Infant Perception and Cognition: Recent Advances, Emerging Theories, and Future Directions 9780195366709, 0195366700

"Lisa M. Oakes is Professor of Psychology and Faculty Researcher at the Center for Mind and Brain, University of Ca

291 41 122MB

English Pages 296 [324] Year 2011

Report DMCA / Copyright

DOWNLOAD PDF FILE

Recommend Papers

Infant Perception and Cognition: Recent Advances, Emerging Theories, and Future Directions
 9780195366709, 0195366700

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Infant Perception and Cognition

Infant Perception and Cognition Recent Advances, Emerging Theories, and Future Directions

Edited by LISA M. OAKES CARA

H. CASHON

MARIANELLA

CASASOLA

DAVID H. RAKISON

OXFORD UNIVERSITY

2011

PRESS

OXFORD UNIVERSITY

PRESS

Oxford University Press, Inc., publishes works that further Oxford University’s objective of excellence in research, scholarship, and education. Oxford New York Auckland Cape Town DaresSalaam HongKong KualaLumpur Madrid Melbourne Mexico City NewDelhi Shanghai Taipei Toronto

Karachi Nairobi

With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam

Copyright © 2011 by Lisa M, Oakes, Cara H. Cashon, Marianella Casasola, David H. Rakison Published by Oxford University Press, Inc. 198 Madison Avenue, New York, New York 10016 www.oup.com Oxford is a registered trademark of Oxford University Press All rights reserved, No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Infant perception and cognition : recent advances, emerging theories, and future directions / edited by Lisa M. Oakes... [et al.], — Ist ed.

p. cm. ISBN 978-0-19-536670-9

1. Perception in infants. BF720,P47154 2010 155,42'23—de22 2010002569

2, Cognition in infants.

1. Oakes, Lisa M., 1963-

This book is dedicated to Les Cohen, mentor and teacher

Preface

‘This volume was inspired by a conference we organized to honor the contribution of Leslie B. Cohen to the field of infant cognition, and it highlights recent advances in our understanding of cognitive and perceptual development in infancy. In organizing, attending, and participating in the conference, held preceding the biennial meeting of the International Conference on Infant Studies in Vancouver in March 2008, we were struck by how modern conceptions and models of many different aspects of the infants’ developing perceptual and conceptual systems derive from a long history of understanding the human information-processing system in general. Moreover, recent advances in these areas incorporate historical contributions gleaned from over 40 years of empirical work with infants, as well as recent advances in computational modeling and neuroscience. Despite the fact that there have been recent volumes focused on specific content areas in cognitive development (e.g., examining different approaches and findings related to categorization, memory, or spatial cognition), few volumes illustrate commonalities across content areas in our understanding of the early emergence and development of cognition. Thus, we planned this volume. At first glance, the contributors seem to have little in common—they have examined the early development of abilities ranging from “low-level” perceptual and attentional processes (e.g., Chapters 1, 2, 3) to “high-level” recognition of the symbolic nature of scale models (e.g., Chapter 12). However, as you read these chapters you will find a common thread in how these researchers think about and investigate these various abilities. Specifically, each of the research programs described here has been influenced to some degree by classic work on information-processing approaches to cognition. It has long been held that how infants perceive and think about the world provides the foundation for all of our cognition, Whether one argues that infants’ cognition is motivated by innate modules that guide perception, or that infants’ thinking about the world is constructed from lower-level processes, the clear consensus in the field is that human cognition builds from infancy. Importantly, it is also becoming increasingly clear that understanding infants’ cognitive and perceptual processes contributes to our understanding of human mental life in general. Recent work has shown that the same mental processes operate in infancy and adults, and the same models of cognitive and perceptual processes can be used (to one degree or another) to explain infant and adult cognition and perception. The contributors to this volume share a deep interest in understanding the human mind in general, as well as a desire

viii

Preface

to understand how cognitive and perceptual processes develop over the course of infancy. THE INFORMATION-PROCESSING LEGACY IN INFANT COGNITION AND PERCEPTION A revolution happened in the field of cognitive psychology with the advent of the information-processing approach to cognition. This revolution also influenced the study of infant perception and cognition, and most (if not all) modern studies of infant cognition are built on a foundation of thinking about the information-processing infant. The information-processing movement made two main contributions to the study of infant perception and cognition. Information-processing theories of infant cognition have shaped many of the questions asked in the field. For example, Cohen's (1973) two-process model of infant attention inspired others to examine components of visual attention (Richards & Casey, 1992), as well as other aspects of attention (for a review

see Colombo, 2001), and to look for measures of those different aspects of attention (see Chapters 1 and 2). Work by Aslin and his colleagues on infants’ detection of statistical regularities (Chapter 6; Saffran, Aslin, & Newport, 1996) and by Younger and Cohen (1986) on infants’ recognition of correlations of features, has contributed to a focus on how infants can extract structure from the input simply by detecting regularities or correlations that exist in that input. Such considerations derive from a view that the infant takes in information from the environment, and processes that information (attends

to it, perceives it, encodes it, remembers it, recalls it, categorizes it, and so on). Cohen’s information-processing model of how, with development, the system deals with larger and larger units or chunks of information (Cohen, 1991), underlies modern day theories that examine the units of information infants perceive, attend to, and remember, and how those units are combined, stored, etc,, by the system (see Chapters 3, 4, 8, 9). The second contribution of the information-processing approach to the study of infant perception and cognition was methodological. The early work on visual habituation was inextricably linked to the study of the infant information-processing system. Cohen (1972) proposed attention-holding and attention-getting processes, and he developed visual habituation tasks to examine them separately. These studies provide the foundation for the looking time techniques used by most researchers studying infant perception and cognition. Horowitz et al. (1972) developed a similar procedure in the context of studying individual differences in how information was processed, The methods developed in this tradition have become the primary tool for understanding everything from newborn categorization of simple shapes, to toddlers’ learning about labels for spatial relations. Although these methodological contributions were critical for opening new avenues for assessing infants’ perceptual and cognitive abilities, aspects of these tools informed theorizing about perceptual and cognitive development itself. For example, as described earlier, Cohen's differentiation of

Preface

ix

attention-holding and attention-getting processes shaped others’ views of infants’ attention (see Chapters 1 and 2). In addition, in adapting this method

to studying infants’ sensitivity to the correlations among features, or the statistical regularities that characterize the habituation set, Cohen and his colleagues (Cohen & Younger, 1984; Younger & Cohen, 1986) introduced what has become widely known as the “switch design.” This standardized technique allows researchers to examine infants’ developing abilities in recognizing how features are combined or bound in stimuli, and has been applied to examine infants’ perception of static features of stationary images, to dynamic features of moving objects, and to cross-modal links between words and objects (see Chapters 7, 8, 9, and 10). Finally, the information-processing approach provided the field an alternative way of considering the constructive nature of infant development. Rather than focusing on the constructive processes introduced by Piaget, theorists and researchers could consider how the information-processing system constructs an understanding of the world by focusing on larger “chunks” of information, or combining information in new ways. This constructive approach has influenced thinking about infants’ emerging perception of objects and faces (Chapters 3 and 4), understanding of number, space, and function (Chapters 5, 8, and 9) and their categorization and language abilities (Chapters 7, 10, and 11). Moreover, this process has been used as a foundation for understanding

cognition at the highest level, including toddlers’ emerging understanding of the nature of symbols (Chapter 12).

The chapters in this volume show how each of these contributions of the information-processing approach have been critical for our understanding in one or more aspect of infants’ developing perceptual and cognitive abilities. Moreover, these chapters illustrate how the field has built upon this foundation and enriched our understanding of the processes of early perceptual and cognitive development by incorporating findings from neuroscience, cognitive neuroscience methods, and computational modeling. For example, the work of Richards (Chapter 2) and Colombo and colleagues (Chapter 1) has

incorporated neuroscience findings and methods into the study of visual attention. Plunkett (Chapter 10) and Shultz (Chapter 7) illustrate how computational modeling can enrich our understanding of behavioral development in infancy. ORGANIZATION OF THIS VOLUME This volume is unique in that it brings together a collection of developmental scientists who exemplify this approach to infant cognition and perception. Each chapter shows how modern understanding and theories of some aspect of cognitive or perceptual development in infancy is based on historical models of the human information-processing system (either in adults or infants), but has evolved beyond these roots to incorporate modern theories, and findings from infants, adults, and neuroscience. Thus, these chapters do not show the full range of theoretical approaches to one aspect of infant cognition.

x

Preface

Rather, they provide an extensive overview of theories of cognition and perception that complexly integrate roots in our understanding of the human information-processing system, behavioral changes in infant cognition and perception, and cognitive neuroscience,

The chapters in this volume are presented in an order that loosely corresponds to “lower-level” cognitive processes, such as visual attention, face perception, and so on, and then moves in later chapters toward “higher-level” cognitive processes, such as linking labels to categories, appreciating the correspondence between symbols and their referents, and so on. Each contributor provides an overview of their research in this area, with an emphasis on how their current theory, model, or understanding has evolved. Each chapter stands alone and provides an excellent overview ofa particular content area. However, by reading several of the chapters in this volume, the reader can gain an appreciation for the commonalities in theories, methods, and approaches across content areas, These chapters illustrate how the same general approach can be used to understand a diverse set of developing abilities and link developmental changes across domains. We believe this is the real value of this volume, ACKNOWLEDGEMENTS

This volume emerged from a preconference to the biennial meeting of the International Society for Infant Studies we held in March 2008. Many people were instrumental in the success of that meeting, and ultimately for the success of this volume. In particular, we thank Janet Werker and Ron Barr for their help and support in planning and executing the preconference. We also received generous support from the National Science Foundation (Grant BCS 0751237) and the University of Texas Psychology Department. Amy Sussman of NSF and Jamie Pennebaker of UT Austin were extremely helpful in helping us with the funding needed for this project. Miye Cohen was an invaluable source of information and guidance through this project. Finally, we thank our contributors to this volume for their excellent contributions and their patience as we asked them to refine and rework aspects of their chapters. We think you will agree that the resulting product illustrates the best work in the field of infant cognitive development. REFERENCES Cohen, L. B. (1972). Attention-getting and attention-holding processes of infant visual preferences, Child Development, 43, 869-879. Cohen, L. B. (1973). A two-process model of infant attention. Merrill-Palumer Quarterly,

19, 157-180.

Cohen, L. B. (1991). Infant attention: An information processing approach. In M. J. Weiss & P. R. Zelazo (Eds.), Newborn attention: Biological constraints and

the influence of experience (pp. 1-21), Norwood, NJ: Ablex.

Preface

xi

Cohen, L. B., & Younger, B. A. (1984). Infant perception of angular relations. Infant Behavior & Development, 7, 37-47. Colombo, J. (2001). The development of visual attention in infancy. Annual Review of Psychology, 52, 337-367. Horowitz, F. D., Paden, L., Bhana, K., & Self, P. (1972). An infant-control procedure for studying infant visual fixations. Developmental Psychology, 7, 90. Richards, J. E., & Casey, B. J. (1992). Development of sustained visual attention in the

human infant. In B. A. Campbell, H. Hayne & R. Richardson (Eds.), Attention and information processing in infants and adults: Perspectives from human and animal research (pp. 30-20). Hillsdale, NJ; Erlbaum. Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-monthold infants. Science, 274, 1926-1928. Younger, B. A., & Cohen, L. B. (1986). Developmental changes in infants’ perception of correlations among attributes. Child Development, 57, 803-815.

Contents

Contributors

XV

Varieties of Attention in Infancy John Colombo, Leah Kapa, and Lori Curtindale Infant Attention, Arousal, and the Brain

27

John E. Richards

A Constructivist View of Object Perception in Infancy

51

Scott P. Johnson

Development of Specialized Face Perception in Infants: An Information-Processing Perspective Cara H. Cashon

The Role of Perceptual Processes in Infant Addition/ Subtraction Experiments

69

85

Alan M. Slater, J. Gavin Bremner, Scott P. Johnson, and Rachel A. Hayes

Perceptual Constraints on Implicit Memory for Visual Features: Statistical Learning in Human Infants

111

Richard N. Aslin

Computational Modeling of Infant Concept Learning: The Developmental Shift from Features to Correlations

125

Thomas R, Shultz

Information-Processing Approaches to Infants’ Developing Representation of Dynamic Features Kelly L. Madole, Lisa M. Oakes, and David H. Rakison

Infant Spatial Categorization from an Information Processing Approach

153

179

Marianella Casasola

10.

The Role of Auditory Stimuli in Infant Categorization

203

Kim Plunkett 11.

The Development of Categorization and Facial Knowledge: Implications for the Study of Autism Lisa C. Newell, Catherine A, Best, Holly Gastgeb, Keiran M. Rump, and Mark S. Strauss

223

xiv

12.

Contents

Emerging Competence with Symbolic Artifacts: Implications for the Study of Categorization and Concept Development

261

Barbara A. Younger and Kathy E. Johnson

Author Index

285

Subject Index

293

Contributors

Richard N. Aslin, University of Rochester Catherine A. Best, University of Pittsburgh J. Gavin Bremner, Lancaster University

Marianella Casasola, Cornell University

Cara H. Cashon, University of Louisville John Colombo, The University of Kansas Lori Curtindale, The University of Kansas Holly Gastgeb, University of Pittsburgh Rachel A. Hayes, University of Exeter Kathy E. Johnson, Indiana University-Purdue

University Indianapolis Scott P. Johnson, University of California

Los Angeles, CA Leah Kapa, The University of Kansas Kelly L. Madole, Western Kentucky University Lisa C. Newell, Indiana University of Pennsylvania Lisa M. Oakes, University of California, Davis Kim Plunkett, University of Oxford David H. Rakison, Carnegie Mellon University Keiran M. Rump, University of Pittsburgh John E, Richards, University of South Carolina

xvi Thomas R. Shultz, McGill University Alan M. Slater, University of Exeter Mark S, Strauss, University of Pittsburgh Barbara A. Younger, Purdue University

Contributors

Infant Perception and Cognition

Varieties of Attention in Infancy John Colombo, Leah Kapa, and Lori Curtindale

INTRODUCTION Attention is a fundamental component in the information processing system (Neisser, 1967; Posner & Rothbart, 2007). Within the literature of cognitive

psychology, the function of attention is generally characterized as a process (or processes) that involve the selection of a subset of stimuli from the many concurrently available to an organism for processing or action (e.g., James, 1890). For the first century of research on this topic in the study of adult cognition, researchers have been preoccupied with two questions. The first concerns the nature of the selection process, and asks whether the selection of one stimulus implies the absolute or relative exclusion of others (e.g., Broadbent, 1958; Sperling, 1960). The second concerns exactly where in the information processing system this selection occurs (Broadbent, 1982; Deutsch & Deutsch, 1963; Treisman, 1969).

Multicomponent Models of Attention With the emergence of cognitive neuroscience in the 1980s and 1990s, the topic of attention gained renewed emphasis as researchers sought to examine the construct from the point of view of brain structure and function (e.g., Parasuraman, 1998; Posner, 2004; Richards, 1998). The notion that attention might be quantified or conceptualized in many ways can certainly be traced to the initial scientific consideration of the attention construct. William James

clearly had this in mind when he alluded to the existence of “varieties of attention” in the Principles of Psychology (1890), and Parasuraman (1984) used this phrase for the title of edited volume that explored the diversity of forms that this phenomenon could take. More than anything, however, the analysis from the cognitive neuroscience standpoint strongly reinforced the general idea that the term attention, at least as it is used in the modern parlance of the cognitive psychologist, actually reflects a construct rather than a single or unitary phenomenon. ‘The available evidence suggests that this construct is comprised of a finite set of independent and dissociable subcomponents, each with independent and largely dissociable neural substrates. To wit, a quick perusal of any of the comprehensive handbooks on cognitive neurosciences (e.g., Gazzaniga, 2004) will

4

Infant Perception and Cognition

reveal multiple chapters on attention (the most recent edition has eight), in which many different central nervous system (CNS) structures and pathways are implicated, Purpose of the Chapter It is our view that this perspective on attention is specifically relevant to considerations of development, particularly during the early part of the lifespan. This chapter has several purposes. First, we will briefly trace the history of the consideration of attention as a construct in the developmental literature; it has been evident for some time that a multicomponent conceptualization of this cognitive function is necessary. Second, using the development and evolution of our own research program here at Kansas over the past 25 years as a basis, we review the development of attention from the multicomponent perspective. Such an examination will require a discussion of the neural basis of these processes, and show the inexorable movement of the field toward cognitive neuroscience. We will end with a discussion of some future questions for the study of attention in infancy and early childhood. MULTICOMPONENT MODELS IN DEVELOPMENT The notion that attention might consist of multiple subcomponents has also long existed in the field of developmental psychology, and attempts to parse attention into subcomponents quickly followed the innovation of methods for studying infant attentional responses. Attention-Getting and Attention-Holding Among the first attempts to derive subcomponents of infant attention came from Cohen (1972, 1973; Cohen, DeLoache, & Rissman, 1975), who proposed that different measures of infant visual fixation might be used to identify different types of attention. Cohen (1972) specifically proposed that some stimuli might be better for attention-getting, while others might be better for attention-holding. Indeed, if we consider these as attentional processes, rather than as the particular effects of stimuli on the organism, these components readily map onto concepts that were subsequently investigated in the adult and developmental literature on attention.

For example, attention-getting is easily translated to the phenomena of visuospatial orienting (e.g., Posner, Inhoff, Friedrich, & Cohen, 1987) and attentional capture (Yantis, 1993; Yantis & Jonides, 1984). The topic of attentional capture within a developmental framework was elaborated elegantly in Dannemiller’s (1998, 2000, 2001) program of work on exogenous attention some decades later. Furthermore, attention-holding might be characterized as an aspect of sustained attention (Fisk & Schneider, 1981; Neuchterlein, Parasuraman, & Jiang, 1983). Although the distinction between these two processes dates back nearly four decades, it has been invoked with some regularity over that time (e.g., Finlay & Ivinskis, 1982; Foreman, Fielder, Price, & Bowler,

Varieties of Attention in Infancy

5

1991; Landry, Leslie, Fletcher, & Francis, 1985; Richard, Normandeau, Brun, & Maillet, 2004), and thus represents a relatively current notion. Attention as a “State” versus “Process”

Ruff and Rothbart’s (1996) seminal volume on the development of attention in infancy and toddlerhood made an important distinction between attention as a state versus attention as a process in development, and the general progression from one to another across age. This is predated by Ruff’s (1986a, 1986b, 1990) analysis of infants’ behavior in free-play object-manipulation sessions, which suggests that attention might be parsed into a short-term phase characterized by initial evaluations of novelty and familiarity, and a longer-term phase of sustained attention. This distinction has generated important studies extending the developmental investigation of attention from infants to toddlers and preschoolers (Kannass, Oakes, & Shaddy, 2006; Oakes, Tellinghuisen, & Tjebkes, 2000; Ruff & Capozzoli, 2003; Ruff & Lawson, 1990; Ruff, Lawson, Parinello, & Weissberg, 1990) and has initiated the investigation of the link between cognitive conceptualizations of attention and important constructs such as temperament and self-regulation (Posner & Rothbart, 2007; Rothbart,

Posner, & Kieras, 2006; Rueda et al., 2005). A DEVELOPMENTAL TAXONOMY OF ATTENTIONAL PROCESSES Recently we (Colombo,

2001a; 2002; Colombo

& Cheatham,

2006) have

proposed a taxonomy of attention in early development. This taxonomy of processes is comprised of at least four independent functions or processes, each mediated by distinct neural structures. Alertness/Arousal

This is perhaps the most fundamental of the processes that we characterize as attention. The background “state” of the organism prepares it for input, and likely initiates neural conditions that facilitate learning and retention. The available evidence suggests that, at least in early development, this function is mediated by various ascending brainstem pathways (e.g., Robbins & Everitt, 1995). This function develops relatively early, but is not necessarily mature at birth. Newborns are capable of alert states, but periods of alertness only account for approximately 20% of a newborn’s time (Colombo & Horowitz, 1987). Time spent in the alert/aroused states dramatically increases in an

infant's first 10-12 weeks. Early state organization is related to later cognitive processes in infancy (e.g., Moss, Colombo, Mitchell, & Horowitz, 1988). Visuospatial Orienting This process involves the disengagement, shifting/moving, and engagement of attention from one locus to another, and is largely mediated by Posner’s (Posner & Petersen, 1990) posterior attentional network. Posner's posterior attentional network is generally synonymous with what is sometimes referred

6

Infant Perception and Cognition

to as the “where” system of attention. The available evidence at this time (see Colombo, 2001a) suggests that functions related to orienting develop during the first 4-5 months of life. Object Perception This process, which may be limited to visual attention, includes the analysis, binding, and recognition of stimulus features, and likely reflects the mediation of extrastriate and temporal structures. This is generally synonymous with what is sometimes called the “what” system of attention. This component includes the detection of the boundaries between and patterns within objects, and a subsidiary aspect of this function relates to the processing of wholes and parts, which has a long tradition as a topic within the field of perceptual development. Again, there is significant development in this domain across the first six or seven months of life (Colombo, 2001a).

Endogenous Attention This complex function reflects the integrated and coordinated activation of various attentional components in the context of the content and status of memory systems (Colombo & Cheatham, 2006); this is presumably mediated by frontal lobe structures (e.g., Fernandez-Duque, Baird, & Posner, 2000; Rueda,

Posner,

&

Rothbart,

2005).

Endogenous

attention

represents voli-

tional attention-holding/inhibition of shifting attention, and emerges later in infancy than the other three forms of attention. There are likely precursors of its presence early in the first year, but the evidence suggests that this function emerges in the latter half of the first year and is predominant in the second and third years. DEVELOPMENTAL AND INDIVIDUAL DIFFERENCES IN INFANT ATTENTION: A MULTICOMPONENT APPROACH For the last 25 years, our laboratory at the University of Kansas has followed a path of investigation that has led us to delineate the developmental functions for the various components of attention during infancy, and to determine their contribution to infant cognition and learning (Colombo, 2001a, 2002, 2004; Colombo & Mitchell, 1990). This program of investigation has specifically focused on the development of visuospatial orienting and object perception, and the individual differences between infants that might account for dissimilarities on measures of these types of attention. Developmental Parameters in Visual Habituation Our program of work began with a simple study of infants’ visual habituation to faces during the first year. The study was conducted to map out the basic psychometric properties of measures taken from the infant-controlled visual habituation paradigm at 3, 4, 7, and 9 months of age (Cohen, 1973; Horowitz, Paden, Bhana, & Self, 1972). In this procedure, a stimulus is repeatedly

Varieties of Attention in Infancy

7

presented to the infant. The duration of infant looks to the stimulus declines over such presentations, presumably reflecting the infants’ learning/encoding of the stimulus. The psychometric properties were reported in a number of subsequent publications (Colombo, 1993; Colombo, Mitchell, O’Brien, & Horowitz, 1987a, 1987b). However, data on the developmental course of such

measures (Colombo & Mitchell, 1990; see also a contemporaneous report by Mayes & Kessen, 1989) provided important clues about the development of attention.

Our initial hypotheses were based on the comparator theory derived from

Sokoloy’s (1963) model

of the formation of internal representations.

According to Sokolov, the strength of the orienting response in habituation (here, this is translated to the length of the infant's look) is thought to reflect

the degree of mismatch between the external stimulus and the habituating organism’s internal representation of it. As such, when organisms learn

more rapidly they should show steeper habituation curves and attain habituation in shorter sequences. Given the expectation that older infants would learn faster and more efficiently than younger infants, we predicted steeper curves and larger magnitudes of decline in the habituation functions of older infants. Instead, we observed exactly the opposite with both longitudinal and cross-sectional samples (Colombo & Mitchell, 1990): habituation slopes were shallower and decrements were smaller for older infants. This finding has been replicated numerous times with other samples (e.g., Colombo, Shaddy, Richman, Anderson, & Blaga, 2004) and with stimuli other than faces (Courage, Reynolds, & Richards, 2006; Shaddy & Colombo, 2004). The finding is explained quite simply by the fact that younger infants show longer initial durations of looking to stimuli than older infants, but decline to more compa-

a

Look Duration (sec)

rable levels when they attain the habituation criterion (see Figure 1.1).

10 5 o+

T 1

2

3

r cl

c2

Presentation Trial

Figure 1.1. Habituation curves for different-aged infants. The x-axis represents habituation trials 1 through 3; “cl” and “c2” represent the two criterion looks. Because these data come from infant-controlled habituation, varying numbers of looks may occur between look 3 and cl, although the average number of looks across all ages is between 6 and 7. These curves are culled from the KU Early Cognition Project database reported on in Colombo et al. (2004).

8

Infant Perception and Cognition

Based on this finding, we reasoned that if the most robust developmental differences in infant visual behavior were attributable to changes in look duration, then the decline in look duration with age must reflect a major change in stimulus encoding. If that were true, then it seemed likely that more rapid encoding was reflected by shorter look durations. We proceeded using Underwood’s (1975) model for using idiographic evidence to support and confirm nomothetic principles, and thus ventured into the domain of individual differences. We formulated a simple hypothesis: if shorter look durations reflected more rapid encoding across ages, it seemed reasonable that individual differences in look duration within ages should also reflect the differences in the efficiency or rapidity of encoding. It is worth noting that the limited psychometric data available at the time supported the notion that individual differences in look duration were relatively stable, with test-retest reliabilities within ages ranging from .40 to .50 (Colombo et al., 1987a, 1987b; the psychometrics are also reviewed in Colombo, 1993). This spurred us on to compare the performance of infants who looked longer versus those who looked more briefly (long-looking and shortlooking infants, respectively). If the hypothesis was correct, then infants who looked for briefer durations should show recognition of visual stimuli after shorter familiarizations than infants of the same age who looked for longer durations. To test these hypothesized differences between short- and long-looking infants, we used a standard procedure for characterizing infants based on look duration in all of the studies described below. In each case, look duration was measured using independent assessments of look duration prior to subsequent tasks, and infants were then divided into groups based on their peak fixation durations. Infants who displayed peak look durations that were greater than the group median were classified as long-looking infants, while those whose peak look duration was less than the group median were shortlookers. The first of these studies (Colombo, Mitchell, & Horowitz, 1988) supported our prediction that infant look duration relates to encoding speed, as did a number of additional studies that followed both from our own laboratory (e.g., Colombo, Mitchell, Coldren, & Freeseman, 1991) and from others (e.g., Courage & Howe, 2001; Jankowski & Rose, 1997; Jankowski, Rose, & Feldman, 2001; Rose, Futterweit, & Jankowski, 1999), which are discussed in some detail

later on. Thus, we felt somewhat confident that both the developmental and individual differences in attentional measures were attributable to individual differences in the rapidity or efficiency of encoding. However, we were also aware that rapidity and efficiency is merely a parameter of an information processing system rather than a process in and of itself, and as such, we did not have an explanation for the phenomenon beyond a default hypothesis about general maturation (Colombo & Mitchell, 1990) or CNS integrity (e.g., Axia, Bonichini, & Benini, 1999).

Varieties of Attention in Infancy

9

Look Duration and Attention to Object Features When individual differences in the rapidity of cognitive performance is observed, acommon assumption is that there should be some cost to improved speed in relation to accuracy in performance (i.e., the speed-accuracy tradeoff; see Briggs & Shinar, 1972; Pachella & Pew, 1968; Swanson & Briggs, 1969). Therefore, we began looking for differences in how infants who varied on look duration might also differ in their processing biases (the terminology here taken from signal detection theory), or in terms of the completeness of their stimulus processing. Discrimination Studies

Although we were unable to design a study that satisfactorily addressed the bias issue, we were able to design stimuli that could be characterized by different stimulus properties: “global” aspects of the stimulus (i.e., the overall arrangement or configuration of the stimulus) versus “local” features (i.e., smaller features that made up the overall shape of the stimulus). In the first attempt to test for this possibility (Colombo et al., 1991), we used one set of stimuli (dots placed in various arrangements) to represent a “global” discrimination, and another set of stimuli (alphabetic letters that varied in a single

feature) to represent a “featural” discrimination (see Figure 1.2). No tradeoff was observed in discrimination; the global task was solved after briefer levels of familiarization, and infants with shorter look durations required less familiarization than infants with longer look durations to perform

Featural Discrimination

Global Discrimination

Figure 1.2. Stimuli used for local (top) and global (bottom) discrimination tasks. Source: From “Individual differences in infant attention: Are short lookers faster processors or feature processors?” by J. Colombo, D. W. Mitchell, J. T. Coldren, and L. J. Freeseman, 1991, Child Development, 62, p. 1250. Copyright 1991 by WileyBlackwell. Reprinted with permission.

10

Infant Perception and Cognition

either type of discrimination, Despite this finding, we were concerned that the use of separate stimulus sets may have obscured possible tradeoff effects, and so we employed the strategy of using hierarchical stimuli (see Figure 1.3), where local elements (Ns or Zs) were arranged within a global configuration (hourglass or diamond), thus allowing either level to be manipulated. In the adult literature, attention to the global configuration will interfere with processing of the local features, but the reverse does not hold; researchers have inferred from this pattern of results that there is a “precedence” for processing the global configuration (e.g., Miller, 1981; Navon, 1977, 1983). Freeseman, Colombo, and Coldren (1993) featured a series of studies using discrimination and generalization tasks. Again, we observed no evidence for a speed-accuracy tradeoff: discrimination and generalization occurred after briefer familiarization times for infants with short-look durations than for infants with long-look durations. In other words, short-looking infants appeared simply to process stimuli faster than long-looking infants. In addition, a complete global-to-local processing sequence was demonstrated for infants with short-look durations, and we saw what we believed to be evidence for the start of such a sequence for infants with long-look durations. Thus, the first two sets of studies involving simple discrimination procedures showed no evidence for any tradeoff in the more rapid processing seen for infants with short-looking profiles. Global-Local Competition Previously, this topic had only been investigated using tasks based on discriminative capacities. However, it was well known that visual patterns could be successfully attained using various capacities or strategies, even under

Zz LLL ZLZ 22. ar as Zz

NNNN i

2222. eZ

NN NNNN

Ze PLE

Lee

N NN NNNNN NN N

Figure 1.3. Examples of hierarchical stimuli used in studies with human infants. Stimuli can be processed on both a global (configural) level (diamond vs. hourglass shape) and a local (featural) level (Ns vs. Zs).

Source: From “Individual differences in infant visual attention: Discrimination generalization of global local stimulus properties,” by L. J. Freeseman, J. Colombo, and J. T. Coldren, 1993, Child Development, 64, p. 1193. Copyright 1993 by Wiley-Blackwell. Reprinted with permission.

Varieties of Attention in Infancy

1

conditions where fundamental sensory functions were significantly impaired (e.g., Ganz, Hirsch, & Tieman, 1972). Thus, another approach was employed in which we tried to determine which visual properties of the stimuli were being processed by these two groups of infants under conditions where both properties were simultaneously available. This was done by placing the two properties in competition with one another for infants’ attention (Colombo, Freeseman, Coldren, & Frick, 1995). For example, it was possible to familiarize infants to a specific global configuration composed of particular local elements (e.g., a diamond composed of Zs). Then, on a subsequent pairedcomparison test we paired the familiar global configuration (diamond) composed of novel local elements (e.g., Ns) against a novel global configuration (e.g., an hourglass) composed of familiar local elements (Zs). Theoretically, if infants were attending to the global property, the novel configuration should attract more attention on the test trials (in spite of the presence of familiar local elements). However, if infants were attending to the local property, the novel local elements should be preferred (again, in spite of the presence of the familiar global configuration). Given that the property being attended to might vary as a function of the amount of study time that infants devoted to it, we manipulated familiarization time across experiments. Given the results of the discrimination studies, the outcome of the competition study was relatively striking: infants with shorter look durations showed preference for global attributes after briefer familiarization times, and preference for local features at longer familiarization times. Infants with longer look durations showed preferences for the local visual property, but only after considerably more familiarization. Additionally, infants with longer look durations never displayed a global preference, regardless of the length of familiarization. This finding was the first to suggest that the differences seen on simpler tasks as a function of individual profiles in look duration might be attributable to differences in infants’ manner of visual intake, with short-looking infants displaying global intake patterns and long-looking infants focusing on local elements. Indeed, older reports from eye movement and scanning studies (Leahy, 1976; Maurer & Salapatek, 1976; Salapatek, 1975) about changes in the

visual behavior in infancy suggested a progression from “focalized” (featural) fixations in early infancy to a broader dispersion of fixations (global) at older ages. Although the ages of infants in these scanning studies were somewhat younger than the ages at which we had seen individual differences in look duration, our findings fit with that general progression. Degraded Stimulus Studies Given this new finding, we quickly realized that hierarchical stimuli presented a number of potential interpretive problems when testing for discrimination of either local or global properties (see Figure 1.4). Specifically, if an infant who focused solely on local elements had encoded only the circled element, then it would still be possible for him to discriminate the second figure in a novel global arrangement, despite the identical local elements, because the local feature is no longer present in the same circled location.

12

Infant Perception and Cognition

Point of NNN Infant's —» N N NNN Fixation NNN

N

Point of Infant's Fixation

NNNNN NN N

NN

NNNNN Figure 1.4. An example of how discrimination of global configuration might be attained using attention only to local elements. An infant relying solely on local visual information might be able to discriminate a “global” stimulus based on the presence or absence of a local element in a particular spatial location in the visual field. Source: From “Individual differences in infant visual attention: Four-month-olds’ recognition of forms connoted by complementary contour,” by J. Colombo, J. E. Frick, J. 8. Ryther, and J. J. Gifford, 1996, Infant Behavior and Development, 19, p. 114. Copyright 1996 by Elsevier. Reprinted with permission,

As a result, we tried to address these issues and continue the program of work with a number of alternative stimulus designs, and eventually turned to the use of degraded stimuli. In the first, we examined the infants’ ability to recognize a degraded visual stimulus when none of the original contours were present (Colombo, Frick, Ryther, & Gifford, 1996). Here, infants were habitu-

ated to a particular shape in which half of the stimulus contours that defined its perimeter were missing. We then tested for recognition of that shape by pairing the same shape (composed of the half of the contours missing from the familiarization stimulus) against a degraded novel shape. This presumably forced infants to use global cues for recognizing the shape, since the familiarization and test representations of the same shape shared no local features (see Figure 1,5). Although infants with short look duration profiles readily generalized from a degraded form to its complement, infants with longer look duration profiles did not—strongly reinforcing our interpretation about the deployment of attention to different parts of the stimulus by the two groups. We continued this line of work with degraded stimuli (Frick & Colombo,

1996) and manipulated stimuli in keeping with critical theories of adult object perception. Biederman’s (1987) recognition-by-components theory of adult object perception held that visual scenes are reducible to a set of primary geometric shapes (geons), and that the junctions between these shapes are critical for object recognition. Stimuli were then manipulated to test infants’ recognition under conditions where the stimuli were degraded at either vertices or at non-vertex locations. The prediction was that infants who looked for longer durations were likely focused on the vertices, and would be unable to recognize vertex-absent visual forms, but that infants who looked for shorter durations

Varieties of Attention in Infancy

13

4 Habituate:



or s

wv”

=* s

,

-

g* Test:

s

w.

a

s :

Sf

4

a



x.

es

fs *%

J

ee

*

Figure 1.5. Discrimination task using complementary contour. Source: From “Individual differences in infant visual attention: Four-month-olds’ recognition of forms connoted by complementary contour,” by J. Colombo, J. E. Frick, J. S. Ryther, and J. J. Gifford, 1996, Infant Behavior and Development, 19, p. 115. Copyright 1996 by Elsevier. Adapted with permission.

would be able to perform the task. The findings were in good accord with the hypothesis; long-looking infants did not show recognition of vertex-degraded stimuli at any level of familiarization, but did recognize stimuli with vertices intact. Short-looking infants performed both recognition tasks, but vertexdegraded discriminations required more extensive familiarization than vertex-present forms. These findings were also consistent with prior reports showing that less mature infants tend to distribute their fixations disproportionately to stimulus angles and vertices (Haith, 1981; Salapatek, 1975). From this set of studies, then, a picture emerged showing that infants with more mature attentional profiles, as characterized by briefer look durations, processed visual stimuli with a more holistic, globally based manner of visual intake. Presumably, this was the result of distributing their attention across the entire stimulus. Infants with less mature profiles, characterized by longer look durations, appeared to be using a narrower, more locally oriented manner of visual intake. Symmetry Studies These differences could also be assessed in terms of whether these infants differentially processed symmetrical and asymmetrical forms. If attention is more broadly and completely distributed across stimuli, it stands to reason that the processing of symmetrical forms will be faster than asymmetrical forms. Across a series of experiments, Stoecker, Colombo, Frick, and Ryther (1998) found that short-looking infants indeed processed symmetrical stim-

uli more quickly than asymmetrical forms. These results, taken together, suggested that differences in infant look duration reflected different attentional schemes that had clear implications for object perception. More mature

14

Infant Perception and Cognition

infants (as marked by shorter duration of looking) appear to distribute their attention more broadly across stimuli; less mature infants (again, marked by longer look durations) appear to distribute their attention disproportionately on local stimulus features. Results from Other Laboratories While these findings were being derived from our own laboratory, the results from other programs of work on the same topic began to emerge. With a few exceptions (e.g., Frick & Richards, 2001) these findings have largely corroborated the pattern of results that we have observed. Jankowski and Rose (1997) specifically investigated several hypotheses derived from this line of inquiry. Using a hand-scoring rubric for coding the location of looking, they found that the distribution of looking across stimuli did in fact vary as a function of look duration, and that—at least at five and nine months of age— briefer look durations were related to more robust recognition performance. In an unpublished dissertation, Krinsky-McHale (1996) measured eye movements during visual habituation with infants at two and three months of age. Although the ages of the infants she studied were younger than those under study in our program, she did find that, at least on the peak look during the sequence, infants categorized as “long-looking” spent more time fixating, and made more fixations to a limited number of stimulus zones, than did infants categorized as “short-looking.” Further, Rose, Futterweit, and Jankowski

(1999) found that longer look durations in 5-, 7-, and 9-month-

olds were associated with greater positive affect, and that behavioral patterns characterized by both shorter look durations and more neutral affect were related to faster visual learning. Jankowski and Rose (2001) provided a piece of especially compelling evidence in support of this program of work. Here, their earlier findings (Jankowski & Rose, 1997) were replicated, as infants characterized by long

look durations showed more narrowly distributed attention across complex visual

stimuli,

and

performed

less well on

subsequent

recognition

tests,

than did infants with shorter look duration profiles. However, these authors sequentially illuminated different quadrants of the stimuli during familiarization, which elicited more widely distributed visual scanning. Under these conditions, the differences disappeared between short- and long-look duration infants, on both the breadth of fixations and on recognition performance. In other words, an environmental manipulation that encouraged broader attention to stimuli overcame these individual differences in visual intake. Look Duration and Visuospatial Orienting Although the results of this work suggested that different forms of visual attention affected object perception and recognition, the question remained unanswered as to why younger infants—and infants with less mature attentional profiles at older ages—appeared to fixate less broadly and perform less reliably on recognition tests, even after extended exposure to the stimuli in question. During the 1980s, Posner's program of work (summarized in Posner &

Varieties of Attention in Infancy

15

Petersen, 1990) on the cognitive neuroscience of cognitive processes had, in fact, identified an important division in the neural pathways that contributed to visual attention. A posterior attention system —similar in both structure and in basic function to the dorsal visual pathway system described by Webster and Ungerleider (1998)—was shown to mediate a number of important aspects of visuospatial orienting. Of particular interest with regard to finding individual differences in infant attention was the description of disengagement of visual attention, presumably mediated by a subset of structures in Posner's posterior attentional system. Indeed, bilateral lesions to the parietal (or occipitoparietal) lobes are known to produce a condition known as Balint’s Syndrome (e.g., Rafal, 1997), a disorder in which individuals cannot disengage visual atten-

tion from an object or stimulus (and which, incidentally, is characterized by an impairment in global processing; see Jackson, Swainson, Mort, Husain, & Jackson, 2004; Shalev, Mevorach, & Humphreys, 2007). A phenomenon called “obligatory attention” had been noted several decades ago (Friedman, 1975; Stechler & Latz, 1966), where young infants’ visual regard is “captured” by stimuli and the infant appears to be unable to terminate the fixation. The phenomenon had not previously been considered within the larger realm of infant cognitive function, although an important passage in Cohen's (1976) chapter on infant visual habituation is particularly relevant to this issue. Here, Cohen attempts to explain the propensity for prolonged fixation in young infants by invoking a construct that is strikingly prescient of the concept of disengagement: “One other point should be mentioned regarding the behavior of our 14-week-old infants. It may be that the reason they have longer fixation times is that they have difficulty releasing their attention from the visual stimulus. We get the subjective impression when observing some of these infants that they look intently for a while, then become increasingly

agitated with their eyes still glued to the pattern, and finally avert their gaze in an inconsistent manner. It is almost as if they wanted to turn away earlier, but couldn't. ... It may still be too early to give research on the attention releasing-process the status ofa new level. However, it is clear that the consistency of turning away from a visual stimulus does reflect an aspect of infant attending which has not been thoroughly investigated in the past.” (Cohen, 1976; p. 235) In 1995, we formally proposed that disengagement, a component of visuospatial orienting, might contribute to some of the correlates of look duration that we were seeing in the development of infant visual behavior across the first year of life (Colombo, 1995), and we set out to investigate this possibility in a series of studies. The first of these (Frick, Colombo, & Saxon, 1999) featured a paradigm initially developed by Hood and Atkinson (1993; see also Matsuzawa & Shimojo, 1997) to measure disengagement. Here, the infant's fixation is attracted to a stimulus in the center of the visual field. At that point, a stimulus is presented in the visual periphery (we used 25° from midline). The infant is naturally drawn to the peripheral stimulus, and the latency of the infant's first eye movement to that stimulus is measured using frame-by-frame

16

Infant Perception and Cagnition

analysis. There were two experimental conditions in the study. In the gap condition (initially termed the “non-competition condition,” since there was no

stimulus competing for the infant's gaze in the visual field), the center stimulus is removed for some time before the peripheral stimulus appears; response in this condition does not require disengagement of attention. In the overlap (or “competition”) condition, the center stimulus remains when the peripheral stimulus is presented; here, the infant must disengage attention from the central stimulus in order to look to the periphery. If disengagement were a contributory factor to the development of infant visual attention, it was predicted that we should see correlations between look duration and latency in the overlap condition, but not in the gap condition. That is, longer-looking infants would be slower than short-looking infants to shift attention to the periphery in the overlap condition, which requires disengagement, but the two groups would not differ in the gap condition. Indeed, for both 3- and 4-month-olds in this study, this prediction was confirmed, with strikingly large effects for infant work: with age partialed from the correlations between look duration and overlap latency, the Pearson correlations ranged from r= +.41 (using all trials from all subjects) to r = +.55 (using data from subjects who completed most trials), with all effects attaining p < .01 values, Importantly, the lack of correlation between look duration and ocular latency where disengagement was not necessary (in the gap condition, where the correlation was r = .05, with age partialed), indicated that this effect was

not simply due to general CNS or oculomotor slowing. A second study (Colombo, Richman, Shaddy, Greenhoot, & Maikranz, 2001) used Richards’ framework for parsing attention (Richards & Casey, 1992) with simultaneous heart-rate (HR) measurement during infant looking. Infant looking is typically accompanied by a deceleration of heart rate, a parasympathetic nervous system response that is generally taken to reflect active engagement with, and active processing of, the stimulus at hand; there is considerable evidence to support this contention (e.g., Richards, 1987, 1989, 1997). Of particular importance, however, is the fact that the decelerative response is not isomorphic with looking. That is, the infant’s heart rate is decelerated during only a portion (usually 50% to 60%; see Colombo et al., 2004) of looking. Richards and Casey (1992) characterize this as sustained attention (SA),

and much of the research that has been done on this topic has focused on the processes occurring there. Looking prior to the onset of the deceleration is thought to reflect simple orienting (OR) to the stimulus. However, looking that continues after the end of the decelerative phase is called attention termination (AT), and can be considered to reflect the same construct as disengagement of attention. We applied this rubric to 4-month-olds’ performance in paired-comparison recognition tasks to determine whether sustained attention (SA) or disengagement (as reflected by AT) was a contributor to look duration, and whether either mediated the relationship between look duration and recognition memory performance (Colombo et al., 2001). To do this, we parsed infant looking during the familiarization phase of the paired-comparison procedure

Varieties of Attention in Infancy

7

and examined whether it predicted looking time and success in recognition. As one might expect, look duration was correlated with the raw time for all three of Richards and Casey’s (1992) HR-defined phases of attention—and, as predicted, we observed look duration to be a significant predictor of success on the recognition task. However, we were somewhat surprised to find evidence that AT, rather than SA, was the attentional phase mediating the look duration/recognition performance. More time spent in AT (i.e., less facility in disengaging attention from stimuli) predicted a lower probability of success on the recognition task, and entering AT as the first predictor in a hierarchical regression dropped the predictive power for look duration below significance. Thus it appears that, at least in early infancy, disengagement contributes significantly to the development of look duration, and may figure prominently in infants’ recognition abilities. Most recently, this line of work was extended to older ages to examine the degree to which the processing of content interacts with disengagement (Blaga & Colombo,

2006), Here, 3- and 7-month-olds were tested in the

overlap paradigm described above, but the content of the center stimulus was manipulated so that it was either always the same stimulus from trial to trial, or changing from trial to trial. The rationale behind this study was to determine whether the changing content of the center stimulus would affect disengagement performance at either age. Indeed, 3-month-olds processed the central target less rapidly than did the 7-month-olds, and novel content in the central stimulus did affect disengagement at both ages. The 3-month-olds were slower to disengage when the central content was changing, and 7-month-olds showed some evidence of this when exposure time to the central stimulus was reduced markedly in a subsequent experiment. However, the age differences in processing the central target did not account for the age differences seen in disengagement as, even in the highest cognitive load conditions, the disengagement performance of 7-montholds never dropped to that of 3-month-olds. In addition, the correlation between look duration and disengagement observed by Frick et al. (1999) was repeated in this study (r = .47, p < .01) for 3-month-olds, but was nonsignificant for 7-month-olds. Altogether, these findings suggested that developmental and individual differences in look duration is determined by both the visuospatial process of disengagement and efficiency in processing, and that the contribution of disengagement wanes toward the end of the first year. SUMMARY

AND CONCLUSIONS

This chapter summarizes over 20 years of research on the development of attention in human infants. The work has been largely influenced by adult work on human information processing (Navon, 1977, 1983; Pomerantz, 2003,

1983; Pomerantz, Sager, & Stoever, 1977), and a focus that emphasizes both the nomothetic and idiographic aspects of behavioral science (Underwood, 1975).

18

Infant Perception and Cognition

What we hope this review shows is that attention, particularly during infancy, is best considered from a multicomponent perspective, The picture revealed by this work is a complex one, where both the behavioral characteristics and the cognitive performance of human infants is best explained by invoking multiple processes that might fall under the general rubric of attention. The distinction between “what” and “where” is a common one in explaining the course of visual processing (Atkinson & Braddick, 1989; Beck, Peterson, & Vomela, 2006; Duncan, 1993; Irwin & Brockmole, 2004), largely because the neurology of visual cognition (see Carlesimo, Perri, Turriziani, Tomaiuolo, & Caltagirone, 2001; Webster & Ungerleider, 1998) strongly supports the existence of separate neural pathways that support the processing of the visuospatial characteristics of stimuli, and the processing of the content of the stimuli per se: a dorsal pathway (visual cortex to parietal cortex and superior colliculus), which codes and coordinates responses based on visuospatial input, and a ventral pathway (visual cortex to temporal cortex and medial temporal structures) are involved in explaining the types of findings reviewed here (see Colombo, 2001b).

The data suggest that changes in performance in infants’ visual processing across the first six months might be attributable to the maturation of these two pathways. In addition, if we may extrapolate from the most recent data (Blaga & Colombo, 2006), we might further suggest that the developmental course of the two systems diverges at about four months, as there is relatively little change in ocular reaction time in overlap paradigm tests, whereas differences in the processing of visual content persist beyond that point. The developmental course of these systems should be viewed in the context of all of the processes that give rise to the construct of attention. As we have noted previously (Colombo, 2001), the modularity of attention in infancy will always make for

a complicated explanation, because infant behavior and development must be explained in terms of the varieties of attention that William James so aptly characterized over a century ago. Why is a younger or less mature infant slower to turn or look away from a visual stimulus than an older or more mature infant? The fact that we must invoke these varieties of attention tells us that there will be no simple answer to such a simple question. A 3-month-old infant is slower than a 4-monthold because the younger infant is both slower to encode the stimulus content, and less able to disengage attention from the stimulus once the encoding is completed. The differences between a 4- and a 7-month-old, however, may be largely explained in terms of differences in encoding efficiency, as it appears that visual disengagement performance is roughly equivalent across these two ages. Disentangling these influences in terms of age or maturation is possible in terms of nomothetic principles; disentangling these influences on an idiographic level (with the intent of understanding how these two processes contribute to the development of later cognitive function) will be much more

difficult. A final point that we hope this review shows is that research on the development of early cognitive function is best executed when informed by adult

Varieties of Attention in Infancy

19

models of information processing and by an understanding of the cognitive neuroscience of the systems under study. Perhaps the field of infant perception looks back now with some discomfort at attempts to explain, for example, the nature of infant visual preferences in terms of “contour,”

“area,” “contour density,” and “complexity” only to learn that the wheel had long been invented in other realms of science, and that the importing of the principles of linear systems theory (Banks & Salapatek, 1981), long known in the study of adult vision, could quickly resolve many of the fundamental questions about the phenomena in question (e.g., Gayl, Roberts, & Werner, 1983).

REFERENCES Atkinson, J., & Braddick, O. J. (1989). “Where” and “what” in visual search, Perception, 18, 181-189. Axia, G., Bonichini, S., & Benini, F. (1999). Attention and reaction to distress in infancy: A longitudinal study, Developmental Psychology, 35, 500-504. Banks, M. S., & Salapatek, P. (1981). Infant pattern vision: A new approach based on the contrast sensitivity function. Journal of Experimental Child Psychology, 31, 1-45.

Beck, M. R., Peterson, M. S., & Vomela, M. (2006). Memory for where, but not what, is used during visual search. Journal of Experimental Psychology: Human Perception and Performance, 32, 235-250. Biederman, I. (1987). Recognition-by-components: A theory of human image understanding. Psychological Review, 94, 115-147. Blaga, O., & Colombo, J. (2006). Visual processing infant ocular latencies in the overlap paradigm. Developmental Psychology, 42, 1069-1076. Briggs, G. E., & Shinar, D. (1972). On the locus of the speed/accuracy tradeoff. Psychonomic Science, 28, 326-328.

Broadbent, D. (1958). Perception and communication. London: Pergamon Press. Broadbent, D, E. (1982). Task combination and selective intake of information. Acta

Psychologica, 50, 253-290. Carlesimo, G. A., Perri, R., Turriziani, P., Tomaiuolo, F., & Caltagirone, C. (2001). Remembering what but not where: Independence of spatial and visual working memory in the human brain. Cortex, 37, 519-534. Cassia, V. M., & Simion, F. (2002). Individual differences in object-examining duration: Do they reflect the use of different encoding strategies? Cognitive Development, 17, 1219-1234. Cohen, L. B. (1972). Attention-getting and attention-holding processes of infant visual preferences. Child Development, 43, 869-879. Cohen, L. B. (1973). A two process model of infant visual attention Merrill-Palmer Quarterly, 19, 153-180.

Cohen, L. B. (1976). Habituation of infant visual attention. In T. J. Tighe & R.N. Leaton (Eds.), Habituation (pp. 207-238). Hillsdale, NJ: Erlbaum.

Cohen, L. B., DeLoache, J. S., & Rissman, M. W. (1975). The effect of stimulus complexity on infant visual attention and habituation, Child Development, 46, 611-617.

Colombo, J. (1993). Infant cognition: Predicting later intellectual functioning. Newbury Park, CA: Sage Publications, Inc.

20

Infant Perception and Cognition

Colombo, J. (1995). On the neural mechanisms underlying developmental individual differences in infant fixation duration: Two hypotheses. Developmental Review, 15, 97-135.

Colombo, J. (2001a). The development of visual attention in infancy. Annual Review of Psychology, 52, 337-367. Colombo, J. (2001b). Recent advances in the assessment of infant cognition:

Implications for LC-PUFA supplementation studies. Lipids, 36, 919-926. Colombo, J. (2002). Infant attention grows up: The emergence ofa developmental cognitive neuroscience perspective. Current Directions in Psychological Science, 11, 196-199,

Colombo, J. (2004). Visual attention in infancy: Process product in early cognitive development. In M. Posner (Ed.), Cognitive neuroscience of attention (pp. 329-341), New York: Guilford Press. Colombo, J., & Cheatham, C. (2006). The emergence of endogenous attention in infancy and early childhood. In R. Kail (Ed.), Advances in child development and behavior (pp. 283-322), New York: Elsevier. Colombo, J., & Fagen, J. W. (Eds.) (1990). Individual differences in infancy: Reliability, stability, prediction. Hillsdale, NJ: Lawrence Erlbaum Associates. Colombo, J., Freeseman, L. J., Coldren, J. T., & Frick, J. E. (1995). Individual

differences in infant visual fixation: Dominance of global local stimulus properties, Cognitive Development, 10, 271-285. Colombo, J., & Frick, J. E. (1999). Recent advances issues in the study of preverbal

intelligence. In M. Anderson (Ed.), The development of intelligence (pp. 46-71). East Sussex: Psychology Press. Colombo, J., Frick, J. E., Ryther, J. 8., & Gifford, J. J. (1996), Individual differences

in infant visual attention: Four-month-olds’ recognition of forms connoted by complementary contour. Infant Behavior and Development, 19, 113-119. Colombo, J., Frick, J. E., & Gorman, S. A. (1997). Sensitization during visual

habituation sequences: Procedural effects individual differences. Journal of Experimental Child Psychology, 67, 223-235. Colombo, J., & Horowitz, F. D. (1985). A parametric study of the infant control

procedure. Infant Behavior and Development, 8, 117-121. Colombo, J., & Janowsky, J. S. (1998). A cognitive neuroscience approach to

individual differences in infant cognition. In J. E. Richards (Ed.), The cognitive neuroscience of attention: A developmental perspective (pp. 363-392). Hillsdale, NJ: Lawrence Erlbaum.

Colombo, J., Laurie, C, A., Martelli, T. A., & Hartig, B. R. (1984). Stimulus context infant orientation discrimination. Journal of Experimental Child Psychology, 37, 576-586. Colombo, J., McCardle, P., & Freund, L. (Eds.) (2009). Infant pathways to language: Methods, models, research directions. Mahwah, NJ: Lawrence Erlbaum Associates. Colombo, J., & Mitchell, D. W. (1988). Infant visual habituation: In defense of an information-processing analysis. European Bulletin of Cognitive Psychology/ Cahiers de Psychologie Cognitive, 8, 455-461. Colombo, J., & Mitchell, D. W, (1990). Individual developmental differences in infant visual attention: Fixation time information processing, In J, Colombo & J. W, Fagen (Eds.), Individual differences in infancy: Reliability, stability, prediction (pp. 193-227). Hillsdale, NJ: Lawrence Erlbaum. Colombo, J., Mitchell, D, W., Coldren, J. T., & Freeseman, L. J. (1991). Individual

differences in infant attention: Are short lookers faster processors or feature processors? Child Development, 62, 1247-1257.

Varieties of Attention in Infancy

21

Colombo, J., Mitchell, D. W., Dodd, J. D., Coldren, J. T., & Horowitz, F. D. (1989).

Longitudinal correlates of infant attention in the paired comparison paradigm. Intelligence, 13, 33-42. Colombo, J., Mitchell, D, W., & Horowitz, F. D. (1988). Infant visual behavior in the

paired-comparison paradigm: Test-retest attention-performance relations. Child Development, 59, 1198-1210. Colombo, J., Mitchell, D. W., O’Brien, M., & Horowitz, F. D. (1987). Stability of infant

visual habituation during the first year. Child Development, 58, 474-489. Colombo, J., Mitchell, D. W., O’Brien, M., & Horowitz, F. D. (1987), Stimulus

motoric influences on visual habituation at three months. Infant Behavior and Development, 10, 173-181. Colombo, J., Richman, W. A., Shaddy, D, J., Greenhoot, A. F., & Maikranz, J. (2001).

HR-Defined phases of attention, look duration, infant performance in the paired-comparison paradigm. Child Development, 72, 1605-1616. Colombo, J., Shaddy, D. J., Richman, W. A., Maikranz, J. M., & Blaga, O. (2004).

Developmental course of visual habituation preschool cognitive language outcome. Infancy, 5, 1-38. Colombo, J., Shaddy, D. J., Blaga, O. M., Anderson, C. J., & Kannass, K. N. (2009).

High cognitive ability in infancy and early childhood. In F, D. Horowitz, R. F, Subotnik, & D, J. Matthews (Eds.), The development of giftedness and talent

across the life span (pp. 23-42). Washington, DC: American Psychological Society Press.

Colombo, J., Shaddy, D. J., Blaga, O. M., Anderson, C. J., Kannass, K. N., & Richman, W. A. (2009). Attentional predictors of vocabulary from infancy. In J. Colombo, P. McCardle, & L. Freund (Eds.), Infant pathways to language: Methods, models, and research directions (pp. 143-168). New York, NY: Psychology Press/Taylor and Francis. Courage, M. L., & Howe, M. L, (2001). Long-term retention in 3.5-month-olds;

Familiarization time and individual differences in attentional style. Journal of Experimental Child Psychology, 79, 271-293. Courage, M. L., Reynolds, G, D., & Richards, J. E. (2006). Infants’ attention to patterned stimuli: Developmental change from 3 to 12 months of age. Child Development, 77, 680-695. Dannemiller, J. L. (1998). A competition model of exogenous orienting in 3.5-month-old infants. Journal of Experimental Child Psychology, 68, 169-201. Dannemiller, J. L. (2000). Competition in early exogenous orienting between 7 and 21 weeks. Journal of Experimental Child Psychology, 76, 253-274. Dannemiller, J. L. (2001). Testing models of attentional capture during early infancy.

In C. L. Folk, & B.S. Gibson (Eds.), Attraction, distraction and action: Multiple perspectives on attentional capture. Advances in psychology (Vol. 133, pp. 265-291). New York, NY: Elsevier Science, Deutsch, J. A., & Deutsch, D. (1963). Attention: Some theoretical considerations.

Psychological Review, 70, 80-90. Duncan, J. (1993). Coordination of what and where in visual attention. Perception, 22, 1261-1270. Fernandez-Duque, D., Baird, J. A., & Posner, M. I. (2000). Executive attention and metacognitive regulation. Consciousness and Cognition, 9, 288-307.

Finlay, D., & Ivinskis, A. (1982). Cardiac and visual responses to stimuli presented both foveally and peripherally as a function of moving stimuli. Developmental Psychology, 18, 692-698.

22

Infant Perception and Cognition

Fisk, A. D., & Schneider, W. (1981). Control and automatic processing during tasks requiring sustained attention: A new approach to vigilance. Human Factors, 23, 737-750.

Foreman, N., Fielder, A., Price, D., & Bowler, V. (1991). Tonic and phasic orientation in full-term and preterm infants, Journal of Experimental Child Psychology, 51, 407-422.

Freeseman, L. J., Colombo, J., & Coldren, J. T. (1993). Individual differences in infant visual attention: Discrimination generalization of global local stimulus properties. Child Development, 64, 1191-1203. Frick, ]. E., & Colombo, J. (1996). Individual differences in infant visual attention:

Recognition of degraded visual forms by 4-month-olds. Child Development, 67, 188-204. Frick, J. E., Colombo, J., & Allen, J. R. (2000). The temporal sequence of global-local processing in 3-month-olds. Infancy, 1, 375-386. Frick, J. E., Colombo, J., & Saxon, T. F. (1999). Individual developmental

differences in disengagement of fixation in early infancy. Child Development, 70, 537-548. Frick, J. E., & Richards, J. E. (2001). Individual differences in infants’ recognition of

briefly presented visual stimuli. Infancy, 2, 331-352. Friedman, S. (1975). Infant habituation: Process, problems, and possibilities. In N. R. Ellis (Ed.), Aberrant development in infancy (pp. 217-237). Hillsdale, NJ: Erlbaum, Ganz, L., Hirsch, H. V. B., & Tieman, S, B. The nature of perceptual deficits in visually deprived cats. Brain Research, 1972, 20, 67-87. Gayl, I, E., Roberts, J, O,, & Werner, J. S. (1983). Linear systems analysis of infant visual pattern preferences. Journal of Experimental Child Psychology, 35, 30-45. Haith, M. M. (1981). Rules that babies look by. Hillsdale, NJ: Erlbaum. Hood, B. M., & Atkinson, J. (1993). Disengaging visual attention in the infant and

adult. Infant Behavior and Development, 16, 405-422. Horowitz, F, D., Paden, L., Bhana, K., & Self, P. (1972). An infant-control procedure for studying infant visual fixations. Developmental Psychology, 7, 90. Irwin, D. E., & Brockmole, J. R. (2004). Suppressing where but not what: The effect of saccades on dorsal- and ventral-stream visual processing. Psychological Science, 15, 467-473, Jackson, G. M., Swainson, R., Mort, D., Husain, M., & Jackson, S. R. (2004). Implicit

processing of global information in Balint’s syndrome. Cortex, 40, 179-180. James, W. (1890). The principles of psychology. New York: Henry Holt. Jankowski, J. J., & Rose, S. A. (1997). The distribution of visual attention in infants.

Journal of Experimental Child Psychology, 65, 127-140. Jankowski, J. J., Rose, S. A., & Feldman, J. F. (2001). Modifying the distribution of

attention in infants. Child Development, 72, 339-351. Kannass, K. N., Oakes, L. M., & Shaddy, D. J. (2006). A longitudinal investigation of the development of attention and distractibility. Journal of Cognition and Development, 7, 381-409. Krinsky-McHale, S. J, (1996), Visual scanning during information processing in infancy. Dissertation Abstracts International: Section B: The Sciences and Engineering, 56(10-B), 5796. Landry, S. H., Leslie, N. A., Fletcher, J. M., & Francis, D, J. (1985), Visual attention skills of premature infants with and without intraventricular hemorrhage. Infant Behavior and Development, 8, 309-321.

Varieties of Attention in Infancy

23

Leahy, R. L. (1976). Development of preferences and processes of visual scanning in the human infant during the first 3 months of life. Developmental Psychology, 12, 250-254, Maikranz, J. M., Colombo, J., Richman, W. A., & Frick, J. E. (2001). Autonomic

indicators of sensitization look duration in infancy. Infant Behavior and Development, 23, 137-151. Matsuzawa, M., & Shimojo, S. (1997). Infants’ fast saccades in the gap paradigm and development of visual attention. Infant Behavior and Development, 20, 449-455. Maurer, D., & Salapatek, P. (1976). Developmental changes in the scanning of faces by young infants. Child Development, 47, 523-527. Mayes, L. C., & Kessen, W. (1989). Maturational changes in measures of habituation. Infant Behavior and Development, 12, 437-450. Moss, M. M., Colombo, J., Mitchell, D. W., & Horowitz, F, D, (1988). Neonatal

behavioral organization and 3-month visual discrimination. Child Development, 59, 1211-1220.

Miller, J. (1981). Global precedence in attention and decision. Journal of Experimental Psychology: Human Perception and Performance, 6, 1161-1174. Navon, D. (1977). Forest before trees: The precedence of global features in visual

perception. Cognitive Psychology, 9, 353-383. Navon, D. (1983). The forest revisited: More on global precedence. Psychological Research, 43, 1-32. Nuechterlein, K. H., Parasuraman, R., & Jiang Q. (1983). Visual sustained attention:

Image degradation produces rapid sensitivity decrement over time. Science, 220, 327-329.

Oakes, L. M., Tellinghuisen, D. J., & Tjebkes, T. L. (2000). Competition for infants’ attention: The interactive influence of attentional state and stimulus characteristics, Infancy, 1, 347-361. Pachella, R. G., & Pew, R. W. (1968). Speed-accuracy tradeoff in reaction time: Effect of discrete criterion times. Journal of Experimental Psychology, 76, 19-24. Parasuraman, R. (1984, Ed.) Varieties of attention. New York: Academic Press. Parasuraman, R. (1998, Ed.). The attentive brain. Cambridge, MA: MIT Press. Pomerantz, J. R. (2003). Wholes, holes, and basic features in vision. Trends in Cognitive Sciences, 7, 471-473.

Pomerantz, J. R. (1983). Global and local precedence: Selective attention in form and motion perception. Journal of Experimental Psychology: General, 112, 516-540.

Pomerantz, J. R., Sager, L. C., & Stoever, R. J. (1977). Perception of wholes and of their component parts: Some configural superiority effects. Journal of Experimental Psychology: Human Perception and Performance, 3, 422-435. Posner, M. (2004, Ed.). Cognitive neuroscience of attention. New York: Guilford Press. Posner, M. I., Inhoff, A. W., Friedrich, F. J., & Cohen, A. (1987). Isolating attentional

systems: A cognitive-anatomical analysis. Psychobiology, 15, 107-121. Posner, M. L., & Petersen, S. E. (1990). The attention system of the human brain. Annual Review of Neuroscience, 13, 25-42. Posner, M. I., & Rothbart, M. K. (2007), Research on attention networks as a model for the integration of psychological science. Annual Review of Psychology, 58, 1-23.

Rafal, R. D. (1997). Balint syndrome. In T, Feinberg & M. Farah (Eds.), Behavioral neurology and neuropsychology (pp. 337-356). New York: McGraw-Hill.

24

Infant Perception and Cognition

Richard, J. F., Normandeau, J., Brun, V., & Maillet, M. (2004). Attracting and

maintaining infant attention during habituation: Further evidence of the importance of stimulus complexity. Infant and Child Development, 13, 277-286. Richards, J. E. (1987). Infant visual sustained attention and respiratory sinus

arrhythmia. Child Development, 58, 488-496, Richards, J. E. (1989). Development and stability in visual sustained attention in 14, 20, and 26 week old infants. Psychophysiology, 26, 422-430. Richards, J. E. (1997), Effects of attention on infants’ preference for briefly exposed visual stimuli in the paired-comparison recognition-memory paradigm. Developmental Psychology, 33, 21-31. Richards, J. E. (1998, Ed.), The cognitive neuroscience of attention: A developmental perspective. Hillsdale, NJ: Lawrence Erlbaum. Richards, J. E., & Casey, B. J. (1992). Development of sustained visual attention in the human infant. In B. A. Campbell, H. Hayne, & R. Richardson (Eds.), Attention

and information processing in infants and adults (pp. 30-60). Hillsdale, NJ: Erlbaum. Robbins, T. W., & Everitt, B. J. (1995). Arousal systems and attention. In M. S.,

Gazzaniga, (Ed.) The cognitive neurosciences (pp. 703-720). Cambridge, MA: The MIT Press. Rose, S. A., Futterweit, L. R., & Jankowski, J. J, (1999), The relation of affect to

attention and learning in infancy. Child Development, 70, 549-559. Rothbart, M. K., Posner, M. I., & Kieras, J. (2006). Temperament, attention, and the

development of self-regulation. In K, McCartney & D. Phillips (Eds.), Blackwell handbook of early childhood development. Blackwell handbooks of developmental psychology (pp. 338-357). Malden, MA: Blackwell Publishing. Rueda, M. R., Posner, M. I., & Rothbart, M. K. (2005). The development of executive attention: contributions to the emergence of self-regulation. Developmental Neuropsychology, 28, 573-594. Ruff, H. A. (1986a). Components of attention during infants’ manipulative exploration. Child Development, 57, 105-114.

Ruff, H. A. (1986b). Attention and organization of behavior in high-risk infants. Journal of Developmental and Behavioral Pediatrics, 7, 298-301.

Ruff, H. A. (1990). Individual differences in sustained attention during infancy. In J. Colombo, & J. W. Fagen (Eds.). Individual differences in infancy: Reliability, stability, prediction (pp. 247-270). Hillsdale, NJ: Lawrence Erlbaum Associates. Ruff, H. A., & Capozzoli, M. C. (2003). Development of attention and distractibility in the first 4 years of life. Developmental Psychology, 39, 877-890. Ruff, H. A., & Lawson K. R. (1990). Development of sustained, focused attention in young children during free play. Developmental Psychology, 26, 85-93. Ruff, H. A., Lawson, K. R., Parinello, R., & Weissberg, R. (1990). Long-term stability of individual differences in sustained attention in the early years. Child Development, 61, 60-75. Salapatek, P, (1975). Pattern perception in early infancy. In L. Cohen & P. Salapatek (Eds.), Infant perception: from sensation to cognition (Vol. 1). Basic visual processes (pp. 133-248). New York: Academic Press. Shaddy, D, J., & Colombo, J, (2004). Developmental change in attention: Dynamic

static stimuli. Infancy, 5, 355-365. Shalev, L., Mevorach, C., & Humphreys, G. W. (2007). Local capture in Balint’s syndrome: Effects of grouping and item familiarity. Cognitive Neuropsychology, 24, 115-127.

Varieties of Attention in Infancy

25

Sokolov, E. N. (1963). Perception and the conditional reflex, New York, NY: MacMillan. Sperling, G. (1960). The information available in brief visual presentations. Psychological Monographs, 74, 1-29. Stechler, G., & Latz, E. (1966). Some observations on attention and arousal in the human infant. Journal of the American Academy of Child Psychiatry, 5, 517-525. Stoecker, J. J., Colombo, J., Frick, J. E., & Ryther, J. S. (1998). Long- and short-looking infants’ recognition of symmetrical asymmetrical visual forms. Journal of Experimental Child Psychology, 71, 63-78. Swanson, J. M., & Briggs, G. E. (1969), Information processing as a function of speed versus accuracy. Journal of Experimental Psychology, 81, 223-229. Treisman, A. M. (1969). Strategies and models of selective attention. Psychological Review, 76, 282-299. Underwood, B. J. (1975). Individual differences as a crucible in theory construction.

American Psychologist, 30, 128-134. Webster, M. J., & Ungerleider, L.G. (1998), Neuroanatomy of visual attention. In

R. Parasuraman (Ed,), The attentive brain (pp. 19-34). Cambridge: MIT Press. Yantis, S. (1993). Stimulus-driven attentional capture. Current Directions in

Psychological Science, 2, 156-161. Yantis, S., & Jonides, J. (1984). Abrupt visual onsets and selective attention: Evidence

from visual search. Journal of Experimental Psychology: Human Perception and Performance, 10, 601-621.

2 Infant Attention, Arousal, and the Brain John E. Richards

Leslie B. Cohen is a pioneer in many areas of infant cognitive development. One line of research “founded” by his early work is the examination of multiple attention types in infants, employing behavioral and experimental manipulations (e.g., Cohen, 1972). This work led to the study of several types of attention in young infants and is now paying off in the identification of brain areas involved in infant attention, Developmental psychologists are beginning to understand how neural activity underlying infant cognitive processes is facilitated by attention. The current chapter is a summary of a line of research that I have been following for more than 25 years on the topic of “multiple attention types.” This work has been inspired by the information-processing tradition that asserts that attention is one way that information is selected for evaluation from a broad range of available stimulation. This approach was studied behaviorally first by Cohen (e.g., 1972), but has expanded to affect nearly all of infant cognitive developmental work. I will emphasize the aspects of my own work that have examined the information processing aspects of infant attention, and the neural basis of attention. This is a selective review that traces some “prescient” conclusions drawn by Cohen regarding the nature of differing attention types, but also expands upon those views to examine the neural basis of attention. The review is selective, in that it neither fully reviews the field of infant attention, nor is it a full review of my work in this area. TWO

PHASES OF INFANT ATTENTION

(WITHIN A LOOK)

“Whatever conclusions are reached from the studies, the present investigation has already demonstrated the feasibility, perhaps even the necessity, of independently assessing attention-getting and attentionholding aspects of infant visual fixations.” L. B. Cohen, (1972, p. 878)

The start of the behavioral work on infant attention types is easily traced to a single study of Cohen’s in 1972 (Cohen, 1972). An important This research was supported by a grant from the National Institute of Child Health and Human Development, R37-HD18942

28

Infant Perception and Cognition

methodological advance in this study was the “infant-control” presentation procedure. The visual patterns were presented, and online evaluation of the direction of infant fixation was made. As long as the infant looked toward the visual pattern, the pattern remained on. When the infant looked away from the pattern, it was turned off. The duration of fixation toward the pattern was both the dependent variable and the variable controlling stimulus duration. The infant control procedure differed from previous presentation methods in which the stimulus was presented for a fixed period of time, the infant looked toward the stimulus and away from it several times, and the dependent variable was the amount of looking time over the entire duration of the fixed stimulus interval (Brennan, Ames, & Moore,

1966; Kagan

& Lewis, 1965). Cohen (1972) argued that the infant control procedure was preferable because it did not confound patterns of looking (and looking away) with the infant’s transitory fixations, and thus offered an improved measure of attention. Duration of fixation toward the stimulus was the measure of “visual attention.” Incidentally, for similar reasons, the infant control procedure apparently was “discovered” independently by Horowitz, Paden, Bhana, and Self (1972).

One of the reasons Cohen gave for preferring the infant control procedure was that it was the most relevant to the distinction of attention types. The total fixation duration ona fixed-duration presentation procedure could come from a number of brief looks, or a single long look. Presupposing that the beginning of fixation is affected by one process, and the total duration of fixation is determined by other processes, the same amount of accumulated fixation for the multiple looks and the single looks would obscure the underlying attention processes. The length of a single look toward the stimulus in the infant control procedure would be a better measure of the processes holding fixation toward the stimulus. The process controlling behavior at the beginning of the look was labeled “attention-getting,” and the process keeping fixation toward the stimulus was labeled “attention-holding.” How were attention-getting and attention-holding examined in this study? Infants at four months of age were presented with visual patterns that consisted of black and white checkerboards. The checkerboard patterns were of varying size and number. Earlier work had hypothesized that there was a relation between infant age and the complexity of visual stimuli that elicited attention. Thus, it was expected that infants should show varying amounts of attention to the patterns. The stimulus was presented when the infant was looking toward a blinking light that was located away from the presentation screen. Cohen measured the latency of the look from “off-screen” to “screen” following the onset of the checkerboard stimulus on the screen. The infant control procedure was used to determine the duration of the stimulus presentation, and this also was the duration of “visual attention.” The most important finding from this study was that different aspects of the stimuli affected look latency and look duration. The size of the squares in the checkerboard was most important in influencing the latency to look toward the checkerboard screen, whereas the number of squares in the checkerboard was most

Infant Attention, Arousal, and the Brain

29

influential in the duration of looking. Cohen concluded from this finding that

this method could be used to measure the attention processes at the beginning of fixation (attention-getting) and attention processes occurring during fixa-

tion (attention-holding). Cohen’s study made two very important contributions. First, the infant control procedure was introduced, and provided a new method for studying infant visual attention. I will come back to the importance of this in the next section. Second, Cohen’s proposal was that attention was not a unitary process. Rather, there are different types of underlying processes that affect attention. The “getting” and “holding” further implies that differing types of attention are being studied sequentially. Attention first begins (attention-getting), followed by attention being sustained by the stimulus (attention-holding). The sequential nature of these attention “types” leads to the notion that attention goes through “phases” during visual fixation, The methodological advance (infant control procedure) made it possible to identify the theoretically distinct attention phases.

MULTIPLE ATTENTION PHASES DEFINED WITH HEART RATE “Taken together, the two studies support Cohen’s hypothesis that infant attention involves at least two different mechanisms: an attention-getting process which determines whether or not the infant will orient toward a stimulus projected in his periphery, and an attention-holding process determining how long his gaze will be maintained once he fixates.” L. B. Cohen (1972, p. 877).

One of the conclusions from the Cohen (1972) study was that there were “at least two” different mechanisms in infant attention, the attention-getting and attention-holding mechanisms. The previous section reviewed the importance of the methods and the theoretical conclusions reached by Cohen on the basis of this work. However, around the same time, psychologists interested in infant cognitive development were using heart rate as a measure of attention (e.g., Graham, 1970, 1979, 1992; Graham, Anthony, & Zeigler, 1983; Porges, 1976, 1980). They concluded that there may be several components of attention that occur in a “stimulus-processing event.” I will review some of this work, and present a model for using heart rate to index multiple phases of attention within an infant’s look in the infant control procedure. I have reviewed this work in several places (Berg & Richards, 1997; Reynolds & Richards, 2007; Richards & Casey, 1992). Sokolov (1963) had asserted that physiological measures could be useful in distinguishing the human response to environmental stimuli. Well-known processes relevant to attention research first studied by Sokolov in adult participants were the orienting response, habituation, and sensitization. Graham and Clifton (1966; nee Rachel Keen) proposed that heart rate could be used to distinguish orienting process (heart rate deceleration) from other activation responses such as defensive responses (heart rate acceleration;

30

Infant Perception and Cognition

Graham & Clifton, 1966). Graham in several places developed this work further (Graham,

1970, 1979, 1992; Graham

et al., 1983). She claimed that

heart-rate responses to a sudden-onset, moderate-intensity stimulus, first produced an automatic interrupt that was a transient detection of the stim-

ulus change. This was indicated by a brief change in heart rate. ‘Then, if the infant continued being interested in the stimulus, a longer-lasting stimulus orienting would occur, indicted in heart rate by a large deceleration, Graham and several colleagues showed that infants’ responses to auditory and visual stimuli showed this pattern of response (Graham, 1970). Much of Graham’s

work examined the responses to relatively brief stimuli, presented for a fixed duration (e.g., the fixed duration stimulus presentation). Porges (1976, 1980) began using heart rate in response to much longer sustained stimuli. He suggested that another, more sustained attention response occurred after stimulus orienting. The work of Graham and Porges implied that heart rate might index a sequence of qualitatively different aspects of attention in response to the presentation of visual stimuli. The similarity of this work to Cohen’s distinction between attention-getting and attention-holding is obvious. The short-latency, transient-detection reaction, indexed by a brief heart rate change, interrupts the ongoing cognitive processes and attracts processing to a new stimulus. This redirects attention toward the new stimulus—“attention-getting.” The stimulus orienting that occurs are the initial stages of information processing, and the sustained stimulus processing to sustained stimuli—reflected in heart rate deceleration and continued heart rate change—reflect the active attention to the stimulus. This continued stimulus processing is similar to Cohen's “attention-holding” phase. I was perhaps the first to use heart rate together with the infant control procedure, with the specific goal to study the more extended aspects of stimulus processing (Richards, 1987). In my studies infants were presented with a range of stimuli, including simple visual stimuli, geometric patterns, visual stimuli linked with auditory stimuli, and complex multidimensional dynamic stimuli. There is a ubiquitous pattern of heart rate change that occurs in the infant control procedure. Figure 2.1 shows the heart rate change in infants ranging in age from 3 to 6 months when presented with simple geometric patterns in the infant control presentation method (Richards & Casey, 1991). There is a large deceleration of heart rate that occurs at the beginning of the look toward the stimulus (or stimulus onset, if the infant is already looking). This is followed by a sustained lowered heart rate for some period of time, after which heart rate returns to its prestimulus level. This all occurs within the look toward the stimulus controlled by the infant’s fixation. At some point the infant looks away from the stimulus. I have proposed a model that hypothesizes multiple phases of attention occurring sequentially during the course of a single look toward the stimulus (Reynolds & Richards, 2007; Richards & Casey, 1992). The “pre-attentive” phase consists of the automatic transient-detection system which directs attention towards the stimulus. Stimulus orienting then occurs, lasting for 4 to 5 seconds, and indicated by the rapid deceleration of heart rate. Stimulus orienting is

Infant Attention, Arousal, and the Brain

31

Pre-

Stimulus

Sustained

Pre-Attention

Attention

Attention

Orienting

Attention

Termination

Termination

2 E

a 2.

@

0 po

-2

Cc s

—4 -8 -10

T

-5

0

5

10

T

T

7

15

20

25

Seconds following stimulus onset

Figure 2.1. The heart-rate-defined phases of attention in 3- to 6-month-old infants. Source: From “Heart rate variability during attention phases in young infants,” by J. E. Richards and B. J. Casey, 1991, Psychophysiology, 28, p. 46. Copyright 1991 by Wiley-Blackwell. Reprinted with permission.

characterized by an initial registration of the physical properties of the stimulus, but only limited information processing occurs during this phase. Following stimulus orienting, given that the stimulus is interesting to the infant, sustained attention begins. Sustained attention is indicated by a prolonged lowered heart rate. The duration of sustained attention is variable, affected by the infant’s age, the complexity of the stimulus, the relation between the background context and the stimulus, and other variables. Sustained attention represents the period of time in which information and stimulus processing occurs. Attention termination is a final phase of attention occurring in the infant control procedure, within a look toward the visual stimulus. This phase is preceded (marked by?) a return of heart rate to its prestimulus level. Immediately after heart rate returns to its prestimulus level, the infant is less responsive to external stimulation than during the other phases. This phase may be followed by continued fixation on the stimulus, with heart rate remaining at the same level as it was during the prestimulus period, indicating a period without active attention engagement—or, the infant is very likely to look away from the stimulus during attention termination, or this inattentive period. Note that the use of the entire duration of fixation in the infant control procedure, as the measure of infant visual attention (“attention-holding” phase), is unlikely to be correct. The model I have proposed suggests that there can be significant periods of time in which stimulus information processing does not occur, or which vary on the amount of information processing that does occur. This model of sequential heart-rate-defined attention phases has many similarities to the conclusions reached by Cohen from his behavioral study.

32

Infant Perception and Cognition

‘The first three phases of the model (attention-interrupt, stimulus orienting, sustained attention) were derived from Graham’s and Porges’ work, and share the similarity of that work to Cohen’s. Both Cohen’s work and this model support the idea that sequential attention phases occur during the course of a look. Cohen’s statement that “infant attention involves at least two different mechanisms” was prescient in its recognition that attention should be parsed into multiple phases. However, the “at least” has turned out to be “at least five phases of attention.” INFANT “ATTENTION-HOLDING” INVOLVES INFORMATION PROCESSING “On the other hand, Cohen (1969) has provided some evidence for the hypothesis that attention holding involves more active information processing and may be influenced more by the variability, amount of edge, or novelty of the pattern,” L. B. Cohen (1972, p. 878) “Everyone knows what attention is. It is the taking possession by the

mind, in clear and vivid form, on one out of what seem several possible objects or trains of thought..., The immediate effects of attention are to make us: a) perceive b) conceive c) distinguish d) shorten ‘reaction

time’—better than otherwise we could...” William James, Principles of Psychology, 1890 Cohen (1972) concluded that the attention-holding mechanism

involved

more information processing, and was more influenced by a variety of experimental factors, than was the attention-getting process. Cohen's conclusion is consistent with William James’ assertion that attention has the effect of

enhancing psychological processes, and he attributed this aspect of attention to the attention-holding aspect of infant attention. In the current section I will present a study demonstrating that information processing occurs primar-

ily during the period of “sustained attention,” defined by heart rate changes (Reynolds & Richards, 2007; Richards & Casey, 1992). I will review one study that showed the effect of information processing occurring during sustained attention on a subsequent measure of recognition memory (Frick & Richards, 2001; Richards, 1997). Infants were first presented with a Sesame Street movie, “Follow that Bird,” on a television monitor. ‘This movie elicits the heart rate changes associated with attention, including sustained attention, attention termination, and periods of inattentiveness. The heart rate changes were monitored in real time by digitizing the ECG, identifying the R-wave in the ECG, calculating interbeat intervals, and determining when a significant deceleration occurred in heart rate (sustained attention) or

when heart rate returned to its prestimulus level after the deceleration (attention termination). Then—at points defined by time, the occurrence of sus-

tained attention, or the occurrence of attention termination—a simple black and white geometric pattern replaced the movie for 2.5 or 5.0s, and then the movie was continued for several seconds. If sustained attention represents the period of time that infants are processing the information in the visual

Infant Attention, Arousal, and the Brain

33

stimulus, then they should garner more information about the stimulus if it is presented during sustained attention than during attention termination. Control trials were included that contained no-familiarization stimulus exposure, or exposure to the stimulus for 20s. The amount of information processing was estimated with a paired-comparison recognition memory procedure (Fagan, 1974). This procedure tests recognition memory by presenting the infant with familiar and novel stimuli. Recognition memory is inferred if the infant looks longer toward the novel stimulus than the familiar stimulus, i.e., “novelty preference.” Figure 2.2 illustrates the results from this study, with the x-axis showing the duration of exposure to the familiarized stimulus during sustained attention, and the y-axis showing the novelty preference). The most obvious result in this figure is the positive linear correlation between the amount of sustained attention exposure and the resulting novelty preference measure. This was true in the trials when the stimulus was presented immediately, for the 20-s procedure, for the heart rate deceleration procedures, and when the presentation occurred 5s after attention termination occurred.

55 Immediate oO

3

50 +

45 4

20-Second

¥

Accumulation

/

, 40

7

3//

y

/

35-4

at.

a

4

=

5

Heart Rate Deceleration. ©

/

/

ae

_-* “

/

< 3

2

2

;/

=

c 2 S

we

'

/ Return of HR to Prestimulus + 5-S /

=E 3



©

ore

No-Exposure Control

/

ae

a“

A

_7” Return of HR to Sao

30

T 0.0-0.5s

gt

~~ T

0.6-2.5s

Prestimulus

T

Tt

T

2.6-5.0s

7 >5.0s

Duration of Familiar Stimulus Exposure during Heart Rate Deceleration

Figure 2.2. Novelty preference (recognition memory) as a function of familiar stimulus exposure during heart rate deceleration (sustained attention) for 3- to 6-month-

old infants. Source: From “Effects of attention on infants’ preference for briefly exposed visual stimuli in the paired-comparison recognition-memory paradigm,” by J. E. Richards, 1997, Developmental Psychology, 33, pp. 28. Copyright 1997 by American Psychological Association.

34

Infant Perception and Cognition

There are two general points to be made from this figure. First, the positive relation between sustained attention and novelty preference occurred even when the infant was exposed to the stimulus for 2.5, 5, or 20s. That is, it was the quality of processing during stimulus exposure that affected the subsequent recognition, not the total quantity. Five seconds of sustained attention in the brief exposure conditions resulted in as much recognition memory as five seconds of sustained attention, in addition to 15s of non-sustained attention (20-s condition). Second, stimulus presentation during attention termi-

nation (“Return of HR to Prestimulus”) actually resulted in longer fixation times toward the familiar stimulus than toward the novel stimulus. This was interpreted as evidence that the attention-termination phase was inhibiting processing of the stimulus (or, allowing only partial information processing) so that the infant actually shows a “familiarity preference.” This implies that there is a separate attention phase beyond “attention-holding,” where fixation is held toward the stimulus, but information processing does not occur. This study confirms Cohen’s proposition that there are multiple types of attention during the course ofa look toward a stimulus, that information processing occurs primarily during “attention-holding,” but adds additional phases of attention beyond the attention-holding stage. Many studies have been done that show this association between information processing and sustained attention. | have reviewed my own work on this issue in several places (Reynolds & Richards, 2007; Richards, 2007). Several of these studies have included a direct measure of information processing, such as recognition memory, and show a similar relationship between sustained attention and the amount of recognition memory, A number of studies also have shown the selective aspect of attention (“It is the taking possession by the mind, in clear and vivid form, on one out of what seem several possible objects or trains of thought."—James, 1890). This is shown, for example,

in a lack of distractibility from looking at a central stimulus by a peripheral stimulus during sustained attention, and not during stimulus-orienting or attention-termination phases (e.g., Richards,

1987). I also have shown that

there is a “top-down” influence of sustained attention on reflex stimulusinterrupt processing, exogenous orienting toward peripheral stimuli, and in eye-movement systems involved in saccadic- and smooth-pursuit tracking. These studies affirm Cohen’s conclusion regarding the processing that occurs during “attention-holding” and the relation of the attention-holding process to several interesting psychological variables. INFANT “ATTENTION-HOLDING” IS NONSPECIFIC AROUSAL: BRAIN BASIS “Back when Les and I were young (in the 1960s and 1970s), my slogan was ‘The way to the head is through the heart’ because I was using heart rate to indicate what was going on in the baby’s head.” (Keen, 2008) How does the brain operate to control infant attention? There has been a general feeling that research on infant attention tells us something about what

Infant Attention, Arousal, and the Brain

35

is happening inside the baby’s head. Many researchers interested in infant attention and its development have moved heavily toward neurodevelopmental models of attention. In this section I will review a model of the neural basis for the heart-rate-defined attention phases. In the following section I will review how these attention phases might influence other types of cortical processing.

The general body of literature on infant attention has not been concerned with the neural basis of attention. On the other hand, the psychophysiological literature concerning attention has had a neural model as its founding principle. The first work using “stimulus orienting,” “orienting reflex,” “habituation,” and “sensitization” was done by Sokolov (e.g., Sokolov, 1963). Sokolov used a wide variety of physiological measures (heart rate, skin conductance, respiration) to measure the human response to environmental stimuli. Sokolov argued that the initial presentation of a “novel” (or not recently presented) stimulus produced a conflict between a “neural model” of the current environment and the sensory processes occurring in the brain. This conflict resulted in an “orienting reflex” that was reflected in a wide variety of physiological systems, e.g., skin conductance changes, heart rate deceleration or acceleration. The repeated presentation of the stimulus resulted in a modification of the neural model of the environment so that the neural model matched the sensory processes, leading to a decrease in the physiological system response to the stimulus. This decreasing response is the definition of habituation. Sokolov was the first “cognitive psychophysiologist” and, through his neural model and measurement of physiological activity, maybe the first “cognitive neuroscientist.” Many researchers interested in infant attention adopted the general research program begun by Sokolov, though not necessarily his emphasis on the brain control of attention. On one hand, Cohen and many workers adopted the orienting reflex and habituation processes as tools to study a wide variety of infant cognitive processes. However, they were unconcerned with the “neural model” and did not use physiological measures routinely in their work. On the other hand, there has been a continuing use of psychophysiological experiments with infant participants. This was probably generated by Graham and Clifton’s (1965; nee Rachel Keen) review of the bidirectional nature of heart rate. And, there has been a continuing use of heart rate to distinguish attention types for infant participants (e.g., see reviews by Reynolds & Richards, 2007; Richards, 2007). I have presented a model for explaining the relation between the heartrate-defined attention phases and neural processes (Richards, 2001, 2007)

and I will briefly review this here. One aspect of attention hypothesized by cognitive neuroscience is the arousal associated with energized activity (Posner, 1995). Arousal in this sense is the enhanced behavioral performance when attention is aroused, and not emotional arousal or sleep-state arousal. This arousal function is controlled by specific and distinct neural systems. In particular, the noradrenergic and cholinergic neurotransmitter systems control the arousal aspect of attention. Figure 2.3 shows the nuclei and distribution of these two neurotransmitter systems. These brain processes have

36

Infant Perception and Cognition

large nuclei centers in the locus coeruleus and reticulated mid-brain area (noradrenergic) and basal forebrain (cholinergic). These neurons have long axonal processes that ascend and have terminal endings throughout the cortex, have a direct influence on the thalamus, and have descending connections to several midbrain systems. When these neurons are active, they release the relevant neurotransmitter in the brain areas where the terminals end. These neurotransmitters are then available for increased neural efficiency of the target areas. Thus, the brain areas receiving the neurotrans-

mitters act more efficiently when these neurotransmitter systems are active. These two neurotransmitter systems are active in response to novel incoming sensory stimuli, as well as from top-down cortical influences. This arousal system “invigorates” or “energizes” cognitive processes, leading to increased processing efficiency, shorter reaction times, better detection, and sustaining of cognitive performance for extended periods of time. The heart-rate-defined attention phases are markers of these two arousal systems (Richards, 2001, 2007). The neural control of heart rate originates from cardioinhibitory centers in the orbitofrontal cortex, via the vagal nerve, to the cardiac pacemaker neurons (Figure 2.3; also see Reynolds & Richards, 2007). When the arousal system is active, reciprocal connections between the two arousal neurotransmitter systems and the cardioinhibitory centers results septal

fornix

cingulum

medial

,

.

forebrain bundle

~*~

basal

to the amygdala

to the

hippocampal

Cholinergic cardioinhibitory from vagal nerve

formation

Tonia spinal cord

Ny Heart

Rate Change © (bpm)

6 -—4

-8 -10

T

-5

T

0

T

5

10

15

a

20

25

Seconds following stimulus onset

Figure 2.3. The cholinergic (upper left) and noradrenergic (upper right) neurotransmitter systems of the brain, These systems have ascending influence on the brain, and descending influence on the cardiac pacemaker cells of the heart. Heart rate changes in infants reflect the arousal properties of these neurotransmitter systems. (See also figure in plate section.)

Infant Attention, Arousal, and the Brain

37

in a large heart rate deceleration in infants.Thus, the initial heart rate deceleration occuring, defining stimulus orienting, marks the onset of the brain arousal system; the continuing lowered heart rate marks a sustaining of the brain arousal; and the return of heart rate to its prestimulus level represents the end of the brain arousal processes. Thus, one might use the terms “attention-holding,” “sustained attention,” “arousal,” and “attentiveness” to refer to the enhancing aspect of this neural arousal system. “Attention termination,” “inattention,” and “inattentiveness” refer to the times when this arousal system is inactive.

INFANT AROUSAL AFFECTS SPECIFIC BRAIN ATTENTION SYSTEMS: DIRECT BRAIN MEASURES “Now I say, “The way to the head is through the hand.”



Keen, 2008

The first three sections of this chapter reviewed behavioral and _psychophysiological studies showing that infant attention consists of multiple phases, controlled by different processes, which have differing levels of information processing. This research was consistent with the conclusions of Cohen (1972) and his evaluation of the significance of his research. The prior section reviewed a model of the neural control of these attention phases. The “functional” part of this model is based on psychophysiological recording, an understanding of the neural control of the physiological indicator of attention, and the demonstration that behavioral processes are differentially affected by the status of attention as measured by the physiological measures. The “neural” part of this model is based upon work with animals, inferences about the physiological system used to measure attention, and the behavioral tasks. These make a coherent explanation of the neural basis of attention. However, they lack direct measures of brain activity. This section will review the effect of the attention phases on direct measures of neural activity, It will also present the first attempts to localize where “inside the baby’s head” these effects take place. Researchers interested in infant cognitive processes often place their work in the context of neural control of behavior. These models often use “marker tasks” for measurement of neural activity rather than direct measures (“The

way to the head is through the hand."—Keen, 2008). Marker tasks are behavioral tasks that have been studied in animal models or invasive preparations, and which have known neural control (Johnson, 1997; Richards, 2001, 2007; Richards & Hunter, 2002). Developmental changes in these behavioral tasks should reflect developmental changes in the brain area(s) that control their

functioning. Marker tasks are therefore useful in “developmental cognitive neuroscience” models of behavior development. However, these measures, including psychophysiological measures such as heart rate, only provide indirect measurement of brain activity. There are several reasons to be cautious about the use of these tasks in neurodevelopmental models of attention (Richards, 2001, 2007, 2010; Richards & Hunter, 2002).

38

Infant Perception and Cognition

One psychophysiological measurement provides a direct measure of neural activity: the electroencephalogram (EEG), or scalp-recorded event-related potentials (ERP). The EEG is measured as electrical potential changes on the scalp, and ERPs are EEG activity that is time-locked to experimental events or cognitive processes. This electrical activity on the scalp is generated by extracellular neural tissue and neural synaptic potentials, probably excitatory post-synaptic potentials, There are times when a large number of neurons in a small area of brain tissue fire relatively synchronously, and the neurons are oriented in the same direction in relation to the skull. When this occurs,

the current generated by the synaptic potential summate and current flows through the cortex, cerebralspinal fluid (CSF), meninges, skull, and skin, and can be measured as changes in electrical potential on the scalp, These are what EEG is measuring. Therefore, the EEG is a direct measure of the temporal flow of underlying neural activity, and the extent of synchronized neural activity occurring in discrete brain areas. I, along with colleagues Greg Reynolds and Mary Courage, have a series of studies showing the effect of the heart-rate-defined attention phases on infant ERP. These studies used “oddball” tasks first used with infant participants by Courchesne (1977, 1979; Courchesne, Ganz, & Norcia, 1981) and later modified

by Nelson and colleagues (e.g., Nelson & Collins, 1991, 1992). In the oddball procedure, one stimulus of a brief duration (500 ms) is presented relatively frequently (“standard stimulus”) and a second stimulus is presented infrequently (“oddball”). In adults, the presentations result in a positive-going ERP component about 300 ms following stimulus onset (P300, or P3), which is larger to the oddball than to the standard stimulus. The studies with infants do not find the P300, but instead report a large negative ERP component occurring about 500 ms following stimulus onset, located primarily in the frontal and central leads, which is larger to the oddball stimulus. This has been labeled the Nc (“negative central”) component. A modification of this procedure is to present a familiar stimulus frequently, a familiar stimulus infrequently, and a series of novel stimuli that are presented infrequently (Nelson & Collins, 1991, 1992).

In this case, often the infrequent and frequent familiar stimuli show the same Ne response, and the infrequently presented novel stimuli result in a larger Ne response. Figure 2.4 shows the ERP recording from this procedure (Reynolds, Courage, & Richards, in press). The infrequent familiar and infrequent novel stimuli resulted in a similar magnitude Nc component peaking about 300-400 ms following stimulus onset, and a sustained Nc for the infrequent novel stimulus lasting for another 200 to 300 ms. The Nc is thought to be a measure of orienting toward the stimulus, based upon a primitive recognition memory system discriminating the familiar and novel stimuli (Nelson, 1994; Richards, 2003). We (Reynolds & Richards, 2005, 2007; Reynolds, Courage, & Richards, in press; Richards, 2003; also see review by Reynolds & Richards, 2007) have modified this procedure to study the heart-rate-defined attention phases. As described in a prior section, for other studies (Richards,

1997) we first pres-

ent a stimulus that elicits the heart rate changes marking stimulus orienting,

of}

*(joAou juanbasyut serpruey yuNbosyur “eres

Zieyaueg

z1esua

Ziewoay—

ArisL-

SANISL

JPAON quaNbasyuj seynuey quanbasyuy seyuuey yuanbasy

oO

-O-Q

Arg By

:

\

Zleiaued

Zies ua) [

“pz aunb14

juoNbasy) 9dA] uONeyUasaad snpnurnys Jaisg pur (dpeds yo

Og

5-2 56

SO

INQ oP

t

ynoKeq JaNJOSUaS

‘dda SNINWIS-1sog ‘suONeyUasaig SNINWIYS jalg

\

Jajuad ‘dyeds Jo JUOI]) UOLEI0] SuLpsodas Aposqdo]9 JO UOTIUN] & sv ‘s}URTUL P[O-YUOU-G7zZ 0) -Gp Woy (qyy) speNUaod-poyepas-yuaay

sus o00ry,


>

Frequent Familiar -10 mV

Noe iZ

infrequent Familiar

ye

wa

Infrequent Novel +10 mV

Figure 2.5. Topographical scalp potential maps as a function of brief stimulus presentation type and attention phase (top figures: attentive; bottom figures: inattentive). Source: From “Attention affects the recognition of briefly presented visual stimuli in infants: An ERP study,” by J. E. Richards, 2003, Developmental Science, 6, p. 319. Copyright 2003 by Wiley-Blackwell. Reprinted with permission. (See also figure in plate section.)

Infant Attention, Arousal, and the Brain

41

sustained attention? The interpretation of the attention is implied from the interpretation given in the previous section. The neurotransmitter systems

controlling arousal are active, leading to an increased amount of the noradrenergic and cholinergic neurotransmitters being present in the cortex, leading to enhanced brain activity. The amplitude of the ERP is not interpreted as necessarily indicating “more” brain activity, though it might. Rather, the higher level of electrical potential measured on the scalp likely reflects increased synchronization of the activity at specific locations in the cortex that are suitably aligned to have current flow reach the scalp. Thus, it is not just that more cortical activity occurs when the arousal system is active, but that there is an increased efficiency of specific neural locations during this arousal. WHAT'S INSIDE A BABY’S HEAD? OR, WHERE IS ATTENTION INSIDE A BABY’S HEAD? Nearly all developmental psychologists acknowledge the importance of the brain in influencing behavior. It is also acknowledged that developments occurring in the brain in young infants may be largely responsible for causing behavioral development. However, as a field we have been content with measurements of brain activity outside the head (direct measures) or on the body (marker tasks) and have used such measurements to infer what is inside

the head. This has changed! The field of infant attention has moved to incorporate models from neuroscience, neural development, and neuroimaging, to study attention. Behavior has not been left behind—rather, changes in attention behaviors are now explained with developmental changes in the brain, neural processes, or the reciprocal effect of neural development with experiential input. These models are often labeled “developmental cognitive neuroscience.” My most recent work has been to use MRI neuroimaging to measure brain structure inside the head, and relate this to neural activity measured with outside-the-head measures. I recently summarized two ways in which information has been obtained to study brain development in infants (Richards, 2010). This information pri-

marily comes from nonhuman animal models of brain development, primarily primates. The majority of our knowledge of the patterns and characteristics of brain development comes from the study of normally developing nonhuman animals. An advantage of this approach is that invasive neural techniques and rigorous experimental control may be used with nonhuman animals that cannot be used with human infants. However, a strong disadvantage to this approach is that it assumes a correlation can be made between ages of nonhuman animal and human infants, that changes in the brain are isomorphic across species and across brain areas, that psychological processes and the changes in these are similar in human and nonhuman animals, and that the complexity of the human brain does not affect the comparability of either brain or behavior development in human and nonhuman species. I assert that each of these disadvantages could enormously affect our neurodevelopmental models, and, unless we have some direct measure of brain development in

42

Infant Perception and Cognition

normal human infants, we cannot know to what extent such incompatibilities exist within nonhuman-animal-model-derived neurodevelopmental models of infant cognitive processes.

The second source of information comes from postmortem studies of young infants. The most well known of these studies is a series of autopsy studies by Conel (1939-1967), who studied the human cerebral cortex. Conel laid out a well-articulated pattern of neuroanatomical and cytoarchitectural change in human infants, Conel’s work is more applicable to humans because he used humans, but it has weaknesses. Longitudinal growth patterns cannot be studied in postmortem studies, it is assumed that the individuals measured at different ages are representative of normal individuals, studies are limited to small samples, and it is not always clear that the cause of death is entirely unrelated to psychological or behavioral changes that may have occured. Notwithstanding any benefits or deficits that the study of postmortem human infants, or invasive studies with nonhuman animals, may bring, both techniques fail to provide the neurodevelopmental status of any particular human infant. Thus they cannot be used to relate the status of that infant’s brain development to the infant's behavioral-developmental status. I have been using structural (anatomical) MRI with human

infants who

also participate in psychophysiological studies of attention (see presentation in Richards, 2010). The information obtained from the specific individual’s neurodevelopmental status can be compared to neural activity measured with the EEG/ERP, or behavioral indices of development (novelty preference). This allows us to “look inside the baby’s head” directly for information about brain control of cognitive processes, I have described this work in several places (Reynolds & Richards, 2009; Richards, 2007, 2010). My use of the structural MRI scans has been to determine the location in which attention is affecting the Nc ERP component in the modified oddball tasks. I previously described the effect of the heart-rate-defined sustained attention on the Ne ERP components as likely reflecting the increased synchronization of the activity at specific locations in the cortex. These locations encompass enough cortical area, and are suitably aligned, to produce an electrical current that flows through the materials of the head to the scalp. A technique called “cortical source analysis” (brain electrical source analysis; Reynolds & Richards, 2007, 2009; Richards, submitted) uses the amplitude and topographic distribution of the EEG on the scalp to infer the brain area(s) that generate the electrical current. The location of the current source and the activity of the current source over time may be calculated. The steps in this analysis using the structural MRI are as follows: First, the structural MRI scan is done. Figure 2.6 (upper left) shows a

single slice

of an MRI volume taken from a 7.5-month-old infant. Second, the MRI volume is segmented into component parts, such as gray matter, white matter, CSF, skull, scalp. Figure 2.6 (left, middle panel) shows the areas from the MRI slice with colors representing the segmented material. Third, a computer file is made, called a “wireframe,” that consists of tetrahedra mapped with the location and material type for the entire head (Figure 2.6, bottom left). Finally, the

Infant Attention, Arousal, and the Brain

43 Control

Posterior Prefrontal



Py

SA

a

Or"

'

Front Pole

HQT

en

ey ey

@

Anterior Cingulate

fp

ay

ery

vr

TN

Inferior Prefrontal x mS

UA eer

sting Po seer gly

Figure 2.6. Cortical source analysis of the Nc ERP component with realistic models of infant heads via structural MRI of infant participants. The left figures represent steps in the cortical source analysis technique, the mid-left figure potential cortical source locations for the Nc, the mid-right panels the topographical potential maps for the projection of the sources on the scalp, and the right panels the temporal activity of different brain areas for the brief stimulus presentation types. (See also figure in plate section.)

wireframe file may be used with computer programs to do the cortical source analysis. This aspect of this work is “neuroimaging.” Figure 2.6 shows some results from a study of infant recognition memory using the modified oddball procedure (Reynolds, Courage, & Richards, in press). The head with spots on it is a representation of the saggital view of the likely locations for the current sources of the Nc ERP component. These locations were identified with the cortical source analysis of the ERP in the modified oddball task. The locations have been grouped into distinct neuroanatomical areas, including inferior prefrontal, frontal pole, anterior cingulate, posterior-superior prefrontal, and central. The dipoles identified by the cortical source analysis then may be used in a quantitative model to generate current that is projected to the scalp surface, and drawn as topographical scalp potential maps. The column of topographical scalp potential maps shown in the right middle section of Figure 2.6 represents these projections from the dipoles in the specified areas onto the scalp. The topographical potential maps from the inferior prefrontal and anterior cingulate match the topographical scalp potential map of the ERP in this study, and, to a lesser extent, so does the posterior-superior prefrontal projections. Finally, the activity of the dipoles

44

Infant Perception and Cognition

may be plotted across time in the same time resolution as the ERP (Figure 2.6, far right). This temporal activity is presumed to represent the temporal extent of the neural activity for the brain area(s) generating the Nc ERP component. The representation of the dipoles inside the head, and the projection of the electrical currents on the scalp, is the “spatial” aspect of this work, and the neural activity unfolding in time is the “temporal” aspect. So far, we have “spatiotemporal infant neuroimaging” of the Nc ERP component. The most important aspect of this work to cognitive development is the functional relation of the putative neural activity to experimental events or cognitive processes. Recall that the Nc is hypothesized to represent the orienting of attention based on a primitive recogniton of the novel stimuli vis-a-vis the familiar stimuli. We can examine the activity of the cortical sources in relation to the experimental events. Figure 2.7 (top figure) shows the activity of the dipoles located in the inferior prefrontal cortex as a function of time and the three stimulus testing conditions. The activity shows a large deflection in the negative direction at about 500 ms, This represents enhanced neural

Inferior Prefrontal

te!re

Infrequent Novel

Prefrontal

Familiar Stimuli

— Attention —— Inattention

10)

Prefrontal

Novel Stimuli

co

Vv

10 |

-10

Figure 2.7. The temporal activity of the cortical current sources for the inferior prefrontal brain areas, as a function of the brief stimulus presentation procedures (top figure); effect of brief stimulus familiarity (left bottom figure, familiar stimuli; right bottom figure, novel stimuli) and attention (solid line, attentive; dashed line, inattentive) on the temporal activity of the current sources located in the prefrontal cortex. Source: Adapted from “Familiarization, attention, and recognition memory in infancy: An ERP and cortical source localization study, “ by G. D. Reynolds and J, E. Richards, 2005, Developmental Psychology, 41, p. 608. Copyright 2005 by American Psychological Association.

Infant Attention, Arousal, and the Brain

45

activity in this area, projected on the scalp as a negative electrical potential. This activity is larger for the novel stimulus than for the two familiar stimuli. We believe that this represents one of the areas of the brain that generates the Nc. The relation to the stimulus conditions affirms the functional significance of the cortical sources in this area of the brain. The activity of the cortical sources can also be related to cognitive processes. Figure 2.7 (bottom figures)

show the summed activity from dipoles located in the inferior prefrontal cortex, anterior cingulate, and posterior-superior prefrontal cortex, as a function of stimulus familiarity and attentiveness (Reynolds & Richards, 2005). The

figure on the lower left shows that the ERP response to the familiar stimuli was not affected by attention status. On the other hand, when the infant was attending, there was a large increase in the activity of the cortical dipoles nearly immediately upon stimulus presentation that lasted for several hundred ms (Figure 2.7, bottom left, solid line). This increased activity represents the enhanced and efficient processing occurring in this cortical area when the arousal system is activity. We have a “spatiotemporal functional neuroimaging” technique to investigate infant attention! SUMMARY

AND FUTURE DIRECTIONS

This chapter reviewed work being done on infant attention that was inspired by the early work of Cohen (e.g., Cohen, 1972). Several aspects of this work were directly influenced by a conception of the sequential unfolding of multiple attention phases, first explicitly summarized by Cohen. Some recent work has been to examine some “stimulus” variables that affect infant attention development, including the role of attention in controlling extended visual fixations in television program viewing (Courage, Reynolds, & Richards, 2006; Hunter & Richards, submitted; Richards & Anderson, 2004; Richards & Cronise, 2000; Richards & Gibson, 1997; Richards & Turner, 2007). This work continues to find aspects of infant attention both consistent with Cohen’s early views on attention, and aspects of attention that were unanticipated in 1972. The second aspect of my work that has been reviewed in this chapter is the development of neurodevelopmental models of infant attention. I reviewed an explanatory neural model for the heart-rate-defined attention phases, as well as some work that looks “inside the baby’s head” to find how the brain arousal system, measured by heart rate changes, affects neural processes involved in cognition and attention. Developmental changes in the brain can now be related to developmental changes in attention, perception, cognition, and behavior with the “spatiotemporal functional neuroimaging” techniques Iam developing. This neuroimaging work requires further advances. Grey Reynolds and | describe in some detail an application of cortical source analysis of ERP to infant participants (Reynolds & Richards, 2009). We note in that chapter that the MRI recording done with infant participants will go a long way to eliminating some deficits of this approach. Specifically, there are topological characteristics of the infant’s brain inside the skull that differ dramatically

46

Infant Perception and Cognition

from adult participants. This requires the application of cortical source analysis with realistic head models from infant participants (Richards, submitted). It also is the case that infant head media differ from those of adults. For

example, the impedances of skull and scalp are much larger in adults than in infants, and a substantial portion of the axons are unmyelinated at birth. ‘The realistic models being used with the MRI recording will allow tests of the effect that these parameters have on cortical source analysis for infant participants. We believe that a large library of structural MRIs done on human infants, the “NIH MRI Study of Normal Brain Development” (Almli, Rivkin, & McKinstry, 2007; Evans, 2006; NIH, 1998) may provide the raw material for examining the characteristics of the head media in infants, and stimulate work that relates brain development in individual participants to the development of attention, perception, cognition, and behavior.

REFERENCES Almli, C. R,, Rivkin, M. J., & McKinstry, R. C. (2007), The NIH MRI study of normal brain development (Objective-2): Newborns, infants, toddlers, and preschoolers, Neurolmage, 35, 308-325. Berg, W. K., & Richards, J. E. (1997). Attention across time in infant development. In P. J. Lang, R. F. Simons, & M. T. Balaban (Eds.), Attention and orienting: Sensory and motivational processes (pp. 347-368). Mahwah, NJ: Erlbaum. Brennan, W. M., Ames, E. W,, & Moore, R. W. (1966). Age differences in infants’ attention to patterns of different complexities. Science, 151, 354-356, Cohen, L, B. (1972). Attention-getting and attention-holding processes of infant visual preferences. Child Development, 43, 869-879. Conel, J. L. (1939). Postnatal development of the human cerebral cortex: The cortex of the newborn (Vol, 1). Cambridge, MA: Harvard University Press. Conel, J. L. (1941). Postnatal development of the human cerebral cortex: The cortex of the one-month infant (Vol. 2). Cambridge, MA: Harvard University Press. Conel, J. L, (1947). Postnatal development of the human cerebral cortex: The cortex of

the three-month infant (Vol. 3), Cambridge, MA: Harvard University Press. Conel, J. L. (1951). Postnatal development of the human cerebral cortex: The cortex of

the six-month infant (Vol. 4.) Cambridge, MA: Harvard University Press. Conel, J. L. (1955). Postnatal development of the human cerebral cortex: The cortex of

the fifteen-month infant (Vol. 5). Cambridge, MA: Harvard University Press. Conel, J, L. (1959). Postnatal development of the human cerebral cortex: The cortex

of the twenty-four-month infant (Vol. 6). Cambridge, MA: Harvard University Press,

Conel, J. L. (1963). Postnatal development of the human cerebral cortex: The cortex of the forty-eight-month infant (Vol. 7). Cambridge, MA: Harvard University Press. Conel, J. L. (1967). Postnatal development of the human cerebral cortex: The cortex of the seventy-two-month infant (Vol. 8). Cambridge, MA: Harvard University Press. Courchesne, E. (1977), Event-related brain potentials: Comparison between children and adults. Science, 197, 589-592. Courchesne, E. (1978). Neurophysiological correlates of cognitive development: Changes in long-latency event-related potentials from childhood to adulthood. Electroencephalography and Clinical Neurophysiology, 45, 468-482.

Infant Attention, Arousal, and the Brain

47

Courchesne, E., Ganz, L., & Norcia, A. M. (1981). Event-related brain potentials to

human faces in infants. Child Development, 52, 804-811. Courage, M. L., Reynolds, G. D., & Richards, J. E. (2006). Infants’ visual attention to patterned stimuli; Developmental change from 3- to 12-months of age. Child Development, 77, 680-695. Evans, A. C. (2006). The NIH MRI study of normal brain development. Neurolmage 30, 184-202.

Fagan, J. F. (1974). Infant recognition memory: The effects of length of familiarization and type of discrimination task. Child Development, 59, 1198-1210. Frick, J., & Richards, J. E. (2001). Individual differences in recognition of briefly

presented visual stimuli. Infancy, 2, 331-352. Graham, F. K. (1970). Cardiac orienting responses as a function of age. Psychonomic Science, 19, 363-365. Graham, F. K. (1979). Distinguishing among orienting, defense, and startle reflexes. In H, D, Kimmel, E, H. van Olst, & J. F. Orlebeke (Eds.), The orienting reflex in humans (pp. 137-167). Hillsdale, NJ: Erlbaum. Graham, F. K. (1992). Attention: The heartbeat, the blink, and the brain. In

B. A. Campbell, H. Hayne, & R. Richardson (Eds.), Attention and information processing in infants and adults (pp. 3-29). Hillsdale, NJ: Lawrence Erlbaum. Graham, F. K., Anthony, B. J., & Zeigler, B. L. (1983). The orienting response and developmental processes. In D. Siddle (Ed.), Orienting and habituation: Perspectives in human research (pp. 371-430). Sussex, England: John Wiley. Graham, F. K., & Clifton, R. K. (1966). Heart rate change as a component of the orienting response. Psychological Bulletin, 65, 305-320. Horowitz, F. D., Paden, L., Bhana, K., & Self, P. (1972). An infant-control procedure for studying infant visual fixations. Developmental Psychology, 7, 90. Hunter, S. K., & Richards, J. E. (submitted). Development of eye movements in young infants in response to complex dynamic stimuli. Johnson, M. H. (1997). Developmental cognitive neuroscience. London: Blackwell, Kagan, J., & Lewis, M. (1965). Studies of attention in the human infant. MerrillPalmer Quarterly, 11, 95-127. Keen, R. (2008). Planning sequential actions in early childhood. Paper presented at the Festschrift for Leslie B. Cohen, International Conference on Infant Studies, Vancouver, Canada. Nelson, C. A, (1994). Neural correlates of recognition memory in the first postnatal year. In G. Dawson & K. W. Fischer (Eds.), Human behavior and the developing brain (pp. 269-313). New York: Guilford Press. Nelson, C. A., & Collins, P. F. (1991). Event-related potential and looking-time analysis of infants’ responses to familiar and novel events: Implications for visual recognition memory. Developmental Psychology, 27, 50-58. Nelson, C. A., & Collins, P. F. (1992), Neural and behavioral correlates of visual recognition memory in 4- and 8-month-old infants. Brain and Cognition, 19, 105-121. NIH (1998). Pediatric study centers (PSC) for a MRI study of normal brain

development. NIH RFP NIHNINDS 98-13, sponsored by National Institute of Neurological Disorders and Stroke, National Institute of Mental Health, National Institute of Child Health and Human Development. Porges, S. W. (1976). Peripheral and neurochemical parallels of psychopathology: A psychophysiological model relating autonomic imbalance in hyperactivity,

48

Infant Perception and Cognition

psychopathology, and autism. In H. Reese (Ed.), Advances in child development and behavior (Vol. 11, pp. 35-65). New York: Academic Press. Porges, S. W. (1980). Individual differences in attention: A possible physiological substrate. In B, Keogh (Ed.), Advances in special education (Vol.2, pp. 111-134). Greenwich, CT: JAIL.

Posner, M. I. (1995). Attention in cognitive neuroscience: An overview, In M.S. Gazzaniga (Ed.), Cognitive neurosciences (pp. 615-624). Cambridge, MA: MIT Press.

Reynolds, G. D., Courage, M., & Richards, J. E. (in press). Infant attention and visual preferences: Converging evidence from behavior, event-related potentials, and cortical source localization. Developmental Psychology. Reynolds, G. D., & Richards, J. E. (2005). Familiarization, attention, and recognition memory in infancy: An ERP and cortical source localization study, Developmental Psychology, 41, 598-615. Reynolds, G. D., & Richards, J. E. (2007). Infant heart rate: A developmental psychophysiological perspective. In L. A. Schmidt & S. ]. Segalowitz (Eds.), Developmental psychophysiology (pp. 173-210). New York: Cambridge Press. Reynolds, G. D., & Richards, J. E. (2009). Cortical source analysis of infant cognition. Developmental Neuropsychology, 3, 312-329.

Richards, J. E. (1987). Infant visual sustained attention and respiratory sinus arrhythmia. Child Development, 58, 488-496. Richards, J. E. (1997). Effects of attention on infants’ preference for briefly exposed visual stimuli in the paired-comparison recognition-memory paradigm. Developmental Psychology, 33, 22-31. Richards, J. E. (2001). Attention in young infants; A developmental psychophysiological perspective. In C. A. Nelson & M. Luciana (Eds.), Developmental cognitive neuroscience (pp. 321-338) Cambridge, MA: MIT Press. Richards, J. E. (2003). Attention affects the recognition of briefly presented visual stimuli in infants: An ERP study. Developmental Science, 6, 312-328. Richards, J. E. (2007). Attention in young infants: A developmental psychophysiological perspective. In C, A. Nelson & M, Luciana (Eds.), Developmental cognitive neuroscience (pp. 479-489). Cambridge, MA: MIT Press.

Richards, J. E. (2010). Attention in the brain in early infancy. In S. Johnson (Ed.), Neoconstructivism: The new science of cognitive development (pp. 1-36).

New York: Oxford University Press. Richards, J. E. (submitted) Cortical sources of ERP in the prosaccade and antisaccade task using realistic source models based on individual MRIs. Richards, J, E., & Anderson, D. R. (2004). Attentional inertia in children’s extended looking at television. Advances in Child Development and Behavior, 32, 163-212.

Richards, J. E., & Casey, B. J, (1991), Heart rate variability during attention phases in young infants. Psychophysiology, 28, 43-53. Richards, J. E., & Casey, B. J, (1992). Development of sustained visual attention in the human infant. In B. A. Campbell, H. Hayne, & R. Richardson (Eds.), Attention and information processing in infants and adults (pp. 30-60). Mahway, NJ: Erlbaum. Richards, J. E., & Cronise, K. (2000). Extended visual fixation in the early preschool years: Look duration, heart rate changes, and attentional inertia. Child Development, 71, 602-620,

Infant Attention, Arousal, and the Brain

Richards, J. E., & Gibson, T. L. (1997). Extended visual fixation in young infants:

Look distributions, heart rate changes, and attention. Child Development, 68, 1041-1056. Richards, J. E., & Hunter, S, K. (2002), Testing neural models of the development of

infant visual attention. Developmental Psychobiology, 40, 226-236. Richards, J. E., & Turner, E. D. (2001). Distractibility during extended viewing of television in the early preschool years. Child Development, 72, 963-972. Sokolov, Y. N. (1963). Perception and the conditioned reflex. New York: Pergamon Press.

49

3 A Constructivist View of Object Perception in Infancy Scott P. Johnson

This chapter describes a body of research whose goal is to understand and explain the development of object perception in infancy. In some ways this is a Sisyphean task, for two reasons: first, “object perception” resists a simple, straightforward definition; and second, the more I have learned about object perception in infancy, the more I have come to realize how far we are from this goal. There is a great deal to be learned about cortical development, visual perception, oculomotor and manual action systems, intermodal perception, learning and memory, spatial cognition, social cognition (e.g., face perception), and other developmental phenomena in infancy, all of which are part of object perception and all of which work in concert to impart the experience we (and infants, at some point) have of objects in our surroundings. Nevertheless, I believe that significant progress has been made recently in understanding the developmental processes by which infants perceive objects in like manner to adults. These developmental processes, described in greater detail subsequently, provide clear evidence in favor of a constructivist view. Representations of objects as complete and coherent entities, persisting across space and time, emerge in the first several months after birth, In part, this occurs via processes of active assembly; self-initiated oculomotor and manual activities that facilitate development of perceptual completion, a key component of object perception. Active assembly can be understood best within an information processing framework, by considering the information available to the infant for object boundaries through vision and tactile sensory systems, and the means by which the infant pieces together this information into a coherent structure. These two means of processing available information (information uptake and information organization), and their development, will be described in detail subsequently. First, I will present in briefa picture of object perception as a psychological construct, and how it might be decomposable into tasks amenable to empirical investigation in infancy. Second, I will describe several theoretical accounts that inform questions of infants’ object perception. Third, I will summarize recent research, from my laboratory and others, that has examined in detail Preparation of this article was supported by NIH grants RO1-HD40432 and RO1-HD48733.

52

Infant Perception and Cognition

the mechanisms of active assembly. Finally, I will consider the state of our knowledge of infants’ object perception in light of recent advances in modeling, and discuss briefly the “bigger picture,” asking in particular what remains to be determined. OBJECT PERCEPTION The input to the visual system, and our subjective experience of the visual world, are quite different. The visual input under most circumstances is continuous and unbroken across the retina. There are no gaps in the input, and few photoreceptors go unstimulated as our eyes take in visual information. Our subjective experience of the visual environment, in contrast, is one of largely empty space interspersed with objects at various distances. Subjective experience is at odds with the visual input in a second way. Objects, though generally experienced as having a regular, solid shape, most often cannot be seen in their entirety because of occlusion—occlusion of far objects by nearer ones, and self-occlusion of the far sides of individual objects due to opacity. Several steps are involved in the processing of visual input leading to the subjective experience of objects. First, input from a visual scene reaching different parts of the retina must be coded according to variations in color, luminance, motion, shape, orientation, distance, and so forth. Next, the outputs of these processes must be recombined into structured units—the building blocks of objects—and, as appropriate, units must be perceived as complete across space and time despite gaps in perception. These gaps may be due to occlusion, to movement of the observer, or to movement in the environment. ‘The process of filling in the gaps is known as perceptual completion, and it includes deduction of 3D shape from limited views due to self-occlusion. Next, higher-order visual processing is performed as necessary, such as recognition of objects, categorization, tracking identity of objects over time, and planning relevant actions based on perceived affordances and the needs of the observer, Object perception, therefore, rests on a foundation of an initial coding of visual features, followed in succession by a linking of features into units, and finally by an interpretation of units as objects that may be recognizable or otherwise relevant to the observer. This way of conceptualizing object perception maps roughly onto processes of lower-, middle-, and higher-level visual processing that has long formed the basis for investigations of visual perception in adults (e.g., Marr, 1982; Palmer, 1999),

My work has been concerned largely with development of “middle vision,” the processes by which the developing visual system assembles visual surfaces in depth from image fragments. I have been particularly interested in perceptual completion, and my work to date has examined three kinds: spatial completion, perceiving the unity of partly occluded surfaces; spatiotemporal completion, perceiving the unity of partly occluded trajectories; and 3D object completion, perceiving the 3D shape of objects seen from a limited vantage point. Development of perceptual completion stems from multiple

A Constructivist View of Object Perception in Infancy

53

mechanisms—in particular, processes of construction initiated by the child (active assembly) and described in greater detail subsequently—and also by learning, which I will not discuss in detail in this chapter, but has been presented elsewhere (Johnson, 2010; Johnson & Shuwairi, 2009; Johnson, Amso, & Slemmer, 2003).

THEORETICAL ACCOUNTS OF INFANTS’ OBJECT PERCEPTION The issue of how infants come to perceive objects can be framed within a larger question posed by Kurt Koffka (1935), who asked, “Why do things look as they do?” This larger question has its origins in the nineteenth-century theory of structuralism, espoused by Wilhelm Wundt in Germany and Edward Titchener in the United States, which was largely consistent with the views of the British empiricists such as George Berkeley, David Hume, and John Locke. The theory of structuralism held that perception arises from assembly of sensory primitives in a given sense modality, through a process of repeated associations of those primitives in time and space. These associations are presumably formed early in life from exposure to structured objects and events. In opposition to structuralist theory, theorists in the Gestalt tradition operating in the early to mid twentieth century, such as Wolfgang Kohler and Max Wertheimer (as well as Koffka) from Germany, argued that structure cannot be reduced to the sum of the parts. Rather, many configurations, such as illusory figures, have emergent properties that are inherently holistic. Gestalt theory was consistent with the views of rationalist and nativist schools of thought, exemplified by such philosophers as René Descartes and Immanuel Kant in his early writings, and held that holistic perception necessarily arises from underlying holistic processes. Holistic perception cannot arise from haphazard formation of associations by fits and starts early in life, so goes the argument, but must be unlearned—originating, for example, in the intrinsic structure of sensory systems and neural circuits in the brain. From this standpoint, therefore, intrinsic sensory and cortical structure in the visual system obviates learning and formation of associations for infants to experience a world of coherent objects. These views were complemented by additional advances in theory in the twentieth century. Three are particularly important for understanding development of object perception. The first is information processing theory, rooted in advances in computer technology, in particular the invention of machines that could be programmed to carry out a potentially infinite variety of procedures based on the instructions and the inputs that were provided. The second is the notion of “ecological optics” espoused by James Gibson (1979). Gibson

suggested that perception is best understood by examining the structure of the perceiver’s environment—for example, the information in light reflected from objects as it is received by the organism. The third is constructivism, advocated by the psychologists Richard Gregory in the UK, and Julian Hochberg and Irvin Rock in the United States—a theory about the mechanisms of perception that extract information from the environment and, importantly, fill

54

Infant Perception and Cognition

in the missing pieces via processes of inference. On a constructivist view, per-

ception of holistic structure from relatively underspecified input (as in the case of occlusion) implies a set of heuristics by which optical information leads to subjective experience, such as the “likelihood principle,” a probabilistic computation concerning which interpretation of a given scene is most likely given current retinal input and past experience (nowadays known as the Bayesian approach). Gibson’s ecological optics is sometimes considered more closely aligned with nativism than with empiricism, but this was not a principal feature of the theory. Information processing and constructivism, strictly speaking, are mute with respect to developmental origins—but, as we shall see, both are invaluable to discovering developmental mechanisms of object perception. The focus on (a) the information in the external environment, (b) the uptake

of available information, and (c) the precise mechanisms by which the organism receives and interprets the information, provides clarity with respect to understanding development of object perception in infancy, and informs the research that I will describe in the following section of this chapter. The task of the developing infant is to use his or her perceptual systems—vision, hearing, touch, and so forth—to explore the world and obtain information about its properties. The information must be attended to, encoded, stored, retrieved, and acted upon, in order to build knowledge of objects and their characteristics. How do these processes develop in infancy? Jean Piaget (1952, 1954) provided the first and, to date, the most fully realized account of the development of object perception in infancy. Piaget described processes of perceiving objects veridically as objectification: concepts of objects as distinct and separate entities, persisting across time and space, and obeying commonsense physical constraints. Objectification is rooted in the child’s recognition of her own body as an independent object, and her own movements as movements of objects through space, akin to movements of other objects she sees. Spatial reasoning and a mature object concept are revealed by changes in infants’ behavior in everyday activities and when participating in certain tasks that Piaget devised. Particularly important is the idea of active search behavior, as infants interact with objects that are partly or fully obstructed from view. Piaget outlined a number of examples, including visual search for an object whose path was unseen (but could be inferred), and later manual search, at first for partly hidden objects (as when partially occluded) and subsequently for fully hidden objects. Object concepts are constructed from the infant’s own behayiors: dropping, stroking, mouthing, searching, and observing. Through these experiences infants gain information about objects, and about the kinds of transformations that objects can undergo and still maintain their properties. Methodological advances in the 1950s and 1960s, and particularly methods based on looking times, led to views of infant object perception that were at odds with Piagetian theory and more consistent with the nativist views described previously, though not necessarily the particular views of the

A Constructivist View of Object Perception in Infancy

55

Gestalt theorists (more on this later). Studies using infant looking times as the dependent measure are founded on the assumption that infants look longer at displays and events that are novel or unusual relative to past experience (Aslin, 2007; Bornstein, 1985)—either experience provided by the experimenter in the laboratory (e.g., familiarization and habituation paradigms) or experience gained by the infant in the real world (e.g., violation-of-expectation designs). Anexample using a habituation design was provided by Spelke, Breinlinger, Macomber, and Jacobson (1992). They habituated young infants (2.5- to 4-month-olds, in different experiments) to events in which an object was moved behind a screen and was then revealed by raising the screen. This was followed by test displays in which the same object was moved behind the same screen and again was revealed (Figure 3.1). But in these test events, a wall or

table was placed behind the screen as well—and its edges could be seen protruding beyond the screen, so that the infant could see it—such that in half the events, it would impede the movement of the object. The infants were reported to look reliably longer at the test events in which the object was shown to have moved, apparently, to the other side of the wall or table—a scenario that is impossible, because objects cannot pass through each other. On the basis of these findings, Spelke et al. suggested that young infants maintain active mental representations of occluded objects under some conditions, and that such representations are part of a system of core knowledge consistent with commonsense, everyday properties of objects, such as their persistence and solidity. Core knowledge, by definition, is increasingly enriched and refined but fundamentally constant across development. The claim was made, in addition, that core knowledge is unlikely to arise in infants from perception and action experience, and instead arises from its own foundations as a system of

Habituation

Consistent

Inconsistent

Habituation

Consistent

Inconsistent

Figure 3.1. Events used by Spelke, Breinlinger, Macomber, and Jacobson (1992) to examine young infants’ representations of hidden objects. Left: habituation events. Center and right: posthabituation events are either “consistent” or “inconsistent” with an expectation of solidity and persistence of objects.

56

Infant Perception and Cognition

“initial conceptions.” This last claim was founded on negative evidence: In the absence of evidence for the means by which concepts may emerge from antecedent perception and action skills, the assumption was made that perception- and action-based accounts are unlikely to succeed, and could be disregarded. Similar criticisms have been levied against learning accounts of infant cognition (e.g., Baillargeon, 1995), and in some cases innate knowledge has been attributed to infants without considering other possibilities, simply on the basis of competence at a particular task (e.g., Wynn, 1992). A second

example

was

described

by Kellman

and

Spelke

(1983), who

habituated 4-month-olds to events where a partly occluded object (a rod) was seen to move back and forth behind a second, occluding object (a box), fol-

lowed by test displays with the rod alone moving in the same fashion as before (Figure 3.2). In these test events, the rod was presented either as whole, or as two aligned rod parts with a gap in the center. The infants were reported to look longer at the two aligned rod parts than at the whole (i.e., a preference for the “broken” vs. the “complete” rod). In this case, the looking preference was assumed to stem from a novelty preference following habituation—the complete rod being more consistent with an impression of rod connectedness in the habituation display—rather than recognition of some impossible scenario as in the Spelke et al. (1992) study described previously. In follow-up experiments, Kellman and Spelke tested the possibility that unity perception (or spatial completion) at four months is consistent with Gestalt theorists’ predictions regarding perceptual organization from individual visual cues, such as common motion, alignment, similarity, symmetry, and simplicity. In contrast to the Gestalt predictions, infants perceived unity only under restricted conditions; namely, when rod parts (or a rod part and a polygon) underwent common motion—not in static displays, or displays where the box moved in tandem with the rod, or when the box moved in front of a static rod. Kellman and Spelke concluded that young infants’ perception of objects’ boundaries and connectedness was best explained by an unlearned conception of the world, as described previously—but not in the manner envisioned by Gestalt theorists, because organization of the displays seemed to accord with only a

-\-

+ \\—

+\\—

+ \\—

Figure 3.2. Displays used by Kellman and Spelke (1983) to examine 4-montholds’ perception of object unity. Left: habituation display. Center and right: posthabituation displays are “complete” and “broken” versions of the rod seen during habituation.Adapted from Kellman and Spelke (1983).

A Constructivist View of Object Perception in Infancy

57

limited set of the cues these theorists suggested were responsible for our everyday experience of the world. Instead, displays were organized principally by motion cues. Motion cues have been suggested to be especially reliable indicators of objects’ unity or separateness, and therefore might have become part of an unlearned object-recognition system via forces of evolution (Kellman & Arterberry, 2006).

A difficulty for claims of innateness founded in infant competence is that evidence for failure of younger infants to exhibit the same level of performance casts doubt on those claims. Newborn infants have been tested for evidence of spatial completion using an habituation design very similar to that employed by Kellman and Spelke (1983), and newborns show the opposite posthabituation preference: a preference for the complete rod, implying perception of disjoint rod parts, not connected rod parts, in the rod-and-box stimulus (Slater et al., 1990). The youngest infants demonstrated to perceive spatial completion in moving rod-and-box displays are aged two months (Johnson, 2004).

There are at least two possible explanations for the data I have just described showing a change in response to partly occluded object displays between birth and 2-4 months. First, spatial completion might develop after birth, consistent with a constructivist view stipulating that percepts of whole objects are the result of integrating smaller units into larger ones; integration processes develop over time. Alternatively, it may be that a nascent object-perception system is innate, yet not functional in young infants because of limitations in detecting visual information that specifies connectedness and, in particular, common motion (Kellman & Arterberry, 1999; Smith, Johnson, & Spelke, 2003). From this standpoint, motion perception triggers a response to unity. Indirect evidence in favor of this possibility comes from experiments on motion-direction discrimination using random-dot kinematograms, which cannot be discriminated without true motion detection—otherwise, each kinematogram appears over time as a shifting collection of randomly placed dots. Infants provide evidence of motion discrimination in such displays at about 6-8 weeks after birth (Wattam-Bell,

1996), timing that is consistent

with the onset of spatial completion as described previously. I recently designed two sets of experiments to determine which of these accounts of spatial completion in infancy—the constructivist view or the nativist view—is correct. In the first set of experiments, I showed that a group of 2-month-olds successfully discriminated between rod-and-box displays with different patterns of motion, although a second group, when tested for perceptual completion in similar displays, failed to provide evidence of completion (Johnson, 2004). That is, the infants perceived motion patterns, but not unity, in the same displays. This finding is inconsistent with the possibility that unity perception stems directly from sensitivity to common motion. In the second set of experiments, my colleagues and I tested infants aged 58 to 97 days for both motion-direction discrimination and spatial completion on the same day (Johnson, Davidow, Hall-Haro, & Frank, 2008). There was a range of performance in both tasks, but no reliable correlation between the two as

58

Infant Perception and Cognition

would be expected on the nativist account. These experiments help narrow the possible explanations of development of object perception—such explanations likely will have to be consistent with a constructivist view to be successful— but do not by themselves specify the precise means by which these developments occur. This is the subject of the following section. CONSTRUCTING

OBJECTS IN DEVELOPMENT:

ACTIVE ASSEMBLY

Perceptual completion is a vital step in object perception, because there is more

to an object than typically can be seen from any one vantage point. Perceptual completion is by definition an active process of filling in the gaps in perception, and two important aspects of this process have been examined in recent experiments: (1) assembly of visible surface fragments into continuous, partly occluded surfaces—spatial completion—from oculomotor patterns, and (2) assembly of limited views of surfaces into volumetric, coherent objects—3D object completion—from visual-manual coordination. These are examples of active assembly of objects, or perceptual completion from behaviors initiated by the infant. Each is described in turn.

Spatial Completion from Oculomotor Patterns Adults and 4-month-old infants construe the occlusion display depicted in Figure 3.2 as consisting of two objects, one (the rod) in motion behind the other (the box) (Kellman & Spelke, 1983). Neonates, in contrast, perceive this display as consisting of three objects (Slater et al., 1990), implying that occlusion perception emerges over the first several postnatal months (Johnson, 2004). That is, piecemeal or fragmented perception of the visual environment extends from birth through the first several months afterwards, at which time spatial completion is possible, The task of the developing visual system, and other perceptual, cognitive, and action systems, therefore, is to put the pieces together. In earlier work, José Nanez and I asked about spatial completion in infants between birth and four months; a second question addressed by this work was whether young infants would perceive completion in 2D, computer-generated displays, as seen in Figure 3.3 (Johnson & Nafiez, 1995). We replicated the Kellman and Spelke (1983) finding that 4-month-olds showed a posthabituation novelty preference for a broken rod stimulus, corroborating the validity

Figure 3.3. Computer-generated versions of the Kellman and Spelke (1983) displays depicted in Figure 3.2.

A Constructivist View of Object Perception in Infancy

59

of conclusions drawn from the use of 2D vs. 3D object displays, a novel finding at the time. Two-month-olds, however, exhibited no reliable posthabituation preference, implying that spatial completion is developing at this point but is not yet complete. Shortly thereafter, Richard Aslin and I examined the possibility that 2-month-olds will perceive unity if given additional perceptual support (Johnson & Aslin, 1995). The relative amount of visible rod surface revealed behind the occluder was increased by reducing box height and by adding gaps in it, and under these conditions 2-month-olds provided evidence of unity perception. With newborns, however, this manipulation failed to reveal similar evidence: Even in “enhanced” displays, newborns seemed to perceive disjoint rather than unified rod parts (Slater, Johnson, Brown, & Badenoch, 1996). These experiments served to pinpoint more precisely the time of emergence of spatial completion in infancy: the first several weeks or months after birth under typical circumstances. As mentioned previously, other experiments evaluated the hypothesis that spatial completion stems more or less directly from discrimination of motion patterns in the visible, moving rod parts (Johnson, 2004). The experiments were designed to challenge 2-month-olds’ detection of these motion patterns in two ways: by increasing box width such that the rod segments were farther apart, and by misaligning the rod segments. Neither manipulation had any effect on motion discrimination, although both manipulations prevented perception of the unity of the rod parts. Spatial completion, therefore, does not appear to stem directly from motion perception, although a number of experiments support the conclusion that common motion is necessary for infants to perceive the rod parts as connected (Eizenman & Bertenthal, 1998; Jusczyk, Johnson, Spelke, & Kennedy, 1999; Kellman & Spelke, 1983). The precise con-

tributions of motion to spatial completion in infants remain unknown. One possibility is that motion serves multiple functions: segmenting the scene into its constituent surfaces, segregating these constituents into different depth planes relative to the observer, and binding moving surfaces together (Johnson et al., 2008). We continue investigating these questions in my lab. Taken together, these experiments suggest that young infants analyze the motions and arrangements of visible surfaces in partly occluded object displays. At birth, infants perceive visible rod segments as separate from one another and from the background. Within a few months, infants integrate surfaces into larger units whose boundaries extend beyond what is directly visible. A vital question, then, concerns the means by which information in the visual environment that specifies these units is extracted and assembled. There is little evidence that these means operate according to predictions from the Gestalt tradition concerning intrinsic holistic perception. More likely is the possibility that filling-in develops from a constructive process. How is integration accomplished? To address this question, Amso and Johnson (2006) and Johnson, Slemmer, and Amso (2004) observed 3-month-

old infants in a spatial completion task using the habituation paradigm described previously. Infants’ eye movements were recorded with a corneal reflection eye tracker during the habituation phase of the experiment.

60

Infant Perception and Cognition

We found systematic differences in oculomotor scanning patterns between infants whose posthabituation test display preferences indicated unity perception, and infants who provided evidence of perception of disjoint surfaces. “Perceivers” tended to scan more in the vicinity of the two visible rod segments, and to scan back and forth between them (Figure 3.4), In a somewhat younger sample, Johnson et al. (2008) found a correlation between posthabituation preference—our index of spatial completion—and targeted visual exploration, operationalized as the proportion of eye movements directed toward the moving rod parts, which we reasoned was the most relevant aspect of the stimulus for perception of completion. Spatial completion was not predicted by other measures of oculomotor performance, such as mean duration and breadth of fixations, and “global” vs. “local” scanning activity. Instead, spatial completion was best predicted by saccades directed toward the vicinity of the moving rod parts. This can be a challenge for a developing oculomotor system, attested by the fact that targeted scans almost always followed the rod as it moved, rarely anticipating its position. How targeted visual exploration itself emerges in infancy to maximize effective uptake of visual information is not yet known, and is an area of active investigation. Spatial completion in very young infants may be challenged by difficulties in accessing relevant visual information. In line with this possibility, Johnson and Johnson (2000) observed scanning patterns in 2- to 3.5-month-olds as they viewed occlusion displays in various configurations, and found that the younger infants engaged in more extensive scanning when the occluder was relatively wide and when rod edges were misaligned. Alternatively, insufficient information acquisition may lead to a “default” response to the visible surfaces only, characteristic of neonates, yielding a novelty preference for the complete rod at test. Either possibility is consistent with the idea that efficient visual exploration is an important mechanism of development in object perception and, in turn, consistent with an information-processing view that emphasizes constructive processes. A relation between targeted visual exploration and spatial completion does not by itself pinpoint a causal role. Direct evidence for such a role would come from experiments in which individual differences in oculomotor patterns were

Figure 3.4. Examples of scanning patterns by infants as they view rod-and-box displays. Each example comes from a single habituation trial. Left: This infant looked

longer after habituation at the broken test stimulus, and this is a “perceiver” of unity. Right: This infant looked longer at the complete test stimulus, a “non-perceiver.”

A Constructivist View of Object Perception in Infancy

61

observed in both spatial completion and some other visual task, and this was recently reported by Amso and Johnson (2006). We found that both spatial

completion and scanning patterns were strongly related to performance in an independent visual search task in which targets, defined by a unique feature (either motion or orientation) were placed amongst a large set of distracters. Infants’ eye movements were recorded as they viewed each stimulus. There were substantial individual differences in successful search, both in terms of detecting the target and the latency to do so, and these differences mapped clearly onto the likelihood of spatial completion. This finding is inconsistent with the possibility that scanning patterns were tailored specifically to perceptual completion, and instead suggests that a general facility with targeted visual behavior leads to improvements across multiple tasks. In summary, the individual differences in targeted visual exploration that we have observed suggest that scanning patterns make a vital contribution to the emergence of veridical object perception. As scanning patterns develop, they support binding of disparate visual features into unified percepts—active assembly of coherent objects from surface fragments. With the emergence of selective attention and other perception-action systems, infants become increasingly active participants in their own perceptual development, rather than passive recipients of information. Active engagement of an infant's visual attention is consistent with a key tenet of Piagetian theory—the central role of the child’s own behavior in cognitive development—and with a constructivist view—the building of structure from constituent pieces. The following section describes another of these perception-action systems, visual-manual exploration, and its role in constructing volumetric objects.

3D Object Completion from Visual-Manual Exploration Solid objects occlude parts of themselves such that we cannot see their hidden surfaces from our present vantage point, yet our experience of most objects is that of filled volumes rather than hollow shells. Perceiving objects as solid in three-dimensional space despite limited views constitutes 3D object completion. Relatively little is known about its development, and whether infants typically perceive objects as whole in 3D space. Kasey Soska and I recently addressed this question with a looking-time paradigm (Soska & Johnson, 2008). Four- and 6-month-olds were habituated to a wedge rotating through 15° around the vertical axis such that the far sides were never revealed (Figure 3.5). Following habituation, infants viewed two test displays

beremeeSReE Leet COED “ieee ESE SPD Figure 3.5. Displays used by Soska and Johnson (2008) to examine 4- and 6-montholds’ 3D object completion.

62

Infant Perception and Cognition

in alternation—one a complete, whole version of the wedge (as adults would likely expect), and the other an incomplete, hollow version, with both undergoing a full 360° rotation revealing the entirety of the object shape. Four-month-olds showed no consistent preference for either test stimulus, but 6-month-olds looked longer at the hollow wedge display, indicating perception of the wedge during habituation as a solid, volumetric object in 3D space, despite restricted views. What are the developmental mechanisms of 3D object completion? One possibility is that emerging motor skills support perception of objects as coherent volumes. Two types of motor skill, both of which develop rapidly at the same time 3D object completion seems to emerge (4 to 6 months), may play a particularly important role—self-sitting and coordinated visualmanual exploration. These skills develop in tandem, and there is evidence that they are related, because independent sitting frees the hands for play and promotes gaze stabilization during manual actions (Rochat & Goubet, 1995). Thus, self-sitting might encourage coordination of object manipulation with visual inspection as infants begin to play with objects, providing the infants with multiple views. In addition, manipulation of objects—touching, squeezing, mouthing—may promote learning about object form from tactile information. To examine these possibilities, Kasey Soska, Karen Adolph, and I observed infants between 4.5 and 7.5 months in a replication of the Soska and Johnson (2008) habituation experiment with the rotating wedge stimuli (Soska, Adolph, & Johnson, 2010). In the same testing session, we assessed infants’ manual exploration skills by observing spontaneous object manipulation in a controlled setting, and obtained parental reports of the duration of infants’ sitting experience. We reasoned that infants who had more self-sitting experience would show a greater tendency to explore objects from multiple viewpoints, and therefore have more opportunities to learn about objects’ 3D forms outside the lab. Thus, within this age range, individual differences in self-sitting experience and coordinated visual-manual exploration were predicted to be related to individual differences in infants’ looking preferences to the complete and incomplete object displays, our index of 3D object completion. Our predictions were supported. We found strong and significant relations between both self-sitting and visual-manual coordination, from parents’ reports and the motor skills assessment, and from 3D object completion performance assessed with the habituation paradigm. We recorded a number of other motor skills to explore how widespread the relations were within the perception-action systems under investigation, such as grasping, holding, and manipulation without visual inspection. None were related to 3D object completion. Self-sitting experience and coordinated visual-manual exploration were the strongest predictors of performance on the visual habituation task. The results of a regression analysis yielded evidence that the role of self-sitting was indirect, influencing 3D completion chiefly in its support of infants’ visualmanual exploration. Self-sitting infants performed more manual exploration

A Constructivist View of Object Perception in Infancy

63

while looking at objects than did nonsitters, and visual-manual object exploration is precisely the skill that provides active experience viewing objects from multiple viewpoints, thereby facilitating perceptual completion of 3D form. These results provide evidence for a cascade of developmental events following from the advent of visual-motor coordination, including learning from self-produced experiences. In principle, 3D object completion might develop from more passive perceptual experiences, but the findings yielded by the Soska et al. (2010) experiment indicate that passive experience may be insufficient to learn about 3D object form. Active exploration provides information to the infant about her own control of an event, while simultaneously generating multimodal information to inform developing object-perception skills. Coordinating visual inspection with manual exploration seems to be critical: only the visual-manual skills involved in generating changes in object viewpoint—rotating, fingering, and transferring while looking—were related to 3D object completion. This kind of visual-manual exploration, therefore, constitutes active assembly of objects, objects as solid and volumetric in 3D space.

FILLING IN THE GAPS IN OUR KNOWLEDGE OF OBJECT PERCEPTION Previously in this chapter, I raised two issues that are worth revisiting in light of the research on active assembly just described. First, I couched the problem of development of object perception in terms of two sets of theoretical perspective: the empiricism-nativism controversy, and 20'" century advances in Gibson’s theory of ecological optics, information processing theory, and constructivism. Second, | provided a broad outline of the steps involved in object perception; clearly there is more to understanding development of object perception than perceptual completion. Each of these issues is discussed in turn. Bringing Clarity to Theoretical Perspectives on Development of Object Perception It is often noted that the extreme positions of both empiricism and nativism are untenable, yet the arguments persist. The research I have described cannot settle the controversy on its own, but I believe it provides clarity with respect to fundamental mechanisms by which infants come to perceive the visual world. An entirely holistic pattern of perception, as envisioned by the Gestalt psychologists, can be rejected as far as young infants are concerned, because visual perception is more piecemeal than holistic early in life. At the same time, perception is never wholly unstructured. Infants begin postnatal life with perceptual and motor capacities sufficient to acquire and retain information relevant to the task of disambiguating sensory input, and with natural proclivities to orient and attend to this information. For example, neonates actively scan the environment (Haith, 1980), and show visual preferences for

64

Infant Perception and Cognition

high-contrast edges and motion, two important sources of information for object boundaries (Slater, 1995), The research I have described is motivated by reciprocal notions of information: the information available in the stimulus, and the uptake of that information, central to Gibson’s theory of ecological optics and information processing theory, respectively. Motion and edge alignment, for example, are requisite cues for infants to perceive spatial completion, and recent studies of eye movements reveal individual differences in infants’ attention to these cues—differences that are related to perception of unified or disjoint surfaces. And the developmental processes that lead to the more adultlike construal of objects as complete, despite occlusion, are revealed to be compatible with constructivist theory, with its focus on the precise means and mechanisms by which missing information is filled in. Completion of objects in 3D space, likewise, proceeds from learning relevant information for object coherence and solidity over time—information from, for example, the typical properties of objects that infants encounter in daily life, obtained through everyday activities such as turning and touching as they inspect objects visually and thereby come to generalize these experiences to novel objects.

Progress in our understanding of developmental mechanisms of object perception could be greatly enhanced with the application of formal models, including connectionist and other kinds of computational models. One such model was proposed by Denis Mareschal and myself (Mareschal & Johnson, 2002). It consisted of a simple “retina” to which events were presented, perceptual modules that processed independently different visual cues, a hidden layer, and a response (output) layer. The models were initially trained by exposure to simple events in a simulated, schematic visual environment consist-

ing of unified and disjoint “objects” moving past and behind an “occluder” (Figure 3.6). The models were endowed with the ability to extract several cues important to spatial completion, such as object motion and edge alignment. The models also possessed a short-term memory, such that when an object became occluded, a rapidly decaying trace of that object’s representation remained. After varying amounts of training, we presented novel test events

that incorporated the visual cues to which the model was sensitive, and we systematically included or omitted cues across displays and observed the models’ responses. After sufficient training, the models responded appropriately

HEE Time step 1

Time step 4

Time step 7

Time step 11

Time step 14

Figure 3.6. Schematic depiction of one kind of event presented to a connectionist model of spatial completion by Mareschal & Johnson (2002).

A Constructivist View of Object Perception in Infancy

65

to unity under conditions of partial occlusion, demonstrating the potential importance of learning in the development of spatial completion. The nature of the training environment (i.e., the cues that were made available) was critical in determining performance. The Mareschal

and

Johnson

model

accords

well with data from

human

infants, and it is important in its demonstration of “learning to perceive.” Yet its scope and contributions are limited, in part because the human visual system does not work or develop in the same way. Our retinas have a fovea and we move our eyes to points of interest in the scene, and visual development in human infants takes place largely at the level of cortex—formation and strengthening of neural circuits within, to, and from visual areas (Atkinson, 2000)—as opposed to updating of weights within fixed connections between modules that is characteristic of many models (Rumelhart & McClelland, 1986), including ours. Progress toward a developmental model of object perception that begins to meet these goals was reported recently by Schlesinger, Amso, and Johnson (2007a, 2007b), who devised a computational model of infants’ gaze patterns based on the idea of “salience maps” produced by visual modules tuned to luminance, motion, color, and orientation in an input image (Itti & Koch, 2000). The input image we selected

was a moving rod-and-box stimulus (Figure 3.7). Salience was computed

Figure 3.7. Input (upper left) and outputs ofa computational model of visual development (Adapted from Schlesinger, Amso, & Johnson, 2007b). Outputs are represented as salience maps, portrayed here as 3D topological surfaces showing regions of activation. After initial exposure (iteration 0, upper right), edge regions become highly salient. Iterations 1 and

10 are shown at the lower left and lower right, respectively.

Between iterations 0 and 10 the map becomes sharpened and defined as the rod is highlighted and the box recedes in salience. (See also figure in plate section.)

66

Infant Perception and Cognition

in part via a process of competition between visual features, as the model received repeated exposures (or iterations) to the images. This was a strategy motivated by patterns of activity in the posterior parietal cortex that are suppressed in response to visual features that remain constant across exposure, while increasing responses to features that change—thus highlighting their salience (Gottlieb, Kusunoki, & Goldberg, 1998). The model had a simulated fovea and the ability to direct “gaze” toward the most salient region in the image. After several iterations, the model quickly developed a salience map in which the rod segments were strongly activated, as activity for the edges of the occluder receded (Figure 3.7)—consonant with the learning process in infants as they begin to direct their attention toward the moving rod (Amso & Johnson, 2006; Johnson et al., 2004). The Schlesinger

model was intended to examine development of visual attention, not spatial completion per se, but given the success of this model, and that of Mareschal and Johnson (2002), a model of “learning from scanning” seems feasible and likely to achieve important insights into human development. Infants’ Object Perception in the Bigger Picture We live in a sensory environment that is complex, rich, and variegated, yet our subjective impression of this environment is generally seamless, stable, and comprehensible. Visual, auditory, tactile, and other sensory inputs are unerringly integrated. We readily detect a friend’s face among strangers; we coordinate oculomotor, locomotor, and manual action systems on the fly as we navigate; we remember where we last placed our keys. All of these everyday behaviors are a part of object perception, and all develop from simpler origins, The cortical, perceptual, motor, and cognitive developmental mechanisms that bring us to these skills pose a great challenge to developmental and cognitive scientists. The experiments on perceptual completion and active assembly provide a small contribution to this larger enterprise, but much work remains—in particular, understanding the cortical foundations of perception and cognition in infants. REFERENCES Aslin, R. N. (2007). What's in a look? Developmental Science, 10, 48-53. Atkinson, J. (2000), The developing visual brain. New York: Oxford University Press. Baillargeon, R. (1995). A model of physical reasoning in infancy. In C. Rovee-Collier & L. P. Lipsitt (Eds.), Advances in infancy research (Vol. 9, pp. 305-371). Norwood, NJ: Ablex.

Bornstein, M. H, (1985). Habituation of attention as a measure of visual information processing in human infants: Summary, systematization, and synthesis. In G. Gottlieb & N. A. Krasnegor (Eds.), Measurement of audition and vision in the first year of postnatal life: A methodological overview (pp. 253-300). Norwood, NJ: Ablex. Eizenman, D. R., & Bertenthal, B. I. (1998). Infants’ perception of object unity in translating and rotating displays. Developmental Psychology, 34, 426-434.

A Constructivist View of Object Perception in Infancy

67

Gibson, J. J. (1979). The ecological approach to visual perception. Hillsdale, NJ: Erlbaum. Gottlieb, J. P., Kusunoki, M., & Goldberg, M.E. (1998). The representation of visual salience in monkey parietal cortex. Nature, 391, 481-484, Haith, M, M, (1980), Rules that babies look by: The organization of newborn visual activity. Hillsdale, NJ: Erlbaum. Johnson, S. P. (2004), Development of perceptual completion in infancy. Psychological Science, 15, 769-775. Johnson, S. P. (2010). Perceptual completion in infancy. In S. P. Johnson (Ed.),

Neoconstructivism: The new science of cognitive development (pp. 45-60). New York: Oxford University Press. Johnson, S. P., Amso, D., & Slemmer, J. A. (2003). Development of object concepts in infancy: Evidence for early learning in an eye tracking paradigm. Proceedings of the National Academy of Sciences USA, 100, 10568-10573. Johnson, S. P., & Aslin, R. N. (1995). Perception of object unity in 2-month-old infants. Developmental Psychology, 31, 739-745. Johnson, S. P., Davidow, J., Hall-Haro, C., & Frank, M. C. (2008), Development of

perceptual completion originates in information acquisition. Developmental Psychology, 44, 1214-1224. Johnson, S. P., & Johnson, K, L. (2000). Early perception-action coupling: Eye movements and the development of object perception. Infant Behavior & Development, 23, 461-483. Johnson, S. P., & Nafiez, J. E. (1995). Young infants’ perception of object unity in twodimensional displays. Infant Behavior & Development, 18, 133-143. Johnson, S. P., & Shuwairi, S. M. (2009). Learning and memory facilitate predictive tracking in 4-month-olds. Journal of Experimental Child Psychology, 102, 122-130.

Johnson, S. P., Slemmer, J. A., & Amso, D. (2004), Where infants look determines how they see: Eye movements and object perception performance in 3-montholds. Infancy, 6, 185-201. Jusezyk, P. W., Johnson, S, P., Spelke, E. S., & Kennedy, L. J. (1999). Synchronous change and perception of object unity: Evidence from adults and infants. Cognition, 71, 257-288. Kellman, P. J., & Arterberry, M. E. (1998). The cradle of knowledge: The development of perception in infancy. Cambridge, MA: MIT Press. Kellman, P. J., & Arterberry, M. E. (2006). Perceptual development. In W. Damon, D. Kuhn, & R. Siegler (Eds.), Handbook of child psychology: Cognition, perception, and language (6th ed, pp. 109-160). Hoboken, NJ: Wiley. Kellman, P. J., & Spelke, E. S. (1983). Perception of partly occluded objects in infancy. Cognitive Psychology, 15, 483-524. Koffka, K. (1935). Principles of Gestalt psychology. London: Routledge & Kegan Paul. Mareschal, D., & Johnson, S, P. (2002). Learning to perceive object unity: A

connectionist account. Developmental Science, 5, 151-185. Marr, D. (1982). Vision. San Francisco: Freeman. Palmer, S. E. (1999), Vision science: Photons to phenomenology. Cambridge, MA: MIT Press.

Piaget, J. (1952). The origins of intelligence in children (M. Cook, Trans.) New York: International Universities Press. Piaget, J. (1954). The construction of reality in the child (M. Cook, Trans.) New York: Basic Books.

68

Infant Perception and Cognition

Rochat, P., & Goubet, N. (1995). Development of sitting and reaching in 5- to 6-month-old infants. Infant Behavior & Development, 18, 53-68. Rumelhart, D. E., McClelland, J. L., & the PDP Research Group. (1986). Parallel

distributed processing: Explorations in the microstructure of cognition, Cambridge, MA: MIT Press. Schlesinger, M., Amso, D., & Johnson, S. P. (2007a). The neural basis for visual selective attention in young infants: A computational account. Adaptive Behavior, 15, 135-148.

Schlesinger, M., Amso, D., & Johnson, S. P. (2007b). Simulating infants’ gaze patterns during the development of perceptual completion. In L, Berthouze, C.G, Prince, M. Littman, H. Kozima, & C, Balkenius (Eds.), Proceedings of

the seventh International Workshop on Epigenetic Robotics: Modeling cognitive development in robotic systems (pp. 157-164). Lund, Sweden: Lund University Cognitive Studies. Slater, A. (1995). Visual perception and memory at birth. In C, Rovee-Collier & L. P. Lipsitt (Eds.), Advances in infancy research (vol. 9, pp. 107-162). Norwood, NJ: Ablex.

Slater, A., Johnson, S. P., Brown, E., & Badenoch, M. (1996). Newborn infants’ perception of partly occluded objects. Infant Behavior & Development, 19, 145-148.

Slater, A., Morison, V., Somers, M., Mattock, A., Brown, E., & Taylor, D. (1990). Newborn and older infants’ perception of partly occluded objects. Infant Behavior and Development, 13, 33-49, Smith, W. C,, Johnson, S. P., & Spelke, E. S, (2003). Motion and edge sensitivity in perception of object unity. Cognitive Psychology, 46, 31-64. Soska, K. C., Adolph, K. A., & Johnson, S. P. (2010). Systems in development: Motor skill acquisition facilitates 3D object completion, Developmental Psychology, 46, 129-138.

Soska, K. C., & Johnson, S. P. (2008). Development of 3D object completion in infancy. Child Development, 79, 1230-1236. Spelke, E. S., Breinlinger, K., Macomber, J., & Jacobson, K. (1992). Origins of

knowledge. Psychological Review, 99, 605-632. Wattam-Bell, J. (1996), Visual motion processing in one-month-old infants: Habituation experiments. Vision Research, 36, 1679-1685.

Wynn, K. (1992), Addition and subtraction by human infants. Nature, 358, 749-750.

4 Development of Specialized Face Perception in Infants: An Information-Processing Perspective Cara H. Cashon

Faces constitute some of the most important stimuli for humans. They provide information about identity and emotions, and convey important communicative and social cues. Considerable behavioral and neuropsychological research with adults indicates that faces are treated differently than other types of stimuli, although the reasons for such findings are hotly debated (e.g., Gautheir & Nelson, 2001; Gauthier & Tarr, 1997; Haxby, Hoffman, & Gobbini, 2000; Kanwisher, 2000; Kanwisher & Yovel, 2006; Tarr & Gauthier, 2000).

Dating back to Fantz’ (1963) early demonstration that newborns look longer at face-like stimuli, research with infants provides evidence that faces enjoy differential treatment from a very young age (see Goren, Sarty, & Wu, 1975; Johnson & Morton, 1991; but also see Macchi Cassia, Turati, & Simion, 2004,

for an alternate explanation). Recent studies on infant face perception also suggest that, during the first year of life, infants are becoming “little experts” on certain types of faces. A common pattern among these results is that younger infants act as “generalists,” ready to attend or process a wide variety of face stimuli, whereas older infants act in a more “specialized” manner, having become attuned to the kinds of faces seen in their environment. The goal of this chapter is to outline recent advances in our understanding of infants’ developing expertise for certain classes of faces, with an emphasis on the mechanisms that may underlie its development. First, I will provide some background on what is known, generally, about adults’ expertise for certain faces. Next, I will present recent research that focuses on the development of that expertise during infancy. I will review findings that illustrate a “broad-to-narrow” developmental pattern across studies related to infants’ preferences, recognition/discrimination, and event-related potentials. Finally, I will review some of my work that shows a pattern of specialization with respect to infants’ processing of upright and own-race faces. As part of this discussion, I will highlight Cohen’s information-processing approach to infant cognitive development, how it has influenced some of the work in this area, and how it may help to shed light on our understanding of the underlying mechanisms involved in the development of infants’ specialization.

70

Infant Perception and Cognition

ADULTS’ EXPERTISE FOR CERTAIN CLASSES OF FACES Considerable research has shown that adults are better at recognizing upright compared to inverted faces (e.g., Searcy and Bartlett, 1996; Yin, 1969), and

own-race compared to other-race faces (e.g., Meissner & Brigham, 2001). Many researchers argue that such differences could be due to differential processing for upright versus inverted, and own- versus other-race faces (e.g., Diamond & Carey, 1986; Tanaka, Kiefer, & Bukach, 2004). Although there is some lack of agreement on the specific terms used in the literature, Maurer, LeGrand and Mondloch (2002) argue that the modes of processing that have been discussed and tested can be summarized as: (1) first-order configural processing (i.e., detecting that the eyes are above the nose, which is above the mouth, and so forth, used to identify that a face is a face); (2) holistic processing (i.e., “gluing”

the features of a face into a Gestalt); and (3) second-order configural processing (i.e., being sensitive to the spacing between features, such as the distance between the eyes),

Consistent with these definitions, researchers have shown that inverting a face disrupts holistic and second-order configural processing (e.g., Bartlett & Searcy, 1993; Diamond & Carey, 1986; Farah, Tanaka, & Drain, 1995; Freire, Lee, & Symons,

2000;

Sergent,

1984; Tanaka

& Farah,

1993).

For example,

Farah, Tanaka, and Drain (1995) used a part-whole paradigm to test the hypothesis that holistic processing is disrupted by inversion, Adult participants were trained on upright faces that were presented either intact (holistic) or as isolated parts. In the test phase, participants were shown either upright or inverted intact faces, and were asked to identify the faces. Farah et al. found that adults who were trained on intact faces were worse at recognizing the test faces when they were presented in the inverted orientation. In support of the notion that inversion disrupts holistic processing, they found that the adults who were trained on face parts did not show a difference in their ability to recognize upright or inverted faces during the test phase. These results indicate that when part-based processing is induced, inversion does not impair performance; however, when holistic processing is engaged, inverting the face stimuli has a detrimental effect. Similarly, recent research shows that adults use holistic and second-order configural processing of own-, but not other-race faces (Michel, Caldara, & Rossion, 2006; Michel, Rossion, Han, Chung, & Caldara, 2006; Rhodes, Hayward, & Winkler, 2006; Tanaka, Kiefer, & Bukach, 2004). For example, again using a part-whole paradigm, Tanaka and colleagues (Tanaka, Kiefer, & Bukach, 2004) found that Caucasian participants, who reported having very little exposure to Asian individuals, showed a deficit in their ability to recognize Asian faces compared to Caucasian faces in the holistic-face condition. However, no effect of race was found in the parts-face condition, indicating that other-race faces may have a similar detrimental effect on holistic processing. Tanaka et al. also found an effect of experience when they tested Asian participants who reported having more exposure to Caucasian than Asian others. These participants did not show an effect for own- versus other-race

Development of Specialized Face Perception in Infants

71

faces. They showed better performance in the holistic than in the parts-face condition when viewing faces of either race. Together, these findings suggest that differential processing by adults may account for differences in recognition of the classes of faces that adults see most often. In addition to these behavioral findings, neuropsychological correlates for adults’ facial expertise have also been found, and provide another possible underlying mechanism. An example of this is with the so-called “fusiform face area” (FFA), an area of the fusiform gyrus in the occipitotemporal region that has been shown, in functional imaging studies, to have higher activation to faces than to other objects (Gauthier & Nelson, 2001; Kanwisher &Yovel, 2006; Kanwisher, McDermott, & Chun, 1997; for review, see Tzourio-Mazoyer,

de Schonen, Crivello, Reutter, Aujard, & Mazoyer, 2002), The FFA has also been shown to have differential activity levels when adults view upright compared to inverted faces (e.g., Yovel & Kanwisher, 2005), and own-race versus other-race faces (Golby, Gabrieli, Chiao, & Eberhardt, 2001). Furthermore,

event-related potential (ERP) studies have consistently shown that the N170, a negative peak responding 170 ms after the presentation of a face, is delayed 10 ms when the face stimulus is inverted and, on occasion, has greater amplitude (for review, see Rossion & Gauthier, 2002). Taken together, these findings

provide evidence that neurophysiological differences exist in adults’ responses to faces of different classes. DEVELOPMENT OF INFANTS’ SPECIALIZATION TO CERTAIN FACES As discussed, there is behavioral and neuropsychological evidence that certain types of faces are treated differently by adults. It is generally agreed that these two phenomena, the “inversion effect” and the “other-race effect,” are the result of experience, and are thought to reflect expertise for these faces. Thus, the research described inevitably leads to the question of just how such expertise develops. There is growing evidence that, during the first year, infants become attuned to the world around them. A classic example can be found in studies investigating infants’ perception of native versus nonnative phonemes. It has been shown that infants around six months discriminate between both native and nonnative speech sound distinctions, but around 12 months they discriminate only between two native speech sounds (Werker & Tees, 1984a), Similar examples have recently emerged for infants’ perception of culturally native musical rhythms (Hannon & Trehub, 2005) as well as cross-species face and voice matching (Lewkowicz & Ghazanfar, 2006). Recent studies on infant face per-

ception reveal that they also become attuned to certain faces. In the following sections, | will outline several ways in which infants show a similar broad-tonarrow, less-constrained to more-constrained pattern in response to faces.

Discrimination/Recognition Inarecent demonstration of the development of the other-race effect in infancy, Kelly, Quinn et al. (2007) found that infants’ abilities to discriminate faces of

72

Infant Perception and Cognition

their own and other races develops between three and nine months. Using a visual paired comparison (VPC) paradigm, Kelly et al. tested Caucasian infants on whether they could recognize a familiarized face when paired with another face. Infants either saw Caucasian, Middle Eastern, Asian, or African faces, At three months, they found that infants were able to recognize the target face regardless of its race. At six months, however, they began to see a bias such that infants recognized the target face in the Caucasian and Asian conditions but not in the Middle Eastern or African conditions. The 9-month-olds, in contrast, were the first to show clear evidence of the other-race effect—they showed evidence of recognizing the target face only in the Caucasian condition. In another study by Sangrigoli and de Schonen (2004), 3-month-olds’ ability to discriminate between other-race faces was found only when infants were exposed to multiple faces during the familiarization period, but not when they were familiarized to one face. While the ages at which these two groups of researchers report finding evidence of the other-race effect differs slightly, differences in their stimuli could account for the differences in results. Whereas Kelly etal. presented infants with photographs of “normal” faces, Sangrigoli and de Schonen showed women wearing shower caps on their heads. Nevertheless, together these findings suggest that the origins of the other-race effect may have roots in infancy, seemingly between three and nine months. Recent research also indicates that during the first year infants become specialized towards human faces (versus monkey faces). Pascalis, de Haan, and Nelson (2002) found that while infants at six months show no difference in

their ability to discriminate between either two monkey faces or two human faces, adults and infants at nine months struggle to discriminate between two monkey faces but not two human faces. Presumably, infants have much more experience with human faces than monkey faces, and thus maintain the ability to discriminate human, but not monkey faces. In a more recent study, Pascalis et al. (2005) tested the effects of exposure to monkey faces on this “other-species effect.” Six-month-olds and their parents were sent home with a picture book of monkey faces, and the parents were asked to expose the infants to these faces on some predetermined schedule. After three months of training, the infants were brought back to the lab and their ability to discriminate between the monkey faces was tested. A separate group of 9-montholds—who received no such picture-book experience at home—was also tested on their ability to discriminate between monkey faces. Pascalis et al. (2005)

reported finding that the 9-month-olds who had had received training could still discriminate between a novel set of monkey faces, but the control group who received no training could not. Such findings support the notion that experience plays an important role in the development of expertise for certain faces (see also Sangrigoli, Pallier, Argenti, Ventureyra, & de Schonen, 2005), Preferences

In addition to studies on infants’ recognition abilities, researchers have also studied infants’ preferences for certain classes of faces. Quinn, Yahr, Kuhm,

Development of Specialized Face Perception in Infants

73

Slater, and Pascalis (2002) found that 3- to 4-month-olds who were raised by

females preferred to look at female faces, whereas those who were raised by males preferred to look at males. Kelly and colleagues recently investigated this issue with respect to race (or ethnicity). Although they found evidence that 3-month-old Caucasian and Chinese infants show a preference for ownrace faces (Kelly et al., 2005, 2007 respectively), when they tested Caucasian newborns, no such preference was found. The authors took this as evidence that experience matters and drives the development of infants’ preferences for certain faces, An important study by Bar-Haim, Ziv, Lamy, and Hodes (2006) supports this view that experience plays an important role in the narrowing effect. In a more direct test of the experience hypothesis, Bar-Haim et al. found that 3-month-old Caucasian Israeli infants, raised in a predominantly Caucasian environment, preferred Caucasian faces, whereas Black Ethiopian infants raised in a predominantly African environment preferred African faces. Their most important findings came from a group of Black Ethiopian infants who were immigrating to Israel and living in mixed-race housing. These infants showed no preference, suggesting that the types of faces infants experience can modify their facial preferences. Event-Related Potentials

In addition to the behavioral evidence described above, there is also evidence with event-related potentials (ERP) that infants become more specialized. De Haan, Pascalis, and Johnson (2002) compared ERP responses of adults and

6-month-old infants when viewing faces of humans and nonhuman primates that were presented in upright and inverted orientations. They found that in adults, the N170 is smaller in amplitude and shorter in latency when viewing upright human faces compared to the other face stimuli. In 6-month-old infants, they found separate components showing adult-like specialization. The N290 component responded to the species of the face, but not the orientation; in contrast, the P400 responded to the orientation, but not the species. In a subsequent study with 12-month-olds, de Haan and colleagues found that both the N290 and the P400 responded to both the species and orientation (Halit, de Haan, & Johnson, 2003). Together, these results suggest that some specialization is occurring in infants’ ERP responses to different types of faces during the first year, however, these responses are completely adult-like. Additional changes must occur before the infant ERP responses are combined into the adult-like N170 response. MECHANISMS UNDERLYING SPECIALIZATION Perceptual Narrowing A common pattern among many of the results discussed is that younger infants acted in a “general” manner, whereas older infants acted in a more

74

Infant Perception and Cognition

“specialized” manner. The studies cited above provide evidence for perceptual narrowing with respect to infants’ facial recognition and discrimination abilities, preferences, and ERP patterns. A similar developmental pattern has emerged recently in two other areas—infants’ perception of culturally native musical rhythms (Hannon & Trehub, 2005), along with cross-species face and voice matching (Lewkowicz & Ghazanfar, 2006). Due to the similarity in timing and perceptual nature of these and many other findings, Nelson (2001) has termed this pattern of findings “perceptual narrowing.” More recently, Scott, Pascalis, and Nelson (2007) have argued that a common neural architecture may underlie such findings. However, the underlying mechanisms involved in this process are not completely understood. Missing from this discussion are the ways in which processing may be changing and becoming more attuned during infancy, as well. In the following section, I will describe research inspired by Cohen's information-processing approach (e.g., Cohen, Chaput, & Cashon, 2002) that begins to address such issues. Constraints on Processing When Les Cohen and I initially started to investigate infants’ processing of faces, we approached the topic from a domain-general, informationprocessing perspective. More specifically, we wondered whether the developmental changes found in infants’ processing of faces would be similar to or different from changes found in other areas. From our perspective, we believed that while the input might be different, the processes would generally be the same. This perspective had its roots in a set of informationprocessing principles of infant cognitive development proposed by Cohen that have been shown to explain developmental changes found across a number of domains (e.g., see Cohen & Cashon, 2001b; Cohen, Chaput, & Cashon, 2002). This approach has been described in detail elsewhere, but I will briefly outline some of the key principles that, as I will describe, apply to infants’ face processing. The principles on which I will focus are: (1) information integration is a key to development and, as such, infants are often found to shift from featural to holistic processing; (2) the integration of lower units of information serve as base units that are integrated and form higher units of information; and (3) presenting additional informa-

tion to the system can overload it, causing it to regress to a previously used, lower level of processing. Infants’ Processingg of 1

Upright g Versus Inverted Faces

Guided by these principles, we investigated whether infants would process upright versus inverted faces holistically, as adults do, and whether their developmental differences would follow a featural-to-holistic pattern. We used a looking-time technique, called the visual habituation “switch” procedure, commonly used to study infants’ ability to associate two or more pieces of information. This design has been used to study infants’ ability to detect correlations in a number of domains, including infants’ perception of line-drawn animals

Development of Specialized Face Perception in Infants

75

(Younger & Cohen, 1986) and infants’ perception of word-object associations (Werker, Cohen, Lloyd, Casasola, & Stager, 1998). Using the “switch” design allowed us to explore whether infants processed faces holistically or not—that is, whether infants associated the specific internal features and external features of a face. In our initial set of studies, infants were shown two color photographs of female faces (presented sequentially) during the habituation phase, and their looking times were recorded. In the test phase, infants were shown three test faces, one at a time, in an order that was counterbalanced across participants. The test faces were as follows: familiar (one of the two habituation faces), “switch” (consisting of the eyes, nose, and mouth of one of the habituation faces, and the region outside the eyes, nose, and mouth of the other habituation face) and novel (an unseen face that consisted of the internal facial features of one of the habituation faces and the external features of the other habituation faces). The most important test faces were the familiar and switch faces (see Figure 4.1), We reasoned that if infants looked significantly longer at the switch test face than the familiar test face, it must be because they were sensitive to the novel combination of internal and external features. This was how we operationalized “holistic” processing. If, however, infants did not look significantly longer at the switch face than the familiar face, we assumed that infants did not notice the novel combination of features. This we refer to as “featural” processing.

Habituation Face 1

Habituation Face 2

Switch Test Face

Figure 4.1. Examples of face stimuli used in the switch face-processing task. Actual

faces used are not shown.

76

Infant Perception and Cognition

In our first study, we tested 7-month-old infants. We chose this age because it is the developmental point where others had found that infants process the correlations between features of novel, complex patterns (Younger & Cohen, 1986). Thus, we hypothesized that ifinfants process faces like objects, we should find that they process faces holistically. However, based on the adult literature, we also reasoned that infants might process upright faces and inverted faces differently. We found that, in fact, 7-month-olds did process upright faces holistically, but they did not do so for inverted faces. This was consistent with the adult literature that showed processing differences for these two classes of face stimuli, and with the infant literature showing that complex patterns are processed holistically around seven months. Ina subsequent set of experiments, we investigated the development of this differential processing of upright and inverted faces. Thus, we replicated the study with infants who were between three and six months. The data are summarized in Figure 4.2, The first thing to notice is that the 3-month-olds did not process the faces holistically, but the 4-month-olds did. This featural-toholistic progression is consistent with Cohen’s (1998) information-processing principle that infants will initially process the parts ofa stimulus, and later the correlations between those parts. In this task, then, infants at three months do not correlate the internal and external features of a face, but at four months of age they do. Based on Younger and Cohen’s (1986) work, as well as others (see Cohen & Cashon, 2001b), we had previously argued that there may be

Featural

65

% Combination Score

Integrative

60

Type of Face

55

05 Upright 0

504--/---

Inverted

45 40 35

3

4

575

625

7

Age in Months * p< .05 (one-tailed) ** p< .01 (one-tailed)

Figure 4,2. Data representing the development of holistic processing of upright and inverted faces between 3 and 7 months of age. The “% Combination Score” reflects the percentage of time infants looked at the switch test face compared to the familiar plus switch test faces. Scores significantly greater than 50% (chance) are thought to reflect holistic (“integrative”) processing. Source: From “Beyond U-shaped Development in Infants’ Processing of Faces: An Information-Processing Account,” by C. H. Cashon & L. B. Cohen, 2004. Journal of Cognition and Development, 5, p. 74. Copyright 2004 by Taylor and Francis. Reprinted with permission.

Development of Specialized Face Perception in Infants

LP

featural-to-holistic shift in processing objects between four and seven months of age. The fact that we found this shift with face stimuli in younger infants around three to four months is noteworthy because it is evidence that while the processes involved in development may be similar across domains, the timetable can shift depending on the stimuli. What is also noteworthy about these findings are that the 4-month-olds did not struggle to process the inverted faces holistically. We wondered if this result was due to the way we created our faces. Specifically, we thought that if infants attended only to eyes and their corresponding eyebrows, they would notice the new combination of features in the “switched” face. Thus, in a subsequent experiment, we created stimuli such that the eyes and eyebrows were never separated; that is, the internal features that were pasted on the other face now consisted of the eyebrows, eyes, and nose. Even with this manipulation, however, the result was the same. Thus, we rule out the possibility that infants were only detecting the correlation between the eyes and eyebrows and conclude instead that they were attending to the correlation between internal (e.g., eyes, nose, and mouth) and external features (the hairline) at this age. Next, we tested 6-month-olds. There was early indication that there may be differences between young and old 6-month-olds, So, we ran complete experiments on younger (mean age 5.75 months) and older (mean age 6.25 months) infants around six months. We found that the 6.25-month-olds did not process either orientation holistically, and the 5.75-month-olds seemed to fall somewhere in between. We took this as evidence that infants at this age had regressed. We argued—from our information-processing viewpoint—that this regression may be related to an influx of new information about upright faces, such as their social importance. As part of this explanation, it is important to note that the regression is temporary and occurs at an age just prior to when infants, for the first time, display differential processing of upright and inverted faces. Thus, we hypothesized that the regression is part ofa reorganization that occurs as infants incorporate new, meaningful information about upright faces. In this way, Cohen’s information-processing principle that the system can be overloaded—when additional information needs to be integrated —gave us a framework to help make sense of our puzzling result. Recently, my students and I have been investigating the question of whether learning to sit, which occurs on average around six months, could be related to this regression in performance on our face-processing task. Using an ageheld constant design, we compared infants’ face processing on the “switch” task based on sitting ability (Cashon, Ha, Allen, & Barna, 2009). In the first part of the experiment, we compared two groups of infants at the younger end of the U-shaped curve—infants around 5.5 months of age. These infants were divided into two groups: those who could not sit up at all (nonsitters) versus those who tried to sit up, but could only do so by putting their hands down or resting on their bellies (near-sitters). No significant differences in age were found between sitting groups. In the second part of the experiment, we compared infants at the older end of the U—infants around 6.5 months of age. These infants were divided into those who could sit autonomously for

78

Infant Perception and Cognition

10s or more (new sitters) versus those who could sit, and had been doing so for four weeks or more (expert sitters). Again, no differences in age were based on

sitting ability. The results of this study support our hypothesis. We found we found that the infants who could not sit up at all (nonsitters) and those whom we

considered experts (expert sitters) process faces holistically; however, those who were learning to sit (near-sitters and new sitters) were not successful at processing holistically. Together, this pattern indicates that learning to sit is related to a regression in infants’ holistic processing of upright faces. These findings are consistent with a growing body of literature connecting infants’ actions or motor skills with changes in cognition or perception (see Rakison & Woodward, 2008). However, a complete understanding of this relationship remains unclear. We hypothesize that learning to sit may provide additional proprioceptive information as well as a new view of the external world, including a new view of faces. According to this view, until infants have incorporated this new information into their representation of faces, their face processing suffers. Clearly, more research needs to be conducted to investigate these ideas further. It would useful, for example, to determine whether learning to sit has similar, non-linear relations with other aspects of infants’ processing of faces—or cognition and perception more broadly. On one hand, the relation may be limited to aspects of face processing that are closely tied to the orientation of faces. On the other hand, learning to sit may be significantly correlated with changes across a wide variety of areas in development. ‘There are many important, unanswered questions as a result of these findings, and clearly more research needs to be done in this area, Across all these experiments, it is apparent that during the first year, infants undergo several important changes in the first year with respect to their holistic processing of upright and inverted faces. Initially, infants develop the ability to process the internal and external features concurrently, i.e., holistically, regardless of orientation. Then, the system appears to become more con-

strained as it follows a broad-to-narrow course. As holistic processing of faces develops, it is used to process both upright and inverted faces. However, with development—possibly due to the development of sitting—infants constrain their use of holistic processing, limiting it to upright faces only. Thus, the narrowing pattern that has been previously described as “perceptual narrowing” need not be limited to infant preferences or discrimination abilities. Infants’ Processing of Own- Versus Other-Race Faces Recently, in a collaborative effort with Marianella Casasola and her lab, we tested the idea that infants’ processing of own- versus other-race faces could also become more specialized —or more narrow—during infancy (Ferguson, Kulkofsky, Cashon, & Casasola, 2009). The results of Cashon and Cohen (2004)

show that around four months, infants do not show differential processing of upright versus inverted faces, but by seven months they do. Thus, in this study, we compared 4- and 8-month-olds on their processing of own- versus otherrace faces. Using a design similar to the previous “switch” task, half the infants

Development of Specialized Face Perception in Infants

79

saw Caucasian faces, whereas the other half saw all African faces. Similar to previous patterns, we found that infants at this younger age processed faces from both ethnicities holistically; however, the 8-month-olds processed only upright, own-race faces holistically. This finding indicates that between four and eight months of age, infants’ processing of faces becomes specialized for own-race faces. This result is very similar to Kelly et al.’s recent finding that at nine, but not three months, infants show better recognition skills for own-race over other-race faces, as well as to our work on the inversion effect. SUMMARY AND CONCLUSIONS In this chapter, evidence was presented indicating that a specialization for own-race, own-species, and upright faces develops during infancy. What do these results have in common, and what might they say about the development of specialization in face perception in infancy more generally? The findings indicate that during the first year of life, infants start out as “generalists,” ready to attend or process a wide variety of faces; but with experience, infants narrow their responses and become more attuned to their environment. This was found to be the case for infants’ preferences, discrimination/ recognition abilities, and ERPs. As reviewed in this chapter, their mode of processing for certain classes of faces—specifically, upright faces and ownrace faces—becomes specialized, as well. With the influence of Cohen’s information-processing approach, we have added an important dimension to the discussion—a focus on developmental changes in processing, and underlying mechanisms of change. The idea that younger infants are less selective in the correlations they will process, and older infants are more selective, is not unique to perceptual tasks during the first year of life. For example, Madole and Cohen (1985) found that 14-month-olds are sensitive to the correlation between the shape ofa feature of an object and its function (a meaningful correlation), as well as the correlation between the shape of a feature of an object and the function of another feature (an arbitrary correlation). They found that 18-month-olds, however, show

evidence of sensitivity only for the meaningful correlation. A similar pattern can be found in infants’ word-object associations. Although 8-month-olds discriminate between two similar-sounding nonsense words (such as “bih” and “dih”) when each is paired with an object, or two less similar-sounding nonsense words (such as “lif” and “neem,”), 14-month-olds can only discrimi-

nate the less similar-sounding words when presented with an object (Rakison, 2005, 2006; Stager & Werker, 1997). Rakison and Lupyan (2008) have made a similar argument regarding a domain-general developmental process in which infants are initially less constrained in the correlations they will detect, but with experience and knowledge become more constrained. In support of his view, Rakison (2006) has recently reported finding that 18-month-olds associate dynamic or static object parts with the onset of motion of an object; however, 20-month-olds detect association only between a moving object part

80

Infant Perception and Cagnition

and self-propelled motion. His recent results add to the growing list of studies showing the narrowing effect in infancy. When considering all of these findings together, the broad age range and diversity of tasks showing a broad-to-narrow, unconstrained-to-constrained, generalized-to-specialized pattern of responses is striking. It suggests that this pattern may represent an extremely pervasive and important learning or developmental process. It may be the case that there are common neural underpinnings to the “perceptual narrowing” findings, as has been argued. However, when studies on changes in infants’ processing of faces in the first year are considered alongside those on infants’ processing of other stimuli beyond the first year, questions are raised about what may be in common to all them. For example, are the underlying mechanisms involved in the development of constraints on facial preferences, discrimination, and processing in the first year all similar, or different? Are those underlying mechanisms related to the developing constraints on the correlations infants detect after the first year? “Narrowing” is clearly not limited to simple facial discrimination studies with infants in the first year. If we hope to deepen our understanding of this developmental pattern, then the mechanisms that underlie or coincide with such changes—such as changes in processing—need to be considered as well.

REFERENCES Bar-Haim, Y., Ziv, T., Lamy, D., & Hodes, R. M. (2006). Nature and nurture in own-race face processing. Psychological Science, 7, 159-163. Bartlett, J., & Searcy, J. (1993). Inversion and configuration of faces. Cognitive Psychology, 25, 281-316. Cashon, C. H,, & Cohen, L. B. (2003). The construction, deconstruction, and reconstruction of infant face perception. In A. Slater & O. Pascalis (Eds.), The development of face processing in infancy and early childhood (pp. 55-68). New York: Nova Science Publishers. Cashon, C. H., & Cohen, L. B. (2004), Beyond U-shaped development in infants’ processing of faces: An information-processing account. Journal of Cognition and

Development, 5, 59-80. Cashon, C. H., Ha, O., Allen, C. L., & Barna, A. C. (2009, April). Processing of upright faces differs by sitting ability for 5.5- and 6.5-month-old infants. Paper presented at the biennial meeting of the Society for Research on Child Development, Denver, CO. Cohen, L. B. (1998). An information-processing approach to infant perception and cognition. In F. Simion & G. Butterworth (Eds.), The development of sensory, motor, and cognitive capacities in early infancy (pp. 277-300), East Sussex: Psychology Press. Cohen, L. B., & Cashon, C, H, (2001a), Do 7-month-old infants process the independent features or facial configurations? Infant and Child Development, 10, 83-92. Cohen, L. B., & Cashon, C. H. (2001b). Infant object segregation implies information integration. Journal of Experimental Child Psychology, 78, 75-83. Cohen, L. B., Chaput, H. H., & Cashon, C. H. (2002). A constructivist model of infant cognitive development. Cognitive Development. 17, 1323-1343.

Development of Specialized Face Perception in Infants

81

de Haan, M., Pascalis, O., & Johnson, M. H. (2002). Specialization of neural

mechanisms underlying face recognition in human infants. Journal of Cognitive Neuroscience, 14, 199-209. Diamond, R., and Carey, S. (1986). Why faces are and are not special: An effect of expertise. Journal of Experimental Psychology: General, 115, 107-117. Fantz, R. L. (1963). Pattern vision in newborn infants. Science, 140, 296-297. Farah, M. J., Tanaka, J. W., & Drain (1995), What causes the face inversion effect?

Journal of Experimental Psychology: Human Perception and Performance, 21, 628-634. Ferguson, K. T., Kulkofsky, S., Cashon’ C. H., & Casasola, M. (2009). The

development of specialized processing of own-race faces in infancy. Infancy, 14, 263-284.

Freire, A., Lee, K., & Symons, L. A. (2000). The face-inversion effect as a deficit in the encoding of configural information: Direct evidence. Perception, 29, 159-170. Gauthier, I., & Nelson, C. A. (2001). The development of face expertise. Current Opinion in Neurobiology, 11, 219-224. Gauthier, I., & Tarr, M. J. (1997). Becoming a “greeble” expert: Exploring mechanisms for face recognition. Vision Research, 37, 1673-1682. Golby, A. J., Gabrieli, J. D. E., Chiao, J. Y., & Eberhardt, J. L. (2001). Differential

responses in the fusiform region to same-race and other-race faces. Nature Neuroscience, 4, 845-850. Goren, C., Sarty, M., & Wu, P. (1975). Visual following and pattern discrimination of face-like stimuli by newborn infants. Pediatrics, 56, 544. Halit, H., de Haan, M., & Johnson, M. H. (2003). Cortical specialisation for

face processing: Face sensitive event-related potential components in 3- and 12-month-old infants. Neurolmage, 19, 1180-1193. Halit, H., Csibra, G., Volein, A., & Johnson, M. H. (2004), Face-sensitive cortical

processing in early infancy. Journal of Child Psychology and Psychiatry, 45, 1228-1234. Hannon, E. E., & Trehub, S. E. (2005). Tuning in to musical rhythms: Infants learn

more readily that adults. Proceedings of the National Academy of Sciences, 102, 12639-12643.

Haxby, J. V., Hoffman, E. A., & Gobbini, M. I. (2000). The distributed human neural system for face perception. Trends in Cognitive Sciences, 4, 223-233. Johnson, M. H., & Morton, J. (1991). Biology and cognitive development: The case of face recognition. Oxford, England: Basil Blackwell. Kanwisher, N. (2000). Domain specificity in fact perception. Nature Neuroscience, 3, 759-763.

Kanwisher, N., & Yovel, G. (2006). The fusiform face area: A cortical region specialized for the perception of faces, Philosophical Transactions of the Royal Society B, 361, 2109-2128. Kanwisher, N. G., McDermott, J., & Chun, M. M. (1997). The fusiform face area: A

module in human extrastriate cortex specialized for face perception. Journal of Neuroscience, 17, 4302-4311. Kelly, D, J., Quinn, P. C., Slater, A. M., Lee, K., Smith, M., Ge, L., et al. (2005).

Three-month-olds, but not newborns, prefer own-race faces. Developmental Science, 8, F31-F36. Kelly, D. J., Liu, S., Ge, L., Quinn, P. C., Slater, A. M., Lee, K., et al. (2007). Cross-

race preferences for same-race faces extend beyond the African versus Caucasian contrast in 3-month-old infants. Infancy, 11, 87-95.

82

Infant Perception and Cognition

Kelly, D. J., Quinn, P. C., Slater, A. M., Lee, K., Ge, L., & Pascalis, O. (2007). The

other-race effect develops during infancy: Evidence of perceptual narrowing. Psychological Science, 18(12), 1084-1089. Lewkowicz, D. J., & Ghazantar, A. A. (2006). The decline of cross-species

intersensory perception in human infants. Proceedings of the National Academy of Sciences, 103, 6771-6774. Macchi Cassia, V. M., Turati, C., & Simion, F. (2004). Can a nonspecific bias toward

top-heavy patterns explain newborns’ face preference? Psychological Science, 15, 379-383. Maurer, D., LeGrand, R., & Mondloch, C. J. (2002). The many faces of configural processing. Trends in Cognitive Science, 6, 255-260.

Meissner, C. A., & Brigham, J, C. (2001). Thirty years of investigating the other-race effect memory for faces: A meta-analytic review. Psychology, Public Policy, and Law, 7, 3-35. Michel, C,, Caldara, R., & Rossion, B, (2006), Same-race faces are perceived more

holistically than other-race faces. Visual Cognition, 14, 55-73. Michel, C., Rossion, B., Han, J., Chung, C., & Caldara, R. (2006). Holistic processing is finely tuned for faces of one’s own race. Psychological Science, 17, 608-615. Nelson, C. (2001). The development and neural bases of face recognition. Infant and Child Development, 10, 3-18. Pascalis, O., Scott, L. S., Kelly, D. J., Shannon, R. W., Nicholson, E., Coleman, M.,

et al. (2005). Plasticity of face processing in infancy. Proceedings of the National Academy of Sciences, 102, 5297-5300. Quinn, P. C,, Yahr, J., Kuhn, A., Slater, A. M., & Pascalis, O. (2002). Representation

of the gender of human faces by infants: A preference for female, Perception, 31, 1109-1121.

Rakison, D. H,, & Woodward, A, L. (2008). New perspectives on the effects of action on perceptual and cognitive development. Developmental Psychology, 44, 1209-1213.

Rhodes, G., Hayward, W. G., & Winkler, C. (2006). Expert face coding: Configural and component coding of own-race and other-race faces. Psychonomic Bulletin & Review, 13, 499-505. Rossion, B., & Gauthier, I. (2002). How does the brain process upright and inverted faces? Behavioral and Cognitive Neuroscience Reviews, 1, 63-75. Pascalis, O., de Haan, M., & Nelson, C, (2002). Is face processing species-specific during the first year of life? Science, 296, 1321-1323. Sangrigoli, $., & de Schonen, S, (2004). Recognition of own-race and other-race faces by three-month-old infants, Journal of Child Psychology and Psychiatry, 45, 1219-1227,

Sangrigoli, S., Pallier, C., Argenti, A, -M., Ventureyra, V. A. G., & de Schonen, S. (2005). Reversibility of the other-race effect in face recognition during childhood. Psychological Science, 16, 440-444, Scott, L. S., Pascalis, O., & Nelson, C. A. (2007). A domain-general theory of the development of perceptual discrimination. Current Directions in Psychological Science, 16, 197-201.

Searcy, J., & Bartlett, J, (1996). Inversion and processing of component and spatial-relational information in faces. Journal of Experimental Psychology: Human Perception and Performance,

22, 904-915.

Sergent, J. (1984). An investigation into component and configural processes underlying face perception. British Journal of Psychology, 75, 221-242.

Development of Specialized Face Perception in Infants

83

Stager, C. L., & Werker, J. F. (1997). Infants listen for more phonetic detail in speech perception than in work-learning tasks. Nature, 388, 381-382. Tanaka, J. W., & Farah, M. J. (1993), Parts and wholes in face recognition, Quarterly

Journal of Experimental Psychology, 46, 225-245. Tanaka, J. W., Kiefer, M., & Bukach, C. M. (2004). A holistic account of the own-race

effect in face recognition: Evidence from a cross-cultural study. Cognition, 93, B1-B9.

Tarr, M. J., & Gauthier, I. (2000). FFA: a flexible fusiform area for subordinate-level visual processing automatized by expertise. Nature Neuroscience, 3, 764-769. Tzourio-Mazoyer, N., de Schonen, S., Crivello, F., Reutter, B., Aujard, Y., & Mazoyer, B. (2002). Neural correlates of woman face processing by 2-month-old infants. Neurolmage, 15, 454-461. Valentine, T. (1988). Upside-down faces: A review of the effect of inversion on face

recognition. British Journal of Psychology, 79, 471-491. Werker, J. F., Cohen, L. B., Lloyd, V. L., Casasola, M., & Stager, C. L, (1998).

Acquisition of word-object associations by 14-month-old infants. Developmental Psychology, 34, 1289-1309. Werker, J. F., & Tees, R. C. (1984). Cross-language speech perception: Evidence for perceptual reorganization during the first year of life. Infant Behavior and Development, 7, 49-63. Yin, R. K. (1969). Looking at upside-down faces, Journal of Experimental Psychology, 81, 141-145,

Younger, B., & Cohen, L. B. (1986). Developmental change in infants’ perception of correlations among attributes. Child Development, 57, 803-815. Yovel & Kanwisher (2005). The neural basis of the behavioral face-inversion effect. Current Biology, 15, 2256-2262.

5 The Role of Perceptual Processes in Infant Addition/Subtraction Experiments Alan M. Slater, J]. Gavin Bremner, Scott P, Johnson,

and Rachel A. Hayes

“,.,£humans innately possess the capacity to perform simple arithmetical

calculations, which may provide the foundations for the development of further arithmetical knowledge.” Karen Wynn (1992, p. 750) “One should be cautious about attributing sophisticated cognitive processes to young infants when simpler processes will suffice.” Cohen and Marks

(2002, p. 200) “It is becoming clear that researchers must address issues relating to familiarization and other basic perceptual processes before assuming more complex understandings or knowledge in infants.” Clearfield and Westfahl

(2006, p. 42)

INTRODUCTION One of the major areas of research into early cognitive development concerns infants’ ability to understand number, given that it leads into later numerical and mathematical competence. Accordingly, there is considerable research on this topic and there is a large body of research suggesting that infants have a least some ability to discriminate between small number sets and between large number sets. As an example, Wilcox and Baillargeon (1998) presented 8- to 11-monthold infants with a sequence in which one object (a ball) disappeared behind a screen, and another (a box) then emerged. This sequence involved either a wide screen that could hide two objects at once, or a narrow screen that could only hide one object. Infants looked longer at the narrow-screen event, and Wilcox and Baillargeon took this as evidence that the infants were aware that

We thank the infants and their parents for participating in the experiments. The research was supported by grants from the Economic and Social Research Council (RES-000-22-1113), the Nuffield Foundation (SGS/32120), and the National Institutes of Health (HD-40432 and

HD-48733).

86

Infant Perception and Cognition

two distinct objects were involved, and that the narrow screen could not hide both objects. In another experiment on object individuation Spelke, Kestenbaum, Simons, and Wein (1995) familiarized 4-month-old infants with one of two

events, both of which involved object movements in relation to two screens: these events are depicted in Figure 5.1. In one event (continuous) the object moved behind one screen, appeared between the two screens, went behind the second screen, and subsequently emerged. The other event (discontinuous) was the same, except that the middle part of the object’s trajectory (between the two screens) was missing. Test events then took place with both screens removed, and involved one or two objects. Infants who had been familiarized with the continuous-trajectory event looked longer at the two-object outcome, while those familiarized with the discontinuous event looked longer at the one-object outcome. The authors took this as evidence that young infants interpreted continuity of motion as indicating a single object, and discontinuity as indicating two objects. The studies by Spelke et al. (1995) and Wilcox and Baillargeon (1998) used a

Violation of Expectancy (VoE) paradigm, based on the assumption that infants will spend longer looking at an unexpected outcome than an expected one, giving clear evidence that young infants are able to individuate objects. These are just two of the many articles that have reported investigations of infants’ detection and responses to number and associated variables, and many of the findings have proven to be highly controversial. These controversies have been at the center of debates between different theoretical perspectives, and this is an excellent context for examining the contributions of those perspectives. The different perspectives can be separated into two different broad theoretical views, emphasizing either an information-processing constructivist, empiricist perspective, or a nativist perspective suggesting innate abilities.

Two of the principles underlying the constructivist information-processing approach are 1) that infants are endowed with a rudimentary, innately determined, and domain-general information-processing system which will improve with development, and 2) “the process is constructive (in) that infants

learn to construct higher, more sophisticated units, from lower, simple units” (Cashon & Cohen, 2003, p. 58). The nativist perspective argues that infants are born with innate abilities or “core knowledge” that is domain-specific and

Continuous Event

le

Discontinuous Event

H I

Figure 5.1. To illustrate the continuous and discontinuous events used by Spelke, Kestenbaum, Simons, and Wein (1995). (See also figure in plate section.)

Role of Perceptual Processes

87

functional early on, and may be elaborated with further development (e.g., Kinzler & Spelke, 2007; Spelke, 1998; Spelke & Kinzler, 2007). Of course, these two broad perspectives have several variations and nuances, which are informed by research findings that are themselves often controversial in that they are sometimes not readily replicable or easily interpretable. Some of these findings and associated controversies are described in this chapter. We begin by describing the evidence for two types of representations of number—one for small item sets, the other for large—together with evidence that these systems are modality-general. This is followed by evidence suggesting that infants may sometimes be responding to continuous variables that are found in displays of discrete items rather than number per se. We then turn to the main focus of the chapter, which is whether infants can add and subtract, or whether their purported arithmetical abilities can be explained in lower-level perceptual terms. It is in this context that the relative contributions of information-processing perspectives are compared with other theoretical views on our understanding of infants’ numerical abilities. TWO SYSTEMS FOR REPRESENTING NUMBER A distinction can be made between two different representations of number, both to be found in human infants and other animal species. With respect to small number sets (four and fewer items) an object-file or object-tracking

system of representation has been proposed (e.g., Simon, 1997; Uller, Carey, Huntley-Fenner, & Klatt, 1999; Xu, 2003). Within the object-file system, each item in a set is represented by a distinct symbol (a file) for that object, and the representation is updated as objects are added to, or taken away from, the set. By representing the total number of items in a set, the individual is able to compare two set sizes and to judge whether a set of items corresponds to an expected numerical outcome. For large number sets (> 4) it has been suggested that an analog-magnitude or ratio system is applicable. This system does not represent the precise number of items in a set, but is able to compare two (or more) sets in terms of the ratio of the set sizes (e.g., Feigenson, Dehaene, &

Spelke, 2004; Mack, 2006; Starkey & Cooper, 1980; Starkey, Spelke, & Gelman, 1990; Xu, Spelke, & Goddard, 2005). Small Item Sets

One of the first demonstrations that infants can discriminate between small number sets but have difficulty with larger sets was by Starkey and Cooper (1980), who found that 5.5-month-olds discriminated 2 from 3 dots, but did not

discriminate between larger numbers. These findings were replicated by Antell and Keating (1983), who tested newborn infants for their ability to discriminate

between 2 versus 3 and 4 versus 6 visual displays of black dots using an habituation paradigm—infants habituated to one display (e.g., either 2 or 3 dots) were then given test trials with a new display (3 or 2 dots). They found that the infants discriminated between 2 and 3 (as indexed by a recovery of attention to the newdot display on the test trials) but not between 4 and 6. These findings mirror

88

Infant Perception and Cognition

those of subsequent researchers, and their conclusion is also similar to those given by later researchers: “We argue that this suggests the ability to abstract numerical invariance from small-set visual arrays and may be evidence for complex information processing during the first week of life” (p. 695). Other researchers have found precise enumeration abilities with human infants for small item-sets. Feigenson and Carey (2003, Experiment 1) hid 1, 2, 3 or 4 ping-pong balls in an opaque box and then allowed 14-month-old infants to reach for them. The infants represented the numerosities of 1, 2 and 3 balls, as indicated by decreasing their searches after retrieving all of the balls they had seen hidden, but they failed at representing 4 — after seeing four balls hidden and retrieving two, they did not then continue searching for further balls. Similar results were found with younger infants. Feigenson, Carey and Hauser (2002) showed crackers being hidden sequentially in two containers to

10- and 12-month-olds, and then allowed the infants to crawl and obtain the cracker(s) from one of the containers. The infants were successful (i.e., they chose the larger quantity) with 1 versus 2 and 2 versus 3, but they failed with the comparison of 3 versus 4, 2 versus 4 and 3 versus 6. Large Item Sets With respect to larger set sizes, it has become apparent that infants (and other species, see below) represent approximate rather than precise magnitudes, and discriminate between number sets on proportionate, or ratio differences; ie., analog magnitudes. Xu and Spelke (2000) habituated 6-month-old infants to visual arrays of either 8 or 16 dots, and on test trials they were shown new displays of the two numerosities. They controlled for continuous variables— total array size, total filled surface area, element size and element density— that would otherwise covary with numerosity. On the test trials the infants looked longer at the display with the novel numerosity, demonstrating that they were discriminating on the basis of numerosity rather than a confounding variable. However, when habituated and tested with 8 vs. 12, or 16 vs, 24, the infants did not discriminate between them, suggesting that the ratio for discriminating numerosities lies somewhere between 2:1 and 3:2. A similar finding—successful discrimination of two ratios that differed by 2.0, and failure with a 1.5 difference—was also reported with 6-month-olds by McCrink and Wynn (2007).

Lipton and Spelke (2003) found that 6-month-olds discriminated large numerosities in the auditory modality that differed by a ratio of 2:1 but not 12 from 8 (3:2), demonstrating that ratio discrimination is not modality-specific (i.e., specific to visual input, a point illustrated further in the next section), and that it is also dependent on the ratio. They also reported that older infants (9-month-olds) successfully discriminated 12 from 8 sounds (i.e., 3:2), giving evidence that “numerosity discrimination increases in precision over development, prior to the emergence of language or symbolic counting” (p. 396). Additional studies confirm this increase in precision with age: 10-month-old infants can discriminate numerosities with a 3:2 ratio (Xu & Arriaga, 2007), and

Role of Perceptual Processes

89

adults with no training in arithmetic are successful with a ratio of 8:7 (Barth, La Mont, Lipton, Dehaene, Kanwisher, & Spelke, 2006), leading to the suggestion that “Abstract numerical quantity representations ... are computationally functional and may provide a foundation for formal mathematics” (p. 199). Numerical Abilities in Nonhuman Animals

It is clear that numerical abilities are not confined to humans, but are found in many nonhuman animal species. For example, Hauser, Carey, and Hauser (2000) tested 200 rhesus macaques for their ability to distinguish between different quantities of apple slices. Each monkey was tested only once, and saw two experimenters place pieces of apple, one at a time, into two opaque containers. The monkeys were then allowed to select one of the containers,

and they successfully selected the container with the largest number of slices when choosing between 1 versus 2, 2 vs. 3, 3 vs. 4 and 3 vs. 5. However, they failed at the comparison between 4 vs. 5, 4 vs. 6, 4 vs. 8 and 3 vs. 8 slices. Beran (2007) reported that rhesus monkeys were also able to enumerate large and small sets of items using analog numerical representations, Others have reported nonverbal numerical judgments in a variety of animals, including pigeons, parrots, racoons, ferrets, rats, cats, New and Old World monkeys, and apes (e.g., Beran, 2007; Brannon, 2006; Cantlon & Brannon, 2006, 2007; Capaldi, 1988; Jordan & Brannon, 2006a, 2006c; Jordan, Brannon, Logothetis,

& Ghazanfar, 2005; Masserman & Rubinfine, 1944; Pepperberg, 2006; Uller, Hauser, & Carey, 2001; Xia, Emmerton, Siemann, & Delius, 2001). These find-

ings are not surprising, and are likely ubiquitous to most animals—the ability to discriminate between one and two portions of food would certainly confer an evolutionary advantage, but an inability to discriminate between “lots” and “lots more” would be of less relevance! Intermodal Representation of Number What is becoming apparent is that infants’ (and other animals’) representation of number is abstract, in that they are able to represent numerosities independently of the sensory modality in which they are presented (e.g., Starkey, Spelke, & Gelman, 1983, 1990). A recent and striking example of intermodal matching was provided by Jordan and Brannon (2006b) who found that 7-month-old infants preferentially attended to dynamic visual displays of human adults that numerically matched the number they heard speaking— i.e., matching the number of entities seen with the number heard — providing evidence that “... by 7 months of age, infants connect numerical representations across different sensory modalities when presented with human faces and voices” (p. 3486). The same group (Jordan et al., 2005) have also shown that rhesus monkeys will preferentially attend to dynamic visual displays showing a number of conspecifics that matched the number of conspecific vocalizations they heard, suggesting that this numerosity matching ability is not confined to humans. Despite these results, there is no reason to assume that infants will always look at the visual display with the same numerosity as the auditory event. For instance, Mix, Levine, and Huttenlocher (1997) and

90

Infant Perception and Cognition

Moore, Benenson, Reznick, Peterson, and Kagan (1987) obtained the opposite auditory-visual matching result, finding that infants looked longer at the display that was not numerically equivalent to the sound pattern. Nevertheless, longer looking at the opposite event still indicates detection of a consistent relationship between visual and auditory numerosities. Kobayashi, Hiraki, and Hasegawa (2005) presented 6-month-old infants with events of either two or three tones, they were then shown either two or three objects, and they looked longer at the numerically nonequivalent event (i.e., the two-tone/three visual objects and the three-tone/two visual objects, a

novelty preference). Ina similar experiment, but using vision and touch, Feron, Gentaz, and Streri (2006) tactually familiarized 5-month-olds with either two

or three objects, presented one by one, in their right hand. Then they visually showed the infants either two or three objects and found greater looking at the nonequivalent visual display (e.g., two-tactual/three visual objects). These several findings confirm the abstract representation of small numbers of objects across sensory modalities (tactile, auditory and visual). Summary In summary, the findings described above demonstrate that in human infants, and in many other animal species, there are two systems for representing and comparing numerical magnitudes: one for the precise representation of small numbers of individual objects (typically four and fewer), and the other for large, approximate magnitudes. These systems appear to be abstract, in that they are independent of the sensory modality in which the items are presented. They also seem to be evolutionarily primitive, emerge early in development, and are independent of both learning and language, although in humans they are likely to constitute the starting point for more sophisticated understandings of number and math. CAN YOUNG INFANTS ADD AND SUBTRACT? Although many researchers have taken these various findings as evidence that at least some numerical abilities are innate (e.g., Antell & Keating, 1983; Wynn, 1992), there have been alternative explanations suggested, such as those based on information-processing and low-level perceptual processes (e.g., Cohen & Marks, 2002; Wakeley, Rivera, & Langer, 2000). These alternatives, and related theoretical issues, are discussed here in the context of infants’ purported arithmetical computational abilities. In addition to being able to represent numbers, it has been suggested that young infants also have the numerical competence to compute the outcomes of addition and subtraction manipulations. This was reported in a highly cited pioneering study by Wynn (1992) in which she claimed that 5-monthold infants can add and subtract. In her addition (1 + 1) condition, infants

were first shown a single doll on a stage. A screen was then lowered to conceal the doll, and a hand appeared holding a second doll; this doll was then placed behind the screen, and the hand emerged empty. In her subtraction

Role of Perceptual Processes

91

(2 - 1) condition, infants were first shown two dolls being placed on the stage

followed by the screen being lowered. An empty hand then appeared, went behind the screen, and emerged holding one doll. A schematic depiction of the addition condition is given in Figure 5.2. If we attribute sufficient numerical competence to infants such that they represent the outcomes of the addition and subtraction events, then the expected outcome in the addition (1+1) condition is two toys, and in the subtraction

(2-1) condition it is one toy: the unexpected outcomes are one toy in the addition condition, and two toys in the subtraction condition, or something other than the arithmetically correct solution. Wynn reported that the infants looked longer at the unexpected outcome and suggests that these findings indicate “... that infants possess true numerical concepts and suggest that humans are innately endowed with arithmetical abilities” (1992, p. 749). Wynn's findings and her interpretation of them have proven to be highly controversial. In

1: An object is placed on the stage

2: A second goes behind the screen 3: The screen is removed to reveal...

...either 2 objects...

...0r 1 object

Figure 5.2. Schematic depiction of a typical addition condition. Note that the firstseen presentation (top) is also the “impossible” outcome at test (bottom). (See also figure in plate section.)

92

Infant Perception and Cognition

the following three sections we discuss these controversies under four headings: (1) Replicability; (2) Addition/Subtraction in Nonhuman Animals; (3) Multiple

Cues for Quantification; (4) Perceptual Versus Cognitive Interpretations. Replicability There have been several failures to replicate some or all of Wynn’s findings. A failure to replicate was reported by Wakeley et al. (2000), who described three experiments, none of which replicated Wynn’s findings. They offer the view that their results, together with those from other studies “... of infant arithmetic, suggest that infants’ reactions to displays of adding and subtracting are variable and, therefore, that infants’ numerical competencies are not robust” (p. 1525). Several partial failures have also been reported. Koechlin, Dehaene, and Mehler (1997), replicated Wynn's finding of greater looking at the “impossible” outcome of subtraction events, but their infants looked equally at “possible” and “impossible” outcomes of addition events; Moore and Cocas (2006) found longer looking at impossible events for female infants, but not for males. Wakeley et al. favor the information-processing hypothesis “... that adding and subtracting are not innate, but rather are arithmetic competencies that develop over the course of several years” (p. 1532).

However, several successful replications of Wynn's findings have been reported using both real three-dimensional displays (e.g., Clearfield & Westfahl, 2006; Cohen & Marks, 2002; Simon, Hespos, & Rochat, 1995; Uller et al., 1999; Walden, Kim, McCoy, & Karrass, 2007; also the present authors’ research, described below) and displays on a computer monitor (Berger, Tzur, & Posner, 2006). The weight of evidence leads Clearfield and Westfahl (2006, p. 35) to the view that “... the behavior Wynn described is both robust and replicable ...” (p. 35). Nevertheless, even when replications are found (such as Cohen & Marks and Clearfield & Westfahl), the experimental designs suggest that something other than arithmetical abilities may underlie infants’ performance on these tasks, a point that is elaborated later. The above comments primarily refer to replications of Wynn's [1 + 1 = either 1 or 2] addition condition and [2 - 1 = either 1 or 2] subtraction condi-

tion—and a caveat is appropriate for attempted extensions that remain within the domain of small numbers but are slightly outside this limited range. In Wynn’s (1992) Experiment 3 she showed 4.5-month-olds a 1 + 1 addition event with an outcome of either 2 or 3, and found that the infants looked longer at the impossible outcome of 3. In follow-up experiments, Wynn (1995) found that 5-month-old infants did not look longer at the impossible outcome of 2 in an addition problem (2 + 1 = 2) when compared with the possible outcome of 2 ina subtraction problem (3 -1 = 2). Additionally, Wynn and Chiang (1998) found that 8-month-olds did not look longer at the impossible outcome

of1 in a subtraction problem (1 - 1 = 1) when compared with the possible outcome of | in an addition problem (0 + 1 = 1). Accordingly, it could be that addition/subtraction problems are most readily solved (or give consistent

Role of Perceptual Processes

93

and replicable results) by young infants when they involve numerosities in the restricted range of one and two items, and that outside of these limits, even within the range 0-3 items, clear evidence of infants’ ability to add and subtract is more difficult to obtain. This view is elaborated upon in the final discussion, together with its implications for theoretical accounts of infants’

numerical competence. Addition/Subtraction in Nonhuman Animals

There are reports that cotton-top tamarins (Uller et al., 2001), capuchin monkeys (Beran, 2008), rhesus monkeys

(Hauser,

MacNeilage,

& Ware,

1996;

Sulkowski & Hauser, 2001) and domestic dogs (West & Young, 2002) are able to compute the outcomes of addition and subtraction operations. With respect to primates, Uller et al. (2001) suggest that “It is likely that these numerical representations are spontaneously available to a variety of primate species and could provide a foundation on which humans’ number sense was constructed over evolution and development” (p. 248). Multiple Cues for Quantification: Number and Continuous Quantity One way in which researchers have dealt with the problems of replicability is by debating the basis of infants’ responding on these tasks. One interpretation of infants’ ability to discriminate between small ( -0.25 —m—0.15

ic et 2 a c

o

=

0.9 +

Correlated

1 Uncorrelated

Test stimulus

Figure 7.5. Mean error on test patterns for 18 networks trained with either deep (0.15, representing 10-month-olds) or moderate (0.25, representing 4-month-olds) score thresholds, simulating differential sensitivity to correlations among older infants,

run with score-threshold settings of 0.15 and 0.25; a few recruited 4 hidden units, There was more variation in networks run at a higher score threshold of 0.5, which recruited 2, 3, or even 1 hidden units. Interaction between Age and Test Stimulus As noted, the most important phenomenon to capture is the interaction between age and the status of the test stimulus. Figure 7.5 shows the mean error on test patterns for 18 networks trained with either deep (0.15, representing 10-month-olds) or moderate (0.25, representing 4-month-olds) score thresholds, interaction F(1, 32) = 14.36, p < .001. At a score threshold of 0.15,

planned dependent t-tests revealed that there was more network error to the uncorrelated test stimulus than the correlated test stimulus—t(8)

= -3.04,

136

Infant Perception and Cognition

p N

Ww

had ny

138

0+

= Correlated

Uncorrelated

Novel

Test stimulus

Figure 7.8. Mean error on test patterns for 18 networks each trained with either low or moderate score thresholds, showing consistent eventual preferences for novel test patterns.

The novel-values condition is merely a control condition, and it is some-

what arbitrary how these novel values are coded in simulations. | coded them as being sharply different from the familiar feature values, but it is possible that infants view them as only slightly different from the familiar. Moreover, there might be a natural limit to attention in the test phase based on exposure time, something that was ignored in the simulations where stimulus error had no such limit. A suitable limit on the amount of attention would reduce attention differences between the novel-values stimulus and the correlated and uncorrelated test stimuli. Discrimination of a Correlated Test Stimulus

To simulate infants’ demonstrated ability to discriminate the correlated test stimulus from the habituation set, we ran 20 CC networks in each of the four conditions listed in Table 7.2 (Shultz & Cohen, 2004). For these simu-

lations, score threshold had the default value of 0.4. In every condition,

there was significantly more error to the correlated novel test stimulus than to the familiar stimulus, showing that the networks clearly discriminated the two test stimuli. This effect is unsurprising for artificial neural networks, because they typically exhibit less error to stimuli that they have been trained on, than to stimuli they have never been trained on. This effect was considerably stronger in networks than in infants, but the important thing is that the effect is in the proper direction and statistically reliable. Because these CC networks learned so quickly (condition means ranged from 3.85 to 8,95 epochs), they almost never recruited a hidden unit. For this reason, it is unnecessary to repeat these simulations with SDCC networks, which are identical to CC networks until more than one hidden unit is recruited.

Computational Modeling of Infant Concept Learning Table 7.3

139

Coverage of the Shift from Features to Correlations

Authors

Effect coverage

Developmental mechanism

Model

Values

Correlation

Shift

| Habituation

Westermann & Mareschal

Shrink-BP

yes

no

no

no

SUSTAIN

yes

yes

yes

no

yes

yes

yes

yes

BP

yes

no

no

no

BP

yes

no

no

no

yes

yes

yes

Yes

Shrink receptive fields Gureckis & Love Recruit more

Oa a

prototypes

Shultz & Cohen Lower score threshold

Thomas Learn longer Shultz & Cohen Lower score threshold Shultz Lower score threshold

SDCC

IS THERE A BEST MODEL? ‘The five computational models reviewed earlier, and the new SDCC model, share several commonalities, They all employ unsupervised or self-supervised connectionist learning, and they attempt to explain apparent qualitative shifts in learning by quantitative variation in learning parameters. Also, the Shrink-BP and SUSTAIN (in the randomized-input version) models both emphasized increased visual acuity as an underlying cause of the shift. Even the learning-depth mechanism in CC and SDCC networks could be interpreted in terms of increasing visual acuity or accommodation; poor visual resolution would retard learning compared to good visual resolution. Finally, the SUSTAIN, CC, and SDCC models all grow in computational power. Comparative patterns of their coverage of the infant data (Younger & Cohen, 1983, 1986) are summarized in Table 7.3. Although commentators emphasized such similarities, and complimented the models on capturing the basic shift from features to correlations (Thomas, 2004;

Younger, Hollich, & Furrer, 2004), a closer look reveals some interesting differences. For example, the Shrink-BP model and Thomas’ static BP model did not include the crucial control employed in Younger and Cohen’s (1986) Experiment 3 that removes the correlated test stimulus from the habituation patterns. As I show later, computer simulations can highlight the importance of this removal because artificial neural networks typically have lower error

140

Infant Perception and Cognition

on trained than on novel test items. This is why training items are routinely kept out of generalization test sets in artificial neural-learning experiments. There are three apparent problems with the SUSTAIN model. One is that the effects obtained by SUSTAIN networks are much smaller than those obtained in infants, requiring 10,000 networks to reach statistical significance (Gureckis & Love, 2004). In contrast, the number of CC and SDCC networks

matched those of the nine infants tested in each condition of Experiment 3 (Younger & Cohen, 1986), and the results were of approximately the same magnitude as those in infants (Shultz & Cohen, 2004). Second, there were sev-

eral parameter values that had to be optimally fit in SUSTAIN, but not in CC or SDCC. Third, like the well-known ALCOVE model for concept-learning (Kruschke, 1996), SUSTAIN employs a separate attention mechanism that CC and SDCC do not require. One way to put it is that SUSTAIN attends in order to learn, whereas CC and SDCC learn to attend. Likewise, the Shrink-BP (Westermann & Mareschal, 2004) model exhibits

a number of problems. Somewhat like SUSTAIN, Shrink-BP required far more networks per condition (1000) than the nine infants and nine CC or SDCC networks per condition. Another problem is that Shrink-BP learned the training set very slowly, taking 1000 epochs versus the 50 taken by CC and SDCC networks. Although it is not generally possible to equate human learning time with that taken by computational models, because humans might engage in indeterminate amounts of extra processing and learning during or between trials, it is clear that CC and SDCC are roughly in the ballpark of the few minutes of habituation used in the infant experiments, whereas Shrink-BP is well outside of that ballpark, Despite covering Experiment 2, which kept the correlated test stimulus in the training set, less than half of Shrink-BP networks showed more error to uncorrelated than correlated test patterns in the more challenging Experiment 3, at the smallest receptive field size of 0.15. Thus, coverage of the most important finding documenting a shift from features to correlations is not robust in the Shrink-BP model. The static BP model (Thomas, 2004) had the same problem, by virtue of not removing the correlated test stimulus from the training set. Again, failing to remove this item sets up a very trivial problem for neural networks. Apart from prototype phenomena, they will inevitably generalize better to items on which they have been trained than to novel items. To see how trivial this problem becomes with the duplication of this stimulus in training and test sets, [ran nine SDCC networks with Thomas’ training and test patterns, using the same parameter settings reported earlier. All of these networks recruited only a single hidden unit and required fewer than one-half as many epochs as did networks in my SDCC model. Even more telling was that the mean network error for the uncorrelated test stimulus was 35 times greater than that for the correlated test stimulus, at both low (0.15)

and moderate (0.25) score thresholds. This is far too large an effect of test stimulus for simulating the infant experiments, F(1, 16) = 145,451, p < .001,

and there was no differential sensitivity to correlation detection, interaction F(1, 16) = 025.

Computational Modeling of Infant Concept Learning

141

Another problem with Thomas’ static BP model is that it required a large difference in training time to manipulate the age difference—100 epochs for older infants versus 16 epochs for younger infants. This is a very large difference in amount of neural-network training, considering that 10- and 4-monthold infants had equal exposure to the habituation stimuli. Finally, there were several other simplifications in Thomas’ model, ensuring that his network’s task was different from, and simpler than, the task faced by infants. We had previously investigated whether static BP encoder networks, equipped like the Thomas model with three hidden units, could cover these key interactions between score threshold (implementing age differences) and test stimulus (Shultz & Cohen, 2004). Our back-propagation simulator was modified to use a score-threshold parameter to determine learning success as in CC, allowing a direct comparison of CC and BP networks. Ordinarily, BP networks stop learning when they reach a particular, programmer-set error level. We varied BP network topology to explore the possible roles of network depth, and the presence of cross connections. The coding scheme that we had used in the CC simulations was also used with BP networks. In all of our BP simulations, we used the default parameter settings of 0.5 for learning rate, and 0.9 for momentum. The learning-rate parameter regulates the amount of weight change, and is typically set to a moderate value to allow reasonably fast learning, but not so fast that weight changes continue oscillating over more optimal values. The momentum parameter provides weights with some inertia or momentum, so that they would change less when the last change was small, and change more when the last change was large. The basic idea behind momentum is to induce larger weight changes when the weight is far from the minimum error, and small weight changes when the weight seems close to the minimum error, We ran nine networks in each condition, and varied score-threshold values from 0.05 to 0.50 in nine steps of 0.05, sampling a wide range of learning depth. There were nine networks at each of ten score-threshold values for each of the two habituation sets, yielding 180 networks, each run for a maximum of 300 epochs. Because most standard BP networks place all hidden units on a single layer, without either cascaded weights between hidden units or cross connections bypassing hidden unit layers, we started with this. In addition to the always-on bias input unit, there were six input units coding the stimuli, three hidden units on the second layer, and six output units to register the network’s response. These 6-3-6 networks were thus essentially similar to Thomas’ (2004) BP network and my CC and SDCC networks, but showed considerable

difficulty detecting correlations between stimulus features, as measured by higher error on uncorrelated than correlated test items, We also tried a deeper BP network to see whether CC’s advantage of greater network depth was sufficient to cover the infant data. These 6-1-1-1-6 BP networks produced about the same results as the flat 6-3-6 networks did. It was extremely rare to find any significant difference in t-tests comparing error on the correlated versus uncorrelated test stimuli. Finally, we tried adding the presence of cross-connections to both the 6-3-6 and the 6-1-1-1-6 BP networks, eliminating direct

142

Infant Perception and Cognition

input-to-output connections as in the CC simulations. Results continued to be negative—no comparison revealed lower error on correlated than on uncorrelated test items, which is the signature of effective correlation detection. For this chapter, I ran another batch of BP simulations with a much lower learning rate of 0,05. By making smaller changes in connection weights, this learning rate would reduce the possibility of networks oscillating across error minima. I used 6-3-6 BP networks, which precisely duplicated the topology of the SDCC networks featured in this chapter. As with our previous BP simulations, | varied score-threshold values from 0.05 to 0.50 in nine steps of 0.05, again sampling a wide range of learning depth. At score thresholds above 0.05, all of these networks learned the training patterns perfectly. Despite that, it was extremely rare for error to differ significantly in correlated versus uncorrelated test patterns. In all, | ran nine networks at nine levels of score threshold, with the two habituation training sets from Experiment 3 (Younger & Cohen, 1986). With three full replications, there were 54 (3 x 18) simulations with

nine networks each. In each of the three replications, there were 2 or 3 out of 18 significant (p < .05) dependent t-tests comparing mean error on correlated and uncorrelated test patterns, Each of these significant t-tests reflected more error to the correlated than the uncorrelated test stimulus, a difference suggesting a similarity effect rather than a correlation effect. However, these rare significant results could not be duplicated across habituation sets and replications, suggesting that these differences are not robust. In other words, BP networks were again unable to capture the results of the 10-month-olds in this critical Experiment 3 (Younger & Cohen, 1986), in

that they did not show a correlation effect. A robust, unconfounded transition from a values effect to a correlation effect apparently requires a network that is capable of growing while learning. Although it is difficult to prove that an algorithm such as BP cannot in principle cover some phenomena, we gave this algorithm a fair chance, running nine networks in each of 72 simulations (2 habituation sets x 9 score-threshold levels x 4 network topologies) in our previous paper, and the present 54 simulations at a lower learning rate. The Habituation Effect

Another point of contrast between models is that when our CC networks with a low score threshold were repeatedly tested over the habituation phase, they showed an early similarity effect, followed by a correlation effect (Shultz & Cohen, 2004). Tests of this prediction found that 10-month-olds who habituated to training stimuli looked longer at uncorrelated than correlated test stimuli, but those who did not habituate actually looked longer at correlated than uncorrelated test stimuli (Cohen & Arthur, 2003). CC and SDCC are, so

far, the only computational models to capture this habituation effect across an experimental session at a single age level. Shrink-BP, static BP, and SUSTAIN would not be able to cover this effect because their mechanisms operate over ages, not over trials. In contrast to these algorithms, CC and SDCC mechanisms operate over both trials and ages.

Computational Modeling of Infant Concept Learning

143

All Models are Wrong Most knowledgeable modelers, myself included, are in full agreement with the maxim from the statistician George Box (1979) that “All models are wrong but some are useful (p. 202).” Many non-modelers might apply a more negative spin to the same observation, creating the logically equivalent “All models are wrong and some are useless.” This is not necessarily a bad thing, so much as it is a natural fact of scientific pursuits. It is equally true that all theories are wrong but some are useful, the essential difference between models and theories being that models are more fully specified than theories are. Because good models are relatively well specified, it is easier to tell whether and how they are wrong—but theories are at least as likely to be wrong, even if it is more difficult to figure out how and where they are wrong. The essential reason that all theories and all models are wrong is that they abstract across the details of the real life that they attempt to characterize and explain. The hope of their proponents is that a model (or theory) could get the essential principles right and thus be useful, even at the expense of missing some less important details. Real life is undoubtedly far too complex to yield easily to a complete explanation by mere humans, even if guided by fancy theories or models. ADVANTAGES OF THE SDCC MODEL In this skeptical but candid spirit, I readily agree that all existing models of the infant shift from features to correlations are wrong, including especially our own CC and SDCC models. But given the problems just identified for each of the existing alternative models, it seems reasonable to conclude that CC and SDCC are the most correct and most useful of the current crop of models. These two constructive neural-network models capture all of the main features of the infant data, while remaining in the approximate ballpark of the infant experiments in terms of effect size, and amount and kind of experience. Their usefulness in generating new explanations and new predictions has already been confirmed. Because developing visual acuity and accommodation across the first year could be expected to enhance learning, the CC and SDCC manipulations of the score-threshold parameter (lower for older ages) are compatible with that

explanation of the features-to-correlation shift. The newer SDCC model might be preferred over the CC model, because SDCC provides equivalent coverage of the infant data with three fewer connection weights. Because this sort of shift occurs in several different domains and at different ages (Cohen, 1998; Younger et al., 2004), the more general explanation offered by deeper learning in SDCC seems preferable to more restricted explanations. Among documented evidence for shifts from learning about stimulus elements, to learning about relations between those elements, are the following. During habituation to angles of different size and orientation, infants at 1.5 months learn about line orientations, whereas 3-montholds also learn about angle size; that is, the relation between line segments (Cohen & Younger, 1984). Habituation experiments on causal launching

144

Infant Perception and Cognition

events revealed that 6-month-olds learned about objects or patterns of movement, whereas 10-month-olds learned about causal relations between object movements (Oakes & Cohen, 1990). There seems to be a shift from process-

ing independent static features of a motion event at 10 months, to noticing correlations among object parts and trajectories at 14 months (Rakison & Poulin-Dubois, 2002). In experiments on infants’ understanding of the form and function of objects, 10-month-olds learned about form, 14-month-olds learned about form and function as independent features, and 18-montholds discovered the relation between form and function (Madole, Oakes, & Cohen, 1993). Finally, 14-month-olds learned associations between words and objects, whereas infants between 8 and 12 months did not learn such associations, even though they seemed to learn information about independent words and objects (Werker, Cohen, Lloyd, Casasola, & Stager, 1998). All of these experiments employed variations of the habituation-dishabituation, switch-design paradigm. Importantly, any explanation for such shifts that is tied to a single perceptual modality, or a particular maturation of that modality in a particular age range, will not be capable of covering this wide range of phenomena. Covering these phenomena, ranging as they do across modalities, ages, and content, will require a more abstract and fundamental process. Depth of learning would seem to be a good candidate for offering such coverage, but critical simulations remain to be done. Limitations of the CC and SDCC Models

But if all models are wrong, what is wrong about the CC and SDCC models of the Younger and Cohen concept-learning experiments? Or, as it is more usually put in the context of model improvement, what are the current limitations of these models? As noted, both CC and SDCC models exaggerate the amount of attention paid to test stimuli with novel feature values. It is likely that this could be fixed by altering the stimulus-coding scheme to render the novel values somewhat less novel, or alternatively, putting a ceiling on the amount of network error (representing interest and looking time). But any such fix has nothing inherent to do with the models themselves. There is also a less-than-perfect match between the training epochs that these models require (50-some) and the number of actual habituation trials (nine or twelve 20-s exposures) given

to infants. As noted, it is notoriously difficult to equate participant trials with model epochs, because the amount of processing cannot be precisely limited or controlled in animal or human participants. Nonetheless, a hypothetical new model might well supplant the current CC and SDCC versions by approaching the actual number of infant trials even more closely. Finally, the relatively deeper learning of older versus younger infants is not so much fully modeled as it is approximated by shortcut parameter settings in the CC and SDCC models. This is equally a problem for the SUSTAIN, Shrink BP, and static BP models reviewed here, as it is for many computational models of different developmental phenomena (Shultz & Sirois, 2008), In CC and SDCC models, age differences

Computational Modeling of Infant Concept Learning

145

were implemented by variation in the score-threshold parameter, which was set higher for younger infants and lower for older infants. This implementation was justified in this (Shultz & Cohen, 2004) and other developmental domains (Sirois & Shultz, 1998) by the well-documented tendency for older individuals to learn more from the same exposure time than younger individuals. However, such parameter manipulation is clearly a shortcut, because the modeling does not specify how and why any extra or deeper processing occurs. As noted earlier, older individuals might pay more attention, rehearse during intertrial intervals, process information faster, or be better able to see and thus learn from visual stimuli. Further psychological and modeling research might yet resolve some of these mysteries but for now, manipulation of score threshold is a convenient shortcut that allows for a clear separation of learning time and learning depth, making it (so far) uniquely possible to account for changes both within and beyond experimental trials in this domain. MISINTERPRETATIONS OF CC MODELS Although it is too early for published commentary on the SDCC model presented here, our CC modeling of concept learning in infants (Shultz & Cohen, 2004) has attracted a number of published comments that seem to merit some response. Batch versus Pattern Mode Gureckis and Love (2004) thought that we (Sirois & Shultz, 1998) and other

modelers (Mareschal & French, 2000) strayed too far from the procedures of psychology experiments by adjusting connection weights in so-called batch mode after each epoch, instead of after each stimulus pattern. Actually, however, there is some controversy about the relative merits of batch versus pattern training in terms of both psychological plausibility and learning effectiveness. Even if pattern training is more plausible at first glance, there is both psychological (Oden, 1987) and neurological (Dudai, 1989; Squire, 1992) evidence for batch learning. For example, batch processing may aid in the transfer of information from the hippocampus to cortical regions. Computationally, batch learning is considered to be more efficient than pattern learning because it avoids the setting and changing of connection weights that typically result from adjusting weights after each pattern. Even in batch-learning algorithms like CC and SDCC, output activations are compared to their target values one pattern at a time. Thus, these algorithms never have to process more than one training pattern simultaneously, although they do need to keep a running sum of network error that is eventually used to adjust the weights. Age and Learning Trials Gureckis and Love (2004) also complained that the number of learning trials experienced in CC networks depends on which age group is being modeled, even for cases in which all age groups in the actual experiment receive an

146

Infant Perception and Cognition

equal number of learning trials. Thomas (2004) noted this without complaint,

and indeed used it as justification for his own manipulation of age in simulating the Younger and Cohen experiments with a static BP network, by varying the number of learning epochs, Although some CC models do indeed exhibit a negative correlation between score threshold and epochs to learn (Sirois & Shultz, 1998), no inconsistency is implied, because manipulation of the scorethreshold parameter is based on the idea that older individuals learn more from the same stimulus exposure time—a phenomenon that is well supported in developmental studies, as noted earlier. Moreover, this claim is simply false for the CC and SDCC models of the infant shift from features to correlations. For example, Figure 7.9 plots the mean epochs to reach various score-threshold values in the SDCC simulations reported here. Although the main effect of score threshold is significant, F(2, 48) = 45.66, p < .001, this is due entirely to the fewer epochs required at the highest score threshold of 0.5, as this was the only condition that differed significantly from others by the Honestly Significant Differences (HSD) post-hoc test, p < .05. Pointedly, networks representing 10-month-olds (0.15) do not take any longer to learn than networks representing 4-month-olds (0.25).

Mean epochs

Epochs to learn in CC-family networks are very often highly correlated with the number of hidden units that need to be recruited. Thus, it is not surprising to find that number of recruited hidden units shows a similar pattern across conditions as number of epochs does, as shown in Figure 7.10. Again, the main effect of score threshold, F(2, 48) = 22.29, p < .001, was due entirely to the highscore-threshold condition being different from the other two conditions, by the HSD test, p < .05. The lack of difference in epochs and hidden recruits between the low and moderate levels indicates that the deeper learning allowed by lowering score threshold is more subtle than merely adding learning trials. Recall that the key finding to simulate in this domain was the interaction between these two levels of score threshold and the type of test stimulus.

0.15

0,25

0.5

Score threshold

Figure 7.9. Mean epochs and standard deviations to reach various score-threshold values in 18 networks, showing that those with low score thresholds (0.15, representing 10-month-olds) do not take longer to learn than those with moderate score thresholds (0.25, representing 4-month-olds).

Computational Modeling of Infant Concept Learning

147

nN

Mean hiddens

w

4

0.15

0.25

0.5

Score threshold

Figure 7.10. Mean hidden units recruited, with standard deviations, for 18 networks each trained with three different score-threshold values, showing that those with low score thresholds (0.15, representing 10-month-olds) do not recruit more than those

with moderate score thresholds (0.25, representing 4-month-olds).

Flat Versus Deep Network Topologies Gureckis and Love (2004) also preferred flat network topologies as constructed

by their SUSTAIN algorithm over the deep, one-unit-per-layer topologies created by the CC algorithm. The relative merits of deep and flat network topologies is a complex problem that has been well analyzed elsewhere (Dandurand, Berthiaume, & Shultz, 2007). While I disagree with any overly simple assertion about the overall superiority of flat or deep topologies, it is worth noting that SDCC automatically constructs a topology to suit the problem at hand, ranging from entirely deep, as in CC, to entirely flat, as in SUSTAIN, and anything in between. As it turned out, for the simulations reported here, all SDCC networks were constructed flat. Model Parsimony Gureckis and Love (2004) argued that SUSTAIN was relatively parsimonious because it provides a mechanistic continuum of concept-learning from infancy to adulthood. This implied algorithm comparison overlooks the fact that CC-family networks are equally applicable to adult learning and children’s learning, and have been successfully applied to many more domains than concept-learning, including language (Oshima-Takane, Takane, & Shultz, 1999; Shultz & Bale, 2001, 2006; Shultz & Gerken, 2005), mathematics

(Egri & Shultz, 2006), and problem solving (Shultz, Rivest, Egri, Thivierge, & Dandurand, 2007) in infants, older children, and adults. The fact that a sin-

gle algorithm, or close family of algorithms, can be applied without major alterations to such a wide range of psychological problems is a strong argument in their favor. The principal alteration required in CC-family applications is defining the particular training and test sets required by the domain of interest.

148

Infant Perception and Cognition

Constructivism

Thomas (2004) argued against our claim that some sort of constructivist network growth was required to capture the Younger and Cohen infant data by noting that his static BP could also do the job, But, as I showed earlier, Thomas’ model did not actually simulate the key features of the infant experiments, His model, which was designed primarily for didactic purposes, made several simplifications, including reducing the numbers of features and feature values, Most importantly, as noted, Thomas avoided the critical Experiment 3 (Younger & Cohen, 1986). Instead he tried a simplified version of Experiment 2 from the same paper, in which the correlated test stimulus was also in the training set. As noted, this greatly simplifies learning problem and avoids the principal challenge posed by the infant experiments. Thomas (2004) employed a 4-3-4 BP network, the numbers referring to the numbers of input, hidden, and output units, But when we tried a 6-3-6 static BP topology on the actual, critical Experiment 3, in this and previous papers, it failed to show any consistent correlation effect, or the transition from values to correlation effects. In their comprehensive critique of the SUSTAIN, Shrink-BP, and CC models, Younger et al. (2004) argued that none of these network models truly implements multiple developing layers. They allowed that our CC model (Shultz & Cohen, 2004) came close, but was limited to just three layers, all of which they claimed were trained at once. They seemed to prefer a system in which each layer had to self-organize and then send the resulting signal on, However, this

ignores the fact that CC and SDCC layers are indeed trained one at a time, with each layer self-organizing, freezing just-learned input weights, and sending its output results on to the next layer. Unlike static BP networks, to which their critique would apply, later hidden units in CC and SDCC are not even in the target network until they are recruited. Also on Younger et al.’s (2004) wish list is a model in which topology change is an emergent property of the system, rather than being made “at the hand of the modelers.” But the ability to automatically add hidden units as needed is an essential operating characteristic of CC and SDCC networks. No interference of the modeler is required, or even permitted. New computational power and the resulting novel knowledge representations are truly emergent properties of the algorithm itself. The future is already here in this regard. The inherent computational constructivism of the SDCC and CC algorithms is quite compatible with the psychological constructivism that Cohen (1998) has been developing based on psychological evidence alone. CONCLUSIONS In the introductory section, I asked whether and how Cohen’s ideas on habituation and dishabituation might survive or be transformed in the contemporary study of developmental science. From the perspective of computational modeling, I would conclude that these ideas have not only survived, but have proved to be inspirational. At least five different computational models have

Computational Modeling of Infant Concept Learning

149

focused on the seven experiments from the two Younger and Cohen papers on

the infant shift from features to correlations. This is rather a lot of modeling attention, given that computational approaches to developmental psychology are still in their own infancy. What has changed is the nature of the explanation offered for this developmental shift. In the mid 1980s, the shift was interpreted as some sort of qualitative change in the way infants learned. All of the current computational models concur in viewing this shift as due to underlying quantitative, not qualitative, changes. Infants eventually come to see correlations among feature values because of improved visual acuity (Gureckis & Love, 2004;

Westermann & Mareschal, 2004), an increase in the number of possible prototypes (Gureckis & Love, 2004), or deeper learning (Shultz & Cohen, 2004; Thomas, 2004). These computational explanations complement the fact that Cohen (1998) himself had been moving toward quantitative explanations of this shift based on accumulating evidence that it occurs in several different domains at a variety of ages.

Something else that the computational modelers agreed on was that they could not model the 7-month-old data in the Younger and Cohen experiments, who showed neither value nor correlation effects. In a proper quantitative explanation of age differences, how could a mid-aged group escape coverage and understanding? Twenty-some years later, their lack of habituation to test stimuli is still mysterious, but what all of the computational models do get right is that if participants don’t habituate to the training stimuli, then they don't show differential responses to test stimuli. Such observations, along with the published commentaries on the simulations (Thomas, 2004; Younger et al., 2004), may contribute to the somewhat popular view that modeling is rather easy, and that any sort of model can work on psychological data. This would not be an especially positive outcome

for modeling, because it would mean that modeling is telling us nothing useful, particularly in discriminating among different theoretical explanations. This chapter stressed that a somewhat deeper analysis reveals that not all of these models are equally successful in explaining these developmental shifts. Of the current crop of models, CC and SDCC, while limited, seem to be the most correct, most useful, and most likely to be capable of extension to related domains. These two constructive neural-network models captured all of the main features of the infant data, while remaining faithful to the infant experiments in terms of effect size, and amount and kind of experience. Moreover, their usefulness in generating so-far-unique new predictions (the habituation effect) has already been confirmed with psychological evidence. The fact that flat-topology SDCC covers the infant data as well as deep-topology CC, means that final network topology is not nearly as important as the ability to grow a network from simple to complex, echoing a conclusion from a variety of developmental domains (Shultz et al., 2007). By comparison, other models, while capturing some of the key phenomena, suffered from statistically weak effects, avoidance of the real challenge posed by the infant experiments, requiring far more stimulus exposures than the infants needed, or positing

150

Infant Perception and Cognition

explanatory mechanisms that are too restricted to be extended to the wide range of phenomena on the shift from learning about elements to detecting relations between elements. Computational modeling might also help to address the theoretical issue that inspired this whole line of research, namely whether perceptual development is more properly characterized by information differentiation (Gibson, 1969) or integration (Hebb, 1949). The CC and SDCC simulations show that a

single computational system can both differentiate and integrate information. If any developmental shift occurred in these simulations, it was always from features to correlations, not the reverse, thus supporting Hebb’s informationintegration idea. But the same networks making that shift also allowed for finer discrimination between stimulus categories, as can be seen in Figure 7.8. Moderate score thresholds allowed differentiation only of novel versus familiar feature values, effectively creating just two stimulus categories. But lower score thresholds allowed, in addition, differentiation of correlated versus uncorrelated feature values, effectively creating three stimulus categories: correlated, uncorrelated, and novel. Future work might profitably explore this possibility that information integration and differentiation can proceed together as development unfolds. REFERENCES Baluja, $., & Fahlman, S. E. (1994). Reducing network depth in the cascadecorrelation learning architecture. (Technical Report No. CMU-CS-94-209).

Pittsburgh, PA: School of Computer Science, Carnegie Mellon University. Banks, M. S., & Salapatek, P. (1983). Infant visual perception. In M. M, Haith & J. J. Campos (Eds.), Handbook of child psychology: Infancy and developmental psychology (Vol, 2, pp. 109-160). New York: Wiley. Box, G. E. P., & Draper, N. R. (1987). Empirical model-building and response surfaces. New York: Wiley. Cohen, L. B. (1979), Our developing knowledge of infant perception and cognition. American Psychologist, 34(10), 894-899. Cohen, L. B. (1998). An information-processing approach to infant perception and cognition. In F. Simion & G. Butterworth (Eds.), The development of sensory, motor, and cognitive capacities in early infancy (pp. 277-300). East Sussex: Psychology Press. Cohen, L. B., & Arthur, A. E. (2003). The role of habituation in 10-month-olds’ categorization: Unpublished. Cohen, L. B., & Younger, B. A. (1984). Infant perception of angular relations. Infant Behavior and Development, 7(1), 37-47.

Dandurand, F., Berthiaume, V., & Shultz, T. R. (2007). A systematic comparison of flat and standard cascade-correlation using a student-teacher network approximation task. Connection Science, 19(3), 223-244. Day, M. C. (1975). Developmental trends in visual scanning. In H. W. Reese (Ed.), Advances in child development and behavior (Vol. 10, pp. 153-193). New York: Academic Press, Dudai, Y. (1989). The neurobiology of memory: Concepts, findings, and trends. Oxford: Oxford University Press.

Computational Modeling of Infant Concept Learning

151

Egri, L., & Shultz, T. R. (2006). A compositional neural-network solution to primenumber testing. In R. Sun & N. Miyake (Eds.), Proceedings of the Twenty-eighth Annual Conference of the Cognitive Science Society (pp. 1263-1268). Mahwah, NJ: Lawrence Erlbaum. Gibson, E. J. (1969). Principles of perceptual learning and development. New York: Appleton-Century-Crofts. Gould, E., & Gross, C. G. (2002). Neurogenesis in adult mammals: some progress and problems. Journal of Neuroscience, 22(3), 619-623.

Gureckis, T. M., & Love, B. C. (2004). Common mechanisms in infant and adult category learning. Infancy, 5(2), 173-198. Hagen, J. W., & Hale, G. A. (1973). The development of attention in children. In

A. D. Pick (Ed.), Minnesota symposia on child psychology (Vol. 7, pp. 117-140). Minneapolis: University of Minnesota Press. Hagen, J. W., Jongeward, R. H., & Kail, R. V. (1979). Cognitive perspectives on the development of memory. In A. Floyd (Ed.), Cognitive development in the school years (pp. 129-161). New York: Halsted. Hainline, L., & Abramoy, I. (1992). Assessing visual development: Is infant vision good enough? In G. Rovee-Collier & L. P. Lipsitt (Eds.), Advances in infancy research (Vol. 7, pp. 39-102). Norwood, NJ: Ablex. Haith, M. M. (1990). Progress in the understanding of sensory and perceptual processes in early infancy. Merrill-Palmer Quarterly, 36(1), 1-26. Hebb, D. O. (1949). The organization of behavior. New York: Wiley. Kail, R, (1991), Developmental changes in speed of processing during childhood and adolescence. Psychological Bulletin, 109(3), 490-501. Kruschke, J. K. (1996). Dimensional relevance shifts in category learning. Connection Science, 8(2), 225-247. Madole, K. L., Oakes, L. M., & Cohen, L. B. (1993). Developmental changes in infants’ attention to function and form-function correlations. Cognitive Development, 8(2), 189-209. Mareschal, D., & French, R. M. (2000). Mechanisms of categorization in infancy. Infancy, 111), 59-76. Mareschal, D., French, R. M., & Quinn, P. (2000). A connectionist account of

asymmetric category learning in infancy. Developmental Psychology, 36(5), 635-645. Mareschal, D., & Shultz, T. R. (1996). Generative connectionist networks and

constructivist cognitive development. Cognitive Development, 11(4), 571-603. Miller, P. H. (1990). The development of strategies of selective attention. In D. F. Bjorklund (Ed.), Children’s strategies: Contemporary views of child development (pp. 157-184). Hillsdale, NJ: Erlbaum. Oakes, L. M., & Cohen, L. B. (1990). Infant perception ofa causal event. Cognitive Development, 5 (2), 193-207. Oden, G. C. (1987). Concept, knowledge, and thought. Annual Review of Psychology, 38, 203-227.

Oshima-Takane, Y., Takane, Y., & Shultz, T. R. (1999), The learning of first and second pronouns in English: Network models and analysis. Journal of Child Language, 26(3), 545-575. Quartz, S. R., & Sejnowski, T. J. (1997). The neural basis of cognitive development: A constructivist manifesto. Behavioral and Brain Sciences, 20(4), 537-596. Quinn, P. C., & Eimas, P. D. (1996). Perceptual organization and categorization in young infants. Advances in Infancy Research, 10, 1-36.

152

Infant Perception and Cognition

Rakison, D. H., & Poulin-Dubois, D. (2002). You go this way and I'll go that way: Developmental changes in infants’ attention to correlations among dynamic features in motion events. Child Development, 73(3), 682-699. Shultz, T. R. (2003). Computational developmental psychology. Cambridge, MA: MIT Press.

Shultz, T. R. (2006). Constructive learning in the modeling of psychological development. In Y. Munakata & M. H. Johnson (Eds.), Processes of change in brain and cognitive development: Attention and performance XXI. (pp. 61-86), Oxford, UK: Oxford University Press. Shultz, T. R., & Bale, A. C. (2001). Neural network simulation of infant familiarization to artificial sentences: Rule-like behavior without explicit rules and variables. Infancy, 2(4), 501-536. Shultz, T. R., & Bale, A. C. (2006). Neural networks discover a near-identity relation to distinguish simple syntactic forms. Minds and Machines, 16(2), 107-139. Shultz, T. R., & Cohen, L. B. (2004). Modeling age differences in infant category learning. Infancy, 5(2), 153-171,

Shultz, T. R., & Gerken, L. A. (2005). A model of infant learning of word stress. In Proceedings of the Twenty-seventh Annual Conference of the Cognitive Science Society (pp. 2015-2020), Mahwah, NJ: Erlbaum. Shultz, T. R., Mysore, S. P., & Quartz, S, R. (2007). Why let networks grow?

In D. Mareschal, S. Sirois, G. Westermann & M. H. Johnson (Eds.), Neuroconstructivism: Perspectives and prospects (Vol. 2, pp. 65-98). Oxford, UK: Oxford University Press. Shultz, T. R., Rivest, F., Egri, L., Thivierge, J.-P., & Dandurand, F. (2007). Could

knowledge-based neural learning be useful in developmental robotics? The case of KBCC. International Journal of Humanoid Robotics, 4(2), 245-279.

Shultz, T. R., & Sirois, S. (2008). Computational models of In R. Sun (Ed.), Cambridge handbook of computational New York: Cambridge University Press. Sirois, S., & Shultz, T. R. (1998). Neural network modeling in discrimination shifts. Journal of Experimental Child

developmental psychology. psychology (pp. 451-476). of developmental effects Psychology, 71(3), 235-274.

Sokolov, E. N. (1963). Perception and the conditioned reflex. Hillsdale, NJ: Lawrence

Erlbaum and Associates. Squire, L. R. (1992). Memory and the hippocampus: A synthesis from findings with rats, monkeys, and humans. Psychological Review, 99(2), 195-231. Thomas, M. S. C. (2004). How do simple connectionist networks achieve a shift from “featural” to “correlational” processing in categorization? Infancy, 5(2), 100-207. Werker, J. F., Cohen, L, B., Lloyd, V. L., Casasola, M., & Stager, C. L. (1998).

Acquisition of word-object associations by 14-month-old infants. Developmental Psychology, 34(6), 1289-1309. Westermann, G., & Mareschal, D. (2004). From parts to wholes: Mechanisms of

development in infant visual object processing. Infancy, 5(2), 131-151. Younger, B. A., & Cohen, L. B. (1983). Infant perception of correlations among attributes. Child Development, 54(4), 858-867. Younger, B. A., & Cohen, L, B, (1986). Developmental change in infants’ perception of correlations among attributes. Child Development, 57(3), 803-815. Younger, B, A., Hollich, G,, & Furrer, S$. D. (2004), An emerging consensus: Younger and Cohen revisited. Infancy, 5(2), 209-216.

septal nuclei

fornix

cingulum

medial forebrain bundle

basal nucleus to the amygdala fre

inhi one

formation

to the card spinal

2

Heart Rate Change

(bpm)

-

:

3

~ ov

,

aa

oS

oS

4

3

~ o = 3

s| { |8

3

or -8 -10

T

-5

T

0

t

5

10

15

T

20

r

7

25

Seconds following stimulus onset

Figure 2.3. The cholinergic (upper left) and noradrenergic (upper right) neurotrans-

mitter systems of the brain. These systems have ascending influence on the brain, and descending influence on the cardiac pacemaker cells of the heart. Heart rate changes in infants reflect the arousal properties of these neurotransmitter systems.

eee Nc During Attention

Frequent Familiar

Infrequent Familiar

Infrequent Novel

Nc During Inattention

A GP C2) Frequent Familiar -10 mV

Infrequent Familiar

Infrequent Novel +10 mV

Figure 2.5. Topographical scalp potential maps as a function of brief stimulus presentation type and attention phase (top figures:attentive; bottom figures: inattentive). Source: From “Attention affects the recognition of briefly presented visual stimuli in infants: An ERP study,” by J. E. Richards, 2003, Developmental Science, 6, p. 319. Copyright 2003 by Wiley-Blackwell. Reprinted with permission.

Control

Posterior Prefrontal

Front Pole

Inferior Prefrontal

Figure 2.6. Cortical source analysis of the Nc ERP component with realistic models of infant heads via structural MRI of infant participants. The left figures represent steps in the cortical source analysis technique, the mid-left figure potential cortical source locations for the Nc, the mid-right panels the topographical potential maps for the projection of the sources on the scalp, and the right panels the temporal activity of different brain areas for the brief stimulus presentation types.

eh wh Figure 3.7. Input (upper left) and outputs ofa computational model of visual development (Adapted from Schlesinger, Amso, & Johnson, 2007b). Outputs are represented as salience maps, portrayed here as 3D topological surfaces showing regions of activation. After initial exposure (iteration 0, upper right), edge regions become highly salient. Iterations 1 and 10 are shown at the lower left and lower right, respectively. Between iterations 0 and 10 the map becomes sharpened and defined as the rod is highlighted and the box recedes in salience.

Continuous

Event

Discontinuous Event

4h db

Figure 5.1. To illustrate the continuous and discontinuous events used by Spelke, Kestenbaum, Simons, and Wein (1995).

1: An object is placed on the stage

2: A second goes behind the screen 3: The screen is removed to reveal...

, ..either 2 objects...

..or 1 object

Figure 5.2. Schematic depiction of a typical addition condition. Note that the firstseen presentation (top) is also the “impossible” outcome at test (bottom).

Figure 5.3. An infant being tested in an addition or subtraction condition.

Figure 6.2.

Example stimuli from Fiser and Aslin (2002b) showing (a) sample base-

pairs of shapes and the third “noise” shape that created less coherent statistics with the other shapes when combined in (b) 3-shape scenes. The white rectangle shows a pair

of less coherent shapes (one base-pair element + a noise element) that was contrasted with a coherent base-pair during the post-habituation test phase. Source: From “Statistical Learning of New Visual Feature Combinations by Infants,” by J. Fiser and R. N. Aslin, 2002, Proceedings of the National Academy of Sciences, 99,

p. 15823. Copyright 2002 by National Academy of Sciences, U.S.A.

(b)

Figure 6.3. Sample images illustrating (a) how the Gestalt principle of good continuation enables the immediate perception of connected contours (the central circular form) in a cluttered array of contours. (b) how a complex, multi-part object presents a

challenge for learning which features define a part. Source: From “Detection of Contour Continuity

and

Closure in Three-Month-

Olds,” by P. Gerhardstein, I. Kovacs, J. Ditre and A, Feher, 2004, Vision Research, 44,

p. 2982. Copyright 2004 by Elsevier. Reprinted with permission. Images reprinted from Michael J. Tarr, Brown University, http://www.tarrlab.org/.

red standard (isoluminant to

luminance

the background)

yellow comparison (isochromatic to the back-

= ground), randomly chosen from the set on the right

Figure 6.4. A two-dimensional stimulus space (color x luminance) illustrating how a standard stimulus (circled in the left plot) can be rendered equisalient along both dimensions when a comparison stimulus (in the right plot) is varied along one dimension (luminance) and paired with the standard stimulus.

Source: From “A new method for calibrating perceptual salience across dimensions in infants: The case of color vs. luminance,” by Z. Kaldy, E. A. Blaser, and A. M. Leslie, 2006, Developmental Science, 9, p. 483. Copyright 2006 by Wiley-Blackwell. Reprinted with permission.

Integration Resolution

i

Figure 6.5. A sample image depicted at two spatial scales illustrating (a) how a small retinal image enables the entire object and all its component features to be accessed in a single fixation (represented by the black rectangle), and (b) howa large retinal image requires the component features of the object to be accesses by a series of fixations separated by eye movements. The smaller image requires excellent spatial resolution to perceive all the features, whereas the larger image requires integration of the more easily resolved features across eye movements. Source: Images reprinted from Michael J. Tarr, Brown University, http://www. tarrlab.org/.

Habituation Events Tight-fit containment habituation events:

ae oR Tight-fit support habituation events:

Test Events Familiar objects in a tight-fit relation Familiar objects in a novel relation

ati Novel objects in a tight-fit relation

Novel objects in a novel relation

Figure 9.1. The habituation and test events used in the tight-fit condition of Casasola and Cohen (2002). Source: From M. Casasola,

J. Bhagwat,

and

A.

S. Burke

Psychology, 45, p. 714. Copyright 2002 by Wiley-Blackwell.

2009,

Developmental

Sample Habituation events:

(a)

i

(b)

zil

Sample Test Events Familiar objects in a support relation Familiar objects in a novel relation

a ee Novel objects in a support relation Novel objects in a novel relation

Figure 9.2, Examples of the habituation and test events in the Casasola (2005a) study.

Source: From “When Less is More: How Infants Learn to Form an Abstract Categorical Representation of Support,” by M. Casasola, 2005, Child Development, 76, p. 283. Copyright 2005 by Wiley. Adapted with permission.

No Label Condition 1.05 °

©

Novelty

®

Familiarity

0.8 4 av

e

55 064 o QO

fereeeeeeee' a

e

.? i"

= o44

.

>

% °

eneneeeey coc

oo] gccccce

@

3

z

0.2 5

0.0

3.0

T

T

T

3.5

4.0

4.5

T

1

5.0

5.5

Euclidean (RMS) Distance

Figure 10.4. Novelty preference for modal stimuli as a function of Euclidean distance in Broad condition (Experiment 1) of Plunkett et al. (2008).

Label Condition

1.05

® Novelty ® Familiarity

0.8 4 wv Y

g5

0.6-

3

=

ow

> So

e L

e

3.0

3.5

ee -

a? ie

ve

oa ee

047

Zz

0.2 4 0.0

T

T

T

T

1

4.0

4.5

5.0

5.5

Euclidean (RMS) Distance

Figure 10.5. Novelty preference for modal stimuli as a function of Euclidean distance in single label, Broad condition of Hu (2008).

Typical Example

Atypical Example

Figure 11.1. Example typical and atypical female faces presented in the gender categorization tasks.

scicmebeemeenl ==

Typical

k

1

Somewhat Typical

Atypical

Figure 11.5. Examples of typical, somewhat typical, and atypical objects.

Destinctive Example

Typical Example

Figure 11.7. Example distinctive and typical faces presented in the facial distinctive-

ness memory tasks.

ye

Figure 12.1. Depiction of the apparatus used in the match-to-sample preferential looking task with infant and parent facing the left and right video screens and a central window where the tester displayed model objects. Source: From “Infants’ Comprehension of Toy Replicas as Symbols for Real Objects,” by B. A. Younger and K. E. Johnson, 2004, Cognitive Psychology, 48, 215, Copyright 2004 by Elsevier. Reprinted with permission.

Figure 12.4. Target stimulus materials used in a naturalistic parent-infant play session involving model animals, furniture, and vehicles: The models were divided into 2 sets of 4 models per category.

8 Information-Processing Approaches to Infants’ Developing Representation of Dynamic Features Kelly L. Madole, Lisa M. Oakes, and David H. Rakison

A primary task for the developing infant is to categorize and represent the large number of complex and wide-ranging objects in the world. Without this ability, memory storage would be overstretched, developing a faculty of language would be impossible, and inductive generalization would often err. The three authors of this chapter—individually and in collaboration—have examined the processes of such categorization and representation. This work generally derives from an information-processing background, and illustrates how research programs built from this foundation can provide developmental scientists with a deep understanding of at least one area of infant cognition. Moreover, we have incorporated neuroscience and computational

approaches to understanding development into our thinking and research, leading to a new understanding of infant perception, representation, and categorization.

Our work can broadly be characterized by two themes that emerge from our information-processing background. The first theme is the importance of understanding infants themselves as information processors. We have focused on understanding the units of information that infants perceive, attend to, encode, represent, and so on. The second theme is the critical role of categorization in cognitive development. In other words, we have focused on this one strategy that the human information-processing system uses to deal with the enormous amount of information it faces at each moment. These themes are apparent in our individual and collaborative work on understanding how infants attend to, process, and use dynamic features such as moving parts and function in categorization. Therefore, although our individual research addresses a relatively diverse set of developmental phenomena, we will illustrate here how the same constructivist approach applies to understanding when and how infants use dynamic features to categorize across those phenomena.

Order of authorship was determined alphabetically.

154

Infant Perception and Cognition

INFANTS AS INFORMATION PROCESSORS Beginning with the seminal work of Piaget (1954), theorists have attempted to characterize the changes that occur in cognitive processes during infancy. The information-processing approach has been one particularly influential perspective. Although the general focus of this perspective is on how information is processed (e.g., perceived, encoded, recalled) by the infant cognitive system, one of the main tenets is that development is a constructive process. Specifically, according to this view, infants’ developing cognition can best be understood by considering them as gradually processing ever more complex information received from the environment (for examples of this kind of developmental explanation see Booth, Pinto, & Bertenthal, 2002; Johnson, 2004; Schwarzer, Zauner, & Jovanovic, 2007; Stiles & Stern, 2001; Westermann & Mareschal,

2004). One particularly influential theory is Cohen's information-processing approach to infant cognition (Cohen, 1988, 1991). According to Cohen, from the first months of life, infants attend to and represent the basic features of objects and events in their world. As they develop, infants become able combine these basic features into more complex wholes. Therefore, to understand how infants perceive the objects and events around them, Cohen (1988) argued

that we must know something about the basic unit of information that infants are capable of processing at any point in time. Cohen and his colleagues demonstrated this process in a wide range of domains. Across studies on the perception of relatively simple forms, categorization of complex objects, and the perception of faces, the accumulated results of Cohen’s work point to a clear trend: Younger infants processed smaller “chunks” and older infants combined those “chunks” to process bigger units (e.g. Cohen & Cashon, 2001; Cohen, Gelber, & Lazar, 1971; Cohen & Younger, 1983). This constructivist view has certainly been challenged (e.g., Gibson, 2000; Quinn, 2004), yet we see repeatedly over the course of development that infants move from processing features as separate units and only later begin to combine these features into objects, and then into categories of objects (e.g., Schwarzer et al., 2007; Younger & Cohen, 1983; Younger & Cohen, 1986). Moreover, this developmental progression has been modeled successfully with Parallel Distributed Processing (PDP) networks, which encapsulate many of the assumptions of the information-processing approach (Shultz & Cohen, 2004; Westermann

& Mareschal, 2004).

Cohen's earliest studies focused on infants’ attention to features that are visually apparent, such as line orientation or the shape of an object part. Yet, these kinds of static features comprise only a small portion of the infants’ visual world. In each of our own research programs—and in work conducted in collaboration with Cohen or with each other—we have extended this work to examine infants’ processing of dynamic features, such as characteristic motion trajectories, actions performed on objects, or the presence of moving parts. In these studies, we began with the basic premise that infants are information-processors, and we examined the units of information that they attend to, perceive, and represent. We have proposed the existence of developmental

Information-Processing Approaches

155

continuity in the categorization process combined with developmental change in the units that are the focus of that process (see, e.g., Oakes & Madole, 2003; Rakison & Lupyan, 2008). In the following sections we describe how our studies of infants’ representation of function and animacy illustrate this perspective.

Infants’ Developing Attention to Object Function Initially, we assumed that object function serves as input to the informationprocessing system in the same way as a more visually apparent feature like shape. That is, like shape or color, we considered function to be a feature of objects that infants can attend to and encode (Horst, Oakes, & Madole, 2005; Madole & Cohen, 1995; Madole, Oakes, & Cohen, 1993; Rakison & Cohen, 1999). Because we adopted the general information-processing approach, we asked not only when do infants attend to function, but also when (somewhat later in development) is function combined with other features to be processed as a larger “chunk.” For example, Madole, Oakes, and Cohen (1993) used a design similar to that used by Younger and Cohen (1986), and an object-examining task (Oakes, Madole, & Cohen, 1991), to test how infants

processed object appearance and function. We found the expected developmental pattern: Younger infants were more attentive to the static appearance features (i,e., the color and overall shape of the object), and older infants were attentive to both those features and the more dynamic functional features (i.e., whether the object made a sound when it was shaken). Moreover, it was only later—at 18 months of age—that infants were sensitive to the combination of the function and appearance. This early study set the stage for examining dynamic features such as function from the perspective of infants as information processors.

More recently, however, we have considered that object function may not be a single, unified feature that serves as input into the information-processing system as we construed it in our original study. Rather, function is emergent from the combination of a number of different features, such as the actions performed on the object and the resulting effect of those actions (Oakes & Madole, 2008). A significant challenge that we have faced is establishing a definition of function that is both general and tractable—indeed, this is a challenge for the field as a whole. We have found the informationprocessing approach to be especially fruitful in providing an empirical basis for considering how to define and study function. Specifically, although we initially treated function as a unit, it is likely that what we call “function” actually reflects the combination of multiple elements. Thus, the development of infants’ processing of function might follow the kind of developmental trajectory observed for infants’ processing of faces (Schwarzer et al., 2007), angles (Cohen & Younger, 1984), and schematic animals (Younger & Cohen, 1983, 1986). That is, we expected that infants would initially be sensitive to the components of function, and only later in development recognize how those components are combined.

156

Infant Perception and Cognition

This hypothesis is consistent with our long-held and more general belief that, as with more static, visually available features, dynamic features are constructed from elements. For example, Oakes and Cohen (1995) argued that

infants’ recognition of the difference between causal and noncausal events was based on their perception of the objects, and the spatial and temporal features of the events. Thus, young infants attend to only these individual features, and older infants divide events into causal and noncausal, based on the combination of these features (see Oakes, 2003, for a discussion), We propose

that infants’ representation of object function follows a similar developmental course. The information-processing approach provides us with a framework for considering that even if infants can treat function as a single unified, coherent feature, this unit likely arose from the representation of components that comprise function. Taking this approach, we have begun to identify the components of object function, as well as to chart the developmental time course of infants’ attention to those components and how they are combined (Oakes & Madole, 2008). In information-processing terms, this could be described as a focus on how infants construct their understanding of function over time. As a first step, we broke down the notion of function into the component parts on which it is based, and examined sensitivity to those components. As we have suggested elsewhere (Oakes & Madole, 2008), function is complexly determined by components such as the actions performed on objects (e.g., grasping, rolling), the ways the physical structure of the objects constrain those actions (e.g., the presence of wheels or a handle), the consequence of acting on objects (e.g., making a noise when rolled), the intentions of the actor (e.g., intend-

ing to make a noise versus accidentally bumping into the object causing it to move), and so on. Our work represents a sort of reverse-engineering, by working from a monolithic and rather vague idea of function and decomposing it into more basic parts; in particular, actions and outcomes.

We have examined infants’ sensitivity to many such features, as well as how infants combine these features. Perone and Oakes (2006), for example,

habituated 10-month-old infants to a hand acting on one or more objects (e.g., squeezing, rolling), with that action resulting in a sound (e.g., clicking, mooing). Following this habituation experience, these 10-month-old infants dishabituated when the hand performed a different action on one of the familiar objects, or when the action resulted in a new sound. Thus, by 10 months of age, infants attend to two critical components of function—the actions and the outcomes. However, a conceptualization of function as nothing more than these features in isolation is insufficient. Instead, function is marked by meaningful associations between such features; for older children and adults, the intended outcome is key in how actions on objects are represented (e.g., Bloom, 1996). For example, when imitating adult models, toddlers are sensitive to the intended outcome, they repeat intentional actions more than accidental ones (Carpenter, Akhtar, & Tomasello, 1998; Meltzoff, 1995), and effectively produce the intended outcome, even if it means performing a different action

Information-Processing Approaches

157

than what was modeled (Gergely, Bekkering, & Kiraldy, 2002). This pattern of results suggests that children understand that the functions of objects are revealed through intentional action—that is, action to produce a particular effect. Even very young children do not represent the actions and outcomes independently; rather, the two components are linked through intentions and goals. Perone and Oakes (2006) demonstrated that infants’ sensitivity to the association between the components of function may depend on the particular features being represented. Infants were habituated to two events that embodied a particular correlation or association between the appearance of the object and the action performed (e.g., in one event, a purple round object was rolled and clicking was heard, and in the other event, a yellow square object was inverted and clicking was heard). Infants dishabituated when the action and appearance relation was “switched” (e.g., they responded when the yellow square object was now rolled). Because both features in this event were equally familiar, infants could only have dishabituated to this event because they had learned the association between the action and the appearance—they expected yellow objects to be inverted. Thus, at 10 months, infants are sensitive to how some of the components of function are combined—in these events, how the physical structure is related to what actions are performed on the object. Interestingly, Perone and Oakes (2006) found that 10-month-old infants did not seem to learn the relation between specific actions and specific outcomes (e.g., squeezing produces squeaking) or between specific object appearances and specific outcomes (e.g., purple objects squeak). Thus, they seem to selectively attend to how the physical features of objects constrain or afford actions performed on them; they were relatively insensitive to the association between specific outcomes and either appearance or action. Apparently,

infants at this age do not perceive all the components of function as a unified whole—they have begun to unify or associate some features, but not all the combinations are equally salient. This work not only revealed some of the ways that infants attend to the features that comprise function, it also led to a puzzle. Originally, we reasoned that the relation between action and outcome is critical for understanding the functions of objects, Actions and outcomes, at least from our adult intuition, are not arbitrarily related; actions typically are related to outcomes through causal mechanisms (e.g., squeezing an object causes it to squeak, and pulling a part causes an object to whistle). The fact that infants were insensitive to this relation was particularly intriguing. This finding suggested that to understand infants’ attention to action-outcome relations we needed to explore their attention to the causal mechanisms underlying that relation. It is not immediately obvious how such understanding would develop in infancy. One possibility is that infants first attend to the relation between action and outcome, and later come to understand the causal mechanism underlying that relation—in a sense, learning the connection between causes and effects might bootstrap their developing understanding of the causal mechanism between those causes and effects. An alternative possibility is that

158

Infant Perception and Cognition

infants’ emerging understanding of causal mechanism in general, guides their attention to action-outcome relations. In other words, infants first learn that actions in general cause outcomes, and they then use that knowledge to learn the specific effects that result from specific actions. A third possibility is that even before infants have developed this kind of causal understanding, “caused effects” may be highly salient, which might facilitate infants’ attention to action-appearance relations. Although Perone and Oakes (2006) found that infants fail to learn that particular outcomes are associated with particular actions, it may be still be important that the actions actually produced outcomes. We know that infants’ developing perception of causality may contribute to the salience of other object features. By 10 months, infants use spatial and temporal features of events to differentiate causal from noncausal relations (Oakes, 2003). Because they perceive causal events as different from noncausal events, components of events (for example, actions) that

appear to be the causes of interesting outcomes grabbing for infants. Extending this finding to Oakes (2006), the effect of the action may have the association between the action performed ance of that object. A

subsequent

set of studies

(Perone,

may be particularly attentionthe events used by Perone and facilitated infants’ learning of on the object, and the appear-

Madole,

&

Oakes,

submitted)

revealed that 10-month-old infants learn that some actions cause an outcome and other actions do not. Using the same design used by Perone and Oakes (2006), Perone et al. found that when infants were habituated to one action

producing an outcome (e.g., squeezing resulted in squeaking) and another action failing to produce an outcome (e.g., inverting resulted in no sound),

they dishabituated to the “switch” (e.g., squeezing producing no sound, or inverting producing squeaking). Thus, although infants at this age do not associate particular actions with particular outcomes, they can learn the difference between causal and noncausal actions. Other experiments showed that infants did not learn the difference between functional and nonfunctional objects in these stimuli (i.e., one object appearance was associated with an outcome during habituation, another object appearance was associated with the lack of an outcome), and that infants can learn about appearances and actions even when no outcomes occur, In conjunction, these results suggest that 10-month-olds infants’ emerging conception of function includes causal information. Together, these studies illustrate how adopting an information-processing view of development has influenced our understanding of infants’ conceptions of object function. Consistent with an information-processing approach, we have observed that the developmental challenge is to process units of information of increasing size and complexity. As infants start to combine these units of information (actions, objects, outcomes), particular features (caused outcomes) or correlations among features (object-action relations) take prior-

ity, at least temporarily. The idea that some features, or correlations among features, may become more important than others is particularly relevant

Information-Processing Approaches

159

when describing the development of infant categorization, and we will return to this idea later in the chapter. Infants’ Developing Understanding of Agency As with function, infants’ perception of agency can be considered from an information-processing perspective. Agency is closely related to function.

Where function refers to some feature of an artifact that allows an actor to produce some outcome by acting on or with that object, agency refers to whether or not objects have the function of causing a change in state. The alternative is that objects are recipients of actions; that is, other objects act upon them. An important feature of agency is the ability of objects—typically animates—to act independently. This aspect of agency is often considered a unitary feature that can be perceived. Objects are either agents or not, and the ability to distinguish between animate and inanimate agents is believed to be a core principle (Gelman & Opfer, 2002). However, like function, agency actually emerges from other components of events. This idea is implicit in many studies of infants’ attention to agency—for example, researchers often make objects appear to be independent agents by adding eyes, independent motion, or contingent interaction (Johnson, Booth, & O’Hearn, 2001). What

these studies have not done, however, is chart the developmental progression from infants’ attention to these isolated components, to the combining of such components, to the perception of a more global feature of agency. Using the information-processing approach, we can track how infants construct their perception of objects as agents—either animates, or causal agents in events—from more primitive elements or features, in the same way that we have explored their construction of object function. For example, the role that an object plays in a causal event is a kind of dynamic cue that is related to agency. Therefore, one approach we have taken is to ask how infants start to learn which things in the world tend to be agents, and which things tend to be recipients. According to one perspective, infants possess specialized mechanisms or innate modules to acquire this knowledge (Leslie, 1995; Mandler,

1992)—from this perspective, agency is a unified feature that is processed as an unanalyzed unit. However, we have argued that agency is derived using more general associative mechanisms (Rakison, 2005a; Rakison & Lupyan, 2008).

Agents

in the world

tend to be animates

with

functional

parts that

move when they are acting as agents. Recipients of an action, in contrast, tend to be inanimate objects with parts that do not move when they are acted upon. Thus, it is possible to examine how infants learn such isolated features, and eventually recognize the correlations among them, to recognize the division of objects in events as agents and recipients. Oakes and Cohen

(1990) (see also Cohen

& Amsel,

1998) showed

that

infants 6 months of age and younger attend to the isolated spatiotemporal features of launching events, and that infants 7 to 10 months are sensitive to how such features, when combined, differentiate causal from noncausal

160

Infant Perception and Cognition

events. Thus, by 10 months, infants can extract the property of an event as causal, an important step in identifying some objects in the event as agents and others as recipients. But, in the real world, agents and recipients are not simply defined by the spatiotemporal features of events—rather, agents tend to be animate (i.e., move on their own, have moving parts). Thus, agency is specified both by the spatiotemporal features of the event and the individual features of the objects themselves. Rakison (2005a) tested infants’ sensitivity to these cues at 12, 14, and 16 months of age—beyond the age when infants use spatiotemporal events to differentiate causal launching events from noncausal launching events. In this study, infants saw causal launching events involving the movement of identical geometric shapes. Each shape had a small part on the top surface that either moved or did not move. Some infants were habituated to relations consistent with those in the real world—i.e., the agent (as

defined from the spatiotemporal features of the event) had a moving part, and the recipient had a nonmoving part. Following habituation to such events, 14- and 16-month-olds, but not 12-month-olds, increased their looking when the relation was switched (i.e., the agent had the nonmoving part and the recipient had the moving part), suggesting they learned the relation between the causal role (as specified by the spatiotemporal cues), and whether or not the part moved. Other infants were habituated with relations that were inconsistent with those in the real world—i.e., the recipient had the moving part, and the agent had the nonmoving part. In this case, only 14-month-old infants learned the relation, and both 12- and 16-month-old infants failed to learn the rela-

tion in the habituation event. Thus, at 12 months infants did not learn the associations between the moving part and either role (agent or recipient). At 14 months, infants learned equally well that the agent or the recipient could have the moving part. At this age, infants were unconstrained in the relations they encoded, presumably because ofa lack of experience with exemplars in the real world. In contrast, infants at 16 months encode only those relations that make sense in the real world. The point is that agency was not a single, unitary feature in these events. Rather, it was complexly determined by the interaction of the spatiotemporal features of the event, and the features of the objects. Younger infants responded to the individual features, or how some subset of the features combined. It was only with additional experience with agents in the real world that infants’ “chunked” multiple cues to differentiate agents from recipients. These results, therefore, illustrate how the information-processing perspective shapes the questions we ask about infants’ developing attention to features in events. These results also point to an important contributor to developmental changes in how infants represent the features in these events—their increasing experience with, and knowledge of, objects and events in the work. In the experiments described here, such experience and knowledge constrained the kinds of relations that infants were willing to learn. Thus, their perception of agency combines not only features of the events, but their existing knowledge and experience. This developmental pattern—sensitivity to features and

Information-Processing Approaches

161

feature relations becoming constrained with increasing knowledge—is general. Madole and Cohen (1995) found a similar developmental trajectory for infants’ association of the form of specific parts and the functions of objects. Rakison (2006) found that infants’ learning of the features that specify selfpropulsion is initially unconstrained, and then becomes constrained by learning. Similar patterns have been observed in infants’ language learning (Stager & Werker, 1997) and gesture (Namy, Campbell, & Tomasello, 2004). In our view, that the developmental trajectory is found across a range of domains and ages suggests that the same information-processing mechanisms underpin learning for each of them (see Rakison, 2006). Conclusions

Clearly, these two lines of research demonstrate that the information-processing approach to infant cognition can be extended to the study of dynamic features. Importantly, these findings do not simply illustrate the utility of an empirical approach; rather, these results have implications for how we think about infants’ cognitive processing in general. Specifically, as we have argued elsewhere (Madole & Oakes, 1999; Rakison & Lupyan, 2008), there appears to be continuity in the process of cognition—what changes with development is the information used in that process. INFANTS AS CATEGORIZERS A second major theme derives from the information-processing approach: the idea of infants as categorizers. As information processors, infants must cope with a potentially overwhelming amount of information, and categorization serves to manage this wealth of information by representationally grouping objects together. We now have three decades of research demonstrating infants’ ability to categorize objects on the basis of surface features (Oakes, Coppage, & Dingel, 1997; Quinn, Eimas, & Rosenkrantz, 1993; Rakison & Butterworth, 1998; Strauss, 1979; Younger, 1985). Recently, categorization of more dynamic, nonobvious, and “conceptual” features has been examined, and this research has demonstrated that infants categorize events based on features such as object function (Horst et al., 2005; Rakison & Cohen, 1999) and causal roles (Booth, 2008). In addition, there is evidence that categoriza-

tion is not static and unchanging. Rather, how infants categorize—and their attention to specific features—varies with the context (Oakes, Horst, KovackLesh, & Perone, 2008; Oakes & Madole, 2008; Rakison, 2007).

The work on infants’ categorization has not been without controversy. In fact, some of the earliest research on infant categorization foreshadows trends and controversies that emerged later in the field. For example, in one of the earliest demonstrations of infants’ ability to group discriminably different entities, Cohen and Caputo (1978) habituated infants to different stuffed animals, and then demonstrated generalization to a different stuffed animal but not a rattle. They suggested that they were “instructing infants to respond to perceptual categories.” Cohen and Strauss (1979) demonstrated

162

Infant Perception and Cognition

that infants can learn to respond categorically to a specific female face presented in different orientations, as well as to different female faces. They suggested that infants were responding to a “conceptual category.” The issue of whether infants are engaging in “perceptual” processes or “conceptual” processes continues today, and has promoted vigorous debate concerning the nature of infant categorization (Mandler, 1992; Oakes et al., 1997; Rakison & Butterworth, 1998). Do infants use perceptual information as the basis for categorization (i.e., information that is easily available through basic perceptual modalities), or do they rely on knowledge of category relations, animacy, or other “conceptual” information that joins category members? One goal of our work is to eliminate this mutually exclusive dichotomy, and instead view early categorization as a developmental process. Working from that viewpoint, we have been attempting to describe the ways that infants apparently move from attending to primarily perceptual information, to a more conceptual understanding of categories. This goal derives directly from the same information-processing approach described above, The question is not simply how do the units of information to which infants are sensitive change over time, but also how infants use those units of information in contexts that encourage or require categorization. Because the goal of the information-processing system is to make sense of and organize the enormous amount of information that adults, children, and infants encounter, the units identified in the first sections of this chapter may be used to form categories. Understanding the role of such features in infants’ categorization has been another goal of our work. This approach moves beyond questions about whether infants’ categories are primarily perceptual or conceptual, and whether understanding perceptual categorization informs our understanding of conceptual categorization. Instead, we see categorization as a process that operates at many different levels, and our goal has been to examine how infants select from the many different types of features that could be used for categorization: static features, dynamic features, readily available features, and less apparent features. As infants select among these features, some features (or relations among fea-

tures) must emerge as more critical than others, depending on the particular items to be categorized. For example, identifying agents of outcomes requires attention to spatiotemporal contiguity but not (necessarily) object color; categorizing animals versus vehicles requires attending to gross differences in shape, texture, and type of movement, whereas categorizing dogs versus cats requires attending to more subtle differences in body shape and movement trajectories. In each of these examples, however, the biases to attend to some features over others in categorizing do not derive from a deep conceptual knowledge of the categories. Rather, these biases emerge from detection of statistical regularities in the world, and infants’ increasing knowledge of and familiarity with objects, events, and relations (see Rakison & Lupyan, 2008). Thus, a major focus of our current research is to lay a foundation for studying infants’ categorization using “high-level” features (e.g., functions that have effects, features that differentiate agents from recipients, or correlations that

Information-Processing Approaches

163

conform to the way the world is structured) that may be most important for some kinds of categorization. Importantly, in our view categorization is a process used by the information-processing system to organize information and provide the basis for meaning and inference (Oakes & Madole, 2000; Rakison, 2005a; Rakison & Butterworth, 1998). Thus, our goal is not to identify the contents of infants’ categories (e.g., do infants “know” the functions of objects, do they “have” a concept of animacy). Rather, our goal is to understand how infants use specific kinds of information when they are categorizing. Thus, our focus is on how categories are constructed online—and over varying timescales—as infants encounter items over trials, moments, days, and weeks (Oakes et al., 2008; Rakison & Lupyan, 2008). In the following paragraphs, we provide two illustrations of how we can use this general perspective to understand infants’ categorization using dynamic features. The Role of Object Function in Infants’ Categorization Our initial work on object function was motivated by a desire to understand how infants’ use function in categorization. Specifically, we asked whether infants use object function to form groups of objects. To answer this question, Horst, Oakes, and Madole (2005) habituated 10-month-old infants with

a series of events that conformed to a “category” from an adult perspective. Half of the infants saw events in which the object had the same appearance on each trial (e.g., always a purple round object), but four different “functions” were presented (e.g., squeezing that produced squeaking on some trials, rolling that produced clicking on some trials, and so on). The reasoning was that in this condition, infants might categorize the events in terms of the object appearance. They might learn to ignore variation in function, and learn that what was relevant was the commonality across trials in object appearance— just as infants who are familiarized with a series of dogs learn to ignore variations that differentiate dogs (e.g., fur color, markings), and selectively attend to commonalities among those dogs (e.g., relative size and spacing of facial features) (e.g., Oakes & Ribar, 2005).

The other infants were habituated to events in which the same function was presented on each trial, but it was performed on four different objects (e.g., purple round on some trials, pink oblong on others, and so on). In this

condition, infants might learn to ignore the variation in appearance and selectively attend to the commonality in function across events. Thus, this condition addresses whether infants categorize these events based on a common function. Infants’ responding during test revealed that when familiarized with four different items—whether those items shared a common function or a common appearance—function was the most salient feature. Infants who were familiarized with four different appearances and one function dishabituated when shown a new event involving a familiar appearance and a new function. Infants who were familiarized with four different functions and a

164

Infant Perception and Cognition

single appearance dishabituated when shown a new event involving a familiar appearance and a new function. What does this pattern mean? Clearly, infants did not respond simply to the category presented during familiarization; instead, the dynamic feature— namely, function—was the most salient cue in both conditions. This finding implies that in the appearance-constant condition, infants did not selectively attend to the common appearance and ignore the variation in function. Rather, they remembered each of the four functions, and additional experiments showed that when given more familiarization, infants responded to both the common appearance and the varying functions. The categorization task here was difficult: infants were required to attend to, perceive, learn about, and remember some aspects of four different dynamic events. In this situation, in which the information-processing system was overloaded, infants failed to organize the information using the less salient but common feature, information about the surface features of the object. Rather, the system defaulted to attending to the more salient, dynamic information about object function. Indeed, young infants find dynamic features more compelling than static features (Shaddy & Colombo, 2004). In some cases, infants actually selectively attend to dynamic features—encoding, for example, actions, and apparently not encoding the objects on which those actions are performed (Perone, Madole, Ross-Sheehy, Carey, & Oakes, 2008), even the face of a woman brushing her teeth or combing her hair (Bahrick, Gogate, & Ruiz, 2002). Thus, our results suggest that when faced with a category of items that have a dynamic component, infants had difficulty overcoming their bias to focus on dynamic features, and did not recognize the commonality in the static features. However, Horst et al.’s (2005) findings do suggest that infants can cate-

gorize these events on the basis of function. That is, presenting infants with four different events did not overwhelm their information-processing system and interfere with their ability to detect the commonality among events that shared a common function. The biases just described for infants’ attention, in

our events, to dynamic versus static features, and their selective attention to the function, may bootstrap their categorization in this context on the basis of the less obvious, more dynamic feature of object function. This logic depends on the notion that functional cues are salient when infants are faced with a categorization task. That is, we argue that this reliance on function is a reaction, in part, to being unable to deal with a large amount of information presented. When infants cannot attend to and remember all the information, they attended to and remembered either the individual functions or the commonality in function across exemplars. Indeed, when familiarized with these same events in a context that places fewer demands on their information-processing resources (in this case, by presenting only a single event during familiarization), 10-month-old infants encoded both the object’s appearance and what function was performed, and dishabituated to changes in each of these features (Horst et al., 2005). Therefore, in this

Information-Processing Approaches

165

context, infants’ attention to function did not inhibit their attention to object appearance. Together, these findings suggest that for 10-month-old infants, function has a special status when they are engaged in categorizing events. Perhaps the increased information-processing demands when processing multiple exemplars causes infants to “fall back” to less developmentally mature patterns of attention—in this case, favoring attention to function rather than balancing attention to the more salient feature of function, and

the less salient feature of object appearance. This pattern of reverting to a less sophisticated processing mode when faced with increased processing demands was observed by Younger and Cohen (1986). They found that 7-month-old infants were unable to use correlations among features in a category, despite the fact that 7-month-old infants were sensitive to feature correlations in individual objects. Moreover, Younger and Cohen found that 7-month-old infants’ performance degraded in other ways when they were familiarized with a category of items—they failed to habituate, and they did not dishabituate even when presented with a completely novel object. Thus, in general, when category tasks are demanding, infants have difficulty detecting commonalities among features or relations that they are sensitive to in other contexts.

It is also possible that our results on the salience of object function reflects the fact that by 10 months of age, infants may, at some level, recognize that function is more fundamental for categorizing objects than is object appearance. Theorists have long assumed that infants’ earliest categories are based on functional information (Mervis & Bertrand, 1993; Nelson, 1979, 1991). It is quite possible that what we have documented is infants’ sensitivity to function as a central feature for forming groups of objects. In summary, our general approach—considering infants as actively categorizing the information they encounter—has led us to ask specific kinds of questions about the role of object function in infants’ categorization. Note that we do not ask whether any specific concept possessed by infants contains functional information (e.g., do infants “know” that phones are for talking into, that cups are for drinking). Although such knowledge is critically important for infants’ emerging understanding of the world, our focus has been on the role of function in infants’ categorization of new information. In this way we gain an understanding of how function and the components of function act as units of information, as infants actively construct their understanding of the world. From our perspective, because “function” may change over time, and because how infants use the units of information is complexly determined by contextual factors, whether or not infants at any age will use function in forming categories will vary with the demands of the task, the type of function, how it is demonstrated, and so on. Indeed, in studies with toddlers in sequential touching tasks (and thus function is primarily discovered through the child’s own manipulation of objects), we have observed that similarities in object shape are more salient than are similarities in function (Ellis & Oakes, 2006; Horst et al., in press).

166

Infant Perception and Cognition

Infants’ Categories for Animates versus Inanimates Function is not the only dynamic or “high-level” feature that infants might use in forming categories of objects. A natural extension of the agency work described earlier is to examine whether infants categorize dynamic stimuli in ways that correspond to the distinction between animates and inanimates. Rakison (2004, 2005a, 2006; Rakison & Poulin-Dubois, 2002) has examined such categorization by infants and toddlers. Again, the goal was not to identify when infants possess concepts that animals move in one way and cars move in another way (although such knowledge likely contributes to how infants respond in these tasks), but rather to identify how different kinds of information are used in categorization tasks, and how the use of that information changes with development, as well as with increased knowledge about the world. Our approach to how infants learn about animacy from an informationprocessing perspective builds on the foundation of the agency work described earlier. Specifically, a potential information-processing “unit” used in categorization is the association between relatively static features and relatively dynamic ones. Such a unit is particularly critical for recognizing differences between animates and inanimates, causal agents and recipients, self-propelled from caused-to-move objects, and so on. For example, wheels, legs, and wings are all associated with different kinds of movement trajectories, and they tend to move when the object to which they are attached engages in movement. Thus, differentiating animates from inanimates may not require highlevel conceptual understanding of the differences between those two classes of things, but rather it may be accomplished by attending to these kinds of feature correlations (see Rakison & Lupyan, 2008). Infants’ use of such rela-

tions in categorizing stimuli may follow the same developmental trajectory as their use of other kinds of information-processing units: infants first attend to the individual features, later begin to recognize the correlations among them, and eventually use those correlations for forming categories of the objects and events they encounter. Rakison has examined just how such units of information are used in infants’ category formation—and how their use of those units changes with development. The relations that specify animacy are quite complex. Specifically, static features might predict dynamic features such as motion trajectories (i.e., the presence of wheels, legs, and wings all predict different kinds of motion trajectory). But, other kinds of associations are also important—specifically, those between motion characteristics of the parts, and the motion trajectories (i.e., object motion trajectories are associated with how wheels, legs, and wings move). Based on the finding that 18-month-old infants in the sequential touching paradigm tend to roll objects with wheels and hop objects with legs, Rakison (2005a; Rakison & Poulin-Dubois, 2002) hypothesized that infants learn about objects’ motion characteristics by associating them with causally relevant and conjointly dynamic parts. How do infants’ learn these complex relations? If the developmental trajectory follows that described in previous sections, we should see that first

Information-Processing Approaches

167

infants detect and represent the individual features, later they combine some of the features, and even later they recognize more complex combinations of features. Of course, the further question is how they use such relations to form categories of objects. These developmental steps have been verified empirically. Early work revealed that when learning about events involving the movement of novel dynamic geometric figures, 10-month-old infants were unable to encode relations among dynamic features (although they could learn the individual static features such as body shape), 14-month-old infants associated object motion with dynamic object parts, and 18-month-old infants extended this correlation to whole objects (Rakison & Poulin-Dubois, 2002). Thus, as has been observed for sensitivity to information-processing units seen in other domains, with development, infants are sensitive to larger and larger units of information, and those units are significant for changing the “meaning” infants make of the events. The next question was how such correlations are used in categorization. That is, rather than asking (for example) if infants can learn that one bird moves in a particular way because of how its wings move, and that one horse moves in a different way because of how its legs move, we can ask whether infants can learn that birds in general move in a characteristic way because of how their wings move, and horses in general move in a different characteristic way because of how their legs move. Using the same dynamic, geometric shapes described in the earlier section on animacy, Rakison (2004) found that

it is not until between 18 and 22 months that infants are sensitive to such correlations in a category context, and that they attend exclusively to the relation between object parts and a motion characteristic in such a context. In other words, infants will extract correlations between moving, apparently functional parts and how an object moves, and use commonalities among objects in those correlations to group the objects or events. This line of work illustrates another aspect of the information-processing approach—how careful and thorough investigations can effectively challenge criticism of this approach to the study of conceptual development. Specifically, the Original Sim criticism of similarity-based approaches (Keil, 1981) states that it is impossible to know a priori which features are significant for category membership, because there are so many available in the environment. The result described here, and those on infants’ attention to correlations involving object function (Madole & Cohen, 1995), suggest that, rather than considering all possible features, infants constrain their attention to particular features in response to statistical regularities they experience. Summary of Our Research Program Our focus on infants as categorizers is an extension of our focus on infants as information processors—in fact, in many regards, our categorization work flows seamlessly from our work on infants’ processing of information about individual objects. In both of these areas we focus on the units of information that are used

168

Infant Perception and Cognition

in those processes. By considering how the kinds of information used by infants changes with increasing age and experience, we gain an understanding into the meaning they make of the world around them. Importantly, this approach provides an alternative account of the apparent “perceptual to conceptual shift” (Quinn & Eimas, 1997; Rakison, 2005b) That is, our work shows that infants can

use “high-level,” dynamic stimulus properties as the basis of categorical groups. However, their ability to categorize objects on the basis of function or motion trajectories need not reflect a qualitatively different “conceptual” understanding of those properties, Rather, our work has shown that such categories are constructed by applying the same kinds of processes that are applied when categorizing line drawings of animals. What changes with development are the units of information that the infant attends to and encodes. COMPUTATIONAL AND NEUROSCIENCE CONTRIBUTIONS TO THE INFORMATION-PROCESSING APPROACH In the preceding discussion we focused on continuity in the processes of information-processing. This focus is distinctly different from those who have argued for qualitative developmental shifts that result in a move from perceptual to conceptual processing (Gelman, 2003; Mandler, 1992). In our view, the apparent shift from perceptual to conceptual understanding does not reflect qualitative developmental changes in representation or processing. Instead, we propose that this change reflects how information is combined and constrained in new ways as a result of changes in how efficiently and effectively the system processes information, as well as changes in access to new information due to the accumulation of experience with objects and events in the world. Thus, we argue for general purpose processes that allow infants to organize, make sense, and construct their understanding of many different domains (see also Quinn & Eimas, 1997; Smith, Jones, & Landau, 1996).

Development, from our perspective, is a gradual process, and qualitative shifts or changes in thinking and representation likely stem from moving across some threshold in that gradual process. Our view has much in common

with recent computational and neuroscience approaches to cognitive

development—and our evolving understanding of the domains described here is advanced by work in those areas. In both computational and neuroscience approaches to such domains, quantitative changes lead to qualitative changes in how information is processed and represented. Advances in each of these areas have increased our sophistication in thinking about how information is processed by the mature system, as well as how that information-processing changes with development. In the following sections we describe how such advances are shaping our work. Computational Advances The seminal work of Rumelhart and McClelland (1986) inspired a number of developmental scientists to implement their theoretical perspectives in

Information-Processing Approaches

169

parallel distributed processing (PDP) and Bayesian models. The rationale for creating a computational account of a theory is that it compels the researcher to make explicit the assumptions of the theoretical account (Hintzman, 1991). Moreover, a computational model can make concrete predictions that can be tested empirically in behavioral work, which in turn can help to unify incongruent findings under a common framework (for a discussion see McClelland, 1988).

Those adopting the kind of information-processing perspective that we support here have implemented their approach predominantly in PDP models (for a review see Schlesinger & Parisi, 2004), The main reason for this bias is that PDP models are more aligned with the assumptions of the information-processing view than are Bayesian approaches; PDP architectures are essentially domain-general associative learning mechanisms that learn gradually over time. Moreover, Bayesian architectures require assumptions that are inconsis-

tent with the information-processing view; they explicitly test hypotheses and have mechanisms devoted to processing specific kinds of information (e.g., Tenenbaum, Griffiths, & Kemp, 2006), The implementation of PDP models has helped in our understanding of early category and concept development. For example, Mareschal and colleagues (French, Mermillod, Quinn, Chauvin, & Mareschal, 2002; Mareschal,

French, & Quinn, 2000) presented a PDP model of early concept learning that relies on the feature values of the stimuli. The model revealed that young infants’ categorization of basic-level categories such as dog and cat can be explained by bottom-up associative processes. Similarly, Cohen, Chaput, and Cashon (2002) showed that information-processing principles implemented in a connectionist architecture can explain how infants learn about causal events. Finally, Colunga and Smith (2005) used a PDP model to show that associative processes—without built-in assumptions about how labels should be interpreted—are sufficient for early word learning. We have also created a PDP model to implement our theoretical approach for how infants learn static and dynamic object features (Rakison & Lupyan, 2008). The simulations modeled a number of behavioral experiments on how

infants learn about objects’ motion paths (Rakison & Poulin-Dubois, 2002), their causal role (Rakison, 2005a), and their ability to self-propel (Rakison, 2006). The simulations show that general learning processes, in combination with a few well-established assumptions about development, attention, and the nature of the input, can account for how and when infants learn about the static surface features of objects, and how those objects move in the world. Moreover, they made predictions about early word learning—in particular, that words should be initially associated with functional parts—that were borne out by novel experimental research with infants. It is worth noting, however, that a model is nothing more than a sufficiency proof; it can show that a researcher's theoretical assumptions can lead to the pattern of development observed in infants, but not that these assumptions are actually correct. Nonetheless, the computational approach to early development can offer developmental scientists a more explicit way of formulating

170

Infant Perception and Cognition

and testing their theories. In this vein, they have provided strong evidence to support a number of information-processing theories, including our own. Neuroscience Advances

A number of findings and theoretical advances in the study of the neuroscience of perception and cognition shape our thinking and research. Although little work is conducted with human infants, our understanding of the development of primate neuroanatomy, as well as the role of specific brain regions in primate and adult human psychological function (as evidenced by techniques such as single-unit recordings and fMRI), have informed our continued work on infants’ attention to dynamic features and their ability to categorize on the basis of those dynamic features. For example, neuroanatomical evidence points toward a distinction between the processing of “what” information in the ventral visual stream, and the processing of “how” or “where” information in the dorsal visual stream (Goodale & Milner, 1992). Because action on objects requires integrating these two sources of information—e.g., you can only intentionally move your eyes to a specific object if you have encoded not only what that object is, and where there are objects in the environment, but where that specific object is located—the differences between these processing streams and how perceivers and actors integrate this information has been of intense interest to researchers (Buxbaum, Kyle, & Menon, 2005; Kaufman, Mareschal, & Johnson, 2003; Mareschal & Johnson, 2003; Valyear, Culham, Sharif, Westwood, & Goodale, 2006). Moreover, there has been debate in the literature about whether the

ability to process these two kinds of information develops simultaneously, or if one kind of information-processing emerges before the other (Atkinson, 1984; Mash, Quinn, Dobson, & Narter, 1998).

Of course, we can only indirectly assess these relations in infants. For example, Kaufman et al. (2003) proposed that infants engage in spatiotemporal processing required for action when viewing potentially graspable objects (e.g., small, moveable objects), and object-recognition processing when viewing non-graspable objects (e.g., large, static objects). Consistent with this proposal, Mareschal and Johnson (2003) reported that 4-month-olds remem-

bered the location of graspable objects but not the identity of non-graspable objects. Note that we cannot actually know whether viewing graspable objects induced more dorsal stream processing, and viewing non-graspable objects induced more ventral stream processing. We must infer this difference from infants’ looking patterns—if they appear to represent information relevant for action in some contexts more than in others, it is possible that those contexts have engaged dorsal processing more strongly. In our studies of infants’ representation of object function, we have shown infants events in which a hand reaches for, grasps, and acts on an object (Horst et al., 2005; Perone et al., 2008; Perone & Oakes,

2006). Thus, infants have

good information that the objects are graspable (they see a hand grasping the objects), and thus our events may invoke more dorsal stream processing. This

Information-Processing Approaches

171

may be why, in part, in our studies, infants often find action more salient than appearance. The dorsal-ventral distinction—as well as emerging communication between those two processing streams—is also important for our work examining infants’ attention to correlations between dynamic and static features, and between moving parts and motion trajectories (Rakison, 2005a). Because

different aspects of those events are presumably processed by different visual streams, the effects we have observed may reflect, in part, the development of these processing streams, as well as pathways that allow communication across the streams. Moreover, because these studies have used abstract geometrical shapes, it is not clear whether, or when, infants see the objects as graspable. If the analysis by Kaufman et al. (2003) is correct, the perception that the objects in the events are, or are not, graspable may change the role of dorsal and ventral stream processing of these events. The possibility that representing the features in our events engages different aspects of the visual system—and may reflect communication between those parts of the visual system—allows us to think somewhat differently about the components that comprise function, and how those components are combined. Moreover, these speculations lead to new hypotheses about the mechanisms of change in infants’ representation of such features. If infants engage in different processing when they see objects with features relevant for action, this processing should be influenced by infants’ developing understanding of action and what features are relevant for action. For example, infants who can reach out and grasp objects may engage in different processing when viewing an object with a handle than do infants who are not yet able to reach out and grasp objects. Thus, infants’ changing motor abilities may contribute to their developing representations of objects. In addition, work on how different ontological categories are represented in the human adult brain (for example, that viewing pictures of animals and pictures of inanimate objects, specifically tools, result in activity in different brain regions, (Devlin, et al., 2002; Martin, 2007) has implications for our

work. Do the developmental changes we have observed for infants’ attention to features related to animacy (Rakison, 2005a, 2006) reflect the increasing specialization of these brain regions? Findings by Pelphrey and colleagues (e.g., Pelphrey, Morris, & McCarthy, 2004)—that regions of the human adult superior temporal sulcus (STS) respond to the intentionality of action being viewed—have implications for how we interpret infants’ perception of our functional demonstrations, in which a human hand intentionally acts on an object producing an outcome. Do infants’ representations of aspects of human action depend on the development of structures of the STS? Or, do our results better reflect the model Tyler and her colleagues (e.g., Tyler & Moss, 2001) have proposed for distributed hierarchical object processing system, in which multiple brain regions are engaged in the processing of objects and the actions performed on them,? The point is not that our work directly addresses such controversies. Rather, the kinds of investigations we conduct, and advances in computational and

172

Infant Perception and Cognition

neuroscience approaches, inform our models of development. Such work gives us insight into potential mechanisms of change, as well as notions about the codevelopment of multiple functions. CONCLUSIONS Our work in general extends classic information-processing approaches to infant cognition, to domains of “high-level” dynamic features. We have applied this approach to infants’ perception, detection, and categorization of causality (Oakes, 1994; Oakes & Cohen, 1990; Rakison, 2005a), animacy (Rakison, 2006), and form-function relations (Madole & Cohen, 1993). In each of these

domains, we have taken a constructivist approach that infant informationprocessors initially deal with relatively small units of information (e.g., the visible details of an object, the shape and color ofa single part, whether or not a part moves), and gradually become able to combine those units into larger units. Moreover, one of the ways infants manage the overwhelming amount of information they encounter is to detect commonalities across instances and events, and form categories of those instances and events. We have shown that this developmental trajectory holds true for a variety of dynamic features that exist in the world. This extension of the information-processing perspective is critically important for understanding how infants move from a primarily perceptual understanding of the world (e.g., what objects look like) to a “deeper” conceptual understanding of the world (e.g., living things), Rather than positing conceptual understanding to the preverbal infant, or separate processes for creating perceptual and conceptual representations (e.g., Leslie, 1995; Mandler, 1992), we argue for general-purpose mechanisms that, over time,

gradually allow for apparent qualitative changes in how infants represent the objects and events in the world. The information-processing perspective we have presented here also fits well with findings from computational and neuroscience studies of cognitive development, though considerable work is needed to discover the whether this overlap reflects the genuine substrate to this developmental process or is coincidental. Nonetheless, our research shows that the view of infants as information processors has provided considerable insight, not only into what information infants represent, but also into the mechanisms that underlie learning in the first years of life and beyond.

REFERENCES Atkinson, J. (1984). Human visual development over the first 6 months of life: A review and a hypothesis. Human Neurobiology, 3, 61-74, Bahrick, L. E., Gogate, L. J., & Ruiz, 1. (2002). Attention and memory for faces and actions in infancy: The salience of actions over faces in dynamic events. Child Development, 73, 1629-1643. Bloom, P. (1996). Intention, history, and artifact concepts. Cognition, 60, 1-29.

Information-Processing Approaches

173

Booth, A. E. (2008), The cause of infant categorization? Cognition, 106, 984-993.

Booth, A. E., Pinto, J., & Bertenthal, B. I. (2002), Perception of the symmetrical patterning of human

gait by infants. Developmental Psychology, 38,

554-563.

Buxbaum, L. J., Kyle, K. M., & Menon, R. (2005), On beyond mirror neurons: Internal representations subserving imitation and recognition of skilled objectrelated actions in humans. Cognitive Brain Research, 25, 226-239, Carpenter, M., Akhtar, N., & Tomasello, M. (1998). Fourteen- through 18-month-old infants differentially imitate intentional and accidental actions, Infant Behavior & Development, 21, 315-330. Cohen, L. B. (1988). An information processing approach to infant cognitive development. In L. Weiskranz (Ed.), Thought without language (pp. 211-228). New York: Clarendon Press. Cohen, L. B. (1991). Infant attention; An information processing approach. In M. J. Weiss & P. R. Zelazo (Eds,), Newborn attention; Biological constraints and

the influence of experience (pp. 1-21). Norwood, NJ: Ablex.

Cohen, L. B., & Amsel, G. (1998). Precursors to infants’ perception of the causality of a simple event. Infant Behavior and Development, 21, 713-732. Cohen, L. B., & Caputo, N. F. (1978, May). Instructing infants to respond to perceptual categories. Paper presented at the Midwestern Psychological Association, Chicago. Cohen, L. B., & Cashon, C. H. (2001). Do 7-month-old infants process independent features or facial configurations? Infant and Child Development. Special Issue: Face Processing in Infancy and Early Childhood, 10, 83-92. Cohen, L. B., Chaput, H. H., & Cashon, C. H. (2002). A constructivist model of infant cognition. Cognitive Development. Special Issue: Constructivism today, 17, 1323-1343.

Cohen, L. B., Gelber, E. R., & Lazar, M. A. (1971). Infant habituation and generalization to differing degrees of stimulus novelty. Journal of Experimental Child Psychology, 11, 379-389, Cohen, L. B., & Strauss, M. S. (1979). Concept acquisition in the human infant. Child Development, 50, 419-424. Cohen, L. B., & Younger, B. A. (1983), Perceptual categorization in the infant. In E. K. Scholnick (Ed.), New trends in conceptual representation: Challenges to piaget’s theory? (pp. 197-220). Hillsdale, NJ: Erlbaum. Cohen, L. B., & Younger, B. A. (1984). Infant perception of angular relations. Infant Behavior & Development, 7, 37-47. Colunga, E., & Smith, L. B. (2005). From the lexicon to expectations about kinds; A

role for associative learning. Psychological Review, 112, 347-382. Devlin, J, T., Russell, R. P., Davis, M. H., Price, C. J., Moss, H. E., Fadili, M, J., et al.

(2002). Is there an anatomical basis for category-specificity? Semantic memory studies in PET and fMRI. Neuropsychologia, 40, 54-75. Ellis, A. E., & Oakes, L. M. (2006). Infants flexibly use different dimensions to

categorize objects. Developmental Psychology, 42, 1000-1011, French, R. M., Mermillod, M., Quinn, P. C., Chauvin, A., & Mareschal, D. (2002). The

importance of starting blurry: Simulating improved basic-level category learning in infants due to weak visual acuity. Proceedings of the 24th Annual Conference of the Cognitive Science Society, NJ: LEA. Gelman, S. A. (2003). The essential child: Origins of essentialism in everyday thought. New York: Oxford University Press.

174

Infant Perception and Cognition

Gelman, S. A., & Opfer, J. E. (2002). Development of the animate-inanimate

distinction. In. U. Goswami (Ed.), Blackwell handbook of childhood cognitive development (pp. 151-166). Oxford, UK: Blackwell Publishing. Gergely, G., Bekkering, H., & Kiraldy, I. (2002). Rational imitation in preverbal infant. Nature, 415, 755. Gibson, E. J. (2000). Perceptual learning in development: Some basic concepts. Ecological Psychology, 12, 295-302. Goodale, M. A., & Milner, D. A. (1992). Separate visual pathways for perception and action, Trends in Neuroscience, 15, 20-25. Hintzman, D, L. (1991). Why are formal models useful in psychology? In W. E. Hockley & S. Lewandowsky (Eds.), Relating theory and data: Essays on human memory in honor of Bennet B, Murdock. (pp. 39-56). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Horst, J. S., Ellis, A. E., Samuelson, L. K., Trejo, E., Worzalla, S. L., Peltan, J. R., etal. (2009). Toddlers can adaptively change how they categorize: Same objects, same session, two different categorical distinctions. Developmental Science, 12, 96-105.

Horst, J. S., Oakes, L. M., & Madole, K. L. (2005), What does it look like and what can it do? Category structure influences how infants categorize. Child Development, 76, 614-631.

Johnson, S. C., Booth, A, E., & O’Hearn, K. (2001). Inferring the goals ofa nonhuman agent. Cognitive Development, 16, 637-656. Johnson, S. P. (2004). Development of perceptual completion in infancy, Psychological Science, 15, 769-775. Kaufman, J., Mareschal, D., & Johnson, M. H. (2003). Graspability and object processing in infants. Infant Behavior & Development, 26, 516-528. Keil, F. C, (1981), Constraints on knowledge and cognitive development. Psychological Review, 88, 197-227. Leslie, A. M. (1995). A theory of agency. In D. Sperber, D. Premack, & A, J. Premack (Eds.), Causal cognition: A multidisciplinary debate (pp. 121-149). Oxford: Clarendon. Madole, K. L., & Cohen, L. B. (1993, March). Infants’ categorization of form versus function. Paper presented at the Society for Research in Child Development, New Orleans, LA. Madole, K. L., & Cohen, L. B. (1995), The role of object parts in infants’ attention to form-function correlations. Developmental Psychology, 31, 317-332. Madole, K. L., & Oakes, L. M. (1999). Making sense of infant categorization: Stable processes and changing representations, Developmental Review, 19, 263-296. Madole, K. L., Oakes, L. M., & Cohen, L. B. (1993). Developmental changes in infants’ attention to function and form-function correlations. Cognitive Development, 8, 189-209, Mandler, J. M. (1992). How to build a baby: II. Conceptual primitives. Psychological Review, 99, 587-604, Mareschal, D., French, R. M., & Quinn, P. C. (2000). A connectionist account of

asymmetric category learning in early infancy. Developmental Psychology, 36, 635-645.

Mareschal, D., & Johnson, M. H. (2003). The “what” and “where” of object representations in infancy. Cognition, 88, 259-276.

Martin, A. (2007). The representation of object concepts in the brain. Annual Review of Psychology, 58, 25-45.

Information-Processing Approaches

175

Mash, C., Quinn, P. C., Dobson, V., & Narter, D. B. (1998). Global influences on the

development of spatial and object perceptual categorization abilities: Evidence

from preterm infants. Developmental Science, 1, 85-102. McClelland, J. L. (1988). Connectionist models and psychological evidence. Journal of Memory and Language, 27, 107-123. Meltzoff, A. N. (1995). Understanding the intentions of others: Re-enactment

of intended acts by 18-month-old children. Developmental Psychology, 31, 838-850.

Mervis, C, B,, & Bertrand, J, (1993). Acquisition of early object labels: The roles of operating principles and input. In A. P. Kaiser & D. B. Gray (Eds.), Enhancing children’s communication: Research foundations for intervention (pp. 287-316). Baltimore, MD; Paul H. Brookes Publishing. Namy, L. L., Campbell, A. L., & Tomasello, M. (2004). The changing role of iconicity in non-verbal symbol learning: A u-shaped trajectory in the acquisition of arbitrary gestures, Journal of Cognition and Development, 5, 37-57. Nelson, K. (1979). Explorations in the development ofa functional semantic system. In W. A, Collins (Ed.), The Minnesota symposium on child psychology: Children’s language and communication. (Vol, 12, pp. 47-81). Hillsdale, NJ: Erlbaum. Nelson, K. (1991). Concept development in the perspective of the recent history of developmental psychology. In F. S. Kessel, M. H. Bornstein, & A. J. Sameraoff (Eds.), Contemporatoy constructions of the child: Essays in honor of William Kessen (pp. 93-109). Hillsdale, NJ: Erlbaum. Oakes, L. M. (1994). The development of infants’ use of continuity cues in their

perception of causality. Developmental Psychology, 30, 748-756. Oakes, L. M. (2003). Development of causal perception. In L. Nadel, P. Chalmer, P. Culivoer, R. L. Goldstone, & R. M. French (Eds.), Encyclopedia of cognitive science (Vol. 1, pp, 456-460), London: Macmillan. Oakes, L. M., & Cohen, L. B. (1990). Infant perception of a causal event. Cognitive Development, 5, 193-207. Oakes, L. M., & Cohen, L, B, (1995). Infant causal perception. In C, Rovee-Collier & L. P. Lipsitt (Eds.), Advances in infancy research. (Vol. 9, pp. 1-54). Norwood, NJ: Ablex. Oakes, L. M., Coppage, D, J., & Dingel, A. (1997). By land or by sea; The role of perceptual similarity in infants’ categorization of animals. Developmental Psychology, 33, 396-407. Oakes, L. M., Horst, J. ., Kovack-Lesh, K. A., & Perone, S. (2008). How infants learn categories. In A. Woodward & A. Needham (Eds.), Learning and the infant mind (pp. 144-172). New York: Oxford University Press.

Oakes, L, M., & Madole, K. L. (2000). process-oriented approach. Child Oakes, L. M., & Madole, K. L. (2003). infants’ category formation. In D. and concept development: Making

The future of infant categorization research: A Development, 71, 119-126. Principles of developmental change in H. Rakison & L. M. Oakes (Eds.), Early category sense of the blooming, buzzing confusion

(pp. 132-158). New York: Oxford University Press.

Oakes, L. M., & Madole, K. L. (2008), Function revisited: How infants construe functional features in their representation of objects. In R. Kail (Ed.). Advances in

child development and behavior (Vol. 36, pp.135-185). San Diego, CA, US: Elsevier Academic Press. xi, 2503). Oakes, L. M., Madole, K. L., & Cohen, L. B. (1991), Infant habituation and

categorization of real objects. Cognitive Development, 6, 337-392.

176

Infant Perception and Cognition

Oakes, L. M., & Ribar, R. J. (2005). A comparison of infants’ categorization in paired and successive presentation familiarization tasks. Infancy, 7, 85-98. Pelphrey, K. A., Morris, J. P., & McCarthy, G, (2004), Grasping the intentions of others: The perceived intentionality of an action influences activity in the superior temporal sulcus during social perception. Journal of Cognitive Neuroscience. Special Issue: Social Cognitive Neuroscience, 16, 1706-1716. Perone, S., Madole, K. L., Ross-Sheehy, S., Carey, M., & Oakes, L. M. (2008). The relation between object exploration abilities and attention to object features in infancy. Developmental Psychology, 44, 1242-1248. Perone, S., & Oakes, L. M. (2006). It clicks when it is rolled and squeaks when it is squeezed: What 10-month-old infants learn about object function. Child Development, 77, 1608-1622. Piaget, J. (1954). The construction of reality in the child. Oxford, England: Basic Books. Quinn, P. C, (2004), Development of subordinate-level categorization in 3- to 7-month-old infants. Child Development, 75, 886-899. Quinn, P. C., & Eimas, P. D, (1997). A reexamination of the perceptual-to-conceptual shift in mental representations. Review of General Psychology, 1, 171-187. Quinn, P. C., Eimas, P. D., & Rosenkrantz, S. L. (1993). Evidence for representations

of perceptually similar categories by 3-month-old and 4-month-old infants. Perception, 22, 463-475. Rakison, D. H. (2004). Infants’ sensitivity to correlations between static and dynamic

features in a category context. Journal of Experimental Child Psychology, 89, 1-30. Rakison, D. H. (2005a). A secret agent? How infants learn about the identity of objects in a causal scene. Journal of Experimental Child Psychology, 91, 271-296. Rakison, D, H. (2005b). The perceptual to conceptual shift in infancy and early childhood: A surface or deep distinction? In L. Gershkoff-Stowe & D. H. Rakison (Eds.), Building object categories in developmental time (pp. 131-158). Mahwah, NJ: Lawrence Erlbaum Associates Publishers. Rakison, D. H. (2006). Make the first move: How infants learn about self-propelled

objects. Developmental Psychology, 42, 900-912. Rakison, D, H. (2007), Inductive categorization: A methodology to examine the basis for categorization and induction in infancy. Cognitie Creier Comportament. Special Issue: The development of categorization, 11,773-790. Rakison, D. H., & Butterworth, G. E. (1998), Infants’ use of object parts in early categorization. Developmental Psychology, 34, 49-62. Rakison, D. H., & Cohen, L. B. (1999). Infants’ use of functional parts in basic-like

categorization. Developmental Science, 2, 423-432. Rakison, D. H., & Lupyan, G. (2008). Developing object concepts in infancy: An associative learning perspective. Monographs of the Society for Research in Child Development, 73(vii), 1-110.

Rakison, D. H., & Poulin-Dubois, D. (2002). You go this way and I'll go that way: Developmental changes in infants’ detection of correlations among static and dynamic features in motion events. Child Development, 73, 682-699. Rumelhart, D. E., & McClelland, J. L. (1986). Parallel distributed processing:

Explorations in the microstructure of cognition. Volume 1. Foundations. Cambridge, MA: MIT Press. Schlesinger, M., & Parisi, D. (2004). Beyond backprop: Emerging trends in connectionist models of development: An introduction. Developmental Science, 7, 131-132.

Information-Processing Approaches

177

Schwarzer, G., Zauner, N., & Jovanovic, B. (2007). Evidence of a shift from featural to configural face processing in infancy. Developmental Science, 10, 452-463. Shaddy, D, J., & Colombo, J. (2004). Developmental changes in infant attention to dynamic and static stimuli. Infancy, 5, 355-365. Shultz, T. R., & Cohen, L. B. (2004). Modeling age differences in infant category learning. Infancy, 5, 153-171. Smith, L. B., Jones, S. S., & Landau, B. (1996), Naming in young children: A dumb attentional mechanism? Cognition, 60, 143-171. Stager, C. L., & Werker, J. F. (1997), Infants listen for more phonetic detail in speech perception than in word-learning tasks, Nature, 388, 381-382. Stiles, J., & Stern, C. (2001). Developmental change in spatial cognitive processing: Complexity effects and block construction performance in preschool children. Journal of Cognition and Development, 2, 157-187. Strauss, M. S. (1979). Abstraction of prototypical information by adults and 10-month-old infants. Journal of Experimental Psychology: Human Learning and Memory, 5, 618-632.

Tenenbaum, J. B., Griffiths, T. L., & Kemp, C. (2006). Theory-based Bayesian models of inductive learning and reasoning. Trends in Cognitive Sciences. Special Issue: Probabilistic Models of Cognition, 10, 309-318. Tyler, L. K., & Moss, H. E. (2001). Towards a distributed account of conceptual knowledge. Trends in Cognitive Sciences, 5, 244-252. Valyear, K. F,, Culham, J. C., Sharif, N., Westwood, D., & Goodale, M. A. (2006). A

double dissociation between sensitivity to changes in object identity and object orientation in the ventral and dorsal visual streams: A human FMRI study. Neuropsychologia, 44, 218-228. Westermann, G., & Mareschal, D. (2004). From parts to wholes: Mechanisms of development in infant visual object processing. Infancy, 5, 131-151. Younger, B. A. (1985). The segregation of items into categories by 10-month-old infants. Child Development, 56, 1574-1583. Younger, B. A., & Cohen, L. B. (1983). Infant perception of correlations among

attributes, Child Development, 54, 858-869. Younger, B. A., & Cohen, L. B. (1986). Developmental changes in infants’ perception of correlations among attributes, Child Development, 57, 803-815.

9 Infant Spatial Categorization from an Information Processing Approach Marianella Casasola

Infant spatial cognition contributes to a number of important abilities that appear later in development, including navigation, map reading, and the acquisition of spatial language. The relation of infant spatial cognition to the acquisition of spatial language has motivated some researchers to study infants’ ability to form abstract categorical representations of the spatial relations between or among objects (e.g., McDonough, Choi, & Mandler, 2003; Quinn, 1994, 2004). Because languages vary widely in how they organize spatial events into semantic categories (e.g., Bowerman, 1996; Choi & Bowerman, 1991; Talmy, 1983, 1985), documenting which spatial relations infants can discriminate and group into categories has been important in informing the controversy about the universality versus language specificity of the underlying concepts expressed across language-specific semantic spatial categories (e.g., Bowerman, 1996; Landau & Jackendoff, 1993; Mandler, 1996). Studies exploring infant spatial categorization are often conducted with the goal of outlining the perceptual and conceptual starting points for the meanings expressed across spatial terms. For this reason, examining the spatial categories formed by preverbal infants and language-producing toddlers offers a particularly rich venue for understanding the relation between cognition and language in early development. The present chapter outlines recent findings on infant spatial categorization, and examines these findings within the context of a constructivist information-processing approach, This approach offers a number of advantages. First, it provides a useful theoretical framework for understanding how infants learn to form abstract categorical representations of spatial relations. Second, its emphasis on process brings to light those factors that influence infants’ ability to form abstract categorical representations of spatial relations and, consequently, can highlight how infants’ categorization of spatial events is shaped by experience. This approach may be especially useful in pinpointing whether linguistic input can shape infants’ formation of spatial categories. Third, the information-processing approach is well suited for arguing for the

Preparation of this article was supported by NSF grants PECASE BCS-0349183, BCS-0751237, and BCS-0721238.

180

Infant Perception and Cognition

fluidity of infants’ spatial categories. As will be shown, infants’ spatial categories are not all-or-none representations that have been acquired or remain to be acquired. Rather, infants’ spatial categories are better understood as developing, explaining why infants form particular spatial categories under some circumstances but not others. Finally, the information-processing approach lends itself well to understanding how the role of spatial language for forming particular abstract spatial representations can differ with the spatial category in question. In making these arguments, I focus on two information-processing principles outlined by Cohen and his colleagues (e.g., Cohen, 1991, 1998; Cohen & Cashon, 2003, 2006). These principles aptly explain some of the variation found in infants’ ability to form abstract spatial categories under different circumstances, and can account for why spatial language may play a more important role in infants’ categorization of some spatial relations, but not others. The approach taken in this chapter may lack the parsimony of showing when infants “have acquired” a particular spatial category at a given development point, and taking a single side in the debate on whether spatial language is a necessary element in forming these categorical representations. However, the approach outlines a more nuanced understanding of infants’ spatial categorization and, consequently, how this ability might contribute to some of the spatial categories formed by young children as they build a spatial lexicon. In sum, I adopt an information-processing approach to argue for the factors that influence how infants learn to form an abstract categorical representation of a spatial relation and for considering differential effects of spatial language in this learning. THE INFORMATION-PROCESSING APPROACH The information-processing approach has a long, rich history in psychology, and has served as the theoretical basis for studying a number of cognitive abilities in both infants and adults, including attention, memory, and categorization. In the realm of infant cognition, the information-processing view has advanced our understanding of how infants process and understand their visual world, and has been pivotal in explaining developmental change across a number of infant abilities. In a recent chapter, for example, Cohen and Cashon (2006) outlined six principles that embody a constructivist, informationprocessing approach. The present chapter focuses on two of these principles, referred to here as the parts-to-whole principle and the regression-under-stress principle. In the parts-to-whole principle, Cohen and Cashon argue that infants first attend to the features in their visual environment, before learning to attend to the relations among those features. The regression-under-stress principle builds on the principle that infants are biased to process input at the highest formed units they are capable of processing. Nonetheless, when taxed, infants regress to processing input at a simpler level, often times reverting to processing input at the level of features rather than the relations among those features.

Infant Spatial Categorization from an Information Processing Approach

181

‘There is a strong body of evidence in support of both principles. With respect to the parts-to-whole principle, infants show across a range of visual input that they first process input at the level of the features, prior to learning to attend to the relations among (or between) those features. For example, at six weeks, infants process the two lines that form a visual angle solely on the basis of the orientation of the individual lines, but at 14 weeks, they process the two lines in relation to the visual angle that they form (Younger & Cohen, 1984). A similar developmental change has been documented with infants’ processing of faces. Infants of three months discriminate the individual features of a face, but it is not until four months of age that they discriminate changes in the relations among those facial features (Cashon & Cohen, 2004) and respond to faces on the basis of the configuration of the features. In learning to form categories of novel animals, Younger and Cohen (1983) showed how the younger infants in their study first discriminated the novel animals on the basis of the individual features, whereas older infants responded on the basis of the co-occurrence of these features, demonstrating the ability to attend to the relations among the features of the line-drawn animals. The same parts-to-whole developmental change remains apparent when infants view dynamic events. Madole and her colleagues presented infants of 10, 14, and 18 months with objects that produced a specific function when manipulated (e.g., a toy that whistled when squeezed). Although infants of 14 months were attentive to changes in both the object’s form (i.e., its shape) and its function (i.e., whistling when squeezed), it was not until 18 months

that infants attended to the relation between the object’s form and tion, linking a specific function to a specific object (Madole, Oakes, 1993). Although these abilities seem quite distinct from each other, lows the same parts-to-whole developmental progression: infants

its func& Cohen, each folfirst dis-

criminate visual events on the basis of the simpler, featural information, while

later in development they demonstrate the ability to discriminate the same events on the basis of the relations among the features. There also is evidence in favor of the regression-under-stress principle. Even when infants can attend to the relations among features, they revert to processing visual events on a simpler basis when faced with a task that is too challenging. Often, this regression involves infants no longer attending to the relations among features and, instead, attending to only the features in a visual event. In the domain of faces, for example, infants demonstrate a period of regression as they become specialized to upright faces (Cashon & Cohen, 2004). Although infants of four months process faces on the basis of

the relations among facial features, interestingly, for a brief period at around six months, infants revert to processing both upright and inverted faces on the basis of the individual features, demonstrating a period of regression. By seven months, however, infants once again display the ability to processes faces on the basis of the relations among features, albeit with upright, but not inverted, faces. These developmental changes suggest that as infants undergo a reorganization in how they process upright versus inverted faces, they become taxed and revert to processing faces at a simpler level. The

182

Infant Perception and Cognition

regression-under-stress principle also has been observed in infants’ causal perception. By about ten months of age, infants perceive a causal relation when the motion of one object appears to cause the motion ofa second object (Oakes & Cohen, 1990), This ability can be disrupted when the task is made more challenging. When 10-month-old infants are familiarized with multiple object-pairs during habituation rather than a single object-pair (Cohen & Oakes, 1993), or when viewing a chain of events that produces an outcome (Cohen, Rundell, Spellman, & Cashon, 1999), they no longer respond on the more conceptual basis of causality, but rather discriminate the events on the simpler basis of the spatial and temporal features of the events. In the section that follows, I document how these two principles also capture infants’ ability to form abstract categorical representations of spatial relations. Before doing so, it is worth addressing the value of examining infant spatial categorization more closely and with respect to these two informationprocessing principles. Because much of the focus in this area of research has been to document how infants’ preverbal spatial concepts relate to the semantic spatial categories acquired in language, relatively little empirical attention has been dedicated to understanding the processes by which infants learn to form abstract categorical representations of spatial relations (work by Quinn and his colleagues, described below, has been pivotal in this regard). Rather, the goal of much of this research generally has been to demonstrate whether infants can form abstract categorical representations of particular spatial relations during the preverbal period, in order to argue for universal conceptual primitives or to outline the conceptual building blocks that form the basis of language-specific semantic categories. These are important goals, and well worth pursuing; however, there is an advantage to taking a step back and considering how infants are learning to form these spatial categories, and how this ability may vary with context. Demonstrating that infants’ spatial categorization can vary with task demands (among other factors) requires rethinking the argument that infants have acquired a particular spatial category by a certain developmental time point. Variability in infants’ ability to form particular spatial categories, whether it be due to differences in task demands or other reasons, leaves the door open for a possible role of spatial language in aiding infants to form some spatial categories under more challenging contexts. These possibilities lead us to reconsider the argument for universal primitives, and consider how best to frame the relation between infant cognition and spatial language in the early development of young children’s semantic categories. INFANT SPATIAL CATEGORIZATION: FROM SPECIFICS TO THE ABSTRACT Quinn and his colleagues were among the early researchers to assess infants’ ability to form abstract categorical representations of the spatial relations between or among objects. In the first of these studies, Quinn (1994) presented

Infant Spatial Categorization from an Information Processing Approach

183

young infants of3 and 4 months with static displays ofa dot in relation toa referent object, a horizontal line. At 3 to 4 months, infants generalized an “above versus below” relation across changes in location on one side of the line, but not to changes that crossed the referent line. However, when presented with a variety of symbols during familiarization (e.g., a dollar sign, a dot, an arrow, and a letter), and then tested with novel symbols (e.g., a triangle and a plus sign), the 3- and 4-month-old infants no longer demonstrated the ability to generalize the above-versus-below spatial relation to the new instances, an ability that infants of6 to 7 months did demonstrate (Quinn et al., 1996).

These results were critical for a number of reasons. First, they showed that even very young infants could generalize a spatial relation to changes in location that maintained

the familiarized spatial relation. Second, the

results outlined an important developmental change. They suggested that infants first learn to recognize a specific instance of a spatial relation prior to forming an abstract representation of the relation—one that allows them to recognize this relation when depicted by novel objects. Casasola and Cohen (2002) noted a similar progression when they tested 10- and 18-month-old infants on their ability to form an abstract categorical representation of containment, support, or tight-fit spatial relations. At 10 and 18 months, infants generalized a containment relation to a new example of the relation, demonstrating the ability to form an abstract categorical representation of containment. However, infants at both ages failed to form abstract categorical representations of the support and tight-fit spatial relations. Infants of 10 months provide no evidence of discriminating these spatial relations from novel relations. Infants at 18 months did discriminate the support and tight-fit spatial relations from novel relations, but only when familiar objects (those seen during habituation) depicted the familiarized and novel relations.

These infants did not generalize these relations to novel instances. That is, 18-month-old infants’ ability to discriminate support and tight fit was tied to familiar objects, analogous to the 3- and 4-month-old infants tested by Quinn et al. (1996) on their categorization of the above-versus-below spatial

categorization task. The parallel in findings across these two studies is striking given the difference in ages, procedure, and stimuli. The same developmental pattern emerged with younger infants of 3 to 4, and 6 to 7 months, tested in a paired familiarization procedure (the Quinn study), and with older infants of 10 and 18 months in a visual habituation paradigm (the Casasola and Cohen study). Furthermore, Quinn and his colleagues (1996) presented infants with static

symbols in relation to a consistent referent (a horizontal line or a row of diamonds), whereas Casasola and Cohen (2002) presented infants with dynamic

videos in which different toys were moved by a hand into their spatial relation to distinct referent objects. Yet, in both sets of findings, infants’ recognition of a particular spatial relation was tied initially to familiar instances (i.e., objects

or symbols that had been presented during the familiarization or habituation phase). It was not until later in development that infants could abstract the relation from a specific example, and generalize it to a novel example

184

Infant Perception and Cognition

of that relation, demonstrating contextual flexibility in their recognition of a spatial relation and, hence, the ability to form the abstract categorical representation.

Although the nitive milestone, egories remains infants as young

ability to form an abstract categorical representation is a cogwhat drives infants to gain this ability with specific spatial catunclear. The ability is not tied to age. While English-learning as six months can form abstract categorical representations

of containment and above-versus-below (Casasola et al., 2003; Quinn et al.,

1996), they do not form an abstract categorical representation of “between” until 10 months (Quinn et al., 2003), and 18-month-olds struggle in forming abstract representations of support and tight-fit (Casasola & Cohen, 2002, Casasola, 2005a, 2005b, Casasola & Bhagwat, 2007, Casasola, Bhagwat, & Burke, 2009). As Quinn et al. (2003) noted, each spatial relation has its own

developmental time course, but each nonetheless progresses through the specific-to-abstract progression. The parts-to-wholes and the regression-understress information-processing principles may help explain this developmental progression, and provide insight into what allows infants to advance from recognizing a spatial relation when depicted by unfamiliar objects, to recognizing the relation when depicted by novel objects. THE PARTS-TO-WHOLE PRINCIPLE IN INFANT SPATIAL CATEGORIZATION In examining infants’ categorization of containment, support, and tight-fit relations, Casasola and Cohen (2002) found that infants at both ages attended most easily to changes in the objects in the events, Although infants at 10 months of age formed the abstract spatial category of containment, they did not respond to changes in the support or tight-fit relations. At 18 months, infants also only formed the abstract spatial category of containment. These infants did, however, discriminate the support and tight-fit relations from novel relations—provided that familiar objects depicted each relation. These findings suggest that the objects in a spatial event may be more salient to infants than particular spatial relations. The finding is consistent with the parts-to-whole progression outlined by Cohen and Cashon (2006). That is, infants first learn to attend to the features of a spatial event (the objects) prior to the relation between those features (the spatial relation). One possibility is that

infants engage in bottom-up processing of spatial events. When first attending to a spatial event, they note the most salient aspect of the event—the objects. Once sufficiently familiarized with the objects, they shift their attention to the spatial relation between those objects. If a spatial relation is easily identified across a variety of objects, then infants can attend to this relation and form the abstract categorical representation, as they did with the containment relation. However, if the spatial relation is less salient than the objects, or more difficult to identify relative to the objects, then infants may only attend to the objects in the events, as may have been the case with the support and tight-fit relations. Perhaps because infants of 18 months could process the events more quickly

Infant Spatial Categorization from an Information Processing Approach

185

than the younger infants, they progressed beyond the objects to attend to the support and tight-fit relation between familiar objects. Nonetheless, even the older infants of 18 months failed to form abstract categories of support and tight-fit, suggesting an attentional or conceptual limitation to their spatial categorization skills. That is, while infants can form some abstract spatial categories, they do not display this ability with all spatial relations. Given that infants attend more easily to the objects in a spatial event than to the spatial relation, the objects in a spatial event may influence infants’ attention to the spatial relation and, consequently, their ability to form the abstract categorical representation. Support for this possibility emerges by comparing findings reported by Hespos and Spelke (2004), McDonough, Choi, and Mandler (2003), and Casasola and Cohen (2002), each examining infants’ ability to discriminate and generalize a tight-fit relation. At first glance, the studies appear to provide contrasting results. Hespos and Spelke, and McDonough et al., both found that English-learning infants could discriminate between a tight-fit and a loose-fit containment event. Infants in their studies also generalized a tight-fit containment relation to a new example of this relation. In contrast, Casasola and Cohen (2002) found that the infants of 10 months pro-

vided no evidence of discriminating a change in a tight-fit spatial relation. The infants of 18 months in their study did discriminate a change in the tightfit relation, but did not generalize the relation to a new example of tight-fit. Thus, two studies showed a preverbal sensitivity to tight-fit relations, while the results of the third study failed to show this same sensitivity. Placing each finding within the context of an information-processing approach reconciles these seemingly conflicting results, and highlights how the objects in a spatial event influence infants’ ability to attend to a tight-fit spatial relation. Although both studies showed that infants could generalize a familiarization event to a new instance of the same relation (e.g., tight-fit), the 5-month-old infants in the Hespos and Spelke study showed a novelty preference for test events with a change in the degree of fit between the objects, while the 9-month-old infants in McDonough et al. showed a strong familiarity preference for the spatial events that depicted the same degree of fit as the events viewed during familiarization. Thus, despite using the same procedure, infants in each study responded differently to test events that presented a novel spatial relation. The reason may lie in the complexity and variability of the objects used in each study. Hespos and Spelke (2004) presented infants with objects that were perceptually simple and highly similar in appearance to one another. The cylinders and containers differed only in size, color, or pattern from the familiarization to the test phase. In contrast, McDonough et al. (2003) used objects that were more complex (actual toys), and also were quite vari-

able in appearance. Given the greater complexity and variability in the objects presented during the familiarization phase, it is not surprising that the 9-month-old infants in the McDonough et al. study showed a familiarity preference, while the infants tested by Hespos and Spelke, although younger, showed a novelty preference. That is, the amount of complexity and

186

Infant Perception and Cognition

variability in the objects dictated infants’ response to a familiar versus novel spatial relation. Similar to McDonough et al. (2003), Casasola and Cohen (2002) presented infants with events in which the objects were quite distinct from each other (see Figure 9.1). In contrast to McDonough et al., however, they also presented different types of tight-fit relations during the habituation phase. Infants in their study viewed both tight-fit containment and tight-fit support events during habituation, whereas infants in the McDonough et al. viewed only tightfit containment events (or, alternatively, only loose-fit containment events)

during familiarization. The greater diversity of the objects, along with the

Habituation Events Tight-fit containment habituation events:

es eB.

Tight-fit support habituation events:

Test Events Familiar objects in a tight-fit relation Familiar objects in a novel relation n

bs Novel objects in a tight-fit relation

Novel objects in a novel relation

Figure 9.1, The habituation and test events used in the tight-fit condition of Casasola and Cohen (2002). (See also figure in plate section.) Source: From M. Casasola, J. Bhagwat, and A. S. Burke 2009, Developmental Psychology, 45, p. 714. Copyright 2002 by Wiley-Blackwell.

Infant Spatial Categorization from an Information Processing Approach

187

diversity in the tight-fit relations, may explain why infants in the Casasola and Cohen (2002) study failed to generalize tight-fit to a new instance while infants in the Hespos and Spelke, and McDonough et al. studies were able to do so. Possibly, as both the objects and the tight-fit relations in the events increase in complexity, infants must expend more cognitive resources toward attending to the objects at the cost of attending to the spatial relation. In line with this argument, recent findings by Son, Smith, and Goldstone (2008) have shown that young children best generalize a single instance to new examples, even ones that are more complex, when the single instance is perceptually simple. In their case, children generalized an object based on shape, and did so much more reliably when the training object was simple. The difference in results reported by the three studies above suggests that the same may be true of infant spatial categorization. Simple instances (i.e., simple objects in the spatial events, and presenting only one type of tight-fit relation), and those that highlight the spatial relation over the object, may engender better generalization to new instances of the spatial relation. These findings suggest that the objects that comprise a relation influence infants’ ability to attend to and generalize a spatial relation to form the abstract spatial category, in line with the parts-to-whole information processing principle. This pattern also suggests that infants may not necessarily immediately recognize a spatial relation, but rather may construct the relation when viewing a spatial event. Although infants of five months can generalize a tight-fit containment event to a tight-fit support event (Hespos & Spelke, 2004), infants’ representation of a tight-fit relation is not sufficiently robust to recognize the relation when depicted by more perceptually diverse examples (Casasola & Cohen, 2002). That is, infants’ spatial categorization of tight fit is

dictated, at least in part, by the complexity and diversity of the objects that comprise the events (and, likely, the diversity of the tight-fit relation itself). In sum, these findings show how infants’ spatial categorization does proceed from parts to wholes, and more importantly how the objects in the events (the parts) influence infants’ ability to attend to the spatial relation (the relations among those parts). REGRESSION UNDER STRESS: HOW INFANT SPATIAL CATEGORIZATION CAN REVERT TO FEATURES The regression-under-stress principle also captures several aspects of infant spatial categorization. Infants’ categorization of a support relation provides evidence consistent with this second information-processing principle. In one study, Casasola

(2005a)

showed

how

the number

of exemplars

(and hence,

the number of objects) presented during habituation, influences 14-monthold infants’ ability to form an abstract categorical representation of support. Infants of 14 months were habituated to either 2 or 6 different object-pairs, each placed in a support relation to a referent object. To enhance infants’ attention to the support relation, the stimuli were designed to be perceptually similar and simple. The objects were fashioned out of Crayola Modeling Magic’ and each figure had the same general shape (a snowman-like body and

188

Infant Perception and Cognition

head), although the individual features of the objects did vary, and vaguely resembled animals (e.g., a worm, duck, cat, dog, toucan bird, elephant) or people (e.g., a woman, a girl). Examples of the habituation events are presented in Figure 9.2a, while sample test events are presented in Figure 9.2b. Sample Habituation events:

aed

esi

Sample Test Events Pallas

objects in a support relation

Familiar objects in a novel relation

Novel objects in a support relation Novel objects in a novel relation

Figure 9.2. Examples of the habituation and test events in the Casasola (2005a) study. (See also figure in plate section.) Source: From “When Less is More: How Infants Learn to Form an Abstract Categorical Representation of Support,” by M. Casasola, 2005, Child Development, 76, p. 283. Copyright 2005 by Wiley. Adapted with permission.

Infant Spatial Categorization from an Information Processing Approach

189

If infants process the objects in an event prior to the spatial relation, as suggested by the parts-to-whole principle, then habituating infants to fewer objectpairs should allow them to be more attentive to the support relation, relative to infants habituated to more object-pairs. In addition, consistent with the regression-under-stress principle, infants who view many object-pairs during habituation may be overwhelmed by the number of objects and respond to the events on a simpler basis than infants viewing only two object-pairs. The results provided support for both information-processing principles. Infants who viewed only two object-pairs in a support relation responded to the test events in a manner consistent with having formed an abstract categorical representation of support (see the left-hand graph of Figure 9.3), These infants looked significantly longer at a novel containment relation than at the familiarized support relation, and did so both when familiar and novel objects depicted each relation. In contrast, infants who viewed six different pairs of objects during habituation failed to provide any evidence of attending to the support relation (see the right-hand graph of Figure 9.3). In fact, these infants did not discriminate any change in the events, suggesting that the task was too demanding. This pattern of results fits well with the regression-under-stress principle outlined by Cohen and Cashon (2006). Infants who viewed only two objectpairs during habituation formed the abstract categorical representation of

(C (© B®

Familiar object & relation Familiar object-Novel relation Novel object-Familiar relation

@

Novel object-relation

4

157

S

Mean Looking Time in Seconds

20-4

Two

Four # of Habituation Events

Six

Figure 9.3. For infants in the Casasola studies (2004, 2005a)—their looking time to the familiar event (white bar), a test event with familiar objects in a novel containment

relation (light gray bar), a test event with novel objects in the familiar support relation (medium gray bar), and novel objects in a novel containment relation (dark bar). The

first and third graphs represent findings reported in Casasola (2005a) and the middle graph represents findings reported by Casasola (2004).

190

Infant Perception and Cognition

support, generalizing support from familiar instances to a novel instance. In contrast, infants who viewed six object pairs during habituation failed to discriminate any change in the events, even a change in the objects. In a second study, Casasola (2004) gave infants four examples of the support relation, using the same object-pairs as those used by Casasola (2005a). With this number of exemplars, infants discriminated the change in support relation, but did so only with familiar objects (see the middle graph of Figure 9.3). Their performance fell in between infants who viewed two exemplars, and those who viewed six exemplars. Thus, how well infants attended to the support relation varied with the number of exemplars presented during habituation (and hence, the number of object-pairs). Few objects, seen more often, was most effective in inducing infants to generalize the relation and form the abstract spatial category. More objects appeared to detract attention away from the support relation and lead infants of 14 months to process the events on a simpler level. In sum, taxing infants with too many objects (which they seemed to process prior to the spatial relation) resulted in infants responding to the spatial events on a much simpler basis. THE INFORMATION-PROCESSING PRINCIPLES IN INFANT PERCEPTION OF OTHER EVENTS The results from Casasola (2005a) seem counterintuitive. Providing infants with more exemplars would seem more effective in facilitating abstraction. Yet, in the case of infants’ ability to process the spatial relation (at least, the spatial relation of support), fewer exemplars resulted in infants’ generalizing the support relation to a new example (that is, in their forming the abstract categorical representation of support). Analogous findings have been reported in other studies as well. For example, the number of objects used in the events also influences infants’ perception of causality. Infants of 10 months will no longer respond on the basis of causality when the task is changed from habituating infants to a single pair of objects to habituating them to multiple pairs of objects (Cohen & Oakes, 1993; Oakes & Cohen, 1990). Maguire, Hirsh-Pasek, Golinkoff, and Brandone (2008) found that older children of 2.5 and 3 years of age best generalized a novel verb when provided with only a single exemplar of the novel action (i.e., a single actor performing the action) than when provided with four actors performing the same novel action. Similarly, Song, Golinkoff, Seston, Ma, Shallcross, and Hirsh-Pasek (2007) found that using a point-light display of an actor was more effective in teaching children to learn a label for the action than having the actor be visible. Along similar lines, Kersten and Smith (2002)

found that presenting preschool-age children with novel agents made it more difficult for them to attend to an action than when viewing the same events with familiar agents. Together, these findings converge to show that the objects (in this case, agents) were more salient than the actions, and that decreasing the number of actors (i.e., objects)—and, consequently, the

Infant Spatial Categorization from an Information Processing Approach

191

demands in processing these objects—results in better learning of the relational information. In spatial mapping tasks as well, children demonstrated more accurate performance when the salience of the objects was minimized. DeLoache, Kolstad, and Anderson (1991) showed that when the objects were similar in appearance, children at 2.5 years of age performed significantly better in a spatial mapping task than peers who viewed perceptually dissimilar objects. Using pictures rather than objects also aids children in this task (Marzolf & DeLoache, 1994), perhaps because pictures are easier to use as representations than actual

objects. Reducing the processing load of the objects, either by making them similar in appearance or making them less interesting as objects, aids children’s attention to the relational information and facilitates representational insight that results in successful performance on the spatial mapping task. Hund and Plumert (2003) found that objects influenced attention to location. In the domain of analogical reasoning, the same pattern of results emerges. Young children process analogies on the basis of the similarity of the objects, and only later in development can they process an analogy on the basis of the relational information (Gentner & Ratterman, 1991; Kotovsky & Gentner,

1996). In their comparison of how math is taught in the US versus in Japan and Taiwan, Stigler and Stevenson (1992) suggested that the Asian teachers’ use of the same set of simple objects to illustrate mathematical principles aids Asian children in learning the principles more effectively than their American peers, whose teachers use a variety of objects in teaching different math concepts. Together, these findings suggest that objects influence attention to the relational information in a range of events in young children and infants. Infants seem biased to attend first to the objects in an event, prior to attending to the spatial relation presented, in line with the two information-processing principles. They also show that infants’ spatial categorization is not an all-or-none proposition. Infants can form particular spatial categories under some circumstances but fail to do so under other circumstances. Although they may be apparent under some testing conditions, infants’ spatial categories are still developing, and can be easily disrupted when infants are presented with a task that is too challenging. The next section examines how spatial language can boost infants’ spatial categorization when the task is too challenging. These findings also provide evidence to show how the parts-to-whole and the regression-under-stress principle capture the effect of linguistic input on infant spatial categorization. HOW THE EFFECT OF SPATIAL LANGUAGE ON INFANTS’ SPATIAL CATEGORIZATION FITS WITH THE INFORMATIONPROCESSING PRINCIPLES Does spatial language contribute to the development of infants’ spatial categories? The question remains at the heart of the controversy concerning the relation between infant spatial cognition and the early acquisition of spatial language (e.g., Bowerman, 1996, Mandler, 1996). On the one hand, researchers

192

Infant Perception and Cognition

have argued that spatial language and spatial cognition develop independently, and that the meanings expressed in language-specific semantic categories stem from existing spatial concepts (Johnston,

1988; Landau & Jackendoff,

1993; Mandler, 1996). On the other hand, a number of researchers have proposed that spatial language can aid infants in developing particular meanings expressed in language-specific semantic categories (Bowerman & Choi, 2003, Choi & Bowerman, 1991). The information-processing view can contribute to this discussion by helping to elucidate whether, and in what manner, spatial language shapes the spatial categories that infants learn to form, How necessary is experience with spatial language in infants’ categorization of spatial relations? As the findings referenced earlier show, infants do not require input from spatial language in forming particular spatial categories. Even when viewing more diverse instances of a spatial relation, infants form the abstract categorical representation. Korean- and English-learning infants of9 to 14 months generalized one type of containment relation (e.g., tight-fit containment) to a new instance of this relation, discriminating tightfit containment from loose-fit containment events with unfamiliar objects (McDonough, Choi & Mandler, 2003). At the same time, infants of six months

showed that they could form a broader, abstract category of containment, one that included both tight-fit and loose-fit containment, but not support relations (Casasola, Cohen, & Chiarello, 2003). The contrast in results demonstrates that infants are flexible in their ability to form narrower groupings of the containment relation (as shown by McDonough et al.) or broader, more diverse groupings of this relation (as shown by Casasola et al.). In the case of containment, then, infants do not depend on spatial language in forming an abstract categorical representation.

However, spatial language can assist infants in grouping together diverse instances of other spatial relations. For example, infants have difficulty forming a spatial category of support composed of different types of support relations. When habituated to a loose-fit resting support relation (a cup placed on an inverted bowl, and a car placed on another car), a tight-fit support relation (a Duplo’ figure placed on a Duplo’ car), and an encirclement tight-fit support relation (a turtle with a hole in its center placed on a pole and on top of other turtles), infants of 18 months failed to generalize the support relation to a new example of this relation (Casasola, 2005b; Casasola & Cohen, 2002). Similarly,

when tested on their categorization of tight fit, infants viewed both tight-fit containment (e.g., a peg placed snugly in the opening of a block) and tightfit support events (e.g., a Duplo block snapped onto a second Duplo’ block) during habituation, and did not form the spatial category. In both cases, infants may have struggled to note the relational commonality of the support or tight-fit relation across the various habituation events. Given the diversity of the objects and the spatial relations, it is not surprising that infants of 18 months have difficulty forming these broader, more perceptually diverse spatial categories. For these spatial categories, spatial language can facilitate infants’ ability to form the abstract categorical representation. Infants of 18 months, for

Infant Spatial Categorization from an Information Processing Approach

193

example, form an abstract categorical representation of support when a specific label is presented during habituation, but they fail to form the category when viewing the events in silence (Casasola, 2005b, Casasola & Bhagwat, 2007). The same results emerge with 18-month-old infants’ categorization of the tight-fit relation. Infants form an abstract categorical representation of tight-fit if provided with a specific term for the relation, but not when viewing the events in silence (Casasola, Bhagwat, & Burke, 2009). The facilitative effect is not due to the addition of linguistic phrases to the habituation events. General, attention-getting phrases lacking a specific spatial term (e.g., “Look! Look at that”) did not lead infants to form a spatial category of support (Casasola, 2005b). Rather, the beneficial effect of language is only apparent when a specific label is presented with each example of the support or tight-fit relation during habituation. How does spatial language exert its facilitative effect on infant spatial categorization? A specific label, presented with each example of a spatial relation, functions to direct 18-month-old infants’ attention to the spatial relation. Recent findings by Casasola, Bhagwat, and Ferguson (2009) provide direct support for this claim. Infants were habituated to events in which one object was placed in a support relation to a referent object, a box or an inverted bowl. During habituation, infants viewed objects from a single category, either animals or vehicles. While viewing each habituation event, infants heard a novel word. In one condition, the novel word was presented as a count noun (e.g., “It’s a toke”) while in a second condition, the same word was presented as a

spatial particle (e.g., “She puts it toke”). Following habituation, infants were tested on their categorization of both the objects and the support relation. The results showed that infants who heard the novel word as a count noun during habituation formed a category of the objects, looking significantly longer at a novel object outside of the familiarized category than a novel object from the familiarized category. In contrast, infants who heard the novel word as a spatial particle did not form the object category. These infants looked significantly longer at both novel objects, regardless of category membership. A different pattern of results was found with respect to infants’ categorization of the support relation. Infants who heard the novel spatial word discriminated between the familiar support relation and a novel containment relation, whereas infants who heard the novel count noun did not provide any evidence of discriminating between the familiar and novel relations. The results indicate that infants of 18 months are sensitive to the syntactic context of a novel word. A novel count noun focuses infants’ attention to the objects, whereas a novel spatial particle instead focuses infants’ attention to the spatial relation. These results are consistent with previous findings reported by Booth and Waxman (2002, 2003) demonstrating that a novel count noun directs attention to commonalities among objects. These findings also provide the first evidence that a spatial word uniquely directs infants’ attention to commonalities in the spatial relation. Although the count noun directed infants’ attention to the objects at the expense of the spatial relation, infants in the spatial word condition were nonetheless attentive to the objects in the

194

Infant Perception and Cognition

events. Infants in this condition discriminated between the familiar and novel objects but did not attend closely enough to the objects to organize them intoa category, presumably because their attention was more focused on the support relation, This result provides further support for the salience of the objects in spatial events: Even when there is a spatial word directing infants to the support relation, they remain sensitive to changes in the objects. In contrast, infants in the noun condition could disregard the support relation, demonstrating that attention to where an object was located could be suppressed. The findings across the two conditions fit well with the parts-to-whole and the regression-under-stress principles. Infants show that they remain sensitive to the objects (the parts) even when discriminating the support relation, but that they are not sensitive to the relations when attending to the objects, In addition, when their attention to the objects is diminished, infants revert to processing the objects on a simpler basis. They discriminated changes in the objects, but did not organize them into a category, consistent with the regression-under-stress principle. Finally, these results provide one mechanism by which spatial language can impart its facilitative effect on infant spatial categorization. By focusing infants’ attention on the spatial relation, spatial language minimizes the potential distraction of the objects in those events and, in this manner, can aid infants in forming the abstract categorical representation.

In sum, novel words presented in distinct syntactic frames can direct infants’ attention to different aspects of the same events. However, even general language phrases, those lacking a specific word, influence infants’ attention to spatial events. General phrases, such as “look at that,” add processing demands that result in infants attending solely to the objects rather than the spatial relation (Casasola, 2005b). It is not necessarily the case that the general language phrases direct infants’ attention to the objects over the support relation in the same manner as a novel count noun. Rather, general language phrases may add a sufficient processing load so that infants have difficulty attending to the linguistic input, the objects, and the spatial relation. Infants consequently process the most salient aspect of the events (the objects) and do not attend to the less salient aspect of the events (the support relation), in line with the regression-under-stress principle. By considering how infants are processing spatial language and spatial events, the information-processing view can account for those instances in which linguistic input facilitates spatial categorization (e.g., when infants are provided with a spatial word) and those instances in which it does not (e.g., when infants are provided with other types of linguistic input, such as a count noun or general language phrases). The information-processing view also predicts that the effect of language will vary with infants’ own acquisition of language. The more familiar a spatial term, for example, the less processing demands it should place on infants and the more likely it will direct infants’ attention to the spatial relation and facilitate their spatial categorization. Consistent with this argument, infants were found to form the abstract categorical representation of a support or tight-fit relation when provided

Infant Spatial Categorization from an Information Processing Approach

195

with a familiar label, but showed greater difficult in forming the category with an unfamiliar label (Casasola, 2005b, Casasola, Bhagwat, & Burke, 2009). Specifically, if infants heard “Look! It goes on” during habituation, they formed the abstract categorical representation of support (Casasola, 2005b). Similarly, if infants heard “Look! It goes tight,” they formed the abstract categorical representation of the tight-fit relation (Casasola et al.,

2008). In contrast, when infants heard a novel word in a general syntactic frame (e.g., “It goes toke”), they failed to form the abstract categorical representation of support—even when the same phrase resulted in them forming this category when hearing the familiar spatial term “on.” Rather, infants required hearing the novel word in a more informative syntactic frame, “She puts it toke.” When presented with this more informative frame, infants formed the abstract categorical representation of support (Casasola, 2005b, Casasola & Bhagwat, 2007). Given the unfamiliarity of the word, infants depended on the syntactic cues as well as the familiar verb “put” in order to direct their attention to the support relation and form the spatial category. A similar set of findings was found with infants’ categorization of the tightfit relation. Infants of 18 months required more experience with a novel label prior to using it to facilitate their spatial categorization of tight fit (Casasola et al., 2009). Thus, label familiarity and, for novel words, syntactic cues, play an important part in facilitating infant spatial categorization. These features influence how well infants can attend to the relevant spatial relation and, consequently, they matter in infants’ ability to then form the abstract categorical representation.

Finally, spatial language facilitates infants’ spatial categorization beyond their presence in a visual habituation task. Infants who are reported by their parent to be producing spatial words, such as “in,” “on,” and “up,” are more attuned to a support relation than those infants who do not. Casasola and Bhagwat (2007) found that infants of 18 months who were producing spatial words discriminated a change in the support relation, whereas those infants who were not yet producing any spatial terms did not. Along similar lines, Korean infants of 18 months form the abstract categorical representation of tight fit, perhaps because many of these infants have acquired the Korean term for tight-fit relations (Casasola, 2006). In contrast, Korean infants of 18 months no longer form a spatial category of containment, one that includes both tight-fit and loose-fit containment events. This difficulty is not due to language environment. Korean infants of 10 months do form a spatial category of containment. Rather, the difficulty may arise because infants may become more attuned to those spatial relations that are described by the spatial terms of their language. Consistent with this argument, Choi (2006) found that English-learning toddlers who were producing spatial words were less sensitive to a tight-fit relation. She proposed that as children begin to acquire the spatial words of their language, they begin to disregard those relations that are not linguistically relevant. These findings highlight how spatial language directs attention to linguistically relevant spatial relations, They also

196

Infant Perception and Cognition

show how spatial language may provide the needed conceptual boost to assist infants in generalizing a spatial relation to a novel instance. The results outlined in this section show that spatial language can facilitate infants’ ability to form an abstract categorical representation of particular spatial relations. These findings provide some of the first evidence that spatial language can facilitate infants’ ability to form abstract categorical representations of particular spatial relations. Spatial language appears to play an important role in grouping together more perceptually diverse events, and assisting infants in forming spatial categories that they may not otherwise form. That is, spatial language can highlight relational commonalities and assist in the formation of those spatial categories comprised of more perceptually diverse events. For this reason, spatial language plays a more central role in the formation of some spatial categories, but not others, As the findings described above show, spatial language imparts its effect by directing attention towards spatial relations and away from objects. Of course, how language is processed matters, and influences the impact it can have on infant spatial categorization. When the spatial categorization task is difficult, spatial language can reduce processing demands and facilitate spatial categorization. Specifically, spatial language functions to direct infants’ attention to a particular spatial relation; it may also highlight relational commonalities. These functions together aid infants in abstracting a spatial relation from particular instances to form the abstract spatial category. The results described above inform the debate over the role of spatial language in infant spatial categorization. The answer suggested by the findings so far, however, is not as parsimonious as the debate. There is evidence in support of both views. On the one hand, infants do form abstract categorical representations of spatial relations without any need for spatial language (e.g., Casasola et al., 2003; Hespos & Spelke, 2004; McDongough et al. 2003; Quinn et al., 2003), in line with the arguments advanced by Mandler (1996), Landau (Landau & Jackendoff, 1993; Munnich

& Landau, 2003) and Quinn (2007).

On the other hand, infants do not group together all types of spatial events into a categorical representation. Their ability to form an abstract categorical representation of support is tenuous—apparent under simplified conditions without the aid of spatial language (Casasola, 2005a), but requiring support from spatial language when the objects and relations are more perceptually diverse (Casasola, 2005b, Casasola & Bhagwat, 2007). Thus, there is empirical support for arguments that spatial language shapes the spatial categories that infants learn to form (Bowerman

& Choi, 2003; Choi & Bowerman,

1991).

Adopting an information-processing approach allows us to predict when spatial language will play a more central role in the formation of spatial categories and when it will not. CONCLUDING THOUGHTS AND FUTURE DIRECTIONS One goal of this chapter was to address how the information-processing view can provide insight into infants’ ability to form abstract categorical

Infant Spatial Categorization from an Information Processing Approach

197

representations of spatial relations, and address developmental changes as well as variations in this ability across tasks. An additional goal was to address the eternal issue of the relation between spatial cognition and spatial language by examining whether spatial language can shape the spatial categories that infants learn to form. Applying the information-processing principles to the study of infant spatial categorization shifts attention away from whether infants do or do not possess a particular spatial category at a particular developmental point. This view instead allows for infants to demonstrate the ability to form a spatial category under certain circumstances, while acknowledging that they may not do so under all circumstances. Consequently, spatial language can serve to facilitate infants’ ability to demonstrate a spatial category under more difficult conditions than they otherwise might. This approach also provides cohesion to what may seem disparate findings, in which infants form a spatial category in one task but not another. The approach similarly explains why language may facilitate infants’ spatial categorization in one set of conditions but not others. The information-processing approach allows us to acknowledge how the task (among other things) may be tapping or taxing infants’ processing abilities. Although the information-processing approach provides a strong fit to the data presented above, it is not unique in its effort to address the underlying processes and mechanisms that account for developmental change in infants’ spatial categorization. Nor is it unique in considering the role of language in the formation of abstract categorical representations of spatial relations. Many similar points are apparent in Gentner’s Relational Shift hypothesis (Gentner, 2003; Gentner & Ratterman, 1991), which offers comparison and structural alignment as mechanisms to account for the developmental changes in young children’s ability to process relational information. Interestingly, in this view as well, the objects play an important role in how relational information is processed. Gentner and her colleagues have reported several findings in which children first attend to the objects in analogies, or in a spatial mapping task, prior to attending to relational information—particularly when an object match is pitted against a relational match (Gentner, 1988, Gentner & Ratterman, 1991; Loewenstein & Gentner, 2005). Gentner’s theory also explicitly addresses how relational language (such as spatial language) contributes to the development of children’s relational concepts. There are many similarities between the two views, and both views offer the advantage of motivating thinking about how developmental change does occur, a point that can be easily overlooked. The findings reviewed above begin to provide some degree of insight into infant spatial categorization, and the role of spatial language in this ability. The results are exciting in showing how infants are learning to form abstract categorical representations of spatial relations, and how spatial language can direct infants’ attention to a particular spatial relation, facilitating their ability to form the abstract categorical representations under more challenging conditions. However, our understanding of these abilities remains limited. Many of the findings point to the role of objects in how infants attend to a

198

Infant Perception and Cognition

spatial relation. Nonetheless, various other factors that influence infant spatial categorization remain to be addressed. For example, the arguments about the role of objects in influencing infants’ attention to a spatial relation assume a single role for all types of objects. However, it may be that objects with static properties might differentially influence infant spatial categorization relative to objects with dynamic features. Along similar lines, studies have yet to address how infants’ attention to spatial relations may differ when viewing agents versus inanimate objects as the figures in the spatial events. In sum, research in this area remains open to many more investigations

for understanding how infants learn to form abstract categorical representations of spatial relations. However, adopting the information-processing view, or any other view that focuses the mechanisms of developmental change, allows us to better understand how infants are learning to form abstract categorical representations of the spatial relations between objects, and how this ability relates to the semantic spatial categories formed later in development.

REFERENCES Booth, A. E., & Waxman, S. R. (2002). Object names and object functions serve as cues to categories for infants. Developmental Psychology, 38, 948-957. Booth, A. E., & Waxman, S. R. (2003). Mapping words to the world in infancy: Infants’ expectations for count nouns and adjectives. Journal of Cognition and Development, 4, 357-381. Bowerman, M. (1996). Learning how to structure space for language: A crosslinguistic perspective. In P, Bloom & M., A, Peterson (Eds.), Language and space: Language, speech, and communication (pp. 385-436). Cambridge, MA: MIT Press.

Bowerman, M., & Choi, S, (2001). Shaping meanings for language: Universal and language-specific in the acquisition of spatial semantic categories. In M. Bowerman & S. Levinson (Ed.), Language acquisition and conceptual development (pp. 475-511). Cambridge: Cambridge University Press. Casasola, M. (2004, May). Is more always better? Learning to form an abstract category of support. Symposium paper presented at the International Conference on Infant Studies, Chicago. Casasola, M. (2005). When less is more: How infants learn to form an abstract categorical representation of support. Child Development, 76, 279-290. Casasola, M. (2005). Can language do the driving? The effect of linguistic input on infants’ categorization of support spatial relations. Developmental Psychology, 41, 183-192.

Casasola, M. (2006, July). How linguistic input shapes infants’ and toddlers’ spatial categorization. Oral presentation given at the Sth International Conference on Cognitive Science (Asia-Pacific), Vancouver, BC. Casasola, M., & Bhagwat, J. (2007). Does a novel word facilitate 18-month-olds’

categorization of a spatial relation? Child Development, 78, 1818-1829. Casasola, M., & Cohen, L. B. (2002). Infant categorization of containment, support, and tight-fit spatial relationships. Developmental Science, 5, 247-264.

Infant Spatial Categorization from an Information Processing Approach

199

Casasola, M., Bhagwat, J., & Burke, A. (2009). Learning to form a spatial category of tight-fit relations: How experience with a label can give a boost. Developmental Psychology, 45, 711-723. Casasola, M., Bhagwat, J., & Burke, A. S, (2009), Learning to form a spatial category of tight-fit relations: How experience with a label can give a boost. Developmental Psychology, 45, 711-723. Casasola, M., Bhagwat, J., & Ferguson, T. (2010). Infants’ attention to “what” versus “where” in dynamic events: Sensitivity to the syntactic contexts of novel words Manuscript under review. Casasola, M., Cohen, L. B., & Chiarello, E. (2003). Six-month-old infants’ categorization of containment spatial relations. Child Development, 74, 679-693. Cashon, C. H., & Cohen, L. B. (2004). Beyond U-shaped development in infants’ processing of faces: An information-processing account. Journal of Cognition and Development, 5, 59-80. Choi, S. (2006). Influence of language-specific input on spatial cognition: Categories of containment. First Language, 26, 207-232. Choi, S., & Bowerman, M. (1991). Learning to express motion events in English and

Korean: The influence of language-specific lexicalization patterns. Cognition, 41, 83-121.

Cohen, L. B. (1991), Infant attention; An information-processing approach. In M. J. Weiss & P. R. Zelazo (Eds.), Newborn attention: Biological constraints and the influence of experience (pp. 1-21). Norwood, NJ: Ablex. Cohen, L. B. (1998). Cohen, L. B., & Cashon, C. H. (2003). Infant perception and cognition. In J. Mistry, R. Lerner, & M. A. Easterbrooks (Eds.), Handbook of psychology: Developmental psychology (Vol. 6, pp. 65-89). Hoboken, NJ: John Wiley & Sons, Cohen, L. B., & Cashon, C. H. (2006). Infant cognition. In W. Damon,

R. M. Lerner, D. Kuhn, & R. S. Siegler (Eds.), Handbook of child psychology: Volume 2, Cognition, perception and language (6th ed, pp. 214-251). Hoboken, NJ: John Wiley and Sons. Cohen, L. B., Rundell, L. J., Spellman, B. A., & Cashon, C. H. (1999). Infants’

perception of causal chains. Psychological Science, 10, 412-418. Cohen, L. B., & Oakes, L. M. (1993). How infants perceive simple causality. Developmental Psychology, 29, 421-433. DeLoache, J. S., Kolstad, V., & Anderson, K. N. (1991). Physical similarity and young children’s understanding of scale models. Child Development, 62, 111-126. Gentner, D, (1998). Metaphor as structure mapping: The relational shift. Child Development, 59, 47-59. Gentner, D, (2003). Why we are so smart. In D. Gentner, & S. Goldin-Meadow (Eds.),

Language in mind: Advances in the study of language and cognition (pp. 195-235). Cambridge: MA: MIT Press Gentner, D., & Rattermann, M. J. (1991). Language and the career of similarity.

In J. P. Byrnes, & S. A. Gelman (Eds.), Perspectives on language and thought: Interrelations in development (pp. 225-277). New York, NY: Cambridge University Press.

Hespos, S. J., & Spelke, E. S. (2004), Precursors to spatial language. Nature, 430, 453-456. Hund, A. M., & Plumert, J. M. (2003). Does information about what things are

influence children’s memory for where things are? Developmental Psychology, 39, 939-948.

200

Infant Perception and Cognition

Kersten, A. W., & Smith, L. B. (2002). Attention to novel objects during verb learning. Child Development, 73, 93-109. Kotovsky, L., & Gentner, D. (1996). Comparison and categorization in the development of relational similarity. Child Development, 67, 2797-2822. Landau, B., & Jackendoff, R. (1993), "What” and “where” in spatial language and spatial cognition. Behavioral and Brain Sciences, 16, 217-265. Loewenstein, J., & Gentner, D. (2005). Relational language and the development of relational mapping. Cognitive Psychology, 50, 315-353. Madole, K. L., Oakes, L. M., & Cohen, L. B. (1993), Developmental changes in infants’ attention to function and form-function correlations. Cognitive Development, 8, 189-209. Maguire, M. J., Hirsh-Pasek, K., Golinkoff, R. M., & Brandone, A. C. (2008). Focusing on the relation: Fewer exemplars facilitate children’s initial verb learning and extension. Developmental Science, 11, 628-634. Mandler, J. M. (1996). Preverbal representation and language. In P. Bloom, M.A. Peterson, L. Nadel, & M. F. Garrett (Eds.), Language and space (pp. 365-384). Cambridge, MA: MIT Press. Marzolf, D. P., & DeLoache, J. S. (1994). Transfer in young children’s understanding of spatial representations. Child Development, 65, 1-15. McDonough, L., Choi, S., & Mandler, J. (2003), Understanding spatial relations: Flexible infants, lexical adults. Cognitive Psychology, 46, 229-259. Munnich, E., & Landau, B. (2003). The effects of spatial language on spatial representations: Setting some boundaries. In S. Goldin- Meadow & D. Gentner (Eds.), Language in mind: Advances in the study of language and thought (pp. 113-155). Cambridge, MA: MIT Press. Oakes, L. M., & Cohen, L. B. (1990). Infant perception ofa causal event. Cognitive Development, 5, 193-207. Quinn, P. C. (1994), The categorization of above and below spatial relations by young infants. Child Development, 65, 58-69. Quinn, P. C. (2004). Spatial representations by young infants: Categorization of spatial relations or sensitivity to a crossing primitive? Memory and Cognition, 32, 852-861.

Quinn, P. C. (2007). On the infants’ prelinguistic conception of spatial relations: Three developmental trends and their implications for spatial language learning. In J. M. Plumert & J. P. Spencer (Eds.), The emerging spatial mind (pp. 117-141). New York, NY: Oxford University Press. Quinn, P. C., Adams, A., Kennedy, E., Shettler, L., & Wasnik, A. (2003). Development of an abstract category representation for the spatial relation between in 6- to 10-month-old infants. Developmental Psychology, 39, 151-163. Quinn, P. C., Cummins, M., Kase, J., Martin, E., & Weisman, S. (1996). Development

of categorical 7-month-old Son, J. Y., Smith, Short-cutting

representations for above and below spatial relations in 3- to infants. Developmental Psychology, 32, 942-950. L. B., & Goldstone, R. L. (2008). Simplicity and generalization: abstraction in children’s object categorizations. Cognition, 108,

626-638.

Song, L., Golinkoff, R. M., Seston, R., Ma, W., Shallcross, W., & Hirsh-Pasek, K. (2006, November). Action stations: Verb learning rests on constructing categories of action. Boston University Conference on Language Development, Boston, MA.

Infant Spatial Categorization from an Information Processing Approach

201

Stigler, J. W., & Stevenson, H. W, (1992). The learning gap: Why our schools are failing

and what we can learn from Japanese and Chinese education. New York: Summit Books. Talmy, L. (1983). How language structures space. In H. L. Pick & L. P. Acredolo (Eds,), Spatial orientation: Theory, research and practice (pp. 225-282). New York:

Plenum Press. Talmy, L. (1985). Lexicalization patterns: Semantic structure in lexical forms. In T. Shopen (Ed.), Language, typology, and syntactic description, Vol. 3: Grammatical categories and the lexicon (pp. 36-149). New York: Cambridge University Press.

Younger, B. A., & Cohen, L, B. (1983). Infant perception of correlations among attributes. Child Development, 54, 858-869. Younger, B. A., & Cohen, L. B. (1984). Infant perception of angular relations. Infant Behavior and Development, 7, 37-47.

10 The Role of Auditory Stimuli in Infant Categorization Kim Plunkett

A substantial body of experimental evidence has demonstrated that labels have an impact on infant categorization processes. Yet little is known regarding the nature of the mechanisms by which this effect is achieved. This chapter distinguishes two accounts—supervised name-based categorization and unsupervised feature-based categorization—and describes a neurocomputational model of infant visual categorization, based on self-organizing maps, that implements the unsupervised, feature-based approach. The model successfully reproduces experiments demonstrating the impact of labeling on infant visual categorization reported in Plunkett, Hu and Cohen (2008). The model predicts that the impact of labels on categorization is influenced by the perceived similarity and the sequence in which the objects are presented to infants, and that the observed behavior in infants is due to a transient form of learning that might lead to the emergence of hierarchically organized categorical structure. The results suggest that early in development, say before 12 months of age, labels need not act as invitations to form categories, nor highlight the commonalities between objects, but may play a more mundane but nevertheless powerful role as additional features that are processed in a similar fashion to other features characterizing objects and object categories. Inan extensive series of studies, Waxman and colleagues have provided evidence for the view that labels have an impact on category formation in young infants (Balaban & Waxman, 1997; Fulkerson & Waxman, 2007; Waxman & Booth, 2003; Waxman & Markow, 1995). Using a novelty-preference procedure, infants are familiarized with a series of objects taken from the same category and then given a choice between two novel objects, one of which is from the familiarized category. During familiarization, the objects are accompanied by a novel label such as “dax,” or a neutral carrier phrase such as “Look at this.” Infants show a preference for the out-of-category object when familiarized with the novel label, but not with the neutral carrier phrase. These findings are interpreted as demonstrating that “labels facilitate categorization,” that labels “act as invitations to form categories,” and that labels “highlight

204

Infant Perception and Cognition

the commonalities between objects.” Their findings suggest that the effects are specific to the consistent use of labels that could be words in the infant's

language. Tones and buzzers don’t achieve the same effect, and the same label needs to be used consistently throughout familiarization. Using different labels doesn’t work. These findings have been reported for infants well before their first birthday, indicating that labels have an impact on infant categorization before they produce their first words and before they have acquired a substantial receptive vocabulary. A contrasting set of studies by Sloutsky and Robinson point to a different conclusion: that novel labels overshadow the processing of visual stimuli by young infants and, therefore, that auditory stimuli (including novel labels) interfere with category formation (Robinson & Sloutsky, 2007; Robinson & Sloutsky, 2004). They base their conclusions on a

series of habituation studies

in which infants are familiarized with compound auditory-visual stimuli and are then exposed to a dishabituation stimulus that changes either the auditory component or the visual component. Infants notice the change in the auditory component but not the change in the visual component. Failure to dishabituate toa change in the visual stimulus is interpreted as a failure to process the visual stimulus because the auditory stimulus overshadows the visual information

during familiarization. It should be noted that familiar auditory stimuli, such as well-known names, do not produce such dramatic overshadowing effects. Furthermore, novel labels interfere with visual processing at younger ages (10 months) but not in older infants (16 months) (Sloutsky & Robinson, 2008).

The finding that novel labels interfere with visual processing in 10-monthold infants does not sit well with the finding that novel labels can facilitate the categorization of objects: auditory dominance effects are more likely to impede categorization than to facilitate infants’ attention to the commonalities between objects. However, there are important differences in the procedures used to test infants in these studies that might readily explain the apparently discrepant findings. The categorization studies are typically conducted in a one-on-one setting, and don’t require infant habituation. Auditory dominance studies are infant-controlled, and testing does not occur until the infant reaches a habituation criterion. The stimuli and timing of events are also quite different. The categorization studies make use of objects or pictures that are likely to be familiar to infants, and that are readily interpretable as single, whole objects, and perhaps more likely to be members ofa category. The auditory dominance studies commonly exploit complex shapes unfamiliar to infants, and that are readily segmented into separate objects with no obvious category alignment. In categorization studies, labels are presented after infants have had an opportunity to examine the visual objects, whereas in auditory dominance studies, labels are synchronized with the onset (and often the offset) of the visual stimuli.

Given the transient character of auditory stimuli compared to visual stimuli in the real world (as opposed to the laboratory), the human cognitive system may have evolved to prioritize auditory stimuli in order ensure that sufficient attentional resources are available for speeded processing. The observed

The Role of Auditory Stimuli in Infant Categorization

205

auditory dominance effects may derive from the synchronization of visual and auditory onsets, a condition which does not typically hold when objects

are labeled for infants. Although it may be possible to reconcile these apparently disparate findings by appealing to differences in stimuli, timing, and experimental procedures, some difficulties of interpretation of the impact of labels on infant categorization remain. Waxman and colleagues report that infants show a novelty preference for the out-of-category object—and, thereby, evidence for categorization—only when the familiarization phase includes a novel label, and not when the familiarization phase involves a neutral carrier phrase, such as “Look at that.” Hence, the locus of the categorization effect would appear to be the novel label. However, it is well established that infants will demonstrate novelty preference effects in experimental situations that indicate category formation, but in the absence of any auditory stimulus (Behl-Chada, 1996; Eimas & Quinn, 1994). For example, after being familiarized with a sequence of cats, infants will prefer to look at a novel dog over a novel cat. Interpretation of the role of the novel label as a facilitator of category formation is compromised by the fact that infants exhibit robust category formation in the absence of any auditory stimulation. An alternative interpretation of the categorization studies is that the neutral carrier phrase interferes with category formation, whereas novel labels don’t block the process. However, this interpretation does not sit well with auditory overshadowing studies, since novel rather than familiar auditory stimuli are likely to interfere with visual processing. Neutral carrier phrases, such as “Look at this one,” are likely to be familiar to infants and should not hinder visual processing. In order to determine the locus ofa labeling effect on infant categorization, Plunkett, Hu and Cohen (2008) adapted an earlier experiment by Younger (1985). In the original experiment, Younger (1985) presented 10-month-old

infants with line drawings of animal-like objects which varied in the length of their legs and necks, the size of their tails, and the spread of their ears

(see Figure 10.1). Infants were familiarized with 8 drawings and then tested with novel drawings that differed from the familiarization set in systematic ways. In one set of drawings called the “broad condition,” the value of feature attributes across drawings were not predictive of each other. For example, long legs could just as well occur with long necks as short necks. However, in another set of drawings called the “narrow condition,” feature values were predictive of each other. For example, spread ears predicted small tails and vice versa. When subsequently tested with pairs of novel objects, with one depicting the average value for each of the features (3333) and one depicting the extreme values of each of the features (1111 or 5555), infants demonstrated a novelty preference for the extreme stimuli in the broad condition, whereas they preferred to look at the average stimulus in the narrow condition. Younger (1985) interpreted these findings as indicating that infants had formed a single category centered on the average when familiarized with broad-condition stimuli. In contrast, they formed two categories when familiarized with narrow-condition stimuli, where the average stimulus belonged

206

Infant Perception and Cognition

Broad Condition

1155

1515

2244

2424 |

4422 |

4242 |

5511

5151

1122

1212

2211

2121

4455 |

4545 |

5544

5454

Narrow

Condition

Figure

10.1. Familiarization stimuli used by Younger (1985) and Plunkett et al.

(2008). Source:

From

“Labels

as

Features

(Not

names)

for

Infant

Categorization:

A

Neurocomputational Approach,” by V. Gliozzi, J. Mayor, J-F. Hu, and K. Plunkett, 2009, Cognitive Science, 33, 711. Copyright 2009 by Cognitive Science Society. Reprinted with permission.

to neither category and the extreme stimuli were close to the averages of the categories. These experiments demonstrated that 10-month-old infants are able to exploit the correlations between feature values to determine category membership. Other experiments by Younger & Cohen (1986) demonstrated that younger infants do not exploit feature correlations in the service of category formation. In Plunkett et al.’s (2008) adaptation, the original experiments were replicated and supplemented with three additional experiments in which the narrow-condition stimuli were accompanied by novel labels during familiarization. In their Experiment 3, two labels (“dax” and “rif”) were paired with

the eight familiarization stimuli such that one of the labels was heard with members of one category (1122, 1212, 2211, 2121) and the other label with members of the other category (4455, 4545, 5544, 5454). In this case, the labels correlated with category membership, just as the visual feature values correlated with each other, so the labels were expected to support category formation and perhaps even enhance it. In Experiment 4, the two labels were pseudo-randomly paired with the narrow-condition familiarization stimuli such that one label was heard with two members from each category, and the other label was heard with the remaining stimuli. Label assignment was now in conflict with the visual feature correlations, so category formation might be disrupted. Notice that Experiments 3 and 4 used identical novel auditory and visual stimuli. Hence, if auditory stimuli overshadow the processing of visual stimuli (Sloutsky and Robinson, 2008), then the same preferences should be

observed in the test phase of each experiment where infants were given a choice between the average and extreme visual stimuli. Finally, in Experiment 5, the same label was used with each of the familiarization stimuli. Infants might choose to ignore the label because it is redundant, and form two categories as

The Role of Auditory Stimuli in Infant Categorization

207

they did with the narrow-condition stimuli in Younger (1985). Alternatively,

the label might overshadow the processing of the visual stimuli, thereby interfering with category formation. Or, the label might highlight the similarities between the familiarization stimuli (as in the Waxman

and colleagues

studies) and encourage infants to form a single category, as they did with the broad-condition stimuli in Younger (1985).

The impact of the labels on the infants’ novelty preferences differed in each of the three experiments. When two labels correlated with visual category membership, infants formed two categories as indexed by their preference for the average stimulus during testing. When the two labels were decorrelated, infants showed no novelty preferences during testing—an indication that no categories had been formed during familiarization, When a single label was used for all familiarization stimuli, infants showed a novelty preference for the extreme stimuli, indicating that a single category had been formed. In other words, infants formed two categories, one category, or no categories, depending on the labeling contingencies to which they were exposed during familiarization. Recall that the same visual stimuli were used during familiarization in all three experiments. Furthermore, the infants viewed the same testing stimuli in all experiments in silence, The variation in their novelty preferences could only be driven by differences in the mental representations they had formed during familiarization. These novelty preferences were reversed when infants were familiarized with two correlated labels versus a single consistent label, and obliterated when two labels were used in an uncorrelated fashion. We can conclude, therefore, that the relative perceptual familiarity of the test objects was affected by the impact of the labels on the categorization of the visual stimuli during familiarization; Novel objects that belong to a represented category are perceived as more familiar than novel objects outside a represented category. Note that the impact of the labels cannot be explained by the overshadowing hypothesis (Sloutsky and Robinson, 2008). Overshadowing would predict that category formation will be overridden in all three experiments, because the novel labels should prevent infants from noticing the statistical correlation among the visual feature values. However, infants failed to demonstrate category formation in only one case—Experiment 4. Even this case indicates that infants are sensitive to the correlations between the labels and the visual stimuli, since they performed quite differently in Experiment 3, which used exactly the same familiarization stimuli but with different patterns of correlation between labels and objects. The results of Plunkett et al. (2008) provide a

clear demonstration that 10-month-old infants can compute detailed statistical correlations between the values of visual features across objects, even when those visual objects are accompanied by novel labels. Moreover, these infants can compute the cross-modal statistics so that the outcome of these computations influence the process of visual object categorization. Simply put, auditory labels impact visual categorization in 10-month-olds. Although the results of Plunkett et al. (2008) fail to support the overshadowing hypothesis, neither do they support an alternative view that “labels

208

Infant Perception and Cognition

facilitate categorization” (Waxman and Markov, 1995), For labels to have a facilitative effect, the outcome of the categorization process should be noticeably different from the outcome in the absence of labels. However, there were no apparent differences between infants’ novelty preferences in Experiments 1 and 3. If labels had facilitated categorization, one should expect a failure to demonstrate a novelty preference in Experiment 1, or an enhanced novyelty preference in Experiment 3 compared to Experiment 1. The best we can conclude is that labels impact the process of categorization by virtue of their correlation with the configuration of feature values defining the set of visual objects. What, then, is the role of the label in these types of experiments? NAMES OR FEATURES? A COMPUTATIONAL INVESTIGATION Demonstrating that infants are sensitive to statistical correlations between labels and category instances does not reveal the nature of the mechanisms that are responsible for computing these statistics. One possibility, suggested by Waxman and colleagues, is that labels act as invitations to form categories by highlighting the commonalities between objects. In this view, labels play a supervisory role through their one-to-many associations with objects. Labels impact the process of categorization because multiple objects are given the same name, and objects that are given the same name belong to the same category. Categories formed in this manner often contain members that share other attributes, but they will always share the same name. A label can function as the name for the category, and may even be understood or produced by the infant to refer to members of the category. Stimuli that do not count as names, such as tones and buzzers, cannot invite infants to form categories and will not have meaning. Let us call this approach the supervised namebased account of category formation. Theories of lexical development, such as Clark’s (1973) semantic feature hypothesis, attribute a similar role to labels in the development of word meaning. An alternative approach assumes that labels are additional features-values that enter into the statistical computations performed by infants during the process of category formation. In this view, labels are nonsupervisory, i.e., they have the same status as other features, and are handled in the same manner and as part of the same statistical computation as other features. Like other features, they may vary in their salience and thereby have a greater or lesser impact on the outcome of computations, Note that redundant features do not help discriminate between categories. Contrastive features are the most informative sources in category formation. In this view, labels that do not vary contrastively across sets of objects will be redundant and fail to contribute to category formation. Let us call this approach the unsupervised feature-based account of category formation. Theories of infant categorization, such as Younger and Cohen's (1986) account of the perception of correlations among attributes, ascribe a similar role to features. Evidence as to whether labels fulfill a supervisory or nonsupervisory role for infant categorization is scant. The categorization studies reported by Waxman

The Role of Auditory Stimuli in Infant Categorization

209

and colleagues involve just a single category of objects and, consequently, do not evaluate whether infants have learned an association between the familiarization label and the object category. It is, therefore, unclear whether the categorization effect described in these studies is caused by the supervisory role of the label being associated with category members, or other factors, such as heightened attention due to the presence of a salient auditory stimulus. The results of Experiment 5 in Plunkett et al. (2008) offer some support for a name-based account. Recall that infants treat the narrow-condition stimuli as members of a single category when they are accompanied by the same label during familiarization, whereas they are grouped into two categories in the absence of any labels. On the feature-based account, the label is redundant and should be ignored. The result of Experiment 5 suggests that the label is playing a supervisory role, and perhaps acting as the name of the category. Hu (2008) reports evidence consistent with a feature-based account in a follow-up study of Experiment 3 described in Plunkett et al. (2008). After

infants had been familiarized with narrow-condition stimuli and two correlated labels, and tested using the standard novelty preference procedure, they were given an intermodal preferential looking task (Golinkoff& HirshPasek, 1987) in which novel but typical instances of the two categories (1111 and 5555) were displayed side by side, and each of the labels was played to the infants. If infants had learned the names for the two categories, then we would expect them to orient preferentially to the category instance associated with the appropriate labels (Pruden, Hirsh-Pasek, Golinkoff, & Hennon, 2006; Schafer, 2005). However, infants failed to demonstrate any preference upon hearing either label. Insofar as a null result can be interpreted as evidence, we may cautiously conclude that infants demonstrated no evidence of learning the names for the categories they had formed during the familiarization phase of the experiment, even though labels clearly had an impact on the category formation process. One interpretation of this finding is that labels fulfilled a nonsupervisory role in Experiment 3, acting simply as additional features that entered into the statistical computations leading to category formation. But why should labels play a nonsupervisory role in Experiment 3, but take on a supervisory capacity in Experiment 5? One possible answer to this question may lie in the constancy of the labeling events. Infants can readily identify that labels are used in a contrastive fashion in Experiment 3, and may engage a different learning strategy than when the same label is used. However, this explanation suggests a somewhat capricious infant, ungrounded in the labeling contingencies of the real world—labels are rarely repeated in the fashion of Experiment 5. How would the infant know which learning strategy to adopt in this situation? Gliozzi, Mayor, Hu and Plunkett (2009) investigated the possibility that an unsupervised feature-based approach could account for all the experimental findings reported in Plunkett et al. (2008). They used a neurocomputational model to simulate the infant patterns of looking behavior. The model handled visual and acoustic information in an identical fashion, and no direct associations were formed between objects and labels. In other words, the learning

210

Infant Perception and Cognition

process was unsupervised. The model consisted of a layer of neurons laid out in a grid-like fashion. Each neuron in the grid received input from an array of visual and auditory sensors. In the simplified version of the model, the grid consisted of a 5x5 sheet of neurons, and each sensory input consisted of 4 receptor cells, as shown in Figure 10.2. The visual stimuli were represented by input vectors with four dimensions, each value in the input vector corresponding to a feature in the cartoons presented to infants (the length of the legs and neck, tail width and ear separation). Each feature was first measured, and then divided by the maximal value the feature can take. A similar approach was adopted by Mareschal and French (2000) in their simulation of Younger’s original experiments. The

acoustic stimuli were also represented by four dimensions, with two dimensions active for one label, and the other two dimensions active for a second label. Each sensory receptor was connected to every neuron in the grid. Before training, the connections were set to small random values, reflecting an initial ignorant state of the system. During training, a single visual/acoustic stimulus was presented to the sensory receptors, e.g., the combination 1212 + “dax.” Activity propagated through the connections to the neural map, resulting ina pattern of activation across the map. The connections to the most active neuron in the map were adjusted so that they began to mirror the feature values of the current input vector. The connections feeding into immediately neighboring neurons were adjusted in a similar fashion. Thus, if the same input pattern was presented to the sensory receptors again, the same set of neurons would be even more active. In fact, each input pattern was presented to the model just once, so the network was exposed to eight different training items, one for each member of the familiarization set. The size of the adjustments to the connections on each trial were positively correlated with the novelty of the input stimulus (how well the input vector matches the connections feeding into the most active unit), and negatively correlated with the length of the input stimulus. In the simulations of Experiments 1-2, only visual stimuli were presented to the receptors, whereas in Experiments 3-5 both visual and acoustic stimuli were used during training. Hence, the rate of adjustment to connections

Visual Input

Acoustic Input

Figure 10.2. Neuro-computational model from Gliozzi et al. (2009). Source: From “Labels as Features (Not names) for Infant Categorization: A Neurocomputational Approach,” by V. Gliozzi, J. Mayor, J-F. Hu, and K. Plunkett, 2009, Cognitive Science, 33, 719. Copyright 2009 by Cognitive Science Society. Reprinted with permission.

The Role of Auditory Stimuli in Infant Categorization

211

tended to be higher when labels were not used. Gliozzi et al. (2009) argue that

labels increase the cognitive load, and lead to slower learning. The training procedure adopted by Gliozzi et al. is essentially identical to that used in self-organizing maps (Kohonen, 1977) where the neural grid forms a topological structure such that neighboring neurons respond similarly to similar input stimuli. The ocular dominance columns in visual cortex are known to obey these self-organizing principles. Clearly, the model used by Gliozzi et al. is not a copy of visual or auditory cortex. However, the model adopts learning procedures that are known to operate in the brain. The most important characteristics of the model, for present purposes, is that it functions in an unsupervised fashion—there is no explicit teaching signal—and it treats all the sensory feature receptors equivalently: individual visual features are treated in the same way as individual acoustic features. In general, the more two stimuli overlap in their feature specification, the closer will be their representations—the pattern of neurons that are most active—on the self-organizing map. Stimuli that tend to have non-overlapping features—such as those from different categories—will be represented at physically distant locations on the map. Note also that just as the model does not distinguish the contribution of acoustic and visual features to the organization of the map, neither does it distinguish between the features within a modality. For example, the model embodies the assumption that the visual features such as leg length and ear separation are equally important for the learning process. This assumption may be incorrect, but was adopted in the absence of evidence to the contrary. It is possible to override this assumption in the model so that certain features are attributed a greater salience than others. After the model had been trained with the relevant input stimuli for each experiment, it was tested with the same stimuli used by Plunkett et al. (2008), i.e., 3333 and 1111 or 5555. However, unlike the novelty preference procedure, the model does not have the facility for presenting two objects simultaneously. Instead, the novel visual stimuli are presented to the model individually with learning turned off. Looking time is measured by calculating how well the visual stimulus aligns with the connections feeding into the most active neuron evoked by the stimulus. A good fit indicates that the model interprets the current stimulus as familiar, resulting in short looking times, whereas a poor fit indexes relative novelty and longer looking times. A good fit would be expected when the input stimulus is similar to objects presented during the training phase of the simulation. Figure 10.3 shows the proportion of time that infants spend looking at the average stimulus in the five experiments of the Plunkett et al. (2008) study,

and the proportional preference for the average stimulus in the five corresponding conditions in the Gliozzi et al. model. Just as 24 infants were tested

in each condition of the experimental study, 24 networks with different initial random connections were tested in the simulations. Looking preferences above 50% in the infants and the networks indicate that the average stimulus (3333) is perceived as more novel than the modal stimuli (1111 or 5555).

212

Infant Perception and Cognition

60 |

GB infants L_] network

Lu

£

ei23> =o

50+

—-—



ae

=

ae

32

x8

8s

5

s

=

45

S a

40 1

2

3

4

5

Experiment

Figure 10.3. Comparison of network performance and infant performance on Plunkett et al. (2008) categorization task. Source: From “Labels as Features (Not names) for Infant Categorization; A Neurocomputational Approach,” by V. Gliozzi, J. Mayor, J-F. Hu, and K, Plunkett, 2009, Cognitive Science, 33, 724. Copyright 2009 by Cognitive Science Society. Reprinted with permission.

Overall, the pattern of novelty preferences in the simulations mimics that of the infants quite closely. In the experiments where the infants showed a significant novelty preference for the average stimulus, so did the networks, and where the infants preferred the modal stimuli over the average stimulus, the networks did, too. The capacity of the model to mimic infant novelty preferences across all five experimental conditions shows that an unsupervised learning device which performs statistical computations on compound visual and acoustic stimuli offers a viable solution to the problem of how labels influence category formation in the infant experiments. In particular, the success of the Gliozzi et al. simulations adds credence to the unsupervised feature-based account of infant categorization: labels need not act as names for the objects in order to have an impact on categorization. Models implement theoretical explanations ofa dataset. The implementation may be successful but the theory may still be wrong. As with any other scientific explanation, the success ofa model depends on the plausibility of its assumptions, its success in accounting for the data, and its capacity to generalize to new circumstances; i.e., to generate novel empirical predictions. Of course, the Gliozzi et al. model was built specifically to investigate a particular dataset. However, it is possible to examine how the model fares when compared to infant performance in different parts of the original experiments. For example, Plunkett et al. report on the time course of infant looking during the familiarization phase of the experiment. Specifically, they found that infants spent more time fixating the familiarization stimuli in Experiment 1 (broad

The Role of Auditory Stimuli in Infant Categorization

213

condition) than Experiment 2 (narrow condition), and that the presence of labels during familiarization (Experiments 3-5) prolonged looking times compared to the absence of labels (Experiments

1-2). Furthermore, infants

spent more time looking at the stimuli during the first three trials of familiarization than the last three trials, in all experiments. It turns out that the Gliozzi et al. model accounts successfully for all of these effects on infant looking times during familiarization. The reduction in looking time for the final trials of familiarization is explained by the model in terms of the decreasing novelty of the input stimuli. Recall that before training, the connections are initialized to small random values. Consequently, any stimulus will appear novel to the network. However, after five training trials, the connections will have adapted to the kind of stimuli that are presented, so that even novel stimuli taken from the same domain will appear more familiar, and hence attract lower levels of looking, than the first few training trials. In essence, the model gets used to seeing these strange cartoon figures. The higher level of looking times in the presence of labels is a direct consequence of the cognitive load imposed by the greater complexity of the compound input stimuli. The learning algorithm for adapting the connections in the map imposes a damping effect on learning for complex (longer) stimuli. Stimuli that appear novel to the network will remain so for longer if they are complex than if they are simple. Compound visual/acoustic stimuli, therefore, produce longer looking times than unimodal visual stimuli. In order to understand how the model explains why infants look longer in the broad condition than the narrow condition, it is necessary to consider characteristics of the familiarization stimuli themselves (see Figure 10.1). The

model treats each visual object as a 4-bit vector, each bit corresponding to one of the visual features, so that any object corresponds to a single point in a 4-dimensional Euclidean space. The familiarization phase of the experiment involves the successive presentation of 8 objects, so familiarization can be likened to traversing a Euclidean pathway connecting 8 points, with learning occurring at each point on the pathway. The distance between any two points in the space is a measure of the similarity of the two objects. The pathway traversed by the model will depend upon the particular random sequence of objects presented—some pathways will be longer than others—even though the same 8 objects are used in a given training condition. A simple calculation of the average Euclidean distance traversed by the 24 networks in the broad condition reveals a significantly greater distance than that traversed in the narrow condition. Since distance reflects perceived similarity and novelty, then the model exhibits longer looking times in the broad condition than narrow condition. If the same calculations are performed for the object sequences to which infants are exposed in the Plunkett et al. (2008) study, then it is also found that broad-condition infants experience longer Euclidean pathways than narrowcondition infants. However, even within an experimental condition, Euclidean pathways differ. The model predicts that infant looking times should also differ within an experimental condition: infants experiencing longer Euclidean

214

Infant Perception and Cognition

pathways should look the longest during familiarization. Gliozzi et al. reanalyzed the infant data and found a significant positive correlation between looking times and Euclidean distances, as predicted, This finding indicates that infant looking times are directly influenced by the particular sequence of the same 8 objects to which they are exposed, suggesting that infants are performing online similarity comparisons of consecutive objects, and these comparisons are driving their looking behavior. Furthermore, the metric of similarity that infants use in the experiments appears to be closely related to that used by the computational model. A surprising success of the model was its ability to mimic the infant behavior in Experiment 5, where one label was used with all narrow-condition stimuli, and infants demonstrated a novelty preference for the modal testing stimuli, indicating formation of just a single category. Recall that the model implements an unsupervised, feature-based account of categorization, and that redundant stimuli on such an account should be ignored by the statistical computations. The label in Experiment 5 is redundant because it is presented with all 8 objects, and yet it has an impact on categorization since in the absence of the label, the networks (and infants) form two categories rather

than one. However, the model was only exposed to a single presentation of each compound acoustic/visual stimulus. When the networks in Experiment 5 were trained for multiple epochs, they learned to segregate the visual stimuli into two categories rather than one, thereby ignoring the label. In other words, the formation of the single category is a transient effect in the model. This finding has several implications. First, it predicts that infants should show similar transient effects, such that if they were continuously trained on the label-object contingencies of Experiment 5, they would eventually form two visual categories. Second, the transition from a single- to a two-category representation in the model implies that the label changes its status from being associated with a single visual category of objects, to being associated with two discriminable visual categories of objects. This suggests that the model has the potential to represent a hierarchy of categorical organization as a result of the introduction of a common label for members of the hierarchy. However, the organizational capacity of the label is obtained at a price. The model suggests that initially, the label obliterates categorical distinctions, and it is only through further experience and internal reorganization that a hierarchical structure is achieved. Nevertheless, these transitions are emergent properties of a self-organizing system which does not require explicit instruction or feedback. The model predicts that the demonstrated impact of labels on categorization in 10-month-old infants does not represent an end point of learning, but rather is a step en route to the development of a more structured system—perhaps a system that underpins the organization of the mental lexicon itself. PERCEPTUAL LOAD HYPOTHESIS We have seen how labels can impact the formation of perceptual categories by infants such that the structure of labeling events influences the number

The Role of Auditory Stimuli in Infant Categorization

215

of categories formed. Specifically, when labels correlate with the structure of perceptual categories, then the perceptual categories are maintained by the infants, whereas if labels are uncorrelated they can either interfere with category formation or lead to the formation of completely new perceptual categories. This position is inconsistent with the view that auditory stimuli overshadow the processing of visual stimuli, but is consistent with the view that labels impact infant categorization (though as additional features rather than names). An additional experiment reported by Hu (2008) questions the validity of this conclusion. Recall that exposure to broad condition stimuli in the absence of labels (Experiment 1 of Plunkett et al., 2008) leads to the formation of a single category. Hu (2008) reports another experiment in which broad-condition stimuli are presented together with a single label during familiarization, thereby maintaining the correlation between category membership and label assignment. A straightforward prediction from the other experiments described by Plunkett et al. is that infants should continue to form a single category for the same reasons that they continued to form two categories in Experiment 3—a correlation between visual category information and labeling contingencies. However, Hu (2008) found that this was not the case. Infants exhibited no systematic novelty preference during testing, indicating that they had failed to form a category during familiarization, even though they showed evidence of category formation with broad-condition stimuli in the absence of a label. This finding is inconsistent with the view that infants exploit the correlation between visual category structure and labeling contingencies. Two explanations seem plausible. The presence of the label persuades infants to treat all familiarization stimuli as equally good members of the category. The category no longer has a family resemblance structure, but is of the classical, definitional kind. In this view, the average stimulus has no special status and the modal stimuli are no longer perceived as particularly novel. However, the attribution of this role to the label would predict that infants in Experiment 5 of Plunkett et al. (2008) would also fail to show a novelty preference, whereas, in fact, they did. An alternative explanation is that the label overshadows the processing of the visual stimuli in line with the Robinson and Sloutsky (2004) account,

precluding the formation of any categories and resulting in failure to demonstrate a novelty preference. At first glance, the overshadowing hypothesis would seem to lack parsimony, as it fails to account for the pattern of findings in the other experiments. However, a closer analysis of infant behavior in the one-label broad-condition experiment reveals why under some circumstances infants may fail to show category formation, whereas an apparently minor adjustment to the stimuli (removing the label) reinstates category formation. Plunkett et al. (2008) demonstrated that infants look longer during famil-

iarization in the broad condition (Experiment 1) than in the narrow condition (Experiment 2), Gliozzi et al. (2009) showed how this could be explained in

terms of the length of the Euclidean pathways infants must traverse when viewing the sequence of familiarization objects—the longer the pathway, the longer

216

Infant Perception and Cognition

the looking time. Now, suppose we examine in detail the Euclidean distances traversed by each infant in Experiment 1 as a function of their novelty preferences in the test phase of the experiment, as shown in Figure 10.4. Keep in mind that each infant is likely to traverse different Euclidean distances during familiarization, because they are exposed to different sequences of the same set of 8 objects. Overall, in Experiment 1, infants show a novelty preference (above 0,5) for the modal stimuli. Figure 10.4 shows that almost all the infants who prefer the modal stimuli have long Euclidean pathways, whereas the rest of the infants are quite varied in RMS values. In other words, the infants who showed a novelty effect at test tended to experience the greatest perceptual shifts during familiarization. Figure 10.5 shows the same type of analysis for the infants in the single-label, broad condition experiment. Now the pattern is reversed. Overall, the infants in this experiment failed to show any systematic preference. However, all the infants who failed to show any novelty preference for the modal stimuli were exposed to high Euclidean pathway sequences. This pattern of results suggests that hearing a label and experiencing large perceptual shifts during familiarization is detrimental to category formation. This contrasting pattern of results can be explained in terms of a perceptual load hypothesis. For simplicity, let us assume that only two factors contribute to the variability in the perceptual load on infants in these experiments. The first factor is the label, such that the presence of a label increases perceptual load. The second factor is the Euclidean pathway traversed by the infant during familiarization, such that longer pathways increase the perceptual load. Finally, assume that category formation and novelty preference is sensitive to perceptual load according to an inverted U-shaped function, a Goldilocks Principle: you've got to get it just right—too No Label Condition 1.05

@

Novelty

e

Familiarity

0.8 5 a 5 -

e

$ ge

0.64

5

=

bette eeeee @.........-. Qccctrt

=e044

a

oo

*e. RRR

ae 7

a[5

o

3

Zz

0.2 4

0.0 3.0

T

T

T

7

:

3.5

4.0

4.5

5.0

5.5

Euclidean (RMS) Distance

Figure 10.4. Novelty preference for modal stimuli as a function of Euclidean distance in Broad condition (Experiment 1) of Plunkett et al. (2008). (See also figure in

plate section.)

The Role of Auditory Stimuli in Infant Categorization

217

Label Condition 1.05

©@ Novelty O Familiarity

0.8 5

gc x

2

=ceenessar 8

*e =suiw eens peers toy 0G cf

=

z2 045

a

So

za

0.2 4 0.0 3.0

T 3.5

T 4.0

45

T 5.0

1 5.5

Euclidean (RMS) Distance

Figure 10.5. Novelty preference for modal stimuli as a function of Euclidean distance in single label, Broad condition of Hu (2008). (See also figure in plate section.)

Novelty Preference

Broad Condition No Label

Broad Condition One Label

Pereceptual Load

Figure 10.6. Inverted U function of infant novelty preference as a function of perceptual load.

much perceptual load interferes with categorization, but so does too little. Infants who showed a novelty preference in Experiment | experienced large Euclidean distances, placing them near the peak of the inverted U-shaped function (see Figure 10.6). However, infants who heard a label and experienced large Euclidean distances were under heavier perceptual load, and so were less likely to show a novelty preference. The perceptual load hypothesis therefore predicts that labels can interfere with categorization when infants are already under demanding processing conditions. This is not the same as the overshadowing hypothesis that predicts

218

Infant Perception and Cognition

that novel labels will always interfere with visual processing. In contrast, the perceptual load hypothesis predicts that labels will facilitate categorization if the visual processing load is reduced; e.g., by using short Euclidean pathways. The perceptual load hypothesis might, therefore, be able to resolve the conundrum whereby auditory stimuli sometimes appear to overshadow the processing of visual stimuli, and at other times facilitate visual processing and categorization. Note that virtually all of the experiments on categorization performed by Waxman and colleagues involve visual objects that are likely to be quite familiar to infants. The visual perceptual load resulting from object processing will therefore be low, and novel labels can enhance processing to the peak of the inverted U-shaped function, resulting in categorization and the expression of a novelty preference. In contrast, the experiments conducted by Sloutsky and colleagues involve entirely novel objects and therefore high perceptual load, so that the addition of a novel label results in perceptual overload and failure to categorize. Under the perceptual load hypothesis, category formation is not just the result of the computation of statistical correlations across the cross-modal stimuli. The outcome of these computations will also depend on the load that individual stimuli place on the perceptual system. Therefore, there is no simple answer to the question as to whether labels facilitate or interfere with infant categorization. The answer must be that it all depends on the status of infants’ current mental representations of the experimental stimuli. Thus, Sloutsky & Robinson (2008) did not find overshadowing effects when infants were pre-

familiarized to the auditory stimuli in their experiments, presumably because pre-exposure decreases the perceptual load. NAMES OR FEATURES? A REPRISE Unsupervised feature-based approaches, supplemented by a consideration of perceptual load effects, can offer an explanation for a wide variety of experimental findings relating to the impact of labels on infant visual categorization. Labels do not need to be names to influence the categorization process. However, this does not mean that labels are not special to the infant. Just as some visual features of an object, such as its shape, may assume a special role in object categorization, so may certain types of auditory signals—such as those phonotactically legal sequences which emerge from the mouths of socially significant others, and act as salient features in directing categorization. Socially significant auditory signals, such as words, may acquire salience even for the young infant, and salient stimuli might be particularly useful in driving statistical computations across features. Fulkerson & Waxman (2007)

report that words but not tones facilitate categorization in young infants, suggesting that words have achieved some special significance by 6 months of age. Such findings are consistent with an unsupervised feature-based approach to infant categorization. Is there any role for a supervised name-based approach to infant categorization? We know that young infants are able to demonstrate an

The Role of Auditory Stimuli in Infant Categorization

219

appreciation of familiar label- object associations (Tincoff& Jusczyk, 1999) and are able to learn novel label-object associations (Pruden, Hirsch-Pasek,

Golinkoff, & Hennon, 2006). It is unclear whether these early labels should be considered as names, since they may not be referential in character and may be limited in scope of generalization. Irrespective of the referential nature and generalizability of early words, the formation of labelobject associations

might

impact

categorization

beyond the statistical role ascribed to based approach. However, this may be ies of concept formation in adults are which labels impact the categorization

in a manner

that goes

labels in the unsupervised featuredifficult to demonstrate. Even studequivocal about the mechanism by process, For example, Yamauchi &

Markman (2000) have demonstrated that labels function in much the same

fashion as ordinary perceptual features in a straightforward classification task, but that labels differ from other perceptual features when subjects are required to make inferences—inferences based on feature information are governed by perceptual similarity, whereas inferences based on knowledge of labels are more categorical. Nevertheless, Yamauchi & Markman (2000) acknowledge that “category labels can be viewed as reliable pointers to systematic knowledge structures that may provide a basis for making predictions about unknown features” (p.793), suggesting that it may be the predictive reliability of labels rather than their status as names that is driving the asymmetry between linguistic and nonlinguistic features in adult categorization experiments.

Lupyan, Rakison, and McClelland (2007) have also shown that labels can facilitate categorization even when the labels are redundant to task success. Furthermore, they showed that categorization performance correlated positively with verification performance (akin to Yamauchi & Markman’s 2000 classification task), indicating that subjects who were best at identifying names of training items also performed best at categorizing them. This result can be interpreted as evidence for name-based categorization. However, the authors note that their findings are compatible with “a general account that naming a category causes items within that category to cohere because the name serves as a reliable cue to class membership” (p. 1082). In other words, even these adult findings are compatible with an unsupervised feature-based account of categorization, but where labels have the status of particularly reliable predictors, presumably as a result of a lifetime of use. Of course, separating the contributions of name-based and feature-based approaches to categorization is complicated by the fact that former involves the latter; names are reliably correlated with category membership. Should we therefore abandon any attempt to resolve this issue? I think not. Names refer. Features do not refer. One might reasonably suppose that the cognitive mechanisms that underpin the referential use of labels are separable from the mechanisms that exploit labels as perceptual features. The identification of these mechanisms, and perhaps their neural instantiation, promises to provide insights into the manner by which labels impact the process of categorization in both infants and adults.

220

Infant Perception and Cognition

REFERENCES Balaban, M. T., & Waxman, S. R. (1997). Do words facilitate object categorization in

9-month-olds. Journal of Experimental Child Psychology, 64, 3-26. Behl-Chada, G. (1996). Basic-level and superordinate-like categorical representations in early infancy. Cognition, 60, 105-141. Clark, E. V. (1973). What’s in a word? On the child’s acquisition of semantics in his

first language. In T. E. Moore (Ed.), Cognitive development and the acquisition of language. New York: Academic Press. Eimas, P, D., & Quinn, P. C. (1994). Studies in the formation of perceptually based basic-level categories in young infants. Child Development, 65, 903-917.

Fulkerson, A. L., & Waxman, S. R. (2007). Words (but not tones) facilitate object categorization: Evidence from 6- and 12-month-olds. Cognition, 105(1), 218-228.

Gliozzi, V., Mayor, J., Hu, J. -F., & Plunkett, K. (2009). Labels as features (not names) for infant categorization: A neuro-computational approach. Cognitive Science, 33, 709-738.

Golinkoff, R., Hirsh-Pasek, K., Cauley, K. M., & Gordon, L. (1987). The eyes have it: Lexical and syntactic comprehension in a new paradigm. Journal of Child Language, 14, 23-46.

Hu, J. F. (2008). The impact of labeling on categorization processes in infancy. Oxford: University of Oxford. Kohonen, T. (1977). Associative memory: A system theoretical approach. New York: Springer.

Lupyan, G., Rakison, D, H., & McClelland, J. L. (2007). Language is not just for talking: Redundant labels facilitate learning of novel categories. Psychological Science, 18(12), 1077-1083. Mareschal, D., & French, R. M. (2000). Mechanisms of categorization in infancy.

Infancy, 1, 59-76. Plunkett, K., Hu, J. F., & Cohen, L. B. (2008). Labels can override perceptual categories in early infancy. Cognition, 106(2), 665-681. Pruden, S. M., Hirsh-Pasek, K., Golinkoff, R. M., & Hennon, E. A. (2006). The birth of words: Ten-month-olds learn words through perceptual salience. Child Development, 77(2), 266-280. Robinson, R. W., & Sloutsky, V. M. (2004). Auditory Dominance and its changed in the course of development. Child Development, 75(5), 1387-1401. Robinson, C. W., & Sloutsky, V. M. (2007). Visual processing speed: Effects of auditory input on visual processing. Developmental Science, 10(6), 734-740. Schafer, G. (2005). Infants can learn decontextualized words before their first birthday. Child Development, 76(1), 87-96.

Sloutsky, V. M., & Robinson, C. W. (2008). The role of words and sounds in infants’ visual processing. Cognitive Science, 32(2), 342-338.

Tincoff, R., & Jusezyk, P. W. (1999). Some beginnings of word comprehension in 6-month-olds. Psychological Science, 10(2), 172-175.

Waxman, S. R., & Markow, D. B. (1995). Words as invitations to form categories: Evidence from 12- to 13-month-old infants. Cognitive Psychology, 29, 257-302. Waxman, S. R., & Booth, A. E, (2003). The origins and evolution of links between

word learning and conceptual organization; New evidence from 11-month-olds. Developmental Science, 6(2), 130-137.

The Role of Auditory Stimuli in Infant Categorization

221

Yamauchi, T., & Markman, A. B, (2000). Inference using categories. Journal

of Experimental Psychology: Learning, Memory, and Cognition, 26(3), 776-795.

Younger, B. A, (1985). The segregation of items into categories by 10-month-old infants. Child Development, 56, 1574-1583. Younger, B. A., & Cohen, L. B. (1986). Developmental change in infants’ perception of correlations among attributes, Child Development, 57, 803-815.

11 The Development of Categorization and Facial Knowledge: Implications for the Study of Autism Lisa C. Newell, Catherine A. Best, Holly Gastgeb, Keiran M. Rump, and Mark S. Strauss

Autism is a pervasive developmental disorder with onset before the age of three years. Recent studies suggest that its prevalence in the United States is as high as | out of every 150 children (Centers for Disease Control and Prevention,

2007), and there is concern that the incidence of autism is rising worldwide (Chakrabarti & Fombonne, 2005). There has been a tremendous amount of research on autism, especially in recent years. However, the theories which attempt to explain autism have failed to address how autism might develop before the diagnosis occurs (i.e., the development of autism during infancy). In addition, there is little theoretical explanation of the development of the diagnostic symptoms of autism. Despite the lack of scientific insight into the early development of autism, there is a growing body of evidence that symptoms of autism are present before the first birthday, The intervention literature has suggested that early intervention is most effective at altering the course of the disorder. Thus, there is a pressing need to diagnosis autism before age two. Research findings from our lab indicate a number of cognitive deficits in individuals with autism that are developing during infancy for typically developing individuals (e.g., gender categorization of faces). We, along with several other labs across the country, are beginning to investigate potential early markers of autism in a high-risk population (infant siblings of children with autism). This paper will illustrate how the current field of infant and adult cognitive theories can elucidate some the mysteries of autistic symptomatology. We will first review the prevailing cognitive theories of autism, followed by a review of the growing body of evidence that autistic symptoms are present during infancy. We will then describe how the current research on deficits in autism fit with theories of the categorization of objects and faces. These findings are presented against a backdrop of an information-processing approach, Finally, This research was supported by NICDH grants: PS0HD055748 & PO1-HD35469. This chapter represents the joint research efforts of current and past graduate students in the Strauss Research lab. Order of the middle three authors reflects seniority. We would like to thank Michelle Reilly for her help on this chapter. Correspondence should be sent to [email protected],

224

Infant Perception and Cognition

we will present some data from our lab that link categorization theories with symptoms of autism and illustrate how the methods can be applied to identifying autism during infancy. COGNITIVE THEORIES OF AUTISM Clinically, autism is characterized by qualitative impairments in social interaction and communication, as well as the display of restricted, repetitive, and stereotyped patterns of behavior, interests or activities (American Psychiatric Association,

2000). To date, most

research

on autism

has focused

on social

deficits, because they are both necessary and unique to the diagnosis. However, there is a growing literature suggesting that individuals with autism also have significant information processing differences (e.g., Frith & Happe, 1994; Mottron & Belleville, 1993; Ozonoff & Strayer, 1997; Plaisted, O’Riordan, & Baron-Cohen, 1989) and, indeed, some believe these processing differences should be considered as part of the diagnostic criteria for autism (Mottron, Dawson, Soulieres, Huert, & Burack, 2006). Several theories have been pro-

posed to be the “core cognitive deficit” underlying autism. These theories include theory of mind (e.g., Baron-Cohen, 1985), executive functioning (e.g., Ozonoff, 1997), and weak central coherence (e.g., Frith & Happé, 1994).

Theory of Mind Account Theory of mind was originally defined as the ability to understand that others have beliefs, desires and intentions that are different from one’s own; to impute mental states to self and others; and to predict and explain behavior based on these mental states (Premack & Woodruff,

1978). More simply, theory of

mind is the ability to take the perspective of someone else and understand that their perspective may be different from one’s own ideas and beliefs. This ability is important because it allows one to represent the thoughts and feelings of another person and understand social behavior. Wimmer and Perner (1983) found that children can successfully pass false belief tasks (requiring

an understanding of someone else’s false belief) at 3-4 years of age, providing evidence that children of that age have a theory of mind. Because autism is a developmental disorder in which children display evidence of a variety of social difficulties, Baron-Cohen, Leslie, and Frith (1985)

predicted that children with autism may not have a theory of mind. They found that only 20% of children with autism (with a mental age above 4 years) were able to pass a false belief test compared to 86% of typically developing children and 86% of children with Down’s syndrome. Thus, they concluded that one of the primary deficits in autism is an inability to attribute mental states to oneself and others, or a lack ofa theory of mind. One strength of the theory of mind hypothesis is that it can be related to the three domains of the symptoms of autism (Baron-Cohen, 2002; Joseph and Tager-Flusberg , 2004; Tager-Flusberg, 2000). Another strength is that the broader theory of mind account that includes potential early precursors

The Development of Categorization and Facial Knowledge

225

of theory of mind, such as joint attention, provides a developmental explanation of difficulties with mental-state understanding from early in life (Baron-

Cohen, 1995; Tager-Flusberg, 2001). However, one weakness of the theory of mind hypothesis is that the deficits are not specific to autism, and have been found in children with mental retardation (e.g., Yirmiya & Shulman, 1996), deaf children (e.g., de Villiers & de Villiers, 2000), and blind or severely visually impaired children (e.g., Peterson, Peterson, & Webb, 2000). Another weakness is that many of the tests of theory of mind do not adequately capture the ability to understand mental states in real-life social interactions, and deficits in individuals with autism may be related to the cognitive load, verbal load, and/or complexity of the inferences that must be made to succeed on theory of mind tasks (e.g., Happé, 1995). In fact, failure on theory of mind tasks may actually be due to language problems or executive function problems rather than a deficit in theory of mind (see next section for review of executive functioning in autism). Executive Function Hypothesis The executive function hypothesis of autism was formulated to a large extent as an alternative to the theory of mind hypothesis. While studying theory of mind, researchers determined that performance on theory of mind and executive functioning tasks are correlated, and in some cases, that executive dysfunction was a better discriminator of autism than problems with theory of mind (Ozonoff et al., 1991). As a result, much research has been conducted

to determine whether executive dysfunction may be the primary cognitive deficit that drives the social and communicative abnormalities in autism. Executive functions are those cognitive functions that involve the ability to maintain an appropriate problem-solving set in order to attain a future goal. These abilities allow one to disengage from the immediate context in order to guide behavior toward a goal (Hughes, Russell, & Robbins, 1994). Executive functioning, however, is a ubiquitous term and lacks a clear operational definition (Bryson, Landry, & Wainwright, 1997). Over the past 20 years, studies of executive functioning have included subsets of the following processes, all of which are thought to be aspects of executive functioning: planning, impulse control, inhibition, working memory, set maintenance, problem solving, cognitive flexibility, set-shifting, organization, self-monitoring, awareness over time, mental representation of tasks and goals, forming abstract concepts, focusing and sustaining attention, self-correcting, rapid retrieval of relevant information, working memory, action monitoring, generativity, and mental operations.

Two executive functions, planning and mental flexibility, have been shown consistently to be impaired in autism (e.g.; Bennetto, Pennington, & Rogers, 1996; Ozonoff, 1995; Ozonoff & McEvoy, 1994; Pascualvaca, Fantie, Papageorgiou, & Mirsky, 1998). In general, however, research on executive functioning has mixed results, and not all researchers agree on whether executive dysfunction is a primary cause or secondary feature of autism. As

226

Infant Perception and Cognition

with the theory of mind hypothesis, there is a lack of specificity of the executive function deficit to autism. Executive functioning problems are evident in a variety of developmental and psychiatric disorders including Attention Deficit Hyperactive Disorder (e.g., Chelune, Ferguson, Koon, & Dickey, 1986), conduct disorder (e.g., Lueger & Gill, 1990), PKU (e.g., Welsch, Pennington, Ozonoff, Rouse, & McCabe, 1990); obsessive-compulsive disorder (e.g., Head,

Bolton & Hyman, 1989), Tourette’s (e.g., Bornstein, 1990) and schizophrenia (e.g., Axelrod, Goldman, Tompkins, & Jiron, 1994). Another weakness of the

executive functioning theory of autism is that executive functioning deficits have not been reliably found in preschool children with autism (Dawson et al., 2002; Griffith, Pennington, Wehner, & Rogers, 1999). If evidence of executive dysfunction cannot be found in young children with autism, one needs to question whether executive functioning is the primary core deficit in autism. It is possible that executive function deficits are characteristic of autism, but are more of a general correlate of developmental neuropathology and/or mental retardation (Tager-Flusberg, Joseph, & Folstein, 2001). Weak Central Coherence Hypothesis Finally, the weak central coherence account of autism was formulated by Frith in 1989 to account for assets in addition to deficits with a single theory. About 20% of children with autism have “islets of ability” in which they excel, and often even reach savant abilities in areas such as music, mathematics, art, and even writing (Hermelin, 2001).

Central coherence is a construct created by Frith (1989) to describe a cognitive processing style present in typically developing individuals that includes a tendency to draw together diverse information to construct higher-level meaning within context, a preference for wholes over parts, a global processing bias, and a focus on gist/meaning. Central coherence is believed to be a continuum in the population; strong central coherence leads to a good understanding of meaning but less attention to specifics, while weak central coherence leads to good knowledge of detail but less understanding of the gist or meaning. Individuals with autism lack the tendency to draw together diverse information to construct higher-level meaning within context, prefer parts over wholes, have a local processing bias, and focus on details (Frith & Happé, 1994). Thus, they have a weak central coherence. The main strength of the weak central coherence hypothesis is its ability to explain the patterns of cognitive strengths and weaknesses that have not been well explained by other cognitive theories of autism. Another strength of Frith’s theory is that it frames the strengths and weaknesses of individuals with autism as a “cognitive style” rather than a “cognitive deficit.” This framework helps people realize that individuals with autism have particular strengths that can be focused on and further strengthened with intervention.

Similar to theory of mind and executive functioning deficits, weak central coherence is not specific to autism, and is thought to be present in a variety

The Development of Categorization and Facial Knowledge

227

of disorders including William’s syndrome (e.g., Bellugi, Lichtenberger, Jones, Lai, & St. George, 2000), schizophrenia (e.g., Chen, Nakawama, Levy, Matthysse, & Holzman, 2003), depression (e.g., Derryberry & Tucker, 1994) and right hemisphere

damage

(e.g., Robertson

& Lamb,

1991). Another

problem is that weak central coherence does not provide any information about why there is a local processing bias or what the central processes are that lead to this bias. Finally, none of the research to date has found any significant relationship between social skills or repetitive behaviors and weak central coherence in individuals with autism (e.g., Morgan et al.,

2003; Pellicano et al., 2006; Turner, 1997). As a result, Frith has recently revised her theory of weak central coherence, suggesting that weak central coherence co-occurs with a deficit in social cognition, such as theory of mind, and can explain the strengths and cognitive processing features of autism that cannot be explained by other cognitive theories (Happé & Frith, 2006).

Although each of the discussed theories can explain certain behavioral aspects of autism, their ability to explain the core behavioral aspects of autism is still quite limited. To some extent this difficulty is because each theory represents broad constructs that are difficult to operationally define. It is difficult to know what exactly is meant by “theory of mind,” “executive functioning,” or “central coherence.” because the definitions tend to vary depending on the researcher, the research question of interest, and the measurements used to address this question. More importantly, despite the fact that autism is a developmental disorder that is present by the age of three years, none of the theories has adequately addressed the role of development in autism. Most studies aimed at examining these theories ignore development and include a single age group (e.g., children, adolescents, or adults), or a wide variety of age ranges (e.g., both children and adults included as a single sample). Only recently have studies included preschool-aged children, and no studies have examined these theories in infants or toddlers. Thus, little to nothing is known about the full developmental trajectory of theory of mind, executive functioning, weak central coherence, or any potential early developmental precursors that lead to the development of these abilities. It is possible that deficits in these abilities, or precursors of these abilities, begin early and change over developmental time, and these changes may occur differentially in typically and atypically developing individuals. Without studying these theories from early in life, with a developmental perspective, the developmental trajectory and potential early developmental impact of these abilities/deficits is unknown, Asa recent review of the cognitive theories concluded (Rajendran & Mitchell, 2007), while each of these theories is useful, “an ideal theory would trace [autism] from infancy through to adulthood and would apply to individuals with autism who have severe learning disabilities as well as those who are higher-functioning. Any new theory would additionally have to integrate the sociolinguistic, perceptual, and sensorimotor aspects of the disorder” (p. 247).

228

Infant Perception and Cognition

EVIDENCE OF AUTISM DURING INFANCY Although Autism Spectrum Disorders (ASDs) are rarely diagnosed before two years of age, 30-50% of parents report abnormalities in their child’s behavior before the time of diagnosis (Werner, Dawson, Osterling, & Dinno, 2000;

Zwaigenbaum et al., 2005). Retrospective studies of children with an ASD, using parent report and analysis of early home videos, suggest that there may be early indicators of an ASD before one year of age. For instance, children with an ASD have been shown to have difficulty orienting to their name by their first birthday. Gomez and Baird (2005) found that more than 60% of parents reported that their autistic children did not react to their own name and “looked through or past people.” This deficit has also been identified in several retrospective studies in which home videos of children between 8 and 12 months were reviewed for atypical behaviors (e.g., Baranek, 1999; Osterling & Dawson, 1994; Werner et al., 2000). Retrospective studies have identified a number of other deficits in children who will eventually receive an ASD diagnosis. These behaviors include orienting to social stimuli, as well as joint attention behaviors. For example, children with an ASD looked at the face of another person less often than did typically developing (TD) children (Osterling & Dawson, 1994) and were less likely than TD children to coordinate smiling with gaze at another person's face (Werner et al., 2000), Osterling and Dawson (1994) also identified deficits

in joint attention behaviors, such as showing and pointing. However, relatively few nonsocial behaviors have been investigated and/or identified. Although retrospective studies have been beneficial for observing early behavioral markers in autistic children, there are problems in standardizing the situations in home videos and relying on parents’ long-term recollection of specific behaviors. Recently, researchers have worked to examine these issues prospectively. Increasing evidence suggests a genetic link in ASDs, indicating that the siblings of children with autism (at-risk siblings) are at a higher risk

for developing an ASD than siblings of TD children (e.g., Jorde et al., 1991; Smalley, Asarnow, & Spence, 1988). Thus, research projects have begun studying the infant siblings of children with autism to potentially obtain a prospective sample of children with autism. Prospective studies have identified areas that may potentially serve as markers of risk in the future. For instance, several studies have identified language deficits in 12- to 24-month-old infant siblings of children with an ASD (Gamliel, Yirmiya, & Sigman, 2007; Mitchell, Zwaigenbaum, Roberts, Szatmari,

Smith,

&

Bryson,

2006;

Yirmiya,

Gamliel,

Shaked,

&

Sigman,

2007). However, most prospective studies which have investigated general cognitive deficits have not found reliable differences between at-risk and typically-developing populations (Gamliel et al., 2007; Yirmiya et al., 2007). One exception is evidence of earlier diagnoses and more striking impairments in children who showed significant decreases in IQ from 12 to 24 months (Bryson, Zwaigenbaum, Brian, Roberts, Szatmari, Rombough et al., 2007).

The Development of Categorization and Facial Knowledge

229

Other prospective studies have investigated deficits in social and emotional communication and cognition. Joint attention abilities are known to be impaired in children with autism; thus, they have provided a valid avenue for research with at-risk populations. At-risk siblings consistently display difficulties in responding to and initiating joint attention behaviors between 12 and 23 months (Cassel, Messinger, Ibanez, Haltigan, Acosta,& Buchman, 2006; Goldberg et al., 2005; Presmanes, Walden, Stone, & Yoder, 2007; Sullivan, Finelli, Marvin, Garrett-Mayer,

Bauman,

& Landa, 2007). These behaviors

potentially have predictive ability; Sullivan et al. (2007) found that performance on a responding-to-joint-attention measure at 14 months predicted eventual ASD diagnosis. Cassel et al. (2006) also identified disruptions in emotional responses during the Face-to-Face/Still-Face paradigm where atrisk siblings smiled less and lacked emotional continuity between phases of the paradigm. Although these are reliable differences between participant groups (ASD and TD sibs), it is not yet known whether these differences are related to an eventual ASD diagnosis. Other behaviors that are commonly identified in children with an ASD have also been investigated in their at-risk siblings. For instance, at-risk siblings who eventually receive an ASD diagnosis display more stereotyped motor behaviors, such as arm waving and putting their hands to their ears, at 12 and 18 months. Also, children’s orienting to their name is potentially a reliable marker of risk at one year of age. Not orienting to one’s name in one or two trials at 12 months is consistently related to developmental delays at 24 months, and has a high specificity for ASD or a developmental delay (Nadig, Ozonoff, Young, Rozga, Sigman, & Rogers, 2007). Relatively few prospective studies have investigated underlying perceptual and/or cognitive processes in autism. One line of research has identified diffculties in visual attention in at-risk siblings that are similar to deficits in visual attention seen in children with an ASD. Merin, Young, Ozonoff and Rogers (2007) identified a subgroup of infants (primarily at-risk infants) in a Faceto-Face/Still-Face paradigm who, at 6 months of age, demonstrated diminished gaze to their mother’s eyes relative to her mouth. This finding relates to data showing that individuals with autism demonstrate fewer fixations to eye regions and more fixations to mouth regions compared to typically developing individuals (Klin, Jones, Schultz, Volkmar, & Cohen, 2002). Individuals with

autism also show a tendency toward “sticky attention” in which they appear to be impaired in their ability to disengage and shift attention (Landry & Bryson, 2004). Ibanez, Messinger, Newell, Lambert, and Sheskin (in press) identified potential early markers of impaired visual attention in a group of 6-month-old at-risk siblings. At-risk siblings shifted their attention less frequently than control siblings, and they had longer gazes away from their parent’s face during a Face-to-Face/Still-Face protocol. These studies represent the only prospective studies to investigate possible underlying perceptual abilities in children with autism, and are the only prospective studies to identify group differences before the first birthday. To date, there are no prospective studies investigating specific cognitive abilities that develop during the first year, such as prototype

230

Infant Perception and Cognition

formation or categorization, that may be related to deficits commonly seen in children with an ASD. If, indeed, autism is a disorder that begins during infancy, it becomes critically important to understand what early underlying perceptual and cognitive abilities may be developing atypically. Both functional imaging and behavioral studies with older children and adults have suggested that individuals with autism may differ from typically developing individuals with respect to cognitive processes that are implicit and automatic, as opposed to more deliberate or conscious (Klinger, Klinger & Pohlig, 2006; Rajendran & Mitchell, 2007; Williams & Minshew, 2007). Thus, there may be differences in the types of implicit processes that are developing in the preverbal child with autism. In particular, there is growing evidence that individuals with autism may differ in their categorization abilities and in the way they process facial information (Gastgeb et al, 2006; Rump et al., in press). Research on categorization abilities during infancy is essential to an understanding of what processes may be disrupted in infants who will eventually receive an ASD diagnosis. As reviewed in the other chapters in this book, there are several core mechanisms responsible for the development of categorization during infancy. Our research with older children and adults with autism points to a few of these key mechanisms as potential early markers of autism. Thus, the rest of this chapter reviews research we have been conducting on the development of categorization and facial knowledge, integrating theory and research on typical populations, with populations with autism. In particular, the results fit well within a general information-processing approach, documenting how the difficulties displayed by individuals diagnosed with autism are related to their processing of faces and other objects. CATEGORIZATION AND FACIAL KNOWLEDGE IN INDIVIDUALS WITH AUTISM Consider the following quote from Temple Grandin, Ph.D, (Grandin, 2006), a

well-known individual who writes and lectures about what her world is like as a highly intelligent person who has autism. When I was a child, I originally categorized dogs from cats by size. That no longer worked when our neighbors got a small dachshund. I had to learn to categorize small dogs from cats by finding a visual feature that all the dogs had and none of the cats had. All dogs, no matter how small, have the same nose. (p. 30)

Temple Grandin has also discussed publicly how she did not know the difference between cats and dogs until she was around five years old, when she explicitly began to figure out the categories using features such as those described above (Grandin, 2008). Grandin’s categorization of dogs versus cats

stands in stark contrast to the significant amount of research (much cited in this book) that demonstrates that preverbal infants are able to catego-

rize objects (see, Rakison & Oakes, 2003. It also suggests that categorization may represent a primary implicit process that is deficient in individuals with

The Development of Categorization and Facial Knowledge

231

autism. Deficiencies or differences in basic categorization abilities could have a profound effect on how one learns about the world and ultimately acquires expertise about objects, social information such as people and faces, and language. As will be discussed, our current research with high-functioning individuals with autism (HFA) suggests that there are general deficits in the way that individuals with ASD process and categorize perceptual information. These deficits manifest themselves in difficulties processing both objects and social information (e.g., face identity and gender categorization), Traditional theories of categorization posit that all categories are defined by a few necessary and definitive features, and have well-defined boundaries (see, Murphy, 2002), According to these theories, all members of a category are considered to be equally representative or typical. Categorization is seen as a process where one must “learn” the rule or rules that define the category boundaries. Using this approach, studies concluded that people with autism can form categories successfully. These studies used only categories that had simple definitive features such as color or size, and did not examine whether individuals with autism process category information in the same manner as typically developing individuals, especially when categories are more complex (Tager-Flusberg, 1985; Ungerer & Sigman, 1987). In studying natural categories, Rosch (1978) discovered that natural categories such as dogs do not have simple criterial features but instead have “fuzzy boundaries.” Categories also have typicality structures in which some of the members are more representative or “better examples” of the category and other members are less representative or “worse examples” and therefore less “typical.” Individuals tend to agree on which members of a category are most and least typical, and in verification tasks, reaction times to verify or identify typical members of a category are faster than reaction times to identify less typical members. Such reaction time differences reflect memory storage—typical exemplars of a category are easier to retrieve than less typical exemplars. Additionally, children learn the names of typical members of novel categories more quickly than less typical members (e.g., Barrett, 1995). Typicality effects seem to be present from infancy, in that 18- and 24-month-old infants look significantly longer at more typical items than less typical items (Southgate & Meints, 2000), Studies of prototype formation in infants also provide evidence for the typicality effect early in life (e.g., Strauss, 1979). Thus, it is possible that while individuals with autism can successfully categorize on the basis of simple definitive features, they may have difficulty categorizing when categorization is based on more complex or less perceptually apparent features (Klinger & Dawson, 1995; Plaisted, 2000). It is also possible that while individuals with autism may be able to categorize typical exemplars, less typical exemplars may pose more difficulty. As category exemplars become less typical, criterial features also become less apparent and decision processes become more difficult. Thus, studies using only typical exemplars of a category may not indicate deficits in these individuals. Studies using atypical exemplars, however, may show categorization deficits as the task becomes more difficult.

232

Infant Perception and Cognition

Several studies support the notion that individuals with autism can successfully categorize when the task is simple or rule-based, but have difficulty when categories are more abstract or complex. Minshew, Meyer, and Goldstein (2002) found that individuals with HFA can group information in a rule-based manner, but have difficulty when the task requires that concepts be abstracted from complex information. Studies have also suggested that individuals with autism are unable to abstract prototypes or average representations of the features of a category. Klinger and Dawson (2001) compared the abilities of children with autism, and typically developing children, to use rule-based and prototype category learning. They found that both groups could categorize using a rulebased strategy when there was a simple distinctive feature, but children with autism were unable to abstract a prototype of animal-like categories. Similarly,

Plaisted

(2000)

conducted

two

studies which

indicated

that

adults and children with HFA were also unable to form prototypes. The ability to compare multiple examples of a category and abstract a prototype or central tendency is inherent in typicality structures. Rosch’s (1978) research has been further supported by more recent investigations of object and face categorization. Infants as young as three months can abstract a prototype from complex stimuli such as faces (Bomba & Siqueland, 1983; Quinn, 1987; Younger & Gottlieb, 1988). Taken together, these results suggest that individuals with autism may engage in different categorization processes than typically developing individuals. With respect to natural categories, it is possible that individuals with autism are able to categorize typical category members efficiently and accurately using simple definitive features, but have difficulty categorizing less typical category members which require a different, more complex processing strategy. Another aspect of categorization that has yet to be explored in the autism research is the developmental course of categorization in individuals with autism. While studies have examined whether children or adults with autism can categorize, no study has examined processing differences across the lifespan. Are categorization differences apparent in both children and adults with autism? If so, are there any improvements with development? How does the development of categorization compare in typically and atypically developing individuals? The following research addresses these important issues. The Categorization of Gender Gender Categorization in Typically Developing Infants and Children One of the basic natural categories that must be learned by all individuals is the category of gender and, in particular, facial gender. Adults are very good at classifying the gender of faces, and they do so very quickly (O’ Toole, et al., 1998). The discrimination of facial gender is based on a very fine-grained discrimination of the features that are maximally distinctive between male and female faces. These features include, among others, nose length, chin width, and eye to eyebrow distance (Brown & Perrett, 1993; Chronicle et al., 1995; Yamaguchi, Hirukawa, & Kanazawa, 1995), Not only are adults very good at

The Development of Categorization and Facial Knowledge

233

classifying gender, but they are also quicker to identify the gender of a face if that face has been rated as being very typical of its gender. For instance, a male face that has been rated by adults as being very masculine is classified as male in a gender identification task significantly faster than a male face that has been rated as being somewhat less masculine (O'Toole et al., 1998).

Although previous research has demonstrated that infants are able to categorize facial gender by as early as six months of age (Leinbach & Fagot, 1993; Walker-Andrews, Bahrick, Raglioni & Diaz, 1991; Yamaguchi, 2000), this research has not looked at the effect of typicality on the ability of infants to categorize gender. Also, previous research has not always controlled for whether infants had to categorize only on the basis on internal facial features or whether hair cues were presented. Finally, all prior research has used static pictures of faces as opposed to naturalistic videos. In a recent study (Newell & Strauss, in preparation), typically developing 5- and 8-month-old infants were habituated with multiple exemplars of either male or female faces until they reached a habituation criterion of a fifty percent reduction in fixation time from the average of the first three presentations. They were then presented with novel exemplars of both male and female faces. All of the exemplars were videos of men and women reciting acommon nursery rhyme (although the sound was excluded). The videos consisted of only the models’ faces; hair cues were hidden with a black cap. Importantly, all of the videos were also rated by undergraduate students for typicality of gender on a 7-point scale. Different groups of infants were tested with an habituation paradigm to faces that had been rated as either “typical” or “atypical” of men and women. Examples of typical and atypical female faces are shown in Figure 11.1. It should be noted that although some faces were rated as “atypical,” all the faces were categorized with 100% accuracy by a second group of undergraduate students. Thus, the study tested the ability of infants to categorize either typical exemplars of gender or atypical exemplars of gender. Results are shown in Figure 11.2. As can be seen, 8- month-old infants demonstrated a novelty response (i.e., increased looking to the exemplars of the gender to which they had not been habituated). However, they demonstrated this evidence of categorization only with the typical exemplars.

Typical Example

Atypical Example

Figure 11.1, Example typical and atypical female faces presented in the gender categorization tasks. (See also figure in plate section.)

234

Infant Perception and Cognition 105

O Novel B Familiar

9

*

*

p