstantial methodology used by cognitive training companies to validate their training programs (Simons et al., 2016). In parallel to the growth in research in cognitive training, the emergence of companies specializing in cognitive training, or “brain training,” has developed into a multibillion-dollar market (Simons et al., 2016). Recently, Akili labs has received approval from the Food and Drug Administration (FDA) to treat individuals diagnosed with ADHD using their cognitive training program (Commissioner, 2020; Kollins et al., 2020). Nevertheless, many companies have made exaggerated claims about potential benefits, triggering punitive intervention by US regulatory agencies (Simons et al., 2016). Like other cognitive training programs and related interventions, the viability of WM training is evaluated by improvement on separate domain-specific measures such as related cognitive-behavioral tasks, observed behavior ratings, self-reports, and/or academics (Melby-Lervåg & Hulme, 2013). The translation of the training program’s benefits can be qualified by the degree of similarity between the underlying mechanisms of the training task and the targeted outcome measure (Barnett & Ceci, 2002). Typically, the translation of benefits from the training paradigm to the outcome measure has been dichotomized as near- and far-transfer (Simons et al., 2016). To illustrate, an effect of near-transfer is commonly examined by testing the effects of a training paradigm (e.g., using a visuospatial span task) on a nontrained variant of the task (e.g., a span task that uses different, nonvisuospatial stimuli). Another example of near-transfer, or the translation of benefits to proximal outcomes, is best exemplified by training on a spatial variant of an n-back task and evaluated by performance
https://doi.org/10.1017/9781108955638.046 Published online by Cambridge University Press
883
884
DOM EN ICO TULLO AND SUSANNE M. JAEGGI
on an object-based variant of n-back. Likewise, researchers examine the translation of WM training benefits to more distal targeted outcomes such as behavioral reports, ADHD symptoms, and academics. Overall, the field of cognitive training is highly contentious with those strongly encouraging more research toward the programs’ viability (Au et al., 2014, 2020; Green et al., 2019; Karbach & Verhaeghen, 2014) and those strongly opposing this area of research altogether (Melby-Lervåg et al., 2016; Melby-Lervåg & Hulme, 2013, 2015; Sala & Gobet, 2017; Shipstead et al., 2010; Shipstead, Hicks, et al., 2012; Shipstead, Redick, et al., 2012). The primary point of contention is related to the extent of benefits observed. Currently, the meta-analyses, systematic reviews, and editorials garnering greatest attention (i.e., citations) argue that there is little evidence for the generalization of WM training and suggest that any evidence is limited to proximal targeted outcomes, and thus, they question the utility of WM training overall (Melby-Lervåg et al., 2016; Melby-Lervåg & Hulme, 2013, 2015; Sala & Gobet, 2017; Shipstead et al., 2010; Shipstead, Hicks, et al., 2012; Shipstead, Redick, et al., 2012). In comparison, there are fewer analyses that have focused on clinical populations, or more specifically, individuals that: (1) are typically underserved and underrepresented in this area of research, and (2) could benefit the most from this alternative and/or supplemental information given the impact of cognitive deficits characteristic of this population (Jones et al., 2020; Shah et al., 2012). The current chapter provides a synthesis of the findings from 19 metaanalyses, published in peer-reviewed journals, on cognitive training that include both children as well as clinical populations1 (i.e., children diagnosed with ADHD or other developmental disabilities; cf. Table 39.1). The meta-analyses were gathered using PsycInfo, PubMed, and Scopus using the search terms: “WM training” or “executive function training” or “attention training” and “meta-analysis” or “systematic review.” We included “executive function training” and “attention training” in addition to “WM training” in order to maximize the reach into domains that capture general WM processes and to reflect the complexity of WM as a construct. Metaanalyses were excluded if they did not include studies evaluating cognitive training with clinical populations (e.g., Au et al., 2014; Sala & Gobet, 2017; Soveri et al., 2017). Furthermore, we excluded meta-analyses that assessed the validity of various interventions (e.g., physical exercise, biofeedback, nutrition) and where cognitive training represented only a small subset of the studies examined (Kassai et al., 2019; e.g., Lambez et al., 2020; SonugaBarke et al., 2013; Takacs & Kassai, 2019).
39.2
Direct Effects: Evaluating Near- and Far-Transfer
The aggregation of findings across the 19 meta-analyses demonstrates agreement in observed benefits of cognitive training in proximal domains;
https://doi.org/10.1017/9781108955638.046 Published online by Cambridge University Press
https://doi.org/10.1017/9781108955638.046 Published online by Cambridge University Press
Table 39.1 List of meta-analyses included and excluded from the current review Included in Current Review Authors (year)
Training target
Population
Age Range
Number of Studies
Aksayli et al. (2019) Cao et al. (2020) Chacko et al. (2013) Cortese et al. (2015) Danielsson et al. (2015) Farcas and Szamoskozi (2016) Gathercole et al. (2019) Kassai et al. (2019) Melby-Lervag and Hulme (2013) Melby-Lervag et al. (2016) Peijnenborgh et al. (2016) Peng and Miller (2016)
WM (CogMed) EF (WM, cognitive flexibility, IC) WM (CogMed) Cognitive functions (broadly defined) WM WM WM EF (WM, cognitive flexibility, IC) WM WM WM EF, Attention
Not specified 3–12 4–17 3–18