Representation, Memory, and Development: Essays in Honor of Jean

Representation, Memory, and Development Essays in Honor of Jean Mandler

This Page Intentionally Left Blank

Representation, Memory, and Development Essays in Honor of Jean Mandler

Edited by

Nancy L. Stein

University

of Chicago

Patricia J. Bauer

University

of Minnesota

Mitchell Rabinowitz Fordham

LAWRENCE ERLBAUM Mahwah, New Jersey

University

ASSOCIATES, PUBLISHERS London

Copyright 0 2002 by Lawrence Erlbaum Associates, Inc. All rights reserved. No part of this book may be reproduced in any form, by photostat, microform, retrieval system, or any other means, without prior written permission of the publisher. Lawrence Erlbaum Associates, Inc., Publishers 10 Industrial Avenue Mahwah, NJ 07430 Cover design by Kathryn Library

of Congress

Houghtaling

Cataloging-in-Publication

Lacey Data

Representation, memory, and development : essays in honor of Jean Mandler / edited by Nancy L. Stein, Patricia J. Bauer, Mitchell Rabinowitz. p. cm. Includes bibliographical references and index. ISBN o-8058-41 96-2 (alk. paper) 1. Cognition in children. 2. Mental representation in children. 1. Mandler, Jean 3. Memory in children. 4. Child development. Matter. II. Stein, Nancy L. III. Bauer, Patricia J. IV. Rabinowitz, Mitchell. BF723.C5 R46 2002 155.4’13

-dc21

2001054337

CIP

Books published by Lawrence Erlbaum Associates are printed on acid-free paper, and their bindings are chosen for strength and durability. Printed in the United States of America 10 9 8 7 6 5 4 3 2 1

Contents -#-

Dedication

vii

to Jean

Preface

xi

George Mandler

I

Being There Conceptually: Simulating in Preparation for Situated Action

Categories

1

Larry Barsalou

2

Building Toward a Past: Construction Reliable Long-Term Recall Memory Patricia J. Bauer

3

of a System

17

The Origin of Concepts: Continuing the Gnversation

43

4

Scripts, Schemas, and Memory

of ‘Ikauma

53

5

On Animates

Things

75

Susan Carey

Robyn Fivush

Rachel Gelman

and Other Worldly

V

vi

l

Contents

6

How to Build a Baby That Develops Atypically

7

Pretense and Representation

8

Early Concepts and Early Language Acquisition: What Does Similarity Have to do With Either?

115

A Stitch in Time: The Fabric and Context of Events

145

10

The Reemergence of Function

161

11

The Procedural-Procedural Distinction

185

Annette KarmilofiSmith

Revisited

Alan M. Leslie

89 103

Laraine McDonough

9

Tamar Murachver

Katherine Nelson and Angelica Ware

Mitchell

12

Knowledge

Rabinowitz

Spatial Language: Perceptual Constraints and Linguistic Variation

199

Conceptual Development of Containment

in Infancy: The Case

223

Memories for Emotional, and Traumatic Events

Stressful,

Terry Regier and Laura Carlson

13

Elizabeth S. Speke and Susan J. Hespos

14

247

Nancy L. Stein

Author Index

267

Subject Index

277

Dedication -ii?-

To Jean

I

n the process of putting together this volume for Jean, we could not help but be struck by the breadth of topics and ideas that the contributors considered and discussed. The variety is indicative of Jean’s wide-ranging, yet deep interests in the field of psychology, in particular, and her interest in ideas, in general. The fact that Jean entered the field of psychology carrying out animal discrimination and learning studies, that with George Mandler, she wrote a book on Thinking (from Association to Gestalt), and that she became involved in the Children’s Gifted Association in San Diego, as her children proceeded through the San Diego-La Jolla School system, made her an ideal collaborator, mentor, colleague, and friend to all the contributors in this volume. Although we all have mentors in the field, and clearly we have colleagues and friends, rarely do we find a person that can play all of these roles equally well. She does each of these things with little effort. Part of her ability lies in a keen intelligence, coupled with a wit that has been known to save many of us from ourselves, an empathic stance (even when she was in the process of giving up smoking), and an ever present sense of morality-even when faced with circumstances that might have driven most of us insane, or at least crazy enough to commit outra--

vii

... VIII

l

Dedication

geous acts. She has the ability to take a risk, to make a stance, to create an idea, especially when data beg for an alternative interpretation. She is able to move on to different areas and to navigate unknown territories, with a sense of excitement that might be daunting to many of us. Each of the chapters in this volume was written by an individual that Jean has touched deeply. In editing the contributions, in organizing a symposium at the International Conference on Infant Studies (July 2000, Brighton, England), and in hosting a dinner in Jean’s honor after the symposium, a sense of enthusiasm and affection was always present. All of us had the desire to honor Jean for her contributions and the many times she has reached out, helped us, or engaged in an act of friendship. To Jean, we wish only the very best to come. She embodies what most of us hope we become- a good scientist, a friend with deep understanding, a person with boundless curiosity, and most of all, a person with a discriminating eye that recognizes the very best in just about everything. --Nancy Stein -Patricia Bauer -Mitch Rabinowi tz September 2001

Preface

Jean

J

ean Matter was born on the 6th of November, 1929, in Oak Park, IL, to a traditional upper middle-class family-an intellectual, but authoritarian, father who was a prominent Chicago bond lawyer, a caring housewife mother, and an older brother. The family traveled widely, and Jean proudly related having visited all 50 states except Alaska. An extended family provided many tales-her grandmother’s memory of her great uncle James Fenimore Cooper and crossing the prairie in a covered wagon, and encountering Native Americans as homesteaders in Nebraska. In her immediate family, Jean chafed from early on as the undervalued female “minority” member of the family. After a successful career at the renowned Oak Park & River Forest High School, she followed family advice and entered Carleton College. Jean enjoyed the Carleton classes and particularly flourished in the Carleton drama department, but she was unhappy about the homogeneity of the student body and the resulting stultifying and conservative atmosphere. After 2 years-the highlight of which was acting in the world premiere of Brecht’s The Caucasian ChaZk Circle-she transferred to Swarthmore College in 1949, and a new world opened up to her. The Swarthmore honors program instigated Jean’s intellectual coming of age as she discovered new frontiers and congenial friends, among them two young intellectually stimulating faculty members-Henry Gleitman and Sidney Morgenbesser. Jean had majored in philosophy at Carleton and continued that at Swarthmore but also wanted to minor in psychology, for which she had to pass a test covering the missed freshman and -

ix

X

l

Preface

sophomore years. Given the appropriate text she was off to Europe for the grand tour, psychology book tucked under her arm, and passed her test into a psychology minor in the fall. The seminars at Swarthmore provided her with a vision of the excitement of scholarly pursuits. She remembers reacting to Wolfgang Kohler’s and Hans Wallach’s seminars with wonder that some “philosophical” questions about perception might actually have empirical answers. That idea made her decide to go on in psychology. In 195 1, Jean received the BA sumltta CUM Zaude from Swarthmore, and she entered graduate school in the department of Social Relations at Harvard, initially in clinical psychology working with George Klein. In part, she picked psychology rather than philosophy and clinical rather than experimental psychology because of the scarcity of positions for women in these other fields. She soon realized that her real interests were in research, and she started doing animal research, on the effects of high drive, with Jerry Bruner, who also became her thesis advisor. Jean greatly enjoyed her social life in Cambridge with good friends such as Carolyn Cohen in biophysics, Alice (Timmie) Kohler in philosophy, and Lotte Lazarsfeld Bailyn (a Swarthmore friend) in psychology. Her work with Bruner produced several publications, and she continued her rat work for her PhD thesis. She decided to study the effects of early experience on behavior, and soon took her thesis prospectus oral, which was the major hurdle on her way to the PhD. As the orals proceeded, one young faculty member kept raising questions and making suggestions. At the conclusion of the examination, Jean wondered to Jerry who the committee member was who gave her so much trouble. Jerry told her not to worry, that the young faculty member was fresh out of Yale and still wet behind the ears. Jean’s thesis, eventually published in Science, showed that irregular feeding schedules in infant rats produced excessive eating and hoarding behavior in adulthood. Although some of the Yalie’s misgivings and suggestions had proved helpful, she did not have much contact with him until some time later when he offered her a research assistantship, working on autonomic nervous system reaction to stress and related topics. Their friendship developed during repeated trips to Quincy, MA, where their new polygraph was being built. Eventually, she married the troublemaker in 195 7. After receiving her PhD in 1956 and the arrival of son Peter in 1958, the family spent 1959 to 1960 in California. After that, it was time to look for new pastures, except that in the mode of the 195Os, little if any attention was paid to Jean’s professional needs-a sin shared by both Jean and George. They ended up at the University of Toronto where Jean was given a nonsalaried research position. She continued her animal research and was supported by research grants, including some

Preface

l

xi

partial salary. Soon after arriving in Canada, Michael joined his brother mother Peter. For the next several years, Jean showed that a working could also be a loving and beloved mother. However, she still did not have a regular faculty appointment and chafed at the secondary position in which her female social role had placed her. She did some research at Toronto and for a few years in California that was concerned with discrimination learning with special reference to overtraining. Some of Jean’s intellectual restlessness during this period motivated her to venture outside the narrow confines of rat running. Her early interest in philosophy and Gestalt psychology at Swarthmore motivated several undertakings in collaboration with George. The first was Thinking, a book on the historical development of research on thought, and the second an account of the fate of the German-speaking psychologists after the advent of Hitler. In 1965, Jean happily seconded the family’s move to La Jolla, CA, and the University of California at San Diego (UCSD). Because of the strict nepotism rules at the time, and with her spouse being chair of psychology, she was given a part-time appointment in the Department of Biology, as a research psychologist, and continued her rat work. At the same time, she enjoyed the California environment, and in particular her garden, which she treasures to this day. However, still she had no regular faculty ladder appointment. The atmosphere of the 196Os, together with her unhappiness at her second-class academic position, made her a committed feminist. She was determined to enter into a regular academic career. The family went on a sabbatical leave in Oxford in 19 71 and 1972 and, in the course of the year, an inquiry from Columbia and Barnard suggested the possibility of a move and an entry into a proper academic career. But then, UCSD finally decided that her appointment should be regularized. She was eventually appointed as an associate professor in psychology in 1973 and promoted to professor in 19 77. Despite the fact that Jean’s career was truncated at its beginning, she very quickly caught up with the academic progression and made important and path-breaking contributions. However, the field in which she was to shine most brightly was still in her future. With the start of her regular appointment there also came important changes in the content of her work. Jean was bored with rat research and was looking for a more demanding and interesting area of work. She was influenced-as was nearly everybody else-by the so-called “cognitive revolution” and wanted to work on human thought and cognition-an interest since Swarthmore. Nancy Stein had come to UCSD as a postdoctoral fellow but was dissatisfied with her situation and was lool&g for new challenges in directions similar to Jean’s. Nancy had some background in

xii

l

Preface

developmental psychology and the two of them set to work in joint mutual education and research-starting at the beginning, that is, in perception. They published several articles on children’s perception, emphasizing recognition of patterns and figures. Clearly, Jean had found her niche in psychology, she now had a regular appointment, she had started on a trajectory on children’s cognition, and she had benefitted from the change in women’s status in the academy. After Nancy left La Jolla to enter into her own distinguished career, Jean continued research on picture perception with Nancy Johnson and other creative graduate students, essentially working on some of the 1,000 words a picture is worth. From a study of perception and pictorial representation, the next obvious step was to look at the understanding of written material-a step up on the scale of human cognition. Jean’s next shift led to the work on people’s understanding of and memory for stories, inspired by the research on schemas by David Rumelhart. That work led to story analyses and recall and-to bring some structure into the understanding of story representation -the development of a widely used story grammar. Her work on story and event structures, with Mitch Rabinowitz, Robyn Fivush, Tamar Murachver, among her students and postdoctorates, carried on until the mid-1980s, resulting in various excursions into cross-cultural problems, aspects of temporal order, story memory, and developmental studies. In 1984, she published the book, Stories, scripts and scenes, on these efforts. Having explored the young mind, Jean naturally looked toward origins, and she became interested in the wellsprings of children’s thought-specifically, the mind of the infant. The arrival of a postdoctoral fellow, Pat Bauer, and a graduate student, soon to be postdoctoral associate, Laraine McDonough, created a highly productive and ingenious research program. This period, stretching to the present, saw numerous important publications on cognitive development, starting with the first “How to build a baby” article in 1988 and followed by the second one in 1992. Starting in the early 1990s when the family spent increasing time in London, Jean benefitted from and enjoyed her association with John Morton, Annette Karmiloff-Smith, and Alan Leslie at the Medical Research Council’s Cognitive Development Unit until its demise at the end of the decade. By the middle 199Os, Jean’s research had concentrated on the conceptual development of the human infant. She helped create the Department of Cognitive Science in 1986 and joined its interdisciplinary program as more appropriate for a study of the infant mind. She joined other psychologists, anthropologists, neuropsychologists, and linguists in the move to the new department.

Preface

l

xiii

Drawing on the various cognitive sciences, in particular cognitive linguistics, Jean developed a major theoretical picture of the development of the infant mind. Her particular emphasis has stressed the distinction between perception and conceptual thought and the processes of categorization and inductive inference. In her retirement, she is writing a book that will summarize her contribution. Having started with work on infant animals and discrimination in simple organisms, Jean has provided us deep insight about the most complex discriminations of the human infant. Jean’s achievements were recognized by her peers, with her elections to the Society of Experimental Psychologists and the American Academy of Arts and Sciences. For the past decade, she and George have spent half a year in London, and delight in having both their sons and their grandchildren living in the same town. -George

Mandler

This Page Intentionally Left Blank

Being There Conceptually : Simulating Categories in Preparation for Situated Action Lawrence W. Barsalou Emory University

A

constant theme in Jean Mandler’s work is that a child’s developing knowledge is grounded in sensory-motor experiences of events (e.g., Mandler, 1987, 1992). Rather than being detached from events, knowledge remains grounded in them. Rather than being amodal, knowledge retains its sensory-motor origins. The essay to follow arises in the tradition of this work and reflects its influence. THE SITUATED VIEW OF CONCEPTS

According to the view developed here, people conceptualize a category differently across situations, with each conceptualization embedded in a background situation. A single situation-independent concept does not represent the category; the concept does not represent the category in isolation, independently of the situations in which it occurs. Consider the category of chairs. According to the situated view, different conceptualizations of chairs are represented in their respective situations. Thus, one situated conceptualization might represent

i

2

.

1.

Situated

Concepts

office chairs in business environments, another might represent easy chairs in homes, another might represent theater chairs in theaters, another might represent airline chairs in jets, and so forth. A single situation-independent concept does not represent chairs across situations, and the conceptualizations do not represent isolated chairs. As the category is encountered in different situations, a situated conceptualization develops for each, linked together in a radial concept, as described later. EVIDENCE

FOR THE IMPORTANCE

OF SITUATIONS

Findings across diverse areas demonstrate the importance of situations in intelligence and behavior. In developmental psychology, the Vygotskian approach has stressed the importance of situations in acquiring cognitive and social skills (e.g., Vygotsky, 1991). From this perspective, Jean Mandler illustrated the importance of situations in children’s ability to remember stories and events (e.g., Mandler & Johnson, 19 7 7; Mandler, 198 7). In the social and personality literatures, situations predict behavior at least as well as traits (e.g., Mischel, 1968). In perception, situations greatly facilitate object recognition when an object occurs in a predicted context (e.g., Biederman, 1981), with Jean Mandler and her collaborators providing some of the earliest demonstrations (e.g., Mandler & Parker, 1976; Mandler & Ritchey, 1977; Mandler & Stein, 1974). In memory, situations play a central role in elaborating perceived scenes (e.g., Intraub, Gottesman, & Bills, 1998) and in retrieving information from memory (e.g., Tulving & Thomson, 19 73). In language comprehension, texts can be incomprehensible when the relevant situation is not known (e.g., Bransford & Johnson, 1973). Indeed, language comprehension can be viewed as preparation for situated action (Barsalou, 1999a). In pragmatics, situations are central to establishing common ground between communicators, both human (e.g., Clark, 1992) and nonhuman (e.g., Smith, 1977). In problem solving and reasoning, it may be difficult to draw valid and useful conclusions without the support of a concrete situation (e.g., Cheng & Holyoak, 1985; Gick & Holyoak, 1980; Johnson-Laird, 1983). In linguistics, the importance of situations has motivated the theory of construction grammar, where syntactic structures evolve out of familiar situations (e.g., Goldberg, 1995). In philosophy, the importance of situations has motivated the theory of situation semantics, where logical inference is optimized when performed in the context of specific situations (e.g., Barwise & Perry, 1983). At a more general level, arguments about the central role of situations in cognition can be found in Barsalou, Yeh, Luka, Olseth, Mix, and Wu (1993), Clark (1997), Glenberg (1997), and Green0 (1998).

Barsalou

l

3

Across these diverse areas, the common theme is that situations are fundamental to cognition. By incorporating situations into a cognitive task, processing becomes more tractable than when situations are ignored. Because specific entities and events tend to occur in some situations more than others, capitalizing on these correlations constrains and facilitates processing. Knowing the current situation constrains the entities and events likely to occur. Conversely, knowing the current entities and events constrains the situation likely to be unfolding. By focusing on situations, the cognitive system simplifies many tasks. Rather than having to search everything in memory across all situations, the cognitive system focuses on the knowledge and skills relevant in the current situation. As a result, it becomes easier to recognize objects and events that may be present; it becomes easier to remember relevant information and skills; it becomes easier to resolve the ambiguities of language; it becomes easier to solve problems and perform reasoning; it becomes easier to predict the actions of other agents. For all these reasons, it would not be surprising if situations turned out to be central for concepts. CURRENT THEORIES

OF CONCEPTS

Most current views implicitly view concepts as unsituated, assuming that concepts have been abstracted from the situations in which they occur. Although these theories could be readily extended to represent situated concepts, they typically do not. Consider classical theories, which typically assume that rules describe the objects in a category independently of situations (e.g., Bruner, Goodnow, & Austin, 1956). For example, a rule might attempt to capture the physical properties of chairs that are necessary and sufficient for membership. Although such rules could also attempt to capture situational properties, they typically do not. Instead, classical theories abstract across situations, rather than establishing rules for subsets of chairs within particular situations. Classical theories to date represent the extreme view of unsituated concepts. Prototype theories similarly tend to assume that unsituated abstractions represent categories (e.g., Hampton, 1979; Rosch & Mervis, 1975). Rather than being definitional, however, these abstractions are statistical, representing the most frequent properties across situations, with situation-specific properties canceling themselves out. Although subprototypes could develop for concepts in particular situations, this possibility has not been explored. The view that categories are embedded in intuitive theories similarly tends to ignore situations (e.g., Murphy & Medin, 1985). Although in-

4

l

1.

Situated Concepts

tuitive theories constrain the form that concepts take, situations have not played a central role in these accounts. Even when concepts are constrained by intuitive theories, they are nevertheless assumed to remain constant across situations (but see Gelman & Diesendruck, 1999, for an account of intuitive theories that is compatible with situated concepts). Exemplar models have much potential for implementing situated concepts but typically have not. Many exemplar models assume that the entire set of exemplars stored in memory for a category represents it on each occasion (e.g., Lamberts, 1994; Nosofsky, 1984). Thus, a fixed representation stands for the category on all occasions, not a situation-specific one. Furthermore, exemplar representations typically only include physical properties of exemplars, not properties of associated situations. Notably, some exemplar models have more of a situated character. For example, Nosofsky and Palermi’s ( 199 7) random-walk model assumes that only a subset of exemplars is retrieved in the current context, such that the category representation changes across contexts (also see Barsalou, Huttenlocher, & Lamberts, 1998). Similarly, Medin and Schaffer’s (1978) context model assumes that context is important in categorization, although context is typically implemented as other properties in the objects being categorized, not as their situational properties. As these theories illustrate, exemplar models can implement situated conceptualization if two conditions are met: (a) situational information is represented in exemplars along with object properties, (b) situation-specific subsets of exemplars are retrieved during categorization. In contrast to the previous four classes of models, connectionist models clearly implement situated conceptualization. Not only do they represent a category differently across situations, they include situational information in these representations. Consider Rumelhart, Smolensky, McClelland, and Hinton’s (1986) account of the room schema. In an auto-associative net, subsets of object properties are linked to subsets of room properties, such that correlated sets of object and situational properties form attractors. When a subset of object properties is activated, related situational properties become active, thereby situating the object. Conversely, when situational properties become active, relevant properties of the object become active, resulting in a situation-specific representation of it. Connectionist models have been explicitly formulated to implement situated conceptualization, and they do so elegantly. CONCEPTS

AS GROUNDED

IN PERCEPTUAL SIMULATION

The importance of situations for concepts follows from the proposal that people represent concepts with perceptual simulations (Barsalou,

Barsalou

l

5

1999b). This next section briefly outlines the theoretical assumptions of perceptual symbol systems, and then reviews some of the empirical evidence for this approach. The following section then illustrates how viewing concepts as grounded in perceptual simulation predicts the importance of situations in concepts. The first assumption of this view is that selective attention focuses on components of experience. During perception of sensory events, people focus on shapes, colors, sounds, smells, etc.; during perception of proprioceptive events, people focus on movements, facial expressions, vocalizations, etc.; during perception of introspective states, people focus on emotions, cognitive operations, beliefs, etc. Once attention selects a perceived aspect of experience, associative areas in the brain capture the respective pattern of activation in the relevant perceptual, proprioceptive, or introspective area. Later, these associative areas partially reactivate these perceptual representations in the absence of perceptual input, thereby simulating the experience of what an external or internal event was like. Using such simulations, people conceptualize objects, external events, and internal events in their absence. Barsalou (1999b) illustrated how these simulation mechanisms implement a fully-functional conceptual system, including the type-token distinction, categorical inference, the productive construction of novel simulations, the representation of propositions, and the representation of abstract concepts. Also illustrated are how these simulation mechanisms could underlie the knowledge that supports basic cognitive processes, including perception, categorization, memory, language, and thought. Additional articles extend this theory (Barsalou, 1993, 1999a; Barsalou & Prinz, 199 7; Prinz & Barsalou, 2000; Prinz & Barsalou, in press). In one of the earliest articles to champion this theme, Jean Mandler highlighted the importance of sensory-motor knowledge in children’s developing concepts (Mandler, 1992). Glenberg (199 7) offered a related proposal.

Empirical Support for Perceptual Simulation in Conceptual Tasks Several lines of empirical inquiry implicate perceptual simulation in the representation and processing of concepts (for a review, see Barsalou, Solomon, & Wu, 1999; also see Goldstone & Barsalou, 1998). Solomon and Barsalou (2001a) demonstrated that when individuals attempt to verify a property of a concept, they search for the property in a perceptual simulation of the respective object. To verify that a house has a roof, for example, individuals simulate a house and a roof, and then search the simulated house for a region that matches the simulated

6

l

1.

Situated Concepts

roof. Under conditions that ensured conceptual processing, perceptual predictors, such as the size and perceptual salience of properties, predicted verification times and errors better than other factors, such as associative strength and word frequency. Furthermore, neutral participants not instructed to use any particular strategy performed the same as imagery participants who received explicit instructions to use imagery. The results of these experiments support the a priori prediction that neutral participants adopt perceptual simulation spontaneously to represent and process concepts. Solomon and Barsalou (2OOlb) similarly found that the specific perceptual forms of properties predict whether they prime each other during property verification. For example, verifying mane for pony benefits from previously verifying mane for horse but not from verifying mane for Zion. Control conditions ruled out the explanation that higher similarity between horses and ponies overall, relative to less similarity between lions and ponies, was responsible. Instead, the detailed perceptual similarity of the two properties was critical. Because perceptual details that are difficult to verbalize predicted conceptual priming, it appeared that participants adopted perceptual simulations spontaneously to represent concepts. Accounting for such results is difficult and post hoc without adopting perceptual simulation as the basis of performance. Wu and Barsalou (2001) explored similar issues in the property generation task. Analogous to Solomon and Barsalou (2001a), neutral participants produced essentially the same distributions of properties as imagery participants, suggesting that both groups scanned perceptual simulations to produce properties. For example, both groups produced essentially the same detailed distribution of instructed to produce word properties for Zawn, unlike participants associations, who produced a different distribution. Furthermore, the perceptual factor of occlusion affected performance for both neutral and imagery participants, again implicating perceptual simulation. When participants produced properties for lawn, they rarely produced occluded properties like roots and dirt. In contrast, when other participants produced properties for rolled-up lawn, they produced roots and dirt much more frequently, because these properties were no longer occluded in simulations. The empirical evidence just reviewed suggests that people run perceptual simulations to represent concepts. When participants list the properties of a concept, they simulate an instance, scan across it, and report the properties attended. When participants verify a property of a concept, they simulate an instance and the property, and then attempt to locate the property in the instance.

Barsalou

PERCEIVED SITUATIONS

AND SIMULATED

l

7

SITUATIONS

If perceptual simulation underlies the representation of concepts, it places an important constraint on them: If a conceptualization attempts to simulate a perceptual experience, then it should typically simulate a situation, because situations are intrinsic parts of perceptual experiences. To make this critical assumption more concrete, consider the nature of perceptual experience. At a given point in time, people perceive the immediate space around them, including any agents, objects, and events within it. Some of these entities and events may be external, whereas others may be internal. Furthermore, this experience is multimodal; it is not just visual, but also auditory, tactile, gustatory, olfactory, proprioceptive, and introspective. Thus, a perceptuaz situation is a perceived region of space that contains agents, objects, and events, both external and internal. Most importantly, even when people focus attention on a particular entity or event in perception, they continue to perceive the background perceptual situation-the situation does not disappear. If perceptual experience takes the form of a situation, and if a conceptualization is essentially an attempt to simulate perceptual experience, then the form of a conceptualization should take the form of a situation. When people construct a simulation to represent a category, they should tend to envision it in a relevant perceptual situation, not in isolation. When people conceptualize chair, for example, they should attempt to simulate not only a chair but a more complete perceptual situation, including the surrounding space and any relevant agents, objects, and events. In principle, it is possible to simulate a chair independently of a situation, and indeed, the ability to focus attention on aspects of situations is a central part of perceptual symbol systems (Barsalou, 1999b). When we actually perceive chairs in the world, though, we never perceive them in a vacuum. Although we focus attention on them, we nevertheless continue to perceive the background situation. This observation motivates that claim that conceptualizations are similarly situated, at least much of the time. Although the simulation of a chair may typically focus attention on the simulated object, the background situation nevertheless tends to be simulated along with it.

Following Yeh and Barsalou (2001), the following lie the view that concepts are situated: 1. A conceptualization situation.

of a category typically

assumptions

under-

includes a background

8

l

1.

Situated Concepts

2. Each conceptualization represents a category in a way that is relevant to the background situation, such that different conceptualizations represent the category differently. 3. The different conceptualizations of a category become linked by analogy or by an essence (real or imagined) to form a radial concept. The first assumption is the one just described, namely, conceptualizations do not simulate a category’s members in isolation but simulate them against background situations. For example, people do not simulate chairs in isolation, but tend to simulate them in their background situations (e.g., living rooms, classrooms, jets, theaters, etc.). The second assumption follows naturally from the first: If conceptualizations include background situations, then each simulated form of a category should include information appropriate for the respective situation. For example, the simulation of a chair in a living room should simulate a living room chair, whereas the simulation of a chair on a jet should simulate a jet chair. As a result, the different conceptualizations of a category differ, not just in situational information, but also in information about category members per se. Although some properties may be relatively common across conceptualizations, others are likely to vary (Barsalou, 1982). The third assumption is necessary to explain how different conceptualizations of the same category become linked together. For example, how do the various conceptualizations of chairs become integrated into a single category? One possibility is by analogy. When a perceived entity accesses a structurally analogous conceptualization in memory, the two become linked (e.g., Brooks, 1978; Gentner & Markman, 1997; Holyoak & Thagard, 1989; Nosofsky, 1984). Perceiving a dining room chair, for example, may activate the conceptualization of an office chair via their shared physical structure, or via the common actions performed on them. As a result, the two conceptualizations become linked in memory. As chairs are increasingly encountered in other situations, the respective conceptualizations become related to similar conceptualizations, thereby forming linked chains. Although core properties could ultimately become established across the various conceptualizations of a concept, they need not be. When they do not, the linked chains of conceptualizations form a radial concept, whereby each conceptualization is closely related to at least one other (Lakoff, 1987; Malt, Sloman, Gennari, Shi, & Wang, 1999). Essences constitute another possible linking mechanism. If all known conceptualizations of a category are believed to share a common essence, they become linked around the essence, even when their

Barsalou

l

9

physical appearances differ (e.g., Gelman & Diesendruck, 1999). Depending on the category, the essence could reflect a real essence that actually exists across instances, or it could simply reflect the belief that an essence exists, even when one does not. Regardless, the point is that the situated conceptualizations for a category could become linked in several ways. Following Barsalou (1999b), the result is a simulator capable of producing many situated simulations of a category. This proposal does not simply boil down to the fact that a category has subordinates; the claim is significantly stronger. A category does not simply take different subordinate forms. Instead, these forms arise to accommodate the constraints of different situations. Conceptualizations of chairs, for example, take different forms because the constraints on having a place to sit vary from situation to situation. Furthermore, the heart of this proposal is that conceptualizations are represented against background situations-they are not simply subordinates represented in isolation. Finally, this framework extends well beyond subordinate categories. Consider cars. This framework predicts that a single subordinate, say sedans, will be conceptualized in a variety of situations, such as driving a sedan, seeing a sedan drive by, repairing a sedan, filling a sedan’s gas tank, and so forth. Rather than conceptualizing sedans in a generic situation-independent manner, people conceptualize them in these various situations, focusing on different perspectives and properties in each. Thus, the theoretical proposal here extends beyond the fact that categories have subordinates. Barsalou et al. (1993) presented the functional specifications of the aforementioned theory, which remains to be implemented computationally. As described earlier, existing connectionist theories offer one natural approach. Implementing this theory as a perceptual symbol system, however, constitutes a significant challenge that lies considerably beyond existing connectionist models (Barsalou, 199913). EMPIRICAL

SUPPORT

FOR SITUATED CONCEPTS

Yeh and Barsalou (2001) reviewed a wide variety of evidence that concepts are situated. Not only do people represent concepts in background situations, they represent them from subjective perspectives. In representing a concept, it is as if people were being there with one of its instances. Rather than representing a concept in a detached isolated manner, people construct a multimodal simulation of themselves interacting with an instance of the concept. To represent the concept, they prepare for situated action with one of its instances (Barsalou, 1999a). This final section briefly illustrates this point with two empirical findings.

10

l

1.

Generating

Situated

Concepts

Category

Members

From Situations

Vallee-Tourangeau, Anthony, and Austin (1998) illustrated how participants imagine themselves in situations to produce exemplars of concepts. These researchers asked participants to generate exemplars from common taxonomic categories, such asfurniture andfruit, and from ad going on a holihoc categories, such as things dogs chase and reasomfor day. After participants finished generating exemplars, they were asked to describe the strategies that they had used. Each strategy was classified as one of the following: 1. Experiential mediation -retrieving an autobiographical memory of a situation that contains individuals from the target category, and then reporting the categories to which these individuals belonged. When generating types offruit, for example, this might involve retrieving a memory of a grocery store, scanning across it, and reporting the types offruit perceived in the produce section. Similarly, when generating types offurniture, this might involve retrieving a memory of a residence, scanning across it, and reporting the types offurniture perceived in the living room. 2. Semantic mediation-retrieving a detached taxonomy that contains the target category, and then reporting its subcategories. When generating exemplars of fruit, for example, this might involve retrieving thefruit taxonomy and reporting subtypes, such as tropical fruit, driedfruit, and citrusfruit. Similarly, when generating exemplars offurniture, this might involve retrieving thefurniture taxonomy and reporting subtypes, such as decorative furniture, storage and seatingfurniture. furniture, 3. Unmediated retrieval-accessing exemplars unconsciously and not being aware of any obvious strategy. On such occasions, participants often made remarks such as, “I just thought of them.” Vallee-Tourangeau et al. (1998) reported that their participants used experiential mediation about 3 times as often as semantic mediation for both common taxonomic and ad hoc categories (unmediated retrieval was used even more rarely). vpically, experiential mediation included situations, namely, memories of events in environmental contexts. One might well expect that participants would report situations for ad hoc categories, given that these categories arise out of goal-directed activity in specific contexts (Barsalou, 1983, 199 1). Much more surprising is the finding that situations were reported just as often for common taxonomic categories, suggesting that they, too, are organized around situations.

Barsalou

l

11

Bucks ( 1998) reported the same pattern of results as Vallee-Tourangeau et al. (1998), again showing that participants used experiential mediation most often to generate the instances of both common taxonomic and ad hoc categories. Related results have been reported by Walker and Kintsch (1985). Together, all of these studies show that participants represent concepts in background situations. When participants receive a concept, they do not process its meaning in isolation. Instead, they often activate a background situation, and then establish the concept’s meaning within this context. Generating

Features

From Situated

Instances

of Categories

As the next studies illustrate, participants also situate concepts with respect to subjective perspectives when asked to produce the features of a single concept. Wu and Barsalou (2001) asked participants to list properties for individual concepts, such as appZe, and for conceptual combinations, such as sliced apple. The instructions explicitly stated that participants should produce properties of the target objects per se. Nevertheless, participants produced many other properties that described background situations and subjective perspectives on these situations. The importance of situations can be seen in the types of properties that participants produced: 1. Taxonomic concepts- neighboring concepts in a taxonomy that contains the target concept. For example, generating the concepts fruit, banana, and Granny Smith for the target concept apple. 2. Entity properties-properties that describe the target object’s surface properties and components. For example, generating smooth, red, stem, and seeds for apple. 3. Situational properties-properties that describe a physical setting or event in which the target object occurs. For example, generating grocery store, fruit basket, slicing, and picnic for apple. 4. Introspective properties-properties that describe an agent’s subjective perspective on the target object. For example, generating delicious and “I like them” for apple. It is not surprising that participants generated entity and taxonomic properties in the feature listing task. After all, this is what they were instructed to do. What is surprising is how often they described situational and introspective properties. Participants frequently described the physical settings and events in which the target objects are typically found (i.e., situational properties). Furthermore, participants often de-

12

l

1.

Situated Concepts

scribed these situations from their subjective perspectives (i.e., introspective properties). Across four experiments, the proportion of situational and introspective properties combined ranged from 26% to 50%. In a given study, roughly two thirds of these properties were situational, and about one third was introspective. These findings illustrate that participants did not simply represent the target objects as detached and isolated. Instead, participants typically imagined being there with the objects, situating the objects in the environment, and viewing them from their subjective perspectives. Once participants had constructed these situated simulations, they scanned across them, producing a variety of properties in the process. Although participants were asked to process isolated concepts, they nevertheless represented them in background situations from subjective perspectives. These findings are consistent with the importance of thematic relations in concepts. Not only do people represent a concept’s structural properties relevant to a taxonomy, they also represent its thematic relations relevant to related situations. It has long been believed that thematic relations are primarily important for children and not for adults (cf. Inhelder & Piaget, 1964; Luciarello, Kyratzis, & Nelson, 1992; Markman, 1981, 1989; Nelson, 1977). However, recent work illustrates the central importance of thematic relations in adult concepts as well (Lin & Murphy, 2001). This importance may further reflect people’s spontaneous inclination to represent concepts in situations. SUMMARY These findings illustrate the importance of situations in the representation of concepts. They also point toward future research that could illuminate the roles of situations in concepts, and the roles of concepts in situations. The human conceptual system probably did not evolve to represent concepts in isolation, or in detached taxonomies. Instead, the human conceptual system probably evolved to support human action in the environment. REFERENCES Barsalou, tion in Barsalou, Barsalou,

L. W. (1982). Context-independent and context-dependent informaconcepts. Memory & Cognition, IO, 82-93. L. W. (1983). Ad hoc categories. Memory & Cognition, 11, 21 l-227. L. W. (1991). Deriving categories to achieve goals. In G. H. Bower

(Ed.), Thepsychologyof learning and motivation: Advances in research and the-

ory (Vol. 27, pp. l-64). San Diego, CA: Academic. Barsalou, L. W. (1993). Structure, flexibility, and linguistic vagary in concepts: Manifestations

of a compositional

system of perceptual symbols. In A. C.

Barsalou

l

13

Collins, S. E. Gathercole, & M. A. Conway (Eds.), Theories of memory (pp. 29-101). London: Lawrence Erlbaum Associates, Inc. Barsalou, L. W. (1999a). Language comprehension: Archival memory or preparation for situated action? Discourse Processes,28, 61-80. Barsalou, L. W. (199913). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577-660.

Barsalou, L. W., Huttenlocher, J., & Lamberts, K. (1998). Basing categorization on individuals and events. Cognitive Psychology, 36, 203-272. Barsalou, L. W., & Prinz, J. J. (1997). Mundane creativity in perceptual symbol systems. In T. B. Ward, S. M. Smith, & J. Vaid (Eds.), Creative thought: An investigationofconceptualstructuresandprocesses (pp. 267-307). Washington, DC: American Psychological Association. Barsalou, L. W., Solomon, K. O., & Wu, L. L. (1999). Perceptual simulation in conceptual tasks. In M. K. Hiraga, C. Sinha, & S. Wilcox (Eds.), Proceedings of the4th ConferenceoftheInternationaZ CognitiveLinguisticsAssociation: Vol. 3. CuZturaZ, typological, and psychological perspectives in cognitive linguistics (pp. 209-228). Amsterdam: Benjamins. Barsalou, L. W., Yeh, W., Luka, B. J., Olseth, K. L., Mix, K. S., & Wu, L. (1993). Concepts and meaning. In K. Beals, G. Cooke, D. Kathman, K. E. McCullough, S. Kita, & D. Testen (Eds.), Chicago Linguistics Society 29: Papers from the parasession on conceptual representations (pp. 23-61). University of Chicago: Chicago Linguistics Society. Barwise, J., & Perry, J. (1983). Situations and attitudes. Cambridge, MA: MIT Press. Biederman, I. (1981). On the semantics of a glance at a scene. In M. Kubovy & J. R. Pomerantz (Eds.), Perceptual organization (pp. 213-253). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Bransford, J. D., & Johnson, M. K. (1973). Considerations of some problems of comprehension. In W. G. Chase (Ed.), I%uaZ information processing (pp. 383-438). New York: Academic. Brooks, L. R. (1978). Nonanalytic concept formation and memory for instances. In E. Rosch & B. B. Lloyd (Eds.), Cognition and categorization (pp. 169-211). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Bruner, J. S., Goodnow, J. J., & Austin, G. A. (1956). A study of thinking. New York: Wiley. Bucks, R. S. (1998). Intrusion errors in Alzheimer’s disease. Unpublished doctoral dissertation, University of Bristol, England. Cheng, I? W., & Holyoak, K .J. (1985). Pragmatic reasoning schemas. Cognitive Psychology, 17, 39 l-41 6. Clark, A. (1997). Being there: Putting brain, body, and world together again. Cambridge, MA: MIT Press. Clark, H. H. (1992). Arenas of language use. Chicago: University of Chicago Press. Gelman, S. A., & Diesendruck, G. (1999). What’s in a concept? Context, variability, and psychological essentialism. In I. E. Siegel (Ed.), Theoretical perspectives in the concept of representation (pp. 87-I 11). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Gentner, D., & Markman, A. B. (1997). Structure mapping in analogy and similarity. American PsychoZogist, 52, 45-5 6. Gick, M. L., & Holyoak, K. J. (1980). Analogical problem solving. Cognitive Psychology, 12, 306-355.

14

l

1. Situated Concepts

Glenberg, A. M. (1997). What memory is for. Behavioral and Brain Sciences,20, l-55. Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Goldstone, R., 8~Barsalou, L. W. (1998). Reuniting perception and conception. Cognition, 65, 231-262. Greeno, J. G. (1998). The situativity of knowing, learning, and research. American Psychologist, 53, 5-26. Hampton, J. A. ( 19 79). Polymorphous concepts in semantic memory. JournaZ of Verbal Learning and Verbal Behavior, 28, 441-461. Holyoak, K. J., & Thagard, P R. (1989). Analogical mapping by constraint satisfaction. Cognitive Science, 13, 295-356. Inhelder, B., & Piaget, J. (1964). The early growth of Zogicin the child. London: Routledge & Kegan Paul. Intraub, H., Gottesman, C. V, & Bills, A. J. (1998). Effects of perceiving and imagining scenes on memory for pictures. JournaZ of Experimental Psychology: Learning, Memory, & Cognition, 24, 186-201. Johnson-Laird, P N. (1983). Mental models. Cambridge, MA: Harvard University Press. Lakoff, G. (1987). Women, fire, and dangerous things: What categories reveal about the mind. Chicago: University of Chicago Press. Lamberts, K. (1994). Flexible tuning of similarity in exemplar-based categorization. Journal of Experimenta PsychoZogy:Learning, Memory, and Cognition, 20,1003-1021. Lin, E. L., & Murphy, G. L. (2001). Thematic relations in adults’ concepts. JournaZ of Experimental Psychology, 130, 3-28. Luciarello, J., Kyratzis, A., & Nelson, K. (1992). Taxonomic knowledge: What kind and when? Child Development, 63, 978-998. Malt, B. C., Sloman, S. A., Gennari, S., Shi, M., & Wang, Y. (1999). Knowing versus naming: Similarity and the linguistic categorization of artifacts. Journal of Memory and Language, 40,230-262. Mandler, J. M. (1987). On the psychological reality of story structure. Discourse Processes,10, l-29. Mandler, J. M. (1992). How to build a baby: II. Conceptual primitives. Psychological Review, 99, 587-604. Mandler, J. M., & Johnson, N. S. (1977). Remembrance of things parsed: Story structure and recall. Cognitive Psychology, 9, 11 l-l 5 1. Mandler, J. M., & Parker, R. E. (1976). Memory for descriptive and spatial information in complex pictures. Journal of Experimenta Psychology: Human, Learning, and Memory, 2, 38-48. Mandler, J. M., 81 Ritchey, G. H. (1977). Long-term memory for pictures. Journal of Experimental Psychology: Human Learning and Memory, 3, 386-396. Mandler, J. M., & Stein, N. (1974). Recall and recognition of pictures by CMdren as a function of organization and distractor similarity. Journal of Experimental PsychoZogy,102, 65 7-669. Ma&man, E. M. (1981). Two different principles of conceptual OrganiZatiOn. In. M. E. Lamb & A. L. Brown (Eds.), Advances in developmental psychoZogy (pp. 199-236). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Ma&man, E. M. (1989). Categorization and naming in children: Problems of induction. Cambridge, MA: MIT Press.

Barsalou

l

15

Medin, D. I,., & Schaffer, M. (1978). A context theory of classification learning. Psychological Review, 85, 207-238. Mischel, W. (1968). Personality and assessment. New York: Wiley. Murphy, G. L., & Medin, D. L. (1985). The role of theories in conceptual coherence. Psychological Review, 92, 289-316. Nelson, K. (1977). The syntagmatic-paradigmatic shift revisited: A review of research and theory. Psychological Bulletin, 84, 93-l 16. Nosofsky, R. M. (1984). Choice, similarity, and the context theory of classification. JournaZ of Experimental Psychology: Learning, Memory, and Cognition, 10, 104-l 14. Nosofsky, R. M., & Palmeri, T. J. (1997). An exemplar-based random walk model of speeded classification. Psychological Review, 104, 266-300. Prinz, J. J., & Barsalou, L. W. (in press). Acquisition and productivity in perceptual symbol systems: An account of mundane creativity. In D. M. Peterson & T. Dartnall (Eds.), Creativity and computation. Westport, CT Greenwood. Prinz, J. J., & Barsalou, L. W. (2000). Steering a course for embodied representation. In E. Dietrich & A. Markman (Eds.), Cognitive dynamics: Conceptual change in humans and machines (pp. 51-77). Cambridge, MA: MIT Press. Rosch, E., & Me&s, C. B. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7, 5 73-605. Rumelhart, D. E., Smolensky, I?, McClelland, J. L., & Hinton, G. E. (1986). Schemata and sequential thought processes in PDP models. In J. L. McClelland, D. E. Rumelhart, & the PDP Research Group (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition: Vol. 2. Psychological and biological models (pp. 7-57). Cambridge, MA: MIT Press. Smith, W. J. (1977). The behavior of communicating: An ethological approach. Cambridge, MA: Harvard University Press. Solomon, K. O., & Barsalou, L. W. (2001 a). Grounding concepts in perceptual simulation: II. Evidence from property verijication. Manuscript submitted for publication. Solomon, K. O., & Barsalou, L. W. (2001b). Representing properties locally. Cognitive Psychology, 43, 129-l 69. lulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processesin episodic memory. Psychological Review, 80, 352-373. VallCe-Tourangeau, F., Anthony, S. H., & Austin, N. G. (1998). Strategies for generating multiple instances of common and ad hoc categories. Memory, 6, 555-592. Vygotsky, L. S. (1991). Genesis of the higher mental functions. In I? Light, S. Sheldon, & M. Woodhead (Eds.), Learning to think: ChiZd development in social context (Vol. 2, pp. 3241). London: Routledge. Walker, W. H., & Kintsch, W. (1985). Automatic and strategic aspects of knowledge retrieval. Cognitive Science, 9, 261-283. Wu, L., & Barsalou, L. W. (2001). Grounding concepts in perceptual simulation: I. Evidence from property generation. Manuscript submitted for publication. Yeh, W., & Barsalou, L. W. (2001). The situated nature ofconcepts. Manuscript submitted for publication.

This Page Intentionally Left Blank

Building Toward a Past: Construction of a Reliable Long-Term Recall Memory System Institute

0

Patricia

J. Bauer

of Child Development,

University of Minnesota

ne of the major themes in developmental science is continuity and change. Developmental scientists are interested in what aspects of function show stability across developmental time and what aspects show substantial and significant change over time. This theme is particularly salient for me as I write this chapter because my association with the individual whom we honor, namely, Jean M. Mandler, was for me personally, a source of profound developmental change in my years as a postdoctoral fellow with her, from 1985 to 1989. By the same token, Jean Mandler contributed to continuity in my developmental processes in that the broader issues and questions on which I started working during my years in residence with her have remained at the center of my research program. In many respects, I view this chapter as a report on the progress that I have made since finishing my postdoctoral fellowship and leaving for the University of Minnesota, approximately a decade ago.

18

l

2.

Long-Term Recall Memory

The work on which I report was begun in Jean Mandler ‘s laboratory in the Department of Psychology, University of California, San Diego (UCSD). Actually, it was begun in the laboratory of her colleague, Elizabeth Bates, because, when I first arrived at UCSD in 1985, Jean Mandler did not even have an infant cognition laboratory. Until that time, most of her work had been with adults and children of preschool age and older. The enormous contributions that Jean Mandler has made to the area of infant cognition loom even larger when her “relative newcomer u status is considered. The work that is the subject of this chapter is on memory development in the period of transition from infancy to early childhood. For researchers of memory development, this is a particularly interesting and challenging age range. It is especially intriguing because of speculation that in this time frame, there occur two highly significant changes in representational or mnemonic function. One change is thought to occur at the beginning of the transition (roughly 6 to 9 months) and the other at its end (roughly 3 to 4 years). The first important change is in the emergence and subsequent consolidation of the ability to maintain accessible memories of specific past events over extended periods of time. From very early in their lives, we see evidence that infants remember-they habituate, they can be conditioned, and they recognize all manner of stimuli in visual paired comparison tasks. What they appear not to do, however, is engage in recall and in particular, long-term recall. To quote Jean Mandler, recall involves “. . . accessing (bringing to awareness) a cognitive structure pertaining to a past experience not currently available to perception” (Mandler, 1984, p. 79). Recall is the process in which we engage when we attempt to conger up a memory for what we had for breakfast this morning, what we had for dinner the last time we visited a favorite restaurant, or where we went on vacation the summer after our senior year in high school. Recall, and in particular, long-term recall, is an ability that in adults, depends upon a multicomponent neural network (discussed later). That network is thought to coalesce in the second half of the first year of life. The change in neural activity heralds significant changes in behavior as the developing substrate becomes sufficiently mature to support long-term recall. The second important change, which takes place at about age 3 to 4, is associated with the development of a particular type of long-term recall, namely, that of autobiographical or personal memory. Most readers are familiar with the phenomenon of infantile amnesia: the relative paucity among adults of verbally-accessible memories for specific events that happened in the first years of life (Freud, 19 16/l 966). Although there is variability, for adults, the average age of earliest verbalizable memory is

Bauer

l

19

3% years (e.g., West & Bauer, 1999). Beginning at age 3% there is a steady increase in autobiographical or personal memories, which are episodic memories that include a sense of personal involvement or participation (Brewer, 1986; K. Nelson, 1993). The emergence of this type of memory is seen by many as a qualitative change. In summary, during the period of transition from infancy to early childhood, there are two points in time at which there is speculation about significant change in representational or mnemonic function. One change occurs around 6 to 9 months of age and is thought to allow for the capacity for long-term recall of previously experienced events. A challenge in investigating this early development is that such young infants are not able to participate in the paradigm of choice for memory researchers, namely, verbal report. Jean Mandler and I met this challenge by developing a nonverbal measure of recall memory. After describing the procedure, I present some data that illustrate continuities in recall processes by very young children compared with older children and even adults. I then provide an integrative summary of the results of separate investigations which, together, tell the story of the emergence and gradual consolidation of long-term recall memory function over the course of the 1st and throughout the 2nd year of life. The second significant change occurs at the end of the transition from infancy to early childhood, at around 3 to 4 years of age, and permits the construction of autobiographical or personal memories. At the close of the chapter I present some data that bear on this second important development.

INITIATING

THE STUDY OF LONG-TERM IN YOUNG CHILDREN

RECALL MEMORY

Any good progress report begins with reference to a starting point. When Jean and I first began our inquiry in 1985, as a field, we knew virtually nothing about long-term recall memory in children younger than 3. From the work of others, some of whom are represented in this volume (i.e., Katherine Nelson & Robyn Fivush), we knew that children 3 and older had impressive long-term memories (e.g., Fivush, 1984; Hudson, 1986; K. Nelson, 1986; Nelson & Gruendel, 1981). By that age, children evidenced well-organized representations of familiar events, such as going to McDonald’s. There also were diary reports and naturalistic observations of mnemonic competence in young preschoolers (e.g., Ratner, 1980). However, whether mnemonic competence extended to children younger than 3 was unknown. The major reason was that both conceptually and methodologically, the ability

20

l

2.

Long-Term Recall Memory

to recall was associated with the ability to provide a verbal report. Because children 3 and older were able to use language to talk about the past it was clear that they could remember it. However, l- to 2-year-olds are not particularly adept at language. Even by late in the 2nd and into the 3rd year, when children have made substantial gains in language, they seem to have considerable difficulty using their language to talk about the past. It is not until they are 3 and older that they become relatively reliable partners in conversations about past events (e.g., Fivush, Gray, & Fromhoff, 1987). To study recall memory in infants and young children, when I first arrived at UCSD, Jean and I developed a technique that did not require children to provide verbal reports of what they remembered. Instead of telling us what they remember, in our procedure, children showed us. The procedure is known as elicited or deferred imitation.’ It involves using objects to produce a unique action or sequence of actions, and then allowing the infant or child to imitate the actions. The phenomenon of imitation after exposure to a modeled event originally was reported by Jean Piaget and to him, was one of the hallmarks of the transition from the sensorimotor to the preoperational period (Piaget, 1952). Independently and unbeknownst to us, Andrew Meltzoff also was working on an adaptation of Piaget’s technique as a means of studying memory in infants (Meltzoff, 1985). In elicited-imitation tasks, props are used to demonstrate a specific action or action sequence. Either immediately after demonstration, after some delay, or both, the child is allowed to imitate the actions. Before the event is modeled, the props are given to the child for a baseline period, during which spontaneous production of the target actions and sequences is assessed (in some laboratories, a betweensubjects manipulation is used to obtain an uninstructed baseline; e.g., Meltzoff, 1985). After baseline, experimenters label the event to be produced, and then narrate their actions as they use the props to produce the sequence. In situations in which immediate imitation is assessed, after modeling, the experimenter returns the props to the child, who is invited to imitate. To test retention of information over time, the child returns to the laboratory after a delay. The props then are given to the child, with no instruction and no modeling. Procedurally, the delayed assessment period is identical to baseline. To the extent that production of target actions and sequences after exposure ‘Prior to my arrival at the University of California, San Diego, Cecilia Shore and I had used elicited imitation to examine children’s abilities to reproduce sequences of action in symbolic play (Bauer & Shore, 1987). It was through the collaboration with Jean Mandler that the technique developed into a tool for examination of memory development.

Bauet

l

21

to the model differs from baseline, either immediately or after the delay, we infer memory for the event. From each phase of testing we determine the number of individual steps or actions of the event the child produced, and the number of pairs of steps or actions produced in the target order, as a measure of ordered recall. There is excellent reason to believe that the technique of elicited imitation taps the same type of memory as verbal report (see Bauer, 1996, 1997, in press-b; Mandler, 1990; Meltzoff, 1990; for detailed discussions of elicited and deferred imitation as measures of recall memory). For present purposes, I merely note that the task passes the “amnesia test.” Thanks to the work of Jean Mandler and her colleagues, we know that adults suffering from amnesia, in whom declarative processes are disrupted, have difficulty performing an age-appropriate version of the task (McDonough, Mandler, McKee, & Squire, 1995). Moreover, the task requires recall, rather than recognition. Consider that, to reproduce an ordered sequence of actions, the child cannot rely on recognition-once modeling is complete, information about the order of the event is no longer perceptually available. To reproduce an ordered sequence then, the child must encode order information, and later retrieve it from a representation of the event, in the absence of ongoing perceptual support (see Bauer, 1996, 1997, in press-b, for further discussion). Because reproduction of temporal order provides the strongest evidence of recall, my primary focus is on the extent to which children adhere to the target order in their reproductions of events. In our first published report on young children’s memories for past events, Jean and I used elicited imitation to examine immediate recall and recall after a 2-week delay, in infants 16 and 20 months of age (Bauer & Mandler, 1989). In the years since that publication, the technique has been adopted quite broadly, as well as extended to test infants as young as 6 months of age (Barr, Dowden, & Hayne, 1996) and as old as 30 months of age (Bauer, Dow, Bittinger, & Wenner, 1998; Bauer & Fivush, 1992), over delays of as many as 12 months (Bauer, Wenner, Dropik, & Wewerka, 2000). It also has been extended to work with theoretically interesting special populations, such as infants born prematurely but otherwise healthy (de Haan, Bauer, Georgieff, & Nelson, 2000). We recently created a variant of the task that holds promise as a measure of working memory (Starr, de Haan, & Bauer, 2000). In short, from humble beginnings in Jean’s lab, the task of elicited imitation has proven to be quite productive indeed. It is not an exaggeration to say that it has become the paradigm of choice for researchers interested in recall memory in the transition from infancy to early childhood.

22

l

2.

Long-Term Recall Memory

DETERMINANTS OF EVENT RECALL DURING THE TRANSITION FROM INFANCY TO EARLY CHILDHOOD Use of elicited imitation as a measure of memory has provided evidence of continuity in recall processes in infants and very young children compared with older children and even adults. Continuity is particularly apparent in two major determinants of event recall: the organization of event representations and the availability of verbal reminders of to-be-remembered events. Early in our work together, Jean and I learned that just as they do in older children and adults, these factors profoundly affect event memory in children in the transition from infancy to early childhood. Because the work has been reviewed elsewhere (e.g., Bauer, 1996, 1997, in press-a), I present here only a brief summary of the issue on which Jean and I worked most closely together, namely, how well the mental representation of the event is organized. The organization of an event representation is influenced by at least two factors: (a) the nature of the temporal relations among the components of the event, and (b) familiarity, or repeated experience with the event (Mandler, 1986). Effects of Temporal

Structure

Younger children, as well as older children and adults, show superior recall of events, the temporal orders of which are characterized by enabling relations, relative to events that are arbitrarily ordered. Enabling relations are said to exist when, for a given end-state or goal, one action in a sequence is both temporally prior to and necessary for a second action in the same sequence. In contrast, for events lacking enabling relations, there are no inherent constraints on the sequence of actions; the actions are arbitrarily ordered (see Bauer, 1992, for discussion). That enabling relations facilitate recall in older children is readily apparent in the literature (see van den Broek, 199 7, for a review). Preschool-age and older children show a consistent pattern of superior ordered recall of events characterized by enabling relations, compared with events that lack such relations, and thus, are arbitrarily ordered (e.g., Hudson & Nelson, 1986). It also is apparent that enabling relations facilitate recall in very young children. In several studies, we contrasted recall of novel events containing enabling relations (e.g., making a rattle of two nesting cups and a rubber ball: put the ball into one cup, cover it with the other cup, and shake the cups to make a rattle) with that of novel arbitrarily ordered events (e.g., making a party hat: put a pompom on the top of a cone-shaped base, attach a sticker to the front of the cone. and attach a colored band

Bauer

l

23

around the base of the cone). The findings from these studies are remarkably consistent. Although children produce an equivalent number of the individual actions of enabling and arbitrarily ordered events, they do not produce an equivalent number of ordered pairs of actions: Children’s ordered recall of novel-enabling events consistently is greater than that of arbitrarily-ordered ones. The advantage is apparent both at immediate testing (e.g., Bauer, 1992) and at delayed testing (e.g., Bauer & Hertsgaard, 1993; Bauer, Hertsgaard, & Wewerka, 1995; Bauer & Mandler, 1989; Mandler & McDonough, 1995). It is apparent even after several experiences of arbitrarily-ordered events in invariant temporal order (Bauer & Travis, 1993). The robust facilitating effects of enabling relations in events cannot be attributed to the problem-solving abilities of young children: Children who have seen the goal-state or end-state of an event (e.g., the completed rattle), but not the sequence of actions necessary to produce it, do not “figure out” how to achieve the goal (e.g., Bauer, 1992; Bauer et al., 1995). Rather, as suggested by Mandler (1986), it seems that one component of an enabling event sequence provides a reliable cue to the next component, resulting in well-ordered recall (see Bauer, 1992; Bauer, Hertsgaard, Dropik, & Daly, 1998; and Bauer & Travis, 1993; for additional discussion). Effects of Familiarity

or Repeated

Experience

Familiarity or repeated experience with an event also influences organization and recall for both younger and older children. Indeed, preschoolers’ competence in event memory first was noted in the context of their reports of familiar events and routines (e.g., going to McDonald’s; Nelson & Gruendel, 1981). Although repeated experience is not necessary for accurate recall (e.g., Hudson, 1986), it aids memory both in terms of (a) the amount of information remembered (e.g., Fivush, 1984; Hudson, 1986), and (b) the length of time over which events are recalled (Fivush & Hamond, 1989). For example, Fivush and Hamond (1989) found that when asked to reenact test events after a 3-month delay, 24- and 29-month-olds who had experienced the events two times recalled more about them than children who had experienced them only once. Moreover, children with repeated experience recalled as much after 3 months as they had after 2 weeks. Similar facilitating effects of familiarity and repeated experience have been observed in children in the 1 st and 2nd year of life. Infants as young as 11 months of age accurately reproduce event sequences depicting familiar routines (e.g., giving a teddy bear a bath; Bauer & Mandler, 1992). Repeated experience is not, however, necessary for immediate or delayed recall of events: (a) 1 1-month-olds show accurate

24

l

2.

Long-Term Recall Memory

immediate reproduction of novel events characterized by enabling relations (Bauer & Mandler, 1992); and (b) after one week, 13- and 15-month-olds recall both the individual target actions and the temporal order of actions of novel events experienced only once (Bauer & Hertsgaard, 1993; Bauer et al., 1995). Nevertheless, particularly over the longer term, repeated experience clearly facilitates recall. For events experienced only once, performance after 1 month falls off precipitously in comparison to performance after 1 week. In contrast, events experienced three times before imposition of a l-month delay are well recalled. Notably, recall after 1 month of events experienced three times is comparable to that after 1 week of events experienced only once (Bauer et al., 1995). Summary

of Determinants

of Recall

In summary, in our early work using the elicited imitation paradigm, Jean Mandler and I and our colleagues demonstrated that factors known to influence recall of events by older children and adults also influence recall in very young children. Specifically, the relational structure of an event has a pronounced effect on recall. Events constrained by enabling relations are well recalled, even over substantial delays. In addition, events with which children are familiar, or have had repeated experience, are well recalled, also over long delays. EMERGENCE AND CONSOLIDATION OF LONG-TERM RECALL MEMORY FUNCTION Development of a technique to study long-term memory in preverbal and early-verbal infants and children was the first step along the road to explication of development of the ability to maintain accessible memories of specific past events over extended periods of time. The second step was demonstration of continuities in recall in younger and older children. The next steps were investigation of the course of remembering and forgetting in the transition from infancy to early childhood and extension of the study of long-term recall ability into the 1 st year of life. Recall in the 2nd and 3rd Years of life: of Long-Term Recall Memory Function

Consolidation

To date, the most comprehensive study of long-term recall memory throughout the 2nd and into the 3rd year of life has been conducted in my laboratory, under the moniker of “Monster II (Bauer, Wenner, et al.,

Bauer

l

25

2000). Children were enrolled at the age of 13, 16, or 20 months. A total of 360 children participated, 180 of whom were 16-month-olds. All of the 20-month-olds and half of the 16-month-olds were tested on event sequences four steps in length; all of the 13-month-olds and half of the 16-month-olds were tested on event sequences three steps in length. At each of three sessions, spaced 1 week apart, the children were exposed to the same six event sequences. Three of the sequences they never were permitted to imitate; three of the sequences they were permitted to imitate one time, at the end of the third exposure session. Because there was virtually no substantive difference in performance as a function of whether imitation was permitted prior to the delay, I make no further reference to this manipulation. The children returned for delayed recall testing after delays of either 1, 3, 6, 9, or 12 months (delay condition was a between-subjects manipulation). The result was a 20-cell design with 18 children per cell. At the delayed recall session, the children were tested for recall of the six event sequences to which they previously had been exposed, as well as on three new events, as a within-subjects control. For all nine of the event sequences, the children first experienced a delayed-recall period during which they were prompted by the event-related props alone, after which they were provided with verbal reminders of the event sequences (see Bauer, Wenner, et al., for details of the procedure). The complete report of the results of “Monster” is available in Bauer, Wenner, et al. (2000). Here I present data that illustrate the consolidation of long-term memory ability over the course of the 2nd year of life. The illustration begins with data on recall by infants 20 months of age at the time of exposure to the event sequences. In Figure 2.1 are the mean number of pairs of actions that 20-month-old children produced in the target or modeled order. The 20-month-olds were tested on four-step sequences. Therefore, the maximum possible number of pairs of actions that they could produce in the target order is 3 (l-2, 2-3, 3-4; credit also would be granted for nonadjacent pairs such as, for example, l-3, in that order). Three important points can be made about the data in Fig. 2.1. First, at immediate testing, children evidenced high levels of ordered recall. The average number of pairs of actions produced in the target order was 2.12, out of a possible 3.00. Second, although the children in the different delay conditions did not have identical levels of ordered recall, the levels nevertheless did not differ reliably. Thus, children in the different delay conditions entered the delay intervals with approximately equal levels of initial learning. Third, across the different delay intervals, forgetting was evident. There was an initial steep decline in performance between the last exposure session and the l-month delay;

r

Bauer

l

27

more information was lost between the I- and 3-month delays; additional information was lost between the 3- and 6-month delays; across the 6- to 12-month delays, children’s performance stayed at roughly equivalent levels. It is noteworthy that this performance curve looks very similar to that which is seen in older children and even adults: there is an initial drop in performance, followed by a smooth, shallow declining function (e.g., Schneider & Pressley, 199 7). In a verbal recall paradigm with an older child or adult, everything produced at delayed testing would be attributed to memory. However, in the nonverbal elicited imitation paradigm, such an attribution is not valid. Over the delay, children get older and “smarter,” and they get better at figuring out what to do with the event-related props presented to them. Therefore, as a control for problem solving, inference, or both, we use performance on never-before-experienced or “new N events, as a within-subjects control. Children are said to recall if their performance on previously experienced event sequences (i.e., those imitated and those only watched) is greater than their performance on event sequences new to them at the time of delayed testing. In Fig. 2.2 is depicted 20-month-olds’ performance on previously experienced and new event sequences in each of the delay conditions. In all of the delay conditions, the 20-month-olds’ performance on previously experienced event sequences was greater than performance on event sequences new to them. Thus, 20-month-olds provided evidence of temporally ordered recall at delay intervals of 1,3,6,9, and 12 months. Evidence that children as young as 20 months at the time of experience of events remember them for as long as 12 months is interesting in its own right. It also challenges assumptions that before they are able to encode them verbally, children are unable to remember the events of their lives. Nevertheless, the length of time over which children remember is not the real story here: Any number of parameters can be varied to produce variation in the length of time over which children are able to recall (see Bauer, 1996, for examples). For purposes of illustrating the consolidation of long-term memory ability over the course of the 2nd year of life, the critical data are those on the variability within the sample. The variability in children’s performance reveals that over the course of the 2nd year, the ability to recall over the long term consolidates and as it does so, it becomes more reliable and more robust. In Table 2.1 are the percentages of children at each age of enrollment who, at delayed testing, performed more pairs of actions in the target order on previously experienced than on new event sequences. An asterisk indicates that the number of children with this pattern is greater than the number that would be expected by chance. For infants who had been 20 months of age at the time of exposure to the

FIG. 2.2 Performance by 20-month-olds at delayed recall on event sequences previously experienced and event sequences new at the time of delayed testing. Data are from Bauer, Wenner, Dropik, and Wewerka (2000).

Bauer

l

29

to-be-remembered event sequences, even after 12 months, fully 67% of the sample had higher levels of performance on previously experienced than on new event sequences. Moreover, for the 20-montholds, in all of the delay conditions, the number of children exhibiting the pattern indicative of recall was reliably greater than chance. Thus, for children who had been 20 months at the time of exposure to the event sequences, not only was absolute performance on old events greater than on new events, but a large percentage of the children exhibited the pattern indicative of recall. At the shorter delay intervals, the percentage of 16-month-olds who showed evidence of temporally ordered recall was roughly equivalent to the number of 20-month-olds who exhibited recall. At the longer delay intervals, relative to the 20-month-olds, fewer of the children who had been 16 months of age at the time of exposure to the event sequences showed evidence of temporally ordered recall. Indeed, for the children who had been 16 months at the time of experience of the events, the percentage of children evidencing the pattern indicative of long-term ordered recall was reliably greater than chance in the l-, 3-, and 6-month delay conditions, but not in the 9- and 12-month delay conditions (the specific values are for 16-month-olds tested on four-step event sequences; the pattern applies to both groups of 16-month-olds). Among the 13-month-olds, there was a more rapid decline in performance across the delay conditions, compared to older children. After 6 months had elapsed, fewer than 50% of 13-month-olds evidenced the pattern indicative of ordered recall. Beyond 1 month, the number of 13-montholds who performed at higher levels on previously experienced than on new event sequences was not greater than chance. TABLE

2.1

Percentage of 13-, 16-, and 20-Month-Olds Whose Performance on Previously Experienced Event Sequences was Greater Than Performance on New Event Sequences Delay Interval Age at Exposure 20 16 13

months months months

1 month

3 month

100”

100”

94*

94*

78”

67

6 month 83”

9 month

12

month

78”

67”

72+

50

61

39

44

39

Note. Data are from Bauer, Wenner, Dropik, and Wewerka (2000). An asterisk indicates that the number of children exhibiting the pattern of ordered recall (i.e., higher level of performance on previously experienced than on new event sequences) was reliably greater than chance.

30

l

2.

Long-Term Recall Memory

The data in Table 2.1 are reflective of increasing reliability in long-term ordered recall across the 2nd year of life. At the short delay interval of 1 month, across age groups, there are roughly equivalent proportions of children who contribute to the “memory” effect: 100% of 20month-olds, 94% of 16-month-olds, and 78% of 13-month-olds showed higher levels of performance on previously experienced than on new event sequences. As delay interval increased, fewer children maintained the information that they had learned about the events. The younger the children were at the time of experience of the events, the faster they “dropped out.” Even at the longest delay interval, a random selection among 20-month-olds would yield a roughly 70% chance that the child would show temporally ordered recall. In contrast, for 13-month-olds, at 6 months and beyond, a random selection would yield a roughly 40 to 45% chance that the child would show temporally ordered recall. In addition to evidence of increasingly reliable recall, Bauer, Wenner, et al. (2000) also provided evidence of increasingly robust recall over the course of the 2nd year. Simply put, older children remembered more than younger children. Moreover, age-related differences in the amount remembered were particularly apparent under conditions of greater cognitive demand. Figure 2.3 is a schematic representation of these patterns. In Panel A of Figure 2.3 is a representation of the results of analyses of covariance in which we controlled for age-related differences in (a) children’s initial learning of the event sequences, and (b) children’s problem-solving ability at the time of delayed-recall testing (i.e., their performance on “new” event sequences). In evaluating age effects in how much is remembered, it is important to control for these two sources of variance: What we are trying to explain are age-related differences in long-term memory, not age-related differences in initial mastery or in problem solving. What the figure indicates are the delay conditions in which, after these sources of variance were controlled, the performance of older children was reliably greater than the performance of the younger children. The data are for the 20-month-olds and the 16-month-olds tested on four-step event sequences. The pattern is basically the same for the 13-month-olds and the 16-month-olds tested on three-step event sequences. The top portion of Fig. 2.3 is a schematic representation of children’s performance when they were prompted by the event-related props alone. The bottom portion of the figure is a representation of children’s performance when they were prompted by the event-related props and by the verbal reminders of the events. Three particularly noteworthy points can be made. First, across delay conditions, there were reliable age differences in the amount of information retained. Age effects were apparent in children’s ordered recall and in their production of the indi-

Bauer

.

31

Panel A Delay interval

Across delay I month

3 month

6 month

9 month

I2 month

J

J

J

J

4

J

4

4

J

J

J

4

-4

J

4

J

J

4

6 month

9 month

12 month

J

J

J

conditions Prompted by event-related Actions Pairs Prompted by event-related Actions Pairs

props alone

props and by verbal reminders

Panel B Across delay conditions Actions

4

Paki

J

Delay interval 1 month

3 month 4

FIG. 2.3 Schematic representation of age-related differences in the robustness of temporally ordered recall over long delays. Panel A indicates the results of analyses of covariance controlling for differences between 16- and 20-month-olds’ initial learning and problem-solving abilities at the time delayed recall was tested. A check indicates that the performance of the older children was reliably greater than the performance of the younger children. Panel B indicates the results of regression analyses of age as a unique contributor to predictions of levels of long-term recall. A check indicates that age was a unique predictor of children’s performance. Data are from Bauer, Wenner, Dropik, and Wewerka (2000).

vidual actions of the events. In both cases, older children performed at higher levels than younger children. Second, when the children were prompted by the event-related props alone, age differences were observed in all delay conditions. Third, when the children were prompted by the event-related props and by the verbal reminders of the event sequences, age differences were observed only at the longer delay intervals of 9 and 12 months. Thus, age effects in how much information is retained over the long term are particularly apparent under conditions of greater cognitive demand. That is, they are observed (a) when chil-

32

l

2.

Long-Term Recall Memory

dren have less support for recall (i.e., when they are prompted by event-related props alone), and (b) at longer retention intervals. There also is evidence that maturational and experiential changes associated with age have more impact on children’s ordered recall than on recall of more perceptually-supported individual target actions. Panel B of Figure 2.3 reflects the results of regression analyses in which we examined the unique contributions of a number of predictor variables. ‘Age” more frequently contributed unique variance to predictions of ordered recall than to predictions of individual target actions. In addition, consistent with the results of the analyses of covariance, age predicted unique variance in ordered recall at the longer retention intervals (6, 9, and 12 months). Together, the data indicate that over the 2nd year of life, there are increases in the reliability and the robustness of long-term recall. The changes are highly suggestive of consolidation of long-term mnemonic function over this space of time. In the following section, I argue that coalescence of the neural substrate that supports maintenance of accessible memories over the long term contributes to the consolidation process. Recall in the 1st Year of life: Memory Function

Emergence

of Long-Term

Recall

If long-term recall ability is consolidating during the 2nd year of life, when is it newly emergent? For a variety of reasons, as argued in Carver and Bauer (1999, in press) and Carver, Bauer, and Nelson (ZOOO), the most likely answer is that it emerges at about 9 months (see also C. A. Nelson, 199 7, for discussion of this time frame). Nine months of age is a likely “target” for the emergence of long-term recall ability because of suggestions that by that age, the complete neural network implicated in long-term recall memory is sufficiently functionally mature (although not fully mature) to support the behavior. Although the story is still unfolding, it is increasingly clear that long-term declarative memory is supported by a network that is comprised of (a) medial temporal lobe structures (including the hippocampus, and entorhinal and perirhinal cortices), and (b) higher cortical association areas, including the prefrontal cortex and limbic/temporal association areas (Bachevalier & Mishkin, 1994; Murray & Mishkin 1998). There are a number of indicators that most (although not all) of the medial temporal lobe components of the declarative memory system develop early (see Seress, 2001, for a review). In contrast, the association areas, and the reciprocal connections between the hippocampus and the neocortex, develop more slowly (Bachevalier, Brickson, & Hagger, 1993;

Bauer

.

33

Bachevalier & Mishkin, 1994). The best available evidence suggests that in the human infant, the entire temporal-cortical network begins to coalesce near the end of the 1 st year of life (C. A. Nelson, 1997); development continues into the early part of the second year (Carver & Bauer, in press). Maturational changes throughout the network continue into adolescence (e.g., myelination in frontal cortex; Johnson, 1997). For researchers interested in the development of long-term recall in the human infant, this time frame has intriguing implications. A major implication is that, relatively early in development, we should see evidence of mnemonic behavior that is supported by the medial temporal lobe structures themselves. In contrast, behaviors that require the entire declarative memory network should appear only as it reaches sufficient functional maturity near the end of the 1st year. Specifically, because of the role played by prefrontal cortex in retrieval of information over the long term, and recall of temporal order information in particular, changes in long-term ordered recall ability could logically be expected to be indicative of functional maturation of the substrate of long-term declarative memory. The predicted time course for such development is the 2nd half of the 1st year of life (see Carver & Bauer, 1999, in press, for discussion). The expectations derived from this line of reasoning are that during the first half of the 1 st year of life, infants will be able to recall past events but their recall will be temporally limited and ordered recall will present a particular challenge. In contrast, during the second half of the 1 st year of life, evidence of long-term temporally ordered recall should become more apparent. In our initial inquiry, Leslie Carver and I tested 9-month-olds’ recall of two-step event sequences analogous to the three- and four-step sequences on which the older children in Bauer, Wenner, et al. (2000) were tested. The infants were exposed to the event sequences at each of three sessions, spaced 2 to 4 days apart. The infants only watched the events being produced; they did not imitate them themselves. Recall was tested 5 weeks later, as was performance on event sequences new to the children, as a within-subjects control. The results were quite consistent with the proposed time frame. In the sample as a whole, the children showed evidence of recall of the individual target actions of the event sequences. However, 14 of 31 infants (45%) showed evidence of temporally-ordered recall memory after 5 weeks; the remaining 17 of 3 1 infants (55%) did not show evidence of temporally ordered recall. We since have replicated this distribution in two separate samples of 9-month-olds (Bauer, Johnson, Carver, Waters, & Nelson, 2000; Bauer, Wiebe, Waters, & Bangston, 2001). Although it is not yet entirely clear precisely what the observed individual differences in long-term ordered recall mean, our interpretation

34

.

2.

Long-Term Recall Memory

is that the data are indicative of differences in infants’ ability to store information over extended periods of time (Carver & Bauer, 1999, in press; Carver et al., 2000). That we are finding differences at 9 months suggests that this is an especially important period in development of long-term recall ability. This conclusion would, of course, be undermined by evidence of long-term ordered recall in children significantly younger than 9 months. Although we have not done the precise experiments, data from Harlene Hayne’s laboratory suggest that if we were to look at a younger age, we would not find such evidence. In Barr et al. (1996), infants as young as 6 months of age were tested for immediate and 24hr delayed recall of the three-step action sequence of pulling a mitten off a puppet’s hand, shaking the mitten (which, at demonstration, contained a bell), and replacing the mitten on the puppet’s hand. They found that 75% of 6-month-olds imitated at least one action after the 24-hr delay. These data thus provide evidence of recall over the short term as early as 6 months. Interestingly, what was not apparent in Barr et al.‘s data was compelling evidence of ordered recall over the delay. Specifically, whereas 75% of 6-month-olds produced one action after the 24hr delay, only 25% of them provided evidence of memory for more than one step of the sequence. Barr et al.‘s data are consistent with the suggestion that relatively early in development, mnemonic behaviors that can be supported by medial temporal lobes structures themselves will be in evidence, whereas behaviors that likely require further maturation of the neural substrate show a more protracted course of development. Summary In summary, across experiments, a picture of the early development and subsequent consolidation of long-term ordered recall memory has emerged. When we consider data from 6-month-olds to N-montholds, over delays of 24 hr to 12 months, temporally ordered recall emerges gradually, over the second half of the 1 st year and into the 2nd year of life. As depicted in Table 2.2, at 6 months of age, 25% of infants retain order information for 24 hr. Although we do not have data on 6-month-olds’ ordered recall at retention intervals longer than 24 hr, based on the results at 24 hr, it is unlikely that the behavior would be readily apparent at appreciably longer retention intervals. At 9 months of age, approximately 45% to 50% of infants retain temporal order information for periods of at least 1 month. The capacity undergoes further development over the next few months, such that by 13 months of age, fully 78% of infants retain temporal order information for 1 month. By the second half of the 2nd year, the competence is even more

Bauer

l

35

TABLE 2.2 Across-Experiment Summary of Patterns of Long-Term Recall by Infants and Children 6 Months to 20 Months of Age: Percentage of Children Showing Evidence of Ordered Recall Delay Interval Age at Exposure 6 months8 9 monthsb 13 month$ 16 months’ 20 months’

24 hour

1 month

6 month

12 month

?

25%

45% 78%

? 39%

94% 100%

72% 83%

Note. ‘Data are from Barr, Dowden, and Hayne (1999). ‘Data are from Bauer, Wenner, Dropik,

(1996). bData are from and Wewerka (2000).

39% 61% 67% Carver

and Bauer

reliable: 94% of 16-month-olds and 100% of 20-month-olds evidence temporally ordered recall for 1 month. At longer retention intervals of 6 months and more, children early in the 2nd year of life less reliably recall temporal order information. That is, by 13 months, roughly 40% of infants show evidence of temporally ordered recall after 6 months. In contrast, by 20 months, roughly 70% of infants show temporally ordered recall, even after as many as 12 months. At the same time that more infants demonstrate the capacity to recall over long delays, there also are age-related increases in the robustness of memory, as evidenced by age-related differences in the amount of information retained. Age differences are especially apparent under conditions of greater cognitive demand (i.e., when recall is prompted by event-related props alone; when recall is tested after long retention intervals; in recall of temporal order information relative to recall of individual target actions). LONG-TERM

RECALL AS EVIDENCED

BY VERBAL REPORT

The results that I have reviewed make clear that by the 2nd children reliably form and retain memories of specific past extended periods of time. The significance of this capacity overstated: among other “minor details,” long-term recall mits the formation of a personal past. Verbal expression graphical or personal memories signals the second important

year of life, events over cannot be ability perof autobiochange in

36

l

2.

Long-Term Recall Memory

mnemonic function in the transition from infancy to early childhood. Although they were derived from a nonverbal paradigm, results from studies such as Bauer, Wenner, et al. (2000) shed light on this important development. In the course of children’s visits to the laboratory to demonstrate their memories nonverbally, the children in Bauer, Wenner, et al. also provided spontaneous verbal “reports’ of’ their experiences. In Bauer, Kroupina, Schwade, Dropik, and Wewerka (1998) we characterized the children’s spontaneous verbalizations as either indicative of memory or not indicative of memory. To be indicative of memory, a verbalization had to include (a) a name or description of an event (e.g., of the rattle: “it makes a loud noise”), (b) a question about or statement regarding a target action of the event while not performing it (e.g., “put the ball in there” [indicating one of the cups]), or (c) a request of an as yet unseen prop or event (e.g., “can I see the rattle?“). We analyzed the verbalizations of 20 children who had been 16 months of age at the time of experience of the event sequences and 20 children who had been 20 months of age at the time of experience of the event sequences. In each age group, 10 of the children had been in the 6-month delay condition and 10 had been in the 12-month delay condition. As illustrated in Panel A of Table 2.3, children who had been 16 months of age at the time of experience of the events provided spontaneous mnemonic verbalizations about event sequences that they had imitated but not about sequences that they had only watched. That is, the number of mnemonic utterances attributed to 16-month-olds on the events that they previously had imitated was greater than the number of mnemonic utterances attributed to the children on the event sequences that were new to them at the time delayed recall was tested (coders were unaware as to whether an event sequence had been imitated, had been watched only, or was new to the children). In contrast, the children produced no more mnemonic utterances on the events that they had only watched than on the event sequences that were new to them. That the 16-month-olds provided verbal evidence of memory on event sequences previously imitated is particularly telling regarding the nature of the memory representation on which performance was based: only declarative memory representations are accessible to verbal report. In contrast to the 16-month-olds who provided verbal evidence of memory of events imitated but not of events only watched, the 20-month-olds provided verbal reports about events imitated and watched (Bauer, Kroupina, et al., 1998). Encouraged by these data, we invited the children to come back to the laboratory at 36 to 42 months, an age by which they could be counted on to be more active participants in memory conversations (see Bauer, Kroupina, et al., 1998, for discussion). At that time, follow-

Bauer

l

37

TABLE 2.3 Average Numbers of Mnemonic Utterances Produced by Children 16 and 20 Event Sequences. Months of Age at the Time of Ex osure to To-Be-Remembered Panel A Indicates the Number o P Spontaneous Mnemonic Utterances at 22 to 32 Months of Age. Panel B Indicates the Number of Elicited Mnemonic Utterances at 36 to 42 Months of Age Event Type Age at Experience

Imitated (Number

Panel A: Spontaneous 16 20

month month

Panel B: Elicited 16 20

verbal

verbal

month month

Note.

Data

are from

Bauer,

New

Watched

“reports”

of Mnemonic

22-32

Utterances)

months

0.93

0.48

0.50

1.30

1.10

0.43

6.48

5.45

5.80

7.80

6.80

3.70

reports

36-42

Kroupina,

months

Schwade,

Dropik,

and Wewerka

(1998).

ing procedures outlined in detail in Bauer, Kroupina, et al. (1998), we elicited verbal reports about the events. As suggested by inspection of Panel B of Table 3, whereas the children who had been 16 months at the time of exposure to the test events did not maintain verbal accessibility of memories into the 4th year of life (i.e., the number of mnemonic verbalizations did not differ across events imitated, events watched, and events that were new), the children who had been 20 months of age did. That is, for the children who had been 20 months of age at the time of exposure to the event sequences, at the age of 36 to 42 months, both for events previously imitated and for events previously only watched, there was evidence of verbally accessible memories of the previously experienced event sequences. These data are evidence of long-term verbal accessibility of memories originally encoded at 20 months and likely without the benefit of language. Although they are not reports of significant personal experiences, they nevertheless bear on the question of the fate of early memories: At least under some circumstances, they are accessible to verbal report, even after long delays.

38

l

2.

Long-Term Recall Memory CONCLUSION

In the most recent phase of her long and productive career, Jean Mandler contributed to nothing short of a revolution in the way in which the field thinks about infant memory. Little more than a decade ago, we knew virtually nothing about long-term recall memory in infants and very young children. Jean Mandler helped to lessen our ignorance by spearheading development of a nonverbal technique that was suitable for the study of event memory in preverbal and early-verbal infants and children. The technique not only meets criteria as a nonverbal analogue to verbal report, but has been shown to yield patterns of recall that are qualitatively similar to those observed in older children and even adults. What is more, it has been used to chart the course of development of the fundamental cognitive competence of the ability to recall the past. As a result of work initiated in Jean Mandler’s laboratory (and independently, in Andrew Meltzoff ‘s), we now know, for example, that the ability to recall specific events over long delays undergoes significant development in the second half of the 1st year of life. Between 6 and 9 months we see an increasing number of children who are able to maintain accessible memories of events, such that by 9 months of age, about half of infants demonstrate ordered recall after 1 month. This likely indicates that by 9 months, for at least half of infants, the entire neural network thought to support long-term declarative memory is Over the course of the second year of life, “up and running.” long-term recall ability becomes more reliable and more robust. At delays of 1 month, 80% to 100% of 13- to 20-month-olds show temporally ordered recall; at delays of 6 months, roughly 40% of 13-month-olds, and 70% to 80% of 16- and 20-month-olds demonstrate ordered recall. At delays of 12 months, roughly 70% of 20-month-olds still remember. Moreover, there are age-related differences in the amount recalled, particularly under conditions of greater cognitive demand (e.g., in the absence of verbal reminders, at long delays, when retention of temporal order information is required). Together, these data illustrate the emergence and gradual consolidation of long-term ordered recall ability. Work inspired by Jean Mandler also has informed our understanding of the transition from exclusively nonverbal to primarily verbal expression of memory. In the supportive context of the laboratory, not only are children able to show us that they remember, nonverbally, they also are able to tell us. They produce verbal reports when we ask them to do so, and even when we do not ask them to. Some of the verbal reports are of events that initially likely were encoded without the ben-

Bauer

.

39

efit of language. That months later, children are able to talk about events for which they did not have language at the time of experience, suggests that at least under certain circumstances, early memories are verbally accessible later in development. This evidence has important implications for the second major development in mnemonic competence during the transition from infancy to early childhood, namely, the offset of infantile amnesia and the onset of autobiographical memory. The significant progress that we have made in understanding important developments in memory in infancy and early childhood is a fitting tribute to Jean M. Mandler. ACKNOWLEDGMENTS The work reported in this chapter was supported by NICHHD (HD-28425) to Patricia J. Bauer. I also thank the many collaborators who have helped to make the research possible. Those most directly associated with the research discussed are Leslie Carver, Patricia Dropik, Maria Kroupina, Jennifer Schwade, Jennifer Wenner, and Sandi Wewerka. I also thank the children and parents who graciously volunteered their time and energy to the efforts reported here. REFERENCES Bachevalier, J., Brickson, M., & Hagger, C. (1993). Limbic-dependent recognition memory in monkeys develops early in infancy. Neuroreport, 4,77-80. Bachevalier, J., & Mishkin, M. (1994). Effects of selective neonatal temporal lobe lesions on visual recognition memory in rhesus monkeys. The Journal of Neuroscience,

14, 2 128-2

139.

Barr, R., Dowden, A., & Hayne, H. (1996). Developmental changes in deferred imitation by 6- to 24-month-old infants. infant Behavior and Development, 19,159-170. Bauer, I? J. (1992). Holding it all together: How enabling relations facilitate young children’s event recall. Cognitive Development, 7, l-28. Bauer, I? J. (1996). What do infants recall of their lives? Memory for specific events by l- to 2-year-olds. American Psychologist, 51, 29-41. Bauer, I? J. (1997). Development of memory in early childhood. In N. Cowan (Ed.), The development of memory in childhood (pp. 83-111). Hove East Sussex, England: Psychology Press. Bauer, I? J. (in press-a). Early memory development. In U. Goswami (Ed.), Handbook of cognitive development. Oxford, England: Blackwell. Bauer, I? J. (in press-b). New developments in the study of infant memory. In D. M. Teti (Ed.), Handbook ofresearch methods in developmental psychology (pp. 000-000). Oxford, England: Blackwell. Bauer, I? J., Dow, G. A., Bittinger, K., & Wenner, J. A. (1998). Accepting and exempting the unexpected: 30-month-olds’ generalization of event knowledge. Cognitive Development, 23, 421-452.

40

l

2.

Long-Term

Recall Memory

Bauer, P J., & Fivush, R. (1992). Constructing event representations: Building on a foundation of variation and enabling relations. Cognitive Development, 7,381-401. Bauer, I? J., & Hertsgaard, L. A. (1993). Increasing steps in recall of events: Factors facilitating immediate and long-term memory in 13.5- and 16.5 month-old children. Child Development, 64, 1204-1223. Bauer, l? J., Hertsgaard, L. A., Dropik, I?, & Daly, B. P (1998). When even arbitrary order becomes important: Developments in reliable temporal sequencing of arbitrarily ordered events. Memory, 6, 165-l 98. Bauer, l?J., Hertsgaard, L. A., & Wewerka, S. S. (1995). Effects of experience and reminding on long-term recall in infancy: Remembering not to forget. JournaZ of Experimental Child Psychology, 59, 260-298. Bauer, I? J., Johnson, K., Carver, L. J., Waters, J. M., & Nelson, C. A. (2000). Behavioral and electrophysiological evidence of developments in long-term recall memory in infancy. Manuscript submitted for publication. Bauer, I? J., Kroupina, M. G., Schwade, J. A., Dropik, I?, & Wewerka, S. S. (1998). If memory serves, will language? Later verbal accessibility of early memories. Development and Psychopathology, 10, 655-679. Bauer, F?J., & Mandler, J. M. (1989). One thing follows another: Effects of temporal structure on l- to 2-year-olds’ recall of events. Developmental Psychology, 25,197-206. Bauer, P J., & Mandler, J. M. (1992). Putting the horse before the cart: The use of temporal order in recall of events by one-year-old children. DeveZopmentaZ Psychology, 28, 441452. Bauer, I? J., 81Shore, C. M. (1987). Making a memorable event: Effects of familiarity and organization on young children’s recall of action sequences. Cognitive Development, 2, 327-338. Bauer, I? J., & lIavis, L. L. (1993). The fabric of an event: Different sources of temporal invariance differentially affect 24-month-olds’ recall. Cognitive Development, 8, 3 19-34 1. Bauer, l? J., Wenner, J. A., Dropik, I?, & Wewerka, S. S. (2000). Parameters of remembering and forgetting in the transition from infancy to early childhood. Monographs of the Society for Research in Child Development, 65 (4, Serial No. 263). Bauer, I? J., Wiebe, S., Waters, J. M., & Bangston, S. K. (2001). Reexposure breeds recall: Effects of experience on 9-month-olds’ ordered recall. Journal of Experimental Child Psychology, 80, 174-200. Brewer, W. F. (1986). What is autobiographical memory? In D. C. Rubin (Ed.), Autobiographical memory (pp. 2549). New York: Cambridge University Press. Carver, L. J., & Bauer, F?J. (1999). When the event is more than the sum of its parts: Nine-month-olds’ long-term ordered recall. Memory, 7, 147-l 74. Carver, L. J., & Bauer, P J. (in press). The dawning of a past: The emergence of long-term explicit memory in infancy. Journal of Experimental Psychology: General. Carver, L. J., Bauer, P J., & Nelson, C. A. (2000). Associations between infant brain activity and recall memory. Developmental Science, 3, 234-246 de Haan, M., Bauer, I? J., Georgieff, M. K., & Nelson, C. A. (2000). Explicit memory in low-risk toddlers born between 27-42 weeks of gestation. Developmental Medicine and Child Neurology, 42, 304-3 12.

Bauer

.

41

Fivush, R. (1984). Learning about school: The development of kindergartners’ school scripts. Child Development, 55, 1697-l 709. Fivush, R., Gray, J. T, & Fromhoff, E A. (1987). ?hro-year-olds talk about the past. Cognitive Development, 2, 393-410. Fivush, R., & Hamond, N. R. (1989). Time and again: Effects of repetition and retention interval on 2 year olds’ event recall. Journal ofExperimentaZ Child Psychology,

4 7, 25 9-2 73.

Freud, S. (1916/1966). introductory lectures on psychoanalysis. Translated and edited by J. Strachey. New York: Norton. (Original work published 1916-1917) Hudson, J. A. (1986). Memories are made of this: General event knowledge and development of autobiographic memory. In K. Nelson (Ed.), Event knowledge: Structure andfunction in development (pp. 97-l 18). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Hudson, J. A., & Nelson, K. (1986). Repeated encounters of a similar kind: Effects of familiarity on children’s autobiographical memory. Cognitive DeveZopment, I, 253-271. Johnson, M. H. (1997). Developmental cognitive neuroscience. Oxford, England: Blackwell. Mandler, J. M. (1984). Representation and recall in infancy. In M. Moscovitch (Ed.), Infant memory: Its relation to normal and pathological memory in humans and other animals (pp. 75-101). New York: Plenum. Mandler, J. M. (1986). The development of event memory. In F. Klix & H. Hagendorf (Eds.), Human memory and cognitive capabilities-Mechanisms and peqormance (pp. 459467). New York: Elsevier. Mandler, J. M. (1990). Recall of events by preverbal children. In A. Diamond (Ed.), The development and neural bases of higher cognitive functions (pp. 485-516). New York: New York Academy of Science. Mandler, J. M., & McDonough, L. (1995). Long-term recall of event sequences in infancy. Journal of Experimental Child Psychology, 59, 457474. McDonough, L., Mandler, J. M., McKee, R. D., & Squire, L. R. (1995). The deferred imitation task as a nonverbal measure of declarative memory. Proceedings of the National Academy of Sciences, 92, 7580-7584. Meltzoff, A. N. (1985). Immediate and deferred imitation in fourteen- and twenty-four-month-old infants. Child Development, 56, 62-72. Meltzoff, A. N. (1990). The implications of cross-modal matching and imitation for the development of representation and memory in infants. In A. Diamond (Ed.), The development and neural bases of higher cognitivefunctions (pp. l-3 7). New York: New York Academy of Science. Murray, E. A., & Mishkin, M. (1998). Object recognition and location memory in monkeys with excitotoxic lesions of the amygdala and hippocampus. Journal

of Neuroscience,

18, 6568-6582.

Nelson, C. A. ( 199 7). The neurobiological basis of early memory development. In N. Cowan (Ed.), The development of memory in childhood (pp. 41-82). Hove East Sussex, England: Psychology Press. Nelson, K. (Ed.) (1986). Event knowledge: Structure andfunction in development. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Nelson, K. (1993). The psychological and social origins of autobiographical memory. Psychological Science, 4, 7-14. Nelson, K., & Gruendel, J. (1981). Generalized event representations: Basic building blocks of cognitive development. In M. E. Lamb & A. L. Brown

42

l

2.

Long-Term

Recall Memory

(Eds.), Advances in developmental psychoZogy (Vol. 1, pp. 131-158). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Piaget, J. (1952). The origins of inteZZigencein children. New York: International Universities Press. Ratner, H. H. (1980). The role of social context in memory development. In M. Perlmutter (Ed.), New directions for child development-Children’s memory (pp. 49-67). San Francisco: Jossey-Bass. Schneider, W., & Pressley, M. (1997). Memory development between two and twenty, second edition. Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Seress, L. (2001). Morphological changes of the human hippocampal formation from midgestation to maturation. In C. A. Nelson & M. Luciana (Eds.), Handbook of developmental cognitive neuroscience (pp. 45-58). Cambridge, MA: MIT Press. Starr, R. M., de Haan, M., & Bauer, P J. (2000). AZZin good time: Temporal integration and working memory in I7- and 20-month-old children. Manuscript submitted for publication. van den Broek, P W. (199 7). Discovering the cement of the universe: The development of event comprehension from childhood to adulthood. In I? W. van den Broek, I? J. Bauer, & T Bourg (Eds.), Developmental spans in event comprehension and representation: Bridging fictional and actual events (pp. 321-342). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. West, T. A., & Bauer, I? J. (1999). Assumptions of infantile amnesia: Are there differences between early and later memories? Memory, 7, 25 7-278.

The Origin of Concepts: Continuing the Conversation Susan Carey

New York University

I

have long been an admirer of Jean Mandler’s work, and recently entered into a public conversation with her through a commentary on her article “Perceptual and Conceptual Processes in Infancy” (Mandler, 2000a). My contribution to this volume is a continuation of that conversation, namely a reply to her replies (Mandler, 2OOOb) in the same issue of the Journal of Cognition and Development. I have left my original commentary intact, assuming that not all readers of this volume will have read my original remarks. THE ORIGINAL

COMMENTARY

(CAREY, 2000)

Jean Mandler ‘s rich and nuanced article develops a four part argument: I. An adequate guish conceptual

characterization representations

of the adult mind must distinfrom perceptual representations.

2. A priori arguments cannot decide the issue of the ontogenetic roots of each type of representation, or of the relations between them. In particular, there is no convincing a priori argument that in43

44

.

3.

Origin of Concepts

fants’ representations are first perceptual and only later conceptual. Furthermore, there is no known mechanism through which conceptual representations may be built from perceptual representations. 3. Data from three paradigms, sequential touching, manual habituation, and imitation, provide convergent evidence that conceptual categories exhibit a different course of development than do perceptual categories. Specifically, conceptual categories emerge at the domain level (e.g., animal vs. vehicle) before the basic (dog vs. cat) or subordinate (poodle vs. collie) levels, whereas most available evidence suggests that categories based on visual similarity are formed at the basic level first. 4. A process of attentive categories.

perceptual

analysis

yields conceptual

It is always easy to quibble with a programmatic statement with the scope of Mandler’s lovely article, and I shall not be able to refrain from doing so. But first, I would like to strongly endorse, quibbles aside, the first three parts of her argument. I reserve serious doubts only for the fourth, and offer a friendly amendment to Mandler ‘s project. Distinguishing

Perceptual

From Conceptual

Representations

In distinguishing perceptual from conceptual representations, Mandler most often appeals to a difference between what entities look like and what kinds of entities they are. Additionally, she claims that conceptual representations differ from perceptual representations in being consciously accessible, supporting problem solving and inference, and being stored in long-term memory. To quibble, it is not clear that these different properties determine a single type of representation, or whether Mandler considers all of these to be properties of conceptual representations. Each may characterize at least some perceptual representations. For example, some clear cases of perceptual representations, such as a toothache or the experience of redness, may be consciously accessible. Nonetheless, the distinction between categories based on what entities look like (or sound like or feel like) and categories based on what kind of things entities are can be cashed out in the tradition of the “theory theory” of concepts (Carey, 1985; Gopnik & Meltzoff, 1997; Keil, 1989). In this tradition, concepts such as those discussed in Mandler’s paper (tiger, mammal, animal, bird, duck, vehicle, car, cup, container, etc.) include both core and peripheral features. The core of the concept includes its causally deepest properties, those properties that determine

Carey

l

45

what kind of thing the entity is and that determine the entity’s particular properties. In this tradition, conceptual categories are those with cores, those for which adults take the stance Medin & Ortony (1989) call “psychological essentialism.” Core properties, or essential properties, are often not perceptually available. For example, in the adult concept of a tiger, its essence is inherited from its parents, is internal, and is not observable (Keil, 1989). If concepts’ cores include nonobservable causal constructs, then concepts for which have cores have a nonperceptual component. At least since John S. Mill, it has been recognized that the attribution of causality goes beyond spatio-temporal analysis. Even in the case of Michotte-like contact causality, the mind contributes the causal attribution; perception merely gives us contact, simultaneity, and so forth. Similarly, perception gives us aspects of paths and contingency; the mind attributes goals to agents. This is a subtle point. The fact that we can see a physical event as a causal interaction does not make the concept causal interaction perceptual, any more than the fact that we can see a certain building as a nuclear reactor makes the concept nuclear reactor perceptual. Perceptual categories, as I and I believe Mandler mean them, are formed from observational properties such as red, square, dog-shaped, and spatio-temporally specified aspects of motion. A Priori Relations Categories

Between

Conceptual

and Perceptual

Here I have no quibbles with the points Mandler develops in her paper and wish only to add the following observation. If causal analysis is at the core of at least some conceptual representations, then a relevant literature as to the origin of conceptual representations is the literature on infants’ appreciation of causality (Gergeley, Nadasdy, Csibra, & Biro, 1995; Leslie, 1988; Spelke, Phillips, & Woodward, 1995). As I read this literature, at least by the time infants are 7 to 12 months of age, the earliest ages in which Mandler ‘s studies are revealing conceptual representations of domain-level concepts, infants appreciate both Michotte-like contact causality (Michotte, 1963), and aspects of agency. Further, both types of causality are attributed to entities on the basis of analysis of their action, not on the basis of what they look like. Gergeley and his colleagues (1995) showed that 12-month-old infants attribute goal directness to dots which appear to chase other dots through gaps in walls or appear to jump over barriers to reach each other; Johnson, Slaughter, and Carey (1998) showed that 12-month-olds follow the focus of attention of faceless, amorphous robots, just so long as those robots in-

46

l

3.

Origin

of Concepts

teract contingently with the baby. Reiterating Mill’s argument: Although perceptual information is required for the infant to identify cases of contact causality or intentional goal directed activity, the causal attributions go beyond the perceptual information. For Conceptual Categories, Precede, Developmentally,

Domain-Level Distinctions Basic-Level Distinctions

Whether a task reflects perceptual or conceptual categorization is both an empirical and a theoretical matter. It depends on how we draw the perceptual or conceptual distinction, and then it depends on what drives the behavior in question. A minor quibble concerning this part of Mandler ‘s argument is that the tasks in which she and her colleagues have shown domain-level distinctions to precede basic-level distinctions (sequential touching, manual habituation, and imitation) do not transparently reflect conceptual rather than perceptual categorization. It is an open question, for example, whether infants of the age Mandler has studied would habituate to a manually presented class of objects united by a clearly perceptual property. Van de Walle (1999) showed that they will; 9-month-olds presented with a series of red horses recover interest when allowed to play with a yellow horse, and generalize habituation when offered a red pig. That the basic level distinction did not determine recovery of interest is, of course, consistent with Mandler’s own data from this age. What is new is that the color distinction did. Manual habituation at 9 months can reveal perceptual categorization. This is not to say that the domain-level distinctions reflected in these tasks do not reflect categories with conceptual cores. Rather, my point is simply that the nature of the representational distinction subserving some behavioral distinction cannot simply be ascertained by considering the nature of the task itself. That being said, I do agree that the imitation or induction paradigm in particular seems on its face to reflect conceptual categorization. ‘IAffording keying” and “drinks from a cup” are not observation properties, and the infants’ generalization patterns were unaffected by perceptual similarity. The convergence of developmental pattern from the three paradigms-much earlier evidence of domain-level distinctions before basic-level distinctions-is a very important set of findings in the developmental literature. For the sake of argument, let us accept these findings as reflecting a distinct course of development of perceptual and conceptual categorization. I certainly believe Mandler is right on this point, although it is difficult to establish this beyond doubt.

Carey

A Process of Attentive Perceptual the First Conceptual Categories

Analysis

l

47

Yields

Mandler does not think that the concept animal, uniting birds, fish, toads, dogs, horses, snakes, and so forth, is innate. She does think that by ages 7 to 9 months, infants have formed such a category, and this domain-level category has a conceptual core (seZf-moving agent). The problem, then, is how the infant goes from the state of not having the domain-level concept of animal to having one. The answer Mandler gives is that the infant produces conceptual categories from perceptual input through a process of active, attentive, perceptual analysis. Visual information is redescribed into a simpler and explicitly realized form, most likely in the format of an image schema, but perhaps in a more abstract format. The content of these explicit conceptual relations includes paths objects take, plus various relations among objects such as containment, support, contact, and contingent relations. I see several problems with this proposal. Most trivially, Mandler offers no evidence for the “active, attentive” character of the process of perceptual analysis. From adult studies of Michotte on contact causality, it is clear that some conceptual attributions that go beyond the perceptual input are automatically invoked. Although there is no evidence on this point, I would think this is also likely to be true for infants in the case of contact causality, as well as in the case of the attribution of agency and goal directedness to the interactions of moving entities. Much more importantly, for reasons that Mandler clearly lays out in her sections on why there is no known mechanism for deriving conceptual categories from perceptual ones, there is equally no known way that perceptual analysis could do the trick on its own. Where do the categories represented in the image-schematic meanings themselves come from? If one cannot derive causality from spatio-temporal descriptions, or agency from spat&temporal descriptions (even those that provide the necessary input for attributions of each type of causality), then the problem of how these concepts arise has not been solved. In sum, although I endorse the first three parts of Mandler’s argument, I do not believe that perceptual analysis can be the mechanism through which conceptual representations are formed. I offer a friendly amendment to Mandler ‘s story, and suggest she look to the literature on core knowledge (cf. Carey & Spelke, 1994, 1996) for the origin of the cores of conceptual categories. By hypothesis, core knowledge derives from innate learning mechanisms in at least two domains-intuitive mechanics, with the concept of an object and contact causality at its core, and intuitive psychology, with the concept of an agent and intentional causality at its core. Core knowledge is appropri-

48

l

3.

Origin of Concepts

ately abstract, takes perceptual information as its input, and outputs the event-level descriptions Mandler needs for her very persuasive account of the distinct developmental course of perceptual and conceptual categorization to fly. MANDLER’S

REPLY (MANDLER,

moob) AND A REJOINDER

Mandler quibbled with some of my quibbles, and I will leave quibbles aside, getting to the heart of our disagreements. The key issues follow: How are we to analyze conceptual categories as distinct from perceptual categories? When we distinguish what things look like, smell like, feel like, sound like, on the one hand, from what kinds of things they are, on the other hand, what distinction are we drawing? What is the relation between the output of perceptual analysis and conceptual representation? I still see a deep tension in Mandler’s views on these matters. Although Mandler rejected my friendly amendment, I will respectfully try again to persuade her to consider it. I believe that part of her rejection is due to misunderstanding of what I am suggesting, misunderstanding I hope I can clear up here. One problem is the unfortunate terminological confusion between two uses of the word “core” : a concept’s “core” (sense 1) and “core knowledge” (sense 2). In the literature on concepts, many writers (beginning, I believe, with Miller and Johnson-Laird, 19 76) distinguish between the core features (sense 1) of a concept and more peripheral features. The core features are the most heavily weighted in determining category membership. Mar-idler states that she is sympathetic to the theory theory of concepts, but doubts that the causally deepest properties of entities constitute their core, those properties that determine kind membership and explain their surface features. I don’t see how it is possible to have sympathy for the theory theory of concepts and to doubt that the causally deepest properties of an entity determine what it is, its kind. According to the theory theory of concepts, categorization decisions are a form of inference to best explanation; given the features of an entity, one makes the best guess as to the kind that would explain the entity’s features. But explanation and causality are inextricably linked (Salmon, 1989). Explanation, like causality, is asymmetrical. We explain an animal’s capacity to bark in terms of it’s being a dog, not vice versa; those features that make something a dog also make it bark. In the case of artifacts, the causally or explanatorily deepest features derive from the intended function of the maker-a cup is a cup because its designer intended it to be used for certain purposes, and this constrains its shape, size, the material from which it is made, and so on. In the case of natural kinds, the causally or explanatorily deepest features are rep-

Carey

’

49

resented in terms of the framework theories in which that entity participates. The stance of psychological essentialism with respect to natural kinds is simply the stance that entities have essential (often hidden, often unknown) features that make them the kind they are and determine their properties. Mandler complained that psychologists have been unduly influenced by philosophical discussions of concepts, thereby confusing ontological questions and epistemological ones. I don’t think this is true. The philosophical literature on natural kinds points out that what makes a dog a dog is a matter of nature, not mental representations. If dogs or gold have essences, this is a fact about the world, not our mind (Kripke, 1972; Schwartz, 1977). But the theory theory of concepts, and the doctrine of psychological essentialism, is firmly about mental representations, about the representations that determine our decisions about category membership and guide our inferences about entities. And there is a great deal of empirical evidence for psychological essentialism (Keil, 1989; Medin & Ortony, 1989). For example, Keil’s (1989) transformation studies show that by ages 7 to 9, children take parentage to fix the kind of an animal, not bodily features or characteristic behaviors. Similarly, Matan and Carey (2001) show that by age 6 (but not at age 4), children take original intended function as determining artifact kind. Finally, elegant work by Ahn, Kim, Lassaline, and Dennis (2000) demonstrated that for adults at least, the cores of concepts are indeed the causally deepest known features of the entities that fall under them. Mandler pointed out that it is absurd to believe that the cores of infants’ artifact categories are such features as original intended function (for artifacts) or parentage (for animal kinds), and I fully agree. Psychological essentialism requires a commitment to their being causally deep properties that determine kind, but a person’s best guess as to what these might be come from the causal or explanatory knowledge they then possess. Inference to best explanation, as a psychological process, is constrained by a person’s current explanatory knowledge. As Mandler pointed out, alI of my own work on conceptual change can be seen (and has been seen by me) as reflecting the joint changes in conceptual cores, conceptual structures, and theories in the course of conceptual development. So my friendly amendment to Mar-idler’s program was certainly not that the core of infants’ concepts are the same as those of older children or adults. So what was I suggesting as a friendly amendment to Mandler ‘s program? To understand this, one must turn to the second sense of the term “core,” in Spelke’s (e.g., Spelke, Breinlinger, Macomber, & Jacobson, 1992) proposals concerning core knowledge (see also, Carey &

50

.

3.

Origin of Concepts

Spelke, 1994,1996). According to the core knowledge hypothesis, conceptual development gets off the ground with the support of domain specific learning mechanisms with the following properties: (a) the entities in the domain are identified by dedicated perceptual analyzers, (b) these analyzers continue to operate throughout life, (c) often, core knowledge is evolutionarily ancient, and (d) often, it is supported by dedicated neural circuits. Three domains of core knowledge that meet all of these specifications are knowledge of objects and some of their mechanical interactions, knowledge of agents and some aspects of intentional causality, and knowledge of number (both analog magnitude representations and systems of parallel individuation of small sets of individuals). My friendly amendment to Mandler’s program is that the systems of core (sense 2) knowledge may provide some of the conceptual primitives that provide the earliest cores (sense 1) of kind concepts. Mandler and I agree that the core of domain concepts like animal (for babies) is likely to be cashed in terms of intentional causality, not biological essentialism; animals are entities that move on their own, have goals, pay attention to entities in the world, and interact contingently at a distance. However, some of these core representations are the output of the dedicated perceptual analyzers in one of the systems of core knowledge-knowledge of agents. Mandler doubts that there is any such innate core system of knowledge, and here is where we disagree. I admit that babies could learn generalizations such as entities that move on their own and react contingently to me from a distance. However, those generalizations do not license attribution of goals or intentions, and there is evidence that infants do so, at least by 9 to 12 months of age (e.g., Gergeley et al., 1995; Johnson et al., 1988). The attribution of goals and intentions goes beyond spatio-temporal analysis. This is the role of the dedicated input analyzers of core knowledge, to take spat&temporal data as input and output abstract conceptual representations. 1 take the computations that yield the perception of Michotte causality as a prime example of such a dedicated input analyzer. That we see the motion transferred from one object to the other does not take away from that point, for in actual causality in the world, there is no such thing as motion transfer, and whatever transfer actually occurs happens instantaneously. Seeing the causality is simply the output of a relevant input analyzer. The causality is still the product of mind; it goes beyond the perceptual properties (simultaneity, contact, etc.) that are its necessary and sufficient input. In sum, Mandler and I have deep disagreements on two related points. My hypothesis is that the cores (sense 1) of the young infants’ global kind categories, categories such as animal, vehicle, and furni-

Carey

l

51

ture, embody causal notions, just as the cores of adult kind categories do. The causal notions that are relevant include attentional states, goals, agency, and function. Importantly, such notions transcend perceptual primitives. The perceptual system can analyze contact, simultaneity, contingency, paths, and the transition from rest to motion, but one cannot state causal notions in terms of such primitives. If the infant attributes causality on the basis of such perceptual information, the causal attribution is the product of a computation that outputs it from perceptual input. Mandler ‘s hypothesis is that the cores (sense 1) of the young infants’ global kind categories can be abstracted from perceptual primitives-for example, self-generated motion, irregular motion, and contingent motion. She may be right, but if she is, the distinction between perceptual and conceptual categories is not as great as I believe it to be. If I am right about the nature of kind concepts’ cores (sense l), then they cannot be formed through a process of perceptual abstraction and schematization, the second point on which we disagree. Perhaps Mandler and I could agree on the following: If there is such a thing as core knowledge (sense 2), then this knowledge provides some of the abstract causal primitives that are at the core (sense 1) of domain-level concepts. Mandler doubts that core knowledge really exists, but if it does, it will play a crucial role in her project of-distinguishing conceptual categories from perceptual ones. REFERENCES Ahn, W., Kim, N. S., Lassaline, M. E., & Dennis, M. J. (2000). Causal status as a

determinant of feature centrality. Cognitive Psychology, 361-416. Carey, S. (1985). Conceptual change in childhood. Cambridge, MA: MIT Press. and Development, Carey, S. (2000). The origin of concepts. Journal ofcognition 1, 37-42.

Carey, S., & Spelke, E. (1994). Domain specific knowledge and conceptual

change. In L. Hirschfeld & S. Gelman (Eds.), Mapping the mind: Domain speciin cognition and culture (pp. 169-200). Cambridge, England: Cambridge University Press. Carey, S., & Spelke, E. (1996) Science and core knowledge. Journal OfPhilosophy of Science, 63, 5 15-5 3 3. Gergeley, G., Nadasdy, Z., Csibra, G., & Biro, S. (1995). Taking the intentional stance at 12 months of age. Cognition, 56, 165-193. Gopnik, A., & Meltzoff, A. N. (1997). Words, thoughts and theories. Cambridge, MA: MIT Press. Johnson, S. C., Slaughter, V., &ICarey, S. (1998). Whose gaze would infants follow: The elicitation of gaze following in 12-month-olds. Developmental Scificity

ence, 1, 233-238.

Keil, F. C. (1989). Concepts, kinds, and cognitive development. Cambridge, MA: MIT Press. Kripke, S. (1972). Naming and necessity. Cambridge, MA: Harvard University Press.

52

l

3.

Origin of Concepts

Leslie, A. (1988). The necessity of illusion: Perception and thought in infancy. In L. Weiscrantz (Ed.), Thought without language (pp. 185-210). Oxford, England: Clarendon. Mandler, J. M. (2000a). Perceptual and conceptual processes in infancy. Journal of Cognition and Development, I, 3-36. Mandler, J. M. (2OOOb).Replies to the commentaries on perceptual and conceptual processes in infancy. Journal of Cognition and Development, I, 67-79. Matan, A., & Carey, S. (2001). Developmental changes within the core of artifact concepts. Cognition, 78, l-26. Medin, D., & Ortony, A. (1989). Psychological essentialism. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 179-195). New York: Cambridge University Press. Miller, G., & Johnson-Laird, I? (1976). Language and perception. Cambridge, MA: Harvard University Press. Salmon, W. C. (1989). Four decades ofscientific exploration. Minneapolis: University of Minnesota Press. Schwartz, S. I? (Ed.). (1977). Naming, necessity and naturazkinds. Ithaca, NY: Cornell University Press. Spelke, E. S., Breinlinger, K., Macomber, J., & Jacobson, K. (1992). Origins of knowledge. Psychological Review, 99, 605-632. Spelke, E. S., Phillips, A., & Woodward, A. L. (1995). Infants’ knowledge of object motion and human action. In D. Sperber, D. Premack, & A. J. Premack (Eds.), Causal cognition: A multidisciplinary debate (pp. 44-78). Oxford, England: Clarendon. Van de Walle, G. (1999, April). Nine month olds show manual habituation to shared object color. Paper presented at the Society for Research in Child Development, Albuquerque, NM.

Scripts, Schemas, and Memory of Trauma Robyn Fivush

Emory University

A

long with growing awareness of the stressful and often traumatic events that many children experience, there has been increasing research attention on how children remember such events. Much of this research stems from forensic considerations, and has focused on the accuracy and suggestibility of children’s testimony (e.g., Ceci & Bruck, 1993; Goodman & Bottoms, 1993). Yet given the concerted research efforts in this area, we still know surprisingly little about the developmental course of children’s memories for stressful and traumatic events. In this chapter I examine theory and data on children’s developing event memory to provide a model for understanding memory for traumatic events. In particular, I argue that the dimensions along which children’s representations of everyday events develop are useful for understanding how more stressful and traumatic events may be represented. Thus, in constructing models of trauma memory, we must consider both the structure of the event in the world and children’s developing representational abilities.

54

l

4.

Memory

of Trauma

EVENT SCHEMAS Pioneering research in the late 1970s fundamentally changed our conceptualizations of early memory. Whereas early research focused on the deficiencies of children’s abilities to organize and integrate presented material (e.g., Kail & Bisanz, 1982), new ideas about how information may be organized in memory led to research demonstrating early competencies. In particular, when children were asked to recall familiar and meaningful events, they demonstrated remarkably well-developed organizational and mnemonic abilities. For example, Mandler (19 78,1983) demonstrated that young children were able to recall presented stories coherently and Nelson and her colleagues (Nelson, 1986; Nelson, Fivush, Hudson, & Lucariello, 1983; Nelson & Greundel, 1979) described young children’s abilities to generate organized and complex reports of routine and familiar events. This research both stemmed from and contributed to growing theoretical formulations of scripts and schemas as abstract constructs that organize the encoding, storage, and retrieval of information. Briefly described, schema theory specifies a hierarchical organizational framework for comprehension and prediction (Bobrow & Norman, 1975). Story and event schemas specify general categories and chronological order in which information about events will be encountered. “Once upon a time” sets the stage such that the listener knows what information will come next: characters will be introduced and they will encounter some problem which they will attempt to solve; if successful, the story will come to a conclusion, if not, another problem solution attempt will be made. These expectations drive both comprehension and recall, and even preschool children will include information from each of the major categories, or nodes, when asked to recall simple stories (Stein & Glenn, 1982). Similarly, schemas, or scripts, for routinely encountered real world events, specify the actors, actions, and objects most and least likely to be present (Nelson, 1986; Nelson & Greundel, 1979; Schank & Abelson, 1977), as the now classic example of the restaurant script illustrates: one knows that there will be tables and chairs, a person who serves food, a menu, and so forth. In addition, the sequence in which core actions will occur is also specified; first one is seated, then one orders, then eats, then pays. Again, even preschoolers will report familiar events in a temporally organized fashion, and include information about core actions and objects (see Fivush, 1997, and Nelson, 1986, for reviews). Thus, event schemas and scripts are powerful tools for organizing information about the world around us, for anticipating and predicting what will happen, and for providing frameworks for understanding events as they unfold.

Fivush

l

55

Over the past 20 years, research generated from script and schema theory has provided us with a rich understanding of children’s developing representations of real world events. Even in the first year of life, infants are able to build up expected sequences of actions, which allow them to anticipate events (Bruner, 1981). However, not surprisingly, there are systematic developmental advances in children’s abilities that influence event comprehension and representation. Further, the structure of events in the world greatly influences how children come to understand those events. EVENT STRUCTURE Some events in the world follow a tight, enabling temporal order. The restaurant script, as described earlier, is a good example. One must get a menu before one orders, one must order before food can be served, food must be served before one can eat, and so on. In contrast, other events are temporally flexible. Although you almost always have a cake at a birthday party, you can eat it before or after you play games, or open the presents. How might the temporal structure of events in the world affect children’s representations of those events? Do children understand the enabling constraints among actions or do they simply represent the temporal order in which actions occur? It is possible that children, especially very young children, do not distinguish between actions that must occur in a particular order due to enabling constraints versus actions that are invariant in temporal order but only due to convention, such as eating salad before the main course in American restaurants. A substantial body of research now confirms that young children are sensitive to the enabling connections among actions in an event sequence. In particular, Bauer and her colleagues (see Bauer, 1996,199 7, for reviews) demonstrated that l- and a-year old children will imitate action sequences in the order in which they were presented. However, actions linked by enabling connections are significantly more likely to be imitated in their presented order than are actions only linked temporally. Further, enabling connections between actions lead to better long-term recall than simple temporal invariance. These results clearly indicate that children are perceiving enabling connections between actions and are using these connections to organize their recall. Indeed, when presented with long and complex sequences, in which some actions are linked by enabling connections and others linked only by temporal connections, children learn the actions connected by enabling relations before they learn actions linked by simple temporal sequence (Bauer & Fivush, 1992). It seems that children use the enabling connections to anchor their developing representation of a complex event.

56

l

4.

Memory

of Trauma

In a study directly comparing children’s developing representations of events linked by enabling connections to temporally invariant but arbitrarily connected events, Fivush, Kuebli, and Clubb (1992) asked 3and S-year-old children to engage in two sets of play activities. One activity was making edible fundough, in which each action had to occur in the specified order, and the other activity was sand play, in which the same actions occurred in an invariant temporal order, but there were no necessary connections among the actions. One group of children experienced the events just once and was then asked to reenact the sequences. At both ages, the enabling connected event was reenacted in its presented order, and the temporally invariant but arbitrary sequence was not. More telling, another group of children experienced the events four times before being asked to reenact them. Here, the children experienced the sand play event in exactly the same order every time they encountered it, yet there were still differences in their final reenactments. Again, at both ages, the enabling connected event was reenacted in its presented order, but the temporally invariant arbitrary activity was not, indicating that experience with an invariant temporal order in and of itself does not necessarily lead to a temporally organized representation. Similarly, Bauer and Travis (1993) assessed the memories of 24month-olds for both laboratory-constructed events and for real-world events that were either connected by enabling relations or were temporally invariant but arbitrary. For both types of events, even after considerable experience, children showed better organized recall of events linked by enabling relations than invariant but arbitrary relations. Clearly, young children are sensitive to the underlying causal structure of events and use this structure to organize their developing event representations. EVENT VARIABILITY In addition to temporal structure, events also differ in the variability of the component activities. Some events, such as going to McDonald’s, include virtually the same activities each and every time they are experienced. Although the actual food eaten may vary (chicken nuggets vs. a hamburger) the other activities (and even the decor across different McDonald’s restaurants) are more or less the same. Other events contain a great deal more variability. For example, at preschool, one may always read a story at snack time, but the story read will vary from day to day, as will the snack eaten. Similarly, one may always play a game after nap, but the game played will change daily. How might children represent this kind of variability? There are

Fivush

l

57

at least two issues embedded in this question. One, how do children represent variability during the first few experiences with an event, and, second, how do children represent variability once an event representation is firmly established? To answer the first question, we need to understand how children represent an event on the very first experience. Do children have a loosely connected set of actions, people, and objects that were encountered with little organizational framework? Perhaps surprisingly, it appears not; even after a single experience with a relatively novel event, children seem to form an organized representation that specifies what actions and objects will be encountered at particular points as the event unfolds in time. For example, Fivush (1984) examined children’s scripts for the school day routine after the first day of kindergarten. Children report the event in a coherent, accurate, temporal order, specifying which activities occurred when. Moreover, children reported the daily activities at a generalized level, reporting “we play” rather than specifying the specific games played, or “we read a story” rather than specifying “we read ‘Curious George Goes to School.“’ This level of report suggests that children are forming more generalized expectations of the kinds of activities that are likely to occur during future occurrences of the event rather than a set of specific activities that occurred in the past. In related research, Ratner, Smith, and Dion (1986; see also Hudson 1990) found that 5-year-olds gave temporally organized verbal reports of a structured play activity after the first experience. And Bauer (1996, 199 7) demonstrated in a series of studies that even toddlers have a temporally organized representation of an event experienced only once, especially if that event follows a logical order. Obviously, children’s event representations continue to develop with increasing experience with similar events, as I will describe later, but importantly, children form an organized representation of an event the first time it is experienced that allows them to anticipate and predict future occurrences. Given this finding, how do children represent events after the first few experiences during which some of the component elements change? Whereas the temporal structure of the event seems to influence children’s representations across a wide developmental age span in similar ways, the variability of an event affects children’s developing representations differently at different ages. More specifically, preschool children seem to be more adversely affected by variability than are older children. In the study by Fivush et al. (1992) described earlier, children were also asked to engage in an event that varied across each specific experience, Although children always made a shape collage, the shape and the materials used changed from experi-

58

l

4.

Memory

of Trauma

ence to experience. After a single experience, children were asked to reenact the event with a novel shape and materials to glue. Fiveyear-olds had no difficulty generalizing their knowledge from the first experience to this second experience with different elements, but 3-year-olds were quite discombobulated. They had difficulty reenacting the event and kept asking the experimenter what these new objects were for. After four experiences, in which the elements varied each time, the 3-year-olds were able to generalize to yet another set of elements, suggesting that younger children need more experience with how events can vary from time to time to build this variability into their representations. Bauer and Fivush (1992) examined this issue more closely by asking 30-month-olds to reenact a series of event sequences after each experience. Children were presented with sequences once and asked to reenact them, and 1 to 3 days later were presented with these same sequences in which some elements varied and asked to reenact them again. This procedure was repeated for a total of four exposures to events that varied somewhat at each exposure. Children had no difficulty reenacting the sequences after the first experience, but they were somewhat disrupted by the second experience in which some elements varied. By the third exposure in which some elements varied, they were again able to reenact the sequences quite well. Thus, it seems that it is in the initial stages of developing event representations that young children may have some difficulty generalizing. This interpretation is supported by Kuebli and Fivush (1994). They asked 4- and 7-year-old children to engage in activities four times in which some elements varied at every experience and some elements varied only at the third experience, with the other three experiences being identical. After all four experiences, children were asked for both free recall and to respond to specific prompts about the variability. Again, 7-year-olds were able to incorporate both recurring variability and the single instance of variability into their representations, but 4-year-olds had more difficulty. When elements changed at every experience, they integrated this into their representation, but when variability occurred only once, they did not free recall this; in fact, they denied any change had occurred even when directly prompted. These kinds of results indicate that, although variability must surely be incorporated into developing event representations, it may be more difficult for younger children to understand and represent this variability than older children over the course of the first few experiences with an event. However, what happens as the event becomes routine and children establish a stable representation? How are variations from an expected schema represented and recalled?

Fivush

.

59

SCRIPTS AND EPISODES Recalling a specific instance of a highly scripted event is difficult even for adults (Graesser, Woll, Kowalski, & Smith, 1980). Think about what you had for lunch last Tuesday. Unless it varied significantly from your normal routine, this is an extremely difficult piece of information to access. Essentially, routine occurrences become absorbed into the script of “what usually happens.” Thus, specific experiences become difficult to recall unless they vary in some significant way from the usual and can be “tagged” in memory. Children, like adults, have difficulty recalling specific actions or objects embedded in routine events unless specifically cued. For example, Fivush (1984) asked kindergarten children what they did yesterday at school, on the 2nd day of school, during the 2nd week, the 4th week, and the 10th week. Children had a great deal of difficulty responding to this question. Many children simply reported they did the same things as they did everyday. Even when asked, “What book did you read at snack yesterday. 7” most children could not differentiate which book was read which day, but when specifically cued with the title of the book, all children could give detailed reports of what the book was about. So it is not that this information is lost in memory, but rather that it is difficult to access using the daily routine as a retrieval guide. In related work, Hudson (1990) studied children’s memories of a creative movement workshop. Some children experienced a single workshop, whereas others experienced four different workshops. When asked to recall the first workshop, children who had only this one experience did quite well. But children experiencing all four workshops had difficulty recalling what happened the first time. Similarly, Farrar and Goodman (1990) asked 4- and 7-year olds to verbally recall events experienced once, twice when they varied from the first to the second experience in some of the elements, or four times when all experiences were identical except for the second experience which varied in some of the elements. Again, after a single experience, the reports were accurate and organized. But after two experiences that varied slightly, 4-year-olds recalled less and made more confusions between the two events. The ‘I-year-olds seemed able to hold the two instances in memory separately. After four experiences, the younger children had a well-developed script for the event, but when asked to recall the one episode that varied from the recurring script, they had great difficulty. Seven-year-olds were substantially better at being able to recall the one episode that deviated from the other three experiences. These results are similar to results presented earlier on children’s abilities to incorporate variable elements into their scripts. Whether children are asked to recall

60

l

4.

Memory

of Trauma

a single episode of a recurring event, or to recall the ways in which specific elements of an event can vary across experiences, young children seem to have difficulty integrating how things change with how they remain the same. However, with increasing experience with a recurring event, even young children begin to represent how that experience may be the same and different across episodes. In the study of kindergarten children’s representations of the school day already discussed (Fivush, 1984), children were asked to report what happens at school four times across the first 3 months of the school year. As already mentioned, even on the 2nd day, children reported the event as a general temporally organized frame that provided children with a predictive schema for what would happen on any given day of school. But this representation did change with increasing experience with the school day event. Children reported more of the component elements with increasing experience, but at the same time the reports also became more schematic. Essentially, children boiled down the school day routine to its bare elements and reported them with little detail. For example, one child reported, “I put my things away. Play. And then I turn over my name (this is how the teacher takes attendance). And then I go to an activity until the teacher comes in. Play, meeting, reading jobs, meeting again, math jobs, snack. Then, after, then minigym. And then lunch. And then we have, and then we go home.” Moreover, children began to incorporate more optional and conditional activities into their reports, including information about actions or actors that might be a part of the daily routine (e.g., “If it’s Friday, we have sharing with snack” or “Sometimes we have science after lunch.“). Increasing inclusion of optional and conditional activities indicates that children are building a more complex representation with increasing experience, including information about what almost always happens, what might happen and the conditions under which certain activities are more or less likely to happen. Overall, then, there are both developmental continuities and differences in children’s event representations. Across a wide developmental age span, children represent an event in a generalized temporally organized schema from the first experience, and actions linked by enabling relations remain better organized and recalled than do temporally arbitrary relations. However, younger children seem to have more difficulty than older children in incorporating the variable aspects of events into their representations, especially during the first few experiences. Younger children are more likely to omit actions and objects that vary across occurrences from their representation, whereas older children seem better able to integrate variable actions. However, with increasing experience with a recurring event, even young children begin to repre-

Fivush

l

61

sent the event in a more conditional and yet also more schematic framework. These conclusions raise an interesting paradox. Although we might expect repetition of an event to facilitate memory, in these cases, repetition seems to hinder memory, at least for reporting specific episodes or elements that vary across experiences. This raises the more general question of how children recall a specific one-time event as compared to a recurring event. In a study designed to address this question, Hudson and Nelson (1986) asked 3-, 5-, and 7-year-old children to report both “what happens” at dinner at home or snack at day camp, or “what happened one time” for these same events. Children’s reports of what happens in general were longer and better organized than their reports of what happened one time. Again, specific occurrences of routine events are difficult to access and recall, but the general script is well-established and articulated. Yet, when Hudson and Nelson then asked these same children to report about a distinctive event that the children had only experienced once, such as a trip to Disneyland, children at all ages were able to give long detailed reports. Thus, although a specific occurrence of a routine event is quite difficult to recall, a distinctive one-time event is not. Indeed, there is now a great deal of research demonstrating that children as young as 2% to 3 years of age are able to give accurate, detailed accounts of events experienced only once (Fivush, Gray, & Fromhoff, 1987; Hamond & Fivush, 1990; Todd & Perlmutter, 1980; see Fivush, 1993, 1997, for reviews). Moreover, events from these early years are retained over very long durations; if the event remains distinctive in the child’s experience, they seem to be able to recall it even after 5 to 7 years have passed (Fivush & Schwarzmueller, 1998; Hamond & Fivush, 1990; Hudson & Fivush, 1991; Pillemer, Picariello, & Pruett, 1994). Thus it is clear that distinctive, highly salient events are well-recalled even very early in development and remain accessible for vivid and detailed recall over periods of many years. The overall conclusions, then, from the research on children’s developing event memories, indicate that even quite young children are able to recall the events of their lives accurately and coherently. From a very early age, children are sensitive to the underlying temporal structure linking actions into event sequences and their event representations seem to be anchored by actions that are linked by enabling connections. With increasing experience with events, children’s representations include more of the component actions, but also include less detail about these actions; event representations become more schematic. Variability within recurring events seems to disrupt forming a stable representation over the first few experiences for very young children, but with increasing experience, variability seems to be incorporated and even ab-

62

l

4.

Memory

of Trauma

sorbed into the more generic representation of “what usually happens.” At the same time, novel events, which remain distinctive in the child’s experience, are retained in quite vivid detail even over many years. This basic framework for understanding the development of children’s event memory has clear implications for memories of more stressful and traumatic events. MEMORY

OF TRAUMA

There is a longstanding controversy in the literature on trauma memory over whether high levels of stress facilitate or hinder memory. Within the clinical literature, there are descriptions of how survivors remember and report horrific experiences such as battle, rape, and severe physical and sexual abuse (see McCann & Pearlman, 1990, and Reviere, 1996, for reviews). These studies provide a great deal of descriptive information about both remembering and forgetting of trauma. Paradoxically, there is evidence in the literature for vivid flashbacks and the sense of reexperiencing the original event, as well as for traumatic amnesia, and the seemingly total loss of memory for these events. The problem, of course, is that it is impossible to do controlled experiments in which some participants are randomly exposed to extreme stress whereas others are not. When this issue is brought into the psychological laboratory, the “traumatic” event is often quite mundane compared to the events described in the clinical literature. How best to resolve this issue is still an open question but several things need to be considered. First, we need to make distinctions among trauma, stress, and arousal. In most of the laboratory-based studies, adult participants are exposed to some stimuli, such as pictures or video, depicting physically disturbing events, such as gory accidents or surgery. Obviously, the participants know they are in a psychology experiment, that they are safe in the confines of a laboratory setting, and that the pictures are not “real.” The stimuli may certainly invoke high arousal (and manipulation checks indicate that they do), but these events cannot be said to be traumatic. The participants are never in fear of their own well-being or that of others. Keeping this in mind, reviews of this literature with adult participants concur that stress, or arousal, seems to facilitate memory, at least for central aspects of the event, although it may hinder memory for more peripheral aspects (Christianson, 1992; Heuer & Reisberg, 1990). One of the prevalent models of stress and memory suggest an inverted u-shape function (Easterbrook, 1959), such that moderate to high levels of stress increase memory to a point, and then as stress becomes more intense, memory is hindered. Although laboratory studies of

Fivush

l

63

stress and memory have not found much support for this model, it must be noted again that these studies do not assess memory effects at the extreme levels of stress. Research with children has followed a somewhat different course. Because of the demands of working with young children, and the ethical considerations of causing them even minor levels of stress in the laboratory, many researchers have turned to real-world stressful events that many children must unfortunately endure, such as painful medical procedures (e.g., Goodman, Quas, Batterman-Faunce, Riddlesberger, & Kuhn, 1994; Ornstein, 1995; Steward, 1993). Because of the continuing controversy over young children’s credibility as witnesses, much of this research has focused on the issue of children’s susceptibility to suggestion during interviewing rather than on examining the development of children’s memories for these types of events per se. TWO tactics are taken in this research. In one, children experiencing a more stressful procedure, such as an inoculation, are compared to children experiencing a less stressful procedure, such as simply having one’s arm rubbed. In a second tactic, children who are all experiencing the same stressful procedure, such as a voiding cystourethrogram (VCUG), which involves catheterization, filling the bladder with fluid, and then voiding on the table, are assessed for their individual levels of stress and this is related to their recall. The problem is that various measures of stress have been used, including doctor’s reports of the child’s stress, parental report of child stress, child’s self-report of stress, and physiological measures such as heart rate and cortisol levels. Not only are few of these measures related to each other, which is a problem in and of itself, but few of these measures are systematically related to children’s recall. Although some relations are found in some studies, no clear pattern across studies has emerged. Moreover, little research has directly compared the same children recalling a stressful event and a more positive event. Thus, from the research conducted to date, it is very difficult to ascertain if there is any systematic relation between stress and memory. Although two recent reviews of this literature concur that, overall, stressful and traumatic events are recalled at least as well if not better than everyday events (Fivush, 1998; Pezdek & Taylor, in press), a great deal more research needs to be done before any firm conclusions can be drawn. TRAUMA

AND WENT SCHEMAS

Although there has been a steadily increasing number of studies examining children’s memories of stressful and traumatic events, there has been little progress in our understanding of how and why specific trau-

64

l

4.

Memory

of Trauma

mas may or may not be well-remembered. To develop models of trauma memory, we must look to more general models of memory development. In the first part of this chapter, I outlined the current state of knowledge about children’s developing event representations; I now turn to specific predictions about children’s representations and recall of traumatic events based on an event representation framework. Memories

of Specific

Traumatic

Experiences

Similar to memories of distinctive positive events, children who experience a single traumatic event are likely to recall that event in vivid detail even over long periods of time, and this will hold true across development. Even children as young as 3 years of age will be able to report a distinctive traumatic event accurately and in great detail, and will retain this memory over many years. Indeed, there seems to be a great deal of evidence for this prediction. The clinical literature is rife with descriptions of children recalling distinctive horrific experiences such as witnessing a parental homicide (Malmquist, 1986), a sniper attack at school (Pynoos & Nader, 1989), and a kidnapping (Terr, 1983). However, it must also be emphasized that although these memories remain vivid and highly accurate, there is also evidence of some reconstructive detail over time. For example, in recalling a school sniper attack, some children placed themselves further away from the spray of bullets and others placed themselves closer to actual danger, possibly to regulate their emotional distance from the event. Furthermore, there is some suggestion that very young children may create a fantasy resolution to a traumatic event. Pynoos, Steinberg, and Aronson (1997) reported that some preschool children who had witnessed severe domestic violence falsely reported how they intervened to save the attacked parent. Thus, although single experiences of trauma seem to be well-recalled, as are single experiences of more positive distinctive events, the emotional needs of the child experiencing trauma may influence memory in ways that are different from more positive experiences. This is an important consideration which should be pursued more systematically in future research. In more controlled studies of children’s memories of stressful experiences, there is again good evidence of detailed and long-lasting memories. For example, Peterson and Rideout (1997) reported that 3- to 1 l-year-old children retain detailed and accurate memories of an injury resulting in emergency room treatment up to 2 years after the event occurred. Surprisingly, as preschoolers grew older, they were actually able to verbally report more about the event, and the additional material was accurate according to reports taken from adult witnesses

Fivus

h

l

65

at the time of the injury. (It should be noted that children younger than about 2% at time of injury also increased in their verbal recall over time, but their reports were quite error-prone.) Similarly, Fivush, Sales, Goldberg, Bahrick, and Parker (2001) interviewed 3- and 4year-old children about Hurricane Andrew, a devastating storm that hit the Florida coast in 1992, both soon after the event and again 6 years later when the children were 9 and 10 years old. Even after this long delay, children recalled the event in vivid detail and verbally recalled more than twice as much information as they had when they were younger. Intriguingly, children experiencing the highest level of stress, with their homes literally being blown apart while they were inside during the storm, recalled less information in free recall than children in a moderate or low stress group. This suggests that at extremely high levels of stress, retrieval may be hindered, but this must be understood in the context of extremely high recall overall for all three stress groups. Further, in these studies there was little evidence of distortion or reconstruction in the service of emotional regulation, so it may only be in the context of therapeutic interviews or interventions that children begin to show these kinds of memory biases. Moreover, memory of a single stressful event must be placed in the larger context of the child’s world. For example, children growing up in inner city neighborhoods are often exposed to various forms of community violence (Richters & Martinez, 1993). In these children’s lives, any one traumatic experience is placed against a backdrop of stressful events, as compared to children growing up in relatively tranquil environments in which a single traumatic event may stand in stark contrast to their daily lives. Children growing up with chronic stress and violence may develop more overarching scripts about the world in which stressful and traumatic events may become the expectation rather than a violation of norms. We know very little about how world views may effect children’s developing representations of specific events. In a study directly comparing children’s memories for a stressful and a positive event among children growing up in violent communities (Fivush, Hazzard, Sales, Sarfati, & Brown, in press), children recalled about equal amounts of information overall about these two kinds of emotional experiences but they recalled different kinds of information. When narrating positive events, children focused more on the actions, objects, and people present than when recalling negative experiences, but when narrating negative experiences, children reported more information about their emotions and internal states than when recalling positive events. Negative experiences were also reported more coherently than positive events. Thus, it may be that negative events lead to a more integrated, internal focus on one’s thoughts and

66

’

4.

Memory

of Trauma

feelings than do positive events. However, it is not clear at this point if this is an effect of experiencing chronic violence or if any negative event would result in a more internal focus. Further, the ways in which adults talk with young children about negative experiences will almost certainly have a profound impact on how children come to represent those experiences. A great deal of research on the development of autobiographical memory has demonstrated that children are learning the forms and functions of talking about the past through participating in adult-guided conversations (Fivush, 1991; Fivush, Haden, & Reese, 1996; McCabe & Peterson, 1991; Nelson, 1996; Reese, Haden, & Fivush, 1993). Two recent studies are beginning to extend this finding to stressful events. Sales, Fivush, and Peterson (2001) found that mothers talked more about the causes of a stressful event, an injury requiring emergency room treatment, than causes of a more positive event, but, surprisingly, talked more about the emotions associated with more positive events. Ackil, Waters, Dropik, Dunisch, and Bauer (1999) found that mothers and children talked more overall about a tornado than about a more positive event, and also talked about the stressful event in more coherent ways. These results suggest that parents may be more focused on helping their children understand why negative events happen and to help them construct a more coherent account of negative than of positive events, and this may partly explain differences in how children themselves recall stressful versus positive events. Overall, then, there is good evidence that single experiences of trauma result in highly detailed enduring representations, but there may also be ways in which traumatic experiences are represented differently than distinctive positive experiences. Through reconstructing some details, providing fantasy resolutions, and by focusing on internal states and emotions, children may focus on the emotional aspects of traumatic experiences and their ability to regulate their negative emotions more so than when recalling emotionally positive distinctive events. Recurring

Trauma

Memories of recurring traumatic events, such as chronic abuse, will be different from memories of single experiences of trauma. Specifically, as children experience recurring episodes of a similar trauma, their representations should become more schematic and less detailed. Memories of recurring trauma should take on a more “script-like” form, with children reporting what “usually happens” and providing information about the actions, objects, and people most likely to occur, but providing little detail about any one specific episode.

Fivush

l

67

Intriguingly, within the clinical literature, Terr (199 1) also proposed different memory outcomes for single and repeated trauma. Type I trauma, which is a single distinctive experience, is expected to lead to vivid and enduring memories, whereas Type II trauma, which is chronic trauma, is expected to lead to more fragmentary memories. However, the mechanisms underlying these differences, according to Terr, is not memory processing per se but coping strategies. Because Type I trauma is a single unexpected incident, it leads to hypervigilance, a focusing of attention on what is occurring, and a focus on the ability to predict possible future occurrences. Thus, attention and processing are high and subsequent memory is quite good. In contrast, traumatic experiences which recur regularly, and which the child cannot control, lead to dissociation during the event as a way of coping. Thus, attention and processing are low and subsequent memory is poor. However, Terr’s approach would suggest that memory of recurring trauma would be fragmentary and perhaps disorganized, whereas an event representation model predicts a highly schematic but very well-organized representation of repeated trauma. Almost no research has examined how children’s memories of a recurring stressful event compares to memory of a single stressful experience of the same type. Goodman, Hirschman, Hepps, and Rudy (199 1) asked children to report about a specific experience of a stressful medical procedure, a VCUG. Although some of the children had experienced previous VCUGs, there were no differences in the accuracy of children’s memory reports as a function of previous experiences. However, the researchers did not assess organization or level of specificity of the report. Similarly, Howard, Osborne, and Baker-Ward ( 199 7) asked child cancer survivors to recall their chemotherapy experiences a year after treatment ended and found that children were able to report quite accurately on these experiences. However, again, the authors did not analyze the type of recall; did children report details of specific experiences or did they organize their reports around what usually happens? Thus, the evidence suggests that single occurrences of a traumatic event are recalled as accurately as multiple occurrences, but the organization and level of specificity of children’s memory reports as a function of number of experiences has not yet been examined. Further, there has been no examination of how children might recall a specific episode of a repeated traumatic event as compared to a general recounting of what usually occurs. It may be the case that children exposed to a recurring traumatic event are less able to describe the specifics of any one experience than children exposed only once. Finally, no one has examined how children might represent a recurring traumatic event over the course of the first few experiences. Ifchil-

68

0

4.

Memory

of Trauma

dren have only had three or four experiences with a specific stressor, they may not have yet developed a fully schematic script, but may still be building their representation of the event and its variability. As reviewed earlier, the prediction would be that preschoolers would have an especially difficult time representing what can vary over the first few experiences, and thus may omit variable actions or objects from their accounts. Further, preschoolers may confuse specific episodes of a recurring event during the first few experiences. Thus, it is likely that, especially for preschool children, reports of traumatic events that have recurred a few times would be less detailed, and more confused than reports of either single occurrences of a trauma or a well-established script of a traumatic event. No research has yet examined how reports of traumatic events change as a function of increasing experience, which is clearly a critical question for understanding trauma memory. The Structure

of the Event

Both single and recurring events which are linked by enabling relations will be very well organized even after the first experience and will continue to be represented in a tight temporal framework, and this will be the case even for very young children. However, events that follow a more variable temporal order, in which actions may occur at different points in the sequence during different occurrences of the event, will be less organized in recall, especially by younger children. In reviewing research of children’s memories of a VCUG, Ornstein (1995) pointed out that children who experienced this painful medical procedure just once recalled it very well, and in fact better than other children reporting a specific well-child visit to the doctor. He argued that it may be because the VCUG procedure follows a causal temporal structure: the child must strip before catheterization, catheterization must occur before filling the bladder with fluid, which must occur before the sonogram picture is taken, which must occur before the child voids the fluid, and so on. In contrast, many actions must occur during a well-child doctor visit (checking heart and lungs, ears, eyes and nose, reflexes, etc.), but the order in which these activities occur is not dictated by any clear enabling relations. Perhaps this is why children recall the VCUG better than the well-child doctor visit. Although this is suggestive, there are alternative explanations for this finding. Because the VCUG is also a single distinctive experience, whereas the well-child doctor visit is a repeated event, it would be predicted that the VCUG would be recalled in more detail than the schematized doctor visit. Also, stress levels would be higher for the VCUG procedure than for the well-child doctor visit, and as alluded to earlier, there is some suggestion that

Fivush

l

69

highly stressful events are recalled better than less stressful events. What is needed is research that compares an equally stressful event that follows an enabling order of actions to one that is more temporally variable to determine whether this factor influences trauma memories. Given the large and robust literature on effects of event structure on young children’s event representations, it is highly probable that this factor will play a role in trauma memories as well. It may also be the case that if the enabling structure of the event is highlighted for children, they may be better able to understand and thus recall the event. Principe (1996) compared children who received a VCUG with a technician who spontaneously explained each step of the procedure to the child with children who received the procedure without this explanatory framework. Children in the first group recalled the procedure more exhaustively and more accurately than children in the second group. Thus, it may not only be the structure of the event in the world, but also the way in which adults help children to understand the structure that will influence the subsequent memory. Just as adult-provided structure in reminiscing about events after they occur may help children form more coherent representations, there is also growing evidence that the ways in which adults verbally structure an event as it is occurring influences children’s recall as well (Haden, Didow, Ornstein, & Eckerman, 1997; Pipe, Dean, Canning, & Murachver, 1996; Tessler & Nelson, 1994). Thus, it is not surprising that in a stressful situation, in which children may be less able to focus their attention, the ways in which adults help them to organize the event would be important. CONCLUSION Applying knowledge of children’s developing event representations appears to be a fruitful model for studying trauma memory. Certainly, an event representation framework points to important dimensions along which we might expect trauma memory to vary, and makes specific testable predictions. However, in trying to integrate the research on developing event representations with trauma memory, several additional factors have emerged which must be considered. First is the role of emotional regulation. Although highly positive events are certainly emotional, there is a critical difference between emotions experienced during positive and negative events. In particular, when experiencing trauma, one must cope with high levels of fear and possibly anger, both as the event is occurring and when subsequently recalling the event. To cope effectively, the child must somehow resolve the negative emotions associated with stressful events. Thus,

70

4.

Memory

of Trauma

we may see emotional experience represented and recalled differently for traumatic events than for equally emotional but positive events. There is some suggestion in the research reviewed here that this may be the case. Second, representations of specific events need to be placed in the larger social-cultural world of the child. Little research has examined the way in which world views or basic world beliefs influence the formation of specific event schemas. A few studies have begun to examine how attitudes about schooling affect the way in which the school day routine is discussed in different social-cultural groups (e.g., Flannagan, Baker-Ward, & Graham, 1995), and this research indicates that larger cultural belief systems influence what aspects of specific events may be highlighted. Similarly, it seems quite likely that children growing up in malignant environments, in which stressful and traumatic events are the norm, would come to represent specific traumas differently than children growing up in more benign environments. Third, we need to consider the role of language. As discussed briefly, recent research has begun to document the pivotal role that language plays in children’s memories of events. Events that are organized and narrated for children, especially young children, as they are occurring and in retrospect, are better recalled than events that are not verbally scaffolded. The ways in which adults talk with children about stressful and traumatic events surely will play a role in how children come to understand and remember these experiences. This raises the intriguing issue of the silencing of trauma (see Fivush, 1998, for further discussion). Many traumatic experiences, especially shameful and secretive traumas such as abuse, are never talked about, and indeed, even when the child tries to recall them, the occurrence of these events is ignored or denied. An intriguing question is how children come to represent events that may be silenced in this way. Finally, we must understand that traumatic events often involve the severing of social and emotional relationships. When parents are the source of the trauma, as in abuse, or do not protect the child from stress, as in painful medical procedures, the child may be more traumatized by the loss of security and attachment than by the actual physical event (e.g., Freyd, 1996). If adults then do not talk with the child about these aversive experiences in ways that can facilitate understanding and coping, this may serve to further traumatize the child, leading to a belief that the world is not a safe place. Thus, although an event representation framework is a useful heuristic for understanding aspects of children’s memories of stressful and traumatic events, we must always keep in mind that trauma is not merely a type of event in the world; rather, trauma fundamentally changes one’s place in the world.

Fivush

.

71

REFERENCES Ackil, J. K., Waters, J. M., Dropik, I?, Dunisch, D. L., & Bauer, F?J. (1999, April). From the eyes ofthestorm: Mother-child conversations about a devastating tornado. Poster session presented at the biennial meeting of the Society for Research in Child Development, Albuquerque, NM. Bauer, I? (1996). Recalling past events: From infancy to early childhood. Annals of Child Development,

11, 25-71.

Bauer, F?(1997). Development of memory in early childhood. In N. Cowan (Ed.), Thedevelopment ofmemory in childhood (pp. 83-l 12). Sussex, England: Psychology Press. Bauer, I?, & Fivush, R. (1992). Constructing event representations: Building on a foundation of variation and enabling relations. Cognitive Development, 7, 381-401. Bauer, I?, & navis, L. (1993). The fabric of an event: Different sources of temporal invariance differentially affect 24-month olds’ recall. Cognitive Development, 8, 3 19-34 1. Bobrow, D. G., & Norman, D. A. (1975). Some principles of memory schemata. In D. G. Bobrow & A. Collins (Eds.), Representation and understanding (pp. 455-461). New York: Academic. Bruner, J. (1981). The social context of language acquisition. Language and Communication, 1, 155-l 78. Ceci , S. J., & Bruck, M. (1993). Suggestibility of the child witness: A historical review and synthesis. Psychological Bulletin, 113, 403-439. Christianson, S. A. (1992). Emotional stress and eyewitness memory: A critical review. Psychological Bulletin, 112, 284-309. Easterbrook, J. A. (1959). The effect of emotion on utilization and organization of behavior. Psychological Review, 66, 183-201. Farrar, M. J., & Goodman, G. S. (1990). Developmental differences in the relation between scripts and episodic memory: Do they exist? In R. Fivush, & J. Hudson (Eds.), Knowing and remembering in young children (pp. 30-64). Cambridge, England: Cambridge University Press. Fivush, R. (1984). Learning about school: The development of kindergartners’ school scripts. Child Development, 55, 1697-1709. Fivush, R. (1991). The social construction of personal narratives. MerriZZPalmer QllarterZy, 3 7, 59-82. Fivush, R. (1993). Developmental perspectives on autobiographical recall. In G. S. Goodman, & B. L. Bottoms (Eds.), Child victims, child witnesses: Understanding and improving testimony (pp. l-24). New York: Guilford. Fivush, R. (1997). Event memory in childhood. In N. Cowan (Ed.), The development of memory in childhood (pp. 139-162). Sussex, England: Psychology Press. Fivush, R. (1998). Children’s memories for traumatic and non-traumatic events. Development and Psychopathology, 10, 699-716. Fivush, R., Gray, J. T., & Fromhoff, F.A. (1987). ?tvo year olds’ talk about the past. Cognitive Development, 2, 393409. Fivush, R., Haden, C., 8~Reese, E. (1996). Remembering, recounting and reminiscing: The development of autobiographical memory in social context. In D. Rubin (Ed.), Reconstructing our past: An overview ofautobiographical memory (pp. 341-359). New York: Cambridge University Press.

72

l

4.

Memory

of Trauma

Fivush, R., Hazzard, A., Sales, J. M., Sarfati, D., & Brown, T (in press). Creating coherence out of chaos: Children’s narratives of stressful and positive events. Applied Cognitive Psychology. Fivush, R., Kuebli, J., & Clubb, I? A. (1992). The structure of events and event representations: A developmental analysis. Child Development, 63,188-201. Fivush. R., Sales, J. M., Goldberg, A., Bahrick, L., & Parker, J. (2001). Weathering the storm: Children’s long term recall of Hurricane Andrew. Submitted manuscript. Fivush, R., & Schwarzmueller, A. (1998). Children remember childhood: Implications for childhood amnesia. AppZied Cognitive PsychoZogy,12, 455-473. Flannagan, D., Baker-Ward, L., & Graham, L. (1995). Talk about preschool: Patterns of topic discussion and elaboration related to gender and ethnicity. Sex Roles, 32, l-l 5. Freyd, J. (1996). Betrayal trauma. Cambridge, MA: Harvard University Press. Goodman, G., & Bottoms, B. (1993). Child victims, child witnesses: Understanding and improving testimony. New York: Guilford. Goodman, G. S., Hirschman, J. E., Hepps, D., & Rudy, L. (1991). Children’s memory for stressful events. Merrill Palmer Quarterly, 37, 109-158. Goodman, G. S., Quas, J. A., Batterman-Faunce, J. M., Riddlesberger, M. M., & Kuhn, J. (1994). Predictors of accurate and inaccurate memories of traumatic events experienced in childhood. Consciousnessand Cognition, 3, 269-294. Graesser, A. C., Woll, S. B., Kowalski, D. J., & Smith, D. A. (1980). Memory for typical and atypical actions in scripted activities. Journal of Experimental Psychology: Human Learning and Memory, 6, 503-S 15. Haden, C. A., Didow, S. M., Ornstein, I? A., & Eckerman C. 0. (1997, April). Mother-child talk about the here-and now: Linkages to subsequent remembering. In E. Reese(Chair), Adult-child reminiscing: Theory and practice. Symposium paper presented at the meetings of the Society for Research in Child Development, Washington, DC. Hamond, N. R., & Fivush, R. (1990). Memories of Mickey Mouse: Young children recount their trip to Disney World. Cognitive Development, 6,433-448. Heuer, F., & Reisberg, D. (1990). Vivid memories and emotional events: The accuracy of remembered minutiae. Memory and Cognition, 18, 496-506. Howard, A. N., Osborne, H. L., & Baker-Ward, L. (1997, April). Childhood cancer survivors’ memory of their treatment afier long deZays. Paper presented at the meetings of the Society for Research on Child Development, Washington, DC. Hudson, J. A. (1990). Constructive processes in children’s event memory. Developmental Psychology, 2, 180-l 8 7. Hudson, J. A., & Fivush, R. (1991). As time goes by: Sixth graders remember a kindergarten event. Applied Cognitive Psychology, 5, 346-360. Hudson, J. A., & Nelson, K. (1986). Repeated encounters of a similar kind: Effects of familiarity on children’s autobiographic memory. Cognitive Development, 1, 253-271. Kail, R., & Bisanz, J. (1982). Information processing and cognitive development. In H. W. Reese (Ed.), Advances in child development and behavior (Vol. 17, pp. 45-81). New York: Academic. Kuebli, J., & Fivush, R. (1994). Children’s representation and recall of event alternatives. Journal of ExperimentaZ Child Psychology, 58, 25-45.

Fivush

l

73

Malmquist, C. I? (1986). Children who witness parental murder: Post-traumatic aspects. Journal of the American Academy of Child Psychiatry, 25, 320-325. Mandler, J. M. (1978). A code in the node: The use of a story schema in retrieval. Discourse Processes,1, 14-35. Mandler, J. M. (1983). Representation. In J. H. Flavell & E. M. Markman (Eds.), Cognitive development, vol. 3 ofR l-i. Mussen (Ed.), handbook of child psychology (4th ed., pp. 420-494). New York: Wiley. McCabe, A., & Peterson, C. (1991). Getting the story: A longitudinal study of parental styles in eliciting narratives and developing narrative skill. In A. McCabe & C. Peterson (Eds.), Developing narrative structure (pp. 217-253). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. McCann, I. L., & Pearlman, L. A. (1990). Psychological trauma and the adult survivor: Theory therapy and transformation. New York: Brunner/Mazel. McDermott, J. (2000). Parent-child conversations about positive and negative events. Unpublished master’s thesis, Emory University, Atlanta, GA. Nelson, K. (1986). Event knowledge: Structures and function in development. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Nelson, K. (1996). Language in cognitive development. New York: Cambridge University Press. Nelson, K., Fivush, R., Hudson, J., & Lucariello, J. (1983). Scripts and the development of memory. In M. T. H. Chi (Ed.), Contributions to human development, vol. 9: 7Fend.sin memory development research (pp. 52-70). New York: Karger. Nelson, K., & Greundel, J. (1979). Generalized event representations: Basic building blocks of cognitive development. In A. L. Brown & M. E. Lamb (Eds.), Advances in developmental psychology (Vol. 1, pp. 131-158). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Ornstein, I? A. (1995). Children’s long-term retention of salient personal experiences. Journal of’I)aumatic Stress, 8, 581-606. Peterson, C., & Rideout (1997, April). And I was very very crying: Children’s memories of minor medical emergencies. In I? Bauer & M-E. Pipe (Chairs), Long-term memory in childhood. Symposium conducted at the meetings of the Society for Research in Child Development, Washington, DC. Pezdek, K., & Taylor, J. X. X. (in press). Memories of traumatic events. In M. Eisen, G. S. Goodman, & J. S. Quas (Eds.), Memory and suggestibility in the@ rensic interview. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Pillemer, D. B., Picariello, M. L., & Pruett, J. C. (1994). Very long term memories of a salient preschool event. Journal ofAppZied Cognitive Psychology, 8, 95-106. Pipe, M-E., Dean, J., Canning, J., & Murachver, T. (1996, July). Narrating events and telling stories. Paper presented at the second International Conference on Memory, Abano, Italy. Principe, G. (1996, March). Children’s memoryfor a stressful medical procedure. Paper presented at the Conference on Human Development, Birmingham, AL. Pynoos, R. S., & Nader, K. (1989). Children’s memory and proximity to vialence. Journal ofthe American Academy ofChild and Adolescent Psychiatry, 28, 236-241. Pynoos, R. S., Steinberg, A. M., & Aronson, L. (1997). Traumatic experiences: The early organization of memory in school-age children and adolescents. In I? S.

74

4.

Memory

of Trauma

Applebaum, L. A. Uyehara, & M. R. Elin, (Eds.), pauma and memory: Clinical and legal controversies (pp. 272-289). New York: Oxford University Press. Ratner, H. H., Smith, B. S., & Dion, S. A. (1986). Development of memory for events. Journal of Experimental Child Psychology, 41, 41 l-428. Reese, E., Haden, C. A., & Fivush, R. (1993). Mother-child conversations about the past: Relationships of style and memory over time. Cognitive Development, 8, 403-430. Reviere, S. (1996). Memory of childhood trauma. New York: Guilford. Richters, J., & Martinez, I? (1993). The NIMH Community Violence Project: I. Children as victims of and witnesses to violence. Psychiatry, 56, 7-21. Schank, R., & Abelson, A. (1977). Scripts plans goals and understanding. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Stein, N. L., & Glenn, C. G. (1982). Children’s concept of time: The development of a story schema. In W. J. Freidman (Ed.), The developmental psychology of time (pp. 255-282). New York: Academic. Steward, M. (1993). Understanding children’s memories of medical procedures: “He didn’t touch me and it didn’t hurt.” In C. A. Nelson (Ed.), Memory and aflect in development: TheMinnesota Symposium on Child Psychology (Vol. 26, pp. 171-226). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Terr, L. C. (1983). Chowchilla revisited: The effects of psychic trauma four years after a school-bus kidnapping. American Journal of Psychiatry, 140, 1543-1550. Terr, L. (1991). Childhood traumas: An outline and overview. American Journal of Psychiatry, 148, 1O-20. Tessler, M., & Nelson, K. (1994). Making memories: The influence of joint encoding on later recall by young children. Consciousness and Cognition, 3, 307-326. Todd, C., & Perlmutter, M. (1980). Reality recalled by preschool children. In M. Perlmutter (Ed.), Newdirectionsfor child development, no. 20: Children’s memory (pp. 69-86). San Francisco: Jossey-Bass.

On Animates and Other Worldly Things Rachel Gelman

Rutgers university-hbv

I

hwmvick

ean Mandler made fundamental contributions to our understanding of the origin and development of concepts. These include her elegant theoretical and experimental work on scripts and schemas, memory, representation, and infant cognition. Her theoretical articles about “how to make a baby” are classics. Mandler started her developmental work relatively late in her scientific career, having focused initially on animal learning. I had the good fortune to be an undergraduate at the University of Toronto, where Jean Mandler had an animal lab. So, I learned firsthand that, from the beginning, she was pondering the roles of attention and concepts. This was not exactly the “in” thing to do at the time. Behaviorism reigned and talk about mental matters was viewed as unscientific by almost everyone. No matter, Jean always was open to the idea that nonlinguistic or prelinguistic individuals might be able to think. Amazingly, a considerable number of students of early cognitive development still reject this possibility. Journals and meetings are full of efforts to explain away the converging lines of evidence. This is especially puzzling in jean’s case. She did not start out with the idea that infants use general

76

5.

Animates

and Others

categories. Instead, she moved toward it because she allows her data to speak to her and because she is careful to design studies that offer an opportunity to consider different theories. Jean is a scientist’s scientist; she does exquisite, careful, replicable research. If anything, one might argue that Mandler ‘s interpretation of her data is on the conservative side. Not only are the infants and toddlers in the experiments using general categories, they are doing so with representations-small, toy replicas-of real-world objects. Converging evidence from the object-manipulation, sequential-touching, and generalized imitation tasks, show that these very young children place replicas of animals, vehicles, and so forth, into separate categories. This is one reason I am motivated to push Mandler ‘s proposal about the origins of these categories even more on the conceptual side than she has. It is not that I want to rule out her idea that infants use information from different kinds of motion path schema (Mandler, in press). I do so because I am concerned that, by themselves, such schema do not form a sufficient base from which to achieve the target inductions. ANIMATE

VERSUS INANIMATE

OR ANIMAL-NOT

ANIMAL?

Mandler has concluded that infants’ global concept of animal is emAnimal conceptual strucbedded in a domain-general, Animal-not ture and that “the distinction is easily learnable from the spatial and movement information presented by the visual system” (Mandler, in press). I agree that infants use global concepts. However, I prefer to say that the origin and development of the animate-inanimate distinction benefits from skeletal, domain-specific causal structures that apply to objects that are separably moveable as a whole. Causal principles that apply to different categories of separably moveable kinds of things in the world facilitate fast learning about the domain-relevant features, attributes, and predicates about global categories (e.g., Gelman, 1990; Gelman, Durgin, & Kaufman, 1995; Gelman, Spelke, & Meek, 1983; Williams, 2000; Keil, Kim & Greif, in press; Williams & Gelman, 1998). Plants, being attached to the ground, are clearly outside the range of items with which we have been dealing. As it turns out, so too are sentient and seemingly self-moving machines. There is no question that various perceptual characteristics of trajectories, of animate and inanimate objects, can influence decisions about their identity. It is one thing, however, to grant that such perceptual information is relevant, and another to say that it forms the basis of the abstraction of the global, conceptual categories animal and nonanimal. There are reasons for my skepticism about the proposal that motion schemas of different kinds serve this function. First, although the per-

Gelman

l

77

ceptual system responds differentially to biological motion as opposed to nonbiological motion, there is no guarantee that such information is abstracted upward, that is, to the conceptual level. The visual system also responds differentially to color and represents spectral information in a three-dimensional color space. However, the dimensionality of the color space does not organize the everyday conception of colors. Similarly, we can move around in the world without bumping into objects, but we have a hard time accessing the implicit physics that our perceptual-motor system uses (Bransford, Brown, 81 Cocking, 1999; McCloskey, 1983). Second, there is the problem of stimulus indeterminacy, a problem that has contributed significantly to my move to say that the issue of relevance is informed by conceptual as well as perceptual matters (e.g., Subrahmanyam & Gelman, in press; Gelman & Williams, 1998; Williams, 2000). Significantly, motion path information is neither necessary nor sufficient for the identification of animate and inanimate objects (Gelman et al., 1995; Williams, 2000). One way to demonstrate this point is to consider Stewart’s theory (1982, 1984). She proposed that we perceive a novel moving object as inanimate when its motion path is consistent with Newtonian laws of motion. If the motion path violates Newtonian principles, then we perceive animacy and attributions like intentions, desires, hunger, affection, and so forth. In Gelman et al. (1995), we were able to present analyses of Stewart’s own data and our follow-up studies that replicated and extended her findings. Repeatedly, we found that adults alter their interpretations of motion paths on the basis of causal considerations about the details of the motion paths they see to possible causal conditions. CAUSAL INTERPRmATIONS

OF MOTION

To obtain evidence for her theory, Stewart showed college student the trajectories of computer generated dots, each of which moved in ways that were either consistent or not with Newtonian mechanics. Gelman et al.‘s (1995) additional analyses of Stewart’s data, in combination with follow-up studies, yielded a number of results that represented departures from Stewart’s predictions. For example, the perception of a path curving upward on the screen lent itself to a wide range of perspectives and causal interpretations, with some individuals attributing the path to an animate event and others attributing it to an inanimate event. Examples of answers included ‘I.. . a bicyclist going around a corner”; I’... a balloon and the wind was blowing and it went like this, this, like a helium balloon. It got caught up in the air”; ‘A horse climbing a

78

l

5.

Animates

and Others

mountain”; and ‘I.. . some kind of magnetic ball that encountered a field that pushed it away.” The aforementioned descriptions are revealing on another level. They embed within them an account of how the “object” might have moved along the path in question. These “hows” were all about unseen causal factors that could support the way the object moved, including wind blowing, a horse climbing, and a magnetic ball in a field. These kinds of answers provide compelling evidence for our argument that the displays were fodder for causal principles, ones that encouraged interpretation of the motion paths in terms of the possible agents and conditions that generated the trajectories. What was perceived was interpreted with reference to the conditions that can cause the seen trajectory, even if this meant inventing invisible forces, mountains, or magnetic fields. One might hold that the preceding argument applies to adults but not young children, who will not have learned enough to relate their perceptions to causally relevant considerations. If so, young children should be more likely than adults to rely on characteristic perceptual information when presented novel displays. This does not appear to be the case. If anything, 3- to 5-year-olds are more likely to “import” unseen or unheard information about energy sources. They do this when they decide whether a motion of an unidentified object was that of an animate or inanimate object moving in the dark, in a particular setting (Williams, 2000); a novel item shown in a photograph can move itself up and down a hill (Massey & Gelman, 1988); and when assigning predicates in either a picture or verbal task (Gelman & Subrahmanyam, in press). Williams (2000) studied groups of adults and 3- and +year-olds’ use of a sequence of motion paths for an object in a given context. The setting information for each trajectory was comprised of a still (frozen) clip from a video of a scene; for example, a high cliff dropping into a moving stream that had some trees along its banks. On some trials, there also was the sound of running water or a wind. Williams ended the presentation of a given context by rendering the screen completely dark. Then a pinpoint of light appeared somewhere on the screen. Pretest training encouraged people to think that the light was attached to a moving object in a darkened scene. For example, the light on the body of an object seemed to move along the top of the just-seen cliff, drop straight down from the cliff, smoothly or not, in silence or not. It also could be on an object that moved in, or just above the river, silently or not, and so on. After viewing a given motion path, participants were shown pairs of photographs, one each of an animate and an inanimate object (e.g., a duck and a leaf) and asked which one they had just

Gelman

l

79

watched move in the dark. Williams expected participants to choose inanimate objects when the just-seen motion paths were force-consistent with the natural setting they just saw. If the motion path was force-inconsistent for the given setting, he predicted they would choose an animate object. So for example, he predicted that a dot that stopped and started, and even changed direction, as it moved along the silent river scene, would be paired with the duck. Adults were especially likely to pair their choice on picture trials with trajectories that were consistent with the forces present in the just-presented scene. Children’s judgments were more variable due to the youngsters’ tendencies to import force conditions that were not part of the setting display. For example, children who paired the leaf, and not the duck, with the trajectory that just moved in a force-inconsistent manner, “imported” a relevant force. They said that “the wind blowed it around,” even though there was no sound of wind during the stimulus presentation. Their use of motion path information was only part of the information they used to construct a coherent account of the event and related causal sources of energy. The trajectories were an important, but not determining source of relevant information for the assignment of animacy. These were related to an overarching concern for relevant object-kind causal conditions. Thus, both children and adults were especially inclined to give motion explanations that were consistent with their judgments. The formative role of causal principles is also illustrated in studies with photographic stimuli. Massey and Gelman (1988) asked preschool children whether novels item depicted in photographs could move themselves up and down a hill. The photographs were of novel animals (e.g., an echidna, a praying mantis, an invertebrate): wheeled objects, rigid complex inanimates (made up of parts that were leg-like, arm-like, etc.), and statues. Some children said the echidna could go up and down the hill because it had feet, even though none were visible. We also heard that a statue could not engage in such self-initiated actions because it did not have feet, even though it clearly did. A wheeled object might go down the hill, if something pushed it, but not up the hill. These comments reveal an active tendency to selectively attend and even reinterpret aspects of photographs with respect to causal considerations. The children were not simply interested in whether the photographed item had certain parts. These had to be intrinsic to animals. Thus, they either denied that statues had legs or pronounced them “not real,” I’pretend,” and the like. They tried, in their own way, to tell us whether an object was made out of the kind of stuff that goes along with the capacity for self-generated actions. As Williams (2000) put it, they were tuned into whether the causal of action was intrinsic to the

kind of object depicted. Not only did these young children relate motion paths to their causal conditions; they seemed to distinguish between the causal conditions for animate and inanimate objects. Subrahmanyam and Gelman (in press) reached a similar conclusion. One of their two studies involved a lengthy interview in which prewere presented a battery of questions schoolers (4- & 5-year-olds) about 19 (verbally presented) objects. These included animates (person, dog, elephant, and bug), a plant and a rock, simple’ artifacts (spoon, chair), “sentient” machines (W, radio, computer, robot), moving machines (car, airplane, and robot), and so forth. One round of questions paired each item with six separate predicates (moves, talks, breathes, has a brain, thinks, remembers); another round asked about the insides and outsides of the objects; and a final round asked about the origin of each item. Our expectations about the pattern of results were based on a consideration of the implications of the principle-first account of the animate-inanimate distinction. As a reminder, I argued that these principles are domain specific, not only because of different sources of causality, but because they are yoked to categorical different kinds of stuff. The fact that animate objects can cause themselves to move or change and inanimate objects cannot goes hand in hand with the fact that inanimate and animate objects are composed of different kinds of matter. Animate objects are made up of biological matter; inanimate objects are not. Although all objects obey the laws of physics, animate objects also obey biochemical ones. In fact, the cause of animate motion and change comes from the internally controlled and channeled release of internally stored chemical energy that is characteristic of biological entities. Animate motions have a quality of function, purpose, or goal-directedness. This is a direct consequence of their governance by biological control mechanisms: these enable adjustments of, and coordination of, component actions, both as a whole and as separate components. These adjustments can affect social as well as nonsocial environments. The effect is an ability to adapt to unforeseeable changes in circumstances and interact with social and nonsocial environments. The cause of inanimate motion is an external force, and there is always a transfer of energy from one object to another or a conversion of potential energy to kinetic energy. This is the case even when a person serves as an agent. My view is that these facts about nature are deeply related to the way people come to distinguish between the general categories of animate ‘The choice of simple artifacts group of adults.

was based ona scaling study with an independent

Gelman

l

81

and inanimate. Sets of different causal principles help us carve the psychological world at its joints, producing distinctions that guide and organize our differential reasoning about entities in one domain versus another. The foregoing underlies my proposal that the real-world distinctions between the causes of animate and inanimate motions and transformations are captured by what I dubbed “innards” and “external-agent” principles. The idea is that higher order causal principles direct attention to a combination of information, including kind of matter, source of energy, and trajectories. As regards objects obeying the innards principle, the energy source for movements, transformations, perceptions, and interactions is intrinsic to biological objects and thus is contained within the objects themselves. In regard to objects obeying the external agent principle, the movements, transformations, and interactions of nonbiological objects are caused by agents and forces of nature that are external to the objects themselves. I make no commitment to what kind of explanation system individuals or cultures marshal to flesh in these principles. I only assume that these domain-specific principles serve to organize the uptake and coherent storage of relevant inputs, to help place young learners on relevant learning paths. Given the assumption that there is the conjoining of material kind with the innards and external-agent principles, we predicted that young children would be disinclined to treat machines as clearly belonging to either the animate or inanimate categories. Basically, the idea is that young children would know that machines are made of the wrong “stuff” and move in the wrong way, although they appear to have the capacity for self-generated motion. Such facts would lead them to use a default strategy, the effect of which would be to start to create a new category (Subrahmanyam & Gelman, in press). To the extent that they could say anything about such devices, we expected them to refer to external agents or mechanisms. We thought it was also possible that they would mention the material as a way of saying that it was not animate. Finally, we did not expect them to be animists. We argued that it takes knowledge of both the animate domain and target objects to achieve animistic analogies. Put differently, the idea is that animism is part of an acquired explanation system. The aforementioned line of reasoning avoids an unacceptable conclusion, this being that our young are endowed with principles that pick out machines per se. It hardly can be the case that there are innate skeletal causal principles for learning to identify machines.2 They are 2Elsewhere, I discuss why only some domain-specific bodies of knowledge benefit from innate skeletal principles (Gelman, 1998; Gelman & Williams, 1998).

82

l

5.

Animates

and Others

cultural artifacts; especially ones designed to support travel and mimic sentient properties. More generally, I do not believe that young children, or even most adults, should be granted the kind of theory-rich knowledge about biology, chemistry, and physics that I make use of to develop the case that there are domain-specific causal principles and that these are skeletal in form. They can serve acquisition even if they constitute but nascent structures. For active minds use whatever mental structures are available to find and assimilate those kinds of inputs that can nurture the development of the knowledge base that can put flesh on these structures. Thus, to start, sets of first principles can serve to pick our relevant data and serve as memory drawers within which to keep together relevant learnings. This is a necessary condition for the acquisition of understanding. Of course, it is one thing to place the material together in memory and another to reorganize it into a theory-rich account of a domain. The literature on this accomplishment makes it clear that acquisition of such understanding can be a long, arduous process (see Bransford et al., 1999, for a review). Considerations like these are why we were conservative about the depth of knowledge we expected. True, Subrahmanyam (Subrahmanyam & Gelman, in press) and I predicted that our preschool subjects would treat machines differently than either animate or nonmachine artifacts. We also expected that they would make some effort to tell us about causally relevant matters. However, we also were disinclined to think they would reveal coherent biological or mechanical understandings of why any class of objects belonged together (Carey, 1985, 1995; Gelman, 1991; Gelman et al., 1995). A sampling of the Subrahmanyam and Gelman (in press) findings serves to illustrate the lines of evidence that together confirmed these predictions. First, as regards the matter of causal principles: Both the children3 and adults answered questions in ways that showed they knew that machines are subject to different causal conditions than are animates-even if the machine can appear to move on its own or display sentient properties. When either age group explained their Yes-No answers about animates, they appealed to the presence of animate parts (e.g., bones, muscles, blood, mouth), causes (e.g., wants to move), and animate notions (e.g., alive, living) to justify their attributions. When talking about machines, they invoked the absence of animate features or the presence of inanimate parts (e.g., wheels), inanimate mechanisms (e.g., batteries, engines), inanimate materials (plastic, metal, hard stuff), and external sources of energy or agency (a pilot). As shown in the top half of Fig. 5.1, these explanation differences are re3There were no reliable age effects for the children.

Gelman

l

83

lated to the pattern of Yes-No answers for the predicate MOVE. These data are plotted as a line graph given that we could use the adult data to generate an ordering of the various items shown on the X-axis. As can be seen, if anything, the adults were more inclined to attribute the capacity for motion to inanimate items -including a doll-than were the children. The reason for this was straightforward: the adults had a

i - -+- -Breathe +Breathe

Adult : Child 1

f I’ I\

FIG. 5.1 Tendency of 4- and 5-year-old that various object kinds can move.

children

as opposed to adults to say

84

l

5.

Animates

and Others

richer repertoire of possible causal conditions, including their own capacity to move a computer, their knowledge about the computer devices that make dolls talk, walk, and so on. Thus, the more participants knew about mechanisms, the more they were able to move toward animistic attributions, regarding the capacity for the seemingly self-generated motion of machines. Here, then, is another way to make the point that the use of motion path is related to causal conditions. As can be seen in Fig. 5.1, the pattern of attributions of animate properties to machines was selective in ways that we expected. It was consistent with the proposal that animates and machines are treated as different kinds of things. Still, knowledge about the different classes of objects would be rather shallow. The bottom half of Fig. 5.1 begins to illustrate this. Although the ability to breathe is selectively attributed to animates and plants, young children are far from certain that all animals and plants breathe. The generalization pattern shown here is the same as ones published by Carey (e.g., 1985). We concur with her account: The children do not have a biological theory, a theory that takes as given the capacity and structural wherewithal to engage in intrinsically generated air-exchange. This conclusion is buttressed by the patterns shown in Fig. 5.2. Notice the characteristic fall-off of the tendency of the children to attribute a brain as well as the capacity to think and remember to a dog, elephant, and bug. We end by returning to the question of animism. Theories that characterize young children as animists place special emphasis on the fact that too broad a range of objects are granted the ability to move (Piaget, 1930). However, as we can see in Fig. 5.1, adults are more prone to do this than are young children. For us, this means that animism is a cultivated mode of thought, one that is dependent on the acquisition of knowledge about both the source and target items. This hypothesis gains support from the pattern of results shown in Fig. 5.2, which illustrates a clear developmental pattern. It was the adults who said computers and robots can remember. However, more importantly for this volume, it rests on the assumption that motion paths are causally categorized by domain-relevant principles. This done, knowledge and explanation systems will develop as a function of our active tendencies to engage in epigenetic interactions with potentially relevant environments. ACKNOWLEDGMENTS Partial support for both the research reported here and the preparation of this chapter was provided by NSF grants DFS-9209741 and SRB-97209741 to the author.

lOOH Brain Adult aBrain Child

80 z3

mThink Adult OThink Child

60

E 40 ai

20 0

.I

I,,

l-l

,

.

n ,......In

l-l

n Remember Adult 0 Remember Child

0

FIG. 5.2 Tendency of 4- and 5-year-old children as opposed to adults to say that various object kinds have a brain, think, and remember. iii

86

’

5.

Animates

and Others

REFERENCES Bransford, J., Brown, A., & Cocking, R. (1999). How people learn: Brain, mind, experience and school. National Research Council, Washington, DC: National Academy Press. Carey, S. (1985). Conceptual change in childhood. Cambridge, MA: Cambridge University Press. Carey, S. (1995). On the origins of causal understanding. In S. Sperber, D. Premack, & A. J. Premack (Eds.), Causal cognition: A multi-disciplinary debate (pp. 268-308). Cambridge: Oxford University Press. Gelman, R. (1990). First principles organize attention to relevant data and the acquisition of numerical and causal concepts. CognitiveScience, 2 4, 79-l 06. Gelman, R. (199 1). Epigenetic foundations of knowledge structures: Initial and transcendent constructions. In S. Carey & R. Gelman (Eds.), The epigenesis of mind: Essayson biology and cognition (pp. 293-322). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Gelman, R. (1998). Domain specificity in cognitive development: Universals and nonuniversals. In M. Sabourin, F. Craik, & M. Robert (Eds.), Advances in psychological science: Vol. 2. Biologica and cognitive aspects (pp. 557-579). Hove, England: Psychology Press Ltd. Gelman, R., Durgin, E. & Kaufman, L. (1995). Distinguishing between animates and inanimates: Not by motion alone. In D. Sperber, D. Premack, & A. J. Premack @is.), Causal Cognition: A multidisciplinary debate (pp. 150-l 84). Oxford, England: Clarendon Press. Gelman, R., Spelke, E., & Meek, E. (1983). What preschoolers know about animate and inanimate objects. In D. Rogers & J. A. Sloboda (Eds.), The acquisition of symbolic skills. London: Plenum. Gelman, R., & Williams, E. M. (1998). Enabling constraints for cognitive development and learning: Domain specificity and epigenesis. In D. Kuhn & R. S. Siegler (Eds.), Handbook of child psychoZogy, vOZ.2:Cognition, perception, and language (5th ed., pp. 575-630). New York: Wiley. Keil, F., Kim, S. N., & Greif, M. (in press). Categories and levels. To appear in E. Forde & G. Humphreys, Category-specificity in brain and mind. Psychology Press. Mandler, J. M. (in press). On the foundations of the semantic system. To appear in E. Forde & G. Humphreys, Category-specificity in brain and mind. Psychology Press. Massey, C. M., & Gelman, R. (1988). Preschooler’s ability to decide whether a photographed unfamiliar object can move itself. Developmental Psychology, 24, 307-317. McCloskey, M. (1983). Naive theories of motion. Mental models. D. Gentner & A. L. Stevens. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Piaget, J. (1929). Thechild’s conception of the world. London: Routledge & Kegan Paul. Piaget, J. (1930). The child’s conception of physical causality. London: Routledge & Kegan Paul. Stewart, J. A. (1982). Perception of animacy. Unpublished doctoral dissertation, University of Pennsylvania. Stewart, J. A. (1984, November). Object motion and the perception of animacy. Paper presented at the meetings of the Psychonomic Society, San Antonio, TX.

Gelman

l

87

Subrahmanyam, K., & Gelman, R., in collaboration with Lafosse, A. (in press). To appear in E. Forde & G. Humphreys, Category-specijkity in brain and mind. Psychology Press. Williams, E. M. (2000). Causal reasoning by children and adults about the trajectory, context, and animacy of a moving object. Unpublished doctoral dissertation, University of California at Los Angeles.

This Page Intentionally Left Blank

How to Build a Baby . . . That Develops Atypically Institute

Annette Karmiloff-Smith

of Child Health, University College, London

A PERSONAL

I

BEGINNING

first admired Jean Mandler, not because I knew her in person, but because of her wonderful titles: “Remembrance of things parsed,” ‘IA tale of two structures,” ‘A code in the node,” “The cradle of categorization,” “ Separating the sheep from the goats” . . . through to the wonderful “How to build a baby.” And it is this latter that inspired the title of my chapter for this volume in her honor. Jean’s impact on my life both personally and professionally goes well beyond cute titles, of course. First, she has been a constantly inspiring colleague on deep issues about how infants and children represent the world. But just as important, Jean has been an inspiring woman who over the years has become a very dear friend. How I look forward to the second half of each year when the Mandlers hit London and move back into their charming Hampstead home. But will Jean ever forgive me for moving away from Hampstead to be within walking distance of my office? As a regular half-yearly visitor to the Medical Research Council’s Cognitive Development Unit (CDU) from 1982 to 1998, there was no topic on which

90

.

6.

Build an Atypical

Baby

Jean didn’t express a critical but constructive opinion. She added a very special dimension to our unit discussions, one that was deeply missed each year during January to June. The CDU closed in 1998, and I often find myself wondering what Jean’s input would be to the work of my team at the Neurocognitive Development Unit at the Institute of Child Health. But I and my colleagues have moved almost entirely into atypical development and to relations between genes, brains, and cognition and, as Jean reminded me in a recent email I am not even sure the study of the brain will help us understand the mind. As for atypical development, I never thought that was the way to go, because we don’t know enough about normal development for it to guide us as to what went wrong. But it is clear that I am hopelessly biased in this regard, and probably am missing important insights as to how to proceed. More importantly, we all have to do what excites us . . . . (J. Mandler, personal communication, April 3rd, 2000)

I offer this short chapter to you, Jean, as a tribute to someone who has always done what excites her and done it so well . . . and to someone who inspired me to dare to set up a baby lab, as you did, Jean, at a time in life when I should have been more focused on planning future retirement than beginning a whole new area of research. I will try to persuade you in the following pages, Jean, that although I agree that normal development cannot always guide us as to what goes wrong, the study of atypical development may even help us in the other direction. Its findings might encourage us to rethink some of our assumptions about normal development. Finally, I will look at how your own work on how to build a (normal) baby who comes to a nonperceptual understanding of objects could inform future work on the study of atypical infants. INTRODUCTION Mandler is right that we still don’t know enough about normal development for it to guide us as to what went wrong in atypical development. Yet, what researchers do already know about normal development is often automatically assumed to hold for atypical development. If scores in a developmental disorder fall within the normal range, it is usually taken for granted that they are equivalent to those of typically developing children. And if an atypical group is below its chronological age, has, for example, a mean Mental Age of 6 years, it is assumed that this is equivalent to the level of a normal 6-year-old. Behind these assumptions are two things: (a) a lack of consideration of the nature of internal representations that sustain overt behavior, and (b) a lack of appreciation of the fact that similar, overt behavior can stem

Karmiloff-Smith

l

91

from very different underlying representations. Even when researchers do consider cognitive underpinnings, much of the work on developmental disorders focuses on middle childhood and beyond. It ignores the importance of the nature of early representations that develop during infancy and how nonperceptual understanding of objects develops. If an uneven pattern of cognitive abilities and deficits is found in adulthood, it is simply taken for granted that a similar pattern holds for the starting state of development. And, if the disorder is genetic in origin, the next assumption is that a mutated gene (or small set of genes) is directly responsible for the adult phenotypic outcome. In this chapter, I challenge these assumptions, showing that the infant start state cannot be derived from the adult end state. I also argue that understanding the relation between the infant start state and the adult end state in atypical development can actually lead to a reassessment of some assumptions about normal development. MODULAR

CONTINUITY

ASSUMPTION

The Modular Continuity Assumption holds that the brain is organized into innate (genetically controlled) mental or neural modules that are present at birth with the same potential for dissociation across the human life span. It is assumed that a transparent relation exists between phenotypic outcomes and genes, with the expectation that the same dissociations observed in the adult steady state must hold during the period in which these abilities emerge during infancy. The relation is assumed to be linear. Such assumptions are so deeply engrained in cognitive neuropsychology and in many practitioners of developmental cognitive neuroscience that researchers are not aware that they are making them. They simply use findings from middle childhood and adulthood to make claims about innately specified modules which are considered to be impaired or intact, depending on the syndrome (Baron-Cohen, 1998; Leslie, 1992; Pinker, 1999; Temple, 1997). These assumptions need to be assessed by the direct study of infancy. Deriving

the Infant

State From the Adult

Phenotypic

Outcome

The pioneering work of Bellugi and her collaborators (Bellugi, Wang, & Jernigan, 1994) on a rare neurodevelopmental disorder, Williams syndrome (WS), pointed to dissociations in cognitive architecture in the adult phenotype which were used to make claims about the modular structure of the mind and brain. Language and face processing were argued to be preserved alongside both general retardation and severe problems with visuo-spatial cognition, number, planning, and problem solving (Bellugi,

92

l

6.

Build an Atypical

Baby

Wang, & Jernigan, 1994). Although language and face processing are in fact relative strengths and weaknesses (Karmiloff-Smith, 1998; Klein & Mervis, 1999; Tager-Flusberg & Sullivan, 1996; Vicari, Carlesimo, Brizzolara, & Pezzini, 1996; Volterra, Cap&i, Pezzini, Sabbadini, & Vicari, 1996), secondary citations have made sweeping claims about absolute strengths in the syndrome (Bickerton, 1997; Pinker, 1994,1999). As Pinker recently stated when contrasting individuals with Specific Language Impairment (XI) and WS: “The genes of one group of children impair their grammar while sparing their intelligence; the genes of another group of children impair their intelligence while sparing their grammar” (Pinker, 1999, p. 262). This implies that the brain is normal in all respects except grammar or vice versa, ignoring the dynamics of brain development as a whole during the lengthy postnatal period. In general, writings by linguists, psychologists and philosophers have used WS to bolster theories of innate and independently functioning modules, some of which are intact and others impaired (e.g., Pinker, 1994,1999). This emanates from a position, held explicitly or implicitly, that behavioral deficits found in the phenotypic outcome of individuals with genetic disorders are direct windows on the initial state, that is, the innate, modular structure of the cognitive system (Baron-Cohen, 1998; Leslie, 1992; Pinker, 1999; Temple, 1997). The pattern of the cognitive system in the phenotypic outcome is treated as if it were static, that is, composed of linear developmental change, with some components of the normal system missing and others intact. The nativist argumentation is based on the model of brain damage to previously normal adults, used by neuropsychologists. But this is a brain that has developed over time and become modularized, not one that necessarily started out as modular. Indeed, it is probable that normal adult brains have a good deal of modular structure to enable them to function rapidly and efficiently. However, an adult brain is very different from an infant brain. Research in several different domains of cognition suggests that the normal infant brain requires many months or years to become like an adult brain, in terms of both localization of function and specialization of the latency and amplitude of electrical activity (De Haan, Johnson, Maurer, & Perrett, 2000; Mills et al., in press; Neville et al., 1989). So, in the case of normal development, the study of infancy is crucial. And this is even more the case for atypical development where the Modular Continuity Assumption is held so strongly, either explicitly or implicitly. Challenges

to the Modular

Assumption

Contrary to claims in the literature is intact, it turns out to be relatively good compared

Language and Number.

language

Continuity

that WS to some

Karmiloff-Smith

l

93

other clinical groups and relatively good compared to their spatial d&itits. However, in general, it is no better than their mental age would predict (Klein & Mervis, 1999). In fact, one of the crucial features of WS language is that in infancy and toddlerhood it is initially seriously delayed (Mervis, Morris, Bertrand, & Robinson, 1999; Singer Harris, Bellugi, Bates, Jones, & Rossen, 1997). We need to understand why the language of people with WS language is so delayed in infancy, and how the infant state relates to the phenotypic outcome not only in language but in other domains of development, as well. In a study of atypical infants, a doctoral student in my lab challenged the Modular Continuity Assumption (Paterson, 2000; Paterson, Brown, Gsodl, Johnson, & Karmiloff-Smith, 1999), which consists in deriving the infant state from the pattern of proficiencies and impairments in the phenotypic outcome. It is known that WS and Down’s Syndrome (DS) display different cognitive profiles in the end state (Jernigan, Bellugi, Sowell, Doherty, & Hesselink, 1993; Klein & Mervis, 1999; Wang, Doherty, Hesselink, & Bellugi, 1992). If the infant state can be directly derived from the phenotypic outcome, then the profiles for WS and DS should be the same in infancy as they are in adulthood. Paterson et al. (1999) purposely chose two tasks-one language-related, one number-related-which could be designed to be as similar as possible for both infants and adults. For number, numerosity judgments were required; for language, receptive vocabulary measures were taken. The domains of vocabulary and language were purposely chosen because in the phenotypic end state it has been claimed that WS shows greater proficiency in vocabulary than DS, and that both syndromes are seriously impaired in number. Paterson asked whether these patterns are also true of WS and DS infants. Paterson (2000) first tested adults with WS and DS on a battery of number and vocabulary tasks. Using a standardized vocabulary test, the British Picture Vocabulary Scale (Dunn, Whetton, & Pintilie, 1982), she showed that WS and DS adults had significantly different scores, with the WS adults clearly outstripping the DS adults. This confirmed previous work. For number, Paterson focused on numerosity judgments because these could be done by both adults and infants. Although some work had been carried out on number conservation with adults with WS and DS (Bellugi et al., 1994), numerosity judgment tasks had not been used. Participants were required to judge which of two numbers (either Arabic numerals l-9 or equivalent numbers of dots) displayed on a computer screen was the larger. Reaction times and accuracy measures were taken. In the normal case, the symbolic distance effect is always apparent: numbers very close (e.g., 8 and 9) take longer for a decision than numbers that are far apart

94

l

6.

Build an Atypical

Baby

(e.g., 8 and 3). Paterson demonstrated that the WS and DS adults performed differently on the numerosity judgment tasks. The DS adults, although having slower reaction times overall, did show the distance effect, taking significantly longer for close numbers. This is similar to the pattern shown by normal controls. By contrast, the WS adults did not show the distance effect. Thus, the phenotype in the adult end state turned out to be as follows: DS performed significantly worse than WS on vocabulary; WS performed significantly worse than DS on numerosity judgments. Having established the phenotypic outcome in the domains of number and language, Paterson et al. (1999) then measured similar abilities in infants with WS and DS, using the preferential looking paradigm. Sixty-five infants between 13 and 36 months were tested, divided into four groups: WS infants, DS infants matched for Chronological Age (CA) and Mental Age (MA) on the Bayley Infant Scales II (Bayley; Bayley, 1993), normal MA control infants (also matched on the Bayley), and normal CA control infants. The DS infants were included to control for general learning difficulties at the same level of intelligence and age. The MA normal controls were equated to the atypical groups for general levels of intelligence, and the CA controls for general levels of experience. For vocabulary, infants saw pictures of two items and simultaneously heard a voice naming one of the pictures (e.g., “Look, look at the car”). It is known that in normal development, infants will look significantly longer at the named picture (Golinkoff, Hirsh-Pasek, Cauley, & Gordon, 1987). For number, infants were familiarized to pairs of two objects (e.g., two trees with two cats, two lions with two cars, etc.) and, after the familiarization phase, were shown one picture with two new objects and one with three new objects. Again, normal infants look significantly longer at the picture with the new numerosity, despite the fact that all stimuli are potentially interesting because they represent new objects (Starkey, Spelke, & Gelman, 1990). Normal infants, then, treat the actual objects displayed as secondary to their extraction of the “two-ness” of the familiarization displays and therefore look longer at the display of three objects. If the use of the Modularity Continuity Hypothesis were justified, then WS and DS infants should show the same profile of cognitive abilities and impairments as adults, with the WS infants outperforming the DS infants on the vocabulary task, and the opposite for the number task. But this was not the case. In fact, the WS and DS groups both performed significantly worse than CA controls on the language task. In other words, the infants from both syndromes were equally impaired on vocabulary, despite the WS adults being significantly better than the DS adults on vocabulary. In fact, both atypical groups of infants per-

Karmiloff-Smith

l

95

formed like the MA controls, that is, at approximately half their chronological age. By contrast, although the WS adults were more impaired than the DS adults with numerosity judgments, the WS infants were unimpaired on the numerosity judgment task. They performed like the CA controls, whereas the DS infants were seriously impaired and did not even reach the level of the MA controls. Again, the pattern in infancy from both syndromes differs considerably from that observed in adulthood. This once more challenges the Modular Continuity Assumption, showing both that infant profiles cannot predict steady state outcomes and that phenotypic end states cannot be used to derive the infant start state. The aforementioned infant and adult findings from two syndromes make it clear that claims about innate modules based on phenotypic outcomes cannot be taken for granted. The infancy and adult data suggest that the learning trajectories of the two syndromes are different in development. Only developmental studies following infants from the initial state through childhood and adulthood can address this question properly. Yet again, this stresses the need to consider the process of development itself when studying developmental disorders and that claims regarding innately specified modules cannot be based on patterns found in phenotypic outcomes (Karmiloff-Smith, 1998; Paterson et al., 1999). Face Processing. Face processing in WS has also been claimed to be intact because of excellent behavioral scores in adulthood compared to the poor scores of adults with DS (Bellugi et al., 1994). Indeed, Bellugi et al. (1991) asserted: I‘... we find in the WS population normal face processing capacities with at floor performance on spatial tasks” (p. 117, my italics), and Rossen et al. (1996) claimed to have found “. . . selective preservation of face recognition in Williams syndrome” (p. 3 71, my italics). Thus, according to the Modular Continuity Assumption, one can automatically assume that infants with WS will display particular proficiency with face processing tasks. But is this so? By 1 month of age, typically-developing infants show a novelty preference for new faces over old faces in a preferential looking task. By 3 months of age, not only do they display a novelty preference but they also show a prototype effect (De Haan et al., 2000). If four faces are displayed one after another and then a choice between a fifth face (not previously seen by the infant but a prototype morphed from the previous four faces) and one of the familiar faces, then 3-month-olds (but not yet 1 -month-olds) treat the morphed face as more familiar than the already familiar face. This is because the prototype of the four faces feels more familiar than any one of the faces. In other words, by 3 months

96

l

6.

Build an Atypical

Baby

normal infants do not simply learn exemplars but build up prototypical representations of previously processed faces. In a pilot study, doctoral student Janice Brown found that WS infants did not show the prototypical effect (Brown, 2000). The preliminary findings suggest that infants and toddlers with WS tend to learn exemplars and may lack the generalization processes necessary to form prototypes. If this initial finding holds, the WS lack of prototype extraction could well be due to a tendency toward featural processing rather than configural processing. It has been shown that infants with WS spend more time than normal or DS controls fixated on faces (Mervis & Bertrand, 1997), so they do not lack experience. The infant processing style, which departs from the typically developing infant, may in fact explain the phenotypical outcome in WS adult face processing rather than the adult behavioral data revealing an intact face processing module already specified in infancy (see also, Deruelle, Mancini, Livet, Casse-Perrot, & de Schonen, 1999; Karmiloff-Smith, 1998). Yet again, we have to bury the myth of what at first blush seemed like an intact face processing module which would operate from infancy onwards in this syndrome. Face processing follows a different developmental trajectory in WS. Children and adults with WS show clear behavioral deficits in visuo-spatial tasks outside face processing. Our lab has also pursued spatial cognition in infancy (see also, Atkinson et al., 1997). In a study of saccade planning, Brown and colleagues showed that infants with WS have very atypical spatial representations for planning visually-guided actions (Brown, 2000; Brown et al., submitted 2000). Saccades in CAand MA-matched infants with DS are executed within body-centered spatial coordinates, as are those of normal controls. By contrast, infants with WS displayed evidence of deficits in saccade planning, suggesting a greater reliance on subcortical processing mechanisms than the other infant groups. So once again, be it face processing or v&m-spatial processing, the WS brain proceeds along an atypical developmental pathway. It is not, then, that one module-face processing-is spared, and the other module-visuo-spatial processing- is impaired. Both domains develop differently in WS. One cannot use the adult phenotypic outcome to make claims about impaired and intact modules in infancy or about their nonchanging developmental trajectories. Finally, evidence from a more general area of cognition-sustained attention-also shows how the adult end state cannot be automatically used to derive the infant starting state. Adults with WS are often reported as having attentional problems. Morris, Demsey, Leonard, Dilts, and Blackburn (1988) found that attention deficit hyperactivity disorder (ADHD) was characteristic of 84% of 4- to 16-year-old WS partici-

Karmiloff-Smith

l

97

pants. Other, more recent studies claim that children with WS are 4 times more likely to have ADHD than children in the general population (Finegan, Sitarenios, Smith, & Meschino, 1994). Likewise, Udwin, Yule, and Martin (198 7) reported that 72% of a sample of 6- to 15-year-old children with WS displayed hyperactivity in school, at home, or both, as rated by parents and teachers on the Rutter Questionnaires (Rutter, 1967). Thus, children and adolescents with WS tend to present with problems of attention. By contrast, the reported incidence of ADHD in people with Down’s syndrome varies, but is much lower than that reported for WS (Cocchi & Favuto, 1997; Green, Dennis, & Bennets, 1989). In general, then, older children and adolescents with WS suffer more from attentional problems than those with DS. To test the validity of the Modular Continuity Assumption, it is therefore crucial to examine attention in infancy in WS and DS. Using an observation and coding system for measuring sustained attention (Ruff & Lawson; 1990), Brown and collaborators (Brown, 2000; Brown et al., 2000) tested fourteen 24- to 36-month-olds with WS, 19 with DS, 17 CA controls, and 16 MA controls (matched on the Bayley). Brown found that the infants with DS were impaired relative to the matched MA group on the number of periods of sustained attention they displayed. They were also impaired relative to the WS infants on total duration of attention. Thus, whereas in adulthood attentional problems seem more serious in WS compared to DS, in infancy, the WS infants perform better than the DS infants. Yet again, it would be erroneous to make claims about the infant start state simply by deriving it from the phenotypic outcome of these two syndromes. Relations

Between

Typical

and Atypical

Development

At the start of this chapter, I suggested that our comparison of atypical adults with atypical infants from various syndromes might lead to a reassessment of some assumptions about the normal case. We saw that in some cases the representational precursors that may underlie subsequent development in a domain are inadequate in infancy. By contrast, in other cases, the infant representational precursors seem to have developed normally, but it is the subsequent process of learning that goes awry. In general, the normal literature may not have sufficiently questioned assumptions about how infant capacities are actually related to adult capacities in a similar domain. It is usually taken for granted that if we measure, say, sensitivity to different numerosities in infants, this has a relation to subsequent number development. Our data from atypical development may encourage re-

98

l

6.

Build an Atypical

Baby

searchers to question this and to take a more developmental approach, charting trajectories from infancy to adulthood using very similar tasks. Our current analyses of atypical development remain fairly coarse. How could Mandler ‘s in-depth analysis of infant representational systems, that is, on how typical babies are built (Mandler, 1988, 1992), inform our thinking about the building of an atypical mind? No research group is studying atypical development with that level of depth. However, it is not excluded. We might ask whether the visuo -spatial difficulties of infants with WS have a direct impact on their preverbal conceptual representations and thereby perhaps explain their delay in language. Do they build image schemas such as PATH, UP-DOWN, CONTAINMENT, FORCE, PART-WHOLE, AND LINK, conceptual notions that Mandler claims are derived from perceptual structure (Mandler, 1988, 1992)? To what extent is basic perceptual structure normal in syndromes like Williams and Down’s? Do atypical infants represent information from an early age at more than one level of description, as both I and Mandler have suggested is the case for normal development (Karmiloff-Smith, 1986,1992; Mandler, 1988, 1992)? As mentioned, current studies of atypical development are at a more coarse level of analysis, and I am still far from using my own work on representational redescription in normal children (Karmiloff-Smith, 1986, 1992) or Mandler ‘s on conceptual recoding (1988, 1992) for understanding atypical development. We still know very little about how atypically-developing infants represent perceptual input and whether they simplify and redescribe perceptual input to understand such concepts as animacy, inanimacy, causality, agency, containment, and support. My intuition is that such studies could be crucial in understanding how the atypical developmental trajectory of language acquisition goes awry, for instance. To my knowledge, no researcher in atypical infant development has cited Jean Mandler ‘s work on the development of representation. Yet, her in-depth, theoretical analysis of such development could play an important role in future studies of how to build a baby . . . that develops atypically. ACKNOWLEDGMENTS The studies reported in this article were funded by Programme and Projects Grants to the author by the UK Medical Research Council, as well as PhD studentships from the Medical Research Council and the Down’s Syndrome Association. My thanks go to Daniel Ansari for his comments on an earlier version of this chapter.

Karmiloff-Smith

l

99

REFERENCES Atkinson, J. A., King, J., Braddick, 0. J., Nokes, L., Anker, S., & Braddick, F. (1997). A specific deficit of dorsal stream function in Williams Syndrome. Neuro Report, 8, 19 19-l 922. Baron-Cohen, S. (1998). Modularity in developmental cognitive neuropsychology: Evidence from autism and Gilles de la Tourette syndrome. In J. A. Burack, R. M. Hodapp, & E. Zigler (Eds.), Handbook ofmental retardation and development (p. 335). Cambridge, England: Cambridge University Press. Bayley, N. (1993). Bayley Scales of Infant Development (2nd ed.). San Antonio, TX: Psychological Corporation. Bellugi, U., Bihrle, A., Jernigan, T., nauner, D., & Doherty, S. (1990). Neuropsychological, neurological, and neuroanatomical profile of Williams syndrome. American JournaZ of MedicaZ Genetics, 6, 115-125. Bellugi, U., Wang, F?I?, & Jernigan, T (1994). Williams syndrome: An unusual neuropsychological profile. In S. H. Broman & J. Grafman (Eds.), Atypical cognitive deficit in developmental disorders: Zmplicationsfor brainfunction (pp. 23-56). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Bickerton, D. (199 7). Constructivism, nativism, and explanatory adequacy. Behavioral and Brain Sciences, 20, 55 7-55 8. Brown, J. H., Johnson, M. H., Paterson, S. J., Gilmore, R., Gsodl, M. K., & Karmiloff-Smith, A. D. (submitted, 2000). Spatial representation for action in infants with Williams Syndrome: Do they have a specific dorsal pathway deficit? Cocchi, R., & Favuto, M. (1997). Hyperkinesis in Down’s syndrome: A survey on 5 10 persons. Italian Journal of lntellective Impairment, 10, 19-23. De Haan, M., Johnson, M .H., Maurer, D., & Perrett, D. I. (2000). Recognition of individual faces and average face prototypes by I- and 3-month-old infants. Manuscript submitted for publication.. Deruelle, C., Mancini, J., Livet, M. O., Casse-Perrot, C., & de Schonen, S. (1999). Configural and local processing of faces in children with Williams syndrome. Brain and Cognition 41, 276-298. Dunn, L. M., Whetton, E., & Pintilie, D. (1982). British Picture Vocabulary Scale. Windsor, Berkshire: NFER-Nelson. Finegan, J., Sitarenios, G., Smith, M., & Meschino, W. (1994). Attention deficit hyperactivity disorder in children with WS: Preliminary findings. Paper presented at Williams Syndrome Association Annual Professional Conference, San Diego, CA. Golinkoff, R., Hirsh-Pasek, K., Cauley, K. M., & Gordon, L. (1987). The eyes have it: Lexical and syntactic comprehension in a new paradigm. Journal of Child Language, 24, 23-45. Green J. M., Dennis, J., & Bennets, L. A. (1989). Attention disorder in a group of young Down’s syndrome children. Journal of Mental Deficiency Research, 33, 105-122. Jernigan, T. L., Bellugi, U., Sowell, E., Doherty, S., & Hesselink, J. R. (1993). Cerebral morphologic distinctions between Williams and Down syndromes. Archives of Neurology, 50, 186-l 91. Karmiloff-Smith, A. (1986) “From metaprocesses to conscious access: Evidence from children’s metalinguistic and repair data.” Cognition, 23, 95-147. Karmiloff-Smith, A. (1998). Development itself is the key to understanding developmental disorders. ‘Pends in Cognitive Sciences, 2, 389-398.

100

l

6.

Build an Atypical

Baby

Karmiloff-Smith, A. (1992). Beyond modularity: A developmental perspective on cognitive science. Cambridge, MA: MIT Press. Klein, B. I?, & Mervis, C. B. (1999). Contrasting patterns of cognitive abilities of 9- and 10-years-olds with Williams syndrome or Down syndrome. DeveZopmental Neuropsychology, 16, 177-196. Leslie, A. M. (1992). Pretence, autism, and the theory-of-mind-module. Current Directions in Psychological Science, 1, 18-21. Mandler, J. M. (1988). How to build a baby: On the development of an accessible representational system. Cognitive Development, 3, 113-l 36. Mandler, J. M. (1992). How to build a baby: II. Conceptual primitives. Psychological Review, 99, 58 7-604. Mervis, C., & Bertrand, J. (1997). Developmental relations between cognition and language: Evidence from Williams Syndrome. In L. B. Adamson & M. A. Romski (Eds.), Communication and language acquisition: Discoveries from atypical development (pp. 75-106). New York: Brookes. Mervis, C. B., Morris, C. A., Bertrand, J., & Robinson, B. F. (1999). Williams syndrome: Findings from an integrated program of research. In H. Tager-Flusberg (Ed.), Nuerodevelopmental disorders (pp. 65-l 10). Cambridge, MA: MIT Press. Mills, D. L., Alvarez, T. D., St. George, M., Appelbaum, L. G., Bellugi, U., & Neville, H. (in press). Electrophysiological studies of face processing in Williams syndrome. Journal of Cognitive Neuroscience. Morris, C., Demsey, S., Leonard, C., Dilts, C., & Blackburn, B. (1988). Natural history of Williams syndrome: Physical characteristics. Journal of Pediatrics, 213, 318-326. Neville, H. J., Holcom, I? J., & Mills, D. M. (1989). Auditory, sensory and language processing in Williams Syndrome: An ERP study. Journal of Clinical and Experimental Neuropsychology, 11, 52. Paterson, S. (2000). The development of language and number understanding in Williams syndrome and Down syndrome: Evidencefrom the infant and mature phenotypes. Unpublished doctoral dissertation, Neurocognitive Development Unit, Institute of Child Health, University College, London. Paterson, S. J., Brown, J. H., Gsodl, M. K., Johnson, M. H., & Karmiloff-Smith, A. (1999). Cognitive modularity and genetic disorders. Science, 286, 2355-2358. Pinker, S. (1994). The language instinct. New York: Penguin. Pinker, S. (1999). Words and rules. London: Weidenfeld & Nicolson. Rossen, M., Klima, E. S., Bellugi, U., Bihrle, A., & Jones, W. (1996). Interaction between language and cognition: Evidence from Williams syndrome. In J. H. Beitchman, N. J. Cohen, M. M. Konstantareas, & R. Tannock (Eds.), Language, learning, and behavior disorders: Developmental, bioZogicaZ,and clinical perspectives (pp. 367-392). New York: Cambridge University Press. Ruff, H. A., & Lawson, K. R. (1990). The development of sustained, focused attention during free play in young children. Developmental Psychology,26,85-93. Rutter, M. (1967). A children’s behaviour questionnaire for completion by teachers: Preliminary findings. Journal of Child Psychology and Psychiatry and Allied Disciplines, 8, l-l 1. Singer Harris, N. G., Bellugi, U., Bates, E., Jones, W., & Rossen, M. (199 7). Contrasting profiles of language development in children with Williams and Down syndromes. Developmental Neuropsychology, 23, 345-370. Starkey, I?, Spelke, E., & Gelman, R. (1990). Numerical abstraction by human infants. Cognition, 36, 97-127.

Karmiloff-Smith

l

101

Tager-Flusberg, H., & Sullivan K. (1996). Theory of mind abilities in young children with Williams syndrome. San Francisco, CA: Cognitive Neuroscience Society. Temple, C. M. (1997). Cognitive neuropsychology and its applications to Children. Journal of Child Psychology and Psychiatry, 38, 27-52. Udwin, O., Yule, W., & Martin, N. (1987). Cognitive abilities and behavioural characteristics of children with idiopathic infantile hypercalcaemia. Journal of Child Psychology and Psychiatry and Allied Disciplines, 28, 297-309. Vicari, S., Carlesimo, G., Brizzolara, D., & Pezzini, G. (1996). Short-term memory in children with Williams syndrome: A reduced contribution of lexical-semantic knowledge to word span. NeuropsychoZogia, 34, 9 19-925. Volterra, V, Capirci, O., Pezzini, G., Sabbadini, L., & Vicari, S. (1996). Linguistic abilities in Italian children with Williams syndrome. Cortex, 32, 663-677. Wang, I? I?, Doherty, S., Hesselink, J. R., & Bellugi, U. (1992). Callosal morphology concurs with neurobehavioral and neuropathological findings in two neurodevelopmental disorders. Archives of Neurology, 4 9, 407-4 11.

This Page Intentionally Left Blank

Pretense and Representation Department

I

Revisited

Alan M. Leslie of Psychology and Center for Cognitive Science Rutgers University

was lucky enough throughout the 1980s and early 1990s to have Jean Mandler as my colleague for about half of each year. Everyone at the erstwhile Cognitive Development Unit in London looked forward annually to Jean, along with spring, resuming her residence among us. Jean has a way of deploying her intellectual commitment, erudition, creativity, and skill in debate so that even if she is skewering your latest proposal it is done constructively and with great charm. I had particular reason to anticipate Jean’s return each year because I share with Jean a concern for the nature of early representation in infancy and the belief that infants are capable of representing abstract properties. Although our numbers are growing a bit (or so I hope), there are still not many people in the world who share that belief. What most people believe is that infants can represent sensory properties and ody sensory properties (for a contemporary expression of this dogma, see the entertaining article by Haith, 1998). According to this view, abstract properties have to be extracted from the statistics of how the world presents sensory properties to the individual infant during ontogenesis. Furthermore, the ontogenetic extraction from sensory properties requires numerous iterations over increasing levels of abstractness, and there-

104

l

7.

Pretense and Representation

fore concept acquisition proceeds by stages. The upshot is that abstract concepts, really abstract concepts, would not, should not, and perhaps cannot, be present early in life. Like Jean, I have focused on a set of highly abstract concepts that are present early in life in the hope that one might be able to understand how such a thing is possible. Rather than start with a preconceived idea of what an abstract concept is, we should be prepared to learn from our studies what such a thing might be. Rather than take as our point of departure the rule that early abstract concepts are impossible, we should be prepared to discover whatever nature is prepared to reveal. Although Jean and I disagree on many things, on these points we agree and are kindred spirits. In 1987, I published a new theory of our ability to pretend (Leslie, 1987); this has come to be called the metarepresentational theory of pretense. I want to revisit some issues that article raised and ask the following: What does our ability to pretend tell us about the nature of early abstract concepts? Why would anyone think that the ability to pretend should tell us anything about concepts? The answer is twofold. First, the heart of the metarepresentational theory is the claim that the emergence of pretense depends, not on the emergence of a new ability as such, but on a new concept, specifically, the concept PRETEND.’ Second, the concept PRETEND makes an interesting case study because it is, infact, a mental state concept. If these claims are correct, then children as young as 2 years old who pretend must possess at least one mental state concept. I stressed that PRETEND is infact a mental state concept for a reason. I do not want to prejudge the question of whether 2-year-olds, in understanding pretense, understand that pretense is a mental state. Actually, I doubt whether 2-year-olds know that pretending really is a mental state; fortunately, they do not need to know this to possess the concept. 1 learned from Jean that when it comes to the existence of early abstract concepts, there are always two paths one can follow. One path is to decide without further ado that a child with minuscule encyclopedic knowledge of the world and with extremely limited general reasoning power could not possibly possess abstract concepts. The other path is to investigate early concepts with an open mind: If we find that a child with no encyclopedic knowledge and little general reasoning can nevertheless possess a mental state concept then we will no doubt learn from that child something we did not know about the nature of concepts. ‘I write concepts in uppercase and italicize the property or relation to which the concept refers. I take concepts to be symbolic entities in a cognitive system, and assume that they represent or designate properties or relations “in the world.”

Leslie

A COMPARISON

l

105

OF BELIEF AND PRETENSE

Here is the basic idea I argue: Although in accounts of behavior there are usually trade-offs between process and representation that make their effects hard to distinguish, sometimes it is possible to distinguish between the two. I argue that the case of beZiefversus pretense provides such an example. We find that the ability to have a beziefrests on a mode of processing, whereas the ability to have a beliefabout a beliefrests on a representation or concept of belief. By contrast, in the sense corresponding to having a belief, there is no such thing as having a pretense, nothing for the case of pretense that rests on a special mode of processing. Instead, the ability to pretend rests on a special representation. Consider the case of beZief. I start with an assumption that should be quite uncontroversial: there is a big difference between having a beZief and having a belief about a belief. No one has yet developed a detailed cognitive model of exactly what it means for an organism to have a belief. However, “having a belief * is a state defined by how a representation is processed (or how an organism is disposed to process a representation). If a given representation is processed in one way, then represents; prothe organism believes whatever the representation cessed another way, the organism desires it, and so on for different modes of processing for each distinct mental state. What we lack is detailed understanding of what differentiates these different modes. However, in each case, the content of the representation will express the content of the corresponding mental state. Thus, the commonsense notion of having a belief is shorthand for a particular kind of processing mode within a cognitive system; the details of the various processing modes that are possible determine the range of mental states in which the organism is capable of being. No doubt the reason this assumption is uncontroversial (extreme reductionism aside) is that it says so little; that’s fine, however, because it says enough for these purposes. By contrast with merely having a belief, the ability to have a belief about a belief requires something beyond a mode of processing: it requires a specific representational ability, namely, it requires possession of the concept BELIEF. At least, this is the case in the sense of “belief about a belief 0 that is intended in “theory of mind” research, In “theory of mind” research, we are concerned with the recursive ability of someone to have a belief that someone believes that I? The reason for this is straightforward. If I am to believe that someone believes that P, then the second occurrence of believe is part of the content of my belief, and therefore part of the representation that I am processing, as a belief. Because representations like this are conceptual, that is, composed of concepts, then, to represent a belief as a belief, I must

106

l

7.

Pretense and Representation

use the concept BELIEF. This reasoning sounds more complex than it really is. This is because in “theory of mind” research we are concerned with the ability of A to attribute a belief to B. When A does this, A has a belief, namely the belief that B has a belief. The term beliefoccurs twice but in different guises, once as the ability of A to have beliefs, and once in A’s representation that B believes something. Let’s make it simpler. If A attributes to B the property of being a dog, that is, A believes that B is a dog, then A possesses and uses the concept, DOG. As psychologists, we might, of course, have the usual worries about whether A reaZZy attributed to B the property of being a dog specifically, and thus whether A really has the concept DOG specifically, as opposed to some other concept whose extension overlaps dogs; or, if we believe that concepts depend on inference drawing, we might worry whether A is really able to draw some fancy dog-related inference about B, and so on. If we cannot satisfy ourselves on these worries, then we will not be satisfied that A really has the belief that B is a dog as opposed to some overlapping and similar, but not quite, B is a dog belief. Conversely, to the extent we can satisfy ourselves, we will grant A the concept DOG. The case of BELIEF is not fundamentally different from the case of DOG. Similar reasoning, attended by similar worries, lies behind the familiar claim that when 4-year-olds pass the “Sally and Ann” false belief task, they attribute a belief to Sally, and that therefore they must possess and use the concept BELIEF. Accepting the aforementioned assumptions, which I take to be uncontroversial, the following picture for belief emerges. Even young infants can have beliefs. However, they are not able to have beliefs about beliefs. Taking young infants as illustration for my argument is not at all critical. Some writers, for example, Haith, 1998, object to the idea that young infants can have beliefs, although Haith, given his beliefs, should at least grant that young infants have sensory beliefs. No matter; my idea can be illustrated just as well if we consider l-year-old infants or adult monkeys. My point is that the capacity to have beZiefs should be developmentally and phylogenetically prior to the capacity to have beliefs about beliefs. The capacity to have beliefs is just the capacity

to have any kind of representational thought at all, to have any concepts whatsoever. Such a broad capacity can exist in many forms and to many different degrees and no doubt stretches far back in human ontogeny and far down in phylogeny. To have beliefs about beliefs, on the other hand, is a very specific capacity and demands that an organism possess a specific concept, namely, BELIEF. Which organisms possess this concept is an empirical question and we do not know the answer. However, it looks likely that there will be only one or two species other than our own who will turn out to possess this concept, or perhaps

Leslie

.

107

none at all other than our own. Likewise, the question of when in human ontogeny BELIEF becomes available is empirical and remains controversial. However, there would be wide agreement, based on research over the last 15 years, that it is not later than 4 years and not earlier than 2. In either case, there is a long period in which the normally developing human enjoys a capacity to have beliefs but lacks the capacity to have beliefs about beliefs. And that period may be very prolonged in abnormal development, for example, in Kanner syndrome (autism). Although there are similarities in the “logic” of the attitudes of pretending and believing (see Leslie, 1994, for discussion), there are striking dissimilarities in their natural histories. There seems to be no prolonged period in which a child is able to “have a pretend” before she is able to have beliefs about pretends. In fact, as I pointed out elsewhere (Leslie, 1987, 1994), the capacity to pretend seems to appear at the same time as the concept PRETEND, around 22 to 24 months. The principle piece of evidence for this is the “yoking” between the appearance of solitary pretending, in which the child plays all by herself, and the appearance of the ability to recognize pretense in other people, in which the child shares pretend play with another person. The ability to recognize pretense in others, and thus to share pretense, would be a truly remarkable ability at almost any age but it is intriguing to find it emerging as early as the 2nd birthday. At least it would be intriguing were we not inured by familiarity to the everyday miracle of early verbal and nonverbal communication. However, interesting although the ability for solitary pretense is, and interesting although the ability to recognize pretense in other people is, the fact that these abilities emerge together in development is the most intriguing fact of all. How odd then that students of early pretense prior Piaget (1955), apparently never noto Leslie (1987), most prominently, ticed, or at least never commented on, the social nature of early pretense. Did Piaget, who spent much time on the carpet with his three children, making the most intricate and insightful observations while interacting with them, never join in their pretend play? The yoking between solitary pretense and the recognition of pretense in others suggests that, in sharp contrast with the case of belief, there is no developmental priority of the capacity to have pretends over the ability to have beliefs about pretends. And the reason for this is straight-

forward. The human ability to have pretends actually consists in the employment of the concept, PRETEND. This striking claim lies at the heart of the metarepresentational theory of pretense. Before discussing this claim more closely, let me say a word about phylogenetic priority. It is wholly an empirical question whether species other than our own can pretend and, if so, whether the same or similar mechanisms underlie their pretense. At this time, it seems to me

108

l

7.

Pretense and Representation

that the evidence for an ability to pretend in other species is not compelling, although not surprisingly the best anecdotes feature great apes (e.g., Whiten & Byrne, 1988). As far as I know, there is not even a hint that another species might recognize pretense in conspecifics. Of course, one might take a totally superficial, behavioral approach to the question and advance examples like kittens chasing balls of wool, “pretending” to hunt. But then one has to ask oneself whether the kitten is pretending that the ball is a mouse, a squirrel, or any definite thing at all? Does the cat ever pretend that the mouse gets away, perhaps with a broken leg? Or one should ask, given that cats care and know a lot about bowls of milk, do cats ever pretend that an empty bowl contains milk? How surprised would one be to return home one day to find one’s cat pretending to lap milk from an empty bowl, perhaps stopping every so often to look up at one askance? Such a marvelous cat would surely be Puss-in-Boots. He might at any moment begin to talk to us. The point about chasing balls of wool, of course, is that cats have specialized mechanisms for hunting prey and these mechanisms are engaged by any right-sized object rushing past them. That cats should be so designed allows for the practice and honing of their survival-critical skills. Their ability for pretense is specialized around this topic. This specialization stands in marked contrast to the human capacity for pretense which is not at all limited to a few topics but is instead productive: the rule for human pretense, at any age, is that whatever we can think about, we can pretend about. The productivity of human pretense strongly suggests that a quite different cognitive mechanism underlies our ability, than the mechanisms underlying putative cases of specialized “pretending” that may be found in other species. Nevertheless, the question concerning the ability for pretense in other species and the nature of underlying mechanisms is, and remains, entirely open. PRETEND AND THE ABILIJY

TO PRETEND

Why would we be so designed that our ability to pretend depends on a concept rather than on a processing mode? There are two points that 1 should clarify before I suggest an answer to this. First, the claim that pretend play depends on the concept, PRETEND, is intended to be an empirical claim. It is not a claim that the nature of pretense makes it logically necessary that we be so designed. I see no reason to suppose that a pretending creature must be designed with PRETEND as one of its concepts rather than with a special mode of processing. However, I believe it is a fact that we are so “designed.” We could conceivably be constructed so that every now and then one of our R enters a special processing mode such that we then representations

Leslie

.

109

“have the pretend that R.” Such an eventuality would be entirely analogous to occasions in which R enters the belief processing mode (for whatever reason) and we find ourselves “having the belief that R.” It is conceivable that pretending might have worked that way. It is further conceivable that, for example, after a period in our lives in which we are thus able to “have pretends,” that is, engage in solitary pretense, we somehow or other come to acquire the concept PRETEND, whereon we become able to recognize pretense in other people and to realize that we can share pretend play with them. Although this is conceivable, and in fact, is the way that we are designed with respect to belief, I don’t think that, as a matter ofempiricazfact, it works that way for pretense. Recognizing this fact leads to a number of fruitful insights about our cognitive organization. Of course, if it is a matter of fact that our ability to pretend depends on the concept PRETEND, then a number of things follow logically. These entailments can be used to help diagnose the presence of the concept, which brings me to the second point I want to clear out of the way. Pretending is a propositional attitude and, in common with other such attitudes, it takes a proposition as its object and this object becomes opaque in the attitude context.2 In my 1987 article, I used this logical consequence to understand some of the representational phenomena underlying early pretend play. I coined the term decoupling to refer to these phenomena. I later developed some of these ideas (Leslie, 1994) to show how pretend play can be understood as an inferencing process in which the inferences “respect” the decoupling structure in pretend metarepresentations. Thus, one might reasonably ask, is such inferencing not a special process that is necessary for pretense? I think decoupling and its related processing are necessary for pretend play but that they are not special to pretense. Decoupling is required for a number of other abilities, including other “theory of mind” abilities such as false belief, and plausibly for various other kinds of hypothetical reasoning, and perhaps even other things-see Cosmides and Tooby’s (2000) interesting discussion. Furthermore, as well as not being special to pretense, I don’t think that decoupled processing is sufficient to give rise to our pretend ability (although it might be for some other creature’s). Now I need to introduce a key distinction that has only been partly explored. There are two divergent, although not unrelated, kinds of ‘The details of the semantics of propositional attitudes don’t really matter to my account, which is just as well because philosophers and others keep changing their minds about exactly what are the semantics. All that matters is that, whatever the semantics of propositional attitudes, that semantics apply to pretending, in the sense of “pretending that P” discussed later.

110

l

7.

Pretense and Representation

pretending. The first is pretending to do X, where X is an action. When I pretend to brush my teeth, what I do is produce a kind of tooth-brushing-like action but the action is fake, not the real thing. In this sense of pretend, the fake action is comparable to pretend candy, for example. The pretend candy is a kind of candy-like thing but, as candy, the thing is a fake. One thing that pretend candy cannot be is real candy. Likewise, I cannot pretend to cut down a tree and by the self-same action reaZZy cut down a tree. The second kind of pretending is pretending that P, where P is a proposition. Interestingly, in the case of pretending that P, P does not have to be false in the way that in pretending to do X, X must be fake. For example, Leslie (1994) showed that 2-year-olds would readily pretend that an empty cup was empty (and for that reason pretend to fill it up again). This suggests that pretending that P is not equivalent to pretending to believe that I? If it was -if pretending that P is really a case of pretending to do X in which to do X is replaced by to believe that P-then the believing would have to be fake and you couldn’t really believe that P at the same time that you pretend that I? This line of reasoning may underlie the intuition that some people have that if you pretend that P, then P must be false. (I have encountered this claim at regular intervals in conversation over the years; an example in print is Perner, 1995 .) You should quickly be able to obtain empirical evidence against this idea if you get an empty cup, pretend to fill it with your favorite beverage, then pretend to pour all the contents over your head, then pretend to refill the cup and then pretend to have a drink. In pretending to empty the cup over your head, I am sure you will pretend that the cup becomes empty and requires refilling. You, of course, will believe throughout that the cup is in fact empty, but this does nothing to block your pretense. Pretending has a complex relation to action. There are many possible sources that a fake X action might have and a pretense-related intention is only one of them (I might be demonstrating a golf swing to you, for instance). The relation between pretense and action has been little studied outside of the Piagetian framework in which the focus was on the increasing complexity of actions the child could string together (e.g., Fein, 1975; McCune-Nicolich, 1981). It wouldbe interesting to see a revived interest in the relation between action and pretense studied within a more contemporary theoretical framework. Actions are movements performed in service of a goal. The goal is determined by the agent who undertakes the action, which means the agent must represent the goal of the action undertaken. The representation of the goal plays a causal role in generating the associated movements (Bernstein, 1967).

Leslie

l

111

An obvious hypothesis about pretend actions is that the goal representation is decoupled. If so, this might explain why the movement undertaken typically does not carry through to the point in the real world that it would normally if it were generated by a regular “coupled” goal representation. For example, if I have a normal goal of drinking from a cup, I will lift the cup all the way to my lips ensuring close contact between cup and lip (for obvious reasons). If I pretend to drink from the cup, typically I will stop short of contact. I may even only outline the action of lifting and drinking in a highly truncated manner of gesturing the cup toward my lips. However, not all movements performed in service of pretend play are truncated or decoupled. For example, if you really did, as I asked previously, pretend to pour a cup of something over your head, then probably you did actually turn the cup upside down. Observations of children pretend playing suggest that full movements and truncated “gesturalized” movements are interspersed. Part of what is going on may have to do with communication if the pretense is shared, as Leslie and Happe (1989) suggested. PRETENDING-THAT

AS MENTAL ACTION

I noted previously that pretending and believing have quite different natural histories. Believing is a state that stems from a mode of processing representations. Eventually in human ontogenesis, the representations that can be belief-processed come to include representations of the believing relation itself, namely, metarepresentations that feature the concept BELIEVE. Pretending, by contrast, does not appear to have a stage in human development in which there is a pretend mode of processing but no concept PRETEND. The human ability to pretend seems to depend on availability of the concept PRETEND. The key evidence in this regard is that there does not seem to be a period in which the human is capable only of solitary pretense but not capable of recognizing pretense in others. There is, however, a clear and prolonged period in which the human is capable only of “solitary” believing-that is, is cais, pable of having beliefs-but is incapable of recognizing belief-that having beliefs about beliefs. The fact that solitary pretending and recognizing pretending in other people emerge together in development strongly suggests that the concept PRETEND is used in both cases. However, it is conceivable that there are two entirely distinct factors: a pretend processing mode and the concept PRETEND, and they just happen to develop at the same time. As a scientist, I find this quite underwhelming, especially when I can account for their yoking by not postulating a pretend processing mode.

112

b

7.

Pretense and Representation

However, there are more than just reasons of parsimony to prefer the unified over a dual factor account. There are further, even more striking, differences between believing and pretending that are accounted for by the unified, but not by the dual, factor account. Having a belief is not something one dues; it is something that happens to one. It is an involuntary state, not something one decides to undertake. For that reason, it is commonly the case that one has beliefs that one is not aware of having. Pretending, on the other hand, falls in the class of voluntary action; one decides to pretend that P or decides not to pretend that P-it is under one’s control and it is not something that happens to one involuntarily. The voluntariness of pretense is nothing more than the ordinary voluntariness with which one, for example, undertakes to lift a cup or to place a marble on the floor. Unlike believing that P pretending that P is not something that simply happens to one, it is something that one undertakes deliberately. And for that reason, one is always aware when one is pretending. Young children, as Piaget (1955) noted, appear to be aware when they are pretending, judging by their “knowing looks and smiles.” The properties of voluntariness, deliberateness, and awareness that accompany pretending-that suggest strongly that pretending-that is a should not be identitype of action. As we saw earlier, pretending-that fied with either pretending to do X or with pretending to believe-that. It belongs to a distinct class of mental action in which an agent deliberately, voluntarily, and with awareness undertakes to hold a specific attitude toward the truth of a proposition. In the case of pretending, we undertake to hold the attitude of pretending that some proposition is true. Pretending is not the only mental action of this kind that we can undertake. We can also suppose, consider, imagine, plot, memorize, and so forth. Deliberate-that is, goal directed-external, physical actions require the representation of their goal. Likewise, deliberate, internal, mental actions also require a representation of their goal. Deliberately undertaking the external action of tying laces requires representing the goal of that action as one of tying laces and therefore requires having the concept, TIE LACES. Likewise, deliberately undertaking the action of pretending that P requires representing the goal of that action as pretending that I? And for this reason, the child who deliberately pretends uses the concept PRETEND-THAT in his goal representation. Naturally, this is also the concept required for representing the mental state of another person who is pretending-that. Recent findings on the existence of “mirror neurons” (e.g., Rizzolatti et al., 1996) suggested that the brain uses the same representation for one’s own action of grasping an object as for representing a similar

Leslie

’

113

grasping action performed by someone else. The same neurons are active in carrying out that action as in observing it being carried out by someone else. The simplest thing to assume is that the activity of the neurons in question in some way reflects the use of the concept GRASP. On this assumption, it is no surprise that GRASP refers to grasping, no matter who carries it out. A similar brain organization may underlie pretending. That is, we should expect that there will be mirror neurons for pretending-that. However, there is no particular reason that I can see for expecting that mirror neurons for pretending must be found in the motor cortex rather than somewhere else. The central idea of the metarepresentational theory of pretense still seems to me to be a fruitful empirical hypothesis. The human ability to pretend depends on the availability of the concept PRETEND. This concept allows the child both to pretend by himself or herself and to recognize pretense in other people. In both cases, the brain is, in a sense, reporting on a mental state. That is what ties pretending to “theory of mind” and makes the study of pretense such an intriguing case study of an early appearing, yet highly abstract, concept. Understanding pretense still poses a deep challenge to past and current attempts to understand the nature of early developing abstract concepts.

Bernstein, N. (1967). The coordination and regulation of movements. Oxford:

Pergamon.

Cosmides, L., & Tooby, J. (2000). Consider the source: The evolution of adaptations for decoupling and metarepresentations. in D. Sperber (Ed.), Metarepresentations: AMuZtidiscipZinary~rspective (pp. 53-115). Oxford: OUP Fein, G. G. (1975). A transformational analysis of pretending. Developmental Psychology, 11, 291-296. Haith, M. M. (1998). Who put the cog in infant cognition? Is rich interpretation too costly? Infant Behavior & Development, 21, 167-l 79. Leslie, A. M. (1987). Pretense and representation: The origins of “theory of mind.” Psychological Review, 94, 4 12-426. Leslie, A. M. (1994). Pretending and believing: Issues in the theory of ToMM. Cognition, SO,21 l-238. (Reprinted from COGNl77ONon cognition, pp. 193-220, by J. Mehler & S. Franck, Eds., 1995, Cambridge, MA: MIT Press.) Leslie, A. M., & Happk, E (1989). Autism and ostensive communication: The relevance of metarepresentation. Development and Psychopathology, I, 205-2 12. McCune-Nicolich, L. (198 1). Toward symbolic functioning: Structure of early use of early pretend games and potential parallels with language. Child Development, 52, 785-797. Perner, J. (1995). The many faces of belief: Reflections on Fodor’s and the child’s theory of mind. Cognition, 57, 241-269. Piaget, J. (1955). The child’s construction of reality. London: Routledge & Kegan Paul.

114

l

7.

Pretense and Representation

Rizzolatti, G., Fadiga, L., Gallese, If., & Fogassin, L. (1996). Premotor cortex and the recognition of motor actions. Brain Research. Cognitive Brain Research, 3, 131-141. Whiten, A., & Byrne, R. W. (1988). Tactical deception in primates. Behavioral and Brain Sciences,I 1, 233-273.

Early Concepts and Early Language Acquisition: What Does Similarity Have to do With Either? Laraine McDonough

Brooklyn College and the City University of New York Graduate Center

R

esearchers have traditionally investigated children’s first words to understand the nature of early cognitive development, but the relation between first words and early concepts turns out not as straightforward as hoped. Over the past 15 years, Jean Mandler and I have examined concept development using various nonverbal measures. We knew that at some point we would have to address the topic of language acquisition because our findings contrasted sharply with the assumption that basic level concepts provide the foundation on which basic level nouns are first understood. However, our time was consumed with research designed to untangle the perceptual and conceptual aspects of early categorization abilities in infants and young children. I fondly regard these experiments as works in progress. More recently, I began to focus on how language and early concepts embellish each other. Armed with the implications of my research with Jean Mandler, I set out to uncover what young children understand about the meanings of the early words they learn. As will be dis115

116

l

8.

Early Concepts

and Early Language

cussed, many researchers have made assumptions about early language use that have not been formally tested, many have designed experiments that confound the bases children use for learning the meanings of their first words, and others have used clever but impractical stimuli to tout the role of similarity in children’s generalizations of basic level terms. But first, a brief overview of our research on preverbal concepts is in order. If I wanted to test the meanings young children entertain when using their first words, I needed to consider what they understand in the first place.

For most who investigate early cognitive development, the role of similarity has been treated with considerable reverence despite the fact that similarity has little a priori value. However, the construct still packs a mean punch after the data are collected and the skeptical reviewers put on their magnifying glasses and see little in infants’ minds and plenty in the researchers’ stimuli. After all, similarity can be a flexible and unreasonable beast (Goodman, 1972). Flexible because we are able to search for similarity in situations where we otherwise would not spontaneously find it. For example, adults are happy to judge that seals are more similar to three-wheel all-terrain vehicles than they are to buses. Their judgments (revealed by self-reports) are based on the observation that seals’ bodies are supported by the ground in three places (two flippers and the lower part of the body), which is similar to the manner in which all-terrain vehicles are supported on the ground via their three wheels. Adults will also rate cows as more similar to chickens than hyenas because cows and chickens are farm animals, but will then turn around and rate cows and hyenas as more similar because they stand on four legs. Thus, similarity can be influenced by gross structural similarities (e.g., three points of ground contact), a structured knowledge base (e.g., farm animals), or overall shape similarities (e.g., items with four legs). Similarity arguments can be unreasonable because views about what constitutes similarity have changed markedly in the past 15 years and are thought to be sensitive to contextual influences and observers’ strategies. When my research with Jean Mandler first began, similarity among objects was commonly defined in terms of their overall shapes (Kemler, 1983; Rosch, Mervis, Gray, Johnson, & Boyes-Braem, I 9 76). To some extent, this view remains, particularly in the research on language acquisition (e.g., Imai, Gentner, & Uchida, 1994). However, when my research with Jean Mandler and Patricia Bauer began to show that young children and infants could categorize what were commonly called “superordinate” categories yet at the same

McDonough

l

117

time had difficulties discerning basic level categories (Mandler, Bauer, & McDonough, 199 1 ), others began to voice concerns about similarity. It would seem infants were not to be privy to concepts. Instead, views on what constituted similarity were supposedly in need of updating. In fact, when Jean Mandler and 1 began our work using the object examination task, a task originally used to test infants’ abilities to distinguish geometric forms (Ruff, 1986), we too had our doubts. However, the results from the object examining task showed categorization of animals, vehicles, furniture, and plants in infants as young as 7 to 9 months with chance performance on tasks contrasting basiclevel tasks such as chairs versus tables, and dogs versus fish (Mandler & McDonough, 1993, 1998a). We proposed that infants were exhibiting conceptual behaviors using the task. However, given that the technique was partly derived from research in which infants’ attention to geometric forms was tested (Ruff, 1986; see also, Ross, 1980), and that geometric forms in and of themselves do not capture the nature of what we commonly think are concepts, interpretation of our data was not as straightforward as we desired. If infants were making such distinctions based on what they knew about these categories, what precisely was it that they knew? Was it possible that similarity was guiding our findings even though we could not find obvious systematic shape or featural commonalties among our stimuli? Our work continued in two directions. First, we set out to test a perceptual explanation of our findings with the help of Peter Eimas and his student, Jonathan Boutelle. Second, we designed a new technique that would bring us closer to figuring out what infants knew about the roles objects play in everyday events. To check to see if the stimuli we used in our categorization tasks could be resolved via perceptual processes alone, we had professional slides made of each exemplar. We sent the slides to the Peter Eimas because he and his colleagues are experienced in testing 3- to 4-month-olds using a preferential-looking technique with pictured stimuli. Consistent with the previous work from the Eimas lab showing categorization of various basic-level categories (e.g., Eimas & Quinn, 1994), the results contrasted with our own in that they did not find clear evidence of categorization of animals and vehicles (Boutelle & Eimas, 1995). Thus, the differences between our findings and theirs were not due to anything unusual about the stimuli used by our labs. Given that the older infants in our tasks showed the opposite pattern of results (succeeding on the animal vs. vehicle task but not on all basic-level tasks), it would appear that different processes are operating at different periods in development in the two different tasks. That is, young infants are limited to responding based on what the depicted ob-

118

l

8.

Early Concepts

and Early Language

jects looked like. After all, 3-month-olds have little real-world experience with objects such as dogs, cats, fish, or birds. More recent research on categorization using the preferential-looking paradigm shows that even 3- to 4-month-olds can go beyond overall shape similarity to categorize perceptually variable mammalian forms that exclude fish and birds (Behl-Chada, 1996). They also can resolve the fine details of the differences in facial structure of domestic cats and categorize dogs and cats using head and facial attributes (Quinn & Eimas, 1996; Spencer, Quinn, Johnson, & Karmiloff-Smith, 1997). Thus, views of similarity in terms of the overall shapes of items have been challenged. The infant perceptual system is more sophisticated than we expected. Similarity slipped in another wrench. If that slippery notion of similarity was guiding categorization in the object examination task, then it is possible that similarity is being judged in a different way at 7 months than at 3 months. One might consider that things that are alike look alike (although the saying is usually the other way around). Perhaps the more infants realize that animals or vehicles are the same kind of thing, the more animals begin to look like each other and vehicles begin to look like each other. In this case, it is possible that conceptual processes are influencing similarity. However, Van de Walle (1999) showed that even though infants do not show basic-level distinctions such as between horses and pigs using the object examination task, they will distinguish objects based on perceptual characteristics such as color. Thus, successful performance on the object examination task can be accomplished on perceptual grounds alone. Clearly, another measure of concept development was needed. Inductive generalization was a technique borrowed from the research on event representation and symbolic play. For example, studies by Bauer and her colleagues (Bauer & Dow, 1994; Bauer & Fivush, 1992) showed that young children not only remembered novel event sequences, but they could also generalize what they learned to other categorically related objects. Research on symbolic play by Killen and Uzgiris (198 1) showed the much younger infants would use appropriate but not inappropriate objects to imitate familiar actions. For example, infants would use a telephone receiver but not a car for pretending to talk on the phone. Thus, imitation of events could tell us if infants knew the appropriate functions of objects and could also allow us to more formally examine what kinds of “fillers” could be placed into the various “slots” in events (see Lucariello & Nelson, 1985). The results from our inductive generalization experiments were clear. As evidenced in their imitations, infants generalized drinking to all animals including fish ( !) and birds. Infants also generalized a keying action demonstrated on a car to all vehicles, even airplanes. They

McDonough

l

119

did not, however, generalize a drinking action to a vehicle nor did they generalize a keying action to an animal. Furthermore, they would not imitate such activities when we modeled them, a replication of Killen and Uzgiris (1981). We then conducted a number of experiments using this technique and varied the properties tested and the exemplars for generalization. With few exceptions, basic-level categories, such as those found using habituation or preferential-looking tasks (e.g., Eimas & Quinn, 1994; Younger & Cohen, 1986) appear not to be used by infants to guide their inductive inferences, even when basic-level properties are tested. For example, on showing a basic-level property such as a dog chewing on a bone or a bird sitting on a nest, 14- and 19-month-olds were content to place other animals on a nest and even to give a bone to a bird. They are also content to imitate sleeping with bathtubs as often as with beds and drinking with fry pans as often as with cups. Not until around 18 to 24 months of age do infants begin to limit their generalizations to basic-level categories. Thus, the initial pattern of generalization of basic-level properties is one of overgeneralization, a pattern also found in early language use (Mandler & McDonough, 1998b, 2000). LANGUAGE

ACQUISITION:

A FEW UNANSWERED

QUESTIONS

With a focus on children’s correct extensions of early words, researchers have been impressed with the expertise children seem to display early on in the language game. Overall, the data appear to support a developmental view that the majority of children’s first words are nouns best described as representing “basic-level” categories. Basic-level terms are used almost exclusively in child-directed conversations, so it makes sense that such labels would be the first to be acquired (Anglin, 1977; Brown, 1958). As tradition would have it, basic-level categories were also thought to be perceptually salient (e.g., Rosch et al., 1976). Because basic-level categories were thought to be the easiest to derive perceptually, they also were thought to be easily learned. Thus, first-learned words were thought to correspond with first-learned categories, leading one to believe that language acquisition is a reasonably good indicator of early cognition. However, do children use perceptual cues such as global shape to guide their generalizations of nouns? Or do they base their generalizations on their conceptualizations of various taxonomic relations? Given the recent evidence that early conceptualizations are more globally defined (e.g., Mandler & McDonough, 1998a, in press; McDonough & Mandler, in press), there is reason to suspect that first-learned words may not match (in the adult sense) early meanings.

120

l

8.

Early Concepts

and Early Language

When researchers switch their focus from correct language use to the errors children make, they find that errors can constitute as much as 30% of a child’s language (Rescorla, 1980), many of which are overgeneralizations/overextensions. An examination of the various kinds of errors children make in early language acquisition highlights the complex mapping involved in terms of coordinating cognition and language. These findings, based primarily on production data, add to the evidence against earlier assumptions that early cognition is most accurately described in terms of basic-level categories. Why Does the Child

Overextend

Nouns?

A child may label a skunk “cat” presumably on the basis that skunks and cats share a number of perceptual features. One interpretation of this kind of overextension is that it reflects a performance strategy of substituting a word which the child feels is similar enough to the meaning that he or she wants to express (Bloom, 19 73). That is, the child knows the word “cat” but has not yet learned the word “skunk.” This view implies that the child knows the differences between cats and skunks but is forced to ignore such differences so that communication needs can be met. Alternatively, such use may indicate that children believe that a skunk is an exemplar belonging to the same category as cat. That is, the child’s meaning of the word cat is not the same as the adult meaning, but is less specified and more general (e.g., Clark, 1973; Mervis, 1987). According to an earlier proposal made by Clark (1973), such underspecified meanings allow overextensions that may seem to be appropriate within the child’s linguistic system. First words may initially have only a few features of meaning (e.g., “cat” may include the feature “four legs” but not the more definitive features of cats such as purrs or meows), and extensions become established (or pruned away) through the gradual accretion or analysis of features which differentiate a particular concept from other related concepts. In this view, all concepts will be overgeneral until the set of contrasting concepts within a domain is formed. However, empirical support for a “feature accretion” view has been problematic (cf. Barrett, 1982; Carey, 1982) and has since been abandoned by Clark (1983). For example, parts which may be considered as defining or criteria1 features are not treated as such by young children when they are removed from one object or added to another object (see Poulin-Dubois, Graham, & Riddle, 1995). The term “bird” is comprehended in conditions in which a bird is shown either with or without wings, but the term “bird” is not comprehended when wings are placed on another kind of animal. Like similarity, criteria1 features are also difficult to pin down.

McDonough

l

121

Another account of overgeneralizations is that the basic-level categories of young children differ from those held by adults (Golinkoff, Mervis, & Hirsch-Pasek, 1994; Mervis, 1987; Mervis & Mervis, 1982). The reason for this difference is that children, whose expertise falls short of that of adults, differentially weigh attributes or functions of category members. Thus, their “similarity metric” is calculated differently so that what is the “basic level” for a child can be much different than the “basic level” for an adult. In principle, this view agrees nicely with the interpretation of the changing nature of similarity discussed in the previous section. Yet, according to Mervis and her colleagues, early extendibility of words can be accounted for in a number of ways including similarity of shape or features. Over time shape is thought to predominate in extensions because children learn to recognize that shape is most frequently associated with same name referents (Golinkoff et al., 1994; Landau, Smith, & Jones, 1988). However, prior to gaining extensive linguistic experience, the basis on which categories are formed and basic-level terms are extended has been controversial. Basic-level categories as traditionally defined are problematic in that “basic level” in the adult sense is not an accurate description of children’s categories when first acquiring language. To add to this problem, we have yet to assess adequately what children think basic-level nouns mean. Another frequently reported finding in production data is that children underextend some words, using them in more limited contexts than adults would (e.g., Clark, 1983). For example, Bowerman’s daughter Eva first used “off V only in the context of removing clothing and other items from the body (Bowerman, 1978). Tests of children’s comprehension of underextensions are difficult because we frequently do not know the exact context in which a word is learned and, because of this, research is scarce (however, see Reich, 1976). Underextensions may occur because children cannot remember the term in new contexts (a production problem) but would recognize its appropriateness if it were used by others (not a comprehension problem). What

Does the Child

Comprehend?

Although no clear answers have been directly forthcoming as to the initial meanings of nouns such as “cat” (however, see Naigles & Gelman, 1995, to be discussed later), researchers have generally assumed that toddlers think such terms mean roughly the same as they do for adults. Indeed, the focus in recent years has been on the extent to which children obey constraints on word meaning from the onset of lexical development (e.g., Backscheider & Markman, 1990; Huttenlocher & Smiley, 1987; Macnamara, 1982; Waxman & Gelman, 1986). However, the

122

l

8.

Early Concepts

and Early Language

claim that children understand words in the same way as adults remains a matter of considerable debate. The kind of assumptions researchers typically make about early language comprehension become apparent when one examines the design of language comprehension tasks. In tests of language comprehension, the young child’s job is to select from various items presented the correct referent for a particular word. The distractor items are typically selected from a different superordinate category from the correct referent. For example, in a study on lexical development in which comprehension of 75 terms were assessed, Reznick and Goldfield (1992) contrasted the correct referent with a distractor from the same superordinate on only three occasions. In their examination of whether or not children know the meaning of words such as “shoe” (a clothing item) or “bee” (an insect or animal), the correct referents were contrasted with “banana” (a food item) or “motorcycle” (a vehicle). It is possible that the contrast items were chosen so as not to be too difficult for young children. However, such contrasts do not tell us whether basic-level categories provide the boundaries for the early meanings of basic-level terms. It is possible, but has not been formally tested, that infants would initially comprehend the meaning of banana as “food” and dog as “animal.” Rescorla (1980) found that overextensions in production are most often made with early acquired words and are made within superordinate boundaries (see also Anglin, 2 977; Bloom, 19 73; Clark, 19 73). Some data on comprehension are available, but how to interpret these data is not straightforward. For example, Behrend (1987) used a preferential-looking measure to assess comprehension of object labels. One label was tested per individual for eight trials. On half the trials, the correct referent was displayed along with three distracters, one of which was an item judged to be perceptually similar to the correct referent and whose label was unknown by the individual (according to parental report). On the remaining half of the trials, the correct referent was not available; the similar item was presented with three nonsimilar, unrelated distracters. However, many of the similar items were also taxonomically related; in fact, 8 of the 10 perceptually similar distracters also reflected a taxonomic relation of a superordinate category (e.g., animate objects, vehicles, food). When parents instructed their infants to find an item that was not present (but a similar one was), all but one of the infants showed a tendency to look at the taxonomically-related or perceptually similar distractor items. One interpretation of this finding is that toddlers’ comprehension of basic-level terms is overgeneral in that basic-level boundaries do not constrain generalizations (cf. Kuczaj, 1982). Another is that toddlers looked at the similar item only because the correct one was not available.

McDonough

l

123

Naigles and Gelman (1995) also tested overextensions of words on trials in which the correct referent was not presented (i.e., anomalous trials). They used a preferential-looking paradigm to examine comprehension of three animal terms: dog, cat, and cow. On hearing the instructions to look at the cat during anomalous trials (i.e., a dog and cow were shown but no cat was available), participants showed no looking preferences: looking to the dog and cow was the same. This finding was thought to reflect correct comprehension of the term cat in that it was not overextended to include dogs or cows. However, on the anomalous trials in which children were shown a dog and cat and were directed to find a cow, they showed a preference to look at the dog. Thus, the data suggested that children might have overextended the term cow to include a dog. Naigles and Gelman explained that this finding of overextension seemed attributable to similarity but was not indicative of “categorical” (i.e., basic-level) immaturity (as proposed by Mervis, 1987). However, given that it is typically thought that early categories are based on perceptual similarity, it remains an open question as to whether such an overextension is similarityor categorybased, because in this case both bases lead to the same prediction. Although the results from both these experiments suggest overgeneralization in comprehension as shown in the anomalous trials, it remains possible that infants knew that the correct referent was not available and simply looked at the next best exemplar when one was available (e.g., Kuczaj, 1982; Mervis & Canada, 1983). It is likely that once infants comprehend a basic-level term they would prefer the correct referent but nevertheless be willing to include other objects that were either from the same domain or were similar to the real referent. That is, they don’t really believe that cows and dogs are labeled with the same term but, as suggested by Naigles and Gelman (1995), they do recognize that the two are similar. PRODUCTION

AND COMPREHENSION

REVISITED

All in all, the previous research is ambiguous in terms of young children’s comprehension of nouns. The research reported below examines more extensively children’s overextension of nouns commonly described as “basic level” (according to the perspective of an adult speaker) in both production and comprehension. In these experiments, anomalous trials are not used: Overextension is tested in the presence of the correct referent to the target item. The nouns tested are drawn from the domains of animals, vehicles, clothing, and food. Distractor items are drawn from either the same domain as the target or from a contrasting domain. One group of toddlers (between 22 and 25 months of age) was

124

l

8.

Early Concepts

and Early Language

tested on nouns typically learned very early in language acquisition (around 16 to 18 months). A second group of the same age was tested on nouns learned later in development (nouns not found on early language assessment questionnaires and appearing with low frequency in children’s books). It was predicted that if children are mapping basic-level nouns to the shapes of items, then their selections would be correct and they would not make overgeneralizations outside the boundaries of basic-level categories. However, if children were mapping basic-level nouns to undifferentiated categories, then one would expect overgeneralizations even though the correct referent is always available. If overgeneralizations are found, it was expected that more overgeneralizations for the early than the late learned nouns would be found, because once the late learned nouns are comprehended, there should be no reason to overgeneralize either noun due either to mutual exclusivity (i.e., children initially expect that each object has one label only; Markman & Wachtel, 1988) or the principle of contrast (i.e., children assume that there are no synonyms in language; Clark, 1987). Stimuli

for Testing

bsic-Level

Nouns

Sixteen test cards were constructed, each with three realistically colored pictures of objects from the animal, vehicle, food, and clothing domains. Eight objects were selected as test items within each domain. Four of the eight objects in each domain were chosen because they had labels that are among the earliest comprehended and produced by toddlers (i.e., typically comprehended around 15 months of age and produced at 20 months of age; Fenson et al., 1994). These are referred to as “early items.” The remaining four objects had labels that are typically learned somewhat later in development (i.e., “late items”). All but one of the pictures (early and late) were selected from Pictures Please, a language book used to test children’s language production and comprehension (Abbate & LaChappelle, 19 79). Each of the pictures was colored in by hand using colored pencils. The colors were selected by how natural they were for each pictured object. A complete listing of the pictures and how they were grouped onto the 16 test cards is shown in Table 8.1. The first page in Booklet A on which is displayed a pig, train, and bus will be used as an example to describe how the cards were constructed. ‘Ityo of the three pictures on each card were selected from the same domain (e.g., train and bus), one of which was an early item (i.e., train) and the other a late item (i.e., bus). The third object, also an early item, was from a different domain (e.g., pig). Placement of the target items on each card was balanced for left, right, and center positions. ‘Trials in which participants were asked

McDonough TABLE

Test Booklets

Early Noun (High Contrast) Booklet cow Carrot Orange Dog

Apple Cat Egg

Early Noun (low Contrast)

late Noun

l%ain Pants Shoe Bicycle Car Shirt

Bus Shorts

Dress Airplane

Sandal Motorcycle Truck Vest Sweater Rocket

B Pig

Bicycle

cow Carrot Orange

*ain Aimlane

8.1

Used for Testing Production and Comprehension of the Early and late Nouns

Shirt Boat Pants Dress Car

125

A

Pig

Booklet

l

Dog Apple Cat Cake

Hippo Moose Celery Beet Fox Strawberry Raccoon Pie

to label or point to the target that was contrasted with two items from another domain (i.e., the pig was requested) were called high-contrast trials. Trials in which participants were asked to label or point to the target that was paired with the distractor from the same domain (i.e., the train was requested) were called low-contrast trials. Participants were first given a production task, after which their comprehension was tested. The opposite order (i.e., comprehension then production) was not given because of the probable influence the experimenter’s use of the words in the comprehension task would have on later production. For example, on hearing the experimenter ask the child to point to a dog, the child may be more willing to produce the word “dog” to one of the items later on. Half the participants were tested using Booklet A, the remaining half were tested using Booklet B.

126

l

Production

8.

Early Concepts

and Early Language

Task

The experimenter showed each card and allowed the child to inspect each individually (e. g . , saying, “Look at these pictures!“). After the child had ample time to look at all three pictured objects on a given card, the experimenter pointed to each picture and asked for its label (e.g., saying, “What is this one?“). If the child did not respond, the experimenter would pretend to guess incorrectly (e.g., saying, “Is it a cup? NO? What is it?“), thus encouraging participants to state the correct label. All responses, correct or not, were praised with comments from the experimenter (e.g., saying, “So that is what that one is called! Good job.“). If the child mislabeled an item, no correction was made. A total of 26 children (ages 22 to 26 months) provided production data. Children labeled an average 16.3 pictured objects (range4 to 24). A label was considered correct if it was the basic-level term commonly used for the object or the correct sound associated with the object (e.g., cat, kitty, and meow were all considered correct labels for the cat; train and choo-choo were considered correct labels for the train). A label was considered an overgeneralization if the label was incorrect but from the same domain as the referent (e.g., the raccoon was called a “cat”). Other incorrect labels (e.g., calling the egg “table”) or inaudible labels were scored as “other errors.” Correct labels were given for 261 items (62% of the total labels produced), overgeneralizations were given for 122 items (299/o), and other error types were given for 39 items (9%). Note that the percentage of overgeneralizations is similar to the previous research on early language production (Rescorla, 1980). Only rarely were superordinate terms offered (one child labeled the moose as an animal; one labeled celery as salad). The data were analyzed by comparing the labels that were produced for the early and late items. Because twice as many early as late items were tested (see Table 8.1), proportion scores were entered as the dependent measure. The proportion of early nouns was calculated out of a maximum of 16 possible; the proportion of late nouns was calculated out of a maximum of 8 possible. As shown in Fig. 8.1, correct labels were generated more often for the early (.60) than late items (.06). This finding is consistent with data obtained from parent inventories by Fenson and his colleagues (1994) and verifies that the division of the pictures into the early versus late items is correct. More overgeneralizations were made for the late (.33) than the early items (. 14), p < .05. (Note that proportions for each category were calculated for each participant and the average proportion scores are presented. No credit was given for items not labeled.) This finding indicates that overgeneralizations can occur for the referents of first words children

.

i-

0.8

McDonough

-

l

127

pzz&iEj

.

0.6

0.4

0.2

0 EARLY ITEMS

FIG. 8.1 Production tion scores).

LATE ITEMS

data of early and late nouns

(based on mean propor-

enter into their lexicons, but more often occur for the referents they learn to label later in development. That is, overgeneralizations occur more often to the items for which a label is not known. Other errors made for the early (.07) and late (.04) items did not significantly differ. For the early items, significantly more correct labels than overgeneralizations or other error types were made. Overgeneralizations and other error types did not differ significantly. A different pattern was found for the late items in that significantly more overgeneralizations than correct labels or other error types were made. These results not only replicate previous reports of overgeneralizations in spontaneous speech (e.g., Rescorla, 1980), but also suggest that overgeneralizations are the most probable type of label children will offer in situations in which the correct label is not part of their productive vocabularies. The data were then broken down by domain (e.g., animal, vehicle, food, clothing) and analyses were conducted to determine if any differences in the number of correct and incorrect selections would be found among the four domains. No significant differences were found in either the first choice data or the first and second choice data combined. Also, no significant differences were found between the two test booklets. This is a consistent finding across the studies reported later, as well.

128

l

8.

Early Concepts

Comprehension

and Early Language

Task

As in the production task, children were tested individually. The comprehension task was administered by asking children to place a sticker (a small post-it note with a squiggle on it) on one of the objects. Trials were ordered so that a high contrast trial was followed with a low contrast trial and so on until the child had seen the test booklet twice and had responded to both the high and low contrast items on each card. If the child became frustrated or bored with this technique, the experimenter ended the session (see results section for average number of trials completed). Because the procedure for the high- and low-contrast trials was the same, a comparison of participants’ selections could be made between them. It was predicted that if children were comprehending the nouns in the same way they produced them, then more errors in terms of overgeneralizations would be made in the low-contrast trials because participants would be more willing to generalize basic-level terms to items in the same domain (e.g., generalize a basic-level noun for a vehicle to another vehicle) than to contrasting domains (e.g., generalize a basic-level noun for a vehicle to an animal). Nouns

Learned

Early in Development

The comprehension data for the early-learned basic-level nouns are based on 16 children (ages 24 to 26 months), 14 of whom contributed to the aforementioned production data. The early items are those listed in the left and middle columns on Table 8.1. It was reasoned that if errors in production are primarily due to retrieval failures, then few errors in comprehension would be found, particularly for nouns learned early in development. If, however, comprehension and production follow the same general acquisition pattern (Mervis & Pani, 1980; Reznick & Goldfield, 1992), then errors in comprehension can be expected to approximate those found in production. Out of 16 test items, participants completed an average of 14.8 comprehension trials (range- 9 to 16 trials; high-contrast trials-M = 7.4; low-contrast trials-M = 7.6). Participants’ selections were made more often to correct referents (M = 12.2) than incorrect ones (M = 2.7). As can be seen in the leftmost columns of Fig. 8.2 (percentage data are displayed so that they are somewhat comparable with the mean proportion data given in Fig. 8.1), more correct selections were made in the trials (M = 5.3; 709/o), p < high- (M = 6.9; 93%) than the low-contrast .Ol, and more incorrect selections were made in the low- (M = 2.3; 30%) than the high-contrast trials (M = .5; 7%), p < .Ol.

McDonough

l

129

80 -.

40

-8

20-

O-

EmLYNouN!3

1

HIGHCDNIFUW

FIG. 8.2

Comprehension

lATENolJNs

LCNV(CXlNRW

of early and late nouns

EARLY NOUNS: HIGH-

(percentage

data).

Next, the kinds of errors made were examined. In the low-contrast trials participants could err in two ways. They could choose the distractor item from the same domain as the referent (i.e., they could make an overgeneralization) or they could choose the distractor item from the contrasting domain as the referent. The data showed the ma(M = jority of errors in the low-contrast trials were overgeneralizations 2.2 out of 2.3; p < .Ol). For example, on trials in which the train was requested, participants were more likely to err by selecting the bus than the pig. Almost no incorrect selections were made on the high-contrast trials (M = 0.5). Nouns Acquired

later

in Development

The same test booklets were used to assess comprehension of the nouns acquired later in development. Eight participants were included in the data presented, six of whom contributed to the aforementioned production task. To make sure they would not find the task too difficult or too discouraging, participants were asked to

130

.

8.

Early Concepts

and Early Language

point to the referent of the early items in the high-contrast trials (i.e., the items listed in the left column in Table 8.1). In the low-contrast trials, participants were asked to point to the referent of the basic-level term that is commonly learned later in development (i.e., the items listed in the right column in Table 8. I). Given that comprehension typically proceeds production, it remains possible that children might know the correct referents to the nouns learned or produced later in development. In this case, they may point to the correct referents without making errors. Alternatively, they may make an inference as to the meaning of the late nouns because it is clear to them what the labels are to the other two contrasting items, both of which are early-learned nouns and prevalent in their production vocabularies (however, see the section later on nonce words). Or, they may overgeneralize the late nouns, thus showing some overlap in the meanings of the early and late nouns. This overlap in meaning would reflect their general knowledge as to the domain to which early and late nouns belong. The results showed that participants made significantly more cortrirect selections in the high- (M = 7.5; 100%) than the low-contrast more incorrect als (M = 5.1; 71%; p < .Ol ) and made significantly trials selections in the low- (M = 2.1; 29%) than in the high-contrast (M = .OO, p < .05). An examination of the incorrect selections made in the low-contrast trials indicated that participants made significantly (M = 1.63) than other errors (M = .50, p < more overgeneralizations .05). This finding was somewhat surprising in that the pattern of overgeneralization between the early and late items is symmetrical (see Fig. 8.2). That is, not only are early-learned terms overextended to items whose labels are learned later in development (e.g., the term train is extended to include a bus), but that later-learned terms were also overextended to the items whose labels were learned early in development (e.g., the term bus is extended to include a train). The results of these experiments challenge the view that children’s first word extensions are primarily based on perceptual features. For example, older children and adults treat airplanes as a basic-level category from which nonairplanes are excluded, but the younger children in these experiments included rockets (displayed in an upright launching position without wings) as acceptable “airplanes.” In the same fashion, vests were considered to be shirts even though they did not have sleeves or collars; hippos were called pigs even though their faces are shaped differently and they do not have curly tails (it should be noted that hippos were also called cows); and red strawberries and green apples are both called apples even though they differ in overall texture, shape, and stem configuration.

McDonough

WHAT NONCE

LABELS TELL US ABOUT

l

131

POSSIBLE MEANINGS

The question arises as to how one compares the results from the aforementioned experiments to the experiments in which nonsense or nonce words are used. Nonce labels in research with younger children (I 5 to 33 months of age) were used to test which hypotheses children entertain when first learning a new term (e.g., Bauer & Mandler, 1989; Waxman & Hall, 1993). Children are shown one object which is labeled with a nonce term (e.g., “fep”). They are then asked to choose another object using the same label (e.g., “Can you find another fep?“). The objects from which they can select are taxonomically related, thematically-related, or unrelated objects that share a perceptual feature with the example. In contrast to the prediction that young children will extend labels to thematic relations (e.g., Dromi, 1987; Smiley & Huttenlocher, 1995), the results from these tasks indicate that children as young as 15 to 16 months of age will more often extend the nonce label to a taxonomic than a thematic item (Markman & Hutchinson, 1984; Waxman & Hall, 1993). This kind of selection provided evidence for what is known as a “taxonomic” bias. The taxonomic bias (constraint or principle) is thought to reflect the child’s belief that words refer to whole objects of a particular kind (Waxman, 1990). However, a frequent finding in this research is that children are willing to extend a nonce label not only to basic-level category exemplars but to other items from the same superordinate class as well (Bauer & Mandler, 1989; Waxman, 1990). If perceptual similarity (either in terms of overall shape or individuated features) is used as a cue for early category formation (in 2-year-olds and younger), then experiments using nonce labels should find superior performance on basic-level compared to superordinate-level tasks. That is, if young children prefer to use similarity for their extensions, then they should also prefer basic-level exemplars in their extensions of novel labels. However, such results have not been found. One unresolved issue in the research using nonce labels concerns a “translation hypothesis.” Just because a nonsense word is presented along with a previously unnamed object does not mean that the child has no idea what to call the object in the first place. On hearing an item labeled “fep” for which the child perhaps already has a label (e.g., dog), does the child simply translate the term fep to mean dog? According to the interpretation that “fep” is readily extended to include other animals (not just dogs), and the assumption that dog is only extended to include dogs, then fep is not simply translated to mean dog but something more like animal, a term that does not enter a child’s lexicon until much later (Waxman, Senghas, & Benveniste, 1997). Yet, as shown in

132

l

8.

Early Concepts

and Early Language

the previous experiments, the meaning that young children initially hold for the term dog is also more widely extended to other animals as well. This finding is consistent with research showing that young children (before the vocabulary spurt) are apparently comfortable having two labels for the same object (Mervis, 1987; Waxman & Hatch, 1992; Waxman & Senghas, 1992). Comprehension

of late-learned

Nouns

and Nonce

Labels

Nonce labels are not only useful to examine the meanings children initially consider but also to test inferences they make during language acquisition when they hear a novel noun in the context of a previously unnamed object. For example, it is possible that a correct selection of the referents to the late nouns in the previous study (listed on Table 1) were attributable to a “fast mapping” strategy (e.g., Au & Markman, 1987; Carey, 1982). That is, participants may point to the bus when it is requested not because they know that “bus” is the correct term for that item, but because they know the correct terms for the other items (e.g., train and pig). Then they tended to use mutual exclusivity (Markman & Wachtel, 1988) to infer that the correct item (i.e., bus) is the one requested. To test for this possibility, nonce labels were used to test another group of participants (e.g., fep: Where is the fep?). In the low contrast trials, participants were asked to point to the referent of a nonce word (e.g., fep). To make sure they would not find the task too difficult or too discouraging, participants were asked to point to the referent of the early items in the high-contrast trials (i.e., the items listed in the left column in Table 8.1). A comparison between the selections to the nonce words and the late nouns should indicate whether the late nouns are comprehended. If the late nouns are unknown and mutual exclusivity guides responses, then no significant differences between the nonce and late-noun trials would be expected. Eight children participated in this study, six of whom contributed to the aforementioned production task. A comparison among these children and those who were tested on the late nouns showed that, based on their production data, the two groups were comparable (correct labels for late items 0.5 vs. 0.4). It should first be pointed out that these participants frequently refused to make selections on the comprehension task when novel labels were used. Out of 58 trials using the novel labels, participants said “no” to 23 of them (40%). This is considerably greater than the percentage of “no” responses made on the high-contrast trials (16%; p < .05). It would appear that participants frequently did not know how to interpret these terms and were unwilling to guess.

McDonough

l

133

The data for the remaining 35 trials (in which participants were willing to guess) were scored as late item (e.g., rocket), early item, (e.g., airplane) and contrast (e.g., egg). As predicted, the results showed that participants tended to make more selections to the late items (M = 2.13; 47%) than either the early (M = 1.25; 28%) or contrast (M = 1.13; 25%) items, p < .05. These results are consistent with the research showing that children will typically choose an item for which they do not have a name when presented with a novel term such as “fep” (Golinkoff et al., 1994). To summarize, the main interest of this experiment was the extent to which toddlers would simply infer that the referent of a nonsense label was the item for which a label was less likely to be known (or unknown). The results showed that participants most often selected the referent of the late noun, a finding that would be predicted by several researchers (e.g., Clark, 1983; Golinkoff et al., 1994; Markman, 1989). However, it was also interesting that participants frequently responded “no” to these requests. Note that the method in this experiment contrasts with previous experiments using nonsense labels (e.g., Bauer & Mandler, 1989). Participants were not provided with an example of an item as the referent for the nonsense label (e.g., the experimenter did not show the participants an item and label it as a fep and then ask them to choose another fep). Instead, the experimenter simply asked for a fep and allowed participants to arrive at their own conclusions with no instructions or feedback. Other possibilities why refusals were often made is that children already comprehended the correct labels for these items, they recognized that the nonsense words were meaningless, or both. UNCONFOUNDING

CONCEPTS

AND PERCEPTS

At this point, the issue of similarity is likely to still be on the reader’s mind. After all, one might argue that the results of the previously discussed findings also reflect perceptual similarity judgments. Generalization of the basic-level terms was not tested across entire domains but most often to items that share some varied perceptual features with the correct referent. If the generalizations found in the previously discussed experiments are derived from perceptual bases, one wonders which features are considered and why? As previously pointed out, the examinations of children’s overextensions did not seem systematic with respect to features. Although shape similarity was also not systematically used, it is possible that if it were an available cue, it would override the generalizations children make. The next study examined the role of overall shape similarity in young children’s extensions of basic-level nouns. The challenge to

134

l

8.

Early Concepts

and Early Language

working out which generalizations are attributable to similarity and which are best accounted for by an understanding of taxonomic categories is that one must find appropriate test stimuli. New test cards were designed on which two of the three items were highly similar in perceptual appearance but were not related by taxonomic class (e.g., layer cake, top hat), and two of the items were taxonomically-related but considerably less similar in appearance (e.g., cake, pie). Although it was not possible to find realistic-looking, unrelated items that were similar to the same items tested in previous studies, as many of the same items were used as possible. The perceptually similar matches were designed by Poulin-Dubois and her colleagues (1995) for use in a preferential-looking study with 16- to 1 &montholds. Their results showed that infants generalized novel nouns to these similar yet unrelated items. That is, their results indicated that infants generalized a word such as “fep” from a layer cake to a top hat (Poulin-Dubois et al., 1995; Poulin-Dubois, Klein, Graham, & Frank, 1993). However, they did not test comprehension and generalization of real words (e.g., cake). After the stimuli were obtained, six adults were asked to provide perceptual similarity ratings of the items. It should be pointed out that the manner in which the pictures were drawn capitalized on similarity due to orientation or size but did not take away from the realistic features of the objects as found in the real world. On every test card, every adult rated the similar but unrelated pair as more similar than the taxonomic pair. This confirmed that the stimuli were not only viewed as similar by infants (Poulin-DuBois et al., 1995) but also by adults who clearly knew the identity of the objects depicted. It was predicted that if perceptual similarity guides children’s extension of the basic-level nouns, then more incorrect choices would be made to the similar than the taxonomic item; yet if taxonomic class guides their extensions, then more choices would be made to the taxonomic item. Note that both the similar and the taxonomic items are technically incorrect choices. As in the previous study in which nonce words were used, a child could answer “no” to these requests. On each of the seven test cards were three realistically colored pictures of objects. ?tvo of the objects were perceptually similar (e.g., layer cake and top hat) and two were related by taxonomic class (e.g. layer cake and pie). Each card served as a low-contrast test (e.g., the child was asked to point to the cake), and a high-contrast test (e.g., the child was asked to point to the hat). Each child was shown the test booklet twice. Low- and high-contrast trials were interleaved so that no more than two trials of one type were given in a row. As in the previous experiments, children were asked to point to or place a sticker on the correct

McDonough

l

135

referent for each noun. Children were praised for all their selections. No corrective feedback was given. The results showed that more correct than incorrect selections were M = 5.8,84%; incorrect made in both conditions: high contrast--correct M = 1.1, 16%; low contrast-correct M = 4.9, 70%; incorrect M = 2.1, 30% (ps < .Ol ). Correct selections did not differ significantly between the contrast conditions; however, significantly more incorrect selections were made in the low- than in the high-contrast trials, p = .05. An examination of the incorrect selections was made next. In the low-contrast trials, children made significantly more errors by choosing the overgeneralization item (i.e., the distractor that was taxonomitally related to the target; e.g., when asked about cake they pointed to the pie; M = 1.9) than to the distractor that was similar but unrelated to the target (M = .29, p < .OOl; e.g., the hat). Incorrect selections in the high-contrast trials (e.g., the hat was requested) were made as often to the distractor that was similar to the target (M = .6) as to the one that .5,p= .75). These results indicate that overall shape simiwasnot(M= larity alone showed little to no influence on the comprehension errors; instead, errors were most highly influenced by taxonomic relations. That is, even though the top hat was highly similar to the layer cake, children made more errors by choosing the pie, thus indicating that they were overgeneralizing the basic-level noun “cake” based on kind (i.e., to another food item). The question raised is why does using real words show different effects in this experiment using a picture-pointing task than found by Poulin-Dubois et al. (1993) using nonce words and a preferential-looking paradigm? It is difficult to know what children in the Poulin-Dubois experiments thought when they looked longer at the similar but unrelated pictures. They noted that children tended to make taxonomic selections as their vocabulary increased, which may be one explanation of the differences. However, it is also the case that the picture-pointing task used in this experiment allowed children to point to the correct referents as well as to respond “no.” As such, it was an explicit task of word knowledge, and interpretation of the findings is more straightforward.

These experiments explored five issues. First, the elicited production data showed the same degree of overgeneralization as found by Rescorla (1980) in her examinations of young children’s spontaneous productions. Second, the pattern of generalization found in the elicited production data was found in the comprehension data as well. It should

136

l

8.

Early Concepts

and Early Language

be pointed out that even though children extend basic-level terms to more than one type of object within a domain does not necessarily mean that they believe that terms such as clog and cat are synonymous with exemplars such as fox and raccoon. More correct referents than overextension items were chosen in response to both the early nouns and late nouns. A comparison between the results of the experiments indicated symmetry in the patterns of generalization that are likely to reflect some overlap in meaning. This overlap in meaning appears to reflect children’s knowledge as to the domain to which the early and late nouns belong. Third, children can use a fast-mapping strategy when figuring out the referent of a nonce label, a finding that replicates other research on word learning (Au & Markman, 1987; Carey, 1982). Yet, children are also likely to make refusals if they know the appropriate term for a referent or are aware that the nonce word is meaningless. These studies also showed that children are not swayed by perceptual similarity when generalizing nouns they already know. In light of the fact that linguistic cues are useful for highlighting superordinate or global categories at 1 year of age (Waxman & Markow, 1995), it would appear that similarity is not a powerful predictor of generalization from the very beginning of language acquisition. The role of similarity unconfounded with taxonomic class was studied in children’s generalizations of the words they have learned. Although the results of this task are clear, it is not without the same problems inherent to all the research which uses this technique. The challenge of working out which generalizations are attributable to similarity and which are best accounted for by an understanding of taxonomic categories lies in the problem one encounters when faced with bending and altering perspectives of reality to make stimuli that best capture which similarity metric children might use in their generalizations of words. I do not believe that any of us are really content with the procedure. If children represent what is being depicted using these stimuli, it is somewhat bizarre to expect them to overlook the identity of objects they already know. Nevertheless, this is what is typically done when one asks young children to entertain hypotheses about nonce words. However, experimenters typically ask children to suspend reality using a puppet (e.g., Waxman & Gelman, 1986) or training trials. One could consider such tasks as tapping into early metalinguistic skills. After all, when used with older children who know a lot about basic-level nouns and categories, these children have likely found that overall shape similarity is highly correlated with word extensions. Thus, older children can suspend reality and generalize a nonce word from a layer cake to a top hat with no problem (Imai et al., 1994). Yet, if the use of nonce labels taps into early metalinguistic strategies, it is likely that such strategies are

McDonough

.

137

implicit in nature. It is as yet unknown when children can explicitly explain the basis on which they generalize words. Our traditional views on categorization and language acquisition tend not to distinguish what something looks like from what kind of thing an object is. Jean Mandler and I agree that this is a serious problem when one is trying to find out when and how early “meaning” is formed in development. The rationale underlying early acquisition of basic-level categories is that early categories are based on perceptual similarity, basic-level category boundaries can be detected early in development, and thus first learned categories are basic-level categories. This could be true if similarity is “unprincipled” or “dumb” and unchanging throughout development. That is, most researchers would agree that the relevant knowledge and understanding of objects is lacking in early development, thus unprincipled similarity is what guides early categorization. Unprincipled similarity is a useful heuristic that is also used by older children and adults when they are faced with puzzling situations (e.g., Keil, 19 8 9; Medin, 19 8 9). However, between early infancy and childhood, infants quickly figure out the nature of kinds. From as early as 7 to 9 months of age, they indicate an appreciation that the common relation among dogs, fish, and birds is that they are all animals (Mandler & McDonough, 1993) and the chairs, beds, and tables are all furniture (Mandler & McDonough, 1998a). Other research suggests that linguistic input may increase the salience of such domains (e.g., Waxman & Markow, 1995). It has also been shown that the generalization of the common properties that bind such domains together (such that animals drink and one can ride in vehicles) is already being realized by around 9 to 14 months of age (Mandler & McDonough, 1996, 1998b, 2000; McDonough & Mandler, 1998). However, by 18 to 24 months of age, children are beginning to understand the properties that distinguish basic-level categories, although their performance is far from perfect. Inappropriate generalizations of basic-level properties are still as high as 25% in 24-month-olds (Mandler & McDonough, 2000). Just like the pattern of overgeneralization found in language production and comprehension in these studies, generalization of basic-level properties and animal sounds tends to follow the same trajectory. When considered as a whole, these findings show that classification through similarity (if that is still what one wants to call it) is made on more principled or smarter grounds than traditionally thought. Even though some developmental researchers have often assumed that the infant does not need to know the function or meaning of an object to categorize it, we do not yet clearly understand what constitutes similarity perceptually or conceptually to the developing child, particularly at the point when first words are being learned. As a col-

138

l

8.

Early Concepts

and Early Language

league admonished me, similarity is not to be sniffed at. Indeed, but understanding its developmental course by first defining what similarity means and then pushing the limits as to what it can and cannot account for is crucial. For example, research by Landau et al. (1988) showed us that similarity in terms of overall shapes of objects (the cue that is most predictive of basic level membership; e.g., Rosch, et al., 1976) becomes increasingly salient once children are well on their way to mastering language. As Mervis and her colleagues (e.g., Mervis, 1987; Mervis & Pani, 1980) pointed out, development is accompanied with changes in how infants and young children weigh attributes or functions of category members. That is, our notions of similarity change with expertise. HOW do early concepts and language acquisition interact? It is clear that when children first start out in the language game, they are not simply pattern recognizers (although this does play a significant role in grammar acquisition; Winter & Reber, 1994), but they also begin the task of mastering what words mean. The task of uncovering preverbal concepts is a tricky one because we cannot assume that infants comprehend the world in the same way we do. For example, my research with Jean Mandler, Soonja Choi, and Melissa Bowerman shows that preverbal infants are adept at forming various spatial categories such as containment, support, tight-fitting containment, and loose-fitting containment (McDonough, Choi, Mandler, & Bowerman, 1999). Yet, when the children are around 20 months of age, the categories onto which they extend early linguistic terms are specific to the language they are learning, whether it is English or Korean (Choi, McDonough, Mandler, & Bowerman, 1999). Adults (and perhaps young children) actually lose sensitivity to categories not saliently marked in their native language (McDonough et al., 1999). Thus, language can play many roles in concept acquisition. It can help young children differentiate global categories through use of mutual exclusivity or the principle of contrast, a process not complete until well after their 2nd birthday. It can teach them that shape similarity is an important feature not only for generalizing basic-level nouns (Landau et al., 1988) but also understanding that different shapes can be a good predictor of different functions and different kinds (Nelson, 1985). As found in the crosslinguistic research, language can strengthen the salience of categories they have already formed and also diminish the salience of those for which labels are not available in the language they are learning. More often than not, it can be troublesome when trying to communicate an idea without appropriate labels in the speaker’s vocabulary or else in the vocabulary of the language spoken. Jean Mandler ‘s struggles in her early descriptions of image schemas (a brilliant idea that was initially misunderstood by some researchers) come to mind as a good example.

McDonough

l

139

Her concept is clear but our language does not have a term readily available to describe it clearly (it takes at least a paragraph to describe image schemas adequately). Finally, as pointed out by Nelson (1985), language also allows children to recall the past in a more efficient manner. With experience talking to others about past events, young children learn both how to structure their narratives and also what is relevant to report relative to the ongoing social situation (Fivush, Haden, & Reese, 1996). However, we still need to know how to interpret what children actually say when recounting a past event. That is, we still need to understand what meanings they consider at different points in development. Given that infanti’ early meanings are overgeneral, then it is no wonder it is so difficult (but not impossible) to obtain clear evidence for episodic or highly specific memories in early childhood. Plenty of issues remain unresolved and more research is clearly needed. For this I am happy because I am at the beginning of my career. Yet, many issues have been illuminated in the past 15 years. For this I am happy because I have had the wonderful opportunity to share in these discoveries with a superb mentor, an exciting, ambitious colleague, and a wonderful friend-Jean Mandler. Dear Jean, I keep you in my heart.

Abbate, M. S., & LaChappelle, N. B. (1979). Pictures, please! ficson, AZ: Communication Skill Builders, Inc. Anglin, J. (1977). Word, object andconceptual development. New York: Norton. Au, T. K., & Markman, E. M. (198 7). Acquiring meanings via linguistic contrast. Cognitive Development, 2, 21 l-236. Backscheider, A. G., & Markman, E. M. (1990). Young children’s use oftaxonomic assumptions to constrain word meaning. Unpublished manuscript, Stanford University, Stanford, CA. Barrett, M. D. (1982). Distinguishing between prototypes: The early acquisition of meaning of object names. In S. A. Kuczaj (Ed.), Language development (pp. 3 13-334). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Bauer, F?J., & Dow, G. A. (1994). Episodic memory in 16- and 20-month-old children: Specifics are generalized but not forgotten. Developmental Psychology, 30,403417. Bauer, R J., & Fivush, R. (1992). Constructing event representations: Building on a foundation of variation and enabling relations. Cognitive Development, 7, 381401. Bauer, I-?,& Mandler, J. M. (1989). Taxonomies and triads: Conceptual organization in one- to two-year olds. Cognitive Psychology, 21, 156-184. Behrend, D. A. (1987). Overextensions in early language comprehension: Evidence from a signal detection approach. Journal of Child Language, 15, 63-75.

140

.

8.

Early Concepts

and Early Language

Behl-Chada, G. (1996). Basic-level and superordinate-like categorical representations in early infancy. Cognition, 60, 105-l 41. Bloom, L. (1973). One word at a time: The use of single word utterances before syntax. The Hague, Netherlands: Mouton. BouteIIe, J., & Eimas, P (1995). Categorization and conceptualization in infancy. Unpublished honor’s thesis, Brown University, Providence, RI. Bowerman, M. (19 78). The acquisition of word meaning: An investigation into some current conflicts. In N. Waterson & C. Snow (Eds.), The development of communication (pp. 263-287). New York: Wiley. Brown, R. W. (1958). Words and things. New York: Free Press. Carey, S. (1982). Semantic development: The state of the art. In E. Wanner & L, R. Gleitman (Eds.), Language acquisition: The state of the art (pp. 347-389). New York: Cambridge University Press. Choi, S., McDonough, L., Mandler, J., & Bowerman, M. (1999). Early sensitivity to language specific spatial categories in English and Korean. Cognitive Development, 14, 241-268. Clark, E. V. (19 73). What’s in a word? On the child’s acquisition of semantics in the first language. In T. E. Moore (Ed.), Cognitive development and the acquisition of language (pp. 65-l 10). New York: Academic. Clark, E. V. (1983). Meanings and concepts. In J. H. Flavell & E. M. Markman (Eds.), Cognitive Development, Vol. III (pp. 787-840). New York: Wiley. Clark, E. V (198 7). The principle of contrast: A constraint on language acquisition. In B. MacWhinney (Ed.), Mechanisms of language acquisition (pp. 7-33). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Dromi, E. (198 7). Early lexical development. Cambridge, England: Cambridge University Press. Eimas, I? D., & Quinn, I?C. (1994). Studies on the formation of perceptually based basic-level categories in young infants. Child Development, 65, 903-917. Fenson, L., Dale, I? S., Reznick, J. S., Bates, E., Thai, D. J., & Pethick, S. J. (1994). Variability in early communicative development. Monographs of the Society for Research in Child Development, 59, l-l 73. Fivush, R., Haden, C. A., & Reese, E. (1996). Remembering, recounting, and reminiscing: The development of autobiographical memory in social context. In D. Rubin (Ed.), Remembering our past: Studies in autobiographical memory (pp. 341-359). Cambridge, MA: Cambridge University Press. Golinkoff, R. M., Mervis, C., & Hirsch-Pasek, K. (1994). Early object labels: The case for lexical principles. Journal of Child Language, 21, 125-155. Goodman, N. (19 72). Seven strictures on similarity. In N. Goodman (Ed.), Problems and projects (pp. B29-B62). New York: Bobbs-Merrill. Huttenlocher, J., & Smiley, I? (1987). Early word meaning: The case of object names. Cognitive Psychology, 3 9, 63-89. Imai, M., Gentner, D., & Uchida, N. (1994). Children’s theories of word meaning: The role of shape similarity in early acquisition. Cognitive Development, 9, 45-75. Keil, F. C. (1989). Concepts, kinds, and cognitive development. Cambridge, MA: MIT Press. Kemler, D. G. (1983). Holistic and analytic modes in perceptual and cognitive development. In T Tighe & B. E. Shepp (Eds.), Perception, conception and development: Interactional anaZyses (pp. 77-102). Hillsdale, NJ: Lawrence Erlbaum Associates. Inc.

McDonough

.

141

Killen, M., & Uzgiris, I. C. (1981). Imitation of actions with objects: The role of social meaning. Journal of Genetic Psychology, 138, 219-229. Kuczaj, S. (1982). Young children’s overextension of object words in comprehension and/or production: Support for a prototype theory of early object word meanings. First Language, 3, 93-105. Landau, B., Smith, L. B., & Jones, S. S. (1988). The importance of shape in early lexical learning. Cognitive Development, 3, 299-321. Lucariello, J., & Nelson, K. (1985). Slot-filler categories as memory organizers for young children. Developmental Psychology, 21, 272-282. Macnamara. J. (1982). Namesfor things. Cambridge, MA: MIT Press. Mandler, J. M., Bauer, I? J., & McDonough, L. (1991). Separating the sheep from the goats: Differentiating global categories. Cognitive Psychology, 23, 263-298. Mandler, J. M., & McDonough, L. (1993). Concept formation in infancy. Cognitive Development, 8, 291-318. Mandler, J. M., & McDonough, L. (1996). Drinking and driving don’t mix: Inductive generalization in infancy. Cognition, 59, 307-335. Mandler, J. M., & McDonough, L. (1998a). On developing a knowledge base in infancy. Developmental Psychology, 34, 1274-1288. Mandler, J. M., & McDonough, L. (1998b). Studies in inductive inference in infancy. Cognitive Psychology, 3 7, 60-96. Mandler, J. M., & McDonough, L. (2000). Advancing downward to the basic level. Journal of Cognition and Development, l(4), 379-403. Markman, E. M. (1989). Categorization and naming in children: Problems of induction. Cambridge, MA: MIT Press. Markman, E., & Hutchinson, J. (1984). Children’s sensitivity to constraints on word meaning: Taxonomic versus thematic relations. Cognitive Psychology, 16, l-27. Markman, E. M., & Wachtel, G. F. (1988). Children’s use of mutual exclusivity to constrain the meaning of words. Cognitive Psychology, 20, 121-15 7. McDonough, L., Choi, S., Mandler, J., & Bowerman, M. (1999). The use of preferential looking as a measure of semantic development. In C. Rovee-Collier, L. Lipsitt, & H. Hayne (Eds.), Advances in infancy research, Vol. 12 (pp. 336-354). Stamford, CT Ablex. McDonough, L., & Mandler, J. M. (1998). Inductive generalization in 9- and 11 -month-olds. Developmental Science, I :2, 227-232. Medin, D. (1989). Concepts and conceptual structure. American Psychologist, 44, 1469-1481. Mervis, C. B. (1987). Child-basic object categories and early lexical developmerit. In U. Neisser (Ed.), Concepts and conceptual development: EcoZogicaZand intellectual factors in categorization (pp. 201-233). New York: Cambridge University Press. Mervk, C. B., & Canada, K. (1983). On the existence of competence errors in early comprehension: A reply to Fremgen 8~Fay and Chapman & Thomson. Journal of Child Language, 10, 431440. Mervis, C. B., & Mervis, C. A. (1982). Leopards are kitty-cats: Object labeling by mothers for their thirteen-month-olds. Child Development, 5.3, 267-273. Mervis, C. B., & Pani, J. A. (1980). Acquisition of basic object categories. Co@tive Psychology, 4, 496-522.

142

l

8. Early Concepts

and Early Language

Naigles, L. G., & Gelman, S. A. (1995). Overextensions in comprehension and production revisited: Preferential-looking in a study of dog, cat, and cow. Journal of Child Language, 22, 19-46. Nelson, K. (1985). Making sense: The acquisition of shared meaning. San Diego, CA: Academic. Poulin-Dubois, D., Graham, S. A., & Riddle, A. S. (1995). Salient object parts and infants’ acquisition of novel object words. First Language, IS, 301-3 16. Poulin-Dubois, D., Klein, B., Graham, S.A., & Frank, I. (1993, April). Shape and the taxonomic assumption. Poster session presented at the 25th Annual Stanford Child Language Research Forum, Stanford, CA. Quinn, P C., & Eimas, I? E. (1996). Young infants’ use of facial information in the categorical differentiation of natural animal species: The effect of inversion. infant Behavior and Development, 19, 383-386. Reich, I?A. (19 76). The early acquisition of word meaning. Journal of Child Language, 3, 117-l 23. Rescorla, L. (1980). Overextension in early language development. Journal of Child Language, 321-335. Reznick, J. S., & Goldfield, B. A. (1992). Rapid change in lexical development in comprehension and production. Developmental Psychology, 28, 406-413. Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-Braem, I? (1976). Basic objects in natural categories. Cognitive Psychology, 382-439. Ross, G. S. (1980). Categorization in l- to 2-year-olds. DeveZopmentaZPsychology, 16, 391-396. Ruff, H. A. (1986). Components of attention during infants’ manipulative exploration. Child Development, 5 7, 105-l 14. Smiley, P, & Huttenlocher, J. (1995). Conceptual development and the child’s early words for events, objects, and persons. In M. Tomasello & W. E. Merriman (Eds.), Beyond names for things: Young children’s acquisition of verbs (pp. 21-61). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Spencer, J., Quinn, I? C., Johnson, M. H., & Karmiloff-Smith, A. (1997). Heads you win, tails you lose: Evidence for young infants categorizing mammals by head and facial attributes. Early Development and Parenting, 6,113-126. Van de Walle, G. (1999, April). Nine-month-olds show manual habituation to shared object color. Paper presented at the Society for Research in Child Development, Albuquerque, NM. Waxman, S. R. (1990). Linguistic biases and the establishment of conceptual hierarchies: Evidence from preschool children. Cognitive Development, 5, 123-l 50. Waxman, S., & Gelman, R. (1986). Preschoolers’ use of superordinate relations in classification and language. Cognitive Development, 2, 139-156. Waxman, S. R., & Hall, D. G. (1993). The development of a linkage between count nouns and object categories: Evidence from 15- to 2 l-month-old infants. Child Development, 64, 1224-1241, Waxman, S. R., & Hatch, T. (1992). Beyond the basics: Preschool children label objects flexibly at multiple hierarchical levels. Journal of Child Language, 19, 153-166. Waxman, S. R., & Markow, D. B. (1995). Words as invitations to form categories: Evidence from 12- to 13-month-old infants. Cognitive Psychology, 29, 25 7-302. Waxman, S. R., & Senghas, A. (1992). Relations among word meanings in early lexical development. Developmental Psychology, 28, 862-8 73.

McDonough

l

143

Waxman, S. R., Senghas, A., & Benveniste, S. (1997). A cross-linguistic examination of the noun-category bias: Its existence and specificity in French- and Spanish-speaking preschool-aged children. Cognitive Psychology, 32, 183-218.

Winter, B., & Reber, A. S. (1994). Implicit learning and natural language acquisition. In N. C. Ellis (Ed.), Zmplicit and explicit learning of languages (pp. 115-l 46). New York: Academic. Younger, B. A., & Cohen, L. B. (1986). Developmental change in infants’ perception of correlations among attributes. Child DeveZupment, 57, 803-815.

This Page Intentionally Left Blank

A Stitch in Time: The Fabric and Context of Events Tamar Murachver

University of Otago, New Zealand

E

vents are central to human development. Through the regularity of actions, infants begin to understand their world. Early social and communication skills emerge through children’s involvement in familiar, predictable action sequences, and later narrative skills are formed from the telling and retelling of events. These events experienced by children become the basis of their self-knowledge. The study of event memories is important to psychologists for other reasons, as well. Understanding children’s event memory touches on questions about the sources of infantile amnesia and about the feasibility of children as expert witnesses. For all these reasons, the study of child development is also the study of children’s ability to remember, retell, and reproduce events. In this chapter, I discuss how aspects of event structure and context-influences outside of the child-influence how the child thinks of and retells an experience of an event. I first review some of the arguments about why event structure matters, and present new data that modify some earlier assertions. I then go on to describe how the context of the event, from physical surroundings to supporting narration, also 145

146

l

9.

A Stitch in Time

impact on school-age children’s memories. The studies reported highlight two main points. First, don’t forget to check the simple reasons, because sometimes the less sophisticated explanations do a good job fitting the data. Second, don’t forget the social context in which events are learned, remembered, and retold. THE STRUCTURE

OF EVENTS

One of the basic principals of memory is that organization improves performance. This is true not only for events, but for other types of information as well. There is a claim, however, that there is something special about events and, in particular, their organization, which distinguishes them from other sorts of organized information (Mandler, 1984). Events are organized temporally. Actions occur across time. Moreover, the order in which they occur across time is often important. So, for example, we order food at a restaurant before we eat it, and not the other way around. Sometimes the relation between actions is arbitrary. When getting ready in the morning, do you make the bed first and then have breakfast, or do you eat breakfast first before making the bed? Sometimes these relations between actions are simply conventional. When an arbitrary ordering becomes a regular way of behaving in a social group, it ceases to be arbitrary, and becomes a convention. In America, children do not open up the presents at a birthday party until after all the guests have arrived, and probably after they have had their party games and cake. In New Zealand, children are accustomed to opening their birthday presents as each guest arrives. Technically, in conventional orderings, there is no logical reason why one action occurs before or after another. Nonetheless, people often construct reasons why their ordering of conventional actions is sensible. However, the type of relation between actions that has received the most attention, both in the memory and story comprehension literatures, is the logical relation between actions. The order in which logically connected actions can occur is constrained. For example, you cannot pay for an item in a store until you have chosen it, because paying is dependent on having a particular item to buy. Because the order of these actions occurring is constrained, and because there is a logical explanation for this order, they are more cohesive and should therefore lead to a more coherent event representation. The more an event is comprised of logically organized actions, the more likely it will be recalled (Fivush, Kuebli, & Clubb, 1992; Murachver, Pipe, Gordon, Owens, & Fivush, 1996; Ratner, Smith, & Dion, 1986) and reproduced in the same order as the original experience (Barr & Hayne, 1996; Bauer & Mandler, 1989).

Murachver

l

147

How do we explain this effect? One explanation is that when actions are logically related, each remembered action serves as a retrieval cue for the next logically occurring action (Mandler, 1983). Moreover, logically related actions are generally united by a higher-level goal in the causal chain, and are thus retrievable by remembering the superordinate goal. For example, I might remember at a restaurant that I gave the waiter cash by starting from the superordinate goal of “paying for the food.” Through a series of priming studies, van den Broek and Larch (1993) showed that both these forms of access are used by adults. There is little question that these effects are reliable, and I believe the explanations offered have much merit. In particular, when logical relations are tested against arbitrary relations using children under the age of 3 years, the events generally consist of a small number of actions (e.g., three), and each successive action involves a continuity of objects. For example, Bauer and Mandler (1989) used a frog jump event of three logically-ordered actions whereby a child leaned a board on a block to make a teeter-totter, placed the frog on the board, and hit the board to make the frog jump. Between each successive pair of actions, at least one object remained constant. Similarly, in their arbitrary events, there was an overlap in the objects used from one action to the next. In the train ride event, for example, the child attached two cars, put the cars on the track, and put a doll into one of the cars. This same degree of object overlap is not always present in the events used with older children (e.g., Murachver et al., 1996; Ratner et al, 1986; Smith, Ratner, & Hobart, 1987). These events tend to involve many more actions and objects (e.g., 20 or more), and the continuity of objects across successive pairs of actions is difficult to maintain. The question is, does object continuity occur with logical sequences more than it does with arbitrary sequences, and can it partially account for the effect of organization on children’s event reports? As an example, in the event, “going to a birthday party,” the more logically linked actions of “blow out the candles on the cake, ” “cut the cake,” and “eat the cake” involve more overlap of objects than the arbitrary sequence, “get a balloon,” ” play pass the parcel,” and “have a treasure hunt.” Object overlap is just one of many forms of overlap that might exist across actions in an event. Actors can overlap, and so too can actions. Many of the events used with children in experimental situations are constructed so that the main actor is the child. Moreover, because memory is often assessed by counting the number of actions recalled or performed, events are designed so that each action is distinct. Thus, action overlap is generally not a feature of the events used with older children. In the context of the types of events used with preschool and

148

l

9.

A Stitch in Time

school-age children, object overlap is probabl .y the most relevan t form of repetition across a series of actions. CONTRASTING LOGICAL CONNECTIONS AND OBJECT OVERLAP To test whether object overlap facilitates memory independent of logical organization, Larry Owens, Mel Pipe, and I constructed two versions of an event. One version was organized; the other version was not, and served as a control. In the organized version, there were four types of scenes experienced by each child, and each scene comprised five actions. The four types of scenes reflected the possible combinations of logical versus arbitrary relation and the presence or absence of object overlap. In other words, one scene contained arbitrary relations with object overlap. For example, within the scene the child might put a hat on the bear, make the bear look through a telescope, and put an eye patch on the bear as part of the goal of getting the crew ready. Another scene contained arbitrary relations with unique objects. While pretending to sail a ship, the child might steer the wheel, turn on the ship lights, and raise the anchor. A third scene contained logical relations with object overlap. Example actions include tracing a skull on paper, shrinking the paper, and placing the paper in the holder to achieve the goal of making the pirate flag. Finally, the fourth scene contained logical relations with unique objects. In this scene, the child might assemble a puzzle, read a poem, and listen to a tape to find clues to where the treasure is hidden. In the control event, the same set of 20 actions were presented, but in an arbitrary order and without object overlap. Each scene involved a different toy animal. For example, in the first control event scene, the bear put a piece of paper in a holder, put on a vest, and steered the wheel of the ship. In Scene 2, the dog turned on the lights, put on a hat, and opened a chest. Thirty-six children between the ages of 61 and 83 months (M = 73.1 months) were visited at their school. Half experienced the organized version of the event and half experienced the control version. Three to 4 days after their event experience, children were interviewed about what had happened during the event. There were two measures of children’s performance that are of interest: how much children reported about the event, and how well they retained the original ordering of this information (i.e., sequencing). An analysis of the quantity of information reported showed a clear effect of object overlap. Children reported more actions from scenes containing object overlap than they did from scenes without object overlap. There was no effect of organization (causal versus arbitrary) on the amount that children recalled.

Murachver

l

149

The analysis of children’s sequencing revealed a slightly different picture. Both object overlap and causal organization helped sequencing. As shown in Fig. 9.1, this effect was additive. Having object overlap and causally-related actions led to the most accurate sequencing, whereas the absence of both resulted in the least accurate sequencing. Analyses of the corresponding actions in the control version indicated that the effects found in the organized event were the results of organization and not the memorability of particular actions within a scene. Why did object overlap influence children’s memory reports? If one action performed on an object could be remembered, it could then cue recall of other actions performed on that object. This resulted in increased recall and more accurate sequencing of actions. It is important to note that this facilitation was not a function of the amount of unique information to be remembered. First, scenes were constructed such that those with object overlap had equal or slightly greater numbers of objects than those without object overlap. Second, the recall of actions, not objects, served as the dependent measure. These results pose no contradiction to those who claim that causal organization improves children’s sequencing. In fact, a number of studies examining event memory in very young children have found organization effects on sequencing and not on quantity recalled (Bauer

FIG. 9.1 Children’s mean sequencing tion and object overlap.

scores as a function

of scene organiza-

150

l

9.

A Stitch in Time

& Hertsgaard, 1993; Bauer & Mandler, 1989). Our results suggest that object overlap enhances the effects of organization on sequencing measures. Moreover, object overlap aids in reporting more information overall. This issue is particularly important when using more elaborate events with older children, where object overlap is not always monitored when other organization variables are manipulated. These results also serve as a reminder. It is easy to become swept away with complex explanations rather than stepping back and looking for simpler reasons. This is particularly relevant in our explanations of human behavior, where this tendency is likely to go unchecked. HIERARCHICAL

STRUCTURE

AND CONTEXT

There are other aspects of event structure that influence the representation and retrieval of events from memory. The overall organization of events, and in particular, their hierarchical structure, are critical to children’s representations (Mandler, 1984). What is meant by hierarchical structure? Most events consist of a number of subgoals. For example, getting ready for school might involve the subgoals of “make your bed, ” “get dressed, ” “eat breakfast,” “wash up,” and “pack your school bag.” Each of these subgoals is achieved by performing a series of specific actions. The event is held together by strands running both horizontally (i.e., logical and temporal relations among actions at the same level) and vertically (i.e., actions at one level achieve subgoals, which in turn help to complete the event). To what extent are children sensitive to hierarchical structure and reproduce this structure in their event memories? Kindergarten children are less likely to use hierarchical structure than are older primary school children and adults (Ratner, Smith, & Padgett, 1990). These younger children are more apt to focus on objects to organize their rewhereas adults are more likely to rely on call and sorting, superordinate goals. As they grow older and more familiar with the event, children’s ability to use hierarchical organization to guide cued recall and sorting increases. Philippa Drew, Mel Pipe, and I decided to approach the problem from a different angle. Rather than compare memory for subordinate and superordinate actions and objects as Ratner et al. (1990) had done, we asked whether children would use information about the goal structure of an event to merge or keep distinct multiple event instances. In other words, we asked, what makes an event one event and not another? Are two experiences the same event if they have the same goal structure? Are they instances of the same event if they have the same actions and objects?

Murachver

b

151

To answer this question, we created two events, “Visiting the Pirate,” and “Visiting the Wizard.” These two events differed in their goals, of which there were four. For example, the first goal of the Pirate event was to “become a real pirate,” whereas the first goal of the Wizard event was to “empower the wand.” There were five actions that had to be performed in each scene to achieve the scene goal. These actions were identical across the two events. The first scene in each event is shown in Table 9.1. The actions are the same across the two events, as is the order of the actions. Two things differ: the goal of the scene, and the overall theme of the event. We made two versions of these events, so that we could vary some of the content across multiple event exposures. In each scene, there were two actions that could vary. These are shown in italics in Table 9.1. Thus, of the 20 actions in the event, 12 were similar across the two versions and 8 differed. One version was referred to as the “Blue Version” and the other as the “Red Version.” The color of the pirate or wizard’s clothing corresponded to the version of the event being experienced. This was important later, when children were asked about what happened in each event experience. Thirty-two children between the age of 5 and 6 years received two event experiences, separated by a week. All children received one Blue TABLE

9.1

Scene 1 of the Pirate-Wizard Version

A

Study Version

t3

Pirate Event Goal-Become a real pirate Decorate the pirate cape Put on cape Read pirate poem Tap captain’s stick Write name in pirate book

Goal-Become a real pirate Put on pirate gloves Put on cape Read pirate poem Tap captain’s stick Hoist pirate flag Wizard

Goal-Empower

the wand

Decorate wizard cape Put on cape Read magic spell Tap magic wand Write

name in spell book

Event Goal-Empower the wand Put on wizard gloves Put on cape Read magic spell Tap magic wand Hoist wizard flag

152

l

9.

A Stitch in Time

version and one Red version, and the order of experience was counterbalanced. Half of the children received the same goals on each experience. For example, they might have a Blue Pirate experience followed by a Red Pirate experience. The remaining children received different goals on each experience. For example, they might have a Blue Pirate experience followed by a Red Wizard experience. All events and versions were equally likely to be in the same or different goal condition, and were equally likely to be experienced first or second. All children were interviewed 7 days after their second event experience. They received a verbal interview, followed by a picture recognition task. The picture recognition task involved 32 hand-drawn, black and white actions. Children were asked whether the action depicted happened in each experience. If a child had visited the Red Wizard, followed by the Blue Pirate, the child was asked, did this happen “when you visited the Red Wizard?” and then did this happen “when you visited the Blue Pirate?” The order of asking about the first or second experience was switched for half the cards. Of the 32 actions depicted, 8 occurred in both (Red and Blue) experiences, 8 occurred only in the Red experience, 8 occurred only in the Blue experience, and 8 were distracter items that never occurred but were plausible (e.g., putting on a ring; feeding the fish). To our surprise, we found absolutely no effect of repeated versus different goals across the two experiences. Across the different combinations of Pirate and Wizard experiences, children reported a greater proportion of actions that were common to the two events (M = 0.56) than ones that varied (M = .16), and they were more accurate in their picture recognition of the common items. In essence, the children were treating the Pirate and Wizard events as instances of the same general event. We then rethought the events as the children had experienced them. All experiences and interviews had occurred at the children’s school, and some of the adults involved in the study overlapped between experiences. The children had many cues to tell them that these were similar experiences, and very few cues, other than event goals and thematic background, to tell them they differed. We repeated the study with an additional 31 children. This time the children were brought to the university by a caregiver and experienced one event in our lab. The other event they experienced at their school. We also made sure that the adults the children met during one experience were different to those they met during the other experience. As much as possible, we tried to maximize the differences between the physical and social contexts of the two experiences.

Murachver

l

153

The picture recognition data, in particular, yielded some interesting results. Children in the same goal condition were more likely to correctly respond to actions presented in both experiences (M = 7.6 7) than were children in the different goal condition (M = 6.75). On the other hand, when children received pictures of actions that occurred in only one experience, there was a trend for those in the different goal condition to be more accurate (M = 4.3 1) compared to those in the same goal condition (M = 3.40). Thus, when children experienced events with different goals in different contexts, they were able to keep these experiences distinct and were less likely to note that actions had been repeated across event experiences. This conclusion was supported by further findings. Children who received different goals were more likely to misidentify “both” items (items that occurred in both events) as occurring only once. This suggests that having different goals, along with different contexts, made it easier for children to differentiate their experiences. We then compared the picture recognition data from the two studies to assess the overall effect of context. Our analyses showed clear effects of context (see Fig. 9.2). Children who experienced different contexts (i.e., lab and school contexts) made more correct responses to actions that occurred in only one experience. Thus, context helped them to differentiate the events. Similarly, when children received different contexts, they were less likely to make errors in responding to items that occurred only once. When they had the same context, they misremembered “once only” items as occurring in both versions, or in the other experience. In other words, they had more difficulty differentiating their experiences. We are left concluding that although children can use information about the overall goal structure of an event, they are inclined to use concrete cues to determine whether an event experience is related to another. A similar reliance on context is notable in young infants and toddlers (Barr & Hayne, 2000; Herbert & Hayne, 2000). Our Pirate-Wizard study suggests that this is not outgrown with infancy. The extent to which children are context dependent is likely to be a function of the complexity of the event relative to the child’s age. For a given event, children need less contextual support to generalize across event experiences with age. Moreover, a child who generalizes across instances of one event might rely on contextual information for another, more complex event. Therefore, we cannot conclude that 5- to &year-old children are dependent on context outright. With events as complex as the Pirate-Wizard ones, they are. We would predict that even adults would be somewhat dependent on context when first experiencing an unfamiliar and elaborate event.

154

’

9.

A Stitch in Time

FIG. 9.2 Mean numbers different contexts.

of errors on “once”

items experienced

in same and

RESEARCH ON THE EFFECTS OF ADULT NARRATION ON CHILDREN’S MEMORIES So far, I have shown that there are aspects of event structure that impact on children’s ability to remember and retell an event. There are also aspects of the event context that impinge on their memories. One aspect of the event context that might also be expected to influence children’s representation of an event is the linguistic support surrounding an event experience. Certainly we know that talking to children about events after they have occurred can influence how they think and talk about the event later. Research programs carried out in the labs of Robyn Fivush and Elaine Reese, for example, showed that maternal narrative style guides children’s representation of events and influences their ability to recall past experiences (e.g., Fivush & Fromhoff, 1988; Reese, Haden, & Fivush, 1993). Similarly, Goodman, Quas, BattermanFaunce, Riddlesberger, & Kuhn (1994) found that children’s memory for a painful medical procedure was influenced by parental discussion after the event. Children whose parents talked with them following the procedure provided more accurate reports of the event compared with those whose parents did not discuss the event.

There is some evidence that narration during an event can influence children’s later memory for the event. In a museum visit, children were more likely to include in their reports information that was labelled by both their parent and themselves (Tessler & Nelson, 1994). This suggests that adult narration might provide information about what to report as well as information about what not to report. Narration by an adult during an event experience provides a structure for what is happening. It might relate the experience to others of the past, or it might help children organize a series of actions into a larger unit (see Fivush, Pipe, Murachver, & Reese, 1997). Narration should influence even the most verbally skilled children. Although preschool age children would be expected to benefit from linguistic support, older, school-age children should also benefit. Children between the ages of 5 and 7 are still learning how to tell stories about their experiences. Moreover, although they have a great deal of experience with the world, they still might not elaborate explanations about what is occurring unless they are guided by a more experienced person. Children in the first few years of school make minimal and haphazard use of memory strategies such as organization and rehearsal (Naus & Ornstein, 1983). Given the paucity of metamemory strategies employed by young school-age children, it would not be surprising that they might refrain from elaboration when confronted with sparse narration for an event. One question not addressed in the Tessler and Nelson (1994) study is whether the effects of narration are long-lasting. If narration plays a role in the representation of the event, these effects should persist over lengthy delays. In addition, narration might not only instruct a child on how to talk about an event; it might change how they tell the event to themselves. Again, this would suggest that the influence of narration during an event might be durable. Glenda Clark and I used two narration conditions with the novel event, “Helping the Zookeeper.” The zookeeper event was comprised of four scenes, each with five actions. In the Full Narration condition, children received an elaborate, explanatory narration, similar to that used in most of our memory events. In Scene 1, “Getting the House Ready,” children were told, “Next we need to put these boxes together like steps, from smallest to biggest, so that Minko can climb up them. Can you put them together against this wall?” In the Empty Narration condition, children received little labeling of actions or objects, and no explanations for why actions were being performed. The adult still talked to the child about the event, but used “empty” phrases such as, “we need to put some of this in here,” and “next we need to put these together from smallest to largest. Can you put them together against this for me?”

156

l

9.

A Stitch in Time

Thirty-three 5- to 7-year-old children visited the zookeeper at their school. They were interviewed 1 week and 6 months later, using both a verbal recall and behavioral reenactment interview. Analyses of children’s verbal reports confirmed that narration could help or hinder what a child remembers about an experienced event. Children who had received full narration reported more information per scene than did those who had received empty narration. Moreover, this result persisted over the 6-month delay between interviews. The narration effect was present only in the verbal interviews. When children were provided with the physical support of objects used in the event, the influence of narration disappeared. Likewise, narration had no discernible impact on children’s errors or accuracy. More recently, Kylie Paterson and I replicated these results and added a third narration condition. In this third condition, some of the actions, objects, and explanations were mislabeled. This condition was included for two reasons. First, by including information that does not actually occur during the event, we can assess the extent to which linguistic information alone enters into children’s reports. Second, we suspect that in some real-life experiences, for one reason or another, adults do not fully inform children of the ongoing activities. The reason might be benign, to protect a child, or it might be otherwise, for example, when the adult is engaged in some criminal activity. Nonetheless, knowing whether children’s memories of events can be distorted by partially incorrect narration during an event experience would be informative. The design was similar to the prior study. Forty-eight 5- to 7-year-old children experienced the novel event, “Helping the Zookeeper.” The children were interviewed 1 week and 6 months following their event experience. Interviews consisted of verbal recall followed by behavioral reenactment. As before, children who experienced full narration provided more complete verbal reports than those who received empty narration. As before, this effect persisted over the 6-month delay. Although similar trends were present in the reenactment data, these were not significant. These children were sophisticated users of language; yet they still benefited from the structure provided by adult narration. When the narration was insufficient, even when the event was fairly simple and understandable, they were unable to provide as complete a report (see Fig. 9.3). We were surprised to find that mislabeled narration seemed to have no ill effects on the amount of correct information children reported. There was some suggestion that after a longer delay, children in this condition needed more prompting to recall as much information as did those in the full narration condition, however. Children in the full nar-

Murachver

FIG. 9.3

Mean numbers

of core actions reported

in verbal

l

157

interviews.

ration condition reported most of their information during free recall, whereas children in the mislabeled condition recalled very little until they were prompted with the scene goal. Surprisingly, narration did not reliably affect the quantity or types of children’s errors. However, there was a trend toward greater errors with full and mislabeled narration compared to empty narration. In addition, narration did not reliably affect accuracy or sequencing. When the narration contained erroneous information, children did not necessarily copy the errors into their own reports. Moreover, they provided reports as complete as did those who experienced full narration. Why might this be? One suggestion is that narration helped children to focus on specific actions and objects, and where mislabeling occurred, perhaps the children simply corrected it. There was little overt evidence of such relabeling, but this explanation cannot yet be ruled out. It might be possible to test this hypothesis by varying the familiarity of the event, and thus children’s ability to fill in additional, correct information. An alternative explanation is that the mislabeled narration was still coherent, and thus helped children to organize the event information. The mislabeled narration was constructed such that incorrect labels, actions, and reasons were plausible and not bizarre. In

158

l

9.

A Stitch in Time

the context of a pretend event, mislabeling “cleaning the cage” as “redecorating the room,” or saying “pour flour over the paper U instead of “shake flea powder over the straw” was still informative and believable. A more detailed examination of the type of the mislabeled component (e.g., object, action, or reason), and the degree of discrepancy between the correct and mislabeled term (mild to more extreme), might clarify why mislabeled narration did not hinder children’s performance in this study.

CONCLUSIONS In conclusion, there is much about the world outside of the child’s mind that can influence what they remember. Children often focus on fairly concrete aspects of events, such as physical or social context, or the presence of certain objects, rather than more abstract features such as goal structure. It is likely, as Ratner et al. (1990) suggested, that this focus shifts with age. However, we have to be careful of assuming too much reliance on abstraction, even from adults in their everyday thinking. Moreover, the same event experienced in a rich linguistic context or a sparse linguistic context is remembered differently, and this effect is persistent over a 6-month delay. This latter finding, in particular, highlights the importance of social interaction in the establishment and maintenance of memories. These narration effects cannot be simply attributed to children learning a particular style of talk from a caregiver, although there is no doubt that children acquire preferred styles of talking about events. We have ample data from our lab investigating speech accommodation in children and adults to show that people will adapt their style of talk to match the style used by another adult (Stephenson & Murachver, 1999; Thomson, Murachver, & Green, 2001). Perhaps it is not surprising then that children adapt their ways of talking about events to match the context created by the adult and child. In this way, narration effects do not have to be due to a set way of talking about events in general, but might be more a way of talking about an event in a particular social context. Event episodes can be thought of as stitches in time, woven across a lifetime. Episodes of similar content are sometimes woven together, depending on both the participants’ age and the contextual similarity between the episodes. Although development brings with it greater freedom from context, the reliance on specific contextual features is not an absolute function of age. It is a function of age relative to the complexity of the event. Events themselves are woven of actions, actors, and objects. The threads that bind these together affect how well they

Murachver

l

159

hold together in the mind of a child. To this tapestry, we add the threads of language. Language links individual actions and it links whole events; it can link events with other events and with other people. It may well be language that turns our patchwork into a quilt. ACKNOWLEDGMENTS I would like to acknowledge the assistance of a large lab team in the collection, coding, and analysis of the data reported in this chapter. These include Glenda Clark, Philippa Drew, Kylie Paterson, Larry Owens, Meagan Stephenson, Elizabeth Hall, Kypros Kypri, Vanessa Chapman, Ruth Walker, Natasha Speight, l%acey Anderson, Chee Leong, Sasha Farry, Deirdre Brown, ‘Bri Patterson, and Megan Gollop. The participation of Glenda Clark, Philippa Drew, Kylie Paterson, and Larry Owens, in particular, has been substantial. My colleagues at Otago have had a profound influence on my thinking and research, and I would like to acknowledge the influence of Mel Pipe, Harlene Hayne, and Elaine Reese, in particular.

Barr, R., & Hayne, H. (1996). The effect of event structure on imitation in infancy: Practice makes perfect? Infant Behavior and Development, 19, 253-25 7. Barr, R., & Hayne, H. (2000). Age-related changes in imitation: Implications for memory development. In C. Rovee-Collier (Ed.), Progress in infancy research (pp. 21-67). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Bauer, F! J., & Hertsgaard, L. A. (I 993). Increasing the steps in recall of events: Factors facilitating immediate and long-term memory in 13.5 and 16.5 month old children. Child Development, 64, 1204-l 223. Bauer, I? J., & Mandler, J. M. (1989). One thing follows another: Effects of temporal structure on l- to 2-year-olds’ recall of events. Developmental Psychology, 25, 197-206. Fivush, R., & Fromhoff, F. (1988). Style and structure in mother-child conversations about the past. Discourse Processes,I I, 337-355. Fivush, R., Kuebli, J., & Clubb, I? A. (1992). The structure of events and event representations: A developmental analysis. Child Development, 63,188-201. Fivush, R., Pipe, M-E., Murachver, T, & Reese,E. (1997). Events spoken and unspoken: Implications of language and memory development for the recovered memory debate. In M. Conway (Ed.), Recovered memories and false memories (pp. 34-62). Oxford, England: Oxford University Press. Goodman, G. S., Quas, J. A., Batterman-Faunce, J. M., Riddlesberger, M. M., & Kuhn, J. (1994). Predictors of accurate and inaccurate memories of traumatic events experienced in childhood. Consciousnessand Cognition, 3, 269-294. Herbert, J., & Hayne, H. (2000). Memory retrieval by 18-30-month-olds: Age-related changes in representational flexibility. Developmental Psychology, 36,473484.

160

l

9. A Stitch in Time

Mandler, J. M. (1983). Representation. In J. H. Flavell & E. M. Markman (I%.), I? H. Mussen (series Ed.), Handbook ofchikf psychology: Vol. 3. Cognitive development (pp. 420-494). New York: Wiley. Mandler, J. M. (1984). Stories, scripts, and scenes: Aspects of schema theory. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Murachver, T., Pipe, M-E., Gordon, R., Owens, J. L., & Fivush, R. (1996). Do, show, and tell: Children’s event memories acquired through direct experience, observation, and stories. Child Development, 6 7, 3029-3044. Naus, M. J., & Ornstein, F?A. (1983). Development of memory strategies: Analysis, questions, and issues. In M. T. H. Chi (Ed.), 7bend.s in memory development research: Contributions to human development series (Vol. 9, pp. l-30). Basel, Switzerland: Karger. Ratner, H. H., Smith, B. S., & Dion, S. A. (1986). Development of memory for events. Journal of Experimental Child Psychology, 41, 41 l-428. Ratner, H. H., Smith, B. S., & Padgett, R. J. (1990). Children’s organization of events and event memories. In R. Fivush & J. A. Hudson (Eds.), Knowing and remembering in young children (pp. 65-93). Cambridge, England: Cambridge University Press. Reese,E., Haden, C. A., & Fivush, R. (1993). Mother-child conversations about the past: Relationships of style and memory over time. Cognitive Development, 8, 403430.

Smith, B. S., Ratner, H. H., & Hobart, C. J. (1987). The role of cuing and organization in children’s memory for events. Journal of Experimental Child Psychology, 44, l-24. Stephenson, M., & Murachver, T (September, 1999). Gender-preferential speech and accommodation in children. Paper presented at the New Zealand Psychological Society Conference, Dunedin, New Zealand. Tessler, M., & Nelson, K. (1994). Making memories: The influence of joint encoding on later recall by young children. Consciousness and Cognition, 3, 307-326. Thomson, R., Murachver, T., & Green, J. (2001). Where is the gender in gendered language? Psychological Science, 2 2, 17 l-l 74. van den Broek, I?, & Larch, R. F., Jr. (1993). Network representations of causal relations in memory for narrative texts: Evidence from primed recognition. Discourse

Processes, 16, 75-98.

10 -4% The Reemergence

of Function

Katherine Nelson Angelica Ware City University of New York Graduate Center

I

ean Mandler and her colleagues have made important discoveries about infants’ categorizing of objects that have the potential for transforming our understanding of conceptual development, word learning, and indeed cognitive development in general. On the basis of their work they have made a set of related theoretical claims: I. Conceptual and perceptual categories develop together from the infant period. This view accords with those who distinguish between perceptual and conceptual categories (e.g., Barsalou, 1999; Bornstein, 1984; Nelson, 1985). It contrasts with those theories of conceptual development that assume that categorization on the basis of similarity of perceptual features is primary, and that only later do children learn to categorize on the basis of essential characteristics (Keil, 198 7; Smith & Jones, 1993). 2. The initial conceptual categories are abstract and global. This is perhaps Mandler ‘s most crucial claim, and it is complemented by the third proposal. Both contest accepted formulations based on Rosch’s pathbreaking work in the 197Os, which claimed that there is a basic level of categorization of objects that is most “natural” and

162

l

IO.

The Reemergence

of Function

that is acquired first by children (Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976). This is the level in an object taxonomy where category instances share most distinctive features with each other and are distinct from other categories at that level, thus the features have high cue validity for identifying category instances. In Rosch’s view, the critical features of the basic level include perceptual attributes, especially shape, and movement. For example, cars and trucks are each basic level categories within the vehicle taxonomy. Specific kinds-subordinate types-of cars (e.g., Camrys, taxis) share car features but have distinctive features differentiating them from each other. Vehicles share only a few features in common, thus are higher order, not basic level, and the features they share are primarily abstract and functional, for example, “used for transportation.” 3. Basic level categories are not primary in development. Rosch’s work, and that of others, indicated that young children could categorize at the basic level, but not at higher or lower levels. Mandler ‘s counter claim that the basic level is not primary in development is based on findings that infants respond selectively to instances of categories such as animate and inanimate, rather than randomly or to narrower categories, such as dog and cat (McDonough & Mandler, 1998). These findings contrast with habituation studies showing that infants can discriminate perceptually between dogs and cats (discussed in a section to follow). Forming global categories such as animate is abstract, not based on specific perceptual features, a conclusion that goes against theoretical claims that conceptual categories develop from concrete to abstract bases. Mandler is not claiming that children form superordinate categories that include basic level categories; rather basic level categories must be differentiated from global categories. 4. Conceptual categories are formed in the effort to establish meaning; thus the abstract basis for infant categories involves the role that objects play in events, a relational basis, not a static inherent perceptual basis. This claim is based in part on findings that infants will allow members of one global category (animate) to engage in the same functions in events (such as drinking from a cup) but will not extend this liberty to members of a different (inanimate) global category. In this chapter, we trace the implications of these claims for word learning and conceptualization beyond infancy. We first review work on infant categorization, focusing on studies investigating the role of function as a basis for infants’ categorization. Function is implicitly related to

Nelson and Ware

l

163

the event role that Mandler and McDonough have identified as the basis for infants’ categorization of objects. However, function has sometimes been defined and studied as a more abstract and sophisticated relation than the event role suggests. Therefore, in the next section we consider some of the uses of function in the developmental literature. Following the discussion of the infant work on form and function, we review work on children’s word learning and concepts focused on these issues. We then relate these to general theories of concept structure. In the final section we focus on theoretical considerations of developments after infancy that must take place if the claims of early global conceptual categories and perceptual identification features are correct. THE USES OF FUNCTION A long-standing tradition in psychology, philosophy, and linguistics prior to Rosch’s work dictated something like the following: concepts are formed from the abstraction of features logically combined into classes. Lexical items (words) map onto concepts that are invariant across languages (Fodor, 1975; Macnamara, 1982). Studies in psychology were designed to determine how adults, children, and other organisms (e.g., rats) formed concepts from features displayed in examples. Studies in linguistics were aimed at identifying the set of universal semantic features sufficient to characterize all words of all languages (Bierwisch, 19 70). From these perspectives, the function of things has little relevance. To achieve a word meaning or to form a concept, one must abstract the defining features or attributes. Failure to do so results in an incorrect meaning or concept. The child who viewed what things could do or what one could do with them might err if the ideas the child formed did not map onto the correct features or attributes. The possibility of wrong steps along the way to the final achievement of the correct concept or meaning was viewed as irrelevant. Developmentalists (Clark, 1973; Piaget, 1962; Vygotsky, 1962) might be interested in these wrong moves, but the theory of concepts would not (Fodor, 1972). Earlier work on children’s categorization within this framework assumed that function is a more abstract, hidden characteristic of objects than perceptual features and therefore a characteristic more difficult for children to use as a basis for classification. Olver and Hornsby (1966), for example, were interested in what attributes children identified as common to two items and what remote kinds of relations they were capable of at different ages. Their paradigm requires children to relate basic level items to one another, imposing a hierarchical relation. When asked what a dog and a cat have in common, children might reply that they are both animals (abstract hierarchical level). Or they

164

l

10.

The Reemergence

of Function

might say they both have fur (perception), or they both run (action), or they are both pets. Olver and Hornsby found that activity and abstract function were later in development than perceptual characteristics. Only older school-age children, for example, could identify the abstract function that items like a newspaper and a telephone had in common (communication). Such results were taken to confirm the abstractness of function and its difficulty for young children. In contrast, observers of children’s spontaneous definitions found that young children are most sensitive to the activities related to objects (arguably their function). When asked to say what a thing is the child is not required to relate two different classes on some abstract dimension but simply to note how something is related to activities in the world, that is, its role in events. Krauss’s (1952) charming collection of such “definitions” (‘;9 Hole is to Dig”) has found support from other, more standardized studies. In 3 974, Nelson put forth the “functional core hypothesis” (FCH) proposing that I-year-old children beginning language were forming concepts about things in the world, and learning and generalizing names on the basis of these concepts. The core of the concept was proposed to be the function of the things, what they did, or what could be done with them. Originally, this idea derived from the analysis of children’s naming practices and actions with objects, and it was then explored in several studies of children’s word learning, word meaning and general knowledge of objects and relations. The FCH proposed that perceptual features were part of the child’s concept, but primarily served as identificational features, on the basis of which children would extend the concept and word (when learned) to new instances. Rosch also had gone beyond perception to emphasize that common movements of objects or actions of people with respect to objects were important features of the object concepts’ internal structure (Rosch et al., 1976). This seemed generally consistent with the notion of function as the core of a child’s concept. Although the FCH respected the dictionary definition of function as “the specific, natural or proper action or activity of anything,” the notion of function as highly abstract was a barrier to its broader reception. Nelson (1979) attempted to clarify the concept of function, emphasizing that the 1974 proposal was based on function from the child’s point of view, thus on activities engaged in by children, not conventional uses or those intended by design. Four variations on this definition were noted: 1. Actions on things, for example, throwing balls. This aspect conforms to Rosch’s emphasis on common motor movements of persons engaged with the objects.

Nelson and Ware

.

165

2. Independent activity of the things themselves, for example, a barking dog. This aspect may help to define animates versus inanimates. 3. Reaction of a thing to an action on it, for example, the rolling of a ball after it is pushed, reflecting a cause-and-effect relation between two actions or between an action and an end state. 4. The use (idiosyncratic or conventional) of a thing for human purposes, for example, drinking from a cup. This aspect reflects the relation of an object to goal activities. Conventional use is the definition often applied to concepts of artifacts in discussions of conceptual understanding and object naming, but it may not be the function that a child will identify as significant. These characteristics “have potential significance; knowledge of function enables the individual to predict future actions and changes, regardless of the present actual state” (Nelson, 1979, p. 49). Or as Mandler pointed out, functional characteristics identify kinds and enable inductions about unseen properties and relations. They also allow inductions about new members of the category-those that have the same functions. During the same time period that the FCH was proposed, Miller and Johnson-Laird (19 76), in a comprehensive treatment of lexical, semantic, and conceptual representations, explicitly specified function together with perception as components of the meaning of object words. They also proposed the notion of conceptual core: ‘A conceptual core is an organized representation of general knowledge and beliefs about whatever objects or events the words denote-about what they are and do, what can be done with them, how they are related, what they relate to” (p. 29 1). They noted that this deeper conceptual core represents a kind of lay “prototheory,” similar in its organizing capacity to theories within scientific domains, thus prefiguring later theories of conceptual structure such as Murphy and Medin (1985). Miller and Johnson-Laird described specific concept types in terms of a schema including functional and perceptual information that could independently identify an instance of the concept. As an example, they noted that the word table could be used in a functional context as something to eat on when the common perceptual features were entirely absent, for example, when looking for a lunch stop on a hiking trip a person points to a rock and says “that’s a good table.” The functional core hypothesis (Nelson, 19 74) was strikingly consistent with Miller and Johnson-Laird’s (1976) ideas about object concepts and labels, although it was developed independently on the developmental level. In retrospect, it can be seen that the FCH reflected

166

l

10.

The Reemergence

of Function

the general zeitgeist of this period, when ideas about concepts and word meanings were beginning to change. Largely through the joint influence of Rosch and Miller and Johnson-Laird, no longer did the classical view of classes as composed of universal and equally weighted features additively combined hold firm. By 1981 prototypes, conceptual cores, and basic level were established parts of the concept and category literature (Smith & Medin, 1981). With this bit of historical background we turn to the literature that emerged in the wake of the FCH on identification of functional properties by infants, studies that tested the origins and limits of the function hypotheses. FUNCTION

IN INFANT CATEGORIZATION

Earlier conceptions of children’s categorization abilities, including classic studies by Inhelder and Piaget (1964) and Vygotsky (1962), implied that prelanguage children (under 12 months of age) lacked the cognitive requisites for categorizing objects. Over the past 30 years, however, numerous studies have found that children younger than 12 months of age can make categorical distinctions among objects. Although infants do not have the developed motor skills that allow them to sort objects into separate categories (as do older children), modern methods have revealed a rudimentary form of categorization ability. Evidence of this form of categorization was first introduced by Ricciuti’s (1965) innovative study of infants’ object-touching behavior. He observed that infants group objects by sequentially touching those that are from a single category and concluded that sequential touching behavior indicates sensitivity to similarity within object categories. Ricciuti’s technique, together with succeeding methodologies, made it possible to investigate the basis for categorization in infants. These procedures measure categorization by (a) observing the infants’ sequential touching and grouping behavior, or (b) measuring attention to familiar category members and interest in members of novel categories. Nelson (1973a) used the Ricciuti paradigm showing that 12- and 24-month-old infants made category distinctions based on objects’ functional and shape attributes. Using a similar paradigm, Starkey (198 1) found that 9- and 12-month-old infants made category distinctions based on perceptual similarities. Other methods devised to study infant categorization include object-examination tasks, preference-for-novelty looking procedures, and the habituation-dishabituation paradigm. Object-examination tasks, first used by Ross (1980), involve presenting infants with the stimulus objects, which they are allowed to examine and manipulate

Nelson and Ware

l

167

simultaneously or one at a time (Madole, Oakes, & Cohen, 1993; Mandler, Bauer, & McDonough, 199 1; Ross, 1980). Preference-for-novelty looking procedures involve presenting a category of pictures or photographs, for example of animals or furniture, for a period of time and, subsequently, presenting a novel category member. Infants younger than 4 months of age cannot effectively handle and examine objects and, therefore cannot participate in object-examining tasks. Researchers have instead used objects, pictures, photographs, or videotaped displays of objects in the habituation paradigm, assuming that picture-looking procedures are effectively measuring young infants’ ability to categorize. The habituation-dishabituation paradigm measures the decline in attention as a result of repeated elicitation to familiar stimuli (habituation), and renewed attention as the infant dishabituates to a change in the eliciting stimulus (Cohen, Gelber, & Lazar, 1971). Madole, Oakes, and Cohen (1993) used real objects and found that lo- and 1 &month-old infants attended to category differences and formed categories during habituation. Eimas and Quinn (1994) found that 3- and 4-month-old infants formed categories of horses and cats when familiarized to photographs of horses, cats, tigers, and female lions. Behl-Chadha (1996) also found that 3-monthold infants showed categorization of pictures of mammals, but did not show categorization of pictures of furniture and vehicles. Although these studies seem to indicate that young infants can make category distinctions, a question remains as to the process (perceptual or conceptual) they are using to differentiate between categories of photographs (Madole & Oakes, 1999). Infants attend to differences in shape more than color, size, and texture (Nelson, 1973a; Ross, 1980; Starkey, 1981), but infants may attend to different properties for conceptual categorization. Nelson (1973a) found that 12- and 24-month-old infants categorized objects primarily by their shape and function (70% of the infants forming groups based on these properties). Madole et al. (1993) found, using an object-examining task, that 14-month-old infants discriminated between objects on the basis of function when presented with artificial objects constructed of Lego pieces, which had a familiarized form but novel function, and that 18-month-olds discriminated on the basis of the correlation between form and function, Tllhroforms (round and rectangular) and two functions (shaking and rolling) were used; each form crossed with a different function, so that one object of each form performed the two different functions. However, it is uncertain as to whether the infants distinguished between the categories of rolling and shaking because both the rolling and nonrolling objects had wheels, suggesting the function of rolling.

168

l

10.

The Reemergence

of Function

Other research suggests that infants younger than 18 months can attend to correlations between object features and function when there is a high degree of familiarity with the functional attributes. Booth (1998) found that when 14-month-olds actively manipulated the stimulus objects, they paid greater attention to the correlations between object features and function. Nevertheless, it is indeterminate as to which perceptual features the infants found the most salient for category distinctions. Booth used two artificial categories, which resembled families of flowers, that varied physically (within their petal complexity and color of petals) and functionally (inanimate versus animate). However, her results are ambiguous as to whether the infants were forming functional correlations with the features of petal complexity, color, or both. The results from the Booth (1998), Madole et al. (1993), and Nelson (1973a) studies indicate the difficulty of separating shape and function independently. As with real artifacts, stimulus objects do not easily vary physical attributes. It is, therefore, often uncertain as to whether infants are discriminating on the basis of shape, function, or the correlation between form and function. Madole and Cohen (1995) found that 18-month-old infants attended to the relation between form and function when it was within, not across, the parts of an object. The infants in this study were habituated to videotaped events of two objects that differed in the configuration of their parts and function. Both objects had wheels, but one was topped with a figure that turned and whistled, and the other was topped with a tree that did not move or make a sound. Depending on the color and texture of the wheels, the object would also roll. Infants dishabituated when the part correlated with a specific function no longer produced the function. Rakison and Butterworth (1998) found that 14-, 18-, and 22-month-old infants in an object-examining task also attended to object parts and formed categories, animals and vehicles, that were based on parts, legs or wheels, respectively. When the part was confounded across categories, such that the animals had wheels and the vehicles had legs, infants treated equivalently the animals with wheels and the vehicles with wheels. The authors claim that these findings imply that infants form dynamic categories on line that are based on the perceptual characteristics of the input. However, it is possible that the infants in this study were displaying a category of “toys with wheels.” In a recent critique of this literature, Mandler (2000) questioned the conclusions drawn. With regard to the picture-looking task, she argued that its passive nature and the young age (2 to 4 months) of the infants in these studies promotes categorization solely at the perceptual level.

Nelson and Ware

l

169

Perceptual categorization may occur early (at 2-3 months of age) based on the physical appearance of objects, including shape, color, size, and texture. Quinn and Johnson (1998) reported that 2-month-olds categorized pictures of mammals as different from furniture, presumably because of the considerable difference in shape among the two categories, suggesting that the infants learned a perceptual prototype or schema. Mandler claimed that “forming these kinds of perceptual categories or prototypes occurs without attention or intention and does not require conceptualization” (p. 7) She claimed that the notable overlap in shape among the objects in the photographs used by Eimas and Quinn (1994) an d o th ers renders them insensitive to higher-level categorization skills. Indeed, Mandler and McDonough (1993) found that infants from 7 to 11 months of age cannot differentiate between vehicles and animals in picture-looking tasks, but can do so when presented with realistic toy models in an object-examining task. Conceptual categories are formed in an effort to establish meaning, or to understand what sorts of things objects are. The ability to conceptually categorize, to go beyond the perceptual information provided by the senses, is a crucial aspect of development. As Mandler (2000) argued, it is a fundamental human capacity that differs greatly from simply being able to tell things apart. “The process is one that constructs concepts, and for infants it is part of creating a central, accessible representational system” (Mandler, 2000, p. 8). Mandler’s (2000) view contrasts sharply with that of other contemporary theories that have claimed that categorization skills in infancy develop from a perceptually centered discrimination to a more abstract, conceptual base (Eimas and Quinn, 1994; Madole & Cohen, 1995; Tversky & Hemenway, 1984). According to these theories, categorization begins at the perceptual level, based on the object’s shape, size, color, and other physical features, and gradually develops into conceptual categorization, which is based on function and the correlation of form and function. The alternative theory put forth by Mandler (2000) is that infants younger than 1 year categorize objects on a global domain level that goes beyond the singular object’s form and function. Infants begin to formulate concepts of things by “setting up kinds,” which are characterized by the roles they play in events. Evidence from Mandler and McDonough (1993) suggested that infants as young as 7 months can categorize at this global, domain-level. Mandler (2000) argued that the findings from studies like Rakison and Butterworth (1998) do not indicate that infants younger than 14 months of age are incapable of forming conceptual, global categories. Rather, she proposed that perceptual and conceptual categories each serve distinct functions. Perceptual categories are used for recognition

170

l

10.

The Reemergence

of Function

and object identification, whereas conceptual categories are used to control inductive generalizations. Mandler and McDonough (1998) found that although infants do use physical features to tell animals apart, they do not rely on them when they are constructing the meaning of an event, for example they did not give a drink of water to a Flying Tiger airplane despite its prominent mouth. Their results suggest an alternative explanation of the Rakison and Butterworth results, that the infants formed conceptual categories based on the functional attributes of the object parts (rolling versus walking). Overall, the infant literature suggests that categorization is a complex task with different bases that may function in dissimilar tasks and at different developmental levels. Mandler (2000) provided substantial evidence that infants not only form perceptual categories (beginning at 2-3 months of age), they can also form conceptual categories (by 7-8 months of age). The studies of children in their 2nd year that provide evidence of the use of both form and function suggest that the global categories that Mandler has identified in younger infants may become differentiated as word learning proceeds during the subsequent period of conceptual development. This implication is examined in the following sections. FORM AND FUNCTION IN WORD LEARNING AND CONCEPTUAL CATEGORIES AFTER INFANCY Mandler and other infant researchers concerned with categorization have not directly addressed the question of how the categorization skills observed in the first 2 years of life relate to the child’s acquisition of words and other linguistic categories (see McDonough, chap. 8, this volume). This was, however, the focus of the FCH (Nelson, 19 74), which was based on direct observation of children’s activities and word learning in the 2nd year (Nelson, 1973b). In the beginning, children seem to prefer to learn words that refer to things that do interesting actions or that are involved in interesting activities, for example, car, dog, kitty, clock, duck, hat. Some of these names were extended by toddlers to other instances that exhibited the same kinds of actions or were involved in similar activities. However, some were extended on the basis of singular perceptual features, or even on the child’s own orientation, as observations from Piaget (1962), Vygotsky (1962), L. Bloom (1973), and Bowerman (19 76) all documented in different ways. That concepts were formed on the basis of function and generalized on the basis of form seemed to bring together these disparate observations. The FCH was originally tested in a study (Nelson, 1973a) that showed that toddlers who knew the word ball when asked to choose a

Nelson and Ware

l

171

among 10 ball-like objects first picked objects that had the shape of balls. Then, after having the opportunity to examine and play with the balls, they homed in on those that could be used as balls, thrown, rolled, and so on, rather than those that were not movable. They apparently generalized their known word for the concept ball on the basis of what things looked like (their shape), but the presumed core their initial choices once its appliof the concept- function-overrode cability to specific examples was gained from experience. A later study found that 20-month-old children learning novel words for novel categories learned best on the objects with which they could do the most actions, that is, that had the most functions from the child’s point of view (Ross, Nelson, Wetstone, & Tanouye, 1986). Nelson (1978) extended this work with children learning their first words to preschoolers’ word meanings and concept structure. ?hro studies of 3- to 5-year-old children analyzed responses to the questions “what is X?” (the definition question) and “what do you know about X?” (the conceptual question). Basic level objects and superordinate categories (e.g., chair, furniture) were used in both studies. In line with predictions from the FCH, the strong findings were that all children predominately provided functional information to both questions, followed by perceptual properties for objects and instances for the categories. Anglin (1977) similarly reported that, in their definitions, young children tended to focus on what things do rather than what they look like. Additional support was found in the analysis of children’s word associations showing that action and function relations (syntagmatic) are the most common responses to common object words (e.g., tableeat) for children younger than 8 years, in contrast to the categorical (paradigmatic) responses (e.g., table-chair) given by older children and adults (Nelson, 19 7 7). Perceptual responses (e.g., table-big) are uncommon at all ages. Several related studies have been based on a study by Labov (1973), who tested the stability of word meanings in adults, demonstrating in a series of experiments that people would use perceptual characteristics differentially in giving names to items (specifically, cup, glass, or vase) in different functional contexts. For example, the same pictured vessel might be called a cup when holding a hot drink and a glass when used for a cold drink. Using a picture-naming paradigm similar to Labov’s and specification of functions that 4- and 5-year-olds would be familiar with, DeVos and Caramazzo found that children also adjusted their naming to functional contexts (cited in Nelson & Nelson, 1978, p. 244) . That is, they were likely to say that a particular vessel was a glass if it was imagined as holding lemonade but a cup if it was imagined as containing hot cocoa. These decisions were modified by

ball from

172

l

10.

The Reemergence

of Function

the height of the vessel; the taller the item, the more likely it was to be named a glass regardless of the drink. Even the youngest children (4and 5-year-olds) modified their decisions according to context, but also took account of the perceptual configuration. Labov’s point was that meanings are not timeless universals attached to context-free words, but depend on functional context of use, a point supported in this study with young children. Other studies, however, found fewer effects of function on children’s naming and concept formation. Andersen’s (1975) experiment, also based on Labov’s (1973) study of context effects on naming cups and glasses, reported that preschool children did not rely on conventional function in naming variations in vessel sizes and forms, although older children and adults did so. Two other studies using the cups and glasses paradigms (Anderson & Prawat, 1983 ; Prawat & Wildfong, 1980) agreed with Andersen, although their results could have reflected age variations in naming preferences. Further, a study by Tomikawa and Dodd (1980) concluded that children preferred form over function in learning names for novel objects. The most cited study countering the FCH was by Gentner (19 78). She presented two “machines” with different perceptual features and different functions to young children and adults with a name (e.g., “Jiggy”) given to each. Children interacted with the objects over the course of a week, and then the functions of the two were switched. After noting the switch, children and adults were asked to identify the “jiggy.” Both children and adults chose the perceptually identical object rather than the one with the same function. ( Note that this paradigm used specific objects, not instances of object categories, possibly suggesting that the word was a proper name, not a common noun indicating a category.) Together these counter-studies seemed to doom the function hypothesis at least for developmentalists, and it appeared to die a quiet and mostly unmourned death, leaving behind what appeared to be unresolved issues of conceptual development, perceptual distinctiveness, familiarity, naming, lexical development, and contextual usage, all of which might enter into the particular outcome on a function versus form test of concept formation. Summing up the case, Merriman (1985) presented a critique of the research literature on the FCH and other contemporary theories. He concluded that the FCH was inadequate to explain the data on word learning, specifically faulting the functional core theory in its applicability to early concepts and words. However, he concluded that there was a case “for the primacy of function over form in the organization of older children’s lexicons . . . [further] Function most likely assumes a

Nelson and Ware

l

173

greater importance in object word meanings as development proceeds” (p. 32). He admitted that neither he nor any other critic had proposed an account that could explain all of the disparate findings in the word learning and generalization studies reviewed. As with the tests of function in infancy, almost all the examinations of the question as to how children use function information involved situations in which function was tested against shape for the extension of words, with each varying independently so that two different shapes could display the same function, and two different functions could have the same shape. This is clearly not the case in the real world of obas modernist aesthetics jects. In most cases, “form follows function,” would have it; or as evolutionary theorists would emphasize, function follows form. That is, a particular function is enabled by its form. Changing the form is likely to undermine the function; changing the function is likely to lead to a change in form. Each constrains the other, reciprocally. This is as true for the objects that infants and young children engage with as for any others, preventing a clear conclusion that one dimension is preferred over the other. Happily, the conclusion that function and the functional core were dead as far as conceptual theory was concerned was an illusion. Although the issue was dormant in the developmental literature, it reappeared in basic theories of conceptual structure in the general cognitive literature, albeit without reference to the earlier developmental proposals. ESSENCES, CORES, THEORIES, AND FUNCTION: THE NEW LOOK Following Smith and Medin’s (198 1) review documenting the demise of defining feature theories and critiquing the family resemblance theory, function as the conceptual core of artifact concepts emerged in the concept literature with adults, responding to the need for positing “deeper V properties for artifacts as well as natural kinds (Barsalou, 1989; Medin & Ortony, 1989; Rips, 1989; see Malt & Johnson, 1992, for a review and references). Recall that Miller and Johnson-Laird ( 19 76) had made function of objects a prime component of their lexical concept theory. Reflecting some of the same ideas, Medin and his colleagues (Medin, 1989; Murphy & Medin, 1985) argued that the way we form categories and relate them to each other is determined by the theories we hold about the real world within particular domains of knowledge. Our concepts are entrenched in and defined by our theories, it is claimed, whereas relations between concepts will be determined by the particular theoretical structure.

174

l

10.

The Reemergence

of Function

Summarizing over several related proposals in this vein, Medin and Ortony (1989) considered the relation of deeper conceptual structure, including function, to perceptual similarity, concluding with three key points: First, people act as if their concepts contain essence placeholders that are filled with “theories” about what the corresponding entities are. Second, these theories often provide or embody causal linkages to more superficial properties. Our third tenet is that organisms have evolved in such a way that their perceptual (and conceptual) systems are sensitive to just those kinds of similarity that lead them toward the deeper and more central properties . .. Appearances are usually not deceiving . . . [organisms] will not be led far astray because many of these surface properties are constrained by deeper properties. (p. 180)

Medin and Ortony (1989) go on to suggest that it is plausible that young children are most sensitive to global surface properties, as these will be those most likely to be connected to the deeper ones and thus will lead them in the direction of the correct categories. Although the functional core plus identification proposal thus has appeared in several guises as a theory with considerable support from research with adults as well as children, Malt and Johnson (1992) questioned its generality. They tested the idea in a series of experiments asking college students to say whether an object with a familiar function but unusual physical characteristics was a member of the category (e.g., chair, boat, desk) associated with the function. They found that participants often rejected members with the allegedly defining function and accepted members that were perceptually similar but did not have the assumed central function. I? Bloom (1996), taking account of Malt and Johnson’s (1992) results, introduced the idea of the historical intentional basis for inclusion of items in a category of artifacts. He proposed that items are held to belong to a class if they were intentionally created to serve as a member of that category. This approach can account for cases where the function is not operative, as in broken objects and pictures of objects, but it may not be successful in accounting for the development of such concepts, as young children are unlikely to take account of the intention of the creator of the object. CATEGORIZATION

AND NAMING

BY CHILDREN:

AN UPDATE

Keil(l987) turned to the “theory theory” of concepts to explain their development, claiming that young children have theories based on perceptual characteristics of the world, and that with development they gain understanding of “deeper,” more definitional features of things. In

Nelson and Ware

l

175

the case of artifacts, the function, intended or common use, of objects is seen as defining the type. This theory is plainly a variation on the functional core plus identificational features hypothesis (Nelson, 19 74), but it reverses the developmental proposal, claiming that children develop the core properties only later, relying at first on perceptual features. Note that this and other current theories are not concerned with what bases the child uses in forming a concept, but with how the child uses the information in induction tasks for generalizing to new instances or new properties. In contrast, the importance of function in the FCH was viewed in terms of its role in establishing the basis for forming the concept in the first place. In other developmental literature, the contrast between perceptual and conceptual bases has reappeared in the form of the perception-function controversy as part of a broader concern with the bases for children’s acquisition and extension of words. Today, the relation of words and things is in hot dispute. Renewed attention to the basis for early word learning and generalization emerged at about the point where Merriman’s (1985) review showed that the semantic feature and functional core hypotheses were running aground. A number of studies (Markman & Hutchinson, 1984; Waxman & Gelman, 1986) suggested that children may rely on different defining characteristics for named in contrast to not-named objects, specifically that they choose category rather than thematic associates when objects are named. The question of whether words have a specific shape-eliciting or taxonomic category effect has been raised by a number of researchers. Bauer and Mandler (1989) showed that toddlers can make category-related choices in the absence of words when provided with suitable instructions. They argued that there is nothing magical about words in relation to taxonomic categories. Although it may be true that object nouns refer to categories, and not thematic relations, and that children of 3 years understand this, this does not show the basis for such choices. Moreover, not all words, even nouns, refer to objects; some refer to thematic situations (e.g., lunch or party) which include a number of thematically-related objects that are not members of a higher-order taxonomic level but could be considered parts of a collection (Markman, 1981). Nelson (1978) referred to these words as “spatiotemporal-organizers” and found that preschoolers readily supplied definitions for them and listed objects that belonged with them, such as foods for lunch. A major issue emerging from the claims for early biases in word meaning is whether shape is the primary dimension that determines children’s and adults’ extension of the reference of words. Jones and Smith (1993) argued against the view that children’s concepts have

176

l

10.

The Reemergence

of Function

“cores” composed of “nonobvious properties” (the latter referring to essences, function, and defining properties), and for the view that people, including children, are sensitive to perception, especially shape, in making category decisions in different contexts. One of the major targets in Jones and Smith’s argument is the idea that perception is a peripheral part of concept structure where function constitutes a core. Smith (1999) claimed that naming is based on a “dumb” perceptual association process rather than a “deep” reasoned inferential process. Thus the controversy returns to the earlier perception-function debate. The Jones and Smith (1993) argument has not gone unchallenged (see commentaries by Barsalou; Gelman & Medin; Mandler; Mervis, Johnson, & Scott, in the same issue of the 1993 journal of Cognitive Development). Kemler Nelson has taken up the challenge from Smith and her colleagues. In a series of important papers she clarified two issues: (a) Do children make different decisions about object category membership when objects are named than when they are not named? (b) Do they base their decisions on shape (or other perceptual characteristics) or on function? (Kemler Nelson, 1999; Kemler Nelson, Russell, Duke, & Jones, 2000; Kemler Nelson & Students, 1995). She showed that when functions are understandable or familiar to young children, they use function in extending names to novel things, even when these do not share perceptual similarity with the item for which the name was learned. She also showed that, contrary to Landau, Smith, and Jones’s ( 1988) claims, when children are given time to reflect on function, and are not pressed for time in responding, they make choices on the basis of function in preference to perception. Even 2-year-olds in her studies rely on function in their choices in preference to perception. Moreover, she found little difference between children’s named and not-named choices. Together, these studies contradict the Smith and Jones (1993) and Landau et al. (1988) claims that children’s word extensions are based solely on “dumb attentional mechanisms.” Rather, according to Kemler Nelson and her colleagues, function is meaningful to children who understand how the artifacts work (Kemler Nelson, Frankenfield, Morris, & Blair, 2000). Kemler Nelson is not the only researcher in recent years who has addressed children’s use of function in word meaning and concept formation. Gathercole (Gathercole & Whitfield, 2000; Gathercole, Cramer, Somerville, & Jansen op de Haar, 1995) explored some of the complexities of the use of function in word extensions by both children and adults. Deborah Kelemen (1999) investigated children’s preference for functional explanations, and Tim German (2000) used a study of functional fixedness to probe the developmental course of functional flexibility in children’s thinking. Space constraints prohibit reviewing these

Nelson and Ware

l

177

and other new research (some of it in dissertation form, such as Eriksson, 1990) focused on the role of function in the thinking, concepts, and language of both children and adults. We welcome this renewed attention to function in developmental research. Yet we would add that what is missing from much of the debate is the function of concepts and the meaning of language. Why should anyone focus on something to form a concept of it? Why have a word in the first place? Why not just point to an object? Smith and Jones (1993) appear to confuse two levels; the mechanism of cognition and the content and function of cognition. We agree that dynamic systems are a good way of conceptualizing cognitive mechanisms, and that they can explain a good deal about people’s online decisions in context. However, humans seek for meaning, and we believe, as does Mandler, that they seek for meaning from the beginning of life. Perceptual information is critical to identifying the meaning of things, but so is inference, memory, interaction, and communication with others. Thus, we think that the emphasis on perceptual information is important but is not the only important variable in how humans deploy their knowledge or form new understandings. IMPLICATIONS FOR CONCEPTUAL AND COGNITIVE DEVELOPMENT: FROM INFANT CATEGORIES TO ADULT CATEGORIES Does Mandler’s infant work tell us something about basic cognitive processes, functions, or mechanisms that is relevant as well to cognitive development in childhood and to cognition in adults? We believe that this survey, incomplete as it may be, and inconclusive as both data and theory appear, does have implications for understanding the nature of conceptual categories and their relations to language. To summarize briefly, Mandler ‘s work (see Mandler, 2000) shows that infants sort objects into general global categories that reflect roles in events. Other work with infants shows that very young infants categorize objects and pictures on perceptual bases as well. When learning first words for objects at the basic level, infants appear to be sensitive to the functions of things (Lucariello, 1987; Nelson, 1974, 1979; Ross et al., 1986). This suggests a continuity with their categorization at the end of the 1 st year, but also implies a differentiation of global categories in response to the naming practices of the linguistic community. Over the 2nd year, toddlers learn to attend to shape as a major important basis for extending count nouns, as Smith (1999) showed, and this is reflected also in many of the word-naming paradigms with young preschoolers.

178

l

10.

The Reemergence

of Function

Young children continue to be attentive to function in defining object concepts and words, as shown in studies by Anglin (19 7 7) and Nelson (19 78), as well as in tasks such as word associations (Nelson, 19 77). And, in making choices among objects that are like in function although dissimilar in shape or form, they may choose function over form when function is familiar or relevant to the choice (Kemler Nelson et al., 2000). Although not reviewed here, function at the level of superordinate categories is also evident in children’s “slot-filler N categories (Lucariello, Kyratzis, & Nelson, 1992), where children relate basic level items at a higher level on the basis of their event roles. This process implies a new integration within the domains that were previously differentiated from global categories in response to naming of basic level categories. Thus function appears to play a role in conceptual structure and word meaning from infancy through the childhood years, although it does not usually override perceptual information, especially shape, in tasks such as the extension of words to new items or the choice of items in response to object names. These developmental observations cohere with general theoretical proposals about conceptual structure and its relation to semantics, especially the proposals of Miller and Johnson-Laird (1976), and the ideas about “deep” levels of artifact concepts analogous to the “essences” proposed for natural kinds. Although the kinds of functions that young children are sensitive to may not be the kinds of intentional design on which adults rely, there is a commonality of attending to deeper levels of meaning than the surface appearances of things may reveal. This does not mean that surface appearance is irrelevant; it is generally correlated with function, which makes tests of the importance of each in people’s choices difficult. Moreover, in making quick decisions about the category membership of something (for example, “Is that a tiger stalking toward you?“), perceptual characteristics and context are likely to be more important than judgments about the essence of the beast. We suggest that the first global categories that divide the world of things into animals and vehicles (or nonanimals) become differentiated on the basis of individual kinds of action potentials as well as shape in response to learning words. These specific lower-level categories are then reintegrated into new higher-level function-organized categories and these into higher-order culturally and linguistically defined taxonomies based on abstract function. Fig. 10.1 provides a schematic view of how such a process might proceed. As these levels become consolidated, a new level differentiating subcategories based on culturally significant functions or categories may emerge (e.g., footwear, meat, literature, etc.) Increasingly, bases for constructing categories, finding new bases, and

Nelson and Ware

l

179

Category Development

Global Collection Superordinate

FIG. 10.1 Suggested developmental course of conceptual category development in relation to word learning development, beginning with global- and event-based categories, followed by differentiation and reintegration into larger, more articulated, and hierarchically organized categories.

composing organizations with different relational structures (theory domains), become more complex and integrative. Throughout this long and complicated process, children, like adults, try to make sense, to find meaning in the encountered world and its language. Words and concepts are in constant contact in the effort to construct meaning. Function represents one of the first and most important components of the “deeper,” more meaningful levels of understanding and interacting with objects. As Werner’s organismic principle proposed (Werner & Kaplan, 1963), the basic developmental process involved in this progression is that of differentiation of global structures, followed by integration, further differentiation, integration, differentiation, integration, and so on, at increasingly complex levels. We began with the statement that Mandler’s work has profound implications for our understanding of cognitive development. It identifies the process of the search after meaning in terms of conceptualization of the things of the world beginning in infancy. The world is not just a scene, but an encounter. Making sense is necessary to act in the world of meaning. However, at the same time, the concepts that infants construct are those that are relevant to their encounters, from the perspec-

180

l

10.

The Reemergence

of Function

tive of the small, unsteady, unskilled, naive child. They rest on the same basic interests (making meaning) and sources (perception, inference) as later concepts, but they still lack one of the most powerful sources for constructing meaninglanguage. Taking this beginning seriously and following it through the acquisition of meaning in language and the opening up of the world to knowledge acquired through language will show both what is the same and what is different about infancy and the toddler years as they change into early and later childhood. Yes the processes are the same or similar, but they change the product in radical ways as language and culture enter the child’s mind. Mandler’s work enables us to see the connections between these developmental levels, as well as the changes they must undergo, more clearly.

Andersen, E. S. (1975). Cups and glasses: Learning that boundaries are vague. Journal of child language, 2, 79-104. Anderson, A. L. N., & Prawat, R. S. (1983). When is a cup not a cup? A further examination of form and function in children’s labeling responses. MerriZZPalmer Quarterly, 29, 3 75-3 85. Anglin, J. (1977). Word, object and conceptual development. New York: Norton. Barsalou, L. A. (1989). Intraconcept similarity and its implications for interconcept similarity. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 76-121). Cambridge: Cambridge University Press. Barsalou, L. W. (1993). Challenging assumptions about concepts. Cognitive Development, 8, 169-l 80. Barsalou, L. W. (1999). Perceptual Symbol Systems. Behavioral and Brain Sciences. Bauer, I? J., & Mandler, J. M. (1989).Taxonomies and triads: Conceptual organization in one- to two-year-olds. Cognitive Psychology, 22, 156-l 84. Behl-Chadha, G. (1996). Superordinate-like categorical representations in early infancy. Cognition, 60, 104-141. Bierwisch, M. (1970). Semantics. In J. Lyons (Ed), New Horizons in Linguistics. Baltimore: Penguin. Bloom, L. (1973). One word at a time. The Hague: Mouton. Bloom, I? (1996). Intention, history, and artifact concepts. Cognition, 60,1-29. Booth, A. (1998). The roleoffinctional information in the development of concepts of objects in infancy. Unpublished doctoral dissertation. University of Pittsburgh, Pennsylvania. Bornstein, M. H. (1984). A descriptive taxonomy of psychological categories used by infants. In C. Sophian (Ed), Origins of cognitive skills. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Bowerman, M. (1976). Semantic factors in the acquisition of rules for word use and sentence construction. In D. M. Moorehead & A. E. Moorehead (Eds.), Normal and deficient child language. Baltimore: University Park Press. Clark, E. V (19 73). What’s in a word? On the child’s acquisition of semantics in his first language. In T E. Moore (Ed), Cognitive development and the acquisition of language. New York: Academic.

Nelson and Ware

l

181

Cohen, L. B., Celber, E. R., & Lazar, M. A. (1971). Infant habituation and generalization to repeated stimulation. Journal OfExperimental Child Psychology, 11, 379-389. Eimas, I? D., & Quinn, F?C. (1994). Studies on the formation of perceptually based basic-level categories in young infants. Child Development, 58, 1523-1531. Eriksson, M. (1990). Functional Relations and the Development of Concepts: Three studies in relation to the functional core model. Doctoral Dissertation. Uppsala University, Sweden. Fodor, J. (1972). Some reflections on L. S. Vygotsky’s Thought and Language Cognition, 1, 83-95. Fodor, J. A. (1975). The language of thought. New York: Crowell. Gathercole, V., Cramer, L., Somerville, S., & Jansen op de Haar, M. (1995). Ontological categories and function: Acquisition of new names. Cognitive Development, 10, 225-25 1. Gathercole, V. C. M., & Whitfield, L. C. (2000). Function as a Criterion for the extension of new words. Journal of Child Language, 28, 8 7-125. Gelman, S. A., & Medin, D. L. (1993). What’s so essential about essentialism? A different perspective on the interaction of perception, language and conceptual knowledge. Cognitive Development, 8, 15 7-l 68. Gentner, D. (1978). A study of early word meaning using artificial objects: What looks like a jiggy but acts like a zimbo? Papers and Reports on Child Language Development, 15, l-6. German, T. I? (2000). Developing knowledge of function and the solution of insight problems. Paper presented at the International Conference on Infant Studies. Brighton UK, July 2000. Inhelder, B., & Piaget, J. (1964). The earlygrowth oflogic in the child. New York: Harper Press. Jones, S. S., & Smith, L. L. (1993). The place of perception in children’s concepts. Cognitive Development, 8, 113-l 39. Keil, F. C. (1987). Conceptual development and category structure. In U. Neisser (Ed), Concepts and conceptual development: Ecological and intellectual factors in categorization. (pp. 175-200). New York: Cambridge University Press. Kelemen, D. (1999). Why are Rocks Pointy? Children’s Preference for Functional Explanations. Poster presented at the biennial meetings of the Society for Research in Child Development. Albuquerque, NM, April 1999. Kemler Nelson, D. G. (1999). Attention to functional properties in toddlers’ naming and problem-solving. Cognitive Development, 14, 7 7-l 00. Kemler Nelson, D. G., Frankenfield, A., Morris, C., & Blair, E. (2000). Young children’s use of functional information to categorize artifacts: three factors that matter. Cognition, 75, l-36. Kemler Nelson, D. G., Russell, R., Duke, N., & Jones, K. (2000). l%vo-year-olds will name artifacts by their functions. Child Development, 71, 1271-1288. Kemler Nelson, D. G., & 11 Swarthmore College Students (1995). Principlebased inferences in young children’s categorization: Revisiting the impact of function on the naming of artifacts. Cognitive Development, 10, 347-380. Krauss, R. (1952). A Hole is to Dig: A First Book of First Definitions. New York: Harper and Brothers. Labov, W. (1973). The boundaries of words and their meanings. In C.-J. N. Bailey & R. W. Ghuy (Eds.), New ways of analyzing variation in English. Washington, DC: Georgetown U. Press.

182

.

10.

The Reemergence

of Function

Landau, B., Smith, L. B., & Jones, S. S. (1988). The importance of shape in early lexical learning. Cognitive Development, 3, 299-33 1. Lucariello, J. (1987). Concept formation and its relation to word learning and use in the second year. Journal of Child Language, 14, 309-332. Lucariello, J., Kyratzis, A., & Nelson, K. (1992). Taxonomic knowledge: What kind and when. Child Development, 63, 978-998. Macnamara, J. (1982). Names for things. Cambridge, MA: MIT Press. Madole, K. L., & Cohen, L. B. (1995). The role of object parts in infants’ attention to form-function correlations. Developmental Psychology, 32,63 7-648. Madole, K. L., & Oakes, L. M. (1999). Making sense of infant categorization: Stable process and changing representations. Developmenta Review, 19, 263-296. Madole, K. L., Oakes, L. M., & Cohen, L. B. (1993). Developmental changes in infant’s attention to function and form-function correlations. Cognitive Development, 8, 189-209. Malt, B. C., & Johnson, E. C. (1992). Do artifact concepts have cores? Journal of Memory and Language, 3,195-217. Mandler, J. (1993). On concepts. Cognitive Development, 8, 141-148. Mandler, J. M. (2000). Perceptual and conceptual processes in infancy. Journal of Cognition and Development, 1, 3-36. Mandler, J. M., Bauer, F?J., & McDonough, L. (1991). Separating the sheep from the goats: Differentiating global categories. Cognitive Psychology, 23, 263-298. Mandler, J. M., & McDonough, L. (1993). Concept formation in infancy. Cognitive Development, 8, 291-318. Mandler, J. M., & McDonough, L. ( 1998). Studies in inductive inference in infancy. Cognitive Psychology, 37, 60-96. Markman, E. M. (1981). ?hro different principles of conceptual organization. In M. Lamb & A. Brown (Eds.), Advances in developmental psychology. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Markman, E. M., & Hutchinson, J. E. (1984). Children’s sensitivity to constraints on w&d meaning: Taxonomic vs. thematic relations. Cognitive Psychology, 16, l-27. McDonough, L., & Mandler, J. M. (1998). Inductive generalizations in 9- and 1 1-month-olds. Developmental Science,I, 227-232. Medin, D. L. (1989). Concepts and conceptual structure. American Psychologist, 44,1469-1481. Medin, D., & Ortony, A. N. (1989). Comments on Part I: Psychological Essentialism. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 179-l 95).Cambridge, England: Cambridge University Press. Merriman, W. E. (1985). How children learn the reference of concrete nouns: A critique of current hypotheses. In S. A. I. Kuczaj & M. D. Barrett (Eds.), The Development of Word Meaning: Progress in Cognitive Development Research. New York: Springer-Verlag. Mervis, C. B., Johnson, K. E., & Scott, I? (1993). Perceptual knowledge, conceptual knowledge, and expertise: Comment on Jones and Smith. Cognitive Development, 8, 149-156. Miller, G. A., & Johnson-Laird, I? N. (1976). Language and Perception. Cambridge, MA: Harvard University Press. Murphy, G. L., & Medin, D. L. (1985). The role of theories in conceptual coherence. Psychological Review, 92, 289-3 16.

Nelson and Ware

l

183

Nelson, K. (1973a). Some evidence for the cognitive primacy of categorization and its functional basis. Merrill-Palmer Quarterly, 19, 21-39. of Nelson, K. (1973b). Structure and strategy in learning to talk. Monographs the Societyfor Research in Child Development, 38, 1-2, Serial No. 149. Nelson, K. (1974). Concept, word, and sentence: Interrelations in acquisition and development. Psychological Review, 8 I, 26 7-285. Nelson, K. (1977).The syntagmatic-paradigmatic shift revisited: A review of research and theory. Psychological Bulletin, 84, 93-l 16. Nelson, K. (1978). Semantic development and the development of semantic memory. In K. E. Nelson (Ed), Children’s language. Vol 1. New York: Gardner. Nelson, K. (1979). Explorations in the development of a functional semantic system. In W. Collins (Ed.), Children’s language and communication: Minnesota symposium on child psychology. Vol. 12. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Nelson, K. (1985). Making sense: The acquisition ofshared meaning. New York: Academic. Nelson, K. E., & Nelson, K. (19 78). Cognitive pendulums and their linguistic realizations. In K. E. Nelson (Ed), Children’s Language. Vol. 1. New York: Gardner Press. Olver, R. R., & Hornsby, J. R. (1966). On equivalence. In J. S. Bruner, R. R. Olver, & I? M. Greenfield (Eds.), Studies in Cognitive Growth. New York: Wiley. Piaget, J. (1962). Play, dreams, and imitation in childhood. New York: Norton. Prawat, R. S., & Wildfong, S. (1980). The influence of functional context on children’s labeling responses. Child Development, 51, 105 7-l 061. Quinn, I? C., & Johnson, M. H. (1998, April). Global before perceptuaz category representations in connectionist networks and Z-month-old infants. Paper presented at the International Conference on Infants Studies, Atlanta, GA. Rakison, D., & Butterworth, G. E. (1998). Infants use of object parts in early categorization. Developmental Psychology, 34, 49-62. Riccuiti, H. N. (1965). Object grouping and selective ordering behavior in early infancy. Merrill-Palmer Quarterly, I 1, 129-l 48. Rips, L. J. (1989). Similarity, typicality, and categorization. In S. Vosniadou and A. Ortony (Eds.), Similarity and analogical reasoning (pp. 21-59). Cambridge, England: Cambridge University Press. Rosch, E., Mervis, C., Gray, W., Johnson, D., & Boyes-Braem, P (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382-439. Ross, G. S. (1980). Categorization in l-to 2-year-olds. Developmental Psychology, 16, 391-396. Ross, G., Nelson, K., Wetstone, H., & Tanouye, E. (1986). Acquisition and generalization of novel object concepts by young language learners. Journal of Child Language,

13, 67-84.

Smith, L. B. (1999). Children’s noun learning: How general learning processes make specialized learning mechanisms. In B. MacWhinney (Ed), The Emergence of Language (pp. 277-304). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Smith, L. B., & Jones, S. S. (1993).Cognition without concepts. Cognitive DeveZopment, 8, 181-188. Smith, E. E., & Medin, D. L. (1981). Categories and concepts. Cambridge, MA: Harvard University Press. Starkey, D. ( 198 1). The origins of concept formation: Object sorting and object preference in early infancy. Child Development, 52, 489-497.

184

l

10.

The Reemergence

of Function

Tomikawa, S. H., & Dodd, D. H. (1980). Early word meanings: Perceptually or functionally based? Child Development, 51, 1103-l 109. Tversky, B., & Hemenway, K. (1984). Objects, parts, and categories. Journal of Experimental Psychology: General, 113, 169-l 91. Vygotsky, L. (1962). Thought and language. Cambridge, MA: MIT Press. Waxman, S., & Gelman, R. (1986). Preschoolers’ use of superordinate relations in classification and language. Cognitive Development, 1, 139-156. Werner, H., & Kaplan, B. (1963). S’bolformation: An organismic-developmental approach to language and the expression of thought. New York: Wiley.

The Procedural-Procedural Knowledge Distinction Mitchell Rabinowitz Fordham University

I

‘ve always thought of academic training in terms of family structures and accordingly, I consider Jean to be my academic “mom. N However, I had to work hard and be very creative to convince Jean to adopt me. When I first met Jean, I was formally enrolled as a graduate student at the University of California, Berkeley, but 1 was visiting at the University of California, San Diego (UCSD) for the year. At that time Jean graciously allowed me to sit in on her lab meetings but told me that she was pretty busy and probably would not have time to work with me on any research projects. As the year progressed, I enjoyed talking with Jean so much that I came up with the following idea-1 reminded Jean that George, her husband, worked with a graduate student named Jan Rabinowitz (no relation) and that they published articles together. I told Jean that if we conducted a study together and published it, there would be multiple articles by Mandler and Rabinowitz, with four different authors. Jean liked the idea so much that we started collaborating. I ended up transferring to UCSD and Jean became my graduate advisor. Our article was published (Rabinowitz & Mandler, 1983) and confusion reigns. 185

186

’

11.

Procedural

Knowledge

Distinction

Actually, my chapter is all about confusion--confusion about words and their referents. When I transferred to UCSD, Jean was working on writing her chapter on “Representation” for the Handbook of ChiZd Psychology (Mandler, 1983). In addressing how knowledge is represented during the acquisition of skill throughout development, Mandler (1983, 1998) argued for a hybrid model, the need to distinguish between two different types of knowledge, and centered on the distinction between procedural knowledge (knowing how) and declarative knowledge (knowing that). She argued that development can be thought of as progressing from a purely procedural system to one that integrates procedural and declarative knowledge. The issue I address is what are the referents for procedural and declarative knowledge? One of the defining features that distinguishes procedural from declarative knowledge is accessibility, which refers to the ability to be aware and introspect about the knowledge. Procedural knowledge is considered to be inaccessible. Mandler (1998) used the example of the ability to tie one’s shoelace. We all know how to do it, but we have a hard time describing the process. On the other hand, declarative knowledge is something that we can think and talk about. We know that we are writing these chapters in honor of Jean Mandler and that they will be published in an edited book. These are facts that can be accessed and discussed. According to Mandler, “Only declarative knowledge is accessible to conscious awareness. Procedural knowledge remains inaccessible (nonconscious). We can only observe the products of our procedures, not the procedures themselves . . . When we do conceptualize a procedure, we construct a theory from observing the outcome of various steps along the way.. .” (Mandler, 1998, pp. 265-266). This characterization of the distinction between procedural and declarative knowledge is well accepted within the literature. However, there doesn’t appear to be the same level of consensus regarding the type of representation instantiated by these two types of knowledge. For example, Mandler (1998) stated: I‘. . . the distinction is implicated in the dispute over symbolic and subsymbolic representational systems, because tasks involving declarative knowledge seem better handled by symbolic systems and tasks involving procedural knowledge better handled by connectionist systems.” (pp. 264-265). Both of the terms symbolic and connectionist systems will be described later in the chapter, but the relevant point here is that Mandler relates the symbolic SYStern (which she says productions exemplify; 1998, p. 255) as instances of declarative knowledge, whereas connectionist systems are examples of procedural knowledge (see also Schacter, 1987; Squire, 1987; for similar descriptions). Contrast this characterization with a description

Rabinowitz

l

187

of the declarative-procedural distinction offered by Anderson (Anderson & Lebiere, 3 998), where procedural knowledge is instantiated by a symbolic system (productions) and declarative knowledge is instantiated by a connectionist type of system (see also Kintsch, 1992, and Rabinowitz & Goldberg, 1995. for similar descriptions). The assignment of systems to instantiate these two types of knowledge (declarative-procedural) is exactly opposite. (I did mention earlier in the chapter that confusion reigns.) Now, the fact that one group of people can take a representational system and characterize it as procedural knowledge and another characterize it as declarative knowledge raises the possibility that the two types of systems may be formally equivalent; that is, they are different in reference to surface level characteristics but the same in deep, structural features. In fact, both connectionist and production systems models exist that argue for a unified theory of representation (Newell, 1990; Rumelhart & McClelland, 1986). In this chapter, however, I will agree with Jean and argue that we need to hypothesize a hybrid representational system, although I feel that a more accurate description would be one that distinguishes between two procedural knowledge systems. The two systems, the same symbolic and network systems that Mandler (1998) discussed, are based on different metaphors of information processing; “the mind is like a computer” and “the mind is like the brain.” Each of these metaphors offers a different computational model of information processing. I will not argue that we need to choose between the two metaphors, but rather that both types of computational models are needed to understand cognition and development. One issue I had in writing this chapter was how to refer to these two models. It’s one thing when you have a hybrid model such as procedural and declarative knowledge; the decision is simply “procedural” and “declarative.” However, what do you do when you are arguing for a hybrid procedural system; “Procedural 1” and “Procedural 2‘7 It doesn’t work. Given that one model is based on the computer metaphor, I refer to that type of procedural knowledge as “symbol manipulation” or “rule-based” knowledge. Given that the second model is based on the brain metaphor, I refer to that type of procedural knowledge as “network” or “retrieval” knowledge. In the remainder of the chapter, I discuss the two different metaphors and the implications of each for characterizing procedural knowledge. I then present a study designed to experimentally distinguish the performance characteristics of the two types of knowledge in terms of learning and transfer characteristics.

188

.

11.

Procedural

Knowledge

Distinction

THE FIRST TYPE OF PROCEDURAL KNOWLEDGESYMBOL MANIPULATION PROCESSES During my formative years, the dominant metaphor for thinking about intelligence was the “mind is like a computer * (Turing, 1963; von Neumann, 195 1). The argument went something like this: people, like computers, could be conceptualized as general information processing devices. People have general hardware constraints, data, and processes that they use to manipulate the data. Accordingly, research on the nature of intelligence attempted to specify the “program,” the “software,” or the rules people use (consciously or unconsciously) to manipulate symbols (data). Procedural knowledge was characterized as these processes or rules that manipulated the data, or declarative knowledge. Researchers who adopt this metaphor frequently describe procedural knowledge in terms of production systems models (Anderson, 1993; Klahr, Langley, & Neches, 1987; Newell & Simon, 1972). A production system consists of a set of rules, called productions, written in the form of condition or action (IF or THEN) pairs. According to Newell ( 19 73) “A production system, starting with an initially given set of data structures, operates as follows. That production whose condition is true of

the current data (assume there is only one) is executed, that is, the action is taken. The result is to modify the current data structures. This leads in the next instant to another (possibly the same production being executed, leading to still further modification. So it goes, action after action being taken to carry out an entire program of processing, each evoked by its condition becoming true of the momentarily current collection of data structures. The entire process halts either when no condition is true (hence nothing is evoked) or when an action containing a stop operation occurs.” (p. 463)

There are two other general features of productions that will become important later in the chapter. First, each production is modular; productions are independent of each other and can be separately added, deleted, or modified. Second, productions are directional and not reversible; they work in the direction of condition to action and can not reverse themselves to go from the action to reinstate the condition (Anderson, 1993; Anderson & Lebiere, 1998). Newell goes on to say that these characteristics provide a general description of production system models, but that there are a variety of ways in which one can implement specific constraints within the system: “Each assemblage . . . yields a different production system with different properties from its siblings. Taken in all, they constitute a family of schemes for specifying information processing systems” (Newell,

Rabinowitz

l

189

19 73, p. 464). Although there have been advances and modification regarding the implementation of production systems, the general family of production systems still follows these general constraints. The productions, then, are explicit representations of the rules that manipulate the data. Once these rules are written down within a production system model they can be discussed. Thus, according to Mandler (1998), these rules “all could be easily described and understood using natural language terms” (p. 255), and this “knowledge is accessible to conscious awareness” (p. 265). This characterization is not accurate. The specification of the productions that produce performance reflects hypotheses that researchers make about how people think and how they learn. It is important to point out the use of the words “reflects hypotheses.” The people who are actually processing the information with these rules do not have access to the rules. The models that are constructed are considered to be hypotheses because the operation of the productions themselves are inaccessible to introspection. Researchers construct a model of the processing from observing the outcome (what gets into consciousness) of the procedures along the way. Thus, a production is an exemplar of the category of procedural knowledge whose main purpose is to create and manipulate symbols. THE SECOND

TYPE OF PROCEDURAL NE-IWORK MODELS

KNOWLEDGE-

As stated earlier, one of the implications of the “mind is like a computer V metaphor is that rules operate on data and these data are stored in some memory location. Retrieval involves going to where it is stored and accessing it. This orientation has led to the investigation of the content and structure of data. Two very important theoretical constructs, associative networks and spreading activation, were derived from this research (Anderson & Bower, 1973; Collins & Quillian, 1969; Norman & Rumelhart, 1975) which, I believe, changed how we think about data and led the field to consider a different metaphor-“the mind is like a brain.” Within an associative network, concepts are represented as nodes that are interconnected by associative links. Retrieval is accomplished by following the appropriate links within the network. Along with the structural description of the knowledge organization, researchers also posited a process that operated on this structure, namely, the automatic spread of activation. The basic assumption of this model is that each node has some level of activation associated with it. When an item in the environment is encountered, the corresponding concept in memory is activated. Activation then spreads from that concept to related concepts across the associative links. The amount of activation that

190

.

11.

Procedural

Knowledge

Distinction

spreads depends on the strength of the associative links between concepts, with stronger associations leading to stronger activation. This spreading of activation is thought to occur automatically without the conscious control of the learner. Spreading activation systems have two different types of associative links-excitatory and inhibitory. Excitatory links increase the level of activation, and inhibitory links decrease the level of activation of a node’s neighbors. The inclusion of both types of links allows spreading activation models to become “constraint satisfaction” models. A number of researchers contend that because of these implicit constraints, rules or procedures can be implicitly represented within these associative links (Rabinowitz, Lesgold, & Berardi, 1988; Rumelhart & McClelland, 1986). The implication of this status is that the system has the capability of making “intelligent” choices regarding knowledge activation. The spreading activation conception of information processing is a precursor to and embodied within the recent work on connectionist or Parallel Distributed Processing models (Rumelhart & McClelland, 1986). In this conception, the computational metaphor is the information processing of neurons in the brain, not information processing of the computer. The brain consists of a large number of highly interconnected simple elements (neurons) that send very simple excitatory and inhibitory messages to each other. These elements update their excitations on the basis of these simple messages. There are a number of basic assumptions concerning the defining characteristics of connectionist models. The main one is that units within the network are subsymbolic and information processing takes place through the interactions of a large number of simple processing units or nodes. Each of these units has some level of activation associated with it and is designed to receive input from its neighbors. As a function of the inputs it receives, the unit computes an output value which it then sends to its neighbors. An additional defining feature of connectionist models is that they are inherently parallel; that is, many units can carry out their computations at the same time. Accordingly, meaning is derived from the pattern of activation across units in the network once the constraints of activation (excitatory and inhibitory) have settled down. Connectionist models, however, vary in the manner in which these defining features are implemented. As with the production system models, connectionist models should be seen as a class of models with each implementation involving a set of slightly different assumptions. However, the knowledge embodied in all connectionist models, as Mandler (1998) pointed out, is procedural knowledge. A person does not have accessibility to the spreading of activation and the computa-

Rabinowitz

tions within the network, only the output. work is also an exemplar of the category whose main purpose is to create a pattern that stand for symbols. THE PROCEDURAL-PROCEDURAL

l

191

Thus, a connectionist netof procedural knowledge, of activation across units DISTINCTION

‘IWO different processing models, each reflecting procedural knowledge, have been presented. Each are based on a different set of underlying computational assumptions. The question that arises is the nature of the relation between these two models. One possibility is that the two are alternative theories in that they both embody hypotheses intended to account for the same set of phenomena. Cognitive processes can be explored by adopting either a symbol manipulation model (productions) or a network model (connectionist) and one would be shown to be more productive in the long road. A second possibility is that each processing model accounts for a different set of phenomena and both are needed, and in fact, you need to propose a hybrid model. For example, previous research has demonstrated that production symbol manipulation models have difficulty handling perceptual and motor tasks, which network models handle easily. Alternatively, network models have been shown to have difficulty handling sequential, planning, and problem-solving tasks that involve generative behavior, that symbol manipulation models handle well (see Mandler, 1998, for a discussion on this topic). The approach taken in this chapter is that a hybrid model of procedural knowledge is needed to understand cognitive processing. In the following section, I argue that the two knowledge systems make different predictions regarding the contexts in which learning takes place, the pattern of learning, and the ability to transfer knowledge, and that predictions from these two processing systems can be experimentally teased apart. DISTINGUISHING

BETWEEN -IWO TYPES OF BEHAMORS

There are two general ways of responding to a situation: figuring out what to do (generating knowledge) or retrieving the response from past knowledge. For example, given the problem “5 + 4 = ?,” a person could count their fingers and derive the answer or they could retrieve the knowledge fact that “5 + 4 = 9.” The premise is that the “figuring out” behavior is related to the “rule-based” procedural system, and “retrieval” is related to the “network” procedural system. Both processes can potentially supply the solution.

192

l

11.

Procedural

Knowledge

Distinction

The first feature that discriminates the two processing systems is the context in which learning occurs. An analysis of the types of practice that facilitates acquisition varies for the two types of processes. Research addressing the acquisition of retrieval shows that acquisition depends on the redundant relation between a stimulus representation and a response representation. When retrieval occurs, there is a correct answer associated with a given stimulus and, very importantly, participants receive frequent practice with the same set of stimuli. For example, the first time a person sees the problem “5 + 4 = -,” they probably have to use a counting procedure or guess the answer. However, if the same problem is repeatedly presented, the person quickly learns to retrieve the answer “9.” In this example, the problem representation stays consistent, the answer stays consistent, and the relation between the representation and the answer remains consistent. This extensive practice allows the association (or the network of associations) between the stimulus and the response representation to be strengthened, and this in turn, facilitates retrieval. Thus, connectionist networks tend to learn as a consequence of extensive rote practice; a network is presented a context, makes a response, and is given feedback regarding the accuracy of the response. If the response is correct, the network is changed by manipulating the weights of the associative links to make the correct response a stronger one. If the response is incorrect, the weights of the associative links that led to that response are weakened. Over a great number of trials, the network learns a set of constraints that allows it to give a certain response in a certain context. This, in essence, describes the process underlying the delta rule (Stone, 1986). Alternatively, research investigating the development of problem solving tends to investigate performance in computer programming, geometry, and text editing. In these domains, participants are not repeatedly presented with the same problem (stimuli) but are presented with a set of different problems requiring the use of the same procedure. For example, a person might first be presented with the problem “5+4= , n and then “3 + 2 = ,” and then “6 + 3 = ,” and so forth. Each time there is a different relation between the problem and the associated answer. However, the need for the procedure remains constant. Given the lack of consistency between the stimulus and response, the associative links between the two representations are not strengthened and consequently, the ability to retrieve an answer is not acquired. However, this type of practice allows people to apply and develop new and more efficient procedures. Thus, the contexts in which learning occurs within the two perspectives vary in their allowance for a redundant mapping between a stim-

Rabinowitz

l

193

ulus and response. In situations where a consistent relation is possible and practiced, retrieval acquisition occurs. In contexts in which a consistent mapping between a problem and response is not allowed, albeit practice on problems requiring similar routes to solution, rule development occurs. These two different types of practice opportunities also suggest that the pattern of learning should differ for the two systems. In the situation where there is a consistent relation between the stimulus situation and response (retrieval), it would be expected that initially, performance would be slow, given that people need to guess or figure out the answer. However, once the relation between the problem and answer is noticed, the individual would simply retrieve the answer and performance should speed up considerably. In situations where there is not the consistent relation between the problem and the answer, it would be expected that there would be a much more gradual learning curve as the rules are being developed and strengthened. Rabinowitz and Goldberg (1995) designed an experimental manipulation to evaluate these predicted patterns. In their study, they compared the development of rule-based and retrieval skill by having participants learn a new task under two types of training conditions. The task used was a variant of the alphabet arithmetic task developed by Logan ( 1988). Specifically, participants were required to solve problems of the following format: Letter1 + Number1 = Letter2. For example, given the problem “L + 5 = -,” the participant should respond “Q,” In the condition where a “rule-based” skill was to be developed, participants received training designed to give them practice counting up the alphabet to derive the answers. This was accomplished by giving participants extended practice with 72 different problems and little practice on any specific problem. The 72 problems were presented six times so that participants had to solve a total of 432 problems. It was presumed that this situation would provide the opportunity for participants to become more facile with the procedure of alphabet counting but would not allow them to become familiar enough with any specific problem so that they would be able to retrieve the answer. In the condition where “retrieval” skill was supposed to be developed, participants were presented with a small set of 12 problems on which they received extended practice; each problem was presented 36 times so that each participant had to solve 432 problems. In this training condition, it was assumed that after a couple of repetitions with each problem, participants would be able to retrieve the answers and not need a rule-based procedure to derive them, Thus, participants in this condition should receive a lot of practice with retrieval and little practice with using the counting procedure.

194

.

11.

Procedural

Knowledge

Distinction

It was predicted that the pattern of knowledge acquisition would vary significantly for these two training conditions. At the start of training, both groups were expected to be equivalent in terms of their reaction times (RTs). Although it was predicted that both group’s RTs would decrease, participants given the retrieval training were expected to show a much faster learning curve and be significantly faster at the end of training than those in the rule-based condition. This pattern was observed. In addition, Rabinowitz and Goldberg (1995) predicted that if participants were deriving answers by counting up the alphabet, it would be expected that the larger the number in the problem, the longer the RT should be. This phenomenon has been labeled the problem size effect (Groen & Parkman, 1972). However, if participants were retrieving the answers to the problems and not using the counting procedure often, the problem size effect would be attenuated. Therefore, it was predicted that the problem size effect at the beginning of training should be evident for both groups. However, at the end of training, the retrieval groups should show a much less pronounced problem size effect than the rule-based group. Once again, this pattern was observed. Thus, the two training groups varied significantly in terms of their performance characteristics during this learning task. After the training task, Rabinowitz and Goldberg (1995) gave the participants one of two different types of transfer tasks. For the first task, participants were given a new set of alphabet addition problems to solve. It was predicted that participants in the rule-based group, who obtained substantial practice using the counting procedure, would show that they would maintain their skill with the new problems that required the use of the procedure. However, it was hypothesized that subjects in the retrieval group, because they received little practice using the procedure, would show little positive transfer and be significantly slower than participants in the rule-based group. As expected, participants provided with the rule-based training were significantly faster than participants in the retrieval condition. For the second transfer task, Rabinowitz and Goldberg (1995) asked participants to switch from alphabet addition to alphabet subtraction. That is, after training on “L + 5 = ?,” participants would then switch to “Q-5 = ?” Retrieval participants should show better transfer than rule-based participants when tested for subtraction on the same problems that they had studied. The logic was that retrieval participants, because of the knowledge they developed, would be able to reason by analogy and say if L + 5 = Q, then Q- 5 must equal L. However, participants in the rule-based condition during training received repeated practice counting up the alphabet. Recall that these types of rule-based procedures are assumed to be directional and nonreversible; that is,

Rabinowitz

l

195

learning how to count up the alphabet should provide little or no positive benefit for counting down the alphabet. Rule-based knowledge differs from retrieval knowledge in that it is committed to a direction of use. In fact, the results from this study showed that participants receiving the retrieval training were significantly faster and more accurate than participants receiving the rule-based training. Thus, the knowledge that was acquired during training varied in terms of the conditions in which it could be subsequently used. THE INTERACTION BETWEEN THE TWO TYPES OF PROCEDURAL KNOWLEDGE What is the relation between these two types of procedural processes? In one sense, as the aforementioned study implies, the two processes can be seen as being competitive. Given the initial stimulus presentation (the addition problem), a race begins with both processing systems reacting to the problem in parallel. If the network system is able to retrieve the answer (settle down quickly) prior to the execution of the procedure (counting), the person retrieves the answer; if not, the person uses a procedure to generate the answer. Siegler (1986) proposed such a model to account strategy choice variations. However, the two procedural processes also can be seen as supporting each other with regard to the acquisition of new skills. For example, in Anderson’s model of skill acquisition, he posited three stages of skill: the declarative, procedural, and compilation stage (Anderson, 1993). Thus, when people are learning a new domain, they have some basic retrievable domain knowledge but don’t really know how to use it. Therefore, when given a problem to solve they rely on that retrievable knowledge and use weak heuristics to figure out what to do. The use of these heuristics imposes a high demand on working memory and does not guarantee success in solving the problem. As people become more familiar with a domain, they develop strategies specifically adapted for a task. The tradeoff is that these strategies impose less demand on working memory and are more likely to lead to successful problem solving, but they are more limited in their generalizability. Finally, the use of the strategy becomes more efficient and requires less effort to apply. Therefore, development occurs in terms of the “rule-based” processes, but the retrieval knowledge acts as the support structure which underlies this development. Rabinowitz and Kee (1994) suggested that the accessibility of this retrievable knowledge also influences the execution of the “rule-based” processes. Alternatively, rule-based knowledge can support the acquisition of retrievable knowledge. As Rabinowitz and Goldberg’s (1995) data illus-

196

l

11.

Procedural

Knowledge

Distinction

trated, retrievable knowledge can develop as a consequence of the initial use of a rule-based procedure. Initially, when participants were given a problem like “C + 3 = ?,” they needed to count up the alphabet to derive the answer. However, after seeing the same problem a number of times, they knew what the answer was and retrieved it. Thus, the initial use of a rule-based procedure leads to the acquisition of an associative network that obviates the need for the rule-based procedure.

DISCUSSION In this chapter I proposed the need for a hybrid model of knowledge representation. One representation is based on the features and constraints of a symbol manipulation system. The other is based on the features and constraints of a subsymbolic network system. There is nothing new or unique about this proposal in that many people, as I presented in this chapter, have made similar proposals-including 1998), specifying the two types of representational Mandler (1983, systems outlined in this chapter. From my perspective, a problem arises when people try to characterize these two types of knowledge systems in terms of procedural and declarative knowledge. First, as I presented within the chapter, there is not general agreement within the field as to which one is declarative knowledge and which one is procedural knowledge. Examples are found within the literature where a symbol manipulation system is characterized as a referent for declarative knowledge and elsewhere, as a referent for procedural knowledge. The same is true for the subsymbolic network representations. This situation reflects an unnecessary state of confusion. Obviously, I think the correct characterization of these two representational systems is that of a proceduralprocedural distinction. A more serious problem, at least from my perspective, is that the procedural-declarative distinction affords (at least to me) an active versus passive processing distinction (Newell, 1973). Newell hypothesized a distinction between structure and process. Knowledge structures (the physical symbols) are static, permanent, and object-like, whereas processes are dynamic, transient, and transformation-like. The application of processes requires certain symbols to be available. This characterization of the process or structure distinction still often defines the relative roles of process and structure. Procedural knowledge is assumed to represent the active process component and declarative knowledge, the passive structure. Even in Mandler’s (1998) chapter, much effort went into defining the processes underlying connectionist (procedural for her) knowledge. There was not a similar emphasis in

Rabinowitz

l

197

the description of declarative knowledge; it was defined more in terms of a description as opposed to how it works. The distinction I am trying to argue for is an active-active distinction with two procedural systems interacting to accomplish cognition. Finally, at the beginning of this chapter I suggested that Mandler (1983, 1998) argued that development can be thought of as progressing from a purely procedural system to one that integrates procedural and declarative knowledge. I think that an interesting topic to be investigated in the future is how these two procedural systems are inteprocess. grated and how this changes within the developmental

Anderson, J. R. (1993). Rules ofmind. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Anderson, J. R., & Bower, G. H. (1973). Human associative memory. Washington, DC: Winston. Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought. Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Collins, A. M., & Quillian, M. R. (1969). Retrieval times from semantic memory. Journal of Verbal Learning and Verbal Behavior, 8, 240-247. Groen, G. J., & Parkman, J. M. (19 72). A chronometric analysis of simple addition. Psychological Review, 79, 329-343. Kintsch, W. (1992). A cognitive architecture for comprehension. In H. L. Pick, Jr., I? van des Broek, & D. C. Knill (Eds.), Cognition: Conceptual and methodological issues (pp. 143-164). Washington, DC: American Psychological Association. Klahr, D., Langley, I?, & Neches, R. (1987). Production system models of learning and development. Cambridge, MA: MIT Press. Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review, 95, 492-527. Mandler, J. M. (1983). Representation. In J. H. Flavell & E. M. Markman (Eds.), Handbook of child psychology: Vol. 3. Cognitive Development (4th cd., pp. 424-494). New York: Academic. Mandler, J. M. (1998). Representation. In D. Kuhn & R. Siegler (Eds.), Cognition, perception, and language (Vol. 2 of W. Damon; Series Ed.), Handbook of child psychology (pp. 255-308). New York: Wiley. Newell, A. (1973). Production systems: Models of control structures. In W. G. Chase (Ed.), L%suaZ information processing (pp. 283-310). New York: Academic. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice Hall. Norman, D. A., & Rumelhart, D. E. (1975). Explorations in cognition. San Francisco: Freeman. Rabinowitz, M., & Goldberg, N. (1995). Evaluating the structure-process hypothesis. In F. E. Weinert & W. Schneider (Eds.), Memory peflormance and competencies: Issues in growth and development (pp. 225-242). Hillsdale, NJ: Lawrence Erlbaum Associates, inc. Rabinowitz, M., & Kee, D. (1994). A framework for understanding individual differences in memory: Knowledge-strategy interactions. In I? A. Vernon

198

l

11.

Procedural

Knowledge

Distinction

(Ed.), Neuropsychologyof individual difirences (pp. 135-148). New York: Academic . Rabinowitz, M., Lesgold, A. M., & Berardi, B. (1988). Modeling task performance: Rule-based and connectionist alternatives. ZnternationaZ Journal of Educational Research, 12, 35-48. Rabinowitz, M., & Mandler, J. M., (1983). Organization and information retrieval. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9,430-439. Rumelhart, D. E., & McClelland, J. L. (1986). Parallel distributed processing: Explorations I the microstructures of cognition. Vol. I: Foundations. Cambridge, MA: MIT Press. Schacter, D. L. (1987). Implicit memory: History and current status. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 501-518. Siegler, R. S. (1986). Unities across domains in children’s strategy choices. In M. Perlmutter (Ed.), Perspectives on intellectual development (pp. l-48). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Squire, L. R. (1987). Memory and brain. New York: Oxford University Press. Stone, G. 0. (1986). An analysis of the delta rule and the learning of statistical associations. In D. E. Rumelhart & J. L. McClelland (Eds.), Parallel distributed processing (Vol. 1, pp. 444459). Cambridge, MA: MIT Press. firing, A. M. (1963). Computing machinery and intelligence. In E. A. Feigenbaum & J. Feldman (Eds.), Computers and thought (pp. 11-35). New York: McGraw-Hill. von Neumann, J. (1951). The general and logical theory of automata. In L. A. Jeffress (Ed.), Cerebral mechanisms in behavior (pp. l-3 1). New York: Wiley.

12 -#Spatial Language: Perceptual Constraints and Linguistic Variation Terry Regier

University of Chicago

Laura Carlson

University of Notre Dame

CHAPTER DEDICATION

W

e offer this chapter as a tribute to Jean Mandler. Her thinking has inspired each of us separately, and has also animated our collaboration. This chapter is the product of that collaboration, and it springs ultimately from one of her arguments, concerning the relation of language and thought. Following Jean, we argue for a nonlinguistic basis for some of the semantic structures of language. We hope, Jean, that you will find in this chapter a reflection of some of your own ideas, and a measure of our intellectual indebtedness to you. INTRODUCTION Different languages package the world differently. For example, a number of languages use the same word for the two colors blue and green (Berlin & Kay, 1969)-but these colors receive two separate labels in 199

200

l

12.

Spatial Language

English. Thus, in this instance, English seems to carve nature into smaller pieces than these other languages. However, English is by no means privileged-there are other languages that make semantic distinctions that English does not. For example, the Mexican Indian language Mixtec distinguishes between spatial location above a horizontally extended object, and location above a vertically extended one (Brugman, 1983)-situations that would both be labeled “above” in English, as shown in Fig. 12.1. Languages also differ in their categorizations of personality types (Hoffman, Lau, & Johnson, 1986), plurality and animacy (Lucy, 1992), and other aspects of experience (Gumperz & Levinson, 1996). What is the significance of such cross-linguistic variation? What does it mean, for our conception of ourselves, that languages differ in the categories they pull out of the world? Primarily, it means that there is no single universal human parsing of experience. Rather, linguistic categories appear to be constructed in part by the semantic conventions of particular languages. Thus, language is accorded a certain freedom of action in its organization of the world around us. However, it is also possible that there are broad universal tendencies underlying this variation. Specifically, there may be culturally-invariant aspects of perception and cognition that leave their imprint on language (Clark, 1973; Kay & McDaniel, 1978; Mandler, 1992, 1996). Such a perceptual core would constrain the semantic partitions that an individual language may make. At the same time, within these con-

0

0

Mixtec: Sni English: above

Mixtec: siki English: above FIG. 12.1

Mixtec and English

spatial

terms.

Regier and Carlson

l

201

straints, the specifics of a given language may also contribute to the shaping of meaning (Regier, 1996). The linguistic expression of spatial relations is a particularly useful domain in which to explore these issues of cognitive constraint in language. There are three converging reasons for such a focus on space in particular. The first is that languages differ in their structurings of space (Casad & Langacker, 1985; Levinson, 1996; Talmy, 1983). An example is the Mixtec/English difference, shown in Fig. 12.1. Thus, we may be assured that the cross-linguistic variation that motivates this inquiry is represented in the spatial domain. The second reason is that spatial cognition is a foundational element of human understanding. Certainly, prelinguistic children demonstrate some very early spatial competence (Bomba, 1984; Needham & Baillargeon, 1993; Spelke, Breinlinger, Macomber, & Jacobson, 1992), and this spatial understanding later undergirds some aspects of nonspatial cognition as well (Lakoff, 1987; Mandler, 1992, 1996). Third, and finally, our current understanding of nonlinguistic spatial cognition is substantial-thus, the search for cognitive constraints on spatial language may rely on this groundwork. Jean Mandler has pursued such a search, for an underlying common cognitive bedrock beneath the spatial systems of language. In particular, she has argued that some of the “categorical or packaging characteristics often ascribed to language” may actually be reflections of preverbal, nonlinguistic structures (Mandler, 1992, 1996, p. 365). It is this central notion, the grounding of linguistic meaning in structures that are not themselves linguistically determined, that has inspired us in the work we present here. For in this chapter, we also argue for such a bedrock. We present a computational model of spatial language that is grounded in aspects of spatial perception that may reasonably be assumed to be universal. These inflexible structures constrain the space of possible spatial meanings. However, other structures are left free to vary, and may be determined either by a specific language, or by the idiosyncrasies of an individual speaker of a language. We have tested this model against fine-grained empirical data (Regier & Carlson, 2001); we present an overview of these empirical tests here. Finally, and most significantly for the purposes of this chapter, we show that the model accounts for empirically observed linguistic variation, both within a single language, and across languages. GROUNDING

SPATIAL LANGUAGE

IN PERCEPTION

One aspect of spatial language that seems likely to be influenced by universal cognitive or perceptual constraints is the set of spatial terms that

202

.

12.

Spatial Language

concern external reference axes-for instance “above,” “below,” “beside,” and similar terms in other languages. Even newborn infants have some ability to detect such spatial relations. Ante11 and Caron (1985) found that newborns could discriminate between cards that displayed a cross above a square and others that displayed a square above a cross. Yet as we saw in Fig. 12.1, languages differ in their categorization of even such very simple spatial relations. Thus, these spatial terms seem to reflect both prelinguistic predispositions, and to some extent, language-specific convention. Simple English projective spatial terms such as “above” display a graded category structure, as illustrated in Fig. 12.2. Here, all three circles may be reasonably said to be “above” the rectangular landmark object to some degree. But of the three, A appears to be the best example of “above,” B a somewhat weaker example, and C the weakest example. There are cross-linguistic differences even in such simple observations: analogous ratings in German appear to drop off more quickly than those in English, as we shall see. What underlies these judgments? Are there identifiable perceptual or cognitive processes that govern this element of spatial language? And might these processes then serve as constraints on linguistic variation? We argue that there are such processes. We begin with two observations concerning nonlinguistic, spatial perception and representation: 1. The first observation is that the apprehension of spatial relations involves attention. Logan (1994) showed that visual search for a target in a field of distracters is slow when the only difference between targets and distracters is the spatial relation (above, below, left, or right) between their elements. This finding suggests that spatial relations do not preattentively “pop out” of the visual field, as some visual

FIG. 12.2

Graded category

A

B

C

0

0

0

structure

for “above.”

Regier and Carlson

l

features do (TYeisman & Gormican, 1988). Rather, the perception spatial relations appears to require the focus of attention.

203 of

2. The second observation is that in some neural subsystems, overall direction is represented as the vector sum of a set of constituent directions. For example, Georgopoulos, Schwartz, and Kettner (1986) examined the area of monkey motor cortex representing the animal’s arm, to determine the cortical representation of intended arm movement direction. They found a population of broadly directionally tuned cells, such that each cell responded maximally when the monkey’s arm movement was in the cell’s preferred direction. The direction of arm motion was accurately predicted by a vector sum over the population of cells as a whole: Za, c i E population Here, ai is a measure of the activity of cell i, and c is the preferred direction of cell i. This vector sum formulation of direction also accounts well for some aspects of motion perception (Wilson & Kim, 1994), suggesting that the representation is a widely used one. We bring together these two observations, concerning attention and vector sum representation, in the Attentional Vector Sum (AVS) model. The model predicts spatial term acceptability judgments, such as those discussed earlier, based on these nonlinguistic considerations. The operation of the model, and its grounding in attention and vector sum coding, are illustrated in Fig. 12.3. In (a), we see an attentional beam focused on the reference object, or landmark. The beam is centered on that part of the landmark top that is vertically aligned with the located object, or nearest to being so aligned. Attentional strength is maximal at the focus of the beam, and then drops off exponentially with distance (Downing & Pinker, 1985; LaBerge & Brown, 1989)-indicated by the concentric circles in the figure. Thus, some parts of the landmark receive considerable amounts of attention, whereas others receive much less. The width of the attentional beam, and therefore the amount of the landmark that is well-attended, varies with the distance between the two objects: the farther apart the two objects are, the wider the beam, and the greater the extent of the landmark that is well-attended. It is natural to broaden the attentional beam with distance in this fashion, as the perceiver must include both objects in the attentional beam. This, then, is the attentional component of the AVS model.

204

l

12.

Spatial Language

00

FIG. 12.3

The Attentional

Vector

0 a

Sum (AVS) model.

The vectorial component is shown in the next frame, (b). We define a vector projecting from each point of the landmark object toward the located object-yielding a population of such vectors. The attentional and vectorial components of the model are joined in an attentionally weighted vector sum. Specifically, we use the same vector sum formula as before, namely xi a, < , but now, ai is the amount of attention paid to landmark point i, and ; is the vector rooted at point i. Thus, each vector is multiplied by the attentional strength at its root. This yields a population of attentionally weighted vectors, as shown in (c). These are then summed, and the result indicates the overall spatial orientation between the two objects, as shown in (d). We then measure the alignment of this vector sum with a reference axis, such as upright vertical for “above.” This is shown in (e). Perfect alignment corresponds to a perfect example of the spatial term in question-and deviations from perfect alignment correspond linearly to poorer examples. There is also a cutoff, perpendicular to the reference axis: for example, in the case of “above,” points that are strictly lower than the highest point on the landmark are considered poor examples of “above,” regardless of the vector sum. Regier and Carlson (2001) provide a more detailed formal presentation of the AVS model.

Regier and Carlson

l

205

This then is the invariant “perceptual core” of the model-which captures presumably universal aspects of perception. However, embedded in this core are several free parameters, that accommodate some linguistic variation, as long as that variation does not violate the central concept of an attentionally weighted vector sum. We shall focus on two such parameters. These are width, the default width of the attentional beam, and slope, the rate at which spatial term acceptability drops off with deviation of the vector sum from alignment with the reference axis. These two parameters will be relevant later, when we test the model against empirically observed linguistic variation. TESTING THE AVS MODEL There is an instructive prediction made by the AVS model, which highlights the interaction of attention and vector summation. Imagine a marble held a half-inch above a large book lying on a table. Now imagine moving the marble around over the surface of the book, but maintaining the half-inch height-and noting how good an example of “above” each positioning produces. The AVS model predicts that at low elevations of this sort, “above” ratings will not vary much as we move the marble around over the book. At higher elevations, however, ratings will vary more, peaking directly above the center of the book. Why should this be? Because at low heights, the attentional beam will be quite narrow, as beam width varies with interobject distance. This means that attention will be restricted to the region of the book directly beneath the marble, and nearby. Critically, the edges of the book will not receive much attention-meaning that for all intents and purposes, the marble is located above a limitless plane. Thus, the vector sum will be nearly upright for a wide range of positionings. At higher elevations, however, the attentional beam will span a larger area, such that the edges of the book receive more attention-and this will affect the vector sum. Under these circumstances, a strong “above” rating is predicted only if the marble is well-centered over the book. This prediction is not made by an intuitively reasonable-seeming competing model, the Bounding Box (BB) model. The BB model holds that a point is “above” a landmark to the extent that it falls within the rectangular region bounded by the vertical line on the right edge of the landmark object, the vertical line on the left edge, and the horizontal line at the top of the object. This region is highlighted in Fig. 12.4. The BB model predicts that the only quantities that will affect “above” ratings are the horizontal and vertical distances of the located point from the three bounding lines. When we apply the example of the marble being moved around over the book at two different heights to this model, we

206

l

12.

FIG. 12.4

Spatial Language

The Bounding

Box (BB) model.

find that the two elevations are distinguished only by the distance from the horizontal line. And that distance will be the same across points at a given elevation. Thus, there is no prediction of greater sensitivity to centeredness at higher elevations, in contrast with the AVS model. Regier and Carlson (2001) tested this contrasting prediction. We showed participants scenes in which a small object was located relative to a large rectangular landmark object. The array of placements of the located object, relative to the landmark, is shown in Fig. 12.5. At each point in time, the located object appeared in one of the indicated locations, and the participant indicated how good an example of “above” the resulting spatial relation was, on a IO-point (O-9) scale. We obtained a mean rating for each position by averaging the responses of all participants to that position. We then determined how well the AVS model, and a set of competing models, including the BB model, fit these empirically obtained data.

Regier and Carlson

l

207

The highlighted positions, shown in black in Fig. 12.5, are the critical ones that allow us to test the prediction. The AVS model predicts that ratings from the higher row will be more sensitive to centeredness than those from the lower row. The BB model and other competitors do not make this prediction. Figure 12.6a displays the experimentally obtained “above” ratings for these critical points- with the top and bottom rows shown separately. As predicted by the AVS model, the top row is more sensitive to centeredness than the bottom row, which is comparatively flat. This difference between rows with respect to centeredness is statistically significant (Regier & Carlson, 2001; Experiment 7). Figure 12.6b shows the fit of the AVS model to these data-it is qualitatively similar, and quantitatively quite close (r2 = 0.888 over these critical points; r2 = 0.985 over all positions shown in Fig. 12.5). No competing model, including the BB model, was able to produce the qualitative effect shown here, and none had a tighter quantitative fit. In a series of experiments, we similarly tested other predictions of AVS and the competing models, relative to a variety of landmark objects. These landmark objects are shown in Fig. 12.7, with sample placements of the located object. The AVS model provided the tightest overall quantitative fit to our data. The AVS model’s fit to all our data, pooled over seven experiments, is shown in Fig. 12.8. The fit is good (r2 = 0.970, over 337 data points), and better than those of competing models (e.g., BB r2 = 0.953). Critically, only the AVS model also passed all of our qualitative tests.

oa

a

0

0

FIG. 12.5

Stimuli

me

for eliciting

a

0

0 spatial

0 term judgments.

a

0

0

208

12.

l

Spatial Language

E8 7.6

7

0.5

1

FIG. 12.6

Linguistic

,

2

Empirical

3

4

and AVS-predicted

5

0

9

ratings.

Variation

Attention and vector sum coding form an inflexible perceptual core in the AVS model. Because there is independent nonlinguistic motivation for these processes, we advance them as possible universal constraints

0

27

1

0

0

0

h FIG. 12.7

Sample spatial stimuli used in testing the AVS model.

Pmdk!ed vs. txqJelimentnny obtahd

FIG. 12.8 (2001).

above raung

Fit of the AVS model to all data collected by Regier and Carlson 209

210

.

12.

Spatial Language

on spatial term flexibility, within these parameters use. We examine across languages. Within-Language

use. The free parameters of the model provide some these constraints. As we shall see, different settings of account for some observed variation in spatial term such variation, first within a single language, then Variation

One potential locus for linguistic variation is interspeaker variability within a given language. It is conceivable that not all English speakers mean the same thing when they use a given word; even among native speakers there may be only rough agreement over word meaning. Such interindividual differences must be subtle for effective communication to take place. However, the possibility does remain that there are detectable differences in the meanings that different speakers attach to the same word. We have found evidence of such individual differences in spatial term use in English (see also, Carlson-Radvansky & Logan, 1997). Specifically, we have found that not all participants in our experiments adhere to the same meaning of the word “above.” Some individuals (22%-589/o, depending on the experiment) considered a scene to be a perfect example of “above” provided that the located object was directly above some part of the landmark. For example, as we shall see, these individuals gave the same rating to the 3 points shown in Fig. 12.9-their responses were “flat” across these points. However, other individuals did not provide such flat responses. In our earlier tests of the AVS model, we averaged together the responses of all individuals, to obtain an overall rating for a given position. We now examine separately the responses of “flat” and “not-flat” individuals. The AVS model can account for both response patterns through different settings of the attentional beam width parameter. As we have seen, beam width is normally modulated by interobject distance. However, it is also controlled by this width parameter-and when the parameter assumes extremely small values, near zero, this keeps the beam very narrow, regardless of interobject distance. Under these circumstances, the attentional beam takes in only those landmark points that are directly beneath the located object, and the vector sum is therefore based only on those points. Therefore, as long as the located object is directly above some landmark point, the attentional vector sum will be essentially perfectly upright, yielding a maximal response. This matches the responses of “flat” individuals. On the other hand, if the width parameter is larger, beam width varies with distance in the manner discussed earlier, such that attention is often deployed broadly over

Regier and Carlson

A

FIG. 12.9 viduals.

Spatial

stimuli

B

that elicit different

211

l

C

responses from different

indi-

the landmark. In consequence, the vector sum integrates over more of the object, yielding “non-flat” responses. The central feature distinguishing these two sets of individuals then is the extent of the attentional beam over the body of the landmark. This seems plausible, but is it an adequate account of the difference? To test this, within each experiment we averaged together the responses of the “flat” individuals, and separately averaged together the responses of the “not-flat” individuals. This yielded two separate sets of data for each experiment. We then fit the AVS model-and its competitors-separately to each dataset in each experiment. AVS was the only model that accounted well for both classes of response. We illustrate its operation by considering the experiment shown in Fig. 12.9 (Regier & Carlson, 2001; Experiment 4). The empirically obtained “above” ratings may be seen in Fig. 12.10a. Sixteen out of 30 (53%) participants gave the same response to all three positions-these are the responses labeled “flat” in the figure. However, the remaining 14 out of 30 (47%) participants gave responses that were not flat across the three positions. The same flat/not-flat distinction was found in several other experiments. The AVS model accounts for this variation, in the manner suggested. We fit the model separately to the two sets of data from this experiment; the results are shown in Fig. 12.1 Ob. These results closely match

212

’

12.

Spatial Language

the empirical findings: 9 = 1 .O for the flat responses, 9 = 0.98 for the not-flat responses. Table 12.1 shows the AVS fit to the empirical data for all experiments in which we found such individual differences. The primary difference between the “flat” and “not-flat” fits was the one suggested: the default width of the attentional beam was very small for Expdmentally 9

8 *

8.5

obtained ‘above’ tatinfg I

. . ..-........-.....-..-...

l . . . . . . _ . . ..-..

1

Flat -.a-Non-mt -.Q.-

_. . . . . . . . . . . . . *

8 ON.-.

7.5 i

--._._.

--._.__

--‘Q.,

7

.\-\ .\-.

\

‘k

‘. .

\

6.5

.

‘.

\

b

6

5.5

A

0 POSiUOll AVSqedkted

C

‘above’ ratinQs

8.5

6.5

6

FIG. 12.10 ferences.

Empirical

and AVS-predicted

ratings,

illustrating

individual

dif-

Y w I

Experiment (n = 56) Experiment (n = 4) Experiment (n = 65) Experiment (n = 25)

Experiment

7

5

4

3

AVS Fit to Empirical

Flat Not-flat Flat Not-flat Flat Not-flat Flat Not-flat

Response

type

Data for All Experiments

0.000 0.897 0.001 0.804 0.162 0.974 0.035 1.317

Width

3 = fi = r2 = r2 = ? = r2 = r” = 9 =

0.986, 0.986, 0.001, 0.981, 0.975, 0.976, 0.987, 0.978,

TABLE 12.1 in Regier and Carlson (in press) in Which

slope slope slope slope slope slope slope slope

= = = = = = = =

Individual

= = = = = = = =

0.150 -0.086 -0.007 0.138 -0.364 -0.509 -0.163 -0.131

Were Found.

0.974, intercept 1.010, intercept 1 .OOl, intercept 0.977, intercept 1.037, intercept 1.069, intercept 1.016, intercept 1.033, intercept

AVS fit

Differences

214

l

12.

Spatial Language

the flat datasets, but was larger for the not-flat ones. The other AVS parameters had roughly equal values across the flat and not-flat datasets for each experiment. Although some of the competing models produced flat responses, and others produced not-flat responses, only the AVS model accounted for both response patterns. Thus, perceptual core notwithstanding, the AVS model is flexible enough to accommodate some empirically observed differences in spatial term use.’ The model suggests that these linguistic differences may stem from underlying nonlinguistic differences in the nature of attentional deployment.

Cross-Language Variation The general picture being promoted is that of an unchanging perceptual core, with flexible parameters that permit variation within the constraints of that core. We have seen that this idea, instantiated in the AVS model, accounts for within-language differences in spatial term use. Can the same idea also account for cross-linguistic differences? It should be able to, if our account is correct-for attention and vector sum coding are assumed here to be universally available processes. Thus, we would expect their influence to appear in spatial terms from other languages as well. Some cross-linguistic variation then may be explicable in terms of different parameter settings within this universal core, as was the case with individual differences within a language. We have initial evidence supporting this idea. Gapp (1995) collected spatial term acceptability data in German. The spatial stimuli he used were similar to the ones we discussed here: a landmark object with a smaller object located relative to it. A sample stimulus is shown in Fig. 12.11. As in our experiments, the position of the located object was systematically varied around the landmark, and acceptability ratings for spatial terms were collected at each position. We focus in particular on ratings for the German word which Gapp translates as “above” in English. This German word is not specified explicitly by Gapp (1995); however, in a related report he translated “above” as “iiber” (Gapp, 1994). Thus, we assume that Gapp’s (1995) ratings are for “iiber.” We compare these with the ratings obtained for English “above” by Logan and Sadler (1996). We use Logan and Sadler’s English data, rather than our own, because their spatial displays closely matched those of Gapp (1995), whereas ours did not. Importantly, Logan and Sadler’s English ‘See http://www.psych.uchicago.edu/-regier/avs/diffs-results.html discussion of our treatment of individual differences.

for a fuller

Regier and Carlson

l

215

0

cl

FIG. 12.11 Sample stimulus term ratings.

used by Gapp (1995) to collect German

spatial

data do appear to be entirely compatible with our own, despite the differing spatial displays. We have found that when the AVS model is fit to Logan and Sadler’s data, it generalizes very well to our English data (Regier & Carlson, 2001). Thus, data from different laboratories investigating the same language agree. Gapp’s (1995) German “fiber fl ratings drop off linearly with angular deviation from upright vertical. This is compatible with the AVS model-recall Fig. 12.3e. Logan and Sadler’s (1996) “above” ratings also drop off linearly with angular deviation. However, the rate of dropoff appears to be different in the two languages, as shown in Fig. 12.12. The figure shows the mean acceptability rating found at a set of angular deviations, for each word. To test the apparent difference, we regressed all ratings within the range shown here on angular deviation and language (0 for English “above”; 1 for German “iiber “). The regression revealed a significant influence of angular deviation (beta weight = -0.6767, p < O.OOOl), and a significant interaction between angular deviation and language (beta weight = -0.6029, p < 0.0001). There was no effect of language, p = 0.400; overall ? = 0.945 3. The critical point for our purposes is that the effect of angular deviation is clearly different in the two languages. One possible objection to this finding is that the comparison between “above” and “iiber U may be inappropriate. Specifically, “iiber N may be more appropriately compared with its English cognate “over,” rather than “above.” It is conceivable then that the difference found here is

216

12.

l

Spatial Language

Eng#sh and Qefman ratings

0

10

20

30

50

60

70

80

AngdrL4lbn

FIG. 12.12 English “above” and German gular deviation from upright vertical.

“iiber”

ratings,

as a function

of an-

fundamentally a difference between “above” and “over,” rather than between German and English. If this is the case, we should find the same difference when we compare “above” and “over,” within English. We were able to test for this possibility because Logan and Sadler (1996) had collected ratings for “over mas well as “above.” We compared their ratings for these two English words. This time the regression revealed no significant interaction between angular deviation and language, p > .5. Thus, the difference did not appear across these two English words. In contrast, when we compared English “over” with German “iiber,” we did find a statistically significant interaction, p < 0.01, as we had before in comparing “above” and “iiber.” This suggests that there is a genuine cross-linguistic difference.2 2There is, however, another possibility, which cannot yet be discounted. Gapp (1995) collected his German ratings on a continuous 0 to 1 scale, whereas Logan and Sadler (1996) collected English ratings using a discrete 1 to 9 scale, which we then transformed to the 0 to 1 range for comparison purposes. This difference in scale might account for the different slopes. However, it is not clear why it would, while preserving the linearity of the dropoff with angle.

Regier and Carlson

l

217

This difference is easily accommodated by the AVS model, as shown in Fig. 12.13. When fit separately to the English and German data, the model predictions reflect the difference in the empirical findings. The separate fits to the English and German data yielded different values for the model’s slope parameter, which controls how quickly ratings fall off with angular deviation from alignment with the reference axis. This difference is shown in Table 12.2. The more strongly negative value for German indicates a steeper dropoff with angle. Other parameters were roughly equal in the two cases. Thus, as in the case of within-language variation, the AVS model can adapt itself to these empirically observed cross-linguistic variations in spatial term use, despite its unchanging perceptual core. In principle, the model should also be able to accurately predict ratings for the Mixtec spatial terms of Fig. 12.1, with which we opened this chapter. Recall that Mixtec makes a distinction between location above a horizontally elongated object, and location above a vertically elongated object. Thus, its use of reference axes differs from English: the use of an upright vertical reference axis in Mixtec is contingent on in the dethe direction of elongation of the landmark. This difference ployment of reference axes is external to the AVS model, which assumes

En#bh and Gemmn ratinge, AVS predktkm

FIG. 12.13

AVS-predicted

English

“above”

and German

“fiber”

ratings

218

l

12.

Spatial Language

TABLE 12.2 AVS Fit, and Values of Slope Parameter, on English and German

Data

AVS fit

language

Slope

English “above” (n = 48) German “iiber M (n = 4)

-.006

r2 = 0.963,

slope = 1.030, intercept

= -0.036

-.Ol

r2 = 0.995,

slope = 0.993,

= 0.004

intercept

that a reference axis has already been established (Carlson, Regier, & Covey, in press). However, the AVS model does predict that once a reference axis has been established, for example, upright vertical relative to a horizontally elongated object for Mixtec “s&i,” the means in which spatial relations are judged and articulated relative to that reference axis will be the same in Mixtec as in any other language. This will reflect the processes of attention and vector sum coding that we have seen in this chapter. We do not have fine-grained spatial term ratings from Mixtec, so we cannot currently test this hypothesis. But the means by which it would be tested are clear-as is the case with the corresponding claims for other languages. We shall be pursuing further cross-linguistic tests of this sort in the near future. DISCUSSION There are two general approaches to questions of linguistic universals (Comrie, 1981), approaches that complement each other well. One approach is to examine many languages broadly, so as to empirically determine which linguistic features are widely shared across languages (Greenberg, 1978). The other approach examines phenomena within a single language in depth, so as to extract a set of underlying principles that govern their behavior (Chomsky, 1986). To the extent that these principles are truly general linguistic rules, they may also govern the structure of other languages -a hypothesis that may then be empirically investigated. There is a potential danger in this second approach. It is possible that the underlying principles divined from the examination of a single language will be language-specific rather than language-general. However, the approach can be considerably strengthened if there are clearly discernible reasons why these particular principles should be expected

Regier and Carlson

l

219

to recur across languages. If the principles are not merely regularities in the linguistic data under examination, but are also independently motivated in their own right, they become significantly more promising as candidates for universality. This is the approach we have taken in this chapter. We began by seeking a characterization of the use of spatial terms in English. Critically, however, we did not simply extract regularities from our English data, hoping to chance upon regularities that are language-general rather than English-specific. Rather, we sought to ground our English data in independently motivated aspects of perception-attention and vector sum coding-aspects that we may reasonably assume are universally available. We argued that these processes form a perceptual core that may constrain linguistic variation. Given this, the goal of this chapter has been to demonstrate that within the constraints of such a core, there is room to accommodate some linguistic variation of the sort observed empirically-both variation within a language, and variation across languages. There are a number of questions that are implicitly posed by the account we have presented here. An obvious question is whether our account will withstand further empirical testing, against data from a broader range of languages. This is a question that would be implicitly posed by any claim of universality, regardless of its nature. However, there is also a rather different sort of question suggested by this work, one that arises specifically from our claim to have grounded aspects of spatial language in perception. If our account of linguistic variation is correct, we may find nonlinguistic correlates of some of the linguistic differences. For example, because within-language individual differences in linguistic response were explained in terms of the default width of an attentional beam, it would strengthen our argument if we also found nonlinguistic evidence of such variation across individuals-particularly if a given individual’s linguistic and nonlinguistic responses agreed as to the width of attentional deployment. Analogously, it is possible that cross-language linguistic differences mirror underlying nonlinguistic differences-potentially caused by the language itself (Whorf, 1956)-within the constraints of an unchanging perceptual core. We are eager to pursue these questions, in an attempt to further explore the nonlinguistic bases of spatial language. ACKNOWLEDGMENTS This work was supported by NIH Grant DC03384 NSF Grant 9727638 to Laura Carlson.

to Terry Regier, and

220

.

12.

Spatial Language

REFERENCES Antell, S. E., & Caron, A. J. (1985). Neonatal perception of spatial relationships. Infant Behavior and Development, 8, 15-23. Berlin, B., 8~Kay, P. (1969). Basic color terms: Their universality and evolution. Berkeley: University of California Press. Brugman, C. (1983). The use of body-part terms as locatives in Chalcatongo Mixtec. Survey of California and other Indian languages, 4, 235-90. University of California, Berkeley. Carlson, L., Regier, T, & Covey, E. (in press). Defining spatial relations: Reconciling axes and vector representations. To appear in E. van der Zee & J. Slack (Eds.), Axes and Vectors in Language and Space. Oxford, England: Oxford University Press. Carlson-Radvansky, L., & Logan, G. (1997). The influence of reference frame selection on spatial template construction. Journal of Memory and Language, 37,411-437. Casad, E., & Langacker, R. (1985). “Inside” and “outside” in Cora grammar. International Journal of American Linguistics, 51, 247-281. Chomsky, N. (1986). Knowledge of language: Its nature, origin, and use. New York: Praeger. Clark, H. (1973). Space, time, semantics, and the child. In T. Moore (Ed.), Cognitive Development and theAcquisitionof Language (pp. 27-63). New York: Academic. Comrie, B. (1981). Language universals and linguistic typology. Chicago: University of Chicago Press. Downing, C., & Pinker, S. (1985). The spatial structure of visual attention. In M. Posner and 0. Marin (Eds.), Attention and Performance-XI (pp. 171-187). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Gapp, K.-I? (1994). A computational model of the basic meanings of graded composite spatial relations in 3-d space (Tech. Rep. No. 111). Universitgt des Saarlandes, Saarbriicken, Germany, Department of Computer Science. Gapp, K.-F?(1995). Angle, distance, shape, and their relationship to projective relations. In J. Moore & J. Lehman (Eds.), Proceedings of the 17th Annual Conference of the CognitiveScienceSociety (pp. 112-l 17). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Georgopoulos, A. P, Schwartz, A. B., & Kettner, R. E. (1986). Neuronal population coding of movement direction. Science,233, 1416-l 4 19. Greenberg, J. H. (1978). Universals of human Zanguage. Stanford, CA: Stanford University Press. Gumperz, J. J., & Levinson, S. C. (1996). Rethinking linguistic relativity. Cambridge, England: Cambridge University Press. Hoffman, C., Lau, I., & Johnson, D. R. (1986). The linguistic relativity of person cognition: An English-Chinese comparison. Journal of Personality and Social Psychology, 51, 1097-1105. Kay, I!, & McDaniel, C. K. (1978). The linguistic significance of the meanings of basic color terms. Language, 54, 610-646. LaBerge, D., & Brown, V (1989). Theory of attentional operations in shape identification. Psychological Reviav, 96, 101-124. Lakoff, G. (1987). Women, fire, and dangerous things: What categories reveal about the mind. Chicago: University of Chicago Press.

Regier and Carlson

l

221

Levinson, S. C. (1996). Frames of reference and Molyneaux’s question: Crosslinguistic evidence. In I? Bloom, M. Peterson, M. Garrett, & L. Nadel (Eds.), Language and Space (pp. 109-169). Cambridge, MA: MIT Press. Logan, G. (I 994). Spatial attention and the apprehension of spatial relations. Journal of Experimental Psychology: Human Perception and Pegormance, 20, 1015-1036. Logan, G., & Sadler, D. (1996). A computational analysis of the apprehension of spatial relations. In I? Bloom, M. Peterson, M. Garrett, & L. Nadel (Eds.), Language and Space (pp. 493-529). Cambridge, MA: MIT Press. Lucy, J. (1992). Grammatical categories and cognition: A casestudyof the Zinguistic relativity hypothesis. Cambridge, England: Cambridge University Press. Mandler, J. (1992). How to build a baby: II. Conceptual primitives. Psychological Review, 99, 587-604. Mandler, J. (1996). Preverbal representation and language. In I? Bloom, M. Peterson, M. Garrett, & L. Nadel (Eds.), Language and Space (pp. 365-384). Cambridge, MA: MIT Press. Needham, A., & Baillargeon, R. (1993). Intuitions about support in 4.5-month-old infants. Cognition, 47, 121-148. Regier, T (1996). The human semantic potential: Spatial language and constrained connectionism. Cambridge, MA: MIT Press. Regier, T., & Carlson, L. (2001). Grounding spatial language in perception: An empirical and computational investigation. Journal of Experimental Psychology: General, 130, 273-298. Spelke, E. S., Breinlinger, K., Macomber, J., & Jacobson, K. (1992). Origins of knowledge. Psychological Review, 99, 605-632. Talmy, L. (1983). How language structures space. In H. Pick & L. Acredolo (Eds.), Spatial orientation: Theory, research, and application (pp. 225-282). New York: Plenum Press. Also available as Tech. Rep. 4, Institute of Cognitive Studies, University of California at Berkeley. Treisman, A., & Gormican, S. (1988). Feature analysis in early vision: Evidence from search asymmetries. Psychological Review, 95, 15-48. Whorf, B. L. (1956). Language, thought, and reality. Cambridge, MA: MIT Press. Wilson, H. R., & Kim, J. (1994). Perceived motion in the vector sum direction. Vision Research, 34. 1835-1842.

This Page Intentionally Left Blank

13 -4% Conceptual Development in Infancy: The Case of Containment Elizabeth S. Spelke Susan J. Hespos

Massachusetts

Institute

of Technology

H

ow do children develop the concepts that are expressed by the individual terms of human languages-concepts of objects such as dog, of events such as lunch, of actions such as jump, and of spatial relationships such as on? The most popular accounts of this ability root the development of concepts either in language itself or in perception. On the first of these accounts, children construct concepts such as dog or on by observing commonalities in the events to which speakers refer when they say dog or on. On the second of these accounts, the child’s perceptual systems naturally are attuned to such commonalities. Accounts that appeal both to language experience and to perceptual biases also have been offered (e.g., Smith, Jones, & Landau, 1996). Jean Mandler (1988,1992,1998) proposed an alternative to all such theories of conceptual development. In her view, concepts develop from a foundational system that is distinct both from the child’s perceptual capacities and from language. This foundational system has five principal properties. First, it is a primitive system with its own developmental origin and initial course. Second, it is a categorical system that 223

224

l

13.

Conceptual

Development

in Infancy

groups together perceptually diverse entities and treats them as equivalent. Third, it is an accessible system that can be used for a variety of purposes including thinking, recalling, and acting in multiple ways. Fourth, it is primarily a spatio-temporal system that operates by detecting the locations and motions of objects in relation to one another. Fifth, it is a conscious system, distinct from the unconscious mechanisms that give rise to the child’s sensory, motoric, and linguistic representations. In this chapter, we consider all these claims in the context of research on infants’ representations of objects. Our review of this research leads us to argue that four of Mandler’s claims are correct: Infants indeed have a conceptual system that is primitive, categorical, accessible, and attuned to spat&temporal relationships among objects. We remain neutral concerning Mandler’s claim that this system operates consciously. Finally, we argue that the conceptual system studied by Mandler serves not only to categorize the spatial relationships among objects but also to represent the unity and boundaries of objects, their identity through time, and their behavior. This system is domain-specific: It applies to inanimate, manipulable objects but not to other important, perceptible entities in the child’s environment such as people. It is one of a set of systems of core knowledge. To make our discussion more manageable, we focus on the development of a small set of concepts that capture spatial relationships between distinct objects, particularly the relationship of containment. We begin by considering the infant’s ability to distinguish containment (the relationship expressed in English by “in”) from a different spatial relationship expressed in English by “behind” and referred to by Baillargeon (1998) and others as occlusion.’ We first ask whether young infants represent the distinction between containment and occlusion prior to language. In light of evidence that they do, we further ask whether this distinction is represented within a system that is categorical and accessible. Then we consider the development of a further distinction that is not captured by the lexicon of English but is captured by simple morphemes in certain other languages: the distinction between tight-fitting and loose-fitting relationships between objects. We ‘Although we follow Baillargeon’s terminology and refer to this relationship as occlusion, note that this term is somewhat misleading. Because occlusion occurs when objects enter into a variety of spatial relationships, including containment, not all occlusion relationships are captured by the “behind” concept. Moreover, because an object is not occluded when it moves behind a transparent object, some relationships captured by the “behind” concept do not involve occlusion. Here we discuss only the simple case studied by Hespos and Baillargeon (2001), in which one object is placed behind a second, opaque object and therefore is occluded by it, and we refer to this case alone as a relation of “occlusion.”

Spelke and Hespos

l

225

ask whether young infants who are exposed only to English make this spatial distinction. In light of evidence that they do, we conclude by asking how infants’ ability to distinguish tight from loose-fitting relationships between objects relates to their ability to represent individual objects and reason about their motions. CONTAINMENT AND OCCLUSION: A CONCEPTUAL DISTINCTION Recent research provides evidence that 2%month-old infants can distinguish between containment and occlusion events and are sensitive to some of the constraints that these different relationships place on the behavior of objects (Hespos & Baillargeon, 2001). Infants were shown an event where a small cylindrical object was lowered either behind or inside a container; next, the container was moved forward and to the side revealing the object behind the container. There is nothing surprising about this outcome for the behind condition. In the inside condition, however, the outcome appears to adults to be impossible, because the contained object should have remained inside and moved with its container. If infants had expectations similar to adults’ expectations for these two events, they were expected to look longer at outcome for the inside condition. The results confirmed this prediction. These findings suggest that infants as young as 2% months of age have different expectations about containment and occlusion events. Is the distinction between containment and occlusion categorical and conceptual for infants, or does it reflect perceptual sensitivity to continuous stimulus variations? This is a question that Mandler has asked repeatedly in other contexts: For example, do babies group diverse animals together because they look similar or because they are represented as members of the same kind (e.g., Mandler, 1992)? To address this issue, Mandler presented infants with superficially very different but conceptually similar entities, asking whether infants would respond similarly in the face of this perceptual diversity. The approach of Hespos & Baillargeon (2001b) is close in logic to Mandler ‘s: They presented infants with perceptually very similar but conceptually different spatial relationships, asking whether infants would respond differently to those different relationships. For example, one series of experiments tested whether 4- to &month-old infants could assess how much of a tall object should be hidden when lowered either behind an occluder or inside a container. In the container condition, infants were presented with events in which a tall object was lowered out of view into a container either of equal height (consistent) or of half its height (inconsistent). Similarly, in the

226

l

13.

Conceptual

Development

in Infancy

occluder condition, infants were shown the same object hidden behind either a tall or short occluder. To make the occluder and container events perceptually as similar as possible, the occluders were constructed by removing the bottom and back half of the containers, so that they formed rounded occluders (see Fig. 13.1). These events were perceptually similar once the object was fully hidden, but they were conceptually different for adults, because the object was hidden inside a container in one and behind an occluder in the other. The results showed that infants detected the violation in the occlusion events at 4% months of age. In contrast, infants failed to detect the violation in the perceptually matched containment events until 7% months of age. These findings suggest that containment and occlusion are categorically distinct classes of events for infants, despite their perceptual similarity, and that representations of objects and inferences about their behavior depend in part on this conceptual distinction. LOOKING

AND REACHING TO CONTAINED VERSUS OCCLUDED OBJECTS: AN ACCESSIBLE DISTINCTION

Is the conceptual distinction between occlusion and containment an accessible one for infants? One hallmark of an accessible representation is that it can guide diverse actions including visual exploration, manual exploration, communication, and any other actions in one’s repertoire. To investigate whether the containment or occlusion distinction is accessible to young infants, therefore, Hespos and Baillargeon (2002) asked if the same distinction that guides infants’ preferential looking in the aforementioned experiments also would guide their predictive reaching in a different experimental context. In these experiments, infants of 5% and 7% months of age were presented with a tall frog and were encouraged to play with it. After a few seconds, the frog was removed and the infants were presented with two occluders or containers that had frog legs wrapped around the sides of them so that the frog’s feet were sticking out in front and could be grasped directly. As in the preferential-looking studies, the occluders and containers were perceptually identical to one another, except that the container had a continuous back and therefore surrounded the frog. In addition, the pairs of occluders and containers were identical except for their height; one was tall enough to conceal the entire frog behind it, whereas the other was one third the needed height. After the infants’ attention was drawn to each display, the apparatus was moved toward the infants, whose reaching was observed. Infants of both ages reached significantly more often to the frog’s legs that protruded from behind the tall occluder than to those pro-

FIG. 13.1 Test displays used in studies of infants’ developing distinction tween containment and occlusion (after Hespos & Baillargeon, 2001b).

be-

227

228

’

13.

Conceptual

Development

in Infancy

truding from behind the short occluder, although, in a control condition (with no initial exposure to the frog), they showed no intrinsic preference for the tall display. These reaching patterns provide evidence that the infants appreciated that only the tall occluder could conceal the frog. In contrast, infants reached more often for the frog inside the tall container at 7% but not 5% months of age. These experiments therefore provide evidence that infants exhibit the same developmental pattern in an object-retrieval task as in a preferential-looking task: The same developing representations of object occlusion and containment appear to guide both visual exploration and manipulation of objects for young infants. Taken together, these findings suggest in turn that the developing conceptual distinction between in and behind is available to guide diverse actions in infancy. This conceptual distinction appears to be an accessible one, consistent with Mandler ‘s arguments for an accessible, developing, conceptual system in infancy. CONCEPTS AND LANGUAGE: TIGHT-FITTING AND LOOSE-FIITING CONTAINMENT Because children do not begin to use terms such as in and behind until well after the 1st year, and because all the children in the aforementioned experiments were well under 1 year of age, it is tempting to conclude from the aforementioned studies that the categorical distinction between in and behind is a primitive distinction that develops prior to and independently of language. Nevertheless, two recent sets of findings appear to cast some doubt on this conclusion. First, research by Jacques Mehler, Peter Jusczyk, and others provides evidence that infants learn a great deal about their native tongue long before they begin to speak. Within the first days of life, neonates recognize their native language and discriminate this language from other languages with different rhythmic properties (Mehler et al., 1988). By 6 months of age, infants relate the words “mommy” and “daddy” to their referents, looking at the appropriate parent when each word is spoken (Tincoff & Jusczyk, 1999). By 7% months, infants are able to learn to recognize new words within a single session, after only brief exposure to the word (see Jusczyk, 1997). Given the frequency of spatial terms such as in, in the spoken language that surrounds the infant, therefore, we cannot exclude the possibility that the 7%month-old infants in Hespos and Baillargeon’s (2002) experiments have learned something about its meaning. Second, research by Choi and Bowerman (1991) yielded provocative findings concerning the effects of language on conceptual development, based on an interesting difference between the spatial vocabular-

Spelke and Hespos

.

229

ies of different languages. Not all natural languages have words that express the containment concept expressed in English by “in.” In Korean, for example, morphemes appended to verbs of motion express a different distinction; the distinction between moving an object into a tight-fitting relationship with a second object and moving it into a loose-fitting relationship. The tight fit and loose fit distinction cross-cuts the English distinction between containment (“in”) and support (“on”): whereas English speakers put the apple “in the bowl” (containment), the book “on the table” (support), and the ring “on the finger” (support), Korean speakers describe both the apple and bowl and the book and table as entering into loose-fitting relationships, whereas a book placed in its jacket and a ring placed on a finger enter into tight-fitting relationships. Importantly, Choi and Bowerman (1991) showed that adult speakers of English and Korean categorize events such as those just described differently from one another, in ways that reflect the categorical distinctions of their language. Moreover, Korean children appear to learn the morphemes that express the tight and loose distinction as quickly as English children learn “in” and “on,” and they make few errors in applying these terms. Korean children do not go through a stage in which they try to fit the terms of their language into the spatial categories of containment and support to which American infants are sensitive. Together, with the findings of Jusczyk and others described earlier, these findings raise the possibility that language guides the development of spatial concepts from a very early point in development, contrary to Mandler’s thesis. There is, however, an alternative interpretation of Choi and Bowerman’s (1991) findings, first suggested by Mandler (1998): “The fact that there is more than one way to express various spatial relationships does not mean that language itself is teaching relations previously unanalyzed by the language learner . . . . Rather, it seems more likely that different languages teach children different ways to express what they have already noticed through perceptual analysis” (p. x). On this view, she continues, “there is a universal set of relational concepts that preverbal children everywhere have analyzed before learning language. Bowerman’s (1989) analyses indicate that this set is larger than that expressed in any given language” (Mandler, 1998, p. 295). On this view, children may come to the task of learning language equipped with all the principal categorical distinctions that the early-developing parts of the lexicon express; not only the distinctions of English but those of Korean and other languages. If that is the case, then prelinguistic children would possess a richer set of conceptual distinctions than those expressed by any language, and learning a language would require that they select, from among the conceptual distinctions

230

l

13.

Conceptual

Development

in Infancy

in their repertoire, those distinctions that the language expresses. As a result of this selection process, adult speakers may show reduced sensitivity to conceptual distinctions that are not marked by their language. If word learning proceeded as Mandler suggests, then the development of lexical semantics would resemble the development of phonology. We now know that infants come to the task of learning the phonological distinctions of their language equipped with all the principal phonetic distinctions expressed by natural languages. Over the course of experience with their ambient language, infants learn to select from among those distinctions the ones that their own language uses to convey differences of meaning. This selection comes to influence speech perception by adults, who remain sensitive to the distinctions expressed in their language but who are generally less sensitive to non-native distinctions. Young infants therefore show greater sensitivity than do their parents to distinctions among speech sounds that signal differences in meanings in unfamiliar languages but that signal no such differences in their native language (Werker & Tees, 1984). As in the case of speech perception, Mandler ‘s suggestion implies, young infants may show greater sensitivity than their parents to conceptual distinctions that are captured by the terms of some unfamiliar languages but that are not lexicalized in the native language. Experiments by McDonough, Choi, and Bowerman (1999) attempted to test both of these predictions by assessing the spatial categorizations of infants and adults in an English-language environment. Participants were presented with a set of complex and heterogeneous events in which two objects entered various spatial relationships. In Experiment 1, infants aged 9 to 14 months were familiarized with six different scenes in which objects entered into either a relation of tight-fitting containment (a relation expressed by “in” in English and by “kkita” in Korean), or a relation of loose-fitting support (a relation expressed by “on” in English and by “nohta” in Korean). Although each infant observed the same spatial relationship on every familiarization trial, the scenes were otherwise diverse and involved a variety of objects and motions. After familiarization, infants were tested with novel scenes exhibiting each of the two spatial relationships. If infants categorized the events on the basis of their spatial relationships, they should have looked longer at the test event that exhibited the novel spatial relationship. Infants at 14 months showed this test preference, providing evidence that they categorized the spatial relationships in a manner consistent with many natural languages including both English and Korean. In contrast, !&month-old infants looked longer at the test scene that exhibited the familiar spatial relationship, and 11 -month-old infants showed no consistent preferences.

Spelke and Hespos

l

231

In Experiment 2, infants of the same ages were familiarized with six different scenes in which objects entered into either a relation of tight-fitting containment or a relation of loose-fitting containment: two spatial relationships that are described similarly in English but differently in Korean. Like the youngest infants in Experiment 1, infants of all ages in Experiment 2 looked longer at the familiar spatial relationship. Al these findings suggest that infants show some sensitivity to both of the contrasting spatial relationships, because in the absence of any such sensitivity infants should have looked equally at the two classes of events. Nevertheless, the direction of sensitivity corresponded to that which is typically found in categorization research only for the oldest infants presented with a contrast that is lexicalized in English. The authors suggested that abilities to categorize spatial relationships begin to develop before language but are fragile until children begin to learn the spatial terms of their language. Experiments with adults supported this conclusion. English-speaking adults, tested in the same manner as infants, showed a preference for the novel test events of Experiment 1 but showed no such preference among the events of Experiment 2. Similar results were obtained in a different categorization task with adults presenting the same stimuli. Like infants, adults appeared to categorize readily only those spatial relationships that corresponded to the terms of their language. Although the findings of McDonough et al. (1999) appeared to provide some support for the thesis that spatial categorization depends in part on the acquisition of language, the authors noted that they are open to other interpretations. In particular, infants may have been confused or distracted by the wide variation among the events that were presented to them, especially at the younger ages. More generally, early-developing categorization abilities may reveal themselves more clearly when infants are presented with simpler events that exhibit a minimal contrast between two conceptual categories, as in Hespos & Baillargeon’s (200 1 a, 200 1b) studies of containment and occlusion. That is what our experiments attempted to do. In the first experiment (Hespos & Spelke, 2001), we tested 5-monthold U.S. infants’ categorization of tight-fitting versus loose-fitting containment relations using a habituation-dishabituation paradigm. First, infants saw a narrow cylindrical object lowered into a series of loose-fitting, medium-sized containers on a series of trials until their looking time declined (see Fig. 13.2a). Next, the infants were presented with six test trials in which the same cylindrical object was lowered, in alternation, into a wide container (1 l/2 times wider, hence also a loose fit) and into a narrower container (1% times more narrow, a tight fit). If infants make a language-independent categorical distinction be-

23i

233

234

l

13.

Conceptual

Development

in Infancy

tween tight- and loose-fitting containment events, then they were expected to look significantly longer at the tight-fit trials. Our results confirmed this prediction. Because the containers differed in size for the tight- and loose-fitting test events, it was possible that the results from the first experiment stemmed from an inherent preference for perceptual aspects of the tight-fitting event. To test this possibility, a second experiment compared infants’ looking times to the same test events after habituation to an event in which a medium-sized object was lowered into the same medium-sized container as in the first study: a “tight-fit” event. The infants in this experiment showed the opposite pattern of behavior, looking longer at the loose-fit test event (see Fig. 13.2b). Together, these findings provide evidence that infants categorized the containment events as tight- or loose-fitting and mapped the categorical distinction seen during habituation trials onto the events that they saw during test trials. Infants living in an English-speaking environment therefore are sensitive to the categorical distinction between tight-fitting and loose-fitting containment relationships, as Mandler had speculated. When exposed to continuous variation in the size of a contained object relative to its container, infants make the categorical distinction captured by the Korean morphemes “kkita” and “nehta.” We conclude that sensitivity to this distinction develops in the absence of any relevant linguistic experience, prior to and independently of the language the child will learn. In two respects, our findings resemble the findings of studies of phonetic discrimination in speech perception. First, we found that infants parse a continuum of spatial variation into categories of spatial relationships between objects, just as prior studies have found that infants parse a continuum of acoustic variation into categories of speech sounds (see Jusczyk, 199 7). Second, we found that infants are sensitive not only to the spatial distinctions that are lexicalized in their native language but also to spatial distinctions that are lexicalized in other, non-native languages. Similarly, studies of speech perception have found that infants are sensitive to the phonological distinctions of non-native languages as well as to the phonological distinctions of their native language. These findings raise the question whether the development of spatial concepts and of speech perception are similar in a third respect: Over the course of development, do speakers lose sensitivity to the conceptual distinctions that are not captured by the lexical semantics of their native language, just as they lose sensitivity to phonetic distinctions not captured by their native language phonology? To begin to address this question, we presented all the same containment events to two groups of English-speaking adults. Adult participants first were shown

Spelke and Hespos

l

235

the same habituation trials as the infants: one group was saw the loose-fitting containment event of our first experiment and the other group saw the tight-fitting containment event of our second experiment. Then both groups of adults were shown the two test trials consisting of the narrow and wide test events that the infants observed. After each test event, participants were asked to rate the similarity between the test event and the habituation events that they had seen. Finally, participants were asked to rank the test trials in similarity to the habituation trials. In contrast to the infants’ patterns of preferential looking, the adults rated the two test events as equally similar to the habituation event. The adults, therefore, did not appear to make the same categorical distinction as the infants. So far, we’ve discussed experiments that support three of Mandler’s claims, providing evidence for an early-developing system of conceptual representation that is categorical, accessible, and primitive in the sense that it develops prior to and independently of language. Now we turn to Mandler’s last two claims, that the system that gives rise to infants’ early-developing conceptual distinctions is primarily attuned to spatio-temporal properties of objects and relationships among objects, and that this system is conscious in a way that distinguishes it from any of the infant’s other systems for representing the environment. SPATIAL CONCEPTS

AND OBJECT REPRESENTATIONS IN INFANCY

Mandler (1988, 1992, 1998) proposed that infants’ earliest concepts develop through a process of perceptual analysis, and that this process initially focuses primarily on the locations and motions of objects. For example, infants conceptualize some objects as animals by noticing their patterns of self-propelled, irregular, and contingent motion; they conceptualize other objects as furniture or kitchen things by noting their characteristic locations and functions; and they conceptualize spatial relationships between objects as containment by noticing the motion of one object from a position outside to a position inside another object. Because the process of perceptual analysis focuses first on large-scale spatial and kinetic relationships, infants categorize objects at the global levels of animals and furniture before they categorize objects at finer levels such as dogs and cats, although infants’ perceptual systems are highly sensitive to the different appearances of dogs and cats (Quinn & Eimas, 1996). Thus, Mandler claims, infants’ concepts depend on a process of spatio-temporal analysis that is distinct from the perceptual processes by which infants detect the fine-grained colors, shapes, and other properties of objects.

236

l

13.

Conceptual

Development

in Infancy

The cases discussed by Mandler are not the only cases, however, in which infants appear to be especially sensitive to spat&temporal information. Infants also rely primarily on spat&temporal information when they perceive the unity and boundaries of objects, when they reach for objects and extrapolate their motions, and when they follow objects through time and determine relations of numerical identity and distinctness (i.e., whether an object seen now is the same one as a given object seen in the past). For example, infants perceive the unity and connectedness of a center-occluded object by detecting the common motion of its visible surfaces (Kellman & Spelke, 1983; Johnson & Naiiez, 1995) but not by detecting the common coloring of its visible surfaces, even when that coloring is highly salient and synchronously changing (Jusczyk, Johnson, Spelke, & Kennedy, 1999). How do these abilities relate to the ability to form spatial concepts? Mandler discussed the infant’s capacity for object representation in two different ways. On one hand, she drew from findings, such as those previously discussed, the conclusion that infants are especially sensitive to spat&temporal information about objects, and she suggested that this sensitivity may explain why their earliest-developing concepts depend on such information. On the other hand, she argued that the processes of perceptual analysis that give rise to the first true concepts are conscious processes, whereas the processes that give rise to object representations are not (Mandler, 1998). In Mandler ‘s view, object perception is an encapsulated and unconscious process, whereas concepts such as animal and containment are accessible and conscious. Are the processes by which infants categorize spatial relations among objects-relations like containment and tight-fit-qualitatively different from the processes by which infants represent objects in the first place? To approach this question, we need first to consider further what are the processes of object representation. One of us has argued (Spelke, in press) that object representation depends on a system of core knowledge, with four central properties, First, the system gives rise to representations that are abstract and largely independent both from sensory systems and from response systems. On the sensory side, object representations are amodal and persist even in the absence of sensory support, when objects are occluded. On the motor side, object representations guide diverse actions including visual exploration, manipulation, and locomotion. Second, the system is domain-specific: It serves to represent objects but not to represent other perceptible entities such as persons, collections, or nonsolid substances (see Chiang & Lynn, 2000; Huntley-Fenner & Carey, 2001; Woodward, Phillips, & Spelke, 1993). Third, the system is informationally encapsulated: It operates on information about the spat&temporal relationships between visible surfaces but not

Spelke and Hespos

l

237

on information about other surface properties such as color and detailed shape, although those properties are detectable by infants and used for other purposes. Fourth, the system provides the representations that are the building blocks for learning and cognitive development. It provides the initial set of concepts that allows children to develop systematic knowledge of their surroundings and to develop abilities to communicate that knowledge by language. All of these properties, we believe, are true of the conceptual system that Mandler has described. Contrary to Mandler, we propose that a single system of representation underlies both the infant’s capacity to represent objects as unitary and persisting and the infant’s capacity to categorize objects and spatial relationships in the ways that Mandler has described. But how can our proposal and Mandler ‘s be distinguished empirically? Mandler has articulated one set of predictions that separate the two views: On her view, infants should be conscious of the processes by which they form concepts, whereas they should not be conscious of the processes by which they form representations of objects. Consciousness is an elusive phenomenon to study at any age, however, and it is especially difficult to study in infants. Here, we propose and test a different prediction that may separate the two views. On our view, the very same principles that govern infants’ construction of representations of the unity, identity, and behavior of inanimate, manipulable objects also should govern infants’ construction of categorical distinctions among sets of objects and spatial relationships. On Mandler ‘s view, in contrast, there is no reason to expect such a convergence. Specifically, one of us has argued that young infants build representations of objects in accord with six constraints on object motion:2 cohesion (objects do not spontaneously break apart as they move), boundedness (distinct objects do not spontaneously merge as they move), continuity (objects do not move on paths with gaps in space or time), solidity (distinct objects do not occupy the same place at the same time), no action at a distance (objects do not influence one another’s motion when they are separated in space), and action on contact (objects do not move independently when they are in contact). If early-developing concepts of objects and their spatial relations are products of the same system, then they should be guided by the same constraints. In light of this prediction, consider the distinction between the spatial concepts in and behind studied by Hespos and Baillargeon (2001a). These two spatial relationships limit the relative motions of objects in very different ways, given the aforementioned constraints. Because of the solidity 2These six constraints can be reduced to the three spat&temporal principles of cohesion, continuity, and contact (Spelke & Van de Walle, 1993). For these purposes, however, it is more useful to consider the six constraints on motion.

238

l

13.

Conceptual

Development

in Infancy

and the action on contact constraints, an object inside a container will move when and where its container moves. Because of the no action at a distance constraint, an object that is behind and spatially separated from a screen will not move when and where its occluder moves. We have seen that infants as young as 2% months are sensitive to these differing limits and make accurate predictions about object motion in accord with these constraints (Hespos & Baillargeon, 2001a; see also, Aguiar & Baillargeon, 2000; Wilcox, Nadel, & Rosser, 1996). Now consider the distinction between tight-fitting and loose-fitting relationships between objects. When two objects fit tightly together, such as a ring on a finger or a cylinder in a cylindrical container just wide enough to hold it, then almost any motion of one object will induce an exactly parallel motion in the other object.3 The constraints of solidity and action on contact ensure that these objects will move together unless one acts specifically to separate them. In contrast, when two objects fit loosely together, such as an apple in a bowl or on a table, the motions of the two objects are only partly constrained by one another. Because the objects are solid, the apple cannot move laterally through the side of the bowl or downward through the surface of the table; because the objects are in contact, motions of the bowl or table will influence the motion of the apple. In neither case, however, will the motions of two loose-fitting objects be strictly parallel. If a bowl containing an apple is suddenly moved, for example, the apple and bowl will undergo both common and relative motions, with the apple both moving with the bowl and rolling against it. Because tight- and loose-fitting support place different constraints on the motions of objects, it is possible that the principles governing infants’ representations of objects and their motions could also lead infants to categorize these spatial relationships differently, into the categories of support, containment, tight-fit, and loose-fit that are lexicalized in various languages. However, do infants in fact respect these principles in their spatial categorizations? Our last experiments were undertaken to address this question. In the first experiment, we used a preferential-looking paradigm to test 5-month-old infants’ expectations about how motion affects loose-fitting containment relations. First, infants saw a narrow cylindrical object lowered into a wide container until their looking time declined. Next, the infants were presented with six test trials that 3With a suitable action, of course, one can remove the ring from the finger, and with purely vertical motion, one can lift the cylinder from its container. In both cases, however, the motions that will serve to separate the objects are highly nonaccidental; random motions of either object will almost always induce a corresponding motion in the other object.

Spelke and Hespos

l

239

alternated between a move-separately event and a move-together event (see Fig. 13.3a). In the move-separately event, the cylindrical object was lowered inside the wide container and then the container remained stationary and the object moved back and forth inside the container (consistent). In the move-together event, the cylindrical object was lowered inside the wide container and then both the object and container moved horizontally as a unitary object (inconsistent). If infants expected the loose-fitting container to allow the object to move with some independence, then they were expected to look longer at the move-together event. Our results confirmed this prediction: Infants looked significantly longer at the move-together than at the move-separately events. In a second experiment, we similarly tested infants’ expectations for the effects of motion on tight-fitting containment relations. Infants saw the same cylindrical object lowered into a narrow container during the habituation and test trials. In the test trials, infants saw the object inside the container moved back and forth horizontally (see Fig. 13.3b). On alternate trials, the object and container moved together (consistent) or separately (inconsistent). If infants appreciated that a tight-fitting container more strongly constrains the motion of its contained object, then infants were expected to show the opposite looking preference from those in the loose-fitting condition and look longer at the move-separately event compared to the move-together event. The results confirmed this prediction. These experiments reveal a close linkage between infants’ ability to categorize spatial relationships between objects and their sensitivity to the ways in which the motions of objects in these relationships are constrained. As in other studies of object perception and object representation, infants’ sensitivity to object motions is captured by a small set of constraints including solidity, no action at a distance, and action on contact. The same constraints on object motion therefore account both for infants’ representations of objects and infants’ categorization of spatial relationships between objects. How does this system compare to the conceptual system that Mandler described in her writings? We already noted that the system has four properties which Mandler emphasized: It is primitive (that is, not derived from other systems or processes like sensory-motor integration or language learning), categorical, accessible to multiple response systems, and focused on spat&temporal information. It may or may not have the fifth property Mandler described: the property of being a conscious system. Although accessibility and consciousness are related properties, they are not the same. Accessibility is a property of functional cognitive architecture: A system of representation is accessi-

240

241

242

l

13.

Conceptual

Development

in Infancy

ble if its outputs are available to a wide range of response systems, regardless of whether the perceiver is aware of those outputs. Consciousness, in contrast, is a property of human experience: A system of representation is conscious if we are aware of its operations and outputs, regardless of our abilities to act on those representations. The subjective nature of conscious experience makes it difficult to determine whether and when infants are conscious, and we remain neutral on this question. Contrary to Mandler, we argue that the spatio-temporal system of object representation has a sixth property. It is a domain-specific system; one among many. Evidence for the domain-specificity of this system comes from a consideration of the spat&temporal constraints that guide it. Constraints such as action on contact and no action at a distance apply to the motions of inanimate, material bodies. They do not, however, apply to the motions of other perceptible entities, including animals and people (Spelke, Phillips, & Woodward, 1995). Infants, moreover, apply different principles when they reason about inanimate object motions on one hand and about human actions on the other: whereas inanimate object motions are seen as subject to the constraints of contact mechanics, human actions are seen as goal-directed (Woodward, 1998), intentional (Meltzoff, 1995), and socially responsive (Johnson, Slaughter, & Carey, 1998). Findings such as these suggest that infants are not endowed with a single process of perceptual analysis giving rise to a single, unified conceptual system but rather with a collection of such systems, each giving rise to a set of concepts within a particular domain. At the foundation of human cognition are multiple systems of core knowledge. CONCEPTUAL

BEGINNINGS

AND COGNITIVE

DEVELOPMENT

We emphasized in this chapter that infants have early developing, primitive conceptual systems, and that these systems both precede and guide the development of language. It does not follow from this view, however, that cognitive development is a trivial process, or that language development fails to affect it. Indeed, we believe the core knowledge thesis may lead to the opposite conclusions. If infants begin with a set of distinct, domain-specific systems of core knowledge, then they have much work to do over the course of cognitive development: They must come to relate these systems to one another, and to the world that the child perceives. The world of objects is not packaged neatly into domains that match the infant’s core systems. For example, children must learn that there are classes of objects-animals-whose behavior is both goal-directed and subject to mechanical

Spelke and Hespos

l

243

constraints: objects that cross-cut the infant’s distinction between intentional beings and inanimate manipulables. Children also must learn properties follow that there are other classes of objects- tools-whose not only from mechanical constraints but from human intentions: objects that are designed to serve human purposes. Such discoveries, we suggest, require that information from distinct core systems of representation be combined together in new ways. The child’s developing language may be central to this developmental process, in two ways. In writing about the relationship between language and thought, Mandler is quite open to the possibility that these developing functions mutually influence one another, and she proposes one way in which language can exert this influence. Although initial concepts, constructed by perceptual analysis, guide the first steps of word learning, the structure of the language to be learned may in turn guide the later elaboration of those concepts (e.g., Mandler, 1998). Words, first acquired in relation to concepts that are constructed by perceptual analysis, may in turn come to influence the process of perceptual analysis itself and the concepts to which it gives rise. For example, a language like Korean, with terms that distinguish tight- from loose-fitting relationships between objects, may call Korean speakers’ attention to the details of those relationships, leading to new perceptual analyses of the relationships between objects and to an elaboration of the tight-loose conceptual distinction. Although the capacity for perceptual analysis is innate and universal, on Mandler’s view, the particular directions that this analysis takes may be influenced by language and by other aspects of experience. Elsewhere, one of us suggested a further way in which language may influence the child’s developing concepts (Hermer-Vazquez, Spelke, & Katsnelson, 1999; Spelke & Tsivkin, in press). As the child comes to master the combinatorial syntax and compositional semantics of her native language, that language may serve as a medium for conjoining concepts from diverse domains and constructing new concepts that cross-cut those domains. In contrast to core-knowledge systems, language is a domain-general system of representation: it allows US to talk about anything we can conceive, regardless of the domain in which those concepts are couched, and it allows us to combine distinct concepts at will. Outside of language, representations of inanimate objects and of persons may be products of core systems that show little interaction. With language, however, we can easily relate them together, entertaining thoughts such as “Mary is a robot” or “This computer is malicious.” We even may learn words for concepts whose features reside in different categories, such as names for tools or animals. Once the child has learned words and expressions that capture core concepts,

244

l

13.

Conceptual

Development

in Infancy

therefore, the child may be able to use the combinatorial resources of his or her language to express new concepts with components in distinct core domains. CONCLUSION The above suggestions are speculative, but we may close on firmer ground. However much children’s concepts and thinking change over the course of development and learning, these concepts are built on foundational systems that first emerge in infancy. Because many of our foundational concepts are clearest during the infancy period, studies of conceptual development in infants may allow cognitive scientists to approach many difficult questions concerning the structure and content of human knowledge at later ages. It is not easy to study conceptual development in infancy, because it is difficult to determine whether any given behavior pattern observed in infants results from representations that are perceptual or conceptual, implicit or explicit, primitive or derived. Fortunately, Jean Mandler has helped us all to think about these distinctions and to craft experiments that bring us closer to understanding the nature of infants’ representations and the origins of their concepts. ACKNOWLEDGMENTS Supported by a grant to ESS from the National Institutes of Health (R37-HD23103) and by a grant to SJH from the McConnell-Pew Foundation (3987507). We thank Lori Markson and Fei Xu for helpful comments on an earlier draft. REFERENCES Aguiar, A., & Baillargeon, R. (2000). Perseveration and problem solving in infancy. In H. W. Reese(Ed.), Advances in child development and behavior (Vol. 27, pp.135-180). San Diego, CA: Academic. Baillargeon, R. (1998). Infants’ understanding of the physical world. In M. Sabourin, E Craik, & M. Robert (Eds.), Advances in psychological science (Vol. 2, pp. 503-529). London: Psychological Press. Bowerman, M. (1989). Learning a semantic system: What roles do cognitive predispositions play? In M. L. Rice & R. L. Schiefelbusch (Eds.), The teachability of language (pp. 133-l 69). Baltimore: Brookes. Chiang, W.-C., & Wynn, K. (in press). Infants’ tracking of objects and collections. Cognition. Choi, S., & Bowerman, M. (1991). Learning to express motion events in English and Korean: The influence of language-specific lexicalization patterns. Cognition, 41, 83-121.

Spelke and Hespos

l

245

Hermer-Vasquez, L., Spelke, E. S., & Katsnelson, A. S. (1999). Sources of flexibility in human cognition: Dual-task studies of space and language. Cognitive Psychology, 39, 3-36. Hespos, S. J., & Baillargeon, R. (2002). Infants’ reasoning about occlusion, containment, and support events: Evidence from object-retrieval tasks. Manuscript in preparation. Hespos, S., & Baillargeon, R. (2001 a). Knowledge about containment events in very young infants. Cognition, 78, 207-245. Hespos, S., & Baillargeon, R. (2001 b). Infants’ knowledge about occlusion and containment events: A surprising discrepancy. Psychological Science, 12(2), 141-147. Hespos, S. J., & Spelke, E. S. (2000, July). Conceptual precursors to spatial Zanguage: Categories ofcontainment. Paper presented at the biennial meeting of the International Society on Infant Studies, Brighton, England. Huntley-Fenner, G., & Carey, S. (2001). Infant representations of objects and noncohesive substances. Manuscript submitted for publication. Johnson, S., Slaughter, V, & Carey, S. (1998). Whose gaze will infants follow? The elicitation of gaze following in 12-month-olds. Developmental Science,1, 233-238. Johnson, S., & Naiiez, J. E. (1995). Young infants’ perception of object unity in two-dimensional displays. Infant Behavior and DeveZopment, 18, 133-143. Jusczyk, I? W. (1997). Finding and remembering words: Some beginnings by English-learning infants. Current Directions in Psychological Science, 6, 170-l 74. Jusczyk, I? W., Johnson, S. I?, Spelke, E. S., & Kennedy, L. (1999). Synchronous change and perception of object unity: Evidence from adults and infants. Cognition, 71, 25 7-288. Kellman, I?, & Spelke, E. S. (1983). Perception of partly occluded objects in infancy. Cognitive Psychology, 15, 483-524. Mandler, J. M. (1988). How to build a baby: On the development of an accessible representational system. Cognitive Development, 3, 113-l 36. Mandler, J. M. (1992). How to build a baby: II. Conceptual primitives. Psychological Review, 99, 587-604. Mandler, J. M. (1998). Representation. In W. Damon (Series Ed.), & D. Kuhn & R. Siegler (Vol. Eds.), Handbook ofchild psychology: Vol. 2. Cognition, perception, and language. New York: Wiley. McDonough, L., Choi, S., & Bowerman, M. (1999). Preverbal categorization of spatial relations across WideZyvarying contexts. Poster session presented at the biennial meeting of the Society for Research in Child Development, Albuquerque, NM. Mehler, J., Juxczyk, I? W., Lambertz, G., Halsted, N., Bertoncini, J., & Amiel-Bison, C. (1988). A precursor or language acquisition in young infants. Cognition, 29, 144-178. Meltzoff, A. N. (1995). Understanding the intentions of others: Re-enactment of intended acts by 18-month-old children. Developmental Psychology, 32, 838-850. Quinn, P C., & Eimas, I? D. (1996). Perceptual organization and categorization in young infants. In C. Rovee-Collier & L. I? Lipsitt (Eds.), Advances in infancy research (Vol. 10, pp. l-36). Norwood, NJ: Ablex. Smith, L. B., Jones, S. S., & Landau, B. (1996). Naming in young children: A dumb attentional mechanism? Cognition, 60, 143-I 71.

246

l

13.

Conceptual

Development

in Infancy

Spelke, E. S. (in press). Core knowledge.American Psychologist, 55,1233-1243. Spelke, E. S., Phillips, A. T., & Woodward, A. L. (1995). Infants’ knowledge of object motion and human action. In D. Sperber, D. Premack, & A. Premack (Eds.), Causal cognition: A multidisciplinary debate (pp.44-78). New York: Oxford University Press. Spelke, E. S., & Tsivkin, S. (2001). Initial knowledge and conceptual change. In M. Bowerman & S. Levinson (Eds.), Language acquisition and conceptual development. Cambridge, England: Cambridge University Press. Spelke, E. S., & Van de Walle, G. (1993). Perceiving and reasoning about objects: Insights from infants. In N. Eilan, R. McCarthy, & W. Brewer (Eds.), Spatial representation (pp. 132-l 61). Oxford, England: Basil Blackwell. Tincoff, R., & Jusczyk, I? W. (1999). Some beginnings of word comprehension in 6-month-olds. Psychological Science, IO, 172-l 75. Werker, J. F., & Tees, R. C. (1984). Cross-language speech perception: Evidence for perceptual reorganization during the first year of life. Infant Behavior and Development, 7,49-63. Wilcox, T., Nadel, L., & Rosser, R. (1996). Location memory in healthy pre-term infants. Infant Behavior and Development, 19, 309-323. Woodward, A. (1998). Infants selectively encode the goal object of an actor’s reach. Cognition, 69, l-34. Woodward, A. L., Phillips, A. T., & Spelke, E. S. (1993). Infants’ expectations about the motions of inanimate vs. animate objects. Proceedings of the Cognitive ScienceSociety. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Memories for Emotional, Stressful, and Traumatic Events Nancy L. Stein University of Chicago

T

his chapter presents an essay on the nature, organization, and early emergence of emotional memories. Specifically, I focus on the understanding process that guides and regulates the formation of emotional memories. I describe the mental inferences and evaluations of an event that lead to the experience of emotion, the types of evaluations children make before they experience specific emotions, and the courses of action they choose, once they express an emotion. I focus on the role that preferences, goals, and violations of expectation play in evoking emotion and planning behavior. One of my goals is to be able to specify, in fairly precise terms, the nature and origins of very young children’s skill at thinking about, remembering, and learning about events that evoke emotion. Research carried out from 1980 to 2000 has changed significantly our conception of the infant and toddler’s capability to understand, remember, and respond to events and other people in their world. I argue that from the very beginning, emotional understanding is goal- and preferencebased. That is, when young children experience and express emotion, they do so because they have some ability to recognize and respond to events that indicate a change in a goal that they value. 247

248

l

14.

Memory

For Emotion

Even the youngest children who experience and express what we consider to be the “basic” emotions make inferences about the harms and benefits of the situations that elicit emotion. Very young children also engage in planful action to attain pleasurable states or to eliminate aversive states. Thus, the mental structures that govern and regulate emotional understanding are in place well before children begin to talk (Stein, ‘Rabasso, & Liwag, 2000). In any attempt to describe the development and maintenance of memories about emotion, the preferences and goals that regulate the experience of emotion play a central role. Emotions are important because they signify that a personally significant event has occurred and that a valued goal has changed in status (Stein, l’rabasso, & Liwag, 1993,200O). The goal can change its status in three ways: failure can occur, success can occur, or the certainty of maintaining, attaining, or avoiding a particular goal, can change. For example, when we are angry, we are angry about something that has happened or something that a person has done to block one of our goals. When we are afraid, we focus on something that has happened to us that will prevent us from achieving a goal or avoiding a potentially harmful situation. When we are sad, we focus on something that we have lost that we believe we cannot replace, and when we are happy, we focus on something that we have achieved, in terms of a personally meaningful goal. So when we talk about emotional experience, we are talking about events that have impacted on personally significant goals that often define the core of existence for the person experiencing the emotion. Shortly after birth, infants experience and express emotion continually throughout the day, and they also engage in courses of action in the service of changing, maintaining, or avoiding specific outcomes that directly affect their well-being. During the course of one day, infants and young children experience and may express emotion more than 200 times. This is a very conservative estimate. In both our previous and recent studies of emotional memories and understanding (Levine and Stein, 2001; Liwag and Stein, 1995; Stein & Liwag, 1997), the average number of facial expressions of emotion during a 30-min narration of four different emotion episodes was 68, about 17 different expressions per narrative. When children reported conflicts they had with a best friend or a sibling (Stein & Levine, 1999), the average number of facial expressions per report was 10. In the process of experiencing and expressing emotion, we often learn for the first time what a child or an adult is thinking and feeling. We learn about their preferences, goals, beliefs, plans of actions, histories of traumas, good times, and just about anything that is personally

Stein

.

249

significant to the person and the situation. Thus, the study of emotional memory is significant for reasons in addition to determining the accuracy of a memory representation. Specifically, emotional memories are one of the best windows into the mind of children and adults, in terms of what many would call “theories” of mind or “theories” of the self and other people. The content of emotional memories can also be used to assess a person’s psychological well-being in terms of current and future states of depression, sadness, fear, anger, and other affective states. If children and adults are allowed to express themselves and are asked the right questions, much of the time they are the best predictors of their own behaviors both in current and future situations (Stein & Persun, 2001; Stein, Sheldrick, & Broaders, 1999). Thus, throughout this chapter, I discuss the three dimensions that we have used to study emotion and memory: 1. The complexity and accuracy of recalling emotion-provoking events. 2. The content of the evaluations (appraisals) that children make about themselves and others when they talk about emotional events. 3. The ways in which reports about traumatic and stressful experience can be used to predict depression, positive psychological well-being, the quality of a relationship, future physical health, and the probability of being able to sustain a relationship with another person. In discussing these three topics, I present data from our studies and compare the data to those from other studies, especially in regard to why our results indicate that emotional understanding of both the self and others emerges very early on, much before the age of 3 years. In discussing each issue, I use data from studies on children, adolescents, and adults. Although the prototypic approach in cognitive development is to focus solely on the very young child, or perhaps on the parent-child relationship, we found it essential to compare the very young child to the adult, to assess better the changes that occur in the representation of emotional and traumatic events. Although developmental differences are present in the recall of emotional and stressful events, striking similarities also exist. These similarities tell us as much and often more about the ways in which personally significant events are organized and remembered, especially in regard to the situations in which accuracy versus inaccuracy occurs. Further, it is essential to examine adult memories for the appraisals and theories of mind that they have about others as it is to examine

250

l

14.

Memory

For Emotion

young children’s appraisals and theories of mind. Although many developmentalists believe in a qualitative stage theory of cognitive development (see Astington & Gopnik, 1991; Dunn, 1999; Flavell, Green, & Flavell, 1995), much evidence has been generated in the last 20 years to question seriously the existing stage theories that characterize qualitative differences in children’s thinking (Donaldson, 1992; Mandler, 1983,1998). FACTORS PRECIPITATING

EMOTION

To explore emotional memories, we first begin with a description of those factors that are necessary to provoke an emotion. First, some type of event occurs to induce change to a personally significant goal. Generally, the change is perceived as goal success, failure, or a degree of uncertainty in regard to the goal’s continued success or attainment. The language that children and adults use to signal these changes is quite constrained and highly similar over age. The following are examples of outcomes that have resulted in success, failure, or an uncertain end. These statements have been generated by both children and adults when a change in one of their goals occurs: l l l l l l l l l l

I got my favorite Ninja toys (3-year-old). I wasn’t allowed to sleep over at Joshie’s (3year-old). He was gonna hurt me if I didn’t shut up @%-year-old). My grandpa is gonna die soon (4-year-old). I got to have Nicki come over (3%year old). I knew the end was near, he changed so rapidly (man). My dad told me that Jason had died (man). I found this gift when I got home (woman). She told me that I had passed the exam (woman). He came with me and not Adele (woman).

The second feature of an emotional experience is its involuntary nature. You can’t plan how you will react to an emotion-provoking event, no matter how much you try. Even if you know that something bad is going to happen, when the event finally does occur, it’s never the same as you imagined it. We have been able to document this repeatedly in our studies of men who are caregivers of men who have AIDS (Stein, nabasso, & Albro, 2001; Stein, Folkman, ‘Irabasso, & Richards, 1997). The process of caring for a person with AIDS involves an ever-worsening situation. To date, there is no cure for the disease. Everyone who contracts the disease dies. The only uncertainty about the outcome is how long and when.

Stein

l

251

Given the inevitability of losing their partners, many caregivers try to prepare themselves by going into an early mourning period, especially if their partner fails to recognize or respond to them in a cogent fashion. Indeed, the month before the death of their partner seems to be the worst (Stein & Wong, 2001). When their partner does die, however, many of those who thought they had already experienced the worst, go through an even more severe form of disbelief. The disbelief is due to the fact that the event provoking an emotion contains novel, unexpected, information that violates a person’s beliefs about the world. Examples of caregivers’ expressions of disbelief follow: I just didn’t expect to be this devastated-it’s the finality of the whole thing (sadness). I thought I was prepared until I realized that he was really gone. It’s just never like you think it will be (extreme anxiety).

l

l

Children l l l l

talk about violations

of expectations

in a similar way:

I wasn’t supposed to get this until Christmas (happiness). She promised that I could do this, but I wasn’t allowed (anger). He’s gotten mad before, but not this mad-he was different (fear). I guess you don’t yell at my dad and tell him what to do (fear, sadness).

The important point is that the emotion-provoking event has unexpectedly violated an expectation (belief) about how the world was, is, or should be. And the violation is accompanied by the perception that the ability to attain or maintain a valued goal has changed. RECALL OF EMOTIONAL

MEMORIES

The similarity between children’s and adults’ experience and memory for emotional events occurs because the same type of causal schema is being used by both to understand, evaluate, and react to an event. The ways in which an emotion-provoking event is understood can be understood best by using the pedagogical question scheme outlined in Table 14.1. In all of our studies carried out on real-world emotional memories, children as young as 3 years of age have been able to generate all parts of an emotion episode, especially the components that correspond to each question listed in Table 14.1. Table 14.2 contains a summary of the mean number of clauses for each narrative category that children generated in response to requests to talk about six different types of emotion episodes. The first recall fo-

252

l

14.

Memory

For Emotion TABLE 14.1

A Focus on the Antecedents What How Who What What

and Causes of the Emotion

happened? have my goals been affected? or what caused the change in my goals? are the consequences for the changes in my goals? beliefs have been violated?

A Focus on the Goals and Beliefs What do I want to do about the change? Do I think my original goal can be reinstated (maintained, Is there a goal that can be substituted for my failed goal? A Focus on Action

attained)?

and Consequences

Did I carry out actions

to attain

What were the outcomes? Were any of the outcomes

my goal?

unintended?

Did the outcomes cause a reappraisal of the goal? How will the outcome affect other goals?

cused on recent events in the last week that children remembered as having made them feel happy, sad, afraid, and angry. The second recall focused on events that made the child feel the happiest, saddest, angriest, and most afraid. Their intense emotional responses could have occurred at any time in the past. The one criterion that children had to use was that they couldn’t think of any other event that had made them feel any happier, sadder, angrier, or more afraid. The third type of recall was focused on recent events in the last week where they observed their best friend expressing happiness, sadness, fear, and anger. The fourth recall was focused on events that made their friend feel the happiest, saddest, angriest, and most afraid the child had ever observed. We also collected children’s memories of events nominated by their mother, where the mother recalled events where she had recently observed her child experiencing the same four emotions. Finally, we asked children to recall mother-nominated events that had caused the child to experience the most intense happiness, sadness, anger, and fear. The data from all six types of recall were collapsed because few differences were found for type of emotion-episode recall. Children, no matter what type of memory they reported, almost always included at

Stein TABLE

Children’s

Ability

to Identify

Components

Precipitating

event

Plans Actions Outcomes Appraisals

of success or failure

253

14.2

of Conflict

Appraisals Change in the status of a personally significant goal Beliefs about attainment Beliefs about violation Assignment of blame Emotion Preferences Goals

l

of Self

Episode

for Self and Other

Appraisals

of Others

1.00 1.00

1.00 1.00

1 .oo 0.82 1 .oo 1.00 0.97 1.00

1 .oo 0.94 1 .oo 1 .oo 0.92 1.00

0.78 1.00 0.89 1 .oo

0.72 1 .oo 0.91 1 .oo

least one clause per category listed in Table 14.2. Developmental differences occurred in the number of instances per category that children recalled, not in the kind of categorical information children recalled. The same was true of children’s recall of past emotion episodes, independent of whether they talked about themselves or their best friend (Levine & Stein, 1997, 2001; Wade & Stein, 1995). Three-year-old children could always include at least one instance of each type of internal state category that older children and adults used (e.g., emotion, mood state, physical state, preference, goal, mental state, personal disposition, blame, belief about themselves and others, and plans of action). Table 14.3 contains a narrative that a 3-year-old reported in describing a recent instance where her friend got angry. As the narrative illustrates, the narrator reported many different evaluations of her friend’s internal states. Almost every one of the categories listed in Table 14.2 is included in this child’s narrative. Further, three different episodes were included in this child’s narrative: the first focuses on what made the narrator’s friend mad in the first place; the second focuses on the friend’s strategy for solving her problem; and the third focuses on the strategy that the friend would use if her first strategy failed. The types of narratives, explanations, and causal links that children make among events in their narratives are far more complex and sophisticated than the types of explanations and understandings that

254

.

14.

Memory

A Three-Year-Old

For Emotion

Child’s

TABLE 14.3 Hypothetical Report about a Friend Getting

Experimenter (E): Pretend that Rachel’s really, really mad. What her feel that way? Child: If somebody wouldn’t let her talk. If they said “No! Wait finished!” and it was going to be until, and they said that going to talk until, until clean-up time. E: And that would make her mad, if someone didn’t let her clean-up time-

would

Angry make

‘til I’m they were talk until

Child: MMMMM! (Shakes head vigorously with a yes). And she would go like MMMMM! “You never let me talk! You always talk.” E: Well, what would she be thinking about if someone wasn’t going to let her talk? Child: She would want to punch him, but she wouldn’t. She really wouldn’t. E: She wouldn’t? What would she do? Child: She would tell the teachers because it made her feel bad. E: Because it made her feel bad, huh? And what would telling the teachers accomplish? Child: The teachers would say, “Take turns. You could urn, tell half of it, and Rachel could tell half of what she wants to say. And you could tell half of it, and she could tell half of it. And you could tell half of it, and she could tell half of it.” If they didn’t let her, she could, I think she would tell the teachers that they keeped on doing it. That’s what I think she’d do, that’s what she always do, lah lah lah lah! Just like my every, just like my every, just like every of my friends. E: Oh really7 Child: All of my friends. They tell the teacher. E: All of your friends do thatChild: All of my friends at my school do that-All the girl friends of mine, at least. They tell the teachers! That’s what you’re supposed to do, and they do it. And the teachers get the kids to share.

children express when presented with narratives that are made up by an investigator A good illustration of the differences between self-generated narratives (our studies) and those generated in response to pictures can be found in a recent study by Lagatutta and Wellman (ZOOl), who investigated whether children could infer the “correct” current emotion of a story character if the event occurred in the past versus in the present. Their concern was that children as young as 3 years might be able to infer the “correct” emotional responses of people in the present, but children might have difficulty if they had to make inferences

Stein

l

255

about a current emotion state caused by an event that occurred in the past. These investigators found that 4- and 5-year-old children were able to make inferences about both past and present events with little difficulty. However, older children were able to make more of these inferences than younger children. In contrast, we found few developmental differences in the ability to talk about antecedents of emotions such as fear, whether or not the event occurred in the child’s recent past or in the more distal past. One reason is that young preschool children have very good recollections of events that have caused them harm, and they keep updating their memories, even when the precipitating event has ceased to operate (Levine, Stein, & Liwag, 1999). As an example, we analyzed recall from both a mother and son, who individually recollected an event where the son was prevented from taking a favorite toy on a trip to visit his grandparents. The analysis of the individual recalls from the mother and from the child showed that although the child’s angry temper tantrum upset the mother, and although the mother remembered this event as one that caused her child to become the angriest she had observed in the recent past, the mother recalled little of the child’s feelings, once the event had ended. The child, however, not only recalled his feelings when the event happened in the past, he also recalled how he currently felt, given that this event occurred in the past, and he reported exactly what he was going to do in the future, given that he was not allowed to take his favorite toy on a past visit. Thus, memories of this past event were continually active in this child’s mind, because his goal was never to let the event occur again. Further, it took the child a full week to stop thinking about the event on a consistent basis. One reason that we underestimate young children’s memory capacity is that we think children incapable of monitoring the status of their goals, and we give them little credit for being able to rehearse the exact nature and content of a past situation. Yet, young preschool children engage in rehearsal activities with great frequency (Nelson, 1989), especially when events have meaning and significance for them. Further, when children engage in meaningful rehearsal, memories become fairly resistant to change or influence from an external source, such as a mother trying to change the mind of her child. Although there are situations where mothers can be influential in determining how their children encode and understand an event (Ochs, Taylor, Rudolph, & Smith, 1992), mothers’ success depends on their children’s receptivity to the mother’s influence, the degree to which the child has already constructed a causal representation of the event, and whether the child is in conflict with the mother at the time of the retelling (Levine, Stein, & Liwag, 1999).

256

l

14.

Memory

For Emotion

When children are in direct conflict with their mothers regarding the correct representation of an event, not only do children adhere to their original representations, but they often flatly deny the validity of their mother’s representation. Thus, the robustness of memory for past events, in terms of causal cohesiveness and accuracy, depends on the meaning that was made of the event at the time of its occurrence, the events’ importance to the survival and well-being of the child, and the degree to which an accurate understanding is necessary for the prevention of future harm. A persistent problem with theories of autobiographical memory, especially in regard to the fate of early memories, is the retrospective nature of most studies. To remedy this problem, we carried out a study that focused on emotional memories of young children, where we interviewed children first when they were between the ages of 3 to 6 years. We then went back and interviewed a small portion (N = 29) of these children 10 years later. In doing so, we explored children’s memories for two types of emotional events- those they recalled at the first visit, as having occurred in the past week before the interview, and those they recalled at the first visit, as being events where they were the happiest, saddest, angriest, and most afraid. When we probed children for their memories 10 years later, we asked them a two-probe question where the second probe increased the amount of information contained in the cue to elicit children’s memories. The first cue focused on the emotion that was originally given to the children to elicit the narratives. The second probe contained the actual event that the child nominated as having caused them to experience each emotion. Table 14.4 contains the proportion of children who could recall each narrative, given the emotion cue and then the emotion plus event cue, if they could not recall the event in response to the emotion probe. The proportions in the second table are additive with respect to the proportions in the first table. That is, the second table contains all of the children who recalled the emotion event, either in response to an emotion probe or in response to an emotion probe plus an event probe. The results are unambiguous in regard to memorability. Those events that were recalled as part of the recent past originally were not nearly as well remembered as those events that represented the most intense experience of an emotion. Furthermore, even when children were able to recall accurately at least 50% of the clauses that were contained in their original narratives, the proportion of new clauses that was not in the original narratives was significantly higher in the narratives for recent events when compared to those for the most intense experience of emotion. In addition, significantly more new information was inconsistent with the original narratives for recent events as opposed to those most intensely experienced.

Stein

Prooortion A. Emotion Tvoe

of Children

TABLE 14.4 Recalling Emotion

Probe Recent Emotional

Events

.17 (5)

Fear

.I7 (5) .31 (9)“”

.72 (21) .76 (22)

B. Emotion Tvoe

.06 (2)

Most Intense Emotional

Sad

Anger

257

Events 10 Years later

.58 (17)*’ .76 (22)

Happy

l

Probe Plus Emotion

Event

and Event Probe

Recent Emotional

Events

Most Intense

Emotional

Happy

.31 (9)

.89 (20)

Sad Anger Fear

.38 (11) .45 (13) .48 (14)

.83 (24) .76 (22) .90 (26)

Event

Tables 14.5 and 14.6 contain a description of the prototypic events that were nominated for recent emotions and the most intense emotions. The important finding in these data is that different types of events served as the prototype for recent events versus those that caused the most intense expression of emotion. For all four emotions, the themes vary and recent events differ when compared with themes from events that caused the most intense emotion. For example, recent memories of anger were reportedly caused by the child being punished or being forced to do something. In memories for the most intense experience of anger, the two most frequent events were those where children were lied to or deceived or where children were physically harmed. Similarly, the prototypic events for sadness were different for recent versus the most intense emotional experience of sadness. The events nominated for happiness and sadness showed more overlap in the types of events over the two situations. The focal events nominated, however, differed significantly across the two conditions, especially for sadness. The events nominated as causing the most intense expression of sadness were similar to traumatic events. The events included a loss or the inability to maintain a central goal that had primary meaning in the child’s life. The event caused irrevocable change such that the probability of attaining many important goals

258

l

14.

Memory

For Emotion TABLE

Events Nominated

14.5

for Most Recent Versus Most Memorable Most Memorable

Exoerience Recent

(N = 29

(A’ = 29)

0 44 28 0

21 0 0 15

40 20

10 20

0 0

13 13

66 33 0

40 0 20

0

13

Happy Child engages in or masters desirable activities

0

21

Child is able to avoid rejection Child experiences relationship Child gets desirable objects

31 44 20

0 0 44

AWY

Child is punished Child is physically harmed by another Children is lied to or deceived

Child is forced to do something Sad Child is separated

from significant

others

Child is punished Child is denied desirable objects Child is unable to engage in desirable Afraid Child perceives a threat

activity

to self or others

Child perceives a threat to valued relationship Child’s emotions are elicited by real or imaginary animate beings Child has nightmares or bad dreams

or harm attainment

was permanently altered. Finally, the event often resulted in failure to construct a viable coping strategy or plan of action for an extended period of time. In expressing emotion on an everyday basis, such as those in recent events, although the emotional reactions are often volatile and intense, attention was more focused on solutions that could be generated to solve the problem or cope with the emotional reaction. Traumatic events, however, such as those nominated as causing the most intense emotional experience, do not have a quick solution. They often require the complete abandonment of a system of meaningful goals. Rather

Stein TABLE

Examples

Happy Sad

l

259

14.6

of Emotional

Events

Recent

Most Memorable

Got Ninja toy

Dad didn’t

Went to shedd Child can’t go to friend’s

Dad didn’t kidnap child anymore Dad was dying from leukemia Parents divorcing, dad won’t take child Brother tried to drown child Parent lied to child about going on a favorite trip

Child didn’t

get favorite

house toy

Anger

Child was sent to room Child had to clean up room

Fear

Monster in room Get hurt in the dark

Grandpa

die

would

die

Would have to stay in the hospital forever (diabetes)

than one goal being jeopardized or blocked, a whole system of related goals is threatened. When many goals are affected by an event, it is harder to generate a plan to solve the problem. In these circumstances, people begin to consciously question the validity of the beliefs they have held about the world, and this questioning leads them to actively express and examine the specific violations that affected their beliefs. These extended thinking processes then allow the memories to be even more tightly bound, in regard to both coherence and uniqueness. EMOTIONS, GOAL-APPRAISALS, AND PSYCHOLOGICAL WELL-BEING The types of coping strategies reported in emotional memories, as well as the types and frequency of emotions themselves, turn out to be sensitive indicators of current and future psychological well-being, as measured by depressive mood, positive morale, and measures of self-esteem. Table 14.7 shows the relation between the proportion of positive emotion state words that occurred during the narration of stressful or traumatic events and the level of depression expressed by the narrator. In all three sets of studies, the data indicate that the greater the proportion of reference to positive emotions, the less depressed the narrator. In Table 14.8, four different types of appraisals made about emotional, stressful, or traumatic experience are presented. All four types

260

l

14.

Memory

For

Emotion TABLE

14.7

Indicators of Well-Being Across the life Span Proportion Expressed When Talking About a Negative, Stressful, Proportion of Positive Emotions Correlated with

Children

Depressive mood Positive morale or Self-Esteem Note.

All correlations

Adolescents

-.35 +.34

are significant

of Positive Emotions or Traumatic Event

Adults

-.38 +.32

-.62 +.68

at the p < .Ol level.

TABLE 14.8 Appraisals Made About Emotional, Stressful, or Traumatic Experience Relation to Depressive Mood (Numbers Expressed as Percentages) High Depressed Appraisals Generated new goals that could be substituted for failed goals Focused on goals that cannot be attained Expressed beliefs about lack of ability to cope with the results of loss or aversive states Focused on positive things that were learned as a function of stress and trauma

Children

in

low Depressed

Adolescents

Adults

Children

Adolescents

Adults

10

23

22

45””

58””

67””

53””

45**

67**

23

29

20

-

32””

43””

-

06

05

-

10

05

-

35””

48””

of appraisals are significantly related to depressive mood and positive morale. The types of appraisals and beliefs that lead to depressive mood and positive morale in adults are similar to those that lead to depressive mood and positive morale in children. The major difference across the life span is not in the appraisals of a goal or the emotion expressed. Rather, children and adults often differ on which goals they consider to be personally meaningful and important. That is, the themes in chil-

Stein

l

261

dren’s lives are different because the goals they pursue are often different from those of adults. However, when a goal is deemed valuable, children and adults make the same types of appraisals about gains and losses. Furthermore, the coping strategies that result in depression are similar over development as are the coping strategies that result in positive well-being. Table 14.8 shows the percentages of participants who generated new goals in reaction to having had their goals obstructed in a permanent manner. The percentage of participants generating new goals was significantly related to whether participants reported high or low degrees of depression, as measured on the Bradburn Affective Balance Scale. The scale was modified for use with young children (Stein & Albro, 1997a,b), but essentially the same items that appear on the adult scale appear on the children’s scale. The wording on the children’s scale corresponds to children’s natural use of emotion terms in everyday interaction (Stein & Levine, 1999). As you can see, the tendency to generate new goals in substitution for goals that have failed is more probable for those participants who are not highly depressed. Table 14.8 also shows that significantly more participants who are high on depressive mood focus more on lost goals than participants who are low on depressive mood. By focusing on lost goals, we mean that participants actively fantasize about reinstating the lost goal. We do not mean that participants are ruminating about past times that were good and meaningful. Although deeply valued goals are gone forever, many individuals hang on to these goals mentally and often have active fantasy sessions that focus on these lost goals. In addition, Table 14.8 demonstrates that for adolescents and adults, the expression of beliefs about the inability to cope with the results of loss or the inability to muster the personal resources to contend with the devastation of a trauma are significantly related to the level of depression expressed by participants. Finally, Table 14.8 shows that a focus on positive outcomes associated with trauma is associated with a lower state of depression, Participants who scored lower on measures of depressive mood were also the ones who mentioned that their stressful or emotional experience, although negative in many respects, had positive benefits and outcomes. The belief that something valuable was learned during a traumatic or stressful event was significantly related to lower states of depression.

CONCLUSIONS Our emotional and stressful narrative data on children and adults strongly support the hypothesis that emotional understanding and

262

l

14.

Memory

For Emotion

memory of everyday situations is a rapidly developing skill in early childhood. By 3 years of age, children have a rich array of knowledge about the causes and consequences of basic emotions. The types of strategies children use to cope with emotion are highly similar to the strategies adults use to cope with their emotions, and memories for those events that encode a high degree of meaning are retained with a good degree of accuracy over time. A strong explanation for the similarities in memory, thinking, and coping across the life span is that knowledge about basic emotion is highly constrained to a small number of operating conditions and a small number of coping strategies. Emotional responses occur in reaction to unexpected changes in valued goal states, and even the very young child can monitor whether his or her favored goals are being attained or obstructed. The strategies used to cope with changes in a goal’s status are also limited to a small number of options. When people fail to attain or maintain goals of value, a decision is first made about the ability to reinstate the goal in question. If people believe that the goal can be reinstated, they activate plans that allow them to proceed with reinstatement. If a valued goal cannot be reinstated, three types of strategies can be activated. The first is to abandon the goal and substitute nothing; the second is to abandon the failed goal and to substitute a new goal for the failed one; the third is to mentally reinstate the lost goal and to fantasize about the conditions that would allow reinstatement, although the reinstatement cannot occur in reality (Klinger, 1977, 1987, 1996). The fact that emotional understanding relies on a very limited set of conditions affords young children rapid access to this knowledge. The emotional understanding that children and adults express is then intimately related to decisions about reinstating or substituting new goals. These reinstatement decisions then regulate the experience of both positive and negative mood states (Stein & ?frabasso, 1989). Although adults frequently have more knowledge than children about obstacles to goal attainment and ways of overcoming obstacles, the experience of goal failure and success results in the same types of appraisals and decision processes concerning coping strategies. An important issue for further research will be to determine the stability of the appraisal and coping strategies activated in emotional situations. A critical issue is whether temperamental differences, rather than development differences, regulate the types of appraisals and choice of coping strategies in adverse circumstances. A second issue is whether participants can learn to change their appraisals and coping strategies. What has become apparent to us, however, in analyzing the narratives of everyday emotional experience, is an extremely productive strat-

Stein

l

263

egy for describing beliefs, desires, emotions, and plans of people engaged in an ongoing emotion situation. The identification of subjective mental states such as beliefs, desires, emotions, and goals are critical because these states not only predict strategic action that is carried out, but they also predict the type of mood state that will be experienced as a result of the emotional experience. ACKNOWLEDGMENTS This research was funded by grants from the National Institute of Child Health and Human Development, Grant No. HD38895 to Nancy L. Stein, HD 25742 to Tom Trabasso and Nancy L. Stein, by a Provost’s Grant to Nancy Stein and Tom Trabasso, by a Social Sciences Grant to Nancy Stein and Tom Irabasso, and by a Smart Grant to Nancy L. Stein. I would like to thank Linda Levine for her collaboration, Tom Trabasso for his conceptual and insightful editorial comments, and Chrystyna Kouros for her help and willingness to compare existing theories of emotion and memory.

Astington, J., & Gopnik, A. (1991). Theoretical explanations of children’s under-

standing of the mind. British Journal of Developmental Psychology, 9, 7-32. Donaldson, M. (1992). Human minds. New York: Allen Lane-Penguin Press. Dunn, J. (1999). Relationships and children’s understanding of mind. Oxford, England:Blackwell Publishers. Flavell, J. H., Green, F. L., & Flavell, E. R. Young children’s knowledge about thinking. Monographs ofthe Societyfor Research on Child Development, 60(l), Serial No. 243. Klinger, E. (1977). Meaning and void: Inner experience and the incentives in people’s lives. Minneapolis: University of Minnesota Press. Klinger, E. (1987). Current concerns and disengagement from incentives. In F. Halisch & J. Kuhl (Eds.), Motivation, intention, and volition (pp. 337-347). Berlin: Springer-Verlag. Klinger, E. (1996). Emotional influences on cognitive processing with implications of both. In l? M. Gollwitzer & J. A. Bargh (Eds.), The PsychoZogyofAction. (pp. 168-192). New York: The Guilford Press. Lagatutta, K. H., & Wellman, H. M. (2001). Thinking about the past: Early knowledge about links between prior experience, thinking, and emotion. Child Development. 72(l), 82-l 02. Levine, L., & Stein, N. L. (1997). “I’m not his brain”: The emergence of young children’s understanding of themselves and others. Paper presented at SRCD, Washington, DC, April. Levine, L., & Stein, N. L. (2001). Young children’s understanding of emotions in themselves and others. Paper presented at the Cognitive Development Society, Virginia Beach, VA, October.

264

.

14.

Memory

For Emotion

Levine, L., Stein, N. L., & Liwag, M. (1999). Remembering children’s emotions: Sources of concordant and discordant accounts between parents and children. DeveZopmentaZ Psychology. 5( 3), 21 O-230. Liwag, M. D., & Stein, N. L. (1995). Children’s memory for emotional events: The Importance of emotion enactment cues. Journal of Experimental ChiZd Psychology, 60, 2-3 1. Mandler, J. M. (1993) Representation. In J. H. Flavell & E. M. Markman (Eds.), Handbook of Child Psychology: Volume 3, Cognitive Development. (4th ed., pp. 420-494). New York: Wiley. Mandler, J. M. (1998). Representation. In D. Kuhn & R. Siegler (Eds.), Cognition, perception, and language, Vol. 2 of W. Damon (Series Ed.), Handbook of child psychology (pp 255-308). New York: Wiley. Nelson, K, Ed. (1989). Narrativesfrom the crib. Cambridge, MA: Harvard University Press. Ochs, E., Taylor, C., Randolph, D., & Smith, R. (1992). Storytelling as theory building activity. Discourse Processes,75( 1), 3 7-72. Stein, N. L., & Albro, E. R. (1997a) The emergence of narrative understanding: Evidence for rapid learning in personally relevant contexts. Contemporarylssues in Education, 60, 83-98. Stein, N. L., & Albro, E. R. (1997b). Building complexity and coherence: Children’s use of goal-structured knowledge in telling good stories. In M. Bamberg (Ed.), Narrative Development: Five approaches (pp. 5-70). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Stein, N. L., & Albro, E. R. (April, 1997~). Children’s understanding of conflict: Evidence from past memories and on-line negotiated resolutions. Paper presented at the Society for Research on Child Development, Washington, DC., April. Stein, N. L., Folkman, S., l’rabasso, T., & Richards, T. A. (1997). Appraisal and goal processes as predictors of psychological well-being in bereaved caregivers. Journal of Personality and Social Psychology, 72(4), 872-884. Stein, N. L., & Liwag, M. D. (1997). A goal-appraisal process approach to understanding and remembering emotional events. In l? van den Broek, I? Bauer, & T. Bourg (Eds.), Developmental spans in event comprehension and representation (pp. 199-236). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Stein, N. L., & Levine, L. J. (1999). The early emergence of emotional understanding and appraisal. Implications for theories of development. In T. Dalgleish & M. Power (Eds.), Handbook of Cognition and Emotion (pp. 383-410). Chicester: Wiley. Stein, N. L., & Persun, N. I. (2001). The longevity of romantic couples relationships. Unpublished paper. University of Chicago. Stein, N. L., Sheldrick, R., & Broaders, S. (1999). Predicting psychological well-being from beliefs and goal appraisal processes during the experience of emotional events. In S. Goldman, P L. Van den Broek, &A. Graesser (Eds.), Essays in Honor of Tom ?+abasso (pp. 279-302). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Stein, N. L., & ‘Irabasso, T (1989). Children’s understanding of changing emotion states. In C. Saarni & I? Harris (Eds.), Thedevelopment of emotional understanding (pp. 50-77). New York: Cambridge University Press. Stein, N. L., Trabasso, T, & Albro, E. R. (2001). Understanding and organizing emotional experience: Autobiographical accounts of traumatic events. Empirical Studies of the Arts, I9( 1), 11 l-l 30.

Stein

l

265

Stein, N. L., Trabasso, T., & Liwag, M. (1993). The representation and organization of emotional experience. Unfolding the emotional episode. In M. Lewis and J. Haviland (Eds.), Handbook of Emotion (pp. 279-300). New York: Guilford Publications, Inc. Stein, N. L., Trabasso, T., & Liwag, M. D. (2000). A goal appraisal theory of emotional understanding: Implications for development and learning. In M. Lewis & J. Haviland-Jones (Eds.), Handbook of Emotion. 2nd Edition (pp. 436-457). New York: Guilford University Press. Stein, N. L., & Wong, A. (2001). The onset of grieving and recovery from depression. Unpublished paper, University of Chicago. Wade, E., & Stein, N. L. (1995, April). Preschool children’s narrations about real and hypothetical emotional events. Paper presented at the Society for Research in Child Development, Indianapolis, IN.

This Page Intentionally Left Blank

Author Index -4%

A Abbate, M. S., 124, 139 Abelson, A., 54, 74 Ackil, J. K., 66, 71 Aguiar, A., 238, 244 Ahn, W., 49, 51 Albro, E. R., 250, 261, 264 Alvarez, T. D., 92, 100 Amiel-Tison, C., 228, 245 Andersen, E. S., 172, 180 Anderson, A. L. N., 172, 180 Anderson, J. R., 187, 188, 189, 195, 197 Anglin, J., 119,122,139,171,178,180 Anker, S., 96, 99 Antell, S. E., 202, 220 Anthony, S. H., 10, 11, 15 Appelbaum, L. G., 92, 100 Aronson, L., 64, 73 Astington, J., 250, 263 Atkinson, J. A., 96, 99 Au, T. K., 136, 139 Austin, G. A., 3, 13 Austin, N. G., 10, 11, 15

B Bachevalier, J., 32, 33, 39 Backscheider, A. G., 121, 139 Bahrick, L., 65, 72 Baillargeon, R., 201, 221, 224, 224n, 225,226,228,231, 237,238,244,245 Baker-Ward, L., 67, 70, 72

Bangston, S. K., 33, 40 Baron-Cohen, S., 91, 92, 99 Barr, R., 21, 34, 39, 146, 153, 159 Barrett, M. D., 120, 139 Barsalou, L. W., 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 161, 173, 176, 180 Barwise, J., 2, 13 Bates, E., 93, 100, 124, 126, 140 Batterman-Faunce, J. M., 63, 72, 154,159 Bauer, F? J., 19, 20n, 21, 22, 23, 24, 25, 27, 30, 32, 33, 34, 36, 37, 39, 40, 42, 55, 56, 57, 58, 66, 71, 117, 118, 131, 133, 139, 141, 146, 147, 149, 150, 159, 167, 175, 180,182 Bayley, N., 94, 99 Behl-Chada, G., 118, 140, 167, 180 Behrend, D. A., 122, 139 Bellugi, U., 91, 92, 93, 95, 99, 100, 101 Bennets, L. A., 97, 99 Benveniste, S., 131, 143 Berardi, B., 190, 198 Berlin, B., 199, 220 Bernstein, N., 110, 113 Bertoncini, J., 228, 245 Bertrand, J., 93, 96, 100 Bickerton, D., 92, 99 Biederman, I., 2, 13 Bierwisch, M., 163, 180 Bihrle, A., 95, 99, 100 Bills, A. J., 2, 14

267

268

.

Author

Index

Biro, S., 45, 50, 51 Bisanz, J., 54, 72 Bittinger, K., 21, 39 Blackburn, B., 96, 100 Blair, E., 176, 181 Bloom, L., 120, 122, 140, 170, 180 Bloom, I?, 174, 180 Bobrow, D. G., 54, 71 Bomba, ,, 201 Booth, A., 168, 180 Bornstein, M. H., 161, 180 Bottoms, B., 53, 72 Boutelle, J., 117, 140 Bower, G. H., 189, 197 Bowerman, M., 121, 138, 140, 141, 170, 180, 228, 229, 230, 231,244,245 Boyes-Braem, F?,116, 119, 142, 162, 164, 183 Braddick, F., 96, 99 Braddick, 0. J., 96, 99 Bransford, J. D., 2, 13, 77, 82, 86 Breinlinger, K., 49, 52, 201, 221 Brewer, W. F., 19, 40 Brickson, M., 32, 39 Brizzolara, D., 92, 101 Broaders, S., 249, 264 Brooks, L. R., 8, 13 Brown, A., 77, 82, 86 Brown, J. H., 93, 94, 95, 96, 97, 99, 100 Brown, R. W., 119, 140 Brown, T., 65, 72 Brown, V., 203, 220 Bruck, M., 53, 71 Brugman, C., 200, 220 Bruner, J. S., 3, 13, 55, 71 Bucks, R. S., 11, 13 Butterworth, G. E., 168, 169, 183 Byrne, R. W., 108, 114

C Canada, K., 123, 141 Canning, J., 69, 73 Capirci, O., 92, 101 Carey, S., 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 82, 84, 86, 120, 136, 140,236,242,245 Carlesimo, G., 92, 101

Carlson, L., 201, 204, 206, 207, 211,215,218,220,221 Carlson-Radvansky, L., 210, 220 Caron, A. J., 202, 220 Carver, L. J., 32, 33, 34, 40 Casad, E., 201, 220 Casse-Perrot, C., 96, 99 Cauley, K. M., 94, 99 Ceci, S. J., 53, 71 Cheng, P W., 2, 13 Chiang, W.-C., 236, 244 Choi, S., 138, 140, 141, 228, 229, 230,231,244,245 Chomsky, N., 218,220 Christianson, S. A., 62, 71 Clark, A., 2, 13 Clark, E. V, 120, 121, 122, 124, 133, 140,163,180 Clark, H., 200, 220 Clark, H. H., 2, 13 Clubb, I? A., 56, 57, 72, 146, 159 Cocchi, R., 97, 99 Cocking, R., 77, 82, 86 Cohen, L. B., 119, 143, 167, 168, 169,181,182 Collins, A. M., 189, 197 Comrie, B., 218, 220 Cosmides, L., 109, 113 Covey, E., 218, 220 Cramer, L., 176, 181 Csibra, G., 45, 50, 51

D Dale, F?S., 124, 126, 140 Daly, B. I?, 23, 40 De Haan, M., 21, 40, 42, 92, 95, 99 De Schonen, S., 96, 99 Dean, J., 69, 73 Demsey, S., 96, 100 Dennis, J., 97, 99 Dennis, M. J., 49, 51 Deruelle, C., 96, 99 Didow, S. M., 69, 72 Diesendruck, G., 4, 9, 13 Dilts, C., 96, 100 Dion, S. A., 57, 74, 146, 147, 160 Dodd, D. H., 172, 184 Doherty, S., 93, 99, 101 Donaldson. M.. 250. 263

Dow, G. A., 21, 39, 118, 139 Dowden, A., 21, 34, 39 Downing, C., 203, 220 Dromi, E., 131, 140 Dropik, I?, 21, 23, 24, 25, 30, 33, 36, 37, 40, 66, 71 Duke, N., 176, 178, 181 Dunisch, D. L., 66, 71 Dunn, J., 250, 263 Dunn, L. M., 93, 99 Durgin, F., 76, 77, 82, 86

E Easterbrook, J. A., 62, 71 Eckerman, C. O., 69, 72 Eimas, I? D., 117, 118, 119, 140, 142, 167, 169, 181, 235, 245 Eriksson, M., 177, 181

F Fadiga, L., 112, 114 Farrar, M. J., 59, 71 Favuto, M., 97, 99 Fein, G. G., 110, 113 Fenson, L., 124, 126, 140 Finegan, J., 97, 99 Fivush, R., 19,20,21, 23,40,41, 54, 55, 56, 57, 58, 59, 60, 61, 63, 65, 66, 70, 71, 72, 73, 74, 118, 138, 139, 140, 146, 147, 154, 155, 159, 160 Flannagan, D., 70, 72 Flavell, E. R., 250, 263 Flavell, J. H., 250, 263 Fodor, J. A., 163, 181 Fogassin, L., 112, 114 Folkman, S., 250, 264 Frank, I., 134, 135, 142 Frankenfield, A., 176, 181 Freud, S., 18, 41 Freyd, J., 70, 72 Fromhoff, EA., 20,41,61, 71,154,159

G Gallese,

V., 112, 114

Gapp, K.-R, 214, 215, 216n, 220 Gathercole, V., 176, 181 Gelber, E. R., 167, 181 Gelman, R., 76, 77, 78, 79, 80, 81, 81n, 82, 86, 87, 94, 100, 136, 142, 175, 184 Gelman, S. A., 4, 9, 13, 121, 123, 142, 176, 181 Gennari, S., 8, 14 Gentner, D., 8, 13, 116, 136, 140, 172, 181 Georgieff, M. K., 21, 40 Georgopoulos, A. I?, 203, 220 Gergeley, G., 45, 50, 51 German, T. I?, 176, 181 Gick, M. L., 2, 13 Gilmore, R., 96, 97, 99 Glenberg, A. M., 2, 5, 14 Glenn, C. G., 54, 74 Goldberg, A. E., 2, 14, 65, 72 Goldberg, N., 187, 193, 194, 195, 197 Goldfield, B. A., 122, 128, 142 Goldstone, R., 5, 14 Golinkoff, R. M., 94, 99, 121, 133, 140 Goodman, G., 53, 59, 63, 67, 72, 154,159 Goodman, N., 116, 140 Goodnow, J. J., 3, 13 Gopnik, A., 44, 51, 250, 263 Gordon, L., 94, 99 Gordon, R., 146, 147, 160 Gormican, S., 203, 221 Gottesman, C. V., 2, 14 Graesser, A. C., 59, 72 Graham, L., 70, 72 Graham, S. A., 120, 134, 135, 142 Gray, J. T, 20, 41, 61, 71 Gray, W. D., 116, 119, 142, 162, 164, 183 Green, E L., 250, 263 Green, J. M., 97, 99, 158, 160 Greenberg, J. H., 218, 220 Greeno, J. G., 2, 14 Greif, M., 76, 86 Groen, G. J., 194, 197 Gruendel, J., 19, 23, 41, 54, 73 Gsodl, M. K., 93, 94, 95, 96, 97, 99, 100 Gumperz, J. J., 200, 220

270

l

Author

index

H Haden,

C. A., 66, 69, 71, 72, 74, 139,140,154,160 Hagger, C., 32, 39 Haith, M. M., 103, 106, 113 Hall, D. G., 131, 142 Halsted, N., 228, 245 Hamond, N. R., 23, 41, 61, 72 Hampton, J. A., 3, 14 Happe, F., 111, 113 Hatch, T., 142 Hayne, H., 21, 34, 39, 146, 153, 159 Hazzard, A., 65, 72 Hemenway, K., 169, 184 Hepps, D., 67, 72 Herbert, J., 153, 159 Hermer-Vasquez, L., 243, 245 Hertsgaard, L. A., 23,24,40, 150, 159 Hespos, S. J., 224n, 225, 226, 228, 231,237,238,245 Hesselink, J. R., 93, 99, 101 Heuer, F., 62, 72 Hinton, G. E., 4, 15 Hirschman, J. E., 67, 72 Hirsh-Pasek, K., 94, 99, 121, 133, 140 Hobart, C. J., 147, 160 Hoffman, C., 200, 220 Holcom, F?J., 92, 100 Holyoak, K. J., 2, 8, 13, 14 Hornsby, J. R., 163, 164, 183 Howard, A. N., 67, 72 Hudson, J. A., 19, 22, 23, 41, 54, 57, 59, 61, 72, 73 Huntley-Fenner, G., 236, 245 Hutchinson, J., 131, 141, 175, 182 Huttenlocher, J., 4, 13, 121, 131, 140, 142 I Imai, M., 116, 136, 140 Inhelder, B., 12, 14, 166, 181 Intraub, H., 2, 14 I Jacobson, K., 49, 52, 201, 221 Jansen op de Haar, M., 176, 181 Jernigan, T. L., 91, 92, 93, 95, 99

Johnson,

D. M., 116, 118, 119, 142, 162,164, 169,183 Johnson, D. R., 200,220 Johnson, E. C., 173, 174, 182 Johnson, K., 33, 40 Johnson, K. E., 176, 182 Johnson, M. H., 33, 41, 92, 93, 94, 95, 96, 97, 99, 100, 142, 183 Johnson, M. K., 2, 13 Johnson, N. S., 14 Johnson, S., 236, 245 Johnson, S. C., 45, 50, 51, 242, 245 Johnson, S. P, 236,245 Johnson-Laird, P N., 2, 14, 48, 52, 165, 173, 178, 182 Jones, K., 176, 178, 181 Jones, S. S., 121, 138, 141, 161, 175, 176, 177, 181, 182, 183,223,245 Jones, W., 93, 95, 100 Jusczyk, F?W., 228, 234, 236, 245, 246

K Kail, R., 54, 72 Kaplan, B., 179, 184 Karmiloff-Smith, A. D., 92, 93, 94, 95, 96, 97, 98, 99, 100, 118,142 Katsnelson, A. S., 243, 245 Kaufman, L., 76, 77, 82, 86 Kay, I?, 199, 200, 220 Kee, D., 195, 197 Keil, F. C., 44, 45, 49, 51, 76, 86, 137, 140, 161, 174, 181 Kelemen, D., 176, 181 Kellman, P, 236, 245 Kemler Nelson, D. G., 176, 178, 181 Kemler, D. G., 116, 140 Kennedy, L., 236, 245 Kettner, R. E., 203, 220 Killen, M., 118, 119, 141 Kim, J., 203,221 Kim, N. S., 49, 51 Kim, S. N., 76, 86 King, J., 96, 99 Kintsch, W., 11, 15, 187, 197 Klahr, D., 188, 197

Author

Klein,

B. F?, 92, 93, 100, 134, 135, 142 Klima, E. S., 95, 100 Klinger, E., 262, 263 Kowalski, D. J., 59, 72 Krauss, R., 164, 181 Kripke, S., 49, 51 Kroupina, M. G., 36, 37, 40 Kuczaj, S., 122, 123, 141 Kuebli, J., 56, 57, 58, 72, 146, 159 Kuhn, J., 63, 72, 154, 159 Kyratzis, A., 12, 14, 178, 182 1 LaBerge, D., 203, 220 Labov, W., 171, 172, 181 LaChapelle, N. B., 124, 139 Lagatutta, K. H., 254, 263 Lakoff, G., 8, 14,201,220 Lamberts, K., 4, 13, 14 Lambertz, G., 228, 245 Landau, B., 121, 138, 141, 176, 182,223,245 Langacker, R., 201, 220 Langley, I!, 188, 197 Lassaline, M. E., 49, 51 Lau, I., 200, 220 Lawson, K. R., 97, 100 Lazar, M. A., 167, 181 Lebiere, C., 187, 188, 197 Leonard, C., 96, 100 Lesgold, A. M., 190, 198 Leslie, A. M., 45, 52, 91, 92, 100, 104, 107, 109, 110, 111, 113 Levine, L., 248, 253, 255, 261, 263, 264 Levinson, S. C., 200, 201, 220, 221 Lin, E. L., 12, 14 Livet, M. O., 96, 99 Liwag, M., 248, 255, 264, 265 Logan, G., 202,210,214,215,216, 216n, 220,221 Logan, G. D., 193, 197 Larch, R. E Jr., 147, 160 Lucariello, J., 12, 14, 54, 73, 118, 141,177,178,182 Lucy, J., 200, 221 Luka. B. J.. 2. 9. 13

Index

l

271

M Macnamara, J., 121, 141, 163, 182 Macomber, J., 49, 52, 201, 221 Madole, K. L., 167, 168, 169, 182 Malmquist, C. l?, 64, 73 Malt, B. C., 8, 14, 173, 174, 182 Mancini, J., 96, 99 Mandler, J. M., 1, 2, 5, 14, 18, 21, 22, 23, 24, 40, 41, 43, 48, 49, 50, 51, 52, 54, 73, 76, 86, 98, 100, 117, 119, 131, 133, 137, 138, 139, 140, 141, 146, 147, 150, 159, 160, 162, 167, 168, 169, 170, 175, 176, 177, 180, 182, 185, 186, 187, 189, 190, 191, 196, 197, 197, 198, 200, 201, 221, 223, 225, 229, 235, 236, 243, 245,250,264 Markman, A. B., 8, 13 Markman, E. M., 12, 14, 121, 124, 131, 133, 136, 139, 141, 175, 182 Markow, D. B., 136, 137, 142 Martin, N., 97, 101 Martinez, F?, 65, 74 Massey, C. M., 78, 79, 86 Matan, A., 49, 52 Maurer, D., 92, 95, 99 McCabe, A., 66, 73 McCann, I. L., 62, 73 McClelland, J. L., 4, 15, 187, 190, 198 McCloskey, M., 77, 86 McCune-Nicolich, L., 110, 113 McDaniel, C. K., 200, 220 McDermott, J., 73 McDonough, L., 21, 23, 41, 117, 119, 137, 138, 140, 141, 162, 167, 169, 170, 182, 230,231,245 McKee, R. D., 21, 41 Meek, E., 76, 86 Medin, D. L., 4, 15, 45, 49, 52, 137, 141, 165, 166, 173, 174, 176, 181, 182, 183 Mehler, J., 228, 245 Meltzoff, A. N., 20, 21, 41, 44, 51, 242,245

272

l

Author

Index

Merdin, D. L., 3, 15 Merriman, W. E., 172, 175, 182 Mervis, C. A., 121, 140, 141 Mervis, C. B., 3, 15, 92, 93, 96, 100, 116, 119, 120, 121, 123, 128, 133, 138, 140, 141, 142, 162, 164, 176, 182, 183 Meschino, W., 97, 99 Michotte, ,, 45 Miller, G., 48, 52, 165, 173, 178, 182 Mills, D. L., 92, 100 Mischel, W., 2, 15 Mishkin, M., 32, 33, 39, 41 Mix, K. S., 2, 9, 13 Morris, C., 176, 181 Morris, C. A., 93, 96, 100 Murachver, T, 69, 73, 146, 147, 155, 158,159, 160 Murphy, G. L., 3, 12, 14, 15, 165, 173, 182 Murray, E. A., 32, 41

N Nadasdy, Z., 45, 50, 51 Nadel, L., 238, 246 Nader, K., 64, 73 Naigles, L. G., 121, 123, 142 Ndiiez, J. E., 236, 245 Naus, M. J., 155, 160 Neches, R., 188, 197 Needham, A., 201,221 Nelson, C. A., 21, 32, 33, 34, 40, 41 Nelson, K., 12, 14, 15, 19, 22, 23, 41, 54, 61, 66, 69, 72, 73, 74, 118, 138, 139, 141, 142, 155, 160, 161, 164, 165, 166, 167, 168, 170, 171, 175, 177, 178, 182, 183,255,264 Nelson, K. E., 171, 183 Neville, H. J., 92, 100 Newell, A., 187, 188, 196, 197 Nokes, L., 96, 99 Norman, D.A., 54, 71, 189, 197 Nosofsky, R. M., 4, 8, 15

0 Oakes, L. M., 167, 168, 182 Ochs, E., 255, 264 Olseth, K. L., 2, 9, 13 Olver, R. R., 163-164, 183 Ornstein, I? A., 63, 68, 69, 72, 73, 155,160 Ortony, A., 45, 49, 52, 173, 174, 182 Osborne, H. L., 67, 72 Owens, J. L., 146, 147, 160

P Padgett, R. J., 150, 158, 160 Palmeri, T. J., 4, 15 Pani, J. A., 128, 138, 141 Parker, J., 65, 72 Parker, R. E., 2, 14 Parkman, J. M., 194, 197 Paterson, S. J., 93, 94, 95, 96, 97, 99,100 Pearlman, L. A., 62, 73 Perlmutter, M., 61, 74 Perner, J., 110, 113 Perrett, D. I., 92, 95, 99 Perry, J., 2, 13 Persun, N. I., 249, 264 Peterson, C., 64, 66, 73 Pethick, S. J., 124, 126, 140 Pezdek, K., 63, 73 Pezzini, G., 92, 101 Phillips, A. T, 45, 52, 236, 242, 246 Piaget, J., 12, 14, 20, 42, 84, 86, 107, 112, 113, 163, 166, 170, 181, 183 Picariello, M. L., 6 1, 73 Pillemer, D. B., 61, 73 Pinker, S., 91, 92, 100, 203, 220 Pintilie, D., 93, 99 Pipe, M-E., 69, 73, 146, 147, 155, 159, 160 Poulin-Dubois, D., 120, 134, 135, 142 Prawat, R. S., 172, 180, 183 Pressley, M., 27, 42 Principe, G., 69, 73

Author

Prinz, J. J., 5, 13, 15 Pruett, J. C., 61, 73 Pynoos, R. S., 64, 73

Q Quas, J. A., 63, 72, 154, 159 Quillian, M. R., 189, 197 Quinn, I? C., 117, 118, 119, 140, 142, 167, 169, 181, 183, 235,245

R Rabinowitz,

M., 185, 187, 190, 193, 194, 195, 197, 198 Rakison, D., 168, 169, 183 Randolph, D., 255, 264 Ratner, H. H., 19, 42, 57, 74, 146, 147,150, 158,160 Reber, A. S., 138, 143 Reese, E., 66, 71, 74, 139, 140, 154, 155, 159,160 Regier, T., 201, 204, 206, 207, 211, 215,218,220,221 Reich, PA., 121, 142 Reisberg, D., 62, 72 Rescorla, L., 120, 122, 126, 127, 135,142 Reviere, S., 62, 74 Reznick, J. S., 122, 124, 126, 128, 140, 142 Riccuiti, H. N., 166, 183 Richards, T. A., 250, 264 Richters, J., 65, 74 Riddle, A. S., 120, 134, 142 Riddlesberger, M. M., 63, 72, 154, 159 Rideout, ,, 64, 73 Rips, L. J., 173, 183 Ritchey, G. H., 2, 14 Rizzolatti, G., 112, 114 Robinson, B. E, 93, 100 Rosch, E., 3, 15, 116, 119, 142, 162, 164,183 Ross, G. S., 117, 142, 166, 167, 171, 177, 183 Rossen. M.. 93. 95. 100

Index

l

273

Rosser, R., 238, 246 Rudy, L., 67, 72 Ruff, H. A., 97, 100, 117, 142 Rumelhart, D. E., 4, 15, 187, 189, 190,197,198 Russell, R., 176, 178, 181 Rutter, M., 97, 100

S Sabbadini, L., 92, 101 Sadler, D., 214, 215, 216, 216n, 221 Sales, J. M., 65, 66, 72 Salmon, W. C., 48, 52 Sarfati, D., 65, 72 Schacter, D. L., 186, 198 Schaffer, M., 4, 15 Schank, R., 54, 74 Schneider, W., 27, 42 Schwade, J. A., 36, 37, 40 Schwartz, A. B., 203, 220 Schwartz, S. I?, 49, 52 Schwarzmueller, A., 61, 72 Scott, P, 176, 182 Senghas, A., 131, 142, 143 Seress, L., 32, 42 Sheldrick, R., 249, 264 Shi, M., 8, 14 Shore, C. M., 20n, 40 Siegler, R. S., 195, 198 Simon, H. A., 188, 197 Singer Harris, N. G,, 93, 100 Sitarenios, G., 97, 99 Slaughter, V., 45, 50, 51, 242, 245 Sloman, S. A., 8, 14 Smiley, P, 121, 131, 140, 142 Smith, B. S., 57, 74, 146, 147, 150, 158, 160 Smith, D. A., 59, 72 Smith, E. E., 166, 173, 183 Smith, L. B., 121, 138, 141, 161, 176, 177, 182, 183, 223, 245 Smith, L. L., 175, 176, 181 Smith, M., 97, 99 Smith, R., 255, 264 Smith, W. J., 2, 15 Smolenskv. I?. 4. 15

274

l

Author

Index

Solomon, K. O., 5, 6, 13, 15 Somerville, S., 176, 181 Sowell, E., 93, 99 Spelke, E. S., 45, 47, 49, 50, 51, 52, 76, 86, 94, 100, 201, 221, 231,236,237n, 242,243, 245,246 Spencer, J., 118, 142 Squire, L. R., 21, 41, 186, 198 St. George, M., 92, 100 Starkey, D., 166, 167, 183 Starkey, I?, 94, 100 Starr, R. M., 21, 42 Stein, N. L., 2, 14, 54, 74, 248, 249, 250, 251, 253, 255, 261, 262,263,264,265 Steinberg, A. M., 64, 73 Stephenson, M., 158, 160 Steward, M., 63, 74 Stewart, J. A., 77, 86 Stone, G. O., 192, 198 Subrahmanyam, K., 77, 78, 80, 81, 82,87 Sullivan, K., 92, 101

T Tager-Flusberg, H. J., 92, 101 Talmy, L., 201, 221 Tanouye, E., 171, 177, 183 Taylor, C., 255, 264 Taylor, J. X. X., 63, 73 Tees, R. C., 230, 246 Temple, C. M., 91, 92, 101 Terr, L. C., 64, 67, 74 Tessler, M., 69, 74, 155, 160 Thagard, I? R., 8, 14 Thal, D. J., 124, 126, 140 Thomson, D. M., 2, 15 Thomson, R., 158, 160 Tincoff, R., 228, 246 Todd, C., 61, 74 Tomikawa, S. H., 172, 184 Tooby, J., 109, 113 lrabasso, T., 248, 250, 262, 264, 265 Trauner, D., 99 Travis, L. L., 23, 40, 56, 71 Treisman, A., 203, 221 Tsivikin. S.. 243. 246

Tulving, E., 2, 15 Turing, A. M., 188, 198 Tversky, B., 169, 184

U Uchida, N., 116, 136, 140 Udwin, O., 97, 101 Uzgiris, I. C., 118, 119, 141

V Vallee-Tourangeau, F., 10, 11, 15 Van de Walle, G., 46, 52, 118, 142, 237n,

246

van den Broek, I? W., 22, 42, 147, 160 Vicari, S., 92, 101 Volterra, V., 92, 101 von Neumann, J., 188, 198 Vygotsky, L. S., 2, 15, 163, 166, 170, 184

W Wachtel, G. F., 124, 141 Wade, E., 253, 265 Walker, W. H., 11, 15 Wang, l? l?, 91, 92, 93, 95, 99, 101 Wang, Y., 8, 14 Waters, J. M., 33, 40, 66, 71 Waxman, S. R., 121, 131, 136, 137, 142, 143, 175, 184 Wellman, H. M., 254, 263 Wenner, J. A., 21, 24, 25, 30, 33, 36, 39, 40 Werker, J. F., 230, 246 Werner, H., 179, 184 West, T. A., 19, 42 Wetstone, H., 171, 177, 183 Wewerka, S. S., 21, 23, 24, 25, 30, 33, 36, 37, 40 Whetton, E., 93, 99 Whiten, A., 108, 114 Whitfield, L. C., 176, 181 Whorf, B. L., 219, 221 Wiebe, S., 33, 40 Wilcox, T., 238, 246

Author Wildfong, Williams, Wilson, Winter,

S., 172, 183 E. M., 76, 77, 78, 79, 81~1, 86,87 H. B., R., 138,203,143 221

Woll, S. B., 59, 72 Wong, A., 25 1, 265 Woodward, A. L., 45, 52, 236, 246

242,

Index

l

wu, L. L., 2, 5, 6, 9, 11, 13, 15 Wynn, K., 236, 244

Y Yeh, W., 2, 7, 9, 13, 15 Younger, B. A., 119, 143 Yule, W., 97, 101

275

This Page Intentionally Left Blank

Subject Index 3%

A Accessible representation, 226-228 Age-related differences, in long-term recall, 30-32, 31, 38 Analogy, in situated concepts, 8 Animate-inanimate distinction, 76-84, 85 Associative network, 189-l 90 Attentional Vector Sum (AVS) model cross-linguistic variations in, 214218,215,216,217, 218 described, 203-205, 204 vs. competing models, 205-207, 206,207,208,209 within-language variations in, 209-214,211,212,213 Atypical development face processing studies in, 95-97 language and number studies in, 92-95 Modular Continuity Assumption and, 91-92 relations to typical development, 97-98 Autobiographical memory changes in, 18-19 verbal report of, 35-37, 37 AVS model, see Attentional Vector Sum (AVS) model

B BB model, 205-206,206

Belief, vs. pretense, 105-l 08, 111-112 Bounding Box (BB) model, 205-206, 206

C Categorization cognitive development and, 177-180,179 conceptual vs. perceptual, 43-5 1, 169-l 70 function in, 166-l 70 Mandler ‘s theories of, 16 l-l 62 Causality conceptual representation and, 4546,48-51 in motion paths, 77-84, 83 Concepts abstract, in infants, 103-l 04 artifact, 173-l 75 language and, 228-235,242-244 in perceptual simulation, 4-6 preverbal, 116-l 18, see also Language acquisition theories of, 3-4,44-45,223-224 Conceptual categorization, vs. perceptual, 43-5 1, 169-l 70 Connectionist models, 4, 190-l 91 Containment as accessible representation, 226-228 conceptual distinction of, 225-226,227 language and, 228-235,232,233 277

278

l

Subject

index

object representation and, 23%242,240,241 Context in event memory, 153-154, 154 in procedural knowledge, 192-193 Core, of concept, 45, 49, 165 Core knowledge, 47-48, 50-5 1, 236-237,242-243

D Declarative-procedural distinction, 186-187, 196-197, see also Procedural knowledge Decoupling, 109 Deferred imitation, see Elicited imitation Delta rule, 192 Domain-level distinctions, 46 Downs Syndrome studies, 93-97 t Elicited imitation, 20-21 Emotional memory, see also Trauma memory emotional provocation and, 250-25 1 goal and preference basis of, 24 7-248 psychological well-being and, 259-261,260 recall studies in, 251-259, 252, 253,2.54,257,258,259 usefulness of study in, 249 Emotional regulation, in trauma memory, 69-70 Enabling relations, effect on long-term recall, 22-23, 55-56 Entity properties, 11 Essences, in situated concepts, 8-9 Event memory, see a/so ‘Trauma memory event schemas in, 54-55 hierarchical structure in, 150-153,151 logical relations and, 146-l 50 object overlap in, 147-150, 149

role of language in, 154-159, 157 specific experiences in, 59-62 temporal structure and, 55-5 7, 60, 61 variability and, 56-58, 60, 61 world views and, 70 Exemplar models of concepts, 4 Experiential mediation, 1O-l 1 Explanation, conceptual representation and, 48-49

Face processing, in atypical development studies, 95-97 Familiarity, effect on long-term recall, 23-24 Feature accretion, in language acquisition, 120 FHC, 164-166, 170-l 73 Form and function, see Shape Function definitions of, 164-l 65 in infant categorization, 166-l 70 language and, 170-l 73, 175-l 77 theories of, 163-l 66, 173-l 75 Functional core hypothesis (FHC), 164-166, 170-l 73 Functional core plus identification hypothesis, 173-l 74

G Goals, emotions and, 247-248, 250-251, 261

H Habituation-dishabituation paradigms function, 16 7 spatial relationships, 231-234, 232,233,240,241 Hierarchical structure, in event memory, 150-153, 151

I Inductive generalization,

118-l 19

Subject Introspective properties, 1 l-1 2 Intuitive theories of concepts, 3-4

L Language, see also Spatial language in atypical development studies, 92-95 cognitive development and, 180 in event memory, 154-159, 15 7 form and function in, 170-l 73, 175-177 in trauma memory, 66, 70 Language acquisition basic-level categories in, 119-l 20 comprehension tests, 121-125, 125,129 nonce label tests, 131-133 overextensions in, 120-l 21, 123-130 production tests, 123-127, 125, 127 research summary, 13 5-l 3 9 similarity in, 116-l 18, 133-l 35 underextensions in, 12 1 Learning patterns, in procedural knowledge, 193-l 95 Linguistic universals, 2 18-2 19 Logical relations, in event memory, 146-150 Long-term recall age-related differences in, 30-32, 31, 38 changes in, 18-l 9 consolidation in, 24-32, 26, 28, 29 determinants of, 22-23 elicited imitation study of, 19-2 1 emergence of, 32-34, 35, 38 verbal report of, 35-37, 37, 38-39

M Medial temporal lobe structures, long-term recall and, 32-34 Metarepresentational theory of pretense, 104, 107, 113 Modular Continuity Assumption challenges to, 92-9 7

index

l

279

infancy study and, 91-92 “Monster” study, 24-25 Motion paths, causal interpretations of, 77-84, 83 Multicomponent neural network, 18-19

N Network (retrieval) knowledge behavior characteristics of, 191-l 95 described, 187, 189-191 relation to symbol manipulation knowledge, 195-196 Nonce label tests, 131-133 Number tests, in atypical development studies, 92-97

0 Object-examination tasks, 166 Object overlap, in event memory, 147-150,149 Object representation, 235-242, 240,241 Object-touching behavior studies, 166 Occlusion, described, 224, 224n, see also Containment Overextensions, in language acquisition, 120-121

P Parallel Distributed Processing models, 190-191 Perceptual categorization, vs. conceptual, 43-5 1, 169-l 70 Perceptual simulation concepts in, 4-6 perceived situations and, 7 Personal memory, see Autobiographical memory Preferential-looking paradigms function, 16 7 object representation, 238-239 Pretense concept vs. processing mode in, 108-l 09

280

l

Subject

Index

as mental action, 11 l-1 13 metarepresentational theory of, 104,107,113 relation to action, 11 O-l 11 types of, 109-l 10 vs. belief, 105-108, 11 l-l 12 Problem size effect, 194 Procedural knowledge hybrid model of, 18 7, 19 1, 196-197 interaction between types of, 195-196 performance characteristics of, 191-195 types of, 187-191 vs. declarative, 186-l 8 7 Production system models, 188-l 89 Psychological essentialism, 45, 49 Psychological well-being, emotional memory and, 259-261,260

R Recall, see Long-term recall Repetition effect on long-term recall, 23-24 effect on trauma memory, 66-68 Retrieval knowledge, see Network (retrieval) knowledge Rule-based knowledge, see Symbol manipulation (rule-based) knowledge

S

importance of, 2-3 simulated, and perceived situations, 7 Spatial language AVS model described, 203-205, 204 vs. competing models, 205-207, 206,207,208,209 cognitive or perceptual processes in, 201-203 concept development and, 228-235,242-244 cross-linguistic variations in, 199-200,200,214-218, 215,216,217,218 usefulness of study in, 201 within-language variations in, 209-214,211,212,213 Spatial relationships, see Containment Spreading activation systems, 190 Stimulus indeterminacy, 7 7 Story schemas, 54 Stress, effect on memory, 62-63 Sustained attention, in atypical development studies, 96-97 Symbol manipulation (rule-based) knowledge behavior characteristics of, 191-195

described, 18 7-l 89 relation to network knowledge, 195-196

T

Schema theory, 54 Semantic mediation, 10 Shape in infant categorization, 167-l 70 in language acquisition, 170-173, 175-176 Similarity, in language acquisition, 116-118,133-135

Situated concepts empirical support for, 9-12 theories of, 4, 7-9 Situational properties, 1 l-12 Situations

Taxonomic bias principle, 131 Taxonomic concepts, 11 Temporal structure enabling relations and, 22-23, 55-56 event representation and, 55-57, 60,61 in trauma memory, 68-69 Thematic relations in concepts, 12 Theory of mind research, 105-l 06 Theory theory of concepts, 4445, 4849, 174-l 75 Trauma memory

Subject Index

effect of stress on, 62-63 of recurring experiences, 66-68 role of language in, 66, 70 of specific experiences, 6466 temporal structure and, 68-69 world views and, 70

U Underextensions, in language acquisition, 12 1 Unmediated retrieval, 10 Unsituated concepts, 3-4

l

281

V Variability, in event representation, 56-58, 60, 61 Visuo-spatial processing, in atypical development studies, 96

w Williams syndrome studies, 92-9 7 Word meanings, function and, 170-l 73, 175-l 77 World views, trauma memory and, 70

This Page Intentionally Left Blank