Advances in Child Development and Behavior Volume 16

ADVANCES IN CHILD DEVELOPMENT AND BEHAVIOR

VOLUME 16

Contributors to This Volume Elizabeth Bates Marjorie Beeghly-Smith

Marc H. Bornstein Inge Bretherton Martha J. Farah Tiffany M. Field Stephen M. Kosslyn Michael E. Lamb Sandra McNew David S. Palermo Mitchell Robinson Robert S. Siegler Tedra A. Walden

ADVANCES IN CHILD DEVELOPMENT AND BEHAVIOR

edited by Hayne W. Reese

Lewis P. Lipsitt

Department of Psychology West Virginia University Morganrown, West Virginia

Department of Psychology Brown University Providence, Rhode Island

VOLUME 16

@

1982

ACADEMIC PRESS A Subsidiary of Harcourt Brace Jovanovich, Publishers

New York London Paris. San Diego San Francisco *SBo Paulo *Sydney .Tokyo .Toronto

BY ACADEMIC PRESS,INC. ALL RIGHTS RESERVED. NO PART O F THIS PUBLICATION MAY BE REPRODUCED OR TRANSMITTED IN ANY FORM OR BY ANY MEANS, ELECTRONIC OR MECHANICAL, 1NCLUDING PHOTOCOPY, RECORDING, OR ANY INFORMATION STORAGE AND RETRIEVAL SYSTEM, WITHOUT PERMISSION IN WRITING FROM THE PUBLISHER.

COPYRIGLIT @ 1982,

ACADEMIC PRESS, INC.

I 1 1 Fifth Avenue, New York, New York 10003

UuifedKiiiPdotn Editiori ~irblisliedbv A C ADEM~CPRESS, INC. (LONISON) 24/28 Oval Road, London N W l

LIBRARY OF

7DX

CONGRESS CATALOG CARD

NUMBER:63-23237

ISBN 0-12-009716-8 PRINTED IN THE UNITED STATES OF AMERICA

82 83 84 85

9 8 7 6 5 4 3 2 1

LTD.

Contents List of Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

ix

Boyd R. McCandless (1915-1975) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xi

Preface .....................................................................

XIII

Erratum ....................................................................

xv

...

The History of the Boyd R. McCandless Young Scientist Awards: The First Recipients DAVID S. PALERMO Text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

I

Social Bases of Language Development: A Reassessment

I. 11. 111. IV. V. VI.

VII. VIII. IX. X.

ELIZABETH BATES, INGE BRETHERTON, MARJORIE BEEGHLY-SMITH, AND SANDRA McNEW Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A Brief History of the Marriage: Love and Epistemology .............. Cognitive Inputs to Language ....................... .............. Social Inputs to Language ............................................... Verbal Interaction: “Motherese” . . . . . . . . . . ............... Conceptual and Methodological Confounds in Social-Causal Theories of Language Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Direction of Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Genetic Confounds . . . . . . . . . . . . . . . . . . . . . .............. Threshold Effects . . . . . . . . . . . . . . . . . . . . . . .............. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . .............. References ..............................................

8 8 12 15

35 48 54

59 61 64 68

Perceptual Anisotropies in Infancy: Ontogenetic Origins and Implications of Inequalities in Spatial Vision I. 11. 111. IV.

MARC H. BORNSTEIN Introduction .......................................................... Two Classes of Perceptual Anisotropy ..................................... Two Classes of Perceptual Anisotropy in Infancy ............................ Discussion and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V

77 79 85 108 115

vi

Contents

Concept Development MARTHA J. FARAH AND STEPHEN M. KOSSLYN I. Introduction . . . . . . . . . . . . ........................ 11. Information-Processing Theones . . . . . . . . . . . . . . . . . . . . . . 111. The Contents of Concept Representations ........................ IV. The Format of Concept Representations.. ......................... V. The Organization of Concept Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VI. Conclusions ...................... References ..............................................

125

130 160 164

Production and Perception of Facial Expressions in Infancy. and Early Childhood TIFFANY M. FIELD AND TEDRA A. WALDEN 1. Adult Facial Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. The Infancy Literature ................................ ............... Ill. The Child Literature. . ............................ ............... Expressions by the Authors . . . . . . . . . . . . . . . . IV. Studies of Infant and C V. An Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References ..............................................

17 I 174 179

208

Individual Differences in Infant Sociability: Their Origins and Implications for Cognitive Development MICHAEL E. LAMB I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. Measuring Sociability ....... IV. V. VI.

Explaining the Relationship between Sociability and Cognitive Performance . . . . . . Origins of Individual Differences in Sociability ............... Conclusion ..........................

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

213 215 226 230 236 237

The Development of Numerical Understandings ROBERT S. SIEGLER AND MITCHELL ROBINSON ................... ........................ 242 11. An Initial Study of Number Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 Ill. Preschoolers’ Knowledge of Counting ........................ 250 IV. Preschoolers’ Knowledge of Numerical ........................ 267 V. Preschoolers’ Knowledge of Addition . . 287 VI. Conclusions: The Development of Nu . . . . . . . . . . . . . . . 299 References . . . . . . . . . ........................ 308

I. Introduction . . . . .

Contents

vii

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

313

Subject Index ...............................................................

323

Contents of Previous Volumes ..................................................

327

This Page Intentionally Left Blank

List of Contributors Numbers in parentheses indicate the pages on which the authors' contributions begin.

ELIZABETH BATES Department of Psychology, University of Colorado, Boulder, Colorado 80309 ( 7 ) MARJORIE BEEGHLY-SMITH Department ofPsychology, University of Colorado, Boulder, Colorado 80309 (7) MARC H. BORNSTEIN Department ofPsychology, New York University, New York, New York 10003 (77) INGE BRETHERTON' Department of Psychology, University of Colorado, Boulder, Colorado 80309 (7) MARTHA J . FARAH Department of Psychology and Social Relations, Harvard University, Cambridge, Massachusetts 02138 (125) TIFFANY M. FIELD Mailman Center f o r Child Development, University of Miami Medical School, Miami, Florida 33101 (169) STEPHEN M. KOSSLYN2 Department of Psychology and Social Relations, Harvard University, Cambridge, Massachusetts 02138 (125) MICHAEL E. LAMB Department of Psychology, University of Utah, Salt Lake City, Utah 841 12 (213) SANDRA McNEW Department of Psychology, University of Colorado, Boulder, Colorado 80309 ( 7 ) DAVID S . PALERMO Department of Psychology, The Pennsylvania State University, University Park, Pennsylvania 16802 ( I ) MITCHELL ROBINSON Department of Psychology, Carnegie-Mellon University, Pittsburgh, Pennsylvania 15213 (241) ROBERT S . SIEGLER Department of Psychology, Carnegie-Mellon University, Pittsburgh, Pennsylvania 15213 (241) TEDRA A . WALDEN3 Mailman Center f o r Child Development, University of Miami Medical School, Miami, Florida 33101 (169) 'Present address: Department of Human Development and Family Studies, Colorado State University. Fort Collins, Colorado 80523. *Present address: Program in Linguistics and Cognitive Science, Brandeis University, Waltham, Massachusetts 02554. 'Present address: Department of Psychology and Human Development, Peabody College, Vanderbilt University, Nashville, Tennessee 37203. ix

BOYDR. MCCANDLESS

Boyd R. McCandless (1 915-1 975)

This volume is dedicated to the memory of Boyd R. McCandless, who was a beloved teacher and important influence on the lives of both present editors of the Advances in Child Development and Behavior. This publication was founded by one of the present editors (LPL) and another McCandless student, Charles C. Spiker. Boyd himself was an author in Volume 10, published in 1975, the year of his death. It seemed fitting to us that a volume be dedicated to Boyd McCandless’ memory, for he toiled tirelessly to build this field of child development and behavior, and he influenced numerous teachers, researchers, and students. Early in 1979, as we thought about the content that we would choose to appear in such a dedicatory volume, it occurred to us that nothing could be more fitting than to invite contributions from the recipients of the Boyd R. McCandless Young Scientist Award created by the Division on Developmental Psychology of the American Psychological Association shortly after Boyd’s untimely death. The history and intent of this Award are described in David S. Palermo’s contribution. The first awards were made at the 1978 meeting of the American Psychological Association, and we decided to invite contributions from these recipients and the second year’s recipients, who were to be announced at the 1979 meeting of the American Psychological Association. We did invite those young scholars, most of whom did not know Boyd but whose scientific lives in child development could not help but have been affected by his nurturance of the field. This sort of intergenerational continuity in science and among scientists always fascinated Boyd. He was rather a keen observer of the history of child development and delighted in reciting the intellectual pedigrees of colleagues. In keeping with the tradition of the Advances, the contributions to this memorial volume are not organized around any theme, other than child development and behavior in general. Indeed, the diversity of topics in this volume is greater than that of some earlier volumes. We feel that this diversity makes a highly appropriate memorial to Boyd McCandless, whose own scholarly work covered a wide range of topics. He was a codeveloper, with Alfredo Castaneda and David S. Palermo, of the children’s form of the Manifest Anxiety Scale. McCandless, Castaneda, and Palermo used the CMAS, as it came to be called, in five studies published in 1956. This landmark series of studies was among the earliest research demonstrating the utility of learning theory for an understanding of child development and behavior, and helped establish experimental child psychology as a science in its own right. Among the other landmarks in the development of xi

xii

Boyd R . MrCandless

that science were Boyd’s publications with Charles C. Spiker on the operational definition of intelligence, in 1954, and on experimental child psychology research, in 1956. Boyd also had a long-continuing interest in delinquency and personality disorders, as reflected, for example, by a 1944 report on predelinquent children coauthored with Sidney W. Bijou, a 1962 paper on delinquency coauthored with John w.McDavid, and a 1972 paper on intervention. He also published extensively on intelligence, measurement, and education; but by far his major research interest and largest volume of publications were in the area of normal personality and social development. Our purpose here, however, is not to review Boyd’s work but rather to memorialize it and his personal contribution to the science of child development and behavior and to the practitioners of that science. To the memory of those contributions we dedicate this volume. Hayne W. Reese Lewis P. Lipsitt

Preface

The present volume in Advances in Child Development and Behavior is dedicated to the memory of Boyd R. McCandless and contains articles by recipients of the Boyd R. McCandless Young Scientist Award (see the dedication preceding this Preface and the History of the Boyd R. McCandless Young Scientist Awards). As in previous volumes, no attempt was made to organize the articles around a particular theme or topic, thus continuing the tradition of the Advances of providing a place for publication of critical reviews and scholarly speculation. The amount of research and theoretical discussion in the field of child development and behavior is so vast that researchers, instructors, and students are confronted with a formidable task in keeping abreast of new developments within their areas of specialization through the use of primary sources, as well as in being knowledgeable in areas that are peripheral to their primary focus of interest. Moreover, journal space is often simply too limited to permit publication of more speculative kinds of analyses that might spark expanded interest in a problem area or stimulate new modes of attack on a problem. This publication is intended to ease the burden by providing scholarly technical and speculative articles in which recent advances in the field are summarized and integrated, complexities are exposed, and fresh viewpoints are offered. The articles should be useful not only to the expert in the area but also to the general reader. Manuscripts are solicited from investigators conducting programmatic work on problems of current and significant interest. Contributions often deal intensively with topics of relatively narrow scope but of considerable potential interest to the scientific community. Contributors are encouraged to criticize, integrate, and stimulate, but always within a framework of high scholarship. Although appearance in the volumes is ordinarily by invitation, unsolicited manuscripts are accepted for review if submitted first in outline form. Whether invited or submitted, all articles receive careful editorial scrutiny, often with outside review. Invited contributions are automatically accepted for publication in principle, but usually require revision before final acceptance. Submitted contributions receive the same treatment except that they are not automatically accepted for publication even in principle, and may be rejected. We wish to acknowledge with gratitude the aid of our home institutions, West Virginia University and Brown University, which generously provided time and facilities for the preparation of this volume. We benefited as well from the ...

Xlll

xiv

Preface

facilities of the Center for Advanced Study in the Behavioral Sciences at Stanford, where one of us (LPL) was located during the early work on the volume. We also wish to thank Dr. Robert K . Moore for his editorial assistance. With the publication of the present volume, Lewis P. Lipsitt will have stepped down as coeditor and Hayne W. Reese will be the sole editor. Correspondence should therefore be addressed to the latter. Hayne W. Reese Lewis P. Lipsitt

Erratum Advances in Child Development and Behavior Volume 15 In the article by William Fowler entitled “Cognitive Differentiation and Developmental Learning,” the text on pages 180 and 181 should appear on pages 178 and 179. The text on pages 178 and 179 should appear on pages 180 and 181.

xv

This Page Intentionally Left Blank

THE HISTORY O F THE BOYD R. McCANDLESS YOUNG SCIENTIST AWARDS: THE FIRST RECIPIENTS

David S.Palerrno DEPARTMENT OF PSYCHOLOGY THE PENNSYLVANIA STATE UNlVtRSlTY UNIVERSITY PARK, PtNNSYLVANIA

In 1974, the Executive Committee of Division 7 of the American Psychological Association discussed the possibility of providing some kind of recognition for outstanding young developmental psychologists. The Committee thereupon voted to establish a set of awards for that purpose. In the following year, Willard Hartup, who was president of the Division, asked me to present a proposal for implementing the awards to the Executive Committee at the 1976 APA meeting. I asked several members of Division 7 on the Pennsylvania State University campus to meet with me to consider the best way to proceed, the criteria for making such awards, and the general procedures for implementing the wishes of the Executive Committee. That group, which included Joseph Britton, Frank DiVesta, Dale Harris, and John Withall, developed a plan which I, in turn, presented to the Executive Committee at its 1976 meeting. The Committee offered advice concerning the proposed implementation procedures, and with the Committee’s comments in mind, I revised the proposal and presented it at the 1977 meeting, where it was approved by the Executive Committee and was subsequently set into motion, when the membership voted approval at the 1977 Business Meeting of Division 7. On December 5 , 1975, the year that the Pennsylvania State group was meeting to formulate the plans for these awards, Boyd R . McCandless died. At a memorial service honoring Boyd on December 8 , Walter Hodges said, Boyd McCandless gave lifc, hope, love, and knowledge to many. Such gifts will never perish. He is in each of our hearts, our intellects, our beings. Those of us he touched are legion. People all over the world have been nourished by his warmth, his advice, his reflection, his collaboration, his steadfast friendship. Intellectually and spiritually it felt so good to be with this man. His nurturance was always available.. . . His scholarly contributions also reflect his great spirit. He rose above his own human vulnerabilities to do research, to write, to speak, to teach about areas of great concern in our own and other societies. Honored by his profession many times. . . he never lost touch with his basic affection and concern for students, colleagues, friends, and family.

I ADVANCES IN CHILD DEVELOPMENT AND BEHAVIOR. VOL. 16

Copynght 0 1982 by Academic Press, Inc. All rights of reproduction in any form reserved. ISBN 0-12-W971&8

2

Duvid S. Palermo

Boyd Rowden McCandless was born in St. John, Kansas, on August 18, 1915. He obtained his bachelor’s degree (in English) at Fort Hays (Kansas) State College in 1936. Thereafter he found his way to the historic Child Welfare Research Station at the University of Iowa, where he earned his master’s degree in 1938 and his Ph.D. in 1941. Iowa became the central locus of his career, both because of the imprint on him of his Iowa training and the faculty members there with whom he was so close, and because he returned there in 1950 to become Director of the Station for a decade. Boyd’s death was a shock to a large number of developmental psychologists, who felt a strong sense of loss both personally and professionally. Boyd had been an important influence on the field of developmental psychology. His influence began when he was a graduate student in the Iowa Child Welfare Research Station where, among other activities, he participated as a research assistant in Kurt Lewin’s pioneering leadership studies. After completing his graduate studies in Iowa, he went to Wayne County Training School in Michigan, where he expanded his clinical and research experience for 2 years. After a 4-year stint in the U.S. Maritime Service during World War 11, he took a position at San Francisco State University, where he rose from Assistant to Associate Professor in the 2 years he was there (1946-1948). During that period, he encouraged two young men, Charles Spiker and Alfred0 Castaneda, to go on to graduate work in child psychology. In 1948 he moved to Ohio State University as Associate Professor. He remained there until 195 1, when he was appointed Director and Professor of the Iowa Child Welfare Research Station. While in Iowa, with the help of his former students Castaneda and Spiker, he developed a nationally recognized program in experimental child psychology. In 1960, he left Iowa to spend 2 years in Pakistan, where he was Professor and Chairman of the Psychology Department’s Institute for Education and Research at the University of Punjab. He returned to this country in 1962 as Professor of Education and Psychology and Director of the University School Clinic Complex at Indiana University. He also served as Chairman of the Special Education Department from 1963 to 1965. In 1966, he moved to Emory University, where he remained as Professor of Psychology and Education and Director of the Educational Psychology program until his death. Boyd’s influence on developmental psychology is clearly evident in his research contributions to the literature, his several excellent textbooks, his distinguished editorship of the first ten volumes of Developmental Psychology (1970-1974), and his presidency, in 1955, of Division 7 of the American Psychological Association. Boyd was also a Diplomate in Clinical Psychology. As is evident from Boyd’s record, he was a leader, as both teacher and researcher, in experimental, clinical, and educational aspects of developmental psychology.

Boyd R . MrCandless Young Scientist Awards

3

The loss of Boyd McCandless was, however, a very personal one to the many students and colleagues whose lives he influenced at the universities where he had taught. In addition, he had affected many others through his editorial labors and the individual contacts he made at meetings and conferences, where he expanded his friendships so naturally and easily. While Boyd was an obvious leader in the field at large, his influence on the younger members of the discipline was particularly notable. Boyd engendered in his students an excitement, a vigor, and a genuine effervescence which led to an unusually hard-working, interested, and productive group who spread into the various specialized areas of the field. Boyd was eclectic, and the diversity of his students’ interests reflects that characteristic. It was Boyd’s sincere interest in each student as a person about whom he cared, regardless of his or her interests, that brought out the best in those students. The qualities that enabled Boyd to encourage and stimulate students and others beginning their careers in the developmental field made it natural for us to accept Lewis Lipsitt’s suggestion to me that the Division 7 awards be named in Boyd’s honor. At its meetings in 1976, the Executive Committee accepted with enthusiasm the suggestion that the newly established award be called the Boyd R. McCandless Young Scientist Award. As approved in 1977, the Boyd R. McCandless Young Scientist Award was to take the form of an individually printed citation to be presented at the annual business meeting of the Division, following the presentation of the G. Stanley Hall Award for outstanding contributions to developmental psychology by a senior member of the discipline. It was stipulated that the criteria for the McCandless Awards should include evidence of a distinguished theoretical contribution, programmatic research effort, or dissemination of scientific developmental information to those outside as well as within the profession. The award is made on the basis of evidence of a continued effort rather than a single outstanding piece of work, and the contributions of the potential candidates in the year preceding the award are the primary focus of the selection committee’s attention, although the literature for at least 3 preceding years is also examined. Candidates are eligible for the Award only during the 7-year period following the receipt of their doctoral degree. A committee, appointed each year by the president of the division, is charged with the task of nominating one to five persons for the Award to the Executive Committee, which makes the final decision. Once the procedure for making the awards had been established, Frances D. Horowitz, who had assumed the presidency of Division 7 at the 1977 meeting, asked me to chair the nominating committee for the first 2 years of the awards. In addition, she asked Rachel K. Clifton, Martin L. Hoffman, Lewis P. Lipsitt, and Richard D. Odom to serve on the first selection committee. As a group, we solicited the names of young persons who fulfilled the criteria that had been established. In that year, 27 nominated persons were found to be eligible. The

4

Duvid S. Palerrno

curricula vitae of those persons were collected and examined by the five commit-

tee members independently. The members each selected the ten best candidates and rank-ordered the ten. The quality of the candidates and the difficulty of the task of making distinctions among them is reflected in the fact that 19 of the 27 persons nominated were on the combined list of the committee. There was, however, very good agreement among the committee members about the top four candidates. The committee therefore recommended to the Executive Committee of the Division that Marc H . Bornstein, Stephen M. Kosslyn, Michael A. Lamb, and Michael P. Maratsos be awarded the first Boyd R. McCandless Young Scientist Awards. At the 1978 American Psychological Association meeting in Toronto, I had the opportunity of making the presentations to those four outstanding persons. It was the culmination of an exhilarating but difficult taskexhilarating because once we turned our attention to the accomplishments of the young developmental psychologists, it became clear that we had a healthy, vigorous discipline with a large number of very talented and productive young researchers who are pushing to new frontiers our knowledge about developmental processes. I personally was particularly pleased by this, perhaps obvious, fact because it reflects so well the spirit of Boyd McCandless and perpetuates his tradition of encouraging the younger members of the profession. The decisions in the second year of the awards were no less difficult and no less exciting. The selection committee consisted of Rachel Clifton, Michael Lamb, Richard Odom, and Charlotte Patterson, and me, remaining as chairperson. Nominations were solicited from the committee members and the general membership. Out of the 30 nominees who were eligible, three emerged as consensual stars among the outstanding group of emerging scholars who came to our attention. They were Elizabeth Bates, Tiffany Field, and Robert S. Siegler. Again, I had the opportunity of presenting the awards to these individuals at the I979 American Psychological Association meeting in New York. Let me sketch briefly the backgrounds of the seven developmental psychologists who have achieved this distinction. They are clearly promising developmental psychologists in the tradition that Boyd McCandless so naturally encouraged among those who preceded them. Marc H. Bornstein was born on November 23, 1947, in Boston, Massachusetts. He received his B.A. degree from Columbia College in 1969 and his M.S. and Ph.D. degrees from Yale University in 1973 and 1974, respectively. His first position was at Princeton University, after a postdoctoral fellowship year spent at Yale and at the Max-Planck Institut in Munich. He is currently in the Department of Psychology at New York University. He has received a number of awards for his teaching and research, which has focused on visual perception and, in particular, the color vision capabilities of infants. In addition, however, he has been concerned with visual and time sensitivity in different social-cultural settings.

Boyd R . M~~Cunillrss Young Scientist A,rards

5

Stephen M. Kosslyn was born on November 30, 1948, in Santa Monica, California. He received his B.A. degree from the University of California at Los Angeles in 1970 and his Ph.D. degree from Stanford University in 1974. His first teaching position was at Johns Hopkins University. He joined the faculty at Harvard University in 1977, where he has remained. He has received grant and fellowship support from the National Science Foundation nearly continuously since he was an undergraduate student. His research efforts have focused on mental representations, with special concern for mental imagery and how children use mental imagery in their cognitive processing. Michael E. Lamb was born on October 10, 1953, in Lusaka, Northern Rhodesia (Zambia). He received his B.A. degree from the University of Natal, Durban, South Africa in 1972. In 1973 he came to the United States and enrolled at Johns Hopkins University, where he received an M.A. degree in 1974. The following year, 1975, he received a M.Phi1. degree from Yale University, and in 1976 he was awarded his Ph.D. degree, also from Yale University. He joined the faculty at the University of Wisconsin in 1976 as Assistant Professor and in 1978 moved to the University of Michigan at the same rank. In 1981 he accepted a position as Professor of Psychology, Psychiatry, and Pediatrics at the University of Utah. In 1976, 2 years prior to receiving the McCandless Award, he was presented the American Psychological Association Young Scientist Award. His research has been directed toward understanding the socialization processes of young children, with special interest in the father’s role in the context of socialization. Michael P. Maratsos was born on June 26, 1945, in San Francisco, California. He did his undergraduate work at Stanford University, where he received his B.A. degree in 1967. He moved to Harvard University from Stanford and was awarded his A.M. in 1968 and his Ph.D. degree in 1972. Since that time, he has been on the faculty in the Institute of Child Development at the University of Minnesota. His research efforts have been directed primarily toward an understanding of the acquisition of language by young children, which has led him to a serious study of linguistics as well as developmental psychology. He was unable to contribute to this volume because of other commitments. Among the 1979 awardees, the first, in alphabetical order, was Elizabeth Ann Bates. She was born in Wichita, Kansas, on July 26, 1947. She received her B.A. degree from Saint Louis University in 1968. She began her graduate work at the University of Connecticut and then moved to the University of Chicago, where she received her M.A. in 1971 and her Ph.D. in 1974 from the Committee on Human Development. Since 1974, she has been a faculty member at the University of Colorado. In addition, she has spent several research summers at the Institute of Psychology, CNR in Rome, Italy. Her research has been focused on the interrelations among cognitive development, symbolic capacity, and the acquisition and pragmatic use of language.

6

David S. Palermo

Tiffany Field was born on January 16, 1942, in La Crosse, Wisconsin. She was an undergraduate student at the University of Cincinnati, where she obtained a B.S. degree in 1963. She obtained an O.T.R. degree at Tufts University in 1965 and an M.A. at the same university in 1973. Between those two degrees she held several positions: she was a faculty member in the Group Psychotherapy and Group Work Training Institute of the Psychiatric Institute Foundation, Washington, D.C., Director of the Rehabilitation Center at the Psychiatric Institutes of America in Washington, and lecturer and consultant to the Educateurs at the Marseille School for Retarded and Emotionally Disturbed Children in France. In 1973 she moved to the University of Massachusetts, where she was awarded a Ph.D. in 1976. After 2 years as a faculty member at the University of Massachusetts and Research Director in the Department of Pediatrics at the Bay State Medical Center, she joined the faculty at the University of Miami Mailman Center for Child Development. Her research in recent years has focused on the psychophysiological and psychological characteristics of high-risk infants. Robert S. Siegler was born on May 12, 1949, in Chicago, Illinois. He attended the University of Illinois, where he received his B.A. degree in 1970. He enrolled at the State University of New York at Stony Brook in 1970 and was awarded a Ph.D. in 1974. He moved from there to become a member of the faculty at Carnegie-Mellon University. He spent one year at the Institute of Child Development at Minnesota, and recently he joined the faculty at the University of Chicago. His research has centered on the cognitive development of children, with particular emphasis on the development of scientific reasoning and related aspects of formal operational thinking and knowledge. AS evidenced in this volume, these seven persons are making significant contributions to developmental psychology. I look forward to the future development of these young scholars as they are joined by others who receive the honor bestowed upon them by Division 7 of the American Psychological Association through the Boyd R. McCandless Young Scientist Awards. Clearly, developmental psychology is in good hands, and our knowledge of child development will grow rapidly if these persons are representative of our future.

SOCIAL BASES OF LANGUAGE DEVELOPMENT: A REASSESSMENT

Elizabeth Bates, Inge Bretherton, I Marjorie Beeghly-Smith, and Sandra McNew DEPARTMENT OF PSYCHOLOGY UNIVERSITY OF COLORADO BOULDER, COLORADO

I. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. A BRIEF HISTORY OF THE MARRIAGE: LOVE AND EPISTEMOLOGY

.

111. COGNITIVE INPUTS TO LANGUAGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1V. SOCIAL INPUTS TO LANGUAGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

8 8 12 15

A. ATTACHMENT AND LANGUAGE .................................. B. PREVERBAL INTERACTION AND LANGUAGE . , . , . . . . . . . . . . . . . . . . . .

17 18

V. VERBAL INTERACTION: "MOTHERESE" . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

35

V1. CONCEPTUAL AND METHODOLOGICAL CONFOUNDS IN SOCIAL-CAUSAL THEORIES OF LANGUAGE DEVELOPMENT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. INTERNAL VERSUS EXTERNAL CAUSES . . . . . . . . . . . . . . . . . . . . . . . . . . . B. STRUCTURE VERSUS MOTIVATION . , . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

48 49 50

VII. DIRECTION OF EFFECTS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

54

VIII. GENETIC CONFOUNDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59

IX. THRESHOLD EFFECTS , . . . . . . , , , . . , , . . . , , , . , . . . . . . . . . . . . . . . . . . .

61

X. CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

64

REFERENCES . . . . . . . . .

68

'Present address: Department of Human Development and Family Studies, Colorado State University, Fort Collins. Colorado 80523. 7 ADVANCES IN CHILD DEVELOPMENT AND BEHAVIOR, VOL 16

Copynght 0 1982 by Academic Press. Inc All rights of repduction in any form reserved. ISBN 0-12-0097168

8

Elizabeth Bates et a1

I. Introduction Around 1960, a “marriage” was formed between child language research and linguistic theory. By 1970, that marriage was in a state of crisis, and researchers in the two fields began to move in separate directions in their quest for descriptive principles and explanatory mechanisms. In particular, researchers in child language began to look outside linguistic theory for the “causes” of development, seeking both cognitive and social influences on language acquisition and language structure. In the 198Os, there are now signs of a possible reconciliation, with researchers once again seeking a unified theory of language acquisition and language structure (e.g., Bates & MacWhinney, 1979, from the “functionalist” viewpoint; Pinker, 1979, 1980, from the “formalist” viewpoint). As any good marriage counselor could tell us, this effort will work only if what has been learned in the intervening years is incorporated into the “remarriage,” including the successes and failures of research on cognitive and social bases of language development. In this contribution, we shall examine one aspect of the “open marriage” of the 1970s. Of the two affairs, one with cognitive and the other with social theories, we shall concentrate on the search f o r social influences on language acquisition. The cognitive dalliance will be reviewed briefly, only to illustrate how easy that whole effort was. This will be a critical reassessment, documenting our own failures and frustrations as well as those of others. The point, however, is nor that the search for social bases for language was in vain. It is possible, in principle, that language acquisition proceeds on the basis of minimal input, buffered from the vagaries of a social world (e.g., Chomsky, 1975; Wexler & Culicover, 1980). However, in a social species such as ours it would be odd indeed if there were no relationship between heart and mind in the acquisition of what is, after all, a communicative system. Our purpose here is to point out just how difficult this kind of research is, describing some conceptual and methodological confusion that has plagued social input studies. Above all, we would like to point out some possible ways around these problems in the future, so that what we learn about social development can be incorporated into a new union between child language research and linguistics. We shall start with a brief history of the union, and the reasons for its dissolution. Then, after a still briefer review of research on cognitive inputs to language, we shall turn to the social issues.

11. A Brief History of the Marriage: Love and Epistemology Several historians have likened the 1960s union between child language and linguistics to a “paradigm shift” or “scientific revolution” in Thomas Kuhn’s

Social Bascs oJ Language Development

9

use of those terms (Kuhn, 1970). A comparison of successive reviews in Carmichael's Manual of Child Psychology (McCarthy, 1954; McNeill, 1970) makes this change extremely clear. Prior to 1954, most relevant studies assumed a passive model of language acquisition in which (1) the principal structures of language are directly available in the environment, in the form of words and word associations, and (2) the mechanisms of learning in general and language learning in particular (e.g., imitation and exercise, reinforcement and behavioral shaping) are also driven by the environment. These were, of course, the epistemological assumptions that underlay most of American psychology at the time. Chomsky 's ( 1957) publication of Synrucric Strucrures triggered a radical shift in the assumptions, methods, and basic epistemology of both linguistic theory and child language research. The new model stressed the highly abstract nature of grammar in natural languages, based on structures so far removed from actual language production (and hence the data available to the child) that no human child could ever acquire them without knowing in advance what kind of things she or he was looking for. Through careful attention to spontaneous speech in children at various stages (e.g., Brown, 1973), and through a variety of experimental probes (e.g., Berko, 1958), child language researchers soon established that Chomsky was at least partially right: children acquire rules rather than associations, and they formulate hypotheses and make creative errors that are only remotely related to the language spoken by their parents (for reviews, see also Dale, 1976; Slobin, 1979). To some extent, Chomsky was a Hegelian hero, triggering an epistemological shift that was long overdue, and one that was occurring in several other fields at the same time. For example, although Piaget had been writing in French since the 1920s, his work was not widely read in American psychology until the 1960s (Flavell, 1963; Furth, 1966). Similarly, the writings of European ethologists (e.g., Lorenz, 1965; Tinbergen, 1951) began to influence American comparative psychology around the same time, with passive-inductive general learning models giving way to research emphasizing species-specific, innate biases toward particular stimulus-response-reinforcement configurations (for reviews, see Schwartz, 1978; Shettleworth, 1972). Finally, computer technology in the 1950s had a profound influence on the studies of information processing in human adults, human children, and nonhuman species. Edward Tolman's ideas concerning goals, expectations, and mental representation (e.g., Tolman, 1932) were once considered radical and/or mystical. They soon became commonplace in the new field of experimental mentalism called cognitive psychology (e.g., Bruner, Olver, & Greenfield, 1966; G. A. Miller, Galanter, & Pribam, 1960). However, the fact that history had prepared us for Chomsky does not diminish the importance of his influence on the new field of psycholinguistics. Although parallel changes were taking place across the board in American psychology, the study of language acquisition and language processing (as Lashley, 1951, had predicted) provided the most detailed and compelling evidence for active, con-

10

Elizabeth Bates et al.

structive, rule-based approaches to the nature of knowledge. Although Chomsky’s Syntuctic Structures and Aspects of u Theory of Syntax make poor reading as love letters, they were the inspiration for the close relationship between child language and linguistics for at least a decade (Chomsky, 1957, 1965). What went wrong with this fruitful union? As so often happens, the initial romance had drawn much of its fervor from the battle against a common enemy: an empiricist approach to the acquisition process that denied the fundamental creativity of the child. When the battle was won, and the two fields were left alone to develop their relationship, some asymmetries of power and interest remained that were difficult to resolve. Psychologists had essentially abandoned their previous processing theory without developing another one. In the meantime, generative grammarians derived their theories-which were not processing models, but synchronic descriptions of a very abstract system-from their own linguistic intuitions. Neither the descriptive data of child language nor the experimental data of adult psycholinguistics were admitted as evidence against models of grammar. If psychologists could confirm the current grammar, so much the better. If, however, their data were not compatible with existing linguistic theories, then the data were deemed irrelevant (reflecting “performance ” rather than “competence”-for a discussion of this issue, see Fodor, Bever, & Garrett, 1974; Slobin, 1979). A further problem was posed by Chomsky’s very strong innateness hypothesis (e.g., Chomsky, 1975). It was a tenet of standard transformational grammar that grammars comprise abstract categories and operations that are only remotely related to other types of cognition, perception, or social functions. The grammar was defned as separate from meaning. This idea seems counterintuitive to many nonlinguists, who cannot conceive of grammar as anything other than a system for mapping meanings into sounds. Consider, however, the example of algebra. Algebra is a useful and beautiful system precisely because it does not take specific numerical content into account. We have rules for transforming and equating expressions that operate entirely on “ x ” and “ y ” without regard to what “x” and “y” stand for. Suppose that algebra were based on the “meaning” of those symbols, with rules written in the form “for every x, unless x is a prime number greater than 7, apply the following.. . . ” The system would be cumbersome, difficult to learn, and difficult to use. Mathematical formulations are evaluated on the basis of their generality, coherence, and elegance. The same principle applies to the selection of competing formulations in generative grammar. The less they are constrained by specific content or meanings, the better they are judged to be. Because grammars are so far removed from meaning, Chomsky concluded that children must have innate, abstract, and languagespecific knowledge to acquire them. This knowledge could not be derived or constructed out of nonlinguistic structures. The innateness hypothesis left developmental psychologists in something of a dilemma. Nothing in their armory of cognitive, perceptual, or social theories was deemed relevant to the language

Socit~lBases of Language Drvelupmcnl

11

acquisition problem. The most important and interesting aspects of language development involved presumably innate structures, which are best understood by linguists, who know how to examine such structures. Worst of all, the grammatical theories themselves kept changing. Of course, from one point of view this is a sign of progress. However, just as engineers would be very upset if Newtonian mechanics were deemed irrelevant (as opposed to relevant only at certain levels of analysis), so psychologists who had placed their faith in generative grammar were embarrassed by providing evidence for the psychological reality of a structure that grammarians had abandoned last week (Valian, 1979). As Newmeyer (1980) has described in his history of generative grammar, around 1968 a rift developed within generative grammar itself. Basically, the disagreement revolved around the role of semantic factors in the structural component of the grammar. One side-referred to for a few years as “generative semantics”-suggested that the “deep structure” of sentences might be described wholly in terms of abstract components of meaning (e.g., causation, location, object-instrument relations). The other side-referred to as “the standard theory, ” “the extended standard theory, ” “the revised extended standard theory,” and so on-suggested instead that basic sentence relations must be described with autonomous syntactic structures (e.g., “subject of a main clause”) that are interpreted by an independent semantic component but are not structured by that component. On the generative semantic side, more and more information that was previously considered “extragrammatical” was written directly into the grammar: presuppositions about real-world relationships, social intentions of the speaker, “deep” cognitive components underlying lexical items (e.g., “kill” = “cause to become not alive”). It is not our business here to describe the fragmentation and reorganization of linguistic theory that followed (see Newmeyer, 1980, for one version of that history). The important point for our purposes is the effect that this rift had on child language research. As Slobin put it, developmental psycholinguists had “pied-piped’’ after linguistic theory for so long that it was difficult to consider doing without it. Yet generative grammar had provided an ill fit to other aspects of developmental psychology. At this point at least a subset of generative grammarians were attempting to write grammars in the very cognitive socialperceptual terms that psychologists had had to eschew to apply linguistic theory. The result was that many child language researchers felt free to seek their own nonlinguistic explanations for linguistic phenomena, using a background of generative semantic theory, together with much more traditional approaches to cognitive and social development (e.g., Bates, 1976; Bowerman, 1976; Bruner, 1975a, 197513; Edwards, 1973; Greenfield & Smith, 1976; MacNamara, 1972). When psychologists began to look “outside” of language proper for functional explanations of language development, they were united by the rather general contrasts between “inside” and “outside. Chomsky had proposed that ”

12

Elizabeth Bares et al.

the child has innate, a priori knowledge and that such knowledge is language specific, that is, not derived from any other psychological domain. The new approach to child language did not deny the fundamental insight gained in the 1960s concerning the creative, active, selective role of the child in language acquisition. Children do apparently have some idea of what they are looking for. The debate now is centered on the nature of that a priori knowledge, and many theorists now argue that children could construct or derive successive theories of grammar out of various kinds of nonlinguistic cues that lead them in a game of “hot and cold” into the basic categories of language. The affairs began at this point. From 1970 onward, a large number of theoretical articles appeared suggesting some possible sources for these nonlinguistic clues to language structure. Some theorists emphasized cognitive factors (e.g., Bates, Camaioni, & Volterra, 1975; Bloom, 1973; Bowerman, 1973; Bruner, 1975a, 1975b; Edwards, 1973; MacNamara, 1972; Nelson, 1974; Ryan, 1974; Slobin, 1973). Most of them drew heavily on Piaget’s theory of sensorimotor and early preoperational development (e.g., Piaget, 1962), arguing that the categories and operations derived from 2 years of experience in manipulating objects and observing the results of actions gave the child a head start toward understanding the referential and relational functions of language. Other theorists suggested that the case for cognition was overstated, and that many clues to language structure may be based instead on the results of social trunsactions that prepare the child to understand communicative intentions (e.g., the difference between commands and declarations of fact), basic reference and symbolization (i.e., the naming game), the semantic relations underlying grammar (e.g., agent-action-receiver relations developed within exchange games with adult caretakers), and the subtle rules of taking turns in discourse (Bretherton & Bates, 1979; Bretherton, McNew, & Beeghly-Smith, 1981; Bruner & Sherwood, 1976; Lock, 1978, 1980; Schaffer, Collis, & Parsons, 1977; Trevarthen & Hubley, 1978). In short, many aspects of language were viewed as implicit in the shared procedures of early social exchange. This social emphasis-often quite critical of the cognitive work (e.g., Lock, 1980)constitutes the second love affair. In both cognitive and social areas, theory has outstripped data to a startling degree. However, for reasons that it is hoped will become clear in this contribution, the case for a cognitive basis to language acquisition has been easiest to establish. Several reviews of the “cognitive hypothesis” are now available (e.g., Bates & Snyder, in press; Corrigan, 1978). Since our emphasis here is on social factors, we need only a very brief review for purposes of comparison.

111. Cognitive Inputs to Language We can discern two general approaches to research on cognitive factors in language acquisition: processing hypotheses and structural hypotheses.

Processing approaches involve demonstrations that basic milestones in language development (e.g., first words and first multiword utterances) are brought about by changes in information-processing mechanisms that are not unique to language. For example, Piaget (1962) has argued that first words, and the naming function in general, are just one manifestation of a more general shift into symbolic representation processing in all areas of cognition including imitation and symbolic play, causality and problem solving, visual imagery, spatial relations, and knowledge of the permanence of objects. The strongest experimental test of this hypothesis would be to manipulate the supposed representational capacity directly and observe the resulting effects on early language. In principle, this manipulation would include ( 1) training the symbolic function in nonlinguistic areas, and/or (2) removing the hypothetical symbolic substrate for language. On ethical grounds, the latter experiments are thankfully out of the question, and the more positive manipulations required for training experiments have proved extremely difficult to design and execute (see Leonard, 1979, for a discussion of some limited results). For these reasons, research on nonlinguistic analogs to the emergence of naming and of sentences has had to rely on correlational methods: “What capacities appear to be yoked to the appearance of first words and/or first sentences in normal children? ” Also, research with language-disordered children, in turn, can provide “mirror-image correlations” (Bates, 1979) showing whether the same nonlinguistic developments are implicated when aspects of language are absent or impaired. Studies of both types are reviewed by Corrigan (1978) with particular regard to naming and object permanence, and by Bates, Benigni, Bretherton, Camaioni, and Volterra (1977, 1979), J . Miller, Chapman, and Bedrosian (1977), and Bates and Snyder (in press) for several aspects of sensorimotor and early representational intelligence. At the one-word stage, Piaget’s proposals concerning a general stage shift have nor been supported. That is, some aspects of sensorimotor functioning (in particular, object permanence in hiding games and spatial relations) show no correlational relationship to the emergence of first words. Other nonlinguistic measures, in particular, symbolic play with objects (e.g., touching a telephone receiver to the ear or a cup to the lips at 13 months), imitation of sounds or gestures, and certain aspects of problem solving through tool use, do correlate significantly with very early language. These are also the same aspects of nonlinguistic cognition that correlate with degree of deficit in language-disordered children (Snyder, 1978; for reviews, see also Johnston, 1981; Leonard, 1979). Such results suggest a revision of the orthodox Piagetian stage theory, in favor of a neo-Piagetian model (e.g., Case, 1978; Fischer, 1980) in which specific tusks are correlated at specific moments in development on the basis of a more limited set of shared structures. The idea of “local homologies” (Bates, Benigni, Bretherton, Camaioni, & Volterra, 1977) or “domain-specific relations” (Fischer, 1980) is best illustrated with one small finding that has appeared in three independent studies (Corrigan, 1978; Halpern & Aviezer, 1976; J. Miller et

14

Elizabeth Bates et al.

al., 1977): although general measures of object permanence do not relate to the appearance of first words, the specific ability to follow several invisible displacements of an object does seem to coincide with the child’s ability to comment linguistically on the disappearance and reappearance of objects (e.g., “all gone”). This particular example will be useful later in our consideration of social inputs to language acquisition. Although more evidence is available concerning relationships between linguistic and nonlinguistic development at the one-word stage, there is also some evidence for nonlinguistic analogs to the passage into multiword speech. Around 20 months, children begin to join two or more words into a single intonational contour, in other words, a primitive sentence (Branigan, 1979). Reports by Nicolich (1977), Fenson and Ramsey (1980), McCall, Eichorn, and Hogarty ( 1977), and Case (1980) have all indicated that children begin to combine rwo or more gesturuf schemes into a single, planned motor unit in play at 20 months (e.g., pouring, then drinking, then wiping the mouth in “pretend lunch”). In our own research comparing language and symbolic play, we have found significant correlations between multischeme gestures and multiword speech-even in partial correlations removing effects of IQ,vocabulary, overall amount of play, and several other potentially confounding variables (Shore, O’Connell, & Bates, 1981). This kind of evidence suggests that the transition from first words into syntax involves some kind of a more general shift in the “chunking” and planning of higher-order motor schemes, rather than the maturation of an innate and language-specific grammatical capacity. Processing studies like these are relevant only to some very general mechanisms shared by language and aspects of nonlinguistic cognition. At the level of specific grammatical structures, reasoning by analogy and correlations must ultimately break down. For example, specific principles that control the ordering of morphemes in English verb phrases (e.g., auxiliaries come before main verbs) surely have no convincing analogies outside of language proper. They, however, do bear a relationship to the meaning component of language, and hence, indirectly to cognition. At the level of grammatical structure, the cognitive approach focuses not on extralinguistic processing mechanisms, but on semantic bases for specific grammatical rules within language. For example, it has been argued that young children can acquire ordering rules for subject, verb, and object in sentences by assimilating them to prior knowledge of the actoraction-patient meanings that correlate with word order most of the time in natural conversation. Such correlations are probabilistic, not absolute. For example, in a sentence like, “John’s consideration of the problem perplexed his co-workers, the subject “consideration” is neither a “thing” nor an “actor. ” Sooner or later children must acquire a grammar that is sufficiently abstract to permit comprehension and production of sentences like this. However, a number of studies in the 1970s demonstrated that the early word order regularities of Englishspeaking children can indeed be described with semantically based rules like

”

Social Bases nf Language Development

15

“actor-action-recipient-patient” (e.g., Bowerman, 1974; Braine, 1976). Also, the acquisition of productive morphology (e.g., plurals, past and present tense) seems to be guided by the child’s knowledge of the meanings that are encoded by these inflections (e.g., Block & Kessel, 1980; Slobin, 1979). The conclusion seems to be that children use cngnitively based meanings to “crack the code” of grammar in their particular language. In this regard, Pinker (1980) concluded: “The close correspondence between syntax and semantics in early child language is probably the most robust empirical finding in developmental psycholinguistics in the past decade” (p. 39). Some investigators, including Pinker (1980) and Brown (1973), believe that children must eventually abandon such semantically based grammars for a more abstract system, and that this “syntactic shift” may involve innate grammatical capacities of the type foreseen by Chomsky. Others maintain that adults continue to map grammatical structures onto a functional or semantic “core” that can handle sentences like “John’s consideration. . .” by a process similar to metaphoric extension (Bates & MacWhinney, 1981; McNeill, 1979; Schlesinger, 1974). In either case, the cognitive approach to the acquisition of grammar has succeeded in reducing at least some of the mystery about the a priori knowledge that children bring into the acquisition process. Evidence for creative errors and rule generation in children need not always lead to conclusions about innate and language-specific structure. To summarize thus far, the cognitive approach to language acquisition has met with some success in the last decade. At the processing level, the rather general cognitive stage hypotheses of Piaget ’s orthodox theory have not been supported by the data on language. However, a neo-Piagetian approach, emphasizing specific language-cognition relationships, has received support at the level of processing mechanisms that underlie first words and first sentences. At the level of specific structures or content in grammar, evidence also suggests that children “bootstrap” their way into the language by assimilating syntactic and morphological structures to existing cognitive-semantic meanings. This does not mean that we have done away with the hypothesis that children have innate, language-specific knowledge (cf. Curtiss & Yamada, 1978; Pinker, 1979). However, many arguments about innateness have been offered by process of elimination: the source of the child’s creative errors could not be determined, and therefore the conclusion was that they must be innate. At least a few of these arguments by mystery have been eliminated. Let us turn now to the case for social inputs to language acquisition.

IV. Social Inputs to Language Piaget has been criticized for neglecting social factors in his theory of cognitive development (e.g., Trevarthen & Hubley, 1978; Turiel, 1978). In turn, language acquisition theories that rely on Piagetian theory are subject to the

16

Elizabeth Bares et al

same criticisms (e.g., Lock, 1978). As discussed in some detail by Bates et al. (1977, 1979), the criticisms of Piaget are more valid in spirit than in detail. Piaget never argued that social factors are secondary or derived, nor that the child builds a theory of the world exclusively out of interactions with nonsocial objects. Instead, the operative structures of Piaget ’s cognitive theory are intended to be abstract and relatively content free, transcending the distinction between social and nonsocial. Operations such as displacement and reversibility should be equally valid for predicting the behavior of animate or inanimate objects under transformation, and hence the child’s mastery of those operations could derive from interactions with people as well as interactions with things. However, most of Piaget ’s examples in three books on sensorimotor development concentrate on the child’s naive experiments with inanimate objects. Moreover, Piaget ignored (with a few exceptions) the types of predictions that hold only for animate beings, that is, the peculiar causal behavior of creatures that have a will of their own versus the passive behavior of blocks and bottles that can be thrown from cribs without protesting. Lock (1978), Bruner (1977), Bretherton et al. (1981), and others have stressed instead the important role of objects that answer back, social objects that foresee the child’s needs and anticipate his or her actions, in the mastery of communication and meaning. Werner and Kaplan ( I 963) made a much more explicit effort than Piaget to consider social factors as “equal partners” in tbe emergence of the symbolic capacity. They stressed the role of the “addressor-addressee” relationship as critical to the development of a concept of reference (i.e., the relationship between subject and object via a symbolic vehicle). A lengthy segment of their book is devoted to a discussion of the dissolution of symbol-referent relationships in schizophrenia, viewed as the product of affective disturbances. By recreating a stable interpersonal relationship, they contended, a therapist can, in turn, recreate the primordial dialogue that permitted language to emerge in the first place. Unfortunately, their proposals did not go much farther than this. How social relationships effect their causal role in creating and maintaining the bond between sound and meaning simply is not clear. Of the “classic” developmental theorists, Vygotsky (1962) was the most direct in elaborating a powerful causal role of social factors in the development of language and thought. Vygotsky essentially agreed with Piaget about the origins of “tool thought”: sensorimotor intelligence derives in the first year or so of life from the child’s own active exploration of things in the world. At the same time, however, he stressed the powerful social motivation of the infant-to imitate and interact with other human beings, sharing the world as she or he sees it. In the period from 2 to 5 years of age, the child acquires language as a tool for social interaction. Gradually, linguistic structures begin to accompany the child’s own private interactions with the world-talking out loud to oneself in play, using language to announce, “guide,” and anticipate actions to come and to note the

Social Bases of Lnnguage Developmenr

17

results of actions that have been completed. Through this process of accompaniment, language gradually takes on a governing role, structuring and directing tool thought. This governing role becomes more rapid and efficient, “talking aloud, ” in turn, becomes abbreviated, and language moves “inside” to become the master of thought from there on. Because the basic categories, relationships, and beliefs of the culture are embodied in language, and because language is acquired as a tool for interaction with that culture, the result is that thought becomes “socialized. Of these three classic approaches to the emergence of early language, Vygotsky’s theory is most compatible with research in the 1970s on the social bases of language. Three major lines of research in the Vygotskian mode will be considered here: (1) correlations between “attachment” and language; (2) studies of preverbal interaction between child and caretaker; and (3) studies of “motherese, the particular type of language that caretakers address to children in the passage from preverbal to verbal communication. In all three cases, investigators have documented in impressive detail the systematic and subtle nature of caretaker-child interaction. Hence, these lines of research are valid and important in their own right, regardless of their implications for language in the child. However, many researchers have gone on to stress how the complex structure of interaction could facilitate the language acquisition process. The problem is, as we shall see, that very little evidence exists that these phenomena do have a measurable effect on the child. The circumstantial evidence is there-motive and opportunity-but the “smoking gun” demonstrating a causal link from social input to language is still missing. ”

”

A . ATTACHMENT AND LANGUAGE

Psychoanalysts beginning with Freud have pointed to the importance of events during the oral period for the development of “ego.” Insofar as language development is a part of “ego,” the outcome of the oral period should have implications for success in language development in the first two years. From a similar perspective, investigators influenced by attachment theory (Ainsworth, 1973; Ainsworth, Blehar, Waters, & Wall, 1978; Bowlby, 1969, 1973) have hypothesized that cognition and language may be influenced by the quality (i.e., harmony) of the mother-child relationship. A detailed review of research on language, cognition, and quality of attachment is available in Bretherton, Bates, Benigni, Camaioni, and Volterra (1979). As reviewed by Bretherton and colleagues, investigations of language-attachment relationships have been based explicitly or implicitly on two hypotheses: 1 . Infants who can feel assured of their mother’s availability and responsiveness, especially in situations of stress, can devote themselves more fully and

18

Elizabeth Bates et a1

enthusiastically to interacting with the physical environment. In other words, because these infants can use the mother as a secure base from which to launch out into the world, they can learn more about the environment and properties of objects by teaching themselves. This hypothesis will be referred to as the uttuchment-exploration hypothesis. 2. If infant and mother can achieve an interactive style in which harmony and affective synchrony predominate, the infant has an increased opportunity to acquire cognitive and communicative skills through interaction with the mother. This hypothesis will be called the uttuchment-teaching hypothesis. Both these hypotheses would lead us to expect significant correlations between measures of quality of attachment and various measures of cognitive and language development throughout infancy and early childhood. Table I, adapted from Bretherton et ul. (1979), summarizes results across all of these studies, listing the kinds of measures used and correlations obtained. We need not repeat the details of this review here. For our purposes, the most striking conclusion that Bretherton and co-workers reached is that very little evidence exists jiw signijicunt correlutions between attuchment and language merisures. Also, curiously, certain cognitive measures (e.g., symbolic play) form significant correlations with both attachment and language even though attachment and language do not correlate with one another. In other words, the language and social measures must overlap with different aspects of the variance in certain cognitive skills. Because the studies reviewed in Table I do not include cases of severe deprivation (e.g., children of psychotic mothers), they apply only to variation within the normal range. Pentz (1975) has interpreted the largely negative results on language and attachment to mean that language development is biologically “buffered” against variability of social input within the normal range. If we take the attachment-exploration hypothesis seriously, we would have to conclude that the amount of security a child obtains in normal circumstances is enough to acquire language, and that more is not necessarily better. if, however, we adopt the attachment-teaching hypothesis, then we would have to conclude that traditional attachment measures (e.g., degrees of proximity to the mother when strangers approach, greeting behavior when mother returns from a brief separation) d o not capture those aspects of the mother’s teaching role that are relevant for language development. This is essentially the conclusion that Bretherton and co-workers reached in their review, leading us to consider other aspects of mother-infant interaction in the first year. B . PREVERBAL INTERACTION AND LANGUAGE

A quite different approach to possible relationships between language and social interaction has emerged from microanalyses of infant social behavior in

TABLE I Findings of Studies Relating Strange Situation Classifications and Ratings with Cognition and Communication ( I Age at strange situation

Relationship of strange situation classifications or ratings with *

At age (months)

Significant relationships with standardized cognitive tests. Piagetian scales. and play measures I2 DQ (Griffiths Scale) I2 12 DQ (GrifFiths Scale) 8.5, 1 1 , 15 IQ (Stanford-Binet) 30 12 DQ (Bayley) 20.5 11.5 Person permanence 10. I 1 11 Person permanence 8 . 5 . 1 1 . 14.5 14.5 11.5 Object permanence 14.5 II Object permanence 17 Means-end relations (Uzgiris-Hunt) 1 1 . I2 18

12 12

18

I2

Tool use (with mother’s help) Quality of exploration Quality of cognitively mature play (combinatonal and symbolic play) Duration of symbolic play bouts Level. breadth. and frequency of symbolic play Frequency of combinatorial play

24 20.5 I2

24 1 1 . I2

Reference

Sample size

Bell & Ainsworth (1972) Bell (1978)

23 33

Main (1973) Bell (1970) Bell (1978) Bell (1 970) Bell (1978) Bretherton, Bates. Benigni, Camaioni. & Volterra (1979) Matas, Arend. & Sroufe (1978) Main (1973) Harmon. Suwalsky. & Klein (1979)

40 33 33 33 33 25

Matas. Arend. & Sroufe (1978) Brethenon, Bates. Benigni. Camaioni , & Volterra ( 1979)

48 25

48 40 36

I1 (corzririued)

TABLE I (continued) Age at strange situation

N

0

Relationship of strange situation classifications or ratings with

Significant relationships with communicative gestures and language 12 Clarity and variety of cornmunicative signals 12 Communicative gestures

12

12 28

Observed vocabulary size and imitation of words uttered by mother Number of morphemes per utterance Percentage of utterances that are questions

At age (months)

11

18 I2

DQ (Cattell) IQ (Stanford-Binet) DQ (Bayley) 1Q (Stanford-Binet) DQ (Bayley) DQ (Bayley)

Sample size

9-12

Bell & Ainsworth (1972)

23

1 1 , 12

25

18

Brethenon, Bates, Benigni, Camaioni, & Volterra (1979) Connell (1977)

20.5 36

Main ( 1973) Pentz (1975)

40 31

Clarke-Stewart (1973) Connell ( 1977)

38 26

Bell ( 1 978)

33

No significant relationships with cognitive tests, Piagetian scales, and play measures 12 DQ (Bayley) 11.5, 18 12

Reference

14

30 24 36 24 3. 8, I 1

Matas. Arend, & Sroufe (1978) Hock (1976)

55

48 164

12 12

12 12

Object permanence, space imitation (Uzgiris-Hunt) Object permanence, combined means-end + space scale Level and duration of symbolic play Combinatorial play

No significant relationships with language measures 12 Language competence 28 MLU. imitation of maternal utterances Comprehension Percentage of utterances that are questions Number of utterances 12 Referential versus expressive style. comprehension of commands Number of different words uttered. I2 number of words per utterance 12 Comprehension, referential and nonreferential words (maternal interview and observation variables)

10. 11, 12

25

13

Bretherton, Bates, Benigni, Camaioni, & Volterra (1979) Clarke-Stewart (1973)

20.5

Main (1973)

40

10, 12

Bretherton, Bates, Benigni, Camaioni, & Volterra (1979)

25

18 28, 36

Clarke-Stewart (1973) Pentz (1975)

38 31

36 28 28, 36 18

Connell ( I 977)

44

20.5

Main ( 1 973)

40

10. 1 1 . 12

Bretherton, Bates, Benigni, Camaioni. & Volterra (1979)

25

38

"Adapted from Bretherton, Bates, Benigni, Camaioni, & Volterra (1979. Table 5.2). 'In all cases where significant relationships were found, securely attached infants (strange situation classification B) always performed at a level superior to anxiously attached infants (strange situation classifications A, C). In those studies where interactive ratings were used. proximity and contact seeking during the reunion episodes of the strange situation correlated positively with cognitive performances, whereas avoidance-resistance correlated negatively with cognitive performance.

22

Elizuberh Bures et al

the first few weeks and months of life (Brazelton, Koslowski, & Main, 1974; Kaye, 1977, 1981; Sander, 1977; Stem, Beebe, Jaffe, & Bennett, 1977). Investigators in these studies have capitalized on new technologies (e.g., videoanalysis, autonomic recording), which permit very subtle dimensions of interaction to emerge: body postures, rhythms, direction of gaze, facial expressions, etc., in reciprocal exchange between mother and child from their very first encounter. A great deal has been learned in a short time about the infant’s predisposition to social interaction (e.g., an early preference for human faces, early recognition of the human voice) and the surprising degree of organization that can be seen in mutual gaze patterns, smiling, vocalization, touching, and virtually every aspect of social exchange in the infant-caretaker relationship. Some implications of these findings for subsequent language development are summarized in volumes edited by Lock ( 1 978), Schaffer (1977), and Lewis and Rosenblum (1977). Although details vary, the logic of the argument generally goes like this: I . The complex structure of mother-infant interaction in the preverbal period must serve some function for our species. 2. A compelling analogy can be observed between preverbal and verbal interaction along several dimensions including turn taking, shared reference to objects and events, the “semantic” structure of certain games and routines such as object exchange (i.e., exercise of agent-action-recipient-patient relationships), the “pragmatic” structure of communication (e.g., “performative” or speech act functions evolving in preverbal requests, commands), and even in “syntactic” structure (nested social sequences and recursive interaction ‘‘ruI es ’’) . 3. Given the above, at least one function of preverbal interaction may be to lead the child into the structure of language. As Shatz (in press) noted in her critique of social-causal theories, this approach to language acquisition is based primarily on “existence” or “plausibility ” proofs: certain phenomena that could affect language development can be shown to exist. Hence, we hypothesize that these phenomena do have an effect on the acquisition process. Bruner and colleagues have been particularly explicit about the different roles that preverbal social interaction could play in language development (Bruner, 1975a, 1975b; Bruner & Ninio, 1978; Ratner & Bruner, 1978). The following quotations characterize some of his theoretical claims (Bruner, 1977): Language acquisition occurs in the context of an “action dialogue” in which joint undertakings are being regulated by infant and adult. The joint enterprise sets the deictic limits that govern joint reference, determines the need for a referential taxonomy, establishes the need for signal-

23 ling intent, and eventually provides a context for the development of explicit predication. Thr cvolrctictn of language irselj; notably its 14tiivrrs.crlstructures, prubrrbly reJI1,ct.sthe reyuirivnents r ~ f ’ s u cjuint / ~ crction in our spvl-1ec.ie.s.(pp. 287-288, italics added) Mother and child develop a variety of procedures for operating jointly and in support of each other. At first, these joint actions are very direct, specially geared to assistance and comfort. In time, thc two of them develop conventions and requirements about carrying out joint tasks. The struc.tura of’ these tusks mcry shupe the structure of itiiricil gruntmur by the nature of the juintly held concepts it imposes. (p. 274, italics added)

Newson ( 1978) was somewhat less explicit about the direct analogy between grammar and interaction, but was equally emphatic about the causal role of communicative exchange: From the baby’s point of view it is only by being continually involved, as a participant actor, within an almost infinite number of such sequences that he is finally brought into the community of language. In short it is only because he is treated as a communicator that he learns the essential human art of communication. (p 42)

As we shall point out in more detail later, several very distinct hypotheses are actually included in these broad theoretical proposals. All of them, however, lead us to expect significant positive correlations between quality, amount, and type of mother-infant interaction in the first year and subsequent developments in language. What is the evidence? Unfortunately, few of the investigators involved in microanalytic studies of interaction also went on to test the relationship to language in a longitudinal, correlational study. This is understandable, in that these microanalyses are in themselves exhausting and time consuming. The interaction studies are of enormous value in their own right, regardless of their implications for language. Nevertheless, some very strong predictions have been offered that must ultimately undergo an empirical test. Table I1 is a summary of all the studies that we have been able to locate that included preverbal interaction measures similar to those discussed by Bruner, Schaffer, Trevarthen, and others, and included correlations of these measures with various aspects of language and general cognitive development. Many of these studies were designed for a different purpose, for example, to investigate social class differences, high- versus low-risk infants, or very general effects of “good mothering” on “infant competence, ” without offering hypotheses about specific language structures. Hence they constitute, in some cases, only an indirect test of the social-causal model. To help the reader to evaluate these findings, we have listed the kinds of measures used in each study in as much detail as possible, with separate sections for statistically significant and nonsignificant findings. As with attachment-language findings, the results in Table I1 are extremely disappointing for social-causal theorists. The significant relationships that have

TABLE I1 Preverbal Interaction Variables Related to Later Language and Cognitive Abilities A.

Preverbal variables Reciprocal interaction

Studied by

Specific measures a

Tulkin & Covitz (1975)

x length of

Cohen & Beckwith 1979)

3 longest interactions, interaction sequences, reciprocal vocalization Social

Cohen & Beckwith 1979)

Play factor Social Play factor

Mutual

Ramey , Farran, & Campbell ( 1979)

gaze factor Mutual social factor Mutual interaction (frequency and duration)

Significant results

At age

N

Context

Related to what later variables

Pos . or neg .

At age

30 (middle class)

Home

ITPA

+

6 yr

1 mo

50

Home

+

24mo

1 mo

50

Home

Gesell DQ, Sensorimotor scale' Gesell DQ, Sensorimotor scale Receptive language " Bayley DQ Gesell DQ Bayley DQ

+ +

+

24mo 24mo 24mo

+

24mo

+ +

+

25mo 24mo 25mo

+

24mo

+

36mo

1Omo

3 mo

50

Home

8 mo

50

Home

Gesell DQ

6mo

29 (exp. group)

Lab free Play

StanfordBinet

Maternal responsivity and attentiveness

Tulkin & Covitz (1975)

Kaye (1980)

Maternal "interactiveness"

Freedle & Lewis (1977)

Tulkin & Covitz (1975)

Response to fret, reciprocal vocalization (5%) Animated facial activity to infant Initiate vocal bouts ( S ) Entertains infant, gives object

30 (middle class)

Home

ITPA, PPVT "

+,+

6yr

6mo

50

Face-toface Play

PPVT

+

34mo

3 mo

97

Home

MLU

f

2 yr

30 (middle class)

Home

PPVT

+

6 yr

97

Home

MLU

-

2 yr

30 (middle class)

Home

PPVT

-

6yr

28

Lab free Play

StanfordBinet

+

36mo

101110

10mo

VI h)

Maternal lack of responsivity

Freedle & Lewis (1977)

Allows infant to end vocal bout (%)

3 mo

Maternal restrictiveness

Tulkin & Covitz (1975)

Prohibition ratio

10mo

Physical contact

Ramey, Farran, & Campbell (1979)

Physical contact (frequency and duration)

6mo

(control group)

(conrinued)

TABLE I1 (conrinued) A.

Preverbal variables

Studied by Tulkin & Covitz (1975)

Specific measures "

X !ength

Significant results At age 10 mo

30

Total time holding infant Kay (1980)

Frequency

6 mo

50

Ramey, Farran, & Campbell ( 1979)

Frequency and duration

4 mo

28

Tulkin & Covitz (1975)

Frequency

Ramey, Farran, & Campbell (1979)

infant plays alone (frequency and duration) Time over 2 ft away, time in playpen Time over 2 ft away

h)

OI

Time no1 interacting

Tulkin & Covitz (1975)

Context Home

(lower class) 30 (middle class)

held

Maternal vocalization

N

10 mo

(control group) 30 (middle class)

Face-toface Play Lab free Play Home

Related to what later variables

Pos . or neg.

PPVT, ITPA

-

PPVT

t

Language production

-

StanfordBinet ITPA

6 mo

28 (control group)

Lab free Play

StanfordBinet

10 mo

30 (middle class) 30 (middle class)

Home

PPVT

ITPA

At age

6 yr

+

26 and 36 mo(comDined) 36mo

+

6 yr

+

36mo

B.

Preverbal variables Reciprocal interaction

Studied by Cohen & Beckwith (1979)

Specific measures

Nonsignificant results

At age

Mutual gaze factor

I mo

Mutual gaze factor Mutual social factor

3 mo

N 50

Context Home

8 mo

4 N

Burchinal & Farran (1980)

Farran & Ramey ( 1980)

Bakeman & Brown (1980)

Kaye ( 1980)

Mutual play (frequency and duration) Dyadic involvement factor Amount joint activity Amount reciprocal interaction

6 mo

51

6 mo

60

0. I. 3,

43

9 mo

6 mo

50

Lab free play Lab free Play Home feeding

Face-toface Play

Related to what later variables Gesell DQ. sensorimotor scale, receptive language Bayley DQ

24 mo

Sensorimotor scale, receptive language Bayley DQ Bayley DO Stanford-Binet

24 mo

Bayley DQ Stanford-Binet

6. 18 mo 48 mo

Bayley DQ Stanford-Binet

12, 24 mo 36 mo

Conversational ability, language production PPVT. puzzle test

26. 30 mo

25 mo

25 mo 6 . 18 mo 24 mo

34 mo

TABLE 11 (continued) B.

Preverbal variables

Studied by Rarney, Farran, & Campbell (1979)

Tulkin & Covitz (1975)

N

m

Maternal responsivity and attentiveness

Freedle & Lewis (1977)

Specific measures Mutual interaction (frequency and duration) R length of 3 longest interactions, interaction sequences reciprocal vocalization VOCd responsivity

Kaye ( 1980)

Bakeman & Brown (1980)

Cohen & Beckwith (1979)

Vocal and emotional responsivity Responsive holding factor

Nonsignificant results

At age

6 rno

N 28 (control group)

10 mo

30 (lower class)

Context

Related to what later variables

Lab free Play

Stanford-Binet

Home

ITPA, PPVT

30 (middle class)

36 mo

PPVT

6 mo

97

Home

MLU

24 mo

6 mo

50

Face-toface Play

26. 30 mo

0, 1 , 3, 9 mo

43

Home feeding

Conversational ability, language production PPVT, puzzle test Bayley DQ Stanford-Binet

1 mo

50

Home

Gesell DO, sensorimotor scale, receptive language

24 mo

34 mo 12, 24 mo 36 mo

Burchinal & Farran (1980)

Tulkin & Covitz (1975)

Social and vocal responsivity factor Contingent responsivity

3 mo

6 mo

51

Responsivity to 10 mo fretting (I), vocal I ) responsivity ( Responsivity to fretting ( I )

30

Lab free Play Home

Maternal restrictiveness

Tulkin & Covitz (1975)

Cohen & Beckwith (1979)

Entertains, gives objects to infant

Control factor

25 mo

Bayley DQ Stanford-Binet

6, 18 mo 24 mo

PPVT, ITPA

(lower class) PPVT

30 (middle class)

wvr. ITPA

Vocal responsivity (76) Maternal “interactiveness”

Bayley DQ

10 mo

3 . 8 mo

30 (lower class) 30 (middle class)

Home

50

Home

ITPA. PPVT

ITPA

Gesell DQ. sensorimotor scale, receptive language Bayley DQ

24 mo

25 mo (continued)

TABLE I1 (continued) B.

Preverbal variables

Studied by Tulkin & Covitz (1975)

C W

Maternal noti restrictiveness

Physical contact

Specific measures Prohibition ratio

Nonsignificant results

At age 10 mo

N

Context

30 (lower class) 30 (middle class)

Home

Related to what later variables ITPA, PPVT

At age 6 Y'

ITPA

Ramey, Farran. & Campbell (1979)

Absence of punishment

6mo

57

Lab free Play

Stanford-Binet

36 mo

Cohen & Beckwith (1979)

Floor freedom factor

8mo

50

Home

24 mo

Tulkin & Covitz ( I 975)

Time with no barriers

10 mo

60

Home

Gesell DQ, sensorimotor scale, receptive language Bayley DQ ITPA, PPVT

25 mo 6 Yr

Ramey, F m a n , & Campbell (1979)

Physical contact (frequency and duration) Responsive holding factor, stressful holding factor

6mo

29 (exp. group)

Lab free Play

Stanford-Binet

36 mo

1 mo

50

Home

Gesell DQ, sensorimotor scale, receptive language

24 mo

Cohen & Beckwith (1979)

2 N 1ci

8

m

% W

8 0

31

TABLE U (continued) B.

Preverbal variables

N w

Time !lor interacting

Studied by Ramey, Farran, & Campbell ( 1979)

Burchinal & Farran (1980) Bakeman & Brown (1980)

Tulkin & Covitz (1975)

Specific measures Infant plays alone (frequency and duration) Noninteraction time Independent infant activity Time over 2 ft away, time in playpen Time over 2 ft away Time in playpen

Nonsignificant results

At age

N

Context

Related to what later variables

At age

Stanford-Binet

36 mo

29 (exp. group)

Lab free.

6mo

51

0, 1.3,

43

Lab free Play Home feeding

Bayley DQ Stanford-Binet Bayley DQ Stanford-Binet

6, 18 mo 24 mo 12, 24 mo 36 mo

Home

ITPA

6 Yr

6mo

9 mo 1Omo

30 (middle class)

30 (lower class)

Play

PPVT

ITPA, PPVT

Maternal involvement

Ramey, Farran, & Campbell (1979) Burchinal & Farran ( 1980) Bakeman & Brown (1980)

Maternal teaching

2

Variability of interaction

HOME‘ maternal involvement Passive involvement Amount maternal ’ ‘effon ’ ’

6mo

57

Lab free Play

Stanford-Binet

36 mo

6mo

51

0, 1 , 3 , 9 mo

43

Lab free Play Home feeding

Bayley DO Stanford-Binet Bayley DQ Stanford-Binet

6, 18 mo 24 mo 12, 24 mo 36 mo

Gesell DQ, sensorimotor scale, receptive language Bayley DQ Bayley DQ Stanford-Binet

24 mo

Bayley DQ

12, 24 mo 36 mo

Cohen & Beckwith (1979)

Intellectual stimulation factor

8 mo

50

Home

Burchinal & Farran (1980)

Toy demonstration

6mo

51

Lab free

Bakeman & Brown (1980)

Same

Play 0, I , 3, 9 mo

43

“These measures are those reported by individual investigators. bIllinois Test of Psycholinguistic Abilities. ‘Casati-Lezine Sensorimotor Series (Kopp, Sigman, & Parmelee, 1974). “Labeling production (Beckwith & Thompson, 1976) and comprehension. ‘Peabody Picture Vocabulary Test. ’Caldwell (1970) Home Observation for Measurement of the Environment Inventory.

Home feeding

25 mo 6, 18 mo 24 mo

34

Elizobeth Butes et al.

been reported are mostly in the predicted direction: “better” relationships correlate with “better” or more precocious linguistic-cognitive development. However, nonsignificant results outweigh the positive findings. Given the wellknown bias of editors against the publication of nonresults, a conservative conclusion would be that the effects of interaction variables on language development are either very weak or very difficult to prove for some other reason. Could we argue that these studies failed to include the relevant dimensions of preverbal interaction, that is, the ones that Bruner, Schaffer, Trevarthen, and others have delineated in terms of analogies to language? Certainly the studies reported can be claimed to have face validity with the variables discussed by most social causal theories: time spent in mutual gaze, frequency and duration of interactions around a toy or other object, amount of “animated facial activity of mother to infant,” and so on. Of all these longitudinal investigations, the most relevant for social bases of language may be Kaye’s (1980) longitudinal study of 50 infants from birth through 34 months. Kaye has been one of the pioneers of microanalysis in mother-infant interaction, and his measures include many different dimensions of exchange during breast feeding, early caretaking, and play. Altogether 48 interaction variables were chosen for his longitudinal analyses, including those that had proved to be interesting in his own previous microanalytic studies and variables that had been described by other researchers. Language measures through 34 months of age included the Peabody Picture Vocabulary Test, measures of language production, and “conversational ability” (involving number of conversational turns taken by the child, linked to utterances by mother and experimenter). When effects attributable to educational status of the mothers were partialed out of 788 longitudinal correlations, only 50 relationships, 6.7%, were significant at the .05 level. Kaye’s (1980) own conclusions were as follows: Considering that many of these tasks and contingency scores were selected deliberately because of their structural similarity to one another-that is, their face validity-we have to conclude that the lack of continuity among the measures is remarkable and compelling. Beyond the social class differences within our sample (and most of our variables were free even of those differences), we have found practically no relationship between the individual performances of our mothers and infants in any one situation at one age and their performances in another situation at another age. (p. 13)

Kaye’s interpretation echoes Pentz’ (1975) explanation of similar results for the attachment-language studies: the evolutionary “fit” of mother to child has become so good that everybody does what has to be done to get communication under way. Variations within the normal range have no measurable effect on basic developments in language. We shall return later to some further methodological and conceptual problems in interpreting the disappointing results summarized in Tables I and 11. At this point, let us turn to the third major line of social-causal research: the effects of “motherese” on language acquisition.

Socicrl Buses i>f Lcinguaxe Development

35

V. Verbal Interaction: “Motherese” One of Chomsky ’s main points concerning the innateness of grammar was that linguistic input to children is much too poor to yield the rich grammatical theory that children ultimately derive: A consideration of the character of the grammar that is acquired, the degenerate quality and WdrrOW!y limited extent of the available data. the striking uniformity of the resulting grammars, and their independence of intelligence, motivation, and emotional state over a wide range of variation. leave little hope that much of the structure of the language can be learned by an organism initially uninformed as to its general character. (1965, p. 58)

However, a large body of research on the nature of adult speech to children has demonstrated that language input is much better than Chomsky believed. Comparisons of adult speech to other adults with their speech to very young language-learning children show that children receive a simplified, repetitive, and often exaggerated form of the adult code (for reviews, see Berko-Gleason & Weintraub, 1978; Snow & Ferguson, 1977). Modifications occur at every level of the system: intonation, phonology, syntax, semantics, gesture, and broader aspects of the organization of discourse. To offer just a few of the more exotic examples, Williams ( 1979) has shown that Spanish, English, and Chinese mothers tend to exaggerate precisely those phonological distinctions, in their respective languages, that prove in spectrographic analysis to be the most difficult to discriminate. Obviously, mothers do not have access to spectrographic analyses in advance, and so the process must be as “natural” as it is precise. Mandarin is a tone language, using the pitch or tone level of a word to make semantic distinctions. Mothers speaking Mandarin apparently emphasize the most difficult tone distinctions for their children. Since the semantic use of tones is a relatively rare feature in human languages, this “decision” by Chinese mothers must be based on a powerful and pervasive tendency that is applicable to m y feature of language that might pose problems for the child. Newport, Cleitman, and Gleitman (1977) have called this special type of language “motherese”-although research has shown that the language is spoken by fathers, strangers, and even 4-year-olds interacting with 2-year-olds (Shatz & Gelman, 1973) and 4-year-olds pretending to talk to 2-year-old dolls (Sachs & Devin, 1976). Because these modifications are so pervasive, some investigators have suggested that motherese evolved in our species as an unconscious, spontaneous “teaching language” (Moerk, 1975). This is essentially the same kind of plausibility argument described earlier for phenomena in preverbal interaction: motherese exists everywhere, and hence must exist for some reason; motherese ~ ~serve l asdan aid to language acquisition, and thus we hypothesize that it does have an effect. But does it? Unfortunately, the existence of the phenomenon has been documented much better than its purported function. A number of investigators

TABLE 111 Talk to Children: Summary of Effects and Noneffects Categorical adult variable General verbal stimulation

Specific adult variable Frequency/ variety of maternal speech

Total number of words spoken by

Experimental method

Age of child

Regression and correlation: infant and mother simultaneously sampled (sample of low socioeconomic status, balanced by race and sex; all firstborns)

17-17.5 mo

Cross- lagged correlations

T , = 10.512.5 mo T 2 = 1718 mo

Correlations between maternal and child measures

T, = 28 mo T , = 36

N

Child measures

Significance-direction

Studied by

Relationship of stimulation Clarke-Stewan 36 Language comto language and general (1973) petence factor, competence: positive which included Within a regression on comprehension. child competence, verbal expressed vostimulation loaded most cabulary, and highly and was particularly response to related to the language questions in a subfactor probe, as well as spontaneous measure of number of words used (part of a general competence factor) Clarke-Stewart Positive relationship; 36 Bayley mental ( 1973) apparent direction of scales effect mother to child; author extrapolates to language per se from Bayley. since language measures were not comparable from T, to T 2 . Pentz (1975) Mother at 28 mo with 31 28 and 36 mo: 36 mo comprehension: amount of speech produced; frepositive

mother

at T , ; cornlations of maternal measures at TI and T2 with child at T , (sample white and middle class)

mo

Noncontingent speech (maternal initiation) Contingency/ responsivity (proportion of child utterances to which mother responds plus proportion of maternal utterances that are responses)

W

4

Syntactic complexity

MLU

Partial correlation with child variables 6 mo later, controlling

T, = 1227 rno

Others: ns quencies of questions and imitations; response contingency; mean length of utterance (MLU) 36 mo only: rest of linguistic comprehension (syntactic contrasts) Language factor: positive

15 MLU Auxiliaries Noun phrase Inflectionsverb phrase

Clarke-Stewart ( 1973)

All ns

Pentz (1975)

All ns

Newpon, Gleitman, & Gleitman (1977)

(continued)

TABLE 111 (continued) Categorical adult variable

W

Specific adult variable

Experimental method for age and initial child level (sample = middle-class girls) Correlations between simultaneously sampled mother-child measures (sample: middle class and advanced) Correlation: simultaneously sampled mother, infant, and older children from rural Africa: Kokwet, Kenya; includes 13 mothers and 12 older children Correlation of child MLU with earlier maternal measures

Age of child

N

Child measures

Significance-direction

Studied by

Verbs-utterance Noun phrasesutterance Varies: 19-32 mo

24-42 mo

T , = 18 mo T , = 27 mo

10 Receptive score MLU Comprehensibility Typeitoken ratio Longest utterance Age 20 MLU and adjusted MLU

MLU

MLU, receptive scores, longest utterance. and age: positive. Comprehensibility and typeitoken: ns

Cross ( 1977)

Mother-infant and older child-infant: positive for both variables

Harkness (1977)

Negative

Furrow ( 1979)

Difference in maternal child MLU

Long utterances ( S ) Single-word utterances (76) Propositionshtterance

Noun phrasesutterance Noun phrase length in morphemes

Maternal at 28 mo with 36 mo comprehension: positive Others: ns Receptive scores, longest utterance, and MLU: negative Others: ns MLU for mother-child at 28 mo, for mother-child at 36 mo, and for mother at 28 mo and child at 36 mo: negative Imitation for mother-child at 28 mo: negative Others: ns Typekoken: ns Others: positive Typeltoken: negative Others: ns Receptive and comprehensibility: positive Others: ns All ns All ns

All ns. 28 mo maternal with 28 mo MLU: negative Others: ns

Pentz ( 1975)

Cross (1977)

Pentz ( 1975)

Cross (1977) Cross ( 1977) Cross (1977)

Furrow ( 1979) Newport, Gleitman, & Gleitman (1977) Cross (1977)

Pentz (1975)

(conrittued)

TABLE I11 (conrinued) Categorical adult variable

Specific adult variable Modifiers/ utterance Pronouns1 utterance

Noun/pronoun ratio Verbs/ utterance

Verb phrase length in morphemes Reverb complexity

Specific linguistic forms

Declarative

Experimental method

Age of child

N

Child measures

Significance-direction

Studied by

All ns

Furrow (1 979)

Receptive, cornprehensibility: positive Others: ns MLU: negative All ns

Cross ( 1977)

MLU: negative Mother-infant: positive Older child-infant: ns MLU and adjusted MLU All ns

Furrow (1979) Harkness (1977)

Age: positive Others: ns

Cross (1977)

All ns

Newport, Gleitman, & Gleitrnan (1977) Harkness ( 1977)

Older child-infant: positive for both Mother-infant: both ns All ns All ns

Furrow (1979) Pentz ( 1975)

Pentz ( 1975)

Cross (1977) Furrow (1979)

Questions Yes-no questions

Imperative

P

Negative imperative WH-questions

Maternal at 28 mo with 36 mo comprehension test Mother-infant and older child-infant: both ns Number of auxiliaries/ verb phrase: positive Others: ns All ns MLU: positive Number of auxiliariesiverb phrase: negative Others: ns All ns All ns Others: ns 36 mo with 36 mo maternal response contingency: negative Older child-infant: negative for adjusted MLU but ns for MLU Mother-infant: both ns All ns

Pentz ( 1 975)

Ns

Furrow (1979) Newport, Gleitman, At Gleitman ( 1977) Cross (1977)

All ns

MLU and age: negative Others: ns

Harkness ( 1977) Newport, Gleitman, & Gleitman (1977) Cross (1977) Furrow (1979) Newport. Gleitman, & Gleitman (1977) Cross (1977) Furrow ( I 979) Pentz (1975)

Harkness (1977)

Cross (1977)

(continued)

TABLE 111 (conrinued) Categorical adult variable

Specific adult variable Where is NP? What’s that?

P N

Auxiliaryfronted questions Tag questions Rising intonation questions Questions Interjections Deixis

Auxiliary deletions in yes-no questions Semantic Imitations relationship of adult utterances to child utterances

Experimental method

Age of child

N

Child measures

Significance-direction

Studied by

All ns Typdtoken and age: ns Others: negative All ns

Cross ( 1 977) Cross ( 1 977)

All ns All ns

Cross (1977) Cross (1977)

All ns All ns All ns Inflectionshoun phrase: positive Others: ns All ns All ns MLU: positive

Pentz (1975) Cross (1977) Cross ( 1977) Newport. Gleitman, & Gleitman ( 1977) Cross ( I 977) Pentz ( 1975) Furrow (1979)

All ns

Cross ( 1 977)

Older child-infant: both negative Mother-infant: both ns

Harkness (1977)

Cross ( 1977)

Paraphrases

Transformed repetitions Total repetitions (includes above types plus four others: partial exact, sequential, and nonsequential) Expansions

Experimental: expansion. modeling, and control 60.5 HR sessions/ 12 weeks Experimental: 12.5 hr sessions

Varies, 29-37 mo

12 6 measures including MLU and imitation

Varies, 30-46 mo

24 Sentence imitation

Semantic extensions Recasts

Experimental 3 groups (recast new sentence. control)

Varies. 32-40 mo

27 MLU (in words). no. elements/ noun phrase.

Receptive, longest utterance, MLU: negative Others: ns Typdtoken: negative Others: ns Receptive, longest utterance. MLU, comprehensibility, age: negative Typeltoken: ns

Cross ( 1977)

All ns

Newport, Gleitman, & Gleitnian (1977) Harkness ( 1977)

Mother-infant: negative Older child-infant: ns MLU and adjusted MLU All: negative All ns except imitation: modeling and control equivalent; expansions lowest Ns

Cross ( 1977) Cazden ( 1965)

Feldman (1971)

Cross (1977) Receptive: positive (with pronominal extensions) Others: ns Recast significantly higher Nelson. than controls on all measures Carskaddon, & but MLU and noun phrase Bonvillian ( 1973) recast versus new sentences

TABLE 111 (continued) Categorical adult variable

Specific adult variable

Experimental method

Age of child

intervention for I 1 weeks. 20 min/session twice weekly (middle class)

P

Linguistic teaching devices

Experimental: 2 groups both without complex questions or complex verb phrases, groups given recasts of one type or the other Sampled over 15 min: due to low frequency. a composite measure was devised. including expan-

2.5 yr

N

Child measures

no. verb elementsherb construction. auxiliaries/ verb construction. sentence imitation I2 Complex verb phrase and complex question usage

Significance-direction

Studied by

approaches significance on verb and auxiliaries; new sentences and controls not significantly higher on any measure

Children receiving verb recasts produced significantly more complex verbs than "question" children and vice versa

Nelson ( 1977)

28 mo maternal Pentz (1975) with child 28 mo MLU: positive Number of utterances, questions. and response contingency: positive 36 rno maternal with 36 mo

sions, recasts, corrections, occasional questions, prompts and checks

Function of adult utterances

imitation and response contingency: positive Others: ns

Semantically new utterances Novel, isolated utterances Continuous dialogue

All ns

cross ( 1 977)

All but typehoken: positive Typeitoken: ns

Cross ( 1 977)

Mother-infant and older child-infant: both ns

Harkness (1977)

Direct request

Mother-child both at 36 mo Pentz ( 1 975) tendency to respond verbally to mother: negative Others: ns All ns

$ Indirect request Elicit information

Focus attention

Labeling

All ns Mother-infant and older Harkness (1977) child-infant: negative MLU and adjusted MLU Mother-child both at 36 mo Pentz (1975) number of child utterances and tendency to respond: negative Others: ns Pentz ( 1975) All ns (corititiued)

TABLE 111 (conrinued) Categorical adult variable

c P n

Age

Specific adult variable Property of object Instruction in object use Reinforce utterance Reinforce act Verbalize child act

Verbalize child feeling Verbalize maternal act

Verbalize

Experimental method

of child

N

Child measures

Significance-direction All ns

All ns All ns

All ns MLU at 36 mo: positive with 36 mo maternal verbalization of either child or maternal act (combined measure) Others: ns All ns All ns MLU at 36 mo: positive with 36 mo maternal verbalization of either child or maternal act (combined measure) All ns

Studied by

maternal feeling Play idea Engage in play Prohibition Permission Score of referential field

All ns All ns All ns All ns

Immediate references

All ns

Nonimmediate events

Receptive. longest utterance, MLU, and age: positive Others: ns Comprehensibility: positive Others: ns

Persons/ objects present Child events

Mother events Child and mother events “Adapted from McNew (1981a). *Details of each experiment are given the first time it appears in the table

Receptive, longest utterance. and comprehensibility: negative. Others: ns All ns Age: negative Others: ns

Cross ( 1977)

48

Elizabeth Bates el al.

have attempted to assess the relationship between adult input and child language development. The correlational and experimental findings are summarized in Table 111, adapted from McNew (in preparation, a). Only a handful of correlational studies exist, and these vary considerably in method and range of measures. Compared with the preverbal results in Tables l and 11, more significant, positive findings in studies of motherese can be observed. However, as we shall point out in more detail later, these results are often very difficult to interpret. Critical methodological and statistical controls are sometimes missing, and the available methods usually do not allow causal inferences. One conclusion is clear from the studies in Table 111: more languuge input from adults is related to more and better language in children. For example, Clarke-Stewart ( 1973) reported significant positive correlations between the total amount and variety of language stimulation provided by the mother, and several measures of linguistic competence in the child at the same session. Pentz (1975) reported significant relationships between frequency of maternal stimulation at early sessions, and measures of child comprehension 8 months later. Finally, as reviewed in Bates ( 1975), children whose linguistic input comes primarily from adults (as opposed to peers and older children) are at an advantage in language learning. However, these quantitative relationships do not prove the structural, “teaching” claims of the motherese literature that simplification, repetition, and exaggeration clarify the language learning process. To establish such claims, relationships must be shown between specijk q p e s of language input and spec $ ~outcomes in acquisition. To summarize, although the social-causal theories are at least as convincing on conceptual grounds as the cognitive-causal theories discussed earlier, their empirical grounding is less clear-cut. Shall we conclude that social inputs to language are minimal? It may be that the preverbal and verbal interaction phenomena uncovered in the last few years exist primarily in the service of socialization itself; variations in language acquisition may not be affected beyond the minimal requirement that children be exposed to some language input. We think, however, that this conclusion is premature. To evaluate the findings summarized in Tables 1-111 in more detail, we need to consider some specific conceptual and methodological difficulties that may be resolvable in future work.

VI. Conceptual and Methodological Confounds in Social-Causal Theories of Language Development Thus far, we have assumed a distinction between social and cognitive influences on language which could be defined as follows: 1 . By “social causes” we refer to the effects of interactions with and knowledge of animate beings.

Social Bases of Language Developmenl

49

2 . By “cognitive causes” we refer to the effects of interactions with and knowledge of inanimate beings. However, this animate-inanimate distinction is not all that separates the two lines of research reviewed so far. Many studies of social factors in language development also involve the following conceptual and methodological confounds: 1 . an epistemology emphasizing “internal” as opposed to “external” causal-

ity; 2 . a failure to distinguish structural and motivational contributions; 3 . the question of direction of effects between mother and child; 4. a confound between genetic and environmental variance; 5 . the issue of “threshold effects.” We shall consider these issues one at a time, discussing some possible ways around each one. A.

INTERNAL VERSUS EXTERNAL CAUSES

Most of the research on cognitive bases of language assumes an epistemological model in which the child plays an active role in scanning the environment and assimilating language to processes and structures that transcend the particulars of language proper. The environment must furnish language data; however, cognitive theorists place little emphasis on the role of the environment in selecting or highlighting particular structures. It is, of course, true that the child’s cognitive clues to language are ultimately derived from interactions with the external world (as Piaget has argued for decades). From the very beginning, however, the child’s knowledge of external objects results from her o r his own actions. The causal flow is from “inside” to “outside,” a kind of reverse S-R model in which the child stimulates and the environment responds. In social-causal research, in contrast, some of that emphasis on active, childdriven processes has been lost. Shatz (1981) has noted that social-causal theorists “focus less on the child and more on the role of external elements as important factors in acquisition” (p. 2). Hence, some of this literature represents a twofold break with Chomskian psycholinguistics: ( 1 ) a rejection of the innateness hypothesis for language-specific structures, and (2) a move away from the emphasis in the 1960s on active selective processes that are initiated and executed by the child. In contrast, cognitive-causal theorists have broken only with the innateness principle. The new empiricism is particularly strong in the motherese literature, which emphasizes an environmentally driven process in which caretakers “preprocess ” language for the child by eliminating difficult forms and highlighting and exaggerating those forms for which the child is ready. The flow of causality here

50

Elizubeth Butes et al.

is from “outside” to “inside,” an approach that is epistemologically much closer to the kinds of S-R models that were attacked by psycholinguists in the I 960s. The identification of “social ” with “external ” is not logically necessary. What we really have are two logically independent dimensions: child-driven versus environmentally driven learning mechanisms, interacting with social versus nonsocial objects of knowledge. The apparently greater empirical success of cognitive versus social-causal models may have nothing to do with the relative contribution of nonsocial and social knowledge to language learning. Instead, the difference between the two lines of research may revolve around their relative emphasis on internal causality. When we correlate the child’s progress in language with factors in the child, we may obtain better results than we do when progress in language is correlated with factors in the environment. The way out of this problem for social-causal studies may be to focus less on the correlation between the behavior ofsocial objects and child language, and begin focusing on the correlation between the child’s knowledge of social objects and related changes in language. We shall return to this point later. B . STRUCTURE VERSUS MOTIVATION

A further difficulty in research on social bases of language involves a failure to distinguish between structure and motivation, that is, between what the child knows and is mble to do (structure) and what the same child wants and is willing to do (motivation). This problem applies both to cognitive-causal and socialcausal research, but it is particularly salient in the social domain, in which emotional factors are of paramount importance. The contrast between structure and motivation is valid first at the level of data collection in particular contexts. For example, timid children may perform badly in experiments and in naturalistic observations when strangers are involved. In our own recent longitudinal study, we used both observational data and maternal interviews to measure aspects of language development such as vocabulary and the use of multiword speech. In addition, we asked mothers to fill out an infant temperament inventory (Rowe & Plomin, 1977), which included a “sociability” scale. From the viewpoint of method, the temperament questionnaire and the language interview should have shared the largest amount of ‘irrelevant” task variance (the quality of observation and the tendency of mothers to answer questions quickly, thoroughly, conservatively, etc.). In other words, we might have expected stronger correlations between “sociability’’ and language as reported by the mother than between “sociability” and the language observations. In fact, the opposite was true. The sociability scale bore no relationship at all to the language measures obtained in the interview but was associated with a number of language observations, with the reportedly “more sociable” children

performing better (Bretherton, McNew, Snyder, & Bates, 1980). This finding suggests that language competence measures obtained through observations are confounded by motivational factors to a greater extent than language competence estimates from diaries and maternal report. Hence, it may be useful in some studies to remove irrelevant temperament effects statistically, through partial correlations. The motivation issue is also valid at a more general level, with effects across many contexts. That is, a given mother-child couple may “look bad” in a language experiment not because of a specifically linguistic difference in their interactions, but because they get along badly at every level. In other words, the child’s apparent incompetence at language, and/or the mother’s apparent “linguistic insensitivity, ” may be epiphenomena1 of a generally bad relationship. For example, in the study by Tulkin and Covitz (1975) summarized in Table 11, a significant negative correlation was observed between “prohibition ratio” at I0 months of age and performance on the Peabody Picture Vocabulary Test at 6 years. Do we really want to argue that saying “no” to children decreases their vocabulary? It is likely that the link is much more indirect: a relationship characterized by high rates of saying “no” may be one in which (because of the child’s character, the mother’s, or both) very little decent conversation ever gets under way. The motivational confound may be superficial; that is, if we measured the child with his or her other parent, babysitter, or some other more compatible partner, we might obtain higher estimates of language competence. However, the motivational confound might have interacted to reduce “true” competence; because the relationship was so bad, the child’s progress in language has suffered across the board. We have no way of knowing which interpretation is correct from performance during a “one-shot ” set of observations. We can bring the structure-motivation distinction to bear on the motherese literature and on the multifaceted proposals by Bruner, Schaffer, Trevarthen, and others concerning the potential relationship between preverbal interaction and language development. At least four different kinds of causal hypotheses emerge.

I . Bctsic motivution to interact. When a warm relationship is established between parent and child, the child is more likely to observe, imitate, and above all interact with adults. In other words, motivational variables insure a willingness to participate in social exchange in the first place. From that point on, more direct causal influences can take place. 2 . Attention-direc.tinS functions. Once the child is engaged in interaction, mothers may provide a “scaffolding” (Wood, Bruner, & Ross, 1976) for the optimal use of infant attention. Through joint activities, the child comes to understand maternal gestures (e.g., pointing, direction of gaze toward referents) and intonation patterns (e.g., the tone of voice used in requests for action versus

52

Elizabeth Bares et al.

commands to stop action). The child’s attention then can be directed more easily toward the particular objects and events underlying adult speech. In a sense, this form of facilitation is a mechanism halfway along the structure-motivation dimension that we have drawn here, that is, motivation of interest and direction of attention toward structure. 3. Internalization of language by unulogy. Bruner, in particular, has argued that a literal structural resemblance between aspects of preverbal exchange and aspects of language can be observed. These analogies apply at the levels of pragmatics, semantics, and syntax alike. For example, semantic relations expressed in language (agent-action-receiver-patient) are implicit in games of object exchange, and syntactic operations like the embedding of clauses in complex sentence structures are directly prepared by the embedding of one action sequence in another during mother-child play. Because such analogies exist, the child can derive certain basic linguistic structures directly by internalizing the rules of preverbal exchange. In other words, the child goes from “doing” to “having” basic structures (see also Bates, 1976, Chapter 1). 4. Direct provision of structure. Internalization by analogy is an indirect process, one that involves very little “teaching” by the mother. In addition, claims have been offered in both the preverbal and the motherese literature that mothers preprocess and time linguistic structures directly so that the child will receive them when she or he is ready to understand them. Not that mothers supposedly read texts on grammar to carry out this work; rather, in the process of trying to get a conversation going with the child, mothers necessarily “scale down” their utterances to the point where the child can just understand them. As a result, language input is timed to be just one step ahead of the child’s productive abilities. What we have here, then, is a gradation from indirect to direct causal relationships between the social environment and language learning. At the indirect end, motivation plays a larger role. It is not always an easy matter, however, to keep these factors empirically distinct within a given study. In the prohibition-vocabulary relationship just mentioned, it seems obvious that the intervening causal connection must have more to do with emotional-motivational factors than with a specific effect of one kind of language structure on another. Shatz (1981) and Newport et al. (1977) have argued that direct linguistic effects are more easily established if a facevalid connection between maternal and child variables exists. For example, Newport and co-workers reported significant positive correlations between the number of auxiliary fronted questions used by the mother at Time 1 (e.g., “Are you making a tower?”), and use of auxiliary verbs by the child at later sessions. This is such a specific relationship that it seems reasonable to interpret it as a modeling effect. However, consider the following motivational interpretation.

Sociul Buses of L a n p u g e Development

53

I . In an harmonious relationship, in which partners are actively interested in each other’s affairs, more conversation about mutual activities is likely. 2. In conversations about each other’s activities, a high proportion of progressive verb forms is likely to occur (“What are you doing?”-‘‘I am doing . . . ”). 3. Hence, the correlation between auxiliary use in mother and child could be epiphenomena1 of the conversational topics selected in different kinds of relationships. Both the modeling hypothesis and the conversational topic hypothesis are equally plausible when all we have are correlational data, unless we have some way to partial out those motivational effects that d o not interest us in a given study. To summarize thus far, we have delineated three logically independent dimensions that have been confounded in some of the literature on social bases of language: social versus nonsocial knowledge, internal versus external causality, and structural versus motivational influences. A hypothetical space comprising these three dimensions is shown in Fig. 1. A given instance of language learning undoubtedly involves a complex interaction of all three. It is possible, however, at least in principle, to determine the degree to which different factors affect the course of acquisition, and hence to locate a given instance at some point in this STRUCTURE

INTERNAL CAUSALITY

SOCIAL OBJECTS

EXTERNAL CAUSALITY

NONSOCIAL OBJECTS

MOTIVATION/ ATTENTION

Fig. I . The sociul bases of Iunguage: three independent dimensions

54

Elizcibeth Bares et al

three-dimensional space. If we cannot find some way to separate these influences empirically, we run the risk of generating conceptually uninterpretable results.

VII.

Direction of Effects

One of the most serious problems in interpreting interaction data is the problem of direction of effects. Implicit in some of the studies summarized in Tables I and I11 is the belief that positive correlations reflect a causal effect of the mother on the child. However, we have just as much reason to believe that such correlations reflect an effect of the child on the adult. Let us consider how the issue of bidirectional effects applies to the literature on motherese. Ferguson (1976) has noted that simplifications, repetitions, and exaggerations occur in the speech of adults to any listener whose ability to comprehend is in question: foreigners with poor control of English, adults with hearing difficulties, retarded adults, and normal listeners in a noisy environment. The purpose of these adjustments is to make ourselves understood. The decision to use “motherese” then rests on some kind of cue indicating comprehension difficulties in the listener. Insofar as the use of speech adjustments is a function of the listener’s inability to comprehend, motherese is essentiully an indirect tneusure of childfnilure. In fact, the “causal flow” creating a significant correlation between child competence and maternal speech could even be unidirectional, from child to parent. At least in principle, we could obtain such correlations even if parents had no effect on their children at all. In this case, however, the correlations should be in a negative direction, that is, more motherese, less language tluency in the child. For instance, Bohannon and Marquis (1977) report two studies in which adults responded to naturally occurring or fabricated “incompetence” in children. In each case the adults reduced the lengths of their utterances when children signaled lack of comprehension. Thus, with less comprehension there was more “motherese.” In some studies, however, positive relationships have been reported. Here, too, the direction of effects is an issue. A precocious child is capable of understanding more complex speech, by definition. Hence, if the precocious child does receive more complex speech from her or his parents, the direction of effects could have been one in which the child elicits a type of language from the adult, and not vice versa. The direction of effects problem may also create false negative, that is, nonsignificant, correlations, when in reality a reciprocal causal influence is the case. While motherese may be an indirect measure of child failure, it may also be an aid to comprehension. Thus, any positive influence thut mothrrese may have on the child is canceled out by the original negutive relationship between motherese crnd childfiiilure. This confound was made particularly clear to our research team in a study by Carlson-Luden ( 1979) of causal understanding and maternal teach-

ing style with 10-month-old infants. Carlson-Luden’s infants were given a series of three causal tasks to solve, involving a lever that set some kind of dynamic effect into motion. Mothers were free to demonstrate the toys for the child at their own discretion. We had assumed in the usual fashion that “good teachers” would produce “good learning.” To our surprise, a large set of significant negative correlations was observed between various maternal interventions and measures of success in the 48 children who participated in the study. Does this mean that maternal teaching impedes learning? That is unlikely. The most sensible interpretation is that mothers intervene to the degree that their children fail to understand the task. Hence, this nonverbal form of motherese was actually an indirect measure of infant incompetence, resulting in negative correlations. It might be that good teachers do ultimately get their points across to children. However, if the positive effect of intervention cancels out the negative effect of responding to child failure, then we have nothing more than a zero correlation. From this viewpoint, the facilitative effect of mother and child would be difficult to detect in any correlational design. The direction of effects problem is particularly serious if measures of mother and child are taken at the same point in time, when the dyads are interacting. For example, Cross (1977) reported a large number of significant positivc correlations between maternal and child speech, including mean length of utterance (MLU). However, she failed to control for both differences in age and differences in overall ability across children. Hence, positive correlations could reflect the fact that older and/or more precocious children elicit more complex speech from their mothers-rather than any kind of causal effect of the mother’s speech on the child. In contrast, Furrow (1979) reported a significant negative correlation between maternal MLU at 18 months and child MLU at 27 months. This result was interpreted to mean that mothers who simplify their speech at the earlier stages produce children who are more precocious later on. However, Furrow did not control for linguistic differences among children at the earlier sessions. Newport et al. (1977) partialed out the child’s initial level of competence at Time l in a set of analyses similar to Furrow’s and found no significant relationship between maternal and child MLU when this control was used. This procedure seems a useful one in that it avoids the problem of the relationship of the child to herself over time. However, if the mothers’ own use of motherese at Time 1 was a reaction to the child’s level of competence at Time 1 , and that reaction provides the best input for the child’s next steps, then equating children for initial language level may result in removing the variance that we are interested in (McNew, 1981b). When a couple is locked into sequences of bidirectional effects, partial correlations removing any link in the chain may result in removing everything else of interest from there on. One method that has been offered as a solution to the bidirectional effects issue is the cross-lagged correlation. Both mother and child are assessed at Time 1 and Time 2. Hence, every correlational comparison involves correlations among all

56

Elizaberh Bates et a1

M = mother C = child

1 =Time 1 2=Time 2

Fig, 2 . Illustration of 11 cross-lugged correlutional design.

four poles, as shown in Fig. 2. We assume first that correlations between measures taken in the same session are uninterpretable in terms of direction of effects ( M , - C, ; M, - C2), Second, we also assume that mothers and children can be consistent with themselves over time even if they have no effect at all on one another. Hence, the two horizontal correlations (MI - M 2 ; C , - C,)are not relevant to causal effects between mother and child. To assess direction of effects in an interaction over time, we are interested in the two diagonal correlations: M I - C, and C, - M 2 .If the mother’s influence on the child is greater than the child’s influence on the mother, then the two diagonal correlations should differ significantly and the correlation should be the larger. Although the logic of the cross-lagged method is appealing, some serious problems still remain. First, many researchers fail to meet all the statistical assumptions of the model (Rogosa, 1981). For example, to test the difference between the two diagonals, there should not be a significant difference between the other two interactive correlations (i.e., mother with child within sessions). Second, the cross-lagged panel method requires very large samples for sufficient power to test the differences between correlations. Many investigators are apparently unaware that the sample size needed to create significant individual correlations is far smaller than the sample that is usually needed for two individual correlations to be significantly different from one another. The sample sizes in the studies reviewed in Tables 1-111 rarely exceeded 30-50.The difference between correlations would have to be very large (around .20) to be interpretable. Finally, the cross-lagged method does not get us out of the “canceling out” problem described above. If motherese is an indirect measure of child failure, but also a facilitative measure, then the diagonal correlations in a cross-lagged design might end up around the zero level even though “real” causal influences were going on. In his review, Rogosa suggested that no current methods are adequate to the task of determining predominant cause from correlation. The most we can establish is that two phenomena vary together in a meaningful and nontrivial way.

Social Bmes of Langucigc Development

57

Two other alternatives to the direction of effects problem are available. One approach is illustrated by Zukow, Reilly, and Greenfield (1981) in a study of how children understand offers. In a detailed microanalysis of behavior sequences in mother-child interaction, these investigators isolated every naturally occurring instance of successful or unsuccessful offers, examining the specific adult behaviors that preceded and followed each instance. Hence, they were able to identify “packages” of adult behaviors (establishing eye contact before an offer is made, calling the child’s name, etc.) at different points in development, demonstrating a reduction in the number of elements that were necessary for successful comprehension at each point. These sequential analyses preserve the real-time causal flow, including the number of child failures that preceded adult interventions. This study strengthens the case that a mother’s use of vocal and gestural “supports” is an indirect measure of how much support the child needs, that is, of child failure. However, if the details of behavioral sequencing are preserved, then it should be possible to separate out supports that occur a f e r child.failure from “free” supports or additional supports above and beyond child failure. That is, we could remove the variance from “failure-induced motherese” and assess the facilitative effect of additional support independently. Unfortunately, the kind of exhaustive microanalysis carried out by Zukow and colleagues is very time consuming, making i t difficult to obtain the number of cases that would be needed for correlational analyses. To establish causal relationships, the best alternative is to abandon naturalistic designs for an experimental approach in which adult interventions are manipulated directly as independent variables. McNew (1981b), reported a study of the effect of maternal gestures on child comprehension, using a method that stands midway between naturalistic and experimental approaches. Infants 20 months old were given standardized presentations of pairs of culturally familiar and unusual commands (e.g., “Kiss the baby” versus “Kiss the ball”). After the standardized presentation of the noncanonical commands by the experimenter, the infants’ mothers were asked to “Get her/him to kiss the ball any way that you would normally get her/him to understand.” In this situation of up-to-themoment information about the child’s comprehension, maternal use of demonstrations was negatively related to previous compliance. Thus, the child had an effect on the mother. McNew also analyzed the likelihood that these maternal demonstrations facilitated later success, partialing out child performance on the standardized trial. In other words, two kinds of variance were removed: ( 1 ) the effect of the child on the mother in “provoking” demonstrations; and (2) the likelihood that children who succeed once can succeed twice. In this second analysis, maternal demonstrations had a significant positive correlation with the child’s eventual success. In other words, gestures do facilitate comprehension. Belsky, Goode, and Most (in press) have also demonstrated a facilitative effect of maternal intervention on infant play and exploratory competence, using an experimental technique to increase maternal interventions beyond “baseline”

58

Elizubrth Bures et al.

levels. In an initial session, measures of maternal intervention and child exploration were taken for both control and experimental couples observed in routine household interactions. Then, in three succeeding weekly sessions, mothers in the experimental group received positive feedback whenever they gave their children explicit and helpful stimulation. This stimulation included tendencies to point to or reposition objects, demonstrations and support for the child throughout an action, verbal supports of instructing and questioning, highlighting object properties, and naming objects. Note that the mothers were not trained to carry out any particular kind of intervention. Rather, they were reinforced for aspects of their own spontaneous behavior. Control mothers received no such feedback. One week after the experimental-control sessions, Belsky et ul. observed both groups again, and found a significant increase in the use of facilitative interventions by the experimental mothers compared with control mothers. Finally, in a follow-up two months later, they reported that children of mothers in the experimental group engaged in significantly more exploratory play and investigations of the unique properties of objects. These results reached significance at the .05 level with a one-tailed test. This is not a robust effect, but it was obtained after quite minimal intervention, and the effects appeared after a rather long lag time. Clearly, mothers can increase their interventions beyond spontaneous or “natural” levels. Furthermore, such increases appear to have positive effects on their children. Perhaps the strongest experimental evidence supporting a causal effect of motherese has been provided by Nelson, Carskaddon, and Bonvillian (1973) and by Nelson (1977). An earlier study by Cazden (1965) had shown that “expansions” by adult experimenters in a preschool setting had no effect on language development. An expansion is a restatement of a child’s limited utterance, in a correct and more complete syntactic form, for example, “He making a picture,” “Oh, he’s making a picture.” Nelson and co-workers camed out a similar experiment, using a different type of expansion which they called a “recast.” A recast sentence preserves the basic semantic relations in the child’s previous utterance, but adds further syntacric information, for example, “I want some milk,” “You want some milk, don’t you?” (see Berko-Gleason & Weintraub, 1978, for a further discussion of the Nelson et al. technique). Cazden’s expansions were essentially corrections of the child’s utterances; the Nelson et al. recasts provided syntactic variation and elaboration on a semantic theme selected by the child. Nelson and co-workers found that 3-year-olds who received recast input for weeks obtained higher scores than children in a control group on five different posttest measures of syntactic development. The original recast experiment showed that syntactically varied feedback can have a generully positive effect on syntactic development. In a sense, this is an experimental version of a fact that we already knew from the correlational literature: mvre linguistic input is related to mvre language development. We still do

not know whether any kind of a direct causal link exists between specific structures. In this light, the follow-up study by Nelson (1977) is particularly important. Nelson selected a group of children 2% years old who failed to produce either complex questions or complex verb phrases in a pretest speech sample. The children were then randomly assigned to one of two training groups, with experimental sessions spanning 2 months: half of the children received recast sentences involving complex questions, while the other half received recasts involving complex verb forms. By the end of the training, children in the verb group produced significantly more complex verbs and significantly fewer complex questions than children in the question group. In other words, the respective experimental interventions had very specific structural effects. The Nelson studies are important on two grounds. First, they support the kinds of specific correlational relationships reported by Newport and co-workers, but with stronger evidence for direction of effects from parent to child. Second, the learning process in the recast experiments is essentially child driven. That is, the child selects the semantic theme, and the adult merely provides variations on that theme. This was also true of Cazden’s expansions. However, her expansions corrected the child’s syntax, and Nelson’s recasts respected the child’s formula while providing further information on how the same theme could be expressed. Two problems affect the generalizability of experimental findings like these. First, these are short-term effects that may not correspond to long-term processes in the real world. Second, we do not know how these artificial changes correspond to the spontaneous behavior of caretakers in natural settings. Nevertheless, specific changes in adult speech can clearly have a positive effect-if the adult gives the child what she or he is looking for. If we put together converging evidence from naturalistic, correlational studies and experimental manipulations, we may obtain the best possible estimate of how mothers and children affect one another in language development.

VIII.

Genetic Confounds

Thus far, we have considered whether a correlation can be interpreted as an effect of the mother on her child, or an effect of the child on her or his mother. There is yet another way to interpret significant correlations between maternal behavior and child language: Similarities within mother-child puirs may reflect nothing more thon shared genetic variance. If verbal fluency is genetically based, and/or if verbal measures are influenced by general intelligence, then positive correlations between mother and child on a variety of verbal measures may have nothing to do with causal influences of one partner on the other. Rather, they may simply mean that smarter mothers have smarter babies. One solution to the genetic confound is to obtain a measure of maternal

60

Elizaheih Bates et al.

intelligence, and remove that variance through partial correlation. However, no one really knows what variables “intelligence” comprised as it is measured by standardized tests, and therefore we run the risk of overkill, wiping out interesting and real causal effects in our efforts to remove an uninteresting one. Consider the following causal chain:

1. The real factor influencing the child’s language level is not the genetic variance in IQ inherited from parents, but the quality and sensitive timing of input that the child receives. 2. However, smarter mothers are more sensitive to the child’s needs and hence more likely to provide this high-quality input. 3. We measure maternal sensitivity and child language across a given time span, and carry out partial correlations between maternal and child variables, removing the variance due to maternal intelligence. 4. Because we have removed variation in intelligence, we also remove variation in maternal sensitivity. 5 . We obtain no correlation between maternal speech and child language, and conclude incorrectly that infant competence is unrelated to maternal input. A better alternative to the problem of genetic confounds is a behavior-genetic design, in particular, an adoption study in which the genetic contribution of the biological parents and the environmental contribution of the adoptive parents are clearly separated (Plomin, DeFries, & McClearn, 1980). We are aware of only one study of this type using the kinds of maternal and child measures relevant to current social-causal theories of language development. Hardy-Brown, Plomin, & DeFries (1981) have examined several aspects of language and communicative development at 1 year of age, in a sample of 50 infants who were adopted at birth. Measures of the child included a word diary collected by the adoptive parent, and several observational measures of frequency and quality of speech and gestural communication (taken from videotapes). Correlations of the child measures with one another were high, justifying the use of a child communicative competence factor (the first principle component of a factor analysis of child measures), in addition to the individual measures, in correlations with adult variables. For the biological mother, the measures included a battery of standardized cognitive tests (visual memory, verbal fluency, and the like), which yielded four specific cognitive abilities factors and one general factor (first principle component). The biological fathers were usually not available for testing. For the adoptive parents, the same battery of standardized tests were administered to both father and mother. In addition, 15 measures of maternal style and sensitivity (similar to many of the measures in Tables 1-111) were taken from videotaped interactions between the child and adoptive mother. Finally, scores for socioeconomic and educational status of both adoptive parents were included.

Socrirl Buses of Language Di~veloptnent

61

All of these measures were entered into correlational analyses with the child variables. Because there were so many more measures for the adoptive parents, the case is clearly biased toward uncovering environmental effects. Nevertheless, Hardy-Brown et a l . ’s results provide much more support for biological effects on language and communication at I2 months of age. Direct comparisons between genetic and environmental influences on rate of communicative development in the first year are provided by relating cognitive abilities of the birth and adoptive mothers with the communicative ability of the adopted infant. Of 84 correlational relationships between cognitive assessments of the infant 19% were significant. In comparison, 4% of these relationships were significant between the infant and adoptive mother, and 6% were significant with regard to the adoptive father. This average of 5% for both adoptive parents can be expected on the basis of chance alone, and suggests that brighter parents may not environmentally enhance the communicative development of their children in the first year. Since biological mothers contribute only half the genetic input to the child, the total amount of genetic input to child competence at I year of age is probably even greater than indicated by the above results. Furthermore, of the positive results obtained with adoptive parent measures, almost all came from the videotaped interaction variables. Considering the points we raised earlier about direction of effects, the few correlations that were obtained with adoptive parents could actually reflect an effect of the child on the parent and not vice versa. Hardy-Brown et a / . noted carefully that environmental effects may take longer than 1 year to establish. Communicative developments at 12 months of age may be tied to a child-driven, maturational schedule; similar measures at 2 and 3 years of age may be more strongly influenced by the environmental contribution of adoptive parents. Nevertheless, in view of the large number of theoretical claims that have been made about “scaffolding” effects in the first year of life, HardyBrown’s findings strike a cautionary note.

IX. Threshold Effects One further problem arises in correlational and experimental studies alike, with and without behavior-genetic strategies to remove genetic confounds: the problem of threshold effects. Almost all the studies that we have reviewed thus far assume a linear, cumulative relationship between input and output: if some maternal input is good, then more is better. This may not be the case at all, however. Perhaps the amount and type of social input that is necessary for normal language development is a minimal threshold amount that every normal child receives. Increases in such input beyond the threshold amount may have no effect at all, so that correlational and experimental measures of input variation in

62

Elizubeth Bates et al.

the normal range yield largely nonsignificant results. This is the kind of difficulty that Pentz (1975) and Kaye (1980) alluded to in discussing the “buffering” of language development, and the optimal “fit” of mother to child across our species. To illustrate the threshold problem, let us start with an extreme example. Virtually no individual variation exists in the tendency for human beings to be born with two arms and two legs. Hence, if we were to take a measure of two armedness in parents and their children, we would find no individual variation in the measures. If no variation exists, no correlation is observed. Could we conclude from the lack of correlation that parental armedness and child armedness are unrelated? Bates (1979) offered another extreme example that pertains to two measures taken within the child. It is certainly the case that having a functioning heart and liver is a prerequisite to the acquisition of language. Any child born without a heart and liver will not live to acquire language. However, if we take a sample of 12-month-olds and measure “degree of organ ownership” and correlate that measure with language, we will obtain a zero correlation. Because no variance exists in the organ measure, no statistical link to the variance in language at 12 months is found. Certainly we would not want to conclude that the two are unrelated. It is simply the case that the threshold amount of heart and liver necessary for language is usually established in children by 12 months of age. The threshold problem is obvious in the case of measures that show no variance at all. It is less obvious when two measures (e.g., maternal sensitivity and child language) d o show a large amount of normally distributed variation at the point in development that interests us. It may still be the case that the relevant variation in maternal sensitivity is very small, so that no one in the sample falls below threshold. Increases beyond that amount are irrelevant, and vary independently from individual differences in child competence. In fact, evidence f o r child effects on caretakers leads to the hypothesis that human children seize and create for themselves some minimal amount of communicutive interaction. Some findings by Newport et al. (1977) are particularly relevant in this regard. As noted earlier, they reported only a few significant correlations between maternal and child speech, after the variance due to differences in child ability at Time 1 were removed. However, the relationships that did reach significance formed a coherent pattern: (1) they were all very specific connections between similar aspects of English grammar, for example, auxiliary-fronted questions by the mother versus development of auxiliary verbs in the child; (2) such specific correlations were found only for aspects of grammar that are peculiar to English. This pattern led Newport and colleagues to argue that universal aspects of language are ‘‘environmentally insensitive, ” while language-specific structures may be “environmentally sensitive. ” They concluded that, “the mother has

little latitude to teach her child about the nature of language; but she can at least improve his English” (Newport et a/., 1977, p. 147). The concept of environmental sensitivity can be related to the threshold issue as follows. English-specific structures such as the different forms of auxiliary verbs carry very little semantic content-at least, not content that is critical to a 2-year-old. These structures may be environmentally sensitive because children are not looking for them. Indeed, if they are going to notice these forms at all in the early stages, it will have to be because parents are hitting them over the head with them. Hence, we could call these “high-threshold” forms. In contrast, if environmentally insensitive structures are the ones that encode universal meanings, then children may acquire them with no more than some minimal threshold amount because they art actively scanning the environment for precisely those forms. Hence, the distinction between high- and low-threshold structures, and environmentally sensitive and insensitive forms, rests on the distinction made earlier between passive and active processes in language acquisition. The threshold issue in language acquisition also brings to mind Harlow’s studies of social development in nonhuman primates. Harlow and Harlow (1969) have shown that infant monkeys that are deprived of normal parenting and/or social experience with peers fail to function normally as adults in sexual encounters and in behavior toward their own offspring. In other words, social experience in infancy is crucial to normal social development. However, this requirement becomes obvious only in extreme cases in which threshold amounts of early social experience are denied. The same message may apply to the study of low-threshold, environmentally insensitive aspects of human language. If we focus exclusively on variation in the normal range, we may nor detect some critical relationships. Fortunately, cases of extreme deprivation Li ki Harlow are rare in our species. We do know that “wild children” who are badly abused or neglected in childhood fail to develop language properly (e.g., Curtiss, 1977). However, these children are so badly impaired across the board that particular effects on language are difficult to separate from general social and mental impairment. However, some less extreme “natural experiments” do permit us to assess certain kinds of social inputs separately from others. For example, blind infants are unable to respond to a wide array of social signals that are available to the sighted child-no matter how skilled or well intentioned the child’s mother might be. Studies by Urwin ( I 978), among others, demonstrate that this deprivation leads to marked delays in the earliest stages of language acquisition, when the “idea” of intentional communication through symbols is first established. Similarly, institutionalized children in the 1930s and 1940s were deprived of many aspects of normal social input, even though their own senses were intact. Spitz (1965) has argued that such deprivation often leads to depression and even marasmus if the deprivation is too prolonged. Under these circumstances language has little chance. Fortunately for our species, children usually see to it that adults provide the

64

Elizabeth Bares et al.

essential daily requirements for adequate communicative development. If more and better input is forthcoming, certain aspects of language may indeed improve. The child may ultimately move from conversationalist to poet in response to a verbally enriched environment. However, supermothers are not essential for the species to go on talking.

X. Conclusion We have concluded that the case for externally driven, structural effects on language development is not very good. Research on social factors in language acquisition has concentrated primarily on these kinds of effects, and as a result, social-causal theories have not yet obtained adequate empirical support. We do not want to argue that such approaches are uninteresting or wrong, and that the social environment plays no role in language development beyond passive provision of minimal language data. Instead, we have tried to point out why social effects are so difficult to demonstrate. Several partial solutions have been offered. First, observational and correlational studies should be supplemented with converging evidence from experiments in which types of adult input are manipulated systematically. Second, it may occasionally be possible to disentangle issues in direction of effects, and/or motivational confounds, through partial correlations from which the variance due to confounding factors has been removed (e.g., “sociability” in infants as it relates to observational measures of language). Third, behavior-genetic designs are useful in removing genetic confounds in correlational data. Fourth, the linear view that “if some input is good, then more is better” may apply only to certain kinds of high-threshold, environmentally sensitive language structures. When the threshold amount of input required for normal development is low, we may have to abandon studies of the normal populations in favor of research in clinical settings. That is, cases of extreme deprivation (blind infants who cannot see their mothers’ cues, psychotic mothers who d o not give their children normal cues) may tell us more about what happens when critical amounts of social input are missing. However, the major recommendation that we would like to make is for social-causal theories that reverse the causal flow, an organismically driven approach emphasizing the role of social factors in the child on language development in general, and in particular on those aspects of language that encode uniquely social meanings. We can offer five specific recommendations for areas of research in which social developments in the child contribute to the acquisition of language. 1 . Imitation as a species-specific motive. Imitation was not a popular topic ~ because it was identified with developmental research during the 1 9 6 0 ~perhaps the kinds of passive, mechanistic, environmentally driven processes that charac-

Social Boses of Language Development

65

terized social learning theory in the 1950s (e.g., Mowrer, 1960). However, as we have discussed in more detail elsewhere (Bates, 1979, Chapter 7), imitation is actually one of the most mysterious processes in all of cognitive development. For example, Meltzoff and Moore (1977) have provided some evidence (albeit controversial) that human neonates are capable of imitating facial movements. Chevalier-Skolnikoff ( 1977), in a comparison of human cognitive development with related achievements in other primates, concluded that imitation separates human from infrahuman more than any other area of sensorimotor development in the first two years of life. We are much better at “aping” than the apes. We know less than we should about the role of socially motivated imitation in the acquisition of languages (for reviews, see Bloom, Hood, & Lightbown, 1974; and R. Clark, 1977). Our own longitudinal studies have yielded significant correlations between rate and quality of imitation in infancy, and many aspects of early gestural communication and language (Bates et af ., 1979). At later stages in development, R. Clark (1977) and Fillmore (1976) have demonstrated how imitation can be used by some children to acquire and use certain linguistic forms long before those forms have been analyzed into a set of productive rules. Ochs (1979) has pointed out a wide variety of social functions that are served by repetition of prior discourse, including holding one’s place in a conversation and confirming the speaker’s point. Finally, considerable individual variation exists in the degree to which children use imitation as a communicative tool. Bates (1979), Kempler (1980), Bloom et al. (1974), and Horgan (1979) have all suggested that this variation in the uniquely social activity of imitating others might be related to a variety of other individual differences in the acquisition of grammar, including relative use of pronouns versus nouns and the degree to which children acquire complex syntax through idioms or formulas. 2. Shared reference as a species-specific motive. Bates er d. (1379) and Hardy-Brown et al. (1981) have shown strong correlational relationships between language development at 13 months and several aspects of preverbal communication: giving, pointing, showing, ritualized requesting (e.g., requesting by means of an abbreviated open-and-shut hand gesture). Bates, Carnaioni, and Volterra (1975) have suggested that such gestures may be a kind of “protodeclarative, ” exercising the social function of shared attention and reference to some object or event long before language is available for encoding reference. Scaife and Bruner ( 1 975) have shown that this reference-sharing function in humans, when children follow the line of adult visual regard toward some “third party,” can occasionally be observed as early as 3 months of age and becomes the norm around 9 months. Although the social motivation to share reference is not directly linguistic ( i s . , children do it months before they acquire speech), such a motive would have important indirect consequences for the language acquisition process. As Savage-Rumbaugh, Rumbaugh, and Boysen (1978) noted, one of the peculiarities of the chimpanzee language protocols is the high concentration of imperatives.

66

E/izuhe/li Eutcs et al

They are definitely not interested in conversation for its own sake, nor do they engage in nonverbal communication with humans strictly for the purpose of sharing attention to events in the world (e.g., pointing and labeling without any ulterior motive). What better way to equip a species to acquire large vocabularies and complex grammar than to give it an insatiable lust for small talk? This desire to understand one another, and to see things the same way, means that language acquisition does not have to be motivated extrinsically (e.g., as a means for obtaining bananas). Imitation and shared reference involve affective/motivational inputs to language development, via some very general processes of shared social exchange. The next three points refer to areas in which language is used to encode specifically social content, like the relationship between object permanence and “all gone” in our reviews of cognitive bases to language. 3. lnientionality and human action. We have made much of the “actoraction-recipient-patient” relationship as an example of a semantic structure that serves as a clue to the acquisition of grammar. What we have not yet stressed is that the concept of animacy itself requires considerable social knowledge, an ability to distinguish between actions that obey Newtonian mechanics versus the less predictable actions of willful beings. Comparative studies of a wide variety of languages have led some linguists to suggest that all human grammars have rules built around an “animacy hierarchy” (Cooper and Ross, 1975; Dik, 1978; Kuno, 1976; Li & Thompson, 1976). Consider, for example, the grammatical notion of “subject of the sentence. ” All languages seem to have something like a subject, corresponding to the point of view or topic of a piece of discourse. Furthermore, it seems to be a universal of human conversation that we talk predominantly about ourselves. Hence, the topic is likely to be a “speaker-like’’ element. Languages respect this high-probability relationship between topic and human agency by developing rules that assign subject roles (first position in the sentence, agreement with the verb in person and number, nominative case marking, etc.) to a hierarchy of elements ranging from “most like the speaker” to “least human, animate, willful, etc.” For example, in Dutch it is not permissible to make inanimate instruments the subject of a sentence (as in “The knife cut the salami”). In Navajo, neutral declarative sentences must be ordered from left to right to reflect a complete power hierarchy: chiefs before peasants, adults before children, humans before animals, and so on. Children manage to map such grammatical rules onto knowledge of human action with relatively little trouble. However, an intelligent Martian with tremendous analytic capacities and full knowledge of the physics and chemistry of nonsocial earth, might have much more difficulty acquiring grammars based on these inherently social categories. The relationship between grammar and a theory of human action holds not o n l y for such grammatical forms as word order and subject-verb agreement, but also for verb morphology. For example, English modal verbs (can, would, should, etc.)

require considerable understanding of such ontological concepts as doubt vs certainty, ability vs intention. The same is true for mastery of verb tense (e.g., past vs present) and verb aspect (e.g., completed vs uncompleted action). An important arena for social-causal theories of language acquisition should be the mystery of how children map their developing “theory of human action” onto all these disparate aspects of grammar. (For reviews of animacy and causal understanding in early childhood and infancy, see Ammon, 1980; Bloom, Hood, & Lightbown, 1974; Bretherton & Bates, 1979; Golinkoff, 1975 ) 4 . Social feelings and social roles. Andersen (1977) has shown that children between the ages of 4 and 6 years have a detailed knowledge of the linguistic stereotypes associated with such social roles as male vs female, parent vs child, doctor vs patient, and doctor vs nurse. This knowledge cuts across a heterogeneous set of linguistic devices including intonation, type of lexical items used (e.g., “tough” words are reserved for males), and tag questions (e.g., women tend to say “isn’t i t ” at the end of sentences). Cremona and Bates (1977) have reported that Italian children before the age of 6 years know that their rural dialect is considered socially inappropriate and inferior compared with the standard dialect. Bates (1976) reported that children as young as I % - 3 years of age have a basic understanding of the concept of politeness. This concept in turn affects the acquisition of several aspects of grammar, including interrogative vs declarative sentence types, verb morphology (e.g., past and conditional are more polite than present tense in requests), and noun morphology (e.g., diminutive endings on nouns create “nicer” forms for use in requests). Here, too, an extraterrestrial being with no knowledge of social roles or the regulation of status through language would be unable to decipher a wide range of systematic variation in the grammar and the lexicon. By contrast, very young children apparently do understand at least the rudiments of these social concepts enough to acquire a rather complete, albeit caricatured, system of sociolinguistic variations between 2 and 6 years of age. Ervin-Tripp and Mitchell-Kernan (1977) and Ochs and Schieffelin (1979) have edited volumes summarizing many aspects of sociolinguistic development from preschool onward. Bretherton and her colleagues (e.g., Bretherton & Bates, 1979; Bretherton & Beeghly-Smith, 1981; Bretherton. McNew, & Beeghly-Smith, 1981) have discussed how these social structures and categories are prepared even earlier, in mother-infant interaction from 9 through 28 months of age. Bretherton et a l . argued that early language protocols demonstrate much richer understanding of how others feel and think than is evidenced in many nonverbal studies of role-taking (e.g., Flavell, Botkin, & Fry, 1968; Selman, 1971). When a 28-month-old boy says to his mother, ‘‘I hurt your feelings ’cause I was mean to you” (Bretherton & Beeghly-Smith, 1981), he shows an appreciation of interpersonal power and responsibility that is certainly adequate for the task of acquiring many socially based aspects of language acquisition. 5 . Discourse and syntax: learning to “stage” utterunces for a Listener. A par-

68

Elizabeth Ba/es et al.

ticularly active area of linguistic research in the last decade has been “text grammar” or “discourse grammar” (e.g., van Dijk, 1980). These grammars go beyond the level of individual sentence structure to describe how native speakers use grammatical devices to create cohesion across long passages of discourse. For example, rules that regulate the use of adjectives and other modifiers involve planning at the discourse level, insuring that the listener knows which topic is being talked about. We say “the red dog” instead of “the dog” because there is some danger of confusion between canine referents. When we use a relative clause, as in “The man whom I told you about the other day called this morning ,” we are using some rather intricate knowledge of our past exchanges with the listener and his current state of awareness to “set up” the subject of the sentence in a recognizable way. The same kind of discourse-level function applies to nouns vs pronouns, definite vs indefinite articles, conjunctions and adverbs (e.g., “but” vs “and,” “then” vs “now”), and for that matter virtually every aspect of the grammar (see Bates & MacWhinney, 1979, for reviews). As discussed by Greenfield and Smith (1976), among others, even the earliest rules of child grammar respect a division between “proposed” and “presupposed information. To make that division properly, the child will have to be able to make a vast set of rapid calculations of listener knowledge. This is inherendy social knowledge, not derivable from any other uspects of cognition. Bretherton, McNew, and Beeghly-Smith ( 198I ) have argued that language acquisition requires the development of a “theory of mind,” defined by Premack and Woodruff (1978) as a system of inferences which enables an individual to impute intentions and other mental states to self and other. As Bretherton et al. (1981) suggested, part of our task in studies of language acquisition is to understand how the child maps this “theory of mind” onto the structures of her or his language. We noted at the beginning of this paper that there is hope for a reconciliation in the “marriage” between psychology and linguistic theory. In the 1970s, while searching for cognitive and social inputs to the acquisition process, many of us set aside the problem of grammar and how it is acquired. As Susan ErvinTripp reminded us in 1977, in her keynote address to the Stanford Child Language Forum, we never did solve the syntax problem. It is time for us to gather together the knowledge that has been gained about cognitive and social inputs to language, and tie that knowledge to the very difficult problem of how formal grammatical knowledge is acquired. We know a lot more now than we did in 1957. And they tell us that love is better the second time around. ”

REFERENCES Ainsworth, M . D. S . The development of infant-mother attachment. In B . M . Caldwell & H . N . Ricciuti (Eds.), Review of child developmenr research (Vol. 3). Chicago: Univ. of Chicago Press, 1973.

Social Bases of Language Development

69

Ainsworth, M. D. S . , Blehar, M. C . , Waters, E . , &Wall, S . Potterns ofuttuchment: Apsychological study of the strange situation. Hillsdale, N.J.: Erlbaum, 1978. Amrnon, M. Develupment in the linguisric expression rfcuusul relutions: Comprehension qffeatures uflexirul and periphrostic cuusatives. Unpublished doctoral dissertation, University of California, Berkeley, 1980. Ammon, P. R. Cognitive development & early childhood education: Piagetian and neo-Piagetian theories. In H. L. Horn, Jr. & P. A. Robinson (Eds.), Psychological pruresses in eurly edui~ation.New York Academic Press, 1977. Anderson, E. Learning to speak with style: A study uj‘the sociolinguistic skills of young children. Unpublished doctoral dissertation, Stanford University, 1977. Bakeman, R., & Brown, 1. Early interaction: Consequences for social and mental development at three years. Child Development. 1980, 51,437-447. Bates, E. Peer relations and the acquisition of language. In M. Lewis & L. Rosenblum (Eds.), Friendship and peer relations. New York: Wiley, 1975. Bates, E. Language and context: The ucyuisition ofpragmatics. New York: Academic Press, 1976. Bates, E. Intentions, conventions, and symbols. In E. Bates, L. Benigni, I. Bretherton, L. Camaioni, & V. Volterra, The emergence of symbols: Cognition and communicarion in infancy. New York: Academic Press, 1979. Bates, E., Benigni, L.. Bretherton, I., Camaioni, L., & Volterra, V. From gesture to the first word: On cognitive and social prerequisites. In M. Lewis & L. Rosenblum (Eds.), Interaction. conversutiun and the development of Ianguuge. New York: Wiley, 1977. Bates, E., Benigni, L., Bretherton, I., Camaioni, L., & Volterra, V. Cognition and communication from 9-13 months: Correlational findings. In E. Bates, L. Benigni, 1. Bretherton, L. Camaioni, and V. Volterra, The emergence of symbols: Cognition and communication in infancy. New York: Academic Press, 1979. Bates, E., Camaioni, L., & Volterra, V. The acquisition of perfonnatives prior to speech. MerrillPalmer Quarterly, 1975, 21, 205-226. Bates, E., & MacWhinney, B. A functionalist approach to the acquisition of grammar. In E. Ochs & B, Schieffelin (Eds.), Developmental pragmatic..^. New York: Academic Press, 1979. Bates, E., & MacWhinney, B. Functionalist approaches to grammars. In L. Gleitman & E. Wanner (Eds.), Language acquisition: The state of the art. London and New York: Cambridge Univ. Press, 1981. Bates, E . , & Snyder, L. The cognitive hypothesis in language development. In I. Uzgiris & J. McV. Hunt (Eds.), Research with scales ofpsychological development in infancy. Urbana-Champaign: Univ. of Illinois, in press. Beckwith, L., & Thompson, S . Recognition of verbal labels of pictures, objects and events by 17-to-30-month-old infants. Journal of Speech and Hearing Research, 1976, 47, 579-587. Bell, S. M. The development of the concept of the object as related to infant mother attachment. Child Development. 1970, 41, 291-313. Bell, S . M . Cognitive development and mother-child inreructiun in the ,first three years qf life. Unpublished manuscript, 1978. (Some findings are cited i n Ainsworth et a l . , 1978.) Bell, S . M.,& Ainsworth, M.D. S . Infant crying and maternal responsiveness. Child Development, 1972, 43, 1171-1 190. Belsky, J., Goode, M. K.,& Most, R. Maternal stimulation and infant exploratory competence: Cross-sectional, correlational, and experimental analyses. Child Development, 198 I , in press. Berko, J . The child’s learning of English morphology. Word. 1958, 14, 150-177. Berko-Gleason, J., & Weintraub, S . Input language and the acquisition of communicative competence. In K. Nelson (Ed.), Children’s kinguage (Vol. 1). New York: Gardner, 1978. Block, E.. & Kessel, F. Determinants of the acquisition order of grammatical morphemes: A reanalysis and reinterpretation. Journal ofchild Languuge. 1980, 7, 181-189.

70

Elizabeth Bares et al

Bloom, L. One wordat rr time: The use oj‘single wurd utterances hefure syntux. The Hague: Mouton, 1973. Bloom, L., Hood, L., & Lightbown, P. Imitation in language development: If, when and why. Cognitive Psychology, 1974, 6 , 380-420. Bohannon, J., & Marquis, A. Children’s control of adult speech. Child Dewlopnimt, 1977, 48, 1002. Bowerman, M. Eurly syntadc develupment: A c,rosslinguistic study with spei,ia/ reference ro Finnish. London and New York: Cambridge Univ. Press, 1973. Bowerman, M. Learning the striwture of causutive verbs: A study in the relrtionship of cognitive, senurntic trnd syntcrctic. development (Papers and Reports on Child Language). Stanford, Calif.: Stanford University, 1974. Bowerman, M.F. Semantic factors in the acquisition of rules for word use and sentence construction. In D. M. Morehead & A . E. Morehead (Eds.), Normal a n d d & + v z ti,hildlanguoge. Baltimore: Univ. Park Press, 1976. Bowlby, J. Attuchment and Iuss (Vol. 1). Attirchmmt. New York: Basic Books. 1969. Bowlby, J . Attachment und loss (Vol. 2). Sepurarion. New York: Basic Books, 1973. Brdine. M. D.S . Children’s first word combinations. Monographs qf rhe Society fiw Reseirrch in Child Developmen/, 1976, 41, (1, Serial No. 164). Branigan, G. Some reasons why successive single word utterances are not. Journal ofChild L a n gur/gi’, 1979, 6, 41 1-421. Brazelton, T. B., Koslowski, B., & Main, M. The origins of reciprocity: The early mother-infant interaction. in M. Lewis & L. A. Rosenblum (Eds.), The eflect of the infurir on irs caregiver. New York: Wiley, 1974. Bretherton, I . , & Bates, E. The emergence of intentional communication. New Directionsfor Child Development, 1979, 4, 8 1-100. Bretherton, I., Bates, E., Benigni, L., Camaioni, L., & Volterra, V. Relationships between cognition. communication and quality of attachment. In Bates ct a / . (Eds.) The emergence of synhols: Cognitiun ond commirniccrtiun in infirncy. New York: Academic Press, 1979. Bretherton, I., & Beeghly-Smith, M. Talking crbout internal states: The acquisition of an explicit “theory ofmind.” Paper presented at the biennial meeting of the Society for Research in Child Development, Boston, April 1981. Bretherton. I., McNew, S., & Beeghly-Smith, M. Early person knowledge as expressed in verbal and gestural communication: When do infants acquire a “theory of mind”? In M. E. Lamb & L. R. Sherrod (Eds .), Infirnt six-ial cognition: E m p i r i d und rheureticirl considerations. Hillsdale, N.J.: Erlbaum, 1981. Bretherton, I . , McNew, S . , Snyder, L., & Bates, E. Individual differences at 20 months: Analytic and holistic strategies in language acquisition. Journal qf Chill Languuge, in press. Brown, R. A first language. Cambridge, Mass.: Harvard Univ. Press, 1973. Bruner, J. From communication to language: A psychological perspective. Cognition, 1975, 3, 255-287. (a) Bruner, J . S . The ontogenesis of speech acts. Journrrl ojChild Languugr, 1975, 2 , 1-19. (b) Bruner, J . Early social interaction and language acquisition. In H . R. Schaffer (Ed.), Studies in mother-infirtit intercrcrion. New York: Academic Press, 1977. Bruner, J . . & Ninio, A . The achievement of antecedents of labeling. ./ournu/ of Child Languuge, 1978, 5, 1-19. Bruner, J.. Olver, R., & Greenfield, P. M. Studies in ccignitiwc~gruwth. New York: Wiley, 1966. Bruner, J. S., & Sherwood, V. Early rule structure: The case of peekaboo. In J. S. Bruner, A. Jolly, & K . Sylva (Eds.), P l r y : Its role in developniiw and evolutiun. New York: Basic Books, 1976. Burchinal, P., & Farran, D. H o w importcmt i s eirrly mother-infunr infertrction fiw development during infancy? Paper presented at the International Conference on Infant Studies, New Haven, April 1980.

Caldwell. B. Instruction munuril inisentonfor infiitits (Honre ohsar~wiori,fi)r t n ~ u ~ ~ r r of’~ ~the n~~nt eriivronnimt). Unpublished manuscript, 1970. (Available from Center for Early Development and Education, 8 14 Sherman, Little Rock, Arkansas, 72202.) Carlson-Luden, V . Cnustil understunding in the 10-month-old. Unpublished doctoral dissertation, University of Colorado, 1979. Case, R. Intellectual development from birth to adulthood: A neo-Pagetian interpretation. In R. Siegler (Ed.), Children’s thinking: Whnt rlevelopsl Hillsdale, N.J.: Erlbaum, 1978. Case, R. Intellectual development in ir$aticy: A neo-Piugetian interpretation. Paper presented at the International Conference for Infant Studies, New Haven, April 1980. Cazden, C. Environmental nssisrance to the child’s acqitisitinri of grummur. Unpublished doctoral dissertation, Harvard University. 1965. Chevalier-Skolnikoff. S. The ontogeny of primate intelligence. Implications for communicative potential. In S . Harnad, H. Steklis, & J. Lancaster (Eds.), Origins of 1angrtuge (itid speech. New York: New York Academy of Sciences. 1977. Chomsky. N. Synttrc,tic, structures. The Hague: Mouton, 1957. Chomsky, N. Aspects o f u rheor? of syntax. Cambridge, Mass.: MIT Press, 1965. i New York: Pantheon Books, 1975. Chomsky, N. Refler~tionso ~ lringiroge. Clark, R. What’s the use of imitation’? Journal of Child Lmguoge. 1977, 4, 341-358. Clarke-Stewart, K . A. Interactions between mothers and their young children: Characteristics and consequences. Monogruphs of the Soriety for Reseurch in Child Development, 1973, 38 (6-7 Serial No. 153). Cohen, S., & Beckwith, L. Preterm infant interaction with the caregiver in the first year of life and competence at age two. Child Developtnent, 1979, 50, 767-776. Connell, D. B. Individual differences in tittuchment behcivior: Long-term .stubility und relutionships to lunguuge development. Unpublished doctoral dissertation, Syracuse University, 1977. Cooper, W . , & Ross, J. World order. In R. Grossman. L. San, & T . Vance (Eds.), Poprr.s,fr.om the purczsession on functionn/ism. Chicago: Chicago Linguistic Society, 1975. Corrigan. R. Language development as related to stage 6 object permanence development. Journul of’ Child Language. 1978, 5 , 173-189. Cremona, C . , & Bates, E. The development of attitudes toward dialect in Italian children. Journul of’ Psycholinguistic Reseurch. 1977, 6 , 223-232. Cross, T . Mothers’ speech adjustments: The contributions of selected child listener variables. In C. Snow & C. Ferguson (Eds.), Talking to children: Lutigiruge input and acquisition. London and New York: Cambridge Univ. Press, 1977. Curtiss, S. Genie: A psycholinguistic study of u modern-cloy “wild child.” New York: Academic Press, 1977. Curtiss. S . , & Yamada, J. Lungircige und cognition. Paper presented at the Third Annual Boston University Conference on Language Development, Boston, September 1978. Dale, P. Languuge development: Strrrcrure undfitnction (2nd ed.). New York: Holt, 1976. Dik, S. Funcrional grummur. Amsterdam: North-Holland Publ. 1978. Edwards, D. Sensory-motor intelligence and semantic relations in early child grammar. Cognition, 1973, 2, 395-434. Ervin-Tripp, S., & Mitchell-Kernan, C. (Eds.). Child discourse. New York: Academic Press, 1977. Farran, D., & Ramey, C. Social class differences in dyadic involvement during infancy. Child Development, 1980. 51, 254-257. Feldman, C. The effects o / vtirioirs types ofudidt responses on the syntuctic ucyuisition of two to three veur olds. Unpublished manuscript, Department of Psychology, University of Chicago, 1971. Fenson. L., & Ramsey, D. S. Decentration and integration of the child’s play in the second year. Child Development. 1980, 51, 171-178. Ferguson, C. Baby talk as a simplified register. In C. Snow & C. Ferguson (Eds.), Tulking t o

72

Elizabeth Butes et al.

children: Language input and acquisition . London and New York: Cambridge Univ. Press, 1976. Fillmore, L. The second time around: Cognitive and social strutegies in second language ucyvisition. Unpublished doctoral dissertation, Stanford University, 1976. Fischer, K. W. A theory of cognitive development: The control and construction of hierarchies of skills. Psychological Review, 1980, 87, 477-531. Flavell, J . H. The developmentul theory of Jeuh Piuget. New York: Van Nostrand-Reinhold, 1963. Flavell, J . , Botkin, P., & Fry, C. The development ($role-taking and communicution skills in young children. New York: Wiley, 196R. New York: McGraw-Hill, Fodor, J. A., Bever, T. G., &Garrett, M. F. Theps~~chologyoflanguage. 1974. Freedle, R., & Lewis, M. Prelinguistic conversations. In M. Lewis & L. Rosenblum (Eds.), Inreruction,conversation, und the development of languuge. New York: Wiley, 1977. Furrow, D. The benejh if motherese. Paper presented at the biennial meeting of the Society for Research in Child Development, San Francisco, March 1979. Furth, H. G. Thinking without lunguage: Psychologicd implicutions of deufness. New York: Free Press, 1966. Garcia, J . , & Koelling, R. Relation of cue to consequence in avoidance learning. Psychonomic. Science. 1966. 4, 123-124. Golinkoff, R . Semantic development in infants. The concepts of agent and recipient. Merrill-Palmer Quurterly, 1975, 21, 181-193. Greenfield, P. M., & Smith, J. H. The structure o f communication in early languuge development. New York: Academic Press. 1976. Halpern, E., & Avieter, L. Psycholinguistic skills and sensory motor development within Piaget's theoreticalfr.umework. Paper presented at the 2 I st International Congress of Psychology, Paris, July 1976. Hardy-Brown, K., Plomin, R., & DeFries, J. C. Generic and environmental influences on rate of communicative development in the first year of life. Developmental Psychology, 1981, 17, 704-717. Harkness, S. Aspects of social environment and first language acquisition in rural Africa. In C. Snow & C. Ferguson (Eds.), Talking to childreti. London and New York: Cambridge Univ. Press, 1977. Harlow, H. F., & Harlow, M. K. Effects of various mother-infant relationships on rhesus monkey behaviors. In B. M. Foss (Ed.), Determincmts ($infant behavior (Vol. 4). London: Methuen. 1969. Harmon, R. J., Suwalsky, J. D., & Klein, R. P. Infants' preferential response for mother versus an unfamiliar adult. Journul if the American Academy ofchild Psychiatry, 1979, 18, 437-449. Hock, E. Alternative approaches f a child reuring and their eflects on the mother-inflint relationship. Final Report to the Office of Child Development, 1976. Horgan, D. Nouns: Low 'em or leave 'em. Address to the New York Academy of Sciences, May 7 , 1979. Johnston, J. The language disordered child. In N. Lass, J . Norther, D. Yoder, & L. McReynolds (Eds.), Speech. language. and learning. Philadelphia: Saunders, 1981. in press. Kaye, K . Toward the origin of dialogue. In H. R. Schaffer (Ed.), Studies in mother-infant inferoction. New York: Academic Press. 1977. Kaye, K . The soc~ic~l conlext o f infant developtnetir (Final Report to the Spencer Foundation, June 1974-November 1979). 1980. Kaye, K. The menml untl social life of babies. Chicago: Univ. of Chicago Press, 1981. Kempler, D. Varititiow in langucige ucquisition (UCLA Working Papers in Cognitive Linguistics). Los Angeles: Univ. of California, 1980.

Social Bases of Language Development

73

Kopp, C., Sigman, M., & Parmelee, A . Longitudinal study of sensorimotor development. Developmental Psychology, 1974, 10, 687-695. Kuhn, T . S . The structure of scientific revolutions (2nd ed.). Chicago: Univ. of Chicago Press, 1970.

Kuno, S . Subject, theme and the speaker’s empathy: A re-examination of relation phenomena. In C . Li (Ed.), Subject and topic. New York: Academic Press, 1976. Lashley, K. S. The problem of serial order in behavior. In L. A. Jeffres (Ed.), Cerebral mechunisms in behavior. New York: Wiley, 1951. Leonard, L. Language impairment in children. Merri/l-Pidmer Quirrterly, 1979, 25, 205-232. Lewis, M.,& Rosenblum, L. A. Interaction, conversdon, and the development of lunguage. New York: Wiley. 1977. Li, C.. & Thompson, S. Subject and topic: A new typology of language. In C. Li (Ed.), Subject and topic. New York: Academic Press, 1976. Lock, A. (Ed.). Action, gesture and symbol. New York: Academic Press, 1978. Lock, A. The guided reinvention of liinguage. New York: Academic Press, 1980. Lorenz, K. King Solomon’s ring. New York Crowell, 1965. MacNamara, J. The cognitive basis of language learning in infants. Psychological Review, 1972, 79, 1-13. Main, M . E.rp/ormtion. play iznd ccignitive funi.tioning as related to child-morher attachment. Unpublished doctoral dissertation. Johns Hopkins University, 1973. Matas, L., Arend, R . A . , & Sroufe, L. A. Continuity of adaptation in the second year: The relationship between attachment and later competence. Child Developmenr, 1978,49,547-556. McCall, R. B., Eichorn, D. H., & Hogarty, P. S.Transitions in early mental development. Monographs of the Society for Research in Chill Development, 1977, 42 (3, Serial No. 171). McCarthy, D. Language development in children. In L. Carmichael (Ed.), Munual of Child Psychology (2nd ed.). New York: Wiley, 1954. McNeill, D. The development of language. In P. H. Mussen (Ed.), Carmichtrel’s Manuid of Child Ps-vchology (3rd ed.). New York: Wiley, 1970. McNeill, D. The conceptual basis of lunguage. Hillsdale, N.J.: Erlbaum, 1979. McNew, S. Language and gesture: Co-occurrences and contingencies in speech to young children. Manuscript in preparation, 1981. (a). McNew, S. Muternal gestures: An e.rploration of contingency. Manuscript in preparation, 1981. (b). Meltzoff, A , , & Moore, M. K. Imitations of facial and manual gestures by human neonates. Science, 1977, 198,75-78.

Miller, G. A , , Galanter, E. H., & Pribram, K. H. Pion und the .wwtiire of behavior. New York: Holt, 1960. Miller, J . , Chapman, R., & Bedrosian, J . Detining developmentcilly disubled subjectsfor research: The relationships between etiology. cognitive development rrnd language iind communicative pcvj?ormimce.Paper presented at the Second Annual Boston University Conference on Language Development, Boston, October 1977. Moerk. E. L. Verbal interactions between children and their mothers during the preschool years. Developmental Psychology. 1975, 1 I , 788-794. Mowrer, 0. Leczrning theor-y und behuvior. New York: Wiley, 1960. Nelson. K. Concept, word and sentence: Interrelations in acquisition and development. Psychologiciz/ Review, 1974, 81, 267-285. Nelson, K. Facilitating children’s syntax acquisition. Deve/opmentrd Psy:yc,hology. 1977, 13, 101107.

Nelson, K., Causkaddon. G., & Bonvillian, M. Syntax acquisition: Impact of experimental variation in adult verbal interaction with the child. Child Deve/opmetir. 1973, 44, 497-504. Newmeyer, F. Linguistic theov in Americu. New York: Academic Press, 1980.

74

Elizciheth Bates et al

Newport, E . , Gleitman, L., & Gleitman, H. Mother I’d ruther do it niysey: Some efect.7 rind nun-effects of motherese. In C. Ferguson & C. Snow ( a s . ) , Tulking 10 children. London and New York: Cambridge Univ. Press, 1977. Pp. 109-149. Newson, J. Dialogue and development. In A. Lock (Ed.), Action. gesture. und symbol: The eniergence of’ liitigutige. New York: Academic Press, 1978. Nicolich, L. Beyond sensorimotor intelligence: Assessment of symbolic maturity through analysis of pretend play. Merrill-Pultner Quurterly. 1977, 23, 89-99. Ochs, E. Introduction: What child language can contribute to pragmatics. In E. Ochs & B. Schicffelin (Eds.), Developnicntul prugmmatics. New York: Academic Press, 1979. Ochs, E., & Schieffelin, 8 . (Eds.). Df~vc~/o/~mentrr/ prrigmarics. New York: Academic Press, 1979. Pentz, T. Facilitation ofktnguuge ucquisition: The role of the mother. Unpublished doctoral dissertation, Johns Hopkins University, 1975. Piaget, J. Pluy, dreums rind imittition in child/iood. New York: Norton, 1962. Pinker, S. Formal models of language learning. Cognition, 1979, I , 217-283. Pinker, S . A theory of the acquisition of lexical-interpretive grammars. In J. Bresnan (Ed.), The nienfal representation uf grummciricul relutions. Cambridge, Mass.: MIT Press, 1980. Plomin, R., DeFries, J. C., & McClearn, G. E. Behuviorulgenetics. San Francisco: Freeman, 1980. Premack, D., & Woodruff, G. Does the chimpanzee have a theory of mind? Behrrviorul arid Bruin Sciences, 1978, 1, 515-526. Ramey, C., Farran, D., & Campbell, F. Predicting IQ from mother-infant interactions. Child Development, 1979, 50, 804-814. Ratner, N., & Bruner, J. Games, social exchange, and the acquisition of language. Journul of’Child Ltirigituge. 1978, 5, 39 1-401. Rogosa, D. A critique of cross-lagged correlation. Psychological Bulletin, 1981, 88, 245-321. Rowe, D. C., & Plomin, R. Temperament in early childhood. Juurriril of Personality Assessmcnt, 1977, 41, 150-156. Ryan, J. Early language development. In M. P. M. Richards (Ed.), The integrution ofthe child into a sociul world. London and New York: Cambridge Univ. Press, 1974. Sachs, J . , & Devin, J . Young children’s use of age-appropriate speech styles in social interaction and role-playing. Journul uf Child Lunguuge. 1976. 3, 8 1-98. Sander, L. W. The regulation of exchange in the infant-caregiver system and some aspects of the context-content relationship. In M. Lewis & L. A. Rosenblum (Eds.), /n/errrction, conversution, rind the rlevc~lopmentof lunpuge. New York: Wiley, 1977. Savage-Rumbaugh, E. S., Rumbaugh, D. M., & Boysen, S. Linguistically-mediated tool use and exchange by chimpanzees (Pan Trogludytes). Behuviorul und Bruin Sciences, 1978.4.539-5.54. Scaife, M . , & Bruner, J. S. The capacity for joint visual attention in the infant.’Nuture, (London). 1975, 253, 265-266. Schaffer, H. R. (Ed.). Studies in mothcr-infant interaction. New York: Academic Press, 1977. Schaffer, H. R., Collis, G . M . , & Parsons, G. Vocal interchange and visual regard in verbal and pre-verbal children. In H. R. Schaffer (Ed.), Studies in mother-infbnt intermtion. New York: Academic Press, 1977. Pp. 291-324. Schlesinger, I. M. Relational concepts underlying language. In R. Scheifelbusch & L. Lloyd (Eds.), Lunguuge perspectives: Acyuisitiun. retardurion und intervention. Baltimore: Univ. Park Press, 1974. Schwartz, R. Numes, uctions rrnd ol!jec/s. Unpublished doctoral dissertation, Memphis State University, 1978. Selman, R. Taking another’s perspective: Role-taking development in early childhood. Child Development. 1971, 43, 1721-1734. Shatz, M . On mechanisms of language acquisition: Can features of the communicative environment account for development? In L. Gleitman & E. Wanner (Eds.), Lunguuge acquisition: The, sfute of the urt. London and New York: Cambridge Univ. Press, 1981, in press.

Socitrl Btrses of' Ltrnguugr Development

75

Shatz. M., & Gelman, R. The development of communication skills: Modifications in the speech of young children as a function of the listener. Monogruphs of rho So(.icty./or Resecrr~hin Child Development, 1973, 38 (3, No. 152). Shettleworth, S . J . Constraints on learning. A d w z c ~ r sin the Study o j B r h ~ v i o r ,1972, 4, 1-68. Shore, C. S . . O'Connell, B.,& Bates, E. First sentenw.s in lungutrge tmd symholicpluv. Manuscript in preparation. 198 I . Slobin, D. I. Cognitive prerequisites for grammar. In C. Ferguson & D. Slobin (Eds.), Studies in child lungirugc, d e ~ d o p m e n t .New York: Holt, 1973. Slobin. D. 1. Psycholinguistics (2nd ed.). Glenwood, 111.: Scott, Foreman, 1979. Snow, C.. & Ferguson, C. (Eds.). Tulking to diildren: Lungurrge inpui und cicquisition. London and New York: Cambridge Univ. Press, 1977. Snyder, L. Communicative and cognitive abilities and disabilities in the sensorimotor period. Merrill-Palmer Quurirrly. 1978, 24, 161-180. Spitz, R. A . Thefirst yeor of life. New York: International Universities Press, 196.5. Stern, D. N., Beebe, B., Jaffe, J., & Bennett. S. L. The infant's stimulus world during social interaction. In H. R. Schaffer (Ed.), Studies in mother-infimr iwieruction. New York: Academic Press, 1977. Tinbergen, N. The srrrdv of instinct. London: Oxford Univ. Press, 1951. Tolman, E. C. P ~ r r p s i v rhehuvior oj'unimuls and men. New York: Century, 1932. Trevarthen, C., & Hubley, P. Secondary intersubjectivity: Confidence, confiding and acts of meaning in the first year. In A. Lock (Ed.), Action, gesture und symbol. New York: Academic Press, 1978. Pp. 183-230. Tulkin. S. & Covitz, F. Mother-it$h inrerurrion u r r d intellectual jiuncrioninR at age six. Paper presented at the biennial meeting of the Society for Research in Child Development, Denver, April 197.5. Turiel. E. The development of concepts of social structure. In J. Click & K . A. Clarke-Stewart (Eds.), The development of social utufersrunding. New York: Gardner, 1978. Urwin. C. The development of communication between blind infants and their parents. In A. Lock (Ed.). Ariion. gesture, and symbol: The emergenre of language. New York: Academic Press, 1978. Valian, V. The wherefores and therefores of the competence-performance distinction. In W. Cooper & E. Walker (Eds.), Sentence processing. Hillsdale, N.J.: Erlbaum, 1979. Pp. 1-27. van Dijk, T. Marrosinrrrures. Hillsdale, N.J.: Erlbaum, 1980. Vygotsky, L. S . Thought and lunguage. Cambridge, Mass.: MIT Press, 1962. Werner, H.,& Kaplan, B. Symbolf;mncrtion. New York: Wiley, 1963. Wexler, K.,& Culicover. P. Fvrmul princ.iples c$ lrorguuge trquisition. Cambridge, Mass.: MIT Press, 1980. Williams, C. Personal communication, 1979. Wood, D.. Bruner, J. S., & Ross. G. The role of tutoring in problem solving. Joirrntrl of Child Psychology cmd Psychiurrv, 1976, 17, 89- 100. Zukow, P.. Reilly, J., & Greenfield, P. Making the absent present: Facilitating the transition from sensorimotor to linguistic communication. In K . Nelson (Ed.), Children's ltrngrctrge (Vol. 3). New York: Gardner, 1981, in press.

This Page Intentionally Left Blank

PERCEPTUAL ANISOTROPIES IN INFANCY: ONTOGENETIC ORIGINS AND IMPLICATIONS OF INEQUALITIES IN SPATIAL VISION

Marc H . Bornstein DEPARTMENT OF PSYCHOLOGY

NEW YORK UNIVERSITY

NEW YORK. NEW YORK

1. INTRODUCTION ......................................................

77

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

79 79 82

11. TWO CLASSES OF PERCEPTUAL ANISOTROPY

A. PERCEPTUAL SALIENCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B . PERCEPTUAL EQUIVALENCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111. TWO CLASSES OF PERCEPTUAL ANISOTROPY IN INFANCY..

...........

A. PERCEPTUAL SALIENCE IN INFANTS .............................. B. PERCEPTUAL EQUIVALENCE IN INFANTS.. . . . . . . . . . . . . . . . . . . . . . . . . IV. DISCUSSION AND CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. REVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. SOME INTERRELATIONS BETWEEN PERCEPTUAL ANISOTROPIES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. PERCEPTUAL-COGNITIVE-SOCIAL DEVELOPMENT: ANISOTROPY AND THE FACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. SOME IMPLICATIONS FOR COGNITIVE DEVELOPMENT . . . . . . . . . . . . . REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85 85 97 108

108 109 1I1 I 1I

115

After all, who can fail to be interested in how things grow, mature, and die? -B.

I.

R. MCCANDLESS A N D M. F. GEIS (1975, p. 7)

Introduction

Physical space extends outward from the central ego equally in all directions, and the identification and discrimination of orientation, direction, and location in space are critical to perception and to perceptual development in virtually all 77 ADVANCES IN CHILD DEVELOPMEN7 AND BEHAVIOR, VOL. 16

Copyright @ 1982 by Academic Press. h c . All right6 of reproduction in any form reserved. ISBN 0-12-009716-8

78

Murc H . Bornstein

visual vertebrates. Yet, psychological perceptions of space are skewed. Certain orientations are favored: For example, vertical and horizontal hold many advantages psychologically, while other orientations are at a disadvantage. Likewise, certain directions and locations in space are favored: For example, up and down are salient relative to left and right. A variation in spatial perception or conceptualization with regard to axis or location or direction is termed an anisorropy. ' What are the principal anisotropies in humans? When and how do they manifest themselves psychologically? What perceptual functions do they serve? How general are they phylogenetically? What are their physical, physiological, psychological, or cultural origins? When do they first occur ontogenetically? Do they develop continuously or discontinuously? What is their developmental role? In this article I attempt to address these questions and provide their answers as best we know them. In the first part of this article, I define two common classes of perceptual anisotropy . They are perceptual sulience of the main orthogonuls, meaning the orientation hierarchy of vertical-horizontal-oblique, and perceptual equivalence on the laterul. meaning the similarity of left-right mirror-image sections or enantiomorphs of a pattern. Here, I also briefly review the circumstances under which these perceptual anisotropies emerge in different species, particularly in humans. My discussion is circumscribed to visual perception, since the visual modality is superior to others in the perception of space (e.g.. Geldard, 1970; J. J. Gibson, 1966). In the second and main part of this article, I review and interpret psychological evidence that points to the existence of perceptual anisotropies among newborns and infants in the first year of life. Like older human children and adults and like other infrabuman species, human infants are not equally sensitive to all spatial orientations or to all transformations of patterns that occur in the perceptual world. The two common classes of anisotropy are especially prominent in infancy. Infants consistently find the main orthogonals salient, as shown in a variety of detection, acuity, information-processing, and preference measures. Infants also see and treat as equivalent lateral mirror-image components of visual patterns. This review of the infant literature is not intended to be exhaustive. It purports, rather, to meet three goals: first, to demonstrate that perceptual anisotropies are present in infants; second, to assess how alike or unalike one another immature and mature perceptual anisotropies may be; and, third, to examine some contemporary interpretations of the origins of perceptual anisotropies in infants. Do infants manifest perceptual anisotropies? Are they Anlagen of adult anisotropies? What are their sources? 'Throughout this article orientation will be specified in degrees rotation clockwise from vertical = 0". Orthogonol will specify one or both of the principal orientations in space, viz. vertical (0")or horizontal (90'); and, unless the context specifies otherwise, oblique will specify the principal nonorthogonals, viz. 45" left and right rotations from vertical.

In the third and final part of this article, I discuss some theoretical interrelations between the two classes of anisotropy, the special role of the face in perceptual anisotropies, and some likely implications of infant anisotropies for the development of perception, cognition, and language in childhood.

11.

Two Classes of Perceptual Anisotropy A.

PERCEPTUAL SALIENCE

Although spatial orientation varies infinitely in all directions, the main orthogonals of vertical and horizontal are principal perceptual orientations in visual space for humans as well as for a large number of other animals. Oblique orientations give rise to distinct perceptual disadvantages. The data that construct these generalizations derive from a wide variety of sources (for some general reviews, see Appelle, 1972; Arnheim, 1974; Essock, 1980; J. J. Gibson, 1934, 1966; Howard & Templeton, 1966; Pick, Yonas, & Rieser, 1979; Rock, 1973). Appelle ( 1 972) termed the general disadvantage for obliques in spatial perception the “oblique effect.” Both psychophysical and perceptual studies in adult humans, for example, have shown repeatedly that detection, discrimination, comparison, and assessment are easiest and that identification is most rapid and accurate for stimuli aligned along the principal orthogonals as opposed to any oblique. Essock (1980) reviewed this literature and distinguished two classes of oblique effect, one obtained on tasks that measure the basic functioning of the visual system and the second obtained on tasks that measure differential capabilities of orientation information processing at higher perceptual levels. Class 1 oblique effects derive from psychophysical studies which indicate that visual acuity or sensitivity for a target is greatest when it is oriented along the vertical or horizontal meridian (e.g., Attneave & Olson, 1967; Berkley, Kitterle, & Watkins, 1975; Camisa, Blake, & Lema, 1977; Campbell, Kulikowski, & Levinson, 1966; Corwin, Moskowitz-Cook, & Green, 1977; Emsley, 1925; Higgins & Stultz, 1950; Lennie, 1974; Taylor, 1963; Timney & Muir, 1976) and that perceptual stability is greatest for vertically and horizontally oriented stimuli (Cosgrove, Schmidt, Fulgham, & Brown, 1972; Craig & Lichtenstein, 1953; McFarland, 1968). This class of effect has been linked to characteristics of the neurophysiological substrate of vision (see later). Class 2 oblique effects derive from perceptual-learning and cognitive studies which demonstrate that orthogonally oriented stimuli are at an advantage in tasks that require discrimination, identification, matching, and recognition of targets. Thus, for example, adults judge verticality and horizontality with greater accuracy than obliquity (Keene, 1963); they match and reproduce verticals and horizontals better than obliques

80

Marc. H . Bornsrein

(Lechelt, Eliuk, & Tanne, 1976); and they frequently perceive and reproduce tipped visual forms as closer to the principal orthogonals (Bornstein, 1974, n. 4; Bouma & Andriessen, 1968; J. J. Gibson, 1934, 1937; Olson & Hildyard, 1977). Obliques also take longer than the main orthogonals to name (Attneave & Olson, 1967) and to recognize and identify or differentiate (Olson & Hildyard, 1977). This class of oblique effect has been linked to encoding and memory processes. Essock ( 1980) also investigated adults ' perceptions of differently oriented stimuli in a series of integrated studies that examined the Class 2 oblique effect. In the first study, Essock asked eight observers simply to report the presence or absence of single lines (2.5" of visual angle) presented tachistoscopically (100 milliseconds). Observers were instructed to disregard line orientation (O", 45", 90°, or 135")or position (at a short distance from the fixation point and at one of eight equally spaced locations around the clock). Essock found that when observers were asked simply to detect a stimulus, all orientations were detected equally quickly. In the second study, Essock required eight observers to make a unique response for each of the four orientations used previously. Here observers took significantly longer to identify obliques than the main orthogonals. In the third study, Essock required eight observers simply to classify lines (regardless of orientation) as either obliques or orthogonals. As in his first study, Essock found no oblique effect. In short, "unique identification of stimulus orientation was necessary to produce an oblique effect, whereas tasks requiring detection or classification of the stimuli did not demonstrate an oblique effect" (Essock, 1980, p. 40). Another good example of the role of this anisotropy in spatial vision may be found in the literature of symmetry perception. Moreover, studies of symmetry point up a second important aspect of this anisotropy, namely that between the two principal orthogonals (vertical and horizontal) vertical is especially salient. Symmetry is perceptually special, but not all orientations of symmetry provoke the same effective perceptual advantages; symmetry about the vertical axis is distinctive. The salience of vertical symmetry was early noted by Mach (1906, p. 107), and it has been repeatedly commented on since (e.g., Goldmeier, 1937; Julesz, 1971; Rock & Leaman, 1963). When pitted against matched horizontal or oblique symmetries, repetitions, and asymmetries, vertical symmetries are preferred (e.g., Mach, 1886/1959; Szilagyi & Baird, 1977); vertical symmetries are detected more easily, identified more accurately, and sorted more quickly (e.g., Corballis & Roldan, 1975; Fitts & Simon, 1952; Fitts, Weinstein, Rappaport, Anderson, & Leonard, 1956; Fox, 1975; Julesz, 1971; Palmer & Hemenway, 1978), even when visual presentation hovers around threshold (Johnson & Uhlarik, 1977); and vertical symmetries are remembered better in tasks that may rely on recognition (Fitts et a / . , 1956) or on reproduction (Boswell, 1976; Deregowski, 197 1).

Symmetries about horizontal or oblique axes are usually preferred, processed, and remembered better than asymmetries, but are less salient than vertical symmetries. Some (unresolved) controversy surrounds the relative perceptual salience or advantage for horizontal versus oblique. One position derives from the concept of the “oblique effect” (Appelle, 1972); it argues that vertical and horizontal symmetries are more salient than are oblique ones (Attneave & Curlee, 1977; Gamer, 1970; Goldmeier, 1972; Palmer & Hemenway, 1978). The alternative position derives from the view that symmetry processing involves first mentally rotating nonvertical patterns to the vertical (Shepard & Metzler, 1971); it argues that the salience of nonvertical symmetries is directly related to the degree to which they are displaced from the vertical (e.g.. Corballis & Roldan, 1975; Johnson & Uhlarik, 1977; Mach, 1886/1959). Vertical is not just special in symmetry. Vertical, in contrast to horizontal, is related to a physical universal, a constant, and a natural referent of directiongravity. Vertical divides equal halves of space (left-right) as opposed to unequal ones (up-down). Vertical is perceptually and cognitively salient: A vertical line appears to be slightly longer than an objectively equal horizontal line; velocity along the vertical appears appreciably enhanced over an objectively equal velocity along the horizontal; location decisions on the vertical are easier than on the horizontal; memory for the location of items in a visual scene is more durable on the vertical than on the horizontal; etc. (e.g., Howard & Templeton, 1966; Rock, 1973). The main orthogonals are as perceptually salient for children as they are for adults. Braine (1978a, 1978b) has proposed a theory of perceptual development that posits three distinct stages in the child’s perception and understanding of orientation. She suggests that children first distinguish upright (and vertical) from diagonal, horizontal, and upside down, which are all part of a single class of nonuprights. Second, they distinguish among these nonupright orientations. Third, they distinguish left from right (see Section 11,B). In other words, Braine argues, vertical and upright are special for children. Braine, Lerner, and Relyea (1980) provide clear evidence for a developmental sequence between Braine’s first two stages, the primary focus of this discussion of perceptual salience. Braine and her colleagues gave four groups of 3- and 4-year-olds each one kind of two-alternative discrimination-learning problem that involved pairs of line drawings of familiar figures oriented upright, sideways, or upside down. For upright-nonupright problems, that is, upright-upside down and uprightsideways, 85% of 3-year-olds and 95% of 4-year-olds reached a preestablished learning criterion, whereas for the nonupright problem, that is, upside downsideways and vice versa, only 30% of 3-year-olds and 75% of 4-year-olds reached criterion. By the time upside down is distinguished from sideways, the vertical is providing key information about shape orientation. [Interestingly,

82

Mtrw H . Romstein

Braine rt id. (1980) note that adults take longer to discriminate upside down from sideways than to discriminate upright from upside down or sideways; according to this analysis, levels of orientation processing in adults correspond to developmental stages of coding orientation in children.] Olson (1970) has extensively studied and confirmed perceptuocognitive developmental lags for the diagonal in young children, and Fisher (1980, 1981) has argued that children can code orthogonal lines absolutely, and hence these directions are resistant to change or to decay on account of configuration, external visual cues, presentation conditions, etc., whereas children code oblique lines relationally and in less resistant ways. Bryant (1973), Corballis and Zalik (1977), and Williamson and McKenzie (1979) have all shown that young children have difficulty matching a remembered oblique. In general, perceptual development seems to be at an advantage with respect to vertical and horizontal perhaps because “the frame of reference of vertical and horizontal [is] characteristic of the environment” (E. J . Gibson, 1969, p. 377). The perceptual advantage for the main orthogonals is also widespread phylogenetically. A large number of different species, including octopuses, goldfish, pigeons, rabbits, rats, squirrels, cats, monkeys, and chimpanzees, show one or another class of “oblique effect” (for general reviews, see Appelle, 1972; Bauer, Owens, Thomas, & Held, 1979). One qualification to these generalizations about the main orthogonals is important to underscore. For each individual at any time, there may exist two verticals. “Environmental vertical” is defined by gravity, is fixed, and is perpendicular to the earth; “retinal vertical” bisects the body independent of the body’s rotation in space (Essock, 1980; J. J. Gibson & Mowrer, 1938; Howard & Templeton, 1966; Rock, 1973). Normally, environmental and retinal verticals are congruent. In this article, I refer mainly to verticals as experienced by the organism. 8. PERCEPTUAL EQUIVALENCE

Clearly, the discrimination of orientation is critical to spatial perception. Visual organisms are highly sensitive to orientation specificity and change, and detection of orientation appears to be a relatively early stage of visual processing by the nervous system (e.g., Hubel & Wiesel, 1968). It is curious, therefore, that discriminations of stimuli from their reflections 180” around the main orthogonal axes (as opposed to oblique ones) represent extremely difficult problems for a great variety of animals, including human children and adults (for some general reviews, see Bornstein, 1981; Bradshaw, Bradley, & Patterson, 1976; Corballis & Beale, 1976; Sutherland, 1961; Tee & Riesen, 1974). Reflections about the

vertical axis are called lateral or left-right mirror images, while reflections about the horizontal axis are called vertical or up-down mirror images. It is anecdotal and legend that children and adults frequently experience difficulty discriminating between left and right. In the Analysis ofSensations, Mach (188611959, p. 110) noted that “children constantly confound the letters h and d, p and q . Adults, too, do not readily notice a change from left to right.. . .” Farrell (1979) recently investigated adults’ coding and discrimination of left and right in a series of integrated studies. In his first study, Farrell contrasted orienting with discriminating left, right, up, and down; his design separated the subjects’ simply copying spatial information in the stimulus (manual orienting) from naming (vocal discrimination), which (because the responses themselves are not mirror images in the latter case) constitutes the best evidence for the ability to tell left from right (Corballis & Beale, 1976). Twelve young adults judged the direction of arrows (1” x I ” of visual angle) presented tachistoscopically (100 milliseconds) by moving a lever in the matching direction or by naming the direction; reaction times were measured. Farrell found that telling left from right in a simple perceptual judgment is significantly harder than telling up from down, but there were no dimensional differences in orienting. In his second and third studies, Farrell showed that the left-right disadvantage does not result from difficulty in accessing linguistic codes for direction, and, in a fourth study, he showed that left-right discriminations are confusing for positional as well as directional information. In short, adults find it more difficult to discriminate left-right than up-down. Sekuler and Houlihan (1968) provided another simple and clear experimental demonstration of this phenomenon. They asked 24 young adults to identify as “same” or “different” with regard to orientation pairs of C shapes (approximately 10” square) that were aligned vertically or horizontally and that were repetitions or mirror images of one another. Reaction time was again the principal measure. Considering pairs oriented horizontally, Sekuler and Houlihan found that subjects took longest to identify left-right mirror images (C 7)while they identified up-down mirror images (n U) and repetitions (C C and fl fl) in significantly shorter amounts of time. (The rates for “same” and “different” responses in this experiment were equivalent.) Data of several other investigators experimentally confirm the difficulty or confusability of left-right mirror images in adults (e.g., Bradshaw ef ul., 1976; Butler, 1964; Pomerantz, Sager, & Stoever, 1977; Rock, 1973; Wolff, 1971). Of course, as Mach and innumerable teachers and parents before and since have observed, young children have special difficulties with left and right. The classic demonstration of left-right mirror-image equivalence in children was that of Rude1 and Teuber ( 1 963). They used a two-choice discrimination-learning paradigm in which the two stimuli were simultaneously presented and horizon-

tally aligned. Children were required to identify the “correct” stimulus. The lateral positions of the stimuli were randomized from trial to trial, and the children were told whether they had chosen the correct or incorrect stimulus after each trial. Rudel and Teuber found that children between 4 and 9 years of age have great difficulty in learning to discriminate left-right mirror-image obliques (/ versus \) and mirror-image C shapes (C versus 7)but readily learned to discriminate horizontal from vertical lines (- versus I ) and a U shape from its inversion or up-down mirror image (U versus n). Rudel and Teuber’s results have been repeatedly confirmed in both Western and non-Western cultures (e.g., Huttenlocher, I967b; Over & Over, 1967; Sekuler & Rosenblith, 1964; Serpell, 1971). Left-right confusions are also widespread among infrahuman species. Different animals, including octopuses, fishes, rats, pigeons, cats, and monkeys, perceptually equate left-right mirror images (for some general reviews, see Bornstein, 198I ; Bradshaw cr ul., 1976; Corballis & Beale, 1976; Tee & Riesen, 1974). Several qualifications of these generalizations about mirror-image equivalences need to be made. First, left-right differentiation is usually easier under conditions of simultaneous discrimination than successive discrimination (Aaron & Malatesha, 1974; Bradshaw ef al., 1976; Bryant, 1973; P. L. Harris, Le Tendre, & Bishop, 1974; Over & Over, 1967; Sidman & Kirk, 1974; Stein & Mandler, 1974; Tee & Riesen, 1974; Wohlwill & Wiener, 1964). Thus, coding in memory is implicated in the difficulty of discriminating mirror images; temporal separation of enantiomorphs significantly impairs discriminability of their orientation. Second, the spatial alignment of stimuli is also a significant factor in mirror-image discrimination: Vertical mirror images have been found to be confusing in animals, human children, and human adults (Butler, 1964; Huttenlocher, 1967a, 1967b; Lashley, 1938; Sekuler & Houlihan, 1968; Sekuler & Pierce, 1973; Sekuler & Rosenblith, 1964; Sutherland, 1961; Wohlwill & Wiener, 1964; Wolff, 1971 ), but vertical mirror images have usually been found to be somewhat less confusing than lateral ones (e.g., Bradshaw et al., 1976; Butler, 1964; Huttenlocher, 1967a, 1967b; Sekuler & Rosenblith, 1964). Significantly, mirror images about oblique axes-C are U are mirror images about the /-are not confusing at all. (In the absence of cues associated with stimulus alignment, however, left-right and up-down problems are solved equally well-Fisher, 1979; Fisher & Braine, 1981.) Third, alignment affects discrimination of mirror images. Thus, lateral mirror images horizontally aligned (C 7) are the most difficult to discriminate, followed by vertical mirror images vertically aligned ( u), n followed by lateral mirror images vertically aligned ),( C and vertical mirror images horizontally aligned (U (e.g., Huttenlocher, 1967b; Sekuler & Houlihan, 1968).

n)

85

111. Two Classes of Perceptual Anisotropy in Infancy A.

PERCEPTUAL SALIENCE IN INFANTS

Studies of infant spatial vision suggest that babies see as perceptually salient principal orthogonal-especially vertical-orientations or alignments of patterns. Studies that tap detection and discrimination, processing, and preference in infants all converge on this generalization, and it seems to hold for simple and complex artificial geometric forms as well as for more meaningful patterns like the human face.

I.

Vertical Is Perceptually Salient, Horizotital May Be, and Obliques Are Least So

a . Detection and Discrimitiation . The extreme perceptual disadvantage of obliques relative to the mainorthogonals, the so-called "oblique effect" identified by Appelle (1972), is usually discussed in the context of sensitivity differentials at a postretinal level and is usually studied (and usually emerges) at threshold levels of stimulus contrast. Studies of infant visual detection and acuity that follow in this tradition show that at threshold babies also discern verticals and horizontals better than they do obliques. In addition, infant orientation discrimination has been studied with suprathreshold stimuli. Infants have been shown to discriminate easily between the orthogonals, between orthogonals and obliques, and even between some obliques, though doubtlessly they fail at increasingly fine orientation discriminations (Fellows, 1968). Leehey, Moskowitz-Cook, Brill, and Held (1975) provided the classic demonstration of oblique effects at threshold in young babies. They studied 24 infants between 1.5 and 12 months of age. Leehey and her colleagues used a twoalternative preferential-looking task with 1 1" square-wave gratings displayed at five spatial frequencies (0.75, 1.5, 3.0, 6.0, 12.0 cycles per degree) that were aligned along the vertical, horizontal, or two 45" obliques. Over 10-second stimulus exposures in which an orthogonal was always paired with an oblique of the same spatial frequency, babies at all ages looked at verticals and horizontals equivalently, at obliques equivalently, but at the main orthogonals more than at the obliques. For each age, the maximum difference in frequency of looking at the main orthogonals versus obliques occurred at or near the acuity threshold spatial frequency. Thus, across the first year of life, vertical and horizontal gratings are easier to detect than are otherwise identical oblique gratings. Gwiazda, Brill, Mohindra, and Held ( 1 978) replicated these results and also found that acuity for vertical gratings increases more rapidly over the first year of life than does acuity for oblique gratings.

86

Murc H . Bornstein

Suprathreshold discriminations of orientation are much more common in the infant literature. Although Ling (1941) early on concluded that 6-month-old infants do not discriminate orientation differences in perceiving form, several investigations since have determined that infants facilely discriminate among many changes in the orientation of a stimulus. Orthogonal, that is, vertical-horizontal, discriminations have been the most prominent and the easiest to demonstrate (e.g., Bornstein, Gross, & Wolf, 1978; Essock & Siqueland, 1981; Fagan & Shepherd, 1979; Gross & Bornstein, 1978a; McKenzie & Day, 1971; Moffett, 1969). For example, Gross and Bornstein ( 1978a) demonstrated infants' simultaneous discrimination between orthogonals in a relatively simple and clear way. Seven 4-month-old babies were familiarized over a series of trials with a pair of identical geometric stimuli (1C ) that were approximately 2 1" on a side. On every other trial after familiarization, either a 90" rotation (u)or a 180" rotation ( 3 )of the original familiarization stimulus was substituted for either the left or right member of the familiarization pair. If infants looked more at the probe than at the familiar stimulus. it would indicate that they discriminated the probe from

75

-PREFERENCE

FOR

I I P R E F E R E N C E FOR

u 1

65

55 Oh

45

PREFERENCE

35 25

15 5 N=7 Fig. I . Foll~~witig,fuinilicrrizuti~~~~ with a pair of C shapes. itgonts prefir u 90" orthogonul rottition {!A)of the stimulus to LI 180"niirror-irnuge rotmion (3) of the .stinrulus. 111this r.rprvirtrent, preference indictrtes discrimincrtion (NS.not signijicmr).

the familiar stimulus. (Of course, failure to show a preference would not necessarily give evidence of inability to discriminate.) The results of the u probes are pertinent here. (Results of the left-right mirror-image probe, 7, will be discussed in Section 111,s.)On probe trials, babies looked more at u than at the original L , that is, they discriminated the 90" rotation (Fig. 1). Orthogonal-oblique and oblique-oblique discriminations-even those above normal limits of resolution-are thought to present increasingly difficult problems for young children. Reasonably fine orthogonal-oblique discriminations have been studied in three infant experiments; both geometric and realistic stimuli have been employed. Using a habituation-test paradigm, Bornstein er ul. ( I 978) showed that 4-month-old babies discriminate vertical from 45". Twenty babies habituated to a 45"-right oblique line (1.5" x 12.5") over 10 successive 10-second trials. On nine test trials afterwards, the babies were shown the habituation stimulus, a vertical version of it, and its 45"-left oblique version. (Results for the mirror image are discussed in Section 111,B.) In the test, the babies dishabituated to the vertical relative to the habituation oblique (Fig. 2). Using a similar habituation-test paradigm, Weiner and Kagan ( 1976) examined orientation discrimination between a horizontal line and one rotated 35" from the horizontal. Their 5-month-old babies successfully discriminated between the orthogonal and the oblique. Finally, using a paired novelty-preference paradigm, Fagan and Shepherd ( 1979) studied orthogonal-oblique discriminations with faces. (Here it must be borne in mind that faces are special perceptual stimuli.) Fagan and Shepherd found that infants between 5 and 6 months readily discriminated a 45O-diagonal from a vertical but failed to distinguish diagonal from either horizontal or from upside down. Thus, the upright vertical face is special and uniquely discriminable from diagonal, while horizontal and diagonal are treated similarly (see Braine, 1978a. 1978b). Faces, like symmetry, represent a clear case in which vertical possesses a higher status in the orientation hierarchy than does horizontal. In short, babies in their first half-year discriminate at least 35"-45" rotations of a stimulus from an orthogonal, especially vertical. Finer discriminations have yet to be tested. Oblique-oblique discriminations (like all linear discriminations) are in fact discriminations between mirror images since any pair of obliques falls equally on two sides of a main orthogonal or a midoblique. The few extant studies of infant oblique-oblique discriminations therefore can be placed into two categories. One category encompasses discriminations between obliques that are mirror images about an ohliqite. Vertical-horizontal and orthogonal-oblique discriminations also fall into this category. (The second category encompasses discriminations between obliques that are mirror images about an orthogoizal; a more complete discussion of them will be presented in Section 111,B.) One empirical demonstration of the infant's ability to discriminate between two obliques was provided by

88

c

35t 0 4

301 ......

1':I

:.:, .........,....., .........._.... ..............., ........,...... .........,...... ........,...... ................ ...............

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

L

0

I

\

45ORIGHT

VERTICAL

45" LEFT

Fig. 2 . Following hubituutioti to (I 4S"-right oblique, injiu7t.s muintuin habituation to the 4 5 " - r i ~ h t oblique. they dishubituute to vertical, und they generulize hnbitucition to (I 4S"-lefi oblique, the lutcrril mirror imugr ($the hubituution stimulus. I n this experini~nt,dishabitucition indicures disc.riminc~tion.(A,ftrr Bornstoin, Gross. & Wolf. 1978, Experiment I I . Copyright Elsevier Sequoici S . A . , used by pertnis.sicin. )

Bornstein et (I/. (1978). We tested 10 4-month-olds for their ability to discriminate two rightward-tilting oblique lines-one tilted 20" right from vertical and the other tilted 70"-that were mirror images about the 45" oblique. Babies were habituated to the 20" line ( I " x 12") over 10 successive 10-second trials and immediately tested twice with the line rotated 70". The babies recovered looking from the end of habituation to the test (Fig. 3), that is, they discriminated 50" of rotation between two obliques in the same Cartesian quadrant. This is the only infant study to my knowledge in this category*; to define early capacity further, ZEssock and Siqueland (1981) found that 2-month-olds failed at a successive oblique-oblique discrimination between gratings that were mirror images about an oblique oriented 22.5" left of vertical; however, their experiment is flawed as a study of orientation discrimination since their stimuli, oriented 22.5" right of vertical and 67.5" left of vertical, confound a left-right discrimination with a pure 90" oblique-oblique orientation discrimination. The experiment probably falls into the category of confusable left-right mirror images (see Section 111,B).

89

50 w 5

-

I-

g40 Y

0

s

5 30

W 0

K W

a

z 20 a

W

H

:

.

..

J

I

4

20° RIGHT

70' RIGHT

Fig. 3. From the end of huhirrcution to (1 20"-right obliquc. infunrs dishubituute to a trst 70°-right oblique. I N rliis experiment. clishabitiiation indicates di,srriminution. (After Bornstein, Gross, & Wolf; 1978. Experiment 111. Copyright Elsevier Seyuoiu S . A . , used by permission.)

future research in orientation discrimination ought to be concerned with the development in infancy of visual resolution of a variety of small orientation differences. Finally, it is worthwhile to consider the role of stimulus complexity or organization in a discussion of infant orientation discrimination. Fisher, Ferdinandsen, McCall, Kennedy, and Appelbaum ( 1977) also assessed oblique-oblique orientation discriminations in 2.5-month-olds in a study of the discrepancy hypothesis. Their stimulus was a multicolored shaft with arrowheads at both ends and was oriented o", 33", 66". and 90". Infants' discriminations among all possible pairs in both directions from vertical were studied in a habituation-test design, which included two orthogonal (90")discriminations. two magnitudes (33" and 66")of orthogonaloblique discriminations (vis-a-vis both vertical and horizontal), and two oblique-oblique (33") discriminations. Unfortunately, the design of the experiment and analysis of the results did not permit separate assessment of orientation discrimination: "We caution against attempting to interpret the unanalyzed mean fixation times. , . . Tne major point of the design is that such means are thoroughly confounded-and therefore not interpretable. . ." (McCall rt d , , 1977, p. 779).

Mtrrc H. Bornstein

90

and Bornstein (198 1 ) compared,symmetry and orientation discrimination in four groups of 12 4-month-old babies. This discrimination experiment used a singlelook infant-control procedure (Horowitz, 1974). Babies were habituated with one symmetry pattern (vertical, horizontal, or asymmetrical) to a fixed criterion; afterward, they were tested with a different symmetry pattern. All possible pairings were tested. The patterns were closed symmetrical or asymmetrical polygons that were equated for contour, perimeter, area (of approximately 20” square), and number of turns. Babies discriminated vertical symmetry from asymmetry and vertical symmetry from horizontal symmetry (and vice versa), but they did not discriminate horizontal symmetry from asymmetry or asymmetry from asymmetry. This pattern of results suggests, first, that the global organization embodied in vertical symmetry promotes perceptual discrimination and, second, that the infant’s perceptual advantage for vertical symmetry reflects an interaction between orientation and the unique qualities of symmetry. Infants seem not to be sensitive to the structural organization in symmetry if that organization is aligned about the horizontal axis (rather than the vertical). Vertical symmetry is special, and vertical is special. In summary, babies discern orthogonals best, that is, they show an “oblique effect, at threshold, but they can discriminate between orthogonals, orthogonals and obliques, and obliques and obliques if above-threshold levels of stimulation are used and the stimuli are simple. Though still untested, the resolution of angular disparity by the infant visual system is probably comparatively deficient relative to adults; the child literature (see later) suggests that it ought to be increasingly deficient away from the orthogonals. When more complex stimuli are used, vertical separates from horizontal and shows a perceptual advantage. This review of detection and discrimination studies suggests that for infants the main orthogonals-particularly vertical-are perceptually special next to obliques. ”

b. Processing. The perceptual salience of vertical, or of the two main orthogonals generally, has also been studied by submitting to experimental scrutiny infants ’ long-term visual processing of stimuli that are constant in form but varying in orientation. Presumably, differences in infants ’ long-term patterns of attention, for example, their rates of habituation, to the same stimulus oriented in different directions indicate something about the relative salience of different orientations for them. When E. J. Gibson, Owsley, and Johnston (1978) habituated infants to two types of rigid motion, they found that the babies’ rate of looking descended to a constant habituation criterion faster for a stimulus rotated around a vertical axis than for one rotated around a horizontal axis. As part of a larger study on the perception of symmetry in infants, Bornstein, Ferdinandsen, and Gross ( 198 1) investigated rate and amount of habituation to

91

Perc~eptualAnisotropies in Injkncy

visual stimuli that varied in symmetry or in orientation of symmetry. Eighteen 4-month-old infants habituated to six-element red geometric patterns that were symmetrical about the vertical or horizontal axis or that were asymmetrical; the stimuli were otherwise equivalent in terms of perimeter, contour, and area (approximately 21" on a side). Rate of habituation was the principal measure; rate was assessed by first dividing the 240-second habituation period into 24 consecutive 10-second segments, then using average looking in the first three segments as a baseline for each child, and finally calculating the number of segments for each stimulus type that each child required to reach a fixed habituation criterion (e.g., three consecutive segments <50% of the baseline). During the course of separate 4-minute exposure periods, the babies reached criterion fastest when they were looking at the vertical alignment; they habituated equally slowly to the horizontal alignment and to the asymmetrical pattern (Fig. 4). Over the 4 minutes, babies also declined most to symmetry oriented along the vertical axis and less, by equal amounts, to symmetry along the horizontal axis and to asymmetry.

1

Fig. 4 . Infints reach a hahitiration criterion ,fastest (*) with vertical symmetry; habituation to horizontal symmety and to asymmetr). occurs ar equiwlent rates. (Two versions of each symmetry type were used; the I K J O versions produced identical results.) In this experiment, rate of habituation indicate.~.fuciliryin stimirlus processing. (Afier Bornstein, Ferdinandsen, & Gross. 1981, Experiment 2 . Copwight 19x1 by the American Psychological Associution. used by permission ,)

Vertical clearly facilitated the perception of symmetry and stimulus organization in babies; these habituation results are consonant with those of the discrimination study of Fisher et a l . (1981) reviewed earlier. Further, the evidence for an early perceptual advantage for vertical over horizontal symmetry indicates in a second way that infants respond to symmetrical organizations in a hierarchical manner with regard to orientation; in many experimental situations, children and adults respond in the same way (see, e.g., Boswell, 1976; Fitts & Simon, 1952; Hogben, Julesz, & Ross, 1976; Rock & Leaman, 1963). Finally, in an unpublished study that continues these symmetry investigations, I found that vertical is so special that it may act as the prototype of a category of near-vertical patterns (Bornstein, 1980). Ten 4-month-old babies habituated over 12 exposures to a geometric pattern that was symmetric about near-vertical oblique axes (5”, lo”, and 15” left and right of vertical), and the babies were subsequently tested with the three habituation obliques, a novel 7” oblique of the same pattern, and the vertical version of the pattern. (The stimuli were the same ones used by Fisher et al., 1981.) In the test, babies treated the three habituation obliques equivalently; they maintained an equivalent habituation to the novel oblique; but they actually suppressed looking at the vertical relative to the familiar and novel obliques. Apparently, familiarization with a variety of near-vertical symmetries constituted “over familiarization with the never-before-seen vertical: The babies behaved as though they abstracted vertical as common to the habituation series and then treated vertical as the prototype of the series showing suppression to it. In summary, infants ’ long-term experience with complex static or dynamic patterns reveals a particular perceptual advantage for vertical. Habituation to vertical is faster and more complete, that is, it is more efficient, and vertical may be the prototype for a category of oblique near-vertical orientations. ”

c . Preference. The “oblique effect” is well known to be a threshold phenomenon. But, when other factors that trade with stressed or threshold performance, as, for example, complexity or age, are manipulated, a vertical or verticalhorizontal salience in orientation perception emerges. Several studies show this in ways startlingly simpler than implied by detection, discrimination, or processing experiments. For example, given equal exposure to suprathreshold contours or patterns oriented in different directions, newborns and infants tend to look longer or prefer the vertical or the vertical and horizontal to obliques. Three studies that used increasingly complex, but static stimuli show consistent preference for vertical or main orthogonal patterns over the first year of life. In the first study, Kessen, Salapatek, and Haith (1972) assessed the visual response of 16 newborn infants to a black-white edge oriented vertically and displaced 15” to the left or right of the center of its circle, or oriented horizon-

tally, 15" above or below the circle center. During 60-second exposure periods, babies 1- to 4-days-old preferentially oriented to and fixated (crossed) the vertical edges, while the horizontal edges exerted "no effect at all. " Even from the first hours of life babies manifest a consistent preferential anisotropy for vertical. In the second study, Bornstein (1978) gauged the persistence of visual attentiveness in 20 4-month-old infants shown four orientations of a red-white 4.5cycle square-wave grating (25" on a side). During two 10-second exposure periods to each orientation, babies looked reliably and consistently longer on average at gratings aligned vertically or horizontally than at gratings aligned along 45" left or right obliques (Fig. 5). In the third study, Bornstein et (11. (1981) examined visual preference in 15 4-month-olds and in 15 12-month-olds shown three different complex sixelement red geometric patterns, one symmetrical about the vertical axis, one symmetrical about the horizontal axis, and one asymmetrical. (These stimuli are described above.) During 10-second exposure periods in which the three patterns were shown pairwise in all possible combinations, 4-month-olds failed to show a preference, but in contrast 12-month-olds reliably preferred the pattern symmetrical about the vertical to both the horizontally symmetrical and asymmetrical variations, which they preferred equally (Fig. 6 ) . The 4-month-old babies did not prefer vertical symmetry; recall, however, that they processed vertical most

Q

$ $

k

L

-

11111

G RAT1N G

*

@

0R1ENTAT 1ON

Fig. 5 . lnfrints prefer horizontcil cind vertical versions . f a gritting to ~JhliqUeones. In this expcrimant. (rmoiint of looking indicrrtcs prqfirence. (Ajirr Bornstein, 1978. E.rperiment I I . C ( ~ p . ~ i g h t Acndemic Press, I n ( . . , used by permission.)

Murc H . Bornstein

94

2

I

I

I

Fig. 6. 1nJirnt.T of4 months show no prcferenre, bur infints of 12 months prefer verticul symmetry (*) to horizontul symmetry mid usymmetry. (Two versions of euch symmetry type were used; the two versions produced idenrital results.) i n this experiment. amount of looking indicutes preference. (After Bornstein, Ferdinundsen. & Gross, 1981, Experitnent 3 . Cop.yright I981 b y the Americun Psyrhologicul Associotion, used by permission.)

efficiently. Perhaps with relatively complex stimuli, infants come to prefer what is clear or what is cognitively easy. These three studies, which show vertical or main orthogonal preference over the first year, all used static stimuli. At least one additional infant study has found similar results with dynamic stimuli. Ivinskis and Finlay (1981) assessed heart-rate changes in 48 4-month-old infants shown black rectangles (2.3" x 17") moving vertically, horizontally, or diagonally across the visual field at slow ( . 2 cycles per second) or at fast ( 2 . 3 cycles per second) rates. During 10-second exposure periods, the vertical and horizontal orientations evoked significantly

Perceptutrl Anisotropics in I n f t n c ~ ~

95

greater average heart-rate deceleration (orienting) at both speeds than did the diagonal orientations of the rectangle. In summary, babies between birth and about 1 year orient to and look longer at, in short, prefer vertical or vertical and horizontal to oblique. Not all studies have directly compared vertical with horizontal or the main orthogonals with obliques; nevertheless, the principle generalizes across the results of four studies conducted with infants over the first year and holds with behavioral or psychophysiological measures for static or dynamic, simple or complex, and chromatic or achromatic suprathreshold stimuli. Studies that show a statistical vertical-horizontal equivalence (e.g., Bornstein, 1978; McKenzie & Day, 1971; Moffett, 1969; Watson, 1966) tend to favor vertical, though this advantage may interact with stimulus complexity or organization, astigmatism in infants, or the age of the infant subjects i n ~ o l v e d . ~

2. Discussion Infants show an early perceptual salience for vertically or for orthogonally oriented visual patterns. This anisotropy emerges in measures of detection and discrimination, long-term processing, and preference. Explanations at any of four levels at least might account for this orientational hierarchy in infants. The first level is neurophysiological. The orthogonals may be attractive to infants simply because their nervous systems are “tuned” to perceiving orthogonals. A direct relation between specific stimulation and central nervous system activity (particularly in vision) is today unchallenged (e.g., Hubel & Wiesel, 1962, 1968; Regan. 1972). In this light, several investigators have suggested that a behavioral “oblique effect” may be directly related to any one or more of the following neurophysiological findings: ( I ) a greater proportion of orientation-sensitive units that analyze central or foveal vision in the striate cortex in primates is sensitive to vertical and horizontal meridians than to oblique ones (e.g., Bouma & Andriessen, 1968; Mansfield, 1974; Mansfield & Ronner, 1978; Marg, Adams, & Rutkin, 1968); (2) cells that respond preferentially to vertical and horizontal are more sharply tuned than cells that respond preferentially to obliques (Nelson, Kato, & Bishop, 1977; Rose & Blakemore, 1974); or (3) vertical or horizontal stimulation yields greater amplitude evoked cortical potentials than oblique stimulation (e.g., Freeman & Thibos, 1973; Frost & Kaminer, 1975; Maffei & Campbell, 1970; May, Cullen, Moskowitz-Cook, & ‘Many young infants are to a degree astigmatic, mainly along the vertical or horizontal axes (Mohindra. Held, Gwiazda, & Brill, 1978). Atkinson and French (1979) have shown that, given a choice, astigmatic infants tend to prefer gratings oriented along the axis of their astigmatism. (Normals show no preference, and single-stimulus designs would be unaffected by infant astigmatism.) In this way, Atkinson and French ( 1979) account for Slater and Sykes’s (1977) otherwise anomalous result that 1 1 of 14 neonates in a paired presentation design preferred horizontal stripe pattcrns to vertical ones.

96

h4~1r-c H. Bornstein

Siegfried, 1979; Yoshida, Iwahara, & Nagamura, 1975). It may be, therefore, that the perceptual bias for orthogonals reflects the general fact that vertical and horizontal are “trigger features” for a greater number of cortical neurons or more finely tuned neurons and that the visibility, discriminability, or amount of visual attention that a stimulus structure elicits is governed by the level of neural excitation that it engenders (Bornstein, 1975, 1978; Haith, 1978, 1980; Karmel & Maisel, 1975).4 A second level of explanation for this selective anisotropy is anatomical. The perceptual bias for vertical, particularly, may be rooted in the gross bilateral symmetry of the organism as a whole or, more specifically, of the perceiving visual system. Mach (1886/1959), Julesz (1971), and Corballis and Roldan ( 1975) have all favored an explanation that maintains that organisms systematically orient themselves symmetrically vis-a-vis external stimulation so as to match ambient information most efficiently (e.g., Regan & Beverley, 1978, 1979). Based on the results of studies that showed rapid detection of symmetry, Julesz (1 97 I ) went so far as to speculate that a template for symmetry detection is embedded symmetrically in the brain and is “built-in. ’ ’ A third level of explanation is motor. The fact that infants scan along the horizontal more extensively and at a greater frequency than along the vertical (Kessen et al., 1972; Salapatek, 1968, 1975; Salapatek & Kessen, 1966) facilitates perception of information oriented along the vertical. Vertical information would in this way draw more attention to itself since intensity change, a basis of perception, would be more frequent on the vertical. A fourth level of explanation is experiential. The special status of vertical (and horizontal) could result from the infant’s immediate and dense experience in a visual world dominated by orthogonal, particularly vertical, contours or forms. It may be important to note in this connection that one unique experience that is structurally universal among children could function determinatively in this regard. The human face is an especially significant and potent form for infants. Prior to 4 months of age, infants are thought to respond to a face on the basis of isolated features (e.g., the eyes), but older babies seem to respond to faces on the basis of wholistic properties (for reviews, see Cohen, DeLoache, & Strauss, 1979; E. J . Gibson, 1969). The vertical symmetry in faces is one such wholistic property that may be very influential in the developing infant’s bias for vertical. Doubtlessly, endogenous neurological factors and predominant orientations in the environment both influence the course of perceptual development. Annis and Frost (1973) and Timney and Muir (1976) have studied orientational anisotropies for acuity in different human groups reared under different ecological conditions 4The cell population bias in primates that favors vertical and horizontal is also present in adult cats and, intriguingly, in kittens less than 1 month old, independent of the nature of their early visual experience (Frkgnac & Imbert, 1978; Leventhal & Hirsch, 1975, 1977).

and found that the degree of anisotropy or "oblique effect" in adults seems to be related both to genetic inheritance and to early experience in a "carpentered environment. "

B.

PERCEPTUAL EQUIVALENCE IN INFANTS

Though infants discriminate relatively small rotations of a pattern acutely, they naturally treat as perceptually equivalent select orientation changes. Left-right mirror-image reflections of a pattern, for example, seem to be perceived as equivalent; up-down reflections are also perceptually similar, but to a lesser degree. Thus, mirror-image equivalence about the main orthogonal axes constitutes a second class of perceptual anisotropy early in life. Again, infant studies of various kinds show this effect with artificial geometric patterns as well as with more realistic stimuli-like faces. Mirror-Imugt. Equivalence Occurs ubolrt the Main Orthogonals, Not about Obliques Infants' equivalent treatment of lateral and similar treatment of vertical mirror images has been studied in tests that require both simultaneous and successive discriminations. Since the typical child or adult study has most prominently pointed to lateral mirror-image equivalence under conditions in which comparison patterns are displaced in time, this review of the anisotropy in infancy begins by examining lateral mirror-image equivalence under these conditions. I.

a . Lateral Mirror Imuges: Geometric Patterns. Essock and Siqueland (198 I ) studied discrimination of square-wave gratings (34" of visual angle) among 36 2-month-olds in a high-amplitude sucking habituation-test design. One condition tested vertical-horizontal (0"-9OO),and a second tested left-right mirror-image obliques (45"-315"). Relative to appropriate no-shift control groups, babies in the orthogonal condition recovered, but babies in the mirrorimage condition did not. (Essock and Siqueland also found that babies failed at a 22.5" right-67.5" left near mirror-image discrimination; see Footnote 2.) Thus, 2-month-olds treat lateral mirror images as equivalent. As part of a larger study of orientation discrimination (discussed earlier), Bornstein rt ul. (1978) tested the ability of 4-month-old infants to discriminate between left and right 45" tilts of a line, that is, between lateral mirror-image obliques. Twenty babies were habituated over 10 successive 10-second trials to a line (1.5" x 12.5") tilted 45" right of vertical, and they were then tested on three trials each with the same line, its vertical, and the line tilted 45" left of vertical. The results showed that following habituation infants ciishabituated to (discriminated) the vertical but that they did not dishabituate to either mirror-image oblique (Fig. 2). In other words, babies discriminated an angular displacement of

98

Murc H . Bomsrein

45” when the two stimuli were an oblique and a vertical, but not one of 90” when the two stimuli formed 45”-oblique mirror images. It is possible (though not likely) that in this study looking during the test phase was independent of prior habituation and that babies simply preferred the test vertical over the test obliques (as babies do in some situations, e.g., Bornstein, 1978); this would imply that babies do not discriminate among obliques. However, the babies did habituate, and they can discriminate between obliques (as, e.g., 20” and 70” lines; see Section III,A, 1). These facts tend to discount any alternative interpretation of the mirror-image equivalence in terms of the inability of 4-month-old infants to discriminate any two obliques. In order to assess the generality of lateral mirror-image equivalence for infants and in order to discern specifically whether an “oblique effect”-the general insensitivity of infants to obliques (see Section II1,A)-contributed to the infants’ equivalent treatment of oblique mirror-image lines, Bornstein er id. (1978) conducted another study. In this one, infants’ perception of the equivalence of mirror images that were not obliques was examined; the stimuli used were C shapes. Furthermore, a different experimental paradigm was used in this study to provide converging evidence on the perceptual equivalence of lateral mirror images for infants: The first study used a within-subject habituation-test design, while the second study compared amounts of habituation to different patterns between groups of infants. Four groups of 10 4-month-olds experienced different stimulus presentation conditions. Group 1 saw a “standard” stimulus ( C ) on each of 18 discrete 10-second trials. Group 2 saw the same standard stimulus ( C ) on some trials and a “context” stimulus, a 90” rotation of the standard (n),on other trials. Group 3 saw the standard (C)and its other 90” rotation (U). Group 4 saw the standard (C) and its lateral mirror image (I). To promote comparability among groups, the standard stimulus was shown the same number of times and on the same trials to each group. Over a fixed amount of time, habituation of visual attention in infants to repeated stimuli is greater than is habituation to varied stimulation (e.g., Bornstein, Kessen, & Weiskopf, 1976; Cornell, 1974; Fantz, 1964); therefore, if infants distinguish rotations of the standard the amount of habituation in Group 1 would be greater than in Groups 2, 3, and 4. To the extent that infants perceive lateral mirror-image stimuli as equivalent, habituation in Group 4 should be similar to that in Group I and greater than that in Groups 2 and 3 . Of the 18 stimulus exposures, 10 were the standard stimulus, which appeared on the first three trials for all groups. To facilitate comparison among groups, each baby’s data were converted to percentage scores using the average of the initial three trials as a baseline, and the relative amounts of habituation in different groups were assessed by comparing the average percentage decrement in looking time from the first three standard stimulus trials to the last three standard stimulus trials in the series. Group 1 (standard stimulus) and Group 4 (lateral mirror images) showed a reliable decrement in attention between the beginning

99

Pi~ri.eptuu1Anisotropies in Infuncy

and end of habituation, that is, they habituated over 18 trials; Groups 2 and 3 (90" rotations) did not habituate (Fig. 7). Within-groups analyses showed that babies treated an intermixture of the standard stimulus and its lateral mirror image just as they did repetition of the standard stimulus alone and that they treated these conditions differently from 90" intermixtures. The results suggest that 4month-old babies view a stimulus and its lateral mirror image as perceptually equivalent or, more conservatively, as more similar to each other than either is to a 90" rotation. Across these three studies, 2- to 4-month-old babies treated lateral mirror images, reflections about the vertical axis, as equivalent, regardless of the form. In the first and third studies, babies differentiated vertical and horizontal, which are reflections about a 45"-oblique axis, and, in the second study, babies differentiated vertical and 45", which are reflections about a 22.5"-oblique axis. Recall, too, from the 20"-70" comparison reported above that babies also discriminate obliques that are mirror images about a 45" oblique. Mirror images about the vertical orthogonal are special in that they are treated as perceptually equivalent; those about oblique axes are not. 4c I-

Z

W

5 wE3 0 K

300 I-- Z 20 Z Y W O

p a

z

10

Z

a w 2

0

STANDARD: CONTEXT: GROUP;

r 1

L

C U

C

3

4

1

Fig. 7. Infirtits show /urge trnd eyuivulent decrements in looking time between the beginning und pnil of htrbituurion t o /hi, ri,petition of u single stundiird ,stimulus (Group I ) irnd t o the intermixing over tritrls (f t i stundord irntl i t s lrtercrl mirror imiige (Group 4 ) . 1tlfirnt.s clc.i.linr less when the .strrn~lurcluncl either of its YO" rotutioris lire intermixed (Groups 2 o r 3 ) . Thr stundurd .srimulu.s crnd the c.onte.rt ,stimrtlrt.s shown to euch group ur(' indicated on the ubscissir. In this i q e r i m e n t , umount of ~ l ~ ~ c r e mindii~utes ent &icirnc). in stimulrr.s processing, (Afrer Bornstein, Gross, & Worf; 1978, E.xpiviment I V . Copyrigkt Elsevier Seyiroiir S.A., used by permission.)

I00

Mtm. H. Bormtf~ill

Human infants, like human children and adults and like other infrahuman species, treat lateral mirror images as perceptually equivalent, especially when the comparative judgment they are required to make is successive. When comparing lateral mirror images presented simultaneously, however, mature animals and human children and adults display much improved discrimination. Thus, while a wide variety of subjects finds it difficult to learn to respond consistently over a series of trials to one member of a pair of mirror images, the same subjects can pick out the "odd" stimulus when presented with two identical patterns and their mirror image (e.g., Bradshaw et a l . , 1976; Over & Over, 1967; Tee & Riesen, 1974; Wohlwill & Wiener, 1964). The difficulty of discriminating mirror images for different organisms seems therefore to involve coding in memory. How well do human infants discriminate simultaneously appearing lateral mirror images? Gross and Bornstein (1978a), in the study of simultaneous discrimination of orthogonals referred to earlier (Section III,A, I ) , also studied infants' simultaneous discrimination between lateral mirror images. The subjects were seven 4-month-olds. They were shown pairs of C shapes simultaneously over a long series of trials, and after the initial sequence either a 90" rotation of the shape (u) or a 180" lateral mirror-image rotation (7)was substituted for either the left or right C on every other test trial. Increased looking at U or 7 relative to C on the probe trials indexes discrimination. Babies discriminated U from C (as discussed above), but they showed no significant preference for 7 over the original C (Fig. 1). This result indicates that babies found the mirror-image transformation 1 equivalent, or at least more similar, to the original C than they found the 90" rotation. Thus, even when two mirror-image enantiomorphs are simultaneously present, babies (unlike children or adults) tend to treat the two as equivalent to each other or, at least, as less different than main orthogonal transformation^.^ Mirror-image equivalence may be stronger in immature organisms than in mature ones. b . Lateral Mirror Images: Faces. The study of mirror-image equivalence in infants has involved the perception of artificial geometric patterns as well as more realistic stimuli-like faces. Experiments on both successive and simultaneous tasks show that infants in the first half of the first year of life treat mirror-image profiles, diagonals, and sideways representations of the same face as equivalent. The fact that infants treat the left and right halves of the face as equivalent is important in light of the theory of mirror-image equivalence advanced in the Discussion (Section 111,B,2). Bornstein er (11. (1978) studied the perceptual similarity for infants of mirrorSAlthough they constitute only weak confirmations, Bornstein ( I 978)- studying 20 4-month-olds, and Atkinson and French (1979), studying 16 3-month-olds, both found no differential preference between pairs of left and right 45"-oblique gratings.

101

Perceptrial Anisorr~tpic~s i n Infuncy

image faces by examining whether infants would treat the right and left profiles of the same person as perceptually equivalent while discriminating between the right profiles of two different people. Ten 4-month-olds were familiarized with the rightward-facing profile of one man (approximately 27” x 23”) during a single 60-second exposure period. Afterward, the babies were tested six times each for 10 seconds with three profiles, namely, the original right profile and two additional ones, the leftward-facing profile of the same man and the rightwardfacing profile of a different man. (The two profiles were of adult males selected as the least similar pair from an experimental collection of male faces; see Goldstein, Harmon, & Lesk, 1971, Fig. 7.) Babies looked at the face during about 40% of the familiarization period. During the test trials, they looked at the original rightward-facing profile and the leftward-facing profile of the same man equivalently, but they discriminated the rightward-facing profile of the new man (Fig. 8). Infants familiar with the right profile of a face treated the left profile of the same face as equivalent, yet they distinctly discriminated the right profile of a new face from both the familiar and “unfamiliar” profiles of the original face. That babies looked longer at the profile of the unfamiliar face than at the familiar profile indicates that 1 minute constituted sufficient familiarization with faces to ensure discrimination. That the babies remembered the original profile is evidenced by their relative inattention to it in the test phase; as has been shown

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

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

............. ............... I

RIGHT

LEFT

FAM I L I A R “FA M IL I AR”

RIGHT NEW

Fig. X. Following fumiliuriz.utiori with thr right profile ojone man, infants muinticin hahituirrion t o the ri,qhtJumiliurprofile und grrrerulize habirurition to the kji ’ 3imiIiur’’ profile qf the sume mun, but they di.shtihituute l o the right profile o f t i new mun. I n this e.xperiment, increased looking indicutes discrimindon. (After Bornstein Gross, & W d f . 1978, Experiment I , Copyright Elsevier Seyuoiu S .A . , used by permi.s.sion.) I

102

Murc H . Bornstein

before, infants' recognition memory for faces is good (e.g., Fagan, 1979). That babies looked at the leftward-facing profile (which they had never seen before) as much as at the rightward-facing profile of the same face (which they had seen before) indicates that they treated the two profiles equivalently. The infants, however, clearly discriminated the new face from the familiar one, independent of the orientation of the familiar face. That the infants treated the left and right profiles as equivalent is unlikely to have reflected an inability to discriminate change in face orientation: Fagan (1976), for example, has demonstrated that infants can discriminate smaller changes in Orientation, such as a full face from a three-quarter view. Thus, under conditions of successive discrimination, infants treat mirfor-image versions of the same face as equivalent. Fagan and Shepherd (1979) studied the ability of 5- to 6-month-old infants to distinguish lateral mirror-image views of the same face when presented simultaneously. Twenty infants were shown a pair of photographs of a face positioned sideways to the right (or sideways to the left) for 30 seconds, and afterward they were tested for 10 seconds with the original position paired with its lateral mirror image; eight additional babies saw faces set on diagonals tilted either to the left or right. In both conditions, babies treated as equivalent left and right min-or-image views of the same face. In other conditions, babies discriminated between 90" and 180" rotations of the same face (see Section III,A,l and below). It is important to note that adults also tend to confuse or equate left and right mirror-image versions of a face (e.g., Bartlett, 1932). I once asked a group of young adults (mean age: 21.9 years) to indicate which direction Washington's profile on the U . S . quarter faces, to the viewer's left or right. Of 31 subjects, 42% indicated rightward, and 58% indicated leftward which, analyzed by binomial expansion, does not differ from chance (Bornstein et a / . , 1978). In summary, under conditions of simultaneous or successive presentation infants between 2 and 5.5 months treat lateral mirror images of artificial stimuli (geometric patterns) and of realistic stimuli (faces) as perceptually equivalent.6 6A recent study by Maurer and Martello (1980) constitutes an exception to this generalization. They habituated 24 I .5-month-old infants to a 45"-left grating (20" of visual angle) and then tested the babies with the habituation stimulus, i t s right mirror image, and a "negative" of it as a control. Babies selectively dishabituated to the test 45"-right mirror image. It could be that the absence of a counterbalanced group-one habituated to 4S"'right and tested with 45"-left-flaws the experiment and that 1.5-month-old babies, like their 2- to 5.5-month-old elders, actually would treat the class of mirror-image patterns equivalently. Alternatively, the result could be real. Maurer and Martello's infants represent the youngest studied to date. It could be that perceptual development proceeds through three stages wherein lateral mirror images are first discriminated, then not, then discriminated again. (Infants' learning that lateral mirror images of an object actually represent the same object may provide a mechanism for going from the first to the second stage.) It could also be that babies younger than 2' months see more with peripheral vision, as Bronson (1974) and others have argued; intriguingly, the orientation anisotropy is known to decrease with increasing retinal eccentricity (see Mansfield & Ronner, 1978). The contradiction in this alternative is that Maurer and Martello used their results, viz. orientation discrimination at 1.5 months, against Bronson and to fonify the view that central visual cortex is functional in the first months of life.

They discriminate mirror images about oblique axes. (Note, however, that stimulus selection is key in such demonstrations, and theoretically it would be possible to choose patterns or faces which are not discriminable.) Faces per se are special stimuli; they may also be special for the theoretical role they play in explaining mirror-image similarity. I shall turn to a discussion of the possible theoretical origins of this perceptual anisotropy after reviewing briefly the data that exist on infants’ equivalent treatment of vertical mirror images. c. Vertical Mirror Itrmges: Geometric Patterns. The foregoing experiments concerned the infant’s perception of lateral mirror images versus oblique ones. As suggested in the Introduction, human children and adults and infrahuman animals also tend to perceive as similar vertical, i.e., up-down, mirror images, although they are usually perceived as less similar than lateral, i.e., leftright, mirror images. Bornstein et nl. (1978) examined infants’ discrimination of vertical mirror images of geometric patterns by comparing habituation amounts in a procedure identical to that used for lateral ones (see above). One group (Group 5 ) of 10 4-month-olds saw a “standard” stimulus on every trial (n);a second group (Group 6) of 1 I 4-month-olds saw the same standard stimulus (n)on some trials, but they saw as its “context” stimulus on other trials, a 180” rotation (vertical mirror image) of the standard (u).As in the study of lateral mirror images, data for each child in this study were converted to percentage scores using the initial three standard trials as the base, and amount of habituation was assessed by comparing the percentage of looking time on these standard trials with that on the last three standard trials. Both groups decreased significantly and by equivalent amounts (Fig. 9). Thus, the group that saw only the standard stimulus and the group that saw the standard intermixed with its vertical mirror image showed similar habituation: Babies must find vertical mirror images as well as lateral ones perceptually similar. (This interpretation is strengthened by the results from the two groups, 2 and 3, which under identical conditions failed to show reliable habituation when their standard stimulus was intermixed with its 90” rotation.) Further evidence for vertical mirror-image similarity in infancy derives from preference studies by Watson ( 1966) and from discrimination studies by McGurk (1972). Watson ( 1 966) showed 48 2- to 6-month-old babies two pairs of simple geometric patterns in a paired-comparisons visual-preference test. One pair consisted of a dot at the top of a circle and its vertical mirror image, and the other pair consisted of a T in a circle and its vertical mirror image. During 30-second exposures, babies peered at the rightside-up and the upside-down pairs of dots and tees equally; that is, they failed in a preference test to discriminate vertical mirror images from one another, though this is not to say that they cannot discriminate between them. McGurk (1972) used a habituation-test design. He found that 3-month-old infants failed to discriminate vertical mirror images of a geometric pattern-unfortunately , the babies failed to habituate clearly in the

M u r c H. Bornstein

104

5 w

3

STANDARD: CONTEXT: GROUP:

0

nn5

1

n U 6

Fig. 9. Infants show equivalent decrements in looking time between the beginning und the end of’ huhitimtion to repetition ofu single stundard stimulus (Group 5 ) and to the intermixing over friulr of ( I srandurd und its vertical mirror image (Group 6 ) . The stundard stimulus and the context stimulus shown to euch group ure indicuted on the ahscisstr. In this experiment, amount ofdecrement indicates efirienry in stimulus processing. (Afrer Bornstein, Gross. & Wolf, 1978, Experiment V. Copyright Elsevier Sequoiu S.A . , used by permission.)

first place. With 6-, 9-, and 12-month-old infants, McGurk found clear discrimination between vertical mirror images. d . Verticul Mirror Images: Fures. Many experimenters have naturally been interested in whether babies “know” an upside-down face from one that is rightside up, and consequently assessing infants’ differential preference for (e.g., Watson, 1966) or discrimination between vertical mirror images of the same face has frequently recurred (e.g., Caron, Caron, Caldwell, & Weiss, 1973; Caron, Caron, Minichiello, Weiss, & Friedman, 1977; Fagan, 1972, 1979; McGurk, 1974). All these experiments show preference for the vertical upright and clear discrimination between vertical mirror images of a face. In this instance, perception of the face is not just like perception of any other patterns (Fagan, 1979). Watson (1966), for example, conducted two preference studies. In one that presented faces in different orientations singly for exposure periods of 20 seconds, he found that 20 4-month-old babies distinguished upright from upsidedown and sideways faces but may not distinguish upside-down from sideways faces; in this study, Watson measured magnitude of smiling and latency to smiling. In a second study that paired a rightside-up with an upside-down face for periods of 30 seconds, Watson ( 1966) found a distinct preference for the normal orientation of a face versus its inversion in 24 2-month-olds, where he had found no preference between vertical mirror images of geometric patterns; in this study, looking time was measured.

Fagan conducted two discrimination studies using vertical mirror-image faces. In one, Fagan (1972) tested 5- to 6-month-old babies for their simultaneous discrimination of upright and inverted versions of the same face following 60 seconds of familiarization with the upright: Babies differentiated the novel inversion. In the second study, Fagan (1979) tested 7.5-month-olds for discrimination of upright or inverted versions of a face following 40 seconds of familiarization with the upright; as in the 1972 study with younger infants, these older infants discriminated an inversion as novel. In summary, young babies may see vertical mirror images of geometric patterns as perceptually similar; older babies discriminate between them. Vertical mirror images of faces are always discriminated. Even babies in their first months treat an upright face as special. For babies, a face and its inversion are not equivalent mirror images as other geometric forms and patterns may be. Of course, it is well known already that adults frequently fail to recognize an inverted face, even of a familiar person (Rock, 1973; Yin, 1969). e . Laterul versus Verticul Mirror Imczges. Are lateral mirror images more similar than vertical mirror images for infants as has often been reported for older children and adults? Group 6, shown the vertical mirror images in Bornstein er crl. (1978), decreased 65% of the amount of decrease in looking produced by Group 4, shown the lateral mirror images. Although the experiments are not precisely comparable since the standard stimuli in the two differed, a comparison of Figs. 7 and 9 reveals that the general order of decreasing amount of habituation among all the groups across these experiments was ( 1 ) standard stimulus only or (2) standard and lateral mirror image, (3) standard and vertical mirror image, and (4) standard and 90" rotations. A test of ordered alternatives (Jonckheere, 1954) showed that the probability that this predicted order could have been obtained by chance is less than 3 in 100. For infants, as for all other species (Goldmeier, 1972), shapes that are symmetrical about a vertical axis, lateral mirror images, are judged more similar than shapes that are symmetrical about a horizontal axis, vertical mirror images. Shapes symmetrical about oblique axes are always easily discriminated. Under conditions of simultaneous or successive presentation, human infants treat lateral mirror images of realistic faces and geometric shapes as equivalent. Young babies may treat vertical mirror-image versions of geometric shapes as similar, though older babies discriminate between them. Young and old alike discriminate between vertical mirror images of the same face and see the vertical upright view as special.

2. Discussion Under a wide set of conditions human infants as young as 2 months of age, like older children and adults and like other infrahuman species, perceptually equate

I06

M i m . H . Bornsfcin

lateral and (to a lesser degree) vertical mirror images. What explains this behavior? The especial similarity of left-right enantiomorphs has historically been explained on the basis of the bilateral symmetry of the brain and body of the perceiving organism (e.g., Mach, 1886/1959; Noble, 1968; Orton, 1937). According to this view, left-right mirror-symmetric patterns perfectly match the receptive apparatus on both sides of the body and thereby confuse the organism with regard to lateral position. According to a modern version of this view (Corballis & Beale, 1976), the “representation” of an asymmetric stimulus in one hemisphere is a lateral mirror image of its representation in the other hemisphere, and this dual representation (somehow) leads to the confusion of lateral mirror images. Only the development of some behavioral or other bodily asymmetry, as, for example, of hemispheric dominance or handedness, can permit or facilitate lateral dissociation. Thus, development is thought to hold a key to the differentiation of left and right. However, data from several different sources argue against this explanation of mirror-image confusion. First, there is no physiological evidence for “mirror representation” in the two hemispheres of the brain nor are there any known interhemispheric connections that could provide this (Allman & Kaas, 1975; Brooks & Jung, 1973; Zeki & Sandeman, 1976). A second difficulty is that when visual information does transfer from one hemisphere to the other, there is no behavioral evidence that it mirror reverses in doing so (Corballis, Miller, & Morgan, 1971; Hamilton & Tieman, 1973; Lehman & Spencer, 1973; Storandt, 1974; but see Corballis & Beale, 1976). Third, mirror-image confusions persist in adulthood, even after the development of lateral asymmetries (e.g., Butler, 1964; Farrell, 1979; Pomerantz et a / . , 1977; Sekuler & Houlihan, 1968; Wolff, 1971). Indeed, fourth, in Gerstmann’s syndrome, where there is left parietal damage and consequent brain asymmetry, left-right mirror confusions are extreme (Critchley, 1953). Finally, there are convincing infant data that cast serious doubt on any purely developmental or experiential hypothesis. Anatomically, human infants are born with asymmetrical brains (Wada, Clarke, & Hamm, 1 9 7 3 , and they show electrographic asymmetries between the two hemispheres (Davis & Wada, 1977; Molfese, Freeman, & Palermo, 1975). Thus, human infants are not bilaterally symmetrical. Behaviorally, infants tend regularly to favor one side (e.g., Bresson, Maury, Pieraut-Le Bonniec, & de Schonen, 1977; Caplan & Kinsbourne, 1976; Gardner, Lewkowicz, & Turkewitz, 1977; Glanville, Best, & Levenson, 1977; Turkewitz, Gordon, & Birch, 1965). As a consequence, the bilateral-symmetry explanation of lateral mirror-image “confusion” is inappropriate for infants, as it is for adults and other organisms. Elsewhere, we have proposed an alternative explanation for this phylogenetically widespread behavior (Bornstein ot a / . , 1978; Gross & Bornstein, 1978a, 1978b), an explanation that also accounts for its early ontogenetic appearance.

The extreme prevalence and phylogenetic ubiquity of left-right equivalence suggest that, far from a confusion, it may reflect an adaptive mode of visual information processing. Presumably, the selective pressure of evolution made it advantageous for the visual system to be able to perform certain types of visual processing, whereas other modes were irrelevant for survival. In the natural world there are rarely mirror images that would be useful for an animal to distinguish. Indeed, with two exceptions there are virtually no mirror images at all. One exception is the two sides or profiles of a face or, more generally, the two sides of a bilaterally symmetrical animal. But here the two sides are two aspects of the same thing, and it would be adaptive to treat them as the samenot to distinguish between them. Another exception is that the silhouette of an object viewed from one side is the lateral mirror image of the silhouette of the same object viewed from the opposite side. Again, it would be adaptive to treat as similar, not distinguish between these mirror images. In other words, it is possible that the confusion of mirror images is not a “confusion” but an adaptive mode of processing visual information. In the natural world, virtually the only mirror images that ever occur represent twin aspects of the same object or organism and therefore need not be distinguished. Lateral mirror-image equivalence reflects bilateral symmetry, not of the perceiving organism as originally thought, but of significant objects or organisms in the perceptual world. It may be advantageous, therefore, to conceive of the difficulty of discriminating mirror images not as a “confusion” at all, but as an adaptive “perceptual equivalence” of a stimulus and its 180” reflection around the vertical (or horizontal) axis. An implication of the evolutionary view is that the perceptual similarity of mirror images may be present early in life and may not require maturation or extensive experience .7 The common bilateral-symmetry-of-the-body theory does not explain the similarity of vertical mirror images either, but it may also be unparsirnonious to presume that vertical mirror-image similarity could have a totally different explanation from lateral mirror-image equivalence. Lateral mirror images are treated as perceptually equivalent, and they also appear frequently in nature, for example, as faces. Vertical mirror images may be treated as similar for the same reason as lateral mirror images: When vertical mirror images occur in the natural world, they too are usually aspects of the same object. However, it may be that lateral mirror equivalence is primary and stronger because lateral mirror images are more common (e.g., as the face, two sides of a bilaterally symmetrical organism, or two views of a silhouette). The lesser similarity of vertical mirror ’E. J . Gibson and Levin (1975) proposed an experiential theory: Mirror-image confusion, they argued, might be a by-product of perceptual learning about object shape constancy since in perceptual development a child learns that an object rotated in various ways is still the same object and thus the child would be prone to equate mirror-image views of it. The fact that infants, before the acquisition of object constancy, perceptually equate mirror images tends to subvert this experiential view.

108

Marc H . Bornstein

images may reflect their derivative nature or simply the availability of additional cues (directly or indirectly related to gravity, for example) that are not available for left-right discriminations. The inequivalence of mirror images about oblique axes further supports the particular salience of the main orthogonal axes. Thus, the anisotropy associated with perceptual equivalence is also orientation specific and hierarchically ordered.

IV. Discussion and Conclusions A.

REVIEW

In this article I have tried to show that infants’ visual sensitivity to patterns and to space may be marked by two classes of perceptual anisotropy. One, perceptuul salience of the main orrhogonuls, is characterized by an orientation hierarchy of vertical-horizontal-oblique; the second, perceptual equivalence on rhe Iureraf, is characterized by the especial similarity of lateral mirror-image enantiomorphs of a pattern. A review of the literature shows that these two classes of anisotropy are well grounded in perceptual, developmental, and comparative psychology; both have been repeatedly shown to exist in adult humans, in human children, and in a variety of infrahuman species. Only the earliest ontogenetic origins of these perceptual anisotropies were unknown. The main purpose of this article has been to fill in this gap in our knowledge. Some studies of infant perception suggest that during the first year, indeed almost from birth, certain orientations are perceptually salient. Measures of detection and discrimination, long-term information processing, and preference with a variety of stimuli converge to support the conclusion that for infants, as for human children and adults and for other infrahuman species, vertical is perceptually salient, horizontal may be, and obliques are less so. Several possible explanations of this perceptual anisotropy have been offered. They range from neurophysiological, anatomical, and motor on the “nativist” side to experiential on the “nurturist” side. Other studies of infant perception suggest that in the first year of life certain rotations or transformations of patterns are perceptually equivalent. Measures of discrimination, information processing, and preference with a variety of stimuli converge to support the conclusion that for infants, as for human children and adults and for other infrahuman species, lateral (and to a lesser degree vertical) mirror images are perceptually equivalent, while oblique mirror images are not. Several possible explanations of this perceptual anisotropy have also been suggested. Two are the view that mirror-image equivalence relates to the mirrorsymmetric organization of the perceiving visual system and the view that since

the most prevalent mirror images are the two halves of phylogenetically significant stimuli-parents and progeny, predators and prey-they usually need not be discriminated, It is not possible, even in these infancy studies, to disassociate nature and nurture, and it is clear that aspects of all these explanations probably hold for both perceptual anisotropies. Presumably both these pervasive phenomena reflect in some degree the fact that orthogonal orientations in our world (horizons, gravity, etc.) are cardinal and, in turn, are reflected in a uniqueness of the brain mechanisms that process visual information oriented along the principal orthogonals. B.

SOME INTERRELATIONS BETWEEN PERCEPTUAL ANISOTROPIES

Perceptual salience and equivalence are interrelated in several different and quite independent ways. A few observations on some of these interrelations are pertinent. Significantly, both perceptual salience and perceptual equivalence start the child off with a bias toward a heightened perception and understanding of the spatial environment into which she or he is born. The perception of vertical is clearly related to upright, to gravity, and to straight ahead. Avoidance behaviors and an integrated reaction to looming show sensitivity to the vertical, and orienting so as to maximize information reception and perception usually finds the organism vertical vis-a-vis a source of information. Thus, spatial sensitivity to the main axes is psychologically relevant. In brief, sensitivity to the vertical (and horizontal) and equivalent treatment of mirror-symmetric patterns about the main orthogonals are present early in development, and both may be immediately advantageous and functionally valuable to the infant. These two perceptual anisotropies are present both in infants and in adults, but the developmental gradients associated with each appear to be oppositely sloped. While the vertical and horizontal play a significant role in infant visual perception and spatial orientation, it would seem that their influence is even more complete and pronounced in mature perception (e.g., Gwiazda er a / . , 1978; Mayer, 1977). This developmental course may reflect increasing experience with the vertical-horizontal “carpentered world” in which most of us are reared, increased attention to vertical and horizontal, or it may be that the neurophysiological biases with which humans begin life mature and develop to be more pervasive in adulthood than in infancy. In contrast, the perceptual equivalence of mirror images seems to be manifestly stronger in its infant form than in its adult form. It may be that during development, particularly in the course of learning how to read, we are forced to “unlearn” the natural constancy of mirror-image equivalence; that is, a b is not a d, nor is a p a q , was does not

I10

Marc H . Bornstein

mean the same as saw, nor does no mean the same as on (Davidson, 1934, 1935; E. J. Gibson & Levin, 1975). Thus, the courses of these two anisotropies cross during development. Adults can simultaneously discriminate differences in object orientation or disregard such differences. Watson (1966, p. 73) asked how these complementary capacities develop: “Does the infant first see form as form and then slowly learn to note object orientation or is orientation initially of prime salience, followed later by the capacity to recognize object form independent of its orientation? ” There are distinct advantages and disadvantages to the primacy of perceptual salience and to the primacy of perceptual equivalence. If perceptual salience and its corollary, orientation selectivity, were to predominate early in life, infants would discriminate among orientations of the same object, but they might not be prone to recognize an object as constant across variations in orientation or perspective. If perceptual equivalence and its corollary, object constancy, were to dominate, it would promote stability in the infant’s perceptual world but at the expense of other, orientation-sensitive capacities such as eye-hand coordination. Though not mutually exclusive, perceptual salience and equivalence reflect complementary, often competing capacities.* Finally, the two classes of perceptual anisotropy, salience and equivalence, are structurally related in that they may share or reflect the action of the same or similar neurophysiological substrates. Substantial evidence has been marshaled to show that the origins of perceptual salience and equivalence in infants may be found in evolutionally adaptive, natural biases of the vertebrate visual system. These biases may condition orientational and directional salience about the vertical (and horizontal) that influences the perception of equivalences about the same orthogonal axes. The term “mirror image” is used to refer almost exclusively to lateral or vertical mirror images. Yet the two may be viewed as a special class of mirror images, namely those produced by rotation about the main orthogonal axes. An infinite number of other “mirror images” are produced by rotations about other axes: Horizontal and vertical line segments (that are easily discriminated) can be described as mirror images about a 45” axis. Any “mirror image” needs rarely to be distinguished in nature, yet oblique ones are not especially confused in discrimination tasks. The specialty in perception of orthogonal orien“wo theoretical questions related to this issue plague the study of orientation discrimination in infancy. If infants discriminate a stimulus rotated in space, does it indicate discrimination of an orientation change of the same stimulus or, rather, perception of a new stimulus in a new orientation? If infants fail at such discriminations, does it indicate lack of perceptual acuity or, rather, simple object constancy? L. J. Harris and Allen (1974) criticized several early infant studies for failing to disambiguate sensitivity to object orientation from object constancy. Bornstein er a / . (1978, Experiment 11), however, found that the same infants discriminated one orientation change of a stimulus but treated another orientation change (presumptive of object constancy) equivalently. Though these twin capacities are present early in life, sensitivity to orientation may antedate mature object constancy.

tations must therefore influence or even determine the prevalence of mirrorimage equivalence. C. PERCEPTUAL-COGNITIVE-SOCIAL DEVELOPMENT: ANISOTROPY AND THE FACE

In a particularly persuasive argument, Sherrod (1981, p. 1 1) has proposed the general thesis that “in addition to being social behavers, caregivers, and attachment figures, people are also physical stimuli composed of parameters such as contour density and color. The most studied aspect of the person as a physical stimulus for baby is the face, and facial schemata, as has been mentioned pussim, have been centrally implicated in perceptual salience and perceptual equivalence. Vertical and symmetrical are both special characteristics of the axial structure of the human face,’ and both vertical and vertical symmetry are perceptually salient for infants. Certainly, very young babies prefer the vertically upright version of a symmetrical face to any other orientation or configuration of elements. All of us already have very strongly engraved in our minds an image of normal en face vertical social interchange between parent and child or of a parent leaning sideways over the crib rail straining to establish a strictly vertical en face dialogue with the baby lying supine in the crib. (Blaise Pascal is reputed to have observed that “our notion of symmetry is derived from the human face.”) The two halves of the face are grossly enantiomorphic consisting of the very mirrorimage sections that are perceptually equated by infants. While it may not be possible to differentiate between the contributions of nature and nurture to the early ontogeny of these perceptual anisotropies, it may be worthwhile to note that the face must play a significant central role originally in promoting perceptual anisotropies or later in providing stimulus for their expression. ”

D. SOME IMPLICATIONS FOR COGNITIVE DEVELOPMENT

The preferences, discriminations, and information-processing capacities in infants that constitute perceptual salience and equivalence forerun well-known anisotropies that characterize visual cognition in children and in adults. In this sense, infant versions of these anisotropies may serve as Anlugen of normal mature perceptuocognitive processes. This article concludes with a considera9Symmetry is carried by low spatial frequencies (Broadbent, 1977; Corballis & Roldan, 1975; Ginsburg. 1973; Julesz, 1971; Julesz & Chang, 1979) where young infants have been shown to be maximally sensitive (Atkinson, Braddick, & Braddick. 1974; Atkinson, Braddick, & Moar, 1977; Banks & Salapatek. 1976). Interestingly, Harmon (1973) has shown that blocking high spatial frequencies enhances facial recognition, and he suggests, with Ginsburg ( 1976). that the basic information in faces is concentrated at low spatial frequencies. Thus, infants would be expected to be especially sensitive to symmetry in faces.

I12

Mort, H .

Bornstoin

tion of some developmental implications of the two classes of perceptual anisotropy in infancy. The perceptual salience of vertical and horizontal in infancy not unexpectedly continues among children and adults who rely on orthogonals throughout the life-span development of spatial perception (e.g., J. J. Gibson, 1966; Howard & Templeton, 1966; Pick et a / . , 1979). In young children, for example, the salience of verticality plays a central role in cognitive judgments, as Piaget (19611 1969) showed in his analysis of the child’s infamous failures of “centering” in conservation tasks. In adults, as documented earlier, vertical and horizontal serve as the prototypical, preferred, and cognitively salient dimensions of orientation and frames of space. For example, adults assess the orientation of lines and angles most frequently with reference to vertical and horizontal ideals (e.g., Jastrow, 1893). Indeed, Wertheimer (1938) originally exemplified his concept of an “ideal type” by appealing to introspective judgments related to the role of orthogonals in the perception of orientation: Specifically, Wertheimer mused that a 5” line is almost vertical, but a vertical line is not almost 5”. Supporting this observation, Rosch (1975) found that adults will judge relationships among perceptual stimuli in ways that show that only certain orientations function as cognitive reference points; thus, lines oriented 10” off vertical or horizontal are judged to be closer to the orthogonals than vice versa in tasks that tap both spatial and linguistic competences, but principal obliques and many off-obliques are equally related. Orthogonal orientations are prototypes, ideal types, and cognitive reference points in spatial judgments. The perceptual equivalence of mirror images may also have immediate implications in the ontogeny of cognition, specifically in the development of the object concept. Contrary to the experiential position that objects in different perspectives and at different distances come to be seen as the same object (e.g., E. J. Gibson & Levin, 1975), it is possible that since infants naturally treat mirror images of an object as equivalent, their strephosymbolia, or proneness to leftright reversal, predisposes them to perceiving certain objects as constant or invariant. Objects like the face, whose lateral halves are mirror-image enantiomorphs, might in this way engage a very early mode of constancy perception. In turn, it is possible that the perceptual invariance of the two halves of the face underlies the normal child’s acquisition of person constancy, first, and later of object constancy. In this light, Bell’s (1970) finding that person constancy develops prior to the constancy of inanimate objects is less than surprising. Thus, this primitive form of object constancy might represent an incipient sign of stability amidst the perceptual flux that otherwise characterizes an infant’s visual world, and in this way mirror-image equivalence may also serve as a core mechanism to which other constancies (e.g., of shape or size) later refer. It is of interest, finally, to note that these perceptual anisotropies may play significant roles in other practical domains of cognitive development, for exam-

ple, those related to reading and to the acquisition of linguistic terms that signify orientation in space. Throughout this article, I have stressed the absence of necessity to discriminate mirror images in the natural world. But of course, in the unnatural man-made world, discrimination of mirror images is crucial: It is a prerequisite to literacy. Western orthography is plagued by mirror images, and consistent left-to-right scanning is crucial to reading. In learning to read and write, letter reversals (e.g., b for d or p for 9 ) . word reversals (e.g., on for no and WLIS for s a w ) , and the failure to progress consistently from left to right represent common errors for the normal child (Davidson, 1934, 1935; E. J. Gibson & Levin, 1975; Orton, 1937). For example, Davidson (1935) found that 77.5% of kindergarten and first-grade children “confused” lateral mirror-image letters. Letter reversal in reading therefore may reflect the normal child’s difficulty in overcoming a nativistic (and once adaptive) mode of visual processing. In this connection, Fisher, Bornstein, and Gross (1980) studied some interrelationships among intelligence, memory for left-right orientation, letter production and reversal, and performance on reading-readiness and reading tests in 50 kindergarten and in 50 first-grade beginning readers. Beginning reading skills were evaluated by the Gates-MacGinitie Readiness Skills Test in kindergarten and the Iowa Test of Basic Skills Primary Battery in first grade. Letter production was measured by asking each child to write 12 asymmetric lower-case letters from dictation: c y d r a p z e f q n b. Left-right memory was tested in a two-choice delayed matching-to-sample paradigm that included a range of nameable and nonnameable stimuli. Finally, intelligence was measured by short forms of the WPPSI in kindergarten and WISC in first grade. Older children performed better than younger on left-right memory and production. On the memory task, first graders scored significantly higher than kindergarteners. As is well known, left-right memory improves with age (e.g., Aaron & Malatesha, 1974; Cairns & Steward, 1970). On the letter-production task, left-right reversals (e.g., h for d ) accounted for most of the errors (8 1% in kindergarten and 90% in first grade), and 88% of the kindergarteners but only 36% of the first graders made reversals. Kindergarteners made four times as many left-right reversals on the average as first graders; as has been found in other studies, younger children naturally reverse letters, whereas older children reverse less often (e.g., Cairns & Steward, 1970; Davidson, 1934, 1935; E. J . Gibson & Levin, 1975; Lepez, 1969). As suggested above, left-right problems are more difficult than up-down. lntercorrelations among the measures showed that performance on the leftright memory task significantly correlated both with performance in kindergarten on the reading-readiness (.36) and word-recognition (.37) measures and with performance in first grade on the reading test (.43). Other investigators have also uncovered a positive relationship between left-right discrimination and early reading ability (e.g., Frith, 197 I , 1974; Liberman, Shankweiler, Orlando, Har-

ris, & Berti, 1971; Lyle & Goyen, 1968; Shankweiler & Liberman, 1972; Staller & Sekuler, 1975). Left-right memory also significantly correlated with letter reversal in kindergarten (.42), but not in first grade. Similarly, letter reversal and reading-readiness skills correlated in kindergarten (.65) but not with reading in first grade (.24). These findings confirm the importance of left-right visual skills in beginning reading, originally suggested by Orton (1937). Of course, other abilities such as verbal mediation may be equally or more important in the later development of reading (Vellutino, 1977). From this line of reasoning, it is possible to project a role for mirror-image equivalence in dyslexia and in literacy. Reversing letters and other left-right difficulties are reportedly common among “developmental dyslexics, ” children who have severe difficulty in learning to read for no known cause (Benton, 1975; Money, 1962; Orton, 1937; Shankweiler, 1963). From this point of view, reversal problems may reflect an especial difficulty in these children in overcoming an otherwise normal inclination to perceive mirror images as equivalent; that is, some developmental dyslexics may show particularly strong or persistent mirror-image equivalence. Just as mirror images should interfere with reading acquisition, learning to read should facilitate discrimination of mirror images. Support for this possibility comes from Rude1 and Teuber’s (1963) study of American children and Serpell’s (1971) parallel study of urban Zambian children. In both cases, the greatest improvement in mirror-image discrimination occurred at the approximate age of initial reading and writing instruction, between 5 and 7 years for Americans and between 7 and l l years for Zambians. Also by this logic, nonliterate adults or adults literate in languages devoid of orthographic mirror images should show greater mirror-image confusion than adults literate in a Western orthography (see Gross & Bornstein, 1978b, Fig. 3; Shapiro, 1970). Finally, it is conceiveable that spatial conceptualizations that characterize these anisotropies precede in parallel the child’s acquisition of linguistic conceptualizations of the same spatial anisotropies. Clark (1973), for example, has argued that children’s earliest lexical-spatial strategies appear to have their basis in natural perceptual preferences and asymmetries of their canonical conceptual representations of spatial relations. Thus, verticality, a fundamental and asymmetric direction in perceptual space, takes preference over horizontality , orthogonals over diagonals, etc., and children acquire spatial terms like “top” before “bottom,” and “top” and “bottom” before “left” and “right.” Later in adulthood, spatial terms that mark verticality are identified faster and are used more accurately than those that mark horizontality (e.g., Just & Carpenter, 1975; Maki, Grandy, & Haupe, 1979), and terms for vertical and horizontal possess nearly universal linguistic primacy (Clark & Clark, 1977), it being widely recog-

nized that the names for the main orthogonals are well advanced over the semantics of nonorthogonality (Olson & Hildyard, 1977). In short, two classes of anisotropy that characterize infant perception, salience and equivalence, persist in childhood and adulthood, appearing in perception, cognition, and language throughout the balance of the life span.

ACKNOWLEDGMENTS Preparation of this anicle was partially supported by The Spencer Foundation. Thanks go to Nelson Cowan and Celia Fisher for comments on an earlier draft and to Mary Ann Oppeman and Barbara Pallotti for aid in preparing the manuscript.

REFERENCES Aaron, P. G.. & Malatesha, R. N. Discrimination of mirror image stimuli in children. Neuropsychologia, 1974, 12, 549-55 I . Allman. J. M.. & Kaas, J. H. The dorsomedial cortical visual area: A third tier area in the occipital lobe of the owl monkey. Brtrin Resiwrch, 1975, 100, 473-487. Annis, R. C . , & Frost, B. Human visual ecology and orientation anisotropies in acuity. S r i e n w . 1973, 182, 729-731. Appelle. S. Perception and discrimination as a function of stimulus orientation: The “oblique effect” in man and animals. Psychokigicul Bulletin, 1972, 78, 266-278. Arnheim. R. A n and visual pcrrepfion. Berkeley: Univ. of California Press, 1974. Atkinson, J., Braddick, 0.. & Braddick, F. Acuity and contrast sensitivity of infant vision. Narurr (Loridon). 1974, 247, 403-404. Atkinson, J., Braddick, O., & Moar, K. Development of contrast sensitivity over the first 3 months of life in the human infant. Vision Reseurrh, 1977. 17, 1037-1044. Atkinson, J . , & French, J. Astigmatism and Orientation preference in human infants. Vision Reseurrli, 1979, 19, 1315-1317. Attneave, F., & Curlee, T . E. Cartesian organization in the immediate reproduction of spatial patterns. Bulletin of the Psychonomic Socier.v, 1977, 10, 469-470. Attneave, F., & Olson, R. K. Discriminability of stimuli varying in physical and retinal orientation. JournuI of’E.rperirnentu1 Psyrholog)~. 1967. 74, 149- 157. Banks, M. S., & Salapatek, P. Contrast sensitivity function of the infant visual system. Visioti Re.sc~urt,h,1976, 16, 867-869. Bartlett, F. C . Remembering: A study in experitnenrul arid social psychology. London and New York: Cambridge Univ. Press, 1932. Bauer, J. A,. Owens, D. A.. Thomas, J . , & Held, R. Monkeys show an oblique effect. Perreption, 1979. 8. 247-253. Bell, S. M. The development of the concept of object as related to infant-mother attachment. Child D n d o p m e n t , 1970, 41, 291-31 1 . Benton, A. L. Developmental dyslexia: Neurological aspects. Advunres in Neurology, 1975,7,1-47. Berkley, M. A,, Kitterle, F.. & Watkins, D. W . Grating visibility as a function of orientation and retinal eccentricity. Vision Researrh, 1975, IS, 239-244.

Bornstein, M . H. Perceptual generalization: A note on the peak shift. Psychologictil Bulletin. 1974, 81, 804-808. Bornstein, M . H. Qualities of color vision in infancy. Journul I$ Experitnentul Child Psychology, 1975, I Y , 401-419. Bornstein. M. H. Visual behavior of the young human infant: Relationships between chromatic and spatial perception and the activity of underlying brain mechanisms. Joidrnd qf Ex/ierimentci/ Child P s y ~ l i o l ~ ~ 1978, g y . 26, 174-192. Bornstein, M. H. Vertictil symmetry is prototypitul symme~ry.Unpublished manuscript, 1980. Bornstein, M. H . Two kinds of perceptual organization near the beginning of life. Minnrsotu Symposia on Child P sycho/iigy , 198 1 , 14, 39-9 I . Bornstein, M. H., Ferdinandsen, K., & Gross, C. G . Perception of symmetry in infancy. Developmenrtil P s y ~ ~ h o l o g1981 ~ , , 17, 82-86. Bornstein, M. H . , Gross, C . G . , & Wolf, J. 2. Perceptual similarity of mirror images in infancy. Cognition. 1978, 6, 89-1 16. Bornstein, M . H . , Kessen, W., & Weiskopf, S . Color vision and hue categorization in young human infants. Jiiurnul of Experimrnfd Psychokigy: Humun Perwption und Perfi,rrnunc.e, 1976. 2, 115-129. Boswell, S . L. Young children's processing of asymmetrical and symmetrical patterns. Journtrl of EXperimiWCi/ Child PS,Vch(J/Ogy, 1976, 22, 309-3 18. Bouma, H., & Andriessen, J . J. Perceived orientation of isolated line segments. Vision Resecirch, 1968,8, 493-507. Bradshaw, J . , Bradley. D., & Patterson, K . The perception and identification of mirror-reversed 1976, 28, 22 I -246. patterns. Quurferly Journal of Experimentnl P.~Vc&~liigj~, Braine, L. G. Early stages in the perception of orientation. In M. Bortner (Ed.), Cognitive growth unddevdopmrnt: Essci~ysin memory of Herbert C. Birch. New York: BrunndMazel. 1978. (a) Braine, L. G.A new slant on orientation perception. Americun P s y h l l o g i s t , 1978, 33, 10-22. (b) Braine, L. G., Lerner, C . , & Relyea, L. Levels in the identifying of orientation by preschool children. Journul of'Experimentci1Child Psychology. 19x0, 30, 111- 185. Bresson, F., Maury, L., Pieraut-Le Bonniec, G . , & de Schonen, S. Organization and lateralization of reaching in infants: An instance of asymmetric functions in hands collaboration. N m r o psychologifi. 1977, 15, 3 I 1-320. Broadbent, D. E. The hidden preattentive processes. Americ,un Psycho/ogi.u, 1977, 32, 109- I 18. Bronson, G . The postnatal growth of visual capacity. Child Development, 1974, 45, 873-890. Brooks, E.,& Jung, R. Neuronal physiology of the visual cortex. In R. Jung (Ed.), Hundhook sensory physiology (Vol. VI1 3B). Berlin and New York: Springer-Verlag. 1973. Bryant P. E. Discrimination of mirror-images by young children. Jorrmcil of CiJmpfJrUtllW LInd Physiokogicul Psychology, 1973, 82, 41 5-425. Butler, J. Visual discrimination of shape by humans. Qrrurtcrl~~ J O U ~ I IofExp".itnuniul U~ PsV&JIO,L~. 1964, 16, 212-216. Cairns, N. U., & Steward, M. S. Young children's orientation of letters symmetry and stimulus alignment. Chikd Devrkolment, 1970, 41, 993-1002. Camisa, J . M., Blake. R.. & Lema, S. The effects of temporal modulation on the oblique effect. Perception, 1977, 6, 165-171. Campbell, F. W . , Kulikowski, J . J . , & Levinson. J. Theeffect oforientation on the visual resolution of gratings. Joitrncil of Physiology (London), 1966, 187, 427-436. Caplan, P. J., & Kinsbourne, M . Baby drops the rattle: Asymmetry of duration of grasp by infants. Child Development. 1976, 47, 532-534. Caron, A. J.. Caron, R. F., Caldwell, R. A., & Weiss, S. J . Infant perception of the structural properties of the face. Developmentcil P s y h o l o g y , 1973, 9, 385-399. Caron, A. J . , Caron, R. F., Minichiello, M. D., Weiss, S . J., & Friedman, S. L. Constraints on the

use of the familiarization-novelty method in the assessment of infant discrimination. Child Dewlopinenr, 1977, 48, 747-762.

Clark, H. H. Space, time, semantics, and the child. In T. E. Moore (Ed.), Cognirive developtnent l i d the ucquisitiun of lunguugr. New York: Academic Press, 1973. Clark, H. H., & Clark, E. V. Psychology und liinguuge. New York: Harcourt, 1977. Cohen, L. B . , DeLoache. J. S . , & Strauss, M. S. Infant visual perception. In J. Osofsky (Ed.), Handbook of infamy. New York: Wiley, 1979. Corballis, M. C., & Beale, I. L. The psychology qflt$ cind right. Hillsdale, N.J.: Erlbaum, 1976. Corballis, M. C., Miller, A., & Morgan, M. J. The role of left-right orientation in interhemispheric matching of visual information. Perception & Psychophysics. 197 1, 10, 385-388. Corballis, M. C., & Roldan, C. E. Detection of symmetry as a function of angular orientation. Journal ofE.xperimentul P.sychology: Human Perrepriot1 and Perforn~rince, 1975, 1, 221 -230. Corballis, M . C., & Zalik, M . C. Why do children confuse mirror-image obliques? Journul o j E.~prrirnenta/Child Psycholog.v, 1977. 24, 5 16-523. Cornell, E. H. Infants’ discrimination of photographs of faces following redundant presentations. Joiirnul of Experimental Child Psvchology, 1974, IS, 98- 106. Corwin, T. R . , Moskowitz-Cook, A., & Green, M. A. The oblique effect in a vernier acuity situation. Pwwptioti & P.svi,huphysii,s, 1977, 21, 445-449. Cosgrove, M. P., Schmidt, M. J . , Fulgham, D. D., & Brown, D. R. Stabilized images: Dependent variable specificity of pattern-specific effects with prolonged viewing. Perception & Pryc-hophysics. 1972, 11, 398-402. Craig, E. A . , & Lichtenstein. M. Visibility-invisibility cycles as a function of stimulusorientation. Awiericun Journul of Psycholojiy. 1953, 66, 554-563. Critchley. M . The puric,tul lobes. London: Arnold, 1953. Davidson, H. P. A study of reversals in young children. Journal of Genetic Psychology, 1934, 45, 458-465. Davidson, H. P. A study of the confusing letters b. d. p and q. JournulcfGenetii~Psychology, 1935, 49, 458-468. Davis, A. E.. & Wada, J. A. Hemispheric asymmetries in human infants: Spectral analysis of flash and click evoked potentials. Bruin und Lungircige, 1977, 4, 23-31. Deregowski, J . B. Symmetry, Gestalt and information theory. Quurrerl-v Joitrnul of Experiinmrul Psycholo,yy. 1971, 23, 381-385. Emsley, H. H. Irregular astigmatism of the eye: Effect of correcting lenses. Trunsui~tions.f‘th<# Opticul Society (London), 1925, 27, 28-41. Essock. E. A. The oblique effect of stimulus identification considered with respect to two classes of oblique effects. Percepprion. 1980, 9, 37-46. Essock, E. A., & Siqueland, E. R. Discrimination of orientation by human infants. Perception, 198 I , 10, in press. Fagan, J. F. Infants ’ recognition memory for faces. Journtil of Esperirnentril Child Psychology, 1972, 14, 453-476. Fagan, J . F. Infants’ recognition of invariant features of faces. Child Devc,loprnent, 1976, 47, 627-638. Fagan, 1. F. The origins of facial pattern recognition. In M. H. Bornstein & W. Kessen (Eds.), Ps~chologic~til clevrlopinlvit j i o r n i t $ h y Hillsdale. N.J.: Erlbaum. 1979. Fagan, J. F., & Shepherd. P. A. Infants’ perception of face orientation. Injunt Behtivior ~ n d Dawlopment. 1979, 2, 227-234. Fantz, R . L. Visual experience in infants: Decreased attention to familiar patterns relative to novel ones. S c i t w e , 1964, 146, 668-670. Human Perception and Farrell, W. S. Coding left and right. Journal of Eqwrin7rntaI Psychology.?); Per/iJrmuniv. 1979, 5. 42-5 1 .

I I8

Mtrrc H . Bornstein

Fellows, B. 1. The discrimination prmess und development. Oxford: Pergamon, 1968. Fisher, C. B. Children’s memory for orientation in the absence of external cues. Child Development, 1979, 50, 1088-1092. Fisher, C. 8. Children’s memory for line orientation: A reexamination of the “oblique effect.” Journal of Experimental Child Psychology, 1980, 29, 446-459. Fisher, C. B. The role of stimulus alignment in children’s memory for line orientation. Journal of Experimenrul Child Psychology, 1981, 30, in press. Fisher, C. B., Bornstein, M. H., & Gross, C . G . LeJ-righr memory in beginning readers. Unpublished manuscript, 1980. Fisher, C. B., & Braine, L. G. Children’s left-right concepts: Generalization across figure and location. Child Development, 1981, 52, 451-456. Fisher, C. B., Ferdinandsen, K.,& Bornstein, M.H. The role of symmetry in infant form discrimination. Child Development, 1981, 52, 457-462. Fitts, P. M., & Simon, C. W. Some relations between stimulus patterns and performance in a continuous dual-pursuit task. Journul of Experimentul Psychology, 1952, 43, 428-436. Fitts, P. M.,Weinstein, M., Rappaport, M.,Anderson, N . , & Leonard, J . A. Stimulus correlates of visual pattern recognition: A probability approach. Journul ofE.rperimenrul Psychology, 1956, 51, 1 - 1 1 ,

Fox, J. The use of structural diagnostics in recognition. Journal ofExperimentul P.~ych/~logyt Humun Perceprion und Perfi,rmunce, 1975, 104, 57-67. Freeman, R. D., & Thibos, L. N. Electrophysiological evidence that abnormal early visual experience can modify the human brain. Science, 1973, 180, 876-878. Frkgnac, Y., & Imbert, M. Early development of visual cortical cells in normal and dark-reared kittens: Relationship between orientation selectivity and ocular dominance. Journal offhysiology (London). 1978, 278, 27-44. Frith, U. Why do children reverse letters’? British Journal offsychology, 1971, 62, 459-4158, Frith, U. Internal schemata for letters in good and bad readers. British Journul offsychobgy, 1974, 65, 233-241. Frost, B . J . , & Kaminer, J. J . The orientation anisotropy and orientation constancy: A visual evoked potential study. Perception, 1975, 4, 51 -58. Gardner, J . , Lewkowicz, D., & Turkewitz, G. Development of postural asymmetry in premature human infants. Dcvelopm~wtulPsychobiology, 1977, 10, 47 1-480. Garner, W . R. Good patterns have few alternatives. Americun Scientist. 1970, 58, 34-42. Geldard, F. A. Vision, audition, and beyond. Contributions to Sensory Physiology. 1970, 4, 1 - 17. Gibson, E . J . Principles of perceptual learning and development. New York: Appleton, 1969. Gibson, E. J . , & Levin, H. The psychology ofreading. Cambridge, Mass.: MIT Press, 1975. Gibson, E. J . , Owsley, C. J . , & Johnston, J. Perception of invariants by five-month-old infants: Differentiation of two types of motion. Developmental Psychology, 1978, 14, 407-415. Gibson, J . J . Vertical and horizontal orientation in visual perception. Psychologicol Bulletin, 1934, 31, 739-740. Gibson, J . J. Adaptation, after-effect, and contrast in the perception of tilted lines. 11. Simultaneous contrast and the areal restriction of the after-effect. Journal ofE.rprrimenral P ~ y C h ~ l o g y1937, , 20, 553-569. Gibson, J. J. The senses considered us prrcrpfual systems. Boston: Houghton, 1966. Gibson, J. J . , & Mowrer, 0 . H. Determinants of the perceived vertical and horizontal. Psychological Review, 1938, 45, 300-323. Ginsburg. A . P. Pattern recognition techniques suggested from psychological correlates of a model of the human visual system. lEEE Proceedings NAECON, Dayton, Ohio. 1973, pp. 309-3 16. Ginsburg, A. P. The perception of visual form: A two dimensional filter analysis. In V. D. Glezer (Ed.), InJwmtrtion processing in the visuul system. Leningrad: I. P. Pavlov Institute of Physiology, 1976.

Glanville, B. B., Best, C. T . , & Levinson. R. A cardiac measure of cerebral asymmetries in infant auditory perception. Der~e/oppmentcdPsycholop, 1977, 13, 54-59. Goldmeier, E. Uber ahnlichkeit bei gesehen figure. P s j c h o / n g i . d e Forschiing, 1937.21, 147-208. Goldmeier, E. Similarity in visually perceived forms. Psyc~hologicdfssues, 1972.8, Monograph 29. Goldstein, A. J . , Harmon, L. D., & Lesk, A. B. Identification of human faces. Procc1ding.s o / t h r IEEE, 1971, 59, 748-760. Gross, C. G . , & Bornstein, M. H. Inf2rnr.v' disi~rirnintifiorioj'nrirror iiniigc,s. Paper presented at the meeting of the International Neuropsychological Symposium, Frascati, Italy, June 1978. (a) Gross, C. G., & Bornstein, M. H. Left and right in science and art. Leontrrdo, 1978, 11.29-38. (b) Gwiazda, J . , Brill, S., Mohindra, I . , & Held, R . Infdnt visual acuity and its meridional variation. Vision Research, 1978, 18, 1557-1564. Haith, M . M . Visual competence in early infancy. In R . Held. H. Leibowitz, & H. L. Teuber (Eds.). Htrnrihook of sensory physiology (Vol. 8). Berlin and New York: Springer-Verlag, 1978. Haith, M. M . Rules thur bubies look l,v. Hillsdale, N.J.: Erlbaum, 1980. Hamilton, C. R . , & Tieman, S. B. Interocular transfer of mirror image discriminations by chiasmsectioned monkeys. Bruiii Reseurch, 1973, 64, 241 -255. Harmon, L. D. The recognition of faces. Scienr$c Anrericcw, 1973, 229, 70-82. Harris, L. J . , & Allen, T. W. Role of object constancy in the perception of object orientation. Hrrrnuri Drrvloptnenr, 1974, 17, 187-200. Harris. P. L.. Le Tendre, J . B., & Bishop, A. The young child's discrimination of obliques. Pvrc-iyxion, 1974, 3, 261 -265. Higgins, G. C., & Stultz. K. Variation of visual acuity with various test-object orientations and viewing conditions. Joio.nul of the Optit,u/ Soc,irty of' Aniericu. 1950. 40, 135- 137. Hogben, J. H . , Julesz, B., & Ross, J. Short-term memory for symmetry. Vision Rescuri,h, 1976, 16, 861 -866. Horowitz, F. D. Visual attention, auditory stimulation, and language discrimination in young infants. Monogruphs cfthe S o c i e p j i ~ Research r irr Child D i w l o p m n t , 1974, 39 (415, Serial No. I 12). Howard, I . P., & Templeton, W .B. Himion sputicil orirnrtrtion. New York: Wiley, 1966. Hubel, D. H., & Wiesel, T . N. Receptive fields, binocular interaction, and functional architecture in the cat's visual cortex. Journul i$Phv.sfo/o~.v(London), 1962, 160, 106-154. Hubel. D. H., & Wiesel, T. N. Receptive fields and functional architecture of monkey striate cortex. Joirrlicr/ of Phgsiology (London), 1968. 195, 2 15-243. Huttenlocher, J . Children's ability to order and orient objects. Child Deceloptnent, 1967. 38. 1 1691176. (a) Huttenlocher, J. Discrimination of figure orientation: Effects of relative position. Juirrncrl /$Cornpiirurir~rtitid Physiologicul Psyc.hology. 1967. 63, 359-36 I . (b) Ivinskis. A,, 8r Finlay, D. C. Curdiuc responses in j i ~ r~?oii?h r old infiints to slirnuli nrovfirg a? three t/i&renf velocities. Paper presented at the International Conference on Infant Studies, New Haven, April 1980. Jastrow. 1. On the judgment of angles and positions of lines. Arnivicm Jc~urtiulof Psychology, 1893, 5 , 214-248. Johnson. R. M., & Uhlarik, J. Tachistoscopic construction of four orientations of a square. Bitlli4n Of //I<' PS~1'honf)IJlil' Society, 1917, 9, 333-336. Jonckheere. A . R. A distribution-free k-sample test against ordercd alternatives. Biometrikrr. 1954, 41, 133-145. Julesz, B. Forincfntions of'c~rlopeanperception, Chicago: Univ. of Chicago Press, 1971. Julesz, B., & Chang. J.-J. Symmetry perception and spatial-frequency channels. Perception, 1979, 8, 711-718. Just. M . A,. & Carpenter. P. A. The semantics of locative information in pictures and mental images. British Jortrneil o f ' P . ~ y i h l o g y .1975. 66, 427-441. Karmel. B. 2.. & Maisel, E. B. A neuronal activity model for infant visual attention. In L. B. Cohen

Murc, H . Bornstein

120

& P. Salapatek (Fds.), Infant perception: From sensation to cognition. New York: Academic

Press, 1975. Keene, G . C. The effect of response codes on the accuracy of making absolute judgments of lineal inclinations. Journal of General Psychology, 1963, 69, 37-50. Kessen, W., Salapatek, P.,& Haith, M. The visual response of the human newborn to linear contour. Journal of E.rperimental Child Psychology, 1972, 13, 9-20. Lashley, K . S. The mechanism of vision, XV. Preliminary studies of the rat’s capacity for detailed vision. Journul of General Psychokogy. 1938, 18, 123-193. Lechelt, E. G., Eliuk, J . , & Tanne, G. Perceptual orientational asymmetries: A comparison of visual and haptic space. Perception & Psychophysics. 1976, 20, 463-469. Leehey, S . C . , Moskowitz-Cook, A,, Brill, S.,& Held, R. Orientational anisotropy in infant vision. Science, 1975, 190, 900-901. Lehman, R. A. W . , & Spencer, D. D. Mirror-image shape discrimination: Interocular reversal of response in the optic chiasm sectioned monkey. Bruin Research, 1973, 52, 233-241. Lennie, P. Head orientation and meridional variations in acuity. Vision Research. 1974, 14, 107111.

Lepez, R. Regarding the question of reversed letters. Revue de Psychdogie AppliyuPr, 1969, 19, 195-21 5 . Leventhal, A. G., & Hirsch, H. V. B. Cortical effect of early selective exposure to diagonal lines. Science, 1975, 190, 902-904. Leventhal, A. G., & Hirsch, H. V. B. Effects of early experience upon orientation sensitivity and binocularity of neurons in visual cortex of cats. Proceedings of the Nationtil Acudemy o j Sciences ( U . S . A . ) , 1977, 74, 1272-1276. Liberman, 1. Y., Shankweiler, D.,Orlando, C., Hams, D. S., & Berti , F. B. Letter confusions and reversals of sequence in the beginning reader: Implications for Orton’s theory of developmental dyslexia. Cortex, 1971, 7, 127-142. Ling, B. C. Form discrimination as a learning cue in infants. Comparative Psychology Monographs. 1941, 17, Whole No. 86. Lyle, J. G., & Goyen, J . Visual recognition, developmental lag, and strephosymbolia in reading retardation. Journd of Abnormal Psychology, 1968, 73, 25-29. Mach, E. [The unalysis of sen.sations] (C. M. Williams, trans.). New York: Dover, 1959. (Originally published, 1886.) Mach, E. [Space and geometry in the light ofphysiologicd, psychological and physic.ul inyiiiry J ( T . J . McCormack, trans.). LaSalle, 111.: Open Court, 1906. Maffei, L., & Campbell, F. W. Neurophysiological localization of the vertical and horizontal visual coordinates in man. Science. 1970, 167, 386-387. Maki, R. H., Grandy, C. A., & Haupe, G . Why is telling right from left more difficult than telling above from below? Journal of E.rperimental Psychology: Humtrn Perception and PerfiJrmtmce, 1979, 5 , 52-67. Mansfield, R . J . W. Neural basis of orientation perception in primate vision. Science, 1974, 186, 1133-1135. Mansfield, R. J. W., & Ronner, S. F. Orientation anisotropy in monkey visual cortex. Brain Rusrarc.h, 1978, 149, 229-234. Marg, E., Adams, J. E., & Rutkin, B. Receptive fields of cells in the human visual cortex. Experientiu. 1968, 24, 348-350. Maurer, D.,& Martello, M. The discrimination of orientation by young infants. Vision Research, 1980, 20, 201 -204. May, J. G . , Cullen, J. K., Jr., Moskowitz-Cook, A , , & Siegfried, J. B. Effects of meridional variation on steady-state visual evoked potentials. Vision Resettrrh, 1979, 19, 1395-1401. Mayer, M. J . Development of anisotropy in late childhood. Vision Research. 1977, 17, 703-710.

McCall, R . B., Kennedy, C. B.. & Appelbauni, M. I. Magnitude of discrepancy and the distribution of attention in infants. Child Derdopment, 1977, 48, 772-785. McCandless. B. R., & Geis, M .F. Current trends in developmental psychology. Advonces in Child Dewlopment cmd Behuvior. 1975, 10, 1-8. McFarland, J. H. ‘‘Parts” of perceived visual forms: New evidence. Perception & psycho physic.^. 1968, 3, 118-120. McGurk, H. Infant discrimination of orientation. Jorirnul of E.rperimental Child Psychology. 1972, 14, 151-164.

McGurk, H. Visual perception in young infants. In B. Foss (Ed.), New perspectives in child dctdopineirt. Baltimore: Penguin Books, 1974. McKenzie, B., & Day, R. H. Orientation discrimination in infants: A comparison of visual fixation and operant training methods. Joitrncil c~fE.rprrimentul Child Psyhology, 1971, 1 I, 366-375. Moffett, A. Stimulus complexity as a determinant of visual attention in infants. Jottrntrl ofE.rperinrentril Child P s y c h o l ~ g v ~ 1969, 8, 173-179. Mohindra. I . . Held, R., Gwiazda, J., & Brill, S. Astigmatism in infants. Science, 1978, 202, 329-33 I . Molfese, D. L., Freeman, R. B., Jr., & Palermo, D. S. The ontogeny of brain lateralization for speech and nonspeech stimuli. Bruin and Langriage, 1975, 2 , 356-368. Money, J. (Ed.). Reuding disability. Baltimore: Johns Hopkins Press, 1962. Nelson, J. I., Kato, H., & Bishop, P. 0. Discrimination of orientation and position disparities by binocularly activated neurons in cat striate cortex. Journril of Neurophysiology, 1977, 40, 260-283. Noble. 1. Paradoxical interocular transfer of mirror-image discriminations in the optic chiasrn sectioned monkey. Bruin Resetrrch, 1968. 10, 127-151. Olson. D. R. Cognitive development: The child‘s ucqukition cfdiagonality. New York: Academic Press, 1970. Olson, D. R . , & Hildyard, A. On the mental representation of oblique orientation. Canadiun Journal of Psvi~hology. 1977, 3 I, 3- 13. Onon, S. T. Reczding, w i r i n g anti speech problems in children. New York: Norton, 1937. Over, R., & Over, J . Detection and recognition of mirror-image obliques by young children. Journal i f Comnpt7rurive und P hysio/ogicff/ P s v r h o / o ~ V ,1967, 64, 467-470. Palmer, S. E.. & Hemenway, K . Orientation and symmetry: Effects of multiple, rotational, and near symmetries. Journul of E.rperinrrnttil Psychology: Hitmcin Perception iind Performunce. 1978, 4, 691 -702. Piaget, J . !The niechirnismr of perception] ( C . N. Seagrim, trans.). London: Routledge & Kegan Paul, 1969. (Originally published, 1961 .) Pick, H. L., Yonas, A,, & Rieser, J. Spatial reference systems in perceptual development. In M. H. Bomstein & W. Kessen (Eds.). Psyc~hologicol dewlopnrent ,from iqfrrncy. Hillsdale: N.J.: Erlbaum, 1979. Pomerantz, J . R., Sager, L. C . , & Stoever, R . J. Perception of wholes and of their component parts: Some configural superiority effects. Jorrrnrrl of Experinwnrnl Psychology: Hionan Perception und Performutice. 1977. 3 , 422-435. Regan, D. Evoked potenrials in psycho log)^, sensory physiology. and clinical fnedicine . London: Chapman and Hall, 1972. Regan, D., & Beverley, K. I. Looming detectors in the human visual pathway. Vision Reseurch, 1978. 18, 415-421. Regan, D., & Beverley, K. 1. Visually guided locomotion: Psychophysical evidence for a neural mechanism sensitive to flow patterns. Science. 1979, 205, 31 1-313. Rock, 1. Orientution undform. New York: Academic Press, 1973.

122

Mtrrc H . Rorrrsfein

Rock, I . , & Leaman, R. An experimental analysis of visual symmetry. Acta Psychologica, 1963, 21, 171 -183. Rosch, E. Cognitive reference points. Cognirive Psycho/ogV, 1975, 7, 532-547. Rose, D., & Blakemore, C. An analysis of orientation selectivity in the cat’s visual cortex. Experimental Brain Reseurch, 1974, 20, 1-17. Rudel, R. G . , & Teuber, H. L. Discrimination of direction of line in children. Journtrl ofCompurafive and Phy.sio/ogicu/ P,syc,hology. 1963, 56, 892-898. Salapatek, P. Visual scanning of geometric figures by the human newborn. Journal r,f’Compurafive und Physiological Psychrilogy. 1968, 66, 247-258. Salapatek, P. Pattern perception in early infancy. In L. B . Cohen & P. Salapatek (Eds.), Infirnt perception: From sensufion to cognition. New York: Academic Press, 1975. Salapatek, P., & Kessen, W. Visual scanning of triangles by the human newborn. Journal of’ Exprrimmtul Child Psychofogy, 1966. 3, 155 - 167. Sekuler, R. W . , & Houlihan, K. Discrimination of mirror-images: Choice time analysis of human adult performance. Quurterly ~ o l i r n f dof Expi~rirnentalPsychology. 1968, 20, 204-207. Sekuler, R . W., & Pierce, S. Perception of stimulus direction: Hemispheric homology and laterality. Ameritnn Journal of Psychology. 1973, 86, 679-695. Sekuler, R . W . , & Rosenblith. J . F. Discrimination of direction of line and the effect of stimulus alignment. Psychonomic Science, 1964. I , 143-144. Serpell, R. Discrimination of orientation by Zambian children. Journal of Compururive crrul Physiological Psychology. 197 I. 75, 3 12-3 16. Shankweiler, D. P. A study of developmental dyslexia. Neuropsychologiu, 1963, 1, 267-286. Shankweiler, D . , & Liberman, I. Y. Misreading: A search for causes. In J. F. Kavanagh & I. G . Mattingly (Eds.), Languuge by ear und by eyr. Cambridge, Mass.: MlT Press, 1972. Shapiro, M. On some problems in the semiotics of visual art: Field and vehicle in image-signs. In A. J. Greimas & R. Jakobson (Eds.), Sign, lmgrcuge, culture. The Hague: Mouton, 1970. Shepard. R. N., & Metzler, J . Mental rotation of three-dimensional objects. Science, 1971, 171, 701 -703. Sherrod, L. R. Issues in cognitive-perceptual development: The special case of social stimuli. In M. E. Lamb & L. R. Sherrod (Eds.), fnfanr social cognition. Hillsdale, N.J.: Erlbaum, 1981. Sidman, M., & Kirk, B. Letter reversals in naming, writing and matching to sample. Child Drvelopment, 1974, 45, 616-625. Slater, A . , & Sykes, M. Newborn infants’ visual responses to square wave gratings. Child Development. 1977, 48, 545-554. Staller, J . , & Sekuler, R. Children read normal and reversed letters: A simple test of reading skill. Quurterly Journcrl of Experimentuf Psychology, 1975, 27, 539-550. Stein, N. L., & Mandler, J. M. Children’s recognition of reversals of geometric figures. Child Developincwt. 1974, 45, 604-615. Storandt, M. Recognition across visual fields with mirror-image stimuli. Perceptuul and Motor 1974, 39, 762. Sutherland, N. S . The methods and findings of experiments on the visual discrimination of shape by animals. E.rperimentu/ P.Yychm/ogictrl Sociery Monogruphs, I96 1, Whole No. 1 . Szilagyi, P. G., & Baird, J . C. A quantitative approach to the study of visual symmetry. Perception & PsychtJphy.Vics. 1977, 22, 287-292. Taylor, M. M. Visual discrimination and orientation. Journul sf the Opticcil Society of Americu. 1963, 53, 763-765. Tee, K. S . , & Riesen, A. H. Visual right-left confusions in animals and man. Advonces i n Psvchobiology, 1974, 2, 241-265. Timney, B. N . , & Muir, D. W.Orientation anisotropy: incidence and magnitude in Caucasian and Chinese subjects. Science, 1976, 193, 699-701.

Turkcwitz, G., Gordon, E. W.. & Birch. H. G . Headturning in the human neonate: Spontaneous patterns. Jourtiul of’Gcnrtic Psychology, 1965. 107, 143- 158. Vellutino, F. R. Alternative conceptualizations of dyslexia: Evidence in support of a verbal-deficit hypothesis. H u r w r d Educ,trtiotiul Review, 1977, 47, 334-354. Wada, J., Clarke, R., & Hamm, A. Cerebral hemispheric asymmetry in humans. Archives ofhleurulogy, 1975. 32, 239-246. Watson, J . S. Perception of object orientation in infants. Merrill-Pulmrr Quurterly, 1966, 12, 73-94. Weiner, K., & Kagan. J . Infanta’ reaction to changes in orientation of figure and frame. Perceplion, 1976, 5, 25-28. Wertheimer, M. Numbers and numerical concepts in primitive peoples. In W. D. Ellis (Ed.), A sourcc hook .f Gesrult psychology. New York: Harcourt, 1938. Williamson, A . M., & McKenzie, B. Children’s discrimination of oblique lines. Journcrl c$E.rprrimrntal Child Psychology, 1979, 27, 533-543. Wohlwill. J . F., & Weiner, M. Discrimination of form orientation in young children. Child Drwlopmenr. 1964, 35, I 1 13-1 125. Wolff. P. Mirror-image confusability in adults. Jortrnrrl of E.rprrirnentd Psychology. 197 1 , 91, 268-272. Yin. R. K . Looking at upside-down faces. Joctrnctlc!f&.q,erinic.nrcrl Psycholugy. 1969,81, 141-145. Yoshida, S.. Iwahara, S., & Nagamura. N. The effect of stimulus orientation on the visual evoked cirid Clinicctl Neurophysiology. 1975. 39, potential in human subjects. Electroencephalr)~r~phy 53-57. Zeki. S. M., & Sandeman, D. R. Combined anatomical and electrophysiological studies on the boundary between the second and third visual areas of rhesus monkey cortex. Proceedings of the Royal Society of .bndo:i, Srrirs B. 1976, 194, 555-562.

This Page Intentionally Left Blank

CONCEPT DEVELOPMENT'

Martha J . Furah aad Stephen M . Kosslyn" DEPARTMkNT OF PSYCHOLOGY A N D SOCIAL RELATIONS HARVARD UNIVERSITY

CAMBRIDGE, MASSACHUSETTS

1. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11. INFORMATION-PROCESSING THEORIES

125

. , .. . , . . .... , . . .. . , . . . , .. , . . . . .

A. THE COMPUTER ANALOGY IN PSYCHOLOGY . . . . . . . . . . . . . . . . . . . . . . B. MENTAL REPRESENTATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

129

. . .

. . .

.

I30 130

.

. . .

.

134 143

111. THE CONTENTS OF CONCEPT REPRESENTATIONS . . . . . . . . . . . . . . . . . . . A. RULE-BASED REPRESENTATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . 8. NON-RULE-BASED SUMMARY REPRESENTATIONS . . . . . . . . . . . . . . . . . C. EXEMPLAR-BASED REPRESENTATIONS . . . . . . . . . . . . . . . . . . . . . . . . . D. THE PROBLEM OF MULTIPLE INTERPRETATIONS OF DATA . . . . . . .

.

127 127

152

IV. THE FORMAT OF CONCEPT REPRESENTATIONS.. . . . . . . . . . . . . . . . . . . . . . .

153

A. IMAGES AND DESCRIPTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B, THE REPRESENTATIONAL-DEVELOPMENT HYPOTHESIS . . . . . . . . . . . .

155

V. THE ORGANIZATION OF CONCEPT REPRESENTATIONS . . . . . . . . . . . . . . . . .

160

VI. CONCLUSIONS.. . . . . . . . . . . . . . . . . . . . . . . . . , . . . . . . . . . . . . . . . . . . . . . . . . . . . .

161

REFERENCES ........................................................

164

158

I. Introduction With the return of mentalistic psychology the study of concepts has again become not only respectable, but of central concern. The past 15 years have seen increasing sophistication in the study of concept representation in adults, and the theories and methods developed in the course of studying the adult's 'Preparation of this article was supported by NSF grant BNS 79-12418. The authors wish to acknowledge Jerome Kagan 's helpful comments and criticisms of this manuscript. *Present address: Program in Linguistics and Cognitive Science, Brandeis University, Waltham, Massachusetts 02554. I25 ADVANCES IN CHILD DEVELOPMENT AND BEHAVIOR, VOL. 16

Copyright 0 1982 by Academic Presr. Inc. All rights of reproduction in any fvrm reserved. ISBN 0-12-009716-8

I26

Martha J . Furcih artti Stephen M . Kosslvn

concepts can now be put to good use in studying how concepts develop over age. In this article we shall review the recent literature on concept representation in adults, and consider some of the implications of different theories of adult representation for theories of concept development in children. In addition, we shall evaluate the strength of the evidence for different theories of concept representation and consider what further research should be carried out to distinguish among the theories. The word “concept” can mean many things both in common English usage and in theories of psychology. One likely confusion within developmental psychology is between the investigation of the “concepts” of conservation, class inclusion, and transitivity, for example, and the study of “concepts” as the representations of categories by which the child groups and divides the objects and events of the world in an adaptive, psychologically efficient way. “Concepts” in the first sense are complex logical and mathematical competences, whose development in the child’s mind follows a particular course that seems quite different from the process of concept formation in the second sense. We shall exclude these logical and mathematical competences from the domain of this article, and concentrate instead on what might, in comparison, seem a mundane and unmysterious accomplishment of the developing child: the formation of concepts of categories. An oft-cited example of this process of concept formation is the learning of the concept “dog.” A child sees dogs, cats, teddy bears, and so on, hears them named, and eventually comes to distinguish between dogs and other creatures, including those never before seen by the child. This last point is critical, for without this ability to generalize correctly, we have an example not of concept formation, but rather of simple memorization. Having a concept of a category allows one to generalize from some particular dogs to all possible dogs. We think that it is hard to appreciate the magnitude of this accomplishment with so familiar a concept as “dog.” As adults, we cannot imagine the child’s view of dogs and other creatures prior to having the concept “dog” (which is the view of things from which he or she must learn the concept) because we cannot imagine a dog without already knowing that it is a dog. To get a better sense of the problem of concept formation, try to remember the last time you were in a museum and saw a collection of art of some unfamiliar period or artist. Afterward, you could have recognized further examples of that style-you had formed a concept of that style of art. What exactly did you retain of the works on exhibit? What made them cohere in your memory? Simple attributes such as the subject matter, color, or even medium might not have been constant in all of the examples. In fact, words alone probably could not capture your concept of that style. How, then, does one represent that concept? Furthermore, what happened between your first view of the first example, when you perceived a work of art with many attributes and qualities but no characteristic style, and your last view of the last example, when you perceived a work of

art exemplifying a style, “fitting i n ” with the other pieces? The answers to these questions are by no means obvious, either through introspection or logic-which is probably why psychologists inherited them from many centuries of philosophers.

11. Information-Processing Theories The information-processing paradigm provides a framework within which numerous theories of concepts have recently been formulated. This framework rests on notions about the nature of mental representation and the processes that store, manipulate, and interpret those representations. Thus, we must begin our discussions of issues concerning concepts and concept formation with a brief explanation of the information processing paradigm. A.

THE COMPUTER ANALOGY IN PSYCHOLOGY

The scientific and historical forces that gave rise to the information processing paradigm are numerous (see Lachman. Mistler-Lachman, & Butterfield, 1979, Chap. 4, for a review). One of the key events was the demonstration by Alan Newell and Herbert Simon (see Newell & Simon, 1972) that the formal analyses of symbol manipulation by computational devices, which had been developed by Turing and others in the first half of this century, could be applied to the thought processes of humans. The core idea in modern information-processing psychology is that thought is symbol manipulation and the sequence of symbol processing underlying performance in any task can be described by a set of formal rules or instructions, like a computer program. An important analogy concerns levels of description in computational systems: Just as a computer’s internal events can be described on two levels, at the level of hardware (e.g., which wires are connected to which other wires, which gates have current flowing through them) and at the level of software (e.g., which subroutine was just called, which symbol is in which list), two levels of description exist for the human brain. The hardware level corresponds to brain physiology, and the software level corresponds to cognitive processes.* On this computational view of thought, then, the primary task of the cognitive psychologist is to describe the functioning, at the level of description of a computer program, of an enormously *Many theorists identify the cognitive level with software per se. This is not quite correct, because software can be “hard wired” into a machine. but the way to describe its operation is not changed. The imponant distinction here is between the /eve/ qf’descriprim at which we characterize a program and the level of description at which we characterize the hardware. In the former case operation is described in terms of algorithms; in the latter it is described in terms of physical law. We are not interested in the laws of neural operation, but rather in what computations the mechanisms of the brain are carrying out (see Fodor. 1981, for a detailed discussion of this distinction).

128

Martha J . Faruh and Stephen M. Kosslyn

complex computer, the human brain (see Fodor, 1981; Kosslyn, 1980, Chap. 5 , for further discussions of this point). The most important distinction derived from the computational view of thought is between structures and processes. Both structures and processes are defined at the level of software and describe entities like those that comprise a large computer program. Structures store information, and may thus function as representations. Processes operate on these representations, accessing the information contained in the representations, or altering their informational content. A computational model of concepts is a model of a kind of representational structure, defined at the level of the brain’s “program.” A computational model of concept formation is a model of the interactions of representational structures and the processes that produce concept representations. The relevant structures are the representations of stimuli and the representations of the concepts themselves. The relevant processes are those that create the concept representation from representations of individual concept instances, for example by the “averaging” of representations of instances to form a “prototype” (i.e., most representative instance). Structures and processes cannot be studied individually, in isolation from one another. The only way to detect a structure, let alone study one, is if some process operates on it. Similarly, the only way that a process can be observed is when it is operating on some structure. A serious problem for the psychologist, therefore, is separating the effects of structure and process when they must always be observed in combination. As a simple example of this problem with a computer, imagine that we want to know the order of a list of numbers in the memory of the computer. Our first inclination might be to simply have the list printed out. But what if the program that accesses the list changes the order? We cannot know whether it does so without knowing what the actual order of the list in memory is, but, of course, that is what we wanted to find out in the first place. The problem is that we are always observing a structure-process pair, and more than one such pair that often can account for a given piece of data (Anderson, 1978; F. Hayes-Roth, 1979; Pylyshyn, 1979). The list of numbers printed out by the computer, for example, could be a result of a list in memory with that order and a printout process that preserves the order, or a list with some other order in memory and a printout process that changes the order.? Later in this article we shall discuss some results on concept representation that can be explained by two or more equally plausible structure-process pairs. ’One might argue that in this case all one needs to do is look at the hardware to determine the actual list order. This is certainly possible, but easier said than done, even in a device of our own design. In the case of cognitive structures instantiated in brains, even if one could locate and observe the neural structures embodying mental representations, the problem of reading them would still remain. That is, we would need to know how to interpret the configurations of neurons and patterns of activation that we would observe.

Luckily, it is often the case in psychology that only one plausible structureprocess pair will account for a given result; the rest can be ruled out on the basis of being unparsimonious or ad hoc. The ability to rule out structure-process pairs on these grounds can be increased by a strategy of experimentation called “converging operations” (see Gamer, Hake, & Eriksen, 1956; Kosslyn, 1980). A structure can be observed in different experiments in which it is operated on by different processes. Characteristics that are observed in all processing contexts can then most parsimoniously be assumed to belong to the structure, not the various processes. The logic of converging operations is widely used in cognitive psychology; it is a sort of lever to get more confirmatory “force” out of evaluations of the degree of parsimony of theories and evaluations of the extent to which they are ad hoc or post hoc. B. MENTAL REPRESENTATION

Why should psychologists try so hard to attribute properties to psychological structures? Why not simply stop our investigations once we know that a given structure-process pair exhibits certain properties, an option discussed by Anderson (1978)? This question has several answers. First, the goal of cognitive psychology is to model the actual computations that underlie human intelligence. In reality, a particular property is possessed by either the structural or the processing component of the structure-process pair. There is a “fact of the matter,” and our goal is to determine what in fact are the contributions of the different structural and processing components. Second, and especially relevant for anyone interested in the nature and acquisition of concepts, there is good reason to try to discover the properties of the representational structures per se. These structures are literally the embodiments of knowledge, and speculation about their nature stretches back before the time of Plato (see the Theuetetus, Sec. 191). Current research on concepts is almost always research on concept representation, as we shall see. Third, the importance of mental representation for developmental psychology is measured by the fact that at least three major developmental theorists, Piaget, Bruner, and Werner, have claimed that changes in the nature of representation over age account for much of the child’s increasing cognitive abilities. Mental representations can be specified with respect to three properties: content, jbrmat, and orgunizution. The content of a representation is the specific information being stored. The original stimulus in large part determines the content of a representation. However, the content is not determined solely by the stimulus: The encoding processes also work to determine the content, given that some information from the stimulus is usually lost during encoding, and some may even be added (as in the case of subjective contours in visual perception). The format of a representation is the formal structure of the symbols and the

130

Martha J . Farah and Stepheri

M . Kosslyti

way they are interpreted. Different formats for representing information include spoken English, Morse code, and magnetic fluxes in a computer. The format in which information is stored in human memory is determined by the perceptual and cognitive processes that encode the information. For example, if the incoming information from the stimulus (light rays, acoustic waves, or whatever) is transformed during encoding and then stored as a description in terms of, say, a feature list, then the format of that representation is a feature list. Possible formats for the storage of information in memory that have been studied include visual imagery, auditory imagery, feature lists, linguistic phrase structures, and dimensions. The organization of a representation is a “finer-grained” characteristic than content and format. If particular contents are encoded in a format, such that the representation has parts, then those parts must be organized in some way. For example, the features in a feature list must be ordered in some way, for instance, by size, color, or any stimulus attribute, or even in a random order. We shall use these three aspects of representations to discuss the various models of concept representation in the remainder of this article. In the course of considering models of adult representation we shall consider the plausibility of the attendant models of concept formation.

111. The Contents of Concept Representations Most of the research on concepts has been focused on the nature of the contents of concept representation. We shall consider the three main classes of models that have been developed to date: those in which concepts are represented by rules of application, storage of individual exemplars, and storage of some kind of non-rule-based summary of exemplars. A.

RULE-BASED REPRESENTATIONS

The psychologists who first began to investigate concepts and concept formation assumed that all concepts have definitions in terms of a set of simple attributes, and that the representation of a concept is something akin to a definition or rule concerning the necessary and sufficient attributes for category membership. So, for example, “dog” might be represented as “four-legged, hairy, barking animal. ” The process of concept formation, it followed, was one of discovering the necessary and sufficient attributes by observing which attributes occurred in all and only members of the category. Research on concept formation was directed toward modeling the strategies of subjects trying to form concepts, to see how they inferred the necessary and sufficient attributes from examples presented to them by the experimenter. The concepts used in these experiments

were invented by the experimenter and typically involved some combination of simple attributes such as shape, color, and size of geometric forms. Out of this tradition of research came some extremely elegant and precisely confirmed models of subjects’ behavior in these situations (e.g., Bourne, 1966; Bruner, Goodnow, & Austin, 1956; Levine, 1971). However, efforts to extend these models to account for the learning of real, naturally occurring concepts were not fruitful. The strictly defined concepts of the psychology laboratory did not seem to be merely simplifications or idealizations of the more complex natural concepts, but were an altogether different kind of concept. The main problem with these artificial concepts was pointed out by the philosopher Ludwig Wittgenstein (1953) in a discussion of the meaning of natural language terms, and strikes at the heart of the idea of rule-based concepts. For almost all natural concepts, necessary and sufficient attributes simply do not exist. Wittgenstein illustrated this point with an attempt to define the word “game” in terms of a criteria1 feature set (features of all and only games). He could find no feature that was shared by all games that was not also shared by some nongames. For example, on one hand, not all games are played on boards (tennis) or involve competition (solitaire). On the other hand, broader characterizations of “game” such as “done for amusement” also apply to nongames (going to the movies). Wittgenstein suggested the term “family resemblance” to describe what instances of a natural concept have in common. Although no single feature may be shared by all members of a family, there is nonetheless a “family resemblance” among the different members. This resemblance comes about because each person shares some features with some relatives and other features with other relatives, although no one common feature occurs in all family members. The result is a network of resemblance or similarity among members of a family, and among instances of a natural concept. The notion of similarity is critical for any theory of concept formation because stimuli will be categorized together only if they are perceived to be alike in the relevant ways (Goodman, 1968). Thus, the rule-defined concepts presuppose a certain notion of similarity (although that word is not usually used for this kind of concept) when the necessary and sufficient attributes for all instances are specified. These attributes are what is common among the instances of the concept, by virtue of which they are similar. In natural concepts the instances are also all similar, but as Wittgenstein noted they need not be all similar in the same way. (The similarity relations among the letter pairs, ab, bc, ac, are like this.) Developmentalists will notice that the latter kind of similarity corresponds to the way children initially group objects, as the classic studies of Vygotsky (1962) revealed: Successive items sorted into a group are all similar, but not in the same respect, forming a one-dimensional similarity network, or a chain. Thus, we have a set of instances cohering by virtue of similarity, but with no single attribute or set of attributes common to all.

132

Murthu J . Faruh unit Stephen M. Kosslyn

The difference between all-or-nothing similarity among instances of a concept (with respect to the relevant aspects) and a family-resemblance similarity network has important ramifications for the process by which concepts are learned. All-or-nothing similarity requires that each stimulus be analyzed into a single set of features in terms of which hypotheses can be framed and tested to determine which features are criterial. Using family-resemblance similarity, in contrast, requires almost no analysis of the stimuli. The learner need only be able to detect similarity, not have any insight into the basis of that similarity; it does not matter in which respect the instances are similar. The family-resemblance property of concepts has sometimes been called “internal structure” because of the network structure of shared attributes within the concept, and one of its consequences is that some instances of a concept may be more “central” or “better” instances than others. A more central instance shares more features with other instances than does a less central instance. For example, in the family-resemblance network of the concept “bird,” a robin is more central than a penguin because it has more features in common with other birds, such as its beak, feathers, size, and flying ability, than does a penguin. The fact that family-resemblance networks result in some exemplars being more central in a category than others is in keeping with the results of much psychological research indicating that natural concepts do have graded membership, and even fuzzy boundaries at which point people are not sure about concept membership (e.g., is a refrigerator a piece of furniture? see McCloskey & Glucksberg, 1978). Internal structure, and the graded membership and fuzzy boundaries that usually follow from it, is a property of natural concepts that is not shared by rule-based concepts, defined by necessary and sufficient conditions such as “all red triangles with blue borders. At first glance, this fact would seem to rule out the hypothesis that concepts are represented in the mind as conjunctions of necessary and sufficient attributes. However, it might be possible to salvage the rule-based representations, in form if not in spirit, by allowing rather complex disjunctive rules to constitute the psychological representation of concepts. In other words, we could give up trying to represent concepts by a single set of necessary and sufficient attributes (because they evidently do not exist for many concepts) but maintain the idea of a rule-based representation, in which some disjunction of sets of features (e.g., “played on a board or done for competition and not for pay”) would be the criterion for concept membership. (Of course, any concept that can be expressed using the propositional calculus can be represented this way. However, just because a concept can be expressed in logical terms does not mean that is is mentally represented that way. The issue is whether the psychological representation is itself a rule.) In the end, it is psychological considerations, and not analytic ones, that lead us to reject rule-based concepts. First of all, Bruner himself found that disjunctive ”

Concept Development

133

rules were much more difficult to learn than conjunctive rules. More importantly, the process of rule learning, conjunctive or disjunctive, is simply not a plausible model for concept formation-especially in children. All rule-learning models are essentially hypothesis generation and testing models. As Fodor (1975) has pointed out, no one has ever conceived of a way for rules to be learned without hypothesis generation and testing (excluding simple memorization of a rule by rote, which is not to the present point, and the learning of behaviors describable by rules, which is also different from learning a rule per se). But children generally do not perform hypothesis generation and testing systematically or efficiently until the onset of formal operations in adolescence (see Flavell, 1963, pp. 203-205), and it is certainly a task that infants and very young children perform poorly if at all (e.g., Ault, 1977; Olson, 1966). Paradoxically, it is during these early years that children acquire concepts at a rate never again approached in their lives: Templin (1957) has estimated that between the ages of 1.5 and 6 years a child learns an average of 5 words per day, or one every few hours. And vocabulary size is probably a conservative index of how many concepts are learned during these years, for not all concepts will be named (while only a few of the words are proper names). What are the alternatives to traditional rule-based models of the psychological representation of concepts? A viable alternative not only must be able to accommodate the data indicating that natural concepts have an internal structure, but also must be compatible with a plausible learning process. Several types of models meet these requirements, and they will be discussed in the remainder of this article. They fall into two general classes, summary models and exemplarbased models. The difference between summary and exemplar-based models is, predictably enough, a difference in the amount of summarization of the exemplars that takes place during concept formation. One of the unexamined assumptions of the traditional approach to studying concepts is that concept formation is a kind of abstraction or distillation of information gleaned from exemplars of the concept. Concept representations were supposed to be more succinct in content than the combined representations of the exemplars from which the concept was learned. Although rule-based models of concept representation are now generally considered inadequate, the “succinctness” assumption has continued to be held to various degrees by different psychologists. One of the basic questions that we can ask about concept formation is, then, how and to what extent is summarization achieved? What content in the exemplar representations is pared away, and what is retained to represent the concept? Probably the majority of research on concept representation and formation has been addressed to this question about content, either directly or indirectly. We shall first consider two kinds of models that feature nonruled based summary representations, prototype models and cue set models, and then discuss the recent literature on exemplar-based models,

134

Martha J . Furuh und Stephen M . Kossiyrl

which depart from the succinctness assumption in that the content of an exemplar-based concept representation is not a summary of exemplars, but simply information about the individual exemplars themselves. B. NON-RULE-BASED SUMMARY REPRESENTATlONS

In prototype and cue set representations, redundant information from exemplars is eliminated. A prototype is either an exemplar that is selected as being the most similar to the remaining exemplars or a representation that is generated by an averaging process; in either case, the prototype is a representation of the (actual or mentally constructed) “most typical” exemplar of a concept. A cue set is simply a set of properties that are characteristic of the exemplars of a concept. Prototypes and cue sets are summaries of information about exemplars, and thus are more economical than exemplar-based representations in that any attribute that characterizes more than one exemplar will occur only once, if at all, in the summary representation (possibly with some sort of weighting information to indicate its preponderance among the exemplars). In contrast, if encodings of individual exemplars are used to represent the concept, each occurrence of that property will be represented. The lighter demands on memory by the summary representations are achieved at the cost of additional processing at the time of learning. This trade-off has developmental implications because the relative difficulty of storing large amounts of information versus processing it may change over age. Prototypes and cue set representations also differ in content from each other, although the exact distinction has not been drawn clearly in the literature. The essential difference seems to us to be that, unlike cue set representations, prototypes must be structurally similar to the exemplars from which they are formed. For example, the prototypical dog must have a head, a tail, four legs, and so on; in short, it must have a token of all of the types of features that any individual dog has. In contrast, a set of cues for the concept “dog” might not include all of these features. Or, it might have more than one token of each feature type, for example, the several head shapes that are characteristic of dogs (long and sharp like a wolfhound, squashed in like a boxer, jowly like a basset hound, and so on). This difference in the content of prototype and cue set representations is a result of the additional constraints that exist for prototypes, namely, that they be structurally similar to the exemplars of the concept. This constraint requires that the learner have or develop a rather specific structural schema for analyzing the instances that belong in the same class, and thus may also have developmental consequences. Traditionally, the difference between prototype and cue set representations has been explained with reference to two superordinate classes of models: distance models, of which prototype models are an instance, and probability models, of which cue set models are an instance (see B. Hayes-Roth & Hayes-Roth, 1977;

Reed, 1972). In distance models it is the “distance” (that is, similarity) of an item to the concept representation, be it a prototype or exemplars, that determines if the item should be considered an instance of the concept; only items falling within some set distance would be considered instances of the concept. Probability models are those models in which the probability that an item is an instance of the concept determines whether it is classified as an instance or not. The source of the probability estimate is the same as the source of the similarity measure, namely, the amount of content in common between the item in question and the representation of the concept. We are departing from this distanceprobability system of classifying models because, as we shall argue later (Section III,B,3), we think it has more to do with differences in the mathematical language that has been used to describe these models than with differences in the putative psychological representations and processes themselves. I . Prototypes A prototype represents a concept by its “central tendency. ” It summarizes information about the instances of a concept in the same way that statistical measures of central tendency, such as the mean, median, and mode, summarize a set of numbers. The notion of central tendency for sets of objects or events is not as clear-cut as it is for numbers, however. It has been interpreted in different ways by different psychologists, partly depending on the nature of the concepts and stimuli studied. One of the early investigations of prototype representations was performed by Posner and Keele (1968). Their stimuli consisted of sets of random dot patterns, one of which was the prototype of a set and the rest of which were computer generated distortions of the prototype. The sets of dot patterns thus generated had an internal structure like that of natural concepts: The instances had varying degrees of typicality (corresponding to varying amounts of distortion from the prototype), and no single defining attribute was common to all instances. Because the distortions from the prototype were random, the prototype was the central tendency of the set, a kind of spatial average of the patterns in the set. Subjects in this study learned the concepts (the pattern sets) by viewing several of the distorted patterns, but not the prototype. After learning two concepts, they were asked to judge the set membership of further patterns. Some of these patterns were old examples that they had seen, others were members of one of the sets but new to the subject, and others were not members of either set. Subjects’ recognition memory for the old examples was tested as well. Posner and Keele ( 1 968) found that subjects could classify the patterns into the correct set, but in addition, they consistently identified the prototype as having been one of the initial examples of the category. This finding indicated to Posner and Keele that, in addition to remembering individual patterns, subjects had formed a representation of the central tendencies of the sets. In other words, subjects constructed the

136

Martha J . Farah and Stephen M . Kosslvtr

prototypes. Posner and Keele (1970) found that subjects who were tested after a 1-week interval showed poor memory for individual patterns, but remembered the prototypes, which also was taken as evidence that learning the concepts involved forming and storing the prototypes. The spatial average conception of prototypes was given a stronger test by Reed (1972), who pitted it against several other models of natural concept representation that lead to similar, but not identical, predictions. Reed taught subjects categories of Brunswik faces (schematic faces that vary along four dimensions: eye height, eye separation, mouth height, and nose length). His procedure was similar to that of Posner and Keele (1968); after studying several instances of the categories to be learned, subjects were asked to place further instances in their proper categories. The models he considered were: ( I ) a spatial average prototype model, like that of Posner and Keele; (2) a cue set model; (3) an exemplar-based model in which instances are classified according to the category of the most similar stored exemplar (the proximity model); and (4) an exemplarbased model in which instances are classified according to the category of the stored exemplars, which are, on average, most similar to the instance (the average distance model). He also tested a weighted prototype model and a weighted average distance model, in which distances along the different stimulus dimensions were differentially weighted to reflect how well they could predict category membership. For instance, a dimension that has similar values within a category, but very different values for different categories, would be weighted heavily, because it will be a good predictor of category membership. Reed (1972) examined his subjects’ classifications of certain key items designed to distinguish among the models, and found that the weighted prototype gave the best account of the data. For many kinds of concepts the operation of averaging does not make sense. What is the average piece of furniture? A spatial average like that of Posner and Keele (1968) would probably result in a nonsense object. It does make sense to speak of a prototypical piece of furniture, however; our intuitions are that it would probably be a sofa or chair. Similarly, the average of a set of colored circles, 90% of which are blue and 10% of which are red, would be a purple circle. However, a purple circle hardly seems prototypical, in that the original set contained no purple circles. A blue circle seems closer to our intuition of the prototype. For concept domains that are naturally described in terms of prothetic dimensions (the “quantitative” dimensions, such as size, displacement, or intensity, which cumulate or increase and are therefore averageable), it makes sense to treat the prototype as the average of the concept instances. For other concept domains, we clearly need a different conception of prototype. A more sophisticated version of the prototype model has been proposed by Rosch (see Rosch & Mervis, 1975), based on Wittgenstein’s notion of family resemblance. She has successfully applied it in the most difficult concept domain

of all to study, the natural semantic concepts named by the words of our language. The typicality of an instance of a natural semantic concept is determined by calculating a “family resemblance score” for it. The family resemblance score of an instance is the weighted sum of its features; the weights correspond to the frequency of occurrence (and thus the characteristicness) of the features among the instances of the concept. Thus, this score reflects how often the features of an instance appear among the other instances of the concept, which is a direct measure of how similar that instance is to the others, and thus of how typical it is of the concept. For example, let us say that the instances of a concept are ABC, BDA, AMN, ABM, and MQB. The family resemblance score of the last string is 3 1 + 4 = 8, because M occurs three times, Q occurs one time, and B occurs four times. The prototype is ABM since it has the highest possible family resemblance score: 4 3- 4 + 3 = I 1 . If two or more contrasting concepts are being learned, then Rosch’s method is to adjust the weights of the features to take into account not only their frequency within a given concept, but also their frequency in contrasting concepts. For example, if the letter A occurred twice in a letter-string concept contrasting with the one given above, then A would be less characteristic of the first concept, and its weight would be changed to reflect this fact, from 4 (its frequency in the first concept) to 4 - 2 = 2 (its frequency in the first concept minus its frequency in the contrasting concept). This featureweighting procedure is similar to the dimension-weighting procedure used by Reed. In both cases the effect is to weight the most characteristic aspects of the concept most heavily. Rosch (see Rosch & Mervis, 1975) did not, of course, know the actual features used to represent natural semantic concepts, or even whether features are in fact used to represent them internally. In order to obtain “features” for each concept instance from which to calculate family resemblance scores, she asked subjects simply to list all the features that came to mind as belonging to various concept instances. The family resemblance scores thus obtained from the associated scores correlated highly with how typical or “good” an instance was judged to be by other subjects. These scores also predicted the time required to verify that an item was an instance of a natural concept (e.g., that a robin is a bird), and accounted for both learning time and verification time for instances of artificial categories with specific featural structures (such as letter strings). One particularly compelling study involved the effects of priming a word-matching task (e.g., are “apple/apple” the same words?) with the name of the superordinate category (e.g., “fruit”). If the words to be matched had a high family resemblance score in the superordinate category, then the match was performed more rapidly than with no superordinate category priming. Responses were actually slowed (relative to no priming with the superordinate category) when the word pair was composed of low family resemblance members (e.g., ‘‘papaya/ papaya”). This result was taken to support the idea of prototype-based concept

+

138

Murthu J . Furuh and Stephen M . Kosslyn

representation for the following reason. If the concept “fruit” is represented as a prototypical piece of fruit, then priming with the word “fruit” should bring that prototype to mind. To the extent that the objects named by the word pair are similar to the instance already in mind, identification (which presumably occurs when subjects judge that two words are the same) will be primed. In the case of peripheral instances, the prototype will not prime identification but will serve only to distract the subject. Other conceptions of “central tendency” have included the modal average model (see Neumann’s attribute frequency model; Neumann, 1974), in which the prototype is the instance possessing the most frequent token of each type of feature or the most frequent value on each dimension, and the least transformation distance model of Franks and Bransford (1971), in which the prototype is the instance requiring the fewest transformations (from some predetermined set of allowable transformations) to generate the other instances of the concept. The evidence for these models is based on the same logic used by Rosch and Mervis ( I 975) and by Posner and Keele ( 1 968). Briefly, the prototype is determined by the experimenter’s model but is not shown to the subjects during the learning phase, and the subjects’ classifications of new instances after the learning phase are shown to be based on their similarity to the prototype. The Franks and Bransford (1971) experimental paradigm differs slightly from the others in that the subjects are presented a memory task rather than a concept-learning or categorization task, and the only classification that subjects perform after learning is to classify the transfer items as “old” or “new.” Nevertheless, the results suggest that subjects learn the concepts whose instances they are told to memorize, and later judge transfer items according to whether they are instances of the concept.

2 . Evidence That Children Use Prototypes Part of the impetus to investigate prototype representations for natural concepts came from developmental considerations. The rule-based concepts seemed to require too much systematic analysis during learning to be a plausible kind of concept representation for children; therefore the proposal that concepts might be represented by prototypes was quickly seized by developmental psychologists. Concept formation in a prototype model simply requires that similarity be detected among the instances of a concept so that the prototype (the actual or generated instance with the greater similarity to other instances) can be selected or formed. Rosch ( I 973) reports that both children (9- to 1 I -year-olds) and adults require more time to verify that an atypical object is an instance of a given concept than to verify a typical object. For example, children and adults both take longer to answer “yes” to the question, “Is a penguin a bird?” than to the question, “Is a robin a bird?” Rosch held constant the frequency of occurrence in English of

the words for the central and peripheral instances, and thus the difference in time to respond to the question cannot be attributed to longer “look-up” time for the peripheral instances themselves. Therefore, Rosch reasoned, the difference must result from the way the decision is made, specifically, from a comparison between the representation of the instance and the prototype. If the instance is very similar to the prototype, the answer is almost immediate. If the instance is peripheral, then a more thorough comparison must be performed, leading to longer decision times. Thus, children seem to represent their semantic concepts the same way as adults, a prototype being used by both. The difference in verification time between central and peripheral items was, in fact, larger for children than adults. This result could imply that the more prototypic exemplars are learned first or that the most typical properties of exemplars tend to be stored first. Anglin (1977) also conducted a study to investigate the role of typicality in concept formation in children. He examined the effects of typicality on the child’s failure to apply a word to an object. This sort of “underextension” is, for obvious reasons, difficult to study naturalistically. Anglin created an experimental situation in which failure to apply a name would be detectable. The children in his study ranged in age from 3 to 6.5 years old. For each of the concepts “animal,” “food,” “clothing,” and “bird” they were shown a series of pictures of instances and noninstances. After seeing a picture they were asked if it was an instance of a concept, for example, “Is this a bird?” Anglin reasoned that if children represented the concepts as a prototype, then they would confidently classify typical instances of the concept because those instances would be very similar to the prototype. They might be less willing to label peripheral instances with the same name, however, because those instances would not resemble the prototype. In addition to the peripheral-central division, Anglin also used a familiar-unfamiliar division, thus forming four subsets of instances. Peripheral instances evoked the most frequent underextensions. Typicality was a far stronger determinant of how likely the children were to correctly name a concept than was familiarity. For example, they more often called a wombat an animal than they did an ant, despite the fact that ants were more familiar. Carey and Smith (1978) found evidence that the similarity of a concept instance to the prototype plays a role in children’s reasoning about concepts. They asked 8-year-old children whether various animals eat, sleep, breathe, have hearts, and so on. The children’s willingness to attribute these properties to a particular animal depended on the typicality of the animal. Typical animals (e.g., dogs, cows) were thought to have many animal properties, while peripheral animals (e.g., fish, flies) were thought to have few. This was taken to indicate that the representation of the concept “animal” with which these properties are associated is a prototype, and that the association between these properties and a particular animal is determined by the similarity between that animal and the

I40

Murrha J . Faruh and Stephen M . Kosslytl

prototype. Carey and Smith also investigated the role of the prototype in learning new facts about a concept by teaching new properties (e.g., “has a spleen”) about typical and peripheral instances, They found that children would generalize knowledge of new properties to other animals if it was originally learned about typical animals, but not if it was learned about a peripheral animal. Lasky (1974) examined the ability of 6-year-olds, 8-year-olds, and adults to abstract prototypes from sets of geometric patterns. The concepts consisted of a prototype pattern (four different geometric shapes in four particular positions in a rectangular field) and instances generated from the prototype by the application of substitution, deletion, and permutation transformations. As noted above, Franks and Bransford (1971) found that subjects seem to abstract the prototype and transformations even when they are told simply to remember the individual instances of a concept for a later recognition test, with no concept learning instructions at all. Lasky replicated the Franks and Bransford findings with his 8-year-old and adult subjects, but found small, systematic departures from the Franks and Bransford predictions for the 6-year-olds. After dividing the transfer stimuli into groups defined by the presence or absence of different shapes in different positions, he discovered that the 6-year-olds had been classifying patterns on the basis of an incomplete prototype, consisting of the most frequent shapes that occurred in each position, but lacking a shape in one of the four positions. The missing shape was much less frequent in the training stimuli than the others. Lasky proposed that the Franks and Bransford transformational model should be viewed as an abstract “competence” model, using Chomsky’s (1965) distinction, and that the psychological processes involved in “performance are the observation of the frequencies of different shapes in the training stimuli, and the formation of a modal average prototype. His conclusion was that the 6year-olds differed from the older children and adults only in the speed and efficiency with which they formed the modal prototype, not in how they formed it. ”

3. Cur Set Representutions In cue set representations, characteristic aspects of the instances of a concept are stored and used to identify new instances. As in Reed’s (1972) weighted prototype model and Rosch’s (1975) family resemblance model, frequency counts are maintained within and across concepts so that the concept learner knows how characteristic a particular aspect is of a particular concept. The significant difference between this sort of model and the prototype models is the lack of constraint on what aspects of the instances may be represented in the concept representation. The cue set need not “look like” the representations of the instances. That is, although the prototypical dot pattern, letter string, or piece of furniture would fit right in, and be “of a kind” with the dot pattern, letter string, or furniture representations from which it was formed, cue set

Cotic.ept Develnpnient

141

representations of these concepts might be quite different. The instances of a letter string concept might be five letters long, for example, but if more than five letters occur frequently among the instances, the cue set representation will contain more than five letters. Although structural relations cun be encoded as cues (e.g., “A follows B” in a letter string), the structural relations among the different aspects of the stimuli need not necessarily be encoded in order to be able to form a cue set representation. For this reason, cue set representations may play a role in development; they may represent intermediate amounts of knowledge about structural relationships (from none to all), and additional information may be added to these representations as it is learned. Cue set models in effect incorporate the same similarity measure as prototype models: cue overlap. Cues need not be discrete features, although for dimensional stimuli, the additional assumption must be added that probabilities generalize along dimensions, so that knowledge that a particular value is highly diagnostic implies that neighboring values will be diagnostic too, to the extent that they are close to the known value. This assumption parallels an assumption of prototype models, namely, that distance from the prototype will predict concept membership. Thus, cue models will be able to mimic all the properties of prototype models, explaining the effects of typicality and so on simply by substituting cue overlap (or weighted cue overlap) for “distance. ” Cue set models usually do not fare well in the face of data, but this failure is not surprising when we look at the specific versions of cue set models that have been tested. In Reed’s (1972) important study he included among the models tested a weighted prototype, which he said “combines aspects of both the cue validity (probability) and distance models”; but he did not test an equally powerful cue set model-one that combines probability and similarity (“distance”) information. The cue set model he tested assigned separate and unrelated probabilities for each value on a dimension, with no attempt at using similarity of values to generalize probabilities. For example, a face with a short nose and a face with a medium nose were equally likely to be classified with a long-nosed face. Not surprisingly, then, this model proved a poor predictor of subjects’ classifications. Franks and Bransford (1971) also tested a cue set model, and found that it fared poorly when compared to their own prototype model. In this case the failure of the cue set model may have resulted from the experimenters’ assumptions about which aspects of the stimuli would function as cues for the subjects. The absence of structural constraints on which aspects of the instances are represented in the cue set presents a special problem for the experimenter who wants to investigate such representations. We cannot know a priori which aspects or combinations of aspects of the instances will be considered cues. That is, we are faced with the problems of (1) not knowing how subjects initially parse the

142

Martha 1. Faruh w i d Stephe17 M . Kosslyn

instances (into features, dimensions, image segments, or whatever), and (2) not knowing how these “atomic” cues are used. “Compound” cues might also be encoded, representing what is sometimes called ‘‘relational ” information. That is, patterns of covariation among the aspects of the instances could also be used as characteristic of concepts. Experimental psychologists are often able to finesse the first problem by using artificial stimuli that have a very obvious “parse,” such as letter strings, which are clearly decomposed into letters. With cue set models, however, the second problem remains. The lack of constraints on which aspects may be represented makes possible a cue set consisting of the power set of all “atomic” aspects. A power set is the set of all possible subsets of a set. If we are to test a cue set model rigorously and fairly, we must know which cues of this set of potentially usable cues are actually used in representing the concept and classifying new instances. B . Hayes-Roth and Hayes-Roth ( I 977) dealt with the problem of defining the “psychologically real cues” by simply assuming that subjects would represent a concept by all possible cues (i.e,, the power set of each exemplar’s attributes) and that they would make classification judgments based on the most diagnostic cue available. Diagnosticity refers to the ability of the cue to predict membership in a particular concept. Thus, the diagnosticity of a cue increases with its frequency of occurrence among the exemplars of a concept, and inversely decreases with its frequency of occurrence among the exemplars of other concepts. Hayes-Roth and Hayes-Roth pitted their “most diagnostic property set model against Franks and Bransford’s (197 1 ) prototype model by independently varying the cue frequencies and the transformational distances from the prototypes of a large set of exemplars of two concepts. The concepts were clubs, the exemplars were club members, and the properties were age, level of education, and marital status. The predictions of the two models were contrasted for both subjects’ recognition (new versus old) of exemplars, and classification (Club 1 , Club 2, or neither). The cue set model provided a significantly better fit to the subjects’ answers than did the prototype model. The Hayes-Roth and Hayes-Roth model is successful, where so many “straw” cue set models fail, because its definition of a cue does not arbitrarily exclude combinations of simple aspects of exemplars, People encode relational information; therefore, it is reasonable to assume that they use such information for representing concepts and classifying new instances. It should be noted, however, that this model is far from being a simple summary model. Subjects must keep track of the relative frequencies of all the properties and combinations of properties possessed by each exemplar of each concept. Perhaps this task was possible in the Hayes-Roth and Hayes-Roth experiment because of the relatively small number of relevant dimensions (three) and the relatively large number of exemplars (1 32) from which subjects could derive frequency estimates.

”

Evidence That Children Use Cite Set Representations Posnansky and Neumann (1976) found evidence that children as young as 7-year-olds classified letter trigrams (such as XZX) on the basis of letter frequency. Their experiment was designed to distinguish between Neumann’s attribute frequency model (which is a modal prototype model similar to the “performance” model proposed by Lasky, 1974) and the Franks and Bransford (1971) transformational distance model. They gave the Franks and Bransford concept learning task to children ranging in age from 7 to 12 years old, and found the attribute frequency model a better predictor of the children’s classifications. Although they did not contrast the predictions of the attribute frequency model, which is a prototype model, with the predictions of a cue set model, we think that their data actually may support a cue set model. Recall that the difference between the two models is that a prototype is structurally similar to the instances of the concept, whereas a cue set model merely reflects the frequency of various cues among the instances of a concept. If we take the structure of the trigrams to be the three letter positions, then the data seem to implicate an unstructured cue set representation rather than a modal prototype. For instance, they found that XXY was more confidently classified as a concept instance than WWX, even though the frequency of each letter in each position was the same. (That is, X occurred in the first position as often as W, and so on.) The total frequency of the letters comprising the strings is not the same, however: Xs were more frequent overall than Ws. Thus, the children may have used overall frequency of cues, as in a cue set representation, rather than a modal prototype.

4.

C . EXEMPLAR-BASED REPRESENTATIONS

In this section we shall discuss the simplest type of model, according to which a concept is represented solely by encodings of at least some of its known instances. Such models deviate from the traditional assumption that concept representations are succinct in that they contain no “summary statement” representing the concept, only a set of encodings of individual instances encountered in the past. Thus, the concept “dog” is represented by encodings of individual dogs that a person knows. A new dog is recognized as a dog if it is similar to one or more of the known dogs. A dog is judged more or less typical, depending on how many known dogs it resembles, and on how close that resemblance is in each case. Learning a concept requires very little cognitive sophistication, simply the ability to remember individuals along with the concept that they exemplify. One common objection to exemplar models is that the number of stored exemplars would grow so large that either the ultimate capacity of memory would be exceeded, or the search processes required to access a similar exemplar would be too time consuming. In reply, it should first be pointed out that the

144

Martha J . Farah and Stephen M . Kosslyn

memory capacity of the human brain is so large that no one (to our knowledge) has even been able to estimate the amount of information that a person may store in a lifetime. Second, we seem to be extraordinarily good at searching our vast memories. For example, Standing, Conezio, and Haber (1970) found that subjects could recognize with great accuracy over 2500 photographs after exposure of only 1 second each. Above and beyond this line of defense, it is reasonable to suppose that the learner becomes selective as the number of stored exemplars grows, so that only exemplars that differ significantly from those already stored are added to the set of stored representations. A set of experiments performed by Medin and Schaffer ( I 978) demonstrates that people use exemplar-based representations under the conditions used in standard concept learning experiments. Their subjects were shown a series of six exemplars, three from each of two categories. After seeing each exemplar the subjects were asked to classify it, and after responding were told the correct category. This procedure continued, the six exemplars being presented in a different random order each time, until the subjects classified them all correctly twice in a row or until 20 repetitions of the exemplars had been given. Afterwards, the subjects were asked to sort a set of transfer items into the two categories, and to give confidence judgments of their classifications. They also made “old”/“new” judgments with confidence ratings. One week later, they performed the classification and recognition tasks again. The concepts used by Medin and Schaffer ( 1 978) were composed of two kinds of stimuli: geometric shapes, which could have one of two forms, colors, sizes, and positions, and schematic faces, the same stimuli used by Reed (1972). The predictions of the “context” model of Medin and Schaffer for the classification and recognition judgments differ from those of the exemplar-based models tested in the past (e.g., Reed, 1972), and also from those of the prototype and simple cue set models described earlier in this article. The difference stems from the way in which similarity between exemplars is calculated. In the other models described in this chapter, similarity is additive. That is, the amount of similarity along each stimulus dimension, or resulting from each feature of the stimuli, contributes fully and independently to the overall similarity between exemplars. Medin and Schaffer used a multiplicative rule for combining components of similarity, by which the similarity resulting from different aspects of the stimuli is multiplied rather than added. The multiplicative rule enables great dissimilarity along one dimension to prevent two exemplars from being considered similar, even if they are highly similar along all other dimensions. This rule seems to capture the way in which similarity is combined psychologically. For example, a mannequin would not be mistaken for a person, despite its high similarity to people on most dimensions, largely because it has zero similarity along the movement dimension.

Concept Development

145

A distinctive consequence of the mutliplicative similarity rule for classification is that an item may be equally similar, on average, to the exemplars of two categories, but if it is highly similar to an exemplar of one category then it will be classified in that category. To see why, consider a simple example given by Medin and Schaffer (1978) in which two categories have only two exemplars each. If an item to be classified is very similar to an exemplar of category A (matches on three out of four dimensions), then its similarity to that exemplar is 1 X 1 X I x s, where 1 is perfect similarity, and s is some amount of similarity less than 1. If the item is not very similar to the other exemplar of A (matches on only one dimension), then its similarity to that exemplar is 1 x s x s x s, and s3. If the item is the net similarity between that item and category A is s moderately similar to both of the exemplars of category B (matches on two dimensions), then it has s2 similarity to each, and a net similarity to category B of 2 s z . Note that the additive similarity, over all components of the exemplars, is the same for both categories (namely, 4 4s).However, in general (except for the uninteresting cases of s = 1 and s = 0) s + s 3 is greater than 2s2, and the net similarity of the item to category A is thus greater than to category B. The Medin and Schaffer (1 978) model thus predicts that when an item is on average equally similar to the exemplars in two different categories, it will nonetheless be confidently classified in the category with the highly similar exemplar (all the other models discussed in this article predict a toss-up between the two categories, with subjects choosing each category equally often, and giving low confidence ratings for their classifications). Medin and Schaffer designed the training and transfer items to incorporate a series of such contrasts, and the data they obtained supported their model. It is interesting, and a bit sobering, to note that Medin and Schaffer (1978) also obtained the standard typicality effects in classification and recognition, with more typical instances being classified and recognized more reliably. Further, they also found better retention of the prototype than other exemplars when subjects were tested one week later. Their model accounts for these effects, which were until now taken as evidence for prototypes, as follows: The prototypic exemplar in concept learning experiments is generally the exemplar having the greatest number of attributes that are highly similar to stored attributes of other exemplars. Thus, the more typical transfer items are classified and recognized more confidently because highly similar exemplars can be found in memory. The finding that memories of all but the most prototypic stimuli fade after a week can be accounted for by noting that, on the average, it is more probable that all occurrences of noncharacteristic attributes (which, by definition, occur less frequently in instances of the concept) will have decayed than every occurrence of a characteristic attribute (which will be included with many exemplars). Thus, the properties of an “average” exemplar will still be stored, although not neces-

+

+

146

Martha J . Farah und Stephen M . Kosslyn

sarily all with a single exemplar, and a prototypical exemplar will still be similar to relatively many attributes of the representation^.^ I . Nonanalydic Exemplar Representations Concept formation in an exemplar-based model is extremely simple: the learner must simply remember the exemplars, Analysis of the exemplars is thus optional. All that is required to classify a new item as an instance of the concept is to perceive similarity in some respect between the item in question and at least one of the stored exemplars. One of the main disadvantages of rule-based representations for concepts is that the learner has to discern the rule-the exact basis for the concept. The simplest model of concept formation would not require the learner to perform any analysis beyond that required simply to encode the exemplars of the concept. Ideally, the learner would not even have to distinguish between the relevant and irrelevant aspects of the exemplars. At first glance this approach seems doomed by the simple observation that if irrelevant aspects of exemplars are not eliminated at input, they will enter into the concept representation and thus into classification judgments using that representation. For example, we would not expect a viable representation of “dog” to include information about a wall behind a given dog or the illumination level. The possessor of the concept knows that these are incidental, irrelevant aspects of the “dog” exemplars that he or she has seen, and thus not include location and brightness information in the representation of “dog,” or use these aspects of a new item in judging whether it is a dog or not. (Which aspects of a concept instance are irrelevant depends on the contrasting concepts in a given concept learning situation. For instance, if the learner was forming the concept dog at a time when the manufacturers of the Gladys the Goose lamp had just come out with a glowing dog model, then brightness would be a relevant aspect.) Somewhat surprisingly, however, this sort of a minimalist model actually seems viable in certain contexts, which will be described momentarily (see Brooks, 1978). It makes sense that irrelevant information is sometimes encoded if for no other reason than the learner sometimes is unable to tell which stimulus aspects are relevant and which are irrelevant-especially at the very early stages of concept development. This might be the case because of insufficient knowledge about the concept domain (e.g., animals are not generally distinguished by brightness) or because the learner’s analysis did not single out the appropriate aspects (e.g., brightness on the one hand and shape and color on the other) in the first place. 4We have taken some liberties in translating Medin and Schaffer’s (1978, p. 214) account of differential retention of the prototype from the abstract mathematical form in which they presented it into a mechanistic form.

The kind of concepts we have been discussing have been dubbed nonanalytic concepts by Brooks (1978). This nomenclature may be confusing to readers who expect this kind of concept to involve no analysis of the stimuli. The difference between an “analytic” and a “nonanalytic” concept is not that no analysis has taken place in the latter case (figure must be separated from ground even here, for example), but rather that in a nonanalytic representation the analysis at the time of encoding did not distinguish the relevant from the irrelevant aspects of the concept instances. An example will make the notion of a nonanalytic concept clearer. Suppose that we go back to the museum with a member of some nomadic tribe, who has never seen art (except perhaps the decorated tools and clothing of his tribe). We take him to the exhibit where we learned about the new style of art. Later we test him on some new paintings, asking him to categorize them in terms of style. The nomad gets quite a few right, although he does make some incorrect classifications. When we examine the basis for each classification, we find that the correct classifications were usually based on some appropriate similarity to the museum paintings, for instance, brush strokes, subject matter, or color palette, and the wrong classifications were based on inappropriate similarities to the museum paintings, for instance, type of frame, position on wall, or artist’s first name. The tribesman evidently adopted a strategy of classifying new paintings on the basis of their similarity to paintings that he remembered from his museum trip, which he knew exemplified the style of painting. However, he used similarity in all respects, including those which are irrelevant to artistic style. In other words, he has a nonanalytic concept of that style of art. Nonanalytic concepts differ so radically from analytic concepts that we are tempted to add another piece of jargon to the field and call them “pseudoconcepts.” They are included in this article nevertheless because they function surprisingly well in place of true concepts, and some preliminary research results suggest that they may play an important role in cognitive development. Brooks ( 1978) has demonstrated that people will form nonanalytic exemplarbased concepts under circumstances in which, because of the complexity of the concept, they cannot analyze the exemplars into relevant and irrelevant aspects. For example, in one experiment Brooks taught subjects complex letter string concepts corresponding to artificial grammars. The instances of these concepts were letter strings in which the order of the letters was constrained by a Markov grammar. There were two different learning conditions in the experiment. The subjects in one condition, the analytic condition, participated in a “reception” concept learning paradigm. That is, they were shown one letter string at a time, for each string were asked to guess which of the two grammars generated it, and were given feedback on the correctness of each classification. The experimenter presented the complete training set of letter strings in this way as many times as

148

Martha J . F a r a h and Stephen M . Kossiyn

necessary for the subject to produce one sequence of errorless classifications. Subjects in the second condition, the nonanalytic learning condition, participated in a paired associates learning paradigm. Rather than classifying the letter strings, they were simply required to memorize each letter string in the training set along with a word associate, so that if they were later given the word they would be able to produce the associated letter string. Thus, subjects in the nonanalytic group were required to encode both the relevant and irrelevant aspects of the instances, while subjects in the analytic group were allowed to analyze the instances as best they could and encode into memory only what they thought relevant. Half of the associated words that were learned by the nonanalytic group were names of cities, and half were names of animals. Brooks foiled any possible attempts by subjects in the nonanalytic group to analyze the categories during memorization by making the city -animal division cut across the two categories defined by the grammars. Thus, half of the animal-letter string pairs contained strings from one grammar, and half contained strings from the other grammar. The same was true for the city-letter string pairs. However, strings from one grammar were systematically paired with old world items (cities, such as Paris, or animals, such as tiger) and strings from the other grammar were systematically paired with new world items (New York, armadillo), which no subject suspected. Brooks (1 978) then informed these subjects that there was an old worldnew world division, and that it corresponded to strings generated by two different grammars. After the initial task, both the analytic learning group and the nonanalytic learning group in Brook’s (1978) experiment were asked to sort a list of new letter strings into three categories: grammar A (old world), grammar B (new world), or neither. Although the subjects who received nonanalytic training protested that they had no knowledge of the two grammatical categories, they performed significantly better than the analytic group (60%correct as opposed to 47% correct, while 33% correct would be expected by chance alone). The performance of the nonanalytic group was remarkably good, not only in relation to the analytic group (which did, after all, have a very difficult concept learning task to contend with), but in absolute terms as well. How did these subjects do so well? According to Brooks’ model, subjects classified new strings according to their overall similarity to one or more of the memorized strings. It is a surprising mathematical consequence of exemplar-based nonanalytic representations that the accuracy of classification increases with the number of irrelevant aspects possessed by the exemplars. To see why this is so, consider what the subject does when confronted with a new item to classify: He or she first searches memory for a stored exemplar that closely resembles the item in question. If one can be found, then the subject classifies the new item in the same category or concept as the similar exemplar in memory. What are the chances that this will be the

Concept Development

149

correct classification? It will be correct when the new item and the exemplar in memory are similar in those aspects that are relevant to concept membership, and incorrect when the two differ in the relevant aspects. Assuming that an exemplar can be found in memory that differs in only one aspect from the item to be classified, then the chances of that aspect being the relevant one (and the exemplar and the item thus belonging in different categories) get smaller as the number of irrelevant aspects gets larger. Brooks has, in fact, found a good correspondence between the accuracy of classification predicted by this account and that obtained in experiments. We should at this point mention the possibility of nonanalytic summary representations of concepts. Although the analytic-nonanalytic distinction is in principle independent of the degree of summarization in a representation, nonanalytic summary representations do not seem viable, because they would lead to frequent misclassifications. The exemplar search process described above results in classifications being made on the basis of irrelevant aspects of the stored exemplar representations by chance some fairly small fraction of the time, but the irrelevant aspects of a summary representation will enter into the classification comparison every time. Thus, if the child began with a nonanalytic exemplarbased representation and began to form a summary representation, there would be pressure to trim away the irrelevant information; what was actually an advantage before would now become a disadvantage.

2 . Determinants of Nonanalytic Concept Formation Nonanalytic concept formation is a surprisingly effective way of coping with complex concepts when analysis is impossible. Brooks (1978) suggested several factors that may determine when nonanalytic concept representations are used. First, the complexity of the concept being learned will determine how likely it is to be learned analytically. The artificial grammars used by Brooks in the experiment just described were very complex and difficult to analyze properly because the stimulus attributes that determined concept membership were complex combinations of features. If such concepts must be used before their analytic basis can be fathomed, a nonanalytic strategy is the subject’s only choice. Another factor that may influence the choice between analytic and nonanalytic concept formation strategies, leading to analytic or nonanalytic concept representations, is the amount of time and cognitive resources available to carry out analyses during learning. Limited resources may be better put to work attempting to memorize both the relevant and irrelevant aspects of the exemplars rather than trying to distinguish between them. Finally, a person may be biased toward certain analyses, and this bias may also influence whether the concept is appropriately analyzed or remains nonanalytic. For instance, knowledge of the general kind of concept, such as an animal, or an artistic style, may guide a person to the relevant aspects of a particular concept (for an animal, not brightness levels but

I50

Martha J . F a r a h a n d Stephen M . Kosslyn

number of legs). People are also predisposed to notice certain aspects of stimuli, for reasons ranging from psychophysical properties of the encoding system (relating to thresholds of perception and noticeable differences) to the stage of development (e.g., the shift from color to form dominance, see Kosslyn, Heldmeyer, & Glass, 1980). 3. Evidence Thut Children Use Exemplar-Based Representations Learning an exemplar-based concept is simplicity itself: All one need do is memorize exemplars. This learning process is clearly within the bounds of the young child’s competence. Further, the fact that nonanalytic concepts can be formed effectively irrespective of concept complexity, limited cognitive resources, and novelty of concept domain all suggest that such concepts may play a role in the thought processes of young children. We know that children are less adept at analysis of complex concepts than adults, and also that their attentional resources and/or effective working memory space is more severely limited (see Case, 1974; Chi, 1976; Pascual-Leone and Smith, 1969; Pascual-Leone, 1970). In addition, they lack the familiarity with many kinds of concepts, and the rnetacognitive wherewithal to use that familiarity, to help them to analyze new concepts properly. Probably because exemplar models have only recently received much attention, developmental research on them is scarce. The literature on discrimination transfer may be relevant to the question of whether children are more predisposed to forming exemplar-based concepts than adults. In these experiments subjects are taught that two stimuli are “positive” and two are “negative,” with the stimuli varying on two dimensions (such as size and color). Subjects are then transferred to new stimuli in which ( I ) the positive and negative values are switched, but the relevant dimension (e.g., size) is preserved, or (2) the values are switched and so is the dimension. In the second case, when the dimension is switched, one of the two previously positive instances remains positive; in the first case, when the dimension is kept the same, neither of the initial positive instances remains positive. Kendler and Kendler (1959), Tighe and Tighe (19651, and others have found that 4-year-olds learn better in this situation when the dimension is switched than when values are simply reversed on the same dimension. This finding implies that the original stimulus-response pairing for the younger children was between individual stimuli and responses, rather than between a category level description of stimuli and responses. Other children and adults, in contrast, learn better when the stimulus-response pairs are reversed but the relevant dimension is preserved. This finding suggests that these people abstract the dimension, and do not simply memorize exemplars of the different stimulus classes (see White, 1965). Kossan (1981) reported a developmental study in which she examined more directly whether young children tend to utilize nonanalytic exemplar-based con-

cepts. Like Brooks (1978), she used a paired-associate learning paradigm to encourage the encoding of both relevant and irrelevant features of her artificial concept exemplars (imaginary animals). The analytic group learned the animal concepts in the traditional reception paradigm, in which subjects are shown concept instances to classify and are given feedback after each classification until they are able to perform the classifications to some criterion of accuracy. This training procedure encouraged analysis for the relevant similarities among the concept instances. Seven- and 10-year-olds learned one of three different kinds of concepts by one of these two training procedures. The three kinds of concepts were: (1) a rule-defined concept possessing necessary and sufficient features; (2) a “natural” concept possessing a Wittgensteinian network of sufficient features; and (3) another kind of “natural” concept possessing a network of sufficient features but also some irrelevant features (features that occur only once in one instance of the concept, and thus are not characteristic of the concept). Kossan (1981) found that for both age groups, the reception paradigm led to better transfer performance on concepts defined by necessary and sufficient features, and the paired-associate paradigm led to better transfer performance on concepts whose exemplars possessed unique features as well as sufficient features. She also noted some transfer to new test items on the basis of the unique and therefore irrelevant features in the paired-associate groups, suggesting that these children did not distinguish between relevant and irrelevant features and thus had formed a nonanalytic concept. Of developmental interest, however, is her finding that for concepts based on a “natural” network of sufficient features only, the younger subjects performed better on the classification task following paired-associate training, which fostered exemplar encoding of a nonanalytic concept, than following the reception paradigm training. In contrast, the older children benefited more from the reception paradigm, which fostered analysis for the sufficient features of the concept. The results of Brooks’ (1978) and Kossan’s (1981) experiments are intriguing, and raise the following question: To what extent do children and adults spontaneously choose nonanalytic strategies when learning real world concepts outside the psychology laboratory? The paired-associate paradigm may not force subjects to form nonanalytic concepts, but it does discourage analytic encoding by placing equal importance on remembering relevant and irrelevant features of the concept instances, as well as diverting resources from the task of analysis for the purpose of memorizing the exemplars. The Medin and Schaffer findings demonstrate that exemplar-based concepts will sometimes be formed without such task pressures, but their concepts and learning situation were still highly artificial. Certain kinds of overgeneralizations in children’s word usage may hint at the spontaneous use of nonanalytic concepts by young children. Children will occasionally overextend a term to an object that is near or touching a true instance of the concept denoted by the term. Anglin (1977) reported that children in one of

152

Martha J . Farah and Stephen M . Kosslyn

his obligatory naming experiments frequently called a vase a “plant.” This misclassification might be analogous to the nomad’s calling an empty frame an example of a certain painting style. However, a child might misuse a word for many possible reasons; therefore, caution is needed in interpreting overgeneralizations as evidence for a kind of concept representation (see Clark, 1973). Perhaps the most interesting questions raised by work on nonanalytic concept formation are about the developmental relationship between nonanalytic and analytic concepts. Kossan (198 1 ) found that young children do indeed perform better using a nonanalytic strategy than an analytic strategy. Eventually, however, the majority of their concepts become analytic. What is the mechanism of this transition? Are nonanalytic concepts simply an interim solution until the ability to form concepts analytically develops, or do the nonanalytic concepts play a functional role in the later development of analytic concepts? If the latter is true, then why? Is it simply a matter of greater familiarity with concept instances brought about by use of the nonanalytic concept? Or does the possession of a nonanalytic concept allow some incubational process to go on, in which the learner gradually discovers the analytic basis for the concept? D. THE PROBLEM OF MULTIPLE INTERPRETATIONS OF DATA

It should be clear at this point that the different classes of models often can account equally well for the same data. In particular, Medin and Schaffer’s (1978) exemplar-based model can account for the large variety of findings originally taken to support prototype models. Recall, for example, the experiment in which Rosch (1975) found that priming a two-word matching task with the name of the superordinate category of the words facilitated the match for typical instances and delayed the match for peripheral instances. This surprising pattern of facilitation and delay was originally believed to be a unique prediction of the prototype model. However, an exemplar-based model with the proper processing assumptions will do just as well. Suppose that a set of exemplars of the superordinate category is activated (brought into short-term memory) when the category is named, and that the priming of an instance is proportional to the number of similar exemplars that are activated. If so, then the encoding of a typical instance will be facilitated because more of the exemplars in short-term memory will be similar to it, while the encoding of an atypical instance will be delayed because the exemplars in short-term memory will be unlikely to be similar to it, and will simply distract the subject. Thus, this model and the prototype model predict the same pattern of facilitation for typical instances and delay for peripheral instances. The way in which exemplar-based models account for the other main findings that were considered to support the existence of prototypes-typicality effects in learning and classification, and the longer retention of prototypic stimuli after learningwas discussed earlier in our review of Medin and Schaffer’s (1978) model. We should also note that the uncertainty is not confined to a choice between

Concept Development

153

exemplars and prototypes. A cue set model can also account for the priming results as well as all the other results that support prototype models. Rather than give all of the details of the appropriate cue set model here, we offer the following generalization, from which motivated readers may infer the details: All of the data originally thought to support the reality of prototypes really just support the reality of typicality. They show that the typicality of concept instances affects the way in which they are processed. And although this is a prediction of prototype models, because more typical instances are more similar to the prototype, it is also a prediction of cue set models, because more typical instances will have more (and more heavily weighted) relevant cues. This prediction also follows from exemplar models, because more typical instances will be similar to more of the stored exemplars. More powerful experiments can be designed to obtain evidence for particular models, rather than simply to reaffirm that typicality is a psychologically relevant property of concept instances. The experiments of Medin and Schaffer (1 978) are one example of a successful approach to this problem. The Hayes-Roth and HayesRoth (1977) experiment on cue set representations is another example of the kind of quantitative methodology needed to distinguish among the highly similar predictions of competing models in this area. Thus, it seems that the state of affairs in this particular area reverses the usual criticism of psychology as a science; rather than a dearth of well-formulated questions and an overabundance of data, we have a set of questions, even paradigms for research, but almost no data! Medin and Schaffer’s (1978) results indicated that under the learning and classification conditions of their experiment, subjects formed exemplar-based representations of concepts. However, it would be hasty to conclude that the many other concept formation experiments that have demonstrated typicality effects also must have involved only exemplar-based representations. Psychologists should expect to find flexibility in the conceptual systems of their subjects. The constantly varying demands on this conceptual system would be well served by an ability to encode and recode concepts as exemplars, prototypes, and cue sets. There is little doubt that each of these competences exists; we can mentally average concept instances and judge the similarity of other items to that average; we can keep track of frequently occurring aspects of concept instances and judge how many of those aspects are represented in a particular item; and as Brooks (1978) and Medin and Schaffer (1978) have shown, we can simply remember the individual exemplars and judge the similarity of a new item to those exemplars.

IV. The Format of Concept Representations The format of a concept representation is sometimes called the “mode” of representation in developmental psychology. We prefer the term “format” because it makes clear the important distinction between the format (the formal

I54

Martha J . Farah and Srepherr M . Kosslyn

structure of the code) and the content (the information stored) of a representation, as described earlier. The format of concept representations has not yet been dealt with seriously in psychology. We have found a dearth of experimental evidence bearing on this issue. When representational formats have been considered at all by psychologists working in this area, the formats have been only vaguely specified, As cognitive psychologists committed to building information-processing models of conceptual development, we have several reasons for believing that the formats of concepts must be studied. First, the more aspects of a representation that we specify, the more specific our empirical predictions will be, and thus the more confirmatory power we can derive from our experiments. For instance, if we know that a certain concept is represented in an unstructured list format, we can make use of the findings of a large literature on memory scanning in formulating specific experimental predictions. We could thus discover whether a concept has been analytically represented (by determining whether irrelevant features of the training instances are being used in comparisons during a classification task) and whether multiple exemplars are included in the representation (by investigating whether having subjects learn more exemplars initially then leads to longer comparison times). Another reason for studying concept representation at the level of format is that we may discover that different kinds of concepts are represented in qualitatively different ways. For example, some concepts may be represented in visual images, others in linguistic strings, and still others in other formats, such as motoric images (as Piaget suggested; see Flavell, 1963). So far, the prevailing assumption in this field has been, to borrow from Gertrude Stein, that a concept is a concept is a concept. Results from experiments on concepts as diverse as random dot pattern categories, natural semantic domains, and artificial grammars are compared and often used interchangeably as evidence for particular models of concept formation and structure. Perhaps one reason we know so little about the way that concepts are formed and represented is that in our analyses of these processes we have not differentiated enough between different kind of concepts. Often in science, correctly grouping the phenomena to be explained is half the battle. The investigation of concept formats promises to be especially revealing in the study of concept development over age. The representational-development hypothesis, put forth in different forms by Piaget, Bruner, and Werner, is essentially a claim about the format of mental representation in children (see Kosslyn, 1978). Piaget and Bruner in particular claimed that many of the phenomena of cognitive development reflect a shift in the representational format, from imagistic representation in young children to symbolic or descriptive thought in older children (usually at around age 7). We shall take a “backward extrapolation” approach to these issues, in which we view the adult as the end product of cognitive development and try to formulate how the adult state was achieved.

Conrepr Dei’elopment

I55

Thus, we shall return to the developmental issues after discussing what is known about the format of concept representations in adults. A.

IMAGES AND DESCRIPTIONS

Many formats for the psychological representation of concepts are possible; in the literature we find mention of feature lists (Rosch & Mervis, 1975), dimensions (Reed, 1972), structural descriptions (Winston, 1975), and images (Posner & Keele, 1968; Reed, 1972). Perhaps the most important distinction for current purposes is between images and descriptions. Although feature lists, dimension values, and structural descriptions do differ from one another in ways that we shall discuss, a far more fundamental, qualitative difference exists between these descriptive formats and the imagery format: Images depict; descriptions describe. In order to depict, the parts of an image must correspond to parts of the depicted object such that the interpart distances are preserved in the representation. An image of a square, for example, includes parts that corespond to the four sides, and these parts are connected at junctions corresponding to the four corners. In contrast, the elements of a description need not correspond in a direct way to the thing described. A description of a square includes terms like “four,” “sided,” and “polygon,” which do not correspond to parts of the square (see Kosslyn, 1980, Chap. 3, for a more detailed development of this distinction). I . Images as Concept Representations In the eighteenth century the philosopher Berkeley pointed out a limitation of images as concept representations: anything picture-like is too concrete to represent a concept, which is, of course, a class. As an example, he asked how one could represent the concept triangle with an image. Any triangle imaginable is a specific kind of triangle-isosceles, equilateral, or obtuse. Therefore, an imagined triangle could only represent a particular kind of triangle. Berkeley argued that images must by nature represent all properties of an object. They cannot represent the triangularity of a shape without also representing its acuteness, not to mention its color, size, and orientation (see also Kosslyn, 1980, Chap. 1 I ; Reese, 1977). But this is only partly true. Images are representations that occur in a particular processing system, one component of which is an interpreter (a “mind’s eye,” if you will) that can selectively attend to particular properties of images. Thus, an image of a particular triangle may represent triangularity to us if it is processed in a manner that takes the three straight sides and comers as essential, and the particular internal angles, color, and so on as irrelevant. “Triangle” is an example of one of those rare concepts that has an exact definition, and seems a good candidate for an explicitly verbal, rule-based representation. As such, it does not do a very good job of illustrating concept repre-

Martha J . Farah and Stephen M . Kosslyn

156

sentation in an imagery format, A more natural example is a concept that is “perceptual, by which we mean that the relevant similarities among concept instances are a result of similarities in depictable aspects of the instances (such as their size, shape, etc.). Posner and Keele (1968) and Reed (1972) have considered the possibility that images are involved in the representation of some concepts. However, their findings on this question have been based solely on subjects’ introspections when queried after the experiments. Posner and Keele asked 12 of their pilot subjects for introspections on the extent to which they had used images and descriptions of the dot pattern stimuli, and found that most subjects reported using descriptions, Reed gave all of his subjects a multiple-choice questionnaire on the strategy they used to learn schematic face categories, but he confounded choices concerning the format of the representation used (image versus description) with choices concerning the content of their representation of the concept (prototype versus exemplars versus cue set). The most common choice was the one involving an “image of the prototype,” but because this was the only choice involving either images or prototypes, we cannot know how many subjects chose this response because it was the only one mentioning a prototype even though they did not use imagery. Farah (1980) found evidence for the use of imagery by subjects performing a categorization task after concept learning when the concepts were random dot pattern categories (similar to Posner’s), but not when the concepts were letter strings (a presumably feature-based concept). The stimuli to be categorized were presented at different orientations, and were preceded on each trial by an orientation cue. The amount of time subjects took to prepare for the stimulus item after seeing the orientation cue was recorded. If images were being used, Farah expected that they would be rotated into position upon seeing the orientation cue, but that no such process would occur if features were used. For dot pattern concepts, the time to prepare for a stimulus increased linearly for orientation cues indicating increasing departures from the standard upright. For the letter-string categories, in contrast, times did not increase. These results are consistent with the notion that the dot pattern concepts were indeed represented in images, which were “mentally rotated” (see Cooper & Shepard, 1973) to the orientation indicated by the preparation cue. The subjects’ postexperimental introspections supported this interpretation of the data. ”

2 . Varieties of Descriptive Formats The notion of description is far less constrained than the notion of depiction. To say that images depict an object is to say that they preserve some of the structural aspects of the stimulus (see Kosslyn, 1980, Chap. 3, for a more formal treatment). This is an additional and very strong constraint on the conditions that must hold between a depictive representation and the object it represents. The variety of forms available to descriptive models is apparent even within the concept

Concepr Development

157

formation literature. Dimensional and featural descriptions are the most commonly encountered descriptive formats in this area. Dimensions are often used to describe stimuli and concepts. A stimulus may be described by its values or “levels” on a set of dimensions, as Reed’s (1972) schematic faces were. A concept can be described in terms of values, or ranges of values, on the relevant dimensions of the stimuli possessed by a prototype, a set of exemplars, a set of cues, or a definition. A dimension is an attribute of a stimulus which, if it exists for that stimulus, has some positive value and never more than one (see Gamer, 1978). For instance, if color is a dimension for a stimulus, then the stimulus must have some color, and being one color precludes being any other color. Nose length is a dimension for Reed’s faces, each face having one nose length and not more than one. Dimensions need not be continuous in Garner’s scheme; the difference between dimensions and features is not simply a difference in continuity or discreteness. For example, Reed’s faces had discrete nose lengths. Rather, discrete and continuous dimensions are subclasses of dimensions, both of which differ from features in the mutual exclusivity of positive values on dimensions. Featural descriptions simply list the features of a stimulus, and a concept can be described in terms of relevant features belonging to a prototype, exemplars, a cue validity list, or a definition. A feature is an attribute of a stimulus which is either present or not. If it is present, it does not have levels of values; hence the mutual exclusivity of levels is not even relevant. Garner (1978) calls features “dissociable elements” because they can be added to or subtracted from a stimulus without changing other aspects of the stimulus. In contrast, if a dimension, such as color, is completely eliminated from a stimulus, that stimulus ceases to exist: One cannot have a shape, size, orientation, location, or texture without also having some color. Two confusions may arise concerning dimensional and featural formats. The first is over the lack of a consistent usage of the two terms. Brooks (1978) analyzed his concepts and stimuli in terms of “features,” but then discussed the analytic-nonanalytic distinction in terms of relevant and irrelevant “dimensions. ” The second possible confusion about these descriptive formats concerns their psychological reality. The fact that a concept is designed by an experimenter using certain dimensions or features does not imply that the actual psychological representation of that concept will contain those, or for that matter any, dimensions or features. We must be careful not to assume, for example, that the feature lists which Rosch (1973) obtained from subjects, and from which she was able to predict concept prototypes, constitute the actual psychological representation of those concepts for her subjects. Rosch was well aware that the format of her feature lists was entirely artifactual; she asked her subjects to list the features of concept instances. In closing this section, a third class of descriptive formats, known as “struc-

158

Mortha J . Faroh ond Stephen M . Kosslyn

tural descriptions, ” is worth mention. These formats have been developed mostly in the field of artificial intelligence, although some researchers in the more computer-influenced areas of psychology have modeled memory representations using structural descriptions (e.g., see Norman & Rumelhart, 1975). Although there has been relatively little empirical confirmation of their psychological validity as a representational format, their udequucy as a format for representing knowledge has been carefully evaluated and seems superior to the more primitive descriptive formats discussed so far. A feature list, for example, does not represent people’s knowledge of most objects or concepts very well. Both feature lists and dimensional schemes leave out information about the relationships between different aspects of their referents. People know more than the fact that faces have eyes, a nose, and a mouth; they also know that the eyes are over the nose, which is over the mouth, and that the features are arranged symmetrically, and so on. Structural descriptions were specifically designed to capture this sort of information. They specify features and various kinds of relationships between them, for instance, hierarchical, spatial, and functional. These kinds of formats have been used by computer scientists programming machines to learn categories (see Boden, 1977; Winston, 1975). B. THE REPRESENTATIONAL-DEVELOPMENT HYPOTHESIS

The representational-development hypothesis is one of the more sweeping claims that have been made about development. The notion that the predominant mode of representation changes over age is intended as a general explanatory principle, a rare bird in the relatively phenomenon-oriented field of psychology. Before turning to evidence for one case of this principle, the shift from an imagery format to a descriptive one, let us ask why we should expect it to be true. Descriptive formats can be used to represent more kinds of information than can images (notably, abstract classes), and thus may be necessary for successful performance on a variety of cognitive tasks. However, they may also be more demanding on the processes that encode (and, possibly, use) the information that they represent. That is, the encoding of a stimulus can be thought of as taking place in stages of increasing abstractness, beginning with perceptual processes which develop a structured, interpreted perceptual representation from sensory data, and then continuing with higher cognitive processes, which may convert the perceptual representation into an abstract symbolic code or description. Although most of the early perceptual processing appears to be “automatic” (see Schiffrin & Schneider, 1977), continued processing is costly in terms of cognitive resources such as attention and short-term memory space. Children, who seem to have less of these resources than adults (or at least are less efficient in their use of these resources than adults), may therefore end the encoding process at the stage of perceptual representations and store information in the image

format that is presumably used at that stage. If so, then it makes sense that young children will often store information in an image rather than a description. The imagery format has other advantages for children who are learning concepts. We have already noted that images carry a lot of implicit information about their referents. This property of images would be useful for the concept learner who is completely naive about the concept domain, a situation in which young concept learners must often find themselves. Recall the museum example in the introduction. If you were setting out to learn a new artistic style, in what format would you encode the examples? If you knew something about artistic styles, you might simply remember facts about their brush strokes, compositions, colors, etc. Art critics and art historians, who have a comprehensive vocabulary for describing such style-relevant properties of paintings, might be able to learn the concept of a new artistic style in this way. However, if you were unfamiliar with artistic style in general, you would not be able to use a descriptive format efficiently for learning the style at hand. To store only a description of something is to commit yourself to selectively encoding some pieces of information about it and losing the rest. A more sensible strategy would be to form images of the art works, thus retaining more information about them in implicit form. The images could then be “mentally examined” together afterward in an attempt to “see” similarities among the instances of the concept. If you could still not discern a unifying similarity or a set of relevant attributes (features or dimensions), the images could then be used in global similarity matches as a nonanalytic concept. Although the additional cognitive processing and greater “commitment” required by descriptive formats are good reasons in principle to expect children’s concepts to be more imagistic than those of adults, the evidence gathered for this position is not compelling. Attempts at empirical confirmation of the shift from predominantly image representation to descriptive representation have generally simply missed the point, confusing content with format. Observations of young children’s vocabulary, episodic memories, sorting behavior, and so on suggest a tendency to attend to the perceptual properties of stimuli, things which, to paraphrase Bruner, “can be pointed at” (Bruner, Oliver, & Greenfield, 1966, p. 22). These observations tell us about the content of children’s thoughts, not their format. There is no reason why these perceptual qualities of stimuli cannot be encoded in a descriptive format (see Kosslyn, 1978). Kosslyn (1976) found evidence that children rely on imagery more than adults do when answering questions about animal concepts. He asked 6- and 9-year-old children and adults to verify properties of animals, such as “claws” for a cat. Kosslyn considered two different ways that subjects could make these judgments. If the information about the body part of the animal is explicitly described in long-term semantic memory, then accessing that descriptive representation will probably be the most efficient way of answering. In addition, it should lead to subjects ’ responding more quickly to questions concerning highly associated

160

Martha J . Farah and Stephen M . Kosslyn

parts (e.g., “claws” for a cat) than to weakly associated parts (e.g., “head” for a cat). This is a robust finding in semantic memory experiments (see Smith, Shoben, & Rips, 1974). Alternatively, a subject could answer these questions by forming an image of the animal and inspecting it for the presence of the queried part. Past experiments have shown that the size of an imaged part affects how long it takes to be “seen” on an image. Thus, if subjects are accessing the information by consulting an image, large parts (e.g., “head”) should be verified more quickly than small parts (e.g., “claws”). The items used in this experiment were all chosen to pit size against association strength: The “true” properties were either small and highly associated parts or large and weakly associated parts, allowing one to discover whether images or ordered descriptions of parts were being accessed. The subjects all began by evaluating one set of items without being given any specific instructions about imagery. They were given another set of animalproperty pairs and asked to evaluate them by forming images and “looking” for the queried parts. Response times to answer the questions were recorded. The data from the properties evaluated when imagery instructions were given replicated earlier findings, with both children and adults verifying large, weakly associated parts faster. The data from the first condition, which did not include imagery instructions, showed an interesting developmental trend. Adults ’ and 9-year-olds’ response times were generally faster in this block and were especially fast for the small, highly associated parts (indicating a semantic memory retrieval process), but the 6-year-olds’ response times were similar in the two instructional conditions, suggesting that they relied on imagery even when not instructed to do so. The results were not conclusive, however: One can always argue that the youngest children were accessing feature lists in both imagery and no-imagery instruction conditions, with properties in the lists being ordered in terms of size. Kosslyn (1978) reviewed the literature on the purported shift from image representation to descriptive representations and found no conclusive data. Once again, then, we are left with more questions than answers. But also once again, we note that the methodologies are available for providing the answers (see Kosslyn, 1980).

V. The Organization of Concept Representations The organization of a concept representation refers to the structural relations that exist among its parts. For example, a list of features has an order, even if it is random. That order is the organization of the feature list. Dimensional representations also possess some order in which the dimensions and their respective values are specified. A structural description has a more complex organization, because its parts can be combined in ways other than simple order or adjacency.

Concept Development

161

To specify the organization of a structural description we need to specify the particular relations among the features. Imagistic representations have organizations to the extent that they are analyzed into parts. Images may be parsed into units and organized in accordance with the gestalt laws of organization. Further, parts of imaged scenes will be organized spatially, perhaps in three dimensions. The kinds of organizations that are possible for a concept representation depend not only on the format of the representation, but also on its content. For example, we will ask different questions about the organization of a set of exemplars (e.g., are they ordered by typicality?) than about the organization of a prototype or cue set representation (e.g., are the parts ordered by diagnosticity?). Because the possible organizations for concept representations cannot be specified without knowledge of the format and content of the representation, questions about concept organization are logically secondary; we need answers to the questions raised in the preceding sections before specific models of concept organization can be formulated and tested. At the present writing, it seems that virtually no work has been done on the organization of children’s concept representations. The work by Collins and Quillian (1969, 1972) and others (see Smith et al., 1974) could be regarded as aimed at the specification of the organization of structural descriptions of concepts in adults, as could much of the work on “semantic memory” (see Glass & Holyoak, 1973, but space limitations prevent us from reviewing this literature here. Suffice it to say that virtually no issues have been settled pertaining to how concept representations are organized.

VI.

Conclusions

In this article we could only sample the range of research that has been carried out on concept formation in adults and children. Even so, our review contains many different models of the process of concept formation, and the number of models multiplies when one considers combinations of possible formats, contents, and organizations, each of which was discussed independently in this article. Further, there is no reason not to expect hybrid concept representations. For example, a concept such as “Van Gogh’s style of painting” might be represented by a collection of exemplars, some in imagery format, some in a descriptive format, as well as by a set of descriptive cues, and an image of some prototypical Van Gogh painting that comes to mind first when his painting style is mentioned. Such multiple, redundant concept representations seem almost inevitable, given the flexibility of our conceptual systems. In a sense, then, it may be misleading to talk about the process of concept formation, or the representation of concepts in the mind. As we argued earlier, all of the models described in this article are clearly within the range of our

I62

Martha J . Farah and Stephen M . Kosslyn

cognitive competence. We should not be surprised to find that “concept learning” is actually an intricate set of processes, some candidates for which have been presented in this article. The question for cognitive and developmental psychologists is now: Which of the models discussed here and elsewhere describe the processes of concept learning at which stages of cognitive development, for which concepts, and under which circumstances? This is an empirical question, of course, but some principled predictions can be made. These predictions are derived by considering the a priori processing characteristics of the various alternatives. All require some encoding of information into memory, some analysis during or after that encoding, and some mental operations on the resultant representations. However, the amount of effort expended on each kind of process differs among the different representations. By considering the patterns of effort, we can formulate predictions about which representation will be most effective for which individual in which learning situation or classification task. For example, exemplar-based representations require the most information to be encoded into memory, but the analysis of the exemplars may be minimal (only that which occurs as a result of perceptual encoding) and the only necessary mental operation on the representations is the association of the concept label with each exemplar. These properties lead us to predict that exemplar representations will be used when processing capacity is in short supply, such as when one learns concepts under conditions of distraction or fatigue. For the same reasons, we would expect young children, who have less capacity to perform the kinds of cognitive processes that are used in forming a summary representation, to use exemplar representations. In addition, an exemplar-based representation of a concept should be likely when long-term memory representations of the instances already exist. Cue set representations are relatively economical in terms of memory load, but require more analysis than exemplar representations because the instances must be parsed into cues during learning. This kind of concept formation also requires the mental operation of keeping track of the frequency of occurrence of different cues in the exemplars and nonexemplars of the concept, so that only the most diagnostic cues will be included in the cue set. The cue set learner has the option of encoding structural relationships among the cues, and to the degree that such relational information is included in the cue set representation even more analysis of the exemplars will be needed. Note that a cue set representation may start out unstructured, and the relational information may be added at a later time. Prototype representations are also sparing of memory. The amount of analysis and further processing that is required depends upon whether the prototype is simply a wholistic average of the exemplars seen, or a weighted prototype. Wholistic averages require only the minimal, automatic analysis of encoding the exemplars, and an averaging operation akin to superimposition. As was noted earlier, most concept exemplars will not yield an intelligible average in this way.

I

Concept Dtveloptncnt

163

A weighted prototype, however, requires analysis of the exemplars into aspects whose frequency within the concept and outside it must be counted. In addition, if the weighted prototype is formed rather than selected, then analysis of the structure of the instances is also required, because the prototype must conform to this structure itself. Under circumstances in which the structure of the exemplars is known or the variation among them is small enough to allow an intelligible average, a prototype could be formed. The learner’s tendency to use a particular concept representation might depend not only on the cognitive demands and the specific learning situation, but also on the anticipated use of the concept. Exemplar representations allow the most leeway for future uses of the concept, because they are the closest to the “raw” exemplars, allowing future recoding into cue sets or prototypes. However, although they are less flexible, summary representations may be more efficient for use in classification and information retrieval. We could go on speculating about the conditions under which different concept representations will be used, but that would just obscure the point that we are trying to make. We are advocating not a retreat from the laboratory to the armchair, but merely some analysis before using the laboratory. Gedanken experiments are a useful exercise before designing real experiments in this domain because, as we have tried to show in the last few paragraphs, many interacting factors could determine which concept representations are used in a given situation. What are the broader implications of our discussion for research on the development of concepts over age? First, we believe that it points out the absolute necessity of a systematic approach. With so many possible models, we shall get nowhere by performing experiments in which two particular models are contrasted because it is likely that neither will be correct. However, we cannot simultaneously contrast all possible models in all relevant circumstances. The most efficient way to advance our knowledge in this domain may be to take advantage of its analyzability into issues of content, format, and organization, and perform experiments designed to distinguish between possible positions on each issue. For example, we can first determine that a particular concept is analytic or nonanalytic for a child of a certain age (perhaps by examining whether subjects’ patterns of classification distinguish between the relevant and irrelevant aspects of the stimuli), then determine whether it involves imagery or some sort of description (perhaps by the mental rotation and short-term memory scanning paradigms mentioned earlier), and then determine the degree and form of summarization used (again, by examining patterns of classification, as Medin and Schaffer, 1978, did). This approach has the additional advantage that, as each component of the model is determined, it gives us more specificity and thus more predictive power for subsequent experiments. For example, let us say that we have determined that a particular concept involves the imagery system, and

I64

Martha J . Farah and Stephen M . Kosslyn

we now want to address the issue of the summarization of that concept. We can now make use of our knowledge of the imagery system to design the experiment about summarization. If several images take more time to rotate than does one (a reasonable and, more importantly, testable assumption) then by varying the number of exemplars in the learning set we can predict increased preparation time in the image rotation paradigm described earlier if the concept is represented as exemplars. A second broad implication of our analysis is that we should be prepared to find flexibility: Different concept representations and concept formation processes may be used by subjects at different developmental stages and/or in different learning situations. Too often psychologists set out to study the way that a task is performed, and miss one of the most interesting and general aspects of human cognitive performance: that there is more than one way to skin a cat. Once we accept this flexibility as a significant characteristic of the way humans think and learn, rather than a troublesome source of variation in our data, it becomes important to understand the factors that control the adoption of one strategy over others. This perspective is motivated by the broad range of representations that all seem within our capacities to use, and by the fact that each of these representations has unique processing characteristics that will be optimal under different circumstances at different ages. Thus, we see a program of developmental research proceeding hand in hand with the study of adult concept formation, with the end result being a set of empirically confirmed models of the concept representation and formation in particular domains, and a set of principles by which the models are related to the stage of development of the learner and his or her learning situation.

REFERENCES Anderson, J . R . Arguments concerning representations for mental imagery. Psychological Review, 1978, 85, 249-277. Anglin, J . M . Word, object, und ronceptuul development. New York: Norton, 1977. Auk, R. L. Children's cognitive development: Piager's theory and the process approach. London and New York: Oxford Univ. Press, 1977. Boden, M . Arfificiul intelligence and natural man. New York: Basic Books, 1977. Bourne, L. E. Humun conceptual behavior. Boston: Allyn & Bacon, 1966. Brooks, L. Non-analytic concept formation and memory for instances. In E. Rosch & B . C. Lloyd (Eds.), Cognition und categorizarion. Hillsdale, N.J.: Erlbaum, 1978. Bruner, J . S . , Goodnow, J . , & Austin, G . A study of thinking. New York: Wiley, 1956. Bruner, J . S., Olver, R . . & Greenfield, P. Studies in cognitive growth. New York: Wiley, 1966. Carey, S . , & Smith, C. The child's concept of animal. Paper presented at the meeting of the Psychonomics Society, San Antonio, Texas, November 1978. Case, R. Structure and strictures: Some functional limitations on the course of cognitive growth. Cognitive Psychology. 1974, 6 , 544-573.

Concept Development

I65

Chi, M. T. H. Short-term memory limitations in children: Capacity or processing deficits? Memory & Cognition, 1976, 4, 559-572. Chomsky, N. Aspects of the theoty of syntax. Cambridge, Mass.: MIT Press, 1965. Clark, E. V. 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 (if languuge. New York: Academic Press, 1973. Collins, A . M., & Quillian, M. R. Retrieval time from semantic memory. Journal qf Yerhul Learning (2nd Verbul Behuvior, 1969, 8, 240-248. Collins. A. M., & Quillian, M. R. How to make a language user. In M. E. Tulving & W . Donaldson (Eds.), Organization of memory. New York: Academic Press, 1972. Cooper, L. A., & Shepard, R. N. Chronometric studies of the rotation of mental images. In. W. G. Chase (Ed.), Visual information processing. New York: Academic Press, 1973. Farah, M. J . Imagery and visual concepts. Unpublished manuscript, Harvard University, 1980. Flavell, J. H. The developmental psychology of Jean Piaget. Princeton, N.J.: Van Nostrand, 1963. Fodor, J. A. The Language of thought. New York: Crowell, 1975. Fodor, J . A. The mind-body problem. Scientific American. 1981, 114-123. Franks, J . J., & Bransford, J. D. Abstraction of visual patterns. Journal ofExperimenta1 Psychology, 1971, 90, 65-74. Gamer, W. R. Aspects of a stimulus. In 8 . Lloyd & E. Rosch (Eds.), Cognition and cutegarization. Hillsdale, N.J.: Erlbaum, 1978. Garner, W. R., Hake, H. W., & Eriksen, C. W. Operationism and the concept of perception. Psychological Review, 1956, 63, 149-159. Glass, A. L., and Holyoak, K. J . Alternative conceptions of semantic memory. Cognition. 1975.3, 313-339. Goodman, N. Languuges of art: A n approach to ( I theory of symbols. New York: Bobbs-Merrill, 1968, Hayes-Roth, B., & Hayes-Roth, F. Concept learning and the recognition and classification of exemplars. Journal of Verbal Learning und Verbal Behavior, 1977, 16, 321 -338. Hayes-Roth, F. Distinguishing theories of representation: A critique of Anderson’s “Arguments concerning mental imagery.” Psychological Review, 1979, 86, 376-382. Kendler, T. S., & Kendler, H. H. Reversal and nonreversal shifts in kindergarten children. Journal of Experimental Psychology, 1959, 58, 56-60. Kossan, N. Developmental differences in concept acquisition strategies. Child Development. 198 I , 52, 290-298. Kosslyn, S. M. Using images to retrieve semantic information: A developmental study. Child Development. 1976, 17, 434-444. Kosslyn, S . M. The representational development hypothesis. In P. Ornstein (Ed.), Memory, developmenr in children. Hillsdale, N.J.: Erlbaum, 1978. Kosslyn, S. M. lmage and mind. Cambridge, Mass.: Harvard Univ. Press, 1980. Kosslyn, S . M., Heldmeyer, K. H., & Glass, A . L. Where does one part end and another begin? A developmental study. In J. Becker, F. Wilkening, & T. Trabasso (Eds.), Information integrtrtion in children. Hillsdale, N.J.: Erlbaum, 1980. Lachman, R., Mistler-Lachman, J., & Butterfield, E. C. Cognitive ps,vcholog~~ and iqfimnation processing: A n inrroducrion. Hillsdale, N.J.: Erlbaum, 1979. Lasky, R. E. The ability of six-year-olds, eight-year-olds, and adults to abstract visual patterns. Child Development, 1974, 45, 626-632. Levine, M. Hypothesis theory and non-learning despite ideal S-R reinforcement contingencies. Psychological Review, 1971, 78, 130-40. McCloskey, M . E., & Glucksberg. S. Natural categories: Well defined or fuzzy sets? Memory & Cognirion, 1978. 6, 462-472.

166

Martha J . Farah and Stephen M . Kosslyn

Medin, D. L.,& Schaffer, M. Context theory of classification learning. Psychological Review, 1978, 85, 207-238. Neumann, P. G . An attribute frequency model for the abstraction of prototypes. Memory & Cognition. 1974.2, 241 -248. Newell, A , , & Simon, H. Human problem solving. Englewood Cliffs, N.J.: Prentice-Hall, 1972. Norman, D. A , , & Rumelhart, D. E. Explorations in cognition. San Francisco: Freeman, 1975. Olson, D. R. On conceptual strategies. In J. S. Bruner, P. Greenfield, & R. Olver (Eds.), Studies in cognitive growth. New York: Wiley, 1966. Pascual-Leone, J . A mathematical model for the transition rule in Piaget’s developmental stages. Acra Psychologica, 1970, 32, 301 -345. Pascual-Leone, J . , & Smith, J. The encoding and decoding of symbols by children: A new experimental paradigm and a neoPiagetian model. Journal of Experimental Child Psychology, 1969, 8, 328-355. Piaget, J . , & Inhelder, B. Mental imagery in rhe child. New Yark: Basic Books, 1971. Posnansky, C. J., & Neumann, P. G. The abstraction of visual prototypes by children. Journd id Experimental Child Psychology. 1976, 21, 367-379. Posner, M. I., & Keele, S. W. On the genesis of abstract ideas. Journal ofExperimental Psychology, 1968, 77, 353-363. Posner, M. I., & Keele, S. W. Retention of abstract ideas. Journal of Experimental Psychology, 1970, 83, 304-308. Pylyshyn, 2. W. Validating computational models: A critique of Anderson’s indeterminacy of representation claim. Psychologicul Review, 1979, 86, 383-394. Reed, S. K. Pattern recognition and categorization. Cognitive Psychology, 1972, 3, 382-407. Reese, H. W. Toward acognitive theory of mnemonic imagery. Journulo~fMenralImugery,1977, 1, 229-244. Rosch, E. On the internal structure of perceptual and semantic categories. In T. E. Moore (Ed.), Cognitive development and the acquisirion of languuge. New York: Academic Press, 1973. Rosch, E. Cognitive representations of semantic categories. Journal of Experimental Psychology: General, 1975, 104, 192-233. Rosch, E . , & Mervis, C. B. Family resemblances: Studies in the internal structure of categories. Cognitive P.syc~hology,1975, I, 573-605. Rosch, E., Simpson, C., & Miller, R. S. Structural bases of typicality effects. Journal of Experimental Psycholi~gy:Humun Perception and Perfbrmance, 1976, 2, 491 -502. Shiffrin, R. M., & Schneider, W. Controlled and automatic human information processing: 11. Perceptual learning, automatic attending, and a general theory. Psychological Review, 1977, 84, 127-190. Smith, E. E., Shoben, E. J., & Rips, L. J . Structure and process in semantic memory: A featural model for semantic decisions. Psycholrigical Review, 1974, 81, 214-241. Standing, L . , Conezio, J., & Haber, R. N. Perception and memory for pictures: Single trial learning of 2560 visual stimuli. Psychoncmic Science, 1970, 19, 73-74. Templin, M. Certrrin lmguuge skills in children: Their development rind interrelationship. Minneapolis: Univ. of Minnesota Press, 1957. Tighe, L. S . , & Tighe, T. I. Overtraining and discrimination shift behavior in children. Psychonomic Science. 1965. 2, 365-366. Vygotsky, L. S. [Thought and language] (E. Hanfmann & G. Vakar, trans.). Cambridge, Mass.: MIT Press, 1962. Werner, H. Comparative psychology IJT mental development. New York: International Universities Press, 1948. White, S. H . Evidence for a hierarchical arrangement of learning processes. Advances in Child Development und Behuvior, 1965, 2, 187-220.

Winston, P. H. Learning structural descriptions from examples. In P. H. Winston (Ed.), The p y h o l o g y o / compufrr vision. New York: McGraw-Hill, 1975. Wittgenstein, L. [Philosophictrl iitvestiKrrrionsI (G. E. M. Anscornbe, trans.). New York: MacmilIan, 1953.

This Page Intentionally Left Blank

PRODUCTION AND PERCEPTION OF FACIAL EXPRESSIONS IN INFANCY AND EARLY CHILDHOOD

Tiffany M . Field and Tedra A . Walden MAILMAN CENTER FOR CHILD DEVELOPMENT UNIVERSITY OF MIAMI MEDICAL SCHOOL MIAMI. FLORIDA

1. ADULT FACIAL EXPRESSIONS

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

A. CATALOGUING FACIAL EXPRESSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. RELATIONSHIPS BETWEEN ENCODING-DECODING AND PHYSIOLOGICAL MEASURES: EXTERNALIZERS AND INTERNALIZERS

11. THE INFANCY LITERATURE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

A. B. C. D.

NEONATAL EXPRESSIONS . . . . . . . . , . , . . . , . , . , . . . . . . . . . . . . , . . . . . . . . INFANT DISCRIMINATION OR DECODING OF FACIAL EXPRESSIONS . PRODUCTION OF EMOTIONAL EXPRESSIONS BY INFANTS . . . . . . . . . . RELATIONSHIPS BETWEEN FACIAL EXPRESSIONS AND PHYSIOLOGICAL RESPONSIVITY OF INFANTS . . . . . . . . . . . . . . . . . . . . . .

111. THE CHILD LITERATURE . . . . . . . . . . . . , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. RECOGNITION OR DECODING OF FACIAL EXPRESSIONS BY CHILDREN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. PRODUCTION OR ENCODING OF FACIAL EXPRESSIONS BY CHILDREN C. RELATIONSHIPS BETWEEN ENCODING AND PHYSIOLOGICAL RESPONSIVITY IN CHILDREN . . . . . . . , . . . , . , , , . , . , . , , , , , , , , . , . , . , . ,

IV. STUDIES OF INFANT AND CHILD EXPRESSIONS BY THE AUTHORS . . . . . A. PRODUCTION AND DISCRIMINATION OF FACIAL EXPRESSIONS BY NEONATES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. ENCODING AND DECODING OF FACIAL EXPRESSIONS BY INFANTS . C. ENCODING AND DECODING OF FACIAL EXPRESSIONS BY YOUNG CHILDREN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

171 171 I72 174 174 175 I76 178 179

179 I8 I 183 185 185

189 191

V. AN INTEGRATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

204

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

208

'Present address: Department of Psychology and Human Development, Peabody College, Vanderbilt University, Nashville, Tennessee 37203. I69 ADVANCES M CHILD DEVELOPMENT AND BEHAVIOR, VOL. 16

I

Copyright @ 1982 by Academic Press, Inc. All rights of reproducrion in any form reserved. ISBN 012-009716.8

I70

T&ny

M . Field und Tcdnr A . Wulden “_ . . the face one sees is not so different from the face one lives behind. ” TOMKINS & MCCARTEK. 1964

Facial expressions are a major channel of interpersonal communication. They occupy a unique role in the understanding and regulation of interpersonal interactions because the face is often impossible to conceal and difficult to control. Moreover, facial expressions may convey underlying affect more readily than the more easily manipulated verbal expressions. Little is known about the early development of facial expressions, despite their clear importance. In a review of earlier research in this area, Charlesworth and Kreutzer (1973) noted that the literature was sparse and that theoretical interpretations of the origins and development of facial expressions had been given relatively minor emphasis. Research since that review has continued to be largely empirical, with relatively little theoretical integration. One possible reason for the absence of theories is the paucity of data on the developmental course of facial expressions. Separate groups of researchers have studied facial expressions at different developmental stages rather than across infancy, childhood, and adulthood, and few researchers have studied both the production and perception of facial expressions. Developmental researchers have referred to these processes in the neonate as reflexive expressions and primitive discriminations. The development of facial expressions in infancy has been represented as a two-part process of imitation and discrimination learning. In childhood, the discrimination and categorization of facial expressions have been investigated, and finally, in adulthood, relationships between encoding and decoding skills and the emergence of externalizerinternalizer personality differences have been the primary focus of studies on facial expression. The literature on adult facial expressions suggests that differential encoding and decoding skills are at least partially a function of differential socialization experiences. If the socialization process includes constraints such as inadequate modeling or negative reinforcement, the child’s encoding and decoding skills may be altered. The child may show limited facial expressiveness, manifesting only internal physiological responses to affective stimulation, as has been described for “internalizers” in the adult literature. Although reflexive expressions and primitive discriminations of facial expressions may be innate processes, the differentiation, refinement, and intensity of these processes, as evidenced by emerging individual and sex differences, may be a function of different socialization processes experienced by infants and children during their early interactions with social models including parents, siblings, teachers, and peers. Using the above framework, we shall review examples from the adult literature that have provided many of the questions and paradigms for researchers of infant and child facial expressions. We then shall present examples of infant and

Fuciiil Expressions

171

child research relevant to the developmental questions suggested by the adult literature. In addition, we shall describe a series of infant and preschool studies in which we assessed relationships between production and perception of facial expressions. Finally, we shall present a developmental model that may integrate these studies and the existing literature.

I. Adult Facial Expressions Research on adult facial expressions suggests the following: (1) eight basic facial expressions-joy, distress, interest, surprise, fear, anger, disgust, and shame-can be reliably elicited and judged in a number of cultures under a number of different conditions; (2) the production and perception or encoding and decoding of facial expressions are related processes; (3) adults may be generally classified as externalizers (facially expressive and relatively less reactive physiologically), generalizers (facially and physiologically reactive), or internalizers (facially less expressive and physiologically more reactive), with associated extroversion-introversion personality characteristics; and (4) individual and sex differences in encoding-decoding skills and externalizerinternalizer characteristics have been ascribed to differential socialization influences. A.

CATALOGUING FACIAL EXPRESSIONS

Several investigators, following the tradition of Charles Darwin, have provided coding schemes for the cataloguing of facial expressions (Ekman & Friesen, 197 1 ; Izdrd, 197 1; Tomkins, 1962). They have photographed models experiencing or mimicking different affects and have described the individual muscle movements of three facial regions in producing the basic facial expressions. The photographs of adults in the United States and other cultures basically support Darwin’s (1877) early catalogue of eight basic facial expressions: joy, distress, interest, surprise, fear, anger, disgust, and shame. Typically, these investigators have trained models to control various facial muscles and to pose both pure facial expressions and “blends” or combinations of facial expressions. The photographs then are categorized by judges also trained in kinesics and the cataloguing of expressions. This research has provided an invaluable catalogue of facial expressions, and a standardized technique for the assessment of facial expressions and sets of stimuli widely used in research on facial expression. However, the use of static displays of exaggerated, peak facial expressions posed by models and rated by judges trained in kinesics has questionable ecological validity, given that peak facial expressions occur naturally for less than .5 second, are surrounded by

172

Tijany M . Field und Tedra A . Walden

movement, and are typically more subtle blends than pure expressions (Ekman & Friesen, 1971). In addition, the real-life expressors and observers are not usually trained in kinesics. Given that both the models and the judges are trained in the same system of kinesics, the high rates of correct judgments of modeled expressions found in these studies are not surprising. Nonetheless, untrained children and adults also show reasonably high rates of correct judgment. In the study by Tomkins and McCarter (1964), for example, the average rate of correct judgments by untrained firemen ranged from .63 for surprise to .99 for joy. In a study by Bassili (1979), students were asked to identify facial expressions from simple white makeup spots applied to parts of an actor’s face. The face was still or moving and was viewed under natural illumination or in darkness. The correct categorization of facial expressions was facilitated by moving the displays and illumination, although surprise and happiness were recognized accurately under all conditions. Because the basic eight facial expressions appear in the repertoire of infants (Hiatt, Campos, & Emde, 1979; Izard, Heubner, Risser, McGinnes, & Dougherty, 1980), children and adults (Izard, 1971), and blind subjects (Thompson, 1941), as well as in different cultures (Chikvishvili, Valsiner, & Lam, 1977; Ekman, Friesen, & Ellsworth, 1972), and because they can be correctly discriminated by naive judges of varying ages, from various cultures, and under various conditions, these investigators have suggested that the production and discrimination of the basic facial expressions may be innate processes. B . RELATIONSHIPS BETWEEN ENCODING-DECODING AND PHYSIOLOGICAL MEASURES: EXTERNALIZERS AND INTERNALIZERS

Research on adult facial expressions has yielded mixed findings, but it generally is suggested that the production and discrimination of facial expressions are related processes (Buck, Miller, & Caul, 1974; Buck, Savin, Miller, & Caul, 1972; Zuckerman, Hall, DeFrank, & Rosenthal, 1976; Zuckerman, Lipets, Koivumaki, & Rosenthal, 1975). Typically, the adult is exposed to slides or videotapes considered to be pleasant and unpleasant. Facial reactions to these stimuli are filmed, usually without the subject’s awareness, and the films are judged at a later time by the same group of subjects. Many of these studies included physiological measures (heart rate and skin conductance) and psychometric assessments of characteristics such as anxiety, self-esteem, and extroversion-introversion so that encoding-decoding skills might be related to the personality characteristics of the subjects. In studies by Buck and colleagues (Buck ef al., 1972, 1974), facial responses were filmed as slides of sexual, scenic, maternal, disgusting, or unpleasant, and unusual or ambiguous (op-art) scenes were presented. In addition, heart rate and skin conductance were recorded, and scales assessing anxiety, self-esteem, and

Foicial Expressions

I73

extroversion-introversion were administered. Negative relationships were reported between the senders’ accuracy of expression and physiological responsivity. More facially expressive adults generally had less physiological reactivity than the less expressive adults. In addition, categorization accuracy was related to extroversion and test anxiety of the sender and self-esteem of the observer, Females were more accurate senders or encoders than males and were more frequently “externalizers. Males were more frequently internalizers, and internalizers scored higher on introversion and sensitization and lower on self-esteem. Jones ( 1 950), in his studies of infants and children, also suggested a negative relationship between facial expressiveness and physiological responsivity , and contrasted externalizers (high facial expressiveness, low physiological reactivity) with internalizers. Jones (1950) suggested that internalizers are “persons who have been discouraged from manifesting emotional responses overtly” (p. 163). Similarly, Buck er al. (1972) related their finding that males are usually more internalizing than females to the possibility that “young boys in our culture are systematically taught to inhibit and mask many emotions” (p. 369). Several alternative explanations have been offered for the apparent negative relationship between facial expressiveness and physiological reactivity. Jones (1950), in discussing internalizers and externalizers, argued that autonomic arousal and facial expressions are substitutable modes of reducing tension. Thus, a person may be externally or internally reactive. Block (1957), however, took issue with Jones’s position, arguing that tension reduction can be brought about by motor behavior or cognition but not by the slower-acting autonomic activity. Schachter (1964) presented a theory of self-attribution of the emotional state, suggesting that emotions are a joint function of autonomic arousal and cognitive attribution to or labeling of the cause of the arousal. Physiologically labile persons may have learned to attribute the causes of arousal to external rather than internal causes. They thus may not perceive themselves as experiencing an “emotion” and may not adopt a face of emotion. Lanzetta and Kleck (1970) advocated a social learning position, suggesting that socialization processes, including modeling and reinforcement, affect the development of facial expressions and may cause them to be intensified or masked. The individual, during socialization, may be punished for overt displays. Arousal is a joint function of the arousal associated with the stimulus and arousal related to response conflict generated by an attempt to inhibit the display. The inhibition of overt responses may result from stressful social rebukes that function as conditioned stimuli eliciting autonomic arousal. An inhibitory response may, itself, become a conditioned stimulus eliciting similar autonomic responses. Because of these experiences, a person may learn to mask facial expressions but display large autonomic responses. Buck el al. (1974) also suggested that the experiences associated with learning to inhibit an overt response may be stressful. The stress stemming from the effort to inhibit expressions, and not the inhibition per se, may be associated with ”

I74

Tiffuny M . Field und Tedra A . Wolden

increased physiological responding. If a child is rebuked by an adult when the child expresses an emotion, that rebuke might both inhibit the overt emotional response and cause an increase in physiological responding. A relationship between behavioral inhibition and physiological responding may emerge without any direct causal relationship between the two variables. Although a negative relationship between facial expressiveness and physiological reactivity has also been suggested for infants and children (Jones, 1950), only two confirmatory reports have appeared in the literature (Buck, 1977; Jones, 1930). That some infants may be internalizers and some externalizers may explain some of the discrepancies in the infant literature between those who find no behavioral-physiological relationships during the recording of infant facial expressions (Lewis, Brooks, & Haviland, 1978) and those who report relationships between facial and physiological responses (Campos, Emde, Gaensbauer, & Henderson, 1975; Provost & Decarie, 1979). Some basic questions derived from the adult literature for the study of infants and children are whether these processes and their relationships can be observed during early development, whether early individual and sex differences can be observed, and whether these appear to be related to differential socialization experiences. Since the orientation of researchers studying infant and child facial expressions has been limited to either perceptual or production skills, answers to these questions can be addressed only by integrating the data of separate studies and making inferential interpretations.

11. The Infancy Literature Until recently, the literature on infant facial expressions was limited to Darwin’s diary of his own infant and a number of studies in the 1920s and 1930s, which were reviewed by Charlesworth and Kreutzer (1973). Although these are studies on older infants, some authors suggested that discriminable facial expressions occur as early as the neonatal stage. A.

NEONATAL EXPRESSIONS

Darwin suggested that many of the basic facial expressions are present shortly after birth. Wolff (1963) reported that expressions such as smiling can be elicited by a variety of sounds, including a high-pitched human voice, during the first week of life. These authors presented their data as well as other data on facial expressions of chimpanzees and blind infants to support a theory that facial expressions are innate, and learning merely refines them. A series of studies by Sherman (1927, 1928) indicated that observers were unable to differentiate among neonatal facial responses to four stimuli: restraint

Ftiriul Expressions

I75

of the head and face, sudden dropping, pricking with a needle, and hunger. Because all of these stimuli would be expected to elicit similar expressions, that is, distressful crying, Sherman should not have expected the judges to differentiate among them. Yet, despite this obvious methodological weakness, these data were accepted as conclusive evidence that neonatal facial expressions could not be differentiated, and since then only a few studies on neonatal expressions have appeared in the literature. Herzka (1965) photographed 38 infants ranging in age from 1 hour to 29 weeks. Brief descriptions of the stimulus conditions thought to have elicited the expression were given, and inferences about the underlying emotional state were made. Expressions ranged from the “angelic smile” to those accompanying intense visual attention to expressions of disgust. The range of expressions reported is not surprising, given that Oster and Ekman (1978), using their fine-grained measurement system, found that all but one of the discrete facial muscle actions visible in the adult can be identified and finely discriminated in full-tern and preterm neonates. Herzka, however, did not provide a precise description of the eliciting stimuli or the resulting expressions, nor were any attempts made to assess the ability of judges to discriminate these expressions. A more systematic approach to eliciting neonatal facial expressions has been developed by Steiner ( 1979). Tastants and food odorants at “suprathreshold” intensities were noted to elicit distinctive facial expressions in neonates. A pleasurable expression characterized by a discrete smile or by sparkling eyes was elicited by sweet taste and pleasant food odors, and an expression of disgust (turned-down mouth corners and closed eyes) was induced by bitter taste and unpleasant food odors. According to Steiner (1980), “the perinatal elicitability of these responses points clearly to the fact that they do not result from cognitive processes or any kind of learning” (p. 365). Thus, except for studies by Steiner, the information to date on the production of neonatal expressions is largely anecdotal or derived from methodologically weak studies, and the discrimination of facial expressions by neonates has not been studied. B.

INFANT DISCRIMINATION OR DECODING OF FACIAL EXPRESSIONS

The most common methods used for the study of the infant’s discrimination of facial expressions are the visual preference and habituation paradigms. Photographs, slides, or live models are used to present the basic adult facial expressions. In one of the few studies to date, 4-month-old infants were habituated to slides of expressions of joy, anger, and neutrality expressed by a male model (LaBarbera, Izard, Vietze, & Parisi, 1976). Each slide remained visible until the infant fixated it and was removed when the infant looked away. Following an intertrial interval, a new expression was presented. As indexed by fixation time

Tiflutiy M. Field and Tedru A . Wuldcn

176

to the new expression, the infants looked significantly longer at the joy expression. Another group, investigating the expressions of sadness, happiness, and surprise in 3-month-old infants, used a slide habituation recovery paradigm (Young-Browne, Rosenfeld, & Horowitz, 1977). The authors argued that this paradigm is “more sensitive than spontaneous fixation in assessing infant visual discrimination” (Young-Browne, Rosenfeld, & Horowitz, 1977, p. 556). The rationale given for the selection of expressions was that toddlers had discriminated among them more often than other expressions (Izard, 1971), as had adults (Ekman, 1972). Half of the infants were repeatedly presented with the modeled expression until a behavioral habituation criterion was reached. Based on a comparison of the recovery scores of experimental and control groups, the infants discriminated between the surprised and the happy expression and occasionally between the surprised and the sad expression. Because no significant differences were noted for mean looking times at the different expressions, the authors suggested that the habituation recovery paradigm is a more sensitive procedure than the visual fixation paradigm. The visual preference methodology was used by Nelson, Morse, and Leavitt (1979) to determine just how sensitive the technique is and to determine whether 7-month-old infants could make generalized discriminations of facial expressions of more than one model. Nelson and co-workers were concerned that in both of the studies just described, the authors had failed to examine the infants’ ability to notice the invariant features that characterize an expression, by not using expressions posed by more than one face. Although Nelson and colleagues demonstrated that infants can discriminate between happy and fear expressions on faces of different actors, the discrimination was affected by the order in which the stimuli were presented. These order effects reflected a faster rate of habituation within this testing paradigm to the happy expressions, suggesting that studies of facial expression discrimination may be affected by differential habituation rates to different expressions. In fact, the more frequent reports of discrimination among positive faces (happy, surprise) than among negative faces (sad, angry, fearful) may relate to the negative expression signaling aversive stimuli that elicit defense-like responses, which are generally habituated more slowly than attentive responses. Nelson et al. (1979) further suggested that the habituation paradigm, given a correction for differential habituation to positive and negative stimuli, may be more sensitive than the visual preference paradigm for studying the early discrimination of facial expressions. C.

PRODUCTION OF EMOTIONAL EXPRESSIONS BY INFANTS

Most investigators of infants’ production of facial expressions have focused on the reliability with which adults can categorize these expressions. Emde, Kligman, Reich, and Wade (1978), for example, filmed infants 2-4 months of age

Faciul E.rpressions

177

playing with their mothers and with strangers. Mothers and nurses gave free descriptions of expressions and forced-choice categorizations of the infants’ expressions while viewing slide and motion picture presentations of the facial expressions observed. Both types of adult ratings of the infant facial expressions could be readily classified within the list of basic expressions if only two categories were added, passive-bored and sleepy. A similar study by Izard et al. (1980) demonstrated that reliability can be increased approximately 15% by training raters on an anatomically based facial movement coding system. Trained judges used facial muscle criteria in selecting expression stimuli from videotape recordings of 1- to 9-month-old infants’ responses to a variety of events ranging from playful interactions to the pain of inoculations. The training effects were approximately equal for emotion labeling (a free-response labeling task), emotion recognition (a selection from a list of emotions), or an emotion-matching technique (a visual comparison using simultaneous presentation of adult and infant expressions). These authors also reported no significant difference in the accuracy with which still (slides) and moving (videotapes) expressions could be rated. To determine whether reliable judgments of infant expressions can be made in the absence of contextual information and can be verified by instrumental behaviors, Hiatt et d.(1979) elicited expressions in a peekaboo game and collapsing toy situation (happiness), a toy-switch and a vanishing object task (surprise), and the visual cliff and the approach of a stranger (fear). Using forced-choice categorizations and confidence ratings of these judgments, raters reliably coded happiness and surprise expressions, but not fear expressions, of the 10- to 12month-old infants in thp, absence of contextual information. These studies suggest that infant facial expressions, previously assumed to be prone to observer bias or subjective interpretation, can be reliably classified using basically the same coding system and expression categories as are used with adults. Although the discrimination and production studies involved different expressions and different-aged infants, “positive” expressions were more reliably discriminated and produced than ‘ ‘negative ’ ’ expressions. The longitudinal study by Sroufe and Wunsch (1972) on the developmental changes in smiling and laughter and the corresponding changes in effective elicitors of these expressions points to the need for a similar longitudinal study on changes in the discrimination and production of facial expressions. A longitudinal study of the development of both discrimination and production abilities might address questions such as whether discrimination and production abilities develop in parallel or at different rates and whether performance in the two areas is related such that “good” discriminators are also “good” producers. The study of discrimination and production as separate processes by separate groups of investigators has contributed to an incomplete understanding of the development of facial expressions.

I78

TiSfny M . Fidd tind Tedru A . Wol&n

D. RELATIONSHIPS BETWEEN FACIAL EXPRESSIONS AND PHYSIOLOGICAL RESPONSIVITY OF INFANTS

In several recent studies, relationships between infant facial expressions and cardiac responsivity have been measured during laboratory situations such as stranger approach (Campos et ul., 1975; Lewis et a / ., 1978; Provost & Decarie, 1979; Waters, Matas, & Sroufe, 1975). Although the data from these studies vary, they generally suggest that negative expressions are associated with heart rate acceleration and positive expressions with heart rate deceleration, at least in infants ranging from 5 to 12 months of age. Wary infant expressions in the Waters et al. (1975) study and distress expressions in the Campos et ul. (1975) study were associated with heart rate acceleration, particularly when the stranger approached the infant in the absence of the infant’s mother. However, the temporal relationship between facial expressions and cardiac responses is not clear in these studies because Waters et al. (1975) did not present time-locked heart rate data and Campos ef al. (1975) presented data on mean facial expression and mean heart rate changes. Provost and Decarie (1979) exposed 9- and 12-month-old infants to a variety of situations that were hypothesized to elicit different facial expressions. For play with mother they hypothesized an interest expression, for being restrained an angry expression, for being approached by a stranger a fear reaction, for being left alone a distress reaction, and for being greeted by the mother an expression of joy. Periods of each facial expression were selected from videotapes, and 5 seconds of heart rate surrounding these periods of facial expressions were averaged. Interest and joy were associated with nonsignficant cardiac deceleration. Anger and distress were associated with clear, significant heart rate acceleration. However, interest and joy could not be differentiated on the basis of heart rate responses nor could anger be distinguished from distress. These authors reported that analyses of individual differences are now being performed because some infants showed heart rate decelerations during interest and joy expressions and others showed cardiac accelerations. Covariation between facial expressions and cardiac activity might be expected within an individual. However, some investigators who have examined individual differences have reported that persons who show strong facial expressions exhibit relatively weak autonomic responses and those who are autonomically reactive show minimal facial responsivity (Jones, 1950). Jones coined the terms “externalizer” and “internalizer” to describe these individual differences in the infants and children whom he studied. Using the galvanic skin response as an autonomic measure, he noted an increase in autonomic activity with mild facial responses and a decrease in autonomic activity during more expressive responses.

Fucial Expressions

179

In light of these results, it is interesting that Lewis et ul. (1978) found relatively little facial expressiveness in a group of infants who showed large heart rate responses and minimal cardiac responsivity in a group who showed large facial responses. The posited relationship between facial wariness and cardiac acceleration did not emerge in the study by Lewis and co-workers, possibly because they had separately grouped high expressive-low autonomically reactive and low expressive-high autonomically reactive infants. These data suggest different patterns of individual response, with an individual tending to be either physiologically or facially responsive. Individual differences in infants’ predominant response patterns may also be described along an externalizer-internalizer dimension.

111. The Child Literature A.

RECOGNITION OR DECODING OF FACIAL EXPRESSIONS BY CHILDREN

The paradigms used to study children’s recognition of facial expressions have been borrowed from the adult literature. The child is shown slides of posed facial expressions, situations that portray characters experiencing different emotions, or a story that portrays an emotion. Dependent measures may include pointing to a representation of the emotion expressed or labeling the expression, and independent measures include age, sex, or culture of the child. Two groups have investigated children’s discrimination of photographs of actors’ posed facial expressions, one with black and white American children 4-6 years old (Glitter, Mostofsky, & Quincy, 1971) and one with American and French children 2.5-9 years old (Izard, 1971). No ethnic, cultural, or sex differences were noted, but older children recognized the expressions more frequently than younger children. Happy expressions were more accurately recognized than anger or surprise. In addition, Izard (1971) contrasted labeling and recognition skills and found that labeling skills were correlated with an intelligence measure within his middle-income sample. Although labeling skills improved up to the age of 5 years, there was no significant change in labeling beyond 5 years of age. The selection of facial expressions to match stories was investigated by Borke ( I 973) with 3- to 6-year-old Chinese and American children and by Ekman and Friesen (1 97 1) with 6- to IS-year-old New Guinea children. Borke reported that by 3 years of age, the majority of both Chinese and American children could correctly match happy and unhappy expressions with stories, but discriminations of fearful and angry expressions developed later. Ekrnan and Friesen’s data

180

Tifuny M . Field and Tedra A . Walden

suggested no age differences between recognition of these emotions by 6- and 15-year-old children, whose average accuracy approximated 90%. Similar age trends were reported for all cultures, and although Ekman and Friesen did not report comparisons by sex, Borke noted that both Chinese and American girls were more accurate than boys in recognizing emotional expressions. Feshbach and Roe (1968) also reported sex differences among 6- and 7year-olds, favoring females, in a study with slide presentations depicting happy, sad, fearful, and angry situations. In addition, an interaction between the sex of the stimulus person and sex of the judge indicated that boys were more accurate when judging boys and girls were more accurate when judging girls. Feshbach and Roe speculated that the similarity between the actor and the judge facilitated empathy. The child, by putting himself or herself in the other’s place, could judge emotions more accurately. Happy situations were again more reliably recognized than sad situations, which, in turn, yielded more accurate scores than fearful or angry situations. Some evidence suggests that young children differ from older persons in the kind of information about faces that is represented and stored in memory. Young children appear to rely more heavily on specific distinctive features of faces (e.g., a mustache or glasses) to discriminate one from another (Diamond & Carey, 1977), while older persons rely more on configural relations among features of faces. Carey, Diamond, and Woods ( 1 980) speculated that young chitdren are deficient in representing configural properties of faces and are therefore less accurate at recognizing a face that they have seen before. A “spurt in growth of representational skills” (Carey et al., 1980) occurring between the ages of 6 and 10 shifts facial encoding from piecemeal to configural processing, thus increasing the recognizability of a face across different expressions, orientations, or other conditions. Thus, in the few basic expressions investigated, cultural differences were noted in recognition, occasional sex differences favored females, and consistent age trends suggested improved recognition of expressions until approximately I0 years of age. Several researchers have suggested that differential ability by sex and age may be related to early socialization influences. Early socialization influences were investigated in a study by Daly, Abramovitch, and Pliner (1980). They measured the relationship between the expressive or encoding skills of mothers and the perception or decoding skills of their 2-year-old children. Mothers were videotaped while viewing slides of pleasant (smiling infants), neutral (furniture), and disgusting scenes (accidents and autopsies). Trained coders rated the videotapes of the mothers’ expressions, and the mothers were classified according to their facial reactions as “good“ or “bad” encoders. Both mothers and their children then coded the videotapes of mothers who were good and bad encoders, and a correlation between mothers’ encoding scores and their children’s decoding scores was noted. This study

Facial Expressions

181

suggests that parents’ ability to produce easily discriminable facial expressions may be a developmental antecedent of children’s discrimination skill. Exposure to social environments that vary in expression of parental emotional expression is a possible cause of individual differences in decoding ability. Children’s decoding skills may be affected by their mothers’ and other caregivers’ encoding skills. Curiously, no relationship was noted between a mother’s and her child’s decoding skills, and unfortunately, the children’s encoding or expressive skills were not examined to provide a measure of the relationship between the child’s production and discrimination skills. B.

PRODUCTION OR ENCODING OF FACIAL EXPRESSIONS BY CHILDREN

Early socialization influences are also suggested by the few studies published on children‘s production of facial expressions. Children produce socially negative expressions (e.g., anger and fear) less accurately than socially positive expressions (Buck, 1975; Odom & Lemond, 1972; Yarczower, Kilbride, & Hill, 1979). Furthermore, negative expressions are less accurately and less intensely posed when an examiner is present than when absent (Yarczower er al., 1979), suggesting that less socially desirable expressions are inhibited in the presence of an adult. In the study by Yarczower et al. (1979), 6-, 1 I - , and 18-year-old subjects were requested to imitate the Izard (1971) adult-posed photographs of surprise, fear, joy, anger, and distress in the examiner’s presence and absence. The reported accuracies of imitations were somewhat low (26, 30, and 42%, respectively, for the 6-, 1 I - , and 18-year-old groups during the examiner-present condition, and 33, 52, and 36% for the examiner-absent condition). The low accuracy may have resulted in part from a methodological feature; the coders were given the eight basic expression labels as options despite the subjects’ having been requested to imitate only five of the expressions. The 1 I-year-olds were significantly better than the 6-year-olds only in imitating negative expressions in the examiner-absent condition, providing some support for Izard’s (197 I ) proposition that “as an individual moves into later childhood, peers and parents begin to discourage the ready display of strong emotions on the face” (p. 192). That socialization may inhibit facial expressions is also suggested by the lower accuracy and intensity of the less socially acceptable fear and anger expressions, particularly in the presence of the examiner. Odom and Lemond (1972) reported that both 5- and 10-year-old children produced negative expressions less accurately, lending support to the position that negative expressions are inhibited. In the only study tapping both discrimination and production skills of children, Odom and Lemond (1972), surprisingly, reported that skill in facial production lags behind skill in facial discrimination, and that the lag is, again surprisingly, wider in the older age group. These

182

T i f f h y M . Field find Tedrri A . Wulden

findings, however, may be an artifact of the design of the study. The stimuli used to assess the children’s discriminations were Izard’s (197 1) actor-posed adult photographs. The children’s production accuracy was assessed by raters who were first trained on the actor-posed photographs and then rated on the children’s posed imitations of those photographs. Thus, production was assessed in terms of adult raters ’ ability to recognize children’s expressions after being trained on the more idealized standardized adult photographs. Comparing the less idealized child productions with the more idealized Izard photographs would understandably contribute to less adequate production ratings. Because the children’s recognition of expressions was based on the idealized photographs, and because idealized forms are more easily recognized than reproduced, recognition scores would predictably exceed production scores. If less perfectly posed photographs, less exaggerated in form, had been used as stimuli for the discrimination task (e.g., children’s posed expressions), or if children’s poses had been used in the training of raters, the production and discrimination scores of the children might have been more similar. In addition, the accuracy of spontaneous poses and posed imitations of adult photographs were lumped together in the children’s production scores. Because the former were notably less accurate (or less exaggerated) than the latter in children’s production of facial expressions, averaging those scores may have artifactually deflated the total production scores. Finally, although these authors were in the unique position of having both discrimination and production data on the same children, unfortunately no analyses of the relationship between early decoding and encoding skills were reported. The investigator who has most directly addressed the questions raised by the adult literature is Buck (1975, 1977), who had previously conducted several studies on the production skills of adults (Buck et al., 1972, 1974). Buck (1975) studied the productions of 4- to 6-year-olds using the same technique used with adults, first presenting pleasant and unpleasant slides and videotaping the children’s spontaneous facial responses and then asking the children to “show me what your face looks like when you feel ‘happy,’ ‘sad,’ ‘afraid,’ or ‘surprised. ’ The slides included pleasant scenes (e.g., the child subject with other children) and unpleasant scenes (e.g., a grotesque clown, a Dr. Suess character, a close-up of a grasshopper, and the painting “Echo of a Scream” depicting a child screaming). In addition, the child’s toy preferences were assessed to determine any stereotyped sex role preferences, and the children’s teachers rated them on the children’s adaptation of the Affect Expression Rating Scale used to assess adult externalizer-internalizer characteristics. The following results were reported: (1) individual differences in nonverbal expressiveness were large, with 8 of the 14 children showing significantly accurate production ability; (2) females produced more recognizable spontaneous expressions to slides than did males; ( 3 ) happiness was more accurately produced than fear or anger in the role-playing “show me what your face looks like when you feel” situation; (4) production ability in ”

the two situations was positively correlated; (5) production ability was positively related to teachers ’ ratings of extroversion (high activity level, aggressiveness, impulsiveness, bossiness, and sociability) and negatively related to introversion (shyness, cooperation, emotional inhibition, and control) in separate analyses of boys and girls; and (6)among girls, teachers’ ratings were more highly related to spontaneous responses to slides than the role-played expression measures, but the reverse was true for boys. Although this study is unique in its measurement of both spontaneous and posed facial expressions and the relationship of these to personality ratings, some methodological weaknesses can be noted. Judges were not blind to the facial expressions being elicited. Because they did not categorize the expression but merely rated the appropriateness of the expressions, they could not determine whether the accuracy of recognizing the posed expressions was greater than would occur by chance. Second, the posed, role-played expressions occurred after the spontaneous expressions to the slides were elicited, which may have artifactually inflated the relationship between posed and spontaneous expressions. Nonetheless, the moderate correlations suggest that relationships between encoding ability and the externalizer-internalizer or extroversion-introversion characteristics previously reported in the adult literature may emerge in early childhood. Buck (1975) concluded: “If one assumes that the behavior of these children is based more upon innate temperamental propensities and less upon socialization than is the behavior of adults, the personality differences in facial expressions may be based upon innate factors while the sex differences (which were relatively weaker in this study) may be based upon socialization” (p. 652). Of course, to dismiss socialization as a possible influence on the individual differences observed, merely because the differences appear as early as age 5 . is premature. These relationships have not yet been assessed prior to age 5, several years during which socialization experiences would be expected to be extremely influential. C.

RELATIONSHIPS BETWEEN ENCODING AND PHYSIOLOGICAL RESPONSIVITY IN CHILDREN

A study by Buck (1977) included skin conductance as a measure of physiological reactivity in the paradigm just described. Mothers categorized their 4- to 6-year-old children’s facial reactions to pleasant, unpleasant, and neutral slides. Skin conductance was quantified by subtracting the number of fast conductance changes exceeding 500 ohms during the 10-second preslide period from similar responses occurring during the 10-second slide period. High communication accuracy, as rated by the mothers, was associated with low skin conductance responding in the children. Expressiveness, rated by the children’s teachers using

I 84

Tiffuny M. Field und Tedra A . Wulden

Buck’s Affect Expression Rating Scale for Children, was associated with high communication accuracy and low skin conductance responding. High ratings of expressiveness (versus inhibition) among boys were related to fewer skin conductance responses to the slides and more accurate communication, whereas high antagonistic (versus cooperative) ratings among girls were related to fewer skin conductance responses to the slides and more accurate communicating ratings. Although boys were rated as marginally more inhibited than girls, the tendency for female senders to be externalizers and males to be internalizers, observed in adult samples, did not appear in these children. The negative correlation between skin conductance responding and communication accuracy is offered as further evidence of the externalizer-internalizer distinction noted in Buck’s earlier (1975) study, in which relationships between communication accuracy and externalizer-internalizer personality traits were reported. Although Buck (1975) again argued that this personality difference may be based on innate factors, he also admitted the possibility that significant social-learning experiences occurring prior to preschool age may account for these relationships, and he suggested that longitudinal studies of the relationships between emotional expressiveness and autonomic nervous system functioning are needed. Buck (1977) discussed two major theoretical positions relevant to his externalizer-internalizer findings; these positions involve different assumptions about the basis of the relation between electrodermal responding and overt emotional expression: Jones ( I 960) has suggested that if overt expression brings social disapproval, such expression will be inhibited and this inhibition will in some unspecified way cause increased use of hidden “internal avenues” of affect discharge. Thus, the inhibition of overt responses increases physiological arousal. In contrast, Eysenck (1967) and Gray (1972) have suggested that the direction of causality goes the other way, that persons who are physiologically arousable are more susceptible to conditioning and are thus more likely to inhibit overt emotional expression. This view identifies socialization as a product of conditioning; with more efficient conditioning (in arousable persons), there is more efficient socialization and, thus, a greater inhibition of overt expression. Despite their differences, both these theories assume that social learning plays a major role in the development of the relationship between skin conductance responding and emotional expression. In both interpretations, the lack of overt expression is said to be due to inhibition caused by social pressures. (Buck, 1977, p. 234)

Thus, separate studies of children collectively suggest some evidence for rather sophisticated encoding and decoding skills in children, a relationship between encoding accuracy and extroversion, a negative relationship between communication accuracy and physiological reactivity (or an externalizerinternalizer dimension), and occasional sex differences with preschool boys being less expressive and less accurate than preschool girls in judging expressions.

Facicil Expressions

I85

IV. Studies of Infant and Child Expressions by the Authors Our review of studies on infants’ and children’s facial expressions suggests that questions about the relationship between perception and production of facial expressions during infancy and childhood are difficult to answer, as the two skills have typically been studied as separate processes. In the following studies we investigated discrimination and production skills in the same infants and children. Children were observed under both laboratory experimental and free-play naturalistic situations, and personality assessments of externalizer-internalizer characteristics were made. These are preliminary studies to a much needed longitudinal investigation of the early development of the perception and production of facial expressions. A.

PRODUCTION AND DISCRIMINATION OF FACIAL EXPRESSIONS BY NEONATES

A study was designed to observe the production and discrimination of facial expressions by neonates (Field, Woodson, Greenberg, & Cohen, 1981). Thresholds to stimulation and physiological activity were recorded to determine the extent to which neonates’ physiological reactivity was related to facial expressiveness. We considered that since the cataloguing of adult expressions had recently been adapted for infants (Oster, 1978), the facial expressions of neonates could be more systematically investigated at this time. Oster and Ekman (1978), using a fine-grained measurement system, have confirmed that all but one of the discrete facial muscle actions visible in the adult can be identified and finely discriminated in newborns. Given these data and the evidence for various neonatal facial expressions offered by Steiner (1 979) and Wolff (1963), we anticipated finding some of the basic facial expressions in neonates. Our interest in recording thresholds to stimulation and physiological reactivity stemmed from Eysenck’s (1967) finding that extroverted, facially expressive adults have high thresholds to stimulation, show low level physiological responses to stimulation, and are difficult to condition. Introverted adults are less expressive facially, have lower thresholds to stimulation, show considerable physiological responsivity to stimulation, and are more readily conditioned. Our questions, then, were: 1. Do infants spontaneously emit any of the eight basic facial expressions or can these be elicited? 2. Can infants discriminate these facial expressions?

186

Tijfkiny M. Firld rind Terlra A . Wrdden

3. Is there evidence for a relationship between facial expressiveness, stimulus thresholds, and physiological responsivity in neonates? We observed the following behaviors in 48 sleeping neonates (M = 28 hours): sleep state, activity level, facial expressions, and heart rate. A series of five pinpricks was then administered as in the Brazelton Neonatal Behavior Assessment Scale (Brazelton, 1973) to determine the sleeping neonates’ threshold to tactile stimuli. The neonates’ cry sounds were recorded, and the latency to cry and cry duration were measured. A series of buzzer tones was then presented, and the infants’ thresholds for the auditory stimulation were calculated and cardiac responses to these were concurrently monitored. These tests were followed by an administration of the Brazelton scale, during which the neonates’ facial expressions were continuously filmed. Finally, the neonates ’ discrimination among three facial expressions (happy, sad, and surprised) was assessed with a trials-to-criterion habituation procedure. The auditory threshold assessment involved presenting a series of six buzzer tones varying in intensity from 80 to 110 decibels (the hypothesized range in which orienting and defensive responses might be elicited). These were presented for 1.5-second periods with an interstimulus interval of 30 seconds. Three orders of stimulus presentation were determined in a Latin square design with the three hypothesized orienting threshold tones preceding the three hypothesized defensive threshold tones. Filming of the facial expressions during Brazelton items was done with a zoom lens on the neonate’s face so that judges might observe facial expressions without stimulus or situation cues. The babies’ faces were filmed during the following stimulus presentations: (1) orienting stimuli-face alone, voice alone, face and voice, inanimate auditory (rattle to each side of face), inanimate visual (red ball tracked across field of vision); and (2) the 2 0 Brazelton reflex maneuvers. The photographer indicated the onset of each stimulus by a visual marker on the film so that films could be coded without the cues of sound effects. Judges then coded the infant’s expression at the time of the indicated stimulus onset as one of the basic eight facial expressions. The neonate’s ability to discriminate among happy, sad, and surprised expressions modeled by the examiner was assessed with a trials-to-criterion habituation procedure. We hypothesized that the neonates would be capable of discriminating among facial expressions because: (1) neonates can discriminate scrambled from regular faces (Goren, Sarty, & Wu, 1975); (2) neonates have been reputed to imitate facial movements such as tongue protrusion and mouth widening (Meltzoff & Moore,. 1977); and (3) slightly older infants (3 months old) have been shown to discriminate happy from surprised faces using a trials-tocriterion habituation paradigm (Young-Browne et a!. , 1977).

Facicil Exprrssiotis

187

The examiner held the neonate in a semiupright position with the newborn’s head supported in her hands (as is done for the Brazelton orienting procedure) at about 10 inches from the examiner’s face. A tongue-clicking sound was made to elicit the neonate’s eye contact, at which point the examiner emitted a happy face (smile), a sad face (lower lip protrusion), and a surprised face (mouth and eye widening) in a Latin square order. The expression was sustained until the infant turned away from the examiner, and repeated trials were presented until the neonate looked at the examiner’s face for less than 2 seconds (habituation criterion). The remaining stimuli then were presented in a similar fashion. Because state was a critical variable in this procedure, the infants were given vestibular stimulation between trials to maintain a quiet, alert state. The length of visual fixations and number of trials to criterion were recorded by a second observer, who stood behind the seated examiner who modeled the expressions in order to remain unaware of the facial expression being modeled. The second observer also recorded eye and mouth movements (e.g., eye widening and lip protrusion of the neonate) as an incidental measure of imitation of facial expressions by the neonates. Preliminary analyses of the facial expressions produced during the Brazelton examination revealed that several expressions occurred at greater than chance frequencies to specific stimuli. The face and voice orientation stimulus and the first animate and inanimate auditory orientation stimuli produced an interested expression. The second animate and inanimate auditory orientation stimuli, in which the stimulus was presented to the opposite side of the head from the initial presentation, and the tonic neck deviation reflex maneuver elicited surprised expressions in the neonates. Sad or precry grimaces were elicited by the tonic neck reflex. Elicitation of the Nroro reflex produced fearful facial expressions. Disgusted faces were elicited during the insertion of a (soapy tasting) finger when evaluating the sucking reflex. Ashamed and angry facial expressions were not observed. While happy expressions (smiles) occurred with moderate frequency during sleep, they were rarely observed during awake states except occasionally during the elicitation of rooting. Preliminary analyses of neonates’ discrimination of facial expressions revealed that neonates showed a gradual increase followed by a decrease in looking time to the facial stimuli over trials, suggesting an habituation effect (see Fig. 1). Increases in mean looking time from the last trial of one expression to the first trial of a new expression were observed, suggesting a dishabituation effect. Longer looking time per trial was observed to the happy than to the sad or surprise expressions. Finally, the infants required a greater number of trials to habituate to the surprised expression. In neonates who were predominantly alert and attentive, scanning patterns emerged. When observing happy and sad faces (for which the mouth region is

188

Tijfuny M. Field und Tedru A . Wulden

Sad

Happy Surprised

1

I

I

1 I

First

Middle

Last

1

TRI ALS Fig. I . Meun looking time (in seconds) of neonutes to huppy (smile). sud (frown). and surprised fucial expressions during first, middle, and last habituation rriuls.

more salient and perhaps most critical in defining the expression), neonates scanned primarily the mouth region with only occasional looks at the eye region. However, when observing the surprised expression (in which both eyes and mouth are widened and therefore salient, and both widened eyes and mouth define the expression), the infants alternately scanned both the mouth and eye region, distributing their looking time almost evenly to the two facial regions. In addition, the observer, blind to the model’s facial expressions, guessed at greater than chance (33%) levels which of the three expressions was being modeled on

Facial Expressions

I89

the first series of trials by merely observing the neonate’s face (56% for happy, 62% for sad, and 77% for surprise). Finally, infants’ cardiac responsivity during the auditory threshold series, behavioral responses to the threshold trials, facial expressiveness, and habituation of facial expressions were related. Neonates who showed the greatest number of facial expressions during sleep, the Brazelton, and habituation trials, exhibited higher thresholds to the auditory stimuli, less cardiac responsivity (or a narrower range of beat-per-minute change from baseline to stimulus heart rate), and slower visual habituation of the facial expressions. These data, although preliminary, suggest that neonates exhibit a repertoire of facial expressions (except for ashamed and angry expressions) that are identifiable and discrete. In addition, they occasionally show “imitative” responses to these expressions. Neonates can discriminate among at least three basic expressions, happy, sad, and surprised. They look longer at the happy expression, and may therefore be said to prefer it, perhaps because happy faces are more familiar. Neonates take a longer time to habituate to a surprised expression, perhaps because of the greater number of salient or defining features and/or its somewhat negative quality. Although the relationships among thresholds, cardiac responsivity , facial expressiveness, and habituation were somewhat surpnsing, they are consistent with Eysenck’s (1967) finding that adults who are exceptionally expressive (extroverted) show high thresholds to stimulation, less autonomic responsivity, and less conditionability . B. ENCODING AND DECODING OF FACIAL EXPRESSIONS BY INFANTS

In a second study we assessed the relationship between infants’ decoding and spontaneous encoding or perception and production skills (Field & Greenberg, in preparation). Spontaneous, face-to-face play interactions between mothers and their 8- and 12-week-old infants were videotaped, and heart rate was monitored continuously. The tapes were coded for the frequency with which the eight different expressions were emitted by the infants during the interactions. For a measure of decoding ability, we tallied the proportion of maternal expressions that were contingently responded to by each infant. Contingent responding was defined as a similar facial expression by the infant within 3 seconds of the mother’s modeled expression, a measure that implies both discrimination and production abilities. Although this procedure is a less precise way of tapping these skills, the study was a preliminary investigation of the repertoire of facial expressions spontaneously emitted by infants during natural interactions and the relationships between encoding and decoding skills of the very young infant. During 90-second spontaneous, face-to-face interactions with their mothers, 8-week-old infants emitted an average of 5.7 facial expressions (see Fig. 2). Of these, about equal numbers were happy, interested, and sad expressions. The

-I

4.0--

Happy

Interested

Sad

Surprised

Disgusted

Angry

Afraid

Ashamed

1

8-week-01 d 12-week-01 d

Fig. 2 . Mean frequency of facial expressions emitted by 8- and 12-week-old infants during spontaneous face-to:face interactions with their mothers.

Fuciol Expressions

191

surprised expression occurred very infrequently. The others among the basic eight facial expressions (angry, afraid, disgusted, and ashamed) were never observed among the 8-week-old infants. The 12-week-old infants averaged 6.3 facial expressions per 90-second period. Happy or smiling expressions occurred more frequently than the other expressions, at twice the frequency of the interested expression and four times the frequency of the sad expression. Surprised and disgusted expressions occurred infrequently, and angry, afraid, and ashamed expressions were never observed. The older 12-week-old infants emitted twice as many happy faces as the younger 8-week-old infants and half as many sad faces as the younger infants. Contingent responding was used as a measure of decoding ability because the infant must recognize the mother’s expression in order to respond with a similar expression. Infant vocalizations in response to maternal vocalizations, smiles i n response to smiles, and smiles plus vocalizations in response to mothers’ vocalizations (which also were accompanied by smiles) were coded. The younger 8-week-old infants emitted contingent vocalizations at greater than chance frequency. Contingent responses of this type occurred with four times the frequency of contingent smiles and contingent smiles plus vocalizations. For the older 12-week-old infants, contingent vocalizations, smiles, and contingent smiles plus vocalizations occurred at greater than chance frequencies. A correlation analysis revealed a significant relationship ( r = .58) between the number of facial expressions emitted and the number of contingent responses to the mothers ’ expressions. The infants who produced facial expressions more frequently also responded contingently to facial expressions more often. These results suggest that at least happy, sad, and interested expressions occur with some frequency during the early interactions of very young infants. They also suggest that a relationship between encoding and decoding skills may emerge at a very early age. C. ENCODING AND DECODING OF FACIAL EXPRESSIONS BY YOUNG CHILDREN

The following series of studies was designed to assess the discrimination and production of facial expressions by preschool children in laboratory and naturalistic situations. In addition, each child’s extroversion-introversion, popularity, and IQ were assessed. Subjects in these studies were 40 middle-income preschool children ranging in age from 3 to 6 years of age. The children represented a number of different cultural groups including Hispanic, French, Chinese, and Indian. Their IQs ranged from 87 to 144 ( M = 116; SD = 16). A11 children attended full-day nursery school in three separate classes in three different schools. In two of the studies we assessed the children’s discrimination of facial ex-

192

Tiyany M. Field und Tedra A . Walden

pressions. In the third study we assessed the relationship between the children’s encoding and decoding skills and the relationship of these skills to externalizerinternalizer personality characteristics. 1 . Study I : Decoding of Stylized Facial Expressions by Young

Children Study 1 involved a matching procedure in which children were twice shown “Creative Playthings” blocks of 12 stylized line drawings of faces (Field & Walden, in preparation). As shown in Fig. 3, these faces differ in the orientation of the eyes (winking or looking to the left, right, or forward), nose (curved or triangular), and mouth (curving upward or downward). For each of the 6 matching pairs the children were presented a face and asked to select the same face from an array of the other 1 1 faces. The number of errors made by individual children ranged from 0 to 11 out of a possible 12, with a mean of 4.8 or approximately 40% for all children. The most common error resulted from failing to match the eyes of the standard (69%of all single-feature errors were eye mismatches). Errors resulting from failing to match the nose were least frequent, and mouth errors were intermediate. Although the findings that nose errors were the least common was expected, this finding may reflect the specific stimuli used. They were highly stylized and perhaps tended to emphasize the nose (see Fig. 3).

Fig. 3. Stylized,facial expression blocks used as stimuli in Study I

Fucinl Expressiotis

I93

Discrimination among the faces was more accurate in older than in younger children and among children who had higher IQ scores. No sex differences in discrimination ability were observed.

2 . Study 2 : Yourig Children’s Decoding of Realistic Drawings of Facial Expressions Study 2 was a conceptual replication and extension of Study 1 (Walden & Field, 1981). The facial expression stimuli were more realistic than in Study 1 and thus may have been more meaningful or ecologically valid stimuli. Because they differed on only one dimension at a time (eyes, mouth, hair), we were able to analyze decoding errors more closely. Furthermore, the children were given various kinds of “prompts” to facilitate their decoding performance. The children were presented sets of drawings of children showing facial expressions of joy, sadness, surprise, and anger (see Fig. 4 for examples of these expressions). Each child was to choose from an array of five choices the other face that “felt the same way” as a standard. The set of choices always contained: (1) an expressive match for the standard; (2) a face that matched only the eyes of the standard; (3) a face that matched only the mouth of the standard; (4) a face

Fig. 4 . Stcrndrrrdfarid expressions (huppy. srrd, sitrprised, crud angry) used ns stimuli in Study 2.

I94

T@in,v M . Field arid Tedru A . Wu[dcn

that matched only the hair of the standard (a feature irrelevant to the affective expression); and (5) a face in which no features matched the standard (see Fig. 5 for a sample array of expressions). In some cases the facial features (eyes, mouth, nose, brows) of the match were exact duplicates of those of the standard, and in other cases the expressive match was a generalized version of the standard’s affective expression with no identical features. In addition, children were given various prompts to encourage selection of the accurate match for the standard presented. Prompt conditions were: ( 1 ) presentation of the standard while pointing out relevant features of the face to focus on (e.g., eyes, mouth, brows); (2) presentation of the standard along with a verbal label of the expression depicted (e.g., “Pick the other happy one”); (3) presentation of a standard face only; and (4) instead of a standard face, presentation of a request to choose the (happy-sad-surprised-angry) face. We reasoned that by examining the experimental cues present in conditions in which children performed more accurately, and comparing them to the cues available in conditions

Fig. 5 . Sornplr set including stirnirlifbr idprrricci1fecirurc.s mid gerterolized niatc./i in Srudy 2 (rop .i5 sfundurd sudfiice; clockwise,from uppm Iiji, irrc.levcintfeatrrre mtrrdt (huir). eyes only mtrtch, mouth only march, identical fearures motch, no murch, generalized matc.h).

in which children performed more poorly, we might obtain information about the locus of deficits in children’s judgments of facial expressions. Results showed that children made fewer errors matching happy expressions than sad, surprised, or angry expressions (see Fig. 6). Happy expressions were almost always matched correctly without any kind of prompting, and happy faces were matched equally well on generalization trials as on identical feature trials. All other expressions were poorly matched on generalization trials. Furthermore, almost perfect performance was observed in both labeling conditions for the “happy” expression, but labeling did not facilitate matching performance on other expressions. This finding suggests that the children may have a better defined and more generalized notion of the concept “happy expression” than of

I

i d e n t i c a l f e a t u r e s matches

general i z e d matches

t u w ci LL 0

u

Happy Sad Surprised Ang rY F i g , 6 . Percentugc of correct mulches selected t ~ preschoolers y for diflerenr expres5ions given generalized and identiid stimuli (note: huppy > sud > surprised = u n g p ut p < .(IS).

I96

Tiflkny M . Field nnd Tedra A . Walden

other expressions. Alternatively, perhaps the upturned mouth of the smile is more salient or more usually a characteristic of happy expressions so that the smile itself is responded to rather than the concept happy. If so, the smile itself would increase the number of correct matches of happy faces. We also found, however, that the children tended to make more of their incorrect matches by choosing a face with the same eyes as the standard than by basing the match on mouths. Twice as many errors were made by choosing a face that had the eyes of the standard than by choosing a face that had the same mouth as the standard. Thus, the mouth itself is not an especially salient feature, but rather, either the smile is an especially salient feature or the category “happy” is more familiar to the children. Supporting the latter hypothesis were data on the labels condition, in which the happy match was correctly identified more often when the happy label was given than in conditions in which the “happy” label was not given. Because the child’s performance in finding “the (other) happy one” was superior to performance in finding “one that feels the same way this one does” even when critical features (mouth and eyes) were pointed out, we may hypothesize that the child has a better understanding of the general expressive concept category “happy” and that the verbal label facilitates access to this category. Two prompt conditions, the standard plus features and standard plus labels, produced fewer errors than standard only or label only conditions, with the best performance in the standard plus labels condition (see Fig. 7). Thus, both pointing out relevant features of facial expressions and providing verbal labels for the standard to be matched improved performance, suggesting that children may spontaneously fail to do either. If children do not systematically focus on and compare the defining features of various facial expressions, emphasizing features may improve attention to relevant details and therefore aid matching. Labeling may facilitate the categorizations of facial expressions by providing a conceptual category to aid both selection and memory processes. The child selecting a match in the label conditions can label each face and then match category labels (a happy face and a happy face) or may match the pictorial images. Thus, the child has several matching strategies at his or her disposal. The two labeling conditions also facilitated matching of generalized, nonidentical versions of expressions, suggesting that providing verbal labels aids recognition of general classes of facial expression (see Fig. 7). The accuracy of discrimination among the facial expressions increased with age and was more accurate in children who had higher IQs. No sex differences were observed. Thus, the results of this study suggest that preschool children discriminate happy expressions from all other expressions. Children appear to have formed a concept category for “happy” expressions in general, and the verbal label “happy” makes the category accessible. Other categories of facial

Fucial Expressions

197

i d e n t i c a l f e a t u r e s matches genera 1 i zed matches

100

90 80 I-

u

70

s

60

w cz ul W

c3

a 5 w 5

50

V

40

n

30 20 10

Standard Only

Standard + Fea t u r es

Standard Labels

+

Labels Only

Fig. 7. Percentage of correct matches selected by preschuolers in the four prompt conditions (note: standard only < standard +features < standard + libels > labels only at p < ,051.

expression lag behind in formation and are more ill defined, and verbal labels are less likely to aid in making the general category accessible. Effects of prompts in increasing accuracy of matching suggest that children’s discriminations may be hampered by a failure to compare systematically defining features of facial expressions when matching similar expressions. When we encouraged children to compare features, matching improved. Because labeling also improved performance, it may be difficult to ascribe the child’s poor discrimination simply to failure to match correctly similar features of these faces. However, labeling may also facilitate the comparison of defining features by “reminding” the child of relevant features of faces that fit the category (e.g.,

I98

T$uny M . Field and Tedra A . Wulden

Fig. 8 . Phorogruphs qf busic eight fiieial expressions itsed NS srimrtli in Siirdy 3 . (Phorographs hy Tifuny Field.)

I99

Fig. 8 (c.orrtir~uerf)

200

T@ny M. Field cznd Tedra A . Wdden

“happy faces have upturned mouths and relaxed eyebrows”). Additional research is needed to explicate the processes by which children learn to discriminate among facial expressions. 3 . Study 3: Encoding and Decoding of Facial Expressions by Preschool Children

The literature on adult facial expressions suggests that expressions and recognition or encoding and decoding of expressions may be related skills, so that “good” encoders are also good decoders. In addition, good encoders are rated as having more extroverted personalities. Several investigators of adult expressions have proposed that individual differences in these skills may gradually evolve .,cross development due to differential reinforcement for facial expressiveness. The purpose of this study was to determine whether encoding-decoding and extroversion relationships are already present in young children (Field & Walden, 1982; Walden & Field, 1982). The same children who had participated in Studies 1 and 2 were videotaped in this study emitting the following expressions: happy, sad, angry, surprised, afraid, interested, disgusted, and ashamed. The expressions were elicited under four different conditions to assess the children’s differential ability in producing facial expressions given different prompts. The conditions were: (1) a request to exhibit an expression identified only verbally; (2) a request to imitate a photograph of a child exhibiting the expression; (3) a request to imitate the same photograph in front of a mirror; and (4)a request to imitate the photograph, with verbal identification of the expression (see Fig. 8 for the set of photographs used). These conditions were counterbalanced using a Latin square design. Adult coders were shown the photographed models and then asked to guess the expressions appearing on the videotapes. The coders were naive to the conditions, expressions, and hypotheses of the study. The same children also rated their own expressions by viewing the videotapes of their expressions and pointing to the model photographs of each expression they had produced. The adult “guesses” were treated as the child’s encoding score and the child’s guesses as the decoding score. These then were compared across conditions and expressions in a repeated measures analysis of variance design. Each child’s encoding and decoding accuracy scores were entered into a correlation analysis with other measures such as child age, sex, and IQ. In addition, each child’s spontaneous expressions were observed during classroom free-play sessions to assess the degree to which the production and discrimination of these expressions in the laboratory are related to the frequency with which these expressions occur spontaneously. Finally, the children’s teachers completed the children’s version of the Buck Affect Expression Rating Scale of extroversion-introversion. Both teachers and children completed

20 I

Fciciczl Expressions

sociometric and affect ratings on their peers. These measures were also included in the correlation analysis. In the laboratory production task adult and child guesses were more accurate when the child had been requested to imitate photographs of other children’s expressions and least accurate when they were provided only with labels of expressions (see Fig. 9 for performance across conditions). The happy expres-

Irni tat ion

mitation n mirror

Irni tat i o n labels

Labels

&

C h i l d Judge

F i g . 9. Moan numlwr ofrrrors mode, by child in jrtdKirig his o r her own c~xpressionstind bv udult judging child s expressions under vcirioirs prompr conditions.

Adult Judge

%

Happy

C h i l d Judge ( s e l f )

dd Ii Sad

Ashamed

Surprised

Interested

Disgusted

Angry

Afraid

Fig. 10. Mean number of errors made by child in judging his or her own expressions and by adult judging child's expressions as a function of c p e of expression.

sion was the most accurately produced and discriminated and the afraid expression the least (see Fig. 10 for differential accuracy across expressions). The children’s and adult’s guesses were highly related (.72), suggesting that children who were good producers were also good judges. Children who were better producers in the laboratory displayed these expressions more frequently during

W

W cz

80

.-

70

--

60

--

so

--

40

.-

30

--

20

--

lo

..

LL

LL

0 W

W

a

I-

z

W u

e W

Neu t r d 1

Inter-

ested

Sad

5

Surpr i s e d

Afraid

Disgusted

Ashamed

E X P R E S S I O N S

Fig. I I . Meun proportion offree ploy intrrciction rime in which birsic e.kpri~s.rion.s were o6srrwd.

204

TiJuny M . Field and Tedru A . Wulden

free-play sessions (except for ashamed, which had a zero incidence of occurrence), See Fig. 11 for the incidence of these expressions during spontaneous play. In addition, those children who were more accurate producers were rated as more extroverted by their teachers (.42). Finally, sex was correlated with the ability to judge facial expressions (.57), with females being more accurate, and IQ was correlated with correct productions ( S 2 ) and extroversion ratings (.53). Children who were more extroverted were better producers (.46)and discriminators (.42). These children were also rated by their teachers as having more positive affect and as being more popular among their peers. These results are consistent with the literature on adult expressions. The relationships among production, discrimination, extroversion, and popularity are difficult to interpret. The child who is more expressive may more readily recognize expressions in other children and respond accordingly to those expressions. The children who are better encoders and decoders and more extroverted may have responded more contingently and more appropriately to their peers’ affective expressions. Frequent spontaneous and responsive expressions during classroom play may contribute to the child’s popularity and may be perceived as extroversion by the teacher. Imitated expressions are more accurately categorized than expressions elicited by verbal labels. Happy expressions are more accurately produced than fearful expressions. These results may be explained by the child’s experience in learning expressions by imitation and the child’s greater exposure to happy than fearful expressions. The early appearance of these encoding/decoding, extroversion relationships suggests that the socialization influences that may affect these skills occur at a very early age.

V.

An Integration

To integrate these studies with the other findings reviewed, we shall return to the original questions taken from the literature on adult facial expressions: 1. Can the basic eight facial expressions be accurately produced and reliably discriminated at an early age? 2. Are production and discrimination or encoding and decoding of facial expressions related processes during early development? 3. Do individual differences in these processes that have been described by an externalizing-internalizing dimension appear during early childhood? The basic facial expressions (except for anger and shame) can be elicited as early as the neonatal stage, as was shown in our neonatal study (Field et ul., I98 1) and in studies by Herzka (1965) and Steiner (1980). Although the processes underlying early mimicry or imitation are not well understood, the infant’s

Facirrl Expressions

205

facial responses to at least the three expressions (happy, sad, and surprised) modeled in our study (Field er al., 1981) suggest a very early ability to accurately replicate facial expressions. The habituation-dishabituation of these three facial expressions (happy, sad, and surprised) during our trials-to-criterion procedure suggests that neonates can discriminate among these expressions. In addition, their differential scanning patterns suggest that they may be detecting some of the salient features of these expressions. The longer looking time in the case of the smile is reminiscent of the findings on older infants looking longer at happy faces (LaBarbera et ul., 1976; Nelson et al., 1979) and may simply relate to the greater familiarity of the happy face, even at the neonatal stage. The greater number of trials to criterion for the surprised face may mean that this expression elicits a defense-like response, which is generally habituated slowly (Nelson et a / . , 1979). Research on older infants has revealed that negative expressions (e.g., afraid, angry) are less accurately produced than the positive expressions (Hiatt et a / ., 1979). In addition, our data (Field & Greenberg, 1981) suggest that 8-week-olds emit approximately equal numbers of happy, interested, and sad expressions during spontaneous face-to-face interactions with their mothers, and 12-weekolds emit approximately twice the number of happy as interested and sad expressions, surprised expressions very infrequently occur, and angry, afraid, and ashamed expressions never occur. Very much the same distribution is noted for the spontaneous play behaviors of preschool children, with happy and interested expressions occurring frequently and afraid, angry, and ashamed expressions occurring very rarely, if at all (Field & Walden, 1981a, 1981b). In addition, the accuracy with which the latter, less frequent expressions can be judged by adults or by the children themselves is much lower than the accuracy with which the more frequent expressions (happy, sad, interested) are judged. Accurate judging occurs whether these expressions are posed under more optimal conditions such as imitation of a photograph or less optimal conditions such as provision of only a label. Better performance in simple matching of line drawings is also characteristic of preschoolers when given happy and sad faces exemplars as opposed to surprised and angry faces (Walden & Field, in preparation). These results may be attributed to the child’s learning expressions via imitation of adult models, who emit happy expressions more frequently than afraid expressions and who provide more reinforcement for positive than negative expressions. Mothers, for example, are more frequently noted to imitate the positive than the negative expressions of their very young infants (Field, 1977; Pawlby, 1977), and the negative facial expressions of children are more inhibited than positive expressions in the presence as opposed to the absence of an adult examiner (Yarczower et al., 1979). Although the combined evidence from separate studies on the production and discrimination of facial expressions might suggest a relationship between these skills, additional studies of both of these skills in the same children are needed.

206

Tjfliiny M . Field m d Tedra A . Wrilden

Our studies of infants and preschool children provide tentative evidence that these skills are related, such that children who are good at posing expressions are also good decoders or discriminators, but given the mixed findings in the adult literature, these results need replication. In the meantime, Tomkins and McCarter’s (1964, p. 128) statement that . . . the face one sees is not so different from the face one lives behind” remains a matter to be further investigated. Although our studies of neonates and young infants provide evidence for accurate production and discrimination of at least some of the basic facial expressions, our studies and those of others (e.g., Buck, 1977; Odom & Lemond, 1972) demonstrate that accuracy increases with age. We found that older children selected more accurate matches in discrimination tasks but were not necessarily more accurate producers of posed expressions. Buck (1977), however, found that females’ spontaneous expressiveness increased with age, but males showed decreasing expressiveness as they grew older. Buck attributed this finding to sextyped socialization experiences in which boys typically learn to inhibit overt expressions of emotion, whereas girls continue to respond relatively freely. The final question derived from the adult literature was whether individual differences in facial expressivity and physiological reactivity can be observed in early childhood. Individual differences in facial expressiveness, physiological reactivity, stimulus thresholds, and habituation were noted in our neonatal study. The neonates who showed relatively greater physiological lability were characterized by low thresholds of responsiveness to stimulation, less behavioral expressiveness and faster habituation. The neonates who showed less physiological lability tended to display higher stimulus thresholds, greater expressiveness, and slower habituation. In addition, preschool children who were more expressive during free play were more accurate producers of facial expressions in our laboratory study and were rated as “externalizers” by teachers on the Buck Affect Expression Rating Scale. This pattern of individual differences in neonatal expressivity-responsivity and preschool expressivity-externalizer personality traits appears to parallel the externalizing-internalizing personality dimension proposed by Jones (1960) and Buck (1977) for older children and adults. Eysenck (1967) had suggested that the differences that he noted between externally and internally reactive adults may be present as early as birth. Although our data suggest differences between neonates on the degree to which their reactions to stimuli are expressed facially or physiologically or both, these data require replication and a longitudinal assessment of the continuity of these styles of expressiveness. Just as temperamental characteristics may not be stable, these styles of expression may be modified by socialization experiences. Jones ( I 960) and Buck (1 975) have proposed that internalizer-externalizer personality characteristics derive from socialization experiences. If overt expression elicits social sanctions, expression will be inhibited, and as Jones (1960) suggested, the use of covert, internal methods of affective expression may be “

( )

/

lnternalizer Low T h r e s h o l d High Physiological

I

PREDOMINANT SOCl ALl ZING ENVIRONMENT

\

ALIZER INTERN

(

I

/

\

Low E x p r e s s i v i t y Fast H a b i t u a t i o n

/

\

;1

\

CHILD

]

lnternalizer -

,

Reactivity Low E x p r e s s i v i t y Poor Decoding I nt r o v e r s i o n

\

/

\

\

/

\ \

I

-Physiological

GENERALIZER

/

High Threshold Low P h y s i o l o g i c a l Reactivity High Expressivity Slow H a b i t u a t i o n

\

Externalizer

0

OW R ePh a c rsi ivoi tl o ygical

_-_------

EXTERNALIZER Good Decoding

Fig. 12. Hypothesized transactional model of individual differences in facial expressivie and physiological reactivie at birth. differences in socialization environment (parents), and early childhood differences in encoding. decoding. and extroivrsion introversion traits based on the child arid adult models of Jones (IY60), Buck (1977). and € v e n d (1967). Some of the potential pathways are depicted by arrowed lines starting &om neonatal predispositions (solid line for internulizer, double line ,for generalizer, and broken line for externalizer).

208

Tiflkny M . Field crnd Tedru A . Walden

used. Combining the innatist position of Eysenck and the socialization influences proposed by Buck and Jones, we would speculate that a predisposition to be more facially than physiologically reactive or vice versa may be apparent at birth, but that socialization experiences may function either to reinforce or to modify those patterns of responsivity. Thus, an individual who might be classified as both an internalizer, based on the adult criterion of high stimulus thresholds, high physiological responsivity, low facial expressiveness, and fast habituation (Eysenck, 1967), may experience the socializing influences of parents who are externalizers. As a child, then, that individual may show both facial and autonomic responsivity , much like 60% of the adult population, the generalizers, are reputed to show. In Fig. 12 we have incorporated the neonatal predisposition to these types of responsivity (Eysenck, 1967), the potential socialization influences (Buck, 1975; Jones, 1960), that is, the types of responsivity of parents, and the types observed in childhood. The potential pathways by which neonates of one type may change or remain the same are depicted by arrowed lines. If externalizer-generalizer-internalizertypes of individuals are fairly evenly distributed at birth, as was suggested by our preliminary neonatal data (Field et al., 1981), yet 60% of the adult population are categorized as generalizers (Eysenck, 1967), we may need to test a transactional model of this kind, measuring the heretofore untapped socialization influences, to understand those changes, The questions of production and perception of facial expressions, their relationships, and the origins of individual differences derived from our literature search and studies outnumber the answers. Clearly, the data are tentative, and “adultomorphizing” just as “anthropomorphizing” is a tenuous process. As expressed in our introduction, we need to conduct longitudinal studies on the concurrent development of facial expression and discrimination processes, the relationships between these processes, and the predispositions and socialization influences that may contribute to individual differences in the production and perception of facial expressions. REFERENCES Bassili, J. Emotion recognition: The role of facial movement and the relative importance of upper and lower areas of the face. Journal ($Personality and Social Psychology, 1979, 37,2049-2058. Block, J. Studies in the phenomenology of emotions. Journal ofAbnorrnal and Socia/ Psychology, 1957, 54, 358-363. Borke, H . The development of empathy in Chinese and American children between three and six years of age: A cross cultural study. Developmenfa1Psychology, 1973, 9, 102-108. Brazelton, T . B . Neonatal behavioral assessmenf scrile. London: Spastics International Medical Publications, 1973. Buck, R. Nonverbal communication of affect in children. Journal of Personality and Social Psychology, 1975, 31, 644-653.

F { ~ ~ i iExpressions il

209

Buck, R. Nonverbal communication of affect in preschool children: Relationships with personality and skin conductance. Journal of Personality and Social Psychology, 1977. 35, 225-236. Buck, R., Miller, R . , & Caul, W. Sex, personality and physiological variables in the communication of affect via facial expression. Journal of Personality and Social Psychology. 1974, 30, 587596. Buck, R., Savin, J . V., Miller, E. R., & Caul, F. Communication of affect through facial expressions in humans. Journul of Personality and Social Psychology, 1972, 23, 362-371. Campos. J . J . , Emde, R. N., Gaensbauer, T., & Henderson, C. Cardiac and behavioral interrelationships in the reactions of infants to strangers. Developmental Psychology, 1975, I I , 589-601. Carey, S . , Diamond, R., & Woods, B. Development of face recognition: A maturational component? Dtvelopmenrul Psychology. 1980, 16, 257-269. Charlesworth, W. R . , & Kreutzer, M. A. Facial expressions of infants and children. In P. Ekman (Ed.), Dunvin und,facia/e.rpression. New York Academic Press, 1973. Chikvishvili, L.. Valsiner, J . , & Lasn, M.On the experimental investigation of emotion categories in two languages. Tartu University Studies in Psychology, 1977, 5 , 25-26. Daly, E., Abrarnovitch, R., & Pliner, P. The relationship between mothers’ encoding and their children’s decoding of facial expressions of emotion. Merrill-Palmer Quarrerly, 1980, M, 25-33. Darwin, C. A biographical sketch of an infant. Mind, 1877, 7 , 285-294. Diamond, R., & Carey, S . Developmental changes in the representation of faces. /ourno/ uf Experim e n d Child Psychology. 1977, 23, 1-22. Ekman, P. Universals and cultural differences in facial expressions of emotion. In J. K. Cole (Ed.), Nebraska Symposium on Motivation (Vol. 19). Lincoln: Univ. of Nebraska Press, 1972. Ekman, P., & Friesen, W. V. Constants across cultures in the face and emotion. Journal of Personality and Sociul Psychology. 1971, 17, 124- 129. Ekman, P., & Friesen, W. V. Unmusking the,fac*e.Englewood Cliffs, N.J.: Prentice-Hall, 1975. Ekman, P., Friesen, W. V . , & Ellsworth, P. Emotions in the human face: Guidelines for research and integrarion of findings. Oxford: Pergamon, 1972. Emde, R. N., Kligman, D. H., Reich, J . H., & Wade, T. D. Emotional expression in infancy: I. Initial studies of social signaling and an emergent model. In M. Lewis & L. A. Rosenblum (Eds.), The development of affect. New York: Plenum, 1978. Eysenck. H. J . The biological busis of personality. Springfield, Ill.: Thomas, 1967. Eysenck, H. J . , & Lewis, A. Conditioning, introversion-extraversion and the strength of the nervous system. In V. D. Neblitsyn & I . A. Gray (Eds.), Biological bases of individual behavior. New York: Academic Press, 1972. Feshbach. N . , & Roe, K . Empathy in six- and seven-year-olds. Child Development, 1968, 39, 133-145. Field, T. Effects of early separation, interactive deficits and experimental manipulations on infantmother face-to-face interaction. Child Development, 1977, 48, 763-771. Field, T.,& Greenberg. R. Facial expressions of 2- and 3-month-o/dinfants during face-to-face interactions. Manuscript in preparation, 1981. Field, T., & Walden, T. Production and discrimination of facial expressions by preschool children. Child Development, 1982, in press. Field, T.,Woodson, R., Greenberg, R., & Cohen, D. Discrimination and imitarion offucial expressions by neonates. Unpublished manuscript, University of Miami, 1981. Gitter, G., Mostofsky, D., & Quincy, A. Race and sex differences in the child’s perception of emotion. Child Development, 1971, 42, 2071 -2075. Goren, C. C., Sarty, M., & WU, P. Y. K. Visual following and pattern discrimination of face-like stimuli by newborn infants. Pediatrics, 1975, 56, 544-549.

210

Tifany M . Field wid Tedru A . Wulden

Gray, J. A. The psychophysiological nature of introversion-extraversion: A modification of Eysenck’s theory. In V. D. Neblitsyn & J . A. Gray (Eds.), Biological bases of individual behavior. New York: Academic Press, 1972. Guillaume, P. Imitution in rhildren (E.P. Halperin, trans.). Chicago: Univ. of Chicago Press, 1971. (Originally published, 1926.) Herzka, H. S. Dus Gesicht des Sauglings: Ausdruck und Reifung. BaseUStuttgart: Schwabe, 1965. Hiatt, S. W . , Campos, J . , & Emde, R. N. Facial patterning and infant emotional expression: Happiness, surprise and fear. Child Development, 1979, 50, 1020- 1035. Izard, C. E. Face of emotion. New York: Appleton, 1971. Izard, C. E., Heubner, R. R., Risser, D., McGinnes, G. C . , & DougheIty, L. M. The young infant’s ability to produce discrete emotion expressions. Developmental Psychology, 1980, 16, 132140. Jones, H. E. The galvanic skin reflex as related to overt emotional expression. Child Development. 1930, I , 106-1 10. Jones, H. E. The study of patterns of emotional expression. In M. Reymert (Ed.), Feelings and emotions. New York: McGraw-Hill, 1950. Jones, H. E. The longitudinal method in the study of personality. In 1. lscoe. & H. W. Stevenson (Eds.), Personality development in children. Austin: Univ. of Texas Press, 1960. LaBarbera, 1. D., Izard, C. E., Vietze, P., & Parisi, S. A. Four- and six-month-old infants’ visual responses to joy, anger and neutral expressions. Child Development, 1976, 47, 535-538. Lanzetta, J., & Kleck, R. Encoding and decoding of nonverbal affect in humans. Juurnal of’ Personality und Social Psychology, 1970, 16, 12-19. Lewis, M., Brooks, J., & Haviland, J. Hearts and faces: A study in the measurement of emotion. In M. Lewis & L. Rosenblum (Eds.), The development of affect. New York: Plenum, 1978. Meltzoff, A. N . , & Moore, M. K. Imitation of facial and manual gestures by human neonates. Science, 1977, 198, 75-78. Nelson, C. A,, Morse, P. A,, & Leavitt, L. A. Recognition of facial expressions by seven-month-old infants. Child Development, 1979, 50, 1239-1242. Odom, R., & Lemond, C. Developmental differences in the perception and production of facial expressions. Child Development, 1972, 43, 359-369. Oster,, H. Facial expression and affect development. In M. Lewis & L. Rosenblum (Eds.), The development ofufecr. New York Plenum, 1978. Oster, H., & Ekman, P. Facial behavior in child development. In W. Collins (Ed.), Minnesota Symposia on Child Psychology (Vol. 11). Minneapolis: Univ. of Minnesota Press, 1978. Pawlby, S. Imitative interaction. In H. R. Schaffer (Ed.), Studies in mother-infant interaction. New York: Academic Press, 1977. Provost, A. M., & Decarie, G . T. Heart rate reactivity of 9- and 12-month-old infants showing specific emotions in natural settings. International Journal ofBehavior Development, 1979, 2 , 109- 120. Schachter, S. The interaction of cognitive and physiological determinants of emotional state. Advanres in Experimental Social Psychology, 1964, I, 49- 110. Sherman, M. The differentiation of emotional responses in infants. Journal of Comparative Psychol0g.y. 1927, 2, 265-284. Sherman, M. The differentiation of emotional responses in infants. Journal of Comparative Psychology. 1928, 8, 385-394. Sroufe, L. A., & Wunsch, J. P. The development of laughter in the first year of life. Child Development. 1972, 43, 1326-1344. Steiner, J . E. Human facial expressionsin response to taste and smell stimulation. Advances in Child Development and Behavior, 1979, 13, 257-295.

Facial Expressions

21 I

Steiner, J. E. The effect of congenital anomalies on innate facial responses to sensory stimuli (taste and smell). In S. Hare1 (Ed.), The at risk infont. Amsterdam: Excerpta Medica, 1980. Thompson, J. Development of facial expression in blind and seeing children. Archives of Psychology. 1941, 264, 1-47. Tomkins, S. S. Affect, imagery, consciousness (Vol. 2). The positive affects. New York: Springer Publ., 1962. Tomkins, S. S., & McCarter, R. What and where are the primary affects? Some evidence for a theory. Perceptual and Motor Skills, 1964, 18, 119-158. Walden, T., & Field, T. Discrimination of facial expressions by preschool children. Child Devekpmetif, 1982, in press. Waters, E., Matas, L., & Sroufe, L. A. Infants’ reactions to an approaching stranger: Description, validation and functional significance of wariness. Child Devrlopmenf, 1975, 46, 348-356. Wolff, P. Observations of the early development of smiling. In 9. M. Foss (Ed.), Determinants of infnnt behuvior If. London: Methuen, 1963. Yarczower, M., Kilbride, J . , & Hill, L. Imitation and inhibition of facial expression. Drvelupmental Psychology, 1979, 15, 453-454. Young-Browne, G., Rosenfeld, H. M., & Horowitz, F. D. Infant discrimination of facial expressions. Child Development, 1977, 48, 555-562. Zuckerman, M., Hall, J . A , , DeFrank, R. S., & Rosenthal, R. Encoding anddecodingofspontaneous and posed facial expressions. Journal of Personality and Social Psychology, 1976, 34, 966977. Zuckerman, M., Lipets, M . , Koivumaki, J., & Rosenthal, R. Encoding and decoding nonverbal cues of emotion. Journul of Personulity and Social Psychology, 1975. 32, 1068-1076.

This Page Intentionally Left Blank

INDIVIDUAL DIFFERENCES IN INFANT SOCIABILITY: THEIR ORIGINS AND IMPLICATIONS FOR COGNITIVE DEVELOPMENT

Michael E . Lamb DEPARTMENT OF PSYCHOLOGY UNIVERSITY OF UTAH SALT LAKE CITY. UTAH

I. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

213

11. MEASURING SOCIABILITY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

215

111. CORRELATIONS BETWEEN SOCIABILITY AND COGNITIVE PERFORMANCE , , , . . . . . . , . , . . , . , . , . . , . . , , . , . , . , , . . , . , . . , . , , , , . . . , , . . . A. TEST SOCIABILITY . , . , . , , . , . , , . , , . . , , , . . . , , . . , . . , . , . , , . . . . . , , . . . , B. SOCIABILITY OUTSIDE THE TEST SITUATION . . . . . . . . . . . . . . . . . . . . . . C . OLDER CHILDREN . . , , . , . , , . . . , , , . . , . , . , . , , . , , . . . , , . , . . . , , . , . , , , . D. SUMMARY.. . . . . , . . . , . , , . , , , . . . , , , , . _. . . , . , , . , , . . . . , . . . . . . , . , . , , .

222 222 224 225 226

IV. EXPLAINING THE RELATIONSHIP BETWEEN SOCIABILITY AND COGNITIVE PERFORMANCE . . , . . . . , . . . , , . , , . . , . . , , . , . , . , . . . . , . , . . , . . . A. REVIEW OF THE EVIDENCE.. , , , . . . . . , , . . , , . , , . , . . , . , . . , , , . , . , , . . . B. SUMMARY.. , , , . . . , , , . , , . . . . . , . , . . . , , _ _ ,. . , , . , , . . . , , . , , , . , . , . , , . .

226 226 230

V. ORIGINS O F INDIVIDUAL DIFFERENCES IN SOCIABILITY , . . . . . , . . . . . . . . A. BIOGENETIC INFLUENCES , , . , . , , , . , , . , . , . , . . , , . , . , . . , . . . , , . , . , . . . 9 . ENVIRONMENTAL INFLUENCES , , . , , , , . . , , . , . . , . . , . , . , , . . . . , , . , , . , C. SUMMARY,, . . _ ., , , . , . , , . . , , , . , . . , , . , , , , . , . . , , . . . . , . . . . . . . . . . . . . .

230 23 I 233 236

VI. CONCLUSION,, . , . . , . , . . , . . , . , . , , , , . . , , , , . . , , , , . , , . , . . . , , . , , . . . . , . . . .

236

REFERENCES . . , . , . , , . . . . , , , . . , , , . , , . . , , , . . , . , . . . , . . . . , , , , . , . . . , , . . . .

237

I. Introduction Most developmental psychologists consider themselves knowledgeable as concerns either social or cognitive development. This bifurcation has never been entirely satisfactory, particularly for students of infancy who ignore the essential 213 ADVANCES M CHILD DEVELOPMEM AND BEHAVIOR, VOL. 16

Copynghl @I982 by Academic Press. Inc. All rights of reproduction in any form reserved. ISBN 0-12-009716-8

214

Michuel E . Lumh

unity of development at their peril. In recent years, consequently, the traditional division has been breached somewhat, with developmental social cognition emerging as a new and popular topic for research and theorizing (Lamb & Sherrod, 1981; Shantz, 1975). In this article, I review evidence concerning one aspect of the interface between social and cognitive development that is not typically viewed as a facet of social cognition. As I suggest, there is now substantial evidence that the infant’s sociability or friendliness is correlated with measures of infant cognitive competence or performance, as well as suggestive evidence that the same relationship exists in preschoolers. Since little is known about older children, however, I concern myself solely with infancy in this article. My goal is to analyze this relationship between aspects of social and cognitive development with a view to determining what this relationship tells us about developmental processes in infancy and early childhood. Of course, psychologists have been aware of an apparent relationship between social experience and cognitive development at least since research on the devastating effects of “maternal deprivation” became widely known (Ainsworth, 1966; Bowlby, 1951). However, most of this research was interpreted so as to emphasize a linear influence of social experience on cognitive development: Institutionalized infants were believed to perform more poorly on measures of cognitive development because they were deprived of maternal care and thus of emotional and sensory stimulation. Intervention projects, whether or not they are successful, build upon a similar assumption: that enriched experiential histories foster accelerated cognitive development. In this article, a different sort of relationship between social and cognitive development is explored. Instead of asking whether social experience affects cognitive development, as I have done elsewhere (Stevenson & Lamb, 1981), I am concerned here with the way in which a trait, sociability, is related to individual differences in cognitive development. My analysis begins with the well-documented fact that sociable, friendly infants perform significantly better than less sociable infants on various measures of cognitive and psychomotor competence. Several possible interpretations of these findings come to mind. First, it is conceivable that the characteristic, perhaps endogenous, sociability of the child leads the child to elicit more than average amounts of attention from parents and caretakers, and so become more cognitively competent. Second, the sociable child may simply lead testers to be more persistent, so that the child achieves higher test scores even though it is no more cognitively competent than average. Third, sociability may be an epiphenomena1 correlate of cognitive development. The goal of this article is to evaluate the explanatory power of these hypotheses, as well as to define what sort of relationships exist, what it tells us about developmental processes, and what implications the relationship has for future research (basic as well as applied) on the nature and determinants of early cognitive development.

Individuul Di#crenc.es in Injunt Sociubiliw

215

In the next section of this article, I discuss the assessment of sociability and the measurement problems that plague much of the research in this area. I also suggest ways in which some of the more obvious problems can be avoided in future research. Then I review the basic evidence in some detail (Section 111). In the fourth section, I consider the interpretation of this relationship: Is there a causal relationship between sociability and cognitive performance mediated via the infant’s willingness to interact with a strange examiner, or is the infant’s actual level of competence involved? If the latter, just how are sociability and cognitive competence related? In the fifth section, finally, I discuss recent attempts to explore the origins of individual differences in infant sociability. Throughout this article, the term “sociability” refers to the friendliness and social attractiveness of the infant. It describes the extent to which others find interaction with an infant enjoyable and enticing. The assessment of sociability is usually based on the tester’s perception or on a third person’s perception of the infant’s friendliness toward the tester. Only occasionally (and recently) has sociability with unfamiliar people other than testers been assessed. Often, the baby’s behavior toward its mother during testing is also recorded by the examiner. Finally, the parents’ perceptions are often tapped by measures of infant temperament.

11.

Measuring Sociability

The fact that aspects of infant social and emotional behavior should be taken into account when evaluating cognitive development was recognized a half century ago, when Bayley developed her assessment scales (Bayley, 1933a, 1933b). In addition to the Mental and Motor Development Indices, Bayley developed the Infant Behavior Record, which comprises a set of rating scales on which the examiner was supposed to record his or her impressions of the infant’s behavior during the test session. Most of these scales had to do with the infant’s emotional state, motivation, social behavior, and “interest in specific modes of sensory experience” (Bayley, 1969, p. 99). The 24 scales included in the 1969 revision of the Infant Behavior Record are listed in Table I . I Four of these items directly assess the infant’s sociability with the tester: responsiveness to examiner, cooperativeness, fearfulness, and general emotional tone. The baby is assigned a score of between 1 and 9 on all but four scales. Scores of 1 define the negative end of each scale and are given to unsociable, uncooperative, inattentive, or crying babies, whereas scores at the other extreme are given to very outgoing, cooperative, eager, or happy babies. Two of the scales are bimodally scored. ‘Twenty-four scales are psychometrically described in the 1969 Manual. Six others appear on the score sheets only, which is why some sources refer to the Infant Behavior Record as having 30 items.

216

Michurl E . Lamb

TABLE I The Infant Behavior Record

Behavior Responsiveness to persons Responsiveness to examiner Responsiveness to mother Cooperativeness Fearfulness Tension General emotional tone Responsiveness to objects Imaginative play with materials Attachment to objects Goal directedness Attention span

Number of points on scale 9 5

5 9

9 9 9 9 2 2 9 9

Number of points on scale

Behavior Endurance Activity Reactivity Sights-looking Listening to sounds Producing sounds-vocal Producing sounds-banging Manipulating Body motion Mouthing or sucking-thumbs fingers Mouthing or sucking-pacifier Mouthing or sucking-toys

9 9 9 9 9 9 9 9 9 or 9

9 9

"From Bayley (1969).

Perhaps because Bayley herself did not accord the Infant Behavior Record the same importance and attention that she paid to the other indices in her assessment battery, did not employ an equivalently large standardization sample, and never attempted to establish psychometric properties as precisely as with the Motor and Mental Indices, the Infant Behavior Record has never'been studied as widely as the Mental Index, which is the most commonly employed measure of cognitive development. The 1969 Manual, for example, does not include definitions of the behaviors rated on the Infant Behavior Record, making it possible for different examiners to use the scales in different, rather idiosyncratic, ways. In the last decade, and especially within the last few years, however, several researchers have examined the relationship between social behavior and the Mental Index more seriously. Most of the research on the relationship between sociability and cognitive development is based on correlations between performance on the Mental Index and ratings on the Infant Behavior Record or similar sets of scales. (Many of the studies reviewed in this article involved early versions of the Infant Behavior Record, which has since been revised quite extensively by Bayley.) Consequently, it is important for readers to know how these items are rated and what problems exist with such means of assessing sociability. Noting the unavailability of substantial information concerning the reliability of the Infant Behavior Record, Matheny, Dolan, and Wilson (1976) investigated reliability systematically using data from the Louisville Twin Study. For these analyses, two investigators independently rated 96 infants on all the Infant Be-

Individual Differences

it7

Infant Sociability

217

havior Record scales at 3, 6, 9, 12, 18, 24, and 30 months of age. Agreement ranged from 67 to loo%, with a median of 87%; it was lowest for the ratings of fearfulness and sociability with people, and highest for the ratings of sociability with mother and stranger. Presumably, the unreliability of the measure of social responsiveness to people stems from the fact that it assesses social responsiveness with mother and examiner jointly even though these two constructs (as assessed on the Infant Behavior Record) are negatively related to one another (Dolan, Matheny, & Wilson, 1974). Other measures of sociability, such as Stevenson and Lamb’s (1979; see also Thompson & Lamb, 1981), have high interrater reliability-around 90%. For the most part, therefore, measurement unreliability is apparently not a major problem for researchers. The chief problem with the Infant Behavior Record and ratings of this sort, as they are typically employed, is that the ratings are made by the same person who is responsible for assessing the child’s cognitive competence. This nonindependence must raise suspicion about any reported relationship between the scales, as the potential for halo effects is high. Interestingly, however, the highest reported relationship between Infant Behavior Record scales and Mental Index scores was obtained in a study in which careful efforts were taken to ensure complete independence of measures by having the ratings of sociability and the Mental Index completed by different assistants (Stevenson & Lamb, 1979). Furthermore, since testers are usually unaware of any hypotheses regarding the relationship between the Infant Behavior Record and the Mental Index, it is unlikely that ratings are assigned in accordance with implicit hypotheses or expectations. In future, efforts should be made to gather the two types of evidence independently, but the nonindependence does not appear to have seriously compromised the data gathered thus far. Second, the unidimensionality or ordinality of some of the scales can be questioned. This is an especial problem with versions of the Infant Behavior Record in use before the 1969 Manual for the Bayley scales was published. Results based on analyses of the Berkeley Growth Study (Bayley, 1968; Bayley & Schaefer, 1964; Crano, 1977), and the National Collaborative Perinatal Project (Lamb, Garn, & Keating, 1981a, 1981b), among other studies, are all based on these earlier versions of the Infant Behavior Record. The problems are readily evident. Consider, for example, the scale, “social responsiveness to mother, used in the National Collaborative Perinatal Project. The scale points are: ”

1.

2. 3. 4. 5.

“Ignores mother during free play, resists contact with mother”; “Hesitates, cooperates in certain tests”; “Accepts, responds adequately to assistance by mother”; “Enjoys contact with mother during testing”; “Demands, clings to mother.”

218

Michiiel E . Lnrnb

The relevance of scale point 2 is obscure, and 5 does not appear to represent highly sociable behavior because such a baby might be considered anxious or insecure. Problems like these limit the face validity of the Infant Behavior Record for assessing sociability, and ensure that any results obtained using the Infant Behavior Record are conservative underestimates. The usefulness of the Infant Behavior Record measures is also compromised in that the range of obtained scores (at least in the standardization sample) is extremely limited (see Bayley, 1969). It is thus not surprising that studies using the Infant Behavior Record have obtained weaker findings than those obtained in studies employing better-constructed measures of sociability. Convinced that sociability was indeed an important construct but dissatisfied with the psychometric properties of the instruments then available, Stevenson and Lamb ( 1 979) developed two new measures of sociability-one designed to assess sociability within, and the other sociability outside, test situations. Our measure of test sociability was based on three scales from the Infant Behavior Record: responsiveness to the examiner, cooperativeness, and general emotional tone. We converted these all to 9-point scales, specifying each point as clearly as possible so as to ensure reliability, while also guaranteeing an effective range of scores so as to buttress the sensitivity and usefulness of the instrument. Stevenson and I reported a very respectable interrater reliability coefficient of .83. In the standardization sample of 40 infants (each tested twice), the revised measure of test sociability yielded a mean of 15, a standard deviation of 3.5, and scores ranging from 6 to 22 out of a possible 3 to 27 (see Stevenson, 1978, p. 117). One-day test-retest correlations (assessed using Pearson product-moment correlations) were .50. One-year test-retest correlations in a subsample were not significant (Stevenson, 198 I). The scales themselves are reproduced in Table 11. In addition to this revised procedure for the measurement of sociability during testing, Stevenson and Lamb ( I 979) developed an alternative measure of infant sociability, this one designed to assess the response to a strange adult outside the testing situation. (Stevenson and I referred to this as “initial sociability,” and I shall follow their lead in this regard.) The procedure was modified in several ways by Ross Thompson before inclusion in our ongoing work; the changes were designed to improve the sensitivity of the instrument without changing the essence of the measure or the construct it purports to measure.2 Stevenson and Lamb (1979) found that our measure of “initial” sociability was correlated ( r = .46, N = 40) with measures of sociability during two test sessions, even when the assessments were 1-3 days apart. Both measures thus appeared to tap the same dimension. One-day test-retest correlations were .73 for the measure of initial sociability; the correlations between initial sociability at home and in the laboratory on adjacent days were .49 and .40 (Stevenson & Lamb, 1979). In a follow-up study, Stevenson (198 1) reported no significant correlation between *The two scoring systems yield highly correlated scores ( r = .93, N = 19).

Individual Differences in Infant Sociahilit?,

219

TABLE I1 Sociability during Testing “ Social orieriration: responsiveness to examiner I . Avoiding or w i t h d r a w 2. Between I and 3 3 . Hesitant 4. Between 3 and 5 5. Accepting 6. Between 5 and 7 7. Friendly 8. Between 7 and 9 9. /in*iring (initiating, demanding) Cooperativeness: cooperation with examiner, based on interpersonal reactions 1 . Resists all suggestions or requests 2. Does not cooperate 3. Refuses 10 cooperate during part of the session; refuses or resists one or two specific tests, or refuses to attempt the more difficult items he or she is likely to fail 4. Between 3 and 5 5 . Responds to o r accepts the test materials or situations; neither cooperative nor resistant in relation to examiner 6. Between 5 and 7 7. Seeins to enjoy the give-and-take with the examiner in the testing situation 8. Between 7 and 9 9. Very readily arid enthusicrstically eriters irrto suggested games or tusks Ceireral ernotional totie: degree of happiness I . Child seems unhappy throughout the testing period 2. Between 1 and 3 3. At times rather unhappy. but may respond happily to interesting procedures 4. Between 3 and 5 5 . Moderately happy or contented; may become upset, but recovers fairly easily 6. Between 5 and 7 7. Gerierally appears to he iri a happy state of well being 8. Between 7 and 9 9. Radiates happiriess; nothing is upsetting; animated “From Stevenson (1978. p. 88)

infants’ sociability at 1 and 2 years of age. One possible explanation is suggested by the findings of a recent study by Thompson and Lamb (1981) in which we found that initial sociability was highly stable over a 7-month period when the quality of attachment was stable ( r = .74), but when the security of motherinfant attachment changed, so too did sociability (test-retest correlation = .18). In this study, security of attachment changed when major changes in family circumstances (employment, unemployment, new baby born, move house) took place (Thompson, Lamb, & Estes, 1982). Interestingly, Stevenson (1981) found

220

Mirhurl E . Lamb

TABLE Ill Procedure for Assessing Sociability outside the Test Situation The Sociability Assessment has been designed to appraise an infant’s reaction to an initial encounter with an unfamiliar female adult stranger with particular attention to components of wariness and affiliation which characterize the baby’s responses to the stranger’s social bids. It consists of a standardized 5-minute laboratory episode in which the stranger makes a variety of social bids to the infant, at first when the baby is on mother’s lap, and later when he or she is on the floor. Finally, it culminates in the stranger picking up the infant. Other components of the infant’s response which are scored include the baby’s initial reaction to the stranger upon her entrance into the room, and the infant’s response to the stranger’s departure at the end of the assessment. ( I ) Buhy’s iniriul reaction to the strarcger 1 . Cries, becomes quite upset 2. Whimpers or fusses 3. Turns to mother or away from the stranger, sometimes with a coy smile 4. Looks at the stranger intently without smiling 5. Smiles at the stranger

(2) Baby’s initiol reaction l o being ciflerrd the keys by stranger oti mother’s lap (score first 10 seconds following initiation of bid) I . Fusses, cries or turns toward mother 2. Refuses keys by looking away, pushing them away, turning toward a toy, or using words (e.g., “No!“) 3. Looks at keys without reaching for them 4. Tentatively reaches for or touches keys 5. Accepts keys from stranger with little hesitation ( 3 ) Bahjl’s reaction to the stranger’s initiatic~noj’a turn-taking game on titother’s lap (score first 30 seconds following initiation of bid. or two attempts to get a game started) 1 . Crying, fussing or other indications of distress 2. Refusal to join in by turning toward mother or another toy, turning away from stranger, just looking at stranger, or using words (e.g., “No!”) 3. Initially reluctant, then participates 4. Immediately joins in and participates with stranger 5 . Actively participates in game with stranger by smiling, initiating exchanges, anticipating turn taking, changing the game, acting playfully or teasing the stranger, etc. (4) Baby’s behuvior whetz givetz floor freedom 1 . Touches mother or requests to return to lap

2. Approaches or turns toward mother 3. Stays where he or she is or moves away from mother to play with toys 4. Approaches stranger 5. Touches stranger or requests to be picked up ( 5 ) Ruby‘s itlitid reuction to being oflercd a to~ybv strungi’r on poor (score first 10 seconds following initiation of bid) 1. Fusses. cries or turns toward mother 2. Refuses toy by looking away, pushing it away, turning toward a different toy, or using words (e.g., “No!”) (cotitinurd)

Indi\*idical Differences in Infilnr Sociubility

221

TABLE 111 (cvntinurd) ~~

~~~~~~~~~~~~~~

~~~~~

~

3 . Looks at toy without reaching for it 4. Tentatively reaches for or touches toy 5. Accepts toy from stranger with little hesitation (6) Baby’s reaction to the stranger‘s initiation of a turn-raking game on the floor (score first 30 seconds following initiation of bid, or two attempts to get a game started) I . Crying, fussing or other indications of distress 2 . Refusal to join in by turning toward mother or another toy, turning away from stranger. just looking at stranger, or using words (e.g., ”No!”) 3 , Initially reluctant. then participates 4. Immediately joins in and participates with stranger 5 . Actively participates in game with stranger by smiling, initiating exchanges, anticipating turn taking, changing the game, acting playfully or teasing the stranger, etc.

(7) Baby’s reaction when the stranger attempts u pick up I. Immediately cries, becomes quite upset 2. Whimpers or fusses; milder distress reaction 3. Tries to get down, wiggles in stranger’s arms, or turns and reaches toward mother 4. Minimal acceptance of pick up, but baby turns away from stranger, avoids eye contact, and/or does not mold to her 5 . Positive acceptance of pick up, with molding, sometimes smiling, etc. ( 8 ) Baby’s reaction to the stranger’s departure 1 . Returns to mother as stranger walks o u t of room, or clings to her 2. Ignores stranger’s departure entirely 3 . Looks up fleetingly as stranger leaves the room 4. Sustained visual regard of the stranger as she leaves 5. Smiles at stranger, waves or says ”bye-bye,” follows stranger to the door, etc.

(9) Owrall Cnprrssion of sociability I . Quite unfriendly, fussy, fearful 2. Between 1 and 3 3 . Generally unfriendly, serious, wary 4 . Between 3 and 5 5 . Neutral, neither friendly nor unfriendly 6. Between 5 and 7 7. Friendly, positive reaction 8. Between 7 and 9 9. Very friendly, outgoing, smiling

the same relationships between sociability and Bayley Test performance among both 1- and 2-year-olds. The measure of sociability outside the test situation has the advantage of being brief and unobtrusive, and so it can be employed frequently in a diverse array of situations. Consequently, in our ongoing attempts to explore the correlates of sociability in other aspects of socioemotional development, we are using this

222

Michuel E . Lumb

measure in preference to the measures of sociibility during testing. The most recent version of the initial sociability measure is described in some detail in Table 111.

111. Correlations between Sociability and Cognitive

Performance A.

TEST SOCIABILITY

In an analysis of data from the Berkeley Growth Study, Bayley and Schaefer (1964) and Bayley (1968, 1969) reported correlations between items on the original Infant Behavior Record and both contemporaneous and subsequent measures of cognitive competence. “The patterns of r’s indicate that at concurrent ages (10-36 months) high mental scores go with extroverted adjusted ratings: that is, active babies who are responsive to persons and perhaps also those who tend to be calm” (Bayley & Schaefer, 1964, p. 44).This pattern of correlations was more characteristic of girls than of boys. In later testing, the sociable male infants performed more poorly: That is, there was a negative correlation between sociability during the first year and mental test scores in early childhood among boys. McCall, Hogarty, and Hurlburt (1972), in an analysis of data from the Fels Longitudinal Study, also reported that sociable infants tended to perform better than unsociable infants, but this result was based only on an item analysis; no ratings of sociability were employed. in a later reanalysis of data from the Berkeley Growth Study, Crano ( 1977) found significant positive correlations between Mental Index scores and ratings of sociability (based on the Infant Behavior Record) made a few months later. Using the logic of cross-lagged panel correlations, Crano concluded that variations in Mental Index performance were causally related to later social behavior rather than the reverse. Comparing correlations between the measures of sociability-“strangeness” (shy-unreserved), “responsiveness to persons” (slight-marked), “amount of positive behavior” (negative behavior-positive behavior), and “emotinnal tone (happy-unhappy)and both preceding and succeeding Mental Index scores, Crano found that only a chance proportion of the differences between correlations were significant, although most were in the direction indicating that Mental Index scores predict sociability rather than the reverse. Crano did not present correlations between contemporaneous measures of sociability and cognitive development, but the means of the lagged correlations with preceding and succeeding Mental Index scores were, respectively, . I 6 and .14 for responsiveness, .21 and .10 for strangeness, .19 and .07 for positive behavior, and .29 and -06 for emotional tone. ”

Individuul Di#erences in Injunt Sociubilie

223

Seegmiller and King ( 1 975) obtained Infant Behavior Record ratings and Mental Index scores for their sample at both 14 and 22 months. At 14 months, cooperation with the examiner and general emotional tone were both positively correlated with contemporaneous Mental scores. At 22 months, Mental Index scores were positively correlated with responsiveness to the examiner and general emotional tone. Unfortunately, Seegmiller and King did not combine these three into one composite scale and did not examine cross-lagged correlations in an attempt to elucidate the direction of effects. Main (1973) reported a significant positive correlation between 20-month Mental Index scores and sociability as assessed by cooperativeness and evidence of a “gamelike spirit” during testing. Using data from part ( N = 110) of the Louisville Twin Study, Matheny, Dolan, and Wilson (1974) summed the scores on three sociability ratings from the Infant Behavior Record (responsiveness to the examiner, cooperativeness, and general emotional tone) that were consistently correlated with Mental Index scores. The resulting composite “extroversion” scores were then correlated with contemporaneous Mental scores from assessments at 6, 12, 18, and 24 months. The sociability (extroversion) and Mental Index scores were significantly correlated at every age among girls, but only at 24 months among boys. In another report, Dolan ef al. (1 974) noted that although the intercorrelations tended to be significant, the components were less consistently intercorrelated, and the extroversion scores were less stable over time, than were scores on another composite measure, “primary cognition. Perhaps this pattern of correlations occurred because the primary cognition measure, composed of more items than extroversion, is simply more robust a measure as a result. Age-to-age correlations for extroversion were significant only between 18 and 24 months ( r = .31 for girls, r = .45 for boys). Using seven rating scales of their own design, Birns and Golden (1972) found significant correlations ( r ’ s = .33, .41) between cognitive performance and the infant’s cooperativeness with the examiner at both 18 and 24 months, but there was no correlation between cognitive scores and a rating of “shyness,” the only other scale tapping sociability. Cooperation at 24 months also predicted 36month Stanford-Binet scores ( r = .44) and this relationship remained significant when 24-month Cattell Infant Intelligence Scale scores were partialed out ( r = .26). Ramey and Campbell (1979) reported that Infant Behavior Record ratings of fearfulness were not significantly correlated with test performance at 6, 12, or 18 months of age. McGowan, Johnson, and Maxwell (1981), in a study of 125 lower-class Mexican-Americans, used the composite measure of extroversion developed by Matheny et al. (1974) and found significant correlations between extroversion and Mental Index scores at 12 months ( r = .52 for girls, r = .35 for boys), as well as between extroversion at 12 months and Mental Index scores at 24 months ”

224

Michurl E . Lamb

among girls (r = .37) but not among boys ( Y = .05). In a multiple regression analysis, however, sociability did not contribute to explanation of the variance at either 12 or 24 months. Sociability scores were significantly stable over time ( r = .46 for girls, r = -34 for boys). Finally, in the most recent reports on this topic, Lamb et al. (1981a, 1981b) examined the correlations between Infant Behavior Record ratings of sociability and performance on Bayley’s tests of mental and motor development using data from 33,000 8-month-old infants. Test performance scores were significantly and positively correlated with sociability regardless of race, social class, or sex. Although a few researchers have failed to demonstrate it, the clear conclusion to be drawn from these studies is that infants who are sociable during testing perform better on measures of cognitive competence than do infants who are less sociable. In all of the studies reviewed thus far, however, sociability items were rated by the examiner at the end of the test session, thus raising the possibility of halo effects. In our first study, Stevenson and Lamb (1979) had sociability rated independently by a hidden observer and still found remarkably high correlations between test sociability and scores on both the Bayley Mental Index ( r = .54) and the Uzgiris-Hunt Ordinal Scales3 (r = .51) among 1-year-olds. Consequently, the observed relationship between test sociability and cognitive performance cannot simply be attributed to a methodological flaw-nonindependence of sources of data. B.

SOCIABILITY OUTSIDE THE TEST SITUATION

Another group of researchers has examined sociability outside the test situation, reasoning that if sociability outside the test session is correlated with measures of cognitive competence, then sociability must be a transsituational and relatively enduring characteristic of the infant rather than a situationally bound phenomenon. Demonstration of such a relationship would increase the perceived psychological importance of the association between sociability and cognitive competence. In most of the relevant studies, the infant’s initial reactions to an unfamiliar adult have been recorded-hence the term “initial sociability. Clarke-Stewart (1 973) developed an approach sequence for the assessment of infant social responsiveness: During the sequence, the infant was looked at, smiled at, talked to, approached, touched, picked up, and finally left. The sequence was repeated four times, so that sociability toward familiar and unfamiliar adults, with and without the mother present was assessed. A composite measure of sociability during the four approach sequences was not significantly correlated with cognitive competence. ”

)The Ordinal Scales were designed by Uzgiris and Hunt (1975) to provide a psychonietrically sound assessment of those aspects of infant cognitive development emphasized by Piaget (1936/1952, I937/1954).

Individual Diflerences in lnfilnr Sociability

225

In a later study, Clarke-Stewart, Umeh, Snow, and Pederson (1980) found that sociability with a stranger (measured by the frequency of smiling, vocalizing, and play) was not correlated with Mental Index scores at 12, 18, and 24 months, although sociability with a stranger in two contexts outside the test situation was correlated with Mental index scores ( r ’ s = .30) at 30 months. Correlations among sociability scores obtained in different situations and with different persons (mothers and strangers) were generally low; they were highest for contemporaneous assessments of sociability with mothers and strangers in a highly structured procedure. These results may indicate that discrete behavioral counts do not constitute good measures of sociability, since we know from other research (e.g., Waters, 1978, 1980) that such measures are inevitably unreliable when sampling intervals are brief. Interestingly, sociability with mother was more consistently correlated with 1Q than sociability with strangers. In contrast to Clarke-Stewart (1973), Beckwith, Cohen, Kopp, Parmelee, and Marcy ( I 976) did indeed find a significant relationship between sociability during unstructured sequences and later test performance, although they explained it rather differently than I do here. During home observations of I-, 3-, and 8-month-olds and their caretakers, observers tallied (among many other behaviors) the number of 15-second time units in which the infant smiled at or vocalized to the observer. In regression analyses, the sociability scores at 8 months were significant predictors of performance on a measure of sensorimotor development (Kopp, Sigman, & Parmelee, 1974) at 9 months. Finally, using an earlier version of the scale described in Table 111, Stevenson and Lamb (1979) found that initial sociability was significantly correlated with performance on both the Bayley Mental Index ( r = .43) and the Uzgiris-Hunt Ordinal Scales ( r = .60) among I-year-olds. Stevenson (1 98 1) found a similar relationship when she retested these infants a year later. Like Bayley and Schaefer (1964), Stevenson and Lamb (1979) reported that correlations between sociability and cognitive performance were higher among girls than among boys. In sum, cognitive performance is indeed correlated with measures of sociability outside the test session. C . OLDER CHILDREN

As indicated earlier, there has been little relevant research involving older children-perhaps because there is no equivalent of the Infant Behavior Record attached to the popular test batteries, and because, with increasing age, group tests come to replace individual tests, making it impossible to assess sociability with an examiner. However, in an interesting experimental study, Zigler, Abelson, and Seitz (1973) showed that familiarity with the examiner and the test situation significantly affected the performance of disadvantaged preschoolers (i.e., children from low socioeconomic backgrounds) on the Peabody Picture Vocabulary Test, whereas these factors did not substantially affect the perfor-

226

M i h e 1 E . Lamb

mance of middle-class children. Zigler and colleagues also reviewed other evidence implicating motivational factors, particularly a wariness of strange adults, in the poor cognitive performance of disadvantaged children. However, Stevenson (1981) failed to replicate this finding in a study involving 1-year-olds, who were tested at home by their mothers and in the laboratory by an unfamiliar examiner. D.

SUMMARY

Although some researchers have failed to replicate the association, many other researchers have shown that infant sociability, whether measured inside or outside the test situation itself, is correlated with measures of infant cognitive performance. Interrelations between measures of sociability and cognitive performance taken at about the same time (i.e., within days of one another) are impressively high, ranging up to about .60. Because the stability of both sociability and cognitive performance is modest (though statistically significant) over longer periods of time, we would expect, as indeed appears to be the case, that correlations between sociability and cognitive performance are attentuated when the time intervening between the two assessments is of extended duration. Given that the relationship between sociability and cognitive performance is robust, it behooves us to determine how the relationship comes about. Competing explanations are the focus of the next section of this article.

IV. Explaining the Relationship between Sociability and Cognitive Performance A.

REVIEW OF THE EVIDENCE

There are three possible ways of explaining the relationship between infant sociability and cognitive performance. These explanations are not mutually exclusive, and each may account for some portion of the variance. At this stage, we can simply assess the relative explanatory power of the putative causal sequences, determining which are of substantial importance and which are trivial. The first hypothesis proposes that the relationship between social and cognitive development reflects a maturational association, with enhanced sociability (like cognitive precocity) being a sign of greater maturity. On the face of it, this explanation seems implausible: A wealth of research on infants’ reactions to strange adults indicates that during the first year infants became less friendly toward strangers, and that wariness of strangers remains characteristic throughout the second year of life (Batter & Davidson, 1979; Sroufe, 1977). However,

lndiviifrtal Differences in lnjunr Sociuhiliq

227

Clarke-Stewart et ul. (1980), in a longitudinal study of 60 infants observed at 12, 18, 24, and 30 months, found that sociability increased with age in both structured and unstructured situations. This finding adds credibility to the hypothesis. The second hypothesis is that sociability is correlated only with infant performance rather than with infant competence per se. This hypothesis implies that sociable, friendly, outgoing babies give a better impression of themselves in a test situation by virtue of their willingness to interact and cooperate with the tester than do shy, somber, retiring infants, who are hesitant or fussy during testing. According to this hypothesis, infants of these two types may be equally competent, capable of performing similarly if both could be tested in optimal circumstances. If this hypothesis were true, then one would want to partial out variance in performance due to sociability before attempting to compare infants or to assess the effects of enrichment-intervention. The methodological implications of such a finding would be substantial, for it would indicate that there was no straightforward linear relationship between cognitive competence and. performance on infant tests. Remember that the correlations between sociability and cognitive performance range up to .6, so that we are talking of a substantial, not a trivial, portion of the variance. One might also wonder what factors other than sociability affect performance on tests of cognitive development. If the error variance is too high, the validity of the tests would be seriously compromised. The third hypothesis is that sociable, friendly babies invite more social stimulation from adults (especially their parents) and that, over time, the supplementary stimulation facilitates and accelerates the infants’ cognitive development. According to this hypothesis, it is cognitive competence, not simply cognitive performance, that is correlated with sociability. The major implication would be that the social stimulation that appears to facilitate cognitive development (see Stevenson & Lamb, 1981) itself varies depending on characteristics of the infant. This effect of endogenous variables upon exogenous “influences” would constitute further evidence of the manner in which children’s characteristics affect their own development and experiences (Lewis & Rosenblum, 1974). In the following pages, I review evidence pertaining to these three hypotheses. Perhaps the first relevant analyses and interpretations were offered by McCall rtuf. (1972). Using data from Gesell assessments of the 158 subjects in the Fels Longitudinal Study, these researchers computed principal component analyses of individual item performance at 6, 12, 18, and 24 months. At 6 months, the first principal component seemed to be defined by items involving behavior associated with the production of contingent, “perceptual consequences. At I2 months the items composing the main developmental trend appeared to represent fine motor skills, the propensity to imitate the examiner, and the learning of simple social-verbal skills. By 18 and 24 months, the major component was defined by the imitation of more complex verbal behavior as well as skill in verbal labeling and comprehension. ”

228

Michael E . Lcimb

Three alternative interpretations of these data were offered by McCall and colleagues: It is reasonable to speculate that the child who precociously develops the propensity to manipulate objects and appreciate their contingent perceptual consequences may soon apply this same strategy to the less consistent contingencies of social situations. A natural outgrowth of reciprocal social behavior is imitation. . . the main behavior trend emphasizes asocial and then social operant learning. followed by the propensity to imitate sensorimotor and then verbal behavior. . . . [Alternately]. . . the exploration of objects, imitation of adult behavior, and display of verbal fluency all require the infant to possess a type of social extroversion that permits him [or her) to interact freely with a strange examiner. . . . Another alternative is that the developmental trend observed in the infant behavior may actually reflect trends and consistencies in parental behavior that are then reflected in the child. (McCall ef a/., 1972, pp. 742-743)

Very similar results were obtained in a later reanalysis of data from the Berkeley Growth Study (McCall, Eichorn, & Hogarty, 1977). In their report, however, McCall et af. (1977) did not offer interpretations similar to those offered in 1972. Instead, McCall et al. (1977) proposed that their data revealed universal stage transitions, comparable to those described by Piaget (1936/1952, 1937/ 1954), which transcended traditional distinctions between social and cognitive development. It is important to note, however, that in neither analysis did McCall and colleagues (1972, 1977) employ rating scales from the Infant Behavior Record or their equivalent, and they presented no data that might permit one to rank the three hypotheses in terms of their explanatory power. In his more recent reviews, McCall (1979a, 1979b) has appeared to favor the explanation offered by McCall et al. in the 1977 monograph-an explanation that essentially portrays the correlation between sociability and cognitive performance as the manifestation of individual differences in an underlying competence construct. This explanation is the same as that provided above by Hypothesis 1. It is not clear, however, that such an explanation can account for the sorts of correlations reviewed in the previous section. Furthermore, the pattern of results reported by McCall and colleagues is also consistent with the other two hypotheses. Thus, McCall et af. (1972, 1977) have not advanced our understanding of the relationship between sociability and cognitive competence. One point against McCall’s interpretation is the finding that sociability, IQ, and other aspects of temperament reveal different patterns of heritability (see Section V), suggesting that different, rather than similar, underlying dimensions are involved. In another recent study Lamb et al. (1981a) reanalyzed data concerning more than 33,000 participants in the National Collaborative Perinatal Project. They focused on data from the Mental tests conducted when the infants were 8 months old. Ratings of sociability were derived from three five-point scales (social responsiveness to examiner, social responsiveness to mother, social intensity)

Individuul Differences in Infaru Sociuhiliry

229

completed by the examiners. Only one of these scales (social responsiveness to the examiner) tapped sociability directly, and the measure of social responsiveness to mother was of dubious validity since highest scores were assigned to those infants who clung to their mothers during this test (see Section 11). Perhaps because of these measurement deficiencies, the correlations between sociability and cognitive performance were lower in this study than in any other: Correlations between Mental Index scores and social responsiveness-examiner, social responsiveness-mother, social intensity, and a composite score were .23, .16, .08,and .23, respectively. (Because of the immense sample size, all of these correlations were highly significant statistically, even though they explain only a minor portion of the variance.) No significant effects of social class, race, or sex were found. Lamb and colleagues then looked separately at the relationship between sociability and performance on two types of test items: those involving verbal behavior and play with the examiner and those not involving such interaction. We reasoned that if a performance effect were involved (Hypothesis 2 above), then sociability should be much more highly correlated with performance on the items demanding a great deal of social interaction than on the nonsocial items. (Of course, all items in infant tests demand some interaction with the tester, so one would not expect the correlations between sociability and performance on the nonsocial items to be zero even if a performance effect accounted for the entire relationship.) Our analyses revealed a correlation of .26 between composite sociability and performance on the 1 I verbal and play items and a correlation of. 17 between composite sociability and performance on the nonsocial items. Correlations with the individual sociability scales were lower, but followed the same pattern. Evidently, sociability was more highly correlated with performance on socially loaded than nonsocial items, suggesting that a performance effect was involved. However, the difference between the two correlations was not large, and the subtraction of the socially loaded items did not eliminate the significant relationship. Thus, a substantial part of the relationship between sociability and cognitive performance appears to be attributable to a real competence effect. Lamb et al. (1981b) used similar reasoning in their second investigation involving data from the National Collaborative Perinatal Project. In this reanalysis, they examined correlations between the same sociability scales and the children’s performance on the Motor scales of the Bayley assessment battery. Since the motor development items demand less interaction with the examiner than do Mental Index items, the correlations between the Motor scores and sociability would be much lower than the correlations between sociability and Mental Index scores if a performance effect is involved, but would be the same or only modestly different if a competence effect is involved. In fact, the correlation

230

Michael E . Lamb

between composite sociability and Motor Index scores was. 18-about the same as the correlation between sociability and the scores for performance in the nonsocial Mental Index items, and only modestly lower than the correlations between sociability and scores on the total Mental Index and on the socially laden items from the Mental Index. Again, the evidence suggests that there is a modest performance effect, but that the bulk of the relationship between sociability and cognitive-motor development is accounted for by a competence effect. Unfortunately, the assessment of sociability in this large national sample was deficient; consequently, it is impossible to determine how important (in absolute terms) the performance and competence effects are. The correlations obtained in analyses of these data are lower than those obtained in any other studies: Taken literally, they imply that the correlation between sociability and cognitive performance scores accounts for less than 5% of the variance-too little to excite much attention among developmental psychologists. Obviously, my involvement in this area signifies a belief that an appreciable and psychologically important relationship exists, and that further large-scale studies designed to assess the magnitude of the relationship should be undertaken. My optimism is buttressed by the fact that higher correlations have been obtained in studies using better measures of sociability. B. SUMMARY

In all of the studies discussed in this section, the investigators relied upon deficient means of assessing sociability, and this limits the conclusiveness of the analyses. However, all three of the hypotheses proposed in this section appear to account for some part of the relationship between sociability and cognitive performance. The studies by Lamb and colleagues indicate quite clearly that part of the variance in test performance is accounted for by the friendliness and cooperativeness of the infant during the test session. The remainder seems to involve a relationship between sociability and competence, but whether this relationship involves the developmental processes implied by the first or the third of the hypotheses cannot yet be determined. Experimental and longitudinal correlational strategies could and should be used to explore this issue more fully.

V. Origins of Individual Differences in Sociability Researchers have explored both biogenetic and experiential influences on individual differences in sociability. Behavior geneticists have led the effort to determine whether some portion of the variance in sociability is heritable; a diversity of approaches, few of them systematic, have focused on experiential influences on sociability. Although, of course, both sources of influence could,

Inclividuttl Difermces in Infiint Socicihility

23 I

and probably do, affect the development of sociability, I shall discuss the two sources separately in order to facilitate presentation of the evidence. A.

BIOGENETIC INFLUENCES

In an early investigation of twins, Freedman and Keller (1963) studied 29 same-sex twin pairs using early forms of the Infant Behavior Record scales. They found evidence of heritability (i.e., the difference between the correlation for monozygotic twins and that for dizygotic twins was greater than chance4) for social orientation and fearfulness, while home observations revealed evidence for genetic influences on the onset and intensity of social smiling and on the extent of stranger wariness. Unfortunately, this study involved a very small sample, and it is not clear that the examiners remained naive regarding zygosity. Two of the most intensive investigations of the heritability of sociability were published by Goldsmith and Gottesman (1981) and Matheny (1980). Large samples of twins were included in both studies, and in both cases sociability was treated as one aspect of infant temperament. Goldsmith and Gottesman analyzed behavioral ratings of some 360 twin pairs included in the National Collaborative Perinatal Project. These ratings were made when the subjects were 8, 48, and 84 months old: The 8-month ratings involved the unrevised Infant Behavior Record scales. At 8 months, only one (intensity of social response) of three ratings potentially related to sociability (social responsiveness-stranger, social responsiveness-mother, intensity of social response) showed even a modicum of heritability (RMZ- RDZ = .25). A statistically determined factor, Interest in/Responsiveness to People, revealed no evidence of significant heritability. Only one of the 48-month ratings appeared to deal with sociability (degree of cooperation); analyses revealed a modest heritability effect (RMZ - R D Z = .18). At 7 years, three scales dealt with sociability (fearfulness, friendly versus shy with examiner, and degree of cooperation); significant heritability was found for Friendly versus Shy with Examiner (RMz - RDz = .30), and this scale revealed the clearest evidence for genetic effects in the study. Of three relevant factors (Active Adjustment-Active, Spontaneous, Friendly; Fearfulhhibited; Agreeable/Cooperative), Active Adjustment ( R M Z- R,)z = .33) and Fearfull Inhibited (RhlZ - RDz = .15) showed evidence of heritability. Notice that the scales tapping different aspects of sociability do not clearly load together on the same factors, suggesting either that sociability is not a unitary construct, or that measurement in this study was deficient. With the exception of one strong effect at 7 years, sociability seemed to be less influenced by heritability than were the other aspects of temperament studied (e.g., Activity level). 4Heritability is, of course, equal to twice the difference between the correlation coefficients for monozygotic and dizygotic twins.

232

Mirhael E . i a m b

Matheny (1980) analyzed Infant Behavior Record ratings at 3 , 6, 9, 12, 18, and 24 months for the 218 twin pairs who were participants in the Louisville twin study. All tests for genetic influences involved factor scores. At 3 months, a Social factor accounting for only 6% of the total variance emerged; at 6, 9, and 12 months sociability with mother and others (generally) formed one factor, while endurance, fearfulness, emotional tone, social responsiveness to the examiner, and cooperativeness formed another more powerful factor, which Matheny labeled Test Affect-Extraversion. Factors similar to the latter were also evident at both 18 and 24 months. Significant heritability on this factor was evident at 6 , 12, and 24 months, but not at 3, 9, and 18 months. In contrast, a Task Orientation factor showed significant heritability at all ages except 9 months, and an Activity factor showed a strong heritability effect from 18 months onward. Another twin study conducted by Plomin and Rowe (1979) involved 46 twin pairs ranging in age from 13 to 37 months. These researchers sought evidence of heritability on 24 discrete behavioral measures of behavior (e.g., “smiles to stranger in Situation 1 ”) during a standardized interactive sequence at home. Plomin and Rowe found evidence for significant heritability on five measures and near-significant effects on two others. Unfortunately, they did not create a composite measure of sociability from their discrete behavioral indices, although such composite measures would probably be more robust, more reliable, and more valid (see Matheny, 1980). Further, although the behavior of each twin was recorded independently, both observers actually observed the two infants, which may have introduced some bias. Finally, Thompson and Lamb (1982) found that the sociability of 19.5-monthold infants was negatively correlated with maternal reports of their infants’ temperament on dimensions of fearfulness and angedfrustration, and was positively correlated with reports of their positive emotionality and activity level. There were fewer significant correlations between sociability and perceived temperament at 12.5 months. Perceived temperament was relatively stable over time and was unaffected by changing life circumstances or caretaking arrangements, suggesting that reported temperament was constitutionally based and, thus, changed slowly in response to experience (Rothbart & Derryberry, 1981). Thompson and Lamb’s data are thus consistent with the results of twin studies revealing heritability as one determinant of sociability. In sum, genetic influences on sociability evidently exist, but these influences d o not appear to be very powerful. The equivocal evidence concerning genetic influences on sociability or extroversion contrasts with the evidence from a variety of studies indicating fairly clearly that sociability-extroversion is indeed heritable among older children (see Scarr, 1969, for a review). In addition, the modest effects reported in studies of infants may have been enhanced by failures to ensure complete independence in the assessments of cotwins. No information on this score was available to Goldsmith and Gottesman (1 98 1) although they

Irulividuol Differences in Infant Sociuhility

233

were informally assured that independence was ensured. Independence of assessments was not mentioned by Matheny (1980) although previous reports from the Louisville Twin Study (Dolan et al., 1974) indicate that independence was ensured. However, some nonindependence probably contaminated the results reported by Freedman and Keller (1963) and Plomin and Rowe (1979). However, the magnitude of genetic effects has undoubtedly been underestimated because of the deficiencies in the measures used. The variances on Infant Behavior Record scales or their equivalents are known to be low, and the lack of variability is likely to have affected the results reported by Goldsmith and Gottesman, Matheny, and Freedman and Keller. Likely problems with Plomin and Rowe’s measures have already been mentioned. Now that we have available suggestive evidence regarding genetic effects on sociability, we need studies employing psychometrically sound and suitable instruments for assessing the magnitude of genetic effects in this area. B. ENVIRONMENTAL INFLUENCES

The existence of a relationship between amount of social stimulation, sociability, and cognitive development was first suggested by researchers exploring the effects of “maternal deprivation” on infants and young children. As Bowlby (1951) noted in his authoritative review, infants raised in institutions from the first months of life displayed a profound lack of social responsiveness as well as severe intellectual retardation. Infants who were institutionalized later than the first half-year of life often passed through a phase of depression and unresponsiveness before becoming indiscriminately sociable; these infants too showed signs of intellectual retardation. Later studies by researchers such as Casler (1965, 1968) showed that the provision of supplementary auditory, visual, and vestibular stimulation could alleviate the progressive retardation as well as the social unresponsiveness-at least in the short term. Likewise, researchers such as Rheingold ( 1956) successfully showed that supplementary auditory and visual stimulation did produce increases in sociable behavior among institutionalized infants. These findings thus suggested that sociability was a consequence of the amount of social stimulation. However, the fact that institutionalization had differential effects on social responsiveness and cognitive development when it occurred after the first 6-8 months suggested that different factors affected social and cognitive development. Bowlby (1951) and later Ainsworth (1966) proposed that the effects of institutionalization on aspects of socioemotional and motivational development were mediated via an effect on mother-infant attachment. Infants who were institutionalized very early were never able to form attachments, whereas those who were institutionalized later in life were able to form attachments, which were later disrupted. In neither case was it clear how the institutionalization or maternal deprivation had affected the children’s de-

234

Mil-hue/ E . Lamb

velopment, why the failure to attach and the disruption of attachment had such different effects, and whether the unresponsiveness of institutionalized infants was related in any way to their deficient cognitive performance. Unfortunately, the lack of social and physical stimulation (in many institutions) was so profound, and the effects were so pervasive, that it is hard to draw firm conclusions about specific influences on social responsiveness from these reports. In addition, it is necessary to assess environmental variables and sociability more precisely than in the studies comprising the maternal deprivation literature if we are to understand the effects of environmental influences on infant sociability. Suggestive evidence for a systematic relationship between maternal behavior and infant sociability was presented by Bayley and Schaefer (1964). Like infant sociability, measures of maternal affection and warmth were correlated with mental test scores between 10 and 36 months of age, especially in girls. This correlation implied that bright sociable girls had warm sociable mothers-a relationship that could as easily suggest genetic influences or environmental influences-or that sociable babies elicited more affection from their mothers. Stevenson and Lamb (1979) reported a similar relationship: Sociable infants had sociable mothers, but this effect was evident only on initial sociability (not sociability during testing). Clarke-Stewart et a / . (1980) reported that sociability with strangers was enhanced by increases in the amount of contact with nonparental relatives, although these investigators .found a negative correlation between sociability and amount of cure by nonparental relatives. More recently, researchers have studied the relationship between security of infant-parent attachment and sociability with strangers, reasoning that infants who are securely attached should generalize their cooperativeness and trust from parents to other unfamiliar adults (see Lamb, 1981). The relevant research substantiates this prediction. In the first study, Main (1973) reported that securely attached I-year-olds were more cooperative and were more likely to evince a “gamelike spirit” during administration of the Bayley Mental Scales at 20 months than were insecurely attached infants. Other Infant Behavior Record ratings did not yield significant differences between securely and insecurely attached infants. Compared to the insecurely attached infants, the securely attached infants were also more friendly toward an adult playmate at 20 months. The securely attached infants approached more often, were more playful and less avoidant, and were more likely to return a ball in play. Like Main ( 1 973), my co-workers and I have used Ainsworth’s Strange Situation procedure (Ainsworth, Blehar, Waters, & Wall, 1978) to assess the security of parent-infant attachment and have then related security of attachment to infant sociability. Unlike Main, however, we have used composite measures of sociability: either the original measure of initial sociability developed by Stevenson (1978; Stevenson & Lamb, 1979) or the revision prepared by Thompson (see Table Ill). Owen, Chase-Lansdale, and Lamb (1981) found that 1-year-old infants who were securely attached to their mothers were more sociable with strange adult

Individual Differences in hlfilnt Sociuhiliry

235

females than were insecurely attached infants. However, we found no significant relationship between the security of father-infant attachment and the infants’ sociability with unfamiliar female adults. By contrast, Lamb, Hwang, Frodi, and Frodi (1982) found in a study of Swedish parents and infants that infants who were securely attached to their fathers were more sociable with strangers, whereas the security of infant-mother attachment was unrelated to sociability. Although the predictive validity of the father-infant attachments did not vary depending on the fathers’ relative involvement in childcare, it is possible that the discrepancy between Owen et a1.k and Lamb et al.’s findings is attributable to the greater involvement of the Swedish fathers. Most researchers attempting to assess the predictive validity of infant-parent attachment have simply compared infants who are securely and insecurely attached. Such a strategy excludes from consideration the fact that there are two types of insecure attachment which involve very different patterns of behavior (angry, resistant behavior or avoidant, ignoring behavior) and that within the secure group, there are those infants whose dominant mode of interaction involves distal communication (the B1 and B2 subgroups) and those whose dominant mode involves a great deal of proximal interaction (the B3 and B4 subgroups). One would expect these to differ in their degree of sociability toward unfamiliar adults, especially since Easterbrooks and Lamb (1979) earlier found that infants from two of the subgroups within the secure category (B1 and B2) were more sociable with peers than those in the other two secure subgroups (B3 and B4). Consequently, in their study, Lamb et al. ( I 982) considered four composite groups: infants who were avoidant (A1 and A2) with their parents; infants w h o were angryhesistant (C1 and C2) toward their parents; the B 1 and B2 (distal, secure) subgroups; and the B3 and B4 (proximal, secure) subgroups. As predicted, infants with B1 and B2 relationships were significantly more sociable than infants in any other group. These results were consistent with those reported by Thompson and Lamb (1 981), who found that infants who had Bl (secure) attachments with their mothers attained the highest sociability scores, whereas those in the A1 and C2 (insecure) categories attained the lowest scores. The infants in this study were seen twice (at 12.5 and 19.5 months) in both the Strange Situation and the Sociability tests. Sociability in the two assessments was significantly stable ( r = .40) but the degree of stability was closely related to the degree of stability in security of attachment, which was in turn related to the occurrence of major changes in family circumstances and caretaking arrangements (Thompson et al., 1982). When such changes did not occur, security of attachment was stable, as was sociability ( r = .74). However, when changes in caretaking arrangements occurred, security of attachment often changed, and there was no significant stability in sociability ( r = .18, n.s.). Since security of attachment appears to be dependent on the intermeshing of parent and infant behavior, Thompson and Lamb’s (1981) findings suggest that the quality of the infant’s social experiences within the family affects the infant’s sociability.

236

Miclirrrl E . Lumb

C.

SUMMARY

The conclusiveness of research on the origins of individual differences in sociability is limited either by the unsystematic nature of research in this area or by the poor quality of the instruments used to assess sociability. It seems, however, that both biogenetic and experiential factors affect sociability, but the relative importance of these factors and the interaction between them remains to be assessed. Even if one combines the proportion of variance known to be explained by genetic factors and by identified experiential influences, most of the variance remains unexplained. Clearly, the need for additional research is profound.

VI.

Conclusion

Although the list of relevant studies continues to lengthen, the evidence regarding the relationship between sociability and cognitive performance in infancy remains frustratingly inconclusive. On the positive side, it is clear that the two constructs are related to one another; the most recent studies, employing the more sensitive measures of sociability, indicate that 25-35% of the variance in cognitive performance can be accounted for by individual differences in sociability. On the negative side, we do not yet know from what this statistical relationship results, what it tells us about developmental processes, and whence individual differences in sociability originate. Studies designed to explain the relationship between sociability and cognitive performance indicate that a performance artifact (i.e., friendly babies give a better accounting of themselves than unsociable infants of equivalent ability) accounts for some, but not all, of the relationship. Sociable infants, it seems, are indeed smarter, though we do not know whether this is because the sociable infants have elicited more stimulation in the past, or, because social and cognitive development inevitably proceed hand in hand, with precocity in one area necessarily accompanying precocity in the other. The former possibility seems likely to be more important, and is theoretically interesting because it exemplifies one way in which infant characteristics help to shape the infant’s formative experiences. As far as the origins of individual differences in sociability are concerned, both genetic and environmental influences have been demonstrated. Unfortunately, the available research explains only a small portion of the variance in sociability. Perhaps a greater portion of the variance will be explained if better measures of sociability are used systematically in the future. Deficiencies of measurement, in fact, constitute the major limitation in almost all studies in this area. Improved instruments are now available, however; I hope that their availability accelerates progress and understanding of this topic.

Individual Diflerences in lnfint S o c i u h i l i ~

237

REFERENCES Ainsworth, M. D. S . The effects of maternal deprivation on cognitive and intellectual development. In 0.J. Harvey (Ed.), Experience, structure, und tidaptcirbility. New York: Springer, 1966. Ainsworth, M. D. S . , Blehar, M . , Waters, E.. & Wall, S . Potterns qf’utruchment. Hillsdale, N.J.: Erlbaum, 1978. Batter, B. S . , & Davidson, C . V . Wariness of strangers: Reality of artifact? Journd qf’ Child Ps,ycholugy mnd Psychiatry, 1979, 20, 93- 109. Bayley, N . The Calijorniafirst-year mental scale. Berkeley: Univ. of California Press, 1933. (a) Bayley, N. Mental growth during the first three years: A developmental study of 61 children by repeated tests. Genetic Psychology Monographs, 1933, 14, 1-92. (b) Bayley, N. Behavioral correlates of mental growth: Birth to thirty-six years. American Psvchologist, 1968, 23, 1-17.

Bayley, N. Buyley Scales of Infint Dewlopinent. New York Psychological Corporation, 1969. Bayley, N., & Schaefer, E. S . Correlations of maternal and child behaviors with the development of es: Data from the Berkeley Growth Study. Monogruphs ofthe Societyfir Reseurch in Child Development. 1964. 29 (6, Serial No. 97). Beckwith, L.. Cohen, S . E., Kopp, C . B., Parmelee, A. H., & Marcy, T. G. Caregiver-infant interaction and early cognitive development in preterm infants. Child Development. 1976, 47, 579-587.

Birns, B., & Golden, M. Prediction of intellectual performance at three years from infant tests and personality measures. Merrill-Palmer Quurterly. 1972, 18, 53-58. Bowlby, J . Mciternul care and mentul health. Geneva: World Health Organization, 1951, Casler, L. The effects of extra tactile stimulation on a group of institutionalized infants. Genetic Psycho1og.y Monogruphs. 1965, 71, 137-175. Casler, L. Perceptual deprivation in institutional settings. In G. Newton & S . Levine (Eds.), Early experience and behavior. Springfield, Ill.: Thomas, 1968. Clarke-Stewart, K. A . Interactions between mothers and their young children: Characteristics and consequences. Monographs of the Society for Reseurch in Child Development, 1973, 3“-7, Serial No. 153). Clarke-Stewart. K . A,, Umeh, B. J., Snow, M. E., & Pederson, J . A . Development and prediction of children’s sociability from 1 to 2% years. Developrnentul Psycholugy, 1980, 16, 290-302. Crano, W. D. What do infant mental tests test? A cross-lagged panel analysis of selected data from the Berkeley Growth Study. Child Development, 1977, 48, 144-151. Dolan, A. B., Matheny, A. P., & Wilson, R. S . Bayley’s Infant Behavior Record: Age trends, sex differences, and behavioral correlates. J.S.A .S.Cntcilog of Se/ected Documents in Psychology, 1974, 4, 9.

Easterbrooks, M. A , , & Lamb, M. E. The relationship between quality of infant-mother attachment and infant competence in initial encounters with peers. Child Development, 1979,50, 380-387. Freedman, D. G . , & Keller, B. Inheritance of behavior in infants. Science, 1963, 140, 196-198. Goldsmith, H. H., & Gottesman, 1. 1. Origins of variation in behavioral style: A longitudinal study of temperament in young twins. Child Development, 1981, 52, 91-103. Kopp, C. B., Sigman, M., & Parmelee, A. H. Longitudinal study of sensorimotor development. Dcvelopmentul Psvchologv, 1974, 10, 687-695. Lamb, M. E. Developing trust and perceived effectance in infancy. In L. P. Lipsitt (Ed.), A d w n c e s in infancy research (Vol. I). Nonvood, N.J.: Ablex. 1981. Lamb, M. E., Garn, S. M., & Keating, M. T. Correlations between sociability and cognitive performance among eight-month-olds. Child Developmen/. 1981, 52, 71 1-713. (a) Lamb, M. E., Garn, S. M.,& Keating, M.T. Correlations between sociability and motor performance scores in eight-month-olds. hlfitnr Behuvior and Development, 1981, 4, in press. (b)

238

Michael E. Lamb

Lamb, M. E., Hwang, C.-P., Frodi, A. M., & Frodi, M. Security of mother- and father-infant attachment and its relation to sociability with strangers in traditional and nontraditional Swedish families. Infant Behavior and Development, 1982, 5, in press. social cognition: Empirical and theoretical consideraLamb, M. E., & Sherrod, L. R. (Eds.) I&t tions. Hillsdale, N.J.: Erlbaum, 1981. Lewis, M . , & Rosenblum, L. A . (Eds.) Effects of the infant on its caregiver. New York: Wiley, 1974. Main, M. Exploration, play, and cognitive functioning as related to child mother attachment. Unpublished doctoral dissertation, Johns Hopkins University, 1973. Matheny, A. P. Bayley’s Infant Behavior Record: Behavioral components and twin analyses. Child Development, 1980, 51, 1157-1167. Matheny, A. P., Dolan, A. B., & Wilson, R. S . Bayley’s Infant Behavior Record: Relations between behaviors and mental test scores. Developmental Psychology, 1974, 10, 696-702. Matheny, A. P., Dolan, A. B., & Wilson, R. S. Twins: Within-pair similarity on Bayley’s Infant Behavior Record. Journal of Genetic Psychology, 1976, 128, 263-270. McCall, R. B. The development of intellectual functioning in infancy and the prediction of later 1Q. In J. D. Osofsky (Ed.), Handbook of infunt development. New York: Wiley, 1979. (a) McCall, R. B. Qualitative transitions in behavioral development in the first two years of life. In M. H. Bornstein & W. Kessen (Eds.), Psychokogical development from infancy. Hillsdale, N.J.: Erlbaum, 1979. (b) McCall, R. B., Eichorn, D. H., & Hogarty, P. S. Transitions in early mental development. Monographs of the Society for Research in Child Development, 1977, 42(3, Serial No. 171). McCall, R. B., Hogarty, P. S., & Hurlburt, N. Transitions in infant sensorimotor development and the prediction of childhood IQ. American Psychologist, 1972, 27, 728-746. McGowan, R. J., Johnson, D. L., & Maxwell, S. E. Relations between infant behavior ratings and concurrent and subsequent mental test scores. Developmental Psychology, 1981,17, 542-553. Owen, M. T., Chase-Lmsdale, P. L., & Lamb, M. E. Mothers’ and fathers’ attitudes. maternal employment, and the security of infunt-parent attachment. In preparation, 1981. Piaget, J. The origins of intelligence in children. New York International Universities Press, 1952. (Originally published, 1936.) Piaget, J. The construction of reality in the child. New York: Basic Books, 1954. (Originally published, 1937.) Plomin, R.. & Rowe, D. C. Genetic and environmental etiology of social behavior in infancy. Developmental Psychology, 1979, 15, 62-72. Ramey, C. T., & Campbell, F. A. Compensatory education for disadvantaged children. School Review, 1979, 87, 171-189. Rheingold, H. L. The modification of social responsiveness in institutionalized babies. Monographs of the Sociery for Research in Child Development, 1956, 21(2, Serial No. 63). Rothbart, M. K., & Derrybeny, D. Development of individual differences in temperament. In M. E. Lamb & A. L. Brown (Eds.), Advances in clevelopmentalpsychology (Vol. I ) . Hillsdale, N.J.: Erlbaum, 1981. Scam, S. Social introversion-extraversion as a heritable response. Child Develupment, 1969, 40, 823-832. Seegmiller, B. R., & King, W. L. Relations between behavioral characteristics of infants, their mother’s behaviors, and performance on the Bayley Mental and Motor Scales. Journd of’ Psychology, 1975, 90, 99-1 1 1 . Shantz, C. U. The development of social cognition. In E. M. Hetherington (Ed.), Review of child development research (Vol. 5 ) . Chicago: Univ. of Chicago Press, 1975. Sroufe, L. A. Wariness of strangers and the study of infant development. Child Development, 1977, 48, 731-746.

Individual Differences in fnjiant Sociabilitv

239

Stevenson. M. B . E&cts I J ] injirnf socicihility cind the curetaking envirunmertt on i@nt cognitive competence. Unpublished doctoral dissertation, University of Wisconsin-Madison, 1978. Stevenson, M. B . A longitudinul study of the relutionship beriveen sociability cind infiint cognitive competence. In preparation. 198 I . Stevenson, M. B., & Lamb, M. E. Effects of infant sociability and the caretaking environment on infant cognitive performance. Child Developmenr. 1979. 50, 340-349. Stevenson, M. B., & Lamb, M. E. The effects of social experience and social style on cognitive competence and performance. In M. E. Lamb & L. R. Sherrod (Eds.), Infunf social cognition: Empirical and theoretical cunsiderations. Hillsdale, N.J.: Erlbaum, 1981. Thompson, R . A . , & Lamb, M. E. Qitulity uj'cittcichtnrnt mid mcinger sociubi1ir.v in ii$zncy. In preparation, 198 I . Thompson, R. A,, & Lamb, M.E. Stranger sociability and its relationship to temperament and social experience during the second year. lnfint Behavior and Develupmenr, 1982, 5 , in press. Thompson, R. A , , Lamb, M. E., & Estes, D. A. Stability of infant-mother attachment and its relationship to changing life circumstances in a representative middle class sample. Child Developmenf, 1982, 53, in press. Uzgiris, I. C., & Hunt, J . McV. Assessment in infancy: Ordinal scales of psychological develupment. Urbana: Univ. of Illinois Press, 1975. Waters, E. The reliability and stability of individual differences in infant-mother attachment. Child Development, 1978, 49, 483-494. Waters, E. Traits, behavioral systems, and relationships: Three models of infant-adult attachment. In K. Immelman, G. Barlow. M. Main, & L. Petrinovitch (Eds.), The development of behavior. London and New York: Cambridge Univ. Press, 1980. Zigler, E., Abelson, W. D., & Seitz, V. Motivational factors in the performance of economically disadvantaged children on the Peabody Picture Vocabulary Test. Child Development, 1973.44, 294-303.

This Page Intentionally Left Blank

THE DEVELOPMENT OF NUMERICAL UNDERSTANDINGS’

Robert S . Siegler and Mitchell Robinson DEPARTMENT OF PSYCHOLOGY CARNEGIE-MELLON UNIVERSITY PITTSBURGH. PENNSYLVANIA

I. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. AN INITIAL STUDY OF NUMBER CONSERVATION 111. PRESCHOOLERS’ KNOWLEDGE OF COUNTING . . .

242 244

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

A. COUNTING FROM ONE ........................................... B. COUNTING ON FROM A POINT BEYOND ONE . . . . . . . . . . . . . . . . . . . . . . C. MODELS OF THREE LEVELS OF COUNTING EXPERTISE.. . . . . . . . . . , ,

250 250 256 259

IV. PRESCHOOLERS’ KNOWLEDGE OF NUMERICAL MAGNITUDES . . . . . . . . . . A. EXISTING RESEARCH ON ADULTS AND CHILDREN.. . . . . . . . . . . . . . . . B. PRESCHOOLERS’ NUMERICAL MAGNITUDE COMPARISONS . . . . , . , . . C. VERBAL LABELING OF NUMBERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. EFFECTS O F TEACHING A LABELING STRATEGY. . . . . . . . . . . . . . . . . . . E. THREE MODELS O F NUMERICAL MAGNITUDE COMPARISON . . . . . . .

267 267 27 1 275 278 280

V. PRESCHOOLERS’ KNOWLEDGE OF ADDITION. . . . . , . . , . . . . . . . . . . . . . . . . . A. EXISTING RESEARCH ON CHILDREN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. PRESCHOOLERS’ ADDITION STRATEGIES . . . . . . . . . . . . . . . . . . . . . . . . . . C. A MODEL OF STRATEGY CHOICE IN ADDITION . . . . . . . . . . . . . . . . . . . .

287 287 290 296

VI. CONCLUSIONS: THE DEVELOPMENT OF NUMERICAL KNOWLEDGE . . . . .

299

REFERENCES ....................................................

308

IThis research was supported in part by NICHHD Grant HD-15285 and by a grant from the Spencer Foundation. Thanks are due to Dr. Taylor and the teachers of the Carnegie-Mellon Children’s School, and to Mrs. Cohen and the teachers of the Carnegie-Mellon Day Care Center. Also deserving considerable gratitude are Cindy Zaks who was the experimenter in all of the studies and Vickie DeRose and Amy Laird who typed more than a reasonable number of drafts of the manuscript. Finally, a number of colleagues read and commented on earlier versions of the article: Diane Mierkiewict, Janellen Huttenlocher, Hayne Reese, and Roman Taraban. Their help is gratefully acknowledged. 24 1 ADVANCES IN CHILD DEVELOPMENT AND BEHAVIOR. VOL. 16

I

Copyright 0 1982 by Academic Ress. Inc. All rights of reproduction tn any form reserved. ISBN 0-12-W97168

242

Robert

S. Siegler and Mitchell Robinson

I.

Introduction

The purpose of this article is to explore ways in which we can characterize people’s understandings of concepts. In particular, we shall attempt to characterize very young children’s understandings of numbers by examining a variety of their numerical skills, by inferring representations and processes that might give rise to each of these skills, and by integrating the models arising from each task domain to build a general model of preschoolers’ knowledge of numbers. At one time, not very long ago, there was a large degree of consensus as to how best to describe children’s knowledge of concepts. Piaget had identified tasks that came to be accepted as indices both of children’s general cognitive levels and of their understanding of specific concepts. Children were said to understand the concept of classes when and only when they could succeed on the class inclusion task, to understand the concept of ordering when and only when they could succeed on the seriation task, to understand the concept of perspective when and only when they could succeed on the three-mountain task, and so on. Even though Piaget himself examined numerous tasks before reaching his conclusions, many other investigations relied on a single one of his tasks as their sole index of understanding of each concept. Accompanying the view that understanding could be assessed by performance on a single task was the view that children possessed a single understanding to measure. This view was reflected in the multitude of studies in which researchers attempted to establish the age at which children master particular concepts and the order in which children master different concepts. It was also reflected in the titles that investigators, both Piagetians and non-Piagetians, chose for their reports. Illustratively, two of the most influential monographs on children’s understanding of number concepts have been those of Piaget ( I 952) and Brainerd (1979). Piaget titled his book The Child’s Conception of Number. Brainerd titled his book The Origins of the Number Concept. The use of the singular in the terms “conception” and “number” and of the definite article together with the singular in the phrase “the number concept” is striking. The implication is that there exists a single number concept to be understood and that children have a particular concept or conception of it. Mathematicians and philosophers have long debated whether there exists any single concept of number to be understood, but the results of the past 20 years of research in developmental psychology have rendered completely untenable the position that children possess a single understanding of number or of other complex concepts. Even within a single basic problem, wide variability in performance, depending on the details of the task, has been the rule. To cite one illustration, although most children do not succeed on Piaget’s number conservation problem until age 6 or 7 years (Beilin, 1968; Miller, 1976; Rothenberg &

Numericul Understandings

243

Courtney, 1969), most children can succeed on variants of it by age 2 or 3 (Bever, Mehler, & Epstein, 1968; Bryant, 1974; Gelman, 1972). To cite another illustration, Trabasso (1978) listed some of the task variables that have been found to influence class inclusion performance: typicality of subordinate class exemplars, presence of contrasting subordinate classes along with the two classes referred to in the question, ratio of majority to minority subclass, whether the members of the sets are physically present, and whether the subordinate classes are quantified before or after pictures are shown. The large effects of many of these variables make it extremely difficult to decide which form of the task is optimally suited for assessing understanding (Flavell, 197 1 ; Siegler, 1981a). The situation becomes even more complicated when we consider not just variants of a single task but also the many possible tasks that might reasonably be said to correspond to any concept. Rather than assessing knowledge of numbers by the number conservation problem, we might assess it in terms of ability to count objects, to compare numerical magnitudes, to understand the relationship between arithmetic and algebra, to understand number theory, and so on. Again, there is no simple principled way to choose. The issue can be considered at a very general level. Braine (1959) and Brown ( 1976) have made quite eloquent statements advocating specific criteria for defining conceptual understanding. Braine argued for a criterion of initial competence, Brown for a criterion of stable usage. The dilemmas that each of these proposals lead to suggest that no single standard of conceptual understanding can be adequate. Consider Braine’s (1959) statement: It is clear that if one seeks to state an age at which a particular type of response develops, the only age that is not completely arbitrary is the earliest age at which this type of response can be elicited using the simplest experimental procedure. (p. 16)

This statement is entirely reasonable, as far as it goes. When one considers the long time period separating initial and mature understanding of concepts, however, a paradox becomes evident. Adopting the initial competence criterion puts us in the position of saying that many concepts develop at relatively young ages, yet of also saying that children fail many reasonable indices of understanding for years thereafter. Stated another way, much, perhaps most, of conceptual growth would be seen as occurring after the concept “develops.” Brown ( 1976) implicitly suggested an alternative criterion in her discussion of the development of seriation: Under optimal circumstances, they can indeed senate a succession of pictures representing a time course. . , . Yet how robust is their concept of succession? Is it truly operative according to the defining features argument? The answer must be “no,” for their concept of order appears to be extremely fragile and is disrupted by seemingly trivial changes in the optimal task. (p. 77)

244

Robert S. Sicgler und Mitchell Robinson

The paradox that is implicit in Brown’s observation is pointed out in Braine’s comment. What exactly does a child understand when he or she can use a concept in some situations but fails to qualify as having an operative understanding of it? As Braine suggested, it does seem arbitrary to identify understanding with anything other than the earliest form of understanding; however, it seems misleading to identify understanding with the earliest form of understanding. How then can we study conceptual development? The above critique suggests a three-stage procedure. First, one examines performance on a variety of tasks corresponding to different aspects of a concept. For the same reasons that it seems desirable for psychometricians to sample numerous content domains in order to infer children’s status on aggregate-level constructs such as intelligence, it seems desirable for cognitive and educational psychologists to sample content within any given concept to infer children’s conceptual understanding. Second, one characterizes the representations and processes that people use to perform each particular task. The reasons for formulating models that distinguish between representations and processes have been discussed previously by Anderson (1976, 1978) and will be discussed further in a later section of this contribution. Third, one integrates the findings from each task into a general characterization of knowledge sufficient to derive each of the particular representations and the processes applied to it. Information-processing models have been criticized for being overly task specific, for not addressing the connections between how people perform any one task and other aspects of their knowledge (Neisser, 1976; Strauss & Levin, 1981). Producing a model that revealed the relationships among several tasks within a given conceptual domain would constitute at least a first step toward meeting this criticism. In the remainder of this article, we shall describe our efforts to apply this research strategy to analyzing preschoolers’ concepts of numbers.

11. An Initial Study of Number Conservation A study of Piaget’s number conservation problem (Siegler, 1981a, Experiments 3 and 4) was instrumental in convincing us of the need to adhere to the above-described procedures in order to characterize children’s knowledge. This study began with a consideration of the role of transformations in conservation problems. Analysis suggested that transformations play a crucial role within the conservation concept. The one sure way to determine whether the value of a particular quantitative dimension will be preserved in a situation is to know the type of transformation that will be performed. Adding something to that dimension necessarily results in more, subtracting something results in less, and neither adding nor subtracting anything results in the same amount as before. Piaget and subsequent conservation researchers focused almost exclusively on

N u m e r i d Understandings

245

transformations that do not affect quantity: pouring water, molding clay, moving objects apart, and so on. It would seem, however, that transformations that d o affect quantity-addition and subtraction-are of at least equal importance. A conserver might reasonably be expected to understand not only that spreading out a row of objects leaves the number of objects unchanged, but also that spreading them and adding an object means that the row now has more objects than before and that spreading them and taking away an object means that it now has less. In Siegler 's ( 198 1 a) study, 3- to 9-year-olds were presented number conservation problems in which the starting configuration always had two equally numerous and equally lengthy rows of objects. Transformations varied in their effects on numerosity and length. Some problems involved adding objects to a row, some problems involved subtracting objects from a row, and some involved neither adding nor subtracting objects. Some problems involved lengthening the transformed row of objects, some involved shortening the row, and some involved moving the objects but ultimately returning the row to its original length. Finally, on some problems the rows had few objects (2-4), and on others they had many (7-9). In all, 48 problems were presented. The pattern of correct answers and errors on these problems was sufficient to allow us to induce the rules that children were using to perform the task (see Siegler, 1976, 1978, 1981a, for descriptions of the rule assessment approach that was used). Most children used one of five rules in approaching the conservation problems. These rules incorporated progressively more complete subordinations of perceptual to transformational criteria. As shown in Fig. l A , the youngest children, most of them 3-year-olds, chose the longer row as having more objects in almost all cases-whenever the problem involved a large number of objects, and even with small arrays, when nothing had been added or subtracted. When the number of items was small and something had been added or subtracted, however, they relied on the type of transformation; addition meant that there was more and subtraction that there was less, regardless of the relative lengths of the rows.2 Somewhat older children used a simplified version of this approach (Fig. IB). When only a few items were presented, they always decided on the basis of the type of transformation. However, when many objects were presented, the children chose the longer row as having more. Yet older children extended the transformational approach one step further (Fig. IC). Now, either if the rows had few items, regardless of the transformation, or if they had many and something had been added to or subtracted from Thildren using Rules I , 11, and I11 may have used set size and type of transformation cues as indices of when to count rather than as direct bases of judgment. Within this interpretation, the entire Rule IV proredure would replace the bottom-most diamond and branches on the left side of Rules I , 11, and 111. Available evidence did not allow discrimination between these two interpretations.

A Rule

I

-

row rowless

row more

-

row more

--same

--more

row

-same

row-

-same

-more

less

Fig. I . Rules on number ronservurion tusk: ( A ) Rule I ; ( B ) Rule 11.

Numericcil UnJer.stunditigs

C

Smal I

Collect con

Rule Jll

rowmore

D

241

-same

row less

-

row -same

-more

Rule Ip: Count Objects in Row I

E

Count Objects in Row 2

Rule

TL

Transformed row -more

Two rows + some

Two rows --same

Row with greater number +more

Fig. I . ( C ) Rule Ill; ( D } Rule IV; ( E ) Rule V

-less

T r a n s f o r m e d row

248

Robcrt S. Siegler rind Miwhell Robinson

one of them, the type of transformation was the determinant of the conservation judgment. If the rows had many items and nothing had been added or subtracted, however, children still chose the longer row as having more. Interestingly, this last holdout from the transformational criterion is precisely the traditional conservation of number problem. Still older children came to use Rule IV on both small and large sets; they answered all types of number conservation problems correctly. This achievement did not conclude the development of the concept, however. Rule IV children solved number conservation problems by counting or pairing the objects in the two rows (Fig. 1D); Rule V children realized even without counting or pairing that adding something necessarily meant that that row had more, that subtracting something necessarily meant that it had less, and that doing neither meant that it necessarily had the same number of objects as before (Fig. 1E). The same 3- to 9-year-olds whose understanding of number conservation was assessed were also tested on the liquid and solid quantity conservation tasks. Two findings suggested that understanding of number conservation was crucial to the development of understanding of the other two conservation problems. First, no children consistently solved liquid and solid quantity conservation problems who did not also consistently solve number conservation problems. Second, among children who consistently solved number conservation problems, those who justified their responses on those problems in terms of the type of transformation were much more likely to solve the other two types of conservation problems than children who justified their responses on the number problems in terms of counting or pairing of objects in the two rows. These findings suggested an overall model of conservation acquisition. After progressing through rules that will solve some but not all number conservation problems, children come to be able to solve all problems by using the external referents of counting a n d o r pairing. Later, they note that these tests always indicate that when something has been added, there are more objects than before, that when something has been subtracted, there are fewer, and that when nothing has been added or subtracted, there are the same number. Therefore, they come to rely on the type of transformation to solve the number conservation problems. Finally, the children apply what they have learned about transformations in the number context to other domains involving transformations but not offering simple external referents, specifically conservation of liquid and solid quantity. This experiment persuaded us of several points. First, it illustrated just how difficult defining conceptual understanding is. If understanding of number conservation were defined as “in at least some situations, relying on the type of transformation despite opposing perceptual cues,” then Rule I would be the point of understanding. If understanding were defined as “for the traditional number conservation transformation, in at least some situations relying on the type of transformation despite opposing perceptual cues, then Rule 11 would be chosen. If it were defined as “ability to consistently solve number conservation prob”

Nutnericcrl Undrrstcindings

249

lems,” then Rule IV would be chosen. If it were defined as “an abstract understanding of the quantitative effects of transformations that does not depend upon requantification and that can be transferred to other types of conservation problems,” then Rule V would be chosen. Each one of these definitions seems to us sensible within particular theoretical contexts; no one of them seems absolutely more sensible than the others. A second lesson that we derived from the conservation study was the desirability of separating children’s representations of information from the processes that they apply to their representations. The rule models specified the processes that children used to solve the conservation problems but only implicitly addressed the content upon which the processes operated. How many objects constituted a small number and how many a large one was never spelled out. Neither was the form of the knowledge about transformations that children transferred from the number conservation context to the liquid and solid quantity ones. For some issues,.such as how many objects constitute a small number, specifying the additional information would have been awkward but possible. For other issues, such as how knowledge of transformations is represented, it is unclear how the information could have been included within the rule format. Anderson’s ( 1976) suggestion that representations and processes (declarative and procedural knowledge) be separately specified held out the promise of more precise yet also more economical description. Within his framework, diverse behaviors might be characterized as the products of different processes being applied to a single strategy-free representation. Specifying the representation would force us to consider the structure of the relevant knowledge domains and also the interrelationships among the knowledge domains drawn upon in solving different tasks. Specifying the processes would force us to consider the ways in which this information is manipulated to meet the demands of particular tasks. Therefore, in the series of studies presented here, rather than only concentrating on the processes (rules) that children use in solving numerical problems, we also attempted to specify their representations of numbers. A third lesson that we drew from the conservation experiments is that even in a single problem such as number conservation, a variety of other types of skills might be crucial to children’s success. Consider the skills that were likely involved in using the correct performance rules, Rules IV and V. Children using Rule IV were hypothesized to judge the results of the transformations by counting each row of objects. After counting, the children needed to compare the magnitudes of the numbers to determine which row, if either, had more. Similarly, children using Rule V needed to know about the directional effects of addition and subtraction in order to decide, without enumerating, which row had more. The likely involvement of these skills in the mastery of number conservation was one incentive to single them out for study. Other motivations were even more compelling, though. Counting, comparing, and adding are basic mathemat-

250

Robert S. Siegler urui Mitchell Robinson

ical skill^.^ They are activities in which very young children are interested, in which they frequently engage, and at which they possess considerable skill; thus, they are useful for illustrating what young children do know (Brown, 1978; Gelman, 1978; Ginsburg, 1977). Each of the skills generates a large amount of observable behavior in preschoolers, though not necessarily in older children and adults; such visible manifestations of reasoning provide at least a translucent window for inferring the representations and processes that underlie behavior. The skills have numerous possible interrelationships, in addition to the common involvement in number conservation. For example, Groen and Parkman (1972) postulated that when people add two integers, they compare the numbers to determine the larger ones and then count up the number of times indicated by the smaller (Groen & Parkman, 1972). Finally, the three types of skills are sufficiently diverse that data on them, together with the already collected data on number conservation, should allow reasonably broadly based inferences about children’s knowledge of numbers. In the next three sections of this article, we describe experiments intended to reveal how knowledge of counting, numerical magnitudes, and addition develops during the preschool period.

111. Preschoolers’ Knowledge of Counting A.

COUNTING FROM ONE

Perhaps the first experience that most children have with most numbers is in the context of the counting string. Although words denoting small numbers may have many semantic referents (e.g., two ears, three people in our family, four years old), words denoting larger numbers such as 1 1, 15, or 23 almost certainly do not. Learning about numbers in the counting context may help children learn about them in other contexts as well. Illustratively, Pollio and Whitacre (1970) reported that the length of preschoolers’ counting strings is an excellent predictor of their ability to establish 1-1 correspondence, to divide objects into equally numerous sets, to insert the missing number into a series, and to count on from an arbitrarily chosen point within the number string. The first goal of this series of experiments, therefore, was to establish the representation and process that children apply in abstract counting (i.e., use of the number string in the absence of objects). To date, researchers have reported three major efforts to describe young children’s abstract counting: Fuson and Richards (1982), Ginsburg (1977), and Greeno, Riley, and Gelman (1981). The alternatives posed by Ginsburg and by 31n the report presented here, the only research on arithmetic that will be discussed involves knowledge of addition. Research on subtraction is now being conducted and is certainly part of young children’s knowledge of numbers.

Numerical Understandings

25 1

Greeno and co-workers are quite explicit in their hypotheses about the underlying representation of the number string and therefore are of the greatest interest in the present context. Ginsburg (1977) postulated a great deal of structure in children’s representations of the number string, beginning in the teens: The beginning of the sequence-the first 12 numbers or so-is completely arbitrary. There is no rational basis for predicting what comes after a certain number. Therefore, children have to memorize the smaller numbers in rote fashion. After a period of time, they discover that the numbers after about 13 contain an underlying pattern. Using it, children develop a few simple rules by which to generate the numbers up to about 100. (p. 9)

In contrast, Greeno and co-workers postulated a representation without any particular structure among the numbers. They hypothesized that numbers are simply connected by the “next” relation. Thus, as they noted, their model resembles Peano’s (1899) theory of numbers in its emphasis on the successor relationship. The models of Ginsburg and Greeno and co-workers represent the logically possible extremes in the amount of structure that children might impose on the number string. A third, intermediate possibility also existed: that children might detect and use the relatively transparent structure that appears in the number string beyond the number 20 but not the less obvious structure that is present in the teens. These three models predicted counts that differed in a large number of qualities. Consider just one, stopping points. If, as in the model of Greeno and co-workers, only “next” connections bind the numbers within the number string, there would be no reason for children to be more likely to stop at any one point than at any other; thus, the distribution of stopping points should be relatively flat across the number string. Alternatively, if, as in Ginsburg’s model, children treat the teens as the first decade with a repetitive structure, then they might be expected to stop differentially at points where the structure did not indicate the name of the next number: 19, 29, 39, and so on. Finally, if, as in the third possibility described above, children abstract the structure of the number string only beyond the number 20, then they would be expected to stop relatively often at 29, 39, and 49, but not at 19. Neither Ginsburg nor Greeno and co-workers presented data, beyond anecdotal reports, that supported their model or failed to support other models. We could not locate any other detailed descriptions of young children’s abstract counting either. Therefore, our first experiment was an effort to obtain a data base from which to generate one or more detailed models of the knowledge underlying preschoolers ’ abstract counting.

I . Method The children who participated in this and in most of the subsequent experiments were 13 3-year-olds, 19 4-year-olds, and 10 5-year-olds. Among the 3-year-olds

252

Roh1,t-t S. Sieglrr and Mitchell Robinson

were 6 boys and 7 girls, among the 4-year-olds were 1 1 boys and 8 girls, and among the 5-year-olds were 5 boys and 5 girls. Near the beginning of the school year, when this first experiment was performed, the mean age of the 3-year-olds was 41 months, that of the 4-year-olds 54 months, and that of the 5-year-olds 65 months. All of the children attended a predominantly upper-middle-class preschool. The experimenter here and in all subsequent experiments was a 29-yearold female research assistant. Each child was brought individually to a vacant room in the school, seated at a small table, and given the following instructions: “Today I want you to count for me. I want you to count as high as you can without stopping.” The child then received the prompt “One, two.” When the child stopped counting, he or she was asked, “Do you know what’s after that?” The experimenter then said the last number the child had mentioned with an intonation intended to encourage the child to continue. If the child continued, nothing was said until he or she again stopped, at which point the child once more was prompted to continue. If the child did not continue counting after a prompt, the experimenter told the child that he or she had done a good job and took the child back to the group. Children counted on four occasions, each occasion separated from the next by about 10 days.

2 . Results As predicted by the third counting model described above, examination of the data revealed three distinct patterns: one for children who stopped counting by the number 19, one for children who stopped between 20 and 99, and one for children who proceeded beyond 100. The heights of the children’s stopping points appeared to reflect different levels of counting expertise; children in the three stopping point-defined groups differed in the distribution of digit place values of their stopping points, in the types of omission and repetition errors that they made, and in their likelihood of introducing nonstandard numbers into the counting string. The groupings were highly but not perfectly correlated with age: the 1-19 group included 10 3-year-olds; the 20-99 group included 3 3-year-olds, 15 4-year-olds, and 8 5-year-olds; and the 100+ group included 4 4-year-olds and 2 5-year-0lds.~ a . Sroppingpnints. As can be seen in Fig. 2 and Table I, the distribution of stopping points differed greatly among children in the three groups. Children 4Children were assigned to these three groups on the basis of a measure of the modal tendency of their counting: the highest category that they reached on two or more of their four counts. On any one count, the highest category that a child reached was defined in terms of the stopping point of the count, specifically the highest number within the highest group of three consecutive correctly ordered numbers. The stipulation of three consecutive correctly ordered numbers was adopted to avoid identifying children’s counting performance with occasional isolated high numbers that they mentioned (e.g., a billion).

F R

E Q

U E N

C Y

15

ii

X

x

X X X

10

X

X X

X X

x x X x xx x xx x xx xx xx x x xx x xxxxxxxxx

5

X

xx

x xx xx xxx xx xx xxxxx

X

”

X

X

X

X

x x

n

X

x

x

x

x

x x

x

xx X.

X

x Xx

) I 2 3 4 5 6 7 8 9 I I I I I I I I I 122222222223333333333444444444455555555556666666666777777777788000808009999999999 0 123456789012 34567890 I 2345678901234567890 I23456789 0 I 2 345678901234567890 I 2345678 901 23456789

STOPPING POINT X X

X

xx

X

IXI

X X X

x xxx

X

x xx

xx

11111 1111111111111111111 IIIIIIIIIIIIIIIII I 00000000001 I I I I I I I I12222222222333333333344 012345678901234567890123456789012345678900

STOPPING POINT Fig. 2. Distribution of stopping poinrs in counting.

254

R ( ~ b c S. ~ t Siegler cind Mitchell Robinson

TABLE I Data on Counting Group

Measure “ ~~

Stopping point Counts that stop at 9 Counts that stop at 0 Children who finish at least one count with 9 Children who finish at least one count with 0

~

1-19 ( N = 10)

29-99 ( N = 26)

14

69 4 96 14

10

40 20

loo+

( N = 6)

38 50

83 100

Omissions Counts including any omission Counts with omission of entire decade Children omitting at least one number Children omitting at least one entire decade

70

75

67

0 I00 0

32 96 46

41 83

Repetitions Counts including any repetitions Counts with repetitions of entire decades Children repeating one or more numbers Children repeating one or more decades

92 0 I00 0

52 II 85 15

54 0 I00 0

0

29

12

0

54

33

83

0

0

I00

Nonstandard numbers Counts including nonstandard numbers Children who used nonstandard numbers at least once Nonstandard numbers concatenating decade name with 10, 1 1 , or 12 (e.g., 20-10) Nonstandard numbers concatenating hundreds name to another number (e.g., 100-200)

67

‘I All measures are percentages, and should be read as “Percentage of. . . . ” For example, the first measure should be read as “Percentage of counts that stop at 9.” These percentages reflect 4 counts per child.

who did not count as high as 20 did not display any obvious regularities in the points at which they stopped. The four most common stopping points of children within this group were 4, 7, 8, and 13. In contrast, an absolute majority of the counts of children who stopped between 20 and 99 ended in a “9” number. The four most common stopping points were 29, 39, 49, and 59. The stopping points of children whose counts exceeded 99 showed yet a third pattern. Here, many counts ended in “9” but even more in “0.” The four most common stopping points were 100, 120, 109, and 129. Next we shall consider three departures from the standard counting list: omissions, repetitions, and nonstandard numbers. Our focus will be on three features

Nitnzerirul Undersfandings

255

of the departures: their prevalence (the percentage of children in each group who exhibited the departure at least once), their frequency (the number of departures relative to the total number of counts by children in the group), and their quality (the specific types of departures that were observed).

b. Omissions. In order to quantify the frequency of omissions, we treated each omission as a single instance, regardless of the span of omitted numbers (i-e., a skip from 14 to 16 was treated in the same way as a skip from 29 to 60). As Table I indicates, almost all children omitted at least one number from their counts. The frequency of omissions was found to decrease significantly, though not hugely, with the child’s expertise in counting. The ratio of omissions to numbers counted was 5% for children who stopped before reaching 20, 3% for children who stopped between 20 and 99, and 1% for children who stopped beyond 99. The contrasts among groups were much more striking in the quality of the omissions. None of the children in the least expert group skipped 10 or more consecutive numbers, but roughly half in the two more expert groups did. The ratio of decades skipped to decades counted was 0 in the group that stopped counting below 20, . I 2 in the group that stopped between 20 and 99, and .07 in the group that counted beyond 99. All but 3 of the 43 decade omissions involved jumps from a number ending in “9” to the beginning of another decade (e.g., 27, 28, 29, 50). This finding resembles previous descriptions of adults’ learning to count in bases 2, 3, and 4 (Pollio & Reinhardt, 1970). There, as well as here, the majority of large omissions involved jumps from the highest value of the base unit to a too-advanced later point. c. Repetitions. Almost all of the children sometimes repeated numbers (Table I ) . However, both the frequency and the quality of the repetitions changed markedly with expertise. Viewing each repetition as a separate instance, regardless of the length of the repeated list, we calculated for each counting group the ratio of repetitions to numbers counted. This ratio decreased from 26% for those whose counts did not reach 20 to 2% for those whose counts ended between 20 and 99 to 1 % for those whose counts reached 100 or more. The quality of the repetitions also varied markedly with expertise. Only 3% of the repetitions of children in the 1-19 group and 0% of the repetitions of children in the l O O f group involved more than three numbers. In contrast, 22% of the repetitions of those who stopped counting between 20 and 99 involved at least nine numbers. All but two of the extended repetitions involved a number ending in “9” as the stepping-off point; the child would reach the “9” number and then regress to a number ending in either “0” or “ I ” in an already completed decade.

d. Nonstandard numDers. A sizable minority of the children used at least one nonstandard number in their counts (Table 1). The prevalence of use of these

Robert S. Siegler

256

and Mitchell Robinson

nonstandard numbers differed considerably among the three groups: none of the children who did not count as high as 20 used any such number, but a majority of those who counted between 20 and 99 and some of those who counted beyond 100 did. The frequency of nonstandard numbers relative to the total of numbers counted was 1 % of the total counts in the 20-99 group and 0% for both the 1-19 and the 100+ groups. The nonstandard numbers that were used invariably involved concatenations of standard numbers. Those of children who stopped counting between 20 and 99 generally came immediately after a number ending in 9 and themselves began with the same decade name followed by a 10 (e.g., 29, 20-10). These nonstandard strings ended soon after they began; all but one ended by the time the child reached the decade name followed by a “12. ” Children who counted beyond 100 never produced this type of nonstandard number, but fairly often produced a variant in which they reached 100 and then concatenated hundreds names (e.g., 100- 100, 100-200). e . Stubiliry. Recall that the children counted on four occasions, each separated from the previous one by roughly 10 days. This procedure allowed us to examine the stability of their performance over a 1-month period. In general, children’s counts remained in the same category over successive occasions. Between Sessions 1 and 2, and also between Sessions 2 and 3,83% of the children remained in the same category; between Sessions 3 and 4,90% did. Over the month-long period between the first and last counting sessions, 79% of the children remained in the same counting category. Differences in classifications between sessions were generally in the direction of improved counting. Of the changes observed in adjacent sessions, 74% were toward a higher stopping point; of the changes observed between the first and last session, 89% were toward a higher stopping point. Thus, many of those cases in which counting fell within different categories on different occasions probably reflected improved skill rather than measurement error or random variability in performance. B.

COUNTING ON FROM A POINT BEYOND ONE

The counting of children who did not reach the number 20 differed in a large number of respects from the counting of children who proceeded beyond that point. Two basic types of explanations for these differences seemed plausible. One was that the children who did not count as far as 20 were unaware of the generative rule for connecting decade names with digits which applies in the 20s and succeeding decades. The other possibility was that the children knew this rule but had no occasion to use it, since their counts ended before the point at which the rule becomes applicable.

Numcricut Understuiulings

251

Asking children to count on from various points within and beyond their previously observed counting range was one way to probe the possibility of such unrevealed knowledge. Trials on which children were asked to count on from points bevond their previously exhibited counting range seemed likely to be especially revealing. If a child who stopped counting at 12 knew the generative rule for counting in the 20s, then starting the child by saying “2 1, 22, 23” and then asking him or her to continue seemed likely to reveal the additional knowledge. Similarly, the beyond-counting-range trials seemed likely to reveal whether the counting of children who continued beyond 20 was based on rote learning or on knowledge of an abstract rule. Only mastery of the abstract rule would seem likely to allow children to consistently reach the ends of decades at points beyond those at which they had previously stopped.

I . Method The same children who participated in the earlier counting experiment, with the exception of three children who had moved or declined to participate, were brought back to the experimental room approximately 2 months after their last spontaneous counting session. First, in order to examine whether the children’s counting ability had changed since the earlier assessment, they were asked to count from one, using the same procedure as had been used earlier. Children were divided into the same three expertise groups as previously-0-19, 20-99, and 100+-on the basis of their previous classifications or on the basis of their new count, whichever led to the higher classification. On the basis of their new counts, 6 of the 39 children moved to a higher group, 4 to the 20-99 group and 2 to the 100+ group. The starting points for the counting-on procedure were varied for the three expertise groups, so as to give children the opportunity to count on both from within and from beyond their previously demonstrated counting range. Those children who had not counted beyond 20 were started successively at 3 1, 1 1,41, and 21. Those who had stopped between 20 and 99 were started successively at 7 1, 3 I , 91, and 5 1 . Finally, those who had counted beyond 100 were started successively at 3 1 , 17 1 , 91, and 15 1. Although the starting points were the same for all children within an expertise group, whether each starting point was within or beyond a given child’s counting range was defined individually for that child in terms of his or her previous counts. The criterion for saying that a prompt was beyond the child’s counting range was that he or she had not counted as high as the first number of the prompt on any of the five previous counting trials. Children of all three levels of proficiency were given the following instructions: Today we’re going to play another number game. I’m going to aay three numbers. What you need to do is to continue counting on from where I stop. If I say “ 2 1 , 2 2 , 23,” you would say

258

Robert S.Sicglcr und Mitchell Robinson

“24, 25” and so on. I want you to count on as high as you can go. The first number is (e.g.,) 71: 71, 72, 73.

When the child stopped counting, he or she was prompted in the same way as in the counting experiment described above. Once he or she failed to continue, the child was given a running start for the next beginning point.

2 , Results u . 0-19 group. Children who had not previously counted beyond 19 almost never were able to continue much beyond the experimenter’s initial prompt (Table 11). Not one of the children counted to the next “9.” Following the starts in the 20s, 30s, and 40s, none of the children counted beyond the following ‘ 3 . Thus, these children did not display knowledge of counting beyond what they had exhibited in their previous counts.

”

b. 20-99 group. In contrast, children who had stopped between 20 and 99 almost always reached the following “9. They did so as often when the following “9” number was beyond as when it was within their previously demonstrated counting range. However, they rarely proceeded beyond that “9” number. Thus, children in the 20-99 group demonstrated substantial ability to complete the decade in which they were started, regardless of whether it was within their ”

TABLE II Data on Counting On Group

Criterion “ 1-19

Criterion digit within child’s demonstrated counting competence * Counts on which child reached next ”9” after starting point Counts on which child reached next ”0” after starting point Criterion digit beyond child’s demonstrated counting competence Counts on which child reached next “9” after starting point Counts on which child reached next “0” after starting point

‘‘ All data are given as percentages.

( N = 6)

20-99 ( N = 25)

loo+ ( N = 8)

0

75

I00

0

17

75

0

90

100

0

6

31

Criteria should be read as “Percentage of. . . . ” ’Within child’s counting competence means that the child has previously counted from one to at least the point of the criterion digit.

Numericnl Understundings

259

previously demonstrated counting competence, but did not exhibit additional knowledge of decade connections. c . 100+ group. Children who had counted beyond 100 showed yet a third pattern. They always completed the initial decade, usually went on to the next decade when it was within their counting range, and at times went on to the next decade even when it was beyond any of their previous counts. In short, they demonstrated considerable knowledge of the within-decade structure and some knowledge of between-decade connections even when these were beyond their previously demonstrated counting range. C.

MODELS OF THREE LEVELS OF COUNTING EXPERTISE

These data reveal a number of characteristics of children’s counting for which any developmental model would need to account. Children who did not count as high as 20 showed no obvious pattern in their stopping points, often omitted or repeated one or a few numbers, did not omit or repeat whole decades, used no nonstandard numbers, and did not count on to the end of the decade when given a three-number running start. Children whose counts ended between 20 and 99 usually stopped at a number ending in “9,” often omitted or repeated entire decades, used relatively many nonstandard numbers, most of which involved concatenating decade names with 10, 1 1 , or 12, and counted on to the ends of decades both within and beyond their counting range. Finally, children who counted beyond 100 generally ended their counts at a “9” or at a “0,” occasionally skipped whole decades, occasionally formed nonstandard numbers by concatenating hundreds names, and, when asked to count on, both completed decades and went on to later decades within and beyond their previously exhibited counting range. Before we describe models to account for each of these three counting patterns, it may be desirable to consider some of the factors that would inherently influence the way in which English-speaking children learn to count. Of particular importance are the information that is present in the counting string and the process through which children most likely acquire that information. The sequential nature of counting seems to us critical in determining the acquisition process. Since counting is a sequential activity, children may not grasp the structure inherent in later parts of the number string until their mastery of earlier parts is fairly complete. Thus, children who do not count as high as 20 may not know that there is any structure to the number string because there is little obvious structure in the part of the string that they do know. Children who count even a few numbers beyond 20 could notice that there is a definite structure beyond that point, involving two special sets of numbers and a generative rule connecting them. The generative rule states that whenever a number from one set (i.e., decade names from 20 to 90) is used or heard, successive numbers can be

260

Robert S . Siegler und Mitchell Robinson

generated by combining that number with members of a second set (i-e., digit names between 1 and 9) until the numbers in the second set are exhausted. Children who count beyond 100 could note an additional rule involving a third set of numbers (i.e., hundreds names between 100 and 900). In this rule, whenever a member of the third set is used, successive numbers can be created by saying the member of the third set before using the procedure for generating numbers below 100. This general description suggests several types of knowledge that children would need to count beyond 100: knowledge of the memberships of the three special sets, knowledge of the two generative rules, and knowledge of the “next” relations specifying connections that are not predicted by the generative rules. At this level of description, counting would seem to be a fairly simple skill. In attempting to build models that generate counting performances of the types we observed, however, the complexity of the knowledge underlying counting becomes apparent. Three models, one corresponding to each level of counting expertise, are shown in Fig. 3. Each of these illustrates representations and processes capable of generating a distinct pattern of counting and counting on. Note that certain parameters of the models will vary for individuals within each group. One child might know that 15 follows 14, while another might think that 16 follows 14; one child might know that 40 follows 39, while another might not; and so on. Thus, the models do not yield completely determinate predictions of individual performance. However, they do yield specific predictions at the level of groups, and the particulars of each model could be filled in to produce separate, determinate predictions for each child. As will be seen, the models are sufficient to account for the main features that differentiate among the three groups of children. Model I depicts the knowledge hypothesized to underlie counts that ended before 20 (Fig. 3A). This model resembles that of Greeno and co-workers in that the representation includes no particular structure beyond “next” connections. The process is also uncomplicated. First a starting point is chosen; unless some specific point is requested, as in the counting-on instructions, children start with “1 .” They then say the next number if they can recall it, and continue for as long as they have “next” connections. When they reach a point where they do not have a “next” connection, they either arbitrarily choose a number or stop. As Fig. 3B illustrates, the representation and process used by children who ended between 20 and 99 (Model 11) are hypothesized to be considerably more complicated. Within the representation, numbers can be tagged as members of two lists: the digit repetition list and the rule applicability list. The numbers 1 through 9 are the ones that most often will be on the digit repetition list, though some children’s lists may be too short or too long. The functions of this list are to indicate which numbers can be connected with decade names and to avoid the need for separate “next” connections between each successive pair of numbers. Decade names starting with 20 can be on the rule applicability list. This list is

Numerical Understarulings

26 I

less likely than the digit list to be complete for Model I1 children. The function of the rule applicability list is to indicate the places where the generative rule, involving the concatenation of the decade name with each of the members of the digit list, can be used.’ The Model I1 process, shown on the bottom of Fig. 3B, operates on the representation as follows. (It is crucial in this example to switch back and forth between the text and the diagram in order to understand either.) Assume that a child is asked to count as high as possible and that this request meets the child’s conditions for using the counting process. Since the child was not asked to count from any specific number, she or he assigns N , the starting point, a value of one. The child says “one. ” The word “one” does not include any rule applicability list member, so the program branches to the reader’s right. The child then chooses two as the next number, since it is at the end of the “next” link from one. The child says “two.” Two also does not include a rule applicability list member, so the child retrieves three and says it. The process continues until the child retrieves and says 20. Twenty is a member of the rule applicability list, so that when the question about the rule applicability list member in the number name is posed, the answer is “yes,” and the program branches to the reader’s left. The first question in this part of the program is whether there is a digit list member in the number name. The name 20 does not contain any digit list member, so the next number on the digit list is set equal to one. One is retrieved as the next number on the digit list and is concatenated with 20 to produce 21. The child says “21 ” and returns to the question of whether there is a digit list member in the name. Now there is a digit list member, one. The question of whether the digit equals 9 is raised; since it does not, the child retrieves the next number on the digit list, two, concatenates it with 20 to produce 22, and says “22.” This process continues until the child has generated and said 29. Then the digit does equal 9. Since the number 29 has a specific “next” connection, the child says “30” and repeats the generative cycle.6 When the child reaches 39, 5The decision to call this list the rule applicability list rather than the decade name list may cause some immediate confusion but should lead to greater clarity in the long run. The set of numbers could not accurately be called the decade name list because 10 is a decade name yet cannot be concatenated with digits to form other numbers. What distinguishes 20, 30, and successive decade names is precisely that with them, the generative rule is applicable-hence the name “rule applicability list members.’’ “The decision to connect each member of the rule applicability list to the preceding “9” number rather than to the preceding rule applicability list member was not arbitrary. We tested several 5-year-olds’ ability to count by 10s. a skill that they presumably would have if the decade names were linked together. Specifically, the experimenter asked children to count by 10s. prompted them by saying “10, 20,” and then asked them to continue. None of the children was able to count in this way, although all had previously counted by ones beyond 30. Thus, it seemed likely that the children’s interdecade connections were between “9” numbers and the next rule applicability list member, rather than between successive rule applicability list members.

262

Robert S . Siegler and Mitchell Robinson

A REPRESENTATION: Next Next Next Next 19-3-43

Next Next Next A 12

9?Oyl

PROCESS:

Listening

I\

stop -

From Number List

V

. I

263

Numericul Undersiundings

8 REPRESENTATION

Neil Next NerlNext Ip2q*4p5*

*ON DIGIT L I S T

Nexl

9 C O

Neil 18 9 -

5: 20**

Next 2 9 7 d c . t .60**

BO** .40**

**ON RULE A P P L I C A B I L I T Y L I S T

PROCESS

Program

Lbslening

"From N"

Yea Number = N

Say First Number

Choose " N e d Number

Say Number

+

Fig 3 ( 8 ) Model I!

See next page for Flg 3 ( C )

264

Robert

S.Sicgler and Mirchrll Robinaon C REPRESENTATION.

Neal NeilNaatNexr

NextNeil

Ie2e3e4*5*

A9e10

*ON DIGIT L I S T

Neal Neat

Neil

Id9?0**

29?0**

* * O N RULE APPLICABILITY LIST

NeXI

59-60**

***ON

Next= lOd?Ol

99

HUNDREDS LIST

PROCESS:

Proqram

F] First Number

~

Ignore Hundreds List Member Until Encounter "Say Number" Inslruclian Then Soy 11 Before Resl

Yes

t

, \

Next Number on Dial1 LISI

93

i

Wan1

on Diqil L1r1 Gel N e i l Number Y

y

h

Z

SlOQ

Choore "Neal" Number

1 Diqil Wilh Rule Lisl

L

Say Number

u

J

Fig. 3(C): Model I11

Say Number

7

Nutnericul Understandings

265

however, there is no specific “next” connection. Therefore, the child reaches an important choice poinr that accounts for many of the distinctive phenomena in this group’s counting data. The child must decide whether or not to continue. A decision not to continue will result in a stop at a “9” number, in this case 39. A decision to continue will result in an arbitrary choice of a number from the child’s rule applicability list. This could result in an omission if the number chosen is too far advanced (e.g.. 39, go), in a repetition if the number chosen is not far enough advanced (e.g., 39, 20), or, fortuitously, in the choice of the correct answer (e.g., 39, 40). The process continues until the child reaches a “9” number without a “next” connection and decides to stop. The Model I1 representation incorporates only two changes from that of Model 11: the addition of the hundreds list, and the completion of the rule applicability and digit lists. The Model I11 process also works in much the same way as that of the previous model. It proceeds identically until the number 100 is reached. Then, the children note that there is a hundreds list member in the number. They ignore the hundreds list member for purposes of forming subsequent numbers, but remember to add it for purposes of saying them. Remembering to do this may be of more than trivial difficulty; young children in our sample who could count beyond 100 often momentarily forgot the hundreds term at one or more points in their counts (e.g., “124, 125, 26, uh, I mean 126”). This is one reason why within Model 111, the instruction to remember to say the hundreds name is separated from the point at which children actually say it. Model 111 children can continue counting for a very long time. Such factors as fatigue and boredom seemed as likely as lack of knowledge to determine their stopping points. Therefore, Model I11 includes a slightly altered stopping rule. When children’s desire to continue becomes sufficiently slight, they stop at the next point at which the job would be complete-either at the next ‘‘9” or at the next “0” (e.g., at 129 or 130). These three models account for the major features of the counting of children who stopped within the corresponding ranges. First, let us consider stopping points. Model I does not imply any particular distribution of stopping points, since there is no obvious way of predicting which particular “next” connections will be missing. Model I1 predicts that stops will occur at numbers ending in “9.” Children stop because the number ending in ‘‘9” does not have a “next” connection. Model I11 suggests that stops should occur at either a “0” or a ‘‘9,’’ points at which the job is complete. Now let us consider omissions. Model I children’s omissions could come from two sources. First, within the Model I representation, children might have an incorrect “next” connection; illustratively, Fig. 3A indicates that they would say “6” after “4.” Second, within the Model I process, children can arbitrarily choose successor numbers when no specific “next” connection exists; within Fig. 3A, this could lead to their skipping from 6 to any other number that they

266

Robert S.S i e g l i v rind Mitchell Robinson

knew. Model 1 does not directly predict the lengths of the omissions that would arise from these mechanisms. However, as exemplified by the particular incorrect link within the Fig. 3A representation, omissions arising from the first source might be expected to be brief, due to children tending to form “next” connections between numbers that were close together in the strings of other people whom they had heard counting. Model I1 predicts that brief omissions should rarely occur (there is still the possibility of an incorrect “next” connection) but instead predicts the omission of entire decades. Such omissions would occur when children reach a number ending in 9 that does not have a “next” connection and arbitrarily pick a higher number from the rule applicability list. Within Fig. 3B, children might reach 39, not have a “next” connection, and arbitrarily skip to 60 or 80. Model I11 predicts no omissions, short or long. The predictions of the three models concerning repetitions are similar to those concerning omissions, not surprisingly since they are hypothesized to be produced by the same mechanisms. Model I could produce repetitions through two mechanisms: incorrect “next” connections, and arbitrary choices of numbers when no “next” connection existed. Model I1 could produce repetitions of entire decades following numbers ending in “9” for which the child did not possess a “next” connection. Model 111 would produce no repetitions of any length. The frequency and types of nonstandard numbers produced by the three models would also differ. Children who used Model I would not be expected to generate nonstandard numbers. Model I1 children who had just begun to use the digit repetition list would produce such numbers if the boundaries of their digit lists were too high.7 For example, rather than the digit repetition list ending at 9, as in Fig. 3B, it might end at 11. If this were the case, we might expect children’s nonstandard numbers to be relatively brief continuations of the standard digit repetition list (it seems unlikely that their boundaries would be off by very much). Children who used Model I11 would presumably have overlearned the digit list, but might have difficulty in remembering the recursive procedure for forming numbers with hundreds list members; this could lead to nonstandard concatenations of hundreds list members. Finally, consider the counting-on data. Model I includes no mechanism for children to count on if they are started beyond their counting range; they would lack the necessary “next” connections. Children using Model I1 would be expected to be able to count on to the end of the decade from any point that was ’Cases in which the child’s digit repetition list did not extend quite far enough may well have occurred, but were difficult to detect from the production data. The most direct data concerning this possibility were the relatively large numbers of stops at 26, 27, and 28 shown in Fig. 2. The children who stopped at these points often did not proceed beyond 29 on any of their counts and had a higher than average frequency of nonstandard numbers. These findings were consistent with the hypothesis that these children were still in the process of mastering the digit repetition list.

Nurnericd Understundings

267

recognized to include a rule applicability list member and to continue to the next decade if the “next” connnection was known. Even if a child using Model I1 did not yet know that a particular number beyond his or her counting range was on the rule applicability list, he or she might infer its status from hearing it concatenated with one, two, and three successively (in the same way that an adult who heard someone say “blipty-one, blipty-two, blipty-three” would probably continue to blipty-nine). Children who used Model 111 would be expected to perform similarly to those who used Model 11. Almost all of these predictions of the models fit the children’s counting. Only one observation proved recalcitrant: the failure of Model I1 children in the counting-on experiment to count beyond the end of the decade within which they were started, even when their previous counting from 1 demonstrated knowledge of the relevant “next” connection. Illustratively, 7 children who had previously counted flawlessly to 49 or higher did not progress beyond 39 when asked to count on from 3 1. This is not an easy finding to explain. The best that we can do is to place it in the general framework of overlearning phenomena. It seems almost certain that the “next” relations connecting digit repetition list members with each other are overlearned to a much larger degree than those connecting “9” numbers with rule applicability list members. Such overlearning may enable young children to access the “next” relations on the digit repetition list even when they have not found their place in the counting sequence; they may approach the task as if the decade name were a nonsense syllable (i.e., the bliptyone, blipty-two example). Admittedly, more research is needed to explain the phenomenon (cf. Fuson & Richards, in press, for a similar finding on a “What is the next number?” task). With this single exception, the Fig. 3 models were consistent with the major features of the data. The distributions of stopping points, omissions, repetitions, and nonstandard numbers were what the models would predict, as were all but the one finding on children’s counting on within and between decades.

IV. Preschoolers’ Knowledge of Numerical Magnitudes A.

EXISTING RESEARCH ON ADULTS AND CHILDREN

A second central aspect of children’s knowledge of numbers is their knowledge of numerical magnitudes. Research in this area was greatly stimulated by the report by Moyer and Landauer (1967) that comparing the magnitudes of digits of discrepant sizes (e.g., 4 and 8) took less time than comparing the magnitudes of digits of more comparable sizes (e.g., 4 and 5). At the time, this symbolic distance effect must have appeared quite surprising, because the exper-

268

Robert

s. Sii,gler

cind Mitchell Robinson

iment was replicated at least five times within the next 5 years (Aiken & Williams, 1968; Fairbank, 1969; Knoll, cited in Moyer & Bayer, 1976; Parkman, 1971; Sekuler, Armstrong, & Rubin, 1971). The symbolic distance effect was especially intriguing because several of the most obvious models of numerical comparison were directly contradicted by the data. One way that numerical comparison problems could be solved would be to count from 1 until encountering one of the numbers; that number would be the smaller, and the remaining number would be the larger (Parkman, 1971). Such a model does not account for the distance effect, however; it predicts that solution time should be a function of the size of the minimum number, but not of the split between the two numbers. Another way to solve magnitude comparison problems would be to count on from one number until the other was found; if it were not found in some period of time, the count would be initiated from the other number. This model would predict a reverse symbolic distance effect, however; the closer the two numbers, the sooner the second number should be encountered. A third possibility, direct associative retrieval, would seem to yield equal solution times for all comparisons, again different from the observed pattern. As Moyer and Landauer (1967) pointed out, however, the psychophysics literature provides a well-documented analog to the symbolic distance effect. Psychophysicists long ago noted that the more discrepant on some dimension the magnitudes of two visible or audible stimuli, the faster they can be compared. Another well-known psychophysical phenomenon also proved to be present in the abstract comparisons; the smaller the size of the smaller stimulus, the quicker is the judgment (e.g., 2 and 4 can be compared more quickly than 4 and 6). This has been labeled the min effect. To account for these parallels, several investigators have hypothesized that physically and symbolically presented magnitudes share a common underlying representation (Banks, 1977; Moyer & Dumais, 1978; Shepard & Podgorny, 1978). For numerosity, this representation is assumed to resemble a logarithmic scale, with the values at the high end of the scale more densely packed than the values at the low end. The symbolic distance effect is said to arise from the fact that it is easier to discriminate between points that are far apart in the representation; the min effect is attributed to representations of magnitudes of the same linear disparity being farther apart at the low than at the high end of the scale. Multidimensional scaling of the results of numerical comparison tasks (Shepard, Kilpatric, & Cunningham, 1975), random number production tasks (Banks & Hill, 1974), and subjective magnitude judgment tasks (Rule, 1969) have provided converging evidence for this view of the representation of numerical magnitudes. Disagreements have arisen over several other characteristics of the representation, however. Two broad classes of models have emerged: analog and discrete. The central assumptions underlying analog models are that representations of all

Numerical Under.standin#s

269

types of magnitudes preserve continuous information about physical size, and that these analog values are compared directly, perhaps through some type of random-walk process (e.g., Moyer & Dumais, 1978; Poltrock, 1980). In contrast, the central assumption underlying discrete models is that magnitudes are grouped into categories, each of which carries a semantic code (e.g., large, small) and that it is the codes which are compared. Illustratively, in one discrete model, Banks, Fujii, and Kayra-Stuart (1976) suggested that each digit is associated with the code “big” with a probability proportional to the logarithm of that number, and with the code “small” with a probability of one minus its logarithm. When asked to compare numerical magnitudes, people generate codes for the two numbers, answer if the codes discriminate between the numbers, and, if they do not, regenerate the codes until they do discriminate between them. Substantial bodies of data have been collected supporting each type of model, though it is our impression that the discrete models have been most consistently in accord with the data on numerical magnitude comparisons. One limitation of both the analog and the discrete models is that their proponents fail to specify whether they are intended as characterizations of long term memory contents or as characterizations of temporary data structures formed for the purpose of performing a particular experimental task. The models are presented as if the analog or discrete representations were retrieved directly from long-term memory, but this view seems implausible. Adults can compare the magnitudes of a vast set of numbers, yet almost certainly do not possess in long-term memory either analog values or categorical labels corresponding to each number that they know. Interestingly, the force of this reservation may not be nearly as great with preschoolers, a population to whom the models have not been applied, as with adults. In a recent experiment (Siegler & Robinson, 1981) we found that preschoolers most often label 5 as a medium-size number and 9 as a big number, regardless of whether the context is the numbers 1-9 or the numbers 1-1 trillion. Adults also label 5 as medium size and 9 as big in the 1-9 context, but, not surprisingly, refer to both as small numbers in the 1-1 trillion context. Thus, although both analog and discrete models probably reflect temporary data structures in adults, they more plausibly reflect long-term memory representations as well as temporary data structures in very young children. As alluded to in the paragraph above, despite the large amount of work on adults’ understanding of numerical magnitudes, very little is known about how such understanding develops. We were able to locate only two published studies involving young children: Sekuler and Mierkiewicz ( 1977) and Schaeffer, Eggleston, and Scott (1974). Unfortunately, each of these studies has important limitations that prevent us from drawing strong conclusions from them. Sekuler and Mierkiewicz ( 1977) presented magnitude comparison problems involving the digits 1-9, with equal numbers of items having each level of split from 1-8. They found that the solution latencies of children as young as 5 years

270

Robert S. Siegler and Mi/che/l Robinson

showed an effect of symbolic distance. However, their analyses averaged items over values of min, and therefore did not distinguish between the effects of minima and those of symbolic distances, a serious defect since the two variables were highly correlated within their problem set. This omission would be unfortunate with adults, where the absolute size of the numbers has been found to exercise an effect independent of symbolic distance, but seems even more serious with young children, whose knowledge of small and large numbers has been found to differ greatly in many situations (e.g., Fuson & Richards, 1982; Gelman & Gallistel, 1978; Ginsburg, 1977; Siegler, 1981a). Schaeffer, Eggleston, and Scott’s experiment exhibits different but equally noteworthy problems. They presented 2- to 5-year-olds ’ numerical magnitude comparisons involving the digits 1-9 and having a split of either 1 or 4 digits. They interpreted their results as indicating that children divide numbers into small and large categories and that they learn to perform numerical comparisons in the order: (1) large distance, with numbers falling into different categories; (2) large distance, with both numbers in the same category; (3) small distance, with both numbers within the small number category; and (4) small distance, with both numbers within the large number category. It is difficult to know how to evaluate this account. Schaeffer et al. (1974) did not describe the data that led them to believe that young children divided numbers into small and large categories. Their classification criteria were inconsistent; in comparisons involving large distances, the boundary between the small and large categories was placed at 5, but in comparisons involving small distances, it was placed at 4. Further, it is not clear why crossing a category boundary was essential in determining ;he difficulty of comparisons involving large distances [Stage 1 versus Stage 2) but not in determining the difficulty of comparisons involving small distances (Stage 3 versus Stage 4). Thus, at present, we know little about how understandings of numerical magnitudes develop. This lack of knowledge is unfortunate, not only because of the inherent place of such knowledge within children’s understandings about numbers, but also because developmental research would seem to have substantial potential to illuminate some of the general issues that have arisen in the magnitude comparison literature. As Banks (1977) pointed out, almost all magnitude comparison studies have involved one of two types of stimulus materials: overlearned material, such as letters and digits, and arbitrary material, such as different-colored sticks or nonsense syllables. Neither of these types of material seems ideally suited to studying acquisition processes. In the first case, acquisition is already complete; in the second, the material being acquired is inherently unrepresentative of the semantically rich material that people learn about in the world outside the laboratory. In contrast, studying children’s acquisition of the ordering of real-world materials such as numbers would seem to overcome both of the problems. The materials are semantically rich (Lehman, 1979), yet can be studied at a point where they have not been entirely mastered.

Numeric,ril Understandings

271

The purpose of our first numerical comparison experiment was therefore to obtain a more detailed picture of knowledge of numerical magnitudes at ages where such knowledge is sufficient to produce above-chance comparison performance but insufficient to produce consistently correct performance. B. PRESCHOOLERS’ NUMERICAL MAGNITUDE COMPARISONS

1 . Method The same 39 children who participated in the counting-on experiment were brought back to the experimental room and given the following instructions: Today we’re going to help Bunny and Monkey (2 stuffed animals that were present) to play a card game. I’m going to hold up 2 cards so that I can see the numbers. 1’11 tell you what the numbers are, and you need to tell me whether Bunny’s number is more or whether Monkey’s number is more. OK, Bunny has 7 and Monkey has 5 . Which one is more: 7 or 5?

Similar questions were asked for all 36 nonidentical pairs of the digits 1-9. If a child’s interest appeared to be flagging, the session was terminated for that day and continued the next. Children were periodically told that they were doing well, regardless of their actual level of performance. Each participant was presented the group of 36 items on four occasions. The items were presented in one randomly generated order on the first and fourth occasions and in the reverse order on the second and third occasions. The order of mention of the numbers within each problem (e.g., “Which is bigger, 9 or S?” versus “Which is bigger, 5 or 9?”) was reversed on the second and fourth trials from the order on the first and third trials. The four presentations occurred at roughly 10-day intervals. Thus, as in the first counting experiment, the testing period lasted about 1 month.

2. Results The period from 3 to 5 years proved to be one in which considerable development occurred in children’s ability to compare digit magnitudes. Three-year-olds were correct on 56% of their comparisons, 4-year-olds on 81%, and 5-year-olds on 90%. In order to obtain an unbiased estimate of the children’s knowledge, the effect of chance was corrected by the formula (I.

Absolute level of pe$ormanre.

knowledge

=

(observed percentage correct - chance percentage correct) ( 1 - chance percentage correct)

Since the chance percentage correct was 50%, inserting into the formula the above data on observed percentage correct indicated that 3-year-olds knew 12% of the comparisons, 4-year-olds 62%, and 5-year-olds 80%.

272

Robert S . Siegler cind Mitchell Robinson

TABLE 111 Percentage of Variance Accounted for by Alternative Predictors of Magnitude Comparison Errors Model Banks Age of child

Linear (minisplit)

Welford function [log (max/split)]

Parkrnan (niin)

[ A (log(rnin))+ B ( 1 - log(max))]

5-year-olds 4-year-olds

52 69

55 70

48 42

41 55

b. Predictors of performance. We next applied the same types of regression analyses to the young children’s error patterns as previously have been employed in examining adults’ reaction times. The variables that were used to predict the percentages of errors on the 36 problems were the size of the minimum number, the size of the maximum number, the distance between the two numbers, and the sum of the numbers. None of these variables proved significantly predictive of 3-year-olds’ performance. In contrast, both the size of the minimum number and the distance between the numbers proved predictive of both 4- and 5-year-olds’ behavior. In the case of the 4-year-olds, the two variables accounted for 69% of the variance in the number of errors on the 36 problems, while in the case of the 5-year-olds, they accounted for 52%. For each age group, the size of the minimum number accounted for the larger percentage of variance: for the 4-year-olds, rmin.split = - -68, while rsplit.min = -48; for the = - .59, while r,plit.min= .28. The linear combination of 5-year-olds, rmin.s,,li, min and split accounted for at least as much of the variance in children’s errors as several nonlinear combinations of the variables that have been proposed as functional models in previous studies; the respective figures are shown in Table 111.

c . Consistency of performance. The children’s percentages of correct answers showed a fair degree of consistency over the four occasions. For example, on the first and second occasions, the 3-year-olds’ performance correlated r = .43, the 4-year-olds’ performance correlated r = .79, and the 5-year-olds’ performance correlated r = .58. Across all three age groups, the correlations between performance on the two occasions was r = .79. The data on consistency of performance on the same problems over different occasions were also revealing. Three-year-olds were correct on all four occasions on 10%of problems, correct on some but not all occasions on 87%, and incorrect on all occasions on 3%. The corresponding percentages for 4-year-olds were 54, 45, and 1%, respectively, and for 5-year-olds 67, 33, and 0%, respectively. The lack of systematically incorrect responding was striking for all three age groups. It suggested that any model of young children’s numerical magnitude comparisons could not be completely determinate in its predictions of the prob-

Numericul Understundings

273

lems on which each child would err; rather, the model would need to include one or more probabilistic processes that would sometimes but not always lead to errors on given problems. d. Representations of numerical magnitudes. What representations of numerical magnitudes might account for the obtained error patterns? One method that can be used to address such questions is multidimensional scaling. As input, multidimensional scaling routines take data concerning similarities or proximities among the possible pairs of stimuli. The percentage of errors can be used as one index of similarity, the assumption being that errors reflect the difficulty of discriminating the points in some type of representational space. As output, multidimensional scaling routines produce spatial arrangements of the individual stimuli that minimize stress, stress being a badness of fit measure. The particular multidimensional scaling procedure that we used to examine children's magnitude representations was the nonmetric version of KYST, with stress formula 1 and the primary method for dealing with ties. In line with Kruskal and Wish's (1978) suggestion, we limited our consideration of possible dimensionalities to 2, since 9 stimuli were being scaled. The input to the scaling algorithm was the percentage of errors that children made in comparing each of the 36 possible pairs of digits from 1 to 9; the output was an arrangement of the 9 digits in a one- or two-dimensional representational space. The most striking result of the multidimensional scalings was that the numbers did not fit especialIy well into the compressive logarithmic function generally believed to characterize representations of numerical magnitudes. Rather, they seemed to fall into clusters, with quite small distances within clusters and quite large distances between them. Consider, for example, the data of the 4-year-olds shown in Fig. 4A. There are some reasons for preferring the two-dimensional representation over the one-dimensional one-notably that the stress declines from .24 to , 1 1-but even the one-dimensional representation does not particularly closely resemble the hypothesized logarithmic representation. Rather, the numbers seem to arrange themselves into four clusters: (1). (2, 3), (4, 5 ) , and ( 6 , 7, 8, 9). This finding motivated us to reanalyze the reaction time data of Sekuler and Mierkiewicz (1977) with slightly older children. The scaling results with their 6-year-olds are shown in Fig. 4B. Again, note that the results do not fit especially well the compressive logarithmic function postulated previously; rather, the numbers seem to fall into a few clusters. In order to objectively test the impression of clustering in our own error data, we next performed hierarchical clustering analyses (using the diameter method: see Johnson, 1967). The results for the 4-year-olds are shown in Fig. 4C.As can be seen in the patterns of ovals (stimuli within the smallest number of ovals are the most similar), the clusters are quite similar to those revealed impressionistically .

A Scalinqs of 4-year-olds' Errors (Our Data) Stress

One Dimensional Solution

I

2

3

5

4

678

Two Dimensional Solution

9

,24

.II

2 3 7

a 9 6

4

0 Scolmgs of 6-year-olds' Solution Times (Data from Sekuler 8 Mirkiewicz (1977) One Dimensional Solution

2 4 3

657

Stress

9

8

.24

Two Dimensional Solution .I0

I

C Hierarchlcal Clustering of 4-Year-OWi Errors (Our Data1

Fig, 4 . Multidimensional scaling solutions f i r mngnitude compurison data: ( A ) sciilings (?f 4-yeur-olds' errors (our data): ( B ) scalings of 6-year-o1d.v' solution times (duta from Sekuler & Mierkiewicz. 1977; (C)hierurchical clustering of4-yeur-olds' errors (our data).

Numericul Undersfandings

275

The emergence of two distinct dimensions in the multidimensional scaling solutions implied that relative difficulty of the magnitude comparison problems was determined by the two numbers being in the same or in different clusters, rather than by any differential distance between clusters. In order to objectively test this hypothesis within our own data, a regression analysis was performed using between- versus within-cluster status as a predictor of the relative difficulty of the 36 comparison items. The cluster membership variable alone accounted for 63% of the total variance in the number of errors, not that different from the 73% that could be accounted for using the absolute distances between digits in the scaling solution as the predictor. We concluded from these data that withinversus between-cluster status of the comparison items was a major determinant of problem difficulty. Examination of the scaling and clustering analyses also suggested possible explanations for the min and split effects. In all of the analyses, the digit “ 1 ” was in a cluster by itself; this placement would contribute to both min and split effects, since comparisons involving 1 have the lowest possible minimum and on average will have the largest possible splits. Also, in all of the scalings the number of numbers within the clusters at the small end of the scale is smaller than the number of numbers in the clusters at the large end; again, this distribution suggests a greater probability of the relatively easy between-category comparisons when mins are small and splits are large. Thus, several sources of evidence were consistent with the notion that young children’s representations of numerical magnitudes were neither undifferentiated nor closely correspondent to a logarithmic function. Rather, the digits seemed to be represented in terms of a small number of clusters. It must be noted, though, that all of these sources of evidence involved the end products of magnitude comparisons. The children’s categorizations of the numbers could be inferred only from their overail performance on the magnitude comparison task, a performance that was produced by a process as well as a representation. In order to obtain more direct evidence about children’s representations of numerical magnitudes, we therefore thought it important to determine (1) whether children’s categorizations of the digits, measured independently of their magnitude comparisons, were related to their magnitude comparison performance; and ( 2 ) if the relationship was present, whether teaching children a new clustering scheme would influence their pattern of correct answers and errors in comparing magnitudes. The two experiments reported below were addressed to these questions. C.

VERBAL LABELING OF NUMBERS

In both Trabasso’s (1977) and Banks, Fujii, and Kayra-Stuart’s (1976) discrete category models, people were said to attach a semantic code to each stimulus being compared. The only empirical support for this claim, however, was that it was consistent with the semantic congruity effect. In the experiment presented

276

Robert S.Siegler und Mitchell Robinson

here, we attempted to study more directly children’s semantic coding by examining the verbal labels that they assigned to numbers. In particular, we were interested in (1) whether the labels children assigned to numbers could potentially have been useful if they applied them in comparing numerical magnitudes; and (2) whether the quality of children’s labeling, considered across the digits, was predictive of performance on the magnitude comparison task. 1 . Method Children were brought back to the experimental room roughly a week after their last magnitude comparison session and given the following instructions: Toddy I’m going to ask you some questions about the numbers from 1 to 9. Some of these numbers are big numbers, some are little numbers, and some are medium numbers. I’m going to say a number, and you need to tell me if the number is a big number, a little number, or a medium number.

Then the experimenter said a number and asked, “OK, is N a big number, a little number, or a medium size number?” The numbers were arranged in random order, save for the restriction that each third of the presentation order needed to include one number from among the digits 1-3, one from among the digits 4-6, and one from among the digits 7-9.

2 . Results u . Absolute level of performunce. One concern that we had before conducting this experiment was that the children would not understand the task. This fear proved groundless. Even the 3-year-olds demonstrated that they understood what they were being asked to do. The mean value of the numbers that they termed small was 3.14, of the numbers that they labeled medium 5.15, and of the numbers that they labeled big 6.00. More 3-year-olds labeled each of the numbers 1 , 2 , and 3 as small than labeled them as big; more of them labeled 7 , 8 , and 9 as big than labeled them small. Thus, even the youngest children’s labeling demonstrated some knowledge about the magnitudes of the numbers. The 3-year-olds’ relatively accurate labeling does not imply that there was no development in skill at applying the labels. Older children’s labelings were much more differentiated in terms of the numbers they included. By age 5, the mean values of numbers assigned the small, medium, and large labels were 2.16, 5.12, and 7.75, respectively. Illustratively, 9 of the 10 5-year-olds labeled the number I “small,” 9 of the 10 labeled the number 5 “medium,” and all 10 labeled the number 9 “big. ”

b. Potential usefulness of labels. The first main issue addressed in this experiment was whether children’s labeling of the numbers could have been

Numerical Understandings

277

useful on the magnitude comparison task. In order to examine this issue, we needed to quantify the implications of the labels for the magnitude comparisons. Suppose that a child used only his or her labels in comparing numerical magnitudes. Such a procedure could lead to correct answers (e.g., if 7 was labeled “big” and 5 was labeled “little”), to incorrect answers (e.g., if 7 was labeled “medium” and 5 was labeled “big”), or to indeterminately correct answers (e.g., if both 7 and 5 were labeled “medium”). Therefore, as a first rough estimate, the quality of each child’s labels was quantified by examining the labels assigned to each of the 36 possible pairs of digits and assigning a score of 1 on pairs on which the labels would lead to the correct answer, a score of 0 to comparisons on which the labels would lead to an incorrect answer, and a score of .5 on comparisons on which the labels would lead to an indeterminately correct outcome.a Under this scoring system, a child who assigned labels to numbers haphazardly would be expected to answer correctly on 50% of the problems, as would a child who assigned the same label to all 9 digits. The best possible score would be produced by a child who labeled I , 2, and 3 as small numbers, 4, 5, and 6 as medium numbers, and 7, 8, and 9 as big numbers; this strategy would produce an expected value of 88% correct performance. This measure indicated considerable improvement with age in the quality of children’s labeling. The 3-year-olds’ labeling would have led to an average of 58% correct answers, the 4-year-olds’ to 75%, and the 5-year-olds’ to 80%. Thus, the 4- and 5-year-olds’ labels were potentially useful on the numerical comparison task; if the children had used them, they would have performed at a level well above chance. The number of correct answers predicted by the above measure of the quality of each child’s labels also proved to be closely related to the number of correct answers the child had produced on the numerical comparison task. Across the three age groups, the correlation was r = .80. The correlations were also substantial within each of the three age groups: r = .73 for the 3-year-olds, Y = .55 for the 4-year-olds, and r = .64 for the 5-year-olds. In addition to the labels predicting children’s number of correct answers on the magnitude comparison task, they also predicted the relative difficulty of the individual problems. Consider the results with the 4-year-olds. First, we calculated the probabilities, over all of the 4-year-olds, that each label was assigned to each number. The probabilities of each possible pair of labels being assigned to XTheestimate of .5 as a child’s probability of correctly answering a problem o n which he or she assigned identical labels to the two numbers was chosen as a first approximation. in lieu of a specific model of the comparison process. The magnitude comparison model that was eventually hypothesized (described below) actually reaches a somewhat different estimate of the probability of a correct answer in such cases, but for reasons that could not be explained at this point. None of the data reported in this section is importantly affected by the differences between the two procedures for estimating the probabilities.

278

Rohwt S. SiegIer tind Mitchell Robinson

each pair of digits were then calculated to yield the values within the equation:

p (correct on each magnitude comparison problem)

p (correctly ordered labels) p (correctly ordered labels) + p (incorrectly ordered labels) An illustration of the workings of this equation may be helpful. If the number 5 was labeled little, medium, and big with probabilities of .2, .5, and . 3 , respectively, and if the number 7 was labeled little, medium, and big with probabilities o f . 1 , .5, and .4, respectively, then the probability of correctly ordered labels would be .38 [ ( . 5 x .2) + (.4 x .2) (.4 x . 5 ) ] , the probability of incorrectly ordered labels would be .23 [ ( . 5 x . l ) + ( . 3 x . I ) ( . 3 X 3 1 . and the probability of a correct magnitude comparison would therefore be .62 (.38/.61). The results of this equation were used to predict the number of errors that the 4-year-olds previously had made on the 28 magnitude comparison problems that did not involve the number 1 (problems involving 1 were excluded because the Fig. 4 clustering results led us to believe that 1 had a separate “smallest number” label that was not tapped by our possible labels of little, medium, and big). Using the above scoring procedure, the 4-year-olds’ probability of labeling each digit as small, medium, or large was found to account for 66%of the variance in their number of errors on the magnitude comparison problems. Thus, to summarize, the labeling experiment demonstrated that the labels children applied to numbers could logically have led to above-chance performance in the magnitude comparison context, that the labels were in fact correlated with individual differences in the percentage of correct answers on the comparison task, and that the labels predicted the relative difficulty of the magnitude comparison problems. A central question remained, however: “Was the children’s labeling of the numbers used in the process by which they compared magnitudes, or did performance on the one task simply predict performance on the other?” This issue was addressed in the next experiment.

+

+

D. EFFECTS OF TEACHING A LABELING STRATEGY

If children’s labels were functionally involved in the magnitude comparison process, then teaching them a specific set of labels for the numbers might be expected to influence their later numerical comparisons. In particular, children who used the labeling scheme that they were taught would be expected to be more successful on between-category comparisons (defined in terms of that labeling scheme) than on within-category ones. A control group that was not taught this particular labeling scheme would not be expected to show as large a between-category versus within-category difference, because they would not be as likely to use this particular set of labels. In addition, if children had previously

Numerical Undersrandings

219

selected labels independently for each number, without regard for the other number’s label, then instructing them in the use of an entire set of labels would be expected to reduce their percentage of incorrect answers (as will be explained in detail in the models below). The primary purpose of the training experiment was to test these predictions.

I . Method Twenty children of mean age 4 years, 10 months were recruited from a day care center serving a predominantly middle-class area. These children were randomly assigned to one of two groups: a label training group and a control group. Children were brought individually to a small vacant room in the day care center. Those in the training group were told: Today we’re going to learn about the numbers from 1 to 9. Some of the numbers are little numbers, some are medium numbers, and some are big numbers. I will tell you which numbers are little, which are medium, and which are big. One, two, and three are little numbers; four, five, and six are medium numbers; and seven, eight, and nine are big numbers. Now I’m going to say a number and you need to tell me if the number is a little number. a medium number, or a big number.

The experimenter then said “N: is N a little number, a medium number, or a big number?” If the child’s label matched the one the experimenter had provided, the experimenter said “Good, N is a (big) (medium) (little) number. ” If it did not match, the experimenter said “No, N is a ( b i g ) (medium) (little) number.” On each trial block, the digits were presented in a different randomly generated order. After the completion of each trial block, the experimenter repeated the initial information about which numbers were big, little, and medium. Then the next trial block was presented. The procedure continued until the child’s labeling of all 9 digits matched the experimenter’s on 2 consecutive trial blocks, or until 10 trial blocks had been given, whichever came first. Children in the control group were exposed to one of two procedures. Those in the labeling control were asked on five successive trial blocks to label each of the digits as little, big, or medium. These children’s procedure was identical to that received by children in the training group except that they were not instructed to use any particular labels before each trial block and were not given feedback about their performance. Children in the contact control were given no particular experience with labeling of numbers; they simply interacted with the experimenter for a period of time equivalent to children in the label training group. The next day, children were brought back to the experimental room and presented the 36 standard magnitude comparison problems. One day later, they were presented the same problems but in reverse order.

2. Results The data of children in the two control groups were indistinguishable; therefore, they were considered together in all subsequent analyses.

280

Robert S . Siegler and Mitchell Robinson

a . Learning of labels. Children in the training group did not require many trial blocks to learn the labels. The mean number of trial blocks before the block of last error was 2.2 (excluding one child who never met the mastery criterion and was replaced by another child). Among the 10 children in the training group, 9 did not make an error after the third trial block.

h. Magnitude comparisons. The training in labeling greatly reduced the number of errors that the children made. Children who were trained to apply the labels were correct on 96% of the magnitude comparisons, versus 79% correct answers among children in the control group. The labeling training also changed the distribution of children’s errors. Children in the trained group made 65% of their errors on the 9 items involving within-cluster comparisons. In contrast, children in the control group made only 29% of their errors on the same 9 items. E. THREE MODELS OF NUMERICAL MAGNITUDE COMPARISON

The three experiments described above provided converging evidence that numerical categories play an important role in preschoolers ’ numerical magnitude comparisons. Multidimensional scaling and hierarchical clustering analyses of preschoolers ’ untutored magnitude comparisons revealed clusters of numbers rather than a regular distribution. The children’s verbal labeling of numbers predicted both their level of skill in performing the magnitude comparisons and the relative difficulty that each magnitude comparison problem posed for them. Teaching the children an overall categorical organization reduced their number of errors on the numerical comparison task and changed their distribution of errors on it as well. Thus, it appeared that any model of preschoolers’ numerical magnitude comparisons should assign an important place to the children’s categorizations of numbers. Figure 5 depicts four such categorical models that would generate magnitude comparison performance of varying degrees of proficiency. First, let us consider Model I (Fig. 5A), which would produce the approximately chance level performance that we observed in most of the 3-year-olds. Although these children’s magnitude comparisons did not indicate any knowledge of numerical magnitudes, other data indicated that they did possess some relevant information. In particular, their classifications of numbers as little, medium, and big in the Fig. 5. Models of tnugnitude comparison. ( A ) Model I . [Note: Here and in Fig. SB, the probabilities for the numbers 2 -9 (ire the group level probabilities that children in the labeling experiment (3-yeur-olds in Fig. 5 A . 4-yeur-olds in Fig. 5B) assigned euch label to ench number. The Fig. 4 scaling und clustering results suggested that the number I was in u sepurute “smallest” category not rupped by the lubeling procedure; therefore, hypotheticul probabiliries huve been ussigned to rhe number I . ] ( B ) Model I I . See p p . 282 and 283 for F i g . 5C and D .

A

Representat ion

(Empirically derived probabilities of each number being assiqned to each category) Number -

Cateqory I

Smallest

6

Small Medium

2

2

2

3

4

5

6

7

8

9

25 58

42 42

08 92

25

58

00 67

17 50

67

67

17

17

17

25

33

33

25

08

Process For Labeling

For Magnitude Comparison

Choose Label Guess

Representa tion

B.

cotegory

Number

2

3

4

5

6

.2

72

61

.I

28

33

22 67

06 72

67 56

56 27 44

06

I1

22

33 44

67 56

I

Smallest Small Medium Biq

7

8

9

.7

Process For Labeling

Far Magnitude Comparison

For First Number

For Second Number

Greater Label Biaaer Number

Robert S . Siegler and Mitchell Robinson

282

Representat ion

C. Cotegoricol Orgoniration

Probability

Number 1

2

3

4

5

A

.s

~SmollsslI Small I Medium 1

B

.2

ISmallesl ISmalll Medium 1

C

.I

0 E

.I

pmalkst I Small (Smallerl I Small ISmallert I Small

.I

I

6

7

8

9

Biq

1

I.

Big Medium I Big Medium I 8ig 1 Medium 1 Biq

PROCESS: For Magnitude Comparison

For Labeling

Choose Categorical Organization

Retrieve Labels First and Second Numbers

Q abelr

Yes Greater Lobel Bigger Number

Fig, 5. (continued) (CJ Model Ill. (Note: The probabilities of each number being assigned to each label were chosen to correspond roughly to the labeling experiment probabilities that the S-year-olds assigned each label to each number. A n exact matching was impossible, because in the labeling experiment, the children were asked to assign lnbels to numbers independently, but wirhin this model, category boundaries are dependent upon each other.)

Numcricul Understundings

D

283

Representation

Cotsgorical Organizalion

Prob-

ability ~A

Numbsr 1

10

2

3

Small

4 I

5

Medium

6

7 I

8

9

Big

I

Catsgory Smallest

7

Small

2

72 61

Medium

I

28 33 67 7 2 67 56 27 44

819

22 06

06

I1

06

2 2 33 44 6 7 56

PROCESS

+

Greater L a b e l Biggsr Number

% Choose L a b e l For F i r s t

Number Choose L a b e l FOI Second Number

\Greater Lobel d Bigger Number

Fig. 5 . (continued). (D)Model 11-111.

284

Robert S . Siegler and Miichell Robinson

labeling experiment corresponded reasonably well to the relative magnitudes of the numbers. The representation of the 3-year-olds’ knowledge (Fig. 5A) is based directly on their performance in the labeling experiment. Viewed vertically, this matrix indicates the probability that each number will be assigned each categorical label (in the context of the numbers 1-9). Viewed horizontally, the matrix indicates each category’s range and the probability that each category will include each number. When asked whether each number is small, medium, or large, children simply retrieve categorical labels with the probabilities indicated in the representation. When they are asked to compare magnitudes, however, children do not utilize any of the knowledge from the representation; they simply guess, thus producing chance-level magnitude comparison performance. Next consider Model 11, hypothesized to underlie the moderately expert performance (60-95% correct) of the majority of 4-year-olds (Fig. 5B). Model I1 incorporates one major change from Model I: the Model I1 magnitude comparison process makes use of the categorical information. As shown in Fig. 5B, when children are presented a magnitude comparison problem, they generate a label for each number. The choices of labels are independent for the two numbers, and occur for each number with the probabilities shown by looking down the columns of the representation. If the labels differ, children choose as bigger the number associated with the larger label. If the labels are identical, children regenerate labels for each number until the labels discriminate between them. Note that, regardless of the particular probabilities assigned to the labels, this type of model almost inevitably leads to errors. Unless the child creates at least ( N - 1) distinct categories for the N numbers in the comparison set, he or she will always have some probability of assigning the larger label to the smaller number. Model 111, intended to characterize the near-perfect performance of most of the 5-year-olds, is quite similar to Banks, Fujii, and Kayra-Stuart’s (1976) model of adult performance. It also represents a logical extension of the Fig. 5B model of moderately skilled performance. The only large change from that model is that the choice of labels on the magnitude comparison task is not at the level of the individual numbers, but rather at the level of the overall categorical organization. As shown in Fig. 5C, the child’s representation includes several alternative divisions of the 9 numbers into categories. The process indicates that each time a magnitude comparison problem is presented, the child chooses one of the categorical organizations with the probability shown in the representation. Once a particular organization is chosen, the assignment of labels to numbers is determined. If the labels differ, the child chooses the number attached to the larger label as being bigger. If the labels are identical, the child chooses a new categorical organization. The process continues until a categorical organization is chosen that assigns different labels to the two numbers being compared.

Nurnc~rieczlUntirrstandings

285

Finally, Model 11-111, a hybrid of Models I1 and 111, is hypothesized to produce the performance of the trained children in the training experiment. This model is diagramed in Fig. 5D.The representation includes the Fig. 5B model probabilities of each number being assigned each label, and also the one categorical organization that the children were taught in the training procedure. The process starts with the Model 111 approach in which a categorical organization is selected. Because the only categorical organization that the children know is the one they were taught, they use it to assign labels to the numbers. If this procedure assigns different labels to the numbers, the children answer correctly. If it assigns the same labels, however, they would not know any alternative organizations, and therefore would revert to the Model I1 process of assigning labels to each number independently until the labels differed. These models give rise to the types of magnitude comparison performance, labeling, and learning that were observed among each age group in each of the three above-described experiments. Model I prescribes the 3-year-olds’ behavior in a rather direct way. As implied by the probabilities given in Fig. 5A, these children can assign labels to the numbers that have some correspondence to their magnitudes. However, they do not use this knowledge in comparing the magnitudes; instead they simply guess, producing on all 36 problems the roughly chance performance that was observed. Model I1 predicts a considerably lower error rate and a distribution of errors that is linked to the characteristics of the problems. Averaged across all 36 problems, the model generates 84% rather than 50% correct answers. The 84% correct figure is quite close to the 4-year-olds’ observed percentage correct, 8 1 %. The symbolic distance effect emerges because numbers that are farther apart are less likely to be assigned incorrectly ordered labels. The min effect emerges for much the same reason; the distribution of labels changes more rapidly between numbers at the small end of the scale than between numbers at the large end. The labeling experiment results of accurate labeling of numbers as small, medium, and large, would arise from children choosing labels for the numbers in accord with the probabilities given in Fig. 5B. Model 111 predicts that no errors should occur, because within each categorical organization, the category boundaries are correctly ordered and nonoverlapping. The effects of categorization will continue to be seen in solution time patterns, however. Between-category comparisons will on average arise earlier in the comparison process when the digits are farther apart and when the magnitude of the minimum number is small. Sekuler and Mierkiewicz’s (1977) data on 6-year-olds and older children are in accord with these predictions. Finally, Model 11-111, the model of what the trained children learned, would produce performance similar to Model 111 on the between-category comparisons and performance similar to Model I1 on the within-category ones. Overall, it generates 91 % correct performance, an accuracy level quite close to the 96% that

286

Robert S. Siegler and Mitchell Robinson

was observed among the trained children. The superior performance predicted by this model on, and only on, the between-category problems explains both why trained children made fewer errors overall than untrained ones, and why the bulk of the few errors they made were within-category errors. In addition to accounting for our data and providing a plausible outline of the development of numerical comparison skills, these models raise two more general questions. First, is the categorization approach generally used in comparing magnitudes across different age groups and across different types of materials, or is it limited to young children’s numerical magnitude comparisons? Second, what are the developmental implications of the parallels, observed here and in many previous studies, between early error patterns and later solution times? The first question concerns the generality of the models across age groups and across types of stimulus materials. Several recent experiments have demonstrated categorical effects in populations other than preschoolers and in domains other than numbers. Kosslyn, Murphy, Bemesderfer, and Feinstein ( 1977) obtained categorical effects with adults for highly overlearned, artificially imposed categories in the context of length judgments. Pliske and Smith (1979) found that adults spontaneously used gender as a basis of categorization in making distance judgments in a situation in which gender divided the lengths into nonoverlapping categories. Maki (1981) provided a similar demonstration of adults using state boundaries as a basis for categorizing cities as being farther east or west. Thus, the division of stimuli into categories for purposes of comparing magnitudes is far from unique to preschoolers as a population or to numbers as a type of stimulus material. It does seem likely, though, that children (and probably adults) who are in the process of learning new dimensional orderings will be especially likely to categorize stimuli as a means of reducing imposing learning tasks to manageable dimensions (as in the adage, “Divide and conquer”). Under such circumstances, differing degrees of information about the individual stimuli might be expected to dictate the form of the categories. Stimuli associated with numerous referents (e.g., the small numbers in the experiment presented here) will be placed in single member or small categories, but stimuli about which little is known (e.g., the larger numbers here) will be lumped into multimember units. The second question concerns the parallels that arose between young children’s error patterns and the previously reported solution times of older children and adults on the same numerical comparison task. This parallel is far from unique to the research presented here; data on such diverse cognitive skills as analogical reasoning, transitive inference, quantification, and attention have shown similar parallels (Chi & Klahr, 1975; Manis, Keating, & Morrison, 1980; Sternberg, 1977; Trabasso, Riley, & Wilson, 1975). The present Models I1 and 111 account for the similarities in the numerical comparison context and by analogy suggest why they may appear in others. Examination of Fig. 5B shows that the representation of the problem dictates the distribution of early errors. With learn-

Numerical Ondersfundings

287

ing and development, increasingly task-appropriate processes are adopted and greatly reduce or eliminate errors (Fig. 5C). However, the representation that these processes operate upon remains relatively unchanged in the sense that the same numbers usually receive the same labels. This leads to relatively long solution times on those problems that earlier produced errors. We do not know at present whether representations of stimuli generally remain more stable over development than the processes that operate upon them; Newel1 ( 1972) pointed out difficulties in even posing the question. Nonetheless, relatively stable representations and relatively rapidly changing processes provide at least one explanation for the parallels between early errors and later solution times.

V. Preschoolers’ Knowledge of Addition A.

EXISTING RESEARCH ON CHILDREN

Young children’s knowledge of addition has been the subject of far more research than their skill in counting or in comparing numbers. No doubt, this large body of work is due in large part to the traditional centrality of addition in the elementary school curriculum and to its importance in everyday life. Two primary questions have emerged within the research: what addition tasks can young children perform, and what processes do they use to perform them? Whut Young Children Know ubout Addition Much early work on addition was devoted to determining which problems young children find easy to solve and which they find difficult. One of the most elaborately reported examples of such a study was that of Knight and Behrens (1928). These researchers presented second graders the 100 addition problems formed by the factorial combination of augend (0-9) plus addend (0-9). They reported that the greater the sum, the more difficult the problem, and also that ties were easier than would have been expected from considering their sums alone. When we reanalyzed the rankings of Knight and Behrens using regression analyses, we found that the size of the smaller number was also predictive of problem difficulty. It accounted for exactly as much of the variance as the sum when all problems were considered (62%) and more of the variance when only the nontie problems were included (81% vs 67%). More recently, interest has shifted to very young children’s knowledge of the principles underlying addition. Smedslund (1966) demonstrated that by 5 or 6 years of age, children possess the most basic knowledge about addition: that adding increases quantity. Subsequent investigations (e.g., Gelman, 1972; Gelman & Starkey, 198 1 ;Siegler, 1981 a) indicated that children as young as 2 and 3 years possess similar knowledge, at least when the augend and addend sizes are 1.

288

Robert S. Sirgler and Mitchell Robinson

small. Some understanding of the principles of inversion and compensation as they apply to addition and subtraction also seems to be present even before children enter school (Cooper, Starkey, Blevins, Goth, & Leitner, 1978; Gelman & Starkey, 1981).

2 . How Children Solve Addition Problems Another issue that has long been of interest is how young children solve addition problems. For many years, lack of revealing methodologies limited insights into these processes. However, the development of chronometric methods proved to be an important breakthrough. Perhaps the best-known model of the addition process, the min model of Groen and Parkman (1972), is based upon a chronometric approach. Groen and Parkman started with the question of whether material as overlearned as addition facts was retrieved from memory directly or whether answers were reconstructed each time problems were presented. To address this issue, they examined first graders’ patterns of solution times for all integer addition problems with sums s 9 , and aiso adults’ patterns of solution times for all integer addition problems with augends and addends 9. In each case, they found that solution times were directly proportional to the size of the minimum number. On the basis of this evidence they formulated the min model. Within this model, the adder chooses the larger of the two numbers and then increments it by one a number of times equal to the smaller number. The amount of time required to choose the larger number is assumed to be constant for all problems, as is the time per increment. Therefore, the only factor contributing to differences among problems in solution times is the number of increments dictated by the minimum number. The min model fit Groen and Parkman’s (1972) data quite well except on ties, where both children and adults were much faster than would have been predicted. Also unexpectedly, the solution times on ties were nearly constant over all of the items that were tested. These data led Groen and Parkman to amend their model so that ties were reproduced directly from “fast-access memory,” while other problems were reconstructed by the incrementing process. Subsequent experiments have provided a mixed record of support and nonsupport for Groen and Parkman’s model as it applies to children. On the positive side, other investigators have replicated the findings that the size of the minimum number is the best predictor of young children’s solution times and that ties are solved uniformly faster than might be expected from the sizes of their minima (Svenson, 1975; Svenson & Broquist, 1975). Verbal explanations have provided converging evidence; when Svenson ( 1975) asked children how they solved specific problems, they often explained that they chose the larger number and counted upward from it (Svenson, 1975). Even preschoolers who were taught addition by a different method often later produced solution times proportional to the minimum number (Groen & Resnick, 1977).

Numericul Undersrudings

289

Not all of the evidence with children has been consistent with the min model, however. The size of the last number seems to play a role above and beyond that of the size of the minimum number (see Groen & Parkman, 1972, Fig. 2; Svenson, 1975). The same children whose verbal statements indicated awareness of the min strategy also indicated reliance on a variety of more specific pieces of knowledge; they said they used ties and the number 10 as reference points from which to solve other problems (Svenson, 1975). Finally, close scrutiny of Groen and Parkman’s (1972) data indicates that their model fit much better at low values of minima than at higher values; the correlation between the size of the minimum and the mean solution time was r = .82 for minimum numbers of sizes 0-2, while it was only r = -11 for minimum numbers of sizes 2-4 (our reanalysis of all cases; if ties are excluded, the correlations are r = .87 and r = .48, respectively). These findings suggest that attributes of particular problems, other than their minimum numbers, influence children’s addition. Because the subsequent data have been only partially favorable to Groen and Parkman’s model, several investigators have proposed alternative models. Svenson (1975) proposed a model basically similar to that of Groen and Parkman except that children would take time to reorder the numbers if the larger number were second; this reordering operation accounted for the effects exercised by the second number independent of those of the minimum number. The model of Ashcraft and Battaglia (1978) represented a more radical departure. It accounted for differences among the solution times of various problems in terms of search and retrieval processes rather than in terms of any reconstructive process. All of these models of how addition is done are based primarily upon chronometric data; another source of evidence about the addition process comes from clinical descriptions. A study of by Ilg and Ames ( 1 95 1 ) is perhaps the most comprehensive of these clinical studies. On the basis of observations of very large numbers of beginning adders, Ilg and Ames hypothesized a four-stage sequence for the development of addition skills. First, children were said to perform all addition problems by counting on from 1 . Slightly later, they were said to know a few addition facts “by heart” and to count on from the smaller number to compute the others. Still later, they were said to count on from the larger number (as in the min model) to solve those problems they had not memorized. Finally, they were said to have memorized many problems and to use a variety of specific strategies to solve the others. For example, they might break up 14 3 into 4 + 3 = 7; 7 10 = 17. This last observation of specific strategy use has been echoed in numerous other clinical descriptions of young children’s addition. Several approaches have been noted: children have been said to rely on 5s, lOs, and ties as reference points from which to calculate answers, to move their feet rhythmically to help them count, to count on from the larger number, and to count on from 1 (Fuson & Richards, 1982; Hebbeler, 1976; Yoshimura, 1974). Although these descriptions of addition strategies have been largely anecdotal, the reports have been persis-

+

+

290

Robert S . Siegler and Mitchell Robinson

tent enough to leave little doubt that young children employ a variety of methods in adding. Considered as a group, these studies reveal a cleavage between the types of models that have been formulated and many of the phenomena that have been observed. On the one hand, the models of addition-Groen and Parkman (1972), Ashcraft and Battaglia (1978), and Svenson (1975)-have all been designed to depict the strategy people use to solve addition problems. On the other hand, the detailed observations of addition have revealed a large number of distinct strategies that are used on particular problems. The discrepancy suggests that the chronometric data on children’s addition may reflect an averaging over different strategies rather than a consistent adherence to any one strategy; this would call into question any single model that purported to be the way in which children solve all addition problems (cf. Estes, 1956; Newell, 1972). The existence of numerous distinct addition procedures would also suggest that, above and beyond determining how children execute any particular addition strategy, we would also need to learn how they choose among alternative approaches. B . PRESCHOOLERS’ ADDITION STRATEGIES

Considering the possibility that children use different addition strategies on different problems raises a large number of questions that have not been addressed previously. Which strategies are used most frequently? What are their accuracy and temporal characteristics? Are choices of strategies systematically related to the particular numbers involved in each problem? Does variable strategy use help children add more accurately and/or more quickly than using the same strategy at all times? If so, why? Videotaping children in the process of adding seemed the ideal way to learn more about their strategies and thus to address these questions. Therefore, this was the approach taken in the next experiment. 1 . Method

The children who participated in this experiment were the same 3-, 4-, and 5-year-olds who participated in the counting and magnitude comparison studies. Each child was brought individually to a videotaping laboratory within the preschool and seated at a table. The room contained a camera mounted in a comer; aside from the camera, all of the videotaping equipment was kept in an adjacent room. Children were given the following instructions: Today we’re going to play another number game. I want you to imagine that you have a pile of oranges. 1’11 give you more oranges to add to your pile; then you need to tell me how many oranges you have altogether. Okay? You have rn oranges, and I’m going to give you n to add to your pile. How many do you have altogether?

Numerical Understandings

29 1

Children were presented the 25 problems formed by the factorial combinations of augend ( 1 -5) and addend ( 1 -5). Each problem was presented twice. The 25 problems were arbitrarily divided into groups of 9, 8 , and 8, with each group being presented to the child in a different session. When all 25 problems had been presented, the cycle was repeated; thus, each child received 50 items spread over six sessions. Each session lasted approximately 15 minutes. No feedback was given at any time, except for periodic assurances that the child was doing well.

2 . Results The analyses of addition performance that will be reported differ in two ways from the analyses of counting and magnitude comparison. First, except for preliminary, aggregate level analyses, the data of the 4- and 5-year-olds are analyzed together because children in the two age groups were quite similar in their absolute level of performance, the predictors of their performance, and the strategies that they used. Second, again except for the preliminary analyses, only cursory descriptions of the 3-year-olds ’ performance will be reported, and no model will be postulated. The absolute level of performance among children in this age group was very low, none of the variables that were examined predicted the relative difficulty of the problems for them, and visible strategies were rarely observed. Thus, we simply do not possess rich enough data to infer much about the 3-year-olds’ knowledge of addition, beyond the facts that they quite often answered correctly 1 + 1, 2 + 1, and 1 2 and that they rarely used visible strategies to do so.

+

u . Overall level ofperformunce. As in the previous experiments on counting and magnitude comparison, the period from 3 to 5 years proved to be one of substantial development. Three-year-olds were correct on 20% of the addition problems, 4-year-olds were correct on 66%, and 5-year-olds were correct on 79%. b. Predictors of error patterns. Stepwise regression analyses were used to examine the distribution of errors on the 25 problems. The initial analysis included five predictor variables: size of the augend, size of the addend, size of the minimum number, size of the sum, and size of the square of the sum. None of these predictors was significantly correlated with the 3-year-olds’ error patterns, each of them failing to account for even 15% of the variance. In contrast, the 4- and 5-year-olds’ error patterns were much more orderly. The sum was found to be the best single predictor, accounting for 49% of the variance. Careful examination of the 4- and 5-year-olds ' data suggested that three tendencies that were not captured in the above list of predictors also accounted for substantial amounts of variance. First, problems involving the number 5 as either augend or addend were much easier than would have been expected from their

Roherr S . SiegIer rrnd Mirchell Robinson

292

45 I 40 0

t

bl-

f?

g

35 30

c 0 a3

25

B 20

0,

6

I

30 25

f 20

e

g2

45 40 35

15

!L

2

10 5

10 51

1

1 2 3 4 5

1 2 3 4 5

Augend F i g . 6 . Percenrage of errors

15

Addend on

arithrneric problems as u funcrion of augend and addend size

absolute magnitudes. As shown in Fig. 6, problem difficulty increased monotonically with augend and addend sizes for the range 1-4, but this trend was broken at the number 5 . Second, as Svenson ( 1975) found with third graders, the size of the last number exercised an effect independent of the sum. Third, as Groen and Parkman (1972) and Knight and Behrens (1928) found with first and second graders, ties were easier than would have been anticipated. A new regression analysis indicated that these four variables accounted for 86% of the variance in the distribution of 4- and 5-year-olds’ errors: sum accounted for 49%; sum and the presence of a five accounted for 74%; sum, fives, and the size of the last number accounted for 80%;and sum, fives, last, and ties accounted for 86%.

c . Consistency of performance. Consistency of addition performance was similar to consistency of magnitude comparison performance. Children’s total number of correct answers showed substantial consistency; over the two occasions, the correlations were r = .90 for the 5-year-olds, r = .71 for the 4year-olds, and r = .77 for the two groups combined. Performance on individual problems, however, showed little consistency; 5-year-olds advanced the same incorrect answer on 3% of trials, and 4-year-olds did so on 4%. As with the magnitude comparisons, these data seemed to demand that in any model of addition performance, errors would be at least in part the products of probabilistic and/or reconstructive processes rather than solely of determinate, reproductive ones.

d . Strutegy use. Scrutiny of the videotapes revealed that children adopted at least four different approaches to solving addition problems: the counting fingers strategy, the fingers strategy, the counting strategy, and (for want of a better name) the no visible strategy approach. These strategies could be identified very reliably; two independent raters agreed on 98 of 100 ratings. The four

Nummicol Understundings

293

strategies that were identified differed in their visible and audible manifestations, in their accuracy, in their temporal characteristics, and in the types of errors with which they were associated. In the counting fingers strategy, children put up the fingers on one hand, then put up the fingers on the other hand, and then counted the two sets of fingers. The counts generally began with the leftmost finger on the left hand, the hand that was usually put up first, and ended on the rightmost finger on the right hand, the hand that was usually put up second. As shown in Table IV, the strategy was used moderately often, was very slow, and produced accurate performance. When errors occurred, they were most often close to the correct solution. In the fingers strategy, children put up their fingers as in the counting fingers strategy but showed no evidence of counting them. The fingers and the counting fingers strategies produced similarly accurate performance, but the time needed to execute the fingers approach was much shorter. This tendency was very consistent over individuals; 20 of the 21 children who used both strategies at least twice had faster mean solution times on the fingers than on the counting fingers trials. Errors on the fingers trials, like those on the counting fingers trials, were usually close misses (Table IV). In the counting strategy, children counted aloud but without any visible referent. In all cases these counts started from 1 . The strategy was associated with moderately long solution times and was the least accurate and the least often used of the four strategies. Errors were often quite distant from the correct answer; fewer than half were within one of it. The fourth category was a catchall for those trials on which the children did not engage in any visible or audible behaviors that seemed related to the addition process; these trials were grouped together as “no visible strategy.” This approach was the most frequently used, the second least accurate, and the most rapidly executed of the four strategies. The accuracy data fell into a distinctly bimodal distribution. Sixteen of the 30 children performed at high accuracy rates on no visible strategy trials; they were correct on 86-100% of trials. The other 14

TABLE IV Characteristics of Arithmetic Strategies

Strategy Counting fingers Fingers Counting No visible strategy

Trials on which strategy used (%)

IS 13

Mean solution Correct time (sec) answers (%) 14.0 6.6

87

89

8

9 .0

54

64

4.0

66

Errors on which answer was within one of correct sum (%) 70 80 44 41

294

Robert S. Siegler and MircheN Robinson

children were much less accurate, ranging from 13 to 67% correct. This bimodality was not an artifact of a small number of observations producing high variability, as the approach was used on an average of 32 trials per child. Both the high-accuracy and the low-accuracy children executed the approach rapidly, and both tended to make errors that were far from the correct answer. e . Strategy use and problem difliirulty. The most striking finding of the experiment involved the connection between strategy use and problem difficulty. As shown in Table V , the percentage of errors that each problem elicited was very closely related to the percentage of trials for that problem on which one of the three visible strategies was used ( r = .91). The relationship was actually attentuated slightly by the tendency of children more frequently to solve problems on which they used a visible strategy; if only the percentage of errors on no visible strategy trials is used to estimate difficulty, the correlation between frequency of visible strategy use and problem difficulty increases slightly to r = .92. Using the visible strategies more often on the more difficult problems proved to be highly adaptive. Performance was more accurate on 24 of the 25 problems when visible strategies were used than when they were not. Moreover, the frequency of use of visible strategies on the 25 problems correlated r = .73 with the gain in accuracy from using a visible strategy (the difference on each problem between error rates when children did and did not use a visible strategy). These results answered two of our original questions. First, use of visible strategies was indeed related to the characteristics of the addition problems, with the more difficult problems elicting a greater frequency of strategy use. Second, employing such visible strategies aided children’s performance, with the greatest benefits accruing on the most difficult problems. Left unanswered, however, was the question of how children chose among the available strategies so as to produce these relationships. Two types of decision processes seemed the most likely. One possibility was that children were consciously aware of problem difficulty and used their assessment of it as the decision criterion, adopting a visible strategy when they judged the problem too difficult to solve without one. This might be labeled the metacognitive interpretation. The second possibility was that the decision to use a visible strategy arose as a by-product of other solution processes rather than through any process involving explicit judgments of difficulty. This might be labeled the by-product of solution processes interpretation. The metacognitive hypothesis is based on the following model of the strategy choice process:

problem difficulty + judgments of problem difficulty --j use of visible strategies

Numerical Understandings

295

TABLE V Erron and Strategy Use on Arithmetic Problems ~~

Problem

Errors (%) ~~

1+1

1+2 2+ I 2+2 5+ I 3+l, 5 2 4 + l , 2+3

+

1 +5

3+2 l+4,5+5,3+3 4+2, 5 t 3 1+ 3 2+4 2+5 3+5,4+4 4+5,5+4 3+4 4+3

3 7 10 12 14 19 20 25 21 31 32 34 36

31 41 41 53 59

Trials on which visible strategy used (%) ~

10 15 22 18 25 24, 32 24, 39 34 39 32. 34. 34 37, 39 31 47 41 56, 34 49.46 54

58

This model implies that the actual difficulty of problems and children’s judgments of their difficulty should be highly correlated (especially given the observed high correlation between problem difficulty and visible strategy use). To test this prediction, we asked a group of 12 5-year-olds, students at a nursery school very similar to the one at which the original experiment had been run, to label each of the 25 problems as easy, hard, or in between. “Hard” ratings were quantified as 2, “easy” ratings as 0, and “in-between” ratings as 1. These ratings were correlated with problem difficulty as estimated from three error and solution time data sets: ours, Knight and Behrens’ (1928), and Groen and Parkman’s (1972). Regression analyses of the difficulty ratings were also run in order to determine whether the same variables predicted the ratings as predicted actual difficulty. The difficulty ratings correlated only moderately with the actual difficulty of the problems as estimated by the errors of our sample ( r = .47),the errors of the Knight and Behrens sample ( r = .50), and the solution times of the Groen and Parkman sample ( r = .31). It might be argued that these relatively low correlations could have resulted from our obtaining the addition performance and the

296

Roherr S. Siegler und Mitchell Robinson

difficulty ratings from different samples of children. However, the correlations were much higher on other intersample comparisons: between the percentages of errors on the 25 problems in our samples and those of Knight and Behrens ( r = .77), between the percentage of errors on these problems in our sample and the solution times in the Groen and Parkman sample ( r = .72), and between the percentage of errors in the Knight and Behrens sample and the solution times in the Groen and Parkman sample ( r = .89). The predictors of the difficulty ratings also differed from the predictors of the errors and solution times; the size of the larger number was the best predictor of the ratings, but, as reported above, the sizes of the minimum number and of the sum were the best predictors of actual difficulty. Thus, the data lent little support to the view that the children’s judgments of problem difficulty were responsible for the relationship between strategy use and problem difficulty. C . A MODEL OF STRATEGY CHOICE IN ADDITION

The second possibility was that the correlation between visible strategy use and problem difficulty was a byproduct of the solution process, rather than resulting from any knowledge of problem difficulty per se. One model that would give rise to such a correlation is displayed in Fig. 7. The representation, shown on the top of Fig. 7, includes the probabilities that the answer to each addition problem can be recalled correctly without recourse to any external strategy and with a confidence that exceeds some arbitrarily set criterion level. These probabilities will differ for each child and for each situation in which a given child finds himself or herself. The ones shown in Fig. 7 were calculated for the group level probabilities within the experiment presented here that on each problem, children used no visible strategy and generated the correct answer. The process that is applied to this representation, shown on the bottom of Fig. 7 , begins with the setting of a confidence criterion by which children decide whether they are sure enough of an answer to give it. Then they try to recall the answer to the p r ~ b l e m If . ~ their confidence in the recalled answer exceeds the 9“Recall” is used here in quite a loose sense. It is entirely possible that on some of the trials where children did not use a visible strategy and answered correctly, they used the min approach described by Groen and Parkman or some other strategy. The evidence was equivocal as to which particular approach they employed. Regression analyses of all trials on which the children used no visible strategy indicated that the sum of the numbers was the best predictor of errors, accounting for 47% of the variance. The size of the minimum number was the next best predictor, accounting for 41% of the variance. When only nontie cases were considered, the size of the minimum number was the best predictor, accounting for 66% of the variance; the sum of the two numbers was the next best, accounting for 52%. Regression analyses of solution times revealed a similar picture. When all correct trials were considered, sum was the best predictor, accounting for 33% of the variance; the minimum number accounted for 23%. However, when just nontie cases were considered, the size of the minimum number accounted for 62% of the variance in solution times, while the size of the sum

Representot ion Sums and (Probability of Recolling Each Sum with Confidence Beyond Criterion) Addend Augend

I

2

3

4

5

I

2 3 4

5

PROCESS:

Set Confidence

Stote Answer -

In Answer>

State Answer

Count Objects In Representotion State Lost Number As Answer

-

Fig. 7 . Model oj'strutegy choice in uddition. (Note: The probabilities in parentheses represent the percentage of trials on which children used no visible strategy and answered correctly. They can be thought of as associative strengths.)

298

Robert S. Siegler and Mitchell Robinson

criterion, they give it as the answer. Otherwise, they augment their representations of the numbers indicated by augend and addend, either externally, by putting up their fingers, or internally, by forming some type of imaginal representation. If their confidence in an answer at this point exceeds the criterion, they give it; otherwise, they count their fingers or the imaged objects and state the last number of the count as their result. This model gives rise to the four approaches that we observed. If children answer at the first “recall” point, they will not have used any visible straiegy. If they put up their fingers but answer without counting them, they will have used the finger strategy. If they image objects corresponding to augend and addend and then count aloud, they would be classified as having used the counting strategy. Finally, if they put up their fingers, count them, and answer after counting, they would be classified as having used the counting fingers strategy. The model also suggests explanations for the relative solution times and accuracy rates of the four strategies. It predicts straightforwardly that the no visible strategy approach should be the fastest, the fingers approach the next fastest, and the counting and counting fingers approaches the slowest of the four strategies; all of the processing steps necessary to execute each of the faster strategies are included within the steps necessary to execute the slower ones. Predictions of relative accuracy also can be derived, albeit not quite as directly. Both the overall low accuracy and the bimodal distribution of accuracies of the no visible strategy approach follow from the view that some children used this approach because they set loose criteria for deciding when they knew the answer, and others used it because they did not need any external aids to retrieve the correct answer. The high accuracy of the counting fingers strategy would have been expected since 4and 5-year-olds have been shown to be very adept at counting the 2-9 objects required by the problems (Gelman & Gallistel, 1978). Several considerations may have contributed to the high accuracy of the fingers approach: pattern recognition of the number of fingers that were put up, kinesthetic cues associated with putting up particular sets of fingers, and longer search time than that typical of the no visible strategy approach.’O Finally, the relative inaccuracy of the accounted for 50%. Another possibly relevant source of data was that on the counting and counting fingers strategy trials, on which the 4- and 5-year-olds were heard counting aloud, their counts always started from one. Thus, it was difficult to tell on the no visible strategy trials whether children generally used the min approach, whether they generally used the sum approach (like the min model but counts start at one), whether some children used one and some the other, or whether the same children sometimes used one and sometimes the other. ‘“Tounderstand why kinesthetic cues associated with raising one’s fingers might be helpful for recalling a sum, consider what would happen if your eyes were closed and someone lifted one or more fingers on each of your hands. With at least some combinations, the number of fingers raised might well “feel” like 2,4, or 10 (for example). Whether children are helped by such information is an open question, but the kinesthetic cues offered at least one explanation for the superior accuracy of the fingers approach.

Numerical Ur~ciersrcirulings

299

counting approach was predicted by Kosslyn’s (1978) finding that children have difficulty maintaining a clear image for as long as it took to execute this strategy, 9 seconds. Perhaps the most important feature of the model is that it allows us to account for the correlation between problem difficulty and strategy use. No explicit knowledge about problem difficulty is required to produce the relationship. Instead, at each step in the solution process, the child considers whether his or her confidence in an answer exceeds the level demanded by his or her criterion. The more difficult the problem, the less likely it is that this will occur. Thus, at least those children whose criteria are relatively high are led to take increasingly effortful steps to solve the more difficult problems. In a sense, these children use internalized strategies when they can and externalized ones when they must. Such flexible strategy selection has obvious advantages. It minimizes effort while maximizing the probability of a correct answer. By simply adjusting the confidence criterion, children can adapt to situations in which accuracy is the critical consideration or to situations in which speed or lack of effort is. Adults have been shown to possess similar propensities to avoid cognitive effort and to adjust strategies to the difficulty of particular problems. Siegler and Atlas (reported in Siegler & Klahr, in press) found that adults computed quantitative solutions to balance scale problems only when the problems could not be solved by simpler qualitative comparisons. Glushko and Cooper (1978) found that even in simple sentence verification situations, adults varied their approach depending upon the task demands. These and the findings presented here suggest that from early in childhood, two systemic principles may govern the construction of information-processing routines: minimize the effort needed to accomplish any particular goal, and maximize each routine’s flexibility to adjust to different task environments. These principles would have obvious adaptive value and would be in keeping with the flexible strategy use that has been so frequently observed. The implication for future research is that examining the ways in which people choose among alternative strategies for solving problems may be at least as informative as focusing on how they execute any given strategy.

VI.

Conclusions: The Development of Numerical Knowledge

At the outset of this article, we proposed to examine several aspects of young children’s knowledge of numbers, to devise models of their knowledge within each task domain, and eventually to formulate one or more comprehensive models, including the information within each of the specific ones. This last goal, the formulation of models that stretch across task domains, has been given

300

Rohcrt S. Siegler and Mitchell Robinson

considerable homage in the abstract by developmental psychologists, but few such accounts have been stated at a sufficiently precise level to be meaningfully evaluated. We believe that the formulation of detailed but encompassing models is crucial to understanding cognitive growth. Therefore, in this last section, we shall focus on the issue of what type of larger system might produce the numerical skills of preschoolers that we observed in each of the particular areas. Newel1 (1973), in his article, “You Can’t Play 20 Questions With Nature and Win, ” eloquently argued the case for building large-scale integrative models. He contended that although cognitive psychologists have succeeded in identifying robust phenomena and in accounting for performance in particular situations, the research has failed to cumulate. Among Newell’s suggested means of escape from this dilemma was devising a single model capable of producing performance on many tasks. This suggestion has been influential in motivating large-scale computer simulations of thought, language, imagery, and memory (Anderson, 1976; Kintsch & van Dijk, 1978; Kosslyn, 1978; LNR, 1975). The models that we shall present of young children’s knowledge of numbers differ from these others in being much less ambitious in scope, in not yet being specified at the level of running computer simulations (programs in OPS5 are currently being written), and in being primarily concerned with development. However, the motivation for building them was the same. Figure 8 outlines our current understanding of preschoolers’ knowledge of numbers. The three models within Fig. 8 are ordered from the least to the most advanced, and correspond to the knowledge that we hypothesize is most often possessed by 3-, 4-, and 5-year-olds, respectively.” A cursory examination of the models reveals two features: they are quite forbidding looking, and they appear rather similar to each other. Because of the models’ forbidding appearance, we will describe one of them, Model 11, at some length. Because of the similarities among the three models, we will characterize the depictions of the least and the “The models in Fig. 8 are intended to provide descriptions of the modal tendencies among 3-, 4-, and 5-year-olds, but should not be taken to imply a lockstep progression among the three skills or a perfect correlation between age and skill. Empirically, each skill possessed a moderately high correlation with age and also a moderately high correlation with the other two skills. Age correlated r = .64 with percentage of correct answers on the numerical comparison problems, r = .68 with percentage of correct answers on the addition problems, and r = .49 with how high children counted. Across the three age groups, the percentage of correct answers on the addition task correlated r = .80 with the percentage of correct answers on the numerical comparison task, the percentage of correct answers on the comparison task correlated I’ = .60 with the highest number counted on the counting task, and the percentage of correct answers on the addition task correlated r = .65 with the highest number counted on the counting task. The correlations between pairs of tasks were also fairly high within age groups, averaging r = .54 for the 9 within-age correlations. Overall, 27 of the 39 children (70%) who performed all three tasks would have been assigned to the same aggregate model by virtue of their performance on each of the three individual tasks.

most advanced knowledge in terms of differences between them and the formulation of intermediate level knowledge. When we examine Model 11, the model of 4-year-olds’ knowledge, a basic hierarchical form becomes evident. Numbers as a class are at the top of the hierarchy, then categories of numbers (e.g., small numbers), and finally individual numbers (e.g., 6). There are connections both across and within levels of the hierarchy. At the top, numbers as a class can be operated upon by a number of processes: they can be counted, their magnitudes can be compared, and they can be added and (presumably) subtracted. That children treat numbers as a class distinct from other classes was evident in what they did not do as much as in what they did. No child ever gave a nonnumeric answer to an addition problem or used any nonnumeric term in counting. Children at times did use nonstandard numbers in their counting strings, but the nonstandard numbers were always combinations of standard ones. Therefcre, we believe that several processes that can operate upon numbers are attached to numbers as a class rather than to particular groups of numbers or to individual numbers. (The details of these processes are omitted from the diagrams in Fig. 8 only because of considerations of space; they are shown earlier in the article, in the figures indicated.) At the next lower level of the hierarchy are categories of numbers, ordered in terms of magnitudes. Both the number conservation data reported by Siegler (1981a) and tne magnitude comparison data reported in the present investigation suggest that these categories possess psychological reality for young children. Illiistratively , the conservation operators that are applied to small numbers differ from those that are applied to large ones; as shown in Fig. 1 B, children apply the correct transformational rules to small numbers of objects, but with larger groups judge the longer row to have more. The numerical categories occupy an intermediate position within the hierarchy. Each category is linked both upward to the class of numbers and downward to individual numbers. The particular probabilities linking the categories to the individual numbers are based on those that appeared in the Fig. 5B representation-the empirically derived probabilities that 4-year-olds assigned each label to each number. (In order to make Fig. 8 relatively readable, all probabilities have been rounded to the nearest tenth and probabilities of 10% or less are not shown.) The lowest level of the hierarchy involves individual numbers. In addition to being tied to the category labels with varying probabilities, the numbers are at times tied to each other by “next” connections. Some numbers are also labeled as members of the digit repetition and rule applicability lists. The smaller ones are involved in specific addition facts that the children are more or less confident of knowing. Although our experiments did not tap other information about indi-

A MODEL

COUNTING PROCESSISHOWN IN FIGURE 3 A ) MAGNITUDE COMPARISON PROCESSISHOWN IN FIGURE 5 A 1 AOOlTlON PROCESS SUBTRACTION PROCESS ON PROCESSISHOWN IN FIGURE IA)

I

w N 0

0 *I72

*,;3

+2=3 Usually sfarl

Number of

covnllng with

WCL.

orms.

Less thon

Number of

Number of

c o n . rides ond stdes ond leqr comers on cornera on

on verson

lrsongle

Nurnbcr of

NUMBER OF FINGERS

Imperr on hond

square

m y other

number

Fig. 8 . Integrative models of preschoolers' knowledge of numbers: ( A ) Model I .

29

30 31

39

100. 1000

6. MODEL

IT NUMBERS &UPON

W

0 w

+1=2(FJ6) +2=3(6B) 3 =4( 45) * 4 = 5 ( 35) 6 5=61491

*I = 3(75) *2=4(69) +3.5(49) +4=6(2 8 ) +5=7(26)

+1=4(45) .2=5(40) +3=6(35) t4=7( 09) +5=B( 13)

usually start countmg

Numbn 01 Of eyes. ears, arms. legs on

Number 01 sides ond

wlh

.I = 5 ( 35) * 2 =6(23)

+3=7(071 + 4 = B (36) .5=9( IS) Numbtr 01 sides ond

corners on

corners on

lr4a"gle

sawre

+I =6149) + 2 =7( 52) t 3 = a( 341 + 4 = 9( O W +5=10(38) Number 01 Impas on hand

CAN BE OPERATED+ BY

COUNTING PROCESS (SHOWN IN FIGURE 38) MAGNITUDE COMPARISON PROCESS(SH0WN IN FIGURE 581 ADDITION PROCESS(SH0WN IN FIGURE 7 ) SUBTRACTION PROCESS

Number of lingers

Old

Number o f loes

person

Less than any olher number N u m b n of heads. bodies. noses. mouths m person

*=

DIGIT R E P E T I T I O N LIST

Doddy 15 30 years

* *=

RULE A P P L I C A B I L I T Y L I S T

Fig. 8 . ( B ) Model I I . Model III on following p a g e .

C MODEL III (COUNTING PROCESSISHOWN IN FIGURE 3 C I

CATEGORICAL ORGANIZATIONl PROBABIL. I T I

A

5

L

I

2.3

4.5

6.7.8.9

B

2

I

I

2

3.4

1

I

2.3.4

5.6

5.6.7.8.9 7.8.9

D

I

I

2.3

4.5.6.7

8.9

1

E

I

I

2.3.4.5

6.7.8

9

I

c

*:DIGIT

REPETITION LIST

**=RULE

APPLICABILITY L I S T

*t*= HUNDREDS

LIST

Fig. 8. ( C ) Model 111.

I I

J

Nicmericul Understundings

305

vidual numbers, informal discussions with preschoolers suggest that they know many other facts about them. For example, a 4-year-old told us that 1 is the number that she starts counting with, that it is the number of heads, bodies, noses, and mouths on a person, and that it is the smallest number. The Model I1 depiction of moderately skilled performance provides a vantage point for considering the more advanced knowledge depicted in Model 111 and the less advanced knowledge depicted within Model I. First, let us consider some properties that are hypothesized not to change within this age and skill range. As mentioned above, the basic hierarchical form of the representation is constant across the three models. The children have knowledge about numbers in general, about categories of numbers, and about particular numbers. Also relatively constant across the models are many of the particular connections within and across levels of the hierarchy: even in Model I, the larger numbers are more often assigned to the larger categories; even in Model I, some of the “next” connections between digits are present; even in Model I, some facts linking individual numbers to other semantic properties (e.g., people have two hands) are known. Development in these (though not all) aspects of the representations appears to be a gradual, incremental process. At the other extreme, development can be seen in sharpest relief in the processes that children apply to their representations. These processes change greatly in all three task domains between Models I and 11, and the processes for counting and comparing undergo large changes between Models I1 and 111 as well. The counting process changes from using only “next” connections in Model I to also using rule applicability and digit repetition lists in Model I1 to using all of the above information and also the hundreds list in Model 111. The magnitude comparison process changes from guessing in Model I to comparing labels attached to individual numbers in Model I1 to comparing labels derived from categorical organizations in Model 111. The addition process changes from sole reliance on memorized facts in Model I to supplementary use of reconstructive strategies such as putting up fingers and counting fingers in Models I1 and 111. The pattern is reminiscent of the often-expressed speculation that development entails at least as great a growth in what children can do with information as in the amount of information that they possess (Bruner, 1973; Piaget, 1972; Simon, 1972). The issue of intertask relationships is addressed implicitly in all three models. The models suggest that preschoolers ’ understandings of counting, comparing, conserving, and adding are linked in some ways, but not in all of the ways that they could be. All of the processes operate upon a common representation, and this seems to produce some commonalities. Most dramatically, both conservation and magnitude comparison processes utilize the categorizations of numbers, and all processes except magnitude comparison make use of the links among the individual numbers. Other intertask connections that could have been present

306

Ruhcw S . Sieglrr

(inti

Mitchell Rohinsr~ri

were not, however. Preschoolers could have used their knowledge of counting to compare numerical magnitudes but they did not seem to. They could have used their knowledge of comparing to add numbers, as in the Groen and Parkman min model. but again they did not seem to. This last finding especially suggests that early mathematical skills may develop in relative isolation from one another; once children are proficient in the individual skills, they may make greater use of the potential interconnections among them. How can we evaluate the quality of these models? Empirically, they predict in detail preschoolers’ counting, comparing, conserving, and adding. Model I1 can again be used to illustrate. When counting, the model will stop at “9s,” will skip and repeat entire decades, will introduce nonstandard numbers if the boundary of its digit list is too high, and will count on at least to the ne3t “9” from points within or beyond its spontaneous counting range. When comparing the magnitudes of numbers, it will err most often on problems with large minima and small splits, will assign labels that correspond reasonably well to the relative magnitudes of the numbers, and will learn from instruction that adds an overall categorical organization to the existing connections between individual numbers and categories. When performing number conservation problems, it will judge small number problems in terms of the type of transformation but will judge large number problems in terms of the relative lengths of the rows. When adding, it will usually recall the answers to the easiest problems without using visible strategies, and will more frequently use such visible strategies as the problem increases in difficulty. Thus, the model mimics a considerable range of preschoolers’ behaviors in manipulating numbers. A second virtue of the models is the quality that Klahr and Wallace (1976) termed developmental tractability. For most of the changes between models, we can easily imagine how the more advanced form could grow out of the less advanced one. In counting, children first learn the “next” connections that are the only relations that bind the first numbers they encounter; then they add to this knowledge information about the cyclical patterns inherent in the next higher numbers they learn; eventually they extend the list membership notion to include the much larger numbers that they encounter yet later. In learning about numeiical magnitudes, children first obtain a rough sense of magnitudes that allows them to assign individual numbers to categories having some correspondence to the sizes of the numbers; then they learn how to use the categorical information to compare magnitudes; finally they impose overall categorical organizations that subsume the connections between individual numbers and categories but avert errors. In learning about number conservation, children first rely on the type of transformation only in limited situations, and gradually expand that reliance to encompass all three transformations and all set sizes. In adding, children first memorize solutions to specific problems, and then learn supplementary reconstructive strategies to use on problems where they cannot retrieve the answer.

Numerical iJnder.s:andings

307

Thus, development in this age range and content area involves few false starts; children build on what they already know to construct increasingly successful approaches. The separation between representations and processes in the model proved useful for specifying the source of this developmental tractability. Modeling approaches that focus solely on processes, such as the rule assessment approach, might have revealed as much about developmental changes in the preschool period, but probably would not have revealed the developmental constancies that also were present. The representation-process distinction also provided a basis for hypothesizing why early error patterns often foreshadow later reaction time patterns: increasingly powerful processes operating upon fundamentally similar representations. A final strength of the models is that they should be easy for us and other investigators to build upon. They can be expanded both outward, to encompass additional aspects of preschoolers’ knowledge of numbers, and upward, to include the more advanced knowledge of school age children. One early sign of this intellectual “developmental tractability” was the ease with which we could integrate the new information that 4-year-olds in the magnitude comparison training experiment were taught with the model of their existing knowledge (i.e., the extension of the Fig. 5B model to produce the one represented in Fig. 5D). We anticipate that it also will be relatively straightforward to add to the present models information about preschoolers’ ability to subtract, to count objects, to subitize, and to estimate the numerosity of large collections. Other reasonable goals include expanding the models upward so as to include more complex addition and subtraction skills, the relationships of addition and subtraction to multiplication and division, and the extension of arithmetic operations to the rational numbers. This is not to underestimate the difficulty of achieving these objectives, but rather to affirm that the present hierarchical models provide a base that is far from closed. All models have weaknesses as well as strengths. The two greatest weaknesses of the present models seem to be a lack of detail concerning how children choose to use a particular process in a particular situation and a lack of flexibility to cope with novel situations. With regard to the first point, the model of each process is introduced by the rather opaque test, “conditions for process x met?” Even the 3-year-olds counted when we asked them to count, compared when we asked them to compare, and added when we asked them to add. We do not understand, however, how they knew to do so. Recent efforts to discover how children interpret arithmetic word problems (Carpenter & Moser, 1981; Greeno et NI., 1981 ; Nesher, I98 1) represent a first step toward modeling how children understand instructions. Without further research on how language understanding occurs, however, this part of the model must remain a black box. The second weakness of the models is their lack of flexibility for adapting to

Robert S . Siegler and Mitchell Robinson

308

novel situations. This weakness applies most directly to the portrayal of magnitude comparison. The links between categories and individual numbers are presented as fixed in all three models. This may be a realistic depiction of the long-term memory contents of preschoolers, but almost certainly would not continue to be a realistic depiction in older children and adults. The lack of even a poorly developed mechanism for taking into account the effects of context is a weak point in the general developmental tractability of the models. In addition, just as children may not apply the same process to solving each addition problem, they may not apply the same process to solving every magnitude comparison problem. They may directly retrieve some pairs, may judge others relative to some common reference point, and may in general use any number of idiosyncratic judgment techniques in the comparison process. Illustratively, when one of us recently asked his 5-year-old son whether 16 or 33 was the bigger number, the child counted out loud from 1 to 16 and then said that 33 was bigger. Our error data, Sekuler and Mierkiewicz’s (1977) reaction time data, and previous observations of this child indicate that the counting strategy is far from the rule in this age range or even for this individual; nonetheless, children may use it, and probably many other approaches, sometimes. Both the strengths and the weaknesses of our models converge on two final points. First, conceptual development is far too complex for us to assess children’s understanding by examining performance on a single task. No one age is the age at which a concept is understood, and there is little meaning to saying that one concept is understood before, after, or simultaneously with another. Conceptual understanding has many facets, and only by investigating a concept both broadly and deeply do we have any hope of discovering what people know about it. Second, it is possible and desirable to build integrative models of children’s knowledge across different tasks corresponding to a single concept. These models help us to realize which aspects of children’s understandings we have accounted for, to notice the aspects that we have not yet addressed, and to face those implications that we did not intend and would like to change. In short, such models can help the work cumulate.

REFERENCES Aiken, L. R . , & Williams, E. N. Three variables related to reaction time to compare digits. Perceptual and Motor Skills, 1968, 27, 199-206. Anderson, 1. R . Language, memory, and thought. Hillsdale, N.J.:Erlbaum, 1976. Anderson, J R. Arguments concerning representations for mental imagery. Psychological Review, 1978, 85, 249-277. Ashcraft, M. H., & Battaglia, J. Cognitive arithmetic: Evidence for retrieval and decision processes in mental addition. Journal of Experimental Psychology: Human Learning and Memory, 1978, 4, 527-538. I

Numerical Understundings

309

Banks, W. P. Encoding and processing of symbolic information in comparative judgments. The Psychology ofleurning and Motivation, 1977, 11, 101-159. Banks, W. P., Fujii, M., & Kayra-Stuart, F. Semantic congruity effects in comparativejudgments of magnitude of digits. Journal of Experimental Psychology: Human Perception and Performunce, 1976, 2, 435-447. Banks, W. P., & Hill, D. K. The apparent magnitude of number scaled by random production. Jourtzul ojExperimeniol Psychology, 1974, 102, 353-376. (Monograph) Beilin, H. Cognitive capacities of young children: A replication. Science. 1968, 162, 920-921. Bever, T. G., Mehler, J., & Epstein, J. What children do in spite of what they know. Science, 1968, 162, 921-924. Braine, M. D. S. The ontogeny of certain logical operations: Piaget's formulation examined by nonverbal methods. Psychological Monographs. 1959, 73 (Whole No. 475). Brainerd, C. J. The origins of the number concept. New York: Praeger, 1979. Brown, A. L. The construction of temporal succession by preoperational children. In A. D. Pick (Ed.), Minnesota Symposium on Child Psychology (Vol. 10). Minneapolis: Univ. of Minnesota, 1976.

Brown, A. L. Knowing when, where, and how to remember: A problem of metacognition. In R. Glaser (Ed.), Advances in instructional psychology. Hillsdale, N.J.: Erlbaum, 1978. Bruner, J. S. Beyond the informution given: Studies in the psychology of knowing. New York: Norton, 1973. Bryant, P. E. Perception and understanding in young children. New York: Basic Books, 1974. Carpenter, T. P., & Moser, J . M. The development of addition and subtraction problem solving skills. In T. P. Carpenter, J . M. Moser, & T. A. Romberg (Eds.), Addition and subtraction: A cognitive perspective. Hillsdale, N.J.: Erlbaum, 1981, in press. Chi, M. T., & Klahr, D. Span and rate of apprehension in children and adults. Journal of Experimenid Child Psychology, 1975, 19,434-439.

Cooper, R., Starkey, P., Blevins, B., Goth, P., & Leitner, E. Number development: Addition and subtraction. Paper presented at the meeting of the Jean Piaget Society, Philadelphia, May 1978. Estes, W. K. The problem of inference form curves based on group data. Psychologicul Bulletin. 1956, 53, 134-139. Fairbank, B. A. Experiments on the temporal uspects of number perception. Unpublished doctoral dissertation, University of Arizona, 1969. Flavell, J . H. Stage-related properties of cognitive development. Cognitive Psychology. 1971, 2 , 42 1-453.

Fuson, K. C., & Richards, J. Children's construction of the counting numbers: From a spew to a bidirectional chain. Paper presented at the meetings of the American Education Research Association, 1980, Boston. Fuson, K. C., & Richards, J. The acquisition anb elaboration of the number sequence. In C. Brainerd (Ed.), Progress in cogniriw development.Vol. I . Berlin and New York: Springer-Verlag. 1982, in press. Gelman, R. The nature and development of early number concepts. Advunres in Child Development und Behuvior 1972, 7 , 115-167.

Gelman, R. Cognitive development. Annual Review oj't'sychology. 1978, 29, 297-332. Gelman, R . , & Gallistel, C. R. The child's understanding qfnumber. Cambridge, Mass.: Harvard Univ. Press, 1978. Gelman, R., & Starkey, P. Development of addition and subtraction abilities prior to formal schooling in arithmetic. In T. P. Carpenter; J. M. Moser, & T. A. Romberg (Eds.), Addifion und subtraction: A cognitive perspective. Hillsdale, N.J.:Erlbaum, 1981, in press. Ginsburg, H. Children's urithmetic: The learning process. New York: Van Nostrand, 1977.

310

Robert S . Siegler orid Mitchell Robinson

Glushko, R. J . , & Cooper, L. A. Spatial comprehension and comparison processes in verification tasks. Cognitive Psychology, 1978, 10, 391 -421. Greeno, J. G., Riley, M. S., & Gelman, R. Young children’s counting and understanding of principles. Unpublished manuscript. Groen, G . J . , & Parkman, J . M. A chronometric analysis of simple addition. Psychological Review, 1972, 79, 329-343. Groen, G . J . , & Resnick, L. B. Can preschool children invent addition algorithms? Journal of Educational Psychology. 1977, 69,645-652. Hebbeler, K. The development ofchildren’s problem-solving skills in addition. Unpublished doctoral dissertation, Cornell University, 1976. Ilg. F., & Ames, L. B. Developmental trends in arithmetic. Journd of Genetic Psychology. 195 1, 79, 3-28. Johnson, S . C. Hierarchical clustering schemes. Psychometrika. 1967, 32, 241 -254. Kintsch, W . , & van Dijk, T . A. Toward a model of text comprehension and production. Psychological Review, 1978, 85, 363-394. Klahr, D., & Wallace, J. G. Cognitive development: An infurmaticin-processing view. Hillsdale, N.J.: Erlbaum, 1976. Knight, F. B., & Behrens, M. S. The learning of the 100 oddiiion combinutions and the 100 subtraciion combinations. New York Longmans, Green, 1928. Kosslyn, S. M. Imagery and cognitive development: A teleological approach. In R. S. Siegler (Ed.), Children’s thinking: Whaf develops? Hillsdale, N.J.: Erlbaum, 1978. Kosslyn, S . M., Murphy, G. L . , Bemesderfer, M. E., & Feinstein, J. J. Category and continuum in mental comparisons. Journal of Experimental Psychology: General. 1977, 106, 341 -375. Kruskal, J . B., & Wish, M. Multidimensional scaling. Beverly Hills, Calif.: Sage Univ. Press, 1978. Lehman, H. Intrrduciion to the philosophy of mathematics. Totowa, N.J.: Rowman & Littlefield, 1979. LNR. Explurutions in cognition. San Francisco: Freeman, 1975. Maki, R. H. Categorization and distance effects with spatial linear orders. Journul of Experimental Psychology: Human Learning and Memory. 1981, I , 15-32. Manis, F. R., Keating, D. P., & Morrison, F. J. Developmental differences in the allocation of processing capacity. Journal of Experimental Child Psychology, 1980, 29, 156-169. Mehler, J . , & Bever, T. G. Cognitive capacity of very young children. Science, 1967, 158, 141142. Miller, S . A. Nonverbal assessment of conservation of number. Child Development, 1976, 47, 722-728. Moyer. R. S., & Bayer, R . H. Mental comparison and the symbolic distance effect. Cognitive PSychology, 1976, 8, 228-246. Moyer, R. S . , & Dumais, S. T . Mental comparison. The psychology of learning and motivution, 1978, 12, 117-155. Moyer, R. S . , & Landauer, T. K. The time required for judgments of numerical inequality. Nature (London). 1967, 215, 1519-1520. Neisser. V . General, academic, and artificial intelligence. In L. B. Resnick (Ed.), The nuture qf intelligence. Hillsdale, N.J.: Erlbaum, 1976. Nesher, P. Levels of description in the analysis of addition and subtraction. In T. P. Carpenter, J. M. Moser, & T. A. Romberg (Eds.), Addition and subtraction: A cognitive perspective. Hillsdale, N.J.: Erlbaum, 1981, in press. Newell. A. A note on process-structure distinctions in developmental psychology. In S. FarnhamDiggory (Ed.), Information processing in children. New York Academic Press, 1972.

Numericul Understandings

31 I

Newell, A . You can’t play 20 questions with nature and win: Projective comments on the papers of this symposium. In W . G. Chase (Ed.), Visuul information prowusing. New York: Academic Press, 1973. Parkman, J. M. Temporal aspects of digit and letter inequality judgments. Journal of Experimentcil P ~ j ~ h ( ~ l o g1971, y . 91, 191-205. Peano, G . Formulaire de mathematic1ues (Vols. 1-5). Turin: Broca, 1894-1908. Piaget, J . The child’s concept ofnumber. New York: Norton, 1952. Piaget, J. lntellectual evolution from adolescence to adulthood. Humcin Development. 1972, 15, 1-12. Pliske, R. M., & Smith, K. H. Semantic categorization in a linear order problem. Memory & Cognition, 1979, 7, 297-302. Pollio, H. R., & Reinhardt, D. Rules and counting behavior. Cognitive Psychology. 1970, 1, 388-402. Pollio, H. R., & Whitacre, J . Some observations on the use of natural numbers by preschool children. Perceptual and Motor Skills, 1970, 30, 167-174. Rothenberg, B. B., & Courtney, R. G . Conservation of number in very young children. Developmental Psychology, I 969. 1, 493-502. Rule, S. J. Equal discriminability scale of number. Journal ofExperimental Ps.ychology. 1969, 79, 35-39. Schaeffer, B., Eggleston, V. H . , & Scott, J. L. Number development in young children. Cognitive Ps.ycho1og.V. 1974, 6 , 357-379. Sekuler, R., Armstrong, R., & Rubin, E. Processing numerical information: A choice time analysis. Journul of Experimental Psychology, 1971, 90, 75-80. Sekuler, R., & Mierkiewicz, D. Children’s judgments of numerical inequality. Child Development, 1977, 48, 630-633. Shepard, R. N., Kilpatric, D. W., & Cunningham, J. P. The internal representation of numbers. Cognitive Psychology. 1975, 6 , 82-138. Shepard. R. N.. & Podgorny, P. Cognitive processes that resemble perceptual processes. In W . K . Estes (Ed.), Handbook of learning and cognitive processes. Hillsdale, N.J.: Erlbaum, 1978. Siegler, R . S. Defining the locus of developmental differences in children’s causal reasoning. Journal of Experimental Child Psychology, 1975, 20, 512-525. Siegler, R. S. Three aspects of cognitive development. Cognitive Psvcho1og.y. 1976, 8, 481-520. Siegler. R. S. The origins of scientific reasoning. In R. S. Siegler (Ed.), Children’s thinking: What develops? Hillsdale, N.J.: Erlbauni, 1978. Siegler, R. S. Developmental sequences within and between concepts. Monographs of the Sociery,for Research in Child Development. 1981, 46 (Whole No. 189). (a) Siegler, R. S. Information processing approaches to development. In W. Kessen (Ed.), Manual C~ Child Psychology: History, theories und methods. New York: Wiley, 1981, in press. (b) Siegler. R. S., & Klahr, D. When do children learn: The relationship between existing knowledge and the ability to acquire new knowledge. In R. Glaser (Ed.), Advcinces in instructionul psychology. Hillsdale, N.J.:Erlbaum, 1981, in press. Siegler, R. S., & Robinson, M. Preschoolers’ knowledge of very large numericul magnitudes. Manuscript in preparation. I98 I . Simon, H. A. On the development of the processor. I n S . Farnham-Diggory (Ed.), Infmnution processing in children. New York: Academic Press, 1972. Smedslund, J. Microanalysis of concrete reasoning. 1. The difficulty of some combinations of addition and subtraction of one unit. Scnndinmian Journal ofPsychology, 1966, 7 , 145-156. Sternberg. R. J. Component processes in analogical reasoning. Psyc‘holo~icdRevirw, 1977, 84, 353-378.

312

Robert S . Siegler and Mitchell Robinson

Strauss, S., & Levin, 1. A commentary on Siegler, R. S. Developmental sequences within and between concepts. Monographs of the Society for Research in Child Development. 1981, 46, (Whole No. 189). Svenson, 0. Analysis of time required by children for simple additions. Act0 Psychologica, 1975, 39, 289-302. Svenson, 0.. & Broquist, S. Strategies for solving simple addition problems. Scandinavian Journal of Psychology, 1975, 16, 143-151. Trabasso, T.The role of memory as a system in making transitive inferences. In R. V. Kail, Jr. & J. W . Hagen (Eds.), Perspectives on the developmeni of memory and cognition. Hillsdale, N.J.: Erlbaum, 1977. Trabasso, T. How do children solve class inclusion problems? In R. S. Siegler (Ed.), Children's thinking: What develops? Hillsdale, N.J.: Erlbaum, 1978. Trabasso, T., Riley, C. A,, & Wilson, E. G . The representation of linear order and spatial strategies in reasoning: A developmental study. In R. J. Falmagne (Ed.), Reasoning: Representation and process. Hillsdale, N.J.: Erlbaum, 1975. Yoshimura, T. Strategies for addition among young children. Paper presented at the 16th annual convention of the Japanese Association of Educational Psychology, 1974.

AUTHOR INDEX Numbers in italics refer to the pages on which the complete references are listed.

A

Bates,E., 8, 1 1 , 12, 13, 14, 15, 16, 17, 18, 19, 2 0 , 2 1 , 4 8 , 5 1 , 5 2 , 6 2 , 6 5 , 6 7 , 6 8 , 69, 70. 7 / , 75 Battaglia, J., 289, 290, 308 Batter, B. S., 226, 237 Bauer, J. A , , 82, 115 Bayer, R. H., 268, 310 Bayley, N . , 215,216,217,218,222,225,234, 237 Beale, 1. L., 82, 83, 84, 106, 117 Beckwith, L., 24, 27, 28, 29, 30, 31, 33, 69, 71. 225, 237 Bedrosian, J . , 13, 14, 73 Beebe, B., 22, 75 Beeghly-Smith, M., 12, 16. 67, 68, 70 Behrens, M. S., 287, 292, 295, 310 Beilin, H . , 242, 309 Bell, S . M., 19, 20, 69. 112, 115 Belsky, J., 69 Bemesderfer, M. E., 286, 310 Benigni, L., 13, 16, 17, 18, 19,20,21,65, 69, 70 Bennett, S . L.. 22, 7.5 Benton, A. L., 114, 115 Berkley, M. A,, 79, 115 Berko, J . , 9, 69 Berko-Gleason, J., 35, 58, 69 Berti, F. B., 113, 114, 119 Best, C. T., 106. 119 Bever, T. G., 10, 72, 243, 309, 310 Beverley, K. I., 96, 121, 122 Birch, H. G., 106, 123 Birns, B., 223, 237 Bishop, A., 84, I19 Bishop, P. O., 95, I 2 f Blake, R.. 79, 116 Blakemore, C., 95, 122 Blehar, M. C . , 17, 69, 234, 237 Blevins, B., 288, 309 Block, E., 15, 69 Block, J . , 173, 208

Aaron, P. G., 84, 113, 115 Abelson, W. D., 225, 239 Abramovitch, R., 180, 209 Adams, J. E., 95, I19 Aiken, L. R., 268, 308 Ainswonh, M. D. S., 17, 19, 20, 68, 69, 214, 233, 234, 237 Allen, T. W., 110, 119 Allman, J . M., 106, 115 Ames, L. B., 289, 3 / 0 Ammon, M., 67, 69 Ammon, P. R.,69 Anderson, E., 67, 69 Anderson, J. R., 128, 129, 164. 244,249,300, 308 Anderson, N., 80, 118 Andriessen, J. J., 80, 95, 116 Anglin, J. M., 139, 151, 164 Annis, R. C., 96, 11.5 Appelbaum, M. I., 89, 121 Appelle, S., 79, 81, 82, 85, 115 Arend, R. A , , 19, 20, 73 Armstrong, R., 268, 311 Arnheim, R . , 79, 115 Ashcraft, M. H., 289, 290, 308 Atkinson, J . , 95, 100, I I I , 115 Attneave, F., 79, 80, 81, 115 Auk, R. L., 133, 164 Austin, G., 131, 164 Aviezer, L., 13, 72

B Baird, J. C., 80, 122 Bakeman, R.,27, 28, 32, 33, 69 Banks, M. S.,I l l , 115 Banks, W. P., 268, 269, 270, 275, 284, 3OY Bartlett, F. C . , 102, I15 Bassili, J., 172, 208

313

3 I4

Author Index

Bloom, L., 12, 65, 67, 70 Boden, M . , 158, 164 Bohannon, J., 54, 70 Bonvillian, M.. 43, 58, 73 Borke, H., 179, 208 Bornstein, M. H., 80, 82, 84, 86, 87, 88, 89, 90,91,92,93,94,95,96,97,98,99, 100, 101, 102, 103, 104, 105, 106, 110, 113, 114, 116. 118. 119 Boswell, S. L., 80, 92, 116 Botkin, P., 67, 72 Bouma, H., 80, 95, 116 Bourne, L. E., 131, 164 Bowennan, M. F., 1 I , 12, 15, 70 Bowlby, J . , 17, 70, 214, 233, 237 Boysen, S., 65, 74 Braddick, F., 11 I , 115 Braddick, O., 111, 115 Bradley, D., 82, 83, 84, 100, 116 Bradshaw, J . , 82, 83, 84, 100, 116 Braine, L. G . , 81, 82, 84, 87, 116, 118 Braine, M. D. S., 15, 70. 243, 309 Brainerd, C. J . , 242, 309 Branigan, G., 14, 70 Bransford, J . D., 138, 141, 142, 143, 165 Brazelton, T. B., 22, 70, 186, 208 Bresson, F., 106, 116 Bretherton, I . , 12, 13, 16, 17, 18, 19, 20, 21, 51, 65, 67, 68, 69, 70 Brill, S., 8 5 , 95, 109, I I Y , 121 Broadbent, D. E., 111, 116 Bronson, G . , 102, 116 Brooks, B., 106, 116 Brooks, J . , 174, 178, 179, 210 Brooks, L., 146, 147, 148, 149, 151, 153, 157, I64 Broquist, S . , 288, 312 Brown, A. L., 243, 250, 309 Brown, D. R., 79, 117 Brown, J., 27, 28, 32, 33, 69 Brown, R., 9, 15, 70 Bruner, J . S . , 9, 1 I . 12,22,51,65, 70, 74, 75, 131, 159, 164, 305, 309 Bryant, P. E., 82, 84, 116. 243, 309 Buck, R., 172, 173, 174, 181, 182, 183, 184, 206, 207, 208, 208, 209 Burchinal, P., 27, 29, 32, 33, 70 Butler, J . , 83, 84, 106, 116 Butterfield, E. C., 127, 165

C

Cairns, N. U . , 113, 116 Caldwell, B., 33, 71 Caldwell, R. A., 104, 116 Camaioni, L., 12, 13, 16, 17, 18, 19, 20, 21, 65, 6 9 , 70 Camisa, J . M . , 79, 116 Campbell, F., 24,25,26,28,30,31,32,33, 74 Campbell, F. A , , 223, 238 Campbell, F. W . , 79, 95, 116. 119 Campos, J . J . , 172, 174, 177, 178, 205, ZOY, 210 Caplan, P. J., 106, 116 Carey, S.,139, 164, 180, 209 Carlson-Luden, V., 54, 71 Caron, A. J., 104, 116, 117 Caron, R. F., 104, 116, 117 Carpenter, P. A , , 114, I20 Carpenter, T. P., 307, 309 Case, R . , 13, 14, 71, 150, 164 Casler, L., 233, 237 Caul, F., 172, 173, 182, 209 Caul, W., 172, 182, 209 Causkaddon, G . , 43, 58, 73 Cazden, C., 43, 58, 71 Chang, J.-J., 111, 119 Chapman, R . , 13, 14, 73 Charlesworth, W. R., 170, 174, 209 Chase-Lansdale, P. L., 234, 238 Chevalier-Skolnikoff, S . , 65, 71 Chi, M. T. H., 150, 165. 286, 309 Chikvishvili, L., 172, 209 Chomsky, N., 8, 9, 10, 71, 140, 165 Clark, E. V., 114, 117, 152, 165 Clark, H. H., 114, 117 Clark, R., 65, 71, 106, 123 Clarke-Stewart, K. A., 20, 21, 36, 37, 48, 71, 224, 225, 227, 234, 237 Cohen, D.,185, 204, 205, 208, 209 Cohen, L. B., 96, 117 Cohen, S . , 24, 27, 28, 29, 30, 31, 33, 71 Cohen, S. E., 225, 237 Collins, A. M.,161, 165 Collis, G. M., 12, 74 Conezio, J . , 144, 166 Connell, D. B., 20, 21, 71 Cooper, L. A., 156, 165. 299, 310 Cooper, R.,288, 309

315

Author Index

Cooper, W., 66, 71 Corballis, M. C., 80, 81, 82, 83, 84, 96, 106, 1 1 I , 117 Cornell, E. H . , 98, 117 Comgan, R., 13, 71 Corwin, T. R . , 79, 117 Cosgrove, M. P., 79, 117 Courtney. R. G . , 242, 243, 311 Covitz, F., 24, 25, 26, 28, 29, 30, 31, 32, 51, 75 Craig, E. A , , 79, 117 Crano, W. D., 217, 222. 237 Cremona, C., 67, 71 Critchley, M., 106, 117 Cross, T., 38,39,40,41,42,43,45,47,55, 71 Culicover, P., 8, 75 Cullen, J. K . , Jr., 95, 96, 121 Cunningham, J. P., 268, 311 Curlee, T. E., 81, 115 Curtiss, S . , 15, 63, 71

D Dale, P.,9, 71 Daly. E., 180, 209 Darwin, C., 171, 209 Davidson, C. V.. 237 Davidson. H. P. A., 110, 113, 117 Davis, A. E., 106, 117 Day, R. H . , 86, 95, 121 Decarie, G. T., 174, 178, 210 De Frank, R. S.,211 De Fries, J. C., 60, 65, 72. 74 De Loache, J. S., 96, 117 Deregowski, J. B., 80, 117 Derryberry, D., 232, 238 de Schonen, S . , 106. 116 Devin, J., 35, 74 Diamond, R . , 180, 209 Dik, S., 66, 71 Dolan, A. B., 216, 217, 223, 233, 237, 2-38 Dougherty, L. M . . 172. 177, 210 Dumais, S. T., 268, 269, 310

E Edsterbrooks, M. A., 235, 237 Edwards, D . , 1 1 , 12, 71 Egglesron, V. H., 269, 270, 311

Eichorn, D. H., 14, 73, 228, 238 Ekman, P., 171, 172, 175, 176, 179, 185, 2 0 ~ . 210 Eliuk, J., 80, 119 Ellsworth, P., 172, 20Y Emde, R. N., 172, 174, 176, 177, 178, 205, 20Y. 210 Emsley, H. H . , 79, 117 Essock, E. A . , 79, 80, 82, 86, 88, 97, 117 Epstein, J . , 243, 30Y Eriksen, C. W . , 129, 165 Ervin-Tripp, S., 67, 68, 71 Estes, D. A . , 219, 235, 23Y Estes, W. K., 290. 309 Eysenck, H. J., 184, 185, 189, 206, 208, 209

F Fagan, J. F., 86, 87, 102, 104, 105, 117 Fairbank, B. A , , 268, 309 Fantz, R. L., 98, 117 Farah, M. J . , 165 Fman, D., 24, 25, 26, 27, 28, 29, 30, 31. 32, 33, 70, 71, 74 Farrell, W. S . , 83, 106, I18 Feinstein, J . J., 286, 310 Feldman, C., 43, 71 Fellows, B. J., 85, 118 Fenson, L., 14, 71 Ferdinandsen, K . , 84, 89, 90, 91, 92, 93, 94, 116. 118 Ferguson, C., 35, 54, 71, 72, 75 Feshbach, N., 180, 209 Field, T., 185, 193, 204, 205, 208, 20Y, 211 Fillmore, L . , 65, 72 Finke, R. A . . 165 Finlay, D. C., 94, I I Y Fischer, K. W.,13, 72 Fisher, C. B., 82, 84, 89, 90, 92, 113, I18 Fitts, P. M.,80, 92, 118 Flavell, J . H . , 9, 67, 72, 133, 154, 165, 243, 309 Fodor, J . A . , 10, 72, 127, 128, 133. 165 Fox, J . , 80, 118 Franks, J. J., 138, 141, 142, 143, 165 Freedle, R . , 25, 28, 72 Freedman, D. G., 231, 233, 237 Freeman. R . B., Jr., 106, 121 Freeman, R. D.. 95, 118

316

Author Index

Fregnac, Y., 96, 118 French, J., 95, 100, 115 Friedman, S . L., 104, 117 Friesen, W . V., 171, 172, 179, 209 Frith. U., 113, 118 Frodi, A. M . , 235, 238 Frodi, M.,235, 238 Frost, B. J., 95, 96, 115, 118 Fry, C . , 67. 72 Fujii, M., 269, 275, 284, 309 Fulgham, D. D., 79, 117 Furrow, D., 38, 39, 40, 41, 55, 72 Furth, H. G., 9, 72 Fuson, K. C., 309

G Gaensbauer, T., 174, 178, 209 Galanter, E. H., 9, 73 Gallistel, C. R., 250, 270, 298, 309 Garcia, J., 72 Gardner, J . , 106, 118 Garn, S. M., 217, 224, 228, 237 Garner, W. R . , 81, 118, 129, 157, 165 Garrett, M. F., 10, 72 Geis, M. F., 77, 121 Geldard, F. A., 78, 118 Gelman, R., 35, 74, 243, 250, 270, 287, 288, 298, 309, 310 Gibson, E. J . , 82, 90, 96, 107, 110, 112, 113, I I8 Gibson, J . J., 78, 79, 80, 82, 112, 118 Ginsburg, A. P., 1 1 1 , 118, 119 Ginsburg, H.,250, 251, 270, 309 Gitter, G . , 179, 209 Glanville, B. B., 106, 119 Glass, A. L., 150, 160, 161, 165 Gleitman, H.,35, 37, 39,40,41, 42, 43, 52, 55, 62, 63, 73 Gleitman, L., 35,37, 39,40,41,42,43,52,55, 62, 63, 73 Glushko, R. J . , 299, 310 Golden, M., 223, 237 Goldmeier, E., 80, 81, 105, 119 Goldsmith, H. H., 231, 232, 237 Goldstein, A. J . , 101, 119 Golinkoff. R.. 72 Goode, M. K., 69 Goodman, N., 131, 165 Goodnow, J., 131, 164

Gordon, E. W., 106, 123 Goren, C. C., 186, 209 Goth, P., 288, 309 Gottesman, I. I . , 231, 232, 237 Goyen, J., 114, 119 Grandy, C. A., 114, 119 Gray, J . A., 184, 210 Green, M. A , , 79, 117 Greenberg, R . , 185, 204, 205, 208, 209 Greenfield, P. M., 9, 11, 57, 68, 70. 72. 75, 159, 164 Greeno, J. G., 310 Groen, G. J., 250,288,289,290,292,295,310 Gross, C. G., 84,86,87,88,89,90,91,93,94, 97, 98.99, 100, 101, 102, 103, 104, 105, 106, 110, 113, 114. 116, 118, 119 Guillaume, P., 210 Gwiazda, J., 85, 95, 109, 119, 121

H Haber, R. N., 144, 166 Hack, E., 20, 72 Haith, M. M., 92, 96, 119 Hake, H. W.,129, 165 Hall, J. A., 211 Halpern, E., 13, 72 Hamilton, C. R.. 106, 119 Hamm, A , , 106, 123 Hardy-Brown, K., 60, 65, 72 Harkness, S.,38, 40, 41, 42, 43, 45, 72 Harlow, H. F., 63, 72 Harlow, M. K . , 63, 72 Harmon, L. D., 101, 111, 119 Harmon, R. J . , 19, 72 Harris, D. S., 113, 114, 119 Harris, L. J . , 110, 119 Harris, P. L., 84, 119 Haupe, G., 114, 119 Haviland, J., 174, 178, 179, 210 Hayes-Roth, B., 134, 142, 153, 165 Hayes-Roth, F., 128, 134, 142, 153, 165 Hebbeler, K., 289. 310 Held,R.,82, 85, 95, 109, 115, 119, / 2 / Heldmeyer, K. H., 150, 160, 165 Hemenway, K., 80, 81, 119 Henderson, C., 174, 178, 209 Herzka, H. S., 175, 204, 210 Heubner, R. R., 172, 177, 210 Hiatt, S. W., 172, 177, 205, 210

317

Author Index

Higgins, G. C., 79, / / Y Hildyard, A., 80, 115, 121 Hill, D. K., 268, 309 Hill, L., 181, 205, 211 Hirsch, H . V. B., 96, 119 Hogarty, P. S . , 14, 73, 222, 227, 228, 238 Hogben, J. H., 92, 119 Holyoak, K . J., 161, 165 Hood, L., 65, 67, 70 Horgan, D.,72 Horowitz, F. D., 90, 119, 176, 186, 211 Houlihan, K., 83, 84, 106, 122 Howard, I. P., 79, 81, 82, 112, 1IY Hubel, D. H., 82, 95, 119 Hubley, P., 12, 15, 75 Hunt, J. McV., 224, 23Y Hurlburt, N . , 222. 227, 228, 238 Huttenlocher, J., 84, 119 Hwang, C.-P., 235, 238

I Irnbert, M., 96, 118 Inhelder, B., 166 Ivinskis, A . , 94, 119 Iwahara, S . , 96, 123 1zard.C. E., 171, 172, 175, 176,177, 179, 181 182, 205, 210

J Jaffe, J., 22, 75 Jastrow, J.. 112, IIY Johnson, D. L., 238 Johnson, R. M.,80, 81, 119 Johnson, S. C., 310 Johnston, J . , 13, 72, 90, 118 Jonckheere, A. R., 105. 119 Jones, H. E., 173, 174, 178, 184, 206, 207, 208, 210 Julesz, B., 80, 92, 96, 111, I I Y Jung, R., 106, 116 Just, M.A . , 114, I20

K Kaas, J. H . , 106, 115 Kagan, J., 87, 123 Kaminer, J. J., 95, 118 Kaplan, B . , 16, 75

Karmel, B. Z., 96, 120 Kato, H., 95, 121 Kaye, K . , 22, 25, 26, 27, 28, 31, 34, 62, 72 Kayra-Stuart, F., 269, 275, 284, 309 Keating, D. P., 286, 310 Keating, M. T., 217, 224, 228, 237 Kee-le, S. W., 135, 136, 138, 155, 156, 166 Keene, G.C., 79, 119 Keller, B., 231, 233, 237 Kempler, D., 65, 72 Kendler, H. H., 150, 165 Kendler, T. S . , 150, 165 Kennedy, C. B., 89, 121 Kessel. F., 15, 6Y Kessen, W., 92, 96, 98, 116. 119, 122 Kilbride, J., 181, 205, 211 Kilpatric, D. W., 268, 311 King, W. L . . 223, 238 Kinskume, M., 106, 116 Kintsch, W., 300, 310 Kirk, B., 84, 122 Kitterle, F., 79, 115 Klahr, D., 286, 306, 30Y, 310, 311 Kleck, R., 173, 210 Klein, R. P., 19, 72 Kligman, D. H., 176. 209 Knight, F. B., 287, 292, 295, 3 / 0 Koelling, R . , 72 Koivumaki, J., 172, 211 Kopp, C., 33, 72 Kopp, C. B . , 225, 237 Koslowski, B., 22, 70 Kossan, N., 150, 151, 152, 165 Kosslyn, S. M., 128, 129, 150, 154, 155, 156, 159, 160, /65, 286, 299, 300, 310 Kreutzer, M. A., 170. 174, 209 Kruskal, J. B., 273, 310 Kuhn, T. S., 9, 73 Kulikowski, J. J., 79, 116 Kuno, S . , 66, 73

L La Barbera, J. D., 175, 205, 210 Lachman, R.. 127, 165 Lamb, M.E., 214, 217, 218, 219, 224, 225, 227, 228, 232, 234, 235, 237; 238, 23Y Landauer, T. K., 267, 268, 310 Lanzetta, J., 173, 210 Lashley, K. S . , 9, 73. 84, IIY

318

Aiirhor Index

Lasky, R. E., 140, 143, I65 Lam, M., 172, 209 Leaman, R., 80, 92, 122 Leavitt, L. A . , 176, 205, 2 / 0 Lechelt, E. G., 80, / I 9 Leehey, S. C., 85, 119 Lehrnan, H., 270, 310 Lehmim, R. A. W., 106, 1IY Leitner, E., 288, 30Y Lema, S . , 79, 116 Lernond, C., 181, 206, 210 Lennie, P., 79, 119 Leonard, J. A., 80, I18 Leonard, L., 13, 73 Lepez, R., 113, / / Y Lerner, C., 81, 82, 116 Lesk, A. B., 101, I I Y Le Tendre, J. B., 84, 119 Leventhal, A. G.,96, l / Y Levin, H., 107, 110, 112, 113, 118 Levin, I . , 244, 312 Levine, M., 131, 165 Levinson, J . , 79, / I 6 Levinson, R., 106, I19 Lewis, A , , 20Y Lewis, M., 22, 25, 28, 72. 73. 174, 178, 179, 210, 227, 238 Lewkowicz, D., 106, 118 Li, C., 66, 73 Liberman, 1. Y., 113, 114, 119, 122 Lichtenstein, M., 79, 117 Lightbown, P., 65, 67, 70 Ling, B. C., 86, / I 9 Lipets, M., 172, 2 / 1 Llg, F., 289, 3 / 0 Lock, A , , 12, 16, 22, 73 Lorenz, K.,9, 73 Lyle, J . G., 114, IIY

M Mach, E., 80, 81, 83, 96, 106, I I Y MacNamara, J., 11, 12, 73 MacWhinney, B., 8, 15, 68, 6 Y Maffei, L., 95, I19 Main, M., 19, 20, 21, 22, 70, 73, 227, 238 Maisel, E. B . , 96, I19 Maki, R. H . , 114, //Y. 286, 3 / 0 Maxwell, S . E . , 238 Malatesha, R. N., 84, 113, / / 5

Mandler, J . M., 84, 122 Manis, F. R . , 286, 3 / 0 Mansfield, R . J. W., 95, 102, / I 9 Marcy, T. G., 225,237 Marg, E., 95, 119 Marquis, A , , 54, 70 Martello. M., 102, 119 Matas, L., 19, 20, 73, 178, 211 Matheny, A. P., 216,217, 223, 231,232, 233, 237, 238 Maurer, D., 102, 119 Maury, L., 106, 116 May, J. G.,95, 96, 121 Mayer, M. J., 109, 12/ McCall, R. B., 14, 73, 89, 121, 222,227,228, 238 McCandless, 9.R., 77, 121 McCarter, R., 170, 171, 206, 2 / / McCarthy, D., 9, 73 McClearn, G. E., 60, 74 McFarland, J . H., 79, / 2 / McGinnes, G. C., 172, 177, 210 McGowan, R . J . , 238 McGurk, H., 103, 104, I21 McKenzie, B., 82, 86, 95, 121, 123 McNeill, D., 9, 73 McNew, S., 12,47, 51, 5 5 , 57,67,68, 70, 73 Medin, D. L., 144, 145, 146, 152, 153, 163, I66 Mehler, J . , 243, 30Y, 310 Meltzoff, A. N., 65, 73, 186, 2 / 0 Mervis, C. 9..136, 137, 138, 155, 166 Metzler, J . , 8 1 , 122 Mierkiewicz, D., 269, 273, 274, 285, 308, 3 / / Miller, A., 106, 117 Miller, E. R., 172, 173, 182, 209 Miller, G. A., 9, 73 Miller, J., 13, 14, 73 Miller, R. S., 166, 172, 182, 209 Miller, S . A., 242, 310 Millward, R. B., 132, 166 Minichiello, M. D., 104, 1 / 7 Mistler-Lachman, J., 127, 165 Mitchell-Kernan, C., 67, 68, 71 Mow, K . , 1 1 1 , 115 Moerk, E. L., 35, 73 Moffett, A , , 86, 95, I21 Mohindra, I . , 85, 95, 109, //Y, 121 Molfese, D. L., 106, 12/ Money, J., 114, 12/

Author Index

Moore, M. K., 65, 73, 186, 210 Morgan, M. J . , 106, 117 Morrison, F. J., 286, 310 Morse, P. A . , 176, 205, 210 Moser, J. M.,307, 30Y Moskowitz-Cook, A , , 79, 85,95, 96, 117, I I Y . 121

Most, R., 6Y Mostofsky, D., 179, 20Y Mowrer, 0. H., 65, 73, 82, I18 Moyer, R. S.,267, 268, 269, 310 Muir, D. W., 79, 96, 123 Murphy, G . L., 286, 310

N Nagamura, N . , 96, 123 Neisser, V.,244, 310 Nelson, C. A . , 176, 205, 210 Nelson, J . I . , 95, 121 Nelson, K., 12, 43, 44, 58, 59, 73 Nesher, P., 307, 310 Neumann, P. G., 138, 143, 166 Newell, A . , 127, 166, 287, 290, 300, 310, 311

Newmeyer, F., 1 I , 73 Newport, E., 35,37,39,40,41,42,43,52,62, 63, 73 Newson, J . , 23, 74 Nicolich, L . , 14, 74 Ninio, A , , 22, 70 Noble, J . , 106, f21 Norman, D. A . , 158, 166

319

P Palermo, D. S . , 106, 121 Palmer, S. E., 80, 81, 121 Parisi, S . A., 175, 205, 210 Parkman, J . M., 250, 268, 288, 289, 290, 292, 295, 310, 311 Parmelee, A. H., 33, 72, 225, 237 Parsons, G . , 12, 74 Pascual-Leone, J., 150, 166 Patterson, K., 82, 83, 84, 100, I16 Pawlby, S., 205, 210 Peano, G., 311 Pederson, J . A., 225, 227, 234, 237 Pentz, T., 18, 20, 21, 34, 36, 39, 40, 41, 42, 45, 48, 62, 74 Piaget, J . , 12, 13, 74, 112, 121, 166, 224,228, 238, 242, 305, 311 Pick, H. L., 79, 112, l l Y Pieraut-Le Bonniec, G., 106, 116 Pierce, S., 84, 122 Pinker, S . , 8, 15, 74 Pliner, P., 180, 20Y Pliske, R. M., 286, 311 Plomin, R., 50, 60, 65, 72. 74. 232, 233, 238 Podgorny, P., 268, 311 Pollio, H. R., 250, 255, 311 Pomerantz, J . R., 83, 106, 121 Posnansky, C. J . , 143, 166 Posner, M. I . , 135, 136, 138, 155, 156, 166 Premack, D., 68, 74 Pribram, K. H., 9, 73 Provost, A . M., 174, 178, 210 Pyiyshyn, 2. W., 128, 166

0 Ochs, E., 65, 67, 74 O’Connell, B., 14, 75 Odom, R . , 181, 206, 210 Olson, D. R., 80, 82, 115. I I Y , 133, 166 Olson, R. K . , 79, 80, 115 Olver, R., 9, 70, 159, 164 Orlando, C . , 113, 114, 119 Onon, S. T., 106, 113, 114, l l y Oster. H., 175, 185, 210 Over, J., 84, 100, 121 Over, R., 84, 100, 121 Owen, M. T., 234, 23X Owens, D. A., 82, 115 Owsley, C. J . , 90, I18

Q Quillian, M. R., 161, 165 Quincy, A , , 179, 20Y

R Ramey, C., 24, 25. 26, 27, 28, 30. 31, 32, 33, 71, 74, 223, 238 Ramsey, D. S., 14, 71 Rappapon, M., 80, 118 Ratner, N., 22, 74 Rced,S. K . , 135, 136, 140, 141, 144, 155, 156, 157, 166 Reese, H. W . , 155, 166

320

Author Index

Regan, D., 95, 96, 121, 122 Reich, J . H., 176, 209 Reilly, J., 57, 75 Reinhardt, D., 255, 311 Relyea, L., 81, 82, 116 Resnick, L. B., 288, 310 Rheingold, H. L., 233, 238 Richards, J . , 309 Riesen, A. H., 82, 84, 100, 123 Rieser, J.. 79, 112, 121 Riley, C. A., 286, 312 Riley, M. S., 310 Rips, L. J., 160, 161, 166 Risser, D., 172, 177, 210 Robinson, M.,269, 311 Rock, I., 79, 80, 81, 82, 83, 92, 105, 122 Roe, K.,180, 209 Rogosa, D., 56, 74 Roldan, C. E., 80, 81, 96, 1 1 1 , 117 Ronner, S. F., 95, 102, 119 Rosch,E., 112, 122, 136, 137, 138, 140, 152, 155, 157, 166 Rose, D., 95, 122 Rosenblith, J . F., 84, 122 Rosenblum, L. A , , 22, 73, 227, 238 Rosenfeld, H. M.,176, 186, 211 Rosenthal, R., 172, 211 Ross, G.,5 1 , 66, 71, 75, 92, 119 Rothbard, M. K . , 232, 238 Rothenberg, B. B., 242, 243, 311 Rowe, D. C., 50, 74, 232, 233, 238 Rubin, E., 268, 311 Rudel, R. G., 83, 114, 122 Rule, S. J., 268, 311 Rumbaugh, D. M., 65, 74 Rumelhart, D. E., 158, 166 Rutkin, B., 95, 119 Ryan, J., 12, 74 S

Sachs, J., 35, 74 Sager, L. C., 83, 106, 121 Salapatek, P.,92, 96, 111, 115, 119, 122 Sandeman, D. R., 106, 123 Sander, L. W . , 22, 74 Sarty, M., 186, 20Y Savage-Rumbaugh, E. S., 65, 74 Savin, J . V., 172, 173, 182, 209

Scaife, M., 65, 74 Scan, S . , 232, 238 Schachter, S., 173, 210 Schaefer, E. S . , 217, 222, 225, 234, 237 Schaeffer, B., 269, 270, 311 Schaffer, H. R., 12, 22, 74 Schaffer, M., 144, 145, 146, 152, 153, 163, I66 Schieffelin, B., 67, 74 Schlesinger, I. M., 15, 74 Schmidt, M. J., 79, 117 Schneider, W., 158, 166 Schwartz, R., 9, 74 Scott, J . L., 269, 270, 311 Seegmiller, B. R., 223, 238 Seitz, V., 225, 239 Sekuler, R. W., 83, 84, 106, 114, 122, 268, 269, 273, 274, 285, 308, 311 Selman, R., 67, 74 Serpell, R., 84, 114, 122 Shankweiler, D. P., 113, 114, 119, 122 Shantz, C. U., 214, 238 Shapiro, M., 114, 122 Shatz, M., 35, 49, 52, 74 Shepard, R. N., 81, 122, 156, 165, 268, 311 Shepherd, P. A., 86, 87, 102, 117 Sherman, M., 174, 210 Sherrod, L. R., 111, 122, 214, 238 Sherwood, V., 12, 70 Shettleworth, S . J., 9, 75 Shiffrin, R. M., 158, 166 Shoben, E. J., 160, 161, 166 Shore, C. S . , 14, 75 Sidman, M., 84, 122 Siegfried, J. B., 95, 96, 121 Siegler, R. S., 243, 244, 245, 269, 270, 287, 301, 311 Sigman, M., 33, 72, 225, 237 Simon, C. W., 80, 92, 118 Simon, H. A., 127, 166. 305, 311 Simpson, C., 166 Siqueland, E. R., 86, 88, 97, 117 Slater, A., 95, 122 Slobin, D. I., 9, 10, 12, 15, 75 Smedslund, J . , 287, 311 Smith, C., 139, 164 ' Smith, E. E., 160, 161, 166 Smith, J. H., 1 1 , 68, 72. 150, 166 Smith, K. H . , 286, 311

321

Author Index

Snow, C . , 35, 75 Snow, M. E., 225, 227, 234, 237 Snyder, L., 13, 51, 69, 70, 75 Spencer, D. D., 106, 119 Spitz, R. A,, 63, 75 Sroufe, L. A , , 19, 20, 73, 177. 178, 210, 2 1 1 , 226, 239 Staller, J., 114, 122 Standing, L., 144, 166 Starkey, P., 287, 288, 309 Stein, N. L., 84, 122 Steiner, J. E., 175, 185, 204, 210, 211 Stem, D. N., 22, 75 Sternberg, R. J . , 286, 311 Stevenson, M. B., 214, 217, 218, 219, 224, 225, 226, 227, 234, 239 Steward, M. S . , 113, 116 Stoever, R. I., 83, 106, I21 Storandt, M . , 106, 122 Strauss, M. S., 96, 117 Strauss, S., 244, 312 Stultz, K.,79, 119 Sutherland, N. S.,82, 84, 122 Suwalsky, J. D., 19, 72 Svenson, O., 288, 289, 290, 292, 312 Sykes, M.,95, 122 Szilagyi, P. G., 80, 122

T Tanne, G., 80, 119 Taylor, M. M.,79, 122 Tee, K. S . , 82, 84, 100, 123 Templeton, W. B., 79, 81, 82, 112, 119 Templin, M . , 133, 166 Teuber, H. L., 83, 114, 122 Thibos, L. N., 95, 118 Thomas, J., 82, I15 Thompson, J . , 172, 211 Thompson, R. A,, 217, 219, 232, 232, 239 Thompson, S . , 33, 66, 69, 73 Tieman, S. B., 106, I I Y Tighe, L. S., 150, 166 Tighe. T. J., 150, 166 Timney, B. N . , 79, 96, 123 Tinbergen, N., 9, 75 Tolman, E. C., 9, 75 Tomkins, S. S., 170, 171, 206, 211 Trabasso, T., 243, 275, 286, 312

Trevarthen, C., 12, 15, 75 Tulkin, S., 24, 25, 26, 28, 29, 30, 31, 32, 51, 75 Turiel, E., 15, 75 Turkewitz, G., 106, 118. 123

U Uhlarik, J., 80, 81, 119 Umeh, B. J., 225, 227, 234, 237 Urwin, C., 63, 75 Uzgins, I. C . , 224, 239

V Valian, V., 68, 75 Valsiner, J., 172, 209 van Dijk, T. A , , 68, 75, 300, 310 Vellutino, F. R . , 114, 123 Vietze, P., 175, 205, 210 Volterra, V., 12, 13, 16, 17, 18, 19,20,21,65, 69, 70 Vygotsky, L. S., 16, 75, 131, 166

W Wada, J. A., 106, 117. 123 Wade, T. D., 176, 209 Walden, T . , 193, 209. 211 Wall, S., 17, 69, 234, 237 Wallace, J . G., 306, 310 Waters, E., 17, 69. 178, 211, 225, 234, 237, 239 Watkins, D. W., 79, 115 Watson, J . S . , 95, 103, 104, 110, 123 Weiner, K . , 87, 123 Weiner, M., 84, 100, 123 Weinstein, M., 80, 118 Weintraub, S . , 35, 58, 69 Weiskopf, S . , 98, 116 Weiss, S . J . , 104, 116, 117 Werner, H . , 16, 75, 167 Wertheimer, M., 112, 123 Wexler, K., 8, 75 Whitacre, J., 250, 311 White, S. H., 150, 167 Wiesel, T. N . , 82, 95, 119 Williams, C., 35, 75 Williams, E. N., 268, 308

322

Author Index

Williamson, A. M., 82, 123 Wilson, E. G., 286, 312 Wilson, R. S.,216, 217, 223, 233, 237. 238 Winston, P. H . , 155, 158, 167 Wish, M., 273, 310 Wittgenstein, L., 167 Wohlwill, J . F., 84, 100, 123 Wolf, J. Z . , 86, 87, 88, 89, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 110, 116 Wolff, P . , 83, 84, 106, 123, 174, 189, 211 Wood, D.,5 1 , 75 Woodruff, G . , 68, 74 Woods, B . , 180, 209 Woodson, R . , 185, 204, 205, 208, 2UY Wu, P. Y. K . , 186, 209 Wunsch, J . P . , 177, 210

Y Yamada, J . , 15, 71 Yarczower, M . , 181, 205, 211 Yin, R . K . , 105, 123 Yonas, A., 79, 112, 121 Yoshida, S . , 96, 123 Yoshimura, T . , 289, 312 Young-Browne, G . , 176, 186, 211 2

Zalik, M. C., 82, 117 Zeki, S . M . , 106, 123 Zigler, E . , 225, 239 Zuckerman, M . , 172, 211 Zukow, P . , 57, 75

SUBJECT INDEX

A

organization of concept representations and, 160- 16 1 Counting from a point beyond one, 256-259 models of three levels of expertise in, 259267 from one, 250-256 Cue set representations, 140-142 evidence that children use. 142-143

Addition existing research on children, 287-290 model of strategy choice in, 296-299 preschoolers’ strategies for, 290-296 Attachment. language development and, 17-18

C D Causes, internal vs external, language development and, 48-54 Children, facial expressions and physiological responsivity and, 183-184 production or encoding of, I8 1-1 83, 191 -204 recognition or decoding of, 179-181, 191204 Cognitive development, perceptual anisotropies and, 1 1 1-1 I 5 Cognitive inputs, to language, 12-15 Cognitive performance, sociability and, see Sociability Computer analogy, concept development and, 127-129 Concept development, 125-127 contents of concept representations and, 130 exemplar-based, 143- I52 multiple interpretations of data and, 152153 non-rule-based, 134-143 rule-based, 130- 134 format of concept representations and, 153I55 images and descriptions, 155-158 representational-development hypothesis and, 158-160 information-processing theories of, 127 computer analogy and, 127-129 mental representation and, 129- 130

Decoding, of facial expressions in adults, 172-174 in children, 179-181, 191-204 in infants, 175-176 Descriptions, as concept representations, 155I56 Discrimination, of facial expressions, in infants, 175-176

E Encoding, of facial expressions in adults, 172-174 in children, 181-183, 191-204 in infants, 189-191 Environment, sociability and, 233-235 Equivalence, perceptual, 82-84 in infancy, 97-108 Exemplar-based concept representations, 143150

evidence that children use, 150-152 Expressions, see Facial expressions

F Face, perceptual anisotropy and, 100-103, 104-105, I I I

323

324

Subject Index

Facial expressions, 170-171 adult, I71 cataloguing, 171-172 relationships between encoding-decoding and physiological measures, 172-1 74 children and production or encoding by, 181 - I83 recognition or decoding by, 179-181 relationships between encoding and physiological responsivity in, 183-184 infants and discrimination or decoding by, 175-176, 185- 189 neonatal, 174-175 production by, 176-177, 185-189 relationships between expressions and physiological responsivity of, 178- 179

G Genetic confounds, language development and, 59-61

social bases of, 8 attachment and, 17-18 historical aspects, 8-12 internal vs external causes and, 49-50 preverbal interaction and, 18-34 structure vs motivation and, 50-54 verbal interaction and, 35-48 threshold effects and, 6 1-64 cognitive inputs and, 12-15 social inputs and, 15-34

M Magnitudes, see Numerical understandings Measurement, of sociability, 2 15-222 Mental representation, concept development and, 129-130 Mirror images, perceptual equivalence and, 97-108 Motivation, structure vs. language development and, 50-54

N I Images, as concept representations, 155-156 Individual differences, in sociability biogenetic influences on, 23 1 -233 environmental influences on, 233-235 Infants, facial expressions and discrimination or decoding of, 175-176, 189-191 encoding of, 189-191 in neonates, 174-175 physiological responsivity and, 178- 179 production of, 176-177 Information processing, concept development and, 127 computer analogy and, 127-129 mental representation and, 129-130 Interaction preverbal, language development and, 18-34 verbal, language development and, 35-48

L

Non-rule-based concept representations, 134143 Numerical understandings, 242-244 addition and existing research on children, 287-290 model of strategy choice in, 296-299 preschoolers’ strategies for, 290-296 counting and from a point beyond one, 256-259 models of three levels of expertise in, 259-267 from one, 250-256 development of, 299-308 number conservation and, 244-250 numerical magnitudes and effects of teaching a labeling strategy on, 278-280 preschoolers’ comparisons, 271 -275 research on adults and children, 267-271 models of magnitude comparison, 280-287 verbal labeling of numbers and, 275-278

P Labeling, numerical magnitudes and, 275-280 Language development direction of effects and, 54-59 genetic confounds and, 59-61

Perceptual anisotropies, 77-79 equivalence, 82-84 in infancy, 97-108

Subject Index

implications for cognitive development, 11 1-115 interrelations between, 109-1 11 perceptual-cognitive-social development and, 111 salience, 79-82 in infancy, 85-97 Physiological responsivity , facial expressions and in adults, 172-174 in children, 183-184 in infants, 178-179 Preference, perceptual salience and, 92-95 Preverbal interaction, language development and, 18-34 Processing, perceptual salience and, 90-92 Prototypes, as concept representations, 135138 evidence that children use, 138-140

325

Sociability, 213-215 cognitive performance and in older children, 225-226 relationship between, 226-230 outside test situation, 224-225 test sociability and, 222-224 individual differences in, 230-231 biogenetic influences on, 23 1-233 environmental influences on, 233-235 measuring, 215-222 Social-causal theories, of language develop ment, 48-49 internal vs external causes and, 49-50 structure vs motivation and, 50-54 Spatial vision, see Perceptual anisotropies Strategies addition, 290-2% model of strategy choice in, 296-299 labeling, 278-280 Structure, motivation vs, language development and, 50-54

R

T Representational-development hypothesis, concept representations and, 158-160 Rule-based concept representations, 130- 134

S Salience, perceptual, 79-82 in infancy, 85-97

Test sociability, cognitive performance and, 222-224

V Verbal interaction, language development and, 35-48 Vision, spatial, see Perceptual anisotropies

This Page Intentionally Left Blank

Contents of Previous Volumes Volumc 1 Responses of Infants and Children to Complex and Novel Stimulation Gordon N. Cantor Word Associations and Children's Verbal Behavior David S . Palermo Change in the Stature and Body Weight of North American Boys during the Last 80 Years Howard V . Meredith Discrimination Learning Set in Children Hayne W. Reese Learning in the First Year of Life Lewis P. Lipsirt Some Methodological Contributions from a Functional Analysis of Child Development Sidney W . Bijou and Donald M . Baer The Hypothesisof Stimulus Interaction and an Explanation of Stimulus Compounding Charles C . Spiker The Development of "Overconstancy" in Space Perception Joachim F . Wohlwill Miniature Experiments in the Discrimination Learning of Retardates Betty 1. House and David Zeaman AUTHOR INDEX-SUBJECT

INDEX

Selected Anatomic Variables A d y d for lnterage Relationships of the Size-Size, Size-Gain, and Gain-Gain Varieties Howard V . Meredirh AUTHOR INDEXSUBJECT INDEX

Volume 3 Infant Sucking Behavior and Its Modification Herbert K a y The Study of Brain Electrical Activity in Infants Robert J . Ellingson Selective Auditory Attention in Children Eleanor E. Uaccobv Stimulus Definition and Choice Michael D . Zeiler Experimental Analysis of Inferential Behavior in Children Tracy S. Kendler and Howard H. Kendler Perceptual Integration in Children Herberr L. Pick, Jr.. Anne D . Pick, and Robert E. Klein Component Process Latencies in Reaction Times of Children and Adults Raymond H. Hohle AUTHOR INDEX-SUBJECT

INDEX

Volume 2 Volume 4 The Paired-Associates Method in the Study of Conflict Aljred Castaneda Transfer of Stimulus Retraining to Motor PairedAssociate and Discrimination Learning Tasks Joan H. Conror The Role of the Distance Receptors in the Development of Social Responsiveness Richard H. Walrers and Ross D . Parke Social Reinforcement of Children's Behavior Harold W . Stevenson Delayed Reinforcement Effects Glenn Terrell A Developmental Approach to Learning and Cognition Eugene S. Collin Evidence for a Hierarchical Arrangement of Learning Rocesses Sheldon H. White

321

Developmental Studies of Figurative Perception David Elkind The Relations of Short-Term Memory to Development and Intelligence John M. Belmont and Earl C . Butterfield Learning, Developmental Research. and Individual Differences Frances Degen Horowitz Psychophysiological Studies in Newborn Infants S. J . Hurt, H.G . Lennrd, and H. F . R. Prechll Development of the Sensory Analyzers during Infancy Yvonne Brackbill and Hiram E. Firzgerald The h b 1 e m of Imitation Justin Aronfreed AUTHOR INDEX-SUBJECT

INDEX

328

Contents of Previous Volumes

vohnnc 5

Volame 8

The Development of Human F e d Activity and Its Relation to Postnatal Behavior T?ypheM Humphrey Arousal Systems and Infant Heut Rate Responses Frances K. Graham and Jan C. Jackson Specific and Diversive Exploration Corinne Hut2 Developmental Studies of Mediated Memory John H . Flavell Development and Choice Behavior in Probabilistic and ProbIem-Solving Tasks L. R. Goukt and Kathryn S. Goodwin

Elaboration and Learning in Childhood and Adolescence Williarn D.Rohwer, Jr. Exploratory Behavior and Human Development Jum C. Nunnally and L. Charles Lomnd Operant Conditioning of Infant Behavior: A Review Robert C. Hulsebus Birth Order and h n t a l Experience in Monkeys and Man G. Mitchell and L. Schroers Fear of the Snanger: A Critical Exmination Harriet L. Rheingold and Carol 0.Eckennan Applications of HuU-Spence Theory to the Transfer of Discrimination W i n g in Children Charles C . Spiker and Joan H . Cantor

AUTHOR INDEX-SUBJECT INDEX Volume 6 Incentives and Learning in Children Sam L. Wit@ Habituation in the Human Infant Wendell E. Jefley and Leslie 8. Cohen Application of Hull-Spence Theory to the Discrimination Learning of Children Charles C. Spiker Growth in Body Size: A Compendium of Findings on Contemporary Children Living in Different Parts of the World H o w d V. Meredith Imitation and Language Development James A. Sherman Conditional Responding as a Paradigm for Observational. Imitative Learning and Vicarious-Reinforcement Jacob L. Gewinz AUTHOR INDEX-SUBIECT INDEX Volumc 7

Superstitious Behavior in Children: An Experimental Analysis Michael D.Zeiler Learning Strategies in Children from Different SocioceonomicLevels Jean L. Bresnahan and Martin M. Shapiro Time and Change in the Development of the Individual and society Klaus F . Riegel The Nature and Development of Early Number Con-

=F

Rock1 Gelman Laming and Adaptation in Infawy: A Comparison of MOdClS Arnold J . Sameroff

AUTHOR INDEX-SUBJECT INDEX

AUTHOR INDEX-SUBJECT INDEX

Volume 9 Children’s Discrimination k i n g Based on Identity or Difference Betty 1. House, Ann L.Brown, and Marcia S.Scan Two Aspects of Experience in Ontogeny: Development and Learning Ham G . Furth The Effccts of Contextual Changes and Deof Component Mastery on Transfer of Training Joseph C. Campiane and Ann L. Brown Psychophysiological Functioning, AmusaI. Attention. and Learning during the First Year of Life Richard Hirschman and Edward S.Katkin Self-ReinforcementProcesses in Children John C. Masters and Janice R. Mokros AUTHOR INDEX-SUBJECT INDEX

Volume 10 Current Trends in Developmental Psychology Boyd R. McCandless and Mary Fulcher Geis The Development of Spatial Representationsof LargeScale Environments Alexander W.Siege1 and Sheldon H . White Cognitive Perspectives on the Development of Memory John W. Hagen. Robert H . Jongeward. Jr.. and Robert V. Kail, Jr. The Development of Memory: Knowing, Knowing About Knowing, and Knowing How to Know Ann L. Brown Developmental Trends in Visual Scanning Mary Carol Day

Contents of Previous Volumes The Development of Selective Attention: From Perceptual Exploration to Logical Search John C. Wright and Alice G . Vlietstra

AUTHOR INDEX-SUBJECT INDEX

Vdumr 11 The Hyperactive Child: Characteristics,Treatment, and Evaluation of Research Design Gladys B. Baxley and Judith M. LeBlanc Peripheral and Newchemical Parallels of Psychopathology: A Psychophysiological Model Relating Autonomic Imbalance to Hypcnaivity, Psychopathy. and Autism Stephen W. Porges

Constructing Cognitive Operations Linguistically Harry Beilin

Operant Acquisition of Social Behaviors in Infancy: Basic Problems and Constraints W. Stuart Millar

Mother-Infant Interaction and Its Study Jacob L. Gewirtz and Elizabeth F. Boyd

Symposium on Implications of Life-Span Developmend Psychology for Child Development: Introductory Remarks Paul B. Boltes

Theory and Method in Life-Span Developmental Psychology: Implications for Child Development Aletha Huston-Stein and Paul B. B a l m The Development of Memory: Life-Span Perspectives Hayne W. Reese

Cognitive Changes during the Adult Years: Implications for Developmental Theory and Research Nancy W. Denney and John C. Wright

Social Cognition pnd Life-Span Approaches to the Study of Child Development Michael 1.Chandler Life-Span Development of the Theory of Oneself: Implications for Child Development Orville G.Brim, Jr.

Implications of Life-Span Developmental Psychology for Childhood Education Leo Montada and Sigrun-Heide Filipp

AUTHOR INDEX-SUBJECT INDEX Volume I2 Research between 1%0 and 1970 on the Standing Height of Young Children in Different Pons of the World Howard V. Meredith

The Representation of Children's Knowledge David KIahr and Robert S.Siegler

Chromatic Vision in Infancy Marc H . Bornstein

329

Developmental Memory Theories: Baldwin and Piaget Bruce M. Ross and Stephen M . Kerst Child Discipline and the husuit of Self An Historical Interpretation Howard Gadlin

Development of Time Concepts in Children William 1.Friedman AUTHOR INDEX-SUBJECT INDEX Volume 13 Coding of Spatial and Temporal Information in Episodic Memory Daniel 6 . Berch A Developmental Model of Human Leanzing Barry Gholson and Harry Beilin The Development of Discrimination l a m i n g : A Levels-of-Functioning Explanation Tracy S.Kendler

The Kendler Levels-of-Functioning Theory: Comments and an Alternative Schema Charles C. Spiker and Joan H. Cantor Commentary on Kendler's Paper: An Alternative Perspective Barry Gholson and Therese Schuepfer Reply to Commentaries Tracy S. Kendler On the DeveIopment of Speech Perception: Mechanisms and Andogjes Peter D.Eimas and Vivien C. Tanter

The Economics of Infancy: A Review of Conjugate Reinforcement Carolyn Kent Rovee-Collier and Marcy J. Gekoski

Human Facial Expressions in Response to Taste and Smell Stimulation Jacob E . Steiner AUTHOR INDEXSUBJECr INDEX Volume 14 Development of Visual Memory in Infants John S. Werner and Marion Perhurter Sibship-Constellation Effccts on Psychosocial Development, Creativity, and Health Mazie Eark Wagner, Herman J. P . Schubert. and Daniel

S.P. Schuben

The Development of Understanding of the Spatial Terms Front and Back Louren Julius Harris and Elkn A. Strommen Thc Organization and Contml of Infant Sucking C. K . Crook Neurological Plasticity, Recovery from Brain Insult, and Child Development Ian Sf. James-Roberts

AUTHOR INDEXAUBJECT INDEX

330

Contents of Previous Volumes

Volume IS

Validating Theories of Intelligence Earl C . Burrofield, Dennis Siladi, and John M. Belmonf

Visual Development in Ontogenesis: Some Reeva,uations Juri AIIik and Joan Valsiner Binocular Vision in Infants: A Review and a Theoretical Framework Rirhard N . A s h and Susan T.Dumais

Cognitive Differentiation and Developmental Learning William Fowier Children's Clinical Syndromes and Generalized ExpeeFred Rorhbaum AUTHOR INDEXSUBJECT INDEX