THE PSYCHOLOGY OF LEARNING AND MOTIVATION Advances in Research and Theory
VOLUME 15
CONTRIBUTORS TO THIS VOLUME
Rob...
42 downloads
343 Views
15MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
THE PSYCHOLOGY OF LEARNING AND MOTIVATION Advances in Research and Theory
VOLUME 15
CONTRIBUTORS TO THIS VOLUME
Robert G . Cooh Aiitiiorig
ti. Greeriwild
Richrrrd J . Harris Juries E . Jcrstrzernbski
Liirie R . Jolirisoiz
Marvirr R . LcrriiO R. E . I,uho~, P c t p r A . Lircus
Doniirric W . Massnro Dorrglcrs L. Nelsori Dotidtl A . Riley P(llt/
Schriirr
Nuticy L . Steirr
Totti Trahcisso
I . Weitier
THE PSYCHOLOGY OF LEARNING AND MOTIVATION Advances in Research and Theory
EDITEDBY GORDON H . BOWER STANFORD UNIVERSITY, STANFORD, CALIFORNIA
Volirtne I 5
1981
ACADEMIC PRESS A SUBSIDIARY O F HARCOURT BRACEJOVANOVICH, PUBLISHERS New York
London Toronto Sydney
San Francisco
CUPYRlGtl~I @ 198 1, 13Y ACADEMIC I'KI:SS, INC. ALL RIGHTS RESERVED. NO PART OF mIs PUBLICATION M A Y BE REPRODUCED OR T R A N S M I U E D IN ANY FORM OR BY ANY MEANS, ELECTRONIC O H MECHANICAL, INCLU1)INC PIIOTOCOPY, RI..CORDING, OR ANY INFORMATION STORAGE AND RE I RII'.VAI. SYSTEM, WITHOUT PERMISSION IN W K l l I N G FROM THE PUBLISHER.
ACADEMIC PRESS, INC.
11 1 Fifth Avenue. New York, New York 10003
Lliritcd Kiirgdom Ediriori published b y ACADEMIC PRESS, INC. (LONDON) LTD. 24/28 Oval Koad. London NWI 7DX
LIBRARY OF CONGRESS CATALOG CARDNUMBER: 66-30104 ISBN 0 1 2 -543315-8 PRIN I L1) IN I IIE UNITED STATES OF AMERICA 81
xz
x3 84
9 8 7 6 5 4 3 2 1
CONTENTS
List of’Contributors .......................................................................................
ix
Contents of Previcius Volumes.........................................................................
xi
CONDITIONED ATTENTION THEORY R . E . Lubow, I. Weiner, und Paul Schnur I. 11. 111.
IV. V. VI.
Introduction.. ............................................................. ...................... Conditioned Attention Theory of Latent Inhibition.. .......................................... ...... Evidence for CAT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Relationship of CAT to Other Learning Paradigms. ........................ Classical Conditioning.. .................................. ........................ Conclusion.. .......................................................................... References ........................................... ............................................
1
3 16 23 34
42
A CLASSIFICATION AND ANALYSIS OF SHORT-TERM RETENTION CODES IN PIGEONS Donald A . Riley, Robert G . Cooh, and Murvin R . Lurnh I. 11. Ill.
IV. V.
VI.
Introduction ........................................................................................... Designs for the Study of Event Memory in Animals ......................................... A Classification of Codes ................................... Evidence for Different Codes ..................................................................... Stimulus Organization Coding ............................................................... Conclusions.. ........... ......................................................................... References ........................................................................................ V
51 53 58 68 76
11
Contents
CI
INFERENCES IN INFORMATION PROCESSING Riclicird J . Htrrris I. I1 . Ill . IV . V.
VI . VII
Inferences and Comprehension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Processes of Interring.,......................................................................... I'hemea and Schemas . . . . . . . . . . . . . . . . . . . . . ............................................. Schemas and Infcrciicea in Social Cognition ................................................ Varielirs 01' Dixourse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Britain Sttidiea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion: Inferring the Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reference\ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
82 XX
92 95 99 108 117 118
MANY ARE CALLED BUT FEW ARE CHOSEN: THE INFLUENCE OF CONTEXT ON THE EFFECTS OF CATEGORY SIZE Dorigitis L Nelsorl I. I1 . 111. IV .
V.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cued Recall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conceptual Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Soine Implication\ ................................................................................... Reference\ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
129 131
133 145 1%
160
FREQUENCY. ORTHOGRAPHIC REGULARITY. AND LEXICAL STATUS IN LETTER AND WORD PERCEPTION
I. II . 111. IV .
Language Processing Model ...................................................................... Orthographic Structure and Recognition ........................................................ Orthographic Structure and Conscious Knowledge .......................................... General Discussion ............................................................................. References ............................................................................................
164 166 190
195 199
SELF AND MEMORY Aiiihoiiy
I. II .
G . Greenwnlcl
1901 -1935: Four Positions ...................................................................... 1977 Onward: Diacovery of the Self-Refeience Effect ......................................
202 210
Ill . IV . V.
Contents
vii
Memory When Self Is Involved .................................................................. Theoretical Synthesis ............................................................................... Scope and Importance of the Self System ...................................................... References ............................................................................................
214 223 229 233
CHILDREN’S KNOWLEDGE OF EVENTS: A CAUSAL ANALYSIS OF STORY STRUCTURE Torn Trubaso. Nancy L . Stein. und Lucie R . Johnson I. I1 . 111. IV . V.
Introduction .......................................................................................... Causal Relationships ............................................................................... Causal Understanding in Preschool Children .................................................. Logical Justification: The Meaning of Why and Because ................................... General Conclusions ............................................................................... References ...........................................................................................
237 249 269 274 277 279
Index ..........................................................................................................
283
This Page Intentionally Left Blank
LIST OF CONTRIBUTORS Numbers in parentheses indicate the pages on which the authors' contributions begin
Robert G. Cook, Department of Psychology, University of California, Berkeley, California 94720 ( 5 I ) Anthony G. Greenwald, Department of Psychology, Ohio State University, Columbus, Ohio 43210 (201) Richard J. Harris, Department of Psychology, Kansas State University, Manhattan. Kansas 66506 (81) James E. Jastrzembski,l Wisconsin Research and Development Center for Individualized Schooling, University of Wisconsin, Madison, Wisconsin 53706 (163) Lucie R. Johnson. Department of Psychology, Bethel College, St. Paul, Minnesota 55 I 12 (237) Marvin R. Lamb. Department of Psychology, University of California, Berkeley, California 94720 (51) R. E. Lubow, Department of Psychology, Tel-Aviv University, Ramat-Aviv, Tel-Aviv, Israel ( I ) Peter A. Lucas.? Wisconsin Research and Development Center for Individualized Schooling, University of Wisconsin, Madison, Wisconsin 53706 (163) Dominic W . Massaro,' Wisconsin Research and Development Center for Individualized Schooling, University of Wisconsin, Madison, Wisconsin 53706 (163) Douglas L. Nelson, Department of Psychology, University of South Florida, Tampa, Florida 33620 (129) 'Present address: Needham, Harper & Steers Advertising, Inc., 300 East Wacker Drive. Chicago. Illinois 60601. 'Present address: Department of Psychology. Carnegie-Mellon University, Pittsburgh, Pennsylvania 15213. 'Present address: Program in Expcriiiiental Psychology, University of California, Santa Crux, California 95063.
ix
X
List of Contributors
Donald A. Riley, Department of Psychology, University of California, Berkeley, California 94720 (51) Paul Schnur, Department of Psychology, University of Southern Colorado, Pueblo, Colorado 81001 ( 1 ) Nancy L. Stein, Department of Education, University of Chicago, Chicago, Illinois 60637 (237) Tom Trabasso. Department of Education, University of Chicago, Chicago, Illinois 60637 (237) I. Weiner. Department of Psychology, Tel-Aviv University, Ramat-Aviv, TelAviv, Israel ( 1 )
CONTENTS OF PREVIOUS VOLUMES Volume 1
R. C . Atkinson and R . M . Shiffrin Mediation and Conceptual Behavior Howard K. Kendler and Tracy S . Kendler Aurlior I d e x S u b j e c t Index
Partial Reinforcement Effects on Vigor and Persistence Abram Amsel A Sequential Hypothesis of Instrumental Learning E. J . Capaldi Satiation and Curiosity Harry Fowler A Multicomponent Theory of the Memory Trace Gordon Bower Organization and Memory George Mandler
Volume 3 Stimulus Selection and a “Modified Continuity Theory. ’ ’ Allan R. Wagner Abstraction and the Process of Recognition Michael I . Posner Neo-Noncontinuity Theory Marvin Levine Computer Stimulation of Short-Term Memory: A Component-Decay Model Kenneth R . Laughery Replication Processes in Human Memory and Learning Harley A . Bernbach Experimental Analysis of Learning to Learn Leo Postman Short-Term Memory in Binary Prediction by Children: Some Stochastic Information Processing Models Richard S. Bogartz Author IndexSuhject Index
Author I n d e x 4 i ~ h j e c tI n d n
Volume 2 Incentive Theory and Changes in Reward Frank A. Logan Shift in Activity and the Concept of Persisting Tendency David Birch Human Memory: A Proposed System and Its Control Processes xi
xii
Contents of Previous Volumes
Volume 4 Learned Associations over Long Delays Sam Revusky and John Garcia On the Theory of Interresponse-Time Reinforcement G . S . Reynolds and Alastair McLeod Sequential Choice Behavior Jerome L. Meyers 'The Role of Chunking and Organization in the Process of Recall Neal F. Johnson Organization of Serial Pattern Learning Frank Restle and Eric Brown Author I r i d e . r S u b j e c t lnde.r
Volume 5 Condttionlng and a Decision Theory of Response Evocation G . Robert Grice Short-Term Memory Bennet B . Murdock. Jr. Storage Mechanisms in Recall Murray Gianzer By-Products of Discriminative Learning H. S . Terrace Serial Learning and Dimensional Organization Sheldon M. Ebenholtz FRAN: A Simulation Model of Free Recall John Robert Anderson Author I n d e x S u b j e c t index
Volume 6 Informational Variables in Pavlovian Conditioning Robert A . Rescorla The Operant Conditioning of Central Nervous System Electrical Activity A. H. Black The Avoidance Learning Problem Robert C. Bolles
Mechanism!, of Directed Forgetting William Epsrein Toward il Theory of Reintegrative Memory: Adjective-Noun Phrases Leonard M. Horowit7 and Leon Manelis Elaborative Strategies In Verbal Learning and Memory William E. Monlague Alithur l i i d ~ ~ t - 4 u l ? j1irde.x ~ct
Volume 7 Grainmaticai Word Classes: A Learning Process and Its Simulation George R . Kiss Reaction Time Measurcments in the Study of Memory Processes: Theory and Data John Theios Individual Differences in Cognition: A New Approach to Intelligence Earl Hunt, Nancy Frost. and Clifford Lunneborg Stimulus Encoding Processes in Human Learning and Memory Henry C. Ellis Subproblem Analysis of Discrimination Learning Thomas Tighe Delayed Matching and Short-Term Memory in Monkeys M. R . D'Amato Percentile Reinforcement: Paradigms for Experimental Analysis of Response Shaping John R. Platt Prolonged Rewarding Brain Stimulation J. A. Deutsch Patterned Reinforcement Stewart H. Hulse Aurhor IndexSubjecl Index
Volume 8 Semantic Memory and Psychological Semantics Edward E . Smith, Lance J . Rips, and Edward J . Shoben
Contents of Previous Volumes
Working Memory Alan D. Baddeley and G r a h m Hitch The Role of Adaptation Level in Stimulus Generalization David R. Thomas Recent Developments in Choice Edmund Fantino and Douglas Navarick Reinforcing Properties of Escape from Frustration Aroused in Various Learning Situations Helen B. Daly Conceptual and Neurobiological Issues .in Studies of Treatments Affecting Memory Storage James L . McCaugh and Paul E. Gold The Logic of Memory Representations Endel Tulving and Gordon H. Bower Subject Ititlea
...
Xlll
Repetition and Memory Douglas L Htntrman Toward a Framework for Understanding Learning John D . Bransford and Jeffrey J . Franks Economic Demand Theory and Psychological Studies of Choice Howard Rachlin, Leonard Green, John H. Kagel, and Raymond C . Battalio Self-Punitive Behavior K. Edward Renner and Jeanne B. Tinsley Reward Variables in Instrumental Conditioning: A Theory Roger W . Black Sii6jett
/iid<>.i
Volume 11
Volume 9 Prose Processing Lawrence T . Frase Analysis and Synthesis of Tutorial Dialogues Allan Collins, Eleanor H. Warnock. and Joseph J . Passafiutne On Asking People Questions about What They Are Reading Richard C . Anderson and W. Barry Biddle The Analysis of Sentence Production M. F. Garrett Coding Distinctions and Repetition Effects in Memory Allan Paivio Pavlovian Conditioning and Directed Movement Eliot Hearat A Theory of Context in Discrimination Learning Douglas L. Medin
Levels of Encoding and Retention of Prose D. James Dooling and Robert E . Christiaansen Mind your p ' s and 4 ' s : The Role of Content and Context in Some Uses of And, Or, and I f Samuel Fillenbaum Encoding and Processing of Symbolic Information in Comparative Judgments William P. Banks Memory for Problem Solutions Stephen K. Reed and Jeffrey A. Johnson Hybrid Theory of Classical Conditioning Frank A . Logan Internal Constructions of Spatial Patterns Lloyd R . Peterson, Leslie Rawlings, and Carolyn Cohen Attention and Preattention Howard Egeth Siibject ltidea
Siihject Itidex
Volume 10 Some Functions of Memory in Probability Learning and Choice Behavior W. K . Estes
Volume 12 Experimental Analysis of Imprinting and Its Behavioral Effects Howard S . Hoffman
Contents of Previous Volumes
XIV
Menior) , Temporal Diacriiiiination. and Learned Structure in Behavior Charles P. Shimp The Relation hctween Stimulus Analyzability and Perceived Dimen\ioniil Structure Barbara Burns, Bryan E. Shepp, Dorothy McDonough, and Willa K . Wiener-Ehrlich Mental Conipariwn Kohert S . Moyer and Susan T . Dumais The Simultarieous Acqtiisitirin nt Multiplc Meiiioric\ Benton J . Underwood and Robert A. Malmi The Updating ot Human Memory Kobcrt A H j o r k S/&
ltid?.l
Volume 13 Pavlovian Conditioning and the Mediation ot Behavior J . Bruce Overmier arid Janice A. Lawry A Conditioned Opponent Theory of Pavlovian Conditioning arid Habituation Jonathan Schull Memory Storage Factors Leading to Intantile Amnesia Norman E. Spear Learned Helplessness: All of Us Were Right (and Wrong): Inescapable Shock Has Multiple EIlCct.\
Steven F. Maier and Raymond L. Jackson On the Cognitive Component of Learned Helplessness and Depression Lauren B . Alloy and Manin E. P. Seligman
A General Idearning Theory and Its Application (o Schema Ahstraction John R . Anderson, Piiul J. Kline, and Charles M. Beasley. J r . Siniilai-ity and Order in Memory Robert G Cruwder Stimulus Classification: Partitioning Strategies and Use of Evidence Patrick Rabbitt liiiniecliatc Memory and Discotirse Processing Robert J . Jarvella Sii/?jou I i t i l c i
Volume 14 A Molar Equilibrium Theory o f Learned Performance Wi lliain Timber lake Fish as a Natural Category for People and Pigeons R. J . Herrnbtein and Pctcr A . de Villiers Freedoin of Choice: A Behavioral Analysis A. Charlea Catania A Sketch of an Ecological Metatheory for Theories of Learning Timothy D. Johnston and M. T . Turvey SAM: A Theory o f Probabilistic Search of Associative Memory Jeroeii Ci. W . Raaijmakers and Richard M . Shiffrin Memory-Based Rehearsal Ronald E. Johiison Individual Dil'l'erences i n Free Recall: When Sonic People Remember Better Than Others Marcia Ozier ltl(/c.r
THE PSYCHOLOGY OF LEARNING AND MOTIVATION Advances in Research and Theory
VOLUME 15
This Page Intentionally Left Blank
CONDITIONED ATTENTION THEORY R . E . Lubow u n d l . Weiner TEL-AVIV UNIVERSITY RAMAT-AVIV, TEL-AVIV, ISRAEL
Paul Schnur UNIVERSITY OF SOUTHERN COLORADO PUEBLO, COLORADO I . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.....
........................... A. Predictions Related to Conditioning of Inattention to S I ....................
11. Conditioned Attention Theory of Latent Inhibition
111.
IV.
V.
VI.
B . Predictions Related to the Modulation of Attention to S , by S I -S, Pairings. . . . Evidence for C A T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Relationship of CAT to Other Learning Paradigms . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Habituation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Sensory Preconditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... C. Learned Helplessness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. The Feature Positive Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Classical Conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A . CAT and Conventional Conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Blocking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C . Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. .... References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
I.
1
3 14 14 16
23 25 28 30 33 34 34
31 39 41 42
Introduction
Latent inhibition (LI) has been defined as a decrement in learning performance that results from repeated nonreinforced preexposure of the to-be-conditioned stimulus (Lubow & Moore, 1959). The conventional procedure to obtain LI involves two stages. In the preexposure stage, the to-be-conditioned stimulus is repeatedly presented without being followed by reinforcement. In the acquisition stage, the preexposed stimulus is paired with a reinforcement. When conditioning to the stimulus is IThe preparation of this chapter was supported, in part,by a grant from the Israel National Academy of Sciences to the first author. 1 THE PSYCHOLOGY OF LEARNING AND MOTIVATION, VOL IS
Copyright 0 1981 by Academic Press. Inc. All rights of reproduction in any form reserved. ISBN 0-12-543315-8
2
R. E. Lubow et al.
assessed and cottipared with that in a control group that did not receive previous stimulus preexposure, a marked decrement in learning to the preexposed stimulus is found. The phenomenon, quite robust, is found across a variety of species and learning tasks (for reviews, see Lubow, 1973; Weiss & Brown, 1974). A number of explanations have been advanced to account for the L1 findings (Lubow, 1973), but the phenomenon has remained largely unexplained. Although the description of LI might be subsumed under the general heading of “inhibitory conditioning, it has been convincingly demonstrated that “latent inhibition” is not “conditioned inhibition” (Reiss & Wagner, 1972; Rescorla, 1971; Solomon, Kiney, & Scott, 1978; Solomon, Lohr, & Moore, 1974). The preexposed stimuli are retarded in the subsequent development of both excitatory and inhibitory conditioning and show no evidence of conditioned inhibition in summation test procedures. Moreover, Hearst ( 1 972) and Halgren (1974) demonstrated that stimulus preexposure retards subsequent discrimination learning regardless of whether the preexposed stimulus subsequently serves as an S + or an S- . Such results have encouraged the speculation that nonreinforced preexposure reduces the salience or attention value of the to-be-conditioned stimulus, without altering its associative strength (Carlton & Vogel, 1967; Frey & Sears, 1978; Halgren, 1974; Mackintosh, 1975a; Mellgren & Ost, 1972; Reiss & Wagner, 1972; Rescorla, 1969; Wagner & Rescorla, 1972). For example, Mackintosh (1975a) proposed that the salience of a stimulus, A , decreases if that stimulus does not uniquely predict the occurrence or nonoccurrence of reinforcement. Similarly, Frey and Sears (1978) proposed that the salience of stimulus A depends upon its current predictive value for reinforcement. To then account for the decremental effects of preexposure on subsequent conditioning, it is commonly assumed that preexposure reduces the value of trlphu, a, the stimulus-specific learning rate parameter in the Rescorla- Wagner model of conditioning (Wagner & Rescorla, 1972).2 None of these accounts of latent inhibition specifies the mechanism by which such salience reduction takes place. Yet the need for such specification has been repeatedly acknowledged (Mackintosh , 1975a; Rescorla, ”
:In a recent paper, Pearce and Hall (19x0) proposed a model of classical conditioning, based on changes in the effectiveness of conditioned but not of unconditioned stimuli, that has implications for latent inhibition. Like the models of Mackintosh (1975a) and Frey and Sears (1978), the Pearce and Hall model postulates that preexposure reduces the associability of the stimulus. However, in contrast to Mackintosh (1975) and Frey and Sears (1978). Pearce and Hall (1980) propose a fundamental revison of the Kescorla-Wagner model of classical conditioning. A complete analysis of the Pearce and Hall model, therefore, is beyond the scope of the present chapter.
Conditioned Attention Theory
3
1971; Rescorla & Wagner, 1972). “It is not sufficient to argue that stimuli uncorrelated with reinforcement are not attended to; one needs to know how trial-by-trial events which compose the uncorrelated treatment are processed by the animal so as to generate failure to attend to the CS” (Rescorla & Wagner, 1972, pp. 94-95). In a review of the latent inhibition phenomenon, Lubow (1973) concluded that there was a need for a model that combines attentional and learning constructs. The present article presents such a model, the Conditioned Attention Theory (CAT) of latent inhibition. This theory shares some of the ideas expressed by Mackintosh (1975a) and by Frey and Sears ( 1 978) insofar as it specifies the conditions underlying changes in attention. However, it also proposes a mechanism by which changes in attention may occur. According to CAT, empirical laws of conditioning provide the missing motor for driving the attentional mechanisms. In what follows, the theory will be described and its postulates formally stated. Experimental evidence supporting the theory will be examined, and the applicability of the theory to phenomena such as habituation, sensory preconditioning, learned helplessness, blocking, the feature positive effect, and conditioning itself, will be discussed.
11. Conditioned Attention Theory of Latent Inhibition In general, conditioned attention theory states that nonreinforced preexposure to a stimulus retards subsequent conditioning to that stimulus because during such preexposure the animal leurns not to attend to it. The theory is based on the use of attention as a hypothetical construct with the characteristics of a Pavlovian response and on the specification of reinforcement conditions that modify attention. The assumption that changes in attention to stimuli are a function of reinforcement conditions may be traced to Lashley (1929) and Krechevsky (1932). Likewise, Lawrence (1949) suggested that the “acquired distinctiveness of cues” might be a gradual learning process subject to traditional analysis. In more recent theorizing, changes in attention as a function of reinforcement conditions are emphasized by Mackintosh (1975a) as well as in the “selective attention” theories of Lovejoy (1966), Sutherland and Mackintosh (1 971 ), Trabasso and Bower (1968), and Zeaman and House (1963). Similarly to selective attention theories, CAT treats attention as a response, occurring upon stimulus presentation, the probability of which is increased when it is followed by reinforcement and decreased when it is
4
R. E. Lubow et al.
not reinforced. However, CAT differs from those theories in a number of important respects: the conditions specified for the changes in the attentional response; the mechanism postulated to govern such changes; and the course of these changes with repeated stimulus presentation. More specifically, CAT makes four critical assumptions about changes in the attentional response, R,. First, with repeated stimulus presentation, the attentional response inevitably declines. Second, the decline in RA is conditionable in the Pavlovian sense. Third, under certain conditions of repeated stimulus presentation, R, may be temporarily increased; this increase of attention is also conditionable. Fourth, a certain minimum level of attention to a stimulus is a prerequisite for its entering into effective associations. With an increase in attention to the stimulus, its associability is enhanced. Before developing the theory further and deriving detailed predictions from it, it is worth contrasting the general approach with that of two other theories that explain similar phenomena-Mackintosh ’s (1 9754 theory of attention in associative learning, and Wagner’s (1976) theory of priming in short-term memory. These theories are reviewed not only for the purposes of providing alternative explanations of the L1 phenomenon, but also as points of departure for further elaboration of CAT. Mackintosh (1975a) pointed out that in selective attention theories, ‘‘while differential reinforcement is assumed to increase attention to relevant stimuli, there is no comparable independent mechanism for reducing attention to irrelevant stimuli” (p. 280). Mackintosh further noted that the studies of LI demand such a mechanism, as they provide ample evidence that nonreinforced presentation of a stimulus retards its subsequent conditioning. Current selective attention theories cannot explain why such retardation should take place. According to Mackintosh, increase and decrease of attention to relevant and irrelevant stimuli may be explained by postulating changes in a stimulus-specific learning parameter, a . He provides a symmetrical set of rules to account for opposite and independent changes in a to relevant and irrelevant stimuli. The parameter a is assumed to increase when a stimulus, A, predicts a change in reinforcement, and to decrease when A signals no such change. The conditions specified by CAT for producing changes in attention are quite similar to those of Mackintosh (1975a, 1978). However, as a result of the conditioning mechanism, which is postulated by CAT to underlie the changes in attention, there are substantial differences between the two theories in terms of level of explanation and, more important, in terms of predictions. CAT, for instance, derives specific predictions for the effects of an addition of a second stimulus that is in a conditioning relationship to
Conditioned Attention Theory
5
the target stimulus during preexposure. Furthermore, because of the postulation of that conditioning relationship as being central to changes in attention, a nonmonotonic curve, describing first the rise and then the fall of attention, is deduced. The derivations of these predictions will be described in the following paragraphs. Similarly to Mackintosh’s theory, CAT postulates that when a stimulus, S , , is repeatedly presented without being followed by reinforcement, a decline in the attentional response to the stimulus takes place. However, CAT makes two additional assumptions about this decline. First, the decline in attention is postulated to be a conditioned decrement and therefore to be subject to the empirical laws of conditioning as they relate to such variables as stimulus intensity or intertrial interval. In this theoretical paradigm, the US supporting conditioning of inattention is the consistent absence of any effective stimulus event following S , . Second, it is postulated that the decline in the attentional response is an inevitable consequence of repeated stimulus presentation. This assumption is critical for predicting the course of changes in attention to a stimulus, when it is followed by another stimulus event. According to CAT, when S, is repeatedly followed by the presence of a second stimulus event, S2, the attentional response to S, will be incremented. The rationale for predicting an increment in R A to s, with a small number of S, -S2 pairings is as follows: During early preexposures, S, by itself elicits R A . This is consistent with the large number of studies that show that the magnitude of LI is a function of the number of preexposures (e.g., Lubow, 1965) and that at least 15-20 preexposures are required in order to reliably demonstrate LI (e.g., Ackil & Mellgren, 1968; Domjan & Siegel, 1971). Likewise, S2 would be expected to elicit R A , at least during the early stages of preexposure. Then, just as in normal classical conditioning where the S2, as for example a mild acid solution in the mouth as a US, would elicit salivation, and when paired with an S, , as for example a tone, the tone would come to elicit salivation, the R A to S, should, after a number of pairings, also come to be elicited by S,. But since S, also by itself elicits R A , the total amount of R A to S, is the summation of that elicited by S, alone and that supplied by the SI-S2 conditioning of R, to S,. Thus, the total amount of R A to SI-S2 after n trials is greater than that to S, alone. When R A to S, is incremented over repeated teals as a result of its being paired with S2, it is considered as a conditioned attentional response. For a conditioned increment in R A to S, to occur, S2 must be temporally contiguous to S, and must itself elicit an R A . S2 may be a different stimulus, a response, a conventional US, or, in general, any stimulus change consistently following S, . Conditioning of attention will take
R. E. Lubow
6
et al.
place whether or not S, is able to support conditioning of an overt response to S , , i.e., the elaboration of a traditional CR to S , . However, since S, is not followed by an S3 over repeated trials, it, too, will lose its power to evoke R A and thus cease to maintain R A to S , . Consequently, the initial increase i n R A to S, will be followed by its decline. The magnitude and duration of the S,-supported increment in R A to S, will be a function of the strength of R A elicited by S, , but over repeated S 1-S, pairings the R A to S , will inevitably decline. The theoretical course of R, to S, for repeated paired presentations of a moderate-intensity S,-S, as well as the course of R A to S, alone is illustrated in Fig. 1 . In regard to Fig. 1 , two points should be noted. First, we of course make no claims for the exact shape of the functions, only for the general prediction that for S, there will be some monotonic decline of R A as a function of the number of preexposures, and that for S,-S, there will be an initial increase in R A followed by a subsequent decrease in R A as a function of the number of preexposures, and further, that the increase in RA will occur relatively early in preexposure. Second, Fig. 1 serves to relate R A more precisely to the concepts of attention and inattention as employed in this paper. Attention and inattention are meant to be on a continuum and not to
"'1
I
Fig. I . Magnitude of the theoretical attentional response (R,) as a function of the number of preexposures of S, and the number ot preexposures of S,-S,. Latent inhibition occurs below line A-D, and facilitation occurs above line A-D. The origin (A) and the end point (E) of the S, and S,-S, functions are the same.
Conditioned Attention Theory
7
refer to two separate processes. For purposes of exposition, however, we shall call levels of RA that are higher than the level of RA elicited on first presentation of S , , conditioned attention, and levels that are lower than that initial level, conditioned inattention. Finally, it is necessary to elaborate on the relationship between the theoretical curves for R, to S, during preexposure and the predicted performance during a subsequent learning task where S, will serve as a to-be-associated event. It is simply assumed that RA must be elicited by any event that is to enter into an association before such an association can be made. Thus, in any transfer test, as for instance in going from the preexposure phase to the acquisition phase in the LI paradigm, the rapidity of conditioning to s, when followed by a US will be a direct function of the RA level to S, as carried over from the last trial of preexposure. Some further comparisons of CAT with current theories relating attention and learning are in order. Whereas in Mackintosh’s (1975a) and Frey and Sears’ (1 978) formulations changes in attention to relevant (reinforced) and irrelevant (nonreinforced) stimuli are assumed to be opposite and independent, increasing when a stimulus predicts a change in reinforcement and decreasing when it signals no such change, CAT assumes that decrease and increase in attention to a stimulus are on a single continuum and do not refer to two independent and separate processes. Repeated stimulus presentation is assumed to result in eventual loss of attention to it, whether it predicts a change in reinforcement or does not. One of the major functions of reinforcement is to slow the decline in attention to the repeatedly presented stimulus. This description of the course of changes in attention is critical for predicting transfer of training effects. Since a function of reinforcement, according to CAT, is to slow the (inevitable) decline in attention to the repeatedly presented stimulus that precedes it, the only proper condition against which this function may be evaluated is a repeated presentation of the same stimulus in the absence of reinforcement. Thus, if a stimulus, S, , is followed by another stimulus, S2, and if the associability of S, is assessed in a subsequent transfer test where S, is paired with a third stimulus, S,, the results obtained in the transfer stage must be compared with those for a control group where S, was preexposed alone rather than with results for a group for which S, is novel. Since in the S,-S2 condition, S, was preexposed, a loss of attention will accrue to S1, and therefore it will have a lower associability than a novel S,. However, S, in the S1- S p condition will have a higher associability than S, preexposed alone, since Sp in this condition serves to attenuate the decline in attention to S, . Thus, only by comparison with the S,-alone presentation can the attention-maintaining property of S, become evident. By contrast, in
8
R. E. Lubow et al.
Mackintosh’s (1975a) and Frey and Sears’ (1978) formulations, increase in attention to a reinforced stimulus, and thus its subsequent extraassociability, is evaluated by comparison with a novel stimulus. In addition to studies of LI that indicate that pairing a stimulus with another stimulus attenuates LI but does not increase its associability as compared with its associability when it is novel, there is now evidence that even repeated pairing of a stimulus with reinforcement results in its loss of associability as compared with a novel stimulus, while at the same time increasing the associability of that stimulus as compared with the condition of its repeated presentation alone (Pearce & Hall, 1979). It appears, then, that the changes of attention during repeated stimulus presentation, as formulated by CAT, provide a potent framework for predicting the subsequently observed effects of such presentations. However, as stated earlier, CAT describes more than the general course of changes in attention. It specifies the mechanism whereby changes in attention take place, and thereby is capable of predicting the variations in the course of transfer as a function of such variables as stimulus intensity and intertrial interval. Thus, not only does CAT predict that a nonreinforced stimulus presentation will result in a decline in attention to the stimulus and therefore a decline in its subsequent associability, it also predicts how such a decline will be affected by certain selected stimulus variables such as stimulus intensity or frequency. In general, CAT uses empirical generalizations from data derived for classical conditioning experiments to account for changes in attention. In this sense, conditioning is viewed as a primary process for determining both the direction and the amount of attention. Here it should be emphasized that CAT relies on the operations and empirical laws of classical conditioning and not on any particular theory of conditioning to account for changes in attention. Thus, CAT is not committed to any given theoretical formulation of conditioning to account for the decline in the associability of a repeatedly presented nonreinforced stimulus. Several such formulations (Mackintosh, 1975a, 1978; Wagner, 1976, 1978) are available. According to Wagner, the decline in a of a stimulus takes place because the stimulus itself is predicted by other cues in the environment. In contrast, Mackintosh maintains that a decreases because the stimulus does not predict any event not already predicted by other cues in the environment. At present, no empirical evidence is available to distinguish between these two alternatives. Indeed it is possible that both conditions will have to be combined to define the concept of “absence of reinforcement.” In other words, absence of reinforcement (the condition for producing a decline in RA) might be treated as a stimulus that is predicted by the context, and by itself predicts nothing beyond the context. Stated
Conditioned Attention Theory
9
more simply, the situations “before” and “after” the stimulus are identical, or the stimulus presentation is followed by the reinstatement of contextual cues. Surprisingly, this issue has not been adequately elaborated in the literature. Although we have refrained from adopting a specific theory of classical conditioning, those theoretical issues that confront classical conditioning theorists clearly have implications for CAT. While the emerging view of classical conditioning is that of learning interevent relations (Rescorla, 1975, 1978) or causative relationships among events (Dickinson & Mackintosh, 1978), it is not at all clear how such learning takes place. More specifically, is the evaluation of the relationship between two events, CS and US, based upon comparisons ucross CS-US trials, or within CS-US trials? In the former case, the critical factor is the repeated occurrence of one event (CS) followed consistently by another (US). In the latter, the critical factor is that the repeated occurrence of one event (CS) is preceded by one stimulus configuration (context) and followed by a different stimulus configuration (context US). Since only the end product of repeated stimulus pairings (i.e., elaboration of a response to the CS) is typically assessed in classical conditioning, the above two possibilities may not be differentiable. In a similar vein, during repeated nonreinforced stimulus presentation, it is not clear whether the critical predictive relationship is that of the repeatedly presented stimulus being notfollowed by an unpredicted event or that of being preceded and followed by the same stimulus configuration (context). Although these formulations are not incompatible with CAT, neither are they helpful, since at present they do not convey the rules of learning (governing the acquisitions of) the particular interevent relationships suggested, which indeed must be acquired to account for the subsequent conditioning. Therefore, again, we prefer for the moment not to commit ourselves to either of these alternatives, but rather to adhere to the operational definition of nonreinforced stimulus presentation as exemplified in the procedures of LI. Thus, the absence of an S2 following S , , or the presence of a repeatedly presented S,, is sufficient condition for the decrementing of RAto S, . Again, this decline in R, is to be considered as the conditioning of inattention to S, . The utility of the theory depends, in part, on being able to state under what conditions S, and S2 will elicit RA. In a narrow sense this has already been stated. When either event is novel, it will elicit RA. Furthermore, since RA is a response that is subject to the laws of conditioning, all the variables affecting conditioning will similarly affect RA (many of these are specified in the next section).
+
10
R. E. Lubow et al.
At this point, the problem of circularity must be addressed. How can we, on the one hand, use conditioning to explain changes in attention and then, on the other hand, use changes in attention to explain conditioning? This is justified on the basis that the empirical laws of conditioning and the theoretical statement regarding the rise and fall of attention are not in a one-to-one correspondence. Thus, for a single example, an empirical law of conditioning would state that the strength of a conditioned response is some negatively accelerating function of the number of CS-US pairings. And, although one may find a variety of empirical CR acquisition curves, they all increase with trials and then reach an asymptote. As illustrated in Fig. I , RA always declines after some number of trials, whether paired with a second stimulus (which might well be a US) or not. Thus, although the rules for the acquisition of an overt CR and the rules for the acquisition of RA are the same, the fates of the two are quite different; the former remains asymptotic, whereas the latter inevitably declines. Since level of R, to a target stimulus is postulated to affect its subsequent conditionability, we may make certain predictions on that basis that indeed hold up, whereas predictions from level of CR will not prove valid. Again, the simplest example comes from the LI paradigm, where there is, indeed, no overt CR level during preexposure from which to make predictions, but from which accurate predictions can be drawn from the theoretical conceptions of the course of RA during the preexposure. A more dramatic illustration would come from a study demonstrating that a large number of CS-US pairings will retard subsequent acquisitions using either the same C S , or the same US, or both. Hall and Pearce (1979) have shown just such an effect. Pairing a tone or light with a weak shock for 52 or more trials interfered with the acquisition of conditioned suppression to the tone or light when it was later paired with a strong shock. In general, it should be noted here that, because of the rise and fall of R, in the S,-S, procedure, many transfer studies, whether they be from L1, sensory preconditioning, blocking, or overshadowing, should be sensitive to variations in the number of trials during the acquisition phase (i.e., during the phase for acquisition of blocking, overshadowing, etc.) and in fact should exhibit either positive or negative transfer depending on the number of such acquisition trials. We shall say more about this, particularly in regard to sensory preconditioning, in a later section. This may be one way of reconciling the apparently discrepant results in regard to so-called posttrial episodes (PTE), where, for example, Mackintosh and his colleagues find that a surprising PTE may facilitate conditioning to an otherwise recalcitrant CS, as in unblocking (Dickinson & Mackintosh, 1979; Mackintosh, Bygrave, & Picton, 1977), whereas Wagner and his colleagues find that a surprising PTE may interfere with conditioning (Donegan, Whitlow, Clr Wagner, 1977; Kremer, 1979).
Conditioned Attention Theory
I1
In summary, then, while the same empirical laws of conditioning govern the acquisition of CR and RA, the theoretical course of RA is different from the empirical course of CR. These differences allow a number of significant predictions that cannot otherwise be generated. Thus, the apparent circularity of CAT is broken. Although up to this point the emphasis has been on the similarity and differences between CAT and Mackintosh’s theory of attention, for a complete account of current theorizing about the effects of repeated stimulus presentations on subsequent associability one must look to the influential work of Wagner (1976, 1978). According to Wagner, when a CS is preexposed in the absence of reinforcement, a context-CS association develops so that the surprise value of the CS in that context decreases. After a sufficient number of nonreinforced preexposures, the CS comes to be expected in that same context. Then, just as an expected US is an ineffective reinforcer (Kamin, 1969a, 1969b; Rescorla & Wagner, 1972), so an expected CS is an ineffective cue; i.e., its associability is reduced. The mechanism controlling the loss of associability to a preexposed CS is termed “priming” of short-term memory (STM). Wagner maintains that, when a stimulus is already represented or primed in STM, the amount of processing (or rehearsal) given to additional presentations of that stimulus decreases. One way in which a CS may be primed in STM is by the retrieval of a context-CS association from long-term memory (LTM). During preexposure, therefore, a context-CS association is established and stored in LTM. On subsequent conditioning trials, the context retrieves a representation of the CS from LTM and thus primes STM. Since priming of the CS in STM entails reduced processing of the CS actually presented during conditioning trials, conjoint processing of CS and US events does not occur, and acquisition is retarded (i.e., LI). Although the Wagner model specifies both the rule governing changes in associability and the mechanism for generating these changes, we find its application to LI unconvincing. The model can be questioned both as an account of the loss of associability during preexposure and as an account of the transfer of that loss to interfere with subsequent conditioning. With regard to the loss of associability during preexposure, one may ask how a context-CS association can be learned if priming in STM prevents rehearsal? Surely, the context is already primed in STM when the CS is preexposed, and so conjoint context-CS processing should be vitiated, undermining the learning of that association. On the other hand, one may question the mechanism for explaining the subsequent interference with conditioning during CS-US trials by asking why a primed representation of the CS in STM cannot be processed conjointly with the presented US? As Wagner (1976, 1978) has argued, learning involves the
12
R. E. Lubow
et al.
development of associative connections between representarions of stimulus elements that are jointly active in STM. That is, associations are always formed between representations of events, not the events themselves, so why should it matter whether the events enter the central processor (STM) via sensory channels or via retrieval from LTM? Thus, on conditioning trials, the context retrieves the CS from LTM to prime a CS representation in STM, and the US representation enters STM via the usual sensory channels. In any case, CS and US representations should be processed concurrently in STM, and conditioning should proceed apace, At best, the Wagner model, as applied to the conditioning phase of the LI paradigm, would suggest that the presented CS is redundant. But if the CS is not processed via the usual sensory channel because it is ulready represented in STM via retrieval-generated priming, then conjoint CS-US rehearsal seems assured. Both CAT and the Wagner model postulate an associative mediator to account for reduced associability, but the nature of the mediator is different: In the latter, the mediator is the context within which the repeated stimulus was experienced, whereas in CAT, the mediator is R, level to the stimulus. Like the Wagner model, CAT proposes that learning occurs even when a stimulus is not followed by another event. Unlike Wagner’s model, however, CAT proposes that learning occurs because a stimulus is not followed by another event. CAT maintains that the conditions of nonreinforced preexposure are sufficient for learning not to attend. Critical tests to differentiate between the STM priming theory and CAT have yet to be performed. Nevertheless, CAT appears to be more parsimonious in that it accounts for the phenomenon of LI with considerably fewer postulates than does the STM model, as well as being able to derive a larger number of predictions (see next section). Finally, one might well ask how CAT differs from the various conceptualizations of the orienting response (OR) and its relationship to learning (e.g., Sokolov, 1963) and, in particular, to the effects of habituation of the OR. First of all, unlike the OR, “attention” and “inattention” as used in CAT are conceptual terms inferred from operations and outcomes within the context of a specified theoretical structure. They refer to hypothetical processes that are not identified with specific overt index responses that are contemporaneous with stimulus application. Thus, while presentation of an effective stimulus elicits a number of overt identifiable response changes from the organism, such as GSR, heart-rate changes, and EEG, and while these indices are undoubtedly indications of stimulus-directed attention, none is the attentional response. The initial presence of the attentional response may be inferred from the presence of these physiological indices, but the changes occurring in the magnitude of
Conditioned Attention Theory
13
these responses do not provide a sensitive indication of the modifications in the attention to the stimulus, and, as is well known, different measures of OR are differently modified during the courses of habituation and/or conditioning (e.g., Germana, 1968). Conditioning of attention and inattention is response-independent; i.e., an occurrence or nonoccurrence of an overt response to the presented stimuli does not provide information about attentional conditioning. For example, Krauter (1 973) presented acoustic startle stimuli prior to their use as conditioned stimuli in a conditioned suppression procedure. He found that the decrement in the ability of the stimulus to enter into associative relationship proceeded independently of its ability to elicit startle reflex. Lubow and Siebert (1969) showed that the effects of a repeated stimulus are independent of the effects such presentations have on unconditioned suppression to the stimulus. In an experiment by Domjan and Siege1 (1971), 5 stimulus presentations resulted in elimination of an unconditioned response to the stimulus, whereas 25 presentations were needed to obtain an associability decrement to the stimulus in the subsequent test. Thus, there does not seem to be a direct relationship between any particular measure of the OR to a stimulus and the subsequent associability of that stimulus. In summary, the effects of repeated stimulus presentations on subsequent learning are explained by postulating changes in attention to the preexposed stimulus. These changes in attention are governed by empirical laws of conditioning such that attention to the target stimulus at first increases (conditioned attention) if it is paired with a second stimulus that also elicits an attentional response, and then decreases (conditioned inattention) when the target stimulus is paired with a second stimulus that no longer elicits an attentional response, or if it is not paired with any stimulus. This formulation alllows for a variety of specific predictions, many of which are not easily derivable from other theories of preexposure effects. In general, all those variables that affect classical conditioning are expected to have the same effect on the conditioning of attention and inattention. If, for example, normal conditioning improves with increasing CS intensity, then in the S,-alone preparation the conditioning of inattention should be a positive function of s, intensity during preexposure. The types of predictions can be divided into two classes: The first includes those that follow from the conditioning of inattention to S, when S1 is repeatedly presented by itself. These predictions, on the one hand, simply affirm that such preexposures result in learning, but, on the other hand, provide some counterintuitive predictions, such as the prediction of more LI with a high-intensity preexposed S, than with a low-intensity S, . The second class includes those predictions that concern the modulation of attention through the S,-S2 manipulation.
14
A.
R. E. Lubow et al.
PREDICTIONS RELATEDTO CONDITIONING OF INAI'TENTION TO S ,
1 . LI will be relatively long-lasting, i.e., distinguishable from sensory adaptation or fatigue-like effects. 2. LI will be relatively stimulus-specific and demonstrate a stimulus generalization gradient. 3. LI will be a positive function of number of stimulus preexposures. 4. L1 will be a positive function of the time between stimulus presentations, i.e., the intertrial interval (ITI) during the preexposure session. 5. L1 will be a positive function of stimulus intensity. 6. L1 will be subject to the effects of external inhibition, i.e., will be disrupted by the introduction of an extraneous event following the S,alone presentations. The event may be a different stimulus, a response, an environmental change, etc. In terms of CAT, the presentation of the extraneous event will serve to restore attention to the presented stimulus and thus prevent establishment of inattention. 7. LI will be subject to blocking; i.e., if a stimulus to which conditioned inattention already has been established is then paired with a second stimulus, conditioning of inattention should develop more slowly to the second stimulus than to a stimulus that has not been paired with a stimulus to which conditioned inattention has accrued. 8. LI will be subject to overshadowing; i.e., if two stimuli of different degrees of salience are preexposed in a compound, there should be less conditioning of inattention to the less salient one, as compared with a situation where that stimulus has been preexposed by itself. 9. LI will be subject to extinction and will exhibit spontaneous recovery * 10. LI should be enhanced as a result of prior apparatus preexposure. Since apparatus preexposure facilitates learning, then preexposure to the apparatus in which stimulus preexposure subsequently will take place should facilitate the acquisition of conditioned inattention. RELATEDTO THE MODULATION OF B . PREDICTIONS ATTENTION TO s1 B Y s, -s2 PAIRINGS
11. As shown in Fig. 1, when S, and S2 are paired over trials, LI will be preceded by a phase of relative facilitation. The facilitatory phase, as stated earlier, is accounted for by the conditioned transfer of RA from S2 to S, . The two sources of RA are additive, thus giving S, from the SI-S2 pair more R, than that of S, alone. Since S, also acquires conditioned inattention with repeated trials, the early facilitative effect wanes and is
Conditioned Attention Theory
15
replaced by LI. Specifically, then, an inverted U-shaped function is predicted. 1la. With a low number of S,-S, pairings, subsequent associability of S, will be better than if S, were not preexposed. 1 lb. The improvement in associability will increase with a relatively low number of S,-S, repetitions. 1lc. With a further increase in the number of Sl-S2 pairings, LI will be obtained and its magnitude increased. 12. The temporal relationship between S, and S2 will affect the amount of attenuation of LI. As in normal conditioning, there should be an IS1 function. Since, however, this function differs for different systems (e.g., eyelid conditioning, GSR conditioning, conditioned taste aversion), the exact shape of the function and its interaction with number of preexposures cannot be predicted. 13. The attenuation of LI by the addition of S2 will be reduced by the prior preexposure of S, alone. As in prediction 11 above, this procedure should produce conditioned inattention to S, . Thus, subsequently, when paired with S , , S2 will have a diminished ability to maintain RA to S , . 14. The efficacy of conditioning of attention to S, will be a positive function of S1 intensity. Just as conditioning of inattention should be enhanced with a higher intensity S, when it is not followed by S2 (prediction 5 ) , so conditioning of attention should be enhanced with a highintensity s, when it is followed by s,. Thus, it is postulated that the effectiveness of S, in maintaining attention to S,, and thus attenuating latent inhibition to it, will be a positive function of S, intensity. Such enhanced effectiveness will be reflected in either a longer lasting or higher magnitude LI attenuation phase. In regard to the subsequent phase of the RA curve, CAT does not yet address itself to the expected rate of decline of attention as a function of S1 intensity when paired with a constant S,. Whether the rate of attenuation is faster or is constant with increased S, intensity is an empirical question, and either way it will not affect the basic theory. 15. The predicted effects of S, intensity during SI-S, preexposure on the shape of the R, curve for S, are somewhat complex, but rather interesting. On the one hand, as S2 intensity increases, there should be an increase in the conditioned attentional response to S,. This might be reflected in a conditioned attentional response of larger magnitude as compared with the response to a lower intensity S, , or the peaking of the response might occur earlier. For instance, the strongest response, instead of occurring at the sixth trial as in Fig. 1, might occur at the fourth trial. On the other hand, since S, is also subject to conditioned inattention, it should suffer a more rapid acquisition of such inattention as S, intensity
R. E. L u b w
16
et al.
increases. Assuming that the effects of increases in S2 intensity are to raise the magnitude of the conditioned attentional response to S, , which is then accompanied by an increased rate of acquisition of inattention, a family of predicted curves, as shown in Fig. 2, would be generated. As can be seen from Fig. 2, with a few numbers of preexposure an increasing S, intensity has a relatively facilitatory effect. With a greater number of trials this effect is predicted to reverse. 16. It follows from the postulates of CAT that the facilitation resulting from the conditioning of attention and the LI resulting from conditioned inattention may be separable. If during S,-S, preexposure S, could be qualitatively changed from trial to trial-as, for example, on trial 1 , S, would be a tone, on trial 2, S, would be a light, etc.-then conditioning of attention should be promoted as a result of the maintenance of RAto S2 over a longer number of trials, and likewise there should be a strong attenuation of conditioned inattention to S2 because on each trial S2 is different. Thus, we would predict only the rising portion of the RA-to-S, curve, which would reach some asymptotic value and remain there as long as the quality of S, was changed from trial to trial. 111.
Evidence for CAT
If one accepts that the 16 specific predictions are derivable from the theory, then, to the extent that these predictions are verified, the theory receives support. Two types of evidence can be presented to test the predictions: There are the data from studies that relate directly to the predictions but were not collected specifically for the purpose of testing CAT; and there are the data from those studies that were explicitly designed to test the theory. Evidence supporting the theory will be adduced from both sources. 1 . The LI effect will persist over relatively long intervals between nonreinforced preexposure and the beginning of acquisition. Siege1 (1970), using rabbits and the conditioned eyelid response, found no significant differences in the LI effect as a function of delay, 0 and 24 hr. A 24-hr delay was used by Lantz (1 973) in a conditioned suppression task, and a 48-hr delay was used by Carlton and Vogel (1967) and Lubow and Siebert ( I 969) to obtain LI. Crowell and Anderson (1 972) demonstrated that a 168-hr interval between stimulus preexposure and conditioning did not diminish the preexposure effects. Using a conditioned taste aversion paradigm with rat pups, Rudy and Cheatle (1979) demonstrated an LI effect with a 7-day delay between stimulus preexposure and testing.
17
Conditioned Attention Theory
-
HIGH INTENSITY s 2
M E W INTENSITY
SZ
0 5 10 15 20 25 30 NUMBER OF PREEXPOSURES OF S I - S ~ Fig. 2. Magnitude of the theoretical attentional response R, to S,, when S, is paired with S,, as a function of S, intensity.
2. The LI effect will be stimulus-specific but will also show a stimulus generalization effect. LI studies that have employed between-subject and within-subject designs indicate that the decremental effect of stimulus preexposure is stimulus-specific (Carlton & Vogel, 1967, Experiment 2; Lubow & Moore, 1959; Schnur, 1971; Schnur & Ksir, 1969; Siegel, 1969a). Furthermore, Siege1 (l969b) demonstrated that LI generalizes as a function of similarity of acquisition phase stimulus to the preexposed stimulus. 3. Latent inhibition will be a positive function of the number of stimulus preexposures. A previous review of the literature (Lubow, 1973) strongly supports this prediction. Twelve studies, which used number of preexposures as an independent variable, were examined (Ackil & Mellgren, 1968; Cantor & Cantor, 1966; Cantor & Fenson, 1968; Chacto & Lubow, 1967; Domjan & Siegel, 1971; Grant, Hake, Riopelle & Kostlan, 1951; Lubow, 1965; Lubow, Markman, & Allen, 1968; Mellgren & Ost, 1972; Miller, 1966; Siegel, 1969a; Wolff & Maltzman, 1968). Of 14 different groups from 6 experiments, only 1 group (Cantor & Cantor, 1966) showed an inhibitory effect with fewer than 17 nonreinforced preexposures. On the other hand, within the narrow range of 16-20 nonreinforced preexposures, 5 of 6 groups from 5 different experiments showed inhibitory effects. This consistency is all the more surprising when one considers the diversity of the 12 studies in both species and
18
R. E. Lubow et al.
learning paradigms. Since 1973, several additional studies have been performed with number of preexposures as a parameter. Lubow, Alek, and Arzy (1975) report a reaction-time experiment with children and adults. In Experiment 2, separate groups received 0, 3, 10, 20, or 30 preexposures. For the children there was no difference between the 0and 3-preexposure groups, but a significant increase in reaction times from 3 to 10, 10 to 20, and 20 to 30 preexposures. Lantz (1973) and Szakmary (1977b), using conditioned suppression in rats, and Clarke and Hupka (1974), using Pavlovian conditioning of the nictitating membrane response in rabbits, report the same general results. Lubow, Wagner, and Weiner ( I 979a) preexposed rats to a tone-light compound for 0, 20, 40, or 80 trials; a conditioned suppression test indicated greatest suppression for 0 preexposures, and significantly less suppression for 20 preexposures, with no differences between 20 and 40 preexposures o r between 40 and 80 preexposures. Weiner (1979) examined the curve for acquisition of LI more precisely. This study used the more typical single stimulus preparation, also with conditioned suppression, and 0, 5 , 10, 15, 20, 25, 35, 45, 60, 75, or 90 preexposures. The data from this study indicated quite clearly that LI is a positive increasing function of number of preexposures. The exact shape of the function, no doubt, is dependent on the species, testing paradigm, and stimulus that are employed. 4. The magnitude of LT will be a positive function of the IT1 employed during preexposure. Crowell and Anderson (1972) provided evidence that spaced stimulus presentation during preexposure results in more LI. Lantz (1973) explicitly manipulated the IT1 during preexposure, using 2-, lo-, 30-, and 150-sec IT1 between stimulus presentations. She found that longer ITIs during preexposure produced a greater response decrement. Similar results were obtained by Schnur and Lubow (1976). 5 . LI will be a positive function of the preexposed stimulus intensity. A positive relationship between the learning decrement in conditioned suppression and the intensity of the preexposed stimulus was obtained by Crowell and Anderson (1972) and by Schnur and Lubow (1976). 6. LI will be subject to external inhibition; i.e., introduction of an extraneous event following repeated nonreinforced stimulus presentation will disrupt L1. Lantz (1973) presented a novel light stimulus following 60 tone presentations. Subsequently, the tone was paired with a shock US in a conditioned suppression paradigm. She found that the light interpolation disrupted LI, as compared with that found in a group where no light stimulus was presented prior to conditioning. Lantz suggested that the light stimulus may serve an attentional function, thus reinstating the effectiveness of the repeated stimulus.
Conditioned Attention Theory
19
Rudy, Rosenberg, and Sandell ( 1 977), using taste aversion learning, demonstrated disruption of LI by exposing the animals to novel environmental stimulation prior to conditioning. The authors concluded that the effect of novel external stimuli is to reverse the prior habituation. Rudy et al.'s study may be related to the perceptual learning studies (e.g., Gibson and Walk, 1956; Oswalt, 1972; see also review in Epstein, 1967; Gibson, 1969), where stimulus preexposure results in facilitation of later learning relevant to that stimulus. These findings appear to contradict the findings of LI, where stimulus preexposure results in subsequent learning decrement to the preexposed stimulus. However, as Lubow, Rifkin, and Alek (1976a) demonstrated, the differences between the two paradigms are due to the differences in the preexposure and test environments. Specifically, in LI, testing takes place in the same environment, whereas in perceptual learning studies, testing takes place in a new environment, both as compared with the preexposure environment. These authors, manipulating the relationship between the preexposure and test environments, demonstrated either facilitative or decremental effects of stimulus preexposure. To account for the facilitative effects of new environments, the authors suggested a general arousal construct whereby new environments independently of stimulus novelty produce conditions conducive to learning. In other words, the new environment acts as an external inhibitor in the same way as the light does in the Lantz study. 7. LI should be subject to blocking, (Kamin, 1969a, 1969b). Rudy, Krauter, and Gaffuri (1976) preexposed animals to light alone and then to a light-noise compound alone. The animals were subsequently conditioned to the noise stimulus by pairing it with shock in a conditioned suppression situation. The authors found that light preexposed prior to compound preexposure interfered with development of LI to the noise. This finding was contrary to what was expected by the authors and was considered by them as puzzling. However, this is precisely the blocking effect that CAT predicts. Specifically, prior conditioning of inattention to the light during its nonreinforced presentations blocks conditioning of inattention to the noise when presented with it in a compound and not followed by reinforcement. Thus, Rudy et al. 's results clearly demonstrate that LI is subject to blocking. 8 . LI should be subject to overshadowing. According to CAT, if a compound stimulus made up of two elements of differential salience is preexposed, there should be more LI occurring to the more salient element than to the less salient element. Furthermore, as the number of preexposures of stimulus compound is increased, there should be a greater gain i n the amount of LI to the more salient element than to the
20
R. E. Lubow et al.
less salient one. In addition, if compound preexposure effects and compound acquisition effects are governed by parallel processes, with an increasing number of preexposures both elements should lose their effectiveness. These predictions were confirmed in a study by Lubow er al. ( 1979a). 9. LI should be subject to extinction and spontaneous recovery. When the preexposed stimulus is repeatedly paired with a US, conditioning to that stimulus is obtained; i.e., the CS-US contiguity extinguishes the decremental effects of the preexposed CS. Wickens, Tuber, Nield, and Wickens (1977) provided evidence for spontaneous recovery of the decremental effect. These authors note that no reference can be made to a specific response recovery, as no such identifiable response was acquired during nonreinforced stimulus preexposure, but rather what is recovered is “an attitude or response of indifference or inaction toward the stimuli given during the original noncorrelated experience” (p. 68). The recovery was specific to the stimulus and/or the response used in the noncontingent situation. In terms of CAT, the inattentional response conditioned to the preexposed stimulus, after being extinguished, is recovered with the passage of time. 10. LI should exhibit an apparatus preexposure effect; i.e., the decrement should be greater when the animal is initially exposed to the environment in which stimulus preexposure takes place. At present there are no available data that bear on this prediction. 11. LI should be attenuated by pairing the preexposed stimulus S , with a second stimulus, S,, or any other event, in a conditioning relationship to it. Lubow ef al. (1975, Experiment 2) demonstrated that children preexposed to paired stimuli (S,-S,) showed less of a decrement in reaction time to S, than did children preexposed to S, alone. A more direct test of the prediction was conducted by Lubow, Schnur, and Rifkin (1976b). In two separate experiments groups of rats were given one of four preexposure conditions: ( a ) no preexposure to the to-be-conditioned stimulus; ( b ) preexposure to the to-be-conditioned stimulus, with each preexposure immediately followed by the second stimulus; (c) preexposure to the to-be-conditioned stimulus and to the second stimulus, in an unpaired relationship; and ( d ) preexposure only to the to-be-conditioned stimulus. Conditioned suppression of licking served as the measure of LI. The specific predictions were that the greatest suppression to S, would be exhibited by the non-preexposed group, and the least suppression (i-e., the greatest amount of LI) by the group exposed to the to-be-conditioned stimulus alone. Suppression to S, in the paired preexposure group should
Conditioned Attention Theory
21
be raised by the addition of S, in temporal contiguity with S, . The addition of S, in the absence of temporal contiguity should not have the effect of increasing suppression as compared with the S, alone. Both experiments yielded the same pattern of results. The addition of S2 in a conditioning relationship to S, during preexposure significantly reduced the amount of LI as compared with either S, alone or S, , S, unpaired conditions. Additional evidence that S, paired with S, attenuates the amount of LI is provided in other studies (Mackintosh, 1973; Szakmary, 1977b). Mackintosh (1973, pp. 82-83) reported that 16 preexposures to a 1-min light produces significant LI in a conditioned suppression test, but that the LI was significantly reduced when the preexposed light was compounded with a tone. Rudy et al. (1977) replicated this finding. Likewise, Szakmary (1977b) reported similar results in conditioned suppression of barpressing for food. Greater response suppression was found to s, in a group that first received S, and S2 paired as compared with a group in which S , and S, were explicitly unpaired. In a second study, in which a 0-preexposure group was compared with an S,-S2 group, no significant differences were found, thus suggesting that the difference observed in Experiment 1 was due to the unpairing of S, and S2 (i.e., LI) and that this effect was attenuated by S,-S, pairings. Lubow et al. (1976b) noted the possibility that the effect of S, was produced as a result of omission of S, during acquisition. Such an interpretation would not be supportive of CAT. In a test of this hypothesis, Szakmary (1977b) demonstrated that paired preexposure of S, and S, attenuated LI of S , whether or not S2 followed S , in the test. Thus, LI attenuation is not a product of stimulus change between preexposure and acquisition. In keeping with the idea that an effective LI-attenuating S, may be any event that is in a conditioning relationship to S , there are studies showing that positive or negative intracranial stimulation (Frey , Maisiak, & Dugue, 1976b) and overt instrumental responding controlling S , onset (Lubow er al., 1976) or S , offset (Weiss & Friedman, 1975) can serve to reduce LI to the same degree as the more usual sensory S,. Frey et al. (1 976b, Experiment 2) found that preexposure to a to-beconditioned tone retarded subsequent eyelid conditioning in the rabbit. However, when during preexposure the tone was paired with either a light flash, positive intracranial stimulation, or negative intracranial stimulation, the LI effect was attenuated. As Frey er al. note, the transfer effects were observed even though none of the stimuli following the CS was itself an effective reinforcer. Thus, once again, it is shown that the pres-
,
22
R. E. Lubow et al.
ence of an event contingent upon the preexposed stimulus attenuates the development of associative decrement to that stimulus whether or not the event by itself supports conditioning. Weiss and Friedman ( 1975) compared L1 in three groups of rats. Group RT received 30 S, presentations that could be terminated by a crossing response. Subjects in group Y T were yoked to group RT and received the same number and pattern of S, presentations, but without control. Group NP received 0 S, presentations. Subsequent acquisition of avoidance responding to S, was found to be significantly poorer in the yoked group, as compared with the RT and NP groups, which did not differ from each other. Similar results, using the above procedure, were obtained in the authors’ laboratory (Lubow, Rollnick, & Weiner, 1977). The subjects’ response may also produce S,, in terms of kinesthetic feedback. In keeping with previous notational systems of this type, we shall call such a response-produced stimulus S,. When paired with S,, S2 should also serve to maintain attention and thus reduce the amount of LI compared with that produced by S , alone. Two experiments reported by Lubow et al. ( I 976b, Experiments 3 and 4), support this notion. Each study had three groups. During the preexposure period the active group could produce S, by pressing on a bar. The passive group was yoked to the active one and received S, every time its partner pressed the bar. Subjects of a third group were placed in the apparatus for the same period of time as were subjects of the active and passive groups, but they did not receive S,. All three groups had the bar present. In both studies, there were no differences in the number of bar presses among the three groups during preexposure. Nevertheless, when S1 was paired with shock and tested in a conditioned suppression of licking procedure, the most LI was shown by the passive group. There was a significant reduction in the amount of LI displayed by the active group. A recent study by Hall and Pearce ( I 979, Experiment 2) reported a very dramatic and long-lasting attenuation of LI following S,-S, preexposure. Unlike other S,-S, studies (e.g., Lubow et al., 1976b), their experiment used a very weak electric shock as Sz. Such an S2 is probably more salient than the tones and lights used in other studies. The study is particularly worthy, since it demonstrated that LI can occur when preexposure involves many pairings between a CS and a weak electric shock. This is exactly what is predicted by CAT. That is, the effectiveness of S2 in maintaining attention to S, diminishes with increasing amounts of preexposure. A retardation in subsequent conditioning is therefore expected and found. All the above studies using either a relatively moderate or a large number of S, -S, pairings during preexposure demonstrated, as predicted,
Conditioned Attention Theory
23
that S, attenuates LI, and suggested, by comparison across experiments, that such attenuation may be diminished by increasing the number of such S,-S, pairings. Only two studies have varied the number of S,-S, preexposures; and of course it is specifically this procedure that would allow the full prediction of an inverted U-shaped function to be tested. The LI-attenuating effect of S2 as a function of a number of S,-S, pairings was investigated by Szakmary (1977a, 1978), using 2 , 4 , and 16 pairings. S, attained its optimum LI-attenuating effect after 4 S,-S2 pairings and waned with a more protracted number of pairings. A study by Lubow et al. (1975) preexposed children to 0, 3, 10, 20, or 40 SI-S2 pairings. Three such pairings facilitated subsequent reaction time to S1 as compared with reaction time in the 0-preexposed group. On the other hand, there were no differences between the 0- and 10-preexposed groups, and the 20- and 40- preexposed groups showed significant amounts of LI.3 12. The temporal relationship between S, and S2 (ISI) will affect the amount of LI and will be related to similar functions in classical conditioning. Some evidence regarding the effect of the temporal relationship between S, and S2 during preexposure was obtained in a study by Lubow et uI. (1 975), where 0.5-sec IS1 between S, and S2 appeared more effective in attenuating the LI effect than 5.0-sec ISI. The data describing the exact S,-S2 time interval function remain to be collected. 13. Exposure of S, prior to its pairing with S , should reduce the ability of 5, to attenuate LI. A procedure to accomplish the above purpose was used by Szakmary (1 978). Indeed, he found that such prior exposure of S, reduced its effectiveness, in the subsequent S,-S2 exposure phase, to attenuate LI as measured in a conditioned suppression test. 14 and 15. At present, there are no studies that bear directly on these two predictions.
IV.
Relationship of CAT to Other Learning Paradigms
The evidence summarized above demonstrates that a model combining attentional and learning constructs, CAT, not only can account for the results of latent inhibition studies, but can generate many new and testa3Holland and Forbes (1980) recently reported attenuation of LI as a result of SI-S, preexposure in a flavor-aversion conditioning paradigm. In two experiments, preexposure of flavor elements produced a greater decrement in subsequent compound conditioning than did preexposure of the compound itself.
24
R. E. Lubow et al.
ble predictions. In what follows, the theoretical framework provided by CAT is applied to findings arising from learning paradigms other than latent inhibition. A proposition forwarded by CAT, for which there is considerable empirical support, is that S, in the LI paradigm may be any event correlated with S, . In LI studies the S1 is, by definition, the to-be-conditioned stimulus, i.e., a “neutral” stimulus conventionally employed as a CS. However, no restriction upon the nuture of S, is logically implied by CAT. In fact, S , , just like S2, may be any event, be it a “neutral” stimulus, a response, a conventional US, or an environment. This general conception allows for relations among a set of interchangeable events, including single stimuli (neutral, appetitive, or aversive) , configurations of stimuli (e.g., compounds, environments), and responses. Organisms may learn to associate any of the above events that occur together, and, conversely, they may learn that events occur in isolation, i.e., not associated with consistent consequences (cf. Mackintosh, 1974; Rescorla, 1975). The two kinds of learning produce opposite changes in the hypothetical state of attention, which result in an increase or decrease of target stimulus associability when conditions of reinforcement are changed in the subsequent transfer test. Existing learning paradigms can be described in terms of different configurations of events, S, and S,. For example, when S, is a neutral stimulus and S2 is an absence of an event, the LI, or habituation, paradigm is obtained. When S, is a typical US, such as a shock, and S, is an absence of an event, the US preexposure, or “learned helplessness,” paradigm is established. According to CAT, in both cases conditioning of inattention to S, occurs irrespective of whether it is, in a different paradigm, labeled as CS or a US. When the preexposed shock is uniquely correlated with a response, the US preexposure effect is attenuated. In CAT terms, the response constitutes an effective S,, which maintains attention to US and thus prevents an associability decrement. When both S, and S, are neutral stimuli, the sensory preconditioning paradigm is in effect. Likewise, in the standard classical conditioning paradigm the S, is a conventional US. This “S,-S,” formulation of CAT allows its application to a variety of learning paradigms. This application is admittedly post hoc, yet it provides a common framework for integration of the different learning paradigms and for the generation of some interesting predictions. The next sections will explore the relationship of CAT to the habituation, learned helplessness, sensory preconditioning, blocking, feature positive effect, and classical conditioning paradigms.
Conditioned Attention Theory
A.
25
HABITUATION
The operational similarities between the latent inhibition and habituation paradigms are obvious; both the preexposure phase of LI and the habituation session involve repeated single stimulus presentation. However, studies of habituation investigate the effects of repetitive stimulus presentation on the contemporaneous response elicited by that stimulus, whereas studies of LI investigate the effects of such presentation upon subsequent associability of the stimulus. A popular model of habituation is the dual-process theory (Groves & Thompson, 1970; Thompson, Groves, Teyler, & Roemer, 1973). However, it is difficult for CAT to make contact with that approach. Dual-process theory uses the inferred constructs of habituation and sensitization to account for and predict the strength of unconditioned responses. It focuses exclusively on response habituation and emphasizes the contemporaneous response, whereas LI deals with stimulus associability in a test that is remote from the preexposure session. These major differences make it difficult, at this time, to meaningfully compare the two theories. CAT and dual-process theory may well bear a relationship to each other, particularly when the latter refers to “conditioning of state” and its relatively long term effects. However, Thompson and his associates have never been completely clear as to the relationship between their theory and learning, either as to how conditioning is used as a mechanism of habituation, or how the habituation theory itself can predict other learning phenomena. Nevertheless, on several occasions, they have alluded to such relationships. The dual-process theory of habituation presented here has certain similarities to dual-process theories of learning, particularly those emphasizing central mediation (Rescorla & Solomon, 1967). We have suggested that alteration of central state of excitation is a necessary prerequisite to the establishment of those aspects of behavioral plasticity that result in increased response, and that “central state” is particularly subject to such elementary Pavlovian procedures as temporal conditioning. It is perhaps a small step from this to Rescorla and Solomon’s view that “the concomitance. . . between CRs and instrumental responding is mediated by a common central state, and the changes in that state are subject to the laws of Pavlovian conditioning.” (Groves & Thompson, 1970, p. 444)
Although a number of other models have been proposed for habituation, of interest to us here are only those that employ conditioning principles or otherwise relate habituation and associability. One such theory has been proposed by Stein (1966), who suggested that an inhibitory system is classically conditionable to the onset of the stimulus. Repeated stimulus presentation results in a rapid activation of an inhibitory system that
26
R. E. Lubow et al.
suppresses the activity of an excitatory system, thus eliminating the arousal reaction. Stein’s model has been criticized by Groves and Thompson (1970) on the grounds that his model requires direct stimulus intensity and frequency functions, whereas an inverse relationship between stimulus intensity and frequency is typically obtained in habituation (see also Petrinovitch, 1973). This criticism obviously applies to all theories of habituation employing classical conditioning principles. However, recent studies of habituation have provided evidence that there are separable short- and long-term habituation effects, the latter bearing a direct relationship to stimulus intensity and frequency. Davis and Wagner (1 968, 1969), exposing animals to different stimulus intensities and testing with a common intensity, obtained a direct intensity-rate of habituation relationship, and Davis (1970), manipulating the ISIs used during stimulus presentation and test, found more habituation with long ISIS. The latter study differentiated between two different processes-short-term habituation, measured in intervals up to 1 min, inverse/j>related to the IS1 length, and long-term habituation, measured after 24 hr, directly related to IS1 length. The two studies obtained an inverse intensity-rate of habituation function and an inverse IS1 length-rate of habituation function when comparisons conventionally evaluated in habituation studies were made. Thus, when stimulus preexposure and measurement of habituation are separated in time, creating a type of transfer test, it becomes possible to identify a second process, in which there is a direct relationship between measures of habituation and such variables as stimulus intensity and frequency. Latent inhibition studies, using another type of transfer testsubsequent acquisition of response to the “habituated” stimulusdemonstrate an additional long-term effect of repeated nonreinforced stimulus presentation: a decrement in the associability of the habituated stimulus. Like the long-term response decrement obtained in habituation studies, the associative decrement to the habituated stimuli in LI experiments bears a direcf relation to stimulus intensity and IS1 duration (Lantz, 1973; Schnur & Lubow, 1976). The direct intensity and frequency functions that are obtained in the remote transfer tests of habituation parallel those found in classical conditioning. This parallel suggests a common process underlying both functions. We propose that this common process involves the conditioning of the mediator variable that we have labeled “attention.” Conditioned attention theory, which relies exclusively on conditioning principles to explain the decremental effect of repeated stimulus preexposure on a remote test, may be applied to habituation data as well as to LI data. Unlike CAT, which can reconcile long-term habituation data with LI, a
Conditioned Attention Theory
21
more recent theory proposed by Wagner ( 1 976) attempts to integrate LI with both long- and short-term habituation phenomena. The model and its comparison with CAT were discussed in the theory section. Here it is interesting to note that, in spite of the differences between the two models, the experimental manipulations proposed “as particularly relevant to evaluation” of the STM model are attenuation of habituation by presenting an S2 following each stimulus presentation, and demonstrations that habituation is context-specific. The two predictions also follow directly from CAT (predictions 11 and 6 in Section 111); in fact, the former is a cornerstone of the theory. Thus, postulating conditioning processes during nonreinforced stimulus presentation generates the two main predictions of Wagner’s model. Similarly, it easily accounts for the direct IS1 and intensity functions obtained on remote tests, such functions being typical parametric facts of classical conditioning paradigms (Kimble, 1961; Mackintosh, 1974). Critical tests capable of differentiating between the two theories are difficult to arrange. However, consider a design in which S2 follows S, on each preexposure trial, but in which S, is equiprobable in the absence of S , , i.e., otherwise expected. According to Wagner’s theory, LI should be attenuated because S, serves as a distractor stimulus, denying the limited capacity of STM to S , . According to CAT, however, LI should be obtained because S, is followed by an otherwise expected event and inattention should be conditioned to S1. Such a test has not been made. Nevertheless, in a related design, Mackintosh (1973) and Lubow et al. (1976b, Experiment 2) reported that LI does occur when S , and S, are equiprobable, and occasionally contiguous, during preexposure. It should be emphasized that CAT addresses itself only to long-term effects of habituation, i.e., effects demonstrable under changed test conditions. These effects, either reduced responsivity or associability of the habituated stimulus, are said to reflect conditioned inattention to the stimulus. In other words, CAT is concerned not with the waning of response to the stimulus at the time of its presentation, but with the loss of its functional impact as a result of nonreinforced presentation. The response decrement, which is the focus of habituation studies, is but one effect of the conditioned inattention to the nonreinforced preexposed stimulus. Another effect is the reduced associability that becomes evident in the LI procedure. A similar view has been expressed by Jeffrey (1976): “In place of response habituation, I have referred to the habituation of the salience of the stimulus” (p. 283). Jeffrey’s model of habituation shares with CAT the assumption that the reduction of attention-getting properties of the stimulus
28
R. E. Lubnw et al.
is critical in the process of habituation; his “targeting reflex,” elicited by the stimulus, is a conceptual cousin to the attentional response postulated by CAT. In addition, Jeffrey postulates the development of a general inhibitory process during habituation. B.
SENSORY PRECONDITIONING
The most critical prediction of CAT, for which there has been ample support in the LI literature, is that any R,-eliciting event in the environment, s,, contiguous to the repeatedly presented stimulus, s,,maintains attention to S1 and thus increases its associability in subsequent learning tasks. One learning paradigm that normally employs an S,-S, preexposure phase is that of sensory preconditioning (SPC). [See Seidel (1959) and Thompson (1972) for reviews of the SPC literature.] The usual SPC paradigm involves three stages. First, two apparently neutral stimuli, S1 and S,, are presented in temporal contiguity (S,-S,) over a number of trials. Second, a CR to the S, is established, using a typical US. In the third stage, the CR to S, , which was not directly paired with the US, is measured. The data, in terms of number or magnitude of CRs in the third stage, are compared with results for a control group that has been treated identically in stages two and three, and that had the same amount of preexposure to S, and S2 in stage one, but unpaired. Sensory preconditioning is demonstrated when the experimental group shows more CRs than the control group in stage three. The similarity between the SPC paradigm and the S,-S2 paradigm of LI is clear. In both, there is a preliminary stage of stimulus pairing, the experimental group receiving neutral stimuli in a conditioning relationship (Sl-S,), and the control group receiving the same stimuli unpaired (S,, S,). In both LI and SPC, a subsequent test is used to assess learning that takes place during the first pairing stage. Finally, the results of SPC studies are the same as those found in LI studies. The S,-S, preexposed groups perform better in the test than the S1, S, unpaired groups. Furthermore, as in the LI literature (Szakmary, 1977b), the magnitude of the SPC effect is a function of the number of S,-S2 pairings, with many pairings eliminating the SPC effect (Hoffeld, Kendall, Thompson, & Brogden, 1960). In fact, if one were not aware of the existence of the SPC phenomenon, one would predict this effect from CAT of LI. According to CAT, in the pairing phase, the S, serves to maintain attention to S, , as compared with S, alone or S , , S2 unpaired. The latter two procedures result in the conditioning of inattention to S , . Thus, the facilitative effect obtained in SPC can be explained solely on the basis that the control group, against which SPC is assessed, is an LT group rather than a “zero” group.
Conditioned Attention Theory
29
However, the major question in regard to the SPC phenomenon is whether, during the pairing stage, an association is formed between the two stimuli by virtue of their contingency (Mackintosh, 1974; Pfautz, Donegan, & Wagner, 1978; Thompson, 1972). Although the answer to this question is often claimed to be confirmatory (Brogden, 1939; Mackintosh, 1974; Thompson, 1972), as Pfautz et al. (1978) have pointed out, there have been no data that support this assertion unquestionably. They addressed themselves specifically to this problem, and the results of their study support the contention that an association is formed between the two stimuli in the first phase of the SPC paradigm. Although Pfautz ef (11. (1978) write about differentiating between maintenance of attention (or protection from habituation) and S, -S, association, they did not, in fact, employ the group necessary to substantiate this division. In their study, subjects were trained with two pairs of stimuli, S, paired with S,, and S, paired with S,. In the subsequent conditioning phase, S,, but not S4, was paired with a US. Tests of S, compared with S, revealed better learning for the former group. Since both S, and S3 in the preexposure phase are protected from habituation or the conditioning of inattention, the differences in the subsequent test must be attributed to the s,-US acquisition stage and, therefore, retrospectively to an association between S , and S,. However, this particular paradigm, while providing evidence for an associative effect during the preexposure phase of SPC, provides no evidence for the absence of attention-maintenance by S, and S,. If another condition were added to the above design in which an S5 and S, were preexposed unpaired, and then placed directly into the test situation, CAT would predict even poorer performance to S6 (or S,) than to S,; and, as Szakmary (1977b) has shown, these predicted differences would be quite apart from what one might expect from a generalization decrement effect. Thus, Pfautz et ul. (1978) have shown that there is an association between S and S2,but they failed to show that there was no maintenance of attention to S , by S,. We would argue from CAT that indeed such attention-maintenance did occur, and that it preceded the formation of the S,-S, association. As we shall argue later, the S,-S, contingency, as it serves to maintain attention, is a basic process underlying all traditional learning paradigms. A similar view has been taken by Brogden (1939) and Thompson (1972). The latter author, in line with his theory of habituation, has appealed to a ‘‘hypothetical state of sensitization ” as the fundamental process of SPC and of learning in general. Although we may accept Thompson’s general characterization of SPC as being a manifestation of a hypothetical underlying physiological state of sensitization (and, indeed, the studies that show facilitated behavior,
,
30
R. E. Lubow et al.
often with relatively few S, - S p pairings, fit nicely with this conceptualization), we believe that CAT provides a more adequate theoretical basis for understanding the phenomenon. This conclusion is based on the fact that CAT not only provides a coherent theoretical framework for integrating habituation, SPC, and LI, but also details a specific set of rules for generating many new predictions concerning these phenomena. C.
L E A R N EHELPLLSSNESS D
Animals preexposed to a series of unavoidable and inescapable shocks exhibit marked interference in subsequent avoidence or escape acquisition tasks. Conversely, animals receiving the same amount of escapable shock show no such interference. This phenomenon has been labeled “learned helplessness” (Maier & Seligman, 1976; Maier, Seligman, & Solomon, 1969; Seligman, 1975). Learned helplessness and latent inhibition paradigms both involve noncontingent preexposure to stimuli later to be used in a learning task, and, in both, subsequent learning decrements to the preexposed stimuli are obtained. The apparent similarity between the two phenomena has been acknowledged by a number of authors (Baker, 1976; Mackintosh, 1974; Wickens et al., 1977). “Just as a stimulus which is uncorrelated with changes in reinforcement is only with great difficulty established as a CS for reinforcements during subsequent conditioning, so, when changes in behavior are uncorrelated with changes in reinforcement, any subsequent association between behavior and reinforcement is hard to establish” (Mackintosh, 1974, p. 218). Wickens el al. (1977) conducted a series of experiments to investigate the effects of prior noncontingent experience with eventual CSs and USs upon subsequent acquisition of conditioned response to these stimuli. The authors concluded that if either the to-be-CS or the to-be-US for a later learned CR had occurred during a preexposure session, it would be somewhat less effective in the establishment of a CR, and especially ineffective in the maintenance of a CR over time. The most popular theoretical interpretation of the learned helplessness phenomenon to date is that of Seligman (1 975) and Maier and Seligman ( I 976). According to these authors, during shock preexposure, animals learn that their responding is independent of shock termination, and this learning interferes with the subsequent response-outcome acquisition. The theory emphasizes lack of response-outcome contingency and the development of a cognitive set of uncontrollability, which affects subsequent behavior through emotional, motivational, and associational interference.
Conditioned Attention Theory
31
Within the general formulation of CAT outlined above, the learned helplessness paradigm constitutes an instance of simple stimulus preexposure, i.e., a situation of lack of contingency between a repeatedly presented stimulus (shock) and other events in the environment. Under such circumstances, according to CAT, conditioning of inattention to the stimulus takes place, reducing its associability in a subsequent testing situation. On the other hand, when animals are given the opportunity to respond to the shock during preexposure, no subsequent decremental effects are found. In terms of CAT, the response constitutes an event of consequence in the environment that is correlated with the preexposed stimulus, i.e., an S,. As such, it maintains attention to the shock and serves to attenuate the subsequent learning decrement. It will be recalled that, when animals are given an opportunity to respond to the CS during its preexposure, latent inhibition to that CS also is attenuated (Lubow et ul., 1976b). Similarly to the theory proposed by Maier and Seligman (1976), CAT emphasizes a learning process during the stage of preexposure, and the transfer of this learning to the stage of subsequent acquisition. However, CAT and learned helplessness theory differ in the identification of what is learned and in the specification of which theoretical construct best describes the acquisition of such learning. Whereas learned helplessness theory emphasizes the lack of contingency between response and outcome, and consequently the development of a cognitive set of uncontrollability during shock preexposure, CAT addresses itself to lack of contingency between a repeatedly presented stimulus (shock) and any other event in the environment, and the development of an associative deficit to the preexposed stimulus due to the conditioning of inattention to that stimulus. What is suggested by CAT is that response-outcome contingency is but one instance of a broader concept of contingency among events, and it can be analyzed in terms of general properties of such contingencies, rather than in terms of a controllability-uncontrollability dimension. An additional difference between the two theories is that helplessness theory emphasizes, in addition to the associative interference, general motivational and emotional deficits due to the experience with uncontrollable USs. CAT, however, focuses on primary learning processes during the stage of preexposure to shock, and only on associative deficits as the effects of such preexposure during subsequent learning about the stimulus. This is not to say that emotional and motivational disturbances due to shock preexposure do not exist. Exposure to shock is an aversive situation, eliciting strong emotional responses. However, if lack of contingency among events is emphasized rather than uncontrollability of
32
R. E. Lubow
et
al.
events, the emotional and motivational effects due to the latter become of secondary importance in producing the observed learning decrement. In fact, without appealing to possible negative emotional and motivational after-effects of shock preexposure, CAT would predict more profound decremental effects due to such preexposure as compared with those obtained with CS preexposure. It will be recalled that one of CAT’s postulates states that conditioning of inattention to the preexposed stimulus is a positive function of stimulus intensity. More LI has been obtained to preexposed stimuli of greater intensity as compared with those of lower intensity (Crowell & Anderson, 1972; Schnur & Lubow, 1976). As the preexposed stimulus in the typical learned helplessness paradigm, shock, is an extremely intense one, CAT would predict greater conditioning of inattention to it and thus a profound learning decrement in a subsequent transfer task. The test of the applicability of CAT to learned helplessness is straightforward. If, as the theory maintains, the decremental effects of shock preexposure are due to conditioning of inattention to the shock, then any event in the environment correlated with the shock should attenuate the learned helplessness effect, eveti though it does not terminate shock presentation, i.e., does not establish response-outcome contingency and does not increase controllability. This critical prediction, derived from CAT, has been subjected to several tests in our laboratory. In one such study, five groups of rats were run. In the preexposure phase, three of the groups constituted the traditional “triadic” design. The animals in the fourth group received a short-duration light as an S,. The light appeared immediately upon each shock offset. Termination of the shock was independent of any response that the animal might make. In the fifth group, the same number of lights and shock was presented, but randomly spaced. Both of these groups were yoked to the other three, and all animals subsequently were tested in an FR-2 avoidance task. The results of the experiment confirmed the CAT predictions: Animals that received light contingent upon each shock presentation did not show a decrement in avoidance learning as compared with the nonshock group or with the group for which shock and light were unpaired. In fact, they performed as well as those of the escape group, i.e., the animals that could terminate the shock by responding during the preexposure. Thus, presentation of an event, S p , in a contingent relationship to the preexposed stimulus, shock, prevented “learned helplessness,” even though the animals in this group did not have the opportunity to “control” shock termination (Lubow, Weiner, Rosenblatt, Lindenbaum, & Margolit, 1979b). We have obtained similar results with humans in a learned helplessness experiment (Lubow, Rosenblatt, & Weiner, 1981), using the basic proce-
Conditioned Attention Theory
33
dures of Hiroto (1974) and Hiroto and Seligman (1975). Yoked groups were preexposed to “uncontrollable” noise, “controllable” noise, or no noise. In addition, as in our previous study with rats, one group was presented with yoked ‘‘uncontrollable ” noise whose offset was immediately followed by a light. In the hand-shuttle-avoidance escape-from-noise task, there was no evidence of learned helplessness in the latter group, whereas the triadic groups showed the usual effect-poorer performance in the “uncontrollable” group as compared with that in the “controllable” and “no-shock” groups. Thus the evidence for the effectiveness of S, in reducing learned helplessness, as predicted by CAT, is quite convincing, occurring in two studies, one with rats and the other with humans. We know of no other theory that would have made such a prediction. D.
THEFEATURE POSITIVE EFFECT
Working with pigeons in discrimination learning experiments, Jenkins and Sainsbury (Sainsbury, 1971, 1973) found that discrimination learning proceeds faster when the particular feature distinguishing positive from negative stimuli appears on reinforced as compared with nonreinforced trials. They termed this the “feature positive effect. ” For example, a feature positive discrimination might involve reinforced trials to AX and nonreinforced trials to X alone, the distinguishing feature, A , appearing on reinforced trials. A comparable feature negative discrimination would involve reinforced trials to X and nonreinforced trials to AX, the distinguishing feature, A, appearing on negative trials. The feature positive discrimination develops rapidly, with responding on positive trials and a rapid cessation of responding on negative trials. The feature negative discrimination develops slowly, if at all. The feature positive effect has considerable generality across species, situations, and experimental procedures (for reviews, see Hearst, 1975; 1978; Hearst & Jenkins, 1974). If one presumes that successful discrimination performance depends upon an animal’s discovering the distinguishing feature between positive and negative stimuli, it is possible to andyze the feature positive effect in terms of CAT. When the distinctive feature appears on positive trials, it is followed by an otherwise unexpected event, food delivery for example. This event should maintain attention to the CS and facilitate the discrimination. On the other hand, when the distinctive feature appears on negative trials, it is followed by an otherwise expected event, namely, reinstatement of contextual cues. Under these conditions, attention will not be maintained, conditioned inattention will develop to the CS, and the development of discrimination will be retarded. Thus, CAT suggests that the feature positive effect is similar to LI, in arising from conditioned
R. E. L u h w et at.
34
attention to the positive feature and conditioned inattention to the negative feature. A further aspect of the feature positive effect is relevant to CAT. According to Hearst (1978), in pigeons, “the development of pecking af the feature on the FP discrimination precedes elimination of responding to the negative display. . . For the FN birds, pecks at the feature occur infrequently and soon disuppear, but pecks continue at other parts of the display-purts cornmon to positive unci negative trials ” (p. 57, italics ours). These data are exactly what would be expected if attention to the CS is maintained by the subsequent presentation of the US, and they imply strong support for a model employing both attentional and learning constructs. CAT also predicts that the feature positive effect would be eliminated if on feature negative trials (AX nonreinforced) an otherwise unexpected event were scheduled to follow A X . This, of course, could be any event, as proposed by CAT. Such an event should maintain attention to the AX stimulus and allow discovery of the distinctive feature, thereby facilitating discrimination. Such an experiment has not yet been done, but an experiment by Newrnan (reported in Hearst, 1978) is suggestive. Newman, using humans, was able to eliminate the feature positive effect by suggesting that critical information was contained on incorrect items. According to Hearst (1978), “establishment of such a set would encourage the subjects to jociis their utterition on features to which they apparently do nut utterid very rnuch under normal conditions” (p. 81, italics ours). Hcarst (1978) summarizes a discussion of the feature positive effect as follows: “In animal studies there is an asymmetry in the FP and FN procedures. . . On AX (feature) trials in the FP case, the stimuli immediately preceding and following the trial are different. . . whereas pretrial and post-trial stimuli are the same for AX trials in the FN case. . . In the FN case nonreinforceinent is merely indicated by a return to the IT1 condition and thus no salient new event occurs after a feature trial in the FN case” (p. 8 I , italics ours).
V. A.
CAI
AND
Classical Conditioning
CONVENI‘IONAL CONDITIONING
If conditioning of inattention can be attenuated by using a relatively neutral S, such as tone or a light in a conditioning relationship to S, , then the pairing of S, with a biologically significant event, such as food or
Conditioned Attention Theory
35
shock, should have a much more profound effect on the attenuation of inattention to S, . Such pairing constitutes, of course, the conventional classical conditioning paradigm, the typical final outcome of which is an elaboration of a CR to the CS. In addition to the specific formulations of Sokolov (1963), in regard to the role of the US in the development of the orienting response, a number of other authors have commented on a phase during early CS-US pairing, which is preparatory to, and independent of, the elaboration of the CR to the CS. For instance, according to Kamin (1969a), the reinforcer has a dual role, which is governed by two independent process-the formation of the CS-US association, and the establishment of the CR to the CS. Likewise, Prokasy (1972) proposed two phases to describe conditioning, with different sets of rules governing response probability in each phase. According to Prokasy, it is highly unlikely that the first modification observed in the CR across trials provides the experimenter with much information about CS-US contingency learning. In a similar vein, Bolles (1972) suggested that, during conditioning, animals form responseindependent representations of their environment, which then play an important role in subsequent response selection. Schneiderman ( 1 973) stated that, ‘‘while reinforcement facilitates response emission, organisms learn more than they show us,” and that the US in conditioning is involved in attention. “Conditioning apparently does not occur unless the US focuses the organism’s attention upon the relationship between the CS and the US” (Schneiderman, 1973, p. 27). Wickens et al. (1977), in their analysis of CR acquisition, maintain that a variety of learning activities occur during the pre-CR period and conclude that ‘‘generalized information about contingency relations is one of the factors that is being acquired during the “pre-K” period in the usual conditioning situation” (p. 67). Finally, Maltzman and his colleagues, in discussing classical conditioning of the GSR, argue that such conditioning involves only discrimination and not response differentiation. “It therefore is only the first phase of what may be a more extended learning sequence when a skeletal motor R or a response other than an index of the OR is conditioned” (Maltzman, Weissbluth, & Wolff, 1978, p. 331). Elsewhere, Maltzman (1977) writes about the conditioning of the GSR to words as being “a problem solving discovery process. . . . It is conditioning of a form of attention” (p. 114). The proposition that both response-independent and response-eliciting processes take place during conditioning is directly derivable from the postulates of CAT. According to CAT, if effective conditioning is to take place when a stimulus is followed by another stimulus, the attentional
36
R. E. Lubow
et al.
response to that stimulus, elicited upon its first presentation, must be maintained. It follows that the conventional US in conditioning has a dual function: First, it maintains the attentional response to the stimulus, and second, it establishes the CR to that stimulus. These two functions of the US define two distinct sequential phases-establishment of conditioned attention to the CS, and elaboration of a CR to the CS. Conditioning of attention to the CS, while supported by the US, is not US-specific, i.e., is independent of the nature of the US (e.g., food or shock). The US serves the function of S, i n the S,-S2 paradigm, i.e., maintains attention to the stimulus preceding it; and conditioning of attention to a stimulus takes place whenever it is followed by an otherwise unexpected event, whether or not the event by itself is able to support an overt conditioned response. On the other hand, CR elaboration is US-specific, i.e., depends on the nature of the US employed. In this stage, the animal learns the appropriate response to the CS, which has been established as an event of significance by the prior conditioning of attention. The nature of the US determines what class of responses will be strengthened. Thus, conditioning of attention is viewed as a prerequisite for CR evocation; with conventional conditioning procedures, both maintenance of attention and CR elaboration are supported by the same US. However, following specifically from CAT, conditioning of attention may take place without giving rise to CR elaboration, as demonstrated in S,-S, studies of LI. Whether or not CR elaboration succeeds, conditioning of attention will depend on US efficacy. Conditioning of attention may be established with any type of stimulus in a conditioning relationship to the CS, including “neutral” stimuli such as light and tones, which are apparently ineffective for establishing CRs (Badia & Defran, 1970; Frey el a / . , 1976; Mackintosh, 1974; Schneiderman, 1973). In his discussion of sensory preconditioning, Mackintosh (1974) similarly concludes that “animals [may] learn that one stimulus follows another, but. . . formation of such an association is not sufficient to produce a CR to the first stimulus. This would imply that a motivationally significant UCS is not required for the formation of an association between CS and UCS, but only for the elaboration of a CR to the CS” (p. 96). The function of S,, i.e., maintenance of attention to the CS, may be seen also in conventional conditioning, when the Sz is presented intermittently or concurrently with a conventional US. In a study by Bruner (1965), rabbits that received an airpuff US on 50% of the trials and light presentation on the remaining trials reached a significantly higher overall level of conditioning than did rabbits that did not receive the light presen-
Conditioned Attention Theory
31
tations. Thus, light served as an effective stimulus to maintain attention to the CS and facilitated CR acquisition with a conventional US. W. Kimme1 (1967), using GSR conditioning, found that subjects required to judge US intensity by moving a lever showed a significantly larger GSR conditioning than did subjects conditioned without the judgment task. Apparently, the motor response functioned as S,, which increased attention to the CS. The rules and/or the mechanism governing the transition from attention-conditioning to CR elaboration are not, at present, clear. One approach to this problem has been taken by H . D. Kimmel and Bums (1975), who proposed a two-stage model of conditioning. The first stage is described as a “preadaptive” or preconditioning one, in which chemical couplings of sequentially repeated events take place. The second stage is described as an adaptive utilization of these couplings; it occurs if some environmental adjustment becomes possible by using the formed connections. In terms of Kimmel and Burns’ proposition, it is possible that what determines whether or not conditioning of attention will give rise to performance is whether or not it enables the organism to produce adaptive responses on the basis of the contingencies existing during this stage. B. BLOCKING
In the blocking paradigm, two stimuli, A and X, are presented in a simultaneous compound and paired with a US. If a CR to one of these stimuli, A, has already been established by prior conditioning trials with the same US, the other stimulus, X , will gain only little stimulus control on compound conditioning trials. It is said that A “blocks” the acquisition of stimulus control by X, the “blocked” stimulus. The blocking phenomenon has been extensively documented (Kamin, 1969a), and it has spawned several important lines of theorizing about attention and classical conditioning (Mackintosh, 1975a; Wagner, 1978; Wagner & Rescorla, 1972). It should be of no surprise that CAT owes much to Kamin’s theoretical analysis of blocking. The application of CAT to the blocking phenomenon is straightforward and will be described after a brief review of two other explanations of blocking. According to Rescorla and Wagner (1972), blocking is the result of a failure by the US to reinforce learning. After conditioning stimulus A, the US is totally predicted by the presence of A. Consequently, the effectiveness of the US to reinforce new learning in the presence of A is diminished, and pairings of AX with the US add little or no associative strength to either stimulus on these compound conditioning trials. A somewhat different interpretation has been offered by Mackintosh
38
R. E. Lubuw et al.
(1973, 1975a, 1978). who proposes that blocking is due to the ineffectiveness of the added stimulus, X , rather than to the ineffectiveness of the US. According to Mackintosh, the presence of element X in a compound that signals no change in reinforcement from that expected on the basis of A alone causes a rapid decline in the salience of X. The animal lparns to igwre X. One implication of this view is that blocking, reflecting a learning process, should not be apparent on the first compound conditioning trial, but should develop over trials. Mackintosh (1975b) reported that blocking is not evident on the first compound trial, but develops subsequently. Mackintosh’s analysis of blocking is similar to that given by CAT, though the latter additionally specifies the mechanism by which attention to the added element is diminished. According to CAT, blocking is a special case of latent inhibition. Blocking occurs because the added element, X , is followed by an event, the US, that is otherwise expected on the basis of prior conditioning to A or, in the terminology of CAT, no longer elicits an R,. Consequently, the attentional response to X is not maintained, and with continued trials of compound conditioning, conditioned inattention develops to X. CAT, therefore, also predicts the gradual development of blocking. Moreover, CAT predicts that the blocking effect is not restricted to situations in which S, is a conventional US such as electric shock or food. Blocking should be obtainable in sensory preconditioning and habituation paradigms, as well. And, as described above, the detrimental effects of nonreinforced preexposure on subsequent conditioning, LI, are also subject to blocking. The role of S, in maintaining attention to a CS, and thus enabling CR elaboration to it, is nicely illustrated in the experiments on “unblocking. ” Kamin (1 969a) demonstrated that the delivery of a second shock 5 sec after each compound presentation resulted in significant conditioning to the normally blocked element, even though the double shock did not function as a stronger reinforcer. According to CAT, unblocking should occur if any otherwise unexpected stimulus follows the compound presentation. Such an event should maintain the attentional response to the added element and allow the US to serve its second role of elaborating a CR to the stimulus. This prediction has been confirmed by Gray and Appignanesi ( 1 973), who demonstrated the “unblocking” effect with a brief auditory and visual stimulus that followed each compound trial. This is a striking illustration of the dual role normally played by the US in conventional conditioning: that of maintaining the attentional response to the CS and that of selecting the CR to be elaborated. Dickinson, Hall, and Mackintosh (1 976) have also demonstrated that “unblocking” is due not to an increase in the effective magnitude of reinforcement but to the otherwise unexpected nature of the added event.
Conditioned Attention Theory
39
The role of the otherwise unexpected event, according to Dickinson et al., “was not itself to reinforce conditioning but to enable conditioning to proceed normally to the otherwise ineffective first shock” (p. 321). In terms of CAT, the otherwise unexpected S2 enables conditioning to take place through conditioning of attention to the added CS. “The surprising event does not itself reinforce that conditioning; it serves to maintain attention to a stimulus that would otherwise be ignored and thus enables the reinforcer actually paired with that stimulus. . . to have its normal effect” (p. 322). Dickinson et al. (1976) concluded that “this implies a rather sharp distinction between the associations that a reinforcer may enter into and the change in the strength of conditioned response that it may support” (p. 322). These two implied actions of reinforcement are precisely those postulated by CAT to be operative at the time of conditioning, the first defining conditioning of attention to the CS, and the second referring to the CR elaboration. CAT postulates that the functions are independent and that the former can be supplied by any S2 that elicits an RA. C.
TRANSFER
In the latent inhibition studies, the addition of a second stimulus, S2, following S , , results in positive transfer to a subsequent learning task. This positive transfer is said to be mediated by a central state of attention established to S, , as a result of its pairing with S2. If, as CAT maintains, the US i n the classical conditioning paradigm similarly serves to establish a central state of attention to the CS, then pairing of this CS with another US in a subsequent learning task should result in a positive transfer of learning. A similar proposition has been advanced by Overmier and Bull (1970), and by Fowler, Fago, Dember, and Hochhauser (1973). These authors suggested that pairing a stimulus with a reinforcer endows it with general attention-getting properties, which facilitate its subsequent associability with other reinforcers. In support of such a proposition, there is a large number of transfer of training studies that demonstrate positive transfer (see Dickinson & Pearce, 1977, for a review of the literature). Interestingly, these authors suggested that most of the studies that obtained facilitation used random or unpaired control groups and “consequently, the facilitation might have arisen from the fact that pairing a stimulus with food reduces the amount of latent inhibition that results from simple preexposure of the CS” (p. 700). In spite of the large number of studies demonstrating positive transfer, the available literature on transfer of training (Dickinson & Pearce, 1977) also indicates that numerous studies obtained negative transfer.
40
R. E. Lubow et al.
Scavio ( I 974), for example, obtained negative transfer using control groups in which the CS either was not presented at all or was presented unpaired with shock. Thus, facilitation was not obtained in this study, even in a comparison with a latent inhibition group. These findings would seem to invalidate the prediction of positive transfer derived from CAT. There is, however, another explanation for these data. Effective USs exert powerful motivational performance effects on behavior. The motivational effects are likely to obscure any attentional influences that these USs exert. In support of this idea, it should be noted that most studies that obtain negative transfer use an aversive US during the first stage, whereas studies that obtain positive transfer employ an appetitive US in this stage (Dickinson & Pearce, 1977). In addition, transfer of training studies typically employ USs of opposite motivational affective value in the two stages of the transfer of training experiment. The nature of the interaction between such USs is complex and may further obscure the attentional properties these USs may possess. In spite of these possibly contaminating factors, it should still be possible, according to CAT, to isolate the attentional function of the US, reflected as an increase of CS associability in the second stage. This is because CAT not only states that the US serves to condition attention to the CS, but also specifies that this process takes place prior to CR elaboration. It follows, then, that if the switch from the first CS-US stage to the second is made early in training, positive transfer should be obtained, regardless of the nature of the US. To test this proposition, the number of CS-US pairings in the first stage of a typical transfer of training paradigm should be varied parametrically. Within such a design, CAT predicts that, whatever the direction of transfer (positive or negative) that is obtained with a protracted number of CS-US pairings, a positive transfer will be obtained with a small number of such pairings. Another possible way to obtain a differentiation between the attentional and the CR-elaborating function of the US is to use the same US at different intensities. The low-intensity US should give a positive transfer over a longer number of first-stage S,-S2 pairings as compared with the same US of strong intensity. This would occur because the lower intensity US would require more CS-US pairings before elaborating the overt CR than would the strong US, thus allowing for the attentional response to be detected more readily. To summarize, then, CAT proposes that transfer of learning may be profitably analyzed by postulating a conditioning of attention that is established to the CS as a result of its pairing with reinforcement in the early stages of training and that mediates the increased associability of this CS
Conditioned Attention Theory
41
when it is subsequently paired with other reinforcers. Trapold and Overmier (1972) advocated a similar approach. They, however, postulated reinforcer-specific mediators and stated that ‘‘there even may be as many mediators as there are reinforcers ” (p. 446). Findings supportive of this view have recently been reported by Dickinson and Mackintosh (1979). CAT does not preclude such a possibility. It states, however, that in addition to such possible mediators, which could explain the different directions of transfer (negative vs positive) obtained with different reinforcers, there is a general, reinforcer-independent mediator, which may be described as attentional and which theoretically should be manifest in any transfer of training paradigm, provided that appropriate care is taken not to mask its subsequent effects by the overt conditioned responding elicited in the first stage.
VI.
Conclusion
We have presented in this article a theory to account for the latent inhibition phenomenon in particular, and other stimulus preexposure phenomena in general, with special reference to such paradigms as habituation, learned helplessness, sensory preconditioning, blocking, and the feature positive effect. The theory is based on the application of conditioning principles during the stimulus preexposure phase of a transfer experiment. By considering the absence of a significant or an otherwise unexpected event following the preexposed stimulus as an unconditioned stimulus for conditioned inattention, 16 specific predictions were derived. Empirical support for these predictions was drawn from a series of experiments specifically designed to evaluate the theory, as well as from experiments designed outside the theory’s framework. The conditioned attention theory, constructed to account for latent inhibition, also was shown to be useful in understanding related preexposure phenomena and classical conditioning. The generality of CAT is complemented by the formality with which it is stated and the ensuing testability. Only the use of the word “attention” to describe the locus of effect may, to some, be suspect. In order to preempt the accusation that the concept of attention, as employed in CAT, is devoid of utility and/or is circular, it is necessary to explicitly address that problem. To begin with, we recognize, together with Mostofsky ( 1 970) and others, that attention is indeed a most abused term. We appreciate Thompson’s (1972) claim that “to assert that an animal makes a differential response to one of two stimuli because he selectively attends to that stimulus represents no significant advance over the assertion that God. . . makes it happen” (p. 121).
42
R. E. Lubow
et al.
However, our use of the term attention is purely as an intervening variable in the same sense that Hull used such concepts as Drive or Habit Strength. To this end, we have clearly prescribed the antecedent conditions by which attention is varied, and the subsequent conditions within which that variation is made manifest; that is, we have described, independently, the rules for manipulating and measuring attention. In addition, however, we do use the attentional construct as a short-hand notation for some underlying process that must necessarily coordinate the functional relationship between input and output. That we have chosen to label this coordinator or mediator “attention” is attributable to the attenuating effects of hippocampectomy (Ackil, Mellgren, Halgren, & Frommer, 1969; Solomon, 1977; Solomon & Moore, 1975; Solomon, Nichols, Kiernan, Kamer, & Kaplan, 1979; Weiss, Friedman, & McGregor, 1974) and electrical stimulation of the hippocampus (Salafia, Allan, & Borkouski, 1978) on LI. Similarly, several studies have indicated that scopolamine may interfere with the acquisition of L1 (Carlton, 1969; Oliverio, 1968). Since the effects of these centrally operating drugs and lesions are widely regarded as resulting in a failure to produce attention decrements, and since the same drugs and lesions also prohibit the development of latent inhibition, it is suggested that during nonreinforced preexposure for the normal subject there is a centrally mediated decrement of attention to the preexposed stimulus, and that this decrement is responsible for the subsequent L1 effect. Nevertheless, we are not firmly committed to the identification of the mediator as attentional. Such a commitment would require a number of studies, with converging operational definitions of attention, which would produce data parallel to those obtained in the CAT studies. We have opted, perhaps prematurely, to use the label “attention” because a number of such parallels are already available, and because it is suggestive of areas uf research that we feel may be important for understanding LI. We would be neither offended nor surprised if others, of a different persuasion, insisted on renaming the theory CXT . . . only tongue-tied! REFEKENCES Ackil. J . E. & Mellgren, R. L. Stimulus preexposure and instrumental learning. P.y%orrotnic Scrt.rrct.. 1968, 11, 339. Ackil, J . E . , Mellgren, K. L., Halgren, C . , & Frominer, 6 .P. Ellech of CS preexposure on avoidance learning in rats with hippocampal lesions. Jouvtrul of’cotrpwuriw and Phvsiologicul P.\yc,holo,qy, 1969, 69, 739-747. Badia, P . , & Defran, R. H. Orienting responses and GSR conditioning. Psychological Review, 1970, 77, l71-18I.
Conditioned Attention Theory
43
Baker, A. G . Learned irrelevance and learned helplessness: Rats learn that stimuli, reinforcers and responses are uncorrelated. Journal of Experimental Psychology: Animul Behavior Processes, 1976, 2, 130-141. Bolles, R. C. Reinforcement, expectancy and learning. Psychological Review, 1972, 79, 394-409. Brogden, W. S . Sensory preconditioning. Journul of Experimental Psychology, 1939,25, 323-332. Bruner, A. UCS properties in classical conditioning of the albino rabbit’s nictitating membrane response. Journal of Experimental Psychology, 1965, 69, 186-192. Cantor, G. N . , & Cantor, J. H. Discriminative reaction time in children as related to amount of stimulus familiarization. Journul of Experimenral Child Psychology, 1966, 4, 150-157. Cantor, G. N . , & Fenson, L. New data regarding the relationship of amount of familiarization to the stimulus familiarization effect (SFE). Journal of Experimental Child Psychology, 1968, 6, 167- 173. Carlton, P. L. Brain acetylcholine and inhibition. In J . T. Tapp (Ed.), Reinforcement and behavior. New York: Academic Press, 1969. Carlton, P. L., & Vogel, J . R. Habituation and conditioning. Journal ofComparutive and Physiological Psychology, 1967, 63, 348-351. Chacto, C., & Lubow, R. E. Classical conditioning and latent inhibition in the white rat. Psychonomic Science. 1967, 9, 135-136. Clarke, M . E., & Hupka, R. B. The effects of stimulus duration and frequency of daily preconditioning stimulus exposures on LI in Pavlovian conditioning of the rabbit nictitating membrane response. Bulletin of the Psychonomic Society, 1974, 4, 225-228. Crowell, C. R., & Anderson, D. C. Variations i n intensity, interstimulus interval and interval between CS exposure and conditioning with rats. Journal of Comparative and Physiological Psychology, 1972, 79, 291-298. Davis, M. Effects of interstimulus interval length and variability on startle response habituation in the rat. Journal of Coinpurative and Physiological Psychology, 1970, 72, 177-1 92. Davis, M., & Wagner, A . R. Startle responsiveness after habituation to different intensities of tone. Psychonomic Science, 1968, 12, 337-338. Davis, M . , & Wagner, A. R. Habituation of startle response under incremental sequence of stimulus intensities. Journal of Cotparutive and Physiological Psychology, 1969, 67,486-492. Dickinson, A . , Hall, G., & Mackintosh, N . J . Surprise and the attentuation of blocking. Journul of Experimental Psychology: Animal Behavior Processes, 1976, 2, 3 13-322. Dickinson, A . , & Mackintosh, N . I. Classical conditioning in animals. Annual Review of Psychol1978.28, 587-612. Dickinson, A . & Mackintosh, N. J . Reinforcer specificity in the enhancement of conditioning by posttrial surprise. Journal of Experimental Psychology: Animal Behavior Processes, 1979, 5 , 162- 177. Dickinson, A . , & Pearce, J . M . Inhibitory interactions between appetitive and aversive stimuli. Psychological Bulletin, 1977, 84, 690-71 1 . Domjan, M., & Siegel, S. Conditional suppression following CS preexposure. Psychonomic Science, 1971, 25, 11-12. Donegan, H. H., Whitlow, J . W., Jr., & Wagner, A. R. Post-trial reinstatement of the CS in Pavlovian conditioning: Facilitation or impairment of acquisition as a function of individual differences in responsiveness to the CS. Journul of Experimental Psychology: Animal Behavior Processes, 1977. 3, 357-376. Epstein, W. Vurieties of perceprual experiences. New York: McGraw-Hill, 1967. Fowler, H., Fago, G. C., Dember, E. A., & Hochhauser, M. Signalling and affective functions in Pavlovian Conditioning. Anirnul Learning and Behavior, 1973, 1, 81-89. Frey, P. W . , Maisiak, R., & Dugue, G . Unconditional stimulus characteristics in rabbit eyelid conditioning. Journal of E.xperimental Psychology: Animal Behavior Processes. 1976.2, 175190.
R. E. Lubnw et al.
44
Frey, P. W . , & Sears, R. J . Model of conditioning incorporating the Rescorla-Wagner associative axiom, a dynamic attention process, and a catastrophe rule. P.s,vchological Review, 1978, 85, 321-340.
Germana, J . Response characteristics and the onenting reflex. Jourrtul of Experirnentul P sjichulugy, 1968, 78, 610-616. Gibson, E. J. Principle.\ of perceptual learning arid tlevelopnwm. New York: Appleton, 1969. Gibson, E. J . , & Walk, R. D. The effect of prolonged exposure to visually presented patterns on karning to discrirnirtalc thci11. Journal <$Cowiparative and Ph,x+dogicul Psychology, 1956, 49, 239-242. Grant, D. A., Hake, H. W . , Riopelle, A. J . , & Kostlan, A. Effects of repeated pretesting with conditioned stimulus upon extinction of the conditioned eyelid response to light. Anrerican Journul q / Psychology, 195 I , 54, 247-25 1 . Gray, ’1’. , & Appignanesi, A. A . Compound conditioning elimination of the blocking effect. Leurning uncl Motivurion, 1973, 41, 374-380. Groves, P. M., & Thompson, R. F. Habituation: A dual-process theory. Psychological Review, 1970, 77, 419-450. Halgren, C . R. Latent inhibition in rats: Associative or non-associative. Journul qfCompurutive und P hvsiological Psychology, 1974, 86, 74-78. Hall, G . , & Pearce, J . M . Latent inhibition of a C S during CS-US pairings. Journal ofExperirnentci1 Psychology: Anirnul Behavior Processes. 1979, 5 , 31-42. Hearst. E. Some persistent problems in the analysis of conditioned inhibition. In R . A. Boakes & M. S . Halliday (Eds.), Inhibition and learning. New York: Academic Press, 1Y72. Hearst, E. Pavlovian conditioning and directed movements. The Psychology oJLeurning wid Motivnlion, 1975, 9, 215-262. Hearst, E. Stimulus relationships and feature selection in learning and behavior. In S. H . Hulse, H. Fowler, & W. K. Honig (Eds.), Cognitive processes in unimal behavior. Hillsdale, N.J.: Lawrence Erlbaum, 1978. Hearst, E., & Jenkins, H. M. Sigti-tracking: The stimrrlus-reinforcer reltrtionship und directed udion. Austin: Psychonomic Society, 1974. Hiroto, D. S . Learned helplessness and locus of control. Joirrnul qfExprrirnentu1 Psychology. 1974, 102, 187-193.
Hiroto, D. S , & Seligman, M. E. P. Generality of learned helplessness in man. J~)crrwa/[,f’~ervoncr/ity and Social Psvchology, 1975, 31, 31 1-327. Hoffeld, D. R . . Kendall, S . B., Thompson, R. F . , & Brogden, W. S . Effect of amount of preconditioning training upon the magnitude of sensory preconditioning. Journal of Experi,nental Psychology. 1960, 59, 198-204. Holland, P. C . , & Forbes, D. T . Effects of coinponnd o r element preexposure on compound flavor aversion conditioning. Animu/ Learning and Behavior, 1980, 8, 199-203. Jeffrey, W. A. Habituation as a mechanism for perceptual development. In T. J. Tighe & R. N. Leaton (Eds.). Habitualion: Perspectives from c h i l l develupnrent, animal behavior and neurophysiologv. Hillsdale, N.J.: Lawrence Erlbaum, 1976. Kamin, L. J. Predictahility, surprise, attention and conditioning. In B. A . Campbell & R . M. Church (Eds.), Punivhrnerir and nversive behavior. New York: Appleton, 1969. (a) Kamin, L. J. Selective association and conditioning. In N . 1. Mackintosh & W . K. Honig (Eds.), Funcl~menrulissues in ussociatrve learning. Halifax, Can. : Dalhousie University Press, 1969. (b)
Kimble, G . A . Hilgcrrd arid Marquic’ condirioning arid leurning. New York: Appleton, 1961. Kimmel, H. D., & Burns, R. A . Adaptational aspects of conditioning. In W. K. Estes (Ed.), Handbook of learning trnd coptilive processes (Vol. 2 ) . Conditioning and behavior /henry. Hillsdale, N.J.: Lawrence Erlbaum, 1975.
Conditioned Attention Theory
45
Kimmel, W. Judgements of UCS intensity and diminution of the UCR in classical GSR conditioning. Journal of Experimental Psychology, 1967, 73, 532-543. Krauter, E. E. Habituation utid latent inhibition. Unpublished doctoral dissertation, University of Rochester, 1973. Krechevsky, I . “Hypotheses” in rats. Psychological Review, 1932, 39, 516-532. Kremer. E. F. Effect of posttrial episodes on conditioning in compound conditioned stimuli. Journal of Experimental Psychology: Animal Behavior Processes, 1979, 5 , 130-141. Lantz, A. E. Effect of number of trials, interstimulus interval, and dishabituation during CS habituation on subsequent conditioning in a CER paradigm. Animal Learning and Behavior, 1973, l , 273-277. Lashley, K . S. Brain mechanisms and intelligence: A quantitative study of injuries to the brain. Chicago: University of Chicago Press, 1929. Lawrence, D. H. Acquired distinctiveness of cues: I . Transfer between discriminations on the basis of familiarity with the stimulus. Journal of Experinienral Psychology, 1949, 39, 770-784. Lovejoy, E. Analysis of the overlearning reversal effect. Psychological Review, 1966, 73, 87-103. Lubow, R. E. Latent inhibition: Effects of frequency of non-reinforced preexposure to the CS. Journal of Comparative and Physiological Psychology, 1965, 60, 454-455. Lubow, R . E. Latent inhibition. Psychological Bulletin, 1973, 79, 398-407. Lubow, R. E., Alek, M., & Arzy, J . Behavioral decrement following stimulus preexposure: Effects of number of preexposures, presence of a second stimulus, and interstimulus interval in children and adults. Journal of Experimental Psychology: Animal Behavior Processes, 1975, 104, 178188. Lubow, R. E., Markman. R . E . , d;- Allen, J. Latent inhibition and conditioning of the rabbit pinna response. Journal 01Comparative and Physiological Psychology, 1968, 66, 688-694. Lubow, R. E., & Moore, A . U . Latent inhibition: The effect of non-reinforced preexposure to the conditioned stimulus. Journal of Cornparative and Phwiological Psychology, 1959, 53, 415419. Lubow, R . E . , Rifiin, B . , & Alek, M. The context effect: The relationship between stimulus preexposure and environmental preexposure determines subsequent learning. Journal of Experimental Psychology: Animal Behavior Processes, 1976, 2, 38-47. (a) Lubow, R. E., Rollnick, A . , & Weiner, I . Stimulus control during preexposure attenuates latent inhibition. Unpublished manuscript, 1977. (Available from authors.) Lubow, R. E . , Rosenblatt, R., & Weiner, 1. The confounding of controllability in the triadic design for demonstrating learned helplessness. Journal ofpersonality and Sociul Psychology, 1981, in press. Lubow, R . E., Schnur, P., & Rifkin, B. Latent inhibition and conditioned attention theory. Journul of Experimental Psychology: Aniniul Behavior Processes, 1976, 2, 163- 174. (b) Lubow, R. E . , & Siebert, L. Latent inhibition within the CER paradigm. Journal of Comparative and Physiologicul Psychology, 1969.68, 136- 138, Lubow, R. E . , Wagner, M., & Weiner, I . Tests of the conditioned attention theory of latent inhibition: Stirnulus cornpound preexposure of two elements differing in salience. Unpublished manuscript, 1979. (a) (Available from authors.) Lubow, R. E., Weiner, I., Rosenblatt, R., Lindenbaum, U . , & Margolit, H. The confounding of controllabilitv in the triadic design.for dernonstrafing learned helplessness. Unpublished manuscript, 1979. (b) (Available from authors.) Mackintosh, N . J . Stimulus selection: Learning to ignore stimuli that predict no change in reinforcement. In R. A . Hinde & J . S. Hinde (Eds.). Constraints on learning. New York: Academic Press, 1973. Mackintosh, N. J . The psvchologv qf animal learning. New York: Academic Press, 1974.
R. E. Lubow et al.
46
Mackintosh, N . J . A theory of attention: Variations in the associability of stimuli with reinforcement. P ~ y c J ~ o l o g i Review, c~~I 1975, 82, 276-298. (a) Mackintosh, N . J . Blocking of conditioned suppression: Role of the first compound trial. J o u r i ~ duf’ Experiiiientcrl P ~ h / ( ~ f Ai iyi i :i r i d Rehni?or Processes. 1975, 1, 335-345 (b). Mackintosh, N . J , Cognitive o r associative theories of conditioning: Iinplicatioiis of an analysis of blocking. In S. H. Hulse, H. Fowler, & W. K . Honig (Eds.), (’ogriitive processes iri ur~irrral hrhtr~~ior. Hillsdale. N.J.: Lawrence Erlbaum, 1978. Mackintosh, N . J . , Bygrave, D. J . , & Picton, B. M . B. Locus of the effect of a surprising reinforcer i n the attenuation of hlocking. Quurterly Journal of Experirnenrcrl Psyvc.holoh.y, 1977, 29, 327336. Maier, S. F., & Seligman, M. E. P. Learned helplessness: Theory and evidence. Jounra/ of Experir t i ~ n r r r lpsycho log^: General. 1976, 1, 3-46. Maier, S . F., Seligman, M . E. P . , & Solomon. R. L. Pavlovian tear conditioning and learned helplessness: Effects on escape and avoidance behavior of (a) the CS-US contingency and (b) the independence of the US and voluntary responding. I n B. A. Campbell & R . M. Church (Eds.), Pu~tI,s/~rrieii/ mid aversive hehavior. New York: Appleton, 1Y69. Maltzman, I . Orienting in classical conditioning and generalization of the galvanic skin response to words: An overview. Jouriitrl of’ Experir~eii~al Psychology: Gerirral, 1977, 106, 1 I I - I 19. Maltzman, I., Wcisnhluth, S . , & Wolff, C . Habituation of orienting reflexes in repeated GSR seniantic conditioning wssions. .lourrial of Experrrnerital Ps,vchology: Geiieral, 1978, 107, 309-333. Mellgren, R. L . , & Ost, J . W . P. Discriminative stimulus preexposure and the learning of an operant discrimination. Jourt~ulof Cornpurutive untl Phy.siologic,ul P 5yi’kJ/r)gy, 1972, 77, 179- 187. Miller, F. D. 7 % ejfec.r,! ~ c~’difereiiriu/ urtfou~f/.s ofsrirrlu/itsfiri~ii/iiirircrtiori on t / i s c r i r ~ ~ i ~ i t r reaction /ive ~ i r wiri r,hildrcri. Unpublished doctoral dissertation, University of Iowa, 1966. Mostofsky, 0 I . The semantics of attention. In D. 1. Mostofsky (FA.). At~erition:C<Jrl/P/ffp<JrCIr! theory and ~ O / ] ~ . Y ~ New . Y . York: Appleton, 1970. Oliverio, A. Effects of scopolamine on avoidance conditioning and habituation of mice. Psychophar~tr~icolog~a, 1968, 12, 214-216. Oswalt, R. M. The relationship between level of visual patlern ditTiculty during rearing and subsequent discrimination in rats. Jaitriicil of’ Coinpcrrcitive and Physiological Psychology, 1972. 81, 122-125. Overmier, J. B., & Bull, J . A., 111. Influences of appetitive Pavlovian conditioning upon avoidance hehaviur. I n J . H . Reynicrsc (Ed.). C u r r u n / issues i r i o r r i i r i r r l letlriiirix. Lincoln: University ol‘ Nebraska Press, 1970. Pcarce, J . M . , & Hall. G . Loss of associability by a compound stimulus comprising excitatory and inhibitory elements. Jourriul of’Experimentul P svchologs: Aiiirntrl Behavior Processes, 1979, 5 , 19-30.
Pearce, J . M . , & Hall. G A model of Pavlovian learning: Variations in the effectiveness of conditioned but not or Unconditioned stimuli. Psru~hologicdReview. 1980, 87, 532-552. Petrinovitch, L. A species-meaningful analysis of habituation. In H . V . S . Peeke & M. J . H e n (Eds.), Hahi/rrcrrion ( v o l . I ) . Behaviorul sricdic~s.New York: Academic Press, 1973. Pfautz, P. L.. Donegan, N . H . , & Wagner, A. R. Sensory preconditioning versus protection from habituation. J o u r i i d qf Enperirnenitrl Psvr-hology; Ariirrwl Behuvior Proc~e.r.ve,s,197X, 4. 286295. Prokasy, W. F. Developments with the two-phase model applied to human eyelid conditioning. I n A. It: Curreirr theon arid research. New H. Black & W . F. Prokasy (Eds.), C/assica/cmdifiorzi~ig York: Appleton, 1972. Reiss, S., & Wagner, A. R . C S habituation produces a “latent inhibition” effect but no active “conditioned inhibition”. Leurnirig and Motivotiorr, 1972. 3, 237-245.
Conditioned Attention Theory
47
Rescorla, R . A . Pavlovian conditioned inhibition. Psycho/ogicu/ Bulletin, 1969, 72, 77-94. Rescorla, R . A . Summation and retardation tests of latent inhibition. Journd of Comparutive and Physiological Psvchology, 197 I , 75, 77-8 1 . Rescorla. R. A . Pavlovian excitatory and inhibitory conditioning. In W. K. Estes (Ed.), Handbook of /earning ut7d cognitive processe.\ (Val. 2 ) . Conditioning and behavior t h e o p . Hillsdale, N.J.: Lawrence Erlbaum, 1975. Rescorla, R. A. Some implications of a cognitive perspective on Pavlovian conditioning. In S. H. Hulse, H. Fowler, & W. K . Honig (Eds.), Coxnirive processes in unimul behavior. Hillsdale, N . J . : Lawrence Erlbaum, 1978. Rescorla, R . A . , & Solomon, R. L. Two-process learning theory: Relationships between Pavlovian conditioning and instrumental learning. Psychological Review, 1967, 74, 151-182. Rescorla, R . A., & Wagner, A. R . A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classic~alconditioning I I : Current research and theory. New York: Appleton, 1972. Rudy, J . W . , & Cheatle, M. D. Ontogeny of associative learning: Acquisition of odor aversions by neonatal rats. In N. E. Spear & B. A . Campbell (Eds.), Ontogeny of learning and memory. Hillsdale, N.J.: Lawrence Erlbaum, 1979. Rudy, J . W . , Krauter, E. E., & Gaffun, A. Attenuation of the latent inhibition effect by prior exposure to another stimulus. Journal uf Experimental Psychology: Animal Behavior Processes, 1976. 2, 235-247. Rudy, J. W., Rosenberg, L., & Sandell, J . H . Disruption of a taste familiarity effect by novel exteroceptive stimulation. Journal of Experiments/ Psychology: Animal Behavior Processes, 1977, 3, 26-36. Sainsbury, R. S . Effect of proximity of elements on the feature-positive effect. Journal of Experimental Analysis of Behavior. 1971, 16, 315-325. Sainsbury, R. S . Discrimination learning utilizing positive or negative cues. Canadian Journal of Psychology, 1973, 27, 46-57. Salafia, W. R . , Allan, A. M., & Borkouski, G. J . Attention of latent inhibition by electrical stimulation of the hippocampus. Paper presented at the meeting of the Psychonomic Society, San Antonio, November 1978. Scavio, M . , Jr. Classical-classical transfer effects of prior aversive conditioning upon appetitive conditioning in rabbits. Journal of Comparative and Physiological Psychology, 1974, 86, 107-1 15. Schneiderman, N . Classiccd (Pavlovian) cmzditionitrg. Monistown, N.J.: General Leaning Press, 1973. Schnur, P. Selective attention: Effect of element preexposure on compound conditioning in rats. Journul of Comparative and Physiological Psychology, 1971, 76, 123-130. Schnur, P., & Ksir, C. J . Latent inhibition in human eyelid conditioning. Journal of Experimental Psychology, 1969, 80, 388-389. Schnur, P., & Lubow, R. E . Tests of the conditioned attention theory of latent inhibition: The effects of IT1 and CS intensity during preexposure. Learning and Morivation, 1976, 7, 540-550. Seidel, R. J . A review of sensory preconditioning. Psychological Bulletin, 1959, 56, 58-73. Seligman, M. E. P. Helplessness: On depression, development und deuth . San Francisco: Freeman, 1975, Siegel, S . Effect of CS habituation on eyelid conditioning. Journal of Comparutive and Phpsiological Psychology, 1969, 68, 245-248. (a) Siegel, S . Generalization of latent inhibition. Journal <$Comparative and Physiological Psychology, 1969, 69, 157-159. (b) Siegel, S. Retention of latent inhibition. Psychonomic Science, 1970, 20, 161-162. Sokolov, Y .N. Perception and the conditioned reflex. Oxford: Pergamon, 1963.
R. E. Lubow et al.
48
Solomon, P. R. Role of the hippocampus in blocking and conditioned inhibition of the rabbit's nictitating membrane response. Journal of Comparative and Physiological Psychology. 1977, 91, 407-417. Solomon, P. R . , Kiney, C. A., & Scott, D. S. Disruption of latent inhibition following systemic administration of parachlorophenylalanine (PCPA). Physiology of Behavior. 1978, 20, 265271.
Solomon, P. R., Lnhr, C. A . , & Moore, J . W. Latent inhibition of the rabbit's nictitating response: Summation tests for active inhibition as a function of a number of CS preexposures. Bulletin of the Psychonurnic Societv, 1974,4, 557-559. Solomon, P. R., & Moore, J. W. Latent inhibition and stimulus generalization of the classically conditioned nictitating membrane response in rabbits (0ryctolngu.s runiculus) following dorsal hippocampal ablation. Journal of Comparative and Physiological Psvchology, 1975, 89, 1192-1203. Solomon, P. R., Nichols, G. L., Kiernan, J. M., Kamer, R. S . , & Kaplan, L. J. Dijferenrialeffects of median arid dorsal raphe lesions in the rat: Latent inhibition and septo-hippoearnpal serotonin levels. Unpublished manuscript, 1979. (Available from authors.) Stein, L. Habituation and stimulus novelty: A model based on classical conditioning. Psychological Review, 1966, 73, 352-356. Sutherland, N . S., & Mackintosh, N. J. Mechanismr of aniniul iliscrirninution learn in^. New York: Academic Press, 1971. Szakmary, G. A. Larent inhibition with sequentially preexposecl stimuli. Paper presented at the meeting of the Psychonomic Society, Washington, D.C., November 1977. (a) Szakmary, G. A . A note regarding conditioned attention theory. Bulletin of the Psychonomic Society, 1977, 9, 142-144. (b) Szakmary, G. A. Lotent inhibition with sequential1.v preexposed .stimuli: More clatu. Paper presented at the meeting of the Psychonomic Society. San Antonio, November 1978. Thompson, R . F. Sensory preconditioning. In R. F. Thompson & M. R. Voss (Eds.) Topics in learning and performance. New York: Academic Press, 1972. Thompson, R. F., Groves. P. M . , Teyler, T. J., & Roemer, R. A. A dual-process theory of habituation: Theory and behavior. In H. V . S. Peeke & M. J. H e n (Eds.), Hahituariori (Vol. I). Behaviorat studies. New York: Academic Press, 1973. Trabasso, T . , (L Rower. G . H . Attention in learning: T h e o n and research. New York: Wiley, 1968. Trapold, M. A,, & Overmier, J . B. The second learning process in instrumental learning. In A. H. Black & W. F. Prokasy (Eds.), C.'lussical conditioning 11: Current theor?, and research. New York: Appleton, 1972. Wagner, A. R . Priming in STM: An information processing mechanism for self-generated or retrieval-generated depression in performance. In T. J. Tighe & R. N . Leaton (Eds.), HabituaHillsdale, tion: Perspectives from child development, animal behavior and tieurophysiolr~~y. N.J.: Lawrence Erlbaum, 1976. Wagner, A . R. Expectancies and the priming of STM. In S . H. Hulse, H. Fowler, & W. K . Honig (Eds.), Cognitive processes in animal behavior. Hillsdale, N.J.: Lawrence Erlbaum, 1978. Wagner, A. R., & Rescorla, R. A. Inhibition in Pavlovian conditioning: Application of a theory. In R. A. Boakes & M. S. Halliday (Eds.), Inhibition and learning. New York: Academic Press, 1972. Weiner, I. The effects of arousal on latent inhihition: Tests of the conditioned attention theory. Unpublished master's thesis, Tel-Aviv University, 1979. Weiss, K. R., & Brown, B. L. Latent inhihition: A review and a new hypothesis. Actu Neurohiolugiae Experimentalis, 1974, 34, 301 -3 16. Weiss, K. R., & Friedman, R. Stimulus controllability and the latent inhibition effect. Acta Neurohiologiae Experimentalis, 1975, 35, 241-254.
Conditioned Attention Theory
49
Weiss, K. R . , Friedman, R . , & McCregor. S. Effects of septa1 lesions on latent inhibition and habituation of the orienting response in rats. Acta Neurohiologiae Experirnentaliu, 1974, 34, 491-504. Wickens. D. D.. Tuber, D. S . , Nield, A . F., & Wickens. C. Memory for the conditioned response: The effects of potential interference induced before and after original conditioning. Journal of Experirnenlal Psychology: General, 1977, 106, 47-70. Wolff, C., & Maltzman, 1. Conditioned orienting reflex and amount of preconditioned habituation. Proceedings of the American Psvchological Association, 1968, 3, 129- 130. Zeaman, D., & House, 6 . J . The role of attention in retardate discrimination learning. In N . R. Ellis (Ed.), Handhook of’ mental deficiency: Psychological theory und research. New York: McCraw-Hill, 1963.
This Page Intentionally Left Blank
A CLASSIFICATION AND ANALYSIS O F SHORT-TERM RETENTION CODES IN PIGEONS] Donald A . Riley, Robert G . Cook, and Marvin R. Lamb UNIVERSITY OF CALIFORNIA BERKELEY. CALIFORNIA
I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. Designs for the Study of Event Memory in Animals . . . . . . . . . . . . . . . . . . . . . . . . . . 111. A Classification of Codes . . . . . . . . . . . . . . . . . . . . . ....................... A. Chaining Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Test Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. Sample Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. Codes and Their Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IV. Evidence for Different Codes.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Evidence for Chaining Codes.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Evidence for Test Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. Evidence for Sample Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V. Stimulus Organization and Coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Stimulus Organization Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VI. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51 53 55 55 56 56 57 58 58 61 63 68 69 75 76 77
I. Introduction One of the most persistent questions in comparative psychology asks what is involved when behavior is controlled by stimuli that the animal has experienced at one time but that are no longer present. One possibility is that animals maintain an internal representation of the stimulus events they experience. The above-chance performance observed on the first trial of the day in a well-learned visual discrimination task requires that some attribute of the nominal stimulus be represented in the animal’s long-term store of knowledge about the events it has experienced during training. Thus, the facts of long-term memory as exhibited in associative ’Support during the writing of this chapter was provided by National Institutes of Mental Health Research Service Award No. 1 T32 MH15860-01 from the National Institute of Mental Health and by National Science Foundation Grant No. BNS7908839 to D. A. Riley. 51 THE PSYCHOLOGY OF LEARNING AND MOTIVATION, VOL 15
Copyright 0 1981 by Academic Press, Inc. All rights of reproduction in any form resewed.
ISBN 0-12-543315-8
52
Donald A. Riley et al.
learning studies seem to demand that animals form internal representations of stimuli, at least when these events are repeated over many trials. In the present article we are concerned with the types of codes that allow animals to bridge the gap between a single event and a subsequent retention test for that event. We define the term “code” as that transformation of a stimulus event that allows it to control behavior at a later time. A memory trace is an example of a such a transformation. In a conditional discrimination task such as delayed matching-to-sample, it is often assumed that the content of the code consists of visual attributes of the sample stimulus (Maki, Riley, & k i t h , 1976; Roberts & Grant, 1976; Ruggerio & Flagg, 1976). In most cases, however, we do not know the content of the code. Our interest in the content of codes stems from research conducted in our laboratory with the matching-to-sample task. In the initial work, pigeons’ matching of element samples consisting of one of two colors or one of two line orientations was better than matching of compound samples consisting of both one of the colors and one of the line orientations. In addition, performance with both element and compound samples increased with increases in sample duration (Maki & Leith, 1973; Maki & k u i n , 1972). The interpretation of these results was that the amount of information per unit time that can be encoded by pigeons is limited. Thus, an increase in information load (element vs. compound samples) or a decrease in processing time (decreasing sample duration) reduces the amount of sample information per element available to the animal to guide choice behavior at the time of test. This has been designated the information overload hypothesis. The results of these and subsequent selective and divided attention experiments implied that the pigeons processed information from the two dimensions of compound samples separately (Riley & Roitblat, 1978). More recently, Lamb and Riley (1981) found that when compound sample stimuli are presented in a unified form, with color and line angle information presented as colored lines, matching performance was better than it was in compound conditions where color and line were separated in space. Performance with unified compound samples was, in fact, about equal to performance with element samples. Our interpretation of this result is that, unlike separated compounds, the elements of unified compounds are stored together in a single code. This interpretation has two implications, which we shall examine in the present article. The first implication is that pigeons code information about sample attributes. This possibility would hardly strike a student of human shortterm memory as surprising, but recent developments in the animal literature suggest that the content of the code, rather than referring to attributes of the sample stimulus, may instead refer to attributes of the test situation
Short-Term Retention Codes in Pigeons
53
(Roitblat, 1978, 1980). To illustrate, suppose a person is required in a memory test to pick from a set of names the one corresponding to a photograph of an individual he has just seen. What is the content of the memory trace: attributes of the picture, or of the person’s name, or both? One cannot tell from successful performance alone, but, given the right experiments, one can draw inferences concerning the content or meaning of the code. The second implication is that codes for the sample, and perhaps other types of codes as well, may be organized in different ways. Thus, certain stimulus configurations may encourage the animal to code a unified set of attributes in a single code, and this may allow more efficient retrieval of those attributes than if they were coded separately. An extension of the second implication is that compound stimuli that are coded as units are likely to be of the sort used by Lamb and Riley, where form and some other attribute such as color are contained in the same event. Such compounds in human perception are referred to as things or objects. How attributes are combined together into objects is a matter of considerable interest in current human information processing research (Treisman, 1977). Casual observation of animal behavior would persuade most of us that, like humans, some animals, including pigeons, detect and use (e.g., seize, jump up on, hide from) objects as such. At present we know little about the processes whereby animals identify stimulus aggregates as objects or how compounds identified as objects are stored in memory. We think that research of the type described in the latter part of this article, which examines the coding of sample attributes and how the organization of sample information affects those codes, will in time lead to an understanding of object perception and memory in animals. The first part of this report will examine the issue of how pigeons code information so as to perform successfully in conditional discrimination tasks such as matching-to-sample. We shall consider the evidence for different classes of codes and attempt to determine the conditions that produce each class. The latter part will be concerned with the effects of stimulus organization on coding. We shall comment on differences in code organization in our discussion of different types of codes, but shall give special consideration later to the possibility that the codes referring to sample attributes may be organized in different ways.
11. Designs for the Study of Event Memory in Animals Since Hunter’s (1913) initial work with the delayed response task, investigators have been searching for experimental procedures that would
54
Donald A. Riley et al.
allow them to make inferences about the code for a single event. Medin and Davis (1974), Ruggerio and Flagg (1976), and Carter and Werner (1978). among others, have reviewed and commented on various paradigms for assessing memory for specific events in the animal’s immediate past. The emphasis in these tasks is on short-term retention of single events that control a subsequent discriminative response. Typically, the animal is presented with an event and at a subsequent time is required to give a well-learned choice response that depends on that event. Correct responding demonstrates a persisting effect of the previous experience. Historically, paradigms used to investigate short-term event memory include the delayed response (Hunter, 1913), reward substitution (Tinklepaugh, 1928), and double alternation problems (Hunter, 1920). Recently the matching-to-sample and oddity-from-sample tasks have supplanted these, mostly because they allow more control over stimulus variables and because they eliminate orientation toward the goal as a potential solution to the problem. In these tasks pigeons are confronted with a three-key response panel with a food dispenser beneath the center key. A typical trial begins with sample presentation on the center key. In the paradigmatic task there are two stimulus values, which are randomly alternated as samples and which both appear as comparison stimuli in the test. Sample stimuli are usually colors or line orientations, although any kind of discriminable event can be used (Maki, Moe, & Bierley, 1977). After an experimenter-determined duration or a required number of pecks, the center key is extinguished and the two side keys are illuminated. A peck to the test stimulus designated as correct yields access to food. The location of the correct choice on the side keys is randomly determined. In the matching-to-sample paradigm the correct alternative during the test is the stimulus identical to the sample. Using Carter and Werner’s (1978) symbol conventions, a typical matching-to-sample trial can be designated as B(B*,R). The first letter stands for the sample, blue, and the following two letters stand for the test choices, blue vs. red, with blue being correct. In the oddity-from-sample paradigm the correct stimulus is the one not matching the sample, symbolized as B(B,R*). A third version of this kind of task is called symbolic matching-to-sample. In this case, the relation between the sample and the correct comparison stimulus is arbitrary. An example of such a trial would be B(V*,H), where V and H represent vertical and horizontal line orientations. In this case the vertical comparison stimulus is correct if the sample has been blue. Three temporal relations between sample and test stimuli are used in the conditional discrimination tasks discussed here. First is the simultaneous task, in which the sample and test stimuli are presented at the same
Short-l’erm Retention Codes in Pigeons
55
time, either with complete simultaneity or with the sample stimulus appearing first, but overlapping in time with the occurrence of the test stimuli. Second are zero-delay tasks, in which the onset of the test stimuli coincides with the termination of the sample. Third are delayed tasks, in which a delay occurs between the sample and the test. In general, investigators have tended to view correct performance that is based on a cue that is no longer present as evidence for a code that contains information about some attrihure ofthe cue. But, as Tulving and Bower (1974) make clear in their discussion of human memory, correct performance on a retention test is not sufficient to determine what the properties of the memory trace are. That an animal can match-to-sample after a delay interval shows that the animal is capable of bridging the gap between the initial event and the choice situation but does not reveal the content of the code.
111. A Classification of Codes We take it as a given that all stimuli detected and processed are transformed by the nervous system. By this we mean that the proximal stimulus is not directly connected with a response, but is instead mediated by a code that is correlated with the proximal stimulus. Here the term “code” will refer to that transformation of the sample stimulus that allows the organism to function appropriately at the time of the test. We shall consider the three types of codes for which there seems to be evidence. Two codes involve maintenance of information about attributes of stimuli occurring within the experimental context, either of the sample stimulus or of the test stimulus. The first type that will be examined, however, is those involving the use of a repetitive behavior chain to mediate delays. A.
CHAINING CODES
When Lawrence (1963), in an early discussion of coding, used the term “stimulus as coded,” he was referring to a set of processes whereby the subject modifies the sensory event into a new event, which represents the stimulus. We are reluctant to make the assumption that the code includes information about the stimulus itself, because in a number of experiments using delayed matching-to-sample it is quite reasonable to infer that no information about the attributes of the sample stimulus are present in the code determining performance at the time of test. Consider, for example, Blough’s (1959) work with delayed matching-to-sample in the pigeon. Blough trained four pigeons to match-to-sample, where the two Sam-
Donald A. Riley et al.
56
ples were a steady light and a flickering one. The birds matching performance differed considerably at delays of 5-10 sec. Two birds performed well with these long delays, and two did not. Observation of the birds that were successful indicated that both engaged in sample-specific behaviors during the retention interval. One bird, for example, always backed away from the flickering key and waved its head slowly back and forth, whereas during the steady sample it pecked rapidly at the top of the sample stimulus aperture. When a bird made the sample-specific response characteristic of the wrong sample during the delay interval, it responded incorrectly to the test stimuli. In addition, when, because of a change in the experimental arrangements, the same sample-specific responses occurred to each sample stimulus, correct performance at long delays disappeared in the previously successful birds. The fact that test responses were highly predictable on the basis of the sample-specific behavior observed during the delay suggests that performance was in some way based on the sample-specific responding. One interpretation of this experiment is that each sample elicits a specific behavior that, either by its stimulus consequences or the response directly, provides the cue that, in compound with the test stimuli, elicits pecking at the correct test. Such a chain of behavior need carry no information concerning attributes of the original sample stimulus. All that is needed is that distinctive conditioned responses be associated with each sample and that these be capable of being maintained until the occurrence of the test. These responses could then set the occasion for discriminative pecking to the test stimuli. If the bird could be asked to reconstruct the sample stimulus, given the above interpretation, it could not, even though it “matches” the sample correctly. B.
TESTCODES
An alternative to the chaining hypothesis is that animals code and maintain attributes of the visual stimuli used in conditional discrimination tasks. Codes of this type would refer to attributes of either the sample stimulus or the test stimulus. To clarify the latter possibility, suppose that a bird is required to peck a vertical test stimulus whenever a red sample has been presented. It could, at some point immediately after seeing the sample, code something analogous to “vertical.” Then, at the time of test, the application of a performance rule such as “Peck the stimulus that corresponds to the code” will result in a correct choice. C.
SAMPLE CODES
A frequently made assumption in the study of delayed matching behavior in animals is that the code refers to attributes of the sample. As we
Short-Term Retention Codes in Pigeons
57
shall see, however, this is not necessarily always the case. In order to assert that the subject matches on the basis of a code referring to sample attributes, alternative codes such as behavioral chains and codes referring to test attributes must be ruled out. D.
CODESA N D THEIRMEANING
It is important to distinguish between the mechanism or structure for the code and its meaning. For the most part we do not know the structure of the code; it could be a neural reverberating circuit or even a distinctive overt response. To distinguish this latter alternative from the chaining hypothesis, it is essential to ask what the meaning of the code is to the subject. For example, humans in a matching-to-sample task might say the word red over and over again during the delay. This is a coding mechanism, or part of a coding mechanism. The meaning of this coding mechanism for the subject cannot be inferred from the repeated utterance itself, It may mean “I have seen a red stimulus.” On the other hand, it could just as well mean “I will select a red stimulus at the time of test.” This second meaning could occur regardless of what color the sample had been. One cannot tell from the structure of the code what its significance to the organism is. The meaning of any code must be determined from appropriate experiments such as those analyzing confusion errors among to-be-remembered items (Baddeley , 1966). In addition to whether the meaning of the code refers to the sample or to the test, or is a chain, other attributes may be part of the code. The code may contain instructions about what response to perform at the time of test. For example, consider a symbolic matching task where a vertical test stimulus is correct whenever the sample is red. At least two types of test codes could arise. The animal may simply code “vertical,” in which case a separate performance rule such as “Peck the coded stimulus” is needed at the time of test. Alternatively, an instruction to respond may be encoded at the time of the sample, resulting in a “peck vertical” code. In this case no separate performance rule would be needed. In the next part of this article we shall be concerned mainly with the meaning of the code: sample, test, or chain. It will become clear, however, that these other properties of the code, such as how the coded information is organized and whether the performance rule is an attribute of the code, are important for understanding how pigeons perform conditional discrimination tasks. In summary, at least three different ways exist by which an animal could bridge the delay between the presentation of sample information and subsequent correct performance. Although we are primarily interested in codes for the sample stimulus, the evidence suggests that
Donald A. Riley et al.
58
animals can bridge delays in any of the ways we have outlined. Consequently, we must have an understanding of the conditions that lead to each sort of coding so that they can be experimentally isolated and studied. In the next section, we evaluate the evidence for each of these types of codes. IV. A.
EVIDENCE FOR
Evidence for Different Codes
C H A I N I N G CODES
As we have indicated, Blough (1959) provided evidence consistent with a chaining interpretation in an early study of delayed matching. Recall that Blough trained four pigeons on a delayed matching-to-sample task. Two birds developed sarnple-specific behaviors and performed well, while two did not and performed poorly. Performance was disrupted in those animals showing sample-specific responses, if those responses were altered. When the chains varied such that the sample-specific responses became more similar, the likelihood of the incorrect chain following a sample increased, also resulting in reduced performance. All these facts support the idea that the chains mediated the delayed matching behavior. The data are, as Blough recognizes, primarily correlational. The chains are not directly controlled by experimental manipulations, but emerge apparently as superstitious responses that become reinforced and correlated with the sample stimuli. Further evidence has supported Blough 's suggestion that overt repetitive saniple-specific behavior can provide the link between the sample and the test. Carter and Werner (1978) review this literature. Consequently, we shall comment extensively only on those articles not discussed by them. A number of studies have shown that requiring sample-specific behaviors facilitates matching-to-sample performance (Cohen, Looney, Brady, & Aucella, 1976; Eckerman, 1970; Urcuioli & Honig, 1980; Zentall, Hogan, Howard, & Moore, 1978). Eckerman (1970) showed that, the farther apart on a horizontal 10-inch key the subject was required to peck in the presence of different samples, the faster the pigeons acquired the discrimination. Cohen el ul. (1976) showed that associating an FR16 with one sample and a DRL3 with the other produced differences in the speed of pecking the two samples, facilitated the learning of a simultaneous matching-to-sample task, and eliminated the usual superiority of matching color as opposed to line orientation samples. Zentall er uI. ( 1 978) showed that requiring responding to one of two
Short-Term Retention Codes in Pigeons
59
different places, depending on which of two samples was present, facilitated not only zero-delay matching-to-sample, but delayed matching as well. Zentall et al. trained birds in four different conditions: ( a ) no pecking at the center key; ( h )five pecks to the center regardless of sample color; ( c ) five pecks to the center when the color was red, but none for blue; ( d ) five pecks to the center for red, five pecks to a key above the center for blue. The condition requiring differential pecking to both keys resulted in the best performance, both in zero-delay and longer delay conditions. Among the other conditions, performance was poorest when no pecks were required, with performance after a l-sec delay close to chance. With delays longer than 1 see, the general superiority of the group required to give differential pecks to the two samples maintained itself, but with marginal statistical reliability. Some animals in all conditions exhibited sample-specific behaviors regardless of the experimental requirements. There was a reliable correlation between this presence or absence of sample-specific behaviors and the level of performance in the delayed matching-to-sample task. These experiments demonstrate that sample-specific responses facilitate at least two aspects of matching-to-sample performance. First, they can aid in the acquisition of matching, especially if the discrimination is difficult. Second, they seem to be an important component of the pigeon’s ability to match with long delays. Experiments of this sort do not rigorously prove, however, that performance is mediated by a chain. It is possible that differential responding aids in the retrieval of some sample or test attribute. However, nothing in these data forces such an interpretation. One experiment from a series by Urcuioli and Honig (1980, Experiment 3) gives unambiguous evidence that matching can be supported solely by the events associated with sample-specific responding. Their experiment had four phases. They first associated a different reinforcement schedule with each of two line orientations (FRlO-vertical, DRL3 sec-horizontal) in a successive discrimination procedure. In the second phase, separate groups learned either simultaneous matching-to-sample or simultaneous oddity-from-sample, with the same line samples and response schedules being in force during sample presentation. After a high level of performance was achieved, a new series of successive discrimination sessions were given, this time associating each schedule with a color (FRlO-red, DRL3 sec-green). Following 1 day of “refresher” training with simultaneous matching-to-sample and simultaneous oddity-fromsample, again with lines used as both samples and comparison items, the birds were given a symbolic matching task with the previously trained colors as samples and the lines as test stimuli. For half the subjects this
60
Donald A. Riley et al.
new task was consistent with previous training, in that the two response schedules were associated with the same line tests as in the first task. The other subjects had inconsistent mapping, with the response schedules reversed. The birds in the consistent condition continued to match at high levels of accuracy. The birds in the inconsistent condition, however, dropped immediately to below-chance levels of performance. The high level of performance in the consistent condition relative to the inconsistent condition demonstrates that the sample-specific behaviors mediated between sample and test events. Since the birds had received no previous training linking the colors to the appropriate comparison test stimuli, a code reflecting the visual attributes of the sample would have produced chance performance. Although matching was very high on the first day of transfer, there was some drop in performance for the consistent group. This raises the possibility that a code referring to the visual stimulus partially controlled the behavior. As the authors point out, however, it is possible that the hues caused a slight change in the sample-specific behaviors, and this change was responsible for the relative deficit. A control condition in Experiment 1 of the same article, which was similar to the consistent condition but used only lines in all phases, produced a level of performance quite comparable to the color transfer performance just alluded to. This suggests that the change from lines to colors was not responsible for the decrement in the consistent color transfer condition. Our conclusion is that these authors have shown control by sample-specific behaviors. This form of mediation is not one that captures visual attributes of the sample stimulus, as evidenced by the transfer tests. Recently Hogan, Pace, and Zentall (1980) have provided a similar demonstration with a zero-delay procedure. They trained pigeons to match-to-sample with different sample-specific responses associated with different colors (FR16-red and blue, DRL3-green and yellow). During training these same colors appeared as test stimuli but always so that colors with different response requirements occurred together fe.g., R(R*,G) and R(R*,Y)J. Thus, both the color and the sample-specific response predicted the correct test stimulus during training. The pigeons were then given trials in which the color of the sample did not provide any information as to the correct test alternative but the sample-specific response did [e.g., R(B* ,G) and R(B*,Y)]. Transfer to these new trials was high, indicating that the sample-specific responses could control matching behavior. This conclusion was further supported by the poor transfer observed on trials in which color information was the only available cue [e.g., R(R*B) and G(G*,Y)]. In summary, there is evidence that behavioral chains can provide the
Short-Term Retention Codes in Pigeons
61
link between the sample and the test in the conditional discrimination task. This was demonstrated by Urcuioli and Honig (1980) in a simultaneous task and by Hogan et al. (1980) in a zero-delay task. The experiments by Blough (1959) and by Zentall et af. (1978) suggest that the same outcome would be found at longer delays. Thus, like Hunter's delayed response task, delayed matching-to-sample allows correct performance on the basis of overt skeletal responses. Apparently, once sample stimuli come to elicit sample-specific responses, those responses can serve as discriminative cues to correct performance without any memory of the sample stimulus. Consequently, if one wishes to study codes for sample or test attributes, care must be taken to ensure that such behavior chains are not responsible for correct behavior during the test. B . EVIDENCE FOR TESTCODES
As we have indicated, a code could contain information referring to the sample or to the test. The existence of test codes would raise a number of interesting questions. For example, do animals initially code sample attributes and then transform this code into one refemng to the appropriate test attributes'? Evidence from human information processing research suggests that codes can, in fact, change over time (Posner, 1969). Our purpose here, however, is simply to determine whether pigeons can generate a code referring to test attributes. Roitblat (1978, 1980) used the analysis of confusion errors to determine if codes in a symbolic matching-to-sample task referred to attributes of the sample or of the test. He trained two pigeons in a symbolic matching-to-sample task in which the correct test stimuli for three different color samples (blue, orange, and red) were three different line orientations (0, 12, and 90 degrees from vertical, respectively). For another pigeon, the line angles were samples and the colors were tests. For both colors and lines, two of the stimuli were more similar to each other than they were to the third. The relations between the stimuli were such that the dissimilar colors were associated with the similar line angles and the similar colors were associated with the dissimilar line angles. If the animal encodes the sample stimulus and maintains that code across the delay, it would be expected that, with increasing delay, confusion errors would increase more between the similar samples than between the dissimilar samples. This would be revealed by increased errors between the dissimilar test stimuli. On the other hand, if the pigeon's code refers to test attributes, confusion errors across time should increase for similar test items, even though the samples in those cases would be dissimilar. Roitblat 's experiment revealed that, with increasing delay, the in-
62
Donald A. Riley et al.
creases in confusion errors were between similar test items rather than between similar samples. This suggests that it is not the sample that is coded in memory, but some critical attribute of the test stimulus (i.e., line orientation for two birds, and color for the third). This finding occurred regardless of whether the samples were colors mapped onto line tests (two birds) or line samples mapped onto color tests (one bird), although the latter bird’s results were not statistically significant. It is unlikely that this effect was produced by chaining. Suppose the animal had adopted a sample-specific response chain to each of the samples. In order to produce the obtained results, the bird would have had to produce chains that were isomorphic with the angular orientations onto which they were mapped because confusion errors were greatest between siinilar test item s . Roitblat argues, on the basis of these data, that the coded information that allows pigeons to perform in delayed matching-to-sample tasks refers to attributes of the test stimulus. While it seems inescapable that the birds in his experiment did in fact code test attributes, it should be clear from our previous discussion of chaining (and the discussion of sample codes in the following section) that this is only one of the ways that animals can mediate the delay between sample and test. That pigeons do not always perform matching-to-sample tasks on the basis of codes refemng to test attributes is demonstrated in an experiment by Wasserman, Nelson, and Larew (1 980). Using a procedure developed by Konorski (1959; Wasserman, 1976), Wasserman ct al. required animals to associate a different test stimulus (e.g., red, yellow, blue, or violet), which appeared on the center key, with each of four stimulus-response sequences used as samples. An example of such a sequence would be: left key illuminated, peck left key, followed by right key illuminated, peck right key (LR). The four sequences were LL, LR, RL, RR. A test stimulus was designated correct for each stimulus-response sample. If the test stimulus presented on a given trial was the one associated with the stimulus-response pattern presented on that trial, pecking at the test was reinforced. If any of the other three tests appeared, pecking was not reinforced. Under these circumstances, pigeons peck in the presence of the correct test stimulus and withhold responding in the presence of the other three. Since all test stimuli followed each kind of stimulus-response sequence, the pattern of confusion errors among test stimuli allows inferences concerning the content of the code mediating performance. Confusions between stimulus-response sequences occurred most frequently if the last response of two different samples was the same (i.e., LL and RL trials were confused, as were RR and LR trials). Both the sample-
Short-Term Retention Codes in Pigeons
63
encoding and chaining hypotheses can account for these data, but the test-encoding hypothesis cannot. The same pattern of confusions was found in all animals, even though the correct test stimulus for each sample differed among birds. If the animals were using a test code, then one would expect between-subject differences in confusion errors. Samples associated with similar tests should be confused regardless of the properties of those samples. No such effects were observed, however. The fact that Roitblat ’s experiment seems to demand the interpretation that animals code test attributes, whereas the Wasserman et al. data cannot be accounted for with this assumption, suggests one condition that determines whether test codes are employed. One difference between these two paradigms is that in the Roitblat experiment each sample was always followed by the same correct test stimulus, whereas in the Wasserman et al. study any one of four test stimuli could be presented following every sample. Thus, a code referring to test attributes in the Wasserman et al. experiment would have to include all four possibilities. For example, an LL sample might require a test code of the form “If red test then peck, if yellow don’t peck, if blue don’t peck, and if violet don’t peck.” Thus, uncertainty as to which test stimulus is going to appear following a given sample may be one condition that discourages the coding of test attributes.
c.
EVIDENCE FOR
SAMPLE CODES
The code most commonly assumed to occur is one that includes information about attributes of the sample. Evidence supporting the assertion that sample attributes can be part of coded information comes from studies involving transfer of conditional discrimination performance to novel stimuli. An understanding of how and when well-learned matching behavior transfers to novel stimuli is of interest for several reasons. If pigeons were found to match-to-sample with novel stimuli on the first trial of transfer, it would have three implications. First, it would indicate that pigeons have a generalizedperformance rule. An example of such a rule would be “Peck the test stimulus that matches the code. ” Second, it would imply that the existence of a code for the new stimulus did not depend upon prior training in the experimental situation. These two implications are contrary to the general model of conditional discrimination performance proposed by Cumming and Berryman (1965) and more recently by Carter and Werner (1978). The third and most important implication of positive transfer, for our purposes, is that it requires that the mediating code refer
64
Donald A. Riley et al.
to sample attributes. Neither a chaining nor a test-coding hypothesis is likely, because the chaining hypothesis would require that a new chain be learned, whereas the test hypothesis requires experience with the test in order to associate attributes of the test with the sample. Neither of these alternatives is possible with novel stimuli, especially on the first trial. Thus, a demonstration of positive transfer to novel stimuli is important to the present discussion because it seems to demand that performance in the task be based on a code referring to sample attributes. While positive transfer has been found in several recent experiments (Urcuioli, 1977; Urcuioli & Nevin, I975; Zentall & Hogan, 1978), early attempts to demonstrate the effect failed to do so (Berryman, Cumming, Cohen, & Johnson, 1965; Cumming, Berryman, & Cohen, 1965). Comparison of these studies suggests a possible reason for the discrepant results. We shall examine each line of research in an attempt to draw firmer conclusions concerning the conditions that support positive transfer. Cumming et al. (1965) trained pigeons to match with blue, green, and red stimuli and then, after criterion performance was reached, replaced the blue stimulus with a yellow one, whenever a blue would have normally occurred as a sample or comparison stimulus. Under these conditions, the birds consistently chose red on the new Y(Y*,R) trials and continued to do so for 2 days. With Y(Y*,G) trials a slight preference for green on the first day of testing yielded to a preference for yellow on the second day and thereafter. To reach an 80% level of performance required 3 days for the Y(Y*,G) trials and 6 days for Y(Y*,R) trials. The interpretation offered by Cumming et al. was that the birds, not having learned a code for yellow, code it as “red,” resulting in the choice of red on Y(Y*,R) trials. This conclusion is also supported by the observation that during the several test days in which the birds choose red on Y(Y* ,R) trials, performance on red sample trials declined. The authors interpret this change as extinction resulting from a change in reinforcement for “If ‘red’ code then choose red,” from 100% to 50%, reflecting the use of the code “red” on yellow sample trials. A second study by Berryman et a l . ( 1965) used oddity-from-sample and found results consistent with the “red” code for yellow sample coding interpretation. This coding assumption accounts for the animal’s behavior on Y(Y*,R) trials, where it is assumed that the bird has not learned a code for yellow, and so yellow elicits the code “red.” It has difficulty in accounting for the Y(Y*,G) trials, however, where by the second day of testing the animal matched yellow correctly. Apparently a “yellow” code had developed. Why was it not used on the Y(Y*,R) trials? One interpretation of the differential rate of acquisition for these two
Short-Term Retention Codes in Pigeons
65
trial types assumes that the animal comes to the experiment with a skt of codes corresponding to afferent processes. During training each of the sample stimuli will elicit a code corresponding to the sample, but it also will elicit other codes corresponding to adjacent hue values, with the probability of a given code’s being elicited increasing as the similarity between the hue referred to by that code and the hue of the sample increases. That is , generalization of codes corresponding to stimulus Values will occur for all the reasons that stimulus generalization of overt responses occurs (Riley & Lamb, 1979). Thus, each sample will elicit a set of codes. For example, in Cumming et al. ’s experiment a novel yellow sample will produce a “yellow” code but also the “red” code, and this “red” code will control behavior because it has most often been associated with the appropriate response and reinforcement. If procedures had been used that resulted in a sharp generalization gradient for the codes elicited by the red stimulus, however, then yellow samples might not elicit code “red” and the code “yellow” could be used, given that the pigeon had a generalized matching rule (e.g., “Peck the test stimulus that matches the code ”). Performance on Y(Y* ,G) trials improved faster than performance on Y(Y* ,R) trials, according to this interpretation, because there is less generalization between “yellow” and “green” codes than between ‘‘yellow” and “red” codes. For example, the point at which, through differential reinforcement, the yellow sample no longer elicited the “green” code might be reached sooner than the point at which the yellow sample no longer elicited the “red” code. This interpretation of the Cumming et al. data implies that positive transfer of conditional discrimination performance to novel stimuli will be observed only if the subject receives differential reinforcement of stimuli along the dimension in question. Marsh (1967), for example, has shown that steep generalization gradients develop for color if stimuli remote from S + are presented but not reinforced. That is, the procedure inhibits responding to stimuli adjacent to S + even though responses to these stimuli are not explicitly extinguished themselves. If the extinction of responses to remote stimuli does not take place, however, the generalization gradients are relatively flat. In the conventional matching-to-sample task, each set of comparison stimuli includes an S + , thus minimizing the necessity for inhibiting responses to S- stimuli. Under these conditions, the “red” code could occur to a relatively broad range of stimuli around red, perhaps including yellow. If the procedures had forced inhibition of responding to S - stimuli for each sample, however, the “red” code might not generalize and the “yellow” code could control choice behavior. Transfer studies by Urcuioli and Nevin (1975), Urcuioli (1977), and
66
Donald A. Riley et al.
Zentall and Hogan ( 1978) have used procedures that require inhibition of responding to S - stimuli. Rather than using the conventional matchingto-sample task, the first two experiments used a modification of Konorski’s successive test procedure, one that required the birds to withhold pecks to S- stimuli. Zentall and Hogan forced processing of S- stimuli by giving “negative instance” trials where no correct stimulus appeared at the time of test and responding had to be inhibited. Both procedures resulted in transfer to novel stimuli. The Urcuioli and Nevin (1 975) and Urcuioli (1 977) procedures were the same except that the former involved matching-to-sample and the latter oddity-from-sample. In both, a sample was presented on the center key and remained on while a single test was presented on either side key. If the test stimulus was correct, a peck was followed by reinforcement. If it was incorrect, the subject had to withhold responding for 4.5 sec, upon which the correct stimulus would appear on the other key. A peck to that stimulus was then reinforced. The samples in the first study were red, yellow, and green, and in the second they were red, green, and blue. After reaching criterion performance, the pigeons were transferred to novel stimuli: blue and violet in the first article, yellow and violet in the second. The animals responded with short latencies to positive probes, indicating that they recognized them as correct, while long latencies, were observed to negative test stimuli-i.e., the animal waited for the correct one. These results were found with both matching and oddity tasks and whether the novel stimuli appeared as the sample, the test, or both. These results support the assertion that the animals were using a single rule in arriving at a correct solution. i’he application of such a rule (e.g., “Peck the test stimulus whose attributes match the code”) would seem to require that the animal code information about the sample stimuli. Thus, it would appear that animals have a code for novel colors and can use it if stimuli on the dimension in question have received the appropriate differential reinforcement. Zentall and Hogan (1974, 1976), using a different transfer paradigm, also showed positive transfer to novel stimuli. Their basic experiment used four groups of pigeons, each trained with one set of two stimuli and then switched to another set of two stimuli. At the time of the stimulus change, each group was either maintained on the same task (i.e., matching or oddity) or switched to the other. Although Zentall and Hogan found superior transfer (i.e., faster acquisition) when the first and second tasks were the same, the differences have been rather small, and a number of objections to their conclusion that the results show transfer of the matching (or oddity) concept have been raised (Carter & Eckerman, 1976; Carter & Werner, 1978; Holmes, 1979).
Short-Term Retention Codes in Pigeons
61
In their most recent report related to this issue, Zentall and Hogan (1978) have found compelling evidence for transfer of both matching and oddity performance, both within the color dimension and across dimensions (form to color). Using basically the transfer paradigm described, the results were particularly clear cut for groups that were required both to match (or nonmatch in the oddity task) when a comparison test occurred, and to withhold responding on trials on which both test stimuli were negative. Thus, if the task was matching-to-sample and the sample was green, two red test stimuli would be presented on a “negative instance” trial. Like Urcuioli and Nevin’s procedure, this task forces the bird to process information from negative stimuli and make discriminative responses to them. The experiments just cited seem to provide evidence that pigeons can encode attributes of the sample. Two considerations prevent us from drawing this conclusion with absolute confidence. First, the data in these studies are aggregated across trials; thus, hidden learning effects may be present. We are inclined to discount this problem, since performance on the first day of the transfer tests is as high as control performance, and there is no evidence of learning during the transfer tests. A second, and more serious, problem is the use of simultaneous tasks in these studies. Because the sample and test stimuli overlap in time, the birds may have responded to a stimulus configuration (Carter, 1977). That is, responding to the sample and two test stimuli as a pattern does not require short-term retention, as the animal would not be treating the problem as a conditional discrimination. The studies by Urcuioli and Honig (1980) and by Berryman et al. (1965), however, also used simultaneous tasks, but they produced results that could have occurred only if the animals had treated the problem as a conditional discrimination. Thus, animals employ a strategy requiring short-term retention of trial events in simultaneous tasks, at least under some conditions. Other studies that have looked for transfer of matching behavior (Farthing & Opuda, 1974; Holmes, 1979) have found little evidence of transfer to novel stimuli. These investigators concluded that the chaining hypothesis is the best explanation for their results. Both studies, however, used the conventional matching-to-sample task. It is clear that this kind of training procedure does not support transfer to novel stimuli. Apparently some kind of S - discrimination training is needed to allow birds to demonstrate the existence of codes for novel stimuli. In summary, there is evidence that pigeons can perform correctly in conditional discrimination tasks when novel stimuli are presented. We take this fact as evidence for the presence of sample codes in such situations. A clearer understanding of the conditions allowing demonstration
Donald A. Riley et al.
68
of such codes will require direct comparison of the different conditions we have discussed. So far we have discussed three types of codes upon which conditional discrimination performance can be based: sample codes, test codes, and response chains. We have presented evidence suggesting that all three types do in fact occur and have tried to examine some of the factors determining when each type will occur. Because so little research has addressed these questions directly, the conclusions drawn must be taken as tentative. It seems clear, however, that the frequently made assumption that conditional discrimination performance is based on a coded representation of the sample is only one of several possibilities that must be examined if the processes involved in conditional discrimination performance are to be understood.
V.
Stimulus Organization and Coding
As we stated at the outset, our decision to examine coding processes in pigeons was prompted by our observations and research on matching-tosample with compound stimuli (Lamb & Riley, 1981). As mentioned previously, Lamb and Riley attributed differences in performance with compound samples whose components were arranged in different ways to differences in the way those samples were coded. We next examine this implication that the organization of the sample stimulus can affect the coding process. Previous work from this laboratory, summarized by Riley and Leith ( 1 976) and by Riley and Roitblat (1978), has found matching with element samples to be better than matching with compound samples. Typically in these experiments, element samples consisting of white lines of a given orientation or a colored field are presented on the center key of a three-key display. Compound samples consist of a line element superimposed on a color element. Since the test stimuli are typically the two elements from oiie dimension (i.e., color or line orientation), subjects are forced to process both elements of compound samples in order to perform correctly at the time of test. Under these circumstances, matching-tosample performance is better with element than with compound samples. Also, performance with both element and compound samples improves with increases in the duration of sample presentation. While the evidence is somewhat equivocal, it seems that the difference in performance between element and compound samples is greatest at intermediate sample durations (about .3- 1 sec) and decreases with longer sample durations. To put it another way, performance with element samples reaches
Short-Term Retention Codes in Pigeons
69
asymptote at shorter sample durations than does performance with compound samples, but asymptotic performance with both kinds of samples seems to be the same. This finding is consistent with the argument, advanced by Maki et al. (1976), that at brief sample durations the requirement that the bird process both attributes of the sample taxes the information processing capacity of the subject. With longer sample durations, the bird has more time to encode both attributes of compound stimuli and should eventuaIly reach the point where the elements of compound samples are as well coded as they are when extracted from element samples. This explanation has been designated the information overload hypothesis. It should be noted that the information overload hypothesis says nothing about the meaning of the code. It merely says that, whatever the code refers to, the rate at which it develops is limited. A.
STIMULUS ORGANIZATION EFFECTS
Some evidence suggests that the nature of the sample display may influence matching performance with compound stimuli. Farthing, Wagner, Gilmour, and Waxman (1977) reported no differences in performance between element and compound samples, except for a small superiority of element over compound performance with zero-delay line tests. But with longer delays no difference was observed between element and compound performance for either color or line dimensions. The most obvious difference between this experiment and those run in our laboratory is that their line displays consisted of a single line bisecting the key, whereas the experiments reported from this laboratory used three parallel lines. Cook (1979), for reasons unrelated to this problem, ran groups of pigeons in two different apparatuses. One of these had displays with three parallel lines, and the other had displays with two parallel lines. While there may have been other differences between these two situations, it is nonetheless interesting to observe that the birds run in the apparatus with the three-line displays showed more regular differences between element and compound performance (Fig. lc), both in color and line, than did the animals with the two-line display (Fig. lb). These, in turn, showed greater differences than were reported by Farthing et al. (1977) (Fig. la). These differences suggest the possibility that the way a stimulus is organized influences matching performance. In an attempt to determine whether differences in the way the elements of a stimulus compound are arranged can affect Performance in the matching-to-sample task, Lamb and Riley (1981) used small line angle
A
....
......._ .. . . .-...... , . _ .... .. ..-._ ... ... .
B
C
Fig. I . Percentage correct with element and compound samples as a function of delay. Circles indicate element trials. triangles compound trials, solid lines color tests. and dashed lines line tests. ( A ) Data from Farthing. Wagner, Gilmour, and Waxman (1977). The sample duration was 3 sec, and angular orientation was conveyed by a single line bisecting the key. (B) Data from Cook (1979). The sample duration was 2.5 sec, and angular orientation was conveyed by two parallel lines. (C) Data from Cook (1979). The sample duration was 2.5 sec. and angular orientation was conveyed hy three parallel lines.
71
Short-Term Retention Codes in Pigeons
and color displays on a square 6.25-cm2sample display key centered between two round keys. Element samples were either red or orange squares, 25 mm in size, or three vertical or horizontal white lines filling a square of the same size. A color and a line element presented together on the same key either far apart, close together, or superimposed constituted the Separated compound conditions. In a fourth, Unified, compound condition, both angular orientation and color information were carried by colored lines on a black background. No white lines were present (see Fig. 2). Test stimuli in all conditions were the same: horizontal vs vertical white lines, or red vs orange squares. Tests followed samples with zero delay. Matching performance was better in the Unified compound condition than in any of the Separated conditions and was, in fact, often as good as performance in the Element condition. Figure 3 shows performance for Elements, Unified compounds, and Separated compounds for each of the three birds in this experiment a s ' a function of sample duration. The pattern of relationships among conditions was the same for all birds, but the temporal parameters for the third bird were different, making it impossible to include this bird's data in the mean. Analysis of the Superimposed condition and its relationship to the other Separated conditions is both complex and not critical to an understanding of the present argument. For a discussion of this condition, see Lamb and Riley (1981). One aspect of these data that is important for our present purpose is that, unlike previous findings, differences in performance with the different sample types does not decline at the longer sample durations. This suggests that at least part of the performance differences are not due to the processes assumed by the information overload hypothesis. As was al-
Separated C'niiipound\
Suprriiiipowd
Far
I iiitied
Fig. 2. Examples of the five types of samples used in the Lamb and Riley (1981) experiment. The black squares represent the response key onto which the samples were projected. The bars represent angular orientation, either vertical or horizontal. The crosshatched lines represent color, either red or orange. Drawings are not to scale. See text for dimensions of response key and sample stimuli.
Donald A. Riley et al.
12
Bird 45
100,
Bird 46
wA Elemenl
-b!
0 Unified
ao
5-
70-
$I
60-
0
Superimposed
0
Close
80
,
g
6 -
" D
111
E
t:
50
$I
'
LO.
50.
04
110
295
805
2185
5915
.25
Sample Duration (Sect
Mean
so
10.0
150
2185
593q
(Birds 45 and 47)
I
. 04
20
Sample Duration (Sec)
Bird 47
I
.75
110
291
I05
Sample Duration (Ser)
2185
(915
04
110
195
805
Sample Duration (SeL)
Fig. 3. Percentage correct as a function of sample duration for the five sample types used in the Lamb and Riley (1981) cxpcriment. Shown are the individual data from each bird in the experiment, along with mean data for Birds 45 and 47. Data from Bird 46 could not be included in the calculation of mean performance because different sample durations were used for this bird than for the other two. (Copyright 1981 by the American Psychological Association. Reprinted hy permission,)
ready stated, differences in matching accuracy caused by a limitation in the rate at which information is coded should be eliminated if sufficient processing time is available. Thus, the asymptotic differences in performance among sample types observed in Lamb and Riley's experiment suggest that the codes for the different sample types differ in some way other than in the rate at which they are developed. The question then
Short-Term Retention Codes in Pigeons
13
becomes, what about the code is changed when the to-be-remembered information is organized in different ways? As before, let us assume that the three possible codes for achieving matching performance are ( a ) sample-specific response chains, (6) encoding attributes of the sample, and ( c ) encoding attributes of the test. Since this experiment was not conducted with the question of the meaning of the code in mind, it is impossible to reject with certainty any of these three types of codes. However, sample-specific chains as an explanation of the superiority of Unifed over Separated compounds, while possible, is unlikely. The animals were first trained to match elements before compounds were introduced. When the compounds were introduced, the sample attributes should have continued to elicit the same response chains, but each compound would then have elicited two independent and possibly competing chains. Since such competition would exist with all compounds, this analysis provides no explanation for the observed differences in compound matching. Alternatively, new sample-specific responses for each compound stimulus might have developed despite competition from the previously learned element chains. However, this possibility offers no good reason for the large transfer effects found on the first day of training with Unified compound samples. One would expect the least transfer of the chains supporting element performance to Unifed compounds, since these were the compounds that least resembled the elements. Nevertheless, performance with Unified compounds was over 70% on the first day of compound training for the one bird for whom compounds were completely novel. This was comparable to the performance of the other two birds (75%), which had received a small amount of experience with the compounds during element training. There is also reason to believe that the performance of the birds in the Lamb and Riley experiment was based on codes referring to sample rather than to test attributes. Recall that a comparison of the experiments of Roitblat (1978, 1980) and Wasserman et al. (1980) led us to suggest that test coding was unlikely if there was uncertainty as to which test stimulus would appear following the sample. This was the case in Lamb and Riley’s experiment. Each compound sample was unpredictably followed by either a color or a line test stimulus. The most reasonable assumption at the present time would seem to be that the birds in the Lamb and Riley experiment matched on the basis of codes that referred to sample attributes. We suggest that the way in which those sample attributes were organized affected how the coded information was organized. No research in the animal literature bears directly on this issue, but research from the human information processing and memory literature is useful to consider. Spyropoulos and Ceraso
74
Donald A. Riley et al.
(1 977) have conducted experiments that are similar to ours i n that they compare memory for unitary and nonunitary stimuli. An example of a unitary stimulus is a red triangle; that of a nonunitary stimulus, an achromatic triangle with a red rectangle along side. For nonunitary stimuli the color is always presented as a rectangle of the same dimensions. Following Asch, Ceraso, and Heimer (196O), they proposed that the properties of unitary compound stimuli are stored together, whereas the properties of nonunitary compounds are stored separately. Further, if one of the attributes of the unitary stimulus is used to categorize the cotnpound, as required by their task, then this attribute, when used as a cue in recall, will have direct access to the trace of that compound and thus to the other element. If, on the other hand, the noncategorized attribute is used as a cue in recall, access to the trace should be poor and search time should be long, because access to one element of the compound does not provide access to the other. This theory, then, assumes that the content of the memory trace consists of organized attributes of the original stimulus. In general, the results of their study support this argument. Subjects were instructed that, as each stimulus was presented, they were to classify it according to instruction on the basis of either form or color and were then to place it in a box bearing the categorized property on its face. Thus, in the example of the colored triangle, there would be a box with a triangle on the face and another with the appropriate color. Instructions for classification occurred just as the item was seen. Either unitary or nonunitary stimuli could be appropriately classified by this scheme in exactly the same way. Following classification of 10 stimuli, either unitary or nonunitary in character, the subjects were required to recall, either on the basis of the categorizing cue or the noncategorizing cue. Subjects in the unitary condition showed high accuracy and short latency of response when the cue for recall was the categorizing cue and the item to he recalled, the noncategorizing cue. If, on the other hand, the cue was the noncategorizing cue, the subjects responded poorly and slowly. Subjects in the nonunitary condition showed no difference between recall to categorized and noncategorized cues. In general, they were slightly better in performance on both than were the unitary subjects when given a noncategorized cue. The results of this experiment and of four others in the Spyropoulos and Ceraso study support the authors' general thesis. The elements of the unitary stimuli seem to be stored together, so that access to the categorized element ensures access to the other to-be-remembered element. Such access does not occur in the nonunitary compounds. It is possible that a similar process is occurring in Lamb and Riley's experiment. That is, birds may code the attributes of Unified compounds as one unit while coding those of the Separated compounds as separate
Short-Term Retention Codes in Pigeons
15
units. There are at least two ways this might facilitate matching-to-sample performance. First, fewer units might be easier to remember than more. Second, fewer units might mean fewer events to search, if such a process is required to perform the matching-to-sample task. B.
EXTENSIONS
The experiments by Spyropoulos and Ceraso (1977) and by Lamb and Riley (198 1) suggest that ostensibly separable attributes are sometimes coded as a single unit. One possible reason for this type of coding was that these conjunctions of stimulus attributes were considered by the subjects to be “objects,” or unified discrete things. Treisman and her colleagues (Treisman, 1977; Treisman & Gelade, 1980; Treisman, Sykes, & Gelade, 1977) have been concerned with the way subjects combine separable attributes of a compound stimulus in perception and memory. Treisman et al. argue that humans, in order to perceive and remember objects as unitary wholes, must, if the individual attributes are separable, have some mechanism for joining them together so that they are correctly perceived and correctly remembered as objects. Thus, if in a picture there is a blue shirt and a green dress, accurate memory depends upon the association of the appropriate color with the appropriate form. It is their general thesis that the identification of such conjunctions depends upon sequentially focusing attention on each such stimulus compound that appears together in the same space and at the same time. They argue that the identification of individual elements, such as different colors or different forms, can be carried out in parallel; but the identification of conjunctions of elements, such as colored forms, requires sequential focussed attention. Evidence from speed of identification tasks and other perceptual tasks support the proposition that an individual color can be identified equally rapidly regardless of the number of distractors, but that the identification of a particular colored form becomes slower in a more or less linear fashion, depending upon the number of distractors. Similarly, in a recognition memory task Treisman et al. have shown that the speed of memory search is directly dependent upon the number of alternative conjunctions that must be searched in order to determine whether a test item matches a previously seen sample. No single fact from this rather complex set of experiments by Treisman and her colleagues forces one to the conclusion that perceptual and memorial object formation depends on sequential examination of conjunctions of attributes and selectively attending to them. Taken as a set, however, the data are persuasive. Focal attention to attributes presented in the same location provides the ‘‘glue” that holds
76
Donald A. Riley et al.
them together in memory. Whether the same sorts of processes occur in other organisms remains to be determined. Experiments that study the effects of stimulus organization on coding, however, may begin to elucidate how and when organisms treat stimulus compounds as “objects. ”
VI.
Conclusions
We have examined three alternative hypotheses about the nature of the code that exists between a sample stimulus and a response in simple conditional discrimination experiments and have been led to the following conclusions. Each of the three types of codes considered (i.e., sample codes, test codes, and conditioned response chains) mediate between the sample and the response to the test under some conditions. Usually, observation of conditional discrimination performance alone is not sufficient to determine which type of code is occurring. Even in a seemingly obvious case such as when overt sample-specific responses are observed, one cannot be sure that the code is the sensory consequences of the chained response. Such a response chain may instead serve to reinstate sample attributes. Again, one cannot determine the meaning of the code to the animal without the appropriate experimental evidence. Evidence of this type comes from experiments where ( a ) the situation is arranged so that the different possible types of codes refer to discriminably different properties such as when sample stimuli are colors and test stimuli are line orientations, and ( h ) the situation is arranged so that error patterns or other response measures that are sensitive to interference or facilitation can be used, and the effects of operations known to influence one code but not another can be analyzed. Experiments discussed in the present article use only confusion errors generated by the similarity among the sample or test items. Experiments where variations in the similarity of prior or interpolated items are used to selectively interfere with the code have been useful, however, in the study of human event memory (Kroll, Parks, Parkinson, Bieber, & Johnson, 1970; Wickens, 1972). Recently, various studies have found that changes in delay illumination can interfere with pigeon short-term memory (Cook, 1980; Grant & Roberts, 1976; Wilkie, Summers, & Spetch, 1981). By using the same or similar paradigms, the analysis of the errors caused by the similarity of interpolated events with the to-be-remembered item could be used to determine the content of the code. There are some clues concerning the conditions that give rise to one code as opposed to another. For example, the relative discriminability of the different codes (Roitblat, 1980), the relative number of events to be
Short-Term Retention Codes in Pigeons
I7
carried by the different codes (Roitblat, 1980; Wasserman et al., 1980), and the extent to which reinforcement contingencies support one code over another (Urcuioli & Honig, 1980) all may determine which kind of code emerges. Because very few experiments have been designed to distinguish between different kinds of codes, this list is suggestive rather than definitive. Codes may differ in ways other than those just referred to. Thus, a sample code may be unified with more than one attribute represented in a single code, or multiple sample attributes may be coded separately. Codes may or may not include information concerning the appropriate response. Again, since these suggestions are based largely on data collected for other reasons, they are not definitive. The discussion in this article has been based, for the most part, on the performance of pigeons in conditional discrimination tasks. We restricted the discussions to this species and set of procedures because the analysis and integration of a large literature using comparable procedures and a single species seemed most likely to lead to a coherent picture of the processes involved in coding. The issues raised here are general, however, and should be examined in other species. It seems likely that informative species differences in coding will appear.
REFERENCES Asch, S . E . , Ceraso, J . , & Heimer, W. Perceptual conditions of association. Psychological Monog r a p h , 1960, 74(3, Whole No. 490). Baddeley, A. D. Short-term memory for word sequences as a function of acoustic, semantic, and formal similarity. Quurterly Journal of Psychology, 1966, 18, 362-365. Berryman, R., Cumming, W . W . , Cohen, L. R . , & Johnson, D. F. Acquisition and transfer of simultaneous oddity. Psychological Reports, 1965, 17, 767-775. Blough, D. Delayed matching in the pigeon. Journul of the Experimental Analysis of Behavior, 1959, 2, 151-160. Carter, D. E. Three models of conrlitionul discriininution learning: What do transfer datu mean? Paper presented at the meeting of the Eastern Psychological Association, Boston, April 1977. Carter, D. E . , & Eckerman, D. A. Reply to Zentall and Hogan. Science, 1976, 191, 409. Carter, D. E., & Werner, T. J . Complex learning and information processing by pigeons: A critical analysis. Jourtiul of the Experimentul Analysis of Behavior, 1978, 29, 565-601. Cohen, L. R., Looney, T. A . , Brady, J . H., & Aucella, A. F. Differential sample response schedules in acquisition of conditional discriminations by pigeons. Journal of the Experimental Analysis of Behavior, 1976, 26, 301-314. Cook, R. G. Element-compound differences in pixeon short-term memory. Paper presented at the meeting of the Western Psychological Association, San Diego, April 1979. Cook, R . G . Retroactive interference in pigeon short-term memory by a reduction in ambient illumination. Journul of Experitnentul Psvchology: Anitnul Behavior Processes, 1980, 6, 326338.
7x
Donald A. Riley et al.
Cumming, W. W., & Berryman, R. The complex discriminated operant: Studies of matching-tosample and related problems. In D. I . Mostofsky (Ed.), Stirnrtlu~gmeralizarion. Stanford, Calif.: Stanford University Press, 1965. Cumming, W. W . , Berryman, R., & Cohen, L K. Acquisition and transfer of zero-delay matching. P\ychologrca/ Reporis. 1965. 17, 435-445. Eckerman, 0 . A . Generalization and response mediation of a conditional discrimination. Jourriuf uf' rhc Expeririieiitd Ariaf~sisof' Behtrviur, 1970, 13, 301 -3 16. Farthing, G. W . , & Opuda, M. J. Transfer of matching-to-sample in pigeons. Jourrrcrl of ihe E.rperirneir/ul A i t a l y s i s of Behavior. 1974, 21, 199-21 3. Farthing, G . W.. Wagner, J . M., Gilmour, S., & Waxman, H. M. Short-term memory and information processing in pigeons. Lecrrning u n d Moriiwriorl. 1977, 8 , 520-532. Grant, D. S., & Roberts, W. A . Sources of retroactive inhibition in pigeon short-term memory. Jounitrl of E rperiiiwiiiuf Pswholog.~:Anirirnl Behuvior Proi.esse.s, 1976, 2, I 16. Hogan, D. E., Pace, G . . & Zentall, T. R . The qffeci ofrfifereritial sartiple rcspouse reyuirenierirs mi whui pi~ecnt.slcurri drrririg rtrtrtc.hirrg training. Paper presented at the meeting of the Midwestern Psychological Association, St. Louis, May 1980. Holmes. P. W . Transfer of niatching perforniance in pigeons. Jourrrd of ihe Experirnentol Analysis o/BehuikJr, 1979).31, 103- 114. Hunter, W. S . The delayed reaction in animals and children. Behavior Mofiogrcfphs, 1913,2, 1-86. Hunter, W. S. The temporal maze and kinesthetic sensory processes i n the white rat. Psychobiology. 1920, 2, 1-17. Konorski, J . A . A new method of physiological investigation of recent memory in animals. BuIIeiiit de I'..lcadertric PoluituiAe tles Scienrest Sin'e t f r s Scienres Birhgiques, 1959, 7, I 15- 1 17 Kroll, N . E A . , Parks, T.. Parkinson, S . R . , Bieber, S . L., & Johnson, A . L. Short-term memory while shadowing: Recall of visually and aurally presented letters. Journal of E.rperirtierirtrl Psychology. 1970, 85, 220-224. Lamb, M. R., & Riley, D. A . Effects of element arrangement on the processing of compound stimuli i n pigeons (Culuiiihia livia). Jouriitrl cfE.xperirnenta1 Psyc~holog~: Arrirnal Behavior Processes, 1981.7, 45-58. Lawrence, D. H. The nature of a stimulus: Some relationships between learning and perception. In S. hology: A sruciy of a scierire (Vol. 5). New York: McGraw-Hill. 1963. Maki, W . S . , J r . , & Leith. C . R. Shared attcntion in pigeons. Journtrl (!/the Exl'crirrterirul Ariul~~sis cf Behot~ior, 1973. 19, 345-349. Maki, W. S . , Jr., & Leuin, T. C . lnforrnation processing by pigeons. Scierrce, 1972,176, 535-536. Maki, W. S . , Jr., Moe, J . C . , & Rierley, C. M. Short-term memory for stimuli, responses, and reinforcers. Joirrriul of f3perirrienld f'syc,hn/ogyt Arrirtiul Behtrvior P roresses. 1977, 3, 156177. Maki. W. S.,Jr., Riley, 1). A . . & Leith, C. R. The role of test stimuli in rrialching-to-compound samples i n pigeons. Airirrtuf Letmirig urid Behavior, 1976, 4, 13-21. Marsh, G . D. Inverse relationship between discriminability and stimulus generalization as a function of nunibcr of test stimuli. Jourriul qf Conrparuiive mid Physiologit~ulP s y h o l o g v , I 967, 64, 284-289. Medins, 0 I-.,B Davis, R. T . Memory. In A . M . Scliricr B F. Stollnitz (Eds.), Brhnvior of iiorihrrrnuri primtrtes. New York: Academic Press, 1974. Poaner, M. I . Abstraction and the process of recognition. The Psychology ofkurriing and Motivariori, 1969, 3, 44-100. Riley, D. A , , B Lamb, M. R . Stimulus Generalization. In A. D. Pick (Ed.), Percepfiori trm/ i t s dedopriicrit. Hillsdale, N . J . : Lawrence Erlbaum, 1979. Riley, D. A , , & Leith, C . R . Multidimensional psychophysics and selective attention i n animals. Psychological Bulletin, 1976, 83, 138-160. -
Short-Term Retention Codes in Pigeons
79
Riley, D. A., & Roitblat, H. L. Selective attention and related cognitive processes in pigeons. In S . H. Hulse, H. Fowler, & W. K . Honig (Eds.), Cognitive processes in animal behavior. Hillsdale, N.J.: Lawrence Erlbaum, 1978. Roberts, W. A., & Grant, D. S . Studies of short-term memory in the pigeon using delayed matching to sample. In D. L. Medin, W. A. Roberts, & R . T. Davis (Eds.), Processesof animalmemop. Hillsdale, N.J.: Lawrence Erlbaum. 1976. Roitblat, H. L. Oirocling processes in pigeon short-term memop. Unpublished doctoral dissertation, University of California, Berkeley, 1978. Roitblat, H. L. Codes and coding processes in pigeon short-term memory. Animul Learning and Behavior. 1980, 8, 341-351. Ruggerio, F. T . , & Flagg, S. F. Do animals have memory? In D. L. Medin, W . A. Roberts, & R . T . Davis (Eds.), Processes of animal memory Hillsdale, N.J.: Lawrence Erlbaum, 1976. Spyropoulos, T . , & Ceraso, J. Categorized and uncategorized attributes as recall cues: The phenomenon of limited access. Cognitive Psychology, 1977.9, 384-402. Tinklepaugh, 0. L. An experimental study of representative factor in monkeys. Journal of Comparative Psyhology, 1928, 8, 197-236. Treisrnan, A. Focused attention in perception and retrieval of multidimensional stimuli. P erreption & Psyhophysicx, 1977. 22, I - I 1 . Treisman, A., & Gelade, G . A Feature-Integration theory of attention. Cognitive Psychology, 1980, 12, 91-136. Treisman, A., Sykes, M., & Gelade, G. Selective attention and stimulus integration. In S. Dornic (Ed.), A/ren/iotr and performonte V I . Hillsdale, N.J.: Lawrence Erlbaum, 1977. Tulving, E., & Bower, G . H. The logic of memory representations. The Psvchology of Learning and Motivation, 1974, 8, 265-301. Urcuioli, P. J . Transfer of oddity-from-sample performance in pigeons. Journal of the Experimental Analysis of Behavior, 1977, 27, 195-202. Urcuioli, P. J . , & Honig, W. K . Control of choice in conditional discriminations by sample-specific behaviors. Journul of Experimental Psychology: Animal Behavior Processes, 1980, 6 , 25 I 277. Urcuioli, P. J., & Nevin, J . A . Transfer of hue matching in pigeons. Journal of rhe Experimental Analysis of Behavior. 1975, 24, 149-155. Wasserman, E. A. Successive matching-to-sample in the pigeon: Variations on a theme by Konorski. Behavior Reserach Methods and Instrumentation. 1976, 8, 278-282. Wasserman, E. A,, Nelson, K . R., & Larew, M. B. Memory for sequencesof stimuli and responses. Journul nf the Experimenral Analysis of Behavior. 1980, 34, 49-60. Wickens, D. D. Characteristics of word encoding. In A. W. Melton & E. Martin (Eds.), Coding processing in humun memory. Washington, D.C.: Winston, 1972. Wilkie, D. M., Summers, R. J . , & Spetch, M. L. Effect of delay interval stimuli on delayed symbolic matching to sample in the pigeon. Journal .f the Experimental Analysis of Behavior, 1981, 35, 153-160. Zentall, T. R., & Hogan, D. E. Concept learning in the pigeon: Transfer to new matching and non-matching stimuli. American Journal of Psychology, 1974, 88, 233-244. Zentall, T. R., & Hogan, D. E. Pigeons can learn identity or difference, or both. Science, 1976,191, 408-409. Zentall, T. R., & Hogan, D. E . Sameidifferent concept learning in the pigeon: The effect of negative instances and prior adaptation to transfer stimuli. Journal of the Experimental Analysis of Behavior, 1978, 30, 177-189. Zentall, T. R., Hogan, D. E., Howard, M. M., & Moore, B. S. Delayed matching in the pigeon: Effect on performance of sample specific observing responses and differential delay behavior. Learning and Motivation, 1978, 9, 202-2 18.
This Page Intentionally Left Blank
INFERENCES IN INFORMATION PROCESSING' Richard J . Harris KANSAS STATE UNIVERSITY, MANHATTAN, KANSAS
I.
11.
111.
IV.
V.
VI.
82 Inferences and Comprehension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. The Interactionist Position . .. . . . . . . . . . . . . _ _ _ . . . . _ .82 84 ................................ B. Construction of Memory Re 85 .......... C. Developmental Considerations . . . . . . . . . . . . . . . . 86 D. Terminology and Taxonomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Processes of Infemng . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 A. Given-New Strategy ........................................... 89 B . Verification and Comprehension Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 C. Biasing Effects of Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 Themes and Schemas . . . . . . . . . . . . 92 A. The Role ofTheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 B. Schema Theory . .... ................ 95 Schemas and Inferences in Social Cognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 A. Personality and Information Roc 96 B. The Role of Source Information 96 C. Causal Inferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 D. Social Prototypes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 E. Impression Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 F. Audience Effects on Prose Processing . . . . . . . . . . . 99 Varieties of Discourse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 A. Discourse Type and Discourse Force . . . . . . . . . . . . . . . . . . . . . . . . . .... 99 B. Structure of Stones . . . . . . . . . . . . . . . . 101 C. Conversations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 D. The Ecological Validity Question . . . . . . . . . . . . . . 104 E. Judgment Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 107 F. Modality Differences . . . . . . . . . . . . . . I 07 G. Systematic Comparison of Discourse T ............................ .... 108 The Britain Studies.. . . . . . . . . . . . . . . . . . 108 A. Experiment l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 B. Experiment2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 C. Experiment 3 . . . 113 D. Experiment4 .......................................................
'Preparation of this article was supported by a grant from the Kansas State University Bureau of General Research. Thanks are expressed to Tony Dubitsky, Kristin Bruno, Gary Gaeth, Richard Ettenson, and Gregory Monaco for their helpful comments on earlier drafts of the manuscript. Portions of this article were presented at the Southeastern Psychological Association Meeting, Atlanta, March, 1981. 81 THE PSYCHOLOGY OF LEARNING AND MOTIVATION. VOL. I5
Copyright 0 1981 by Academic Ress, Inc. All rights of repmduction in any form resrved. ISBN 0-12-543315-8
Richard J. Harris
82
E. Experinlent 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F. Experiment6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ........... 13 Gcneral Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VII. Conclusion: Inferring the Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .................
113 114
I17 118
I. Inferences and Comprehension We are all very proud of our reason. and Yet we guess at fully one-half we know. ECCLESIASTICUS
The ancient writer of Ecclesiusticus in the Apocrypha was more right than most psychologists might have realized until recently. While the “guesses are typically rather reasonable and predictable inferences, our knowledge is in fact an emergent product of the interaction of stimulus information with our stored knowledge. A major part of this comprehension process is the drawing of inferences, the subject of this article. If truth is beauty, war is hell, and a rose is a rose, then comprehension is drawing inferences. In the 3 years since the “Information Processing between the Lines” paper (Harris & Monaco, 1978), research and theorizing on inferences has bloomed and blossomed into a rich, cognitive garden. ”
A.
THEINTERACTIONKT POSITION
This article will assume an interactionist position on comprehension and memory. As new material is being understood and integrated into memory, it is constantly interacting with information already in memory. The stimulus information, as it is being interpreted and encoded, and the information already stored i n memory are both affected and altered by each other. I . Assirnilutiori und Accoinmodntion
This interaction of stimulus input and the stored knowledge of the comprehender may be viewed in the Piagetian terms of assimilation and accommodation (Waern, 1977a, 1977b). To the extent that our knowledge affects how we comprehend the input, it is assimilation; insofar as prior knowledge is changed by the input, it is accommodation. Just as in activities of the sensorimotor or concrete operations stages, the two pro-
Inferences in Information Processing
83
cesses are simultaneous and complementary in language comprehension. The language comprehender constructs the meaning through assimilation of the stimulus input and accommodation of his or her stored knowledge (Carpenter & Just, 1977; Dooling & Christiaansen, 1977). This derivation of meaning requires two component processes, the application of knowledge and the making of inferences (Warren, Nicholas, & Trabasso, 1979). 2.
What tnferences Do
Inferences fulfill two general functions (Warren et al., 1979). First, they make connections between propositions in the input and between propositions and knowledge already in memory. This allows for integration of new material in memory representations of previously learned information and also helps to provide some organization and structure to the information. Second, inferences fill in empty “slots” in the overall structure. For example, if a strongly implied instrument for some action is not explicitly mentioned, it may be inferred and added to the memory representation just as if it had appeared explicitly (Corbett & Dosher, 1978; M. K . Johnson, Bransford, & Solomon, 1973; Singer, 1979). Although a natural and integral part of the comprehension process, inferences may be drawn especially profusely in response to inadequately developed prose in an attempt to fill the many open slots to make an otherwise elliptical passage more comprehensible (Glenn, 1978). The study of inferences drawn in comprehension has been a major emphasis in cognitive psychology over the past 5 years (Danks & Glucksberg, 1980). In some sense, knowledge that we have in any situation is always incomplete and must be “filled in,” using inferences. As Collins, Warnock, Aiello, and Miller (1975, p. 414) have said, “People can often extract what they do not know explicitly from some forms of implicit knowledge by plausible but uncertain inferences. ” Even very incomplete (or total lack of) knowledge about something can be the basis for an inference, as in answering a question such as 1. Is Burundi a major oil-exporting country?
Upon accessing information (or lack thereof) about Burundi in one’s memory, one may find no information about oil. Thus one may infer that, since there was nothing about that in memory, then probably Burundi is not a major oil-exporting country, since there would probably be something in memory if it were. Although not actively and extensively explored until recently, in-
a4
Richard J . Harris
ferences have long been known to at least some psychologists. An early memory study (Binet & Henri, 1894; see translation by Thieman & Brewer, 1978) observed the drawing of inferences, as did Buhler (1908). In Bartlett’s (1932) classic book on memory he reports examples of memory distortion of both verbal and pictorial material that clearly reflect the use of inferences. B.
CONSTRUCTION OF MEMORY REPRESENTATIONS
Some early studies showed that information appearing in several s e p rate but thematically related sentences is integrated into a single memory representation, which is often difficult to subsequently distinguish fmm the input sentences in later memory tasks (e.g., Bransford & Franks, 1971). Along a similar line, if subjects are presented with Sentences describing transitive or set relations, they construct memory representations and then later are unable to distinguish original input sentences from other sentences equally consistent with the constructed memory representation (Barclay, 1973; Barclay & Reid, 1974; Bransford, Barclay, & Franks, 1972). These constructed memory representations often contain inferred material that has been totally integrated with the information stated directly (Brewer, 1977; Dawes, 1966; Hanis, 1974, 1977; Harris & Monaco, 1978; M. K . Johnson et ul., 1973; Loftus & Palmer, 1974; Marschark & Paivio, 1977; Spiro, 1980). Some of the earliest modem studies to demonstrate inferential processes were sentence-memory studies. In these studies subjects heard lists of sentences and later remembered sentences that were in fact implications inferred from the stimulus material. This was a robust effect occurring in recall (Brewer, 1977), recognition memory (M. K. Johnson et ul., 1973), and truth-value judgment (Harris, 1974). It was later demonstrated that subjects also remember the speaker’s implicit intention (Jarvella & Collas, 1974; Kemper, 1980; Schweller, Brewer, & Dahl, 1976) and may also confuse this type of information with the original stimulus (e.g., remembering tulking about high taxes as complaining about high taxes). In a study of the effects of the context of comprehension on memory, R . C. Anderson and Ortony (1975) found that cues implicit in the input served as effective retrieval prompts in cued recall, e.g., bottle for (2a) but basket for (2b): 2a. The container held the cola. 2b. The container held the apples.
Inferences in Information Processing
85
These data may be interpreted as reflecting that the type of container was inferred in comprehension in light of the context, though there are other interpretations. This effect was replicated with abstract sentences by Marschark and Paivio (1977) and Masson (1979). An important but largely unexplored question is how many of all the possible inferences are typically made in comprehension. Clearly it is not tenable to propose that all possible inferences are routinely made from a linguistic input. Just as clearly, however, many are. Warren et al. (1979) and Goetz (1979) suggest that a comprehender makes only those inferences that are important and relevant to the progress of the discourse, i.e., that establish information necessary to understand what happened and why. To examine this question, Corbett and Dosher (1978) had subjects read one of a trio of sentences such as 3. The athlete cut out an article with scissors. 4.The athlete cut out an article with a razor blade. 5. The athlete cut out an article for his friend. In a cued-recall task, the high-probability instrument (e.g., scissors) was an effective recall prompt not only for the case where it had been present at encoding (3), but also for cases where subjects had heard a lowprobability instrument (4)or no explicit instrument (5). However, subjects were able to discriminate whether the instrument had been presented at encoding. The authors conclude that such implicit instruments are not necessarily inferred at encoding, though they may be. Even if they are not, however, they may still be effective retrieval cues. C. DEVELOPMENTAL CONSIDERATIONS There have been a number of studies of inference-drawing in children as young as nursery-school age (Brown, 1976; Bruno, 1980; Freedle & Hale, 1979; Glenn, 1978; Harris, 1975; Hildyard, 1979; Hildyard & Olson, 1978; Paris & Carter, 1973; Paris & Lindauer, 1976; Paris & Upton, 1976; Stein & Glenn, 1979). Although the tasks have differed considerably and results have not been totally consistent, it is clear that children also draw inferences much as adults do, and in many cases they are even more prone to do so. At the other end of life, there is some evidence that such inference processes may be some of the first cognitive abilities to deteriorate in old age (Cohen, 1979). Developmental aspects of inference-drawing will not be considered further in this paper, however.
Richard J. Harris
86
D.
TERMINOLOGY A N D TAXONOMY
I.
Terms in the Literature
One of the major difficulties in dealing with inferences is the fact that there is no consistent terminology or taxonomy in the literature. Typically there is a distinction made between inferences that must be true and those that are strongly implied but need not necessarily be true. Harris and Monaco (1978) called these two types logical and pragmatic; Chaffin (1979) referred to the same two inference classes as necessary and invired, both types in contrast to directly asserted information. Kintsch (1 974) and others more informally have merely referred to information that is explicit or imp/icif in the input. Hildyard and Olson (1978) use pragmatic inference roughly as do Harris and Monaco, but they call the necessary type of inference proposifional and also discuss a third kind, enabling inferences, which allow ongoing comprehension to occur by relating concepts or events in the discourse (e.g., anaphoric and pronominal reference). Singer (1976) refers to pragmatic inferences as context inferences. Warren et a/. (1979) break inferences into three classes: logical, informarional, and value. Using a more pragmatic, speech-act rationale, Clark (1 977) distinguishes between authorized inferences, which the speaker or writer intended the comprehender to draw, and unauthorized inferences, inferred by the comprehender but without the sender’s intent. In empirically developing a taxonomy of inferences based on responses on a story-memory task, Owens, Bower, and Black (1979) identified five types of inferences. First were gap-filling inferences, reconstructing information implied but not explicitly present in the story. A second type specified how one event in a sequence enabled another, e.g., remembering Nancy boiled some water as Nancy turned on the stove to boil some wafer. Third, subjects inferred the location or a referent. Fourth, they produced superordinate or summary action statements. i .e., compressed several specific actions into a general statement of the overall activity. Finally, about half of Owens et al.’s inferences involved inferred motives. Some of these involved a thematic interpretation of ambiguous statements, e.g., remembering “usual medical procedures” as “pregnancy tests” after being told the character was concerned about being pregnant. Other motive inferences involved inferences about thoughts, motives, and feelings of the character. These will be discussed later in the section on social cognition.
Inferences in Information Processing
87
2 . A Matter of Degree? Although it is an idea that is probably frequently assumed, Singer (1979) was one of the few to explicitly propose and test the notion of a continuum or levels of inferences strength, for example
6. The 7. The 8. The 9. The 10. The
boy boy boy boy boy
cleared the snow with a shovel. cleared the snow from the stairs. cleared a path on the stairs. cleared the bottles from the stairs. used a shovel.
The information in (10) is directly stated in (6), very strongly implied in (7), somewhat weakly implied by (8), and very weakly if at all implied by (9). Although an intuitively appealing and probably correct notion, there has been almost no research testing this notion of a continuum of inference strength. Warren et al. (1979) have categorized “degrees of inferential constraint. ’’ A first-order inference would be totally undetermined; i.e., it would not have to be made to allow meaningful processing of the material. For example, inferences elaborating details of objects, persons, or events mentioned would be first-order inferences. Second-order inferences are determined (i.e., must be true) but are irrelevant to comprehension-e.g. upon hearing mention of a room, inferring it had walls and a ceiling. Third-order inferences are both determined and relevant and for that reason are perhaps the most important to the comprehension process. They must be made to comprehend the flow of the story. Fourth-order inferences are overconstrained and redundant, essentially duplicating information explicitly stated or implied earlier.
3. A Working Definition For the purpose of discussion, the term inference will be used generally for any construction of meaning that a hearer or reader draws from a passage when he or she goes beyond what is explicitly given. For the most part inferences will be considered to have a probabilistic character, in that they may be true but do not necessarily have to be so, in contrast to directly asserted information. Unless specifically noted as such, the term inference will not refer here to the logically necessary type.
88
Richard J. Harris
11. Processes of Inferring One of the most basic but most difficult questions in studying inferences is determining exactly how they are constructed in comprehension. The next section of the article explores work on the actual processes involved in drawing inferences from spoken and written language. A.
GIVEN-NEWSTRATEGY
Haviland and Clark (1974; Clark, 1977; Clark & Haviland, 1977) have proposed the Given-New strategy, one of the few specific models of how information is inferred and added to memory during the process of comprehension. In every utterance that we make, certain information is given-i.e., we assume that the hearer is already aware of it-and other information is new-i.e., we assume the hearer does not already know it. (The same principle holds for written language.) Given information appears in certain characteristic syntactic structures in the sentence (e.g., prenominal adjectives, relative clauses, noun phrases starting with the definite article the), whereas new information typically occurs in other syntactic constructions (e.g., predicate adjectives, direct objects, noun phrases starting with the indefinite article a , postverbal prepositional phrases). The information in the two sentences below is identical except for the fact that what is given in (1 1) is new in (12), and vice versa. 1 1 . The cat on the porch is skinny. 12. The skinny cat is on the porch.
In (1 1) the speaker assumes that the hearer knows there is a cat on the porch and is commenting that this cat is skinny, whereas for (12) the assumption is that the hearer knows of a skinny cat and is being told where this cat is. There is a three-stage process in using the Given-New strategy to compute antecedents and add them to memory. First, the input sentence must be separated into given and new information. Because of the characteristic syntactic structures that typically carry each type of information, the surface structure provides some strong cues to this. Second, memory must be searched for a unique antecedent for the given information. For example, upon hearing (1 1) we would search for information in memory about a cat on the porch, whereas for (12) we would search for a skinny cat. Finally, the new information in the input sentence would then be added to the structure in memory, thus altering that encoded representation.
Inferences in Information Processing
89
The second stage in the process can be extended if no unique antecedent can be readily computed. In such cases, and they are common, additional inferences must be made. Haviland and Clark (1974) have shown that such inferences take time, which is measurable in a reactiontime study. It takes longer to comprehend the second sentence in a twosentence pair if the antecedent is not explicitly stated in the first sentence, e.g., having to infer in (14) that beer must have been among the picnic supplies. 13. The picnic supplies were in the trunk. 14. The beer was warm. B.
VERIFICATION A N D COMPREHENSION STUDIES
Inference processes have been very widely studied by using reactiontime methodology, usually involving the logic that the more inferences that are required, the longer comprehension will take. Several studies have measured reaction time to comprehend or verify inferences drawn from text (Chaffin, 1979; Garrod & Sanford, 1977; Haberlandt & Bingham, 1978; Hayes-Roth & Thorndyke, 1979; HayesRoth & Walker, 1979; Kintsch, 1974, 1976; McKoon, 1977; Singer, 1976, 1979; Springston & Clark, 1973; Walker & Meyer, 1980a; Yekovich & Walker, 1978; Yekovich, Walker, & Blackman, 1979).
1 . Computing Anuphoric Reference Some of these studies deal with inferences made to compute anaphoric reference, the “process of referring back to previously introduced concepts for purposes of restatement or elaboration” (Yekovich & Walker, 1978, p. 265). For example, Garrod and Sanford (1977) found that the reading time for (1 6) was faster if the first sentence ( I 5) contained bus, a more typical vehicle (cf. Smith, Shoben, & Rips, 1974), 15. A bus came roaring around the comer. 16. The vehicle narrowly missed the pedestrian. than if tank were the sentential subject. They concluded that the anaphore vehicle was identifed as bus (or tank) during the reading process, but that it took longer to identify it with tank, a fact that was reflected in the longer reading time for that condition. In explaining this type of phenomenon, Haberlandt and Bingham
Richard J. Harris
90
( I 978) claim that certain inferences are activated by the first sentence in the set, and that subsequent sentences are processed faster if their content is consistent with the inferences triggered by the previous sentence. The less consistent the content i s - e . g . , tank is less consistent with vehicle than bus is-the more inferring is necessary, and the longer time comprehension thus takes. In a similar vein, Chaffin (1979) argues that the less strongly implied an implication is, the more world knowledge must be brought to bear in order to verify it, a process that can add time to the comprehension process.
2.
Orher Rrlrvcitir Vnriahles
Other variables found to influence such comprehension/integration time include the temporal proximity and similarity of wording of the two sentences (Hayes-Roth & Thorndyke, 1979), position in the hierarchical structure of the prose (McKoon. 1977; Walker tk Meyer, 1980a), the appropriateness of presentation of given vs. new parts of the sentence (Haviland & Clark, 1974; Yekovich et a l . , 1979), reading strategy used by subjects (Chaffin, 1979), the number of different arguments (Kintsch, 1976), degree o f strength of the inference (Singer, 1979), overall semantic cohesiveness of the sentences (Haberlandt & Bingham, 1978), and the relationship to a formal logic model (Springston & Clark, 1973). For a recent critical review of integration studies, see Walker and Meyer (1980b). 3. Sentence Verification
A closely related line of research has used reaction-time methodology to measure the latency to verify the truth of a sentence either asserted or implied (explicitly or implicitly stated) to be true in a prior sentence. Kintsch (1974, 1976) found that implicit true inferences, tested immediately after the stimulus, took longer to verify than did explicit ones but that after a 15-min delay there was no difference, suggesting that, after the inference is incorporated into the memory representation, there is no longer any mental distinction between asserted and implied information. Baggett (1975) replicated this finding, using pictorial materials put together to form a story. Singer (1976) found a difference when the test followed the antecedent by 9.5 sec, although when they were very close (0-2.5 sec) there was little difference in verification time. McKoon (1977) found a shorter latency and greater accuracy in verifying sentences testing topic information than for those testing detail information, but only after a delay of .5 hr; there was no difference with immediate testing.
Inferences in Information Processing
C.
91
BIASING EFFECTSOF QUESTIONS
A somewhat different approach to studying the processes involved in integrating new information into memory has been taken primarily by Elizabeth Loftus and her colleagues in an interesting and important series of studies on the biasing effects of questions on eyewitness memory. Merely asking (17) instead of (18) elicited more false recognitions of remembering nonexistent broken headlights from a film of an auto accident (Loftus & Zanni, 1975). 17. Did you see the broken headlight'? 18. Did you see a broken headlight? The definite article, usually a syntactic signal of given information, is used inappropriately in (17), thus placing false information in the given format. Loftus and Palmer (1974) used the same procedure and asked subjects either (19) or (20). 19. About how fast were the cars going when they hit each other? 20. About how fast were the cars going when they smashed into each other? Subjects questioned with (20) reported a faster mean speed estimate and 1 week later were more likely to remember having seen nonexistent broken glass. Similarly, word choice for the name of a linear scale can imply the size of the range of numerical answers to a question; e.g., How short was the rnan? implies that the man was in the short range, whereas How tall was the man? does not imply an answer in any particular range of the scale (Harris, 1973). Thus, merely the choice of a word can imply information affecting an immediate judgment and also be encoded into memory, where it may be later retrieved. implying false information through questions may happen more generally. Loftus, Altman, and Geballe (1975), Loftus (1975), and Loftus, Miller, and Burns (1978) introduced false information covertly through the wording of the questions about films of classroom disruptions and auto accidents and found that it affected subjects' later recall and recognition. Kasprzyk, Montano, and Loftus (1975) showed that such biased questions can go beyond affecting memory, in that they can increase the number of guilty verdicts selected by subject-jurors in a simulated trial situation. Lipton (1977) showed that the least structured questions elicit less (but more accurate) information, whereas more highly structured questions elicit more information, but more open to distortion. Powers,
Richard J. Harris
92
Andriks, and Loftus (1979) showed that male and female eyewitnesses drew inferences differentially , depending on whether observed characters were of the same or opposite sex.
111. Themes and Schemas One of the most important determinants of what inferences are drawn is the overall theme of the material. A comprehender’s expectation of what the passage is about can serve as a guiding force in directing inference construction. A.
THEROLEOF THEME
The overall theme of a passage can markedly affect the inferences we draw from that passage as well as almost every other aspect of comprehending it. Some years ago, several studies showed that it was very difficult to understand or remember much from prose that apparently had no theme (Bransford & Johnson, 1972; Dooling & Mullet, 1973; M . K. Johnson, Doll, Bransford, & Lapinski, 1974); this effect was replicated pictorially with cartoons (Bower, Karlin, & Dueck, 1975; Bransford & McCarrell, 1974). Supporting information is remembered better if it is relevant to the overall theme (Bransford & Johnson, 1973; Kozminsky, 1977) or to prior world knowledge (Moms, Stein, & Bransford, 1979), and new but thematically consistent material is often falsely recognized as having been previously presented (Kennedy, 1973; Sulin & Dooling, 1974).
I.
Prior Knowledge
Inferences drawn are dependent on one’s prior knowledge of the general semantic domain that is the topic of the prose. For example, subjects with a high degree of knowledge of baseball remembered more from a summary of a baseball inning, which they were better able to integrate with a goal structure in their stored knowledge of the game of baseball (Chiesi, Spilich, & Voss, 1979; Spilich, Vesonder, Chiesi, & Voss, 1979). In a similar vein, Vesonder (1980) had science and nonscience students read articles from Science magazine, edited to eliminate jargon. In a subsequent free recall task, scientists recalled more material than nonscientists did, presumably owing to the use of their stored scientific knowledge to organize the new information around. Kintsch and Vipond (1979) suggest that a crucial role that such background knowledge plays
Inferences in Information Processing
93
is in allowing us to construct more and better inferences during comprehension. Prior knowledge can also come in the form of a point of view, which then helps organize information thematically as it is comprehended. For example, Pichert and Anderson (1977) and R. C. Anderson and Pichert (1978) gave subjects a story to read about two boys playing in a house. One group was told to read the story from the point of view of a burglar considering robbing the place; these subjects tended to remember details about valuable objects, isolation from surrounding houses, and other such details of relevance to a potential burglar. The other group of subjects read the story from the point of view of a real estate salesperson; these subjects remembered details such as size and number of rooms, condition of the house, and quality of the yard. A useful analogy can also be a helpful device for structuring information thematically. For example, Schustack and Anderson (1979) had subjects study short biographies of fictitious characters with strong parallels to real people. When the name of the modern analogical person was given at both study and test time, recognition memory was improved. 2 . Levels of Processing
Thematic effects seem to occur only during rather deep semantic processing (Craik & Lockhart, 1972). For example, Britton, Meyer, Simpson, Holdredge, and Curry ( I 979) measured latency to detect a click while reading prose passages. Latencies, assumed to be longer when greater cognitive capacity was required, were longer if a thematic title was presented but only if that title increased the meaning of the passage. In a systematic study using four tasks differing in the level of processing required, Schallert (1976) presented subjects with stories with two possible alternative themes. Presence of a thematic title improved memory but only if the ongoing task during reading involved relatively deep processing. The presence of a relevant context can also affect the processes that occur. For example, Clark and Lucy (1975) concluded that conversationally conveyed requests like (21) are comprehended via a two-stage process of computing first the literal meaning and only later, if necessary, the conveyed meaning. However, Gibbs ( 1 979) found that presenting requests like (21) but preceded by the appropriate context for 21. Why don’t you color the circle blue? the conveyed meaning reversed the results, with that conveyed meaning being computed faster, thus showing the importance of context in guiding
Richard J. Harris
94
the comprehension process. Subjects apparently inferred the appropriate context immediately and thus did not even have to deal with processing at the literal level.
3. Cultural Expectations The input itself interacts with stored knowledge, and this knowledge includes culture-specific expectations. Although this issue was discussed by Bartlett (1932), there has not been much cross-cultural research on this topic. In one exception, Tannen (1979) examined the expectations of Greeks and Americans in viewing films and then telling a third party about the film. Tannen noticed that the two cultural groups drew different inferences. Although they used only American subjects, Kintsch and Greene ( I 978) found better recall and better writing of summaries of stories from our culture than of those of Alaskan Indian myths or Apache folk tales. B.
SCHEMA THEORY
An important development in the last 5 years or so in cognitive psychology has been the growth and proliferation of schema theory. A schema is “a unit of organized knowledge about events, situations, or objects” (Moates & Schumacher, 1980, p. 33). Two other closely related terms,frume (Minsky, 1975) and script (Bower, Black, & Turner, 1979; Schank & Abelson, 1977), are very similar in both definition and application, and the three will be considered together here as schema theory. A fundamental assumption of such an approach is that spoken or written text does not in itself carry meaning but rather provides directions for listeners or readers on how to use their own stored knowledge to retrieve and construct the meaning (Adams & Collins, 1979, p. 3). The goal of schema theory is thus to specify the interface between the comprehender and the text. This involves both bottom-up (data-driven) and top-down (conceptually driven) processes, with knowledge and input interacting at all levels in the process of reading or listening, e.g., word and letter level, syntactic, semantic, and interpretive levels. For a thorough critique of schema theory in cognitive psychology, see Thorndyke and Yekovich ( 1980) and Rumelhart (1 980).
I.
Scripts
Owens et al. (1979, p. 190) define a script as “a memory structure representing a person’s generic knowledge of a stereotyped activity
”
Inferences in Information Processing
95
(e.g., eating a meal in a restaurant, visiting a doctor). The knowledge of this activity can then be used to guide the inference processes operating during comprehension. Inferences consistent with the script, or those filling in slots with information not explicitly stated, are especially likely to be drawn. For the most detailed theoretical development of scripts in psychology, see Bower ef al. (1979).
2. Extensions to Perception There have been attempts to extend schema-frame-script theory beyond the linguistic realm, Minsky ’s (1975) theoretical article on frames deals with verbal but also visual perceptual information, both of which he claims have a common underlying representational system (i.e., frames). Friedman (1979) has developed the visual perception aspect further by sketching a frame theory approach to perception and studying recognition memory for pictures. Even Bartlett (1932) considered verbal, pictorial, and social schemas.
IV.
Schemas and Inferences in Social Cognition
Schema theory has had broad application beyond the area of prose processing. One of the most currently active areas of studying schemas and inferences is in social cognition (Taylor & Crocker, 1981), growing out of the recent marriage of information processing and social psychology. There have also been other recent attempts to study inferences in other social-personality contexts. These areas will be examined first. A.
PERSONALITY A N D INFORMATIONPROCESSING
There have been a number of recent calls for greater communication between cognitive and personality and social psychology (Bower, 1978; Kanouse, 1971; Keenan & Baillet, 1979; Mischel, 1979; Nisbett & Ross, 1980; Sentis & Burnstein, 1979; Snyder, 1981; Zajonc, 1980). Even personality traits can interact with cognition. For example, Zehr and Kimble ( 1 980) found that highly self-conscious subjects remembered less about people encountered in a social situation than did low self-conscious subjects. Memory is enhanced by relating new information to one’s own experience. For example, Rogers, Kuipers, and Kirker ( 1977) and Bower and Gilligan (1979) found that subjects remembered trait adjectives and brief events better if they related them to themselves and their own selfconcept or personal experience. Such “personal schemas ” can guide inference processes in social situations.
Richard J. Harris
96
B . THEROLEOF SOURCE INFORMATION
In their Annual Review articles on attitude change, Eagly and Himmelfarb (1978) and Cialdini, Petty, and Cacioppo (1981) comment on the importance of information processing in understanding attitude change, one of the traditional subareas of social psychology. An example of this cognitive-social courtship can be seen in an extensive and thorough theoretical and empirical article developing and testing mathematical models of how source credibility information is integrated in attitude formation (Bimbaum & Stegner, 1979). Subjects judged the value of several used cars, given one estimate of the price while systematically varying the source of that estimate (friend of buyer, friend of seller, or neutral), his competence (high, medium, or low), and the point of view the subject was given (buyer, seller, or neutral). Results from five experiments generally showed a stronger bias effect of whether the estimator was a friend of the buyer or seller at high-competence levels than at low levels; i.e., the bias mattered much less if the source competence was low than if it was high. Bimbaum and Stegner conclude that the bestsupported information processing model of the effect of bias is that the bias of the source causes a shift in the value of information provided by that source. In a study relating source credibility and memory, Monaco (1979) had subjects read two paragraphs about Grover Cleveland. Half the subjects were told that the author was a professor of history, and half were told that the passage was written on an exam by a freshman history student in a U . S . History class. Subjects then rated the passages on a number of dimensions and had a final incidental-learning free recall test. Results showed low-source-competence subjects rated the passages as showing poorer organization and having a duller writing style than did highsource-competence subjects. Similarly, more propositions were recalled correctly by the high-source-competence group than by the low; this could not have been due to attentional differences, since neither group was expecting a recall test. Similarly, Wilkes and Alred (1978) had subjects read some positive or negative prior information about a student’s adventures in secondary school and higher education, followed by a target paragraph of his childhood experiences. Results showed that interpretation of the target was biased by the polarity of the prior information. C.
CAUSAL. INFERENCES
One of the most common types of social inferences made is that of infemng a causal relationship of events. Nisbett and Wilson (1977) say
hferences in Information Processing
91
that we often do so because it is very difficult to report accurately on the effects of stimuli on complex inference-based responses. Tversky and Kahneman (1977) found that evidence that is readily incorporable into some causal schema is weighted disproportionately heavily, relative to other important but noncausal information, since this causal schema guides the judgment and gives it direction. Owens et al. (1979) conducted an interesting causal attribution study about how the inference of a motive for behavior can guide the inferences drawn from a person’s actions. Subjects read a story about Jack or Nancy (same story except for name) doing a series of routine actions such as fixing coffee, attending a class, and visiting a doctor. Half the subjects (the “context subjects ”) received initial information describing that context and providing a motive for later behavior (e.g., Nancy’s just discovering that she was pregnant after having an affair with a professor). On a free recall task 30 minutes later, the context subjects recalled more episodes and in better order than did subjects not receiving the context. They also made more intrusions representing distortions to fit the inferred motive schema (e.g., remembering “usual medical procedures” as ‘‘pregnancy tests”). Owens et al. concluded that the knowledge of a motive helps integrate episodes into a coherent whole in comprehension. D.
SOCIAL PROTOTYPES
Some studies of social cognition by Mark Snyder and his associates (Snyder, 1981; Snyder & Cantor, 1979; Snyder & Swann, 1978) concluded that people select and test information about others i n attempting to confirm rather than disconfirm some hypothesis (e.g., initial profile of an extrovert). They also selectively remember information consistent with the hypothesis. In this act of seeking to confirm rather than disconfirm hypotheses about behavior, there is a striking parallel to people’s behavior in deductive reasoning situations (Wason & Johnson-Laird, 1972). In fact, a hypothesis can be accurately tested faster by seeking disconfirming information, but we tend instead to seek confirming information. Cantor and Mischel (1977, 1979) report that subjects falsely recognized sentences consistent with prior personality descriptions (e.g., extrovert), whether such a description was given explicitly or only implicitly (i.e., implied). The traits thus function as personality prototypes affecting the organization of information. In this sense they are like the scripts of Bower et af. (1979), the themes of Bransford and Johnson (1973), or the perceptual prototypes of Posner and Keele (1968). All guide inference processes and bias comprehension and memory in the direction of the prototype.
98
Richard J. Harris
These social hypotheses being tested can also draw in part on stereotyped beliefs. For example, Snyder and Uranowitz (1978) had subjects read a story about “Betty K.,” followed by information that Betty was heterosexual, homosexual, or no additional information. One week later subjects answering multiple-choice questions about events in Betty’s life selectively affirmed events consistent with the stereotypical ideas of her given sexual preference. E.
IMPRESSION FORMATION
Many cognitively oriented social psychologists are studying impression formation using memory measures and focusing on cognitive inference processes (Cantor & Mischel, 1977; Hamilton, Katz, & h i r e r , 1980; Hastie & Kumar, 1979; Lingle, Geva, Ostrotn, Leippe, & Baumgardner, 1979; Lingle & Ostrom, 1979; Rothbart, Evans, & Fulero, 1979; Rothbart, Fulero, Jensen, Howard, & Birrell, 1978; Snyder & Uranowitz, 1978; Tsujimoto, 1978; Woll, Weeks, Fraps, Pendergrass, & Vanderplas, 1980; Woll & Yopp, 1978). The focus of these studies is on how information is combined to form an impression of a person. Although a social process, it is one of information processing as well. One area where impression formation takes on considerable practical significance is the case where jurors form impressions of witnesses in a courtroom trial. Conleq, O’Barr, and Lind (1978) and Erickson, Lind, Johnson, and O’Barr (1978) have shown that the speech style of a witness may imply much about his or her character, and this in turn affects what is encoded into a juror’s memory, which later affects a verdict decision. F.
AUDIENCE EFFECTS ON PROSEPKOCESSING
Although the rhetorical situation has seldom been considered seriously by psycholinguistic models of prose processing, Freedle, Naus, and Schwartz (1977) present the outlines of a model of prose processing from a psychosocial perspective. They argue that characteristics of the speaker and hearer are important to a prose processing model. To illustrate the importance of these concepts, subjects were presented with two stories, one written for a child and one for an adult. The subjects then retold these stories to a child or adult. Freedle ef al. found that the original stories were markedly transformed if the hypothetical audience was changed for the retelling. For example, the adult story was greatly simplified if told to a child, and the children’s story was made more complex (e.g., reflecting additional inferences of motives) if retold to an adult.
Inferences in Information Processing
V.
99
Varieties of Discourse
Returning to linguistic materials, we shall examine differences among diverse types of prose materials and their effects on inference processes, in preparation for some experiments examining these issues. A.
DISCOURSE TYPEA N D DISCOURSE FORCE
In an important theoretical article developing a psychologically relevant taxonomy of discourse types, Brewer (1980) has suggested that there are three general types of written discourse, distinguished by different cognitive structures underlying each type. Descriptive discourse deals with a stationary perceptual scene. Narrative discourse describes a series of events occurring in time and related through some thematic or causal chain. Finally, expository discourse represents underlying abstract logical processes. As well as discourse types as such, however, Brewer discusses four types of discourse force, analogous to speech-act force of sentences (Austin, 1962; Searle, 1969). The discourse force is an interaction of the communicative intent of the language producer and the perception of the comprehender. Brewer’s four types of discourse force are informative, entertaining, persuasive, and literary-esthetic . Each of the three types of discourse may occur with any of the four types of discourse force. In fact, a given piece of prose may have more than one discourse force, e.g., a piece of satire that both entertains and persuades. This taxonomy, while probably incomplete in places, offers a very useful framework for beginning to investigate varieties of discourse, most of which, except for narratives, have not been extensively studied. B.
STRUCTURE OF STORIES
I.
Episodes
Surely the type of discourse most carefully studied has been the story, or narrative (Black & Wilensky, 1979; Bower, 1976; N. S. Johnson & Mandler, 1980; Kintsch, 1977, Kintsch & van Dijk, 1978; Mandler & Johnson, 1977; Rumelhart, 1975, 1977; Stein & Glenn, 1979; Thorndyke, 1977). Structural units of stories are called episodes, which may be divided into expositions, complications, and resolutions. The comprehender uses this structural information or story schema to interpret
Richard J. Harris
100
stones by expecting them to contain episodes with expositions, complications, and resolutions. Among other influences, the story schema affects the type and nature of inferences drawn, to “fill in the gaps” in the story structure. The story itself does not have to overtly specify the causal connections; these are “automatically supplied by the listener” (Mandler & Johnson, 1977, p. 131). This “automatic supplying” of information is the inference process. Episodes may also serve as units of processing for memory encoding (Black & Bower, 1979). 2.
Macrostructures
In a series of articles, Kintsch (1974, 1976, 1977; Kintsch & van Dijk, 1975, 1978; Kintsch & Vipond, 1979; Miller & Kintsch, 1980; see also Vipond, 1980) has developed a text base theory with semantic units called propositions, an overriding general organization called a macrostructure, and a more specific organization of individual ideas called a microstructure. Stories are made up of episodes, which in turn typically have expositions, complications, and resolutions. Episodes may be embedded i n one another, as in the case where the complication of one episode is actually an entire second episode in itself. This macrostructure thus organizes the text into subunits. This organization guides comprehension, including inferential processes to fill in expected information not explicitly present. As an example of the use of macrostructures, Bower (1974; see also Thorndyke, 1977) had subjects learn a short biographical story and two additional passages with the same macrostructure but different details. In testing memory for the first story, subjects showed a positive transfer of the overall macrostructure and the unchanged details but a negative transfer of details that differed in the later passages.
3. Hierarchical Model The importance of the organization of prose material has been stressed by Meyer (1975, 1977). She has developed a hierarchical model of prose processing, whereby information higher in the hierarchy, i.e., structurally and thematically more important information, takes a primary role in comprehension and memory and tends to be remembered better. It is, however, very difficult to manipulate hierarchical position while holding content constant (cf. Monaco & Harris, 1978). 4.
Other Approaches
Graesser and his colleagues (Graesser, 1978; Graesser, Robertson, & Anderson, 1981; Graesser, Robertson, Lovelace, & Swinehart, 1980)
Inferences in Information Processing
101
have used the technique of asking why and how questions about narratives and common activities to expose the organization of plot and actions in the underlying structure. They have used this technique to study how inferences are incorporated in memory representations of prose. They suggest that this may be a useful way to attack unanswered questions such as the psychological differences of various prose types and the precise processes of construction of memory representations for prose. There have been other models of prose processing, some of which have tried to extend their generality-e.g., Schank’s (1972) conceptual dependency model, Warren et al’s (1979) goal-oriented model, Frederiksen’s (1975, 1977) model of various semantic processing units in comprehension, Black and Bower’s (1980) hierarchical state transition network model, and Clements’ (1979) notion of staging, which identifies relative prominence given to various segments of discourse. Many of the major points of these models are very similar to those already discussed. C. CONVERSATIONS Another type of discourse where the study of inference has received some attention is conversations, although the approaches have been highly varied and often difficult to compare.
I.
Differencesfrom Other Discourse
In a theoretical article on conversations, Schank (1977) discusses rules and topics in conversations. He also identifies supertopics, the superordinate subjects being discussed, and metatopics, i.e., inferred comments or inferred motives. Because it is a social as well as a psycholinguistic act, there may be more metatopics in conversations than in other types of discourse. In highlighting differences of conversations and other discourse, Bates, Masling, and Kintsch (1978) tested subjects’ recognition memory for a 20-min videotaped soap opera, chosen for its resemblance to natural conversation. They found better memory for surface structure information than in laboratory studies and concluded that in natural settings the surface form may in itself be an important part (or even the whole point) of the utterance, rather than merely a vehicle for carrying the meaning. Keenan, MacWhinney, and Mayhew (1977) found similar results comparing memory for statements either high or low in “interactional content,” which they define as “information about the speaker’s intention, his beliefs, and his relations with the listener” (p. 550). Using statements occurring in natural conversation at an informal lunchtime meeting,
Richard J. Harris
I02
Keenan et al. found better memory for both form and content for statements high in interactional content than for those low in it, with memory for form being surprisingly good. In natural conversations there may be more opportunities for inference than in single-author prose or certainly than in reading. In addition to text cues, there are all the pragmatic conventions and rules of conversational implicature (Grice, 1975) covering expectations brought to the dialogue. 2.
Thernes in Conversations
As in other types of prose, theme plays a major role in conversations. In a reaction-time study of inferences drawn in conversations, Dubitsky (1 980) found that subjects took longer to verify inferences inconsistent with the context than those consistent with it. Subjects listened to a series of dialogues that were topically ambiguous (e.g., could be interpreted as talking about either archery or playing the violin). After listening to each dialogue, subjects judged the truth value of inferences either consistent or inconsistent with the topical interpretation that the subject made. Inferences consistent with the theme were verified more quickly than were those not consistent with it, thus replicating the facilitative effect of contextual information on the comprehension of stories. 3.
Courtrooin Testimony
The inference-drawing phenomenon has been illustrated with a specialized kind of conversation by Harris, Teske, and Cinns (1975) and Harris (1978). These studies had subjects listen to an excerpt from courtroom testimony where the witness either stated something directly, e.g. , (22), or merely implied it, e.g. (23). 22. I picked up the phone and called Mr. Lacey. 23. I picked up the phone to call Mr. Lacey.
Subjects simulating jurors consistently remembered the inference in both cases, with no apparent realization that it had only been implied in one condition. This occurred both immediately and 2 days later, and in a condition where subjects were specifically warned not to make unwarranted inferences. Findings were replicated when subjects met in groups of two or three to decide on a single group response. 4.
Sociolinguistic Approaches
More sociolinguistic approaches to conversation have been taken by others. For example, Weiner and Goodenough (1 977) and Goodenough
Inferences in Information Processing
103
and Weiner (1978) have developed a game model of conversation, which organizes comments into units called “moves,” some of which may pertain directly to the topic being discussed, and others that are signals for the other person to begin speaking, change the subject, agree with the other speaker’s expressed desire to change the topic, etc., such as “okay” or “You agree, right?” Stiles, Waszak, and Barton (1979) coded and scored naturally occurring class and laboratory conversations on a dimension they called “presumptuousness, defined as ‘‘the expression of understanding of another’s experience or viewpoint in social interaction” (p. 158). Results showed that professors ’ speech was more presumptuous than students ’, suggesting that the privilege of overtly presuming and stating knowledge of another person is a prerogative of social rank-e.g., “I can see why you feel upset,” “You must be reaily happy about that.” ”
D.
THEECOLOGICAL VALIDITY QUESTION
Studies of prose and inferences have increasingly taken seriously the application of their findings outside the laboratory. 1.
“Newspaper Studies”
The importance of examining the ecological validity of prose studies was illustrated in creative studies by Graesser, Higgenbotham, Robertson, and Smith (1978) and Thorndyke (1979), both of which used newspaper articles as stimuli. Reading the National Enquirer was examined by Graesser et al. under two conditions-ne where the subject spontaneously picked up the paper in a waiting room with no knowledge of being in an experiment, and the other where the subject was asked to read the newspaper as part of an experiment. Subjects in the former group tended to concentrate on more active, narrative parts of articles and select articles on familiar topics, whereas the latter group focused more evenly on the material, although there was no difference in the number of ideas they chose to read overall. Thorndyke (1979) used current events and feature stories from The New York Times and The Los Angeles Times as stimuli. Recall was significantly lower for the original story than for condensed, narrative, or outline versions of the same material, although the reading time per proposition was fastest for the original news story. 2.
Surface Structure Salience
There is some evidence that people remember more of the surface structure of language in naturalistic settings than has been observed in
Richard J. Harris
104
strictly laboratory studies (Bates et al., 1978; Keenan et al., 1977). For example, Kintsch and Bates (1977) tested subjects’ recognition memory for statements made in the classroom lecture. Memory for the exact wording was stronger after both 2 and 5 days than was the case in laboratory studies. This was especially true for jokes, announcements, and other statements that were not part of the overall structure of the discourse. Such information was actually remembered better than were statements of either topics or details, which did not differ from each other, as one would expect from hierarchical models of prose processing (e.g., Meyer, 1975), where topics would occupy a higher position in the hierarchy than details.
3. Inferences in Advertising In an application of cognitive psychology to consumer research, Harris and colleagues (Bruno & Harris, 1980; Hams, 1977; Hams, Dubitsky, & Bruno, 1982; Harris, Dubitsky, Connizzo, Letcher, & Ellerman, 198lb; Harris, Dubitsky, Perch, Ellerman, & Larson, 1980; Harris, Dubitsky, & Thompson, 1979) have shown that people are very prone to draw inferences about products from advertisements and to remember those inferences as directly stated facts. This occurs both in the laboratory (Harris, 1977) and in real television viewing (Harris et at., 1980). Although it is difficult to train people not to make such inferences, training sessions have shown some success (Bruno & Harris, 1980; Hams et al., 1979, 198I), especially during successive repetitions of advertisements and tests (see Harris et a l . , 1982, for a review of this work). There is some evidence that we are even more prone to draw inferences from advertisements than from other types of prose, although a carefully controlled study of this issue is difficult to design (Bruno, 1980; Preston & Scharbach, 1971). The series of studies described later in this article will examine the issue more closely. E. JUDGMENT RESEARCH
Although not dealing as specifically with language, some of the research in human judgment shows the operation of inference processes to be essentially parallel to those observed in psycholinguistic tasks.
I.
Diagnosis Studies In some very interesting and creative research, Arkes and Harkness
(1 980) demonstrated the use of inference in diagnostic procedures.
Inferences in Information Processing
105
Speech pathology students were given 12 symptoms, 8 of which were related to Down’s syndrome (mongolism). After the first 4 related symptoms were presented, subjects were asked to diagnose the hypothetical patient. In a memory test 12 days later, subjects who had diagnosed the case as Down’s syndrome were less able to correctly reject new symptoms related to Down’s syndrome than were subjects not making that diagnosis. In other words, the diagnosis had acted as a theme or schema to organize information and direct inference-drawing comprehension and direct reconstruction during retrieval. Arkes and Harkness obtained similar results using na’ive subjects diagnosing plumbing problems after a training session on home pipe blockages. In another study examining inferences made during diagnoses, Monaco (1978, Experiment 3) had subjects read a series of conversations between a doctor and a patient and made a series of ongoing estimations of the subjective probability that the doctor believed the patient had a tumor. By comparing the probability estimations after each conversation, each of which contained some new diagnostic information about the patient’s condition, Monaco concluded that as many as 38% of the subjects showed strong evidence of using an information integration averaging model (e.g., N. H. Anderson, 1974) for combining the diagnostic information and inferring what the doctor had believed, and 94% showed some evidence of averaging. Such an approach may eventually shed light on exactly how information is integrated in drawing inferences. 2. Heuristics
In a series of studies examining judgments under uncertainty, Tversky and Kahneman (1 973, 1974; Kahneman & Tversky , 1972) demonstrated that people often base their subjective probability estimations on informal heuristics. These generally involve inferring that a limited sample of information is representative of the whole (representativeness), assuming that a few easily retrievable instances of an event accurately characterize the entire population of events (availability), or being unduly biased by initial information (anchoring-adjustment). Tversky and Kahneman have shown that both naive and expert judges make mathematically inaccurate judgments in a wide variety of tasks, instead using these three informal heuristics. However, later careful research (Beyth-Marom & Fischhoff, 1977; Fischhoff, Slovic, & Lichtenstein, 1979; Olson, 1976) has suggested that under some circumstances people may be able to estimate probabilities much more accurately than Tversky and Kahneman suggested.
Richard J. Harris
1 Oh
3 . Hiridsight
Fischhoff ( 1975, I977b, 1 980) and associates (Fischhoff & Beyth, 1975; Slovic & Fischhoff, 1977; Wood, 1978) have studied and identified the hindsight bias, whereby people show distortions in memory, seeing past events as much more highly predictable than is warranted. Once a person knows that some event has occurred, it is very difficult not to overestimate its predictability. This is part of the general trend of inferring and imposing orderly structure, especially causal relationships, on events we experience. When subjects estimate the probability of events before they happen and then recall those probabilites after the fact, remembered probabilities are higher for events thought to have occurred and lower for events thought not to have occurred (Fischhoff & Beyth, 1975). This generalizes to overestimating the predictability and replicability of the results of psychological experiments, given very limited preliminary results (Slovic & Fischhoff, 1977). Like other information processing biases, the hindsight effect is very difficult to train out, though Slovic and Fischhoff had some success by forcing subjects to generate rationales for why nonpredicted results might have occurred. This technique of generating and explaining alternative scenarios has also been used successfully in training improvement in soil texture judging (Gaeth, 1980) and recognition of implied claims in advertisements (Bruno & Harris, 1980). In all these cases, the training involves a monitoring of fairly automatic inference processes. This offers a way that circumvents Nisbett and Wilson’s (1977) concern that most cognitive processes are not easy to introspect and report on accurately. 4.
Ovrrcotfidence
In addition to making many errors in subjective probability judgments, people are vastly overconfident in the correctness of their flawed judgments (Einhorn & Hogarth, 1978; Fischhoff, Slovic, & Lichtenstein, 1977; Kahneman & Tversky, 1979; Koriat, Lichtenstein, & Fischhoff, 1980). Part of this may reflect the general tendency to infer organization and impose structure on input. “In their search for coherence, people often see patterns where none exist” (Kahneman & Tversky, 1979). Such tendencies are strong, however, and very difficult to “train out,” although some progress has been made by forcing subjects to generate outcomes ontradictory to those they predict and then justify these outcomes (Koriat er al., 1980; Slovic & Fischhoff, 1977). It clearly is very difficult to make subjects aware of their inference processes (cf. Bruno & Harris, 1980; Gaeth, 1980; Harris et id.,1979). Integrative processes in
Inferences in Information Processing
107
comprehension are so natural that people do not realize their need to “deprocess” (Fischhoff, 1977b). In a sense, “turning off” natural inference processes is a metalinguistic (metacognitive) task rather alien to natural ongoing information processing. 5 . Applications Inferences that people make in decision situations often have ramifications for judgment processes in real-world settings. Such applications include cost-benefit analysis (Fischhoff, 1979), medicine (Elstein, Shulman, & Sprafka, 1978), societal risk-taking (Slovic & Fischhoff, 1981; Slovic, Fischhoff, & Lichtenstein, 1976) estimation of probability of natural hazards (Slovic, Kunreuther, & White, 1974), nuclear power risks (Slovic, Fischhoff, & Lichtenstein, 198I ) , decisions to pollute (Dawes, Delay, & Chaplin, 1974), effect of media coverage on inferences (Combs & Slovic, 1979), product safety (Fischhoff, 1977a), seat belt usage (Slovic, Fischhoff, & Lichtenstein, 1978), C0,-induced climatic change (Fischhoff, 1981), and jury decisions (Reyes, Thompson, & Bower, 1980). Faulty inference processes are thus widespread both inside and outside the laboratory. F.
MODALITYDIFFERENCES
In a way it is surprising that there are not many direct studies of modality differences in prose processing. One such study (Kintsch & Kozminsky, 1977) had subjects either read or listen to three stories from Boccaccio’s Decameron and afterward write a summary of each story. Although the summaries were very similar overall for the two groups, there were slightly more propositions in those of the oral group; however, the authors concluded that overall the same comprehension skills were involved in both reading and listening. Baggett (1979) had subjects watch a dialogueless movie (The Red Balloon) or read a text about the characters and episodes in the film. Results showed that perceived episodes were equivalent and recall protocols were very similar. COMPARISON OF DISCOURSE TYPES G . SYSTEMATIC
One of the few studies to systematically examine the relative difficulty of different discourse types is a developmental study by Freedle and Hale (1979). In this study kindergarten and fourth-grade subjects heard either an expository paragraph followed by a story, or vice versa. The stimuli were controlled for content, in that half the subjects heard given informa-
Richard J. Harris
I08
tion as a story and half heard it as an exposition. Recall was better in the narrative-expository order than for the reverse for the kindergartners (no difference for fourth graders). The authors concluded that it may be possible to teach the use of a new or more difficult schema by presenting an easier schema first, followed by the more difficult one. They concluded that some positive transfer occurred from the narrative to the expository schema, even though the content was different.
VI.
The Britain Studies
Although several different types of discourse have been examined in the literature, it is difficult to make meaningful comparisons because there has been little research systematically comparing different varieties of discourse. For this purpose a series of studies (“the Britain Studies”) was conducted to clarify some of these issues. A.
EXPERIMENT 1
I.
Method
a . Subjects. A group of 165 native English-speaking undergraduate psychology students served as subjects and received course credit for participation. Subjects were run in groups ranging from 13- 15 in sessions lasting 50-60 min. 6 . Design. This was a 3 X 2 between-subjects design. The variables manipulated included the type of prose (narration, description, or advertisement) and presentation modality (oral or written). c . Materials. The materials consisted of 12 sets of three passages each. Each set of passages told about a different place in England, Scotland, or Wales, but each member of the set did so in a different type of discourse. One passage in each set (narration) described the place as a first-person story told informally through one particular traveler’s eyes. A second passage (description) was purely descriptive prose in the third person. The third passage in each set (advertisement) was written as a travel brochure trying to convince the reader to visit that place. All three passages in each set contained the same basic core information about the particular place, though wording, style, and minor details differed. For example, advertisements used hyperbole and more flowery language and did not discuss negative aspects of the place, whereas the narration included unfortunate as well as fortunate experiences that happened to the writer there. There was no overt attempt to persuade in the narration and
Inferences in Information Processing
I09
description passages. All three passages in any given set were approximately the same length, though the lengths of the sets vaned. A sample set of materials appears in Table I. d. Procedure. Each subject received all 12 passages in one of three types of prose: narration, description, or advertisements. The type of prose was presented either orally (recorded on tape) or written, each passage on a separate typed page. After each passage the subject filled out three rating questions. These questions used a 7-point scale to evaluate the subject’s interest in the
TABLE I STIMULUSMATERIALS:SAMPLESET (PENMACHNO) 1, Description (Experiments 1, 2, 3, 6 ) Penmachno is a small village of 100-200 people in North Wales, located 2 miles off Highway A5. Except for the presence of cars, the town does not appear to have changed much for over a century. It is the epitome of a lovely country village-a beautiful natural setting in a lush green valley, narrow winding little streets, and a generally sleepy atmosphere. It is largely Welsh-speaking, as a visitor to the town’s one all-purpose store can tell. Similarly, the epitaphs on the tombstones in the churchyard cemetery are all in Welsh, though some of the words are covered by the 2-ft-high grass that has grown up there. The town has largely been inhabited by a few families for generations; a glance at the surnames in the cemetery reveals a few names occurring repeatedly with few otherslots of Roberts, Hughes, Davies, Richards, Williams, and Owens. 2. Narration (Experiments 1-6) Upon approaching it, I could see immediately’that Penmachno itself was an unassuming little village of perhaps 100-200 people in the lush green valley in North Wales. Except for the presence of cars it looked about the way it must have looked when my Roberts ancestors left there for America in 1846. Basically, it looked about the way I always picture my ancestral village-beautiful natural setting, neither very rich nor very poor, narrow winding little streets, and a generally sleepy atmosphere. It was a Welsh-speaking area, judging from the store I stopped at (the town’s one little all-purpose store) and the epitaphs in the churchyard cemetery. The graveyard in this little old church was badly overgrown, but I trouped around some through the 2-ft-high grass. Although I didn’t find any graves of known ancestors, as I returned the 2 miles back to Highway A5, I still felt as though I had recovered a small part of my past that morning. 3. Advertisement (Experiments 1, 2, 3, 6) When in lovely North Wales, be sure to include a visit to the idyllic and unspoiled village of Penmachno. Except for the presence of cars, this village of 100-200 sturdy Welsh men and women looks about the same way it did over a century ago. It is the sort of magic place you have probably pictured your European ancestors as arising from-a beautiful natural setting in a lush green valley, hardy citizens, quaint little narrow winding streets, and a generally completely peaceful air about it. Penrnachno is largely a Welsh-speaking area, but don’t worry; its friendly inhabitants have all studied English as well. You might want to wmder its narrow lanes or visit its single all-purpose store. You can even visit the old Welsh cemetery in the churchyard and do some rubbings of historic tombstones. Penmachno is conveniently located just 2 miles off Highway A5; it is sure to hold a very special place in your heart.
(Continued)
110
Richard J . Harris
TABLE I (Continued) 4 . Conversation (Experiments 4 and 5) B: I want to hear about your visit to the village where your greatgrandfather was horn. A: Oh, that was one of the most exciting parts of my trip to North Wales. Even as 1 approached it, I could see that Penmachno was an unassuming little village of maybe 100-200 people in a lush green valley, even though it's only about 2 miles off Highway AS. R: It sounds just like we tend to picture some ancestral village. A: You know, that's exactly how it struck me-a beautiful natural setting, narrow winding little streets, neither very rich nor very poor, and generally kind of sleepy-looking. Except for a few cars, it must have looked about the same when my Roberts ancestors left there for America in I 846. n: Did you talk to anybody there? A: It was a Welsh-speaking area. I could tell that from the tombstone epitaphs and listening to the people in the town's one little all-purpose store. B: So you found some graves of your ancestors? A: Not that I recognized, but I did tromp around and look for a while through the tall grass in the old churchyard cemetery. B: You must have felt like Alex Haley returning to Africa. A: Oh, ldon't know about that, but 1 felt like I got alittle more in touch with my~. past that morning. 5 . Truth-Value Test Statements (Experiments 114) 1. There is much Welsh spoken by residents of Penmachno. (true) 2. Penmachno is a large industrial city in North Wales. (false) 3. There is a cemetery next to the church in Penmachno. (true) 4. Penmachno has few accommodations for tourists. (implied) 5. Penmachno is in the county of Gwynedd near the border of Clywd. (indeterminate) 6 . The area around Penrnachno is basically desert. (false) 7 . Penmachno is located about 2 miles off a major highway. (implied) 8. Penniachno is known for its weaving crafts. (indeterminate) 6 . Factual Test Question (Experiments 4 and 6 ) The old Wclsh cemetery described was in (a) Conwy (b) Penmachno (c) Snowdonia (d) Betws-y-coed
passage, its perceived difficulty, and the subject's desire to visit the place described in the passage. Following the rating task, subjects were given a second answer sheet containing eight statements, which they were asked to judge on a 5-point scale from false (1) to true ( 5 ) , with the midpoint (3) being defined as indeterminate in truth value. These eight statements contained two items of each of four types. Two were true based on the passage, i.e., paraphrases of material in the passage. Two were false, i.e., contradictions of material in the passage. Two were indeterminate in truth value, i.e., discussion of some aspect of the particular place that was not even mentioned in the passage. Finally, two statements were implied by the passage to be true, though they were not so stated directly. All i t e m were written to be, as much as possible, equally true, false, indeterminate, or implied true in all three versions of a given passage. Sample test statements appear in Table I.
Inferences in Information Processing
111
The instructions to the subjects varied, depending upon the prose condition. Description subjects were informed that the passages were encyclopedia descriptions of various places in England, Scotland, and Wales. Narration subjects were told that the passages were written by a professor about his trip to Britain. The advertisement subjects were informed that the passages were excerpts from travel brochures. All subjects were told to fill out a rating scale and content questionnaire immediately after hearing or reading each passage. They were all cautioned not to read the content questions before the passage had been heard or read and not to go back and change any of their answers. Thus, for each of the 12 passages, each subject did the following: ( a ) heard or read the passage, (6) filled out the rating scales, and ( c ) answered the content questions. This was not a timed experiment, so the subjects were encouraged to work at their own pace.
2 . Resu Its and Discussion Separate 3 x 2 analyses of variance (3 prose types X 2 presentation modalities) were performed on each of the seven dependent measuresi.e., responses to true, false, indeterminate, and implied test sentences, and ratings of interest, difficulty, and desire to visit. Results appear in Table 11. The analysis of the mean truth response to true test statements showed only a main effect of prose type, F(2,159) = 5.50, p < .005, with TABLE I1 EXPERIMENT 1 : MEANRATINGS Discourse type Description
Content questions (1 = false, 5 True False Indeterminate Implied true Ratings ( 1 = low, 7 Interest Difficulty Visit
=
=
Narration
Advertisement
Oral
Written
Oral
Written
Oral
Written
4.39 2.37 3.04 4.03
4.44 2.28 2.90 3.92
4.22 2.42 3.09 4.00
4.21 2.32 2.88 3.85
4.31 2.38 2.95 3.82
4.35 2.13 2.80 3.83
4.28 2.48 4.33
4.42 2.37 4.80
4.51 2.60 4.74
4.42 2.18 4.80
3.99 2.72 4.12
4.40 2.06 4.75
true)
high)
112
Richard J. Harris
narrations rated the least true and descriptions the most true, though all ratings were strongly toward the true end of the scale. The analysis of the truth responses to false test items showed only a main effect of presentation modality, F(1,159) = 4.44, p . < .04,with responses being more false in the written than in the oral condition. Similarly, the analysis of the indeterminate test items revealed only a main effect of modality, F ( 1 , 159) = 8.77, p < .W, with somewhat more true responses in the oral than in the written condition, although, as expected, all the means hovered around the scale midpoint of 3, thus empirically confirming that the truth of such items was indeed indeterminate. The analysis of the implication test items showed no significant main effects or interactions, although directionally descriptions were judged most true and advertisements the least true. For the rating tasks, there were no significant differences in the interest ratings across conditions, with all cell means being near or slightly above the scale midpoint of 4. Analyses of the difficulty ratings showed a main effect of modality, F(1,159) = 5.92, p < .02, with oral passages rated slightly more difficult than written, especially for narrations and advertisements, although all cell means were considerably on the easy end of the scale. The advantage of the written version could have been due to the confusion of the many unfamiliar place names in the oral condition. Finally, analysis of the desire-to-visit ratings also showed a main effect of modality, F(1,159) = 4.72, p < .03,with all the written versions and the oral narration passage being more convincing than the oral description or advertisement. B. EXPERIMENT 2
The first study was replicated by Hams, Stout, and Chastain, using a within-subjects design. Because of clearer effects in the written than in the oral condition in Experiment 1, all passages were presented in written form in Experiment 2. All 45 subjects read four descriptions, four narrations, and four advertisementsof the materials described above. The same seven dependent measures were used as in the earlier study, except that the data were also analyzed for gross serial position effects (first half vs. last half of list). Analyses of the responses to true items showed a main effect of position, with items for the first six stories rated more true (p < .005). Analyses of responses to false items showed lower mean responses in the last half of the stories (M = 2.01) than in the first half (M = 2.43). Neither indeterminate nor any of the three rating measures showed any main effects of sequence, and none of the seven measures showed any
Inferences in Information Processing
113
effects of prose type. Thus it appeared that the within-subjects design obscured any differences in prose type. C.
EXPERIMENT 3
To test the possibility that accompanying pictures might alter the processing of the passages, a third study (by Harris and Sanders-Ware) tested the effect of illustrating the passages. The 39 subjects received an exact replication of the within-subjects design of Experiment 2, with the following two exceptions. The stimulus passages were all presented orally on tape, and each tape-recorded presentation was accompanied by three color slides of places described in the passage. The slides were the same for all three prose types. Except for responses to false items being more false for the last half of the passages than for the first half, there were essentially no significant effects in this study. D.
EXPERIMENT 4
A fourth experiment (by Hams and Stout) used the oral narration passages from the earlier studies plus a new set of passages where the information in each narration version was turned into a conversation between two people, one of whom had recently visited Britain. Using a between-subjects design, half of the 88 subjects heard a sequence of 12 narrations, and the other half heard a sequence of 12 conversations. One additional dependent measure was included in this study to examine possible longer term effects of prose type. At the end of all 12 passages and test, subjects received a final 12-item, multiple-choice, content-memory test, with one question from each passage. This was included in case a possible effect of prose type on memory might not occur in testing immediately after hearing the passage but would differentially affect longer term retention. Results, presented in Table 111, were analyzed by eight one-way analyses of variance for each of the eight variables. Except for responses to true items being more true after conversations than narrations, there were no significant effects. E.
EXPERIMENT 5
A fifth study (by Harris) used the materials of the fourth experiment but substituted a free recall measure for the eight dependent measures used previously. Because of time and tedium factors, only 5 of the 12 passages
Richard J. Harris
I14
TABLE 111
EXPERIMENT 4: MEAN RATINGS Discourse type
Content questions ( 1 True False Indeterminate Implied true
=
Ratings ( I = low, 7 Interest Difficulty Visit
=
Conversation
Narration
4.42 2.22 2.98 4.03
4.20 2.30 3.02 3.94
3.99 2.26
4.55
4.40 2.46 4.59
9.67
I I .33
false, 5 = true)
high)
Factual test (total correct of 12)
could be used here. Protocols from the 24 subjects were scored for number of propositions recalled correctly, roughly in the sense of Kintsch (1974). Only propositions occurring in both the conversation and narration versions were scored. Results showed a mean number of 12.93 propositions recalled per passage from the conversation group and 12.66 for the narration group, with the difference being nonsignificant. Three of the five stories showed greater mean proposition recall in the conversation group; the other two showed greater recall in the narration group. F.
EXPERIMENT 6
I.
Method
The sixth experiment (by Hams and Dubitsky) was conducted to see if subjects could remember the discourse form of the input information. The procedure was identical to that of Experiment 2, the withinsubjects study using descriptions, narrations, and advertisements, with the following exceptions. Only the written condition was used. While reading each passage, subjects performed the rating task, but the content-question dependent measure was not used. At the end of the last passage, however, subjects were given a list of the 12 places described, along with a brief identifying phrase for each, and asked to indicate if they had heard about that place in a description, narration, or advertisement. They also rated their confidence in their answer on a 1-5 scale.
Inferences in Information Processing
115
After this they received the same 12-item content-memory test used in Experiment 4. 2.
Results and Discussion
The data from the interest, difficulty, and desire-to-visit ratings were each analyzed by a one-way analysis of variance, as were the number of correct responses to the factual test. Means appear in Table IV. Results of the interest ratings were significantly different, F(2,46) = 44.86, M S E = .48, p < .001, with descriptions rated most interesting and advertisements least interesting. Narrations were almost exactly halfway in between. Results from the difficulty rating showed, as in Experiments 1-5, that all passages were rather easy, but narrations were significantly more so than the other two types, F(2,46) = 12.97, MSE = .37, p < .001. This is consistent with Freedle and Hall’s (1979) finding that narrations are easier than descriptions for young children and with the common-sense impression that information in the form of stories is easier than when it is presented as straight “dry” facts. Analysis of the desire-to-visit ratings showed a significant difference, F(2,46) = 25.66, MSE = .56, p < .001, with descriptions being more convincing than advertisements or narrations. This is an especially interesting finding, since the discourse force of the advertisements was persuasive and specifically designed to convince the reader to visit there. Results of the number of correct responses on the factual test showed a significant difference, F(2,46) = 4.05, MSE = .46, p < .025, with advertisements being a little harder than descriptions or narrations, which did not differ from each other. Results of the discourse-type memory task appear in Table V. For TABLE IV EXPERIMENT 6: MEANRATINGS Discourse type
InteresP Difficulty” Visit“ Factual test
Description
Narration
Advertisement
4.08 2.04 4.14 3.04
3.04 1.38 2.91 3.80
2.18 2.23 2.68 2.58
1 = low, 7 = high. *Total number correct (out of 4) ‘I
Richard J. Harris
TABLE V EXPERIMENT 6: DISCOURSE-TYPE MEMORYRESPONSE TOTALS’ Remembered as Stimulus prose type
Description
Narration
Advertisement
Description Narration Advertisement
33 (3.27) 23 (2.96) 28 (2.89)
35 (3.43) 59 (4.20) 17 (2.94)
28 (3.29) 14 (2.57) 51 (3.65)
,IMean confidence rating ( I -5) in parentheses.
descriptions, discourse-type identification was at chance level, whereas for narrations and advertisements, there was a significantly greater-thanchance correct identification of the input discourse type, X 2 ( 2 ) = 35.44 and 18.81 (both p s < .OOl), respectively. Confidence was also higher for these two cells than for any of the others. G.
GENERAL DISCUSSION
Why were there such meager effects of prose type in the Britain Studies? Several comments may be offered here. First of all, results suggest that, to a large extent, not only does surface structure matter relatively little in memory but also the discourse form may not be all that important, at least not in this situation. Similarly, modality seems to have mattered little. This generally supports the results of Baggett (1979) and Kintsch and Kozminsky (1977). There may have been some problems with the materials. The unfamiliar places and difficult geographical names may have been confusing to the subjects and not held their interest as well as other material might have. It may have been that the advertisement passages, written by the author, differed in some material way from those in real travel brochures, but what that difference might be is not obvious. In these studies, however, the advertisements were not particularly effective in convincing subjects, nor were they especially well rated on any other dimension. None of the studies showed any main effects or interactions of prose types on inferences drawn. The inferences were readily drawn from all types of discourse, as reflected in the high mean responses to implication items in all the studies, but prose type did not differentially affect this measure.
Inferences in Information Processing
117
The fact that Experiment 6 produced so many more significant effects in the rating measures than did any of the other studies suggests that these measures are more sensitive and useful in comprehension than in memory tasks. Experiments 1-5 were all memory studies, and, even though the delays were rather short, the judgments had to be made from memory. However, Experiment 6 allowed subjects to make the judgments while looking at the passage. The high ratings for the descriptions on interest and desire-to-visit were rather surprising, especially in light of the complete failure of subjects to remember the type of discourse descriptions had appeared in. The sizable differences in memory for the different types of prose shown in Experiment 6 offer a promising area to examine further. Although working with different types of discourse, Dubitsky and Hams (1981) also found a large effect like this in a study using radio dramas told as stories or acted as drama, with the drama portions easier to remember as having occurred in that form. Clearly this type of effect deserves further careful study. In spite of the somewhat disappointing nature of the results of the memory studies, the use of materials controlled for content and systematically varying in discourse type and/or discourse force is still a promising methodology that has been all-too-seldom employed in prose research, owing to the heavy emphasis on structural variables at the expense of considering stylistic features of text. The present methodology could be used to examine many of Brewer’s (1 980) types of discourse and should eventually lead to some firmer conclusions about relative discourse differences in information processing.
VI.
Conclusion: Inferring the Future
Clearly the processes of inference-drawing are widespread and an integral part of the comprehension process, not some extra additional operation occasionally performed if necessary. Although the demonstration of inference-drawing in comprehension and its effects on reconstructive memory were landmark discoveries 6-8 years ago, by now they have become generally accepted and recognized as factors that any global theory of information processing must seriously deal with, It is becoming increasingly difficult for cognitive psychologists to totally ignore questions of ecological validity. The study of inferences is a good avenue from which to explore more applications of information processing research to real-world problems. In so doing it is naturally led closer to problems in applied social psychology. Research on inferences
1 I8
Richard J. Harris
may provide one specific avenue for the integration of themes from cognitive and social psychology. The coming years should see an increasingly specific delineation of the processes involved in drawing and using inferences in both the cognitive and the social domains. The two areas should benefit each other, as the precision of cognitive psychology and the broader scope of social psychology nourish and enrich each other. Inference processes may be a clear set of examples of unconscious processes, which Shevrin and Dickman (1980) argue must be included in any global theory. As more and more specific operations in inference processing are discovered and explored, it is hoped that some of the vagueness and imprecision currently associated with the term will dissipate, as common processes become identified and are shown to operate across a prescribed range of situations. There are still many unanswered questions about the differential effects of various discourse types and discourse forces on inference-drawing and comprehension more generally. The different types of prose as outlined by Brewer (1980) have barely begun to be explored, especially in highly controlled designs systematically examining various discourse types and forces. Clearly, results from the Britain Studies suggest that there are many situations and measures where it makes little difference what the prose type is, especially in memory. Also clearly, however, there are some situations, most notably remembering the form of the discourse (Experiment 6), where it makes a considerable difference. Continuing to sort out these factors will be the focus of much future research. The next decade promises to be an important and fascinating one in information processing psychology. As global models of cognition continue to proliferate and become increasingly subject to empirical testing, currently elusive concepts like schema, script, discourse force, and inference will become increasingly important but also, one hopes, increasingly precise. However these models develop, an important component of any of them will necessarily deal with the processes by which we interpret information beyond what is present in the stimulus material. REFERENCES Adams. M . J . , L Collins, A. A schema-theoretic view of reading. In R . 0. Freedle (Ed.), New directions in discourse proce.wing. Norwood, N.J.: Ablex, 1979. Anderson, N . H. Information integration theory: A brief survey. In D. H . Krantz, R . C. Atkinson, R . D. Luce, & P. Suppes (Eds.), Contemporary developments in muthema?icul psychology (Vol. 2 ) . San Francisco: Freeman, 1974. Anderson, R . C . , & Ortony, A. On putting apples into bottles-a problem of polysemy. Cognitive P.c.whu/ogy. 1975, 7 , 167-180.
Inferences in Information Processing
119
Anderson, R. C . , & Pichert, J. W. Recall of previously unrecallable information following a shift in perspective. Journul of Verbal Learning and Verbal Behavior, 1978, 17, 1-12. Arkes, H. R., & Harkness, A. R. The effect of making a diagnosis on subsequent recognition of symptoms. Journal of E.xperimentcr1 Psychology: Human Learning and Memory, 1980, 6 , 568-575. Austin, J . L. How to do things with w o r h . London and New York: Oxford University Press, 1962. Baggett, P. Memory for explicit and implicit information in picture stories. Journal of Verbal Learning and Verbal Behavior, 1975, 14, 538-548. Baggett, P . Structurally equivalent stories in movies and text and the effect of medium on recall. Journal of Verbal Learning and Verbal Behavior, 1979, 18, 333-356. Barclay, J . R. The role of comprehension in remembering sentences. Cognitive Psychology, l973,4, 229-254. Barclay, J . R . , & Reid, M. Characteristics of memory representations of sentence sets describing linear arrays. Journal of Verbal Leurning and Verbal Behavior, 1974, 13, 133-137. Bartlett, F. C. Remembering. London and New York: Cambridge University Press, 1932. Bates, E . , Masling, M . , & Kintsch, W . Recognition memory for aspects of dialogue. Journal of Experimental Psychology: Human Learning and Memory, 1978, 4 , 187-197. Beyth-Marom, R., & Fischhoff, B. Direct measures of availability and judgments of category frequency. Bulletin of the Psychonnmir Society, 1977, 9, 236-238. Binet, A., & Henri, V . La memoire des phrases (memoire des idees). A n n i e Psychologique, 1894.1. 24-59. Birnbaum, M. H . , & Stegner, S. E. Source credibility in social judgment: Bias, expertise, and the judge’s point of view. Journal of Personality and Social Psychology, 1979, 37, 48-74. Black, J . B . , & Bower, G . H . Episodes as chunks in narrative memory. Journal of Verbal Learning and Verbal Behavior. 1979, 18, 309-318. Black, J. B., & Bower, G. H. Story understanding as problem solving. Poetics, 1980, 9 , 223-250. Black J . B . , & Wilensky, R. An evaluation of story grammars. Cognirive Science. 1979.3, 213-230. Bower, G. H. Selective facilitation and interference in retention of prose. Journal of Educational Psychology, 1974, 66, 1-8. Bower, G . H. Experiments on story understanding and recall. Quarterly Journal of Experimental Psychology, 1976, 28, 51 1-534. Bower, G . H. Experiments on story comprehension and recall. Discourse Processes, 1978, 1 , 21 1-231. Bower, G . H . , Black, J . B . , & Turner, T. J . Scripts in memory for text. Cognitive Psychology, 1979, 11, 177-220. Bower, G. H., & Gilligan, S . G. Remembering information related to one’s self. Journal of Research in Personality, 1979, 13, 420-432. Bower, G. H., Karlin, M. B . , & Dueck, A. Comprehension and memory for pictures. Memory & Cognition, 1975, 3, 216-220. Bransford, J. D., Barclay, J. R., &Franks, J . J . Sentence memory: A constructive versus interpretive approach. Copirivr Psychology, 1972, 3, 193-209. Bransford, J . D., & Franks, J . J . The abstraction of linguistic ideas. Cognitive Psychology, 1971,2, 331-350. Bransford, J. D . , & Johnson, M . K. Contextual prerequisites for understanding: Some investigations of comprehension and recall. Journal of Verbal Learning and Verbal Behavior, 1972, 11, 717-726. Bransford, J. D., & Johnson, M . K. Considerations of some problems of comprehension. In W. Chase (Ed.), Visual informatiori processing. New York: Academic Press, 1973. Bransford, J . D . , & McCarrell, N. S. A cognitive approach to comprehension: Some thoughts about understanding what it means to comprehend. In W. B . Weimer & D. S . Palermo (Eds.), Cognition and the symbolic processes. Hillsdale, N.J.: Lawrence Erlbaum, 1974.
120
Richard J. Harris
Brewer, W . F. Memory for the pragmatic implications of sentences. Memory 6; Cognition. 1977, 5 , 673-678. Brewer, W . F. Literary theory, rhetoric, stylistics: Implications forpsychology. In R. J . Spiro, B. C. Bruce, & W. F. Brewer (Eds.), Theoretical issues in reading comprehension: Perspecrives f r o m cognitive psychology, linguistics, artificial intelligence, and education. Hillsdale, N .J .: Lawrence Erlbaum, 1980. Britton, B. K . , Meyer, B . J . F., Simpson, R., Holdredge, T . S . , & Curry, C. Effects of the organization of text on memory: Tests o f two implications of a selective attention hypothesis. Journal of Experimental Psychology: Human Learning and Memory, I979,5, 496-506. Brown, A. L. Semantic integration in children's reconstruction of marrative sequences. Cogilirive Psychology, 1976, 8, 247-262. Bruno, K . J. The discrimination of assertions and implications: A training procedure for adults and adolescents. Journal of Educational Psychology, 1980, 72, 850-860. Bruno, K . J . , & Hams, R. J . Effect of repetition on the discrimination of asserted and implied claims in advertising. Applied Ps~~cholinguistics, 1980, 1, 307-321. Buhler, K . Tatsachen und Probleme zu einer Psychologie der Denkvorgange 111. Uber Gedankenerinnerungen. Archiv f u r die Gesainte Psychologie, 1908, 12, 24-92. Cantor, N., & Mischel, W . Traits as prototypes: Effects on recognition memory. Journal ($Personality and Social Psychology, 1977, 35,38-48. Cantor, N., & Mischel. W . Prototypicality and personality: Effects on free recall and personality impressions. Journal of Research in Personoliry, 1979, 13, 187-205. Carpenter, P. A , , & Just, M. A. Integrative processes i n comprehension, In D. Labergc & S . J . Samuels (Eds.), Busic processes in reading: Perception and coinprehension. Hillsdale, N.J.: Lawrence Erlbaum, 1977. Chaffin, R . Knowledge of language and language about the world: A reaction time study of invited and necessary inferences. Cognitive Science. 1979, 3, 31 1-328. Chiesi, H. L., Spilich, G. J . , & Voss, J . F. Acquisition of domain-related information in relation to high and low domain knowledge. Journal of Verbul Learning unrl Verbal Behavior. 1979, 18, 257-273. Cialdini, R . B . , Petty, R . E., & Cacioppo, J . T. Attitude and attitude change. Annual Review of Psychology, 1981.32, 357-404. Clark, H. H. Inferences in comprehension, In D. Laberge & S. J . Samuels (Eds.), Basicprore.rses in readinxr Perception trnd comprehension. Hillsdale, N.J.: Lawrence Erlbaum, 1977. Clark, H. H., & Haviland. S . E. Comprehension and the Given-New contract. In R. 0. Freedle (Ed.), Discourse procluction and comprehension. Norwood, N.J.: Ablex, 1977. Clark, H. H., & Lucy, P. Understanding what is meant by what is said: A study in conversationally conveyed requests. Journal of Verhiil Leurning and Verbal Behuviur, 1975, 14, 56-72. Clements, P. The effects of staging on recall from prose. In R . 0. Freedle (ed.), New directions in t1iscour.w processing. Norwood, N.J.: Ablex. 1979. Cohen, G. Language comprehension in old age. Cognitive Psychology. 1979, 11, 412- 429. Collins, A,, Warnock. E. H., Aiello, N., & Miller, M. L. Reasoning from incomplete knowledge. In D. Bobrow & A. Collins (Eds.), Representarion arid understanding: Stuifirs in rognirivu science. New York: Academic Press, 1975. Combs, B., & Slovic, P. Causes of death: Biased newspaper coverage and biased judgments. Journalism Quarterly, 1979, 5 6 , 837-843. Conley, J . M . , O'Barr, W . M . , & Lind, E. A . The power of language: Presentation style in the courtroom. Duke Low Journul, 1978, 1978, 1375-1399. Corbett, A. T . , & Dosher, B . A . Instrument inferences in sentence encoding. Journul uf Verbal Learning untl Verbtrl Behmior, 1978, 17, 479-491. Craik, F. 1. M., Lk Lockhan, R. S . Levels of processing: A framework for memory research. Juurticil of Verbtrl Leurning utid Verbal Behavior, 1972. I I , 67 1-684.
Inferences in Information Processing
I21
Danks, J . H., & Glucksberg, S. Experimental psycholinguistics. Annual Review qf’ Psychology, 1980, 31, 391-417. Dawes, R . M . Memory and distortion of meaningful written material. British Journal of Psychology. 1966, 57, 77-86. Dawes, R. M., Delay, J . , & Chaplin, W. The decision to pollute. Environment and Planning A , 1974, 6, 3-10. Dooling, D . J . , & Christiaansen, R. E. Levels of encoding and retention of prose. The Psycholog)) of’ Learning and Motivation, 1977, 1 1 , 1-39. Dooling, D. J . , & Mullet, R. L. Locus of thematic effects in retention of prose. Journal ofExperimental Psvchology, 1973, 97, 404-406. Dubitsky, T. M. The effects of contextual knowledge on drawing inferences from conversations. Unpublished master’s thesis, Kansas State University, 1980. Dubitsky, T. M., & Harris, R. J . Memory for information and discourse form of radio dramas. In preparation, I98 1. Eagly, A. J., & Himmelfarb, S. Attitudes and opinions. Annual Review ofPsJvhology, 1978, 29, 517-554. Einhom, H. J., & Hogarth, R. M. Confidence in judgment: Persistence of the illusion of validity. Psychological Review, 1978, 85, 395-416. Elstein, A . S., Shulman, L. S . , & Sprafka, S. A . Mediculproblern solving: An analysis of clinical reasoning. Cambridge, Mass.: Harvard University Press, 1978. Erickson, B., Lind, R. A , , Johnson, B. C., & O’Barr, W. M. Speech style and impression formation in a court setting: The effects of “powerful” and “powerless” speech. Journal of’ Experimentul Social Psychology, 1978, 14, 266-279. Fischhoff, B. Hindsight# foresight: The effect of outcome knowledge on judgment under uncertainty. Journal oj Experiinenrul Psjrhology: Human Perception and Performance, 1975, 1, 288-299. Fischhoff, B . Cognitive liabilities and product liability. Journd of Products Liability, 1977, I , 207-220. (a) Fischhoff, B. Perceived informativeness of facts. Journal of Experitnental Psychology: Human Perception and Performance. 1977, 3, 349-358. (b) Fischhoff, B. Behavioural aspects ofcost-benefit analysis. In G. T. Goodman & W. D. Rowe (Eds.), Energy risk management. New York: Academic Press, 1979. Fischhoff, B. For those condemned to study the past: Reflections on historical judgment. In R. A. Schweder & D. W. Fiske (Eds.), New directions for methodology of behavioral science: Fallible judgment in behavioral research. San Francisco: Jossey-Bass, 1980. Fischhoff, B. Hot air: The psychology of COy-inducedclimatic change. In J . Harvey (Ed.), Cognition, social behavior und the environment. Hillsdale, N.J.: Lawrence Erlbaum, 1981, in press. Fischhoff, B., & Beyth, R. “I knew it would happen”: Remembered probabilities of once-future things. Organizational Behavior and Human Performance, 1975, 13, 1- 16. Fischhoff, B . , Slovic, P., & Lichtenstein, S. Knowing with certainty: The appropriateness of extreme confidence. Journal of Experimental Psychology: Human Perception and Perjormance, 1977, 3 , 552-564. Fischhoff, B., Slovic, P . , & Lichtenstein, S. Subjective sensitivity analysis. Organizational Behavior and Humun Performance. 1979, 23, 339-359. Frederiksen, C. H. Representing logical and semantic structure of knowledge acquired from discourse. Cognitive Psychology, 1975, 7,371-458. Frederiksen, C. H. Semantic processing units in understanding text. In R. 0. Freedle (Ed.), Discourse production and comprehension. Norwood, N.J.: Ablex, 1977. Freedle, R . 0.. & Hale, G . Acquisition of new comprehension schemata for expository prose by transfer of a narrative schema. In R. 0. Freedle (Ed.), New directions in discourse processing. Norwood, N.J.: Ablex, 1979.
I22
Richard J. Harris
Freedle, R., Naus, M., L Schwartz, L. Prose processing from a psychosocial perspective. In R . 0. Freedle (Ed.), Di~c.ocrrsrprudurfion nnrl roi,iprehe,l.sioii. Norwood, N . J . : Ahlex, 1977. Friedman. A. Framing pictures: The role of knowledgr i n automatized encoding and memory for gist. Journtrl ofE.rperirnc~n/tilP bwhology: Crrrrrctl, 1979, 108, 3 16-355. Gaeth, G . J . A c~ornptrrisoriof’ liv~rurecirri1 iutrrttc/iw /raining designed to reduce rho ir~fluenc~e u/ irrtrrferiny nrtireritr1.s: Air (tpplirw/iun ro soil .sc,irnce. Unpublished master’s thesis, Kansas State University. 19x0. Garrod, S . , L Sanford, A. Interpreting anaphoric relations: The integration of semantic information while reading. Jnurnal of Verhtrl Lettruiiig und Vrrlxil Rrhnvior. 1977, 16, 77-90. Gibhs, R. W., J r . Contextual effects in understanding indirect requests. LXscourse Processes, 1979, 2 , 1-10, Glenn, C. G. The role of episodic structure and story length in children’s recall of simple stories. Journul of Verhtrl Letiruiug t i r i d Vrrhirl Rehovior, 1978, 17. 229-247. Goetz, E. T . Inferring from text: Some factors influencing which inferences will he made. Discourse Proci,c.ses. 1979. 2 , 179-195. Goodenough, D. R., & Weiner, S. L. The role of conversational passing moves in the managemcnt of topical transitions. Discour-sr P r o r e . m s , 1978. 1, 395-404. Ciracsser, A . C. How to catch a fish: The representation and memory of common procedures. l l i w o u r t e Prorrs.se.r, 1978, 1, 72-89, Graesser, A. C., Higginbotham, M. W., Rohertson, S. P., & Smith, W. R . A natural inquiry into the Nuiional EnqriircJr: Sell-induced versus task-induced reading comprehension. DiscvrrrAr P ruces.srs. 1978, 1, 355-372. Gracsser, A . C., Rohertson, S. P.. & Anderson, P. A. Incorporating inferences in narrative representations: A study of how and why. Cognirive PJychologv, 1981, in press Graesser, A. C .. Rohertson S. P., Lovelace. E . , L Swinehart, D. Answers to why questions expose the organization of story plot and predict recall of stories. Jourunl of Verbal Letmirig tiiitl Vrrbtrl Behln~ior-.IY80, 19, 110-1 19. Grice, H. P. Logic and conversation. In P. Cole & J . L. Morgan (Eds.), Sytitu.x und sr!ncrntics (vol. 3). Sprech trcts. New York. Academic Press, 1975. Haherlandt, K . , & Bingham. G Verbs contribute to the coherence of bricf narratives: Reading related and unrelated sentence triples. Journtil of Vrrhirl1,eurniny arid Verhnl Behavior, 1978, 17, 419-425. Hamilton, D. L . , Katz, L. B., & Lrirer, V . 0. Organizational processes in impression formation. In R. Hastie, T . Ostroin, E. Ebbesen, R. Wyer, D. Hamilton, & D. Carlston (Eds.), Person niemory. Hillsdale, N.J.: Lawrence Erlhauin, 1980. Harris, R. J . Answering questions containing marked and unmarked adjectives and adverbs. Journal u/ E.rprriinrir/a/ P.n.chologv, 1973, 97, 399-401 . Hams, R. J . Memory and comprehension of implications and inferences of complex sentences. Journal of Verbal Lectrniug ant/ Vrrhal Brhm~ior,1974, 13, 626-637. Hams, R. J , Children’s comprehension of complex sentences. Jourrlnl of Experirnentrrl Child PSYc.hi>/ogy, 1975. 19, 420-423. Hams, R. J . Comprehension of pragmatic implications in advertising. Jonrnal of Applied PsYchology, 1977, 62, 603-608. Hams, R . J . The effect of jury size and judge’s instructions on memory for pragmatic implications from courtroom testimony. Bulletin of /he Psvchonornic~Society, 1978, 11, 129.132. Harris, R. J . , Dubitsky, T . M . . L Bruno, K . J . Psycholinguistic studiesof misleading advertising. In R. J . Hams (Ed.), I17fornitrtioiiprixxwing rrsearch iri aclvcrtising. Hillsdale, N . J . : Lawrence Erlhaum, 1982, in prcss. Harris, R. J . , Dubitsky, T M., Connizzo, J . . Letcher, L. E . , & Ellermaii. C . S . 7rizinirrg consumers ahout rnislrciding advertisiilg: 7’run:f’erof training and effec/s of specialized knowledge. In
Inferences in Information Processing
123
Current lssues and Research in Advertisirig I Y H l . Ann Arbor: University of Michigan Graduate School of Business Administration Division of Research, 1981. Harris, R. J . , Dubitsky, T . M., Perch, K. L., Ellerman, C. S . , & Larson, M. W. Remembering implied advertising claims as facts: Extensions to the “real world.” Bulktin of the Psychonomic Socipty, 1980, 16, 317-320. Hams, R. J . , Dubitsky, T. M., & Thompson, S . Learning to identify deceptive truth in advertising. In J . H. Leigh & C. R. Martin, Jr. (Eds.) Current issues and research in adverrising. Ann Arbor: University of Michigan Graduate School of Business Administration Division of Research, 1979. Hams, R. J . , & Monaco, G . E. The psychology of pragmatic implication: Information processing between the lines. Journal of Experimental Psychology: General, 1978, 107, 1-22. Harris, R. J., Teske, R. R., & Ginns, M. J . Memory for pragmatic implications from courtroom testimony. Bulletin of the Psychonomic Society, 1975, 6 , 494-496. Hastie, R., & Kumar, P. A. Person memory. Personality traits as organizing principles in memory for behaviors. Journul of Personality and Sociul Psychology, 1979, 37, 25-38. Haviland, S. E., & Clark, H. H . What’s new? Acquiring new information as aprocess in comprehension. Journal a / Verbal Learning und Verbal Behavior, 1974, 13, 512-521. Hayes-Roth, B., & Thorndyke, P. W. Integration of knowledge from text. Journal of Verbal Learning und Verbal Behavior, 1979, 18, 91-108. Hayes-Roth, B., & Walker, C. Configural effects in human memory: The superiority of memory over external information sources as a basis for inference verification. Cognitive Science, 1979, 3 , 119-140. Hildyard, A. Children’s production of inferences from oral texts. Discourse Processes, 1979, 2, 33-56. Hildyard, A , , & Olson. D. R. Memory and inference in the comprehension of oral and written discourse. Discourse Processes, 1978, 1 , 91-1 17. Jarvella, R. J . , & Collas, J . G. Memory for the intentions of sentences. Memorji & Cognirion, 1974, 2 . 185-188. Johnson, M. K . , Bransford, J . D., & Solomon, S . K . Memory for tacit implications of sentences. Journal of Experimental Psychology, 1973 98, 203-205. Johnson, M. K., Doll, T. J . , Bransford, J . D., & Lapinski, R. H. Context effects in sentence memory. J o u r n d of &perimenfa/ Psycho/ogy, 1974, 103, 358-360. Johnson, N. S . , & Mandler. J . M. A tale of two structures: Underlying and surface forms in stories. Poetics. 1980, 9, 51-86. Kahneman, D., & Tversky, A. Subjective probability: A judgment of representativeness. Cognitive P.vchn/ogy. 1972, 3, 430-454. Kahneman, D. & Tversky, A . Intuitive prediction: Biases and corrective procedures. TIMS Studies in Munugenient Sciences, 1979, 12, 313-327. Kanouse, D. E. Language, labeling, and attribution. In E. E. Jones et a / . (Eds.), Attribution; Perceiving the causes of behavior. Morristown, N.J.: General Learning Press, 1971. Kasprzyk, D., Montano, D. E . , & Loftus, E. F. Effects of leading questions on jurors’ verdicts. Jurirnetrics Journal, 1975, 16, 48-51. Keenan, J . M . , & Baillet, S. D. Memory for personally significant events. In R. S . Nickerson (Ed.), At!enrion und performunr,e VIII. Hillsdale, N.J.: Lawrence Erlbaum, 1979. Keenan, J. M . , MacWhinney, B., & Mayhew, D. Pragmatics in memory: A study of natural conversation. Journal qf Verbal Learning and Verbal Behavior, 1977, 16, 549-560. Kemper, S . Memory for the form and force of declaratives and interrogatives. Memory & Cognition, l 9 8 0 , 8 , 367-371. Kennedy, A . Decision latencies to thematic and nonthematic distractors in prose. Journal ofExperimental Psychology, 1973, 98, 432-434. ~
Richard J. Harris
124
Kintsch, W . The representarion of’meuning in menwry. Hillsdale, N.J.: Lawrence Erlbaum, 1974. Kintsch, W. Memory for prose. In C . N . Cofer (Ed.), The sfrucfure of hutnun memory. San Francisco: Freeman, 1976. Kintsch, W. On comprehending stories. In P. Carpenter & M. Just (Eds.), Cognitive processes in comprehension, Hillsdale, N.J.: Erlbaum, 1977. Kintsch, W., & Bates, E. Recognition memory for statements from a classroom lecture. Journal of Experirnental Psychology: Human Learning wid Memory, 1977, 3, 150- 159. Kintsch, W., & Greene, E. The role of culture-specific schemata in the comprehension and recall of stories. Discourse Processes, 1978, 1, 1-13 Kintsch, W., & Kozminsky, E. Summarizing stories after reading and listening. Journul of Educurionul Psychologv, 1977, 69, 491-499. Kintsch, W., & van Dijk, T. A. Comment on se rappelle et on resume des histoires. Larigage, 1975, 40, 98-1 16. (Eng. transl.) Kintsch, W., & van Dijk, T. A. Toward a model of text comprehension and production. Psychological Review, 1978, 85, 363-394. Kintsch, W., & Vipond. D. Reading comprehension and readability in educational practice and psychological theory. In L. -G. Nilsson (Ed.), Perspectives on menioq research. Hillsdale, N.J.: Erlbaum, 1979. Koriat, A., Lichtenstein. S., & Fischhoff, B . Reasons for confidence. Journul of E.xperirneriral Psychology: Hurnan Leurning and Memory, 1980, 6, 107-118. Kozminsky, E. Altering comprehension: The effect of biasing titles on text comprehension, Memory & Cognilion. 1977, 5, 482-490. Lingle, J . H . , Geva, N . , Ostrom, T . M., Leippe, M. R., & Baumgdrdner, M. H. Thematic effects of person judgments on impression formation. Journal of Personuliry and Social Psychology, 1979, 37, 674-687.
Lingle, J. H . , & Ostrom, T . M. Retrieval selectivity in memory-based impression judgments. Journal uf Personality and Sociul P,Yychology, 1979, 37. 180-194. Lipton, J . P. On the psychology of eyewitness testimony. Journul ofApplied Psychology, 1977,62, 90-95.
Loftus, E. F. Leading questions and the eyewitness report. Cognitive Psychology, 1975,7, 560-572. Loftus, E. F., Altman, D., & Geballe, R . Effects of questioning upon a witness’s laterrecollectiuns. Journal of Police Science and Administration, 1975, 3, 162-165. Loftus, E. F . , Miller. D. C . , & Bums, H . J . Semantic integration of verbal information into a visual memory. Journul of Enperimentcrl Psvchology: Hurnan Leurning and Memory, 1978,4, 19-3 I . Loftus, E. F.. & Palmer, J. C . Reconstruction of automobile destruction: An example of the interaction between language and memory. Journtrl of Verbal Learning untl Verbal Behavior, 1974, 13, 585-589.
Loftus, E. F., & Zanni, G. Eyewitness testimony: The influence of the wording of a question. Bulletin of the Psychonomic Society, 1975, 5 , 86-88. Mandler, J. M., & Johnson, N. S . Remembrance of things parsed: Story structure and recall. Cognitive Psychologv, 1977, 9, I 1 1-151. Marschark, M . , & Paivio, A. Integrative processing of concrete and abstract sentences. Journal of Verbal Leurning and Verbul Behavior, 1977, 16, 21 1-231. Masson, M. E. J. Context and inferential cuing of sentence recall. Journal of Verbul Leurning arid Verbal Behavior, 1979, 18, 173-186. McKoon, G . Organization of information in text memory. Journtrl qf Verbal Learning and Verhul Behuvior, 1977, 16, 247-260. Meyer, B. J. F. The orgunization of prose and its effecrs on memory. Amsterdam: North-Holland Publishing, 1975. Meyer. B. J. F. What is remembered from prose: A function of passage structure. In R . 0. Freedle, (Ed.), Discourse producrion and comprehension. Norwood, N.J .: Ablex, 1977.
Inferences in Information Processing
I25
Miller, J . R., & Kintsch, W . Readability and recall of short prose passages: A theoretical analysis. Journcil of E.rperiineiito1 Psyc.ho1og.v: Huinan Learning cind Meinory, 1980, 6 , 335-354. Minsky, M. A framework for representing knowledge. In P. Winston (Ed.), The psychology of computer v i ~ i ~ ~New i i . York: McGraw-Hill, 1975. Mischel, W. On the interface of cognition and personality: Beyond the person-situation debate. American Psychologist, 1979, 34,740-754. Moates, D. R . , & Schumacher, G . M. An introduction t o wgiiitive psychology. Belmont, Calif.: Wadsworth, 1980. Monaco, G . E. Constructiorr reforinulated. Unpublished doctoral dissertation, Kansas State University, 1978. Monaco, G. E. The effect ofperceived comprehensibility upon memory f o r prose. Paper presented at the meeting of the Rocky Mountain Psychological Association, Las Vegas, April 1979. Monaco, G. E . , & Harris, R . J . The influence of narrative structure on memory. Bulletin of the Psychonotnic Society, 1978, 11, 393-396. Morris, C . D., Stein, B. S . , & Bransford, J . D. Prerequisites for the utilization of knowledge in the recall of prose passages. Journal ofExperirnerita1 Psychology: Hutnan Learning and M e m o y , 1979, 5 , 253-261. Nisbett, R. E . , & Ross, L. Hurwan inference: Strategies and shortcomings of social judgment. Englewood Cliffs, N.J.: Prentice-Hall, 1980. Nisbett, R. E. & Wilson, T. P. Telling more than we can know: Verbal reports on mental processes. Psvchological Review, 1977, 84, 231-259. Olson, C. L. Some apparent violations of the representativeness heuristic in human judgment. Journal of Experimental Psychology: Humtin Perception and Performance, 1976, 2 , 599-608. Owens, J . , Bower, G . H . , & Black, J . B. The “soap opera” effect in story recall. Memory & Cognition, 1979, 7 , 185-191. Paris, S . G . , & Carter, A. Y . Semantic and constructive aspects of sentence memory in children. De~~elopinentul P.sycho1og.v. 1973, 9 , 109- I 13. Pans, S . G . , & Lindauer, B. K . The role of inferences in children‘s comprehension and memory for sentences. Cognitive Psychology. 1976, 8 , 217-227. Pans, S . G., & Upton. L. R . Children’s memory for inferential relationships in prose. Child Development, 1976, 47, 660-668. Pichen, J . W.. & Anderson, R . C. Taking different perspectives on a story. Journal ofEducutiona1 Psychology, 1977, 69, 309-315. Posner, M. I . , & Keele, S. W . On the genesis of abstract ideas. JournalofExperimental Psychology. 1968, 77, 353-363. Powers, P. A.. Andriks, J . L., & Loftus, E. F. Eyewitness accounts offemales and males. Journalof Applied Psychology, 1979, 64, 339-347. Preston, I . L . , & Scharbach, S. E. Advertising: More than meets the eye. Journal of Advertising Research, 1971, 11 (3). 19-24. Reyes, R. M., Thompson, W. C . , & Bower, G . Judgmental biases resulting from differing availabilities of arguments. Jouriiui of Personafity and Socirif Psychology, 1980, 39, 2- 12. Rogers, T. B., Kuiper, N . A., & Kirker, W . S . Self-reference and the encoding of personal information. Journal of Personality and Social PsYcho/ogy, 1977, 35, 677-688. Rothbart, M.. Evans, M., & Fulero. S . Recall for confirming events: Memory processes and the maintenance of social stereotypes. Journul of Experimental Social Psychology, 1979, 15, 343355. Rothbart, M . , Fulero, S . , Jensen, C., Howard, J., & Birrell, P. From individual to group impressions: Availability heuristics in stereotype formation. J/JUrnd of Experimental Social Psychology. 1978, 14, 237-255. Rutnelhart, D. E. Notes on a schema for stories. In D. Bobrow & A. Collins (Eds.), RepreJentation and understanding. New York: Academic Press, 1975.
I26
Richard J. Harris
Rumelhart. L). E Understanding and summarizing stories. In D. La Berge & J . Sainuels (Eds.), Rasic processe.s In retiding: Perception untl cor?iprrhensiu,i. Hillsdale, N . J , : Lawrence Erlbaum. 1977. Rumelhart, 1). E. Schemata: The building blocks of cognition. In R. J . Spiro, B. C. Bruce, & W . F. Brewer (Eda.), T h c o r d c u l issires in rrutling c~oinprehrrrsirni. Hillsdale, N.J.: Lawrence Erlbauin, 1980. Schallen, D. L. Improving memory for prose: The relationship between depth of processing and context. Joitrtiul of' Verbul Leurning trntl Verbul Behuviur. 1976, 15, 62 1-632. Schank, R . C. Conceptual dependency: A theory of natural language understanding. Cugnitiiv PsK~hulog,v,lY72, 3, 552-631, Schank. R . C . Rules and topics in conversation. Cognitive Scietice, 1977, 1 , 421-441. Schank, R. C., C Abelson, R. P. Scripfs, pluns, goal.r, arid unrfer.wanding. Hillsdale, N.J.: Lawrence Erlhaum, 1977. Schustack, M. W.. C Anderson, J . R. Effects of analogy to prior knowledge on memory for new information. Juurntrl of Verhnl Leorning and Verhul Behmvior. 1979, 18, 565-583. Schweller. K . G . . Brewer, W. F., & Dahl, D. A . Memory for illocutionary forces and perlocutionary effects of sentences. Journtrl uf Verbul Leurning und Verbal Behavior, 1976, 15, 325-337. Searle, J . R. Speech ucts. London and New York: Cambridge University Press, 1969. Sentis, K. P., & Burnstein, E. Remembering schema-consistent information: Effects of a balance schema on recognition memory. Journal of Personulify untl Suciul Psychulugy, 1979, 37, 2200-22 I I . Shevrin, H., & Dickman. S. The psychological unconscious: A necessary assumption for all psychological theory'? Aniericun P.sychologisl, 1980, 3 5 , 42 1-434. Singer, M. Context inferences in the Comprehension of sentences. Currudirrri Jourriul ufPsyc.hology, 1976, 30, 39-45. Singer, M. Processes of inference during sentence encoding. Meirrury & Cugnitiun. 1979, 7, 192200. Slovic, P . , & Fischhoff, B. On the psychology of experimental surprises. Juurnitl of Experiwwtul P.syc,hologj : Hionan Percqmon and Perforincirice, 1977, 3 , 544-55 1 . Slovic, P., & Fischhoff, B. How safe is safe enough'? Determinants of perceived and acceptable risk. I n I.. Gould & C. A . Walker (Eds.), Too hot to handle: Socicrl unrl policy issues in the mcrnrrge~nc~n~ u / rtrdiotrc/ive M W S I Y S . New Haven: Yale University Press, 1981. in press. Slovic, P.. Fischhoff, B., & Lichtenstein, S . Cognitive processes and societal risk taking. In J . S. Carroll & J . S . Payne (Eds.), CUgliitio?i und suc,itrl behavior. Hillsdale, N . J : Lawrence Erlbaurn, 1976. Slovic, P . . Fischhoff, B., & Lichtenstein, S. Accident probab es and seat belt usage: A psychological perspective. Awiclerr/ Analy.sis urid Preiwition. 1978, 10, 28 1-285. Slovic, P., Fischhoff, B., & Lichtenstein, S. Nuclear Power: Images of disaster, h d / t d ! i cf the Atomic Scicn~ists.1981, in press. Slovic, P . , Kunreuther. H., & White, G.F. Decision processes, rationality, and adjustment to natural hazards. In G . F. White (Eds ), N u / i c i n d haznr(l.s: f.ocul. national, and global. London and New York. Oxford University Press, 1974. Smith, E. E., Shoben, E. J . , & Rips, I.. J . Structure and process in semantic memory: A featural model for semantic descriptions. Psychologicul Rei'icw, 1974, 81, 214-241. Snyder, M . Seek, and ye shall find: Testing hypotheses about other people. In E. T. Higgins, C. P. Herman, & M. P. Zanna (Eds.), Sucicrl cvgnitiox The Ontcrrin syw~~io.,osium ON p~rsunalityunrl sociul p . s ~ c h o l o y ~ Hillsdale, . N . J . : Lawrence Erlhaum, I Y 8 1 , in press. Snyder, M., &k Cantor. N . Testing hypotheses about other people: The use of historical knowledge. Journul of E.xprrirnerifal Social Psychology, 1979, 15, 330-342. Snyder, M . , & Swann, W. B . , Jr. Hypothesis-testing processes in social intcraction. Jourrrrrl uf Personalit! and Socrrrl Psvcliolugv, 1978, 36, 1202- 12 12.
Inferences in Information Processing
127
Snyder, M . , & Uranowitz, S . W. Reconstructing the past: Some cognitive consequences of person perception. Jourrial of Personality and Social Psychology. 1978, 36, 94 1-950. Spilich, G . J . , Vesonder, G . T . , Chiesi, H. L., & Voss, J . F. Text processing of domain-related information for individuals with high and low domain knowledge. Journal qf Verbal Learning und Verbal Behavior, 1979, 18, 275-290. Spiro, R. J. Accommodative reconstruction in prose recall. Journal of Verbal Learnrng and Verbal Behavior, 1980, 19, 84-95. Springston, F. J . . & Clark, H. H. And and o r , or the comprehension of pseudo-imperatives, Journnl of Verbal Leurning and Verbal Behavior. 1973. 12, 252-272. Stein, N . L., & Glenn, C . G . Analysis of story comprehension in elementary school children. In R. 0. Freedle (Ed.), New dirccrions in discourse processing. Norwood, N.J.: Ablex, 1979. Stiles, W. B., Waszak, C. S . , & Barton, L. R. Professorial presumptuousness in verbal interactions with university students. Journal of Experimental Sociul Psychology, 1979, 15, 158-169. Sulin, R. A , , & Dooling, D. J . Intrusion of a thematic idea i n retention of prose. Journal of Experimental Psychology, 1974, 103, 255-262. Tannen, D. What's in a frame'? Surface evidence for underlying expectations. In R. 0.Freedle (Ed.), New directions i n discourse processing. Norwood, N.J.: Ablex, 1979. Taylor, S . E., & Crocker, J. Schematic bases of social information processing. In E. T. Higgins, P. Hermann, & M . P. Zanna (Eds.), Social cognition. Hillsdale, N.J.: Lawrence Erlbaum, 1981, in press. Thieman, T. J . , & Brewer, W . F. Alfred Binet on memory for ideas. Genetic Psychology Monogruphs, 1978, 97, 243-264. Thorndyke, P. W. Cognitive structures in comprehension and memory of narrative discourse. Cognirive Psychology. 1977, 9 , 77-1 10. Thorndyke, P. W. Knowledge acquisition from newspaper stones. Discourse Processes, 1979, 2 , 95-1 12. Thorndyke, P . W . , & Yekovich, F. R. A critique of schema-based theories of human story memory. Poerics. 1980, 9, 23-49. Tsujimoto, R . N . Memory bias toward nonnative and novel trait prototypes. Journal ofPersonality and Social Psychology, 1978, 36, 1391-1401. Tversky, A . . & Kahneman, D. Availability: A heuristic for judging frequency and probability. Cognitive Psychology. 1973, 5 , 207-232. Tversky, A . , & Kahneman, D. Judgment under uncertainty: Heuristics and biases. Science, 1974, 185, 1124-1131. Tversky, A., & Kahnernan, D. Causal schemata in judgments under uncertainty. In M. Fishbein (Ed.), Progress iri sociul psychologj. Hillsdale, N.J.: Lawrence Erlbaum, 1977. Vesonder, G. T. The role of knowledge in the processing of experimenral reports. Paper presented at the meeting of the Midwestern Psychological Association, St. Louis, May 1980. Vipond, D. Micro and macroprocesses in text comprehension. Journal of Verbal Leurning and Verbal Behavior, 1980, 19, 276-296. Waern, Y. Comprehension and belief structure. Scundinaviun Journal of Psychology, 1977. 18, 266-274. (a) Waern, Y . On the relationship between knowledge of the world and comprehension of texts. Scnnrlinin~ianJournal oj'P.\ychology, 1977, 18, 130.139. (b) Walker, C. H . , & Meyer, B. J. F. Integrating different types of information in text. Journal of Verbul Learning and Verbal Behavior, 1980, 19, 263-275. (a) Walker, C. H . , & Meyer, B. J . F. Integrating information from text: An evaluation of current theories. Review of Educarional Research, 1980,50,(2). (b) Warren, W. H., Nicholas, D. W., & Trabasso, T. Event chains and inferences in understanding narratives. In R. 0. Freedle (Ed.), New direcrions in discourse processing. Norwood, N.J.: Ablex, 1979.
128
Richard J. Harris
Wason, P. C . , & Johnson-Laird, P. N. Psychologv of reasoning. Cambridge, Mass.: Harvard University Press, 1972. Weiner, S . L., & Goodenough, D. R. A move toward apsychology ofconversation. In R. 0. Freedle (Ed.), Discourse production and comprehension. Nonvood, N.J.: Ablex, 1977. Wilkes, A . L., & Alred, G . Prose and prejudice: Some effects of priming context on the immediate recall of information. British Jourrial of Psychology, 1978, 69, 123-133. Woll, S . B . , Weeks, 0.G . , Fraps, C. L., Pendergrass, J . , & Vanderplas, M. A. Role of sentence context in the encoding of trait descriptors. Journul ofPersonuli/y w i d Socicil Psychology, 1980, 39, 59-68. Woll, S . , & Yopp, H. The role of context and inference in the comprehension of social action. Journal of Experimerirul Social Psychology, 1978, 14, 351-362. Wood, G . The knew-it-all-along effect. Journul of’E.rperirnrn/nlPsychology: Hurnuu Perreption tint/ Perfonnunre, 1978, 4, 345-353. Yekovich, F. R., & Walker, C. H. Identifying and using referents i n sentence comprehension. Journal of Verbal Letirning und Verbul Behuvior. 1978, 17, 265-277. Yekovich, F. R . , Walker, C. H.. & Blackman, H. S. The role of presupposed and focal information in integrating sentences. Jortrncrl of’ Verbtrl Lrtrrniug cmd Verhctl Behtivior. 1979, 18, 535-548. Zajonc, R . B . Feeling and thinking: Preferences need no inferences. Americcrii Psychologist, 1980, 35, 151-175. Zehr, H. D., & Kimble, C. E. Selfconsciouswss, iqforrnurion h i t / , und mernorr in u socicrl si/uation. Paper presented at the meeting of the Midwestern Psychological Association. St. Louis, May 1980.
MANY ARE CALLED BUT FEW ARE CHOSEN: THE INFLUENCE OF CONTEXT O N THE EFFECTS OF CATEGORY SIZE’ Douglas L. Nelson UNIVERSITY OF SOUTH FLORIDA, TAMPA. FLORIDA I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. Method . . . . . . . . . . . . . . . . . . . .....................
It[. Cued Recall.. . . . . . . . . . . . . . .
.................................... A. Context Cues Absent at Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .................. B . Sensory Context Cues Present at Study. C . Associative Context Cues Present at Study.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. Nominal vs Functional Set S i z e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I V . Conceptual Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Structural Assumptions. . . . . ...... ......... B . Processing Assumptions . . . . ...................... C. Retrieval Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V. Some Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ................. References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
129 131 133 133
142 144 145 146 150
153 158
160
I. Introduction It is early Saturday morning and Professor Forgetum quietly dozes by the side of his swimming pool, his Journal of Experimental Psychology silently absorbing the beads of perspiration on his chest. He is from a large but appropriately unknown university and believes that he must work on weekends. Suddenly, his wife appears. “Professor, I want you to pick up several things at the store for me while I finish splitting these logs. ” Momentarily startled, but obedient, he gives up his morning’s work, puts on a shirt, and asks what she needs. She replies, “A small ham, any kind of jam, a bag of ice, salt, a pound of butter, and some cheese. And, oh yes, a package of frozen lobster and a can of mushroom soup. ” He cannot seem to find his pen to write the list on his hand, so he ‘Preparation of this article was supported by NIMH Grant 16360 from the National Institute of Mental Health. I am particularly indebted to Cathy McEvoy for her help i n preparing all aspects of this work. I29 THE PSYCHOLOGY OF LEARNING AND MOTIVATION. VOL 15
Copyright Q 1981 by AcAdems Prers. Inc All right, of reproduction in any form reserved. ISBN 0-12-543315.8
130
Douglas L. Nelson
decides to use another strategy: rehearsal. He begins muttering: “HAM J A M . . . I C E . . . SALT.. . BUTTER.. . CHEESE.. . LOBSTER.. . SOUP.” He repeats these words, silently after a while, as he drives. However, down the street, he notices his neighbor’s wife and remembers how‘much he was impressed by her quiet manner and wit at the Christmas party. From this point his mind wanders a bit. Upon reaching the store he feels confident and awake and he swiftly picks up the items, hesitating only in front of the frozen-food counter and in front of the shelves of soup. He returns home with BUTTER, CHEESE, HAM, LAMB, ICE, SALT, SHRIMP, BREAD, WINE, and a can of Minestrone soup. As she unpacks the bag, his wife begins muttering. Professor Forgetum’s wife is fretting for obvious reasons. From her point of view her husband’s recall errors are costly. However, from the point of view of a cognitive psychologist, his errors as well as his correct recalls are interesting. Note that he remembers Butter, Cheese, Ham, Ice, and Salt, for 63% correct recall. However, he substitutes Lamb for Jam, Shrimp for Lobster, and Minestrone for Mushroom. No comment need by made about Bread and Wine. Examination of the grocery list does not immediately suggest why some words were remembered and others forgotten. All represent relatively common and familiar items of experience, all are highly meaningful, and all could be classified as concrete, easily imaged, or easily propositionalized, depending upon theoretical predilections (Anderson, 1978; Paivio, 1971). However, there are at least two differences among these items, one fairly obvious and one more subtle. The obvious one involves context. l w o of the remembered items, Butter and Cheese, are dairy products frequently found together in the store. They share many meaning features, and, in a very real sense, each word provides a semantic context for the other. One of the forgotten items, JAM, rhymes with the word HAM. It is possible that the substitution error, Lamb, is related to the sensory context shared by the items. Thus, Professor Forgetum’s recall performance may be somehow related to the nature of the specific context in which the items were experienced. His performance may also be related to a more inconspicuous difference. Some of the items are more likely to activate a greater number of related experiences than others. That is, some of the items are associated with nominally larger category sizes. Although neither was encoded within a specific sensory or semantic context, both SALT and ICE were remembered. It can be shown that both words normatively produce relatively small semantic categories of about 7-8 items. Remembering that these items were on the list in the presence of retrieval cues available in the grocery store may be related to this fact. Failing to recall Lobster may also be related to this fact. Its semantic category size is estimated at about
Context and Category Size
131
24 items, including crab, seafood, tail, fish, shrimp, and Newburg. Similarly, as he stands before the vast shelves of soup, he incorrectly chooses. However, it should be noted that Cheese also produces a relatively large associative category, and yet it is remembered. Of course, as already indicated, this item was encoded in the presence of a specific semantically related context. The point is that Professor Forgetum’s cued recall performance may be related both to the nature of the specific context and to category size. To anticipate some of the discussion to follow, these two factors do not seem to be independent. The major purpose of this article is to review and summarize our cuing research on the nature of specific encoding contexts and the relationship between this factor and category size. In addition, a conceptual framework for describing these findings will be offered. However, before attempting either of these tasks, it would be useful to provide a brief description of the general methodology common to all the experiments that will be discussed.
11. Method Each cuing experiment can be subdivided into three distinct phases: the scaling of materials, the study trial, and the test trial. The initial scaling phase provides quantitative estimates of both nominal psychological similarity and nominal category size. Large groups of subjects are asked to produce the first word that comes to mind that rhymes with or, for different groups of subjects, that is associatively or meaningfully related to the context item. For both rhymes and associates, the frequency with which one word is given as a response to another provides a directional measure of their perceived similarity as it exists prior to the experiment. Associates given most frequently, in the form of either rhymes or meanings, are considered to be psychologically more similar or stronger because they share more common and fewer distinctive features with the context cue (A. Tversky, 1977). For associative meaning, this definition appears obvious and reasonable. That rhymes also vary in psychological similarity is not so obvious. For example, by counting letters, it would seem that HAND is equally related to both SAND and STAND. However, controlled association norms show that the probability that HAND elicits SAND is .42, and the probability that it elicits STAND is .06. Like meaning-related associates, rhymes vary in perceived similarity. Category or set size is operationally linked to a different aspect of the normative data-to the number of different but relevant targets that have some likelihood of being activated by the context cue. Hence, by counting the number of different words elicited by a word ending, a rhyme, an
132
Douglas L. Nelson
associate, or a taxonomic category name, an estimate of nominal category size can be obtained. The smaller the number of different items, the smaller the nominal set will be. For example, consider the word SALT. The word ending sound OLT generates only seven different rhyming words, including HALT, MALT, and FAULT. Similarly, the word itself essentially elicits the same rhymes, with each elicited item having about the same probability in both sets. When a different group of subjects was asked for meaning-related associates for the word SALT, only 8 different items were given by at least 2 subjects, whereas for the word FISH a total of 18 different words was provided by 2 or more of the 165 subjects. Finally, similar procedures can be used to estimate category size for taxonomic categories; e.g., American Coin yields an estimated set size of 6 items and Kind of Wood produces a set of 26 different responses. The primary motivation for resorting to these nonnative procedures was to obtain reasonable estimates of preexperimentally acquired information linked to different types of cues. Presumably, comparisons in cue effectiveness then could be made for qualitatively different types of retrieval cues without confounding these variations with strength and category size. Showing that one type of cue is superior to another under some set of conditions is hardly an impressive demonstration unless it can also be shown that the two cues were equally prepotent. Of course, this normative approach carries two assumptions. First, the use of the norms implies that encoding reflects an interaction between current experience and previously acquired related experiences. Later recall for the current experience will be a function of the results of this interaction. Second, the normative approach requires the assumption of scale equality for all types of cues. It assumes that the metric underlying the normative procedures is the same for qualitatively different cues. This assumption is reasonable in the sense that all subjects produce the first word that comes to mind. Nevertheless, a caveat is in order. Because subjects focus on different aspects of a cue, equal probabilities obtained under different instructional sets do not guarantee an equal metric. That is, just because BOY elicits TOY and MALE with an equal probability of .41, this is not complete assurance that the two types of relationships are equally similar or equally strong. Thus, the validity of the scale remains to be determined. However, using the norms has always seemed preferable to ignoring potential confoundings, or, alternatively, to avoiding cue-type comparisons altogether. The second phase of each cuing experiment consists of the study trial. List words, called targets, are presented for intentional learning for fixed amounts of study time, usually 3 sec, During this phase specific context cues can be absent or present. It they are absent, each target word is studied in isolation-that is, in the absence of any specifically biasing
Context and Category Size
I33
contextual cues. For example, the word CHEESE would be shown individually as one member of a list of words. Subjects are left to form their own encoding biases, and, presumably, most are naturally biased toward focusing upon the meaning of each item, as well as upon potential relationships among the list items (e.g., Einstein & Hunt, 1980). If context cues are present, each target word is studied in the presence of a specific cue: a rhyme, an associate, a taxonomic category name, or, finally, the word ending itself. For example, pairs such as SNEEZE CHEESE or BUTTER CHEESE might be shown. The subjects are still told that they must remember the target word, but they are also told to pay attention to the relationship between the context cue and its target, since this relationship will facilitate later recall. Thus, the presence of such contextual cues biases encoding operations toward either sensory or semantic information shared by the cue and its target. The final phase of each cuing experiment consists of the test trial. In the typical experiment, all subjects are given explicit test instructions immediately following the presentation of the last item in the study list. During this trial, a retrieval cue or prompt is presented to aid recall of each individual target word. Subjects are always made aware of the dimension on which the cue and target are related, and this episode is usually self-paced. With this procedure, test cues can be rhymes, associates, etc. These cues can be extralist cues presented only during the test trial, or they can be identical to the study context cues. 111.
Cued Recall
The use of the cuing procedure has yielded complex and interesting patterns of findings, particularly in relation to the interactive effects produced by variations in context and functional category size. What follows is a review of the basic findings. This review is organized around the nature of the contextual cues available during the study episode, rather than by a chronological ordering of the actual experiments. Within each context variation, category size effects for both test cues and target words will be discussed with emphasis given to comparing sensory and semantic cue-types. A.
CONTEXT CUESABSENTAT STUDY
I. Test Cues Variations in the category size evoked by different types of test cues have very consistent effects on amount recalled. Nelson and McEvoy
Douglas L. Nelson
134
(1979b, Experiment 1) contrasted the effects of extralist word ending cues with extralist taxonomic category cues, e.g., IME as compared with A N AMERICAN COIN as cues for the target DIME. Target words were either strongly or weakly related to their respective cues, and, at each level of strength, the test cues produced either small, medium, or large categories, respectively averaging 6, 13, or 25 different items. Target set size was held constant. Hence, each subject in the experiment studied a list of 24 words in the absence of specific contextual cues, and then they were given either sensory or semantic test cues. Finally, it should be mentioned that amount of study time was manipulated, with target words presented at a fast (1.5 sec) or at a relatively slow (3.0 sec) rate. The results are shown in Table I, pooled over study time. The statistical analysis of these data indicated that recall varied only with two factors, strength and category size. Type of cue had no effect, nor were any of the interactions reliable. Thus, when cues were presented only at test as extralist retrieval prompts, recall was better for strong than for weak targets. It also improved as category size decreased. Cues that normatively evoked smaller sets were more likely to be effective than were those eliciting larger sets. In a second experiment, similar set size effects were obtained by using rhyme and associatively related cues. Other experiments indicated that these category size effects also seem to be independent of the length of the retention interval and presentation modality. Nelson and McEvoy (1979a) compared extralist ending and taxonomic category cues on an immediate test and after a 15-min or a 30-min filled delay. The filler task required subjects to perform a paired-
TABLE I P R O B A B I I JOF T YCORRECT RECALLAS A FUNCTION OF TYPEOF CUE, STRENGTH, A N D CATEGORY SIZE"
~~
~~
Word endings Strength Strong Weak Mean
Tdxonomlc
Small
Medium
Large
Small
Medium
Large
Mean
.90 .71
.85 .54
.76 .54
.87 .67
.83
.77
.51
.49
.83 .58
.81
.70
.65
.71
.67
.63
Adapted from Nelson and McEvoy (1979b). Copyright 1979 by the American Psychological Association. Reprinted by permission.
Context and Category Size
135
associate task in which the stimuli were odors and the responses were numbers. As shown in Table 11, there were no differences between the two cue-types on the immediate test. This result was as planned (Underwood, 1964). However, over the delay, the effectiveness of the ending cues declined at a faster rate than the effectiveness of the taxonomic cues. Other comparisons in this experiment suggested that the effectiveness of ending cues was related to the encoding of visual information. For half of the subjects the cues and targets were presented visually; for the remaining half, both were presented auditorily. Examination of the results shown in Table I11 indicates that modality had no effect on the influence of the taxonomic cues. However, when word endings were used as cues, visually presented targets and cues were associated with higher recall than auditorily presented items. Furthermore, in a follow-up experiment involving switches in presentation modality between study and test, recall was highest when both targets and text cues were presented visually. Auditory presentation of any or both components produced lower and equivalent levels of performance. Recall in these experiments was also a function of cue strength and category size. As usual, normatively strong targets were more likely to be recalled, and recall increased as set size decreased from large (.55), to medium (.61), to small (.68). As previously mentioned, neither variable interacted with delay, nor did category size interact with any other variable in the experiment. This absence of interactions with category size was regarded as disappointing, since such interactions might have revealed something about the processes underlying the effect. However, TABLE I1
PROBABILITY OF CORRECT RECALL I N THE VISUAL PRESENTATION A S A FUNCTION OF CONDITIONS CUE-TYPEA N D D E L A P Cue-type Retention interval
Ending
Taxonomic
Immediate Delay ( I 5 min) Delay (30 min)
.I2 .54 .48
.68 .64 .63
‘I Adapted from Nelson and McEvoy (1979a). Copyright 1979 by the American Psychological Association. Reprinted by permission.
Douglas L. Nelson
136
TABLE 111
PROBABILITY OF CORRECT RECALL AS FUNCTION O F CUE-TYPEA N D PRESENTATION MODAL IT^
A
Cue-type Presentation modality Visual Auditory
Ending
Taxonomic
.63
.66
.50
.65
" Adapted from Nelson and McEvoy (1 979a). Copyright 1979 by the American Psychological Association. Reprinted by permission.
these experiments did suggest two important conclusions about the effectiveness of sensory cues. First, at least some of their redintegrative power is achieved through the encoding and retention of visual information. Second, given equivalent levels of initial recall, sensory cues seem to lose their effectiveness at a faster rate than semantic cues. The findings considered thus far suggest that cue set size effects seem to be independent of several variables, including cue-type, target strength, retention interval, and modality. Professor Forgetum seems condemned to committing recall errors for cues subsuming large numbers of instances. If these were all the data, it might be tempting to conclude that category size effects could be explained by the most simple generation-recognition type of model (Bahrick, 1970; Nelson, Wheeler, Borden, & Brooks, 1974b). In this case, the cue merely serves to generate the relevant set of instances, with the probability of correct recognition then determined by the number of competing alternatives within the set. However, as the next sections will show, the situation is more complex. Category size effects seem to be very much a function of the focus of encoding-retrieval operations. 2.
Turget Words
Variations of category size are not limited to the test cue. Target set size can be manipulated as well. For example, Nelson and Friedrich (1980) used standard cuing conditions and simultaneously varied both the ending and the associative set sizes of target words presented during the study trial. One-quarter of the targets had small ending and small associative sets, one-quarter had small ending and large associative sets, etc.
Context and Category Size
137
Test cues were either word endings or meaning-related associates. We were generally interested in whether there would be any effects of target category size at all. We were particularly interested in determining whether the manipulation of category size in one domain would have an effect when the retrieval cue was from the other domain. That is, would associative set size of the target influence recall when word endings were shown as cues? Would ending set size of the target influence recall when associates were presented as cues? Affirmative answers to one or both of these questions would indicate that category size effects can reflect processes occurring during the study trial. They would also indicate that retrieval cues can be effective by redintegrating information outside their own immediate domains. The results of several experiments produced apparently clear answers to both questions (Nelson and Friedrich, 1980). Recall was consistently higher for targets having smaller associative sets, and this effect was obtained with both types of test cues-meaning-related associates and word endings. In Experiment 4 of that article, the probabilities of recalling targets having small and large associative sets were, respectively, .76 and .69. Hence, considering the first question, the answer appears to be affirmative. Word ending cues must somehow provide access to semantic information encoded about the target word during the study trial. Thus, some of the effectiveness of sensory cues must be attributed to their capacity to reactivate semantic information encoded about the target, and, as other findings suggest, some of this effectiveness can be attributed to visual and phonemic information that becomes less and less helpful over time. The answer to the second question at first appeared to be negative. Ending set size of the target influenced recall, but only when ending cues were presented at test. With word endings as retrieval prompts, the probabilities of recalling targets having small and large ending set sizes were .81 and .69; with associates as test cues, these probabilities were .70 and .72. This finding could mean that meaning-related associates do not help to redintegrate encoded sensory information or, perhaps, that sensory information is not really encoded at all. Alternatively, by way of anticipating other results, this finding could mean that rhyme-related items might be activated but are not encoded during the study trial under the present conditions. The phrase “present conditions” turns out to be critical. When specific contextually related cues are absent during the study trial, as they were in the conditions being described, subjects may ordinarily focus their processing operations upon the meanings of the individual items simply out of habits acquired before coming to the laboratory (Nelson, 1979). Such a
Douglas L. Nelson
I38
focus may have the effect of activating and encoding a great variety of previously acquired experiences that are semantically related to the current experience. Since no specific semantic context is available that might indicate what particular semantic representation is most relevant, a broader semantic encoding is the result. All past knowledge is potentially relevant. This interpretative analysis also has implications for the sensory set size effect. If subjects do not focus upon sensory information, then category size effects linked to the sensory domain will not be in evidence. In the absence of orientation toward sensory features during the study trial, previously acquired sensory information may not be encoded, and, therefore, there would be no reason to expect an effect of category size with meaning-related test cues (see Eagle & Ortof, 1967; Underwood & Freund, 1968). However, when word endings are presented as retrieval cues, attention is focused upon this domain and the effect does appear. Of course, when word endings are used as test cues, set size of the cue is necessarily confounded with set size of the target, and it is impossible to detennine whether this effect is uniquely associated with events taking place during the study trial, the test trial, or both of these phases. This issue, however, may be of secondary importance if the whole effect is simply dependent upon the specific focus of processing resources. It may occur during any phase. The only necessary condition may be to bias encoding toward the sensory domain. This bias might be achieved by instructions at test indicating that each cue rhymes with one of the targets or, as suggested in the following section, by presenting each target word at study in the context of a rhyme. B.
SENSORY C O N T E X T C U E S PRESENT A T STUDY
1.
Test Curs
The cues to be used as retrieval aids at test can also be present during study as contextual cues. When these items are rhymes or word endings, they presumably focus processing operations on the sensory features of the target. This focus, however, does not result in an exclusive code, one that contains only sensory information. Target words having single meanings are twice as likely to be recalled as are target words having dual meanings, even though both targets are encoded in a rhyme context and are cued with appropriate semantic prompts (Nelson, Walling, & McEvoy, 1979, Experiment I ) . For example, the shift from TOWER FLOWER at study to ROSE at test results in recall of about 51%. The shift from PECK DECK at study to CARDS at test results in recall of only
Context and Category Size
I39
25%. This difference presumably occurs because subjects were just as likely to encode either the “boat” or the “cards” meaning of DECK during the study trial. For those who encoded the “boat” meaning, CARDS would not be a very effective retrieval cue (Tulving & Thomson, 1973). Thus, these results suggest that focusing upon sensory information does not preclude the encoding of semantic information. This result raises the legitimate question of the consequences of this encoding bias. If it does not prevent semantic encoding, what does it do? The answer to this question seems to depend somewhat upon whether word endings or rhymes are used as contextual cues. However, regardless of which cue is used, the influence of category size is evident. As part of the experiment described in Section III,A, Nelson and McEvoy (1979b) presented word ending cues during the study phase as contextual cues and again at test as retrieval cues. From the results displayed in Table IV, it can be seen that recall varied with normatively measured target strength and with category size. Thus, the patterns of recall obtained with ending cues available during both phases were very similar to those found when these cues were shown only during the test trial as extralist cues. In a second experiment, Nelson and McEvoy ( 1 979b) found comparable effects when rhymes were shown at study and again at test. However, if anything, the magnitude of the set size effect appeared to be somewhat greater under this condition than when they were shown only at test. These findings are illustrated in Table V . Thus, at least when rhymes are used as contextual cues, focusing processing resources upon sensory information during both study and test phases seems to augment the sensory set size effect relative to presenting the rhymes only as extralist cues.
TABLE IV PROBABILITY OF CORRECT RECALL FOR WORDENDING CUESPRESENTED AT STUDY A N D TESTAS A FUNCTION OF STRENTH A N D CATEGORY SIZE“ Category size Strength Strong Weak Mean
Small
Medium
Large
Mean
.93 .81
.91 .76
.87 .70
.90 .76
.87
.84
.78
“ Adapted from Nelson and McEvoy (l979b). Copyright 1979 by the American Psychological Association. Reprinted by permission.
Douglas L. Nelson
I40
TABLE V
RECALLFOR R H Y M EA N D WORDENDING CUESAS A FUNCTION OF CUE-PRESENCE AT PROBABILITY OF CORRECT
S T U D Y A N D CATEGORY S I Z P
Cue-type Rhymes
Word endings
Cue-presence
Small
Large
Small
Large
Study and test Test only
.72 .44
.51
.86 .82
.7 I .70
.38
“Adapted from Nelson and McEvoy (1979b). Copyright 1979 by the American Psychological Association. Reprinted by permission.
2.
Turget Words
Recall that, as mentioned earlier, when contextual cues are absent during study, the size of the sensory category produced by the target words influenced recall only when word ending cues were shown at test. When associative cues were presented, sensory set size failed to influence recall (Nelson & Friedrich, 1980). As suggested, the failure to find an effect with meaning-related cues may have occurred because subjects never focused upon sensory information during any phase of the experiment. To evaluate this possibility, rhymes were presented as contextual cues during the study trial, but recall was cued at test with meaningrelated associates (Experiment 5 ) . Therefore, the target words and the associative cues were identical to those used in the earlier study, with targets having small ending and small associative sets, etc. The only known difference was that contextual cues were now present and they rhymed with their targets. The results indicated that recall now vaned with ending set size of the target. The probability of recalling targets having small sets was .64, and the probability of remembering targets having large sets was a reliably lower .56. The suggestion is that sensory set size of the target words can influence recall. Since meaning-related associates were used as cues, this effect cannot be attributed to any confounding of cue and target sensory set size. Thus, under the right conditions, semantically related cues vary in effectiveness, depending upon sensory information encoded about the target during the study trial. One of the right conditions includes focusing processing resources upon the sensory domain. Such focus apparently
Context and Category Size
141
occurs when the target is encoded in the presence of a rhyming word, and it does not seem to occur in the absence of this context unless the test cue itself directs attention to this domain. Finally, one implication of these results is that semantic cues, like sensory cues, operate by providing access to information encoded within an alternative domain. In general, the effectiveness of any type of retrieval cue seems to be at least partially determined by the extent to which it recapitulates the original encoding experience, regardless of the domain of that experience (Tulving, 1979). The presence of a rhyme encoding bias during the study trial apparently leads to the encoding of other rhyming words that are not physically presented but that are members of the same category. Given HAM JAM, the word LAMB may become part of the original encoding experience. Knowing this, Professor Forgetum’s wife might have been slightly more tolerant of his error. The assumption here is that the focused activation of the category name, in this case AM, leads to the encoding of other members of this category. The likelihood of encoding any single instance is presumably related to the strength of that instance within that category, and the number of instances encoded is related to the number of possible instances. The fewer the number of encoded instances, the less will be the competition when recall for any single instance is demanded. Assuming that categorical activation has taken place, recall should vary with nominal category size. Other findings suggest that the sensory set size effect is apparently independent of both type of test cue and encoding strategy (McEvoy & Nelson, 1981). The lists for this experiment contained 27 highly concrete rhyme pairs that could be subdivided into sets having small, medium, or large category sizes. This variable was crossed with encoding strategy, with interaction imagery or repetition, and with type of test cue-original cues, rhyme switch cues, or associate switch cues. As usual, a single study-test trial was used, but each rhyme pair was presented for 4 sec during study. The results indicated that recall was best with original cues (.66) and lower for both rhyme switch (.42) and associate switch (.36) cues. As shown in Table VI, recall was also better under interaction imagery instructions and for smaller sensory set sizes. These values are pooled over cue-type, and it should be noted that all statistical interactions produced F s less than unity. Hence, while imagery instructions facilitated overall levels of performance, recall continued to vary with category size. While the instructions encouraged subjects to focus upon the imaginal or semantic attributes of the words, the mere presence of the rhyme context was apparently sufficient to focus processing upon sensory information as well. Both sensory and semantic information was apparently encoded,
Douglas L. Nelson
I42
TABLE VI
PROBABILITY OF CORRECT RECALLA S A FUNCTION OF INSTRUCTIONSA N D SENSORY SETSIZE Sensory set size Instructions
Small
Medium
Large
Mean
Imagery Repetition
.59 .46
.52 .4I
.50 ,38
.53 .42
Mean
.53
.46
.44
and this information appears to have been recapitulated by both original and switch cues. The main point of this section is that recall will be a function of sensory set size when the target is encoded in the absence of any specific contextual cues, but only if the test cue itself focuses upon sensory information. Recall will also be a function of sensory set size when the target is encoded in the presence of a rhyme context. In this case, it does not seem to matter what type of cue is presented at test, sensory or semantic, nor does it seem to matter whether the encoding strategy stresses repetitious rehearsal or interaction imagery. The common element running through all these results seems to be the emphasis upon sensory information, regardless of whether the test or study phase is involved.
c.
ASSOCIATIVE C O N T E X T CUES PRESENT AT S T U D Y
I. Test
Cues
As already indicated, when contextual cues are absent during the study phase, category size effects are consistently obtained for all types of extralist cues. In contrast, these effects are attenuated or eliminated altogether when semantic cues are present during the study phase. Under these conditions, Nelson and McEvoy (1979b) found that type of cue interacted with category size. As shown in Table VIT, recall varied with target strength, but it did not vary with category size. Recall from large taxonomic categories was essentially equivalent to recall from smaller categories. Comparison of the values shown in the bottom rows of Tables IV and VII indicates that category size effects are obtained when sensory but not when semantic cues are present during both study and test phases. Furthermore, only under these conditions were taxonomic category cues
Context and Category Size
I43
TABLE VII PROBABILITY OF CORRECT RECALLFOR TAXONOMIC CUES PRESENTED AT STUDY A N D TESTAS A FUNCTION OF STRENGTH A N D CATEGORY SIZI? Category size Strength Strong Weak Mean
Small
Medium
Large
Mean
.92 .75
.95 .81
.93 .82
.93 .79
.83
.88
.87 ~
~~
“Adapted from Nelson and McEvoy (1979b). Copyright 1979 by the American Psychological Association. Reprinted by permission.
more effective than word ending cues, specifically when category size was large. Finally, similar results were found for rhymes and meaningrelated associates (Nelson & McEvoy , 1979b, Experiment 2). 2.
Target Words
Nelson and Friedrich (1980, Experiment 4) found that target words having small associative sets were easier to recall than were those having large sets, regardless of the type of cue presented at test. However, this effect was limited to conditions in which context cues were absent during study. When meaning-related associates were presented as contextual cues, the effects of associative set size of the target were no longer apparent. Recall was essentially equivalent for targets with small and large associative sets, respectively averaging .77 and .79. This null effect was obtained when test cues were identical to the study context cues, and when these meaning-related study context cues were switched to word endings. As the above findings suggest, associative set size of the target essentially has no effect upon recall, regardless of type of test cue, when meaning-related contextual cues are available during the study phase of the task. Joined with the results of manipulating semantic category size for the cue, the implication is that the presence of meaningful context acts as a limiting condition for category size effects. In the absence of such context, recall varies with semantic set size; in its presence, it does not. Professor Forgetum’s failure to recall LOBSTER, which has a relatively large semantic set, and his success in recalling CHEESE, which also has a
144
Douglas L. Nelson
large semantic set, might have been related to the fact that LOBSTER was encoded in the absence of a specific context, whereas CHEESE was encoded along with the semantically related item BUTTER. D.
NOMINAL vs FUNCTIONAL SETSIZE
Nominal set size has been defined by counting the total number of different words elicited by some type of cue in a controlled association task. Functional set size refers to the number of concepts actually encoded during the presentation of a word, either alone or within some specific context. The nominal-functional distinction is important, since variations in both the availability and the type of contextual cues seem to influence functional set size. In other words, not all the items belonging to the category defined by a single instance are encoded, and the ones that are encoded are determined by contextual variations. Implicit in this distinction is the assumption that the memory representation for some specific word is not invariably limited to encoding specific sensory and semantic information corresponding to the physically presented item (Anisfeld & Knapp, 1968; Bower, 1967; Underwood, 1965, 1969). Instead, a number of related sensory and semantic concepts are presumed to be implicitly activated and encoded, with the number and type of concepts determined by contextual factors. This assumption forms the basis of the functional set size hypothesis. Accordingly, one important determinant of cued recall performance is the number of encoded concepts. The fewer the encoded concepts, the better is the recall for a single specific concept. In general, fewer concepts are encoded for small than for large sets, thus reducing competition at recall. That is, given that the test cue provides access to some portion of the sensory and semantic representations encoded during study, the likelihood of retrieving the specific representations corresponding to the physically presented item will be greater when there are fewer competing alternatives. The major role of context is to serve as a set of instructions for encoding various types of information automatically activated by the target word upon its presentation (e.g., Hillinger, 1980; Meyer, Schvaneveldt, & Ruddy, 1974; Swinney, 1979). Context serves to focus processing resources within a particular domain of features. When a target word is encoded in the absence of any specific contextual cues, its associative but not its sensory set size influences recall. The recall of familiar words having smaller nominal associative sets is better than recall of those having larger sets. However, remembering words belonging to large rhyme categories is just as likely as remembering words belonging to small rhyme categories. In the absence of rhyme context cues, sensory set
Context and Category Size
145
size influences recall only when sensory cues are presented at test. These findings suggest that when the environment fails to emphasize a specific domain of information, subjects ordinarily focus on meaning. Semantically related concepts but not rhyme-related concepts are encoded in proportion to their number. When the environment fails to provide an appropriate context, such context appears to be provided internally by encoding previously acquired semantic knowledge that is related to the item being processed. When target words are encoded in the presence of rhyme contexts, the ending set size of the target now influences recall-even with meaningrelated test cues. Considered in light of what happens in the absence of contextual cues, this result implies that rhyme-related concepts can be encoded, but that this encoding appears to be contingent upon a biasing rhyme context during the study trial. Only when processing operations are focused upon sensory information will sensory set size effects be in evidence. Finally, when familiar words are encoded in the presence of meaningful contexts, associative set size of the target no longer influences recall, no matter what type of test cue is used. Set size of the cue is also eliminated under these conditions. Only those specific concepts that are shared by the contextual cue and the target are apparently encoded. In this condition functional set size is smallest, since the fewest competing meanings are encoded. Recall should be and is equivalent for words having small and large associative sets; in fact, recall is generally higher under this condition than any other. The contextual conditions favoring the encoding of meaning-related and rhyme-related items appear to be reversed. Associative set size effects are obtained in the absence of biasing semantic contexts, and they are not obtained in their presence. Word ending set size effects are not found in the absence of biasing rhyme contexts, and they are found in their presence. The implication is that the presence and the nature of contextual cues determine what is encoded about the target experience. These cues merely act to focus attention to the semantic or to the sensory domain.
IV.
Conceptual Framework
Our conceptual framework has primarily evolved from the foregoing research. The basic orientation of this viewpoint is that the encoding of a familiar experience involves a complex interaction between current and previously encoded related experiences. The products of this interaction include both qualitative and quantitative aspects. Qualitative aspects refer
I46
Douglas L. Nelson
to the different types of information that are activated and encoded and include inforniation about appearance, name, and meaning. Quantitative aspects refer to the number of related past experiences that are functionally encoded. Thus, familiar experiences are analyzed through time, and this analysis results in a hypothetical mnemonic code, a code that is multifeatured in nature and that corresponds to the type and extent of the analysis (Craik & Tulving, 1975). Presumably, both type and extent can be influenced by the encoding context; therefore, the content or structure of the code is at least partially controlled by the contextual emphasis. Subsequent retrieval will be a function of the results of this encoding interaction, as well as of the relationship between this encoded information and the retrieval cue used to prompt recall. The framework described here represents an extension of the Sensory-Semantic Model (Nelson, 1979). For present purposes, this approach can be described by three major assumptions that concern structure, processing, and retrieval. A.
STRUCTURAL ASSUMPTIONS
The general consensus among cognitive psychologists is that the encoding of a familiar word involves the tagging of its features as they are represented in memory (e.g., Anderson & Bower, 1974; Bower, 1967; Tulving & Thomson, 1973; Underwood, 1969). A number of features have been ostensibly identified, and, somewhat arbitrarily, they include visual, phonemic, and meaning features. These can be classified into two general types, sensory and semantic. Sensory features describe words as they appear in the physical world, their visual configurations (e.g., Gibson, 1971). Although separable effects can be demonstrated, phonemic features are also included in this category (e.g., Nelson et al., 1974a). Meaning features characterize the significance or the potential interpretation of the stimulus and include its semantic, associative, and taxonomic implications. We assume that separate systems are involved in encoding each type of feature, and that each system consists of organized associative networks. The central organizational principle for each network is psychological similarity, with similarity defined in terms of A. Tversky 's (1977) contrast model. Accordingly, similarity is defined as the difference or contrast between the common and distinctive features of two or more stimuli. Similarity increases with the number of common features, and it decreases with the number of distinctive features. As suggested, cue-totarget similarity is operationally linked to and quantified by controlled association norms. A cue that elicits a target with a high probability is
Context and Category Size
147
considered to be psychologically more similar to that target than to one that is elicited with a low probability. While the similarity may be of a different type, we assume that it serves the same organizing function for both sensory and semantic domains. Thus, for both domains, the frequency with which one word is given as a response to some cue provides a unidirectional measure of their perceived similarity. Finally, as indicated, any particular cue presumably defines a category of a finite size, which is also estimated with controlled association norms. For associative categories, the criterion of membership is determined by strength: At least two subjects must give the same response before it is counted as a member of the category. For rhyme and taxonomic categories, even idiosyncratic responses are included if they meet the logical requirements for category membership (Nelson & McEvoy , 1979b). Figure 1 illustrates what the associative networks might look like for the word SALT when OLT, SALT, and A Type of Spice are presented as cues in controlled association tasks asking for a rhyme, an associate, as an instance of the taxonomic category. These cues are represented in rectangular nodes. The target nodes are represented in the ovals at varying distances from the cue. For example, when OLT is presented auditorily as a cue, the most frequently given response is HALT and one of the least frequent is WALT. Following Collins and Loftus (1975), these items are respectively represented as closer to or farther away from the cue. Psychological similarity is physically represented by distance. Furthermore, this distance is directional, moving from the cue to the target.
A TYPE OF SPICE
(T-)( =I(=) Fig. I . Hypothetical sensory-semantic networks for the word SALT.
148
Douglas L. Nelsun
Obviously, the distance from the target back to the cue might be different. It should also be noted that visual and phonemic codes have been collapsed for purposes of illustration. Presumably, each type of information is represented within a separate system. When SALT is presented as a cue and an associate is requested, we presume that it readily elicits the concept of saltness within the semantic system and that this concept activates related concepts within that system in proportion to their perceived similarity. The suffix “ness” is included to distinguish the semantic representation from the sensory representation and to suggest that the meaning code is abstract and potentially independent of the form of the input (e.g., Anderson & Bower, 1973; Nelson, Reed, & Walling, 1976; Pylyshyn, 1973). Note that within this system all the connections are between the concepts. Once a particular concept has been activated, its sensory representation can also be activated, if appropriate. There must be two-way communication between all types of codes. It might also be noted that, for SALT, the number of similar sensory representations and the number of similar semantic representations are quite limited. Thus, it evokes a relatively small sensory set and a relatively small semantic set. Other words evoke large sensory but small semantic sets, etc. However, it should be noted that each of these individual words can be subsumed by higher order concepts. The word SALT is a member of a relatively large taxonomic category, A Type of Spice, which contains approximately 20 exemplars. It is also a member of even larger categories, such as Something Eaten. The point is that the potentially relevant set is defined by the cue itself. All cues presumably serve a self-addressing function (e.g., Shiffrin & Atkinson, 1969). Figure 1 demonstrates another point. Interconnections are probably more likely among semantic concepts than among sensory representations. For instance, links between OCEANNESS and WATERNESS intuitively seem more likely than links between SALT and VAULT. Given the cue SALT and a semantic bias, it may activate saltness, which may activate waterness, which in turn may activate oceanness and OCEAN. The retrieval of OCEAN might be accessed through activation of an intermediate link. Given the same cue, SALT, and a sensory bias, the only route to WALT would be through the concept shared by both words, OLT. Although this suggestion is highly speculative at this juncture, the organization of sensory categories might be less complex. A modest correlational study was undertaken to explore this speculation further (Nelson, 1980). Assuming that subjects in a normative task retrieve instances through the category concept, positive correlations should be found between the sets of responses produced by the concept name itself and the sets produced by each of its instances. For example,
Context and Category Size
149
product-moment correlations should be high between the frequencies of responses produced by the cue OLT and by either its strongest instance, HALT, or a weak instance like WALT. Regardless of what cue is used to provide access to the category, MALT should be one of the most frequent responses. Similar expectations hold for taxonomic categories, but, to the extent that interitem connections are more likely, correlations between the sets of responses produced by the category name and by its instances should be reduced. To evaluate these expectations, rhyme norms were obtained for each of 35 word endings, and instance norms were collected for 44 different taxonomic category names. Then two additional groups of subjects provided rhyme instances either to the strongest instance from each ending category or to one of its weakest instances. Similarly, another two groups were given either the strongest or one of the weaker taxonomic category exemplars and were asked to generate another instance from the same category. Correlations between the sets produced by word ending categories and their strongest and their weaker instances averaged, respectively, .79 ( S D = .17) and .64 ( S D = .31). Thus, the correlations were high and positive, but somewhat higher for stronger instances. They also tended to be somewhat higher for small ( r = .83) than for large sets ( r = .66) and, even though all rhyme cues were presented auditorily, for categories having a single ( r = .81) compared with a variable spelling ( r = .57). Correlations between the taxonomic categories and their strongest and weaker instances were somewhat lower, as expected, but they presented a similar pattern. The average correlation between the sets produced by the category names and their strongest instances was r = .57 ( S D = .40); with their weaker instances this correlation dropped to r = .35 ( S D = .36). Finally, as with the rhymes, the correlations were somewhat higher for comparably small ( r = .69) as compared with large ( r = .51) sets. These correlations suggest that both types of information are highly organized. Associative structure as assessed by this procedure does not appear to change greatly when the same categorical information is accessed by cues of varying strength. MALT remains as a frequent response, regardless of whether the category is accessed by OLT, HALT, or WALT. Likewise, LETTUCE is a frequent response when accessed by GREEN VEGETABLE, BEANS, or CELERY. However, since the magnitude of the correlations seems to be higher for sensory categories, at least for those examined here, these categories appear to be less complexly organized, with fewer interitem links. Access from one rhyme instance to another is primarily through the concept shared by all items in the set.
I SO
B.
Douglas L. Nelson
PROCESSING ASSUMPTIONS
Access to any of the representations in the system presumably begins with an analysis of the physical features of the nominal stimulus (e.g., Craik & Lockhart, 1972; Nelson, 1979). In a visual task, encoding begins with an analysis of visual features. With familiar verbal stimuli, the results of preliminary analyses can provide direct access to semantic features or indirect access following phonemic activation (e.g., Kleiman, 1975; Meyer et a/., 1974; Rubenstein, Richter, & Kay, 1975). The relative order of access to semantic information from words presumably depends upon both task and context. For instance, if the items appear in sentential context, access can apparently be direct (Kleiman, 1975). If the items are encoded in the absence of meaningful context or if their serial order is being learned, the activation of phonemic features prior to semantic activation appears to be more likely (e.g., Nelson et al., 1974a; Nelson, Reed, & McEvoy, 1977). Order of access seems to differ for pictorial stimuli (Nelson et a l , , 1977; Seymour, 1973). However, regardless of the nature of the stimulus, the order of access assumption is not intended as a stage-by-stage, fixed sequence. Encoding does not proceed in a series of dichotomous steps with the requirement that each successive step be completed before the next begins. Encoding is assumed to be continuous, with several types of features being analyzed in parallel (e.g., Posner & Warren, 1972). Thus, a subset of features can be sampled, and this subset can be sufficient for activating information for succeeding codes. Access to established representations within either the sensory or the semantic systems is presumed to be reconstructive (e.g., Brown & McNeill, 1966; Horowitz & Prytulak, 1969; Jones, 1976). Any familiar stimulus is viewed as a retrieval cue for activating previously encoded information, Accordingly, retrieval represents a major component of the encoding of any familiar stimulus. When an item is presented in a memory task, its sensory and semantic features are analyzed, and the results of this analysis serve to activate the encoded results of past analyses. The Hoffding step (Martin, 1971) is theoretically accomplished through the principle of perceived similarity. The encoding operations applied to any nominal stimulus will lead to the activation and encoding of specific past information only to the extent that the results of the current analyses are perceived as similar to those past related experiences. Not all past knowledge will invariably be encoded (e.g., Light & Carter-Sobell, 1970; Tulving & Thomson, 1973), and, therefore, the size of the functionally encoded set can differ from the size of the nominal set. The principle of perceived similarity suggests that processing analyses
Context and Category Size
151
can be selectively focused toward either common or distinctive features. Additional attention to particular features increases the chances that those attributes will become focal mnemonic codes (Nelson, 1979). These codes presumably can be controlled by externally directed contingencies. Encoding operations can be influenced by instructions to use different learning strategies (e.g., Paivio, 1971), by requiring different judgments about the stimuli (e.g., Craik & Tulving, 1975; Hyde & Jenkins, 1973), or by using different types of contextual cues (e.g., Nelson et al., 1974b). These operations also can be influenced by the task itself (whether it involves recognition, paired associate, ordering, etc.), by item characteristics, and by subject characteristics and expectations (e.g., Frost, 1971; Nelson et al. 1977; B. G. Tversky, 1973). The focus of stimulus analyses influences what is encoded about the experience. However, as noted earlier, the encoded representation for any nominal stimulus is not solely determined by the focus of these analyses. Even though processing operations may be selectively directed toward either the sensory or the semantic domain, information within the opposing domain appears to be encoded. Operations directed toward semantic attributes do not activate only semantic codes. Sensory information appears to be encoded as well (Nelson, 1979). Similarly, operations directed toward sensory features do not result in exclusive sensory codes. Semantic information is also encoded (Nelson et al., 1979). Such between-dimension muItiple encoding is compatible with the assumption of two-way and probably automatic communication between established sensory and semantic representations. These considerations suggest that selectively directed processing analyses result in focal but not in exclusive codes limited to particular domains. Focal emphasis within a single domain of processing probably comes after activation of basic information in each domain (e.g., Swinney, 1979). Moreover, the findings discussed earlier in this chapter indicate that the focus of stimulus analyses strongly influences both the qualitative and the quantitative nature of what is encoded. Attempts to remember a word presented in the absence of specific contextual cues result in a relatively narrow sensory encoding and a relatively broad semantic encoding. When the environment fails to provide an appropriate context, such context seems to be provided internally by activating and encoding previously learned semantic concepts. The number of encoded concepts is proportionally related to the number of concepts in the set. Fewer concepts are encoded for smaller than for larger sets. Thus, given SALT during the study trial, some portion of its associatively related concepts is presumed to be activated and encoded, including pepperness, sugarness, waterness, etc. Furthermore, this study trial encoding does not seem to
152
Douglas L. Nelson
include physically similar words like HALT, MALT, FAULT, etc., even though these representations may be momentarily activated (Hillinger, 1980; Meyer et al., 1974). The encoding of sensory information appears to be limited to the specific features of the nominal item; that is, it is so limited when specific rhyme context cues are absent. When rhyme context cues are present during the study trial, related sensory representations appear to be both activated and encoded. Under these conditions sensory set size of the target influences recall even when meaning-related cues are shown at test. Thus, given HALT SALT at study, MALT, FAULT, VAULT, etc., are now apparently encoded. As with semantic concepts, the number of sensory concepts that are encoded is proportionally related to the number of concepts in the set. Fewer sensory concepts are encoded for smaller sets than for larger ones. Finally, it should be noted that associative set size of the target still has some influence on recall when the target is encoded in the presence of a rhyme context (Nelson & Friedrich, 1981 , Experiment 5 ) . The meaning of the target is still relevant, even though the environmental cues may bias encoding toward sensory information. When meaning-related context cues are present during the study trial, encoding is biased toward very specific semantic features of the target word. Neither ending set size nor associative set size of the target has any effect, nor does associative set size of the cue. When associatively related contextual cues are present in the environment of the study trial, recall of targets having large associative sets is just as likely as recall of targets having smaller sets. Similarly, recall associated with cues that nominally activate large sets is just as good as recall with cues that nominally generate smaller sets. Semantically related pairs of words apparently interact and mutually modify one another (Jacoby, 1974; Nelson, 1979). This mutual interactiveness has the effect of restricting the number of functionally encoded alternatives. The encoded set seems to be more precisely limited to the specific alternatives that are common to the presented cue and its target. Given PEPPER SALT, the meaning-related concepts of sugarness, waterness, oceanness, etc., are less likely to be encoded. The latter are apt to be distinctive features of saltness not shared by the concept pepperness. This interpretation suggests that the major role of context is to serve as a set of instructions to focus on either specific or general sensory information, or specific or general semantic information. This inference, of course, is based upon the effects of category size under various contextual conditions. Still needed is a processing assumption that will explain why encoding is not always limited to the specific representations linked to the presented word. This need might be fulfilled by the concept of spreading
Context and Category Size
153
activation (Collins & Loftus, 1975). The processing of a word at least momentarily activates related sensory and related semantic information (Hillinger, 1980; Meyer et al., 1974; Neely, 1977). The products of this activation presumably can be encoded, but whether they are encoded will be dependent upon the task, the instructions, and the specific nature of the contextual cues. Thus, contextual cues and other task variables act to focus encoding resources on specific types of information activated upon presentation of the nominal item. This focus in turn controls the breadth of sensory and semantic encoding.
c. RETRIEVALASSUMPTIONS Figure 2 provides an illustration of some of the stages that are involved in achieving access to a recently encoded experience in the presence of some retrieval cue. As shown, the first stage involves presentation of the test cue itself. When a test cue is presented, it is presumably encoded, and, as if it were an item appearing on the study trial, this encoding begins with an analysis of visual, phonemic, and semantic features. The results of this analysis set the stage for the second phase, a generation phase during which sensory or semantic categories likely to contain encoded target information are activated. Finally, provided that target inTEST TRIAL PHASES PRESENTATIONOF TEST CUE
GENERATION OF SETS DEFINE0 B Y CUE
RECOGNITION OF STUDY TRIAL ENCODING
I
SENSORY SET CONTAINING TARGET
1
ENCOOEO SENSORY INFORMATION
ENCOOEO SEMANTIC INFORMATION SENSORY OR SEMANTIC CUE
SEMANTIC SET CONTAINING TARGET
ENCOOED SENSORY INFORMATION
ENCOOEO SEMANTIC INFORMATION
Fig. 2 . Hypothetical test trial sequence of phases in cued recall
154
Douglas 1,. Nelsnn
formation has been accessed, the third phase can be initiated, which requires recognition of sensory and semantic information encoded during the study trial. What follows is a description of the generation and recognition phases, including an attempt to localize the effects of various manipulations, at different phases of the cued recall process. Following the initial activation of information about the cue, encoding is assumed to be directed toward a particular domain, toward either sensory or semantic networks. The focused domain is usually determined by instructions to the subject about the relationship between each cue and the sought-after target. Subjects might be told that all the cues rhyme with the targets or that they are all meaningfully related to them. In this case, the focus of the generation phase appears to be easily limited to the emphasized information. Intrusion of information from the opposing domain does not seem to occur even when the potential for interference is built into the list, as when it consists of pairs like STREAM BEAM and HOOK BROOK. However, if subjects must randomly switch their focus from one domain to another during the test trial, then interference from the opposing domain will be in evidence even when the subject is told how each cue is related to its target (Nelson, McEvoy, & Friedrich, in press). Thus, unless the test episode requires switching between domains of information, subjects can apparently direct retrieval efforts to either sensory or to semantic networks with some precision. We assume that, once the appropriate unidimensional domain has been selected, information linked to the nominal cue provides access either to prestored codes or to subsets of such codes. An extralist cue operates by providing access to a category within the network, and this information in turn activates codes linked to it. For instance, from Fig. 1, the rhyme cue SALT might be presented as a cue for the list word MALT. To be successful, this cue would have to activate the categorical concept OLT, which in turn would activate related sensory units, possibly including HALT, FAULT, and MALT. Similarly, the semantic cue SALT as a prompt for the target OCEAN would tend to activate saltness, and this concept in turn might activate several related concepts, including pepperness, sugarness, oceanness. Extralist taxonomic category cues presumably operate in the same manner during this phase. For extralist cues, both cue-to-target similarity (strength) and category size are critical determinants of the likelihood of activating the target during the generation phase. We assume that cues are more likely to provide access to those targets that are perceived as more similar. In addition, given equivalent similarity, we assume that the chances of accessing the target are proportionally greater when the category containing it is smaller than when it is larger. These two assumptions serve to explain
Context and Category Size
155
why recall is a consistent function of both strength of the target and category size in relation to the test cue. If the target is not accessed by the cue, the likelihood of correctly initiating the recognition phase will be zero. Furthermore, the emphasis upon the utilization of prestored codes under these conditions is intentional. Category size effects in cued recall are obtained with extralist rhymes, associates, word endings, and taxonomic category cues, and these effects seem to be independent of target strength, presentation rate, test rate, and modaIity . Thus, for extralist cues not involved in cue switches from study to test, variables that influence the overall amount of recall fail to have any influence upon the effects of cue set size. Furthermore, these effects are of the same order of relative magnitude on an immediate as on a delayed test. They did not increase over the length of the retention interval as might be expected if such effects were more episodically determined. One implication of these findings is that cue set size effects are independent of what is encoded about the target during the study phase. This suggestion does not mean, however, that what is encoded during the study trial is without importance. The effects of test cue set size can be modified by simply presenting these cues along with their targets during the study trial. When rhymes are available during study and presented again at test as cues, the sensory set size effect increases in magnitude relative to what happens with extralist rhymes. When identical semantic cues are available during both phases, the semantic set size effect is reduced or eliminated altogether. Thus, the likelihood of generating the target experience can be influenced by events taking place during the study trial. Encoding information about the cue and its relation to its target affects the likelihood of generating the target during the test phase. This influence can be negative, as when rhymes are used, or it can be positive, as when meaning-related cues are involved. Thus, the generation phase is not solely dependent upon the retrieval of only prestored information. The encoding of test cues during the study trial, like the encoding of targets, can produce at least a temporary modification of prestored information. This modification, however, is far from complete. Recall remains as a direct function of preexperimental strength even when sensory or semantic cues are present during the study trial. This conceptualization of the generation process suggests that access to the target can be attained by activating prestored categories of information or by reactivating encoded subsets of those categories. This generation process, however, does not seem to be constrained within a rigid format in which each potential response is generated in order of strength with each generation followed by a recognition check (e.g., Bahrick, 1970; Nelson et al., 1974b). Such a process would require interactions between
I56
Douglas L. Nelson
target strength and category size. Recalling a strong target that is high in the hierarchy should be equally likely for small and large categories. Recalling an item that is weak or low in a small category should be more likely than remembering one that is equally weak but a member o f a large category. The results, however, indicate that these factors do not interact under a wide variety of encoding and retrieval conditions (Nelson & McEvoy, 1979a), including manipulations of test rate. Target strength has parallel effects regardless of category size. Furthermore, encouraging subjects to use a generate-recognize strategy when recall has failed produces mixed results, with both facilitative (Rabinowitz, Mandler, & Patterson, 1977) and detrimental effects obtained (Nelson & McEvoy. 1979a, Experiment 3). One implication of these findings is that, unless the functional set has been reduced to a single item, potential targets are not generated one at a time in some serial fashion. Furthermore, the notion that all instances of the set are generated seems equally unlikely. If they were, set size of the test cue should not influence recall, nor should target strength. The target would invariably be included in the set, virtually guaranteeing access. Thus, the most reasonable assumption seems to be that some portions of the potential targets are generated in parallel by the cue. Multiple activation at retrieval may not be unlike that occurring during the study trial. The smaller the functional category and the more strongly the target is related to the cue, the greater the likelihood of its inclusion in the sample. If the target is not included in the initial sampling, the second phase of the recall process cannot accurately be initiated. In this case, subjects presumably return to the test cue and generate another sample, a process that could continue indefinitely. However, assuming sampling with replacement, regenerating the same candidates becomes more likely with repeated generation, and, eventually, the criterion for stopping recall will be reached (Rundus, 1973; Shiffrin, 1970). Assuming that the target is generated as a member of either the sensory or the semantic set defined by the cue, the next stage in the recall process can proceed. Accurate performance during this phase requires recognition of the study trial encoding. As suggested by Fig. 2, correct recognition will be based upon accurate recapitulation of information encoded about the target during the study trial-information that can include either or both sensory and semantic features. This recognition process is assumed to be redintegrative in nature (Horowitz & Prytulak, 1969). This assumption is necessary because target information generated by the test cue is unidimensional and because it presumably overlaps with, but may not be identical to, the information encoded about the target during the study trial. Thus, the assumption is that the target produced in the generation
Context and Category Size
157
phase acts as a retrieval cue for recapitulating recently encoded information. This information must be integrated and formed into a single response, which passes the decision criteria for the recognition phase of recall (e.g., Kintsch, 1978). We assume that the likelihood of correctly redintegrating this single response is an inverse function of the number of similar sensory and semantic functionally activated during the study trial. This complex chain of events is easiest to demonstrate through examples. If the word SALT is encoded during the study trial and MALT is used as a rhyme test cue, its effectiveness will be partially determined by the likelihood of accessing the sensory network, and within this network, by the likelihood of accessing the OLT category, and within this category, by the likelihood that the sensory unit SALT will be generated. Similarly, if OCEAN is used as a semantic test cue, its effectiveness will depend upon the chances of accessing the semantic network, the oceanness category, and the semantic unit saltness. Finally, if generated within either network, correct recognition will depend upon the number of similar sensory and semantic codes functionally activated during the study trial. If encoded in the absence of contextual cues, present findings indicate that the sensory encoding will be relatively specific and the semantic encoding relatively broad, including a number of competing meaningrelated concepts. However, the chances of correctly redintegrating the word SALT would still be greater than the chances of accurately redintegrating an item having an even larger associative set. The smaller the pool of competing codes within either domain, the greater will be the probability of redintegrating the single specific sensory-semantic code linked to the item actually presented by the experimenter. Presumably, the number of competing codes can be controlled by the presenceabsence and the nature of the contextual cues present during the study trial. Just as the presence of rhyme context cues during the study trial increases the number of competing sensory codes, the presence of meaning-related context cues decreases the number of competing semantic codes. Hence, the recognition phase of the recall process invoives an attempt to regenerate the study trial experience and, out of this information, to produce the single item actually presented. The fact that both sensory and semantic set size effects can be obtained with either sensory or semantic cues suggests that access to the study trial experience can be attained from either domain. That is, sensory cues give rise to encoded semantic information, and semantic cues give rise to encoded sensory information. Cues are effective in part because they facilitate the retrieval of information lying outside of their defined domains. This makes sense in the present framework. We assume that the unidimensional representation activated during the generation phase
Douglas L. Nelson
15R
serves as a redintegrative retrieval cue for reconstructing the original encoding experience, an experience that is presumably represented within both sensory and semantic networks. Nevertheless, once the initial encoding experience has been reactivated, the problem for the rememberer is to produce a single response. The process by which such selection is made is completely unclear to us at this time. The only facet that seems clear is that this process seems to vary in difficulty, depending upon the number of competing codes within each domain. V.
Some Implications
One implication of the foregoing conceptual framework lies in its complexity. Professor Forgetum should have either searched more diligently for a pen or, alternatively, he should have given more credence to the Ninth Commandment. Other somewhat more important implications of the data per se concern the encoding specificity hypothesis (Tulving, 1979; Tulving & Thomson, 1973) and the levels of processing conceptualization (Craik, 1979; Craik Rr Lockhart, 1972; Craik & Tulving, 1975). The primary assumption of the encoding specificity hypothesis is that, in order for any retrieval cue to be effective at test, it must have been encoded during the study trial. Of course, when contextual cues are absent during this phase, the encoding of any particular extralist cue cannot be precisely specified. However, it would seem reasonable to assume that the likelihood of encoding any given extralist cue is greater when it is a member of a small as compared with a larger set. With this corollary assumption, the encoding specificity hypothesis serves as an excellent explanation for effects linked to associative set size of the target. In the absence of contextual cues, subjects should be more likely to encode the extralist meaning-related test cue if the target normally evokes a smaller associative set. The result is that targets having small sets are easier to recall. Similarly, this hypothesis explains the reduction of associative set size effects when the meaningrelated test cues are also present during the study phase. Under this condition, the cue is most likely to be processed and should be highly effective in reactivating the episodic semantic encoding of the target. Furthermore, under this condition, switches in semantic aspects of the cue should and do produce substantial decrements in both recall and recognition (Fisher & Craik, 1977; Nelson et al., 1979; Tulving & Thomson, 1973; Tulving & Wiseman, 1975; Wallace, 1978). Presenting semantically related pairs of words during the study trial represents one condition in which the encoding of even strongly related items is not very likely
Context and Category Size
159
unless they are directly related to the pair being processed. Thus, these considerations suggest that the encoding specificity hypothesis can be used to explain associative set size effects. Despite this success, however, this hypothesis does not seem to provide adequate account of other findings, particularly those dealing with manipulations within the sensory domain. The results of several studies suggested that, when words are encoded in the absence of contextual cues, rhyme-related items that might potentially serve as extralist cues are apparently not encoded. According to the encoding specificity view, extralist rhyme cues should be ineffective. As potential cues, they would not have been encoded during the study episode. However, extralist rhyme cues are effective relative to free recall, and sometimes more effective than semantic cues (e.g., Nelson et al., 1974b). Furthermore, it is difficult to understand how the encoding specificity hypothesis would explain why rhyme set size effects are found when the rhyme cue is present during both study and test phases, let alone results showing that these effects appear to increase in magnitude under this condition. Finally, it is not clear how this idea would explain how associative set size of the target inffuences recall when word endings (or rhymes) are used as extralist cues. To us, the most reasonable way to escape these difficulties was to choose a conceptualization that could rely either on prestored information or on encoded subsets of such information. Thus, cues can be effective because they have been specifically encoded during the study phase of the task, or because they provide access to prestored codes, which, in turn, can provide access to specifically encoded information. As suggested, the data of our experiments also have implications for the levels of processing framework (see, e.g., Craik, 1979). Much, but not all, of the research underlying this conceptualization involves the presentation of rhyme pairs or meaning-related pairs during an encoding phase, followed at test by the presentation of the study context as a retrieval cue (e.g., Craik & Tulving, 1975; Fisher & Craik, 1977; Morris, Bransford, & Franks, 1977; Nelson et al., 1974b). Depending somewhat on the presentation rate, semantic cues are typically more effective than sensory cues. This result has been interpreted as favoring the importance of the principIe of “depth” of encoding, with deeper and more semantic encodings inherently superior to more shallow sensory encodings (e.g., Fisher & Craik, 1977). The results of the present experiments, however, suggest that the depth principle cannot be properly evaluated under these conditions. Presenting pairs of rhymes during the study phase apparently encourages broad sensory processing, with many rhyme-related words activated and encoded. In contrast, presenting pairs of associatively related words apparently encourages a narrow and specific encoding, with
1 60
Douglas L. Nelson
few alternatives encoded. Regardless of the type of code involved, the greater the number of competing items, the poorer the subsequent recall will be. Thus, under the present conditions of comparison, the type of encoding (sensory or semantic) will be confounded with the number of alternatives likely to be encoded. Semantic cues are more effective, but this effectiveness might just as easily be attributed to the number of competing codes activated during the study trial as to the depth of encoding. We are not advocating abandonment of the levels of conceptualization on the basis of this single comparison. What we are advocating is that investigators who claim interest in the nature of representational processes be wary of the potential complexities inhering in the different systems. Results showing that both sensory and semantic cues facilitate the retrieval of information from the opposing domain provide one more inkling of these complexities. What should be abandoned is the tendency to render this rich intricacy down to a few simple principles.
REFERENCES Anderson, J . R . The slatus of arguments concerning representations for rnenlal imagery. Psvc.holo~~ictrl Review. 1978, 85, 249-277. Anderson, J . R . , & Bower, G . H. Humtrri assuc~ia~ive inrwiory. Washington, D.C.: Winston, 1973. Anderson, J . R . , & Bower, G . H. A propositional theory of recognition memory. Mrniory & Cogniriun , 1974, 2 , 406-4 12. Anisfeld, M.,& Knapp, M. Association. synonymity, and directionality in false recognition. Journtrl of E.xperimmttrl Psychology, 1968, 7 7 , 17 I - 179. Bahrick, H. P. Two-phase model for prompted recall. P s y c h o l o g i c d Review. 1970, 77, 215-222. Bower, C. H. A multiconiponent theory of the memory trace. The Psychology of Learning arid Morivarion, 1967, 1, 229-325. Brown. R . . & McNeill. D.The “tip of the tongue” phenomenon. Jourtial of Verbal Leitrni~igarid Verbal Behavior, 1966, 5 , 325-337. Collins, A. M . ,& Loftus. E. F. A spreading-activation theory of semantic processing. Psychological Review. 1975, 82, 407-428. Craik, F. I. M. Levels of processing: Overview and closing comments. In L. S. Cermak & F. 1. M . Craik (Eds.), Levels r,f processing a d human memory. Hillsdale, N.J.: Lawrence Erlbaum, 1979. Craik, F. 1. M., & Lockhart, R. S. Levels of processing: A framework for inernory research. Jourtid of Verbol Leirrning and V e r h l Behavior, 1972, 11, 67 1-684. Craik. F. I. M . , & Tulving, E. Depth of processing and the retention of words in episodic memory. Journtrl of E.rperirnenrtrl P.\yc,hology: Genernl. 1975. 104. 268-294. Eagle, M.. & Ortof, E. The effect of level of attention upon “phonetic” recognition errors. Jourtial qf Vvrbtrl Learning and Verbal Behavior, 1967, 6 , 226-231. Einstein, C. 0 . .& Hunt, R. R. Levels of processing and organization: Additive effects of individual item and relational processing. Jourriiil qf E.xperirnental P sych ology: Human [.rartiing arid Memurv, 1980, 6 , 588-598.
Context and Category Size
161
Fisher, R. P., & Craik, F. I . M. Interaction between encoding and retrieval operations in cued recall. Journal of Experinrental Psychology: Humun Learning and Memory, 1977, 3, 701 -71 I . Frost. N. Clustering by visual shape in the free recall of pictorial stimuli. Journal of Experimental P svchology, 197 I , 88, 409-4 13. Gibson, E. J. Perceptual learning and the theory of word perception. Cognitiw P.rycho/og~,1971,2, 351-368. Hillinger, M. L. Priming effects with phonemically similar words: The encoding-bias hypothesis reconsidered. Meinor! & Cognition, 1980, 8, 115-123. Horowitz, L. M . , & Prytulak, L. S . Redintegrative memory. Psychological Review. 1969, 76, 5 19-53 I . Hyde, T. S., & Jenkins, J . J . Recall for words a h a function of semantic, graphic, and syntactic orienting tasks. Journal of Verbal Learning and Verbal Behavior, 1973, 12, 471-480. Jacoby, L. L. The role of mental contiguity in memory: Registration and retrieval effects. Journal of Verbal Learning und Verbul Behuvior, 1974, 13, 483-496. Jones, G. V. A fragmentation hypothesis of memory: Cued recall of pictures and of sequential position. Journal of Experiinental Psychology: General, 1976, 105, 217-293. Kintsch, W. More on recognition failure of recallable words: Implications for generation-recognition models. Psyc-holigical Review. 1978, 85, 470-473. Kleiman, G. M. Speech recoding in reading. Journal of Verbal Learning and Verhul Behavior, 1975, 14, 323-340. Light. L. L . , & Caner-Sobell, L. Effects of changed semantic context on recognition memory. Journal of Verbal Learning and Verbal Behuvior, 1970, 9, 1 - I 1 . Martin, E. Verbal learning theory and independent retrieval phenomena. Pvchological Review, 1971, 78, 314-332. McEvoy, C. L., & Nelson, D. L. Encoding strategy and rhyme set size. In preparation, 1981. Meyer, D. E., Schvaneveldt, R . W . , & Ruddy, M. G. Functions of graphemic and phonemic codes in visual word recognition. Memory & Cogni~ion,1974, 2, 309-322. Moms, C . D., Bransford, J . D., & Franks, J. J. Levels of processing versus transfer appropriate processing. Journal of Verhul Leurning und Verhuf Behavior, 1977, 16, 519-533. Neely, J. H. Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limited capacity attention. Journul ojExperimentu1 Psvchology: General. 1977, 106, 226-254. Nelson, D. L. Remembering pictures and words: Appearance, significance, and name. In L. S . Cermak & F. I . M. Craik (Eds.), Levels ofprocessing and human mernory. Hillsdale, N.J.: Lawrence Erlbaum, 1979. Nelson, D. L. The rhyme riddle. Paper presented at the First Colorado Conference on Learning and Memory, Boulder, June 1980. Nelson, D. L., Brooks, D. H . , & Borden, R. C. Theeffectsofformal similarity: Phonetic, graphic or both'? Journal of Experirnentul Psychology 1974, 103, 91-96. (a) Nelson, D. L . , & Friedrich, M . Encoding and cuing sounds and senses. Journal of Experimental Psyc.hologyc Human Learning and Memory, 1980, 6 , 7 17-73 I . Nelson, D. L., & McEvoy, C. L. Effects of retention interval and modality on sensory and semantic trace information. Memory & Cognition, 1979. 7, 257-262. (a) Nelson, D. L., & McEvoy, C . L. Encodingcontext and set size. JourriolofE.rperirnentu1 Psychology: Human Learning und Memory, 1979, 5, 292-314. (b) Nelson, D. L . , McEvoy, C. L . , & Friedrich, M . A . Extralist cuing and retrieval inhibition. Journui of Expet.imenfa1 Psychology: Leurning. Memory and Cognitiotr , in press. Nelson. D . L . , Reed, V. S . , & McEvoy, C. L. Learning to order pictures and words: A model of sensory and semantic encoding. Journal of E.rperirnenta1 Psychology: Huinun Learning arid Memorv, 1977, 3, 485-497.
I62
Douglas L. Nelson
Nelson, D. L . , Reed, V . S., & Walling, J . R . The pictorial superiority effect. Journal of Exp. f.vycho/ogv : Hutnun k r r n i n g trnd Mcniory, 1976, 2, 523-528. Nelson, D. L., Walling, J . R., & McEvoy, C. I.. Doubts about depth. Jurrrncrl ufExperirnental Psychology: Huintrn Leurning and Menwrv, 1979, 5 , 24-44. Nelson, D. L., Wheeler. J. W., Jr., Borden, R. C . , tk Brooks, D. H . Levels of‘processing and cuing: Sensory vs. meaning features. Jorcrnd of Experintenral Psy-holoxy. 1974, 103, 971-977. (b) Paivio, A. / m i p r y und verbal procrsses. New York: Holt, 1971. Posner. M. I . , & Warren, R. E. Traces, concepts, and conscious construction. In A. W. Melton & E. Martin (Eds.). (’oding proc.esses in hutnun nieniury. Washington, D.C.: Winston, 1972. Pylyshyn, Z. W. What the mind’s eye tells the mind’s brain: A critique of mental imagery. Psyckological Br~llerin.1973. 80, 1-24, RabinowitL, J . C., Mandler, G . , & Patterson, K . E. Deterininants of rccognition and recall: Accessibility and generation. Journul uj‘ Experinrentul Psychologv: General, 1977, 106, 302-329. Rubenstein, H , Richter. M. L., & Kay, E. J . Pronounceability and the visual recognition of nonsense words. Jorrrntrl of’ Verbul Learning and Verhtrl Brhovior, 1975, 14, 65 1-657. Rundus, D. Negative effects of using list items as recall cues. Journul of’Vcrhal Learning and Verhcil Rehtrviur. 1973, 12, 43-50. Seymour. R. H . K . A model for reading, naming and comparison. Brifish Journ~lo f f s y c h o l o g ~ ~ , 1973. 64, 35-49. Shiffrin, R. M. Memory search. In D. A . Norman (Ed.), Models of hunrcrn nwnory. New York: Academic Press, 1970. Shiffrin, R. M., & Atkinson, R. C. Storage a i d retrieval processes in long-term memory. Psy&/og;u r l Rei~iew,1969, 76, 179-193. Swinney. D. A. Lexical access during sentence comprehension: (Re) Consideration of context effects. Jorrrnul I$ Vc,rbtrl Leurning untl Verbal Behavior. 1979, 18, 645-659. Tulving. E . Relation between encoding specificity and levcls ofprocessing. In L. S. Cermak Br F. I . M . Crdik (Eds.), Lcrrls qf’procc,.s.sir~g und hurnun rneniory. Hillsdale, N.J.: Lawrence Erlbaurn, 1474. Tulving, E., & Thomson. D. M. Encoding specificity and retrieval processes in episodic memory, P.vyc.hologicul Rei’iew, 1973. 80, 352-373. Tulviiig. E . , & Wiscman. S . Relation bctwccn rccognition and recognitinn failure of recallable words. Bulletin i$ rlte P.vychonornic Society, 1975, 6, 79-82. Tversky, A. Features of similarity. Psyc+wlogicul Review. 1977, 84, 327-352. Tversky, B. G . Encoding processes in recognition and recall. Cognirive Psyc’hology, 1973, 5, 275 287. Underwood. B . J Degree of learning arid the ineasureriient of forgetting. Jorrmul of‘Verhul Leurrtirig uml Verbti/ Beha\ior. 1964, 3. I 12- 129. Underwood, B. J . False recognition produced by implicit verbal responses. Jorrrnul ofE.rperimcvlral f.vycholog\. 1965, 70, 122-129. Underwood. B. J . Attributes of memory, P.sychologit~ulRei.iew, 1969, 76, 559-573. Underwood. B. J . , & Freuntl, J . S. Errors in recognition learning and retention. Journal ofExperinwnrul Psychology, 1968, 78, 55-63, Wallace, W. P. Recognition failure of recallable words and recognizable words. Journal of Experimen/ul P s y c ~ h o l o , q i Hrontrn : LetrrniJrg und M e n w r y . 1978, 4, 44 1-452.
FREQUENCY, ORTHOGRAPHIC REGULARITY. A N D LEXICAL STATUS IN LETTER A N D W O R D PERCEPTION' Dominic W . Massuro, James E . Jastrzembski, and Peter A . Lucas2 UNIVERSITY O F WISCONSIN MADISON. WISCONSIN
I. Language Processing Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. Orthographic Structure and Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Experiments I add 2: Bigram Frequency vs Regularity . . . . . . . . . . . . . . . . . . . . . B. Experiment 3: Log Bigrani Frequency v5 Regularity . . . . . . . . . . . . . . . . . . . . . . . C. Experiment 4: Replication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. Experimenl 5: Lexical S t a t u s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E. Experiment 6: Frequency, Regularity, and Lexical S t a t u s . . . . . . . . . . . . . . . . . . . 111. Orthographic Structure and Conscious Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . . . . Experiment 7: Ovcrt Judgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I V . General Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Related Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
164 166 168 178 183 185 188
I 90 190 195 195 198 I99
Perceiving and understanding involves a wonderfully adept integration of our immediate environment with prior knowledge and experience. The knowledge of the reader, for example, is as important as, or more important than, the information on the printed page. One compelling issue in reading research is how the reader's higher order knowledge of the lanIThis research was supported by funds from the National Institute of Education to the Wisconsin Research and Development Center for Individualized Schooling, University of Wisconsin, Madison, Wisconsin 53706. Kenneth R. Paap and Sandra L. Newsome made helpful comments on an earlier report of this research. 2D. W . Massaro is now affiliated with the University of California, Santa Cruz, California 95064; J . E. Jastrzembski is with Needham, Harper & Steers Advertising, Inc., 300 East Wacker Drive, Chicago, Illinois 60601; and P. L. Lucas with the Department of Psychology, Carnegie-Mellon University, Pittsburgh, Pennsylvania I52 13. Correspondence concerning this work should be directed to Dominic W . Massaro, Program in Experimental Psychology, University of California, Santa Cruz, California 95064. I63 THE PSYCHOLOGY OF LEARNING AND MOTIVATION. VOL. 15
Copyright 0 1981 by Academic Press, Inc All nghts of reproduction In any form reserved. ISBN 0-12-543315-8
Dominic W. Massaro et al.
I64
guage interacts with lower level perceptual analyses. The specific question addressed in the present article is how the reader’s knowledge about orthographic structure is combined with the information derived from visual featural analysis in letter and word recognition. Visual featural analysis refers to the evaluation of the component visual properties of letters leading to letter and word recognition. Orthographic structure refers to the spelling constraints in a written language. Given the considerable amount of predictability in English writing, we ask how the reader utilizes this orthographic structure in word recognition.
I.
Language Processing Model
Evaluation of the contributions of visual features and orthographic structure to word recognition can be facilitated by a detailed description of the processes involved in reading. The description we use is part of a more general model of language processing (Massaro, 1975, 1978, 1979a; Massaro, Taylor, Venezky , Jastrzembski, & Lucas, I980b). According to the model illustrated in Fig. I , reading can be viewed as a sequence of processing stages. At each stage of processing, memory and process components are represented. Each memory component (indicated by a rectangle) corresponds to the information available at a particular stage of processing. Each process component (indicated by a circle) corresponds to the operations applied to the information held by the memory component. The memory components are temporary storages except for long-term memory, which is relatively permanent. It is assumed that long-term memory supplements the information at some of the processing stages. During reading, the light pattern reflected from a display of letters is transduced by the visual receptors as the feature defection process detects and transmits visual features to preperceptual visual storage (see Fig. 1).
w LONG TERM MEMORV
PRINTED TEXT
-
Fig. 1 . Schematic representation of stages of processing in reading
Letter and Word Perception
165
As visual features enter in preperceptual visual storage, the primary recognirion process attempts to transform these isolated features into a sequence of letters and spaces in synthesized visual memory. To do this, the primary recognition process can utilize information held in long-term memory. For the accomplished reader this includes a list of features for each letter of the alphabet along with information about the orthographic structure of the language. Accordingly, the primary recognition process uses both visual features in preperceptual storage and knowledge of orthographic structure in long-term memory during the synthesis of letter strings. The primary recognition process operates on a number of letters simultaneously (in parallel). The visual features detected at each spatial location of the letter string define a set of possible letters for that position. The primary recognition process chooses from this set of candidates the letter alternative that has the best correspondence to the detected visual features. However, the selection of a letter can be facilitated by the reader’s knowledge of orthographic structure. We assume that orthographic structure is utilized in the following manner. Upon presentation of a letter string, the primary recognition process begins integrating and synthesizing featural information passed on by feature detection to preperceptual visual storage. Featural information is resolved at different rates, and there is some evidence that gross features are available before the more detailed features (Massaro & Schmuller, 1975). As a result, the primary recognition process is faced with a succession of partial information states. These partial information states are supplemented with knowledge about orthographic structure. Assume, for example, an initial th has been perceived in a letter string, and the features available for the next letter eliminate all alternatives except c and e . The primary recognition process would synthesize e without waiting for further visual information, since initial the is acceptable, whereas initial rhc is not. The primary recognition process transmits letter information to synthesized visual memory. Figure 1 shows how the secondary recognition process transforms this synthesized visual percept into a meaningful form in generated abstract memory. We assume that secondary recognition attempts to translate the letter string into a word. The secondary recognition process makes this transformation by finding the best match between the synthesized letter string and a word in long-term memory. Each word in long-term memory contains both perceptual and conceptual codes. The word that is recognized is the one whose perceptual code gives the best match and whose conceptual code is more appropriate in a particular context. Analogous to primary recognition, knowledge of orthographic structure also can contribute to secondary recognition; word recognition can occur without complete recognition of all of the component letters.
I66
Dominic W. Massaru et al.
Given the letters hea and the viable alternatives h and d in final position, only d makes a word, and therefore word identification (lexical access) can be achieved (Massaro, 1977). 11. Orthographic Structure and Recognition
Our goals in the present series of experiments are to provide a better understanding of primary and secondary recognition and to determine which aspects of orthographic structure the reader knows and uses. To assess how readers utilize knowledge about the structure of written language, it is necessary to state various descriptions of this structure and then to determine how well these descriptions capture reading performance. Venezky and Massaro (1979), Massaro, Venezky, and Taylor (1979), and Massaro er a1. ( I980b) have defined two broad categories of orthographic structure: statistical redundancy and rule-governed regularity. The first category includes all descriptions derived solely from the frequency of occurrence of letters and letter sequences in written texts. The second category includes all descriptions derived from the phonological constraints in English and from scribal conventions for the sequence of letters in words. Although a change in one category would not necessarily affect the other, the two categories overlap to some degree. Even with some overlap in these categories, it is of interest, first, to learn whether one of these categories reflects the manner in which readers store knowledge of orthographic structure and, second, to determine precisely which specific description within that category has the most psychological reality. Massaro et a / . (1979, 1980b) contrasted a specific statisticalredundancy description with a specific rule-governed description by comparing letter strings that varied orthogonally with respect to these descriptions. The statistical-redundancy measure was the summed token position-sensitive single-letter frequency. The rule-governed regularity measure was a preliminary set of rules similar to those presented in Table I of this article. Nonword letter strings were selected that represented the four combinations formed by a factorial arrangement of high or low frequency and regular or irregular. In a series of experiments utilizing a target-search task, subjects were asked to indicate whether or not a target letter was present in each of these letter strings. This task is assumed to measure the visual recognition of the letters in the letter string. Both accuracy and reaction-time measures indicated psychological reality for both the single-letter frequency and our regularity descriptions of orthographic structure.
Letter and Word Perception
167
Massaro et a / . (1980b) formalized the language processing model to provide a quantitative description of the facilitative effect of orthographic structure on task accuracy. The basic assumption of the model is that knowledge of orthographic structure contributes an independent source of information about the letter string. By an independent source of information, we mean that knowledge of orthographic structure does not modify or direct the feature detection process. Rather, information about visual features and orthographic structure accumulates from sources that do not interact. Since information about structure supplements featural information, fewer visual features are necessary to resolve well-structured than poorly structured strings. The model was applied to the target-search task by formalizing a decision algorithm assumed to be used by the subject when faced with partial information. The model provided a good quantitative description of the accuracy results. The parameters of the model were psychologically meaningful, and the parameter values corresponding to the number of letters seen in the test string provided a quantitative measure of the contribution of orthographic structure. According to the model, readers were able to recognize two additional letters in brief presentations of well-structured strings as compared with poorly structured strings. This is a substantial effect, considering that two letters represent one-third of the six-letter test string. These results indicate that we had developed good initial approximations of both a description of orthographic structure and the means by which structure and visual features combine during word recognition. This bolstered our hope that a precise description of orthographic structure can eventually be determined and that a thorough understanding of the word recognition processes in reading can eventually be obtained. The factorial design of the Massaro et al. (1980b) experiments contrasted just one measure of rule-governed regularity with one measure of statistical redundancy. Therefore, a large number of post hoc correlational analyses was conducted to evaluate a wide range of measures of orthographic structure. This was a first step toward refining our initial measures of orthographic structure. Through these correlations, it was possible to determine the refinements needed to reach our goal of a psychologically real description of orthographic structure. The dependent measure was the performance on each of 200 test items. Positionsensitive summed log bigram frequency provided the best statisticalredundancy description of performance on the individual items. Furthermore, an improved rule-based regularity measure also provided a very good description. However, the description given by the regularity measure correlated very highly with that given by the best frequency-based measure. For this reason, it was not possible in these experiments to
Dominic W. Massaro et al.
I68
decide whether either regularity or statistical redundancy is sufficient to account for the reader’s utilization of orthographic structure. Our goal in the present experiments is to refine our measures of structure in a further attempt to contrast a more powerful statisticalredundancy measure with an improved rule-governed regularity measure. Although statistical redundancy and regularity are highly correlated, a design involving orthogonal contrasts might be sufficient to distinguish between them. We follow this logic in the present studies by factorially contrasting bigram frequency and regularity measures in a target-search task. Words of high and low word frequency are also included as test items in order to assess the role of lexical status and word frequency. As with the previous experiments (Massaro et ul., 1980b), it again will be necessary to examine post hoc correlations to determine whether some other measure might provide a better description. By refining and repeatedly testing measures of structure, we hope to determine those properties that best reflect the reader’s knowledge of orthographic structure. A.
EXPERIMENTS 1 A N D 2: BIGRAM FREQUENCY vs REGULARITY
1 . Method
u. Sithjects. Nine subjects were used in the first experiment and eleven were used in the second. All were Introductory Psychology student volunteers who received credit toward their course grade for participating. Additionally, they were all native English speakers, right-handed, had normal or corrected to normal vision, and had not participated in any of the other experiments. b. Stimuli and Appuratus. A sample of high-frequency words was obtained from a list of all six-letter words from Kucera and Francis (1967), subject to the constraints that the words had a frequency greater than or equal to 50, were not proper nouns, and did not have repeated letters. A similar list of words with a frequency of exactly three was used to obtain low-frequency words. For each word in these two lists, all possible 720 anagrams were generated and each of their summed-positional bigram frequencies was calculated. The bigram frequencies were based on counts given by Massaro et a / . (1980b), which were derived from the Kucerd and Francis (1 967) word list. Forty high-frequency and 40 low-frequency words were selected along with four anagrams of each word. The anagrams were selected so that they formed a factorial arrangement of high and low summed-positional bigram frequency and of being orthographically regular and irregular. Orthographic regularity was manipulated in
I69
Letter and Word Perception
TABLE I THERULESFOR CHOOSING REGULARAND IRREGULARLETTERSTRINGS -
Letter strings were regarded as regular if they were phonologically legal and contained common vowel and consonant spellings. A letter string was regarded as orthographically irregular if it contained at least one of the following spellings: a . Phonologically illegal initial or final cluster (e.g., rlhued or eigopn) b. Orthographically illegal spelling for an initial final consonant or consonant cluster (e.g., xeoich or tmoreh) c. An illegal vowel spelling (e.g., caeinm) d . A phonologically illegal medial cluster (e.g., ilrmed)
the same manner as in previous experiments (Massaro et al., 1979, 1980b). The rules for choosing regular and irregular strings are given in Table I. Some examples of the words and their respective anagrams are presented in Fig. 2. Number and person have high word frequencies, and hurdle and pigeon have low frequencies. The letter string rumben is a regular-high anagram of the word number, and helrud is a regular-low anagram of hurdle. The number in each cell gives the average summedpositional bigram frequency for the items of that class. For example, the irregular-high anagrams of high-frequency words have an average count of 5738.
r/ number
Words
person (6060) hurdle pigeon (4468)
Summed Positional Bigram Frequency High
,
,
runemb roneps (1415) .______helrud ginope (1106)
Regular Orthographic Regularity
Low
~
bemrnu
I
___________._
brnemu
Irregular eiopng (4920)
eigopn (1113)
Fig. 2. Example of the test words and their corresponding anagrams from Experiments 1 and 2 . Within each of the five squares, the top two items correspond to high word frequency and the bottom two items correspond to low word frequency. The number in each of the ten cells represents the summed-positional linear bigram frequency.
Dominic W. Massaro et al.
I70
Twenty arbitrarily chosen high-frequency words and their anagrams as well as 20 low-frequency words and their anagrams were selected as stimuli for the first experiment. The remaining 20 high- and 20 lowfrequency anagrams were used with new subjects in the second experiment. The letter strings for the two experiments are presented in Massaro, Jastrzembski, and Lucas ( 1 980a). The visual displays were generated by a DEC LSI-11 computer under software control and presented on Tektronix Monitor 604 oscilloscopes (G. A . Taylor, Klitzke, & Massaro, 1978a, 1978b). These monitors employ a P3 I phosphor that decays to . 1 % of stimulated luminance within 32 msec of stimulus offset. The alphabet consisted of lower-case letters without serifs resembling the type font Univers 55. For an observer seated comfortably at an experimental station, the six-letter displays subtended about 1.9 degrees of visual angle horizontally, and the distance from the top of an ascender to the bottom of a descender was about .4 degree. Up to four subjects could be tested in parallel in separate rooms. c. Procedure. A trial (see Fig. 3) began with the presentation of a 250-msec fixation point. The fixation point was replaced by a test letter string, i.e., a word or an anagram, for a duration of 10-39 msec. The duration on a particular trial for each subject was determined by his or her accuracy. The duration was adjusted every 20 trials by a modified version of the PEST algorithm (M. M . Taylor & Creelman, 1967) in order to keep the subject’s average accuracy at about 75%. A masking stimulus followed the onset of the test string after a 70-msec interval. Therefore, the blank interval between the test stimulus and the masking stimulus was (70 - t ) msec, where t was the duration of the target string. The masking stimulus was composed of six nonsense letters. Each nonsense letter changed from trial to trial and was composed of a montage of randomly selected features of the test letters. The feature density of a nonsense letter was equal to that of the letter g . The size of the nonsense letters was equivalent to that of the test string. The duration of the mask was adjusted along with the duration of the test string. The mask remained on the screen for (40 - I ) msec, giving a range of durations of 1-30 msec. The
I +
fixation point
250 ~
,
letter
,
masking
s t [ r 1 g 5 1
-
~
~
target let?
+ 1 5 4000-1 8
L, 0
Time [msec)
Fig. 3 . A schematic representation of the perceptual recognition task used in Experiments 1-6.
Letter and Word Perception
171
mask was followed by another blank interval and then the target letter. The second blank interval lasted 180 msec minus the duration of the mask. Therefore, the interval between the onset of the test letter string and the target letter was always 250 msec. The target letter remained on the screen until all subjects responded or for a maximum of 4 sec. Finally, the interval between trials was 500 msec. Subjects were instructed to indicate whether the target letter was present in the test string and to be as accurate as possible. The experiment consisted of a session of 100 practice trials with a practice list that was comparable to the experimental list, and two sessions of 400 experimental trials each. Within each session, each item was tested once as a target string and once as a catch string. On target trials, the target letter was selected randomly with replacement from the six letters in the test string. For catch trials, a target was selected randomly from the set of 26 letters weighted by their probability of occurrence in the stimulus set. If the selected letter was present in the test string, additional drawings with replacement were made until an appropriate target letter was selected. Some letters did not occur in any of the test strings and therefore were never tested. A short rest break intervened between the two experimental sessions. The total time for the three sessions and the rest break was about 75 min. Both experiments were conducted in exactly the same manner except that different subjects and different items were used in each.
2.
Results
a . Analyses of Vuriance. Two analyses of variance were performed on the percentage accuracy scores. In the first analysis, word frequency, type of test letter string, target or catch trial, and subjects were factors. In the second analysis, the word data were eliminated, and regularity and bigram frequency were factors in the design. Figure 4 shows the average percentage correct on target and catch trials as a function of letter-string type in Experiment 1 . There were large differences among the various types of letter strings, F(4,32) = 130.7, p < .001. Regular items resulted in a 9.3% accuracy advantage over irregular items F(1,8) = 74.7, p < ,001; and items of high summed-positional bigram frequency had a 2.5% advantage over items of low summed-positional bigram frequency, F ( 1,s) = 11.4, p < .01. The advantage of high bigram frequency was limited to regular items, F(1,8) = 10.0, p < .05. The difference in accuracy between words and the regular-high anagrams was 12.0%, F ( 1 , 32) = 23.3, p < ,001. There was no significant difference in accuracy between target (72.4%) and catch (77.2%) trials, F < 1, and this variable did not interact with letter-string type, F < I .
Dominic W. Massaru et al.
I72
I
1
ORD
I
I
I
I
I
I
R-H
R-L
I-H
I
I
1-1
Fig 4 Percentage correct as a function of display type for larger and catch trials in Experiiiient I .
Figure 5 gives the average percentage correct for the high and low word frequency words and their anagrams as a function of letter-string type. There was an overall 2.7% advantage for the low-frequency words and their anagrams, F ( 1 , 8 ) = 9.86, p < .015, and word frequency also interacted with letter-string type, F(4,32) = 6.18, p < .001. The overall effect of letter-string type was 28.3% for the high-frequency words and their anagrams and 19.5% for the low-frequency words and their anagrams. This difference reflected the fact that high-frequency words were more accurate than low-frequency words, but that the reverse was the case for the four types of anagrams. Word frequency did not interact with target vs catch trials, nor was there a three-way interaction with these variables and letter-string type ( F s < I). Figure 6 gives performance for target and catch trials as a function of letter-string type in the second experiment using new items and new subjects. There were large differences among letter-string types, F(4,40) = 92.76, p < -001. Regular items resulted in 8.7% greater accuracy than did irregular items, F(l ,lo) = 49.1, p < .001. Items of high summedpositional bigram frequency resulted in 3.3% greater accuracy than did items of low summed-positional bigram frequency, F(l ,lo) = 12.2, p < .05, but the advantage occurred only for regular items, F(1,lO) = 8.2, p < .025. The difference between words and regular-high anagrams was 13.1%, F ( I ,40)= 18.0, p < .001). There was no significant difference in accuracy between target (74.1%)and catch (78.4%)trials, F < 1, and this variable did not interact with letter-string type, F < 1 .
Letter and Word Perception
I73
-
60 -
50
-
WORD
A-H
R-L
I-H
I-L
Fig. 5 . Percentage correct as a function of display type for items corresponding to high and low word frequency i n Experiment ! .
Figure 7 gives average percentage correct for the high and low word frequency words and their anagrams as a function of letter-string type. There was a 2.5% advantage for items of the high-frequency words, F(1,lO) = 4.87, p < .052, but word frequency did not enter into any interactions. The overall effect of letter-string type was 24.3% for highfrequency items and 25.8% for low-frequency items. The interaction of I
I
I
I
I
----- ........... *
I
OAD
I
I
I
I
A-H
R-L
I-H
I-L
Fig. 6. Percentage correct as a function of display type for target and catch trials in Experiment 2.
Dominic W. Massero et al.
I74 100
I
501
I
WORD
I
R-H
I
I
R-L
1
I
I-H
I
I-L
Fig. 7. Percentage correct as a funcrion of display typc for items corresponding to high and low word frequency in Experiment 2.
word frequency and letter-string type found in the first experiment and shown in Fig. 4 was not replicated in the second experiment, F(4,40) = 1.15, p < .25. h. Correlations and Regressions. The factorial design is limited in terms of providing a quantitative assessment of the importance of frequency and regularity measures of orthographic structure. The present design contrasted just one frequency measure against just one regularity measure. Therefore, post hoc correlational analyses were carried out to provide an analysis of a range of descriptions of orthographic structure. The independent variables used in this analysis included a number of measures based on frequency counts for letters, n-grams, and words, in addition to a few quantitative measures based on orthographic rules. The dependent measure in all cases was average accuracy for each six-letter test item. The accuracy scores were obtained by averaging across subjects and across target and catch trials. Each of the two experiments used 40 words, 20 each of high and low word frequency, and four corresponding anagrams for a total of 200 stimulus items per experiment. Each subject had been presented with each item twice as a target trial and twice as a catch trial. Accordingly, the accuracy score for each item in the first experiment was based on 36 observations (4 replications X 9 subjects), and the accuracy score for the second experiment was based on 44 observations (4 replications x 11 subjects). c . Frequency Measures. The source of the frequency measures is
Letter and Word Perception
175
based o n a word corpus compiled by Kucera and Francis (1967). This corpus consisted of 500 samples of approximately 2000 words each selected from 15 categories. Massaro et al. (1980b) used these words to derive the frequencies of occurrence of single letters, bigrams, and trigrams. Tables were prepared by counting the occurrence of each n-gram at the position it occurred in words of a given length. The counts were token counts based upon the total number of occurrences of the words containing the n-gram. A position-insensitive count (but still word length dependent) was also obtained for each n-gram by summing across the position-dependent counts. The single-letter tables and bigram tables for word lengths 3 through 7 are presented in Massaro et al. (1980b). Type counts are based on the number of word types that contain a given n-gram; these counts may also be relevant descriptors of frequency-based measures of orthographic structure (Solso & King, 1976). However, Massaro et al. (1980b) found that the correlations between comparable type measures and token measures were very high. Measures based on single letters, bigrams, and trigrams, both position sensitive and position insensitive, correlated between .84 and .99. With such high correlations no meaningful discrimination between type and token measures can be made unless test items are selected with this contrast in mind. For this reason we shall discuss only measures based on the token counts derived by Massaro et a f . (1980b). The present analysis will be restricted to position-sensitive counts. Massaro et af. (1980b) found that position-sensitive counts give consistently better descriptions of performance than do position-insensitive counts. For single-letter frequency, for example, the correlation with average accuracy was only .2 for position-insensitive counts but .62 for position-sensitive counts. The advantage of position sensitivity was attenuated, however, as the length of the n-gram increased. While the effects of frequency seem to be psychologically real, it is not necessary that the mental representations of frequency directly reflect the frequency of objective counts. One alternative scale that has been successful in other research is a logarithmic (base 10) scale (Solomon & Postman, 1952; Taylor, 1977; Travers & Olivier, 1978). Furthermore, a logarithmic representation is consistent with recent studies of number representation (Shepard & Podgorny, 1978) and with many other psychological scales. Therefore, we computed all our frequency measures on the basis of both linear frequencies and log frequencies. Since counts were sometimes zero, the log of zero was defined as zero. Therefore, the two sets of measures being correlated were sums of position-dependent single letters, bigrams, and trigrams derived from either linear-frequency or log-frequency tables. d . Regularity Measures. To provide a quantitative measure of the
Dominic W. Massaro et al.
I76
regularity of each of the 400 stimulus items, a simple count of the number of orthographic irregularities for each item was computed, based on the rules developed by Massaro et ul. (1 98Ob). The rules are given in Table 11. This measure of regularity provided a reasonable description of performance in the Massaro et a/. (1980b) studies. We shall refer to this measure as Regularity(1). One critical feature of the rules for Regularity(1) is that letter strings are treated as monosyllabic and many legal and occurring medial consonant clusters are treated as irregular. For example, the word person would be considered to have an irregularity, since according to rule 2 the medial consonant cluster rs would not be legal in initial position. However, the consonant cluster rs is regular in medial position when considered as part of a two-syllable word. Therefore, a second quantitative measure of regularity was derived that removed the constraint that the letter string must be considered as a monosyllabic string. This measure is referred to as Regularity(2). The rules for Regularity(2) were identical to those for the first measure except that the application of the rules and the counting of the violations were carried out in order to minimize the number of violations for any given letter string. When possible, a syllable boundary was assumed in order to avoid a given violation. As an example, the medial consonant cluster tnd in the string lirnder would be an illegal consonant cluster in the TABLE I1 THERULESFOR A N IRREGUL.ARITY COUNT" I . Segment string into vowel and consonant substrings. Treat final -1e as if it were 4.Treat h between vowels as a (legal) consonant. 2 . For each consonant string, determine minimal number of vowels that must be inserted to make the string pronounceable. Initial consonant clusters must be legal in initial position. Final consonant clusters must be legal in final position, including those followed by final e. Medial consonant clusters must be legal in initial position. 3 . Rate each resulting consonant substring for position-sensitive scribal regularity (count one for each irregular substring). 4 . For each vowel substring, determine minimal number of consonants that must be inserted to create scribally regular sequences. Mark as irregular illegal initial and final vowel substrings. 5 . Count number of inserted vowels and consonants plus number of scribally irregular consonant and vowel substrings. This yields an irregularity index. 6 . The vowel string (10, uc, oe, and ye (among more obvious cases) would be illegal vowel strings. y would be illegal as a vowel in initial position, and i , u , u. oa, and o would be illegal in final position. ue is legal as is y as a single, noninitial vowel. I . h is not allowed in final position unless preceded by c , g. or s. 8. y and w between vowels are to be counted as consonants. ~
After Massaro, Taylor, Venezky, Jastnembski, and Lucas (1980b).
Letter and Word Perception
177
same syllable because of the phonological rule governing the place of articulation of nasals followed by stops in a single syllable. The nasal and the following stop must share place of articulation; therefore mb and nd are possible, but not md or nb. A syllable boundary between m and d in limder is possible, however, resulting in a perfectly legal two-syllable string with no violations. Similarly, in the string nurdgi the medial consonant cluster rdg is legal with a syllable boundary between d and g . The only violation is i in final position. e . Frequency vs Regularity. The correlations of several measures with average accuracy are presented in Table 111. The correlation needed for statistical significance at p = .01 with 198 degrees of freedom is .18. Of central interest is the relative ability of bigram frequency and regularity measures of orthographic structure to predict performance. Two dummy variables were created to contrast these two measures while equating for the range and levels of each measure. The dummy regularity variable assigned a 1 to words and regular nonwords, and a 0 to irregular nonwords. The dummy frequency variable assigned a 1 to words and high bigram frequency nonwords, and a 0 to low bigram frequency nonwords.
TABLE 111 CORRELATIONS OF SEVERAL PREDICTOR VARIABLES WITH OVERALL ACCURACY PERFORMANCE I N EXPERIMENTS 1 AND 2
Dummy regularity Dummy frequency Single letter Linear Log Bigram Linear Log Trigram Linear
Experiment 1
Experiment 2
.49 .26
.31
.51
.32 .36
-28
.35 .54
.29
.47 .60
.46
.45
.51 .59 .68 .51 .55
.49
.63
.64
Word frequency
Linear Log Dummy Regularity count(1) Regularity count(2)
.55 .60
.52 .54
178
Dominic W. Massaro et al.
In both experiments the regularity variables correlated much higher (.49, .51) with performance than did the frequency variable (.26, .37). It is not possible to choose between regularity and frequency measures of orthographic structure. Although the regularity counts give higher correlations than do linear frequency counts and log single letter counts, the log bigram and log trigrain counts give the highest correlations. Both frequency and regularity measures account for a significant portion of the variance in performance. Regularity and frequency measures are positively correlated with each other. As an example, log trigram frequency and Regularity(2) correlate .47 and .46 for the items in Experiments 1 and 2, respectively. A multiple regression was carried out treating the summed frequency counts and the irregularity counts as independent variables. The best combination of predictors was log trigram frequency and Regularity(2), which accounted for 45% of the variance in Experiment 1 and 49% in Experiment 2 . J’. Word Frequency. Linear word frequency correlated .45 and .51 and log word frequency correlated.55 and.59 with performance in the two experiments. The correlations with performance on just the 40 word items were .46 and .51 for linear and log frequencies in the first experiment and .34 and .35 in the second experiment. Although it is possible that word frequency makes an independent contribution to performance, the high correlations between word frequency and sublexical orthographic structure measures preclude resolution of this issue. Log word frequency was highly correlated with both log bigram frequency (.50, .52) and log trigram frequency (.77, .74). Lexical status rather than word frequency may be the critical variable producing the large advantage for word items. A dummy word frequency variable which assigned a 1 to words and a 0 to nonwords was more highly correlated (.60, .68) with performance than was log word frequency. B.
EXPERIMENT 3: Lo<; BICRAMFREQUENCY vs REGULARITY
The creation of the stimulus set was identical to that of the previous two experiments except that log bigram rather than linear bigram counts were used and the strings were controlled more exactly for regularity. Figure 8 gives examples of the five classes of items and the average log bigram frequency for each class. The complete list of letter strings is presented in Massaro et al. (198Oa). In the studies of Massaro et al. (1980b) and Experiments 1 and 2, log-frequency measures were consistently more highly correlated with performance than were linear-frequency measures. Furthermore, Massaro
Letter and Word Perception
period should
179
Summed Positional Frequency
magnet
High
(I3420)
Regular Orthographic Regularity Irregular
I
+ I
rodipe shulod
dripoe lohuds
____________ (:i,",",",' ______ ______ ___________________ (8523) ____
11_____
tarngen
nidcoe nerntag
(11 143)
(7842)
prdioe dhouls
11
dpireo louhds
!!!!S2?!________ -LEE!__________ cnoied ntagem
(11083)
endcoi nagtme
(7883)
Fig. 8. Examples of the test words and their corresponding anagrams from Experiments 3 and 4. Within each of the five squares, the top two items correspond to high word frequency and the bottom two items correspond to low word frequency. The number in each of the ten cells represents the summed positional log bigram frequency.
et al. (1980b) found that the log counts were superior to a range of power-function transformations of the linear counts. This result provides additional evidence that, if frequency of occurrence is important, log frequency appears to be the best descriptor of this variable. A count of the number of irregularities in each letter string was determined by using the rules for Regularity(3) presented in Table IV. The TABLE IV THERULESFOR REGULARITY (3) FOR THE SELECTION OF THE ITEMUSEDIN EXPERIMENTS 3-6 1 . For each string, rate for position-sensitive scribal regularity and pronounceability (count one for each violation). Treat final -le as if it were -el. Treat h between vowels as a (legal) consonant. 2 . For each string, rate for position-sensitive scribal regularity and pronounceability (count one for each violation). Initial consonant clusters must be legal in initial position. Final consonant clusters must be legal in final position, including those followed by final e. 3. For each string, rate for position-sensitive scribal regularity and pronounceability (count one for each violation). The vowel strings ao, ae, oe, and ye (among more obvious cases) would be illegal vowel strings. All vowels are illegal in initial position and y would be illegal as a vowel in initial position. The vowels i, u , a, oa, and o would be illegal in final position. ue is legal as is y as a single noninitial vowel. h is not allowed in final position unless preceded by c, g. or s. y and w between vowels are to be counted as consonants.
1 no
Dominic W. Massaro et al.
rules for Regularity(3) were the same as those for Regularity(2), except that vowels as initial letters violated one of the rules and therefore were counted as irregularities. Given this formula, it was now possible to equate the number of irregularities for the anagrams that differed only in log bigram frequency. In our previous studies, the number of irregularities tended to correlate negatively with frequency, and some of the effect of frequency could have been due to differences in regularity. This possibility was eliminated in the present study by equating the highand low-frequency anagrams of a given test word for the number of irregularities. Consider the test word period shown in Fig. 8 . The regular-high and regular-low anagrams (rudipe and dripae) do not have any irregularities. The irregular-high and irregular-low anagrams (prdioe and dpireo) have two irregularities each. This design might provide a more definitive contrast between frequency and regularity. 1.
Method
u . Subjects. Nine University of Wisconsin summer school student volunteers were used as subjects and paid $9.00 for their participation. All were native English speakers, had normal or corrected vision, and had not participated in any of the other experiments. h. Srirnuli and Appurufus. Words were selected in the same manner as in the previous two experiments. The high-frequency words had a Kucera and Francis (1967) frequency of at least 50, and the low-frequency words had a frequency of 3. Because of a selection error, one low-frequency word had a frequency of 4.For each of the 80 words, four anagrams were selected so that they formed a factorial arrangement of high and low summedpositional log bigram frequency and of being orthographically regular or irregular. For each set of four anagrams, the number of irregularities were matched exactly for the regular conditions and then again for irregular conditions. Finally, an additional sample of words, 13 high and 13 low in word frequency, and their anagrams were selected as practice items. The 80 experimental words and their anagrams were divided into two lists. List 1 contained one-half of the high word frequency items and one-half of the low word frequency items. List 2 contained the remaining items. Stimuli were presented in the same manner and on the same equipment as in the previous experiments with only one exception. The range of durations for the test letter strings was 5-39 msec. Because of the algorithm used, decreasing the lower limit for the duration of the test string increased the maximum duration of the mask to 35 msec.
Letter and Word Perception
181
c . Procedure. The experiment was conducted in a manner similar to that of the previous experiments. The presentation of the test string, masks, and target letters was identical to that of Experiments 1 and 2. Subjects were tested on two consecutive days. At the beginning of each day, subjects began with a practice session of 260 trials. Two experimental sessions of 400 trials each followed the practice. Five of the subjects received all 200 items of List 1 on Day 1 in the first session as both target and catch trials. These subjects then received the List 2 items in the second session. On Day 2, List 2 was presented in the first session and List 1 in the second session. For the remaining four subjects, the order of the lists was reversed. 2.
Results
a . Analyses of Variance. Figure 9 shows the average percentage correct on target and catch trials as a function of letter-string type. There were significant differences, F(4,32) = 127.1, p < .001, among the five types of letter strings. Words had a 16% advantage over the regular-high anagrams, F(1,32) = 55.1, p < .001. There was 4.0% advantage of regular strings over irregular strings, F ( 1,8) 32.5, p < .001, and a 1.4% advantage of high log bigram frequency strings over low log bigram frequency strings, F(1,8) = 2.6, p > .2.
50
WORD
R.H
f7.L
1.H
1.L
Fig. 9. Percentage correct as a function of display type for target and catch trials in Experiment 3
I82
Dominic W. Massaro et al.
One disquieting aspect of the results is the extreme asymmetry in performance on target and catch trials and the interaction of this variable with display type, F(1,8) = 12.8, p < .007, and F(4,32) = 9.5, p < .001. Subjects were extremely conservative in their willingness to indicate that a target letter was present. This result could reflect our failure to instruct the subjects specifically about the relative frequency of target trials as we did in the previous two experiments. Figure 10 gives average percentage correct for the high and low word frequency words and their anagrams as a function of letter-string type. High-frequency words and their anagrams were recognized 3.2% more accurately than were low-frequency words and their anagrams, F ( I ,8) = 69.03, p < ,001, The interaction between word frequency and the five types of items was not significant, showing that this difference was not unique to the word items. Therefore, some variable other than word frequency must be responsible for the difference. However, one caveat is to realize that performance may not be on an interval scale, which weakens any interpretation of the lack of interaction. One solution would be to monitor each display type independently and to adjust the stimulus values to give an average of 75% correct for each display type. If word frequency still does not interact with display type when average performance is about 75% correct at each display type, then the conclusion reached here would be reinforced. b. Correlations ~ i r i i lRegressioiis. The correlations of several variables with overall performance are presented in Table V . For all freI
I
1
I
I
\
I
DRD
I
I
I
I
R-H
R-L
I-H
I-L
Fig. 10. Percentage correct as a function of display type for items corresponding tn high and low word frequency in Experiment 3.
183
Letter and Word Perception
TABLE V
CORRELATIONS OF SEVERAL PREDICTOR VARIABLES WITH OVERALL ACCURACY I N EXPERIMENTS 3 AND 4
Single letter Linear Log Bigram Linear Log Trigram Linear Log Word frequency Linear Log Regularity count(3)
Exp. 3
Exp. 4
Ave.
.29 .36
.26 .28
.31 .36
.38
.33 .43
.40
.59
.36 .48
.45 .60
.44 .60 .40
.37 .50 .29
.46 .63 .39
.53 .44
.54
quency measures, the log measures correlated more highly with performance than did the linear measures. Log trigram frequency predicted performance better than did the other sublexical measures. Log word frequency was correlated with accuracy (.60), but also was correlated with log bigram frequency (.61) and log trigram frequency (.go). Among just the high-frequency words the correlation with performance was - .05 and - .13, respectively, for linear and log word frequencies. (Correlations among the low-frequency words would not be meaningful, since all the items had the same Kucera and Francis frequency of occurrence.) The lack of a significant correlation between performance and word frequency within the class of words replicates previous results (Manelis, 1974) and makes it unlikely that word frequency can account for the effects of orthographic structure. Lexical status alone might be an important variable, however. The dummy variable of word or nonword gave a highly significant correlation of .60 with performance. Summed log trigram frequency and Regularity(3) accounted for 39% of the variance. C.
EXPERIMENT 4: REPLICATION
I.
Method,
Eight new University of Wisconsin undergraduates from Introductory Psychology who met the same requirements as in the previous experi-
Dominic W. Massaro et al.
I84
ments were used as subjects. Experiment 4 was an exact replication of Experiment 3 with one exception. The instructions were modified to inform subjects that a target letter would appear in the test string on 50% of the trials. It was expected that this manipulation would attenuate the asymmetry in the number of positive and negative responses.
2. Results a . Analysis o j Variance. Figure 1 1 shows the average percentage correct for target and catch trials for the five letter-string types. The significant differences among the letter-string types, F(4,28) = 102.7, p < .001, completely replicate Experiment 3. There was a 15.0% advantage of words over regular-high anagrams, F ( 1,28) = 52.4, p < .001; a 2.4% advantage for regular strings, F(1,7) = 10.2, p < .025; and only a 0.7% advantage for-high-frequency strings, F(1,7) = .84. Although the responding asymmetry was substantially reduced, there was still a tendency for subjects to remain conservative in their willingness to indicate that a target letter was present, F(1,7) = 6.1 1 , p < .05. The range of performance across letter-string types was 23.0% for target trials and 13.3% for catch trials. Figure 12 presents the percentage correct for the letter-string types as a function of word frequency. There was an overall effect of 20.5% for high word frequency items and 15.8% for low word frequency items, F(4,28) = 2.19, p < . l o .
90
I ,I
60
50
MNlD
R-H
R-L
I-H
I-L
Fig. 1 1 , Percentage correct as a function of display type for target and catch trials in Experiment 4.
Letter and Word Perception loo
@!f
185
I
I
I
I
I
I
WORD
R-H
R-L
I-H
1.L
I
I
I
60 -
Fig. 12. Percentage correct as a function of display type for items corresponding to high and low word frequency in Experiment 4.
h. Correlations and Regressions. The correlations of several measures with overall performance in Experiment 4 are presented in Table V. As with the previous analyses, log measures predicted performance better than did linear measures. Trigram frequency was the best of the three frequency measures, but only slightly better than bigram frequency. Log word frequency was correlated S O with overall performance. Summed log trigram frequency and Regularity(3) accounted for 25% of the variance. Table V also presents the correlations of the same variables with the average performance across Experiments 3 and 4. Since Experiment 4 was a replication of Experiment 3, the performance on each item was averaged across the two experiments. As might be expected’from increased reliability, these correlations are significantly larger than those for either of the experiments considered separately. D.
EXPERIMENT 5 : LEXICAL STATUS
In Experiments 1-4, a large advantage was found for words as compared with the best regular-high anagrams. One way to account for this effect is by the lexical status of the words. Since words are represented in the reader’s lexicon, they may be retrieved on the basis of partial visual information. For example, the partial information sho-l- given presentation of should might lead to recognition of the word should. Lexical access would allow determination of the two unknown letters. On the
186
Dominic W. Massaro et al.
other hand, the partial information shu-a- given presentation of shulod cannot access any lexical entry, and the missing letters cannot be determined. Consequently, on a word trial there is a better chance that all the component letters will be available for comparison against the target letter. In contrast, the same partial information about an anagram will not lead to recognition of all the letters in the test string. As a result, fewer letters of anagrams will be available for comparison against the target letter. This account is consistent with the model articulated in Section I; the secondary recognition process can lead to word recognition without complete recognition of all the component letters. A second explanation of the word advantage is that words differ from even the best anagrams with respect to sublexical orthographic structure. For example, the bigram frequency of the words in Experiments 3 and 4 averaged almost three log units more than that for the regular-high anagrams (see Fig. 8). Perhaps accuracy was greater for words because words contained more frequent bigrams. To choose between these two explanations in Experiment 5 , the words of Experiments 3 and 4 were replaced with regular anagrams, which were matched with the words on log bigram frequency. If log bigram frequency was the basis of the word advantage, then a similar advantage should be observed for these regular-very high (R-VH) anagrams.
I.
Method
Experiment 5 was conducted in the same manner as the previous experiments. Regular-very high anagrams with log bigram frequencies similar to the words of Experiments 3 and 4 were used along with all the anagrams of Experiments 3 and 4. The summed bigram frequencies for the regular-very high anagrams were 14.940 and 13.407, respectively, almost identical to those of the words (see Fig. 8). The regular-very high anagrams are listed in Massaro et ul. (1 980a). The regular-very high anagrams of period and coined were poried and conied, respectively. The experiment was identical to Experiment 4 except that the regular-very high anagrams were used in place of the word items. Seven new subjects were tested. 2,
Result.7
Figure 13 shows the differences among the five types of letter strings, F(4,24) = 3.3, p < .05. There was a 0.3% advantage of regular veryhigh anagrams over regular-high anagrams, F( 1,6) < I . Regular anagrams gave a 3.6% advantage over irregular anagrams, F ( l ,24) = 13.8, p < .005. There was a -0.2% effect for log bigram frequency, F ( 1,6) < 1. There was only a 3.0% advantage of catch over target trials, F ( I ,6) <
Letter and Word Perception
<,-
c
-----a
187
Catch Target
90
6 #
E80-
0 W
B
B
#!
70-
60 -
501
I
R-VH
1
R-H
I
R-L
I
I-H
I I
I-L
Fig. 13. Percentage correct as a function of display type for target and catch trials in Experiment 5.
1, but this variable interacted with the type of letter string, F(4,24) = 3.3, p < .05. This test reflects the presence of a 7.4% increase in accuracy from the worst to the best structured strings for target trials, but an absence of an effect of orthographic structure for catch trials (Fig. 13). Figure 14 presents the average accuracy as a function of letter-string
Low Frw. 90
p"
60 -
sol
I
R-vn
I
R-H
I
R-L
I
I-n
I-L
Fig. 14. Percentage correct as a function of display type for items corresponding to high and low word frequency in Experiment 5.
Dominic W. Massaro et al.
188
type according to whether the items correspond to high or low word frequency. Neither the 0.6% difference nor the interaction with display type was statistically significant. The results of Experiment 5 support the idea that lexical status makes a significant contribution to perceptual recognition in the target search task. The contribution of lexical status in the earlier experiments cannot be attributed to sublexical orthographic structure differences in log bigram frequency. That is, the reader takes into account not only the frequency of occurrence of letter sequences and the regularity of these sequences, but also whether or not a particular sequence is represented i n a word. Frequency and regularity allow well-structured anagrams to be better recognized than poorly structured anagrams; in addition, lexical status allows a perceptual advantage of words over equally well-structured anagrams. E.
EXPERIMENT 6: FREQUENCY, REGULARITY, AND LEXICAL STATUS
Experiments 1-5 were successful i n demonstrating frequency, regularity, and lexical status as psychological measures of orthographic structure. In a final evaluation of the relative contribution of these measures, the perceptual recognition task was replicated with five display types. The display types were chosen to give a large range of regularity and bigram frequency. The comparisons among display types and the post hoc corre-
50
A WOAD R-VH
R-L
I-H
VI-VL
Fig. 15. Percentage correct as afunction of display type for target and catch trials in Experiment 6.
Letter and Word Perception
I89
lations will be used to measure the relative contributions of lexical status, regularity, and frequency in perceptual recognition. I.
Method
The R-VH anagrams from Experiment 5 and the words, regular-low anagrams, and irregular-high anagrams from Experiment 3 were used as items along with a new type of anagram that was both very irregular and very low in log bigram frequency. The very irregular, very low (VI-VL) anagrams mostly had three or four irregularities and average log bigram frequencies of 4.845 and 4.971 for the high and low word frequency items, respectively. The very irregular, very low anagrams are listed in Massaro et a / . (1980a). The anagrams pdreio and dcoeni were derived from the words period and coined, respectively. The procedure of Experiment 4 was replicated exactly. Eleven new subjects from the Introductory Psychology subject pool were used. 2 . Results As can be seen in Fig. 15, accuracy uniformly increased with better structured letter strings; F(4,40) = 69.4, p < .001. Words had a 10.8% advantage over the regular-very high anagrams, F(1,40) = 15.9, p < ,001. Regular-very high anagrams gave a performance advantage of 4.9% over regular-low anagrams, F(1,40) = 3.1, p < .086; a 7.6% advantage over irregular-high anagrams, F(1,40) = 7.9, p < .01; and a 8.9% advantage over very irregular, very low anagrams, F(1,40) = 10.7, p < .005. The 6.9% advantage of catch over target trials was not significant, F ( 1 ,lo) = 1 . 5 , p > .25, and this difference did not interact with display type, F < 1. Figure 16 reveals a 2.7% difference between levels of word frequency, F(1,lO) = 32.17, p < .001, but the advantage of high word frequency occurred only for the words and the regular anagrams. b. Correlafions and Regressions. The correlations of several predictor variables with overall performance are presented in Table VI. As usual, the log measures predict performance better than the linear measures do. Bigrams and trigrams were similar in predictive ability. Log word frequency correlated .42 with overall performance, but also correlated .48 with summed log bigram frequency and .75 with summed log trigram frequency. Summed log trigram frequency and Regularity(3) accounted for 37% of the variance. a. Anulyses of Variance.
Dominic W. Massaro et al.
I90 100,
I
I
I
L--
- 05
WOAD
A-VH
R-L
I
High Freq.
I-H
VI-VL
Fig. 16. Percentage cnrrect as il function of display type for iLerris corresponding tn high and low word trequency i n Experiment 6 .
111.
Orthographic Structure and Conscious Knowledge
The perceptual recognition task allows us to assess the degree to which readers utilize orthographic structure in visual processing of letter strings. An overt judgment task has been used to assess the degree to which this knowledge is available for a conscious report (Massaro et a l . , 1980b; Rosinski & Wheeler, 1972). Massaro et al. (1980b) asked whether subjects could discriminate among the items on the basis of rule-governed regularity or on the basis of statistical redundancy. Subjects were presented pairs of letter strings and asked to choose the member of each pair that most resembled written English. The instructions emphasized either a regularity or a statistical-redundancy criterion. Subjects' judgments seemed to reflect some knowledge of both rule-governed regularity and statistical redundancy. The present experiment uses this task with our new items derived from improved frequency and regularity measures. These items should provide a more definite contrast between frequency and regularity descriptions of orthographic structure. 7: OVERTJUDGMENTS EXPERIMENT An overt judgment task is used in the present experiment to assess the degree to which regularity and frequency are coiisciously available. Sub-
191
Letter and Word Perception
TABLE VI CORRELATIONS OF SEVERAL
PREDICTOR VARIABLES WITH OVERALL ACCURACY I N EXPERIMENT 6 Single letter Linear
m
Bigram Linear Log Trigram Linear Log Word frequency Linear
k z Regularity count(3)
.35 .37 .40 .50 .41 .59 .38 .54 .44
jects are given pairs of letter strings and asked to choose which letter string most resembles English spelling. Some subjects are instructed to base their decision on the frequency of occurrence of letter sequences in English spelling; other subjects are instructed to respond on the basis of the regularity of letter sequencing. The seven types of letter strings varying in lexical status, regularity, and log bigram frequency were paired with each other in the task. The degree to which subjects can follow instructions and discriminate among the types of items should reveal which aspect(s) of orthographic structure is (are) consciously available and capable of report. 1 . Method
a . Subjects. Sixteen Introductory Psychology students who met the same requirements as in the first six experiments were used as subjects. b. Stimuli and Apparatus. The 280 letter strings represented all seven categories of items used in the previous experiments. Accordingly, 40 words and their corresponding R-H, R-L, I-H, and I-L anagrams of Experiment 3 , 40 R-VH anagrams of Experiment 5 , and 40 VI-VL anagrams of Experiment 6 were selected. The irregular items chosen from Experiment 3 had two irregularities. Seven categories, and allowing the two letter strings of a pair to be from the same category, result in 28 unique pairs of categories. These 28 pairs were sampled randomly without replacement in each block of 28 trials. The actual items from each of
192
Dominic W. Massaro et al.
the categories for each pair were randomly selected with replacement for each group of subjects. Eventually, each subject was presented with 840 pairs for judgment, resulting in a total of 30 observations for each pair. The two strings of each pair were arranged side by side on the CRT. The horizontal visual angle of each letter string was 1.9 degrees with a 2.7degree separation between strings. Subjects selected the one of the two strings that more closely resembled an English word. Subjects indicated their choice by pressing one of two keys located beneath the string. Each trial began with a 250-msec fixation point followed by the two letter strings. The strings remained on the CRT until all the subjects responded or for a maximum of 4 sec. The 840 pairs were presented in two sessions of 420 trials each. Each session lasted about 20 min. Of the 16 subjects, 8 were given the regularity instructions and 8 were given the frequency instructions. The regularity instructions described regularity of letter sequencing, and subjects were asked to choose the more regular of the two letter strings. The frequency instructions described the frequency of letter groups, and subjects were asked to choose the more frequent of the two letter strings. The exact instructions were given in Massaro er al. (1980a).
2. Results For each subject, the proportion of times that each of the seven Categories was chosen as most like English over the other six categories was computed. These proportions were entered into an analysis of variance with instructions, category type, and subjects as factors. Figure 17 presents the percentage of choices of most like English as a function of category and instructions. There was a large decrease in choices with decreases in orthographic structure, F(6,84) = 351, p < .001. However, instructions had no influence on performance and did not interact with structure, F s < 1 . All differences between adjacent categories in Fig. 17 are statistically significant, except for the small difference between R-H and R-L items. Table VII presents the proportion of times each of the seven classes of items was chosen over the other six classes. Some effects of instructions not apparent in the average proportions shown in Fig. 17 are seen in these results. The R-H items were picked over the R-VH items 39% of the time for regularity instructions and 25% of the time for frequency instructions. The three classes of irregular items (I-H, I-L, and VI-VL) were chosen an average of 13% of the time over the R-L items with regularity instructions and an average of 18% of the time with frequency instructions. Regularvery high items were chosen over words 17% of the time for regularity
193
Letter and Word Perception
0
i WORD R V H R.H
1.H
R.L
1.L VI.VL
Fig. 17. Average percentage choices of each of the display types in the overt judgment task in Experiment 7.
TABLE VII
THEPROPORTION
OF TIMES THE ROW ITEM W A S CHOSEN OVER THE COLUMN ITEM
FOR
REGULARITY A N D FREQUENCY INSTRUCTIONS Regularity instructions
Word R-VH R-H R-L I-H I-L VI-VL
Word .55 .I7 .I2 .15 .02 .01
.w
R-VH
R-H
R-L
I-H
I-L
VI-VL
.52 .39 .32 .I1 .09 .02
.50 .50 .24 .I5 .09
.5 I .I7 .I1 .06
.48 .36 .28
.58 .30
.53
Frequency instructions
Word R-VH R-H R-L I-H I-L VI-VL
Word .55 .25 .I8 .II .07 .02 .03
R-VH
R-H
R-L
I-H
I-L
VI-VL
.50 .25 .32 .I4 .09 .05
.51 .44 .27 .I7 .09
.55 .25 .20 .09
.50 .38 .24
.52 .35
.49
Dominic W. Massaro et al.
194
instructions and 25% of the time for frequency instructions. These results are consistent with the idea that a dimension of orthographic structure carries somewhat more weight when that dimension is stressed in the instructions in the overt judgment task. An analysis of variance also was conducted on the reaction times of the choice responses. Response type (selected vs. unselected) was included as a factor to assess the differences in reaction times between choosing a given category as most like English relative to the average reaction time for choosing the other six categories. Figure 18 presents the reaction times as a function of instructions, response type, and category. Overall reaction times were 232 msec longer for frequency than for regularity instructions, F(1,14)= 6.9, p < .025. Reaction times increased with decreasing orthographic structure, but only for the selected response type, F ( 6 , 8 4 ) = 12.7 and 21.0, p s < .001. A comparison between the perception task and overt judgment task shows that the latter is a much more sensitive measure of the reader’s knowledge of orthographic structure. Subjects are able to discriminate among certain classes of items in the overt judgment task that are responded to equivalently in the perceptual accuracy task. For example, R-VH and R-H were differentiated in the overt judgment task but were responded to equivalently in the perceptual accuracy task of Experiment 5 . The same was true of the I-Hvs. I-L and the I-H vs. VI-VL contrasts. Other results were exactly parallel in the two tasks: Words have an advanI
l8
t
I
1
I
I
Regularity
I
1
I
I
I
1
I
I
1
I
I
Frequency
o Unselected
Fig. 18. Average reaction times for choosing each of the display types (selected) and for choosing the other six categorics (unselected) in Experiment 7 .
1
I
Letter and Word Perception
195
tage over regular items, and regular items have an advantage over irregular items. In summary, the results from the overt judgment task paralleled the results from the target-search task. Evidently, readers not only use their knowledge of orthographic structure during the word recognition process, but also are aware of this knowledge and can use it in tasks requiring decisions after the word recognition processes have been completed. As suggested by the model, orthographic structure appears to exert an influence on several stages of language processing (Massaro, 1980).
IV.
General Discussion
The orthogonal contrasts of lexical status, word frequency, positionsensitive frequency, and regularity provided evidence for the following conclusions. Lexical status produced a perceptual advantage of words over equally well-structured anagrams. Word frequency added very little, if anything, beyond that accounted for by lexical status. In the factorial design , regular anagrams were recognized significantly better than irregular anagrams, whereas log bigram frequency had no influence when regularity is controlled. However, the post hoc correlations of these measures of orthographic structure with perceptual recognition of each of the 400 test items revealed that log trigram frequency correlated very highly with performance on the individual letter strings. In this regard, frequency measures allow a fine-grained description of orthographic structure that provides a good index of performance on individual letter strings. The binary classification of lexical status and the small range of the number of irregularities limit the usefulness of these measures as descriptions of a relatively continuous variation in orthographic structure. Some frequency weighting of a regularity description might lead to an improved measure of structure. Until such a description is developed, however, it appears necessary to include lexical status, frequency, and regularity to account for those components of orthographic structure that are psychologically real. The results of the present studies also are relevant to previous studies of orthographic structure. A.
RELATEDRESEARCH
The utilization of orthographic structure in reading was first studied by Miller, Bruner, and Postman (1954), who had subjects reproduce letter sequences presented tachistoscopically. The strings were eight letters and corresponded to different approximations to English based on Shannon’s
I96
Dominic W. Massarn et al.
(1948, 1951) algorithms. Miller et a / . found that performance improved with better approximations to English. By correcting for the relative information per letter in the strings, the amount of information transmitted was shown to be equal for the various approximations. This result is consistent with more recent empirical and theoretical work demonstrating that orthographic structure provides an independent source of information to the reader (Massaro, 1979a, 1979b; Massaro etal., 1980b). In fact, the approximation-to-English algorithms may be viewed as early descriptions of orthographic structure. Accordingly, the more recent studies replicate and extend the Miller et ul. results. The major advances in the recent studies are the more precise descriptions of structure (see Massaro et al., 1980b, Chap. 3) and the quantitative modeling of the processes by which visual information combines with structure during word recognition (Massaro, 1979a, 1979b). Related research by Gibson and her colleagues evaluated the role of word length and pronounceability in a full report of letter strings by both hearing and deaf readers (Gibson, Pick, Osser, & Hammond, 1962; Gibson, Shurcliff, & Yonas, 1970). They found that the number of errors increased with increases in word length and decreased with increases in pronounceability. In the post hoc regression analyses, word length accounted for 72% of the variance and pronounceability accounted for another 15%. Position-sensitive and word-length-specific bigram and trigram frequencies were significantly poorer predictors of performance. However, these counts cannot be used in any straightforward manner for items of various letter lengths. Words of different lengths do not occur with equal frequency, and the less frequent word lengths will naturally have bigrams and trigrams with smaller counts. Therefore, this comparison cannot be considered an adequate test between pronounceability and frequency measures of orthographic structure. The finding that deaf and hearing readers were influenced similarly by pronounceability argues that orthographic structure rather than pronounceability is the important structural variable. Manelis (1974) found an advantage of four-letter words over pseudowords in tachistoschopic recognition, but failed to find a significant correlation between recognition and summed linear bigram and trigram frequencies, as measured by Mayzner and Tresselt (1965) and Mayzner, Tresselt, and Wolin, (1965). In a more recent study, McClelland and Johnston (1977) independently varied position-sensitive bigram frequency and lexical identity in four-letter strings in a Reicher-Wheeler forced-choice task (Reicher, 1969; Wheeler, 1970). In addition, a full report of the four letters either preceded or followed the forced-choice response. The forced-choice responses revealed no effect of either bigram
Letter and Word Perception
197
frequency or an advantage of words over orthographically regular pseudowords. Forced-choice responses also showed a 13% advantage of words and pseudowords over single letters, replicating the word-letter difference of Reicher (1969). The full report score replicated the absence of a bigram frequency effect found for the forced-choice task, but showed an advantage of words over pseudowords. Also, the full report score revealed a large word-frequency effect. In post hoc analyses, McClelland and Johnston report that bigram frequency did not correlate with perceptual accuracy, whereas single-letter position frequency was highly correlated with accuracy. Using a different task, Henderson and Chard (1980) presented items either high or low in both position-sensitive single-letter and bigram frequencies in a lexical decision task. Their results indicate that second and fourth graders were faster in rejecting low-frequency than highfrequency six-letter nonwords. In a related study, Bouwhuis (1979) found that single-letter positional frequency correlated with lexical decisions for three-letter items in Dutch. Reaction times to words decreased whereas reaction times to pseudowords increased with increases in single-letter frequency. Similarly, subjects tended to respond “word” more often to both words and pseudowords if the items were of high single-letter frequency. In contrast, these correlations were considerably diminished when bigram positional frequency was used as the predictor variable. Bouwhuis’ results, when compared with those of Henderson and Chard ( 1 980), imply that the power of the bigram frequency measure with our six-letter items may not generalize completely to smaller letter-string lengths. That is, bigram frequency appears to have more predictive power than single-letter frequency when the items are six letters in length, but the reverse is indicated when the items are three or four letters in length. However, such a conclusion is tenuous for two reasons. First, singleletter and bigrams measures are highly correlated even for small letterstring lengths. Second, experiments demonstrating the predictive power of single-letter frequencies have used linear rather than log counts. Since log counts are uniformly better predictors of the data, it will first have to be shown that log counts do not change the relative power of the two frequency measures. Of the several studies (Bouwhuis, 1979; McClelland & Johnston, 1977) investigating the relative contributions of single-letter and bigram frequencies, only the present studies directly compare linear and log single-letter and bigram counts. The present studies found that log counts are consistently better than linear counts and that bigram counts are better than single-letter counts. In the first of two experiments investigating other structural variables, Spoehr (1978) showed that report accuracy in the Reicher-Wheeler task
Dominic W. Massaro et al.
I98
was lower for five-letter, one-syllable strings made up of five phonemes than for those made up of four phonemes. Performance for words such as rhuriip and pseudowords such as sherk averaged 76% correct, whereas performance was 4% worse for words such as spank and pseudowords such as frost. Average accuracy on words was 7% greater than on pseudowords. In the second experiment, two-syllable words were recognized 13% more poorly than were one-syllable words when phoneme length was equated. Although Spoehr (1978) showed that positionsensitive bigram frequency of the letters could not account for the observed differences, log counts might have been more appropriate. Furthermore, since our counts were derived from the considerably larger Kucerd and Francis (1967) corpus, they are likely more reliable than the Mayzner and Tresselt ( 1965) counts that Spoehr employed. Accordingly, Spoehr’s results are not necessarily inconsistent with the present results. Although recent experiments have failed to find sigfiificant effects of position-sensitive bigram frequency in the perceptual recognition of letter strings, these studies all used linear rather than log counts. We have found much larger effects with log than with linear counts. Linear and log counts correlate .84and .66 for single-letter and bigram position-sensitive counts for four-letter words, .86 and .76 for five-letter words and .85 and .76 for six-letter words in the Kucera-Francis corpus. Therefore, there is sufficient room for improvement of log over linear counts in accounting for perceptual recognition. It is necessary to evaluate log as well as linear counts to provide a sufficient test of frequency measures of perceptual recognition. B.
SUMMARY
The present research assessed the role of orthographic structure in the perceptual recognition and the judgment of letter strings. Lexical status, word frequency, bigram frequency, log bigram frequency, and regularity of letter sequencing were varied across a series of seven experiments. Six-letter words and their anagrams were used as test stimuli in a targetsearch task. Words were recognized better than their corresponding equally well-structured anagrams, but word frequency had small and inconsistent effects. Orthographically regular anagrams were recognized better than irregular anagrams, whereas log bigram frequency did not have an effect. In contrast, post hoc correlations revealed that log trigram frequency did correlate significantly with individual item performance. In a final experiment, subjects judged which of a pair of letter strings most resembled English in terms of either the frequency or the regularity of letter sequences. The results revealed an influence of essentially the same
Letter and Word Perception
I99
dimensions of orthographic structure as was revealed by the perceptual recognition task. The results provide evidence for lexical status, regularity of letter sequencing, and frequency of letter sequencing as important dimensions in the psychologically real description of orthographic structure. REFERENCES Bouwhuis, D. G. Visual recognition in words. Unpublished doctoral dissertation, Katholieke Universiteit, Nijmegen, Netherlands, 1979. Gibson, E. J., Pick, A . , Osser, H . , & Hammond, M. The role of grapheme-phoneme correspondence in the perception of words. American Journal of Psychology, 1962, 75, 554-570. Gibson, E. J., Shurcliff, A . , & Yonas, A . A utilization of spelling patterns used by deaf and hearing subjects. In H. Levin & J . P. Williams (Eds.), Basic studies on reading. New York: Basic Books, 1970. Henderson, L . , & Chard, J. The reader’s implicit knowledge of orthographic structure. In U . Frith (Ed.), Cognitive processes in spelling. New York: Academic Press, 1980. Kucera, H . , & Francis, W . N. Computational analysis of present-day American English. Providence: Brown University Press, 1967. Manelis, L. The effect of meaningfulness in tachistoscopic word perception. Perception & Psvchophysics, 1974, 16, 182-192. Massaro, D. W. Primary and secondary recognition in reading. In D. W. Massaro (Ed.), Understanding language: An information processing analysis of speech perception, reading, and psycholinguistics. New York: Academic Press, 1975. Massaro, D. W. Reading and listening (Tech. Rep. No. 423). Madison: Wisconsin Research and Development Center for Individualized Schooling, December 1977. Massaro, D. W. A stage model of reading and listening. Visible Language, 1978, 12, 3-26. Massaro, D. W. Letter information and orthographic context in word perception. Journal of Experimental Psychology: Human Perception and Performance, 1979, 5 , 595-609. (a) Massaro, D. W. Reading and listening (Tutorial paper). In P. A. Kolers, M. Wrolstad, & H. Bouma (Eds.), Processing of visible language (Vol. I ) . New York: Plenum, 1979. (b) Massaro, D. W. How does orthographic structure facilitate reading? In J. F. Kavanagh & R. L. Venezky (Eds.), Orthography, reading, and dyslexia. Baltimore: University Park Press, 1980. Massaro, D. W., Jastrzembski. J. E., & Lucas, P. A. Frequency, orthographic regularity, and lexical status in letter and word perception (Tech. Rep. No. 550). Madison: Wisconsin Research and Development Center for Individualized Schooling, 1980. (a) Massaro, D. W., & Schmuller, J . Visual features, perceptual storage, and processing time in reading. In D. W. Massaro (Ed.), Understanding language: A n information processing analysis of speech perceprion, reading, and psycholinguistics. New York: Academic Press, 1975. Massaro, D. W . , Taylor, G. A., Venezky. R. L . , Jastrzembski, J. E.. & Lucas, P. A. Lerter and word perception: Orthogmphic structure and visual processing in reading. Amsterdam: North-Holland, 1980. (b) Massaro, D. W., Venezky, R. L., & Taylor, G. A. Orthographic regularity, positional frequency, and visual processing of letter strings. Journal of Experimental Psychology: General, 1979, 108, 107-124. Mayzner, M. S . , & Tresselt, M. E. Tables of single-letter and digram frequency counts for various word-length and letter-position combinations. Psychonomic Monograph Supplements, 1965, 1, 13-32.
Dominic W. Massaro et al.
200
Mayzner, M. S . , Tresselt, M. E . , & Wolin, B . R . Tables of trigram frequency counts for various word-length and letter-position combinations. Psychonomic Monogruph Supplements, 1965, 1 , 33-78.
McClelland, J . L. & Johnston, J. C. The role of familiar units in perception of words and nonwords. Perception & Pswhophysics, 1977, 22, 249-261. Miller, G. A . , Bruner, J . S.,& Postman, L. Familiarity of letter sequences and tachistoscopic identification. Journal of Generul Psychology, 1954, 50, 129- 139. Reicher, G. M. Perceptual recognition as a function of meaningfulness of stiniulus material. Journul of Experimental Psychology, 1969, 81, 275-280. Rosinski, R. R., & Wheeler, K. E. Children’s use of orthographic structure in word discrimination. Psychonomic Science. 1972, 26, 91-98. Shannon, C. E. A mathematical theory of communication. Bell Systenz Technicul Journul, 1948,27, ~
379-423; 623-656.
Shannon, C. E. Prediction and entropy of printed English. Bell System Technicul Journul. 1951,30, 50-64.
Shepard, R. N . , & Podgomy, P. Cognitive processes that resemble perceptual processes. In W . K . Estes (Ed.), Hundbook o] learning and cognitive processes: Human information processing (Vol. 5 ) . Hillsdale. N.J.: Lawrence Erlbaum, 1978. Solomon, R . L., & Postman, L. Frequency of usage as a determinant of recognition thresholds for words. Journal of Experinientul Psychology. 1952, 43, 195-201. Solso, R. L., & King, J . F. Frequency and versatility of letters in the English language. Rehuvior Research Merhods & Instrumentation. 1976, 8 , 283-286. Spoehr, K . T . Phonological encoding in visual word recognition. Journal of Verbul Learning oncl Verbal Behavior, 1978, 17, 127-141. Taylor, G. A . The contribution of’orthugruphic strucrure fu the perception of letter strings. Unpuhlished doctoral dissertation, University of Kansas, 1977. Taylor, G . A , , Klitzke. D., & Massaro, D. W . Considerations in software character generation. Behavior Research Methods & Instrunrentation, 1978, 10, 787-788. (a) Taylor, G . A , , Klitzke, D., & Masssaro, D. W. A visual display system for reading and visual perception research. Behavior Research Methods & Instrumentation, 1978, 10, 148-153. (b) Taylor, M. M., & Creelman, C. 0. PEST: Efficient estimates of probability functions. Journal offhe Acouslical Society of America, 1967, 41, 782-787. Travers, J . R., & Olivier, D. G . Pronounceability and statistical “Englishness” as determinants of letter identification. American Journal of Psychology, 1978, 91, 523-538. Venezky, R. L . . & Massaro, D. W . The role of orthographic regularity on word recognition. In L. Resnick & P. Weaver (Eds.), Theory prccctice of early reading. Hillsdale, N . J . : Lawrence Erlbaum, 1979. Wheeler, D. D. Processes in word recognition. Cognitive Psychology, 1970, 1, 59-85.
SELF A N D MEMORY Anthony G.Greenwald OHIO STATE UNIVERSITY COLUMBUS, OHIO
I . 1901-1935: Four Positions.. .... ..... .... .... A. Freud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Claparede . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. Bartlett . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. Koffka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... . .. , . E. Summary of the Early Views of Self and Memory . . . . . . . . . . . . . . . . . . . . . . . . 11. 1977 Onward: Discovery of the Self-Reference Effect ... A. Rogers and Kuiper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Markus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. Keenan and Baillet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. Bower and Gilligan.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E. Comment on Interpretations of Self-Reference Effects . . . . . . . . . . . . . . . . . . . , , 111. Memory When Self Is Involved.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . , , A. Self-Generation of Study Material Facilitates Recall . . . . . . . . . . . . . . . . . . . . . , . B. Egocentric Perspective Facilitates Recall . . . . . . . . . . . . . . . . . . . . . , . . . , . , . . , , C. Ego-Involvement: Persisting Tasks Facilitate Recall . . . . . . . . . . . . . . . . . . . . . . , IV. Theoretical Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Three Self/Memory Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . , . , , B. Properties of the Self System. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . , C. Explanation of the SelfiMemory Effects in Terms of the Self System . . . . . . . . , V. Scope and Importance of the Self System.. . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Implicit Theories of Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . , , . B. The Self System and the Episodic-Semantic Distinction . . . . . . . . . . , . . . , . , , . . C. Theoretical Status of the Self.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . , . , D. Unfinished Business . . . . . ....................................... References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
202 202 203 204 207 209 210 210 21 I 21 1 212 213 214 214 218 220 223 223 224 225 229 229 230 23 I 232 233
Theories that propose an important role for self (or ego) i n memory have been in the wings and background of psychology since the beginning of the twentieth century. It has only been quite recently, however, that experimental studies of memory have produced data that demand a center-stage role for the concept of self in the study of memory. This chapter reviews highlights of the theoretical history of the relation of self and memory and draws together the diverse lines of recent research that provide a basis for further theoretical development. In synthesizing this 201 THE PSYCHOLOGY OF LEARNING AND MOTIVATION. VOL 15
Copyright 0 1981 by Academic Press, Inc All nghis of reproduction in any form reserved ISBN 0-12-543315.8
Anthony G. Greenwald
202
work, I suggest that ordinary, voluntary recall is a property peculiarly characteristic of the selfsystern. Indeed, the major function of the self system may be to provide the organization that preserves access to information from the varyingly distant past.
I.
1901-1935: Four Positions'
Seeds of present interest in the role of self in memory can be found in the works of four psychologists who had little in common-working in four different intellectual traditions, in four different countries, and in three different languages, and basing their conclusions on widely diverse types of data. A.
FREUD
Clinical observations of dramatic pathologies of memory provided the basis for some of the earliest speculations about the role of self in mernory. It is well known that Freud interpreted ordinary forgetting in terms of motivational conflicts, and that he extended this type of analysis to his account of repression-an occasionally massive forgetting assumed to be symptomatic of conflicts involving sexual motivation. Freud's theories of repression, its basis in psychosexual conflict, and the role of ego as the agency of repression are too well known to require review here. Also, these theories are readily accessible both in the original (Freud, 1915/ 1957; 1923/1961) and in many secondary sources (e.g., C. Brenner, 1957; Erdelyi & Goldberg, 1979). Instead, we consider some of Freud's earliest speculations on the role of self in memory-which came from his self-observations of everyday lapses of memory. I was unable to find a patient's name which had a certain reference to my early life. The analysic had to be followed over a long devious road before the desired name was discovered. The patient expressed his apprehension lest he should lose his eyesight; this recalled a young man who became blind from a gunshot, and this again led to a picture of another youth who 5hot himself, and the latter bore the same name as my first patient, though not at all related to
'This historical survey presents four positions through prominent spokesmen whose works have survived. It is certainly not a comprehensive survey, and may not properly credir the origin of various ideas to their true originators. The reason for mentioning this fact i s to note that the problem of retrieving the theoretical past of a discipline is similar to that of retrieving a personal past. The effort of tracing through libraries to locate publications that have not been retained by the current citation network resembles that of searching for personal memories that have fallen out of the current retrieval network-- which, in this chapter, is identified as the self.
Self and Memory
203
him. The name became known to me, however, only after the anxious apprehension from these two juvenile cases was transferred to a person of my own family. Thus an incessunt streum of “.self-reference”~owsthrough m y thoughts concerning which I usually have no inkling, but which betrays itself through such name-forgetting. It seems as if I were forced to compare with my own person all that 1 hear about strangers, as if my personal complexes became stirred up at every information from others. It seems impossible that this should be an individual peculiarity of my own person; it must, on the contrary, point to the way we grasp outside matters in general. (Freud, 190111917, pp. 41-42, italics added)
From a present-day perspective, Freud’s observations on everyday lapses of memory may be more interesting for their suggestions about the route by which forgotten information can eventually be retrieved, than for his theory of the motivational conflict that presumably caused the forgetting. In the quoted example, the retrieval route involved self-related associations. Freud’s hunch about the “incessant stream of ‘selfreference’ will be seen to fit very well with recent research findings. ”
B.
CLAPAREDE
The dramatic pathology of the Korsakoff syndrome suggested an important role of self in memory to the medically trained Swiss psychologist Edouard Claparede (191 1/1951 ) . His case observations and conclusions are given here at some length. The patient was a woman hospitalized at Asile de Bel-Air. She was 47 at the time of the first experiment, 1906. Her illness had started around 1900. Her old memories remained intact: she could correctly name the capitals of Europe, make mental calculations, and so on. But she did not know where she was, though she had been at the asylum five years. She did not recognize the doctors whom she saw every day, nor her nurse who had been with her for six months. When the latter asked the patient whether she knew her, the patient said: “No Madame, with whom have I the honor of speaking‘?” She forgot from one minute to the next what she was told, or the events that took place. She did not know what year, month, and day it was, though she was being told constantly. She did not know her age, but could figure it out if told the date. 1 was able to show, by means of learning experiments done by the saving method, that not all ability of mnemonic registration was lost in this person. What is worthy of our attention here was her inability to evoke recent memories voluntarily, while they did arise automatically, by chance, as recognitions. (pp. 68-69)
If one examines the behavior of such a patient, one finds that everything happens as though the various events of life, however well associated with each other in the mind, were incapable of integration with the me [ego] itself. The patient is alive and conscious. But the images which he [or she] perceives in the course of that life, which penetrate and become more or less fixated in his organic memory, lodge there like strange bodies; and if by chance they cross the threshold of consciousness, they do not evoke the feeling of “me-ness” which alone can turn them into “memories. ”
Anthony G . Greenwald
204
We can distinguish between two sorts of mental connections: those estahlished ttturually hrfweerr r.e~~rssertrutioris, and thosc cstablished hetween represcnttrtions and /ha m e , the personality. In the case of purely passive associations or idea-reflcxcs, solely the first kind of connection operates; in the case of voluntary recall and recugnition, where the tne plays a role, the second kind of connection enters. In relation to the tnr as center, the connections of the second kind may he called egorsrtmrir functionJ, those o f the first rnurginul. (p. 71)
With some change of language, Claparede’s observations might have a very contemporary ring. Try, for example, substituting “episodic” and “semantic” (Tulving, 1972) for “egocentric” and “marginal” in the last sentence quoted. (This relationship will be returned to in Section V , B.) Claparede’s use of the “saving method” to demonstrate that his Korsakoff patient was capable of “mnemonic registration,” but with “inability to evoke recent memories voluntarily, ” provides a methodological parallel to some of the most recent work on amnesia (e.g., Cohen & Squire, 1980). C. BARTIXTT
In his Remembering: A Study in Experimental und Social Psychology, Bartlett (1932) introduced the concept of schema to the study of human memory (see Hastie, 1980, for a useful survey of the schema concept). Also in that book, Bartlett rejected a role of the self in memory, as will be seen. A review of his position is important to the study of self in memory, to understand both his concept of schema and his reasons for deciding to do without the self. Bartlett ’s concept of schema was based importantly on the neurological studies of Head ( 1 920), who sought to explain the organism’s sense of its location in space. Head (1920) postulated and defined the schema as “a postural model of ourselves which constantly changes” (p. 605; also quoted in Bartlett, 1932, p. 199). Thus, for example, if while walking I turn my head to the right, the postulated schema registers this change, which in turn allows me to judge that an object popping up in front of my nose is to the right of my line of travel rather than straight ahead. Bartlett (1 932), with this inspiration, defined the memory schema as an active organization of past experiences, which must always be supposed to be operating in any well-adapted organic response. (p. 201) In its schematic form the past operates en musse, or, strictly, not quite en mcrsse, because the latest incoming constituents which go to build up a ‘schema’ have a predominant influence. (P. 202)
Self and Memory
205
Bartlett ’s thus crediting the latest incoming stimulus with a privileged place in memory shows the influence of Head’s concept, in which the most important aspect of the schema is its use of information about the latest action to modify itself. It is generally appreciated that Bartlett used the concept of schema to explain a constructive character of memory. He took the fact that experiences are often simplified (or otherwise systematically modified) in memory to mean that a person often infers the past with the aid of a schema, rather than being able to consult an exact record of events. However, Bartlett’s theory of the schema’s operation is not generally known. This rather complex (and unsatisfying) theory is quoted here. What, precisely, does the ‘schema’ do? Together with the immediately preceding incoming impulse it renders a specific adaptive reaction possible. It is, therefore, producing an orientation of the organism towards whatever it is directed to at the moment. But that orientation must be dominated by the immediately preceding reaction or experiences. To break away from this the ‘schema’ must become not merely something that works the organism, but something with which the organism can work. . . . So the organism discovers how to turn round upon its own ‘schemata’, or, in other words, it becomes conscious. It may be that what then emerges is an aftitude towards the massed effects of a series of past reactions. Remembering is a constructive justification of this attitude; and, because all that goes to the building of a ‘schema’ has a chronological, as well as a qualitative, significance, what is remembered has its temporal mark; while the fact that it is operating with a diverse organised mass, and not with single undiversified events or units, gives to remembering its inevitable associative character. (Bartlett, 1932, pp. 207-208)
The above passage shows Bartlett struggling a bit to escape the domination “by the immediately preceding reaction or experiences” that was transported into his theory from Head’s. In this passage there is considerable vagueness associated with the terms “turn round on its own ‘schemata’, ” “attitude,” “diverse organised mass, ” and with the appeal to consciousness. Nevertheless, Bartlett ’s calling attention to constructive aspects of memory has been of great significance. Because of this significance, it is useful to give as clear a statement as possible of Bartlett’s theory. Interestingly, and with homage to Bartlett, it may be that the following statement of his theory is in part a reconstruction, based on a schema that combines his original statement along with more recent statements by others into a “diverse organised mass.” In contrast to Head’s use of schema to designate the current status of an ever-changing entity, Bartlett used schema to characterize the common core of a series of similar past experiences. Bartlett’s “schema” therefore operated at a higher level of abstraction than Head’s. We can compare the two by thinking of repetitions of a complex movement such as a difficult dive involving twists and somersaults. Head’s schema is the diver’s
206
Anthony G . Greenwald
rapidly changing sense of position and location, whereas Bartlett’s schema is the changing pattern of the entire dive on successive repetitions. Head’s schema enables the diver, in the middle of the dive, to judge when to straighten out for entry into the water, whereas Bartlett’s schema allows the diver, between dives, to judge that it is necessary to straighten out earlier on the next dive than on the last one in order to achieve proper entry. In Bartlett’s theory of the use of schemata in recall, some current stimulus (such as an instruction to recall a particular story) acts on the schema to elicit an citritude-a conscious feeling that guides use of the schema to infer the experience that originally gave rise to the schema. Because the schema is a “diverse organised mass,” rather than an exact record of prior events, recall is often “schematic” rather than literal. The theory relies implicitly on associative processes-which allow current stimuli to evoke the schema and also serve’to piece together the components of the schema-but assumes that the organizational aspects of the schema are not reducible to principles of rote association. Bartlett ( 1 932) was willing to assume that the person’s memory shows organizational properties at a very high level, but chose to argue against labeling this high-level organization as a “self”: The materials dealt with by different ‘schemata’ overlap. and both the ‘schemata’ and the appetites, instinctive tendencies, attitudes, interests and ideals which build them up display an order of predominance among themselves. Moreover, this order remains relatively persistent for a given organism. This is equivalent to saying that recall is inevitably determined by temperament and character. All these cun.siderarions, however, give us no justificution fur speaking of some inrungible and h,vporhericnl Sew which receives and rnaintuins innurnerahle Iruces und re-rtirnu1afe.sthem whenever the need arises. All that we can say for certain is that the mechanism of adult human remembering demands an organisation of ‘schemata’ which depends upon an interplay of appetites, instincts, interests and ideals peculiar to any given subject. Equally, of course, we have so far no ground for denying the existence of a substantial, unitary Self, lurking behind all experience, and expressing itselfin all reactions. We know only that the evidence of the experiments which have been considered does not necessitate such a hypothesis. (pp. 308-309, italics added)
It is apparent from this and other remarks that Bartlett (1932, pp. 308-31 1 ) intended to reject the transcendental self, or “pure ego,” just as William James (1890) had earlier done. His observation of a high level of organization among appetites, interests, and attitudes is nonetheless comparable to the sort of evidence on which others have based an empirical conception of self.2 ZThetranscendentaYempirica1 distinction is approximately the same as knowedknown or subject/ object. In philosophy, the transcendental self is often identified with Kant’s idea of self as the agent of perception, and the empirical self with Hume’s view of the self as a bundle of perceptions. The
Self and Memory
207
D. KOFFKA In his major work, Principles of Gestalt Psychology, Koffka (1935) devoted over 60 pages to the hypothesized role of ego in memory (esp. pp. 319-342, 514-528, 591-614). Koffka distinguished between self and ego, the former being defined as a central subsystem of ego, but his analysis of memory was given in terms of the broader system, ego. The following passages give some of the central statements of Koffka’s theory of the role of ego in memory. In our theory the total field of excitation is divided into two major sub-systems, each containing numerous sub-systems of its own: the Ego and the environment. And the trace field which is created by the excitation field contains the same dichotomous organization. (Koffka, 1935, pp. 520-521) If an environmental trace is in close connection with the Ego system it will not only be in communication with the particular time structure of that system with which it communicated at the time of its formation; but because of the coherence of the whole temporal Ego system it will be in communication with later strata also. (p. 522)
Although Koffka (1935) provided no graphic representation of the ego and environment systems, his words directly suggest the representation given in Fig. 1 . Figure I also incorporates information from Koffka’s descriptions of temporal and motivational aspects of the ego-environment structure, and of the centrality of ego in the psychological field. In these passages a trace is a memory representation or record, which is left as a residue of the activity of a perceptual process. The temporal stratification of traces is one of the factors which determine their availability. A trace within its stratum is connected with the Ego of the same stratum, but may be far removed from the Ego of a later stratum. (p. 525) Mere temporal sequence, however, is but one factor in this complex dynamic connection. Availability of the trace. . . depends upon proper connection between the trace system and the Ego.. . . If a trace is derived from a process which was directly connected with a person’s interests, then it will have its place in a field formed by processes of high intensity and will be in particularly close connection with the Ego system. (p. 526) Inasmuch as the Ego is, as a rule, more or less in the centre of its environment, we can picture the Ego part of the trace column as its core and the environmental part as a shaft, keeping in mind that core and shaft support each other. (p. 609)
As it did for Bartlett, the concept of attitude played an important role in Koffka’s theory of memory. “The effect of attitude [is] to put a process in concept of a transcendental self is also often associated with the idea of a soul (for which the Greek is psyche). William James (1890, Chap. 10) provides a lucid introduction to the transcendentali empirical distinction.
Anthony G. Greenwald
208
communication with traces which, without such an attitude, they could not ‘find”’ (Koffka, 1935, p. 607). Figure I includes a portrayal of Koffka’s proposed role of attitude in memory, illustrated in terms of an experiment in which figures seen in a laboratory session yesterday are to be recognized in a second session today. The attitude created by the task of today’s experiment “has the character of a quasi-need, it corresponds to a tension in the Ego part at the tip of the column [the part presently being formed]. This tension can be relieved only through that part of the trace column which contains yesterday’s figures (Koffka, 1935, p. 609). The attitude thus enables connection of today’s excitation field with the trace field created by yesterday’s experiment, despite intervening trace strata, such as one associated with a concert attended last night. ”
Fig. 1 . Simplified representation of Koffka’s conception of an ego-centered memory structure. Egu is at the center, and environment at the periphery, of the excituliun ,field. Residues of past excitation fields form /race fields that are arrayed in temporal strata. The cumulation of trace fields (the trace culurnn) defines ego as a memory structure that forms a core within the shufi defined by the traces of environmental excitation. The three strata shown illustrate Koffka’s example relating today’s experiment, a concert attended last night, and yesterday’s experiment (see text). The present excitation field (A) is assumed to be in more direct contact with an attitudinally similar past trace field (B) than with a temporally closer (but attitudinally dissimilar) one (C), which may have direct communication (D) with other ones dynamically similar to i t.
Self and Memory
209
Accurate recognition of a figure presented today requires that both the trace of yesterday’s figure and the process of the corresponding one presented today be in contact with the ego fields of their respective strata. Koffka’s memory theory was designed to encompass Bartlett ’s findings, and therefore the two theories contain no mutually incompatible assumptions. Koffka’s much more detailed account filled important gaps in Bartlett ’s formulation as represented by the latter’s unelaborated appeal to the “diverse organised mass” of a schema, to attitudes, and to the role of consciousness in enabling the organism to “turn round” on its schemata. Interestingly, Koffka concluded that he could not account for (among other things) the effect of attitude on recall except by providing a central role for ego i n memory-in contrast with Bartlett’s rejection of self. Again, this is not a mutual incompatibility. As previously noted, the self that Bartlett rejected was a transcendental knower (see Allport, 1965, p. 129), whereas the self (ego) that Koffka required was an organization of knowledge (see Greenwald, 1980). Koffka’s formulation of the ego can be regarded, therefore, as a translation and elaboration of Bartlett’s ( I 932) speculation about the dependence of human remembering on “an organisation of ‘schemata’ which depends on an interplay of appetites, instincts, interests and ideals peculiar to any given subject” (p. 309). OF E. SUMMARY
THE
EARLYVIEWSOF SELFA N D MEMORY
Freud, Claparede, Bartlett, and Koffka agreed in asserting that ordinary, voluntary memory depended on a very high level of organization among residues (traces or memories) of past experience. Three of the four (all except Bartlett) believed that this organization was at least in large part the same as that which we call “self. Freud and Koffka had the most detailed theories of the role of the self in memory. Freud’s theory of repression (which has not been described in any detail here) spawned many laboratory investigations, which, however, proved frustratingly inconclusive and often nonsupportive (see review in Holmes, 1974). Koffka’s theory of ego as a memory structure received much less empirical attention than Freud’s. Perhaps partly as a consequence of not having received much test, Koffka’s theory remains consistent with much of the existing evidence. It is clear that the early theorists were concerned with a level of organization that is hierarchically superordinate to that with which more recent work on organization in memory has been concerned (e.g., Tulving & Donaldson, 1972). However, in the last few years, memory research has begun to address phenomena at the highest levels of organization. It is to some of this work that we now give attention. ”
Anthony G . Greenwald
210
11.
1977 Onward: Discovery of the Self-Reference Effect
Theoretical developments concerning the role of self in memory were scarce in the four decades following Koffka’s major theoretical effort. That hiatus has ended now, as is evident in a set of related contributions that appeared starting in the late 1970s. These studies, to be reviewed in this section, are outgrowths of cognitive psychologists’ recent efforts to develop detailed working models of human memory. The theorists’ backgrounds are in personality, social psychology, and experimental psychology. This convergence from several directions on the relationship between self and memory suggests that new theoretical developments in this area will be widely useful. A.
ROGERSA N D KUIPER
Starting from an interest in determining how respondents to personality inventories accessed the information needed to generate self-report judgments, Rogers (1974) turned his attention to a task in which subjects judged whether or not trait-word stimuli were self-descriptive. Subsequent experiments by Rogers, Kuiper, and Kirker (1977) and Kuiper and Rogers (1 979) examined both latencies of such judgments and subsequent performance on unexpected tests for recall of the judged stimuli. Compared to a variety of other judgments on similar stimuli, the selfreferent judgments were made more rapidly and also led to greater (incidental) recall of the trait stimuli. The authors’ initial interpretations of these results employed the depth-of-processing or degree-of-elaboration explanation that had been proposed for other results using the judgment/ incidental recall procedure (Craik & Lockhart, 1972; Craik & Tulving, 1975). The results were taken to suggest that self-reference provided the basis for even deeper or more elaborated processing than did the semantic judgment task with which it was compared. The self [is] an abstract structure that contains both general trait-like entries and some specific behavioral exemplars or instances. This memory structure is active during the input and interpretation of self-related information and provides a degree of ”meaning” or embcllishment to the inconling information. (Kuiper & Rogers, 1979. p. 51 1 )
Most recently, Rogers (1 98 1 ) has added the conclusion that an evaluative, or affective, component of self-referent judgments is implicated by the latency and memory effects associated with such judgments. Affect exerts its major effects during the encoding of personal information. . . . The person can be thought of as “maintaining a watching brief” for indicators of self-relevant events. When such an indicator is encountered, the person’s attention is directed toward it. . . . The encoding
Self and Memory
211
of a unit of personal information will be a combination of the actual cognition plus an affective signal or tag.. , . The strength of the [affective] signal will vary directly as the degree of self-relevance-and will effect [sic] the strength of the memory trace left by the encoding operation. The superiority of self-referent memory performance for evaluative tasks. . . results from the rich and strong two-factor 1i.e.. cognitive and affective] trace left by a self-referent judgment. (pp. 207-209)
B.
MARKUS
Like Rogers and Kuiper, Markus (1977, 1980) has experimentally studied latencies for self-report judgments. Her focus has been on the way in which a person’s idiosyncratic knowledge structures (self-schemas3) influence judgment and recall of information. Markus’s use of “schema” is in the tradition of Bartlett (1 932), although she takes her exact definition from Neisser’s (1 976) recent description of the schema as a structure that ‘‘accepts information as it becomes available at sensory surfaces and is changed by that information; it directs movement and exploratory activities that make more information available, by which it is further modified” (p. 54). A finding that Markus has replicated in several contexts is that persons who are “schematic” for a trait (they judge both that the trait is self-descriptive and that it designates an important characteristic) make self-relevant judgments bearing on that trait more rapidly than they do for other traits, and also more rapidly than do people who are not schematic for the trait. Markus (1980) has summarized her views on the role of self-schemas in judgment and memory as follows: Thinking about the self, like thinking about any object, involves providing meaning to an incoming stimulus so that it can be represented in memory. Meaning is produced by fitting or assimilating various components of the stimulus to the knowledge structures contained in memory, and then adjusting or accommodating these structures so that they can adequately represent the relevant data. . . . Because even the most other-directed among us spend so much time thinking about the self, self-structures are frequently used and available for future use in information processing. Thus, we are likely to pay more attention to information about the self and to expend more effort reflecting on it. Often we may even change data so that it fits with our self-concept or ignore or selectively forget information if it does not match our ideas about ourselves. (p. 130)
C.
KEENAN A N D BAILLET
A number of researchers reacted to the original report by Rogers et a1 . (1 977) of an effect of self-reference on memory by asking whether simi)The only plural of schema given in standard dictionaries is schemaru. Nevertheless many contemporary psychologists, including Markus, use schernas as the plural form. The two forms are used interchangeably in this chapter.
Anthony G. Greenwald
212
larly high levels of recall would not be found also for traits judged in terms of their applicability to other people (rather than to self). Of these several studies (Bower & Gilligan, 1979; Kuiper & Rogers, 1979; Lord, 1980), the two experiments by Keenan and Baillet (1980) may have the greatest impact in suggesting further theoretical development. Their first experiment showed that both speed of judgments of trait applicability and subsequent incidental recall for the judged traits increased directly with the degree of familiarity of the person judged, over seven levels of familiarity (Jimmy Carter, teacher or boss, favorite character, friend, parent, best friend, self). Their second experiment showed that the effect of familiarity on recall occurred for judgments on evaluative dimensions (e.g., Does “rude” describe: you, [your parent, your favorite character, Jimmy Carter]?) but not for judgments on factual dimensions (e.g., Do you [your parent, etc.] have gills‘?) These findings by Keenan and Baillet actually provided a major basis for Rogers’ (198 I ) having concluded that there was an affective component of self-referent judgments. Keenan and Baillet, however, although noting the possibility of an affect-based account, preferred a cognitive interpretation. The essence of their cognitive interpretation was that evaluative judgments, but not factual ones, employ a “rich” conceptual structure. We suggest that the richness of the conceptual structure to which an event is encoded may provide a more general definition of encoding elaboration than has been offered to d a t e , . . because it applies across a wide range of semantic encoding tasks and it can be determined ahead of time rather than post hoc. (Keenan & Baillet, 1980, p. 667) As research into memory for personally and socially significant events proceeds, . . . it may be that the data will call for models [that] incorporate the constructs of motivational psychology
into the mechanistic process models of cognition. For now, however, the data can be adequately explained using only cognitive constructs; they may raise the possibility of a motivational account, but they do not compel one. (p. 668)
D.
BOWERA N D GIILIGAN
In their investigation, Bower and Gilligan (1979) sought to show that the heightened recall of trait stimuli originally found by Rogers et al. (1977) did not depend on judging the applicability of traits to self. They succeeded in finding two other tasks that produced comparably high incidental recall of trait stimuli: ( a )judging relevance of traits to remembered personal experiences, and ( b )judging relevance of traits to remembered experiences involving one’s mother. Their interpretation was offered in terms of a semantic network model (HAM-Anderson & Bower, 1973) of encoding and memory, and they concluded that the self-concept behaved in memory in the fashion expected by this model.
Self and Memory
213
There is nothing special about the self schema as a mnemonic peg; any well-differentiated person will do. (Bower & Gilligan, 1979, p. 429)
Bower and Gilligan (1979) thus subsumed the role of self in memory under the generalization that “good memory depends on relating the inputs to a well-differentiated memory structure” (p. 420). Note that their conclusion can be used either to suggest that the study of memory needs no special treatment of the self or, alternatively, to suggest that the self should be accorded a special role because it may be the best-differentiated cognitive structure available. E.
COMMENT O N INTERPRETATIONS OF SELF-REFERENCE EFFECTS
The researchers whose works have just been reviewed have no disagreement about the major self-reference findings, which can be summarized briefly as follows: Information judged for self-relevance is better recalled than information judged on other dimensions; judgment of relevance of information to persons other than oneself facilitates memory as a function of familiarity with the person being judged; and the beneficial effect on memory of self-reference or familiarity disappears when the judgment has no evaluative content. These same researchers, however, differ in their conclusions about the status of self as a psychological construct. Rogers and Kuiper and Markus have interpreted the selfreference findings as manifestations of an organization (self or self-schema) that has unique psychological properties, whereas Keenan and Baillet and Bower and Gilligan have preferred not to attribute special properties to the self as a psychological construct. These differences of opinion about the status of self are reminiscent of Bartlett’s and Koffka’s differing conclusions about the existence of self as a special entity. As noted earlier, Bartlett and Koffka apparently agreed in finding evidence for high-level organizational processes in memory, but differed on the appropriateness of using the designation “self” for this organization. The present differences of interpretation about the selfreference effect also represent, in part, differing preferences for theoretical labels. Despite their apparent conceptual differences, Keenan and Baillet and Bower and Gilligan can agree with Rogers and Kuiper and Markus that the self-reference findings demonstrate the operation of a high level of cognitive organization. However, there is another aspect of disagreement that cannot be set aside so easily-namely , that concerning the possible involvement of affective processes in the self-reference effect. Rogers (1981) has suggested that affect plays an important role in processing self-relevant information, whereas Keenan and Baillet
Anthony G. Greenwald
214
( 1 980)-whose
research provided some of the most important evidence on which Rogers’ conclusion was based-felt that it was premature to conclude that affective processes are involved in the self-reference effect. Perhaps relevant to this issue are the recent suggestions by Bower (1980) and by Clark and Isen (1981) that affects or emotions can be treated as conceptual nodes in a memory structure, such that analyses of their effects in memory may be given in terms of standard cognitive principles. It will be interesting to follow the future development of this issue, which promises to touch on important theoretical matters. 111. Memory When Self Is Involved
In addition to the self-reference effect, which has inspired the theoretical efforts just reviewed, several other well-established findings suggest a special role of self in memory. These findings are here grouped into categories of self-generution, egocentric perspective, and egoinvnlvernent effects. In order to maintain a broad scope in this review, the coverage of these topics is representative rather than thorough. The common theme that can be found in the results to be reviewed is that memory is superior when self is involved in information processing than when self is not involved. The aim of the review in this section is to provide a basis for later translating the assertion “self is involved” from empirical operations into psychological conceptualization. A.
SELF-GENERATION OF STUDYMATERIAL RECALL
FACILITATES
I . The Self-Generation Artifact in Law-oj-Efiect Experiments As is the case with many interesting phenomena, the first empirical appearance of the effect of self-generation in facilitating recall was an unwanted experimental artifact, not initially identified as interesting in its own right. Thorndike (e.g., 1932) conducted many experiments that used the general procedure of presenting a stimulus item (e.g., a word in a foreign language) together with several alternative (e.g., translation) responses in a multiple-choice format. On retests, subjects repeated rewarded responses (ones that had been called “correct”) at a rate well above chance. This finding was interpreted as supporting the law qf effect-an automatic strengthening effect of reward on stimulus-response connections. Unfortunately it was also true that unreinforced responses
Self and Memory
215
were repeated at a substantially above-chance rate, which suggested to Thorndike that a law of exercise (strengthening as a consequence of unrewarded practice) was needed in addition to the law of effect. As a result of subsequent investigations (see reviews in Nuttin & Greenwald, 1968, pp. 145-154; Postman, 1947, 1962), it became apparent that the assumed chance-repetition rate (i .e., repetition probability of l / n , where n is the number of choice alternatives on each trial) was not appropriate as a baseline against which to compare reward effects. That is, subjects did not select randomly among alternative responses, and a response that was especially likely to be chosen on an acquisition study trial was likely also to be selected at an above-chance rate on a subsequent test trial. Even instructions that the multiple-choice experimental items constituted an extrasensory perception task (e.g., Wallach & Henle, 1941) were insufficient to induce chance selections among response alternatives. A method of eliminating this self-generation artifact was eventually achieved by providing the subject the illusion of choosing responses, the verbal content of which was actually under the experimenter's control (Greenwald, 1970; Nuttin & Greenwald, 1968, pp. 153-155). As is now clear, the purity of method achieved with this innovation was purchased by eliminating from the trial-and-error learning experiment an interesting process (self-generation) that should not have been considered just a troublesome artifact. Before attempting to characterize this process further, let us consider some not-obviously related phenomena of more recent discovery. 2.
Cognitive Response Learning in Persuasion
In the two decades after World War 11, researchers at Yale University, under the leadership of Carl Hovland, conducted an intensive program of laboratory studies of communication and persuasion (e.g., Hovland, Janis, & Kelley, 1953). Those studies followed up on the researchers' earlier wartime efforts to develop and test indoctrination programs, and were guided by learning principles developed in the work of Clark Hull. A major frustration of the Yale communication research program was recurring evidence that the persuasion produced by effective communications did not conform to a simple learning model. According to that learning model, the persuaded audience members should have been those who best learned the arguments that had been presented by the communicator; instead, persuaded and nonpersuaded subjects typically showed equal levels of learning and subsequent memory for communicated arguments. The conceptual puzzle posed by such resu1ts"appears now to have been resolved by the proposition that, during a persuasion
216
Anthony G. Greenwald
episode, the audience generates covert evaluative responses that do not necessarily agree with the communication. If these cognitive responses-whether they be counterarguments or supporting thoughtsare later more easily retrieved than are the communicator’s arguments, then persuasion can have an orderly learning/memory interpretation in terms of cognitive responses (Greenwald, 1968). Greenwald and Albert (1968) gave an intial demonstration that subjects are much better able to recall their own generated thoughts on a controversial topic than to remember comparable statements that were produced by other subjects. This preferred status in memory of cognitive responses, relative to communicated arguments, has subsequently been demonstrated many times (see Petty, Ostrom, & Brock, 1981).
3. SelfGeneration in Verbal Learning At about the same time that persuasion researchers were investigating memory for audience cognitive responses, verbal learning researchers began to investigate the memorial consequences of an active role of the subject in generating to-be-remembered material. The first such study (Bobrow & Bower, 1969) was inspired by developments in psycholinguistics. Bobrow and Bower found that, when subjects actively generated a sentence linking two concrete nouns (e.g., the nouns DOG and ROPE might be linked by “The DOG bit the ROPE”), they were subsequently better able to produce the second noun when cued by the first than were subjects for whom the experimenter had provided the linking sentence. Bobrow and Bower concluded that the actively generating subjects remembered better because they comprehended better; in more recent terminology, this explanation corresponds to the idea of deeper or more elaborated processing (Craik & Tulving, 1975). Interestingly, Bobrow and Bower (1 969) rejected as “simply too vague” (p. 458) the competing hypothesis that “the act of successfully searching for a sensible connective to link [the two nouns] parallels or is equivalent to the process of constructing a scheme for retrieving [the second noun] from memory when given [the first]” (pp. 455-456). Results that are presently to be reviewed suggest that this search-parallels-retrieval interpretation may be the more accurate. The most direct evidence for the facilitating effect of self-generation on recall has come from some recent experimental studies by Slamecka and Graf (1978). Their general procedure involved presentation of a stimulus word and a rule by which this stimulus was related to a response wordfor example, the stimulus word rapid and the rule synonyrn. The main experimental variable was variation of whether the subject’s response word was read or generated. In the generate condition, the stimulus was
Self and Memory
217
followed by the first letter of the desired response--e.g., rapid-fi in the read condition the subjects were instead presented with the full paire.g., rapid-fasf. In five experiments, the generate condition was consistently superior to the read condition in recognition, cued recall, and free recall of the response words (not in free recall of the stimulus word^).^ The comprehension or depth-of-processing interpretation suggested by Bobrow and Bower (and again, later, by Erdelyi, Buschke, & Finkelstein, 1977) was considered problematic by Slamecka and Graf. The latter authors observed that a depth-of-processing interpretation predicts that ( a )the beneficial effect of generation should extend to the stimuli of their paired associates, and ( h ) the generation effect should be greater for a semantic (e.g., synonym) rule than for a phonetic rule (rhyme), which presumably requires shallower processing. Neither of these expectations was supported by the data of Slamecka and Graf’s experiments. 4.
Reward-Bused Education and the Self-Generation Effect
We can now spell out the relation between the self-generation effect and the role of reward in learning. The frequently observed asymmetry of reward and punishment effects-that is, the fact that rewards increase performance probabilities and punishments do not equivalently decrease them-has been used widely as a basis for advocating reward-only training procedures (especially by Skinner, e.g., 1953). Thorndike (e.g., 1932) had explained the reward-punishment asymmetry by means of the law of effect-the assumption of a special connection-strengthening property of rewards. In contrast, the interpretation suggested by the selfgeneration effect looks less to the effects of the reward than to the (covert) events that precede it. When a response is rewarded, the learner’s task requires subsequent reuse of the structures that generated the (rewarded) response. It may therefore be hypothesized that the virtue of reward rests on the fact thut the educator who wishes to rely on reward is obliged to design training situations so as to invoke response-production mechanisms that will later be used in reproducing the desired response, Not all training situations that result in the learner’s producing a correct response will capitalize on the value of self-generation. Jacoby (1978), for example, found that problem solutions are better retained when the 4The task of proofreading these pages suggested that authors’ proneness to miss errors in their own text can provide another illustration of the self-generation effect. In this case the originally generated text is s o easily retrieved that it appears to be there-n the page being proofreaddven when it is not. The common suggestions to the authoriproofreader of reading the pages in reverse order or reading the text aloud may be effective in part because such techniques disrupt the use, during proofreading, of the mechanisms involved in originally generating the text.
Anthony G . Greenwald
218
learner produces the correct response by working through to the solution rather than retrieving it from memory. Also consistent with the value of self-generation in education are findings that show self-generation of incorrect responses to interfere with acquisition of correct responses. This has been shown particularly clearly in studies that yoke an observational learner with a task performer. After an error, the performer (who generated the error) has more difficulty performing correctly on a subsequent trial than does the observer (e.g., d’Ydewalle, 1979; Marx & Witter, 1 972). 5.
Towcrrd Interpretation
The essence of the self-generation effect, in the several results that have been reviewed, is that material actively produced by the subject has a privileged place in retrieval, when compared with material passively received. The hypothesis that the self-generation procedure serves to guarantee the availability of a mechanism that must be used in eventual retrieval is compelling. In agreement with Bobrow and Bower (1969), however, this explanation must be judged unsatisfyingly vague as long as the mechanism shared by acquisition and retrieval is left unspecified. Hypothesis: The shared mechanism that produces the self-generation effect is the self system. Critique: This hypothesis, too, is vague-as long as the self system is not well defined. Replv: Let us proceed to work toward a more precise conception of the self system. B.
EGOCENTRIC
PEKSPECTIVE FACILITATES RECALL
We remember the past egocentrically or autobiographically, recalling events in terms of our relation to them. However intuitively plausible this egocentricity of memory may appear, it is not a necessary truth. It is possible, for example, to conceive of an organization of past experience that is more like that of some reference work, such as a history text or the index of a thesaurus. [Tulving (1972) characterized semantic memory as “a mental thesaurus. ”1 The importance of the egocentric character of memory has become apparent particularly with the recent conduct of several imaginatively devised experiments that have compared memories for material acquired with and without an egocentric perspective.
I.
Enhcrncrd Recall qf Own Contributions to Group Performance
In an experiment by M. W. Brenner (1973), a group of subjects sat around a large table and, in turn, read aloud words that had been prepared
Self and Memory
219
on index cards that they turned over just before reading them. Subjects expected to be tested for memory of all words read by those around the table. A powerful effect found by Brenner was that subjects remembered best the words they themselves read to the group; a lesser effect was subjects’ reduced recall of words read by just-preceding and justfollowing persons (relative to the average of all words read by others). In studies using both natural and experimental groups, Ross and Sicoly (1979) found that people recall their own contributions to a group effort more readily than they recall the contributions of other group members. For example, married subjects recalled more instances of their own performance of household chores than of their spouses’, and subjects in two-person laboratory groups remembered more of their own statements than of their partner’s from the previous day’s group discussion. Ross and Sicoly suggested that their subjects might have attended more to their own efforts than to their partner ’s-a plausible observation that, it should be noted, might apply equally to Brenner’s finding. But we should not dismiss the effect of egocentric perspective on recall as reflecting no more than routine variations in attention. Consider the related selfreference effect that was found by Rogers et al. (1977) and others. In the absence of evidence to the contrary, one might guess that subjects attend more to self-referent judgments than to other judgments. But there is evidence to the contrary. Self-referent judgments take less time, on the average, than do other types of judgments (Keenan & Baillet, 1980; Kuiper & Rogers, 1979). We should therefore be reluctant to interpret Brenner’s and Ross and Sicoly’s findings in terms of variations in quantity of attention, until there is evidence that supports this hypothesized mediator. 2 . Empathy and Memory-Extension Perspective to Others
of the Egocentric
In an interesting variant of his 1973 study, M. W. Brenner (1976) had subjects arrive for the experiment in dating pairs and seated members of each pair at opposite positions around the table. His results indicated that the two effects of egocentric perspective-enhanced real1 of own-read words and reduced recall of words read just before and just after one’s performance-occurred also (albeit in reduced magnitude) for memory of words read by the dating partner and the persons just preceding and following the partner (cf. Keenan & Lindauer, 1981). In the studies following up on the original Rogers et al. (1977) report of the facilitating effect of a self-reference orienting task on recall, similar facilitation was found to occur for orienting tasks that involved encoding in relation to familiar others (Bower & Gilligan, 1979; Keenan & Baillet, 1980; Kuiper
Anthony G. Greenwald
220
& Rogers, 1979; Lord, 1980). Keenan and Baillet’s finding of a graded reduction in the recall-facilitation effect as the referent person was decreasingly familiar is particularly consistent with the suggestion that memory is facilitated by empathic extension of the egocentric perspective. Perhaps the fullest demonstration of the consequences for memory of extending the egocentric perspective to others is in a story-recall experiment by Owens, Dafoe, and Bower (1977). By varying the content of a 300-word introduction to a 1200-word story involving two men and a woman, Owens et u1. succeeded in inducing their subjects to empathize with one or the other of the two men. The effect of this empathy was apparent in imagined physical perspective (subjects tended to imagine the scenes of the story from the position of the character with whom they identified), in interpretation of ambiguous events of the story (subjects saw “their” character as less responsible than the other male for various mishaps that occurred), and in recognition memory for story content (subjects tended to give false positive recognition responses to statements that agreed with their induced perspective). C.
EGO-INVOLVEMENT: PERSISTING TASKS FACILITATE RECALL
I.
Dejinition of Ego-Involvement
In a recent paper (Greenwald, 1981a) 1 have tried to establish a consistent definition of ego-involvement as the person’s engagement in u persisting tusk. (As I noted in developing this definition, any discussion of the consequences of ego-involvement must first deal with the existence of several mutually contradictory usages of ego-involvement.) A persisting task can be distinguished from a recurring task. This is the distinction, for example, between working toward a bachelor’s degree (a task that persists for several years) and going to class (which recurs daily for several years). Persisting tasks can be located at a higher hierarchical level in a task structure than related recurring tasks, as in the example just given. Similarly, for example, the persisting task of building a house hierarchically subsumes the recurring task of hammering a nail. Persistence occurs in degrees, such that (for example) the task of getting a good grade in a specific course is intermediate in persistence between obtaining an undergraduate degree and going to class. As can be seen from the examples just given, persisting tasks can often be described as more irnportunt than the tasks that they subsume-that is, more important than the less persisting tasks that occur at lower hierarchical levels in a task
Self and Memory
22 1
structure. Thus, tasks are ego-involving (by definition) to the extent that they are persisting, and this will often mean also that they are both highly placed in a hierarchical task structure and important.s 2 . Irrelevance of Certain Procedures to Assessing the Effect of Ego-Involvement on Memovy
Some important previous reviews (esp. Osgood, 1953, pp. 571-587; Rapaport, 1942/1971) have based evaluations of the effects of egoinvolvement on memory on studies using procedures that are excluded by the present definition of ego-involvement. For example, on the assumption that a task that is temporarily important to the subject is egoinvolving, Osgood included in his review a study (Biel & Force, 1943) that compared memory for intentionally learned material with that for material incidentally learned to the same criterion. An unanticipated retention test after a 19-day delay showed no difference in memory, suggesting no effect of “ego-involvement on memory. However, although the intentional learning instructions enhanced the importance of the material at the time of initial acquisition, there was no persisting task in regard to this material during the retention interval; therefore Biel and Force’s intentional learning condition doesn’t meet the present conceptual criterion for ego-involvement. Also included in Osgood’s review was Levine and Murphy’s (1943) study of learning and memory as a function of attitudinal agreement or disagreement with the material to be learned. Levine and Murphy found that procommunist and anticommunist material was learned faster and forgotten more slowly by subjects for whom the material was politically agreeable than by ones for whom it was disagreeable. Again, the procedure does not meet the present conceptual criterion of ego-involvement, because the subjects were given no persisting task to which the agreeable material was more relevant than was the disagreeable materiaL6 Both Osgood and Rapaport also included in their reviews studies on relative memory for pleasant and unpleasant experiences, with ”
‘The persisting-task definition of ego-involvement fits well with Koffka’s analysis (see Fig. I ) , in which ego functions in memory by means of motivational links between temporally separated strata in the trace column. Koffka’s analysis straightforwardly produces the expectation that egoinvolvement (task persistence) should facilitate long-term retrieval. Yt is conceivable that some aspects of Levine and Murphy’s materials did make the agreeable information more useful in regard to some (unspecified) persisting task. However, there is no empirical indication that, in general, attitudinally agreeable information is more relevant to persisting tasks than is disagreeable information. There are many results showing, to the contrary, that subjects often find disagreeable information as useful as, and sometimes more useful than, agreeable information [see Wicklund and Brehm‘s (1976) discussion of the lack of support for cognitive dissonance theory’s hypothesis of selective self-exposure to agreeable information].
222
Anthony G . Greenwald
the former assumed to be more ego-involving. Again, this is an empirical variation that doesn’t fit with the persisting-task conception of egoinvolvement. If one assumes (as seems plausible) that all affective experiences are generally more relevant to persisting tasks than are nonaffective ones, then the findings of many of this last group of studies (that both pleasant and unpleasant experiences tend to be retained better than neutral ones) can be taken in support of a role of ego-involvement in facilitating memory. 3. Evidence That Does Bear on the Persisiing-Task Dejinition of Ego-Involvement
The question to which we seek an answer is: Is information better retained when it has future usefulness (relevance to a persisting task) than when it does not? Remarkably, and despite the obvious importance of this question, few texts on human memory report any evidence relevant to it. Aall (1913) was apparently the first to demonstrate that students remembered more studied material, several weeks after an exam, if they had been led to believe that the material would continue to be useful after the exam. Parallel results have been obtained, in recent experiments involving much shorter retention intervals, by Jacoby, Bartz, and Evans (1978, Experiment 2) and by d’Ydewalle, Degryse, and DeCorte (1981), the latter of whom also provided a review of research relevant to this phenomenon. The well-known results of Zeigarnik (1927, 1938) also bear on the question of task persistence and memory. Zeigarnik assumed that, when a task in which the subject was absorbed was interrupted prior to its completion, the task thereby acquired a persisting character. The “Zeigarnik effect” of better recall for such interrupted tasks than for corresponding completed ones is, with this assumption, consistent with the proposition that material associated with persisting tasks is better remembered. Nuttin (1953; available in translation in Nuttin & Greenwald, 1968) introduced the distinction between open and closed tasks (see also Nuttin, 1976). In closed tasks, subjects believe that the initial encounter with task stimuli is also the last; open tasks are ones for which the subject expects that the same (or related) stimuli will later be reencountered. An example of a closed task is a series of (say, psychophysical) judgments of once-presented stimuli. The same procedures can become part of an open task if the subject expects the judgment stimuli to be presented once again.’ Nuttin’s conception of the open task has been perhaps the most ’It may be helpful to point out the difference between an open task and an intentional learning task. In the interval between study and test, an intentional learning task i s an open task, but it becomes a closed task as soon as the last test trial is completed-that is, when the subject expects to have no
Self and Memory
223
influential force in suggesting the persisting-task definition of egoinvolvement. Nuttin regarded the effect of persisting tasks on learning as the essential mechansim of human learning. The persisting task tension explanation allows substantial reconciliation among seemingly disparate results obtained by the Lewinian school and stimulus-response theorists. The fact that interrupted or unsuccessful responses are frequently found to be better recalled than completed or rewarded ones (in the former case [viz., Zeigarnik]) can be reconciled with the fact that completed or rewarded responses are found more likely to be repeated (in the latter) when the typical experiments of each school are analyzed in terms of the persisting task orientation of the learner. In experiments of the Lewinian school, it is the interrupted or failed task that is generally of greatest future significance while, in those of the S-R school, it is generally the rewarded response that guides the learner to future successful performance. (Nuttin & Greenwald, 1968, p. 102)
Nuttin’s repeated findings of superior retention of information acquired in the context of open tasks (Nuttin & Greenwald, 1968, Chap. 6) have received further corroboration in subsequent investigations of ‘‘directed forgetting” (reviewed in Bjork, 1972; Epstein, 1972). In the directedforgetting experiments, items (words or paired associates) are designated as to-be-remembered or as to-be-forgotten, usually by means of a cue that is presented together with or after the item. This procedure directly maps onto Nuttin’s open-closed distinction, in that an open task is created for the to-be-remembered items, and a closed task for the to-be-forgotten ones. Among the findings from the directed-forgetting procedure is a clear superiority of memory for the to-be-remembered items in comparison with the to-be-forgotten ones (as can be determined when the experimenter gives an unexpected recall test for the latter). In summary, the proposition that persisting tasks facilitate retention has received a wide variety of empirical support (see additional discussion in Nuttin, 1976).
IV. A.
Theoretical Synthesis
THREES E L F / M E M O R Y EFFECTS
The following generalizations summarize the results just reviewed. 1 . Material that is actively generated by the learner is more easily recalled than is material passively received (the self-generution effect). 2. Material that is encoded with reference to self is more easily recalled further use for the studied information. Also, as Nuttin pointed out, an intentional learning task is often a mixture of open and closed tasks. The subject usually has an open task in regard to correct or rewarded responses, and a closed task in regard to incorrect responses.
224
Anthony G. Greenwald
than is material otherwise encoded (the egocentric perspective or sewreferetzce effect). 3 . Material associated with a persisting task is more easily recalled than is material associated with a completed task (the ego-involvement effect). The aim of the remainder of this report is to use these findings to build on the body of existing theory (already reviewed in Sections I and 11) concerning the role of self in memory. The major conclusions of this effort will be that ( a ) the three self/memory effects have a common underlying explanation, and ( h ) this explanation has broad implications for the study of organization in memory. B.
P K O P ~ K T IOF E STHE SELFSYSTEM
Let us first review the interpretive principles that have been most frequently suggested in existing accounts of the three self/memory effects. The principle most often appealed to has been depth (or eluhorution) ofprocessing, which has been used in several discussions of both the self-generation and egocentric perspective effects; a principle of enhanced or selective utterzriorz has also been used to interpret some egocentric perspective effects; and the gestalt psychological principle of tusk tension has been used to explain the ego-involvement effect. Although these explanations possess some intuitive appeal, they leave important questions unanswered. For example: What cognitive operations are implied by the principles of enhanced attention or task tension‘? How might such operations differ from those iniplied by the principle of depth (or elaboration) of processing? How can the depth (elaboration) explanation account for the rapid processing of self-referent judgments, or for the fact that self-referent processing enhances memory only for affect-involving judgments? There have been attempts to answer some of these questions (esp. by Bower & Gilligan, 1979; Keenan & Baillet, 1980; Rogers, 1981), but these have not sought to address the complete set of self/ memory effects. The following description of a self system characterizes a set of cognitive mechanisms that is proposed to be able to generate the full range of self/memory effects. Consider a cognitive organization-which we shall call a self system-that has the Properties S 1-S3: S 1 . Self-activation: The organization’s availability is not dependent on external stimulation. S2. Ordered uccess: The organization’s activity can be characterized as an ordered search (or activation) of its components; the order is a consequence of the organization’s structure.
Self and Memory
225
S3. Self-environment interaction; The pattern of activity within the organization is determined by the interaction of its (relatively fixed) structure and the (relatively variable) structure of environmental input; further, the organization’s structure is modified by its activity. Properties S2 and S3 are found in a variety of psychological theories. Property S3 (self-environment interaction), for example, is central to both Bartlett’s and Koffka’s conceptions of memory. This property, along with S2 (ordered access), can be found in most contemporary interpretations of memory that employ a mechanism of search through a semantic network. The self-activation property ( S l ) is not found as an explicit assumption of existing theories of memory, but this property is critically necessary in order for the self system to have the capability of explaining the self/memory effects. If there is a concrete inspiration for the crucial self-activation property, it is the bootstrapping or self-loading feature of many computer systems, by which a relatively short, immediately accessible, and invariant program is used to read into working memory a much larger, less readily accessible, and modifiable program (the operating system). In existing computers, this self-loading feature is usually under an operator’s control, making it not strictly self-activating. However, the removal of this feature from external control is readily possible in principle, and, as I have argued elsewhere (Greenwald, 1981b), this removal portends a substantial change in the power balance between computer operating systems and their human operators. OF T H E SEI.F/MEMORY EFFECTS I N TERMS c . EXPLANATION OF T H E S E L F SYSTEM
In order to show how Properties Sl-S3 can be used to account for the self/memory effects, it is convenient to use, for illustration, a more tangible system that has a parallel set of properties. Properties D1 and D2 describe a desk system.
D l . Activation: The contents of the desk are always available when wanted. : contents (items) of the desk are more D2. Ordered u ~ ~ c e s sSome accessible than others, by virtue of privileged location (for example, desk top, front of center drawer). Properties D1 and D2 parallel S 1 and S2 of the self system. (A parallel to S3 is not needed for present use of the analog system.) To explain the self-generation, egocentric perspective, and ego-involvement effects in
Anthony G . Greenwald
226
memory, we shall consider parallel effects that might be obtained in analog experiments using the desk system. The success of this explanation of the self/memory effects is to be measured by the plausibility of the parallel effects hypothesized for the desk system experiments. (The desk system is convenient as an illustration because its contents are more publicly accessible than are those of the self system and, therefore, their role in explaining the desired effects may be more easily appreciated.) I . Self-Generution
The procedural feature shared by experiments showing the selfgeneration effect is that the subject is led to construct a response to a presented stimulus, rather than having the (same or comparable) response provided by the experimenter. The fact that the critical variation is in the source of the response can obscure the possibility that the events that immediately precede the response are the ones that are critical to explanation of the effect. In the self-generate condition, but not in the control condition (according to the present hypothesis), the subject is obliged to associate the response with some easily accessible component of the self system, which then can serve as a retrieval aid at test time. Consider the analog “desk-generation’’ experiment, with yourself as subject, seated at your desk. The instructions for the desk-generate condition are: “Here is a list of words that I want you to remember. Try to form associations of the to-be-remembered words with the items in your desk as you please. ” If you are in the control condition, the experimenter places a heterogeneous collection of objects on your desk and asks you to associate these extraneous objects with the to-be-remembered words. After a fixedduration study period, the extraneous objects are removed for the control subjects, and all subjects are allowed to examine the desk contents during the subsequent recall test. Of course, the desk contents are not likely to be helpful if you are in the control condition. The analog of the selfgeneration effect that should be obtained in this experiment is due to the subject’s use, during initial encoding, of retrieval aids that will continue to be available (Property S1 or D1) during the subsequent recall test.
2.
Egoceritric Perspective
We may go directly to the desk-system analog experiment, since its explanation builds directly on that of the self-generation analog experiment, The “desk-centric perspective” condition is the sume as the deskgenerate condition of the preceding experiment. The experiment differs in selection of the control condition with which the experimental treatment is compared. In the control condition you spend the study period seated at
Self and Memory
221
the experimenter’s (or someone else’s) desk and you are instructed to use the items available therein or thereon as associative cues for the to-beremembered words. Unexpectedly, at the time of the recall test, you are moved to your own desk and permitted to use its contents as retrieval aids. Your own desk’s contents may be helpful, as a function of the degree of their overlap with the contents of the experimenter’s desk. [Recall Keenan and Baillet’s (1980) finding that recall is facilitated in an orderly way by the subject’s familiarity with the person concerning whom trait applicability judgments were being made. In the desk analog, overlap of contents functions as an analog of familiarity.]
3. Ego-Involvement The preceding two experiments can be considered to provide analogs of the ego-involvement effect-to the extent that ego-involvement is understood as degree of involvement of the self system in encoding. The experimenal treatment in the desk-system analogs of the self-generation and egocentric perspective experiments employed the learner’s desk contents in encoding, whereas the control conditions did not. In order, however, to conduct an analog experiment that is faithful to the present persisting-task conception of ego-involvement , a more subtle manipulation, which uses the ordered access (D2) property of the desk system, must be devised. Again the experimental (desk-involvement) treatment can be the same as in the preceding experiments, and only the control condition changes. In this control condition, the experimenter (without informing you of this fact) selects relatively inaccessible items from your desk and asks you to use these objects as retrieval aids. For the subsequent retrieval test these items are carefully restored to their original positions in your desk, with the usual instruction that you are free to use the items in the desk as memory aids. Now, if you (as control subject) do not detect the experimenter’s strategy for selecting items as retrieval aids, your eventual recall performance should be much inferior to that of an experimental subject who, in normal search through the desk, readily stumbles upon items that are useful in retrieval. What is it about this experiment that lets it provide a faithful rendition of the persisting-task conception of ego-involvement? It is the assumption that the easiest-to-find items in your desk will, by virtue of your natural use of the desk, be items that are of greatest persisting usefulness. 4.
The Common Principle
The argument that the three self/memory effects have a common explanation has been made by devising three analog experiments that share the
22x
Anthony G. Greenwald
same experimental treatment. The conimon principle underlying the three self/memory effects is that all the effects depend on the use, in encoding information, of a system with Properties S1 and S2. The effects, that is, depend on the fact that the instructions in the experimental treatments (if they are followed) oblige the subject to make use-and it is very likely unwitting use-of the self system in encoding information. Note that typical self/memory experiments have little ecological validity, in that the instructions used to bring the self system into the encoding process do not correspond closely to natural inducements to use the self system. I assume that natural inducements to use the self system are environmental indications that the information is associated with some persisting task (in other words, ego-involving). 5 . tncreusing the Approximation of the Anulog Experiment5 to the Se1f System
A small modification of the desk-system experiments can result in Property D2 (ordered access) becoming a much more powerful determinant of results. The change is to oblige the experimental-treatment subjects to rely on rnemory of the desk contents at the time of the recall test. With this change the subject should, optimally, mentally (rather than visually and manually) search the desk contents at study time, attempting to use the items that come first to mind as retrieval aids whenever possible. This change substitutes the concept of accessibility of the desk contents in memory for their physical accessibility in the desk. The change increases the resemblance of the desk system to the self system. Once having thus changed from the desk-in-office version of the desk experiment to its desk-in-memory variation, we can easily make further changes to replace the desk contents by any other easily remembered set of items, such as the street names in the neighborhood of one’s home or office, the names and occupations of one’s neighbors, or objects that have names that rhyme with the first ten digits. The relation of these tactics to familiar mnemonic aids, such as pegwords or the method of loci, should be apparent. A conclusion of this analysis is that effective mnemonic aids are ones that possess Properties S1 and S2-they have assured availability (self-activation) and an internal structure that produces retrieval aids in reliable order (ordered access). 6 . Function of Property S3
Property S 3 acknowledges the complexity of the self system’s interaction with its environment. In the present state of this formulation of the
Self and Memory
229
concept of a self system, Property S3 serves jointly as an expression of faith (that a system of the sort proposed has sufficient complexity to be a powerful and flexible retrieval device) and ignorance (as to the details of its operation). The faith is justifiable by pointing to the accomplishments of existing artificial intelligence systems (e.g., Anderson, 1974; Lehnert, 1978) in which large structured memories, when interrogated by variable probes, retrieve details suitable to the probe. (The ignorance can be accepted on faith.) It is interesting to compare Property S3 with Pribram’s (Pribram, Nuwer, & Baron, 1974) proposal of hologram-like recording in the brain. In making a hologram, a three-dimensional object interacts with a coherent laser beam to produce a filmed record that distributes information about any portion of the object through the entire surface of the record (the hologram). The hologram (or a part of it) can then be used, together with another laser beam, to reconstruct a three dimensional image of the photographed object. In the case of the self system, the “beam” that is projected onto some complex environmental input is itself a complex structure (the perceiver’s self system). The perceptual and memorial capabilities of two such complex interacting structures must be only mildly suggested by the hologram metaphor.
V.
Scope and Importance of the Self System
There is some danger that the analysis presented in the preceding section might succeed too well, leaving the impression that the self system is “nothing but” an effective mnemonic device. Although the organization of an effectively functioning memory may indeed be the main function of the self system, it should not be assumed that this is any minor accomplishment. A.
IMPLICIT THEORIES OF MEMORY
How many readers would agree with the following statement? All attended experience is registered in memory. Memory is impe$ect, however, because associative interference (and perhaps other processes) can obstruct retrieval of these established memories. This statement expresses a position that can be called the taken-for-granted theory of memory, according to which the real theoretical action in the study of memory concerns the process of forgetting-explaining, that is, how some memories, once established, manage to become inaccessible (see the discussion of this point in Loftus & Loftus, 1980). The influence of the taken-for-granted theory can be observed in the massive attention to inter-
Anthony G. Greenwald
230
ference theory in recent decades, as well as in the development of interest in intentional forgetting, which is well expressed in Bjork’s observation: Intentional forgetting is a frequent event in one’s everyday life; i t is probably, in fact, more frequent than is intentional remembering. We overhear conversations, we see things in newspapers and store windows, we add up numbers, we dial phone numbers, we pay attention to advertisements, and so on-nearly all of which we have no use for beyond the point at which we attended to them. To the degree that we have any intentions at all with respect to that information, we intend to forget it rather than reinernber it. (Bjork, 1972, p. 218)
On the other hand, recent attention to the role of encoding processes in memory, as well as to the use of mnemonic strategies, are developments that move away from the taken-for-granted theory. These efforts look to the active role of the learner, at the time of initial receipt of information, in constructing the routes that can eventually permit retrieval. We might, following these developments, give serious consideration to the reverse of the taken-for-granted theory of memory, which might be called the forgetting-for-granted theory. According to this alternative implicit theory of memory, the brain, at birth, has essentially no capacity for retention-no ability to relate new input to previous input. The important theoretical action, by this theory, is in explaining the origin and functioning of the capacity to perform ordinary acts of recognition and recall, in explaining how these acts become so routine that we begin to wonder how things are ever forgotten. It is from the vantage of the forgetting-for-granted theory that the mnemonic accomplishments of the self system are awesome. In this view, the self system functions as a trap that selectively latches onto potentially important (ego-involving) aspects of experience in a way that permits the effect of these experiences to be cumulative. But (it may be argued) I appear to be claiming for the self system precisely the function that is usually attributed to episodic memory (Tulving, 1972) in the contemporary view. Nothing could be closer to the truth. Indeed, I attempted to lay the foundations for drawing the connection between the self system and episodic memory by focusing (in Section I) on those aspects of Claparede’s and Koffka’s theories that attributed to the self (ego) system the properties of episodic memory. B.
THESELFSYSTEM AND
THE
EPISODIC-SEMANTIC
DISTINCTION
Claparede distinguished between egocentric and marginal systems of memory, crediting egocentric memory with the capacity (apparently lacking in his Korsakoff patient) of ordinary, voluntary memory for personal
Self and Memory
231
experiences. Similarly, Koffka distinguished between an ego system and an environmental system, suggesting that the former organized experience in terms of interests and also carried the temporal structure of remembered experience. I suggested earlier (Section I,B) that Claperede’s egocentric-marginal distinction corresponds to Tulving’s episodicsemantic distinction.* However, Koffka’s distinction is a different one. In Koffka’s conception, environmental traces were conceived as peripheral to the self (ego) system, whereas semantic memory may better be conceived as a foundation on which the self system resides, a foundation that has itself been built by the self system and that continues to grow through the activities of the self system. This conception of the episodic-semantic relation, it should be noted, is consistent with the earlier (Section V,A) suggestion that the self system retains potentially important experiences, thereby allowing them to have cumulative impact. To amplify, when any pattern of events is frequently repeated in experience, the cumulation of impact can be assumed to render the relationship among the components of the pattern more automatic-which is to say, more accessible without an active attentional process (see LaBerge, 1974). Such preattentive processing for meaning is not one of the properties of semantic memory that were stressed by Tulving (1972), but neither is it incompatible with Tulving’s discussion. (Tulving addressed the functioning of semantic memory as an object of attention, rather than as the substructure of attention.) Tulving ’s conception of the episodic-semantic distinction can be viewed as suggesting a (semantic) substructure of memory that is in some sense even more central or fundamental than the self (episodic) system. The suggestion that the semantic system is, in effect, built by and from the episodic (self) system is also consistent with recent arguments that episodic and semantic memory are not fundamentally different (e.g., Anderson & Ross, 1980). C. THEORETICAL STATUS OF
THE
SELF
In discussing the properties of the self system, I have so far avoided taking a position on the issue that divided Bartlett from Koffka, and that threatens to polarize contemporary researchers-whether or not such a thing as the self exists. I implied earlier that the issue was in part a pseudo-issue, since all the researchers and theorists whose work has been *In Tulving’s usage, episodic memory consists of events recalled in terms of the time and place of original experiences; semantic memory consists of knowledge of the interrelationships among events and facts, abstracted from the experiences on which this knowledge is based.
232
Anthony G. Greenwald
reviewed seem to agree that the memory phenomena associated with the self require the assumption of a great degree of organization in memory. They differ only in preference for using a special label to designate this organization, and in their conclusions as to whether or not this organization operates according to unique principles. The functions of the self as a memory system are, in large part, amenable to explanation by applying familiar principles of memory to a “rich” (Keenan & Baillet, 1980) or “well-differentiated” (Bower & Gilligan, 1979) conceptual structure. Nevertheless, it does not seem justified to conclude that this particular rich and well-differentiated structure has no special (emergent) properties or that no special designation (self or ego) is justified. Among the apparent emergent properties of the self system are ( a ) the self-activation property described in Section IV,B, (6) the tendency for memory to be biased toward retrieving favorable information about oneself (Greenwald, 1980), and ( c ) the privileged treatment accorded to certain classes of encountered information, such as information that is relevant to persisting tasks or to self-evaluation. None of the preceding discussion justifies a conclusion in favor of the conception of the self as a transcendental entity that receives and possesses knowledge. Of course, those who wish to reject the transcendental view should feel some obligation to explain one particularly intriguing “emergent” property of the self system-its tendency (in the normal case) to perceive itself as unitary and real. D.
UNFINISHED BUSINESS
What are the early experiences that are critical to initial establishment of the self system (apparently within the first two years of life-see Gallup, 1977; Lewis & Brooks-Gunn, 1979)? How does the self as a memory system interrelate with the self as a judgment system that is predisposed to cognitive biases (see Greenwald, 1980)? Is it possible to interpret high-level functional disorders of memory, such as amnesias and multiple personalities, as consequences of exotic or damaged organization of the self system (see Kihlstrom & Evans, 1979)? Can the effects on memory of drugs such as alcohol and barbiturates be interpreted as a suppression of the normal function of the self system (see Hull, 1981)? What are the important dimensions of individual differences in structure of the self system? My treatment of self and memory has stayed away from such complex questions. I can justify this aversion in terms of a conviction that the focus on basic issues concerning self and memory is a necessary preliminary to treatment of the these deeper questions. That is, interpretation of the self
Self and Memory
233
as a memory system seems a promising starting point for dealing with phenomena involving emotion, personality, and pathology. It is fascinating to consider that the goal of using memory as an entering wedge to these important problems returns us, albeit facing in the other direction, to the starting point of this chapter-Freud's and Claparede's use of psychopathological phenomena as an entering wedge to the study of human memory.
ACKNOWLEDGMENTS Work on this chapter was aided substantially by grants MH 31762 and MH 32317 from the National Institute of Mental Health. I thank several colleagues who kindly commented on an earlier drafl-Steven J. Breckler, Gery d'Ydewalle, Janice M. Keenan, Nicholas A. Kuiper. Elizabeth F. Loftus, Charles G . Lord, Joseph R. Nuttin, and Anthony R. Pratkanis. Even though I have been unable to resolve all the issues they raised, the final version of this article owes much to the stimulating comnients they provided.
REFERENCES Aall, A . Ein neues Gedichtnisgesetz'! Experimentelle Untersuchung iiber die Bedeutung der Reproduktionsperspektive. Zeirschrijt ,fur Psychologic,. 19 13, 66, 1-50, Allport, G. W. Puttern cmrl firmvtb i n per.sonuliry. New York: Holt, 1965 Anderson, J . R. Retrieval of propositional information from long-term memory. Cognitive Pvychology, 1974, 5, 451-414. Anderson, J . R . , & Bower, C . H. Hurnun ussocirhv rrre.rnor,v. Washington, D.C.: Winston, 1973. Anderson, J . R . , & Ross, B. H . Evidence against a sernantic-episodic distinction. Journal of' Esperirnentul Ps\vhology: Hutnun Leurnirrg and Memory, 1980, 6, 441 -466. Bartlett, F. C. Remembering: A stu& in esperimenial and social psychology. London: Cambridge University, 1932. B i d , W. C . , & Force, R. C. Retention of nonsense syllables in intentional and incidental learning. Jourtiul of E.x/wriinentul P syhology , 1943, 32, 52-63. Bjork, R. A. Theoretical implications of directed forgetting. In A . W . Melton & E . Martin (Eds.), Coilirig processes in humun ~ n r n ~ o Washington, ~v. D.C.: Winston, 1972. Bobrow, S . A . , & Bower, G . H. Comprehension and recall of sentences. Journal ofEsperirnenral Psychology, 1969, 80, 455-461 Bower, G . H. Einofionul i n W d und rnernorv. Address presented at the meeting of the American Psychological Association. Montreal, September 1980. Bower, G . H . . & Gilligan, S. G . Remembering information related to one's self. Journal of Re.seurc,h in Persoriulity. 1979, 13, 420-432. Brenner, C . The nature and development of the concept of repression in Freud's writings. P.sychounulvtic Stuify of the Child, 1957, 12, 19-46. Brenner, M . W. The next-in-line effect. Journul of Verbal Learning and VerhuI Behavior. 1973.12, 320-323. Brenner, M . W. Memory und interpersonul relations. Unpublished doctoral dissertation, University of Michigan, 1976.
Anthony G. Greenwald
234
Claparcde, E. Recognition and “me-ness. ” In D. Rapaport (Ed.), Orgunizurion and pathology U f thought. New York: Columbia University Press. 1951. (Original French publication, 191 I . ) Clark, M., & Isen, A . M . Feeling states and social behavior. In A. Hastorf & A. M. lsen (Eds.), CuRnitive sotial psyhulogy. New York: Elsevier/North-Holland, 198 1 , Cohen, N. J., & Squire, L. R . Preserved learning and retention of pattern-analyzing skill in amnesia: Dissociation of knowing how and knowing that. Science, 1980. 210, 207-209. Craik, F. I. M.. & Lockhart. R. S . Levels of processing: A framework for memory research. Journul id Verhtrl LeurriinR und Verhal Behavior, 1972, 11, 671-684. Craik, F. I. M . , & Tulving, E. Depth of processing and the retention of words in episodic memory. J f J U I ‘ n i d yf’ &periffientd Psycholugy: G e n e r d , 1975, 104, 268-294. d’Ydewalle, G. Performing versus observing ”right” and “wrong” responses in incidental and intentional learning. Atneric,un Jourtral of Psychology. 1979, 92, 427-436. d’Yitlewalle, G . , Degryse, M . , & DeCorte, E. Expected time of test and the acquisition of knowledge, British Jourwul of Educuttonal Psychology, 1981, 51, in press. Epstein, W. Mechanisms of directed forgetting. The Psyrhology c!fLeurning and Motivmtion, 1972, 6 , 147-191. Erdelyi, M . , Buschke, H., & Finkelstein, S. Hypermnesia for Socratic stimuli: The growth of recall for an internally generated memory list abstracted from a series of riddles. Memory & Cognition, 1977, 5 , 283-286. Erdelyi, M. J., & Goldberg, B. Let’s not sweep repression under the rug: Toward a cognitive psychology of repression. In J. F. Kihlstrom & F. J . Evans (Eds.), Functional disorders of m m o r y . Hillsdale, N.J.: Lawrence Erlbaum, 1979. Freud, S . The psychopathology oJrvervday life. In A . A. Brill (Ed. and trans.) New York: MacmilIan, 1917. (Original German publication, 1901.) Freud, S . Rcprcsaion. In J. Strachey (Ed.), The srundard edition of the complete psyrholoigitcil works ofSigmutzd Freud (Vol. 14). London: Hogarth Press, 1957. (Originally published, 1915.) Freud, S . The ego and the id. In J . Strachey (Ed.), The .staridart/ edition y f t h e corttplefeps~vcholugical works of Sigrnuntl Freud (Vol. 19). London: Hogarth Press, 1961. (Originally published, 1923.) Gallup, G . G. Sell-recognition in primates: A comparative approach to the bidirectional properties of consciousness. Aiirerirun Psychologist, 1977, 32, 329-338. Greenwald. A. G . Cognitive learning, cognitive response to persuasion and attitude change. In A. G. Greenwald, T. C. Brock, & T . M. Ostrorn (Eds.), Psyrhologicnlfountlationsof’artirudes. New York: Academic Press, 1968. Greenwald, A. G . Difficulty of associative performance following training with negative instances: A note on punishment effects. Journul of E~/ucuriomlPsychology. 1970, 61, 255-259. Greenwald, A. G . The totalitarian ego: Fabrication and revision of personal history. America,? Psvchulogist, 1980, 35, 603-618. Greenwald, A. G . Ego task analysis: An integration of research on ego-involvement and selfawareness. In A. Hastorf & A. Isen (Eds.), Cognitive soriul psychologv. New York: Elsevieri North-Holland. 1981. (a) Greenwald, A. G. Environmental stmcture and cognitive stmcture. In J . H. Harvey (Ed.), Cognirion, soi.iu/ hehtrvior, c u i d the frivirorirnent. Hillsdale, N.J.: Lawrence Erlbaum. 1981 . (b) Greenwald. A. G . , 1L Albert. R . D. Acceptance and recall of improvised arguments. Journal of P e r w t i u l i t ~um/ Socid Psvchology, 1968, 8, 3 1-34. Hastie, R. Schematic principles i n hurnan memory. In E. T. Higgins, C. P. Herman, and M . P. Zanna (Eds.), Socitrl cognition: The Ontario Symnposiu,n. Hillsdale, N.J.: Lawrence Erlbaum, 1980.
Head, H. Studies r)r rreurology. London: Oxford University Press, 1920. Holnies. D. S . Investigations of repression: Differential recall of material experimentally or naturally associated with ego threat. Psdzological Bullefirr. 1974, 81, 632-653.
Self and Memory
235
Hovland, C. I . , Janis, I . L., 81 Kelley, H. H. Communication andpersuasion. New Haven, Conn.: Yale University Press, 1953. Hull, J. G. Alcohol and self-awareness: A theoretiml integration. Unpublished manuscript, Indiana University, 1981. Jacoby, L. L. On interpreting the effects of repetition: Solving a problem versus remembering a solution. Journal of Verbal Learning and Verbul Behuvior. 1978, 17, 649-667. Jacoby, L. L., Bartz, W. H., & Evans, J . D. A functional approach to levels of processing, Journal of Experimental Ps-ychology: Human Learning and Memory, 1978, 4, 331 -346. James, W. Principles of psychology (Vol. I). New York: Holt, 1890. Keenan, J . M., & Baillet, S. D. Memory for personally and socially significant events. In R . S. Nickerson (Ed.), Attention and performance V f f f . Hillsdale, N.J.: Lawrence Erlbaum, 1980. Keenan, J. M., & Lindauer, B. K. The next-in-line effect: A methodfor studying social development and irs effrect.s on memory. Unpublished manuscript, University of Denver, 1981. Kihlstrom, J . F., & Evans, F . J . (Eds.) Functional disorders ofmemory. Hillsdale, N.J.: Lawrence Erlbaum. 1979. Koffka, K. Principles of gestalt psychology. New York: Harcourt, 1935. Kuiper, N . A., & Rogers, T. B. Encoding of personal information: Self-other differences. Journal of Personality and Social Psychology, 1979, 37, 499-5 12. LaBerge, D. Acquisition of automatic processing in perceptual and associative learning. In P. M. A. Rabbitt & S. Dornic (Eds.), Attention and performance V . New York: Academic Press, 1974. Lehnert, W. The process of question answering. Hillsdale, N.J.: Lawrence Erlbaum, 1978. Levine, J . M., & Murphy, G. The learning and forgetting of controversial material. Journal of Abnormal anti Social Psychology. 1943, 38, 507-517. Lewis, M . , & Brooks-Gunn, J . Sociul cognition and the cirquisition of self. New York: Plenum, 1979. Loftus, E. F., & Loftus, G. R. On the permanence of stored information in the human brain, Americun Pswholngist, 1980, 35, 409-420. Lord, C. G . Schemas and images as memory aids: Two modes of processing social information. Journal of Personality and Social Psychology, 1980, 38, 251-269. Markus, H. Self-schemata and processing information about the self. Journal of Personaliry and Social Psychology. 1977, 35, 63-78. Markus, H . The self i n thought and memory. In D. M. Wegner & R. R. Vallacher (Eds.), The selfin social psychology. London and New York: Oxford University Press, 1980. Marx, M. H . , & Witter, D. W. Repetition of correct responses and errors as a function of performance with reward or information. Journul of Experimental Psychology, 1972, 92, 53-58. Neisser, U. Cognition and reality. San Francisco: Freeman, 1976. Nuttin, J . R . Tiche, reussite. et Pchec. Louvain: Publications Universitaires, 1953. Nuttin, J . R . Motivation and reward in human learning: A cognitive approach. In W. K . Estes (Ed,), Handbook ofleurning and cognitive processes (Vol. 3). New York: Wiley, 1976. Nuttin, J . R . , & Greenwald, A . G. Reward and punishment in human learning. New York: Academic Press, 1968. Osgood, C. E. Method and theory in experirneniul psychology. London and New York: Oxford University Press, 1953. Owens, J . , Dafoe, J . , & Bower, G. H. Taking u point of view: Character identification and artrihutionul processes in story comprehension and memory. Paper presented at the meeting of the American Psychological Association, San Francisco, August 1977. Petty, R . E . , Ostrom, T. M.. & Brock, T. C. (Eds.) Cognitive responses in persuasion. Hillsdale, N.J.: Lawrence Erlbaum, 1981. Postman, L. The history and present status of the Law of Effect. Psychological Bulletin, 1947, 44, 489-563.
236
Anthony G . Greenwald
Postnian, L. Rewards and punishments in human learning. In L. Postnian (Ed.), Psychologv if?the rnaking. New York: Knopf, 1962. Pribram, K . H . , Nuwer, M., & Baron, R. J The holographic hypothesis of memory structure in brain function and perception. In D. H. Krantz, R . C . Atkinson, R . D. Luce, & P. Suppes (Eds.), Corirrrnporary dcvulopment.v in ~ncrthemtr/ictrlpsvcholugy (Vol. 2 ) . San Francisco: Freeman, 1974. Rapaport, D. En?otion.r and mernon. New York: International Universities Press, 1942. (Republished, 1971.) Rogers, T . B. An analysis of two central stages underlying responding to personality itenis: The self-referent decision and response selection. Juunitrl uf' Resectrch iri Per.soriahty, 1974, 8, 128-1 38. Rogers, T. B. A model of the self as an aspect of the human information processing system. In N. Cantor & J . F. Kihlstrom (Eds.), Cogriition, socicil iriterucrion andpersortaliry. Hillsdale, N.J.: Erlbaum, 1981, in press. Rogers, 'I. B . , Kuiper. N . A , , & Kirker, W . S. Self-reference and the encoding of personal inforination. Jourrictl uf Personality and Sorial Psychology, 1977. 35, 677-688. Ross, M., & Sicoly. P. Egocentric biases i n availability and attribution. Journal of Personality rind Sociiil Psvchology, 1979, 37, 322-336. Skinner, B . F. Sciericr and human hrhavior. New York: Macmillan, 1953. Slamecka, N . J . , & Graf. P. The generation effect: Delineation of a phenomenon. Jourritrl qf Experimeniul Psychology: Hurnun Learning and Merrrory. 1978, 4, 592-604. Thomdike, E. L. The fundumentcrls of' learning. New York: Teachers College, 1932. Tulving, E. Episodic and semantic memory. In E. Tulving & W Dunaldsun (Eds.), Orgunizurion of memory. New York: Academic Press, 1972. Tulving, E., & Donaldson. W. Urgariirutioit (fmeirrory. New York: Academic Press, 1972. Wallach, H., & Henle, M. An experimental analysis of the law of effect. Journal oj E.rprrirnentcd Psvchology, 1941, 28, 340-349. Wicklund. R . A . , & Rrehni, J . W . Perspectives on cognitib'e rlrssonarice. Hillsdale, N . J . : Lawrence Erlbaum, 1976. Zeigarnik, B . Uber das Behalten von erledigten und uncrledigtcn Handlungen. Ps~chulugischu Forsrhung, 1927, 9, 1-85. Zeigarnik, B. On finished and unfinished tasks. In W. D. Ellis (Ed.), A Juurce hook uf Gestult psyhology. New York: Humanities Press, 1938.
CHILDREN'S KNOWLEDGE OF EVENTS: A CAUSAL ANALYSIS OF STORY STRUCTURE' Tom Trabasso and Nancy L. Stein THE UNIVERSITY OF CHICAGO CHICAGO. ILLINOIS
Lucie R . Johnson BETHEL COLLEGE ST. PAUL. MINNESOTA
I . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding Stories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. Causal Relationships. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Study I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111. Causal Understanding in Preschool Children . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Study 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IV. Logical Justification: The Meaning of Why and Because . . . . . . . . . . . . . . . . . . . . . . Study 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V . General Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
237 238 249 252 260 270 274 275 277 279
I. Introduction In the last few years, cognitive psychologists have been interested in examining the process by which we understand and use information from written and oral discourse. In particular, investigators in this area have examined how events in a story are understood and remembered. Developmental psychologists have also expressed a deep interest in stories, not so much from the processing point of view, but more from a concern about the age at which children acquire skills necessary for the attainment of a skilled level of literacy and social-personal knowledge. Few inves'We wish to thank Robert Siegler for his thorough reading and editorial comments on an earlier version of this manuscript. The writing and research reported in this articlewas supported by National Institute of Education grants NIE-G-77-0018 and NIE-'3-79-0125 to Tom Trabasso. In addition, the authors were supported, in part, by the National Institute of Education Contract No. US-NIE-C-40076-0116 to the Center for the Study of Reading, University of lllinois at Urbana-Champaign. 237 THE PSYCHOLOGY OF LEARNING AND MOTIVATION, VOL. 15
Copyright 0 1981 by Academic Press, Inc. All rights of reproduction in any form reserved. ISBN 0-12-543315-8
238
Tom l’rabasso et al.
tigators have integrated issues from both of these areas so that a theoretical and pedagogical statement can be made. In this article, we attempt to address each issue and then show how the combination would benefit both fields. First, we examine models of story comprehension in relation to the specific constraints assumed to influence the process of constructing a coherent representation of stories. In doing so, we investigate the type of causal knowledge acquired about stories and whether this knowledge corresponds to hypotheses derived from current models of story understanding. Second, we focus on the development of story understanding by examining whether the very young, school-aged child has the ability to understand different types of stories. We explore current research on children’s causal understanding of narrative events, and test several hypotheses regarding their knowledge of narrative sequences. At the close of the article, we attempt to integrate theoretical and development issues by discussing issues related to the acquisition of story schemata and the conditions under which learning from text occurs. UNDERSTANDING STORIES
A working assumption in current theories of story comprehension is that prior schematic knowledge of stories is used to facilitate the process of understanding and constructing a representation of any new incoming information. Without the use of previously acquired knowledge, it is believed that little memory for incoming information would exist. Schematic knowledge is thought to be important, not only for the process of encoding incoming information, but also for the retrieval and recall of knowledge already encoded, such as recalling a story one heard long ago or writing a novel story, based upon experiences from many different realms. The process by which this knowledge is retrieved is still an issue of much study and current debate. There is agreement, however, that schematic knowledge helps a reader or listener to generate expectations about what types of information should occur in a text. These expectations are thought to guide comprehension and production at all levels of information retrieval. For example, it has been argued that comprehenders have specific expectations about the types of information that should occur in stories, the types of causal relationships that should link different parts of a story, the order in which events should occur, and the specific semantic content that should occur in a given sequence (N. S. Johnson & Mandler, 1980; Mandler & Johnson, 1977; Rumelhart, 1975, 1977;
Children’s Knowledge of Events
239
Schank & Abelson, 1977; Stein, 1979; Stein & Glenn, 1979; Stein & Trabasso, 1981). During comprehension, if incoming information fails to match the expectations generated by the use of schematic knowledge, the comprehender will then attempt to use his previous knowledge of story sequences to construct a plausible representation of the disparate information. For example, if stories are heard in an order other than the observed order of occurrence, either the subjects should change the observed order to match what they expected to hear, or they should mark the inverted order of incoming story information so that they can accurately retrieve it when necessary. These transformations occur because it is thought that the comprehender is limited in terms of what can be integrated into already existing knowledge of narrative events. Similar explanations exist in describing what happens when incoming information is inconsistent with expectations about either the content or the structure of a particular portion of a story sequence. If incoming information is at all inconsistent with expectations about the class or category of information in a text, the listener or reader should attempt to elaborate upon incoming information so that a coherent representation can be constructed. For example, if a critical part of the story was deleted, the comprehender should attempt to use his schematic knowledge of story structure to make an inference about types of knowledge that would be appropriate to fill in for the deleted information. If appropriate knowledge can be retrieved, the comprehender will make the inference, which then becomes part of the current representation of incoming story events. If all the parts of a story are present, but the content of the incoming information is inconsistent with what was expected, the comprehender may attempt to make inferences about why the particular story events occurred. If appropriate reasons can be inferred, then these reasons should be included in the representation of the story. Otherwise, the comprehender may change or elaborate upon the original event and change it to one more appropriate to his state of existing knowledge. If these types of strategies are not used, the comprehender may be unable to construct or retrieve information from the original sequence. To date, many investigators have attempted to test these assumptions, but because of the complexity and multifaceted nature of the comprehension process, the more crucial studies have not been completed. Most current studies have shown that memory, as measured by story recall, is indeed affected by both the structure and the content of a text (Baker & Stein, 1981; Mandler, 1978, 1981; Stein, 1979; Stein & Trabasso, 1981; Thorndyke, 1977). However, relying upon recall alone is a rather unsatis-
240
’lorn Trabassa et al.
factory and limiting method of ascertaining whether or not current schematic approaches to story comprehension have any psychological validity. More direct assessment techniques are necessary, especially if the notion of a psychological “expectation” is to be understood in any detail. In the studies to be presented in this article, we have relied upon a generation paradigm, as well as the use of probe questions, to assess the expectations formed during comprehension. In some instances, we report recall results, but only to add supportive evidence to our major hypotheses under consideration. The primary expectation investigated in our studies concerns the generic type or the category of information listeners and readers expect to occur in story sequence. In almost every current description of story structure (Black & Bower, 1980; N. S. Johnson & Mandler, 1980; Mandler & Johnson, 1977; Rumelhart, 1975, 1977; Stein & Glenn, 1979; Thorndyke, 1977) a set of hypotheses has been formulated about the generic type of information that should occur in a story. In the Mandler and Johnson (1977) and Stein and Glenn (1979) systems, the constraints on story sequencing and the inclusion of critical parts of a story are relatively similar. The story is conceptualized in terms of a hierarchial network of story categories and the logical relations linking the categories. The categories represent the different types of generic information that occur in most stories. The logical relations connecting the categories specify the way in which information in one category is related to information in another category. We shall use the Stein and Glenn (1979) system to illustrate the expectations assumed to be operating during comprehension and generation of story sequences. A story can be divided initially into a Setting plus an Episode. The setting introduces the protagonist and can contain one or more statements about the physical, social, or temporal environment in which the remainder of the story occurs. The Setting information is thought to constrain the type of events that follow, but supposedly does not directly cause the occurrence of subsequent events. The Episode is the higher order unit of analysis and consists of a sequence of five categories. The category that begins the Episode is the Initiating Event. Information in this category consists of events or actions that function to evoke a desire in the protagonist to achieve a goal or change of state in the protagonist’s world. This desire or goal is included in the second category, labeled the Internal Response. This category not only contains the goal of the protagonist, but also contains the emotional response, which often regulates the choice of the protagonist’s goal and the cognitions or plans related to how goal-
Children’s Knowledge of Events
24 I
attainment will occur. The function of the Internal Response category is to motivate the protagonist to carry out a set of overt actions, in the service of attaining the goal. The actions are labeled Attempts, and they result in states, state changes, or actions that signify whether or not the goal has been attained. This category is labeled the Consequence. The major function of the Consequence category is that it results in a Reaction to the events in this category. Reactions include emotional reactions or actions of the protagonist or others to the goal-attainment of the protagonist, or a moral statement, often summarizing the worthiness of participating in the specific activities at hand. Clearly, there is a definite order proposed as to when each of these categories occurs in a story sequence: Setting
I
Initiating Event Internal Response
1
Attempt
I
Consequence Reaction The arrows signify either a direct causal link connecting two adjacent categories, or an enabling relationship between two categories such that one category sets up the preconditions for the occurrence of the next. From this diagram, we can see that, proceeding in a forward direction through the episode, there are specific expectations about what types of information should follow, given the occurrence of a previous category. The predictions about subsequent occurring statements, however, are somewhat more complex than the previous diagram suggests. Let us proceed through the episode, in order to make the specific predictions necessary. If the Setting category of a story is presented, and a subject is asked to
242
Tam Trabasso et al.
continue the sequence by generating the very next piece of information that should occur in the story, three responses are hypothesized to be appropriate : 1 . Setting statements, representing further elaboration of the focal cat-
egory. 2. Initiating Event statements. 3. Internal Response Statements. The second response is the modal response predicted. However, the subject may decide that not etrough setting information has been included in the focal stimulus category and proceed to elaborate upon the given events before moving to the next category in a sequence. The grammars do allow several statements to be included in a Setting category, as long as the statements represent descriptions specified by the definition of Setting statements. Internal Responses are also considered a correct response under certain conditions (e.g., the Initiating Event can be omitted). Both N. S . Johnson and Mandler ( 1 980) and Stein (1 979) discuss those conditions. Johnson and Mandler argue that lnitiating Events (or Beginnings in their terminology) may be deleted under two conditions: 1. The beginning may be deleted in the first episode of a story, where the protagonist’s internal response is not linked to any observable specific external event. They cite Propp ( I 928/1968), who gives examples of stories beginning with the perception of a Lack, whereupon the protagonist sets out to achieve a desired goal. 2. The beginning may be deleted when the content of such a category would be the perception or appraisal by the protagonist concerning events that have occurred in previous episodes.
The explanation for these deletions rests upon principles of inferability , which govern the ability of a comprehender to retrieve the underlying structure of a story. If statements from the Initiating Event category are given, three types of events can follow: 1 . Initiating Event statements. 2. Internal Response statements. 3. Attempt statements.
Both N. S. Johnson and Mandler (1980) and Stein (1979) argue that Internal Responses can be deleted if an initiating event is present in the
Children’s Knowledge of Events
243
episode, and both investigators understand the paradoxical nature of this deletion. For example, it has been argued that the goal (part of the internal response category) is central to the definition of an episode, for it provides the framework for understanding both prior and subsequent events in the story sequence. However, the type of goal motivated by the occurrence of an initiating event is very limiting, and the comprehender can often guess the appropriate goals of a protagonist. The protagonist’s goal can also be inferred from the overt actions carried out in the story sequence. Thus, by combining the initiating event with the attempt, a comprehender can easily infer the appropriate goal in a large number of stories. If events in the Internal Response category are presented, then two responses are considered to be appropriate: 1. Internal Response statements 2. Attempt statements. Here neither N . S . Johnson and Mandler (1980) nor Stein (1979) allows for any deletions. Thus, Consequence statements would be inappropriate. Black and Wilensky (1979), however, argue that the comprehender can delete attempts and proceed to consequences, under conditions where the attempt can easily be inferred. These differences are, in our opinion, testable, and will be discussed later in the article. When an Attempt is presented, two responses are proposed to be appropriate: 1. Attempt statements. 2. Consequence statements.
Again, the deletion of the following Consequence category is not permitted in a story sequence. Thus Reaction statements would be inappropriate. When the Consequence is presented, one of four events is appropriate: 1. Consequence statements. 2. Reaction statements. 3 . Initiating Event statements. 4.Internal Response statements.
Statements following consequences lead to the greatest amount of variability in responses. As is true for all episodic categories, the consequence can be further elaborated upon, or subsequent Reaction statements
Tom Trabasso et al.
244
can be given. However, another set of responses is appropriate. If a subject chooses to continue the story sequence by adding another episode, then the Consequence statements can serve as an Initiating Event. In this case, it would be appropriate to respond with Internal Response statements. The subject may also decide to continue the story by adding on a second episode. In this case, Initiating Events would be appropriate. In order to facilitate understanding for these predictions, Table I includes a summary of the expectations proposed. In order to investigate these expectations, we shall present four studies, varying in the degree of contextual support provided by the stimulus events. However, before we begin our presentation, a critical discussion of the developmental issues at stake is essential. To an investigator working with adults, the ability to understand the task is taken for granted, as is the assumption that adults have acquired the necessary operations to respond adequately to the task. However, these assumptions cannot be made about children. The following review will illustrate our point.
I.
Developmental Issues in Story Understanding
The earliest work on narrative comprehension was completed by Binet and Henri in 1894. In their studies, Binet and Henri found that children’s memory for stories was not verbatim, and that deletions and transforma-
TABLE I
EXPECTATIONS ABOUT THE TYPEOF EVENTS THATCAN OCCURI N A FORWARD-GOING STORY SEQUENCE 1 . Setting
c
Setting Initiating Event Internal Response
2. Initiating Event
c
Initiating Event Internal Response Attempt
3 . Internal Response
+
Internal Response Attempt
4. Attempt 5 . Consequence
-
-
Attempt Consequence Consequence React ion Initiating Event Internal Response
Children’s Knowledge of Events
245
tions occurred in memory for accurate recall. These investigators also found that certain events were remembered better than others. Thus, they anticipated Bartlett ’s ( 1932) ideas on schema-driven comprehension and the findings of many recent developmental studies of story memory (Mandler & Johnson, 1977; Mandler, Scribner, Cole, & DeForest, 1980; Stein, 1979; Stein & Glenn, 1979). Another early study or! children’s story understanding was completed by Piaget (192511955). In his book, The Language and Thought of the Child, Piaget describes how the child attempts to communicate and understand events like those portrayed in narrative sequences and everyday interaction. The central issue under consideration was whether young children had the prerequisite cognitive abilities to understand and communicate text in a coherent fashion. His main contention was that the young child below the age of 6 or 7 years was egocentric in his thinking and was therefore not capable of true logical thought. His explanation for this egocentrism was that the young child did not have the cognitive capabilities to make himself clearly understood or to take another person’s point of view, a critical dimension of nonegocentric thought. Piaget (1925/1955, 1928a/l969a) relied primarily upon the construct of juxtaposition to support his notions about egocentric thought in the young child. In effect, Piaget assumed that juxtaposition caused the child to make fragmentary and incoherent statements rather than coherent stories and explanations. When ‘statements were juxtaposed, they were said to lack temporal, causal, and logical relations (see Piaget, 1955, p. 130, for a stronger statement on these claims). In support of his claim, Piaget provides evidence from the data he collected on the narrative abilities of young children. Four factors are examined: 1. The temporal sequence in which story events are retold by children. 2. The type of anaphoric reference and pronominal substitutions occurring during story retelling. 3. The understanding of causal relations as manifested in the use of explicit markers connecting two events in a story and the order in which children recall cause and effect. 4. The frequency of children’s using consequences to justify why a preceding event occurred. As might be expected, Piaget claims that the young child cannot correctly order temporal events, confuses the pronominal references used during recall, does not use explicit causal connectors in recalling narrative events, and frequently uses consequence information to justify the occurrence of a prior event.
Tom Trabasso et al.
246
2.
Plan of Reseurch
In order to provide the reader with current findings on children’s knowledge of causal sequences and their ability to use this knowledge in prediction or recall studies, we shall use Piaget’s four criteria for causal understanding as an organizing device. In discussing temporal ordering and pronominal references, we shall reexamine Piaget ’s original data in the light of modern research findings. For the spontaneous use of connectives, the order in which cause and effect are stated, and the use of logical justification to explain the cause of an event, we shall present new empirical evidence. Our new studies will also be used to examine children’s expectations about story structure, primarily regarding the type of information that should occur in stories. We shall present findings about expectations when minimal contextual support is provided (single sentences) vs. expectations when a more elaborated context is provided (more than one category of information). Data will also be cited that relate to story expectations inferred from children’s own generated stories, their answers to probe questions, and their memory for stories with deleted information. 3.
Teinpnral Sequence of Events
The main evidence for the young child’s inability to order correctly events in a narrative sequence comes from Piaget’s (1928bh969b) studies on event memory. He examined children’s skill at reconstructing sequences by asking them to retell stories or to reconstruct the order of a series of pictures. In both instances, he reported that children below the age of 7 or 8 years failed to maintain the correct order of events. From this, Piaget concluded that for the young child events are linked together on the basis of personal interest and not on the real-time order of events (Piaget, 1928b/1969b, p. 272). Fraisse (1963) echoed Piaget in commenting that children’s memory for stories is “completely jumbled up, for they have not learned to reconstruct their past; this is shown by the haphazard way in which they retell stories, for the order of events depends more on their interests or on incidental association than on reality” (Fraisse, 1963, p. 254). Intrigued by these findings, Brown and her colleagues carried out an extensive series of studies investigating children’s memory for ordered sequences of events (see Brown, 1976, for a complete review of this work). Children of preschool age were able to reconstruct accurately a series of ordered pictures with few or no errors in their sequences. When the events were given in a random order, the youngest children ( 5 years of age) made more errors and often attempted to reconstruct the sequence so that it would correspond to the natural order of the events rather than to
Children's Knowledge of Events
241
the random order. However, when Brown assessed the children's skill at recalling the correct order of events, she found children often made temporal sequence errors. Data from recent studies (Day, Stein, Trabasso, & Shirey, 1979; Mandler & Johnson, 1977; Nezworski, Stein, & Trabasso, 1979; Stein & Glenn, 1979), however, show that children as young as 4 years of age can recall the correct order of story events, provided the sequence corresponds to the standard episodic order of stories. It should be noted that children do not recall all the events presented in a story, but those events they do recall are ordered according to the episodic structure. The evidence for these claims stems from the correlations found between the temporal order of the story events in stimulus materials and the order of events as they occur in recall. Stein and Glenn's (1979) data on this issue are striking. For first-grade children, who were asked to recall four stones, the correlations ranged from .81 to .97, with a median correlation of .95 and a mean of .93. The results from the fifth-grade children were similar, the range of correlations being from .95 to .99 (median = .99; mean = .98). In the face of these results, we decided to reanalyze Piaget 's own data, which served as the basis for his claims on the children's inability to reconstruct the order of narrative events. One story that Piaget used is given in Table 11, along with two protocols collected from children in his TABLE I1 THE STORYOF NIOBI? Original version: Once upon a :irne, there was a lady who was called Niobe, and who had twelve sons and twelve daughters. She met a fairy who had only one son and no daughters.'Then the lady laughed at the fairy because the fairy had only one boy. Then the fairy was very angry and fastened the lady to a rock. The lady cried for ten years. In the end, she turned into a rock and her tears made a stream which still runs today (Piaget, 1925/1955, p. 82). Protocols: 1. Gio (8 years). Once upon a time there was a lady who had twelve boys and twelve girls and then a fairy, a boy and a girl. And then Niobe wanted to have some more sons than the fairy. Then she (1) was angry. She (2) fastened her (3) to a stone. She (4) turned into a rock and the tears made a stream which is still running today (Piaget, 1925/1955, p. 116). 2. Met (6 years, 4 months). The lady laughed at this fairy because she (1) had only one boy. The lady had 12 sons and 12 daughters. One day, she (2) laughed at her (3). She (4) was angry and she ( 5 ) fastened her (6) beside a stream. She (7) cried for fifty months and it made a great big stream (Piaget, 1925/1955, p. 121). ~
" The numbers
~~
indicate where (lady or fairy?) was inserted in the experiment.
Tom Trabasso et al.
24H
study. While the story is not well formed (see Mandler & Johnson, 1977; Stein, 1979; Stein & Trabasso, 1981, for more detailed criticism on this feature), we found only one temporal ordering error in all of Piaget’s own data. For the four children he reports data on, the average correlation between the order of events in the story test and the order in which the events were recalled is .98! Piaget presents the texts from other stories he used but does not present enough data to allow an examination of the generalizability of the results found from the data in the Niobe study. However, the data from the Niobe story clearly do not support Piaget’s or Fraisse’s arguments. 4.
Ariuphoric Reference
Piaget (1925/1955) and, more recently, Brown (1976) claim that children frequently use indeterminate pronouns; i.e., “pronouns, personal and demonstrative adjectives, etc., ‘he, she’ or ‘that, the, him’ etc. are used right and left without any indication of what they refer to” (Piaget, 1925/1955, p. 116). The support for these assertions stems from the protocols presented in Table 11, which the reader can reexamine. Our analysis of both protocols suggests that both children’s usage of pronouns is neither random nor irregular. A simple rule accounts for the twelve pronominal references: 1 . The current pronoun refers to the protagonist last named.
or 2. The current pronoun refers to the referent of the current pronoun; 3 . The one exception to this rule is the following: If the preceding pronoun is a recipient of an action, then switch reference to the other protagonist.
For an example, we use Gio’s protocol to illustrate the use of this rule. The first pronoun (she) refers to the protagonist last named, the fairy. The second pronoun (she) refers again to the fairy, who was the referrent of the last pronoun. The third pronoun (her) refers to Niobe, because this pronoun is the recipient of an action and requires a switch in reference. The fourth pronoun (she) then refers to Niobe, who was the referent of the last pronoun. In order to test the ease with which adults could determine the correct protagonist from children’s pronominal usage, we asked adults to tell us which character was being referred to in these passages. Since the children and the experimenter were familiar with the original text of the passage, we gave the text to 26 college students so that they would have
Children’s Knowledge of Events
249
prior knowledge of the story. After reading the story, these students circled one of two noun referrents (lady or fairy) placed at the numbered locations in the protocols given in Table 11. Another 22 college students completed the same task, having no prior knowledge of the story. The percentages of correct referrent identification were 97 and 90 (number of pronouns identified equaled 312 and 264), attesting to the ease with which subjects could identify the correct protagonist referred to by a pronoun. We are aware that these pronominal identifications are influenced by factors other than syntactic structures of sentences. However, the ease of formulating a rule for pronominal use plus the high accuracy of correctly identifying the way in which children referred to the two story protagonists again illustrates that children’s language usage is not random. They in fact communicate fairly well, and adults understand them.
11.
Causal Relationships
In his study of the child’s knowledge of the causal relationship linking narrative events, Piaget (1928ai1969a) relied mainly upon two types of evidence to make his assertions: 1. The overt marking of relationship between successive events in spontaneous speech and descriptions of picture sequences. These markings were noted by the use of causal (because, since, if, etc.) vs. temporal (and, then) connectives. 2. The order in which causes and consequences occurred when children were asked to respond to an event with a because occumng at the end of the stem (e.g., The man fell off the bicycle because. . . ).
Piaget claimed that the young egocentric child did not explicitly mark event relations and confused the meaning of temporal and causal connectors (e.g., young children equated the meaning of “and” with “because”). He also stated that these children juxtaposed (reversed) causeconsequence relationships by giving consequences as explanations for the occurrence of an event. In terms of his argument about improper marking or the low frequency of marking in young children’s spontaneous utterances, Piaget (1928d 1969a) reports that the percentage of because connectives increased from 1.2% to 6. I % over the age span of 3 to 7 years ( N = 2 children per year of age). In another study completed by Margairez and Piaget (1925), 1080 children, aged 4 to 12 years ( N = 120 per age group), made up
250
Tom Trabasso et al.
stories to a pair of pictures taken from a 1911 Polish intelligence test. Again, Margairez and Piaget reported an increase in the use of causal connectives (from 0.8% to 2.3%). In his argument that young children juxtapose cause and consequence, Piaget (1 928a/l969a) gives only examples from protocols where children were given stems with because occurring at the end of the stem. Children were claimed to respond in the following way: The man fell off the horse because he broke his arm. In surveying his results, Piaget claimed that, before the age of 7 or 8 years, children were incapable of differentiation between possible relations such as cause, consequence, or logical justification (see Piaget, 1969a, p. 119, for the exact quote). He went on further to elaborate on his mechanism of juxtaposition as the inversion of causal relations, the result of children both juxtaposing successive statements and omitting or mixing connectives to mark the relations between the sentences. Only gradually would the young child come to differentiate the causal, consecutive, and purposive (finalistic) relations by the age of 7 or 8 years. Following this differentiation, the child would then become capable of logical justification (see Piaget, 1969a, pp. 20-21, for the exact quote). Although Piaget's work has spawned a large interest in aspects of physical causality (Bullock & Gelman, 1979; Gelman, 1978; Mendelson & Shultz, 1976; Shultz & Mendelson, 1975; Siegler, 1975, 1976; Siegler & Liebert, 1974) and in the causal representation of short narrative sequences (Corrigan, 1975; Hood & Bloom, 1979; Katz & Brent, 1968; Kuhn & Phelps, 1976; Kun, 1978), none of these studies has examined, replicated, or extended Piaget 's methods and observations. Those narrative studies related to this work have used two- or three-sentence tasks, requiring either ( a ) the selection of an antecedent or consequence or ( b ) the acceptance or rejection of an A+B relation (or the reverse pairing of two statements). As such, these are comprehension tasks and do not directly test children's usage of causal relationships.2 In our first reported study, we use Piaget 's generation paradigm. One difficulty with Piaget 's original work and subsequent studies on narrative causal understanding is that there is no systematic sampling of the different types of events that might occur in a narrative. This type of sampling is critical because the type of event given might regulate the type of answer children produce. Certain types of events might increase the frequency of juxtaposition, whereas responses to other types of events would contain few if any juxtaposed responses. The same is true for the 'One exception is a study b y Hood and Bloom (1979), who found that 2- and 3-year-olds use causal connectives in spontaneous speech.
Children’s Knowledge of Events
25 1
spontaneous usage of causal connectors. Some types of events might predispose the child to proceed in a backward fashion, necessitating the usage of a causal marker to signal the shift in the normal direction (e.g., forward). Other events might elicit responses indicative of a forwardgoing chain, which could be either causal or temporal in nature. To investigate the spontaneous usage of causal connectors, the frequency with which children juxtapose antecedent-consequence relationships, and their ability to distinguish antecedent-consequent relationships, we extended, in our first study, Piaget’s (1928/1969) procedure in two ways. First, in order to measure children’s spontaneous usage of connectives, we gave children stems without any marker attached to the end. The child had to produce another statement and could choose the directionality of the response as well as whether markers were to be used. Inclusion of this condition also allowed us to test Werner and Kaplan’s (1 963) arguments that consequences should be the normal direction of retrieval rather than proceeding in a backward fashion. This implication can also be derived from the current story work (e.g., the description of the episodic sequence is formulated in a complete forward going chain). A second condition included in our study was the presentation of stems with because attached to the end (e.g., John ran after the ball because . . . ), identical in form to Piaget’s stems. This condition allowed us to assess children’s skill at producing appropriate antecedents to focal events vs. their tendency to juxtapose and produce logical justifications (consequences instead of antecedents). The third condition was the presentation of stems with and then attached to the end (John ran after the ball and then. . . ). The inclusion of this condition allowed us to test several hypotheses. First, we could look at the responses to the different categories of story stems in relation to the expectations children develop about the sequence of story events, as outlined in Table I. Although the measurement of sequence expectations is somewhat difficult with single-sentence cues, there still should be differentiation in responses depending upon the content and form of the stem. Children’s knowledge of events should be constrained in a manner similar to that outlined in the episodic sequence. Comparison across conditions (e.g., between Because and And Then conditions) provides another method of determining whether or not children can differentiate antecedents from consequences. From the descriptions given in current models of story comprehension, it can be hypothesized that the category or type of response to a particular focal event should rarely overlap in the two different conditions. Take the following example as an illustration. Suppose a subject is presented with an action of a protagonist: “John ran after the ball. . . .“ In the Because
Tom Trahasso et al.
252
condition, children would most likely give a goal as the reason for the action (e.g., because he wanted to catch it), whereas in the And Then condition, children would most likely generate an outcome or consequence, represented by an end state or action (e.g., and then he caught it). Comparison between the two conditions should also allow an assessment of the symmetry in the type of sequential chains generated in the forward and backward conditions. Table I details the type of sequence expected in the forward direction. Piaget (192811969) has suggested that going backward from the end point of a narrative should prove symmetrical to the forward-going sequence. Thus, if the forward event sequence is A-+B+C+D, then the backward sequence should be D+C-+B-+A. We would take exception to Piaget’s argument on the grounds that ‘‘backward reasoning in narratives usually represents an attempt to find the “reason for” or the direct cause of a focal event, whereas supplying the event that follows a focal event does not necessarily constrain the subject to a direct causal link. Focal events in a narrative can enable other events to occur but do not necessarily “cause” the occurrence of subsequent events. Our generation paradigm allows us to examine this issue, as well as the more central questions discussed previously. The methods for this study are as follows. ”
STUDY 1
I.
Method
a. Suhjecrs. The children were 135 boys and 135 girls from predominantly middle-class families in St. Paul, Minnesota. They were recruited from kindergarten and second- and fourth-grade classrooms. Kindergarteners ranged in age from 5 to 6 1 0 (mean age = 5:9), second graders from 7 to 8:8 (mean age = 7:8), and fourth graders from 9 to 1O:O (mean age = 9:8). b. Materials. Twenty-eight different types of sentences were presented as focal events. Each type of sentence was classified into one of the seven following categories: A. Qualitative states 1. Physical state of person, e.g., Mike was tall. 2. physical state of object, e.g., The tower was tall. 3 . personality trait, e.g., Mike was friendly. 4. psychological trait, e.g., Mike was smart. environmental state, e.g., It was hot. 5.
Children’s Knowledge of Events
253
B. Attributive states 6. 7. 8. 9.
person-object, e.g., Jerry had a book. person-person, e.g., Jerry had a grandmother. object-object, e.g., The building had an elevator. object-person, e.g., The building had a janitor.
C. Physical events 10. object acts on object, e.g., The branch fell on the car. 11. object acts on person, e.g., The branch fell on Tammy. D. Reactions 12. emotional reactions, e.g., Mark was sad. 13. involuntary actions, e.g., Mark fell down.
E. Cognitions (verbs used: thought about, remembered, forgot) 14. 15.
16.
17.
18.
cognitions about actions, e.g., Paul forgot to carry the paper. cognitions about actions, e.g., Paul forgot to carry George. cognitions about people or objects, e.g., Rachel thought about her sister. cognitions about people or objects, e.g., Rachel thought about her watergun. cognitions about environmental states, e.g., Sheila forgot that it was a hot day.
F. Goals 19. desires about acting on objects, e.g., Paul wanted to carry the 20. 21. 22. 23. 24.
paper. desires about acting on people, e.g., Paul wanted to carry George. desires about obtaining objects, e.g., Paul wanted a watergun. desires about obtaining people, e.g., Paul wanted his mother. decisions about acting on objects, e.g., Paul decided to carry the paper. decisions about acting on people, e.g., Paul decided to carry George.
G. Actions 25. action on an object, e.g., Ann pushed the table. 26. action on a person, e.g., Ann pushed Nancy. 27. attempted action on an object, e.g., Ann tried to push the table. 28. attempted action on a person, e.g., Ann tried to push Nancy.
To enable us to test for generalization of responses in each category, six different stimuli were generated for each of the 28 types of events.
Toni Trabasso et al.
2 54
Then six lists of 28 events were constructed. Children were randomly assigned to the lists so that each list was heard by the same number of children within each grade level and each experimental condition, and by an equal number of boys and girls. Within each experimental condition, there were 30 children from each grade level. c . Procedure. Children were seen individually in an experimental room. Seated at a table across from the experimenter, each child was given the following instructions: “I am going to tell you about something that happened in a story. I would like you to continue the story with just one sentence. Tell me the first thing that you can think about.” Then, if the child has been assigned to the Neutral condition, the child was told: “Here is an example. Suppose 1 said, ‘John spilled his milk.’ Now you say something.” In the Because condition, the example was “John spilled his milk because. . . ,” whereupon the child was asked to finish the sentence. Children in the And Then condition were asked to finish the sentence “John spilled his milk and then.. . .” After each child had successfully answered the statement in the the example, 28 sentences were presented, one at a time, using the same procedure as in the example. The order of presentation for the 28 events was random for each child. The session lasted about 15 minutes.
2.
Results
a . Use .f Explicit Connecting Markers. The neutral condition, where children could proceed in either direction in response to the simulus event, allowed children to mark spontaneously the relationships between events. In order to understand the results from this condition, two analyses were carried out. First, we determined how many responses could be considered antecedents and how many consequents. Then we determined what type of explicit markers were used for antecedent and consequent responses. To determine the directionality of response, two raters scored the event pairs as to the temporal order of occurrence. For the reliability assessment, 350 statements were scored, with the percentage of agreement being 97. Most responses were easily scored. To the statement “John played with the skateboard,” the answer “He fell and hurt his knee” was scored as a consequence. However, a statement such as “He wanted to have fun” was scored as an antecedent. Certain types of responses were less clear. To the statement “It was Sunday,” a frequent response was “They went to church.” This response was judged as a consequence, even though the two events overlapped in time. Sunday was judged as a precondition for going to church.
Children's Knowledge of Events
255
The results showed that for all grade levels the most common direction of response was a forward direction, with 75% of all responses being classified in the consequence category. There were no developmental differences in directionality of responses. Thus, the hypotheses generated by Werner and Kaplan (1963) and many story researchers (N. S. Johnson & Mandler, 1980; Mandler, 1981; Mandler & Johnson, 1977; Stein, 1979; Stein & Glenn, 1979; Stein & Trabasso, 1981) are basically substantiated by these findings. The responses classified as antecedents, although smaller in frequency, were quite interesting. Children produced antecedent responses most frequently to Reactions (52%), which included emotional states, Goals (31%), and Cognitions (24%). Thus, statements perceived to be internal to an actor were those where antecedent responses were generated, whereas antecedents were generated less frequently to states, actions, and physical events. These findings are in keeping with the description of the episode where Reactions occur either after the Initiating Event or at the end of the episode. In either position, there is a preceding event that directly causes the emotional reaction. On the other hand, states, actions, and physical events most often begin the episode, suggesting that, although direct causes could be inferred for these statements, the more likely response would be to infer consequences. Our second set of analyses focused on the frequency and type of explicit connecting markers used by children to link two events. When the absolute number of times children marked events was tabulated, we found that older children marked event relations more frequently than did younger children. The percentages of sentence pairs marked were 35% for 5-year-olds, 71% for 7-year-olds, and 69% for 9-year-olds [F(2,87) = 16.37, p < .01]. These results are not surprising, and they validate Piaget 's ( 1928d 1969a) initial claim that spontaneous marking of event relations increases as a function of age. However, Piaget 's stronger claims concerned the younger child's understanding and use of connecting markers such as because, and then, but, and since. Thus, our second analysis pertained to the type of markers used when antecedents vs. consequences were generated. Although older children did mark spontaneously more event relations, the critical issue does not revolve around the absolute number of markers used. Rather, the more relevant question is: When younger children did mark event relations, did they use the markings appropriately? That is, did they use different markers for antecedents than for consequences. By Piaget 's account, there should be no difference in the type of markers used to link antecedents or consequences, because the young child supposedly makes no differentiation between the two sets of responses.
Tom Trabasso et al.
256
Table 111 contains the number of event relations marked for the two types of responses given to stimulus events. As we have shown in the table, although children produce more consequences than antecedents to stimulus events, they tend to mark events relatively more frequently when they generate an antecedent rather than a consequent. The percentages of marked antecedents vs marked consequents for each age group are 56 vs 27, 78 vs 67, and 85 vs 65. This finding supports the notion that the forward direction is the more expected and frequent, and therefore the child does not have to use explicit devices to signal directionality. However, when children proceed in a backward direction, because it is the less conventional response, they mark their utterances to remain clear to their listeners. This is true for kindergarten children as well as the older children in the study. Table I11 also contains the proportion of times children used the most frequent connectors for each type of response generated. As we have illustrated, the most frequent connectors used were and, because, and but. All other connectors were classified under the Other category for this analysis. The data in this table illustrate two findings. First, the type of response children produce clearly influences the specific connector used to mark event relationships. For antecedent responses, children most frequently marked event relations with because. The connectors and and but were infrequently used. For consequent responses, however, the most frequent markers used were and and bur. Thus, even kindergarten children are making clear distinctions among the types of markers used for each response. Furthermore, when all connecTABLE 111 ANTECEDENT A N D C O N S E Q U E N T RESPONSESA N D PROPORTION OF SPECIFIC CONNECTORS USEDFOR EACHTYPEOF RESPONSE
SFONTANEOUS MARKING FOR
N o . of events Age
5 1 9
5
I 9
Unmarked Antecedent responses 79 54 27 Consequent responses 483 195 227
Conditional proportions
Marked
And
Because
But
Other
102
50 59 51
7 0
156
13 1 3
0
30 40 42
I76 393 430
25 54 56
8
36
30
I
29 25
17
198
1
16
Children’s Knowledge of Events
251
tives were considered (the Other category included), we found that children rarely used the same connectives to mark antecedent and consequent responses (the exception to this was the use of the marker so). For antecedent responses, the connectors were because, or else, so that, to (in order to), for, and when. For consequent responses, connectors used were and, and then, before, instead of, by, and that. Our results, then, show that as children get older they spontaneously mark more events. However, our data also show that young children are not insensitive to event-order, and they mark event relationships correctly. When children give antecedents to stimulus events, they mark them as frequently, using because as much as do older children. Thus, Piaget ’s observation that young children do not use because appropriately is not supported by our data. b. Comparisons between the And Then and Because Conditions. Our next question of interest concerned children’s ability to generate correct responses in the And Then and Because conditions. That is, could children generate correct antecedents when provided with an explicit marker such as because, and could they generate correct consequents when provided with an and then connector? For the Because condition, responses were scored as correct antecedents if the following criterion could be applied: Could the generated response be a reason or a motive for, or could it produce, the focal event? For the And Then condition, the question was: Could the focal event have resulted in the response? The claim being tested in this analysis was that young children would not give appropriate antecedents in the Because condition. Rather, they would ‘‘juxtapose ” their responses, giving consequences instead. Piaget ’s claim (1 928/1969) has received some support from results obtained by Corrigan (1975) and Kuhn and Phelps (1976), where these investigators found a high incidence of confusion of antecedent-consequent relations. The results from this analysis are summarized in Table IV. The first finding of interest is that children of all ages were highly accurate in producing the appropriate response to an event with an explicit marker attached. In the Because condition, kindergarten children gave correct responses 86% of the time. The contention that young children juxtapose events received little support from these data, even though there were developmental differences in the number of correct responses [F(2,174) = 15.33, p < .01], and in the decline for the number of reversals produced in the Because condition [F(2,174) = 20.04, p < .01]. The excellent discrimination between Because and And Then replicates Kun’s (1978) results with children from 4 to 8 years of age. Kun used three narrative events, which were causally linked and supported a picture sequence presented to children (e.g., Scott pulled the dog’s tail-the dog
Tom Trabasso et at.
258
TABLE IV PERCENTAGE OF TYPESOF ANSWERS I N THE EXPLICIT CONNECTIVE CONDITIONS FOR EACH AGEGROUP Condition And Then
Because Type of answer Antecedent Consequent Nonehelevant
Age:
5
I
9
5
I
9
86
94 4 2
96
4 94
2 97 0
2 98
6 9
3 I
3
0
bit Scott-Scott cried). Focusing on the middle event, Kun asked a Why? and a What happened next? question. She found no age differences in correct selection of antecedent or consequent responses. The mean percentages of correct answers were 92 and 94 for the respective Why? and What happened next? questions. These results are similar to ours, even though we did find some developmental increase in correct responses. Our lowest mean for correct responses was 86%. Kun’s results and ours differ from those of both Corrigan (1975) and Kuhn and Phelps (1976). However, a short task andysis of their studies may give insight into the different results. In Comgan’s ( I 975) study, the children had to reject a reversal in order to be correct. For example, if a child was presented with the statement “The dog barked because the man was scared,” the correct answer had to be given by saying “No.” This task demand goes against a positive response bias of younger children (Trabasso & Nicholas, 1980). Also, children could have perceived a situation where this order of the two events was correct. In the Kuhn and Phelps (1976) study, children had to choose pairs of sentences that “went best with a picture.” The major problem in the Kuhn and Phelps studies is that these investigators did not clarify what types of responses they expected. Children can accept as a plausible justification of an event the subsequent consequence. Thus the sentence pair “The water spills because the chair gets wet” may describe a picture as well as “The chair gets wet because the water spills.” When Kuhn and Phelps explicitly marked the temporal order of events with “first” and “second,” the percentage of correct responses rose from 56 to 84 for their 5-year-old children. The addition of these temporal markers may have enabled the young children to understand that a cause was the appropriate answer, not a justification or verification of content against the picture.
Children’s Knowledge of Events
259
In order to better understand the “difficulty” children did experience in our own study, we next examined the items to which children gave the most incorrect responses. We wanted to determine whether specific classes of events proved to be difficult or whether errors were found in all categories of events. From the summary of results presented in Table V, it is evident that two types of events are accounting for most of the errors made by children. The two types of events are Qualitative States (John was tall) and Cognitions (John forgot to bring the balloon). As can be seen in Table V, the errors for these events occur only in the Because condition, where the developmental differences are most marked in providing a consequence (reversal) for a Qualitative State and an irrelevant response for a Cognition. Reversals occurred in response to Qualitative States because children gave justifications or proof for the occurrence of a state (It was a hot day because the ice melted; It was Sunday because we went to church). These types of responses clearly indicate that children interpret because in more than one way. Justifications may be as appropriate as causes, especially when children find it difficult to generate a correct antecedent event. We shall discuss this in more detail later in the article. The large number of irrelevant responses to cognitions may indicate that young children either have difficulty generating antecedents
TABLE V
PERCENTAGE OF APPROPRIATE OR REVERSAL RESPONSES AND ERRORS FOR QUALITATIVE STATES A N D COGNITIONS COMPARED WITH ALL OTHER SENTENCE CATEGORIES Condition Because Age: Qualitative States Appropriate Reversal None/imelevant Cognitions Appropriate Reversal Nonehelevant All other categories Appropriate Reversal None/imelevant
And Then
5
1
9
5
1
9
60 19 21
15 21 4
a1 12
95 1 4
91 2 1
99 1 0
13 6 21
92 0 8
98 2 0
93 5
91 2 1
91 2 1
93
98
I
94 4
4
1
99 1 0
91 3 0
98
3
1
2
2
2 0
Torn Trabasso et at.
260
to cognitions or have difficulty understanding the meaning of the verbs used in these statements (Wellman & Johnson, 1979). c . Diflerentiation of’Event Responses. Our next set of analysis focused on the classes of events generated in the Because and And Then conditions. One of our previous claims was that children should generate different classes of events in the two conditions, with goals and states being the predominant mode of response in the Because condition, whereas actions and end states or reactions should be more frequent in the And Then condition. In order to determine whether or not these differentiations occurred, all responses from both conditions were classified into seven major categories similar to those used to construct the stimulus material: (1) States, (2) State Changes, (3) Physical Events, (4)Reactions (Emotions), ( 5 ) Cognitions, (6) Goals, and (7) Actions. The inter-rater reliability for classification was 92%. The distribution of answers in the two conditions is given in Table VI. The results indicate that answers given i n the two conditions were significantly different, with actions (41%) and end states (27%) appearing more frequently in the forward And Then condition than in the Because condition. Conversely, goals (24%) and states (43%) were given more frequently in the Because condition. Thus, children generate events that occur near the end of an episode in the And Then condition, whereas they generate events that occur near the beginning of an episode in the Because TABLE VI
DISTRIBUTION OF RESPONSES FOR EACH EXPLICIT MARKERCONDITION
PROEAEILITY
Condition Type of answer States Physical Psychological State Changes Physical Psychological Events Emotions Goals Cognitions Actions
Because
And Then
32 11
0s 01
04 00 04 00
09 09 05
24
05 21
I8
04 02 41
Children’s Knowledge of Events
261
condition. These data offer support for our contention that stories are problem-solving sequences involving intentional goal-directed actions, with beginnings and endings (Black & Bower, 1980; Stein, 1979; Stein & Glenn, 1979; Stein & Trabasso, 1981). Our next analysis focuses on an intercorrelation matrix completed on both the stimulus events and responses given to these events. A matrix was set up to determine the similarity in responses to stimulus events across age groups and conditions. The resulting correlation matrix was then subjected to a complete link (diameter, maximum distance) cluster analysis (Friendly, 1977). The results of this analysis are shown in Fig. 1. The category labels that run down the left-hand column of the figure represent the types of events given as stimulus materials. The scale at the top of the page represents the strength of correlation for responses to those events across grade level and condition. The first important finding is that responses to events are highly differentiated when the two conditions are compared. There is virtually no relationship between responses given in the Because condition and those given in the And Then condition. Second, responses to events across age groups are highly similar. This can be seen by the tightness of clusters between the grades. Thus, children from 5 through 9 years of age give the same kinds of antecedent and consequent responses. It can also be seen from Fig. 1 that certain events form identifiable clusters. For example, the set of Cognitions, Goals, and Actions have both common antecedents and consequents. Qualitative and Attributive States have common consequents but are differentiated with respect to antecedents. Physical Events and Reactions are moderately discriminated in both conditions and are fairly separated from other types of events. These clusters provide some insight into the psychological meaning of the episode, as we shall illustrate in our final set of analyses. The fact that the degree of differentiation is greater for antecedents than for consequences also suggests that there is an asymmetry in the types of chains generated when proceeding in a forward vs a backward direction. To summarize the data presented in Table VI and Fig. 1, antecedents are clearly differentiated from consequences, with respect not only to the temporal relation among events, but also to the class of responses given to each stimulus event in the two conditions. When children complete stems explicitly marked with and then, they respond as if they are proceeding forward in a narrative and produce events that occur at the end of a narrative (e.g., actions and state changes). However, when generating responses in the Because condition, the most frequent type of events given are those that occur at the beginning of an episode. The final set of analyses for this study examined the types of events that
Tom lrabasso et al.
262
10
Correlations B e t w r r n Answers b0 .4 0 .20
.OO
b
I
I
Cognition
?I3
5
Goal Action Action Goal Goal Action Reaction
Q
aD
L
Qualitative State Phyricol Event Attributive State
Jl
Action
I
Cognition
c
0
z b TI
Qualitative Stote
C
Attributive State Reaction Physical Event
Qo
I
Fig. 1 , Cluster analysis on age, stimulus events, and the Because-And Then conditions
Children’s Knowledge of Events
263
precede or follow each of the specific classes of stimulus events. These analyses allowed us to examine whether children’s expectations concerning the sequence of episodic events were similar to those outlined in Table I. In order to complete these analyses, a 6 X 6 matrix was generated for each age group and each condition. Each matrix contained the six categories of stimulus events paired with the same six categories of responses generated to each class of events. By creating these matrices, we were able to examine the most frequent category of responses given to each stimulus event. Table VII summarizes the normalized rank order of category responses for each category of stimulus materials. The three different matrices for each grade in each condition were combined because there were no differences in responses across grade level. Table VIII includes the highest ranking responses for each given stimulus category. The format of the table is similar to the one used in Table I . Responses given to each stimulus category can then be compared with the predictions generated from current models of story understanding (N. S. Johnson & Mandler, 1980; Mandler & Johnson, 1977; Stein, 1979; Stein & Glenn, 1979). Our discussion of the results focuses first on the And Then condition because current models make more specific predictions concerning the TABLE VII RANKORDEROF CATEGORY RESPONSES TO THE STIMULUS CATEGORIES FOR THE “BECAUSE”A N D “AND THEN” CONDITIONS Responses given Stimulus category
States
Events
Reactions
Cognitions
Goals
Actions
“Because” condition States Events Reactions Cognitions Goals Actions
5
5.5
4
6 1
1
4 4
4.5 2.5 2.5
4
I 4
States Events Reactions Cognitions Goals Actions
1 2.5 2.5
2 3.5 3.5
5 6 4
3.5 5.5 2
3.5 2 1 6 5
4 3
“And Then” condition 5 2 1
3 5 5
’2 6
3 1
1
4 2
6 4.5
5 2
2 5 5 5
5 6 4
4.5
1
3 2
1
4.5 4.5 2 4.5
5 6 4 2 1
3
Tam Trahasso et al.
264
TABLE VIII THEHIGHESTRANKEDRESPONSES GIVEN TO EACHCLASS OF STIMULUS MATERIAIP And Then condition
Because condition Stimulus States
-
Response
Stimulus
-
Goals
States
Events Actions
Events
Reactions
States Actions
Reactions
Cognitions
Cognitions States
Cognitions
Goals
Reactions Events
Goals
Goals
Actions
Events
-
Actions ~
-
-
Response
States Events Events Goals Reactions Reactions Cognitions Actions
-
Actions Goals
-
Cognitions
~~
“The arrow indicates the causal relation
forward sequence of events in a story. For States, the most frequent responses were other States and Physical Events. Each of these responses corresponds identically to the predictions derived from current story grammar work. Normally, States are considered part of the Setting category, which begins the story. The most frequent responses expected to this type of statement were other Setting statements or Initiating Events. The fact that State responses were the highest frequency answers indicates that children tend to focus upon descriptive accounts before proceeding to Initiating Events. The fact that Physical Events were the second highest response, however, indicates that children are aware of and expect Physical Events to occur and change the current state of affairs. The most frequent responses to Events were Events, Reactions, and Goals, all of which were predicted by the rewrite rules in current story grammars. Again, the tendency for children to elaborate within the same category is apparent from the responses. However, the next most frequent responses were Reactions (emotions) and Goals, indicative of statements from the Internal Response category. The surprising result here is that Actions (Attempts) were not mentioned frequently in response to Events, suggesting that, although children could delete Internal Responses from their output, their strong preference is to include them, rather than infer
Children’s Knowledge of Events
265
them. This may be due to the fact that children are limited to singlesentence responses, which do not provide enough information for either child or the listener to infer the correct internal response. It is clear from these responses, however, that even 5-year-old children understand and expect internal responses to occur as a function of a physical event’s occurring in the environment. The responses given to Reactions, which included emotional states and involuntary actions, were primarily other Reactions. Thus, children tend to stay within this category, indicating that little planning occurs as a result of a single Reaction’s being given. Most often another emotion or involuntary responses was produced. Table IX illustrates a concrete, modal response to all the stimulus categories for both the And Then (consequent) and Because (antecedent) conditions. For the stimulus sentence “Fred was scared,” classified as a Reaction, a concrete modal response was “He cried,” a Reaction. The modal responses to Cognitions and Goals were similar. For both stimulus categories, children tended to elaborate upon the given stimulus by producing statements that were identical in category, namely Cognitions and Goals. However, the other most frequent response was the category Action, which corresponds to the Attempt category. For the stimulus category Action, the most frequent responses were Cognitions, with Actions occurring but less frequently than expected. TABLE IX MODALANTECEDENT A N D CONSEQUENT EVENTANSWERS FOR STIMULUS CATEGORIES Antecedent (Because)
Stimulus (focal event)
Consequent (And Then)
They wanted to catch somebody (Goal) To talk to other people. (Goal) She dropped the glass. (Action) It was real dark. (Physical Event) He thought she died. (Cognition) She was happy. (Reaction) He wanted it. (Goal)
The rope was strong. (Qualitative State) The truck had a radio. (Attributive State) The broken glass cut the car’s tire. (Physical Event) Fred was scared. (Reaction) Danny thought about his grandmother (Cognition) Mary wanted to bring the balloon to the party. (Goal) Joe ran after the ball. (Action)
It broke. (Physical Event) The speaker blew up. (Physical Event) It popped. (Physical Event) He cried. (Reaction) He went to see her. (Action) Shewanted toplaywithit. (Goal) He couldn’t find it. (Cognition)
266
Tom Trabasso et al.
This is the only category of response that, on the surface, does not correspond directly to the expectations generated from the grammars, as outlined in Table I. However, the failure to obtain the responses predicted is more a function of our scoring system than of the children’s failure to generate appropriate answers. A glance at Table IX, containing the most frequent modal responses to stimulus events, serves as an example of this scoring difficulty. For the stimulus statement “Joe ran after the ball,” classified as an Action, the most frequent response was “He couldn’t find it. ” The response was classified as a Cognition because of the verb find, but the statement is also representative of an end state, signaling an abortive attempt at goal attainment. Many of the children’s responses were similar to the example given in Table IX, indicating the need for a more precise scoring system for the consequence category of the episode. Modal antecedent responses were tabulated in the same manner as consequent responses. All relevant data can be found in Tables VII, VIII, and IX. From these tables, it is evident that the backward causal chain is quite different from the chain generated in the forward-going (And Then) condition. Only certain types of relationships were found to be inherently “reversible,” namely Actions and Goals. That is, Goals lead to Actions, and Actions occur because of particular Goals. Thus, Piaget’s statement about ‘‘reversibility of thought” regarding the organization of narrative events is not supported by the data. Many reasons could be given to explain the lack of symmetry, including the methodology used to measure the relations among events. Whatever the explanation might be, however, it is clear that the concept of reversibility is more complicated than supposed, and that even the oldest children in our sample (e.g., 9-year-olds) do not produce responses that are indicative of “reversible” thought. Despite the lack of symmetry between the causal chains in the two conditions, the data from the Because condition were consistent across age groups and parallel to Graesser’s (1981) findings on adult responses to Why questions. Graesser found that the most common response to Why questions for actions performed by an animate being was a goal; the most common response to Why questions concerning physical events was another physical event. These findings suggest that causal reasoning is different for animate and inanimate objects, and lend support to the importance of the concept of intentional and goal-directed behavior underlying the interpretation and organization of social knowledge. d . Modul Expectuiions Derived from Story Generation Tasks. In our final effort to examine expectations about story categories, we present data from a study completed by Glenn and Stein (1981), who had children tell their own stories. In their study, Glenn and Stein gave children two statements representing the Setting category and had them generate the
Children’s Knowledge of Events
261
entire remainder of the episode. By allowing children to construct an entire sequence of events, we can begin to determine how the process of adding more information to an ongoing sequence influences expectations about subsequent categories. As we previously indicated, it was difficult to draw any firm conclusions regarding expectations children have concerning the categories at the end of the episode. Our scoring procedures did not allow an accurate account. However, this problem can be overcome when entire stories are being scored because the entire sequence can be used to determine the appropriate category membership of each statement. Table X summarizes the most frequent types of responses that occurred to each different story category. In this summary, we have included only those responses that included a category statement other than the focal event. As an example, for the Setting category we have not included the number of setting responses given to setting statements. Rather, we have concentrated on the intercategory expectations. We chose to delete continuations of the stimulus category because the story grammar predictions generated for a given sequence of events, although inclusive of the possiTABLE X
PROBABILITY OF PARTICULAR CATEGORY RESPONSES TO EACH STIMULUS CATEGORY IN STORIES
GENERATED BY CHILDREN“ Probabilties
Stimulus category Setting Initiating Event
Internal Response
Category responses Initiating Event Internal Response
Grade:
K
3
5
.I8
.I7 .23
.91
.22
Internal Response Attempt Consequence
.56
Attempt Consequence
.44
.oo
Attempt
No subsequent category Consequence
.68* .32 .I0 .90
Consequence
No subsequent category Reaction initiating Event Internal Response
.21* .46* .15 .I2
“Data from Glenn and Stein (1981). *p < .05.
.08 .53
.I3 .25 .02
.07
.84
.88
.16 .03 .91
.12 .04 .96
.I5 .60 .I0
.08
.15
.40
.62 .10 .20
2hX
Tom l’rabasso et al.
bility of a response’s continuing in the saFe stimulus category, are primarily focused on intercategory expectations. Therefore, we included only those subsequent statements that had a different category membership than the focal event under consideration, The first results of interest are the responses immediately following Setting statements. Over 75% of all responses were classified as Initiating Events, supporting notions about the content of the initiating events. Those responses that were classified as Internal Responses were goals from which states of lack or deprivation could easily be inferred (e.g., The fox really wanted to catch that rabbit for dinner). Thus, the deletion rule proposed by N . S . Johnson and Mandler (1980) is used, especially when children tell stories about states of deprivation. Similarly, the results showed that, after generating an Initiating Event, children often use a deletion rule, leaving out the Internal Response category, and proceed directly to the character’s Attempt. The Internal Response is deleted more frequently than the Initiating Event, indicating that the content in this category can be more easily inferred than information i n the Initiating Event category. These data are strikingly similar to those occurring in story-memory studies, where the Internal Response is frequently deleted during recall, even though this information is included in the original text (Mandler & Johnson, 1977; Nezworski er al., 1979; Stein & Glenn, 1979). The responses that occurred after the Internal Response category are provocative, because they are indicative of the first set of developmental differences occurring in story knowledge, and the data, on the surface, suggest the need for another deletion rule with respect to the Attempt category in the episode, similar to the suggestion made by Black and Wilensky (1979). Deletions of the Attempt category, however, occur as a function of the age of the storyteller and as a function of including Internal Response statements in a story sequence. As Table X illustrates, younger children are more likely to delete the Attempt category than older children. Also, if the absolute frequency of deleting the Attempt category is calculated, only 20% of all stories contain these deletions, a lower percentage than those occurring if an Internal Response is included in the story sequence. The explanation for the deletion of the Attempt category becomes quite apparent after some of the protocols of younger children were examined. Often, when telling a story, these children will include a plan in their story, where the character rhinks about what he wants to do. If a plan is included, most children proceed directly to the consequence, obviously thinking that the listener will have concluded that the plan had been carried out. This occurred in 63% of the stories where Attempts were deleted from the sequence.
Children’s Knowledge of Events
269
These results suggest that the requirements for what constitute a wellformed episodic sequence should be modified so that plans for goal attainment can be substituted for the actual overt series of actions initiated by the protagonist. This is somewhat different from Black and Wilensky’s (1 979) suggestion because it concentrates on substitution rather than deletion. Our suggestion also signifies that Attempts often contain the novel or unexpected information in a story, suggesting that this category can rarely be deleted entirely. Responses subsequent to consequence statements were as expected, with the Reaction category being the most frequent type of response. Here, again, we see a second set of developmental differences. Kindergarten children are more likely to delete the end category of the episode than older children. This finding was independent of whether children generated another episode following the first episode. That the younger children delete Reactions or endings in the Mandler and Johnson (1977) grammar may be indicative of less integration’s occurring during the generation task. Older children were more likely to give “summarizing statements” than younger children. It is difficult, however, to provide an explanation for these data, and further study will be necessary to confirm our findings. To summarize these data, it is evident that the two generation studies reported, one with single-sentence responses, the other with entire episodic sequence responses, strongly support the story grammar predictions concerning the types of information occurring in story sequences and the order in which these events should occur. Our data suggest the necessity for only one revision: the requirement that an Attempt category, per se, be included in the episodic sequence. We propose that the emphasis be put on the inclusion of plans for action or overt attempts, rather than just the latter category.
111. Causal Understanding in Preschool Children Since we were able to illustrate that 5-year-olds have little difficulty differentiating antecedents from consequents, we decided to investigate the young preschool child’s ability to generate and differentiate between these two types of events. Because it was impossible for 3-year-old children to attend to the large number of stimulus materials generated for the first study, we decided to limit the materials to emotional-state terms. Several factors influenced our decision. First, studies by Borke (1971) have shown that children as young as 4 years of age can correctly identify those situations that elicit the emotions of happiness, sadness, fear, and anger. However, Borke has also indi-
Tom Trabasso et al.
270
cated that children sometimes “confuse” the situations that elicit the negative emotions, namely anger and fear. In fact, Borke suggests that it takes somewhat longer for children not to confuse these emotions. We chose emotions because of this study, since it was apparent that even young 3-year-old children could give verbal responses to questions concerning consequents. By allowing children to “generate” their own knowledge of emotional-state terms, we felt that we could better test whether or not these children tended to “confuse” certain emotional terms. One problem with Borke’s (1971) material was that she had no theoretical rationale for explaining why certain events should arouse one emotion or another. In a generation study, responses would not be considered incorrect. Rather, we could cluster the common responses, similar to the methodology used in our first study, and determine how much overlap was present between arztecedents of emotional states. In addition to examining children’s skill at differentiating among antecedents, we also wanted to examine how well these children differentiated the consequents of different emotional states. Neither Borke ( I 97 1) nor Feshbach and Roe ( 1 968) have examined this issue in exploring children’s knowledge of emotional terms. Thus we chose to use a paradigm similar to the one used in the first study, with one exception. Instead of asking children to complete stems with explicit markers attached, we decided to present the specific emotional term, and ask Why and What questions, as described in our subsequent method section. The fuller version of our study is described in a masters’ thesis by Surbey (1979). STUDY
1.
2
Methods
Thirty-two children, aged 3:O to 4:6 years (mean = 3: lo), were shown drawings of children displaying neutral facial expressions and postures. Six emotional-state terms were used, three of which were positive: Happy, Excited, Surprised, and three of which were negative: Sad, Angry, Scared. The children were assigned to one of two conditions, called Cause and Consequence. Each child was shown a picture and told the beginning of a story, such as: “This is a picture of Jennifer. One day Jennifer got very, very angry. She was so angry that her mother and father could tell she was angry. And all of her friends could tell she was very ungry. In the Cause condition, the child was then asked, “Why do you think Jennifer was so angry?” In the Consequence condition, the child was asked, “And then what do you think Jennifer did?” ”
Children’s Knowledge of Events
271
Each child responded to all six emotions, each of which was paired with a different picture. The order of emotions was random, as was the assignment of emotion to pictures, the latter of which consisted of girls and boys whose names were also randomly assigned with sex to picture. 2.
Results
The answers in each experimental condition were first scored for correct directionality of response. In the Cause condition, we asked the question: Could the child’s responses make Jennifer (child pictured on the stimulus card) feel the appropriate emotion? For the Consequence condition, we asked the question: Is it possible that Jennifer was so (appropriate emotional response inserted) that she did (child’s answer)? Responses that were not causal in nature or were inappropriate were not scored as correct. Two raters scored the responses, obtaining 91% agreement on the Causes and 96% on the Consequences. Table XI summarizes children’s skill at producing accurate causes and consequences to the six emotional states. For causes, children were fairly accurate, ranging from 56 to 94% correct (mean = 72% correct), with a higher accuracy for negative emotional states than for positive emotional states. For consequences, children were just as accurate, with a mean of 73% correct. There were no differences in the accuracy of responses due to the positive or negative emotion given. These data support Borke’s (1971) contention that young children are capable of differentiating between different emotional states. However, our initial analysis suggests that young children have more difficulty TABLE XI PERCENTAGE OF ANSWERS JUDGEDAS CAUSESA N D CONSEQUENCES FOR EMOTIONS~ Cause
Consequence
Happy Surprised Excited
.S6 .69 .66
.73 .66 .69
Sad Angry Scared
.72 .75 .94
.72 .78 .81
.72
.I3
Data from Surbey (1979).
212
Tom Trabasso et al.
determining the causes of positive emotions, rather than negative emotions. In Borke’s study, she suggested the opposite, that children did not differentiate well between events that led to anger or sadness. However, Borke had only one positive emotion included in her study, so that no real basis of comparison could be made. In the present study, it is clear that 3-year-old children could give more correct causes to negative emotions. This type of analysis, however, does not indicate how much differentiation there was among the responses given to positive and negative emotions, nor does it show the amount of differentiation between cause and consequence. In order to answer these questions, all responses were classified by thematic content (someone having a party, arriving, having a conflict or a fight with parents or sibling, or getting hit, animate beings such as large dogs and monsters, etc.). There were 33 categories in all, containing from 2 to 34 responses. Two judges rated the responses independently, with the percentage of agreement being 98. We then calculated the distributions for the number of children who gave each type of answer for each emotion in each condition. These distributions were then correlated in a 12 x 12 matrix. The correlations were then subjected to a complete link analysis (Friendly, 1977), which ties together clusters at the value of the lowest correlation between a new event and any event already in the cluster. The hierarchical structure that resulted is shown in Fig. 2. Causes were correlated with consequences - .25, indicating that there was excellent content differentiation between antecedent and consequent responses. Likewise, the set of positive emotions was well differentiated from the negative ones for both antecedents and consequents ( r = - . I 9 and .OO, respectively). These results are consistent with current adult models of emotional-state term differentiation (Dahl & Stengel, 1978; de Rivera, 1977). As for differentiation within classes of emotional-state terms, negative emotions were far more differentiated than positive emotions, just the opposite of what Borke’s data would suggest. We should note, however, that children’s responses to sadness and anger did overlap, similar to responses found by Borke. Our interpretation of these findings, however, is quite different from Borke’s. In a recent article, Dahl and Stengel (1978) argue that two types of emotional responses can be given to stimulus events, ond dealing with feelings toward an object and the other dealing with feelings toward oneself (Dahl’s it vs. me distinction). When children give a response such as anger to an event, Dahl and Stengel argue that an emotion is being expressed toward the object causing the anger. When children give a response such as sadness, they are expressing their feeling toward themselves (e.g., concern for how things are going for them). Thus, it is
273
Children’s Knowledge of Events
CAUSE Happy Excited Surprised Sad Angry
---+I
Scared
-.25 Excited Happy Surprised
.Q
-
Sad Angry Scared
- 111 i
+1.0
01) Correlation
Fig. 2. Cluster analysis on emotional states and the causes-consequences conditions
entirely plausible that children would produce similar events for anger and sadness, simply because events can produce more than one emotion-one toward the object and one toward the self. This hypothesis has not been investigated and deserves further study. Similar comments can be made about the overlap that occurs between the three positive emotions. These emotions might share features with one another, or more than one emotion may be experienced in response to an event. Before concluding that children do not differentiate well among certain emotional terms, we need to construct a theoretical framework that contains predictions about the ways in which events are evaluated emotionally. According to Dahl and Stengel (1978), the three positive emotions share features (all are considered positive, active, me emotions) and therefore would overlap in terms of the events that produced them.
Tom Trahasso et al.
274
In support of our contention, we include a summary of Davitz’s (1969) work with adults. He had adults rate 50 emotional terms for the presence or absence of 556 descriptive items (e.g., my head spins, there is a radiant glow). A factor analysis yielded two main dimensions, negative and positive, similar to ours. A further analysis yielded more detailed categories, the main ones being happy or cheerful terms, sad or depressed terms, and a set of negative terms including angry and afraid. These results are similar to ours, with the exception that children combine sad and angry more than adults do. Despite the clear lack of distinction on this one dimension, however, our data clearly indicate that even 3-year-old children are surprisingly good at identifying antecedents and consequents of emotional reactions, and that further studies should be carried out to determine their range of causal knowledge concerning other social events.
IV.
Logical Justification: The Meaning of Why and Because
Our final study in this article focuses on children’s interpretations of Why and Because. The question Why?, when asked of both physical and psychological events, normally requests a causal, antecedent, or purposive explanation (Lehnert, 1978), which itself requires logical implication (H. M. Johnson, 1939). The connective “because” implies a similar response. A second interpretation of “why” and “because” is to give a “reason for” or to “justify” what occurs. For example, the event “The man fell off the horse because. . .” could be interpreted as requesting either antecedent or logical proof (justification). Although Piaget ( 1928/ 1969) dismissed justifications as valid responses to “why ” or “because” (e.g., he claimed justifications were not truly logical), each type of response constitutes either a respective necessity or sufficiency explanation for an event (H. M. Johnson, 1939; Mackie, 1979). As a result of our interest in this question, we camed out a study, in collaboration with Simon (1 980), to examine how children would respond to Why and Because terms. Children were first read stones containing three events, A-B-C. Focusing on B, the experimenter asked: ( a ) Why B? ( h ) B Because; or (c) How can you tell that B occurred? An additional factor varied was the degree of causal connection between the three events. In one condition, the events A-B-C were all causally related, in another only events B-C were related, and in a third only events A-B were related. Our hypotheses were the following: If children interpret Why and
Children’s Knowledge of Events
215
Because to mean the same thing, then their responses to these questions should be identical independent of the causal links between events in a given trio. This prediction contradicts Piaget ’s assertion about the equivalence of Because and And Then for the young child. The factor that should regulate whether children proceed in a forward vs. backward direction is the degree of causal connection between the events in the trio. If there is a direct causal connection between A and B, children should give A in response to Why questions and Because stems. If there is a direct causal connection between only B and C, then children should attempt to use C as a justification for the occurrence of B. We are suggesting that children use logical justification when there is no evident cause (or they don’t know a relevant one). STUDY
I.
3
Method
Forty-eight kindergarten and 48 third-grade children (median ages, 5:7 and 9:3 years, respectively) participated in the experiment. In order to provide a wide sample of event sequences, 21 stories were constructed, which had tight causal links between each of the events included in the sequence. This was done by testing adjacent events with Why A? and How do you know that A was the case? when A and B were the two adjacent events. An example of one of our causal sequences is as follows: Setting 1: Setting 2: Initiating Event: Internal Response: Goal: Attempt: Consequence: Reaction 1: Reaction 2:
Clarence was a very clumsy giraffe. Clarence was always hitting his head on high things. One day, Clarence hit his head on a cloud. Clarence realized that being tall wasn’t much fun. Clarence wanted to be short. Clarence asked the doctor to help him. The doctor gave Clarence some shrinking medicine. Now Clarence is very short. But now he hits his head on low things.
From each of the 21 stories, seven successive three-event sequences were constructed (e.g., Setting 1 , Setting 2, Initiating Event; Setting 2, Initiating Event, Cognition; Initiating Event, Cognition, Goal;. . . Consequence, Reaction 1 , Reaction 2). These were the basic triplets used in the study. In order to construct the triplets with only two events directly
Tom Trabasso et al.
216
related, one event (either A or C) was replaced with an event riot related to B. The grammatical category of the replaced event was held constant. Thus, we generated three kinds of triplets: Causal Antecedents and Consequents (A-B-C), Causal Antecedents (A-B only), and Causal Consequents (B-C only). Each child received seven different triplets from seven different stories, sampled randomly without replacement, and with the constraint that all seven successive triplets in the nine-event episode be represented. Children were randomly assigned to one of the three conditions, varying the causal relations among events. Each child was tested individually. After the experimenter read the triplet to the child, the child repeated the presentation to the experimenter. This procedure was repeated as often as necessary to ensure that each child could repeat all three events. Then, focusing on the middle event, children were asked the Because, Why? and How can you tell? questions. For example: Clarence wanted to be short. B. Clarence asked the doctor to help him. C. The doctor gave Clarence some shrinking medicine. A.
The questions were:
I . Clarence asked the doctor to help him because. . . . 2, Why did Clarence ask the doctor to help him? 3. How can you tell that Clarence asked the doctor to help him? The latter two questions were counterbalanced within each condition. 2 . Results
The data were first scored in terms of the frequency of choosing A, C, or other responses (children frequently gave plausible answers as alternatives to A and C). There were no main effects, owing to differences in the content of the triplets or in the content of the Why and Because responses. However, both of these differed from How can you tell? as to the number of A or C answers. Collapsing over all triplets, and the Why/Because questions, Tables XI1 and XI11 summarize the results of interest. First, in Table XII, the children clearly select causally related events as antecedents in responses to Why/Because questions. In contrast to the results in Table XIII, they virtually never gave consequents as answers to Why/Because. These data are consistent with Kun’s (1976) finding on pictured triplets of events and rule out the argument that children merely select an antecedent event, regardless of its causal relation (cf. Bullock & Gelman, 1979; Siegler, 1975).
211
Children's Knowledge of Events
TABLE XI1
PROBABILITY OF ANTECEDENT RESPONSE*' Condition Grade
How can you tell?
Whyhecause
A-B relation:
Causal
Noncausal
Causal
Noncausal
Kindergarten Third
.60 .13
.09 .33
.37
.04
.20
.09
Data from Simon ( 1 980)
With respect to consequents, contrasting the How can you tell? results for the kindergarten and third-grade children, a developmental interaction was found. The 5-year-old children prefer to use causal antecedents as proof (albeit to a lesser degree than they do when answering Why/ Because questions) more than the 9-year-olds. However, the older children are only slightly better at selecting causal consequents as proof (a net 16% vs 1% when one subtracts noncausal responding). Thus, the children clearly interpret Why and Because to request causal antecedents. When forced to justify or provide proof, the children in both groups were less able but the younger ones preferred necessary, causal antecedents, whereas the older children showed a slight preference for sufficient, causal consequents.
V.
General Conclusions
Our primary conclusion from these studies is that children as young as 3 to 5 years of age understand a good deal about causal relations between TABLE XI11
PROBABILITY OF CONSEQUENT RESPONSE Condition Grade
Why/because
How can you tell?
B-C relation:
Causal
Noncausal
Causal
Noncausal
Kindergarten Third
.05
.03
.02
.01
.06 .28
.05 .22
278
Tom Trdbasso et al.
events. In fact, they understand and communicate far more about these relations than they have been given credit for doing. Our procedure of allowing the children to generate causal antecedents and consequents may well be part of the key to revealing their knowledge. Specifically, our reanalysis of Piaget’s ( 1 925/1955) original protocols showed that young children reproduce the temporal order of narrative events quite faithfully and that they use rules for pronominalization that allow accurate identification of their referents. Both of these reanalyses and our experiment on pronominal reference suggest that 5-year-olds are competent and not “egocentric” communicators. Second, when asked to generate antecedent or consequent events, children are able to do so accurately, differentiating between these causal relations. There was little or no evidence that children ‘‘juxtapose consequents with antecedents. Rather, consequents are sometimes given as evidence when antecedents are not known for states. Only two types of events, Qualitative States and Cognitions, caused any problems in generating causes. These difficulties appear to be knowledge-based rather than dependent on the presence or absence of logical thought. The youngest children resembled the older age groups in their ability to differentiate between different kinds of events, be they physical or psychological. The kind of event generated was clearly influenced by causal and content constraints, both of which were reflected in descriptions of how story knowledge is represented. The primary antecedent causes for events are states, goals, and actions, with goals being most frequent, The primary consequents are actions and state changes. When the kind of events generated are normalized, the forward (And Then) sequence reflects almost directly the setting and episodic chain of a story schema (Mandler & Johnson, 1977; Rumelhart, 1975; Stein & Glenn, 1979; Thomdyke, 1977). When backward, antecedent causes are sequenced, goals become the modal cause, consistent with the general view that stories are goal-directed, problem-solving sequences (Black & Bower, 1980; Rumelhart, 1977, 1980; Stein, 1978). Similar findings were reported from story generation tasks. Thus, the knowledge a child has about narrative events was shown to correspond to knowledge about story schemata, which describe narrative events and the causal relations among the events. This knowledge seems to occur quite early in development. Our study focusing on understanding of emotional states suggests that children as young as 3 years of age are quite sophisticated about events that precipitate and follow emotional reactions. These young children make differentiations and classifications of emotional states along lines found in theoretical analyses of emotional terms from the psychoanalytic perspective (Dahl & Stengel, 1978) and by adults in general (Davitz, 1969). ”
Children’s Knowledge of Events
219
Finally, children’s ability to use the consequences as justification seems to occur much later than their ability to provide causal antecedents and consequents of an event. They clearly understand questions to produce antecedents, and they can generate consequents when proceeding in a forward direction. However, children preferred answering “HOWdo you know” questions with antecedent responses rather than consequents. These results may support Piaget’s claim that the ability to give “proof” or direct evidence for an event may be rather late in developing. We shall have to wait for further investigation, because, again, the question posed may prove to be the difficulty rather than the inability to give proof. It is clear from our results, however, that young children rarely juxtapose antecedents and consequents. If anything, the youngest children prefer causal antecedents when given a choice between antecedents and consequents. Only in situations where the cause of an event is difficult to generate do children give consequents as evidence of proof or justification. Thus the ability to reason logically, in terms of both causes and justifications, may be a function of a developing knowledge base, rather than a qualitative shift in the type of cognitive operations used. To date, we have found few developmental differences in narrative comprehension. These data, however, should not be used to suggest that there are no developmental differences in story understanding. We (Stein & Trabasso, 1981) have suggested three areas where critical differences may exist:
I . Procedural knowledge necessary to activate story schemata (e.g., knowing when, how, and where to use appropriate schematic knowledge). 2. Conceptual knowledge of stories (e.g., those dimensions that must be included in a piece of prose in order to classify it as a story). 3 . Specific content knowledge about story themes (e.g., knowledge children have about preconditions necessary for successful goal attainment). These areas may be more important and critical for investigation of children’s developing knowledge of stories. Certainly, the differences are not to be found in children’s lack of ability to construct causal representation of narrative sequences. REFE RENCES Baker, L., & Stein, N. L. The development of prose comprehension skills. In C. Santa & B. Hayes (Eds.), Children’s prose comprehension: Research and Practice. Newark, Del. : International Reading Association, 1981 .
Tom Trabasso el al.
280
Bartlett, F. C. Remembering: A stuiiy in experirnentcrl and social psychology. London: Cambridge University Press, 1932. Black, J . B., & Bower, G. B. Story understanding as problem solving. Poetics, 1980.9, 223-250. Black, J . B., & Wilensky, R. An evaluation of story grammars. Cognitive Science, 1979, 3, 213-230. Binet, A,, & Henri, V. La memoire des phrases (memoire des idees). L’AnnCe Psychologie, 1894,1, 24-59. Borke, H. Interpersonal perception of young children: Egocentrism or empathy? Developmentul Psvchology, 1971, 5, 263-269. Brown, A. The construction of temporal succession by preoperational children. In A. D. Pick (Ed.), Minnesota Symposium on Child Psychology Vol. 10). Minneapolis: University of Minnesota, 1976. Bullock, M . , & Gelman, R. Preschool children’s assumptions about cause and effect: Temporal ordering. Child Developinent, 1979, 50, 89-96. Comgan, R. A scalogram analysis of the development of the use and comprehension of “because” in children. Child Development, 1975, 46, 195-201. Dahl, H. D., & Stengel, B. A classification and partial test of de Rivera’s decision theory ofemotion. Psychounulysis und Contetnporury Thoughr. 1978, I , 269-3 12. Davitz, J . R . 7he language of emotion. New York: Academic Press, 1969. Day, J . , Stein, N . L., Trabasso, T . , & Shirey, L. A srudyof,finferentiulcomnprehension:The use oJa story schema to remember pictures. Paper presented at the biennial meeting of the Society for Research in Child Development, San Francisco, March 1979. de Rivera, J. A structural theory of emotions. Psychological Issues, 1977, Monograph No. 40. Feshbach, N. D., & Roe, K . Empathy in six and seven year olds. Child Devrlupment, 1968, 39, 133-145. Fraisse, P. The psychology oftime. New York: Harper. 1963. Friendly, M. la. In search of the M-gram: The structure of organization in free recall. Cognirive Psychology 1977,9, 188-249. Gelman, R. Cognitive development. Annual Revien, of Psychology, 1978, 29, 297-332. Glenn, C. G., & Stein, N. L. Syntactic structures and real-world themes in stories generated by children (Tech. Rep.). Urbana-Champaign: University of Illinois, Center for the Study of Reading, 1981, in press. Graesser, A . C. Prose cornprehension beyond the word. Berlin and New York: Springer-Verlag, 1981. Hood, L., & Bloom, L. What, when, and how about why: A longitudinal study of early expressions of causality. Monographs ufthe Socieryfor Reseurch in Child Developmen/, 1979,44(6, Serial No. 181). Johnson, H. M . Rival principles of causal explanation in psychology. Psvchologicul Review, 1939, 46, 493-516. Johnson, N . S . , & Mandler, J . M. A tale of two structures: Underlying and surface forms in stories. Poetics, 1980, 9, 51-86. Katz, E. W., & Brent, S . B. Understanding connectives. Journal of Verbul Learning and Verbul Behavior, 1968, 7, 501-509. Kuhn, D., & Phelps, H. The developinent of children’s understanding of causal direction. Child Ikvelopment, 1976.47, 248-25 I . Kun, A. Evidence of preschoolers’ understanding of causal direction in extended causal sequences. Child Development, 1978, 49, 218-222. Lehnen, W. G. The process of question crnswering. Hillsdale, N.J.: Lawrence Erlbaum, 1978. Mackie, J . L. The cement of the universe: A study of causation. Oxford: Clarendon Press, 1979. I
Children’s Knowledge of Events
28 I
Mandler, J. M. A code in the node: The use of story schema in retrieval. Discourse Processes, 1978, 1, 14-35. Mandler, J. M. Structural invariants in development. In L. S . Liben (Ed.), Piaget and the foundations of knowledge. Hillsdale, N.J.: Lawrence Erlbaum, 1981. Mandler, J. M . , & Johnson, N. S . Remembrance of things parsed: Story structure and recall. Cognitive Psychology, 1977, 9, 11 1-151. Mandler, J. M., Scribner, S . , Cole, M., & DeForest, M. Cross cultural invariance in story recall. Child Development. 1980, 51, 19-26. Margairez, E . , & Piaget, J. La structure des recits et I’interpretation des images de David chez I’enfant. Archives de Psychologie, 1925, 19, 21 1-239. Mendelson, R., & Shultz, T. R. Covariation and contiguity as principles of causal inference in young children. Journal of Experimental Child Psychology, 1976, 22, 408-412. Nezworski, T., Stein, N. L., & Trabasso, T. S t o n structure versus content effects on children’s recall of evaluative inferences (Tech. Rep. No. 129). Urbana-Champaign: University of Illinois, Center for the Study of Reading, June 1979. Piaget, J . The lmguage and thought of the child. New York: Meridian, 1955. (Originally published, 1925). Piaget, J. Judgment and reasoning in the child. Totowa, N.J.: Littlefield, Adams, 1969. (Originally published, 1928.) (a) Piaget, J . The c h i l J s conception of titne. London: Routledge & Kegan Paul, 1969. (Originally published, 1928.) (b) Propp, V. Morphology of the folktale. Austin: University of Texas Press, 1968. (Originally published, 1928.) Rumelhart, D. E. Notes on a schema for stories. In D. G. Bobrow & A. Collins (Eds.), Representation and understanding. New York: Academic Press, 1975. Rumelhart, D. E. Understanding and summarizing brief stories. In D. LaBerge & J. Samuels (Eds.). Basic processes in reading: Perception and comprehension, Hillsdale, N.J.: Lawrence Erlbaum, 1977. Rumelhart, D. E. On evaluating story grammars. Cognitive Science, 1980, 4, 209-241. Schank, R. C., & Abelson, R . P. Scripts, plans, goals and understanding. Hillsdale, N.J.: Lawrence Erlbaum, 1977. Shultz, T. R., & Mendelson, R . The use of covariation as a principle of causal analysis. Child Development, 1975, 46, 394-399. Siegler, R. S. Defining the locus of developmental differences in children’s causal reasoning. Journal of Experimental Child Psychology, 1975, 20, 5 12-535. Siegler, R. S . The effects of simple necessity and sufficiency relationships on children’s causal inferences. Child Development, 1976, 47, 1058- 1063. Siegler, R. S., & Leibert, R. M. Effects of contiguity, regularity, and age on children’s causal inferences. Developmental Psychology, 1974, 10, 574-579. Simon, R. A . The clevelopmenr of causol reasoning in the narrative. Unpublished master’s thesis, University of Minnesota, 1980. Stein, N. L. The comprehension and appreciation of stories: A developmental analysis. In S . Madeja (Ed.), The arts cognition, and basic skills. St. Louis: Cemrel, 1978. Stein, N. L. How children understand stories. In L. Katz (Ed.), Current topics in early childhood education (Vol, 2). Norwood, N.J.: Ablex, 1979. Stein, N. L., & Glenn, C. G . An analysis of story comprehension in elementary children. In R. D. Freedle (Ed.). Nerc, directioris in discourse processing (Vol. 4). Norwood, N.J.: Ablex, 1979. Stein. N . L.. & Trabasso, T. What’s in a story: Critical issues in story comprehension. In R. Glaser (Ed.), Advances in the psvchology of insrruction. Hillsdale, N.J.: Lawrence Erlbaum, 1981,
2x2
Tom Trabasso et al.
Surbey , P. D. Prc)st,hool rhildrtw’.~iirtdersruwding of eti~otiimilsfares in irrrrrs of cuiisrs i t r i d cortseyrtertwc. Unpublished master’s thesis, University of Minnesota, 1979. Thorndyke, P. W. Cognitive structures in comprehension and memory of narrative discourse. Cugnirive P~vchology,1Y77. Y, 77-1 10. Trabasso. T., & Nicholas. D. W . Toward a taxonomy of inferences for story comprehension. In F. Wilkening, I . Becker, & T. Trabasso (Eds.), Irfortrrcrfiort iriregruriori by children. Hillsdale, N.J.: Lawrence Erlbaum, 1980. Wellman, H . M . , & Johnson. C. N. Understanding of mental processes: A developmental study of “remember” and “forget. ” Child Developmertr, IY79, 50, 79-88. Werner, H . . & Kaplan, B . Sytnbuljormariott. Ncw York: Wiley, 1963.
Coding, see Short-term retention codes Comprehension, inferences and, 82 developmental considerations and, 85 interactionist position on, 82-84 memory representations and, 84-85 terminology and taxonomy and, 86-87 Conditioned attention theory (CAT), 1-3 classical conditioning and blocking, 37-39 conventional, 34-37 transfer, 39-41 evidence for, 16-23 of latent inhibition, 3-13 conditioning of inattention and, 14 modulation of attention and, 14-16 relationship to other learning paradigms, 23-24 feature positive effect, 33-34 habituation, 25-28 learned helplessness, 30-33 sensory preconditioning, 28-30 Context, category size and, see Category size Conversations, inferences and, 101- 103 Courtroom testimony, inferences and, 102 Cultural expectations, inferences and, 94
A Accommodation, comprehension and, 82-83 Advertising. inferences in, 104 Anaphoric reference, inferences and, 89-90 Assimilation. comprehension and, 82-83 Attention, .see Conditioned attention theory Audience effects, on prose processing, schemas and inferences and. 98
B Blocking, conditioned attention theory and, 37-39
C CAT, x r Conditioned attention theory Category size, 129-131 conceptual framework and, 145-146 processing assumptions and, 150-153 retrieval assumptions and, 153-158 structural assumptions and, 146-149 cued recall and, 133 associative context cues and, 142-144 context cues absent and, 133-138 nominal vs. functional set size and, 144145 sensory context cues present and, 138-142 method for study of, 131-133 Chaining codes, 55-56 evidence for, 58-61 Classical conditioning, conditioned attention theory and blocking, 37-39 conventional conditioning, 34-37 transfer, 39-41
D Diagnosis, inferences and, 104-105 Discourse, inferences and comparisons of discourse types and, 107-108 conversations and, 101-103 ecological validity and, 103-104 judgment research and, 104-107 modality differences and, 107 structure of stories and, 99-101 type and force of discourse and, 99 283
284
Index
E Episodes. inferences i n , 99-100 Expectations. cultural, inference< and, 94
F Feature positive ettect, conditioned attention rheory and, 33-34
G
role of source information in, 96 social prototypes and, 97 98 varieties of discourse and comparisons of. 107-108 conversations, 101-103 ecological validity and. 103-104 judgment restarch and, 104- I07 rnodality differences and, 107 structure of stories and, 99-101 type and force of, 99 Informalion, source, schemas and inferences in social cognition and. 96 lnforniation processing. inferences in, see Inferences
Given-new stratcgy, inferences and. RR-89
H Habituation, conditioned attention theory and, 25 -28 Heuristics. inferences and, 105 Hindsight, inferences and, 106
I Impression formation, schemas and inferences in social cognition and, 98 Inattention, .see Conditioned attention theory Inferences Britain studies of, I O X - 1 17 comprehension and, 82 developmental considerarions and, 85 interactionist position on, 82-84 memory representations and. 8 4 4 5 terminology and taxonomy and, 86-87 processes and, 88 biasing effect of questions and, 9 1-92 given-new strategy and, 88-89 verification and comprehension studies and, 89-90 schema theory and, 94-95 theme and. 92-94 in social cognition, 95 audience effects on prose processing and, 98 causal inferences, 96-97 impression formation and, 98 personality and, 95
J Judgment, inferences and. 104-107
K Knowledge conscious, orthographic structure and, 190I 95 prior, inferences and, 92-93
L Language processing model. 164-166 Latent inhibition. conditioned attention theory of, 3-13 conditioning of inattention and, 14 modulation of attention and, 14-16 Learned helplessness, conditioned attention theory and, 30-33 Learning cognitive response, in persuasion, 215-2 16 verbal, self-generation in, 216-217 Letter and word perception, 163-164 language processing model and, 164- I66 orthographic structure and conscious knowledge and, 190-195 orthographic structure and recognition and, 166- 168 bigram frequency vs. regularity and, 168I78
Index frequency, regularity, and lexical status and, 188-190 lexical status and, 185-188 log bigram frequency vs. regularity and, 178-183 replication and, 183- I85 related research, 195-198
285
Memory, srr Self and Memory; Short-term retention codes Memory representations, inferences and comprehension and, 84-85
persisting tasks and, 220-223 self-generation of material and, 214-218 Recognition, orthographic structure and, 166I68 bigram frequency vs. regularity and, 168178 frequency, regularity, and lexical status and, 188-190 lexical status and, 185- I88 log bigram frequency vs. regularity and. 178-183 replication and, 183-185 Retention. w e Short-term retention codes Reward-based education, self-generation effect and, 217-218
0
S
Orthographic structure. see Letter and word perception Overconfidence, inferencea and, 106-107
Sample codes, 56-57 evidence for, 63-68 Schemas, inferences and, 94-95 audience effects on prose processing and, 98 causal inferences, 96-97 impression formation and, 98 personality and, 95 role of source information in, 96 social prototypes and, 97-98 Self and memory, 201-202 early views of, 209 Bartlett, 204-206 Clapdrede, 203-204 Freud, 202-203 Koftka, 207-209 facilitation of recall and egocentric perspective and, 2 18-220 persisting tasks and, 220-223 self-generation of material and, 214-2 18 scope and importance of self system and, 229, 232-233 episodic-semantic distinction and, 230-23 I implicit theories of memory and, 229-230 theoretical status of self and, 231-232 self-reference effect and, 2 10 Bower and Gilligan, 212-213 interpretations of, 213-214 Keenan and Baillet, 21 1-212 Markus, 21 1 Rogers and Kuiper, 210-21 I theoretical synthesis of
M
P Personality, information processing and, 95 Persuasion, cognitive response learning in, 215-216 Prose processing, audience effects on, schemas and inferences and. 98
u Questions, biasing effect of, 91-92
R Recall category size and, 133 associative context cues and, 142-144 context cues absent and, 133-138 nominal vs. functional set size and, 144145 sensory context cues present and, 138-142 facilitation of egocentric perspective and, 2 18-220
286
Index
explanation 0 1 effects, 225-229 propertic5 of self system, 224-225 three effects. 223-224 Scnsnry preconditioning. conditioned attention theory and, 28-30 Sentence verification, inferences and. 90 Short-term retention codes, 51 -53 classification of, 55 chaining. 55-56 meaning and, 57-58 sample, 56-57 test, 56 designs for study of. 53-55 evidence for chaining codea, 58-61 sample codes, 63-68 test codes. 61 -63 stiniulus organization and, 68-69 effects, 69-75 extensions, 75-76 Social cognition, schenias and inferences in, 95 audience effects on prose processing and, 98 causal inferences, 96-97 impression formation and, 98 personality and, 95 role of SOUICC information in, 96 social prototypcs and, 97-98
Stimulua organization, coding and, 68-69 effects. 69-75 extensions, 75-76 Story comprehension, 237-249, 277-279 causal rclarionshipa and, 249-269 causal understanding and, 269-274 logical justification and, 274-277 Story Structure, inferences and. 99 101
T Test code\, 56 evidence tor, 61 -63 Theme.;, inferences and, 92-94, 102 Transfer, conditioned attention theory and, 39-4 I
v Validity, ecological, inferences and, 103-104
W Word perception, see Letter and word perception