Changing Minds Changing Tools Changing Minds Changing Tools From Learning Theory to Language Acquisition to Language Change Vsevolod Kapatsinski The MIT Press Cambridge, Massachusetts London, England © 2018 Massachusetts Institute of Technology All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retrieval) without permission in writing from the publisher. This book was set in ITC Stone Sans Std by ITC Stone Serif Std. Printed and bound in the United States of America. Library of Congress Cataloging-in-Publication Data Names: Kapatsinski, Vsevolod, author. Title: Changing minds changing tools : from learning theory to language acquisition to language change / Vsevolod M. Kapatsinski. Description: Cambridge, MA : The MIT Press, [2018] | Includes bibliographical references and index. Identifiers: LCCN 2017047700 | ISBN 9780262037860 (hardcover : alk. paper) Subjects: LCSH: Language acquisition--Psychological aspects. | Linguistic change-- Psychological aspects. | Language and languages--Psychological aspects. Classification: LCC P142.K26 2018 | DDC 401/.9--dc23 LC record available at https://lccn.loc.gov/2017047700 10 9 8 7 6 5 4 3 2 1 Contents Acknowledgments vii Introduction 1 1 The Web in the Spider: Associative Learning Theory 11 2 From Associative Learning to Language Structure 37 3 What Are the Nodes? Unitization and Configural Learning vs. Selective Attention 69 4 Bayes, Rationality, and Rashionality 99 5 Continuous Dimensions and Distributional Learning 127 6 Schematic Structure, Hebbian Learning, and Semantic Change 155 7 Learning Paradigmatic Structure 179 8 The Interplay of Syntagmatic, Schematic, and Paradigmatic Structure 201 9 Automatization and Sound Change 235 10 Bringing It All Together 269 Notes 305 References 317 Index 375 Acknowledgments This book would not have happened without the contributions of many people. My biggest debt of gratitude is to my mentors, particularly Joan Bybee, who shaped my thinking on these issues when I was an MA student at the University of New Mexico. Joan has led the revolutionary “quan- titative turn” in theoretical linguistics, turning it from a field in which “those who count don’t count” into a field where probabilistic patterns are a primary topic of investigation and quantitative factors such as fre- quency of use (in particular contexts) are seen as driving language change and shaping language structure. I am also very grateful to David Pisoni, who provided me with excellent training in experimental methods and a very supportive environment at the Speech Research Laboratory at Indiana University during my PhD years. Being at Indiana in the second half of the aughts exposed me to a great deal of research that forms the foundation of this book. Of others at Indiana, I am particularly grateful to my coadvisor, Ken de Jong, who showed me how interesting phonetics could be if one approaches it from the perspective of learning, and to John Kruschke for giving me an early education in Bayesian methods. I am also in debt to Harald Baayen for teaching me much about the operation of the Rescorla- Wagner model and for generously sharing code implementing the model. My two visits to Tübingen and subsequent conversations with Harald have greatly increased my understanding of what error-driven models can and cannot do. This book would also not be possible without my graduate students, Zara Harmon, Paul Olejarczuk, Amy Smolek, and Hideko Teruya. Without their insights and hard work, I would have nothing to write. Zara was the one who came up with the “associative triangle” framework and the insight that frequency might need to be logarithmically scaled to explain the puzzling results on skewed distribution learning (chapter 5). She is primarily respon- sible for the research on probability matching summarized in chapter 2, the viii Acknowledgments study on effecting attention shifts via distributional learning in chapter 5, and the work on entrenchment vs. extension in word learning (chapter 6). She has also taken my research program on the role of automatization in disfluencies and extended it in ways I did not think possible, only some of which are reported in chapter 9. Paul is primarily responsible for the work on distributional learning from skewed categories in chapter 5 and for the biased belief updating data and model in chapter 4. His skill in experimen- tal phonetics and speech synthesis has made it possible for us to branch out into the “low-level” mappings between acoustic cues and phonetic catego- ries. Amy is responsible for documenting the inductive and channel biases that constrain the acquisition of paradigmatic mappings, bringing exper- imental rigor to this line of research (chapter 7). Finally, Hideko’s inge- nious demonstration that we would all say schemar if we did not know the word well enough, though it did not make it into this book, has instilled in me the belief that sublexical and lexical schemas compete in production (chapter 7). This book would be very short indeed without the team. Finally, this book would also not be possible without my mother, Varvara Kapatsinskaya, and my grandfather, Mikhail Fuks, who are responsible for both biological and environmental contributions to everything I have ever learned.