% Second Edition @

Learninc

1% “~ ec Waa lo tk wg a rf lL,
1 a) || Lev | ate ( |aeye ‘Cri
a8 - &@ | o | Bl
on |

JOHN R. ANDERSON
 7
Ae be
9
ian “a_.
q

a
oe
t ’ ¥ ~
amar :
i} 7 f
-
Van 1 _ : om a
Me / 2s
a..,
* iv ~_ i
4 oo 4 if —,
7 rom ond Ps 7 iy - 4 = ee —7a c,
™, é :
a i ’

©
4
7

ste

=
te
uns
»,

_
=

e
:

/
LEARNING AND MEMORY
AN INTEGRATED APPROACH
 LEARNING
AND MEMORY
AN INTEGRATED APPROACH

Second Edition

John R. Anderson
Carnegie Mellon University

JOHN yo sove Wied

NewYork ¢ Chi
Brisbane ¢ Sinae
on ste oe
ACQUISITIONS EDITOR Ellen Schatz
MARKETING MANAGER Charity Robey
SENIOR PRODUCTION EDITOR Deborah Herbert
DESIGNER David Levy
ILLUSTRATION EDITOR Anna Melhorn
PHOTO EDITOR Lisa Gee
COVER PHOTO Elle Schuster/The Image Bank

This book was set in 10/12 ITC Palatino Light by LCI Design and printed and bound
by Courier-Westford. The cover was printed by Phoenix Color Corp.

This book is printed on acid-free paper.

Copyright © 1995, 2000 by John Wiley & Sons, Inc. All rights reserved.

No part of this publication may be reproduced, stored in a retrieval system or transmitted

in any form or by any means, electronic, mechanical, photocopying, recording, scanning
or otherwise, except as permitted under Sections 107 and 108 of the 1976 United States
Copyright Act, without either the prior written permission of the Publisher, or
authorization through payment of the appropriate per-copy fee to the Copyright
Clearance Center, 222 Rosewood Drive, Danvers, MA 01923, (508)750-8400, fax
(508) 750-4470. Requests to the Publisher for permission should be addressed to the
Permissions Department, John Wiley & Sons, Inc., 605 Third Avenue, New York, NY
10158-0012, (212)650-6011, fax (212)850-6008, E-Mail: PERMREQ@WILEY.COM.
To order books or for customer service please call 1(800)225-5945.

Library of Congress Cataloguing in Publication Data:

Anderson, John R. (John Robert), 1947—
Learning and memory: an integrated approach / John R. Anderson.-
—2nd ed.
S annk,
Includes bibliographical references and indexes.
ISBN 0-471-24925-4 (cloth : alk. paper)
1. Learning, Psychology of. 2. Memory. 3. Conditioned response.
I. Title.
HD318.A53 1999
153.1—de21 99-32553
CIP
Printed in the United States of America

Ai) PS YH 5 ZY ey Di
To Lynne, for Everything
 4 =

ar atieri( 6 se cry cone. _ an = :

ie ORKUT | re ee .. waging ee oo a Cl '
Sey ree ee eS ge a bot

nesnats wites. ° | ’
lL Sea a sa Ve >: ae :
pI? FO Laks Law
CWeR fp 2 & ie= -onmiee
y Tee teggr Geak : .

5 '

‘Pad Doce tome ot & 10GTT OD rs o> oe pennies

bs Comtn Yadad Thy gen eee oe 7 —
z

Ths Lt @ ree?
; ated Sem sented *
ae ‘. \
— =—

+ —
ane a

Lapayht ©?_— WD ber hegre Vi pe -_ x

Siege ine Darks) uenyy AW aha tals =] ou re
Soo part eadmet us cary Fm remap qatar er§ ss
alia
ih ory tees oe by ee rere, Heep, onsltss we preectyoteyy
4 herent, aaupeip? ab, ponmed vive Feedors 48 andSata the205%
Feitty Ak atau eires tha pele wetterRiccipepen eho hipearose-— -
esa nernin Pag payed athe appriariare ger-erpe Gat te iw Reg :
Daten —s
(Agnarece Carte) Ruscwutsd res olan =
a OG ol aie ro tht Pullabet oheaiur eel
ata cong {fener b/w Biky fi ae + "
10158 GRR OMSL we CLO B-\iuit: Fea aan gL
iy nicer besteOwfoe eusatemnor-vepvied taal CebllSCreaa RRs = a

Librarys of Congtam Cutals eueng 0 Pebiligatiog Lrtut

Hens ep mony, 2 phat R. Gobet Boles), 9G 5 i+
LewrGag aul meta actin epeatent agp! each,
Sedet | : ;
i re
cr.
lecierhis relearn reterery es andl 2 >
(Ady SPL GON2S +6 tod of. apes ae
1 Lqweriing Tayxteiiouy ¢ 2 Meneey al

mom We
1 he
= S

—— = SS
~~ S71 a i

i Kd

Petares oy he Urned Gene oa

i¢ : :
643 2 «+ “
A - 1

7, V ’
“
my
at
7 vit
_
i
 Preface

Research on learning and memory has been part of psychology since it began as
a science in the 1800s. At the height of the behaviorist era, around 1950, learning
was perceived as the key issue in psychology. Behaviorists believed that under-
standing any aspect of human behavior depended on understanding how that
behavior was acquired. Learning was pushed somewhat from center stage by the
cognitive movement in the 1960s. Cognitive psychology emphasized understand-
ing the functioning of the mature cognitive system rather than understanding the
learning that led to the system. The cognitive movement also marked a major
schism in the field. Traditional learning research continued on animals, while
research on human memory became a major part of cognitive psychology. These
two research traditions have evolved as almost completely separate disciplines.
Research on animal learning and research on human memory must
address many of the same issues. For this reason their continued development
independent of one another is not satisfactory. Many colleges recognize the rela-
tionship between the two areas by offering a Learning and Memory course rather
than separate courses for learning and memory. The scientific community is also
starting to recognize that pursuing independent paths is not desirable. Animal
learning research has become more and more cognitive over the decades, and
animal learning theory is no longer cast in such stark behaviorist terms that it is
incompatible with the cognitive perspective. In addition, researchers on animal
learning have profitably borrowed methods and theory from human memory
research. On the other side, research on human memory is turning more and
more to issues concerning the neural basis of memory, which inevitably require
the use of animal models. Cognitive psychology has become increasingly inter-
ested in the adaptive function of human memory, which again brings in the bio-
logical perspective. Finally, cognitive psychologists are beginning to recognize
the centrality of learning to the understanding of memory. In the last ten years,
the two areas of research have come closer together.
This textbook on learning and memory examines the current state of the
traditional learning and cognitive fields, and identifies the exciting opportuni-
ties for the synthesis of ideas. Learning and memory are brought together in one

Vil
 Preface

textbook because of my firm belief that one cannot properly describe research in
either field without describing research in the other. Ideally, there should be
only one field to describe. However, research and theory still tend to divide into
that concerned with animal learning and that concerned with human memory.
In addition to reviewing the behavioral research and theory in the various
areas, the discussion of this book emphasizes several themes.
Two of these themes
contribute to the integration of research on animal learning and human memory.
One theme, that of understanding the neural basis of learning and memory, is
presented throughout the text at appropriate places. This accurately reflects the
current state of learning and memory research, as well as the excitement about
new advances in understanding the neural basis of learning and memory.
The second major theme is appreciating the adaptive character of learning
and memory. Learning and memory are processes that arise in all species as solu-
tions to the problem of adapting to the changing structure of the environment.
This functional view of the system is stressed throughout the book, and empha-
sis is placed on how different mechanisms can achieve the same function.
The first edition of the book was, in part, an effort to gain a perspective on
my own intellectual history. I entered psychology in the 1960s, interested in ani-
mal learning. I joined the cognitive movement and spent years doing research
on human memory. Dissatisfied with the rather narrow perspective this research
gave on issues, I gradually shifted to research on cognitive skills and education-
al applications. As part of this experience, I have come to appreciate that human
learning and memory are part of an adaptation of the human to the environ-
ment. These adaptational interests brought me back to the research on animal
learning that I had left more than 20 years earlier. I found that there were many
new and deep analyses in that field, many of which were quite compatible with
the issues that I was dealing with in my own research. Although the first edition
was a textbook and not a monograph, these perspectives did clearly color the
story that I told in that book.
In preparing the second edition I had two major goals. The obvious one
was to update the book to include the new directions in research over the past
five years. Although there are a wide variety of updates, it is clear that research
from cognitive neuroscience forms the largest fraction of the additions. The sec-
ond goal is perhaps not so obvious. The first edition had some unnecessary fea-
tures of a research monograph. I have tried to smooth these out and make the
book more even at the undergraduate level. However, I have been mindful not
to sacrifice the intellectual character of the first edition. Some of my morte grat-
ifying moments of the past few years have involved the compliments I have
received on the book’s character.

Plan of the Book

Chapter 1 reviews the history of learning and memory research. It explains ideas
and paradigms that have dominated the field and why they became prominent.
 Preface

It also reviews some of the ideas that have failed. These past efforts have set the
stage for current research, enabling rather dramatic progress toward an under-
standing of human learning and memory.
The rest of the book describes the modern understanding of learning and
memory. Chapters 2 through 4 are devoted principally to research on animal
learning: Chapter 2 addresses classical conditioning, Chapter 3 describes instru-
mental conditioning, and Chapter 4 discusses issues of reinforcement. Although
there have been some useful human studies in these areas, animal research is
the basis for most current progress. This section of the book reflects this fact,
while also noting applications of this research to humans. The principal contri-
bution of this research to human learning is probably the establishment of a
better context in which to understand human learning and memory.
The remainder of the book focuses on human learning and memory,
although in many instances the later chapters turn to animal research for per-
spectives. Chapters 5 through 8 review the rather sophisticated understanding
of the nature of memory that has evolved over the past four decades. Chapter 5
examines how information is processed and stored in temporary memories
when it is initially received. Chapter 6 investigates how a permanent record of
this information is built up in long-term memory. Chapter 7 explores how infor-
mation is maintained over potentially long periods of time and what underlies
forgetting. Chapter 8 looks at the different ways in which information can be
retrieved when it is needed.
The final three chapters examine issues of learning and memory in a larg-
er context. Chapter 9 reviews the learning phenomena that arise when poten-
tially complex skills are acquired, such as learning how to use a word processor.
This topic contrasts with most of the research described earlier in the book,
which is concerned with learning simple facts and behaviors that are more
tractable for experimental study. Chapter 10 focuses on issues of inductive
learning, how people discover things about the structure of their environment.
This chapter returns to some issues of the early conditioning chapters, but it
employs the perspective of the human situation. The final chapter is concerned
with the major application of research on learning and memory to education,
bringing together many of the ideas discussed in the preceding chapters.
This book incorporates a number of features to help achieve instructional
and learning goals. The terminology of the field can be a bit overwhelming. |
have tried to minimize unnecessary jargon, but part of the educational mission
of this book is to introduce the student to the important concepts. Each impor-
tant new term is highlighted in the text and explained. These definitions are
brought together in a glossary at the end of the book. I have also concluded each
section of every chapter with a summary statement that identifies the main
point of the section. This provides students with a way to check their under-
standing of the sections and to quickly identify the content of a chapter. To assist
the instructor, Michael Toglin of SUNY at Cortland has prepared an Instructor's
Manual that includes chapter outlines, teaching tips, media resources, and a test
bank.

1x
 Preface

Acknowledgments
My experience with Wiley has been exceptional. Karen Dubno was editor for the
first edition, and Ellen Schatz was editor for the second edition. They have been
earnest both about the need to be true to the intellectual character of the field
and about the needs of communicating that character to an undergraduate
audience. They recruited a set of reviewers who helped greatly in defining and
fine tuning this book. The reviewers of the first edition were Victor Agruso,
Drury College; Robert Allen, Lafayette College; John Anson, Stephen F. Austin
State University; William Baum, University of New Hampshire; Steven
Coleman, Cleveland State University; Robert Crowder, Yale University; Ellen
Gagné, Catholic University; Peter Graf, University of British Columbia; James
Grau, Texas A & M University; Robert Green, Case Western Reserve University;
Mike Grelle, Central Missouri State University; Douglas Hintzman, University
of Oregon; David Hogan, Northern Kentucky University; John Jahnke, Miami
University; Douglas Mandra, Francis Marion University; Philip Marshall, Texas
Tech University; Ralph Miller, SUNY-Binghamton; Tom Moye, Coe College;
Mitch Rabinowitz, Fordham University; Ken Salzwedal, University of
Wisconsin-Whitewater; Michael Scavio, California State University-Fullerton;
Richard Schmidt, UCLA; Robert Schneider, Metropolitan State College; Steven
Sloman, Brown University; John Staddon, Duke University; Edward Wasserman,
University of lowa; Fred Whitford, Montana State University. The reviewers of
the second edition were: Gary L. Allen, University of South Carolina; Sheree
Barron, Georgia College and State University; E. John Capaldi, Purdue
University; David M. Compton, Georgia College and State University; Edward
A. Domber, Drew University; L. Sidney Fox, California State University at Long
Beach; Kevin J. Kennelly, University of North Texas; Robert Madigan, University
of Alaska at Anchorage; Richard L. Marsh, University of Georgia at Athens;
Henry Morlock, State University of New York College at Plattsburgh; Gabriel
Radvansky, University of Notre Dame; Suparna Rajaram, State University of
New York at Stony Brook; Caren M. Rotello, University of Massachusetts;
Kenneth D. Salzwedel, University of Wisconsin at Whitewater; Mark C.
Samuels, New Mexico Tech; Micheal J. Scavio, California State University at
Fullerton; Sonya M. Sheffert, Central Michigan University; Chehalis M. Strapp,
Western Oregon University; Paul D. Young, Houghton College
Besides these many reviewers the expositional structure of the book owes
much to three individuals—Lael Schooler a former graduate student, Jay
Anderson my son, and Ann Boyton-Trigg, Wiley’s developmental editor for the
first edition. Finally, I thank my secretary, Helen Borek, who was responsible for
so many details of both editions while keeping track of all the other aspects of
my professional life.

JOHN R. ANDERSON
 Contents

] PERSPECTIVES ON LEARNING AND MEMORY 1

Learning and Adaptation 1

Behaviorist and Cognitive Approaches 3
Definitions of Learning and Memory 4
History of Research on Learning and Memory 6
Hermann Ebbinghaus (1850-1909) 7
Ivan Petrovich Pavlov (1849-1936) 9
Edward L. Thorndike (1874-1949) 12
Clark L. Hull (1884-1952) 16
Edward C. Tolman (1886-1959) 17
B. F. Skinner (1904-1990) 20
The General Problem Solver (Newell & Simon, 1961) 24
A Model of Memory (Atkinson & Shiffrin, 1968) 27
Neural Basis of Learning and Memory 30
The Nervous System 30
The Neuron 33
Neural Explanations and Information-Processing Explanations 35
Outline of the Book 37
Further Readings 38

2 CLASSICAL CONDITIONING 39

Overview 39
The Phenomena of Classical Conditioning: Eye Blink in Humans
39
Sensitization and Habituation 41
Conditioning and Awareness 42
What This Chapter Covers 43

xi
 Contents

Neural Basis of Classical Conditioning 44

Simple Learning in Aplysia (Sea Slug) 45
Classical Conditioning of the Eye Blink in the Rabbit 47
S—-S or S-R Associations? 49
Response-Prevention Paradigm 50
US Devaluation Paradigm 51
Sensory Preconditioning Paradigm 51
Second-Order Conditioning Paradigm 52
Conclusions 53
What Is the Conditioned Stimulus? 53
What Is the Conditioned Response? 54
Association: The Role of Contingency 58
Rescorla’s Experiment 58
Conditioned Inhibition 60
Associative Bias 61
Conclusions about the Nature of the Association 62
Conditioning to Stimulus Combinations 63
Blocking 63
Configural Cues 64
Conclusions 65
The Rescorla—Wagner Theory 65
Application to Compound Stimuli 66
Application to Blocking and Conditioned Inhibition 68
Problems with the Rescorla-Wagner Theory 70
Neural Realization: The Delta Rule 72
Final Reflections on Classical Conditioning 75
Further Readings 77

INSTRUMENTAL CONDITIONING 78

Overview 78
Classical and Instrumental Conditioning Compared 79
What This Chapter Covers 80
What Is Associated? 80
Associations Between Responses and Neural Outcomes 81
Secondary Reinforcement 82
What Is the Conditioned Stimulus? 83
Generalization 84
Discrimination 86
Spence’s Theory of Discrimination Learning 88
Relational Responding: Transposition 89
Dimensional or Attentional Learning 90
Configural Cues and Learning of Categories 93

Xii
 Contents

What Is the Conditioned Response? 94

Maze Learning 95
Response Shaping and Instinctive Drift 97
Autoshaping 98
Association: Contiguity or Contingency? 99
Superstitious Learning 101
Partial Reinforcement 102
Learned Helplessness 103
Associative Bias 104
Instrumental Conditioning and Causal Inference 106
Application of the Rescorla-Wagner Theory 107
Interpretations 108
The Hippocampus and Conditioning 109
The Nature of Hippocampal Learning 112
Long-Term Potentiation (LTP) 114
Long-Term Potentiation and Hippocampal Learning 115
Final Reflections on Conditioning 116
Further Readings 117

REINFORCEMENT AND LEARNING 118

Some Basic Concepts and Principles 118

Rational Behavior 119
Effects of Reinforcement on Learning 121
Reward and Punishment 123
Aversive Control of Behavior 125
Punishment 125
Negative Reinforcement 129
The Nature of Reinforcement 130
Drive-Reduction Theory 130
Premack’s Theory of Reinforcement 132
Neural Basis for Reinforcement 134
Equilibrium Theory and Bliss Points 134
Studies of Choice Behavior 137
Schedules of Reinforcement 137
Variable-Interval Schedules and the Matching Law 139
Momentary Maximizing 140
Probability Matching 142
Optimal Foraging Theory 143
Effects of Delay of Reinforcement 145
Mechanisms of Choice 148
Human Decision Making 148
Final Reflections 150
Further Readings 151

Xill
 Contents

5 TRANSIENT MEMORIES 152

Conditioning Research Versus Memory Research 152

Animal Research Versus Human Research 153
Sensory Memory 155
Visual Sensory Memory 155
Auditory Sensory Memory 157
Conclusions about Sensory Memory 159
The Rise and Fall of the Theory of Short-Term Memory 160
Effects of Rehearsal 160
Coding Differences 162
The Retention Function 164
Conclusions about Short-Term Memory 166
Rehearsal Systems 166
The Phonological Loop 167
The Visuo-spatial Sketch Pad 169
Working Memory and the Central Executive 171
The Sternberg Paradigm 172
Rehearsal Processes in Lower Organisms 175
The Neural Basis of Working Memory 178
Neural Imaging of Working Memory in Human 180
Final Reflections 183
Further Readings 184

6 ACQUISITION OF MEMORIES 185

Stages of Memory 185

Practice and Trace Strength 186
The Power Law of Learning 187
Repetition and Conditioning 191
Long-Term Potentiation and the Environment 192
Significance of a Power Function 195
Elaborateness of Processing 197
The Generation Effect 198
Differences Between Elaboration and Strength 200
Incidental Versus Intentional Learning 201
Implications for Education 202
The Structure of Memory 203
The Brain and Memory 203
An Abstract Representation of Permanent Memory 205
Priming 206
Chunking 207
Representation of Knowledge 210
Memory for Visual Information 211

XIV
 Contents

Effects of Imagery 214

Meaningful Memory for Sentences 215
Differential Decay of Sensory and Semantic Information 216
Kintsch’s Propositional Theory of Text Memory 218
The Bransford and Franks Study 219
Memory Representation in Other Species 221
Sequential Memory of Pigeons 221
Representational Structures in Primates 222
Final Reflections 223
Further Readings 225

/ RETENTION OF MEMORIES 226

Overview 226
The Retention Function 227
Decay: The Power Law of Forgetting 228
Degree of Learning and Forgetting 231
Environmental and Neural Bases for the Power Law of Forgetting 232
Spacing Effects 234
Spacing Effects on the Retention Function 237
Spacing Effects in the Environment 238
Interference 239
Item-Based Interference 241
A Theory of Associative Interference 243
Relationship to the Rescorla-Wagner Theory 245
Recognition Memory and Multiple Cues 246
Item Strength and Interference 248
Interference with Preexperimental Memories 249
Context-based Interference 252
Is All Forgetting a Matter of Interference? 254
Retention of Emotionally Charged Material 256
Freud’s Repression Hypothesis 256
Arousal and Retention 257
The False Memory Syndrome 259
Eyewitness Memory and Flashbulb Memories 260
Final Reflections 262
Further Readings 264

§ RETRIEVAL OF MEMORIES 265

Overview 265
The Relationship Between Various Explicit Measures
of Memory 266
Recognition Versus Recall of Word Lists 268
 Contents

Retrieval Strategies and Free Recall 270

Mnemonic Strategies for Recall 271
Evaluation of the Generate-Recognize Theory 273
Measuring Recognition Memory: The High-Threshold Model 276
Signal Detectability Theory 276
Conclusions about Recognition Versus Recall 279
Interactions Between Study and Test 279
Context Dependency of Memory 279
State-Dependent Memory 280
Mood-Dependency and Mood-Congruence Effects 282
Encoding-Specificity Principle and Transfer-Appropriate Processing 284
Reconstructive and Inferential Memory 285
Inferential Intrusions in Recall 287
Conclusions About Study—Test Interactions 289
Explicit Versus Implicit Memories 290
Feeling of Knowing 290
Familiarity 291
Retrieval Facilitation 294
Interactions with Study Conditions 295
Amnesia in Humans 298
Selective Amnesia 301
Final Reflections 302
Further Readings 303

SKILL ACQUISITION 304

Overview 304
Power Law Learning 307
Stages of Skill Acquisition 310
The Cognitive Stage 311
Difference Reduction 312
Operator Subgoaling 314
The Associative Stage 319
The Conversion of Problem Solving into Retrieval 320
Production Rules 322
The Knowledge-Intensive Nature of Skill 324
The Autonomous Stage 325
The Motor Program 326
Noncognitive Control 328
Generality of Motor Programs 329
Learning of Motor Programs 330
Tuning of Motor Program: Schema Theory 331
The Role of Feedback 334
Final Reflections 336
Further Readings 337

Xvi
 Contents

10 INDUCTIVE LEARNING 338

Overview 338
Concept Acquisition 340
Concept-Identification Studies 341
Hypothesis Testing 343
Natural Concepts 346
A Schema Theory: Gluck and Bower 349
An Exemplar Theory: Medin and Schaffer 350
A Pluralistic View of Concept Acquistion 352
Causal Inference 354
Statistical Cues 355
Cues of Spatial and Temporal Contiguity 357
Kinematic Cues 361
Understanding Complex Systems 362
Conclusions about Causal Inference 364
Language Acquisition 364
Character of Language Acquisition 366
Theories of Past-Tense Acquisition 368
A Critical Period for Language Acquisition 370
Innate Language-Learning Abilities 371
Animal Language Learning 372
Final Reflections 375
Further Readings 376

17 APPLICATIONS TO EDUCATION 377

The Goals of Education 377

Reading 379
Mathematics 380
Psychology and Education 383
The Behaviorist Program 383
Mastery Learning 386
The Cognitive Approach 387
Reading Instruction 387
Nature of the Adult Skill 388
Phonetic Decoding Skills 391
Comprehension Skills 393
Conclusions about Reading Instruction 396
Mathematics Instruction 397
Basic Arithmetic Facts 399
Multicolumn Subtraction 400
Algebraic Word Problems 401

xvii
 Contents

Geometric Proof Skills 405

Intelligent Tutoring Systems 407
The Role of Mathematics in Life 411
Final Reflections 413
Further Readings 414

GLOSSARY 415

BIBLIOGRAPHY 426

PHOTO CREDITS 467

AUTHOR INDEX 469

SUBJECT INDEX 481

XVili
Pee. AND and,
AN INTEGRATED APPROACH
 rite =o

{) Yo
5KOna <]
Ti ees.
Gj

ye
SAOATHA ‘Wath
495 7 > =% en
~ <«@s

OGRAPHY “Die:
PHOTO CRE/NTS te?

AUTTIDEINDEX 46°

CT INDE x at eS

7 fay : -
/ ; i « a a
— -_ 2 ere i- mt 4 - ov
a

<<

3 ” _—

ii “yg .

’ atdes

i ae a
gt nee

¥
7 iri one

ni 45 7
 Perspectives on Learning
and Memory

Learning and Adaptation

Learning is a crucial activity in a human culture. The very existence of a culture
depends on the ability of new members to learn sets of skills, norms of behav-
ior, facts, beliefs, and so on. People create educational institutions devoted to
learning and invest a substantial fraction of their resources in them. They spend
a large proportion of their lives learning to do things rather than doing them.
ey can learn to
live in the world of the Stone Age tribes in New Guinea and in the weightless
world of an astronaut orbiting the earth. Of course, humans have no monopoly
on learning. Primitive creatures are capable of some degree of learning, as are
certain computer programs today. However,

becomes activat
Through |

be innately specified? S
 CHAPTER 1 Perspectives on Learning and Memory

to enable behaviors to be shaped through an evolutionary process. When the

changes,ment
environ the behaviors that served onegeneration will not serve
the Berra hore beetle newroutesto food every year, and
humans must adapt to technological revolutions every generation. The advent
of the automobile, for instance, required that humans learn a set of behaviors
not anticipated in their evolutionary history. behavi
Species’ ors
are shaped by
learning to the extent that their environments are complex and changing. The
more variable the environment, the more plastic the behavior must be. _
Species can be placed on a dimension of behavioral plasticity. For some
species most behaviors
are innately specified; others arecapableof learning a
ereat many new behaviors. The species with the greater learning capability is not
necessarily at the advantage in terms of its ability to survive. As an example of
three creatures on this continuum of behavioral plasticity, consider the cock-
roach, the rat, and the human. The'cockroachis capable of learning only the sim-
plest things, such as avoiding a dangerous area; the rat can learn a lot more about
the nature of its environment and for this reason has been a favorite laboratory
animal for studies of learning; the human is proportionately still more plastic.
Despite their vastly different learning abilities, all three creatures inhabit modern
cities, and one has not proved notably more successful than the others in terms
of the survival of its species. Within the city, they occupy different niches that are
quite different in the range of possible behaviors. Cockroaches live mainly with-
in walls and survive using basic instincts, such as fleeing light and seeking tight,
crowded places, that have served them for 320 million years (and that serve them
well in modern apartments). The behavior of rats is richer. Rats are capable of
exploiting the knowledge they acquire about their environment, such as various
paths between locations and where food is to be found. The behavior of humans
is even more complex, particularly considering the human potential for using a
wide variety of artifacts, from light switches to pesticides, Itis the potential com-
plexity of the behavior that creates the demand for learning.
Anespecially important dimension of complexity in the human environ-
ment produced
is bythe artifacts or tools created by humans themselves. City
dwellers (and to a large extent rural dwellers) live in an environment almost
totally of their own fashioning and one that is far different from the environ-
ments of just 100 years ago. A common belief is that the human capacity for
complex learning is responsible for tool use, but archaeological evidence sug-
gests the reverse. Small-brained human predecessors started using tools. Only
after tool use was well established did brain size increase in our evolutionary
ancestors. Tool use created a more complex environment that required greater
learning capacity. Once learning capacity increased, tools became even more
complex, creating a_snowball_effect:
morecomplex environments demanded
more_learn which created
ing, more complex environments, and so on. The
snowball has in some sense spun out Of control in modern society—technology
has created an environment of great dangers (drugs, environmental hazards,
nuclear weapons, etc.), which we have not learned to manage and which we
have had no time to adjust to through evolution.
 Behaviorist and Cognitive Approaches

#
Behaviorist and Cognitive Approaches
The title of this book is Learning and Memory. The following section of this chap-
ter offers definitions of these two terms. Though related in their meanings, the
two words refer to separate lines of research in psychology. Learning
is associ-

viorof ar ce to what
pat Dehappening te mindof thearganisin. They held that speaking about
things happening in the minds of lower organisms, such as rats, was unscientif-
ic, and they thought only a little better of attributing minds to humans.
Behaviorism dominated American psych
Some ofthe Key’ideas of behaviorism are reviewed iina later section of this chap-
a iioners.

n learning took‘place withnonhuman

° ie iouiatn arose at the turn of the century when there was still great
excitement about the new ideas-surrounding evolution. Darwin had
argued that humans were continuous with other animals; it was believed
that the laws of learning that held for animals would also hold for
humans.
e Animals might allow researchers to study learning in a purer form, uncon-
tied by bycul
culture and lan
eg

straints
ints than ee
seperformed on humans. .
A major theoretical shift‘in
in psychology began in the 1950s based, in part,
on the beliet et = behaviorists had created too simple a picture of human

greater “proportion of ravines since the 1950s, — ee behavioral

research, a correspondingly smaller proportion. Cognitive ps} og udic
but they did so in the guise of so-called memory experiments on
 CHAPTER 1 Perspectives on Learning and Memory

human subjects. A typical experiment might involve having subjects study a sec-
tion from a textbook such as this one and later testing the subjects to see what
they cou

the pairingfs on and memory in the title of this t book. The separation of
these two traditions is fundamentally artificial and has begun to break down.
Much current research on“animal learning has a strong cognitive orientation,
and there has been a resurgence of more behavioristic learning theories in
research on human memory.
This book covers the research from both traditions but takes note of the
syntheses occurring in the field today. Throughout, the emphasis is on the sig-
nificance of research results for understanding human learning and memory.
The basic purpose of this chapter is to describe the traditional approaches to
studying learning and memory, thereby setting the stage for the remaining
chapters, which center on what these approaches have revealed. Before pro-
ceeding, we must turn to the thorny issue of defining the terms learning and
memory.

Definitions of Learning and Memory

Most people feel they have a good sense of what is meant by the terms learning
and memory. However, it can be frustratingly difficult to specify the precise
meaning of these terms. The following is the most commonly offered definition
of learning:

ere The qualification that change is Selatively_permansnt is

designed to exclude certain transient changes that do not seem like earning.
 ® Definitions of Learning and Memory ¢

Fatigue is a simple example of what learning theorists want to exclude. A per-

son who performs a task repetitively may become tired, resulting in a change in
performance. With rest, the individual returns to the original performance level.

® Behavioral. Until recently, psychologists have needed some external mani-

festation of the learning in the individual’s behavior. If a person learns some-
thing, but it does not affect that person’s behavior be ept secret, how
is a psychologist to know that it was learned?

s we will discuss in subsequent chap-

ters, recent advances in neural recording have allowed psychologists to”see” the
learning that is happening in the minds of their subjects.

° Potential. Not everything we learn has an impact on our behavior. An indi-

vidual may learn another person’s name but never have occasion to use it. Thus,
psychologists do not demand a spontaneous change in behavior, only a change
in the potential for behavior. The psychologist must devise a behavioral test to
tap this potential and show that learning has taken place. For instance, to deter-
mine whether a pet a eee a a it is often necessary to offer that pet a
a tinction be 1 le and _performance

Experience
7 we age, our bodies develop and our " potential oF ores chaos but we
e would not want to consider physical growth as learning. Similarly, a serious
x injury might substantially change a person’s potential for behavior, oe we
be wu not want to consider EELS, an arm asseein ng. The 1 xperience is

This definition of memory depends on the definitionoflearning. However, it

RG that is
i not includ in the definition of learning, the term

and that people just a | ways of behaving. Behaviorists preferred theories

that were couched only in behavioral terms and distrusted references to men-
talistic constructs such as a memory record. The reluctance to discuss such

©
 CHAPTER 1 Perspectives on Learning and Memory

memory constructs has largely disappeared, now that researchers are beginning
to understand the neural changes that embody the memory records. When it is
possible to speak of some precise neural change rather than some vague men-
tal change, the behaviorists’ distrust of memory constructs begins to evaporate.
A portion of this book is concerned with the neural basis of memory.
MASONS SALA ERIESENATE
LAN NEILL S ALE IIL.

Reareay FACS toHEDprocess me maven of ean | to

experience, and memory refers tothe permanent 1records;
that
underlie this adaptation.
yee AER HSRC TAIN NE RESRE CREOLE LN OE TERESA ARSATES INE OEE SESOLES ODT SE ERIN
ELE LOL ELLE LLANES ELIDA LE ELLA

History of Research on
Learning and Memory
Psychology as a scientific field is only a little over 100 years old. From the begin- .
ning, learning has been an important area of research. One reason for the early
interest in learning was Charles Darwin's theory of evolution. The publication of
his On the Origin of Species in 1859 captured the imagination of the intellectual
world with its emphasis on how natural selection had changed species so that
they were better adapted to their environment. Learning theorists saw their ¢
research as the obvious extension of Darwin’s. Whereas Darwin was concerned 3)
with adaptation across generations of a species, learning theorists focused on 5 “t
the ongoing adaptation of an individual member of a species within its lifetime. A ¥
Understanding the relationship between species-general acapiation and indi-
vidual learning is still a current research topic.
Three research enterprises begun at the turn of the twentieth century
influenced much of the subsequent history of research on learning and memo-
ty. One was a series of studies undertaken byGerman the psychologist
Hermann
Ebbinghaus, who used himself as his sole subject. The second was a
sods ofstudies comdictad byaRussian physiologist,
physiologist Ivan
Ivan
Pavlov,
Pa on condition-
ingin dogs. The third was aseries of studies directed by anAmericanpsychol-
ogist, Edward —Thomdiks, on_trial--and- error Jeaming 1 in_cats. Pavloy | and
h

started a tradition of research that, after the cognitive revolution in the 1950s
and 1960s, became the dominant paradigm for the study of human memory. The
history of research on learning in the United States is really a history of the
research traditions started by these three individuals and how these traditions
interacted with the intellectual mood of American psychology.
The research of these three pioneers and of other influential psychologists
is presented next. The purpose of this historical review is twofold. First, it intro-
duces the methodologies
that are part ofcurrent research practice, Second, it
sets th
thee background
bac againstwhich current research and theory can be appreci-
 AST BAY
History of Research on Learning and Memory

ated, in particular, by showing that a number of ideas about learning and mem-
ory)were sorcidaed before the field settled on the current conceptions.
a aa a -

1e point where he was able to repeat the lists twice in order with-
out error. He then looked at his ability to recall these lists at various delays. He
measured the amount of time he needed to relearn the lists to the same criterion
of two perfect recitations. In one case it took 1156 sec to learn the list initially and
only 467 sec to relearn the list. He was interested in how much easier it was to
relearn the list. In this example, he had saved 1156 — 467 = 689 sec. Ebbinghaus
"expressed this savings as a percentage of the original learning: 689 + 1156 = 64.3
"* percent. Figure 1.1 tion of the delay

In another experiment, Ebbinghaus relearned the lists of nonsense sylla-

bles each day for six days. Figure 1.2 shows the number of trials needed to

60

sofo)
eo)

savings,
Time
%

w je)

FIGURE 1.1 Ebbinghaus’s retention function

showing the percentage of time saved as a func- 20 Nei ateyediemaT Y
tion of delay. Ebbinghaus used delays from 20 200°. 400. 600. 800
minutes to 31 days. Fouts of. delay
 CHAPTER 1 Perspectives on Learning and Memory

500

400

oS
%® 300
2)

2
2 200
ai no

s
iS
100

FIGURE 1.2 Ebbinghaus’s practice data, —

showing the total number of trials needed to 0
master a set of lists as a function of the number Sal BoE gly hae eS
Days of practice
of days of practice.

ich haves . Ebbinghau eoretical expla

or the en didnot have much influence on the reaming researen hal
followed.2 However, he did sow the seeds ofatradition of research on human
_memory that eventually became more prominent than the learning research on
animals.
EEEES ONESIES ee EEN EN CNN RENN CIRCE SOD SU Rea RN Re ER EE RESTS RCE

Ebbinghaus established Serenata nethouclone for study-

ingg
phenomena,
memory
memory such
such as
as the
the retention curve and the.
learning curve
HN ROS DRANG UN ESR RR TDN OS RG DN EN DIE NERO
NASB INS

‘Smaller numbers imply better performance in Figure 1.2, which uses a dependent
measure of trials to relearn; larger numbers imply better performance in Figure 1.1,
which uses a dependent measure of percentage of time savings.
*Ebbinghaus’s theory of remote associations was an exception. It was the dominant
theory of serial list learning until the modern era, when it was replaced by the chunk-
ing hypothesis (see Chapter 6). A series of papers discussing Ebbinghaus’s contribu-
tions was published in 1985 in the Journal of Experimental Psychology on the one hun-
dredth anniversary of his treatise.
 History of Research on Learning and Memory

Ivan Petrovich Pavlov (1849-1936)

_ As part of
his research, he put meat powder in adog’s mouth and measured salivation.He
discovered that after a few sessions precise ‘Measurement was ines
a

avlov’s original research involved food and salivation, a variety |

of USs and URs have been used to develop CRs. A frequently used paradigm

Time

(a) Initial pairing

Experimenter’s presentation CS US |
sequence: (bell) (food) |
aS \

Organism’s response: UR
(salivation) ;

(b) Conditioned response

Experimenter’s presentation cS US C)
sequence: (bell) (food)

Organism’s response: CR UR
(salivation) (salivation)

(c) Extinction \

Experimenter’s presentation cs
sequence: (bell) \

Organism’s response: is CR
(salivation)

FIGURE 1.3. Experimental procedure in classical conditioning. (a) CS is paired with

US that evokes a UR; (b) as a result the CS acquires the ability to evoke the CR; (c) the
CS can continue to evoke the CR for some time after the US is removed but will even-
- tually extinguish.
 CHAPTER 1 Perspectives on Learning and Memory

Pavlov observing the conditioning of a dog.

with humans involves conditioning an eye blink (UR), which occurs in response
to a puff of air to the eye (US). A light or tone repeatedly paired with the puff of
air acquires the ability to evoke an eye blink in the absence of the original US.
Eyelid conditioning is also a frequent paradigm with nonhumans.
A classical-conditioning paradigm that has received considerable research
focus in the last few decades involves the conditioned emotional response
(CER). When an animal, such as a fat, is presented *with an aversive stimulus, for
example, a mild shock, it responds in a characteristic way. Its heart rate,acceler-
ates, its blood pressure elevates, and it releases certain hormones. It also tends to
freeze and halt whatever response it has been performing. Parts of this response
pattern can be conditioned to a CS, such as a tone. To measure the CER,
researchers train an animal to perform some task, such as pressing a lever for
food; the degree to which the animal freezes and so reduces its rate of lever press-
ing when the CS is presented is taken as a measure of the strength of the CER. ||
Chapter 2 discusses contemporary research and issues involving the clas-
sical conditioning paradigm, but some of the basic phenomena established by
Pavlov(1927) are worth noting here: Stinaice
in Pie
4-Acquisition. The magnitude of the conditioned response can be mea-"-..
sured as a function of the number of pairings between the US and the CS.
The CR doessuddenly
not appear in full strength. Figure 1.4 illustrates
that the strength of the CR gradually increases withth
repetition. This is
referred to as the process 0 isition. The typical conditioning curve

10
 History of Research on Learning and Memory

15

i [e)

Salivation
CS
to
ol

osetia
se i RR
12-34. 5-6 7 — 89101 T 124s Ta 16 134056.
7 8 79°10
Acquisition trials Extinction trials

FIGURE 1.4 Acquisition and extinction of a conditioned response. (After Pavlov,

1927.) Source: From Introduction to psychology, Eighth Edition, by Rita L. Atkinson,
Richard C. Atkinson, and Ernest R. Hilgard. Copyright © 1983 by Harcourt Brace &
Company, reproduced by permission of the publisher.

obtained during acquisition shows a little increase at first, then a larger

increase, until some asymptotic Ievelisreached at which the rate of
increase tevels off. The pattern of initial slow conditioning, then rapid, and
then slow again is offen summarized in the term S-shaped curves.
Chapter 6 compares the conditioning functions (or curves) obtained in
classical conditioning with learning curves, such as those obtained by
Ebbinghaus (Figure 1.2). They are similar but not the same in that the con-
ditioning function often starts off with rather slow change, whereas the
learning curve almost always shows its most rapid change at first.
(2. Extinction. What happens when the US is no longer paired with the CS?
Figure 1.4 shows that the magnitude of the conditioned response gradually
decreases with the ntimber of trials in which
no US occurs. This is referred
to asprocess
the of extinction. The extinction function for conditioning is
similar to the retention
or forgetting function for memory (e.g., Figure 1.BLY.
However, as is true of the relationship between the acquisition functions for
memory and conditioning, there are differences. The most important is a
methodological difference between the experiments that produce the two
nctions. A forgetting function is obtained by waiting, without presenting a
stimulus; but extinction requires presentation of the CS without the US.
"6 Spontaneous Recovery. Some time (e.g., a day) after a series of extinction
trials, the CS can be presented again without the US..The
The magnitude of
of the
CR offen shows some recovery. This spontaneous recovery is one differ-
ence between forgetting and extinction, because there is seldom, if ever,
any spontaneous recovery from forgetting.

i
 CHAPTER 1 Perspectives on Learning and Memory

(im oral Ordering. Conditioning is strongest when the CS precedes the

US and often failsto occur if the USprecedes the CS. For instance, eyelid
conditioning is unsuccessful if the CS (tone) follows the US (puff of air).
Depending ontheresponse, theoptimal interval between the CSand the
US ‘can vary from 0.5 sec to 30 sec or more. The order dependency of con-
ditionmng isnot found inthe results from a typical memory experiment. In
a memory experiment, a subject might have to learn to say one word like
dog in response to another word like cream. The subject’s learning of this
fact does not depend on the order in which the two words are studied.
>»

Pavlov speculated about the neural basis of classical conditioning. He proposed

that neural excitation flowed from an earlier and weaker center in the brain to a
later and stronger center; that neural excitation in the brain center aroused by
the conditioned stimulus flowed to the brain center aroused by the uncondi-
tioned stimulus. The CS excitation evoked the response when it arrived at the
US center. Pavlov embellished this physiological proposal with many speculative
ideas that found little subsequent support. Several alternative theoretical analy-
ses have been offered over the years. Altho isagreeing with Pavlov on many
details, researchers have tended to regard classical conditioning as a direct
reflection of automatic neural processes of association. Because it is thought to
reflect automatic learning, classical conditioning has been a favorite paradigm in
studies seeking
to understand the neural basis of learning, as shown in Chapter
2. Classical conditioning has also gained popularity in such physiological
research because it can be displayed in primitive organisms, which are often
easier subjects for physiological study. For example, Chapter 2 includes a dis-
cussion of conditioning in the sea slug, whose nervous system is much easier to
study than that of mammals.
Despite the tendency to see classical conditioning as reflexive and auto-
matic, Chapter 2 shows that modern understanding of the phenomenon often
views it in a more cognitive and less reflexive light. Even today, however, classi-
cal conditioning is often regarded as the paradigm of choice for the study of
simple and basic learning processes.
SELENE
AR ELSI STEM STI AES MES EERE ERENN OR eee oe Sd

Pavlov discovered that when a neutral stimulus (CS)ispaired

with a biologically significant stimulus (US), the CS acquires
the ability to evoke
SOLEIL
associated
responses with the US.
ION GRRE PME ETAT SN IE LENE TN TCL SE EN CLR OMRON IPL ER RU eens

Edward L. Thorndike (1874-1949)

Thorndike studied a rather different learning situation than did Pavlov. His orig-
inal research was reported in 1898. Figure 1.5 illustrates Thorndike’s experimen-
tal apparatus, called a puzzle box. He placed a hungry cat in such a box with
some food outside: If the cat hit an unlatching device (e.g., a loop of wire), the
door would fall open and the cat could escape and eat the food. Cats were given

12
 History of Research on Learning and Memory

FIGURE 1.5 One of the four puzzle boxes used by Thorndike in his doctoral thesis.

repeated trials at this task, and’ Thorndike was interested in how quickly they
learned to get out of the puzzle box. Thorndike’s observation was that cats
would at first behave more or less randomly, moving about the box, clawing at
it, mewing, and so on, until they happened to hit the unlatching device by
chance. Over trials the random behavior gradually diminished as the cats head-
ed for the unlatching device sooner and hence were able to leave the box soon-
er. He referred to this as 'trial=and=errorlearningr Figure 1.6 shows typical learn-
ing curves relating number of trials to time to get out of the box.
Arguments continue as to whether Thorndike’s cats gradually learned
(over a series of trials) to get out of the box or whether they suddenly caught on
and learned on a single trialy Thorndike chose to see gradualness in these learn=
ing curves and proposed that the correct response of hitting the unlatching
device was gradually strengthened to the stimulus situation of being in the puz-
zle box, so that it came to dominate the other random responses. He thought
that this strengthening process was automatic and that it require
did not any
ivit.on the part of the cats.
The kind of learning process Thorndike studied is referred to as‘imstru-
ro mental conditioning, incontrast with Pavlov’s classical conditioning. In both
caséS, aresponse is learned toa stimulus situation. In classical conditioning, the
Ce timulus is the CS and the response is theCR. In Thorndike’s puzzle boxes, the
stimulus isthe puzzle box and the response is
the appropriate unlatching action.
~~ Both Kinds of Tearning show the phenomena of practice, extinction, and spon-
taneous recovery. In instrumental
re ee
conditioning, the response is performed to
i

18
 CHAPTER 1 Perspectives on Learning and Memory

120

60 60

0 0

e2 60
a: 180
raf
8
ray
2
2 0 120
=

60 60

0 )

Trials Trials

FIGURE1.6 Learning curves for five cats in Thorndike’s puzzle boxes. Source: From
Psychological Monographs, Volume 2 (Whole No. 8) by E. L. Thorndike. Animal intelli-
gence: An experimental study of the associative processes in animals. Copyright © 1898 by
The Macmillan Company. In the Public Domain.

obtain the

more volitiona O earnite but many share Thorndike’s belief that it is

every bit as automatic as classical conditioning. :
Thorndike remained an active researcher and theoretician throughout his
life. He was especially interested in applications of learning theory to education
and is associated with a number of principles of learning. The following three
principles are particularly important and identify issues that remain fundamen-
tal ee

ing.one he originally
formulated his tnsndastzechapsiioroans such as
— tood, as ralStimulus-response_ connections. (or associations), and»
learn punishments,
me such as shock,«weakened.them.
Thorndike that punishment was relatively
Later evidence convinced
ineffective in weakening

14
 History of Research on Learning and Memory

responses, but he maintained his belief that reinforcement was absolutely

critical to learning’ The position that reinforcement was necessary to
learning, common to a number of learning theories of his time, provoked
numerous attempts at experimental disproof. The most notable research
efforts were the latent learning experiments, discussed later in this chap-
ter in the section on Edward Tolman’s work.
(2 Law Exercise.
of Thorndike originally proposed that repeated practice of
a stimulus—response association strengthened it. Later he retracted this

d He pointedto experiments
such as that of Trowbridge and Cason (1932) in which subjects practiced
drawing 4-inch lines without any feedback and failed to improve in accu-
- : é:
“lead
improvement—for
to example, rehearsing material for a test. Indeed,
as discussed in Chapter 6, practice seems to be a fundamental variable in
_>learning and memory.
3. Principle of Belongingness, In 1933, Thorndike accepted the notion that
a) B.

some things are easier to associate than others because they belong
together, This was a concession to the
Gestalt psychologists
of the time
who argued that it is easier to associate things if they are perceived as
belonging together. Thorndike accepted the principle of belongingness
with some reluctance because it seemed to involve a cognitive component
in the mechanical process of forming an association. This idea of belong-
ingness has played a large role in modern theories of learning and mem-
ory. Animals have certain biological predispositions to associate things; for
instance, as discussed in Chapter 2, rats are especially prepared to associ-
ate taste with poisoning. Also, in the arena of human memory, the way”
e- for instance, as
shown in Chapter 6, memory for a pair of words is greatly enhanced if they
are imagined in an interactive visual image.
Thorndike
and Pavlov provided much of the inspiration:
for the behaviorist
movement
that dominated American psychology in the first half of the twenti-
eth century under-
ae:

most famous of the early behaviorists, was greatly influenced by Thorndike and
Pavlov. Watson argued that mental constructs such as decision making and
memory were excess baggage and that all human behavior could be understood
as the result of learned associations between stimuli and responses.

3As discussed in Chapter 4, Thorndike’s assessment of reinforcement and of the inef-

.
: j

fectiveness of punishment was incorrect.

15
 CHAPTER 1 Perspectives on Learning and Memory
sy PER ESE DCR

Thorndike thought that a stimulus—response Id be

of the
a reinforcement followed emission
formedwhenever
response
in the presence of the stimulus.

Clark L. Hull (1884-1952)

From roughly 1930 to 1970, American psychology was dominated by a series of
erand learning theories."These theories came to overshadow the ideas of
Thorndike and Watson, who were considered intellectually shallow by compar-
ison. Certainly, the grandest of the grand was Hull’s behavior theory, which was
not only impressive in its own right but became the reference point for new the-
oretical ideas in the 20 years after his death. A group of learning theorists called
neo-Hullians (e.g., Abram Amsel, Frank Logan, Neal Miller, O. H. Mowrer,
Kenneth Spence, and Allan Wagner) tried to extend his theory in various ways.
The basic goal of Hull and the other theorists was to develop a systemat-
ic theory of classical and instrumental conditioning to explain all behavior—
human and animal. The details of their theories
are of historical interest only,
but the concepts and issues that they defined remain important to research on
learning. Hull’s final theory (Hull, 1952a) involved many elaborate equations but
can
See be summarized
severeby this one:
— GEDx0=T!
Each symbol in this equation reflects a critical construct in Hullian theory, and
it is worthwhile going through them one at a time.

E—Reaction Potential. The ultimate goal ofthe Hullian theory was to

predict something called reaction potential, which determined the probability,
speed, and force with whichabehavior would be performed inresponséto a
stimulus. The organism was viewed as having a set of potential responses, each
with its own force or reaction potential, striving to become the actual behavior
of the organism in that situation. So, in running a maze, a rat might have the
potential responses of turning left, turning right, and stopping to scratch itself.
The response with the strongest reaction potential would be the response that
the rat exhibited. According to the equation above, this reaction potential was a
function of the controlling factors—the H, D, K, and I in Hull’s equations.

H—Habit Strength. Astrength of association was built up between stimulus

and response through past reinforced trials. Thus, Hull’s theory embraced
Thorndike’s law of effect in positing that reinforcement was necessary for learning.

D—Drive. Acco
to Hull,
rdin behavior-was
g ggt simply a function of habit
strength, as with Thorndike’s law of effect. A rat satiated after many reinforcing
experiences would no longer run the maze for food. Hull proposed that the

16
 History of Research on Learning and Memory

drive state of the organism was an energizer for habit

i ote tha
habit strength multiply in this equation.

[eelicenitine Maniaaton Hatsstrength and drive were still not enough

or be , na maze for food would soon stop running no
matter how well it knew the path or how hungry it was if the food were
removed. If the amount of food were decreased, the rat’s performance would
decline; if it was increased, Bees would teas <n
nt anc reward. Note that inc e
otivation also bears.a STIR relationship to reaction potential.

of inhibition originated with Pavlov and continues to play an . important role in

modern theories of learning. In Hull’s equation it is subtracted from the effect of
the other factors.
The fundamental issue with which Hull and other learning theorists
struggled was how to relate learning and motivation. Learning was not enough.
wanhehavign Daebadiabemaiuaiony! a behaviorist, could not allow for
an organism to mentally consider and weigh its options before making a deci-
sion. The equation presented here was proposed to relate the various factors.
Hull aspired to a highly formal theory of learning. He proposed a set of basic
postulates of learning and then attempted to derive predictions from the postu-
lates. His effort and others like it were greeted with great enthusiasm in the field
of psychology, for they were considered a sure sign that psychology had passed to
the stage of being a true science. In retrospect, these efforts were incomplete and
flawed as a logical exercise. Inconsistency and incompleteness became apparent
with the development of computers capable of carrying through all the deriva-
tions and delivering specific predictions about behavior. Nonetheless, modern
theories of learning and memory still show the influence of Hullian theories.

Edward C. Tolman (1886-1959)

Learning theories such as Hull’s or Thorndike’s were not without their critics.
The most influential critic of the time was Edward Tolman. He came from with-
in the behaviorist eee enasooe a language ene neoot inderstood and

17
 CHAPTER 1 Perspectives on Learning and Memory

2)

S
3 6 No food reward
So
oO

g
=x
4

XS
; = Regularly rewarded

No food reward
until day 11

R21 139 4h 6 67 Beto VOM el 21S 14 3S: 167

Days

FIGURE 1.7. Average number of incorrect choices for three groups of rats that are
running a maze. Source: From E. C. Tolman, and C. H. Honzik, “Introduction and
removal of reward and maze performance in rats.” University of California Publicaltions
in Psychology. Copyright © 1930 by University of California Press. Reprinted by per-
mission.

his most famous demonstrations involved maze learning by rats, which was the
popular experimental paradigm of the time.
The first ioned
with respect to Thorndike’s law of effect. The basi iment by Tolman and
Honzik (1930b) involved three groups of rats running a maze with 14 choice
points. Rats were put in at one end of the maze and were retrieved when they
got to the other end. All rats ran the maze once a day for 17 days. For one group,
food was always at the end of the maze; for another group, food was never at
the end of the maze; for a third group, food was introduced on the eleventh day.
Figure 1.7 shows the performance of the rats in terms of how many wrong
choices they made before reaching the end of the maze. The group given food
on the eleventh day dramatically improved its scores on the twelfth day and
even performed slightly better than the group that was reinforced all along.
According to Tolman, the unreinforced rats were learning all the while.

18
 History of Research on Learning and Memory

So

Black curtain Black curtain

he
Fo

FIGURE 1.8 Maze used to test the relative ease of learning either the response that
brings reward or the place at which the reward is found. Source: From E. C. Tolman
and D. Kalish. Journal of Experimental Psychology, Studies in Spatial Learning II. Place
learning versus response learning. Copyright © 1946 by the American Psychological
Association. Reprinted by permission.

existence of a cognitive map was demonstrated in experiments on place learn-

ing, another famous set of demonstrations by Tolman. Figure 1.8 shows the
maze used in one of the experiments. The rat was put in S, or S,, and food was
available at F, or F,. One group of rats always found food by turning to the right.
Thus, if they started in S,, they found food in F,, whereas if they started in S,,
they found food in F,. The wes eroup of rats Br ee ne food in i by no mat-
ter where they started, This a t or le

s tesponse-learning rats do better. Restle (1957) suggested that rats could

learn to respond to either cue—place or direction of turn. Which was easier
depended on the relative saliency or prominence of the two cues.

ot For instance, rats in the maze experiments learned that going to a spe-
_ cific location would lead them to the goal box. These expectations remained pas-
sive until some goal energized them into action. These goals and MERs antici-
pated many ideas of the cognitive era, but the fundamental problem with
Tolman’s theory was that he never explained how goals energized these MERs.
This problem led Edwin Guthrie (1952), another learning theorist of the time, to
complain that Tolman left his rat in the maze buried in thought. A later section
in this chapter describes how Newell and Simon’s computer simulation theory
of problem solving provided the link missing in Tolman’s theory.

19
 CHAPTER 1 Perspectives on Learning and Memory
Sec tac ere

B. E Skinner (1904-1990)
B. F. Skinner was a behaviorist whose approach was as different from Tolman’s as
one could imagine. Skinner's influence lasted long beyond the heyday of behav-
iorism. Because of his popular books, Walden Two (1948) and Beyond Freedom and
Dignity (1971), he became synonymous with behaviorism in much of the popu-
lar culture. He is often called el EEEER cons: he carried behavior-
ism to one of its extremes. Not only did Skinnerdlacktolerance for mentalistic
constructs such as memory, but he also had little patience for many of the theo-
retical constructs that occupied other behaviorists. For instance, he criticized'the
‘concepts of drive and habit strength that were part of Hullian theory because
they referred to internal states rather than to observable stimuli and responses.
GTS AEE itismajor SDM eltothe stud cabinstrument i

nb out, and so on. Learning changed the relative frequency of

these various
v responses; if lever pressing was followed by food, it would come
to be a more dominant response. _
Skinner is famous for developing the’Skinner box*(so-called by others)
\dproof box containsa lever that rats can pressto =
deéliver a pellet offood? Thetypical dependent measure was how often the rat
pressed the lever. A similar device was developed for pigeons, in which they
pecked at a key. Some important behavioral phenomena were discovered in
these environments, and much of the data discussed in Chapters 3 and 4 comes
from the use of these devices or variations of them.
Although Skinner denied a role for stimulus-response bonds, he had to
allow external to to have some role in controlling behavior. According to
Skinner, exte : :

the situation for a response but was not associated

with pier
it seem toibe a subtle distinction, but it typifies Skinner’ soe de.

Probably mostrae of Skinner’s research was his ee

~
reinforcement
(e.g., Ferster & Skinner, 1957), which is considered in some detail

20
 History of Research on Learning and Memory

Es
Fe
ee

A rat pressing a bar in a Skinner box. é : i :

A pigeon pecking a key in
an operant chamber.

in Chapter 4. This work was concerned with how various contingencies between
reinforcement and response affected the frequency with which the response
was emitted. For instance, in an example of what is known as fixed-interval
reinforcement, a rat might be given a pellet for its next response after 2 min had
passed since its last pellet. Figure 1.9 illustrates a fixed-interval schedule in
which a reinforcement is delivered every 2 min. Note that the total number of.

40

Oo
NO
Ww ro)

number
Total
of
responses
= [o)

FIGURE 1.9 Hypothetical behavior of an

organism under a fixed-interval reinforce- 0
ment schedule in which it receives a rein- 2 4 6
forcement every 2 minutes. Min

2h
 CHAPTER 1 Perspectives on Learning and Memory

organism behaved inner way; he was content with knowing what kind of
behavior could be expectéd from many different organisms (including humans)
given a fixed-interval schedule.
Figure 1.10 is a striking illustration of the generality of Skinnerian analy-
sis. Weisberg and Waldrop (1972) plotted the number of bills passed by the U.S.
Congress as a function of month. They argued that adjournment was a major
reinforcement for congressional representatives; thus, there should be an

1000 1000

750 750

Te!
oO
a
S
xz)
= 81st Congress
S 500 Jan. 1949
— Jan. 1951 500

s
s=)
£
=J
Oo

250 250

ene 0
TEES NUS ALS NPIS AveS StOlm Naps lina) MierAvee Vue oe a AGES ani INIT)
[espareetis Be1st session Ba A ee by eet ee2nd session eee |
Months

FIGURE 1.10 Cumulation of bills passed by the U.S. Congress. Source: From P.
Weisberg and P. B. Waldrop. Fixed-interval work habits of Congress. Copyright ©
1972 by Journal of Applied Behavior Analysis. Reprinted by permission.

22
 History of Research on Learning and Memory

increase in their behavior (bill passing) just before the reinforcer. Indeed, their
behavior showed the same scalloped form as that of rats in a Skinner box. The
exact same mechanisms could not be operating in Congress as in the rats, but
Skinner was not interested in mechanisms—only in the generality of behavioral
laws.
For Skinner, understanding was not found in an explanation of what was
happening inside the organism. A person did not understand a behavior unless
that person knew how to train an organism to perform the behavior. The

ieee. lot of practical knowledge emerged from Skinner's and his stu-
dents’ laboratories about how to control beha
+] at is. an existing

presses, a rat could be aed to press a lever harder. First, presses exceeding
low level of force would be rewarded; as the rat pressed harder, the criterion for
reward would be raised. The rat’s response would gradually move in the desired
direction.

ing and response chaining, truly amazing Re pro-

duced-For example, a pig was trained to go through a complete morning rou-
tine, including making breakfast, picking up dirty clothes, and vacuuming
poland& Breland, ee =
made

y d Onan auc maniedticaton

and psychotherapy, where the goal is to shape appropriate human behavior. Part
of the popularity of Skinner’s work resulted from its practical successes, but con-
“troversy also resulted because many people thought it ignored essential aspects
of human personality and emotion, inappropriately
trying to turn human beings"
into robots.
cients eS
The major sc pith sail approach, however, was its _
; | : fion. proble
This m to a head
came
withttheAT aE ae Behation (1957),in which Skinner tried to provide
an analysis of language and language acquisition. The linguist Noam Chomsky
(1959) published a highly influential critique of the work, arguing that the the-

41 wish I were as effective in training my children.

23
 CHAPTER 1 Perspectives on Learning and Memory

ory was incapable of accounting for the complexities of human language.

C CTE z ; ]
T1dJ

Skinner never responded to the Chomsky critique, although others did (e.g.,
MacCorquodale, 1970), and he lived to see the cognitive approach supplant the
behaviorist approach, partly because of the Chomsky critique. Skinner com-
plained bitterly to the end that the criticisms were unjust and that cognitive psy-
chology was full of fanciful mechanisms that failed to achieve the control of
behavior that he took as the true measure of scientific understanding.
Use ESRC eee Aa NEI SO FEES ERA NI SSE IE EIS ESOT LA RE CEO NEES EBS ELIE DEE OLDEST LEDGES SE LLC ELE LINED SLES

The General Problem Solver (Newell & Simon, 1961)

At about the time behaviorism was beginning to experience difficulties, a new
method of theory construction based on computer simulation was gaining
attention in psychology. The approach was introduced by Allen: Newell and
Herbert,Simon, two collaborators at Carnegie Mellon University, who were also
leaders in the field o
ey incorpora ny ideas from artifici
intelligence into their theories of human cognition, at the same time incorpo-
rating ideas from their theories of human cognition into their work on artificial
intelligence.
Newell and Simon brought a new definition of rigor into the field that
changed the levelof theorizing even among those who disagreed with them.

evious mathematical theories were either logically

flawed, like Hull’s, or very simple, like the theories described in an influential
book by Atkinson, Bower, and Crothers eS on mathematical learning theory.

omputer
simulation techniques have had a profound effect on the character of theorizing
in psychology. As in all fields of science, they have enabled exploration of com-
plexities that formerly had to be ignored. Many modern theories discussed in this
book depend on computer simulation techniques, including theories of animal
conditioning, human memory, and the neural basis of learning.
Newell and Simon’s use of the computer was more than mere simulation,
however.
Sasa TS eS SoHo ©
metaphor aspect of their theories remains controversial, and its acceptance is
difficult for most psychologists, who believe that the human brain is very differ-

24
 History of Research on Learning and Memory

ent from a computer and that theories based on analogy to computers are like-
ly to be misleading (e.g., Rumelhart & McClelland, 1986).
Newell and Simon’s influence led to the development of a number Of sim-
ulations of cognition and learning at Carnegie Mellon and elsewhere. However,
their greatest contribution was not to the study of learning per se but to prob-
lem solving. A difficulty in earlier theories of learning was determining
the rela-
tionship between knowledge (what the organism learned from experience) and
-behaviar How did a creature’s acquisition of new knowledge relate to behavior?
As indicated, some behaviorists such as Thorndike and Hull, merged the two
issues and argued that behavioral tendencies were learned—that there was no
difference between knowledge and behavior. Tolman’s major criticism was
directed against this position, but he was unable to derive a coherent alterna-
tive. Newell and Simon in their theory of problem solving showed how knowl-
edge could be decoupled from behavior and still result in behavior. Along the
way, they showed that rigorous and precise theories of behavior could allow
mentalistic constructs. More than anything else, this demonstration destroyed
the prohibitions against mentalism that Watson had introduced to the field 50
years earlier. In eliminating these prohibitions, Newell and Simon established
the basis for the cognitive revolution that has transformed all of psychology,
including learning theory.
pie
of Newell
center
The cework was the General Problem
and Simon’s
Solver, or GPS (Newell & Simon, 1972).GPS was a computer simulation that
used a way of deploying knowledge in problem solving called means-ends
analThe ysis . in applying means-ends analysis are the
basic steps following:

1. Identify the major difference between the current situation and the goal;
ifm a.- é@##° ~~ ieee
2. Seleet-some-action
that is relevant to eliminating
that difference; that is,
select some means relevant tothat end. Newell and Simon
used the term
operator to refer to the actionormeans. An operator
is much-ike-an-oper-
ant in Skinner’s theory.
Qo. If the operator can be applied, apply it. Ifnot, make the goal to enable the
operator and start over again at step 1; that is, make the means thenew
end.
—

Newell and Simon (1972) give the following everyday example of means—ends
analysis:

I want to take my son to nursery school. What's the difference between

what I have and what I want? One of distance. What changes distance?
My automobile. My automobile won’t work. What is needed to make
it work? A new battery. What has new batteries? An auto repair shop.
I want the repair shop to put in a new battery; but the shop doesn’t
know I need one. What is the difficulty? One of communication. What
allows communication? A telephone...and so on. (p. 416)

25
 CHAPTER 1 Perspectives on Learning and Memory

In the Newell and

Simon example, the focus switches from the goal of getting the son to nursery
school to the means, which is a functioning automobile. Thus, the»means».

nse st. Subgoaling is discussed

more fully in Chapter 9, which considers problem solving in detail. There: can be
. In the example here, taking the child to nursery
school has a functioning automobile as its subgoal, which has a battery as its
subgoal, which has an automobile repair shop as its subgoal, which has a tele-
phone as its subgoal.

ic. Newell and Simon

(1972) showed that their program was not only capable of solving complex
problems in logic but that it went through the same steps undertaken by
humans solving those problems. GPS realized a degree of intelligence
unmatched by previous theories in psychology.
Although GPS was not specifically concerned with learning, it is fairly
clear how to conceive of learning within the theory:
Learning
is involved in
iri d. Operators are like
Tolman’s MERs in that they encode potentially useful knowledge about the
world. In Tolman’s latent learning situation, the rats might learn that making a
certain turn in a maze changes their position in the maze. In the absence of any
goals, however, this knowledge remains dormant and latent. When they realize
that food is in a certain location, they have a goal, getting to that food, and they
can treat their knowledge as operators relevant to that goal..Bach turnin the
socnaenniitidptiiammanith ~driesionsiia ras ‘

is is what Tolman was unable to do.

It is questionable whether what a rat does corresponds to the means—ends
method of problem solving, which, as seen in Chapter 9, is more appropriate for
describing human (and perhaps primate) cognition. However,

ods. Many problem-solving methods have been proposed in recent years.

Chapter 9 discusses another method, difference reduction, which seems more
appropriate for modeling lower organisms.

26
 History of Research on Learning and Memory

A Model of Memory (Atkinson & Shiffrin, 1968)

RichardAtkinson and Richard Shiffrin published a theory of human memory in
1968 that captured the then current wisdom about the nature of human memo-
ty. Their work typifies much of the research of the modern era and influenced
of both animal learning and human

ple remember a seven-digit phone numbe | e di iculty when

they have to add a three-digit area code. ; the phone
number is quickly forgotten if the person is distracted. Th ———

e€ petinahenit TepOsit

as generally thought that knowledge had to be rehearsed in short-term

memory for a while in order for it to get into long-term memory.
These basic ideas about the distinction between short-term memory and
long-term memory had existed for a number of years; Broadbent (1957) was one
of the first to describe them. Atkinson and Shiffrin crystallized these ideas into
a precise theory, expressed as both a mathematical model and a computer sim-
ulation model, and demonstrated that the theory could account for the results
of various experiments current in the study of human memory.
Figure 1.11 illustrates the basic theory. Information comes into short-term
memory from the environment through various perceptual processes. Short-
term memory has several slots, often specified around four, in which it can hold

Rehearsal

Short-term Long-term
Incoming information Transfer memory
memory

Displaced

FIGURE 1.11 The Atkinson and Shiffrin (1968) theory relating short-term and long-
term memory. Incoming items enter short-term memory and can be maintained there by
rehearsal. As an item is rehearsed, information about it is transferred to long-term mem-
ory. Another item coming in can displace an existing item from short-term memory.

D7.
 CHAPTER 1 Perspectives on Learning and Memory

1.0 13}

0.9 11

0.8 2
2 9
3 ®
£07 s
9 i)
eal 37
8
s 0.6
.~ E=
a } 5 5
es |= 2

3
0.4

0.3! ry ear Carear

5 10 5 20 2) 10 15 20
Serial input position Serial input position
(a) (b)
FIGURE 1.12 (a) The mean probability of recall as a function of its serial position in
the input, and (b) the mean number of rehearsals of an item. (From Rundus, 1971. )

these elements. The subject engages in a rehearsal process of reviewing or

rehearsing the information held in short-term memory. Every time the informa-
tion is rehearsed there is another chance for it to be transferred into long-term

experimen
for aa Roce to recall the ‘words inany ¢order.
experiment results in what is
is called they
Figure 1. 12a
FoRa
list 0

Foee CEMENT marae eet,

28
 History of Research on Learning and Memory

The recency effect is easiest to explain in the Atkinson and Shiffrin theo-

The last word is definitely still in the buffer. The next tolast word is
in the-buffer
unless it was deleted to accommodate the last. The decreasing
recall farther from the end reflects the decreasing probability that the item is still
in short-term memory.
According to Atkinson and Shiffrin’s theo

they are pushed out by an intervening word. Rundus (1971) asked subjects to
rehearse out loud and was able to show that the probability of recalling a par-
ticular word could be predicted from the number of times it was rehearsed. As
postulated by Atkinson and Shiffrin, the words at the beginning of a list received
more rehearsals. Rundus’s results are displayed in Figure 1.12b, which illustrates

The research paradigms on which this theory was based were rather sim-
ple experiments like this:free=recall experiment. They reflected a return to the
kinds of experimental paradigms introduced by Ebbinghaus almost a century
earlier. As more complex experiments have been performed, doubts have risen
about the Atkinson and Shiffrin theory. -
. The Atkinson and Shiffrin

iti As people began to look at memory in

more realistic situations, these problems became more and more apparent.
Chapter 5 specifically reviews some of the evidence against the theory.
Although few, if any, researchers still believe the original Atkinson and
Shiffrin theory, it remains very influential. Variations on it can be found in many
current theories, including a more recent theory developed by Shiffrin called
SAM (Gillund & Shiffrin, 1984), which is discussed in Chapter 6. The fact that
psychologists have been able to find problems with the original theory and pro-
pose better theories represents a triumph for modern psychology. This shows
that psychology has moved beyond its former indecisive, verbal arguments to
precise statements that have enabled theories to be tested and rejected. With
such theoretical precision comes scientific progress.
 CHAPTER 1 Perspectives on Learning and Memory

Neural Basis of Learning and Memory

Since learning obviously takes place in the nervous system, the reader may have
found it strange that there was almost no discussion of the neural basis of learn-
ing in the theories of learning and memory. presented. Until recently, not
enough was known about the nervous system to pursue the issue. However,
rapid advances in our understanding of the nervous system and in the research
techniques that enable this understanding have resulted in a surge of research
in this area. This new research is one reason for the recent rapprochement of the
learning research focusétl on animals and the memory research focused on
humans. Researchers in human memory have become aware that they can
begin to understand the neural basis of memory, but to do this they have to rely
to a large extent on research with nonhuman subjects.
Most chapters of this book discuss some of the relevant research on the
neural basis of learning and memory. These discussions assume a basic famil-
iarity with the nervous system. Therefore, this chapter concludes with a short
review of the nervous system from the perspective of understanding research on
learning and memory.

The Nervous System

2s. Almost all learning of any note

brains of several organisms. The
human brain has a volume of about 1300 cc, which is very large, particularly in
relation to the size of the human body. One difficulty in understanding the brain
is that it is a three-dimensional structure; many important areas are hidden
inside it. Figures 1.14 and 1.15 present two views of the brain: Figure 1.14 shows
the outside, and Figure 1.15 shows the inside of the brain as if sliced in half so
that one can see some of the internal structures.
The brain can_be divided into the cerebral cortex and the subcortical
Its
size increases dramatically with a rise up the phylogenetic scale. The human cor-
tex has a surface area of 2200 to 2400 cm. To fit into the human skull, it has to
be folded up, which accounts for the many folds that distinguish the human
brain from the other brains in Figure 1.13.
The cortex engulfs many of the lower brain structures, so that they are
invisible from the outside. The lower parts of the brain tend to be found in more
primitive species that have no or only a poorly developed cortex. Many of these
lower areas SUppOn basic functions. For instance

30
 Neural Basis of Learning and Memory

FIGURE1.13_ Representative brains of different animals show how large the human
brain is compared with those of other animals.

tem refers to parts of the brain that are at the border between the cortex and
these lower structures. The limbic system and in particular the hippocampus,
which is embedded within the temporal lobe, are impo are
discussed inma wpters.
The hippocampus is not shown in Figures 1.14 or
1.15 because it is neither a structure on the outside nor a structure at the center;
rather, it is inside the temporal lobe between the central structures and the sur-
face of the cortex.
The cortex itself can be divided by major folds into four regions, shown in
Figure 1.14. The occipital lobe is devoted mainly to vision, The temporal.lobe
has the primary auditory areas and
is also involved in the recognition of objects.
The parietal lobe is involved with a number of higher-level sensory functions,
including spatial processing.The frontal lobe can be divided into the motor cor-
tex, which is involved with movement,
and theprefrontal cortex. The prefrontal
cortex is much larger in primates than in other animals,
in apes (such as chim-
panzees) than in other primates (such as monkeys), and in humans than in apes.
It is thought to be important to planning and problem solving. We will discuss its
role in Chapter 5 on memory and in Chapter 9 on problem solving. Most areas
of the cortex are thought to be capable of supporting various sorts of learning.

31
 CHAPTER 1 Perspectives on Learning and Memory

Motor cortex 2 :
Primary somatic
Central sulcus sensory cortex

Parietal-temporal—
occipital association
cortex
Prefrontal Bs
association es Occipital lobe
cortex LP E

a) SSS*
Primary visual
Broca's area cortex

K
——
Temporal | lobe
lob ASSWw” Sylvian fissure
Primary
auditory cortex
Wernicke's
area

FIGURE 1.14 A side view of the cerebral cortex. Source: From E. R. Kandel, J. H.
Schwartz, and T. M. Jessell. Principles of neural science. Third Edition. Copyright ©
1991 by Appleton and Lange. Reprinted by permission.

Corpus
callosum

Optic nerve

Pituitary

Hypothalamus

Midbrain Cerebellum

Pons Medulla

FIGURE 1.15 Major components of the brain. (From Keeton, 1980.) Source:
Reproduced from Biological science, Third Edition, by William T. Keeton, illustrated by
Paula DiSanto Bensadoun, by permission of W. W. Norton & Company, Inc.
Copyright © 1980, 1979, 1978, 1972, 1967 by W. W. Norton & Company, Inc.

on
 Neural Basis of Learning and Memory

The cerebral cortex and subcortical areas can be divided into

different regions that serve different functions.
ae im ROUTE LTE SEEN AA ASS ERNE MERU HENS ae ON NRT ETRE

The Neuron

urons. There are estimated to be some 100 billion

neurons in the human brain. Neurons come in many shapes and sizes; Figure
1.16 shows some of the variations. Most neurons consist of some basic compo-
tC) CUTE

stem to an part. Axons vary in length from a few millimeters to

The longest axons stretch from the brain to various locations in the
spinal cord.)

Dendrite
Receptor
cell

Cell body

Dendrite

Cell body
Peripheral
branch

ae @)

Myelin sheath |
Axon

Schwann cell

pene Central
branch

Neuromuscular
junction

FIGURE 1.16 Some of the varieties of neurons. (From Keeton, 1980.) Source:
Reproduced from Biological science, Third Edition, by William T. Keeton, illustrated by
Paula DiSanto Bensadoun, by permission of W. W. Norton & Company, Inc.
Copyright © 1980, 1979, 1978, 1972, 1967 by W. W. Norton & Company, Inc.

es)
 CHAPTER 1 Perspectives on Learning and Memory

Cell body

Myelin sheath Axon hillock

Arborizations ee
sats EE IE ae

Dendrites Nucleus

FIGURE1.17_ Aschematic representation of a typical neuron. Source: From The nerve

impulse by B. Katz. Copyright © 1952 by Scientific American, Inc. All rights reserved.

Axons contact other neurons by means of arborizations (tiny branches) at

their ends. They typically contact the dendrites of other neurons. They do not
actually touch; there is a gap of perhaps 10 to 50 nm (a nanometer is one-bil-
lionth of a meter). This
near-contact
point of is called a synapse. In the mature
adult one axon may synapse on a thousand or more other neurons, and one
neuron may received synapses from a thousand or more axons. Thus, the ner-
vous system is characterized by a great many interconnections among neurons.
The axon of one neuron communicates with another neuron by releasing
chemicals called neurotransmitters. When the neurotransmitters reach the
other neuron, they change the electrical potential at the membrane of the neu-
ron where thé axon synapses, The inside of the neurorris typically about 70 mV
(millivolts) more negative than the outside. The difference results because the
concentration of chemical substances on the inside differs from that on the out-
side of the membrane. Outside the neuron there is a concentration of positive
sodium ions and negative chloride ions; inside there is a concentration of posi-
tive potassium ions and proteins with a negative charge. The distributions are
not equal, and the inside is negatively charged compared with the outside.
Depending on the nature of the neurotransmitter released by the axon, the
potential difference can decrease or increase. Neurotransmitters that decrease
the potential difference are called excitatory, and neurotransmitters that
increase the difference are called inhibitory.
in —
Ifthere are enough excitatory inputs onto the cell body and dendrites of a
neuron and
difference
the in *Sa are ie SEIS
membrane ‘becomes suddenly permeableto sodium ions and they rush in,caus-
ing the inside to become more positive than the outside---This entire process may
only take about 1 msec before it reverses and returns to normal. This sudden
change
called-an-action
is potential. It begins atthe axon hillock and travels
down the axon. The rate at which an axon potential travels down an axon varies
from 0.5 m/sec to 130 m/sec, depending on the character of the axon. For exam-
ple, the more myelin (myelin is a natural insulation around the axon) the axon

34
 Neural Basis of Learning and Memory

has, the more rapidly the action potential moves down the axon. When this
moving action potential, called the nerve impulse, reaches the ends of the
axon, it causes the axon to release neurotransmitters, thus starting. a new cycle
of communication among neurons. The time for information to progress from
the dendrite of one neuron through its axon to the dendrite of another neuron
is roughly 10 msec.
It is thought that all information processing in the nervous system
involves this passage of signals among neurons. As you read this page, neurons
are sending signals from your eye to your brain. As you write, signals are sent
from the brain to the muscles. Cognitive processing involves sending signals
among neurons within the brain. At any one time, billions of neurons are active,
sending signals to one another.
Neurons can be thought of as more or less active. Activity level refers both
to the degree of reduction in the difference in membrane potential and to the
rate at which nerve impulses are sent down neurons. The rate at which nerve
impulses are generated along the axon is called the rate
of firing; it is general-
lythought
that the number of firings,not the temporal pattern of the firings, is
__
important. Neurons can fire at the rate of
100/sec or more. Generally, the more
active a neuron is, the stronger the message it is sending. For instance, the way
in which a motor neuron tells a muscle to increase the force of its action is by
increasing its rate of firing.
Learning involves a change in behavior and so must involve some change
in the way neurons communicate. It is currently believed that changes in such
communication involve changes in the synaptic connections among neurons.
Learning takes place by making existing synaptic connections more effective.
The axon may emit more of a neurotransmitter, or the cell membrane may
become more responsive to the neurotransmitter. Recall that neurotransmitters
have either excitatory influences, reducing the difference in membrane poten-
tial, or inhibitory influences, increasing the difference; inhibitory influences can
be as important as excitatory influences. Many cells have spontaneous rates of
firing, and learning can involve lowering these rates.
IS HERI ESSER ASTANA SLES ESI AIR OLESEN NM REIN CELE REELS eee er Dd

aoa,
BE acommunicate
CE Wap one
Ln, DIES mS at synaptic
PUOICT GS SVT GP Tie
COMEconnec-
tions where one neuron may inhibit or excite the neural activi-
tyofanother neuron. BRS WALEED ESSE RSBAS REESE SME NORIO ARMS AE NRE VINO ISEIELE

Neural Explanations and

Information-Processing Explanations
It is impossible to study directly what is happening in 100 billion cells that are
all crammed into the human skull and can only be seen through a microscope.
However, scientists have found various ways of making inferences about what
is happening at the neural level. In one method, studies look at mass actions in

35
 CHAPTER 1 Perspectives on Learning and Memory

regions of cells. This can be done by measuring electrical potentials on the scalp
or by measuring changes in blood flow with new imaging techniques. Such
techniques allow researchers to see what regions of the brain tend to be more
active in particular tasks. For example, during a spatial reasoning task, areas of
the brain that perform spatial reasoning are more active. In another method, sci-
entists insert electrodes in lower animals to record what is happening in specif-
ic cells. They then infer from the patterns recorded in a hundred or so cells what
is happening in the remaining neurons in that region. Another methodology
used with lower organisms is selective removal of areas of the brain. For exam-
ple, as described in Chapter 3, much has been learned about the role of the hip-
pocampus in memory by studying organisms from which it has been removed.
Humans who have suffered damage to specific regions of the brain from acci-
dental injuries can also be studied. Finally, scientists can study the connections
among neurons and how neurons interact with each other. From this informa-
tion, they can devise computer simulation models of possible patterns of inter-
action among subsets of neurons.
Brain study is one of the most rapidly growing areas in psychology and has
provided genuine insights into what may underlie different learning phenome-
na. However, we are still far from an adequate understanding of the neural basis
of learning or memory. Thus, the majority ofthis book isdevoted. to.behavioral

These theories are ‘often called information- -processing theories

t inthat they‘talk
about the processing of informationin the abstract. For instance, in a discussion
of how experience strengthens a piece of knowledge so that it is processed more
rapidly and reliably, there may be no mention of the possible neural realizations
of the knowledge or its strengthening. Theories cast in such terms have always
been part of the field of learning and memory, although they were not called
information-processing theories before the advent of the cognitive approach.
Neural and information-processin lanations offer two levels of
ine ae ane a ea of learning and
memory. Information-processing theorists
are interested in ideas about_the
neural realizations of their theories, whereas researchers on the neural basis of
learning and memorylookto information-processing theories to help them
make sense of their data. Information about what is happening in a few neurons
an ae, Oe

or in a particular region ofthe-brain is not useful without a bigger picture in

which to place itss interpretation. Thus, Prescot propeeeeeraining andmeny
oty depends on advancing both the neural and the information- ee sad the-
ories-andunderstanding their interrelationships: —-—--__.

Tyfornatieen Beate tebe “rift

%5 oiderstimd theaaa,
changes brough arning, whereas neural arsine
try to understand how these changes ar in the
byain. Vert et oO

36
 Outline of the Book

Outline of the Book

This chapter has provided a basic review of the background needed to under-
stand current research in learning and memory. The rest of this book presents
what is currently known about learning and memory. The next three chapters
are devoted largely to animal research, which has certain advantages over
research on humans. The researcher can exercise more complete control over
the learning history of a nonhuman—controlling its environment from birth
and subjecting it to manipulations that would be unethical if the subject were
human. Also, to the extent that the creature is simpler, the researcher may be
able to look at a purer form of learning, without the complex cognitive process-
es and strategies of humans. Chapter 2 examines classical conditioning, which
provides a basic analysis of how associations are formed. Chapters 3 and 4 focus
on instrumental conditioning, which is concerned with how learning is used to
achieve critical biological goals.
Four central questions should be kept in mind when reading these chap-
ters on animal learning. First, to what degree is animal learning like human
learning? There are some remarkable commonalities in the behavioral manifes-
tations of learning. Second, what is actually happening in the animal during a
learning experiment? The traditional view that simple learning processes are
occurring has largely been replaced by the view that animals try to adapt to their
environment. Third, what is happening in the nervous system to produce such
learning? Here animal research is at a considerable advantage over human
research because physiological experiments can be performed on animals that
cannot be performed on humans. Finally, what is the relationship between
learning and motivation? This question has been central to the psychology of
learning.
Chapters 5 through 8 consider the current concept of memory, which is
based largely on research with human subjects. Human research has two
advantages over animal research. Humans can follow complex instructions and
therefore yield richer data about the learning process; and the results obtained
are closer to what we are presumably interested in, that is, human learning out-
side the laboratory. Chapters 5 through 8 present what is known about how
knowledge is encoded, stored, maintained, and retrieved. Chapter 5 discusses
sensory and working memories, which are systems for encoding information
currently being processed. Chapter 6 discusses how information is originally
encoded into long-term memory. Chapter 7 considers how information is
retained, and Chapter 8 discusses how it is retrieved. Although most of the
research presented is from humans, these chapters show that much of it extends
to other animals. Thus, the principles of memory, though perhaps easier to study
in humans, also apply to many species.
The last three chapters consider important extensions of the research on
learning and memory. Chapter 9 considers skill learning, such as the learning
involved in operating a computer system, and demonstrates that profound
changes occur in a skill with extensive practice—something that is ignored in

37
 CHAPTER 1 Perspectives on Learning and Memory

most traditional research on learning and memory. Chapter 10 reviews induc-

tive learning, which is concerned with how we form inferences, such as what is
or is not a dog, or what is or is not a correct syntactic structure in a new lan-
guage. Issues of inductive learning are of great concern not only in psychology
but in philosophy, linguistics, and artificial intelligence as well. The final chapter
discusses the applications of research on learning and memory to the problems
of education.

Further Readings
Several books recount the history of psychology, including Leahey (1992) and
Wertheimer (1979). Boring (1950) remains a classic review of the early history of
experimental psychology. Bower and Hilgard (1981) provide an excellent dis-
cussion of the major theories of learning. Kandel, Schwartz, and Jessell (1991)
offer a thorough discussion of the nervous system and the neural basis of learn-
ing and behavior. Gazzaniga, Ivry, & Mangun (1998) and Banich (1997) provide
overviews of the neural basis of cognition.

38
 lassical Conditioning

Overview
Twenty-five years ago I ate crab for the first time. Shortly thereafter, I developed
a horrific stomach flu (I will spare you the details) which had been going around.
To this day I get nauseous when I am offered crab. This is an example of classical
conditioning. As discussed in the previous chapter, aconditioned stimulus (CS—
this case, crab) had been paired with an unconditioned stimulus (US—in this
case, flu) that produced an unconditioned response (UR—in this case, extreme
nausea). The result was that the CS acquired the ability to evoke a conditioned
response (CR—again nausea in this case). As we will discuss later in this chapter,
food aversion can be a particularly powerful version of classical conditioning. It
certainly is in my case—I am getting nauseous just writing this paragraph.

The Phenomena of Classical Conditioning:

Eye Blink in Humans
Most contemporary research on classical conditioning has been conducted on
animals, but in the past a great deal of research was conducted on humans. A
popular paradigm for studying human conditioning was eyeblink conditioning.
Imagine what it would be like to be a subject in a procedure such as eyeblink
conditioning. In a typical eyeblink-conditioning experiment, the subject is fitted
with a padded headband containing a nozzle pointed toward the eye, and a
wire measuring muscle activity in the eyelid is taped to the eyelid. The US is a
puff of air directed toward the outside of the cornea, and the CS is a light or
tone. The US normally evokes a UR of an eye blink, and conditioning is con-
cerned with how the CS comes to evoke a similar CR. Figure 2.1 shows a dia-
gram describing this situation much as Figure 1.3 describes Pavlov’s condition-
ing of salivation in dogs. However, in this case we are using the diagram to rep-
resent the situation of human conditioning of the eye blink.
In human eyeblink conditioning, one can demonstrate all of the basic phe-
nomena of classical conditioning. For instance, consider acquisition and extinc-

39
 CHAPTER 2 Classical Conditioning

Time ——AAARL-AN|NV\_\YNYraa—_—

(a) Initial pairing

Experimenter’s presentation cS US
sequence: (tone) (air puff)

Organism’s response: UR
(eye blink)

(b) Conditioned.response

Experimenter’s presentation CS US
sequence: (tone) (air puff)

Organism’s response: CR UR
(eye blink) (eye blink)

(c) Extinction

Experimenter’s presentation CS
i ae ,
sequence: (tone)
\ Tey

Organism’s response: CR
} (eye blink)

FIGURE 2.1 /Experimental procedure for classical conditioning of the eye blink.
Compare with Figure 1.3.
=

tion of the conditioned response. In one experiment (Moore & Gormezano, 1961),
the interval between the CS and the US was 500 msec. Subjects were given 70
acquisition trials in which the CS was followed by the US and then 20 extinction
trials in which the CS was presented alone. Figure 2.2 shows the percentage of tri-

100 Beginning of
extinction

CRs
of
Percent

FIGURE 2.2 Probability of a condi-

tioned eye blink during 10-trial
blocks of acquisition and 5-trial
blocks of extinction. (From Moore & LOMS20)4 30 P40 SO MCON70 Neo) 30
Gormezano, 1961.) Trials

40
 The Phenomena of Classical Conditioning: Eye Blink in Humans

100

80

= 60
So

& 40

20

FIGURE 2.3 Percentage of condi-

tioned eye blink as a function of é
CS-US interval. (From McAllister, 500 1000 1500 2000 2500
1953.) CS-US interval, msec

als in which subjects emitted a CR in anticipation of the puff of air. As in the sali-
vation data presented in Figure 1.4, these human subjects showed a rather stan-
dard conditioning curve, in which the probability of the CR increased, followed by
a standard extinction curve, in which the probability of the CR decreased. If one
waits for a while after extinction, one finds spontaneous recovery of the extin-
guished CR. That is, when tested again without the US, the probability of the CR
increases from its former extinguished level (Grant, Hunter, & Patel, 1958).
Another standard parameter of a classical conditioning experiment is the
interval between the CS and the US. McAllister (1953) varied the time between
a CS of atone and the US of an air puff using intervals of 100, 250, 450, 700, and
2500 msec. Figure 2.3 shows the results in terms of percentage of conditioned
responses after 20 conditioning trials. Nearly maximal conditioning was
achieved in the intervals between 250 and 700 msec, which are typical values for
optimal conditioning in many, but not all, conditioning paradigms.

Sensitization and Habituation

Eyeblink conditioning can be used to show two other learning phenomena that
should be distinguished from true classical conditioning.
‘tion. Inthe design of Figure 2.1 the tone always precedes the puff of air. Figure
2.4
: is no relationship between the CS and US?
In such an experiment, a subject will be exposed to a series of air puffs and hear

41
 CHAPTER 2 Classical Conditioning

a series of tones, but there will not be any relationship between them.
Nonetheless, the subject will show some increase in the tendency to show a CR
of blinking when the tone occurs (Grant & Norris, 1947). Indeed, the experiment
can be run with the US presented alone a number of times before the CS is first
given. Then, upon first occurrence of the CS there is an increased tendency to
emit
an eye blink. Thi

lus. It is as ifthe subject has become

& Well-designed studies of classical

ere there is no relationship as well

tioning experiments te.g., Beecroft, 1966; Oldfield, 1937). In this case, the mag-
nitude of the response is decreasing as the subject becomes exposed to the US.
It is as if the subject is becominggiiiaae? and no longer has as strong a reac-
tion to the US. As a consequence, the size of the conditioned response can
decrease. For instance, Grant (1939) found a gradual decrease of both the fre-
quency and amplitude of the eyelid response to the CS.
Both nd
Habituationvare classified as
ecause they do not depend on the relationship between the CS and
= — ERTS sgRAPER “

\ abit e From
experiment to experiment, their net effect may be either an increased respon-
siveness or a decreased responsiveness (Groves & Thompson, 1970). In all cases,
it is necessary to separate these nonassociative forms of learning from the asso-
ciative learning that defines classical conditioning.

Two other forms of learning are sensitization, which is

increased responsiveness to many stimuli because of US expo-
sure, and habituation, which is decreased _responsiveness
to the
US because of US exposure. —

Conditioning and Awareness

The conditioned eye blink is adaptive. By blinking in anticipation of the puff of
air, the subjects are protecting themselves from an aversive stimulusThe . fact
that the eye blink is adaptive has raised the issue of whether human subjects are

42
 What This Chapter Covers

voluntarily choosing to blink so that they can avoid the light. Human subjects
report being aware of the CS—US relationship and blinking in response to the
US (Grant, 1973). However, they tend to be unaware that they are blinking to
the CS in anticipation of the US. It has nonetheless been argued that certain
eyeblink CRs look like eye blinks that subjects give when instructed to blink; in
particular, the eye closes more sharply, more rapidly, and stays closed longer
when it is a voluntary response (Spence & Ross, 1959). Considerable controver-
sy remains over whether it is possible to discriminate between automatic and
voluntary eye blinks (Gormezano, 1965; Ross, 1965).
At one time, classical conditioning was considered automatic, and instru-
mental conditioning was considered voluntary. This viewpoint led to the argu-
ment that the purported voluntary eye blinks in a classical conditioning para-
digm were really instances of covert instrumental conditioning in which the
human was responding for the reward of avoiding the aversive puff of air. This
distinction has not proved useful and is not pursued here. Whether voluntary or
automatic, certain behavioral regularities tend to be associated with condition-
ing. Research has focused on understanding these regularities. An important
theme in modern research is that classical conditioning servesan adaptive func-
tion, whatever the awareness of the organism or the degree
of voluntariness in
responding.

Classical conditioning shows similar behavioral properties

whether or not the subject is aware.

What This Chapter Covers

One of the striking features of classical conditioning is its ubiquity. Virtually all
organisms can be conditioned. Dogs and humans have already been mentioned.
This chapter reviews classical conditioning in organisms ranging from sea slugs
to rabbits, as well as some of the enormous variety of stimuli that can be used
for the CS and the US.
Classical conditioning is often considered the paradigm of choice for
studying how associations are made. The similar propertiesit displays over a
wide range of situations and organisms might lead to the conjecture that learn-
ing is taking place according to the same neural mechanisms in all these situa-
tions. Recent research on the neural basis of some instances of classical condi-
tioning in certain organisms is discussed next. The research shows that different
neural mechanisms underlie classical conditioning in different organisms and,
indeed, that different mechanisms underlie different types of conditioning in
the same organism. Creatures have found various ways of forming associations
in different situations. Constant across these circumstances is the need to form
associations. The constancy in the behavioral manifestations of classical condi-
tioning reflects the constancy of this need.

43
 CHAPTER 2 Classical Conditioning

After this review, the chapter discusses the behavioral properties of classical
conditioning. In particular, subsequent sections address the following questions:
What is associated?
What is the conditioned stimulus?
What is the conditioned response?
What is the nature of the association?

The answers are surprising in light of earlier views about classical conditioning.
The Rescorla—Wagner theory, an elegant but simple theory that captures much
of the complex structure of the data, is presented later in this chapter.

At as HERBTISR spi asean wave of organismsjai sim-

ilar classical conditioning phenomena when a neutral CS is
followed by a US. —

Neural Basis of
Classical Conditioning
As reviewed in Chapter 1, Pavlov speculated, incorrectly, about what was hap-
pening in the nervous system to underlie classical conditioning. More recently,
researchers have traced the neural bases of certain instances of classical condi-
tioning in certain organisms. This research offers a glimpse of the neural mecha-
nisms behind the behavioral phenomena and may provide a sobering influence
on overly simplistic or overly grandiose interpretations of classical conditioning.
As we have seen, neural information processing takes place through trans-
mission of signals among neurons, which are the individual cells that make up
the nervous system. Neurons transmit signals from one part of the nervous sys-
tem to another by sending electrical pulses along their axons. The axon of one
neuron makes contact with the cell bodies of other neurons. The point of contact
between the axon of one neuron and the cell body of another neuron is called
the synapse, and communication is achieved by transmitter chemicals, called neu-
rotransmitters, going from the axon of one neuron to the other neuron. These
neurotransmitters can increase or decrease the electrical potential of the neuron.
If enough electrical potential accumulates-on-its cell body, a neuron sends-a sig-
nal down its axon.It is_generally-believed_that learning involves changes in the
effectiveness of synaptic connections among neurons, so that one neuron comes
to produce greaterchanges in the
thelectrical
calpotential of
olanother.

acer ae aia tcc rene masceeteennnt essen tee aaa *

44
 Neural Basis of Classical Conditioning

Simple Learning in Aplysia (Sea Slug)

Some of the most influential research on neural mechanisms has been done on
at haveysimple nervous, systemsywith very large neurons. The
simplicity of their nervous systems allows researchers to understand in detail
how the systems work; the large neurons make studying what is happening to
an individual neuron relatively easy.
One of the creatures that has received extensive study is the sea’'slug,
Aplysiavcalifornica (see Figure 2.4). The gill (the respiratory organ of the Aplysia)
and the siphon (a fleshy spout, surrounding and enveloping the gill, used to expel
water) of Aplysia have withdrawal reflexes that can be evoked by touching the
siphon (or other nearby parts, such as the mantle). This is largely controlled
by direct synaptic connections’between ca are excited by the
tactile’Stimulation’and the motor neurons that control the'teflexyFigure 2.5 shows
schematically the synaptic connections for the gill-withdrawal reflex. The sensory
neuron from the siphon skin synapses directly onto the motor neuron for with-
drawing the gill. Thus, touching the siphon stimulates the sensory neuron, which
stimulates the motor neuron, which in turn evokes the gill-withdrawal reflex.
The withdrawal reflex to tactile stimulation is not very strong and tends to
weaken with repeated touching. The strength of the response can be enhanced
by a classical conditioning procedure that pairs tactile stimulation to the siphon
(CS) with shock to the tail (US). After five such pairings, the tactile stimulation
(CS) evokes a much stronger withdrawal reflex than is present without pairing
with US. Carew, Hawkins, and Kandel (1983) compared using a CS of tactile
stimulation to the siphon with a CS of tactile stimulation to the mantle. If the
siphon stimulus is paired with the US, it evokes a stronger withdrawal reflex
than that evoked by the mantle stimulus; if the mantle stimulus is paired with
the US, it comes to evoke a stronger reflex than that evoked by the siphon stim-
ul. Thus,gg ges iaers ARPA. atc SS TOGA
The mechanism of conditioning appears to involve the facilitating
interneurons shown in Figure 2.5. These are neurons onto which sensory neu-
rons form the tail synapse. They, in turn, synapse on the axons of the sensory
neurons from the siphon, providing an example of on a:
en shock to the tail activates these interneurons, which
operate on the synaptic connection between the sensory neurons from the

Mantle

FIGURE 2.4 The Aplysia.

Source: From Cellular basis of
behavior by Eric Kandel. Copy-
right © 1976 by W. H. Freeman
and Company. Reprinted with
Foot Eye
permission. ba

45
 CHAPTER 2 Classical Conditioning

(ie)Sensory
neuron

Facilitating
interneuron

Motor
neuron
Sensory
neuron
FIGURE 2.5 Neural connec-
tions underlying the condi-
tioning of the gill-withdrawal
reflex in Aplysia.

siphon and the motor neurons. They change the synaptic connection by increas-
ing the release of the neurotransmitter from the sensory neuron coming from
the siphon. This process is referred to as because it
enhances a process that is occurring on the axon side of the synapse.
studied the-structural basis for these changes in

chemical chain reaction that results in more release sites at the synapse of the sen-
sory neuron from the siphon. This Wl facilitation iis maximal if the senso-

e state that results in presynaptic facilitation.

Classical conditioning has been studied in another invertebrate, the nudi-
branch mollusk, Hermissenda crassiconis (Alkon, 1984). Classical conditioning in
this animal also involves a change in neurotransmitter rele
differ from those at work in Aplysia.

46
 Neural Basis of Classical Conditioning

lassical conditioning in Aplysia is produced by enhancing

the neurotransmitter release betwee om
thé CS and the motor neuron producing the CR.

prlisher sett
Classical Conditioning of the Eye Blink in the Rabbit NO
Thompson (e.g., 1986) engaged in an extensive project studying the neural basis
for a more complex example of classical conditioning, the conditioning of the
eye blink in the rabbit. In the standard procedure
ormezano, Kehoe, & Marshall,
1983). The CS is a tone, which also comes to evoke an eye blink (CR), although
at greater latency. (The latency for the UR is about 20 msec from air puff to blink,
whereas the latency for the CR is about 70 msec from tone to blink.)
Figure 2.6 illustrates some of the complex circuitry that is relevant to under-
standing this instance of classical conditioning. Sensory neurons from the cornea
synapse onto the fifth cranial nerve, from which neurons go to the sixth and sev-
enth cranial nerves, from which motor neurons go to produce the eye blink. This

Cerebellum

6th and 7th

cranial nerves
UR, CR
(eye blink)
US
(puff to
cornea) 5th cranial
CS (tone) HENS
Ventral cochlear
nucleus

FIGURE 2.6 Simplified diagram of the neural circuitry responsible for eye blink
conditioning.

47
 CHAPTER 2 Classical Conditioning

circuit, which produces the UR, takes about 20 msec. A second, longer circuit
goes from the fifth cranial nerve to an intermediate neuron, and from there to the
cerebellum (a subcortical structure—see Figure 1.15). Among other neurons in
the cerebellum, this path synapses onto cells in a structure called the interposi-
tus. There is also a path of synapses going back from the interpositus to the eye
blink, which produces a discrete genoa component to the eye blink
ff of ai (

‘Honing! There is a circuit by which sensory neurons encoding the tone (CS)
synapse onto the ventral cochlear nucleus, and from there by intermediate neu-

. It is generally thought
C\ that the since lesions (selective
removal of neural tissue) to the cerebellum eliminate eyeblink conditioning. In
addition to the paths by which the CS projects to the Purkinje cells, there are
paths by which the US projects to Purkinje cells. The Purkinje cells synapse onto
the interpositus, which is part of both the CR and the UR paths. The logic by
which the CS is thought to evoke the CR is somewhat complicated:
1. The Purkinje cells normally tend to inhibit the interpositus from evoking
the response.
2. Learning involves developing inhibitory connections from the CS path to
the Purkinje cells.
inhibiting the Purkinje cells, which nor-
| = (In effect, two neural neg-
atives are SOMME to make a neural positive.) Electrodes put into the cere-
bellum to record from the Purkinje cells show reduced firing after conditioning,
which is in line with this proposal. This situation is different and more complex
than that of conditioning in Aplysia. The path of conditioning in Aplysia involves
two neurons directly associated so that the sensory neuron turns on the motor
neuron. In the rabbit, the path involves more than a half-dozen neurons and has
the sensory stimulus turning off the cells that normally turn off the eye blink.
Although not as thoroughly studied, other instances of mammalian classical
conditioning appear equally or more complex and involve different neural struc-
tures from the spinal cord to the cortex and different neural circuitry and mod-
ifications (Thompson, Donegon, & Lavond, 1988). No single neural process
underlies all classical conditioning.
Despite the diversity of neural realizations, classical conditioning presents
a rather consistent picture at the behavioral level. These behavioral regularities
are identified next.

1 Electromyographic recordings measure electric activity associated with contraction

of muscles.

48
 S-S or S—R Associations

Classical rhein a Veeeyeni ales nlaoainee cere-

bellum, where paths involving the CS, the US, and the CR
meet, and it appears to involve learning to inhibit inhibiting 7

Se ae rn

S-S or S—R Associations?

Every time I pick up the leash, my dog bounds to the door. This is an instance of
classical conditioning in which the CS is the leash, the US is starting a walk, and
the CR and UR are going to the door. An interesting question that has occupied
researchers is, what is the CS (leash) associated to? Figure 2.7 shows two pos-
sible answers to this question. Mere om

onditioning in Ap ysia, ; e are between the S-S position

and the S-R position may seem obvious. The sensory neuron encoding the CS
from the siphon was directly connected to the motor neuron that produced the
gill retraction, a clear victory for the S—R position. However, Aplysia has a ner-
vous system very different from that of mammals. In the case of the circuitry
underlying the eye blink in the rabbit, the issue is far from clear. The CS, US, and

(a) SS, ea he
to door)

Se ees
(leash)
eee
;

~
(b) us ——————> UR )

SS teh 9
bye to walk) (go to door)

= se (leash)
FIGURE 2.7. The two possibilities for association formation in classical condition-
ing: (a) the CS is associated to the UR, and (b) the CS is associated to the US.

49
 CHAPTER 2 Classical Conditioning

UR all met in the cerebellum, it could be argued that the CS and the US were
ed at the Purkinje cells.

[by th Although human subjects

=
may have conscious expectations,
ascribing such conscious thoughts to animals, such as rabbits, is problematic.
Hence, the word e used. Some rather simple associative mechanisms can
produce behavior that mimics conscious deliberation. One example is provided
by the Rescorla-Wagne r theory, which is reviewed later in this chapter.
Although the neural evidence is unclear, several behavioral paradigms
provide information relevant to deciding between an S-S position and an S-R
position. These include response prevention, US devaluation, sensory precondi-
tioning, and second-order conditioning. Each of these paradigms is considered
in the following subsections.
Bsa ae UNAM SS NTI OE EA REA EDS RESIGN DN RON BONE ES IE LOT ECON TE TELNAES LS SETI LN NTL OO EELS SOOT

Response-Prevention Paradigm

response-preven paradigm. For instance, Light and Gantt (1936) paired a

CS of a buzzer with a US of shock to dogs’ paws. This US normally produces a
UR of leg lifting, which gets conditioned to the CS. However, they prevented
this response by temporally damaging the spinal motor nerves that went to the
affected legs. Thus, the dogs never had an opportunity to perform the response
and so to form an association between CS and UR. The spinal nerves recovered,
and within a couple of months the dogs had complete use of their limbs. Then
when presented with the CS of a tone, they displayed the CR of leg lifting. More
ecent studies have shown similar conditioning when an animal was prevented
from responding by use of drugs (Fitzgerald, Martin, & O’Brien, 197

It could be-argued-that-this interpretation of the S-R association is too

peripheral. Even-if-these-organisms were~prevented_from_executing—motor—
responses, their central nervous systems might well have been sending signals
for the response, which were blocked from producing the response
bythe
experimental interventions. It could thus be argued that the stimulus was asso-
ciated directly toarepresentation of the response in the central nervous system.
The other paradigms provide more definitive evidence in favor of the S—S posi-
tion.

50
 S-S or S—-R Associations

SN
RP Ral LOMES IRIE EERE RNC ON ERAS ART UL TRB aaa SN aed SUN

We. k evi or the S—S position is on i condition-

ing still occurs when the animal i iss prevented from making a
response.
SALLE ERNE ERI TEL
TIS LIE LENO TEL NINE GILT OB IGT
AT IEIE IPE SESS ES AS CARAS PANUSE IEBRET GRRE TSIM PE UT SI

US Devaluation Paradigm
A different test of the S-S versus S—R alternatives involves devaluing the US _
after classical conditioning has taken place (e.g., Holland & Rescorla, 1975;
Rescorla, 1973). For instance, aCS of light can be paired with a US of food for
hungry rats. The CS comes to evoke a number of responses associated with the
US of food, including increased activity (Holland & Rescorla, 1975). The rats can
then be satiated. When the rats are no longer hungry, will the CS (the light) still
evoke activity? If the CS were directly associated to the CR, it would still pro-
duce increased activity. On the other hand, if the CS had producedincreased
activitythrough anticipation of the US of food, and the US had lost its power to
produce increased activity, the CS would lose its power to evoke activity. In fact,
it loses itsability toevoke activity implying that S=Sassociations were formed.
This type of test is referred to as a devaluation paradigm. The US is
devalued. Ifthe CSisdirectly associated to the CR, i t if
devaluation, but if it is associated to the US, it will be affected. As another exam-
ple, consider the conditioned emotional response, or CER, behaviors, such as
freezing, shown by animals such as rats to CSs in anticipation of aversive stim-
uli, such as shock (see Chapter 1). The shock US can be devalued by repeated
administration (in the absence of the CS), which makes the rat less sensitive to
it (an example of habituation introduced earlier with respect to the human eye
blink). When the CS is finally presented after such US devaluation, the rat
shows a reduced CER. Generally, the response to the CS is reduced when the
US is devalued, indicating that the CS was associated to the US.
asec ee eee sce CARSINS UR UU EU EI SAE UH EN ELON EEE RIT EEO RAS OOOO seats eens

If the US is devalued after papel HIAe responseto

othecsi
is
reduced, suggesting an =S- association, hia
ESTRUS AU RR EG RE AE TE IE BERS EE NE
Tap RAR IESE ERS

Sensory Preconditioning Paradigm

a sensory preconditioning paradigm (e.g., Rizley & Rescorla,
using nts
Experime
1972) also suggest that the CS and US become associated. A typical sensory pre-
conditioning experiment (see Table 2.1) has two phases. In the first phase, one
neutral stimulus, such as a ig t oy occurs Jus wre another neutral stim-
ulus, such as a tone (CS,). In the second phase, CS, is paired with a US. For
ance, the tone (CS,) may
Lee precede a shock to the fe , which produces the UR
of le rawal. After a while, the tone acquires the abilitytoeevokeleg with-
io

51
 CHAPTER 2 Classical Conditioning

TABLE 2.1 Three Paradigms and the Associations Formed to the CS

Paradigm Phase 1 Phase 2 CS, Association CS, Association
Standard — CS,-US S-5 =
(to US)
Sensory
preconditioning CS,-Cs, CS,-US 5-S 5-S
(to US) (to CS,)
Second-order
conditioning CS,-US CS,-CS, S-S S-R or S-S
(to US) (to CR or CS,)

What happens if the other neutral stimulus, the light (CS,), is then pre-
sented? If classical conditioning involves associating responses to stimuli, noth-
ing will fips cbceauedtttaclia Honea engined aaieget a OS the
other hand, if stimuli are directly associated,_the
thelight
light (CS.)
(CS,) will
will evoke
evoke the
the antic-
ant
ipation of the tone (CS,), because of their pairing, which in turn will evoke the
5 iT : F
anticipation of the shock (US). Thu

SLURS AONE ORNL NUT SETA NTN IN,

Second-order Conditioning Paradigm

An alternative paradigm, the second-order conditioning paradigm (see Table
2.1), provides evidence for S—R associations, Holland and Rescorla (1975) per-
formed an experiment with rats, inwhich alight (CS,) waspaired
with aUSof
food. The CS comes to evoke a CR of increased activity. Then a second stimulus,
a tone (CS the first, the light (CS,). Thus, as in the precondi-
tionifig experiment, there are a CS,-CS, pairing and a CS,-US pairing. However,
in the second-order conditioning paradigm, the order is reversed: first CS,—US
andthen CS,-CS;, Asinthepreconditioning paradigm, thesecond-order CS, (in
this example, the tone) acquires the ability to evoke the CR (in this éxample,
increased activity)-This phenomenonis not evidence for an S—S association,

Holland and Rescorla used a devaluation paradigm to determine whether

the associations were S-S or S-R. They devalued the US of food by satiating the
rats. As noted earlier, US devaluation reduces the conditioned response to the

52
 What Is the Conditioned Stimulus?

first-order CS,, suggesting an S-S association between CS, and the US.
However, it does not lead to reduced responding to the second- as CS,, sug-
gesting a direct association between CS, (tone) and the CR (activity). If the first-
order association is extinguished by Ee ane light without food, the second-
order association is not extinguished and the tone continues to produce
increased activity (e.g., Amiro & Bitterman, 1980). Thus, the surprising result
from this and other second-order conditioning experiments seems to be that
first-order associations are S-S and second-order associations are S-R.
However, some second-order conditioning experiments (e.g., Rashotte, Griffin,
& Sisk, 1977) have found evidence for S—S second-order associations.

In.second-order conditioning experiments, second-order asso-

ciations tend to be S-R, incontrast to first-order associations,
Pie are ES2
aes
"pera ean SiRoeaE ARERR RSET TINIE TSE TSE UN SASSI LOE SEE TIERS INE IOI ATES BCR

Conclusions
The current conception (e.g., Holland, 1985a) is that both the stimulus and
response aspects of subsequent events compete for association to the stimulus.
Different paradigms produce S-S or S-R associations, depending on whether
the subsequent stimulus or response aspects are more salient or prominent. A
preconditioning expe
experiment provides evidence for S-S associations because no
salient responses are associate er
conditioning experiment supports S—R associations because the prior oe order
conditioning usually gives the CS, response characteristics that are more salient
than_its stimulus characteristics. First-order associations are more often S-S in
character because the US is typically so salient. Table 2.1 attempts to summarize
the results. There are threeparadigms:‘standard,
s which involves one pairing,
and1 (2)the sensory preconditioning and (3) second-order conditioning, which
‘involve two pairings but differ in their ordering. S-osassociations
ns appear to be
the rule* ePL Se Tespe foCS, association sin

What Is the Conditioned Stimulus?

What exactly is the conditioned stimulus that gets associated in classical condi-
tioning? If an organism forms an association to a tone of a particular pitch or
frequency (for example, 1000 Hz), will the organism display the association to a
slightly different pitch (1010 Hz), a very different pitch (4000 Hz), an entirely
different sound (a dog barking), or a flash of light? Intuitively, it would seem

53
 CHAPTER 2 Classical Conditioning

that a slightly different stimulus, such as 1010 Hz, should elicit the association
but that a very different one (the flash of light) should not.
Siegel, Hearst, George, and O’Neil (1968) conditioned the eyelid response
in rabbits to tones of different frequencies. Different rabbits experienced tones
with frequencies of 500, 1000, 2000, 3000, or 4000 Hz paired with a puff of air to
the eye. They were then tested with a variety of stimuli at different frequencies.
Figure 2.8 shov

2CC ore differer ginal t g stil . For instance, a rab-

bit trained with 1000 Hz responded most often to a test of 1000 Hz, next most
frequently to 500 or 2000 Hz, then to 3000 Hz, and least frequently to 4000 Hz.
Vy 1POPV IM.

humans, Gynther (1957) trained an eyeblink response to a CS involving a dim

light to the right. After conditioning, when initially tested with a dim light to the
left, his subjects showed a strong tendency to respond as well. However, the US
(puff of air to the eye) was never associated with the light to the left. After 100
discrimination trials (50 with each CS), subjects discriminated between the two
stimuli and showed a much greater tendency to blink to the right light.

ot the | \ Ep enomena of
s generalization
and discrimination are discussed at greater length in the
next chapter inasmuch as they have received more attention in research on
instrumental
wiance conditioning.
ue

Organisms naturally generalize a CR to a range of similar

CSs, but they can be trained to change the range
ofCSs to
which they emit the response.
RECENT LEAEA ERE EINER PRESS IS EDT SEES SSNESE SEE EET ERE EEN SELON LDN OEE OREEACCES

What Is the Conditioned Response?

Two possibilities exist as to the nature of the CR in classical conditioning. First,
the CR may be a version of the UR. Perhaps the CS causes the organism to expe-
rience some internal imageof the US, which then produces 3sthe same behavior
for whatever reason itwas produced bythe US.Theother view isthat the CS is
informational;
it allows the organism to anticipate that the US will occur and
thus to take appropriate|actioninanticipation oftreUS-In-this ew theCeis
a preparation for the US, not a response
toit. It isnoteworthy that the CRs that
are conditioned are adaptive; that is, the CRs tend to prepare the organisms for
the US. As noted earlier in this chapter, it is to the human’s benefit to blink in
anticipation of a puff of air. It is also to the dog’s benefit to salivate if food is
coming and to flex its leg in anticipation of shock. The original US—UR reflex is

54
 What Is the Conditioned Response?

fe))(=)

f (e)

500 Hz

ie)(2)

o (eo)

Bh(ce
1000 Hz

20

60

40
2000 Hz

20
Total
%responses,

60

40
3000 Hz

fl
al
20

FIGURE 2.8 Mean percentage of oO)[e)

total generalization test responses for
groups trained at each CS value. (The
arrow indicates the frequency of the 40
training stimulus for each group.) 4000 Hz
Source: From S. Siegel, E. Hearst, N.
George, and E. O’Neal. General- 20
ization gradients obtained from indi-
vidual subjects following classical
conditioning. Journal of Experimental
Psychology, Volume 78. Copyright ©
1968 by the American Psychological oOOW
Association. Reprinted by permis- So
oom Glowooow
© OOOF

sion. Frequency Hz

55
 CHAPTER 2 Classical Conditioning

in the organism to begin with because it is adaptive. Thus, the CS-CR connec-
tion also tends to be adaptive.
One kind of evidence for the preparatory character of the CR comes from
cases in which the CR is not the same as the UR. A striking example concerns
heart-rate changes in response to electric shock. The unconditioned response to
electric shock is heart-rate acceleration; but in some organisms the conditioned
response is heart-rate deceleration (Obrist, Sutterer, & Howard, 1972;
Schneiderman, 1973). In anticipation of a shock, the organism relaxes, which
may reduce the perceived magnitude of the pain. This relaxing reduces the heart
rate. Similarly, some organisms, such as rats, tend to show increased activity in
response to shock but freeze in response to a stimulus associated with shock
(Rescorla, 1988b). Thus, the behavioral responses to the CS can be quite the
opposite of the responses to the US, but both can be viewed as adaptive
responses. The shock is a noxious stimulus that requires escape, whereas the CS
is treated as a warning that may require a different response. For example, in
nature when an animal sees a predator (CS), freezing may help it avoid detec-
tion, but once the predator attacks (US), flight is the appropriate response.
A number of unconditioned responses involve a biphasic structure in which
an initial response is followed by an opposing response. One of the most dramat-
ic involves responses to narcotics, such as heroin. The initie
initial
tialrresponse to heroin, a
feeling of euphoria, isfollowed by a second, opponent
response, which tends to
counteract the initial response and produces unpleasant withdrawal symptoms.
Anothér example of Pe gaa See eee The
feelings of terror the skydiver has just before the dive are followed by an antago-
nistic response of pleasure when the dive is successful. Solomon and Corbit (1974)
suggest that the antagonistic opponent response is caused by the body trying to
avoid extreme arousal states, which are demanding of resources. Also, in the case
of heroin, the increased pain threshold it produces can also be dangerous.
Wagner (1981) proposed that in the case of such biphasic URs, the com-
pensatory second response is conditioned to the CS because it isan appropriate
response to blunt the effect of the
US-On the other hand, when the UR is
asic, like an eye blink, and does not involve an opponent process, the
UR is conditioned because the UR is the appropriate response in anticipation of
such a US.* Wagner called his theory SOP, forsometimes opponent process,
because
luse sometimes (i.(i.e., in the case of biphasic responses) the conditioned
response-is the opponent process.
It has been suggested that the conditioning of the opponent process is
responsible for drug tolerance (Siegel,-1983). With repeated use of a drug, the

“However, even in the case of the eye blink, Wagner distinguished two components
and argued that only the second component becomes conditioned. His position is
supported by Thompsons research on the circuitry underlying eyeblink condition-
ing reviewed earlier. This research showed that the unconditioned eye blink had two
components: one with a latency of 20 msec and one with a latency of 70 msec. The
longer-latency component forms the basis for the CR.

56
 What Is the Conditioned Response?

high becomes weaker and weaker; consequently, the addict needs to take more
and more of the drug to produce the same high. It is claimed that the stimuli that
accompany drug use (for example, a needle) serve as a CS that becomes ¢ondi-
tioned to a second, opponent process. Thus, the conditioning of the opponent
process is producing the drug tolerance. The CR evoked by these stimuli is the
opponent process that negates the effect of the drug. There is evidence that much
of the tolerance built up to drugs such as morphine can be removed by changing
the stimuli that accompany their administration (for a review, see Siegel, 1983).
Akins, Domjan, and Guitierez (1994) provide an interesting example of
how the CR adapts to the circumstances of the US. They were looking at condi-
tioning of sexual behavior in male quail where the US was access to a female and
the CS was the appearance of foam block with bright orange feathers where the
female would eventually appear. This CS evoked a CR of searching behavior,
which is different from the UR to the US, which would be to engage in courtship
and copulatory responses. Again we can see that the CR is behavior in anticipa-
tion of the US. In this case, the quail is looking for the female that the CS signals.
In addition to this basic result, Atkins et al. manipulated the delay between
the CS and the US in the conditioning experiment. They used either 1 minute
or 20 minutes. These are long CS—US intervals compared to the experiments we
have considered so far (e.g., Figure 2.3). As we will see later in the chapter, how-
ever, conditioning can sometimes be obtained at even longer CS—US intervals
than 20 minutes for some kinds of US. Atkins et al. were interested in the dif-
ference in the nature of the conditioned searching behavior at these two delays.
They classified the searching behavior as focused (the quail actually approached
the CS) or general (the quail just ran around). Figure 2.9 compares the percent-
age of general versus focused search in the two conditions. As can be seen,

14 (-

-= Close to CS
-e Far from CS

fon)
(oe)

Moves
minute
per

FIGURE 2.9 Amount of move-

Lg
Le
Afe
ee
an
a | | |
ment close to the CS and far
from the CS as a function of the Short CS-US Long CS-US Control
duration of the CS-US interval. Condition

Vi
 CHAPTER 2 Classical Conditioning

search focused near the CS predominated when the CS-US interval was 1
minute, whereas there was general search everywhere (equal amount of move-
ment close and distant from goal) when it was 20 minutes. Again, this makes
sense. If the CS signals a mate in 20 minutes, then the female can be anywhere
and the quail needs to find her, whereasif the mate is about to appear in a
minute, the quail should approach the CS and wait for his impending opportu-
nity. Thus, not only is the CR an adaptation to the US, but the nature of the
wag coon can vary spa with “otlike the CS-US interval.
aN

The CR asnachesisViePiente ne same asee UR) iis often an

adaptive response in anticipation of the US.

Association: The Role of Contingency

An idea that extends back at least to Aristotle’s writings on associations is that

Earlier in this cen-

ry, contiguity was thoug. be critical for conditioning. Evidence for its
importance was the strong influence that the CS—US interval can have on con-
ditioning. For instance, Figure 2.3 showed that eyeblink conditioning was max-
imal if the CS occurred very close in time to the US. However, as we have just
seen with respect to Figure 2.9, in some of the more recent research, condition-
ing has been obtained at long delays. Thus, the importance of contiguity has
been aa eaein recent ee: of conditioning.

or instance, when my two sons watch TV, they argue. However, my two sons
argue a lot when they are involved in other activities, too. Thus, the mere co-
occurrence of TV and arguments does not mean that there is a contingency (i.e.,
watching TV causes them to argue). The probability of arguments must be
greater when the TV is on for there to be a contingency. Stated symbolically, a
predictive or contingent relationship requires
P(argument |TV) > P(argument | TV )
where P(argument |TV) is the probability of an argument when the TV is on and
P(argument |TV ) is the probability of an argument when the TV is not on.

Rescorla’s Experiment
Rescorla (1968b) conducted an experiment to determine whether contiguity or
contingency is essential in classical conditioning. He intermittently presented a
2-min tone while rats were pressing a bar. In different conditions, he presented

58
 Association: The Role of Contingency

110));=

r + P(USICS) = 0.4
0.8
ie ~e-P(USICS) = 0.2
~- P(USICS) = 0.1

5
fa
FIGURE 2.10 Dependence of condi-
tioning on both P(US| CS) (likelihood
of the US during the CS interval) and
P(US1CS) (likelihood of the US in the
absence of the CS). (Adapted from . 0.0 95 yO. fe pO:2g 00:3. hae 0,4
Rescorla, 1968.) P(US| CS)

shock during 10, 20, and 40 percent of these 2-min tone intervals. Rescorla was
interested in the degree to which the rats would show a CER (freezing) and
decrease their rate of bar pressing during a tone interval.
Rescorla also varied the probability of shock during the 2-min intervals
when no tone was present, creating three separate conditions with 10 percent,
20 percent, or 40 percent probability of a shock during the no-tone intervals.
Figure 2.10 shows the results for various combinations of probability of shock in

f they pressed P times in the presence of a tone and A times in the

absence of a tone, the measure of suppression is the proportion (A — P)/(A + P).°
Since P and A are measured over the same periods of time, they should be equal
if the rat is unaffected by the tone, and the suppression rate should be zero. To
the extent the animal is conditioned to the tone, P should be near zero, and the
suppression rate should be near one, indicating that the animal is freezing in the
presence of the tone.
. Consider the degree of conditioning displayed when the probability of a
shock, P(US|CS), was .4 during a tone interval. The amount of conditioning is a
function of the probability of a shock when no tone is present, P(US|CS). When
the probability of a shock in a no-tone interval was zero, the classical condi-
tioning result of large response suppression occurred. The degree of suppression
was near one, indicating that the rats almost never pressed the bar in the pres-
ence of the tone. However, as the probability of a shock in the no-tone interval
increased to .4, the level of suppression decreased. When the probability of
shock was the same in the presence or absence of a tone (i.e., both were .4), no

3This is an algebraic transformation of the measure reported by Rescorla.

59
 CHAPTER 2 Classical Conditioning

conditioning occurred; the degree of response suppression was Zero, implying

that the rate of bar pressing was the same in both the presence and absence of
the tone. The other conditions in Figure 2.10 displayed similar results. The
amount of conditioning is not a function of contiguity (frequency with which
shock and tone co-occur) but rather of contingency (difference between proba-
bility of shock given tone and probability of shock given no tone).
cn agree eA NLWSERE EESZIEE LL
LLS NE

Conditioned Inhibition
The previous paradigm demonstrated that organisms show conditioning when
the probability of the US is greater in the presence of the CS. What if the proba-
bility is lower? it ce ere Ik 1 S

The Rancid paradigm or d SRE conditioned inhibition involves

using two CSs:a CS+, which is positively associated to the US, and a CS-, which
is negatively associated 1to the US:-In-an experiment by Zimmer-Hart and
Rescorla-{t974), when a clicker (CS+) was presented, a shock followed, but
when both a clicker (CS+) and a flashing light (CS—) were presented, no shock
occurred. (The CS— was never presented alone during this initial training.) The
animal came to show conditioning of the CER to the CS+ alone but not to the
combination of CS+ and CS-.
How does the animal treat the CS— in t

instance, Zimmer-Hart arid Rescorla also conditioned a 1200-Hz tone to shock.

When the tone and arta light (CS-) were eee Skee together, the CER was 7)

Sometimes the response that is being conditioned is bidirectional, and the

“organism can indicate conditioned inhibition by doing the opposite of the CR.

0 For instance, Wasserman, Franklin, and-Hearst (1974) trained pigeons to associ-

ate food with an experimental context (CS+). A light (CS—) signaled that there
would not be food and so became a conditioned inhibitor. In the experimental
context (CS+), the animals approached the feeding apparatus, but in the pres-
ence of the light they actively withdrew from the light. Another example involves
taste preference. Association of taste with illness decreases the preference for the
taste, but the association of the flavor with the absence of illness actually increas-
es the preference for that flavor (Best, Dunn, Batson, Meachum, & Nash, 1985).

60
 Association: The Role of Contingency
SERINE MAAAENRERUN NOE EH AR ES RE ARSON SDA AN DUI AREAS AEDT ES
HSRE SARDITOOOS SMILE EAN ASSO ON EISEN ANTI POE TREN SRNREE EBA

Organi late a CS with the absence

of a US, and the CS then becomes a conditione ibitor for
responding.
———

wens EERIE, EIR RAT ALTE OLE

Associative Bias
This chapter has reviewed the evidence that organisms are sensitive to the sta-
tistical regularity between a CS and a US. There is also evidence, however, that
they are predisposed to associate certain CSs to certain USs independent of
their: : reference for certai tions isreferred to as
associative bias, which is similar to ppormcigs s concept ofbelongingness dis-
cussedin the first Chapter.
uppose you heard a loud noise that sounded like an explosion followed
by the shaking of the ground. How likely would you be to think the explosion
sound and the shaking of the ground were related? Suppose, on the other hand,
you heard a bird’s song followed by the shaking of the ground? How likely would
you be to think that the bird sound and the shaking were related? Presumably,
you would think the first pair was more probably related than the second pair.
Certain pairs of stimuli are more likely to be related than are other pairs of stim-
uli, and organisms condition more readily with such pairs of stimuli.
In a well-known study, Garcia and Koelling (1966) used rats whose only
access to water was from a water spout. While drinking the water, the rats were
exposed to a compound CS consisting of a flavor component (a saccharine
taste) and an audiovisual component (light flash and click). Different groups of
rats then received either a shock or an injection of a drug that produced nausea.
Garcia and Koelling tested the rats separately with saccharin-flavored water and
with the audiovisual stimulus to see which had a greater impact on water intake.
Rats that had received a shock later showed greater CER (i.e., reduced drinking)
to the audiovisual stimulus, whereas rats that had received the injection showed
greater CER to the flavor stimulus. Thus, rats were more prepared to associate a
CS of light with a US of shock and a CS of flavor with a US of poisoning. A great
deal of subsequent research has focused on conditioned taste aversions of a dis-
tinctive taste (CS) with a poisoning (US). Rats (and many other organisms,
including humans) learn such taste aversions after a single CS—US pairing and
after intervals between the CS and US up to 24 hours (Etscorn & Stephens,
1973). It is unusual for conditioning to operate over such long time delays (e.g.,
see Figure 2.3). Organisms presumably have this unusually strong propensity to
associate taste and poisoning because it is adaptive.
The associations that organisms are prepared to form are somewhat species
SEE coxon, Dragon, andKralTGS7D compared usingthe CS of water taste
(sour) or water color (dark blue) in conditioning to a US of poisoning .They com-
pared rats and quail and found that rats displayed greater aversion to taste and
quail displayed greater aversion to color. This result makes sense when the two

61
 CHAPTER 2 Classical. Conditioning

species are compared. Rats are nocturnal animals with an excellent sense of taste
and smell but with poor vision; quail are daytime animals with excellent vision, and
they use this vision to select appropriate foods (e.g., seeds that are not poisonous).
In some situations, a conditioned food aversion can be a serious problem.
For instance, cancer patients undergoing chemotherapy develop strong aver-
sions to the foods they eat because chemotherapy produces nausea; thus, they
do not eat adequate food. One means of combating this problem is to schedule
eating and chemotherapy so that food intake does not precede nausea. Another
possibility for patients is to consume a bland diet, because nondistinctive stim-
uli do not appear to become conditioned as readily as distinctive stimuli. Yet
another possibility is to have patients consume a very distinctive-tasting food
before the onset of nausea so that the taste aversion will only be conditioned to
that stimulus. For example, the problem of taste aversion with children was
reduced if they received a distinctive-tasting “mapletoff” ice cream before
chemotherapy (Bernstein, Webster, & Bernstein, 1982).

Conclusions about the Nature of the Association

ganis fe racterized as forming a statistical inference. Bayesian sta-

tistical models fo e about probabilistic relationships consider not only
the data but also prior beliefs about what relationships are likely. The funda-
mental relationship in Bayesian statistics is*
Ga Posterior belief = Prior belief « ae er ea:
That is, theposterior, orfinal, belief ina hypothesis (e.g., that the CS predicts
the US) is a product of the prior belief and the strength of the evidence for the
a

“For those who prefer the odds formula in terms of probabilities:

P(HIE) _ P(H)|P(EIH)
P(HIE) P(H) P(EIH)
- where P(H|E) is the posterior probability of the hypothesis given the evidence;
P(H |E) is 1—(P1H); P(H) is the prior probability of the hypothesis; P(H) is 1 -— P(A);
P(E|H) is the conditional probability of the evidence given the hypothesis; and
P(E |H)is the conditional probability of the evidence given that the hypothesis is false.

62
 Conditioning to Stimulus Combinations

hypothesis. Organisms in a conditioning experiment can be viewed as acting

according to this statistical prescription. Posterior belief would map onto
strength of conditioning between the CS and the US, prior belief onto a bias to
associate the CS and the US, and evidence onto the degree of contingency
between the CS and the US.

Conditioning to Stimulus Combinations

The research just reviewed shows that organisms respond to the predictiveness
of individual stimuli. This sensitivity to the predictive structure of the stimulus
situation has been further documented by a number of lines of research study-
ing conditioning when multiple stimuli are present as part of the CS.

Blocking
Experiments have shown that an association is not formed with one CS if
another CS is more informative. In one experiment, Kamin (1968) contrasted
two groups:

Control. The animals experienced eight trials in which CSs of noise and
light were followed by shock.
Experimental. The animals received 16 trials in which just the noise was
followed by the shock. Then, like the control condition, the animals
received eight trials in which a CS of noise and light was followed by shock.

Kamin conducted separate tests to determine whether the CER could be evoked
to noise or to light. He found that the CER had been conditioned to both noise
and light in the control group but only to noise in the experimental group. Thus,
for the experimental group, the CER was not conditioned to the light even
though there were eight reliable pairings of light and shock, which normall
would have produced conditioning.

In a variation of this paradigm, Kamin (1969) presented a more intense

shock to the light-tone combination than to the tone alone. The animals were
first given 16 trials of tone-shock pairings with a shock intensity of 1 mA (mil-
liampere), followed by eight trials of tone and light followed by shock with an
intensity of 4 mA. In this condition, rats showed significant conditioning of the

63
 CHAPTER 2 Classical Conditioning

CER to the light. The light was now informative because it signaled a more
intense shock.
In another variation of this paradigm, Wagner (1969) trained three groups
of rabbits in an eyeblink-conditioning paradigm in which the US was a 4.5-mA
shock to the area of the eye:

Group 1. Two hundred trials of eaeand tone followed by shock.

Group 2. Two hundred trials of light and tone followed by shock, inter-
mixed with 200 trials of light followed by shock.
Group 3. Two hundred trials of light and tone followed by shock, inter-
mixed with 200 trials of light followed by no shock.

In contrast to Group 1, which defines the reference condition, Group 2 showed

little conditioning of the eyelid response to the tone. As in Kamin’s studies, this
result can be interpreted to mean that the light was the better predictor. On the
other hand, Group 3 showed even greater conditioning of the tone to the eye-
lid response than did the reference Group 1. The rabbits in this group were get-
ting evidence that the light was not a good predictor of shock and so came to
treat the tone as the sole reliable predictor of shock.

Configural Cues
In the experiments discussed thus far, separate stimuli developed separate asso-
ciations to the US. However, it ispossible to conditiona1an-organism to
respond
to
onl “fa particular configuration of stimuli_are present. Organisms can be
trained to respond when both stimuli A and B are — Sas when just one
is present. Although this result could mean that the AB combination is associ-
ated, it could also indicate that A and B are separately associated to the US but
are too weakly associated to evoke the CR individually and are only strong
enough to do so in combination. Whereas it is possible to explain this result in
terms of separate associations, in some situations the only possible explanation
is that a configuration of stimuli have become associated to the response and
that the individual stimuli are not associated separately. For example, organisms
can also be trained to respond when A is present and when B is present, but not
when AB is present (see Kehoe & Gormezano, 1980, for a review). If there is a
positive strength association between A and the US and between B and the US,
an even stronger association would be expected when both A and B are present.
However, it appears that associations can be learned to cue combinations.In
this case, the cues A + noB and noA + B become associated to the US, but the
combination A + B does not. Another situation that shows configural associa-
tions involves four cues—A, B, C, and D. Organisms can be taught to associate
8

64
 The Rescorla-Wagner Theory

A + Band C+D to the US but not A + C or B + D. Again, it is not possible to

account for this result in terms of associations to the individual stimuli.
must be formed to the combinations. Such stimulus combinations
Associations
are referred
configural
to as cues.
This PO ee ae Use ae cue (A) ina
combination masks the other (B). In the typical blocking paradigm, the organ-
ism comes to
respond to A alone and to AB in combination, but not to B alone.
Ina typicalconfigural
figural conditioning theorganism
conchtioning experiment, theorganism comes
comesto respond
forespe
to A alone to B alone, but not to AB (although as described earlier there are
other ways to demonstrate configural. conditioning). Configural conditioning
also differs from a conditioned inhibition paradigm, where one cue negates the
effect of the other and acquires the properties-of-a. conditioned inhibitor(the
organism léarns to respond to A alone, but-not to.AB or to B-alone}na typical
conditioning
configural maintain positive igm,
both cues-parad effects but only
when presented
iets com alone.
La EES ote pistesrmnese seine 9 2 DAR ear Sea tN

Co i Jett re i iations to com-

binations of stimuli that are di te ciations to
sits individual sbi,
ee a ee ENE LNB LEI SDS APIECE BERATING ELE GE ENERO.

Conclusions
In some cases (blocking and conditioned inhibition), it seems that the response
to the combination of stimuli can be predicted from the response to the indi-
vidual stimuli, but in other cases (configural cues) it cannot. The next section
reviews the Rescorla—Wagner theory, which accounts for many of the occasions
when the response to the stimulus combination can be understood in terms of
the response to the individual stimuli.

The Rescorla—Wagner Theory

In 1972, Rescorla and Wagner proposed a theory that successfully predicts many
phenomena of classical conditioning. Their theory shows how simple learning
mechanisms can be sensitive to the contingency between the CS and the US. It
also illustrates the principle that simple mechanisms can produce the highly
adaptive statistical sensitivities documented in the previous section. Although
the theory is 25 years old, it has gained renewed currency as a popular theory of
neural learning

65
 CHAPTER 2 Classical. Conditioning

the rate
Pe of learning,
ee oes a, and
the difference. between the maximum strength possible-and+the-current
strength, (A-V).
(A— |

a aad Cre

where AVis the change instrength, One ccan view A as reflecting the strength
of the actual US and V gece SN aaa be. The

eee eeren7ee learning ttO the Seat that the USis

is Thee
ing (ce,NOt prredicted by the CS).
plication of this theory to a simple situation in which the
US and the CS are paired 20 times. Assume that the maximum strength of asso-
ciation, A, is 100 and that the rate of learning, « is .20. The initial strength of
association is zero. After the first learning trial, the increase in the strength of
association is
AV =.20 (100
—0) = 20
The strength of association is the sum of this increment plus the prior strength
of 0: 20 + 0 = 20. In the next trial, the same formula applies, except that the prior
strength is now 20 instead of zero, and so the increment is

AV =.20 (100 - 20) =

The total strength is this result plus prior strength: 16 + 20 = 36. This process can
be continued, calculating for each trial the total amount of strength. Figure 2.11
shows the growth of strength over the first 20 trials; the growth looks like a typ-
ical conditioning function (e.g., see Figures 1.4 and 2.2).
EZRA RENAE am REO i ie ec ee eee tg

Sassen eRe ae ALENT

Application to Compound Stimuli

The Rescorla—Wagner theory is similar to several mathematical theories devel-
oped to account for learning, and it is not surprising that it succeeds in account-
ing for the approximate form of the conditioning curve. The theory derives spe-

66
 The Rescorla-Wagner Theory

vie

80 |-

s l
3 60
ae) E
40+
a
S ie

20 r

FIGURE 2.11 Growth of associative

strength with repeated CS-US pair- 0
ings, according to Rescorla-Wagner 5 10 15 20
theory. CS-US pairings

cial interest from how it deals with compound cues such as those used in block-
ing experiments. Before considering blocking experiments, consider a simpler
situation. Suppose that two stimuli, A and B, are simultaneously presented as
the CSs for the US. For example, the US might be food, A might be a tone, and
B might be a light. The Rescorla—Wagner theory holds that the total strength of
association between the compound cue and the US, which can be denoted Vyg,
is the sum of the strengths of the individual associations of A and B to the US,
which can be denoted V, and V,. That is,

Vip =Vat Vp
When A and B are paired with the US, they will grow in strength of association
to the US as in the case where there is onlyone CS:
AV, = O&A — Vaz)

é Vz = o0(A — Vip) ae Hy
On the first trial, since theré-is_no prior conditioning, these equations become
(assuming o = .20 and A = 100):
AV, = .20(100 — 0) = 20
A Vz = .20(100 — 0) = 20 "
Thus after the first trial, the individual stimuli have strengths of 20 and Va, is
the sum, or 40. On the next trial, the equations become

AV, = .20(100 - 40) = 12

A Vz = .20(100 — 40) = 12

67
 CHAPTER 2 Classical Conditioning

100 -—

Compound

association
of
Strength

FIGURE 2.12 Comparison of 0

growth in strength to cue alone ver- 2 a OF ne 10
sus cue in compound combination. CS-US pairings

and the individual stimuli have strengths 20 + 12 = 32. Figure 2.12 shows the
growth in the strength of association to a single cue when the cue is part of a
compound versus when it is alone. The cue strength of A can only reach 50
because it must share the strength of association with B, which will get the other

e preceding equa ons, 0 is the same forA and B. But Rescorla and
re allow
possibility
for the that the stimuli may varyintheir salience and
therefore one-stimulus may have a moreé rapid learning ra rate.>.Suppose that the
rate-oftearning forA is .4 and for Bit is .1. Figure 2.13 plots the rate of strength
accumulation for the two stimuli. As shown, A overshadows B and acquires
most of the strength. When one stimulus is sufficiently less salient than anoth-
er, it fails to be conditioned when presented in a compound cue, even though it
can be conditioned when presented alone (e.g., Kamin, 1969).
HAN OSS ERR ANOLE NER PENNE TREE ITNT ES NAR EIEN NESEY SEEMING MDS ILIA BIEL SEDENTARY
ADELE ECS Ss

INE REST SEAN EEE ISSR ARRON NUE EEE ESS TUTE
SSE ESR

Application to Blocking and Conditioned Inhibition

The Rescorla-Wagner theory can be applied to the phenomenon of blocking.
Recall the original Kamin experiment in which rats were given an initial 16 tri-
als pairing A with the US followed by eight trials pairing A and B with the US.

68
 The Rescorla-Wagner Theory

:
60 }—-
More salient r

S i
1
2
7)
g 40
&
8
Ay he

<x

Less salient
20 r

FIGURE 2.13 Growth in strength 0 te |

of association to a more versus a less 2 4 6 8 10
salient stimulus. CS-US pairings

The effect of the first 16 trials is that A has already acquired most of the avail-
able strength and nothing is left for B. That is, the effect of the conditioning is to
set the strength of association to A to full value.
V, = 100
V, will start with an associative strength of 0, but the stimulus combination will
have a strength of 100. That is,
V, + Vz = 100
Since the US strength A is also 100, there will be no difference between it and
the strength of the compound stimuli. Since there is no difference between the
US and the strength, there will be no learning and V, will stay at 0.
The Rescorla—Wagner theory also explains why B conditions if a stronger
shock is used for the AB combination than for A alone. Rescorla and Wagner
postulated that the value of A was related to the intensity of the US. Thus, if a
higher value is used for the shock, as in Kamin (1969), the value of A is larger
and strength is available to be conditioned to B. As noted earlier, B can be con-
ditioned if the shock intensity is increased.
The Rescorla—Wagner theory can also predict the phenomenon of condi-
tioned inhibition discussed earlier. In the paradigm, A (e.g., a clicker) is associ-
ated with the US (e.g., a shock) but AB (e.g., clicker plus flashing light) is not.
The Rescorla—Wagner theory implies that the organism will learn strengths of
association, V, to the clicker and V, to the light, such that
V, = 100
V,+Vzp=0

69
 CHAPTER 2 Classical Conditioning

100

CS+ (A)

50

association
of
Strength

Trials

FIGURE 2.14 Growth in strength of association for the positive and negative CSs in
a conditioned inhibition paradigm. Source: W. Schneider, “Behavior Research
Methods, Instruments, and Computers,” “Connectionism: Is it a paradigm shift for
psychology?” (1987). Reprinted by permission of W. Schneider.

since A is always associated with shock but the AB combination never is. The
only way these equations can hold is for Vz = —100. Thus, B has a negative
strength of association that corresponds to the result of conditioned inhibition.
Figure 2.14 shows the growth in strength to the positive CS+ (e.g., clicker) and
the negative CS-—(e.g., light) over trials, assuming that the trials presented to the
organism alternate between A (CS+) and AB (CS+ and CS-). The learning
curves go to +100 and —100.
SARUM NIA NPA MOM UZ EN NS PLIERS ARR LATIN

The Rescorla—Wagner theory Meds blocking and condi-.

tioned inhibition because of its assumption that stimuli com-
pete eeeassociation to ue US.
NATAL ERNE RES TEONSTL NLIE ELON TALE LSE NENT ELISE MI IPE TN ETE NINETEEN TIE EIEN ALESIS

Problems with the Rescorla-Wagner Theory

The Rescorla-Wagner theory explains a wide variety of “penmentel calsbut
there are some thingsitdoes not explain. -A-phenomenon that
ffect of st

70
 The Rescorla-Wagner Theory

On. LNe Organism

Nas Deel

Se tCopmenve rune conditionin: 1 However Ptheraieis no

place jin the Rescorla-Wagner theory. for. Sheena [he _CS—US strength
starts pe ee ee in the preexposure trials, it stays
atzero. Thus, the Rescorla-Wagner theory predicts that conditioning should pro-
ceed as ifhad_been-ne-preexposure
there trials. It has been suggested that the
rate of learning, «, should reflect the salience of the CS and‘that the effectofCS
Ngee ara i ant yeatosonalBS it less salient. pieenes (1978) fale an exten-

ave focused on the ais of the CS in ee nciiriies Their eroposalliis that

the organism will pay attention to the CS and learn CS-US relationships to the
extent that the CS is followed by unexpected US events. In the case of latent
inhibition, the animal has had a history of the CS not being followed by signif-
icant US events and has learned to ignore the CS. Thus, in these theories the
competition occurs, not between different associations to a single US, but
ial CSs for attention of the org

on configural cues indi raysinde

ulus combinations can be conditioned. Such are can Saeae be SPapainodated
within the Rescorla—Wagner theory by introducing compound stimuli as CSs dif-
ferent from the individual stimuli. Thus, not only are A and B treated as stimuli
that can be conditioned, but so is the compound AB. This idea was proposed by
Spence in 1952 and suggested by Rescorla and Wagner, and it has been actively
developed by Gluck and Bower (1988) in an application of the Rescorla-Wagner
theory to human learning. However, the introduction of compound stimuli weak-
ens the predictive power of the Rescorla-Wagner theory because there is no basis
for predicting whether the cues will be treated separately or configurally.
asso
The data on ciat also present problems
bias ive for the
Rescorla—Wagner theory because they indicate that learning is not just a func-
on of the CS or the US, but depends on the inte
tion interaction between the two. Some
pairs belong together and are easier to associate; for example, rats are especial-
lyprepa red
to “conditi sisoning to a CS of taste. The
Rescorla—Wagner theory can accommodate this phenomenon by %assuming tl that
the learning rate, &, varies ; with the CS—-US combination. Asthe
in case
corof com-
pound cues, however, such a maneuver weakens the theory because bec the theory
offers riobasis forknowing how toassign o’s toCS—US combinations.
As we have observed, the Rescorla—Wagner theory is not without its prob-
lems, and alternative theories conditioning
of have been proposed. We have

im
 CHAPTER 2 Classical Conditioning

already mentioned the theories that focus on the attention the organism pays to
the CS. Yet other theories propose that the organism compares the relative effec-
tiveness of various CSs in predicting the US and shows conditioning to the
extent a CS is more effective than the alternatives (Gibbon & Balsam, 1981;
Miller & Matzel, 1989). However, these theories have their own difficulties.
Suffice it to say that no theory can capture the many phenomena that have been
documented with classical conditioning.
We have focused on the Rescorla-Wagner theory because it has remained
the reference theory for the field for more than a quarter of a century. In 1997 at
the meeting of the Psychonomics Society, a symposium was held to commemo-
rate the twenty-fifth anniversary of that theory. Few theories survive that long in
psychology, and at that symposium Rescorla offered some speculations to explain
the durability of the theory. He proposed that one of its advantages was that it was
relatively simple and captured many of the phenomena of interest to the field. It
helped set the research agenda for the next 25 years as people explored the
strengths and weaknesses of the theory and developed alternatives. It captured
the basic idea that conditioning involved learning to predict the US and that there
would be learning to the degree that the US was not predicted and so was sur-
prising. The basic learning rule has been used by a number of neural models. At
the same symposium, Gluck proposed that the Rescorla-Wagner learning rule
described one kind of low-level neural learning but that often other kinds of neur-
al learning were taking place as well. In the next section we consider how the
Rescorla—Wagner rule has been used as a model of neural learning.

The Rescorla—Wagner theory c ome of the ways

organisms are more sensitive to the statistical
Aacib relationships
sient iL
among stimuli, but organism 4 ivi an the
theory can capture.

lees
Neural Realization: The Delta Rule
The Rescorla-Wagner theory corresponds to a current idea about how learning
may take place at the neural level, which has played a particularly important role
in a theory of neural processing called connectionism. This theory stresses the
importance of synaptic connections among neurons. Figure 2.15 illustrates a
typical connectionist processing module where neural activation comes in along
a set of input neurons (at the bottom), each of which synapses onto a set of out-
put neurons. Every input neuron is associated to every output neuron. The pat-
tern of connections presents a more complex situation than that found in
Aplysia (Figure 2.5), where one neuron (a motor neuron) becomes more active
when another neuron (a sensory neuron) becomes active.
Researchers have maintained that the greater complexity of such networks
is required in order to reflect the greater complexity of mammalian behavior and

72
 The Rescorla-Wagner Theory

<q

SI Ps Output
neurons

BO
ASIESESS
ES
isS24
BB ea?
BSGEES
ce
RS
BS <
<q
<J
<q

EKE
KEKE
KKK
EKE
KEKKKEK
e LO
/y
KOK
KO
Linesman
bogLgBh:valitiglent |
Input neurons

FIGURE 2.15 Schematic representation of a neural net, where input neurons come
in (at the bottom) and synapse on output neurons (at the right). All input neurons
have synaptic connections to all output neurons. Source: From W. Schneider.
Connectionism. Is it paradigm shift for psychology? Behavior Research Methods,
Instruments & Computers, Volume 19, pp. 73-83. Copyright © 1987. Reprinted by per-
mission of Psychonomic Society, Inc.

learning. It has been argued that human cognition corresponds to patterns of fir-
ing over large numbers of neurons. The learning problem for such a network is to
learn strengths of association among neurons such that when a particular pattern
of activation occurs on the input neurons, a desired pattern of activation appears
on the output neurons. Seen in terms of classical conditioning, each input neuron
corresponds to a CS and each output neuron corresponds to a US. One proposal
for how to achieve such association of patterns is connectionist learning rule,
called the delta rule, which is based on the Rescorla—Wagner equation.
Figure 2.16 illustrates an application of this modeling approach to a med-
ical diagnosis problem studied by Gluck and Bower (1988). Their subjects stud-
ied records of fictitious patients who suffered from four symptoms (a bloody
nose, stomach cramps, puffy eyes, and discolored gums) and made discrimina-
tive diagnoses as to which of two hypothetical diseases the patients had. In this
case, the four symptoms are the inputs and the two diseases are the outputs. For
each patient, the input neurons corresponding to that patient’s symptoms
would be active and the goal would be to have the output neuron correspond-
ing to that patient’s disease active. This problem can be mapped onto the
Rescorla—Wagner theory by having the inputs correspond to CSs and the out-
puts to USs. Just as a rat is trying to predict shock in a particular stimulus situ-
ation, subjects are trying to predict a disease given certain symptoms. The delta

73
 CHAPTER 2 Classical Conditioning

Rare
disease

Common
disease

FIGURE 2.16 An adaptation of

Figure 2.15 to the Gluck and
Bower experiment in which sub-
jects associate symptoms to dis- Bloody Stomach Puffy Discolored
eases nose cramps eyes gums

rule treats the neural learning as if each input-to-output association is being

learned as a separate CS—US association. In the case of Figure 2.16,4 x 2=8
associations are being learned.
In the context of neural modeling, the Rescorla—Wagner equation, or the
delta rule, is used for adjusting the strength of the synaptic connection between
the input neuron i and an output neuron j. The delta rule is stated as
AA,
=0 A,(T;-A)
where A A,. is the change in the strength of synaptic connections between input
i and output j; o is the learning rate; A; is the level of activation of input neuron
i; A, is the level of activation of output neuron j; and T; is the target or desired
activity of j.This equation should be compared with the Rescorla-Wagner equa-
tion, which states

AV=a
(A- V)
where AV corresponds to AA,.; « corresponds to aA;; A corresponds to Tj; and
V corresponds to A.. As in the Rescorla—Wagner theory, learning is proportional
to the difference T;— A,. According to the delta rule, learning is also proportion-
al to A,, the level of activation of the input neuron. Recall that one proposal for
an extension of the Rescorla—Wagner theory was to make learning proportional
to stimulus salience.
In the case of a network, as shown in Figures 2.15 and 2.16, where there
are m possible inputs and n possible outputs, the system uses the delta rule to
simultaneously learn some men synaptic connections. The behavior of the whole
system can be complicated, but it still reproduces the basic competitive learning
behavior of the original Rescorla—Wagner theory.
The actual Gluck and Bower experiment was quite complex. Subjects saw
hundreds of patient descriptions reflecting different combinations of symptoms.
Each combination had a different probability of each disease. Overall, one dis-
ease was much rarer than the other, and each symptom had a differential asso-
ciation with the disease. The subjects were supposed to learn from this experi-
ence how to predict a disease given a pattern of symptoms. Figure 2.16 illus-

74
 Final Reflections on Classical Conditioning

trates the final strengths of association that should be learned in accordance

with the delta rule. These strengths of association did an excellent job of pre-
dicting subjects’ classification behavior and how they rated each symptoms to
whether it was diagnostic of the rare disease or the common disease.5 As seen
from the synaptic weights in Figure 2.16, a bloody nose was treated as more
symptomatic of the rare disease (.44 strength versus .01 strength), whereas the
other three symptoms were treated as more symptomatic of the common dis-
ease. In their ratings of the symptoms after the experiment, subjects agreed that
only a bloody nose was predictive of the rare disease.
The delta rule has become an important construct in theories of neural
learning. It has been used in a wide variety of models of neural processing and
has been imported by computer science to build models of machine learning. As
this example illustrates, it has also been used to predict complex human learn-
ing. At the same time, it has a number of difficulties similar to the difficulties
that the Rescorla—Wagner theory has with classical conditioning. For instance, it
cannot explain the learning of the configural cues. Gluck and Myers (1993) sug-
gest that it is a good model of cortical learning but that other sorts of learning
are controlled by a subcortical structure called the hippocampus. The next chap-
ter discusses the role of the hippocampus in learning.

The Rescorla—Wagner theory corresponds to a popular theory NN .

of competitive learning among neural elements, called the
delta rule.

Final Reflections on
Classical Conditioning
Classical conditioning is a phenomenon defined by experimental procedures:
the US is made contingent on the CS, and, as a consequence, the CS acquires
the capacity to elicit a CR. Early in the history of classical conditioning,
researchers tended to view the learning that was taking place as an unconscious
and automatic consequence of the contiguity of CS and US. Classical condi-
tioning was attractive in part because it was seen to embody pure and simple
learning, uncontaminated by cognition on the part of the organism. Just as early
researchers had hoped, classical conditioning studies on some animals, particu-

°Gluck and Bower’s model actually has a single output, which varies from +1 to -1,
depending on the probability of the rare disease. The model in Figure 2.15 is formal-
ly equivalent but more in keeping with the Rescorla—Wagner theory. In this model,
each disease is predicted separately in a 0-1 scale, and the maximum strength of
association is 1. The Gluck and Bower values can be determined by subtracting the
strength of the common disease from the strength of the rare disease.

79
 CHAPTER 2 Classical Conditioning

larly simple animals such as the Aplysia that have no central nervous system,
have provided insight into associative learning free of cognition. However,
recent research has also demonstrated that classical conditioning in more com-
plex organisms, including humans, frequently involves cognitive influences.
As discussed at the beginning of the chapter, different processes in the ner-
vous system underlie different instances of classical conditioning. Despite these
many differences, there are strong behavioral similarities across most forms of
conditioning: similar acquisition and extinction histories, spontaneous recovery,
generalization, temporal parameters, relationships between the CR and the UR,
and so on. The reason for the similarities lies in the fundamental adaptiveness of
classical conditioning. Classical conditioning allows organisms to respond adap-
tively in anticipation of a biologically significant UR. The common functionality
of classical conditioning across species underlies its common behavioral proper-
ties. An analogy can be drawn from the relationship between the eye of the
mammal and the eye of the octopus. The two have independent evolutionary his-
tories and are formed from different tissue, but they function almost identically.
They function so similarly because both have to deal with the same problem of
extracting information from light. Likewise, classical conditioning is similar
across organisms in terms of its behavioral properties because it serves the same
function of allowing the organism to respond in anticipation of the US.
Thus, certain generalities at a behavioral level in classical conditioning
may not be supported by generalities at a physiological level. In science diverse
underlying processes often give rise to common higher-order generalities in the
behavior of the system. For example, in economics industries producing very
different products display the same economic realities. Similarly, in biology dif-
ferent organisms display similar relationships. For instance, as reviewed in
Chapter 4, rather different creatures follow rather similar principles of foraging
for food.
This book identifies many generalities in learning that apply to diverse
species. The study of learning and memory has traditionally been about identi-
fying these generalities and understanding them. Much of psychology is devot-
ed to understanding generalities at the behavioral level. However, it should not
be concluded that all species learn the same way at the behavioral level. Species
show different preferences for various CS—-US combinations, reflecting differ-
ences in their biological makeup and in their environment. The closest thing to
a universal claim that can be made about classical conditioning is that it tends
to reflect an adaptive response by the organism to CS-US contingencies.
However, even this claim depends on carefully choosing the meaning of “adap-
tive.” For instance, it is really not adaptive for chemotherapy patients to develop
food aversions, although their tendency to form food aversions might be adap-
tive in other contexts, where it would lead them to avoid poisonous food.

The behavioral properties of classical conditioning reflect its

adaptive character.

76
 Further Readings

Further Readings
A number of texts are devoted to discussion of classical and instrumental con-
ditioning, such as that by Domjan (1998). Wasserman & Miller (1997) provide a
recent review of current research on classical conditioning. Miller, Barnet, and
Grahame (1995) have written a review of the Rescorla—Wagner theory. The
September 1992 issue of Scientific American was devoted to the relationship
between brain and mind and presented a number of relevant articles, such as
one by Kandel and Hawkins on the biological basis of learning and another by
Hinton explaining recent developments in connectionist theories of learning.
Among them is a mechanism called backprop, which is a popular extension of
the delta rule. Thompson, Donegon, and Lavond (1988) published a fairly
exhaustive review of the psychobiology of learning. Bernstein and Borson (1986)
reviewed research on learned food aversions. Journals that frequently publish
research on classical conditioning include the Journal of Experimental Psychology:
Animal Behavior Processes; Behavioral Neuroscience; and Animal Learning and
Behavior.

77
 Instrumental Conditioning

Overview
This chapter discusses research and theory on instrumental conditioning. In
instrumental conditioning, an o ism is reinforced if it makes a response (R)
in a certain stimulus (S) situation. For instance, Thorndike’s cats were reinforced
by escape and food if they hit the correct knob in his puzzle box. Just as classi-
cal conditioning is associated with Pavlov,4nstrumental conditioning is some-
times associated with Thorndike, but the association is not as strong because its
use and study did not really originate with Thorndike. In contrast with classical
conditioning, the discovery of which came as something of a surprise, instru-
mental conditioning is what everybody means by learning. It has been used by
teachers and parents since time immemorial, and there has never been any lack
of speculation as to how it should be used. Thorndike was simply the first to
propose a scientific theory of its operation.
Most of what happens in a classroom can be thought of as instrumental
conditioning. Consider a child learning that the sum of 3 and 4 is 7. The stimu-
lus can be thought of as “3 + 4;” the response as,“7;” and the reinforcer, the
teacher’s approval. Or consider a student learning to read a word. The stimulus
is the orthographic representation, the response is saying the word, and the

of their children’s behavior can be conceived of asinstrumental conditioning—

for example, parents rewarding children with money for cleaning their rooms.
Although these instances of human learning can be considered instrumen-
tal conditioning, they differ in an important way from the situation of Thorndike’s
cats. In the examples given here, humans are told the contingencies that are
operative, whereas Thorndike’s cats had to discover them. Sometimes humans
do find themselves in instrumental conditioning situations in which they must
discover the contingency. For instance, many students feel they have to discover
by trial and error what kind of an essay will earn a high grade from a teacher.
This chapter focuses on instrumental conditioning in animals. The issues
involved in instrumental conditioning in humans occupy center stage in the

78
 Classical and Instrumental Conditioning Compared

later chapters on memory, skill acquisition, and inductive learning. However, as

shown later in this chapter, humans placed in the same instrumental condition-
ing paradigms as animals do produce similar behavior.

Classical and Instrumental

Conditioning Compared
Contrasting the procedures used in classical and instrumental pene Ta
helps define them both. In classical conditionin thee erimenter sets u
certain contingen such

r. Forr instance, if a rat is in a Skinner box and aspresses a lever, a pellet will
appear in the feeder. Thus, the difference is that in instrumental conditioning,
the reinforcer (which is like a US) is contingent on the conjunction of stimulus
and response, whereas in classical conditioning it is contingent only on the
stimulus. Thus, in instrumental conditioning the organism can control whether
the reinforcer occurs.
If i ms in either situation, it begins to behave
as if it had figured out the experimenter’s contingency. In the case of classical
conditioning, it begins to perform a response (the CR) in preparation for ee US.
Int e case of instrumenta conditioning, i the respo: ands:

organism is earning to form an association between an antecedent contig

tion oning, a response as ‘yell
and a consequence that can ie redicted from these antecedents. Thus, Both
paradigms involve learning environmental contingencies.
between the two ir tt of the S

| Much debate has occurred over whether the process of learning is the
d instrumental conditioning. Cl al
1 instrumental

conditionin¢g volunt However, as noted in the previous chapter,

TE which pehaviots are automatic and which are voluntary can be prob-

19
 CHAPTER 3. Instrumental Conditioning

lematic. Interest in this distinction has waned, and attention has shifted to the
behavioral similarities between these two types of conditioning, with the implic-
it assumption that the two kinds of conditioning involve the same learning
process. Both kinds of conditioning show the same effects of practice, both
extinguish in the same way when the contingency is eliminated, and both show
spontaneous recovery. Both kinds of conditioning can be hurt if a delay is placed
in the contingency. Both paradigms result in successful conditioning only if
there is a contingency among the elements (not just a contiguity). With respect
to stimulus control, both show blocking effects, both can show configural learn-
ing, and both show similar generalization and discrimination processes. In addi-
tion, both show effects of associative bias. Since classical and instrumental con-
ditioning are so similar, this chapter essentially uses research on instrumental
conditioning to expand on the nature of conditioning in general.
eee SE PORT SLE IDRES EE AON

Instrumental
ee
and classical
a
conditioning
ae
share ee
many similar
behavioral properties.
iO LLL OE TNE AM DSN SE LETT LEE TOONS ESERIES IES ETRE SHG UI SIN er om oN ease ap

What This Chapter Covers

The chapter focuses on the same four questions that organized much of the dis-
cussion in the previous chapter. i :
What is associated?
What is the conditioned stimulus?
What is the conditioned response?
What is the nature of the association?

After addressing these questions, this chapter considers the similarity between
conditioning and causal inference and the evidence about the important role of
a particular brain structure, the hippocampus, in conditioning.

What Is Associated?
! N.Q1t1lonir VCS a STIMULUS TOL C IWC
by a reinforcement-For instance, a dog might lear sspond to the stimulus
“sit” with the response of sitting and receive food as a reward. As |in the case of
classical conditioning, a number of possibilities exist regarding what is associat
ed to what. One possibility is that the stimulus becomes associated to the
response. In this case, the reinforcer would stamp in the association but would
not pe bartol theassociation, This was the original idea of Thorndike and some
+L ns
vas

80
 What Is Associated?

For instance, Tinklepaugh (1928) showed that monkeys registered disappoint-

ment when an expected reinforcer (slice of a banana) was replaced by a less val-
ued reinforcer (lettuce). One monkey threw down the lettuce (which it would
normally eat) and shrieked at the experimenter in anger. This result would seem
to imply that the reinforcer iis ae of the association the animal had learned.

nelle to two different responses (lever pressing and chain pulling). When fed
with one kind of food pellet outside the experiment, they performed a predomi-
nance of responses that yielded the other kind of food pellet. Colwill and Rescorla
argued that organisms develop expectations; that is, if a certain response is emit-
ted in the presence of a certain stimulus, it ome a certain Sasa

pushing a tod to the left and to the right. One response was always rewarded
with food and the other with a sugar solution. Then one of the reinforcers was
paired with an injection of lithium chloride to produce a taste aversion to that
reinforcer. The rate of response associated with the devalued eens

One might argue from these studies that what the animal has really learned
is a two-term association between the response and the reinforcement. The stud-
ies just cited do not show that the animal will make the res ponse only in a partic-

the appropriate condition for that response is met. For instance, suppose the rat
learned that pulling a handle would produce food pellets in the presence of a tone
but liquid sucrose in the presence of a flashing light. If satiated on sucrose, this ani-
mal would stop pulling the lever only in the presence of the flashing light.
MELE NES OEE SELLER ERLE LL DELLE ENOL RL,

Associations Between Responses and Neutral Outcomes

The discussion thus far has reviewed evidence that organisms can learn associ-
ations between responses and reinforcing stimuli. What about associations
between responses and neutral stimuli? Organisms can learn about associations

81
 CHAPTER 3. Instrumental Conditioning

between stimuli and other neutral stimuli in classical conditioning (see the dis-
cussion of sensory preconditioning and second-order conditioning in Chapter
2). Can they similarly acquire such neutral associations in an instrumental con-
ditioning paradigm? In one experiment by St. Claire-Smith and MacLaren
(1983), as part of their free exploration of a Skinner box rats learned that a lever
press produced a noise. The experimental group was then trained on pairings of
the noise with food without a lever present in the box, and the control group
was trained on pairings of light and food. When the lever was reintroduced (but
no food was given), rats.in the experimental group pressed the bar more often
than did the control rats who had not learned the noise—food pairing. As a result
of their earlier free exploration, they appeared to have learned that lever press-
ing produced the noise. Putting this result together with the classical condition-
ing of noise and food, they acted as if the inferred that lever oe might also
produce food. Thus, it :

The ability to form associationsSins TNS eT tcomes

isgue cut Sete
reinforcement. Consider a rat learning to run a maze. If must learn a sequence of
associations of the sort that making a turn in a certain direction in a certain part
of the maze leads to another part of the maze. There is nothing inherently rein-
forcing about such turn—maze associations. Only the final turn becomes directly
associated with food (even though it may not be directly associated with food but
instead with a part of the maze associated with food). However, the rat has to
learn all these associations in order to put them together to run the maze. The
latent learning experiments (see discussion under Tolman in Chapter 1) showed
that rats could learn all these neutral associations before they learned there was
food in part of the maze. When they learned where food was, they could recruit
this neutral information to help them get to it.
LEN LL TILLER I LEE T ESET LIEEE IEE LEE SS ELEN TELE
EEE DRS LENE ESE EL LOR IEEE NESE NUON,

Or anisms can learn th 1 onses rode pune

outcomes and combin Is 1 ation with other experiences
to obtain pernforcemiag.
ALAIN EEN ALT ELEN NTI NET N TT NNIN TE AITE

Secondary Reinforcement
The previous section described a situation in which the rat first learned the asso-
ciation bar press—noise and then the association noise-food. This situation is sim-
ilar to sensory preconditioning in classical conditioning in that the organism first
learns a neutral association, then a biologically significant association, and finally
puts them together. Reversing the order of learning the associations a in the
equivalent of second-order conditioning; the animal first learns iologically
significant association noise-food and then bar press—noise. The sae acquires the

82
 What Is the Conditioned Stimulus?

abilitytoreinforce thebarpressforaryanimal trainedwithsucha procedure,and

the anim. press the bar just for the click without food (Skinner, 1938). The
noise is said to be a secondary reinforcer, or a conditioned reinforcer.
e classic example of a secondary reinforcer is mo s, which
can be extremely reinforcing but has no biological function in and of itself;
human beings have learned to associate money with more primary reinforcers.
Examples of the many other such secondary reinforcers in human society
include letter grades in courses and promises of favors. In an experiment
Saltzman (1949) presented rats with food in a white goal box. Then he intro-
duced them to a T-maze, where the rats had to choose between a path leading
to a white box and a path leading to a black box. The rats learned to take the
path that led to the white box even though the box did not contain any food.
The white box had acquired the ability to reinforce behavior. With enough expo-
sure to the white box, in the maze without food, the rats extinguished and no
longer chose that path. In like manner, when the currency in a particular coun-
try deflates to the point where it is useless, people cease seeking the money.
The functions of secondary reinforcers such as money are clear in the
human world. (It turns out that chimps are also capable of treating coins and
other tokens as money; see Cowles, 1937; Wolfe, 1936.) Secondary reinforcers
are_promises of primary reinforcement, and people know that they can be
exchanged for primary reinforcers. It is not clear whether it is always appropri-
ate toattribute’ such2’
cognitive explanation tasecondary reinforcers inlower
animals, but it does appear that for many species secondary reinforcers are good
at bridging delays in reinforcement. For instance, if a 5-sec delay is inserted
between pecking and reinforcement, a pigeon will not peck a key at a substan-
tial rate. On the other hand, if a green light comes on immediately after the peck
and the pigeon has seen the light paired with food, the pigeon will learn to peck
rapidly at the key. The green light becomes a secondary reinforcer that enables
the pigeon to bridge the delay in reinforcement (Staddon, 1983).

A secondary reinforcer is a previously neutral stimulus that

has acquired the ability to reinforce behavior as a consequence
bbbeing paired with a prima

What Is the Conditioned Stimulus?

As reviewed in Chapter 2, some, though not all, variations
on the original stim-
ulus_are effective in producing the response. The extension of the conditioned
response to new stimuli is called generalization, and the restriction of the con-
ditioned res onse from other stimuli is called discrimination. The phenomena of
stimulus gen nd stimulus discrimination occur in instru al
conditioning just as the do i
inc ; stud-
ied much more e i ntal conditioning.

83
 CHAPTER 3 Instrumental Conditioning

Generalization
In a prototypical study of stimulus generalization, Guttman and Kalish (1956)
trained pigeons to peck at a key of a particular color (measured by wavelength).
During 60-sec intervals, the key was lit with a certain color, and pecking produced
a reinforcement of food. These intervals were separated by 10-sec intervals of total
darkness during which the pigeons did not respond. Following the experiment,
the key was illuminated at different wavelengths, and the number of key pecks
was recorded to test for generalization. Four conditions were defined by the wave-
length of the original key:'530 nm (green), 550 nm (greenyellow), 580 nm (yellow),
or 600 nm (yelloworange). After training, the pigeons were tested without rein-
forcement. Figure 3.1 shows, for each training condition, the number of respons-
es for different test wavelengths. Pigeons showed maximal response when the
test wavelength matched the wavelength on which they were trained. Their rate
of responding decreased as the difference increased between training and test
wavelengitt These:regulis donot, simpli rehec e Se Sta the
study stimulus from the test stimulus—that is, that pigeons responded to a test
color to the degree that they thought it was the study color; pigeons are capable
of making much sharper discriminations than those illustrated in Figure 3.1. In
some sense, pigeons were registering their’ opinion” on whether this difference in
wavelength was likely to be relevant to their reinforcement.
The curves in Figure 3.1 are often referred to as generalization gradients.
Many generalization gradients are not as steep as those depicted in Figure 3.1.
Figure 3.2 from Jenkins and Harrison (1960) illustrates a generalization gradient
from an experiment in which pigeons were trained to peck when a key was lit
and a 1000-Hz tone was on and then were tested for tones that varied from 300
to 3500 Hz. The data are plotted in terms of the percentage of all responses

500

600

400

3 530 580
a. S 300
g 550
Ke)
o
200
FIGURE 3.1 Pigeons are %
trained to peck at lights with
wavelengths of 530, 550, 580, 100
and 600 nm. The curves show
the total responses to stimuli of
similar wavelengths. These are 0
cumulative responses for 6 min. 460 500 540 580 620
(From Guttman & Kalish, 1956.) Wavelength, nm

84
 What Is the Conditioned Stimulus?

Ww jo)

NOoO

FIGURE 3.2 Rate of respond- an[e)

ing to tones of various frequen-
cies for pigeons trained to total
of
Percent
responses
respond to Plines of 1000: H. 0
(From Jenkins & Harrison, 0 1000 2000 3000
1960.) Frequency, Hz

250

200

Yn
B
S 150
ce

g
a)
®
—€ 100
=
=

50

FIGURE 3.3 Gradients of

inhibition for three pigeons fol- 0
lowing learning where only 570 480 510 540 570 600 630 660
nm was not reinforced. Wavelength, nm

given to that tone.! The generalization gradient curve is nearly flat, showing lit-
tle decrease in response as the tone varied from the training stimulus of 1000
Hz. Pigeons were registering their” opinion” that the actual pitch of the tone was
irrelevant to whether reinforcement would be delivered. The pigeons behaved as
if the only critical feature was that the key was lit and that it did not matter what
the tone was. In effect, they ignored the pitch of the tone. /
Figure 3.1 shows a positive generalization gradient, but negative general-
ization gradients are possible, too. Terrace (1972) created a situation in which
pigeons could receive reinforcement for pecking when the light was homoge-
neous white light and not when the light was a specific color (570 nm). They
were then tested with lights of specific colors. Figure 3.3 shows their rate of
responding as a function of wavelength. The minimum rate of responding

! The original Jenkins and Harrison data included a no-tone condition, which is not
shown.

85
 CHAPTER 3 Instrumental Conditioning

occurred at the nonreinforced frequency; the rate gradually recovered as the

wavelength moved away from this frequency.
Organisms have biological predispositions to treat certain dimensions as
significant amlCartan diferences onthesedimensions asimportant indefining
the CS, while theyignore other dimensions and differences. Organisms may pay
attention to different aspects of a stimulus in different situations. For instance,
Foree and LoLordo (1973) trained pigeons with a combined CS of light and
tone. When the pigeons were reinforced with food, it was the light that con-
trolled their behavior. When they were reinforced with shock, it was the tone.
This finding may reflect the fact that visual identification is critical to identifying
food for pigeons but sounds often signal danger. We have already discussed
such associative biases with respect to classical conditioning, and we will have
more to say about them later in this chapter.
seis a = A RS RIES IEPA RU ST EEE

Organisms spontaneously generalize the CS, ignoring certain

dimensions and certain differences in other dimensions.
A SESE EE IT RENE SI BP SCOTIA
SERED SESS ONSITE UE UES ER LEN LEAVES
LEE INGE ELLE LEE LEN LLL LEELA LEE BLESEDLLBEEEELEBED EDD

Discrimination
Although organisms have biological predispositions to attend to certain dimen-
sions Britciteererigea GRU 19apenore“omereothey ualitche epee Eee if
experience contradicts their biases. For instance, what happens if the organism
is exposed to multiple stimuli that it initially treats as equivalent, but learns that
some are accompanied by reinforcement and others are not? The simplest pos-
sibility is an experiment in which the presence of a stimulus is associated with
reinforcement and its absence is not. Jenkins and Harrison (1960) looked at
what would happen in such a condition. Recall from Figure 3.2 that, when there
was only a positive stimulus of 1000 Hz, pigeons pecked at the lighted key no
matter what the frequency of the tone. Jenkins and Harrison compared this con-
dition with a condition of differential training: when the key was lit and there
was a 1000-Hz tone, the pigeons were reinforced for pecking the key, but when
the key was lit and there was no tone, they were not reinforced for pecking the
key. Figure 3.4 shows the results. There are strong generalization gradients
around 1000 Hz. The effect of the discrimination training was to indicate that
the tone was relevant.
This experiment compared the presence of a tone with the absence of a
tone, in contrast with many other experiments in which different values of a
stimulus were positive and negative. In another experiment by Jenkins and
Harrison (1962), pigeons were first reinforced for pecking in the presence of a
1000-Hz tone and not in the absence of a tone, as described earlier. Then the
pigeons were trained to respond to a 1000-Hz tone but not to a 950-Hz tone.
Figure 3.5 compares the generalization gradients of a pigeon before and after
learning that the 950-Hz tone was negative. The generalization gradient is much
steeper after the animal was trained to discriminate between a 1000-Hz tone

86
 What Is the Conditioned Stimulus?

40

205

10/=
FIGURE 3.4 Generalization gradi- total
of
Percent
responses
ents following differential training Pe
with a 1000-Hz tone. Individual gra- 0
dients are based on the means of 0 1000 2000 3000
three generalization tests. Frequency, Hz

and a 950-Hz tone. This pigeon actually showed maximum response to a tone
of 1050-Hz, which is away from the negative 950 Hz. This kind of” overshoot’is
common in human behavior. If students observe that a 400-word essay got a C
and a 500-word essay got an A, they might write a 600-word essay. To explain
this phenomenon, the next section considers a popular theory of discrimination
wale:
Tran SR oa en eT ELLE SSL LI ELT ELLE
LL LAER LLL LDL E ELLEN DEERE LE EAE

ee canaie tridrted to discriminate among sti

ues and to respon
‘SR
RPE ARRAN NNR NRO EI RN APR GI TEIN TT NESE ET NESE LI ES STEEN NEN LATTE LTT BLN ENE IEREI SIE NE ELI ERIN,

1000 Hz
positive stimulus
100

-o- 950 not trained

30 -e 950 negative

60

40
min
Responses/

FIGURE 3.5 Generalization gra- 20

dients obtained from a pigeon
trained to respond to a 1000-Hz
tone and then later trained to dis-
criminate it from a 950-Hz tone. 0 1000 2000 3000
(From Jenkins & Harrison, 1962.) Frequency, Hz

87
 CHAPTER 3 Instrumental Conditioning

Spence’s Theory of Discrimination Learning

Spence (1937), a learning theorist strongly influenced by Hull (see Chapter 1),
developed a theory of how training on positive and negative stimuli combined ~
to produce a net gen t. Although more modern versions of his
theory feature various technical differencesthat make them more sophisticated
and accurate (e.g., Blough, 1975), Spence’s theory is described here because it
contains the essential ideas and is the original proposal. Earlier we learned that
if an animal is reinforced for the response in the resence of astimulus, itbuilds
4 pUSHIE FENGaMZaWON gradient (Figme 371)around theeimulus and Wan
animal isnot reinforced for the response in the presence
of a stimulus, it builds
a negative generalization @fadient (Figure 3.3)around that semulus- Spent
pence’s
basic tdea was that behavior indiscrimination training isjust acombination of
these positive and negative generalization_gradients. Figure 3.6 illustrates his
analysis. Suppose a circle of 256 cm? is the positive stimulus and one of 160 cm?
is the negative stimulus. Figure 3.6 illustrates the positive generalization gradi-
ent around 256 and the negative generalization gradient around 160.
Subtracting one from the other produces the net generalization gradient. Note

10
Positive effect

Net effect

Magnitude
excitation
of

Negative effect

39 62 100 160 256 409 655 1049

eect)

© 1Orcs ic DONA
Stimulus size
FIGURE 3.6 Spence’s theory of how inhibitory influences from the negative stimu-
lus subtracted from excitatory influences of the positive stimulus yield a net general-
ization gradient.

88
 What Is the Conditioned Stimulus?

that the positive peak of this gradient has been shifted from 256 in a direction
away from the negative stimulus. This is the prediction of a peak shift—the
stimulus that evokes the most responding is not the positi ini imulus
but one shifted away from it and the negative stimulus. This prediction is some-
what counterintuitive, since the organism 1s responding
i ae oaceipamied GithanteaSemele haeboca hohe TNS be
more to a stimulus that

diction is typically confirmed in discrimination experiments of this sort. Figure

3.5 from Jenkins and Harrison (1962) is one example of this peak shift; the
pigeon responded more to a 1050-Hz tone than to the 1000-Hz tone with which
it had been trained.
ey

Spence proposed that discrimination learning resulted from

subtracting generalization gradients for nonreinforced stimuli
om generalization gradients for reinforced stimuli.

Relational Responding: Transposition

Spence extended his theory to a simultaneous presentation procedure in which
the organism must select between two stimuli. Suppose that an organism is
trained to discriminate between stimuli of 160 cm* and 256 cm? given the gen-
eralization gradients illustrated in Figure 3.6 and is then given a choice between
two stimuli of 256 and 409 cm2. Because of the peak shift, the organism should
select 409 rather than the original positive 256. A number of experiments sup-
ported this prediction of a preference for the shifted stimulus rather than the
original.
This result was explained in another way by Kohler (1955) and other
Gestalt psychologists. Transposition wastheterm Kohler usedtoindicate that
the organism had transferred the relationship between one pair of stimu

organi wastesponding fo
thelationship PebuseaEh tenth
and had learned to select the larger. A long history of controversy has sur-
rounded relational accounts an unts like that of Spence, which propose
that the organism responds to the absolute value of the stimulus. This contro-
versy has been settled with the conclusion that both sides are right. Under
appropriate circumstances an organism can be trained to respond to a relation-
ship between two stimuli, and under other circumstances it can be trained to
respond to the absolute properties of the two stimuli.
An experiment by Lawrence and DeRivera (1954) provides an example of
animals responding relationally. Figure 3.7 illustrates the stimuli used: cards of
two shades of gray. In Figure 3.7 these shades are indicated by the numbers 1
through 7: 1 is white, 7 is black, and the other numbers denote the various
shades between. The bottom half of the card was always 4 and the top half var-
ied. When the top half was lighter (1 to 3), rats were trained to turn right; when
it was darker (5 to 7) they were trained to turn left. The critical test occurred after

89
 CHAPTER 3 Instrumental Conditioning

Training stimuli for right turn

Training stimuli for left turn

6
Test stimuli

vate
FIGURE. 3.7 Stimuli used by Lawrence and DeRivera
(1954). The numbers 1 through 7 denote shades of gray.

training.The rats were presented with a card with 3 on top and 1 on the bottom.
Both 3 and 1 were associated with moving right, but the top was darker than the
bottom and this relation was associated with turning left. The rats responded to
the relational information and turned left. In contrast, when they were tested
with a 5 on the top and 7 on the bottom, they went right, again confirming the
relational theory.
The fact that organisms can encode and respond to either relational or
Aisharea mecerttreng resem SciYaiccai cam problem
IS TN in the
eTdiscussion
of what constitutes the conditioned stimulus. It is not immediately apparent
ee adc aC
be way(6g,absolutesis)an-anoMr-a-atTen ayGelalve ize),
Without Knowing how the stimulus is encoded, it is not possible to know what
patterns of generalization and discrimination will take place. Researchers and
theorists typically assume what seems to be the obvious encoding. But what
seems obvious to the experimenter may not seem obvious to the organism.
Chapter 6 has more to say about how information is represented, particularly in
the human case.

The Gestalt psychologists proposed that organisms responded

to the relationship between stim e
absolute values.
SSSR ay schiocenncinunconren ” tes SINT SRCUNRES

Dimensional or Attentional Learning

Thus far we have focused on patterns of generalization and discrimination along a
single dimension. However, most stimuli have many dimensions. For instance,
visual stimuli have color, size, shape, and position in space. In addition, there are
various background contextual stimuli, such as the appearance of and possible
sounds in the laboratory. How is the organism to identify which dimension or

90
 What Is the Conditioned Stimulus?

dimensions determine reinforcement? The last chapter described one theory of

dimensional combination for classical conditioning, the Rescorla-Wagner theory.
According to that theory, various dimensions or stimuli divided a total associative
TW Teee ereaecontsten oh theUSIneffed,fhey
competed for association to the US. A similar process seems to occur in instru-
mental conditioning in which stimuli compete for association to the reinforcer.

(Mackintosh, 1974). Blocking phenomena cost Wher onesiamalne ordimer

sion becomes so strongly associated that it blocks out other dimensions. The
blocking data are among the strongest data in support of the Rescorla—Wagner
theory. On the other hand, in classical conditioning, there is also evidence that
learning cannot always be simply a matter of responding to individuakdimen-
sions because animals can be trained to respond to various combinations of
dimensions but not to the individual dimensions (Razran, 1971).
Instrumental conditioning paradigms have been used to explore a some-
Weta oad tecnica oe acne Cannio eT
encoding capacity and canonly payatfention fosomany dimensions atatine.
With_experience, they can change which dimensions they-attend to. For
instance, flat generalization gradients can be transformed into peaked general-
ization gradients by discrimination experiments that simply make the dimen-
sion relevant (contrast Figures 3.2 and 3.4).
Another kind of evidence for dimensional learning (sometimes called
attentional learning) comes from experiments that involve learning multiple,
successive discriminations. The basic paradigm is illustrated in Figure 3.8. A
value on one dimension is reinforced. (In the training example of Figure 3.8, this
is red on the dimension color.) After mastering this discrimination, the subject
is transferred to a condition in imension

another dimension is used (a nonreversal shift—in Figure 3.8, squares now

become positive). Reversal shifts might appear more difficult because the organ-
ism must eos in the completely Eee way. On the other hand, nonre-

attention to a new dimension. Most betes and higher apes find reversal shifts
easier, whereas very young children and nonprimates find nonreversal shifts
easier (Mackintosh, 1975). Interestingly, adult humans with damage to their
prefrontal cortex often have difficulties in reversal conditions as well (Owen,
Roberts, Hodges, Summers, Polkey, & Robbins, 1993). The frontal cortical areas
are much expanded in primates and mature in children later than most other
neural structures. Chapters 6 and 9 will elaborate on the role of the frontal cor-
tex in primate and human learning.
Figure 3.8 also illustrates intradi shift, which requires the sub-
ject to learn to discriminate betw as blue and green) on the
previously relevant dimension. Thissituation iscontrasted withlearning new
values on
onthe
theother
otherdimension
mens (extradimensional
shift). Intradimensional shifts
shifts(c.g., Mackintosh & Little,
afealmost always easierthan-extradimensional

91
 CHAPTER 3 Instrumental Conditioning

Training

Positive |F|

Negative

Reversal shift

Positive Y

Negative
iva
Nonreversal shift

Positive ¥.

eld
Negative

Intradimensional shift

Positive

Negative | G
tele
aieie
Paleer
Extradimensional shift

FIGURE 3.8 Schematic representation of stimuli and Positive

reinforcement contingencies for reversal shifts, nonre-
versal shifts, intradimensional shifts, and extradimen-
sional shifts.
Negative
balabor
1969). The_contrast_ between intradimensional and_extr
unlike the contrast between reversal and nonreversal shifts, does uir ia)
that the organism respond to the same stimuli in differe there
are no competing responses to the stimuli from the original training. A major
function of the prefrontal cortex appears to be the inhibi jate
responses (Dempster, iamond, 1989; Roberts, Hager, & Heron, 1994).
Trust peatthatone
thinghatalorganismsear what mension
are relevant Therefore, all organisms find intradimensional shifts-easierthan
extaddanensgnalshitsDegiise Cheyoe ie ee eel
dimensions. In the case of organisms with developed and intact prefrontal cor-
tices, they can inhibit sonia responses and so also find reversal shifts easi-
er than nonreversatsht fts-Lowerorsanisms-dotiot stiow such dramatic results
but still find intradimensional shifts easier than extradimensiona] shifts.

Organisms can learn which stimulus dimensions are relevant

in discrimination learning.
a at Sc une eo acco UNL Natermed

92
 What Is the Conditioned Stimulus?

Configural Cues and Learning of Categories

In the chapter on classical conditioning, we discussed the evidence that organ-
isms could learn to respond to configurations of dimensions as well as individ-
ual dimensions. Similar demonstrations of configural responding exist in the
instrumental conditioning domain. In Chapter 2 we described configural cues as
if they were the exception. However, some theorists (e.g., Pearce, 1994) have
argued that animals always r onfiguration of dimensions
that-a stimulus presents. The different generalization gradients they show on
different di ifterent similarity of these idimen-
sional stimuli when they are contrasted on single dimensions.
OTe a eeeerage ne ecg crag cial inieates at
organisms respond to categories of objects rather than single dimensions of
gbipets- Portrstance-wherwesee-arObjectsuchasachairweareresponding
to it as a configuration of dimensions that indicate a chair rather than any sin-
gle dimension. Apparently, other organisms also see the world in terms of cat-
egories rather than single dimensions. Figure 3.9 shows some of the stimuli
shown to pigeons in a discrimination experiment by Herrnstein, Loveland, and
Cable (1976). Some pigeons were trained to peck at instances of the category
“tree”; they were trained with some 700 slides of trees and nontrees. The only
characteristic that the positive pictures had in common (and that discriminated

FIGURE 3.9 Four typical pictures used in the experiment by Herrnstein, Loveland,
& Cable (1976). (Negative stimuli are on the left; positive on the right.)

93
 CHAPTER 3 Instrumental Conditioning

them from the negative pictures) was that they involved a tree. The positive pic-
tures could not be discriminated from the negative pictures on the basis of sim-
ple features. Thus, pigeons could only make this discrimination if they knew
what a tree was. For humans, this is a relatively easy discrimination because they
possess the category of trees. It also turned out to be a fairly easy discrimination
for pigeons. Pigeons not only were able to learn to make such discriminations,
but they learned in fewer trials than needed in the simple one-dimensional
problems described earlier. Moreover, after being trained to discriminate one $et
of pictures of trees from a set of pictures of nontrees, the pigeons were capable
of generalizing this ability to new pictures that had not been used for training.
Wasserman, Kiedinger, and Bhatt (1988) demonstrated category learning
by pigeons in a slightly different paradigm. Pigeons were trained to peck at four
different keys according to the rules:

Peck key 1 if the stimulus was one of 10 cat pictures.

Peck key 2 _ if the stimulus was one of a second set of 10 cat pictures.
Peck key 3 _ if the stimulus was one of 10 flower pictures.
Peck key 4 if the stimulus was one of a second set of 10 flower pictures.

Pigeons got quite good at discriminating keys 1 and 2 from 3 and 4, corre-
sponding to the cat—flower distinction. However, they had great difficulty in dis-
tinguishing key 1 from 2 (the cat pictures) or key 3 from key 4 (the flower pic-
tures). They found it difficult to learn discriminations within a category. Humans
would show similar patterns, finding discriminations between categories easy
and discriminations within categories hard.
Chapter 10 provides much more information on concept learning, focus-
ing mainly on human learning of concepts. The experiments just described illus-
trate that lower animals as much as humans see the world in terms of categories
and specific objects and not in terms of single dimensions like colors and
shapes. Often this meaningful representation of the world is much more salient
than the simple dimensional representation, and animals find it easier to learn
discriminations when the discriminating factor is a salient category.

Animals easily learn to respond to complex dimensional com-

binations that define significant categories.

What Is the Conditioned Response?

The next question to address concerns the nature of the response. The tradi-
tional view was that a specific response was being learned. As early as the 1920s,
however, researchers began to see problems with that particular point of view.
Muenzinger (1928) trained guinea pigs to press a bar and found that sometimes

94
 What Is the Conditioned Response?

they pressed with one paw, sometimes with another, and sometimes even with
their teeth! Macfarlane (1930) taught rats to swim through a maze for food and
then found that they were capable of running the maze for food. Lashley (1924)
taught monkeys to solve a manipulation problem with one hand and found they
could generalize the solution to the other hand when the first was paralyzed. It
seems that organisms come to some representation of the functional structure
of their environment and sélect their responses appropriately. Thus, the guinea
pigs in Muenzinger’s experiment wére learning not that a particular response
was associated with reinforcement, but rather that depression of the bar was
associated with reinforcement. As in classical conditioning, the response is the
organism’s adaptation to what it has learned about the environment.
Sra eerie vecrerreeet GRE rte OFtheTeopOTce Tene definition of
an operant. Different responses that had identical effects in the environment
(had identical reinforcement consequences) were defined to be instances of the
same operant. Organisms can be trained to discriminate among responses that
appear to have equivalent effects on the environment (e.g., use of the left hand
versus the right hand to press a bar) if the experimenter sets up reinforcement
contingencies that differentiate them. However, they behave as if their default
assumption were that actions with equivalent effects on the world produce
equivalent rewards—certainly a plausible default assumption.

rganisms tend not to discriminate among responses that are

equivalent in their ef ect onthe-environment.
suas

Maze Learning
Maze learning by rats provides some of the strongest evidence that the organism’s
response is an adaptation to what it has learned about its environment. Rats are
animals whose natural environments are much like mazes, and they are skillful at
learning complex mazes, challenging humans in their ability. As noted in the dis-
cussion of Tolman in Chapter 1, rats’ ability to navigate in mazes depends in part
on their developing cognitive maps. They learn the locations of food and other
objects in space and traverse the maze to get to those locations. However, there is
also evidence that rats can learn the specific turns involved in navigating a maze.
More recent research has revealed some of the other ways in which rats
cope with mazes. Research (e.g., Olton, 1978) has been conducted with a radial
maze such as that shown in Figure 3.10; the rat is put in the center of the maze,
and food is placed at the end of each of the eight arms. Rats on their first
encounter with this maze tended to perform very well, visiting about seven of the
eight arms in their first eight choices. The rats displayed an amazing ability to
avoid revisiting these arms.* How were they able to explore these mazes so effi-
ciently? One might think that the rats had some systematic plan such as going

2 Only in the second edition did I notice the pun in this sentence.

=)3)
 CHAPTER 3 Instrumental Conditioning

FIGURE 3.10 A top view of a radial maze.

Source: From D. S. Olton and R. J. Samuelson.
Remembrance of places passed. Spatial memory
in rats. Journal of Experimental Psychology. Animal
Behavior Processes, Volume’2. Copyright © 1976 by
the American Psychological Association.
Reprinted by permission.

through all the arms in a left-to-right order. This does not seem to be the answer,
however, because they did not display any specific order in visiting the arms.
Rather, the evidence is that rats have good memories for locations and avoid
repeating visits. This is an adaptive trait in their natural environment, where they
need to keep track of where they have been and consumed food. If they have
depleted the food in a particular location, there is no point in repeating the trip.
Other research on rats has compared their ability to learn shift versus stay
strategies in a T-maze (Haig, Rawlins, Olton, Mead, & Taylor, 1983). AT-maze (see
Figure 3.11) is a simple maze in which a rat runs from a start box to a choice point,
at which it must go in one of two directions. There are goal boxes to the left and
to the right, and one of them contains food. Shift and stay strategies refer to two
different principles experimenters have used to determine which goal box to place
food in. The strategies differ in terms of where to look for food after the first trial.
If the rat is being trained with a stay strategy, it finds food if it goes down the path
that had food before. In a shift strategy, it finds food if it goes down the other path.

FIGURE 3.11 An example of a T-maze.

96
 What Is the Conditioned Response?

Rats find it much easier to learn the shift strategy. This result is just the opposite
of what would be predicted if rats were learning specific responses, but it is exact-
ly what would be predicted from their foraging habits in the wild, where they need
to avoid food locations that they have already depleted. Several other species also
show this tendency to learn better with a shift strategy (e.g., Kamil, 1978).
Interestingly, animals find shift strategies harder to learn when they are not
allowed to deplete the food in the goal box (Haig, Rawlins, Olton, Mead, & Taylor,
1983); then they have a reason to return to the same part of the maze.

Rats navigate in various environments according to a cogni-

tive map, which includes where significant objects like food
are to be found.

Response Shaping and Instinctive Drift

In our discussion of Skinner in Chapter 1, we introduced the idea of response
shaping. This is a way to train animals to produce specific behaviors they are
unlikely to emit naturally. The basic idea is that the animal always emits some
range of behaviors and shaping involves selectively reinforcing ever closer
approximations to the target response. For instance, Skinner succeeded in train-
ing pigeons to play a simplified Ping-Pong game. Initially, he reinforced them
whenever they faced the Ping-Pong ball; later he withheld reinforcement until
they approached the ball; and still later he reinforced them when the beak made
contact with the ball. Eventually, the pigeon was only being reinforced for actu-
ally hitting the ball. Parents use similar procedures to reinforce children’s behav-
ior such as their social skills. Parents begin reinforcing simple“hello’s”,“ please”,
and “thank-you” and eventually (if their training skills are good) they have off-
spring who are graceful members of society.
As parents will report, however, even the most careful shaping schedules
sometimes backfire and the learner slips into undesired behavior. At least with
lower organisms, part of the problem is that the organism’s instincts about
appropriate responses can get in the way of such response shaping. Chapter 1
mentioned that a pig was trained to go through an elaborate set of procedures
mimicking the morning routine of a human. Pigs are normally easy to train, but
the trainers described a problem with what they termed instinctive drift
(Breland & Breland, 1961). They wanted to train a pig to take a large wooden
coin and place it in a piggy bank. The pig was able to learn this behavior quite
well given food reinforcement, but after a few weeks, instead of putting the coin
in the bank, it would repeatedly drop the coin, root it (dig or turn it up with its
snout), and toss it up in the air. The pig became useless as a performer, and the
Brelands had to train another pig, which soon developed the same problem.
This behavior is part of the natural food-gathering behavior of pigs. They had
come to regard the coins as a food and consequently began to behave toward
the coins as they did toward food.

97
 CHAPTER 3 Instrumental Conditioning

The Brelands reported a variation of this problem when they tried to train
raccoons to place coins in a container. The raccoons began engaging in behaviors
that corresponded to washing and cleaning the food. Although the intrusive
behavior was different from that of the pigs, the behavior was part of the species-
specific food-gathering behavior of raccoons. Thus, organisms’ instinctive response
patterns can overwhelm responses carefully shaped by instrumental conditioning.

Attempts to shape behaviors in organisms may be frustrated

by species-specifis response patterns.
Sg §9-R NSS VERN SE
DETR SUES SEY SSE RE OE SE ETOCS

Autoshaping
The previous subsection described how conditioning efforts can be frustrated by
organisms’ biological predispositions toward appropriate response patterns. A
somewhat different result can also happen: experimenters can train a behavior
without trying. A much-studied example, called autoshaping, was discovered
by Brown and Jenkins in 1968 in their work with pigeons. At irregular intervals
they illuminated a response key and then followed the key with food from a
grain dispenser. Although the birds did not have to peck at the key to obtain the
food, they all started to peck at the key. They wound up behaving as if there had
been a contingency between pecking and food.
Considerable effort has been made to understand why the pigeons would
peck at a key when it was unnecessary. One enlightening experiment was per-
formed by Jenkins and Moore (1973). They deprived the pigeons of either water
or food, and then they used the autoshaping procedure of the illuminated key
followed by the appropriate reinforcer. All the pigeons began pecking at the key,
but the way in which they pecked at it differed depending on the reward. When
a pigeon had been deprived of food and the reward was food, the pigeon pecked
with an open beak and made other movements similar to those that pigeons
make when they are eating. When the reward was water for a pigeon deprived
of water, the bird pecked at the response key with a closed or nearly closed beak;
again, this and other features of the pecking movement were like the move-
ments that pigeons make when drinking.
These results can be interpreted as examples of classical conditioning. That
is, the lit key is a CS that predicts the US of food or water, and the animal is giv-
ing a conditioned response of pecking to that CS. Although this interpretation
may be basically correct, it fails to capture the full complexity of autoshaping
behavior. A good example of this complexity was observed by Timberlake and
Grant (1975) in a study of autoshaping in rats. Two groups of rats received differ-
ent CSs presented in advance of the delivery of a food pellet. For one group the
CS was a block of wood; the rats came to gnaw at the wood. For the second group
the CS was another rat; in this case the rats approached the other rat and engaged
in various social behaviors, such as sniffing and grooming. Thus, depending on the
CS, rather different behaviors were autoshaped. The difference makes sense if the

98
 Association: Contiguity or Contingency?

Pigeons pecking for water (top row) and for food (bottom row).

eating behaviors of rats are considered. Rats usually eat in groups and display
social behaviors to other rats while eating; they also gnaw at inanimate objects as
part of their eating behavior. There is a complex species-specific pattern of eating
behavior, and different aspects of it are selected by different stimuli.
Generally, the lesson of the research on autoshaping and instinctive drift
is that animals come to learning situations with strong patterns of instinctive
behavior. These patterns may cause the organisms not to learn what the exper-
imenter intended but rather something else. With respect to the issue of what
the conditioned response is, this research shows that the organism is not just a
bundle of simple muscle movements waiting to be conditioned to a stimulus.
Rather, the responses are parts of existing behavioral systems, and their condi-
tioning cannot be understood unless these systems are understood. Timberlake
(1983, 1984) uses the term behavior systems analysis to refer to this approach
that emphasizes the natural, unlearned organization of behavior for a species.

Autoshaping occurs when a stimulus evokes some species-spe-

cific behavior because of its association with a reinforcer.

Association: Contiguity or Contingency?

One issue involved in the case of classical conditioning is whether the learning
is produced because the CS and the US are contiguous or because they are con-
tingent. The corresponding issue in the case of instrumental conditioning is
whether learning is produced because the response and reinforcer are contigu-
ous or because they are contingent. Again, contiguity is the requirement that the

99
 CHAPTER 3. Instrumental Conditioning

two occur in close temporal proximity; contingency is the further constraint of a

predictive relationship between the two. For example, drinking a glass of water
and feeling healthy may be contiguous, but this does not mean that drinking
water produces a feeling of being healthy, because the person may usually feel
healthy. For there to be a contingency, the probability of feeling healthy would
have to be greater after drinking a glass of water than otherwise.
Experiments have varied the probability that the reinforcer would be deliv-
ered when the response was made versus when the response was not made—on
the analogue of the Rescorla experiment discussed in Chapter 2. For instance,
Hammond (1980) trained rats to press a bar for reinforcement in an experiment
that involved four phases. Figure 3.12 illustrates the results in each phase.

Phase 1. If the rats pressed the bar in any 1-sec interval, they had a
chance of a reinforcer. Hammond shaped the rats to a point where they
were receiving reinforcers after only 5 percent of these response-filled, 1-
sec intervals. The rats were making about 3000 bar presses an hour.
Phase 2. Hammond began giving reinforcements 5 percent of the time
when 1 sec passed and no response had been made. He still gave a reward
5 percent of the time when a response was made, but the reward was no
longer contingent on response—it was equally likely whether or not a
response had been made. The rats’ rate of responding dropped off rapidly
until they were making virtually no responses. Thus, even though the
same degree of contiguity of response and reward was maintained, rats
stopped responding because there was no longer a contingency.

Phase 1 Phase 2 Phase 3 Phase 4

/K— .05 - .0 —>}<——. 05 ~ .05 ——>}«——..05 - .0 —>«——.05 - .05 ——>|

5000

= 4000
wn
a
S 3000
[oy
3
= 2000

1000

fy lleeeesthml f sini ia ls OY nA ret

5 10 IRS} 20 25) SOS 235s F404 at5 50 55 = "60 65
Sessions

FIGURE 3.12 Responses per hour for rats when there is a contingency between
pressing and reinforcement and when there is not. Source: From L. J. Hammond.
Journal of the Experimental Analysis of Behavior, The effect of contingency upon the
appetitive conditioning of free-operant behavior, 34, 297-304. Copyright © 1980 by
the Society for the Experimental Analysis of Behavior, Inc. Reprinted by permission.

100
 Association: Contiguity or Contingency?

Phase 3. Hammond stopped giving reinforcers when the rats did not
respond, and the response rate of the rats picked up.
Phase 4. Hammond removed the contingency again, and the response
rate went down again.

These animals were shown to be sensitive to the experimenter’s contingencies,

just as animals were shown to be sensitive to CS—US contingency in classical
conditioning.

Organisms display conditioning when there is a contingency

between response and reinforcement.

Superstitious Learning
Some of Skinner’s famous experiments (Skinner, 1948) on what has been called
superstitious learning were thought to be evidence that contiguity was suffi-
cient for learning and contingency was not necessary. Food was made available
to pigeons from a feeder at fixed intervals (e.g., 15 sec for some, longer for oth-
ers) regardless of what they were doing. Although there was no contingency
between behavior and reinforcement, pigeons in this situation developed high-
ly routinized behaviors. One pigeon turned counterclockwise; another thrust its
head into the upper corners of its cage. Skinner reasoned that these systematic
behaviors appeared because of accidental contiguities between what the pigeon
was doing and the delivery of food. For instance, when the food was delivered,
the pigeon might be hopping from one foot to the other. The contiguity between
this response and the food would increase the pigeon’s tendency to hop from
one foot to the other and would thus increase the chance that the pigeon would
be engaged in this behavior the next time the food was delivered, increasing the
tendency for the behavior even more, and so on, until the pigeon would always
be hopping from foot to foot. Thus, even though there was only accidental con-
tiguity between behavior and reinforcement and there was no contingency, con-
ditioning would occur. In effect, the pigeons developed the superstition that
their behavior was necessary for the reinforcement. Skinner speculated that this
might be the cause of superstitious behavior in humans, such as rain dances to
produce rain; sometimes rain dances are indeed followed by rain, but, presum-
ably, they do not produce the rain.
Subsequent research and analysis have raised doubts about Skinner’s
interpretation of these experiments. Staddon and Simmelhag (1971) repeated
the superstition experiment and replicated many of Skinner’s results. However,
they demonstrated that the situation was more complicated than Skinner real-
ized. They noted that the pigeons’ behavior could be divided into two categories.
Immediately after receiving a reinforcement, pigeons displayed interim behav-
iors. There was a wide variety of such behaviors, including the sort Skinner
reported. After a while, pigeons began to engage in terminal behaviors, clearly

101
 CHAPTER 3 Instrumental Conditioning

in anticipation of the next feeding. This terminal phase always involved some
variety of pecking.
Staddon and Simmelhag’s results present serious difficulties for any
attempt to explain superstitious behavior as learning by contiguity. First, there is
no reason for two segments; second, there is no reason for all pigeons to peck
in the terminal segment, which is contiguous with the reinforcement. Staddon
and Simmelhag argued that terminal behaviors should be understood as exam-
ples of autoshaping, which, as we have discussed, is perhaps best thought of as
a classical conditioning phenomenon.
Although each pigeon evolved systematic interim behaviors, these behav-
iors were not contiguous with reinforcement and thus whatever caused them
was not learned by contiguity. Therefore, what was contiguous was not instru-
mentally conditioned but was classically conditioned, and what might be instru-
mentally learned was not contiguous. Staddon (1983) suggested that these
interim behaviors often served other functions, such as grooming or exercise.
According to this view, human behavior is often analogous to that of rats
in these experiments. Many of us eat on rather fixed schedules. When food is not
likely, we often engage in predictable interim behavior (e.g., studying or watch-
ing television). When food is likely, we engage in predictable terminal behavior
in anticipation of the food (e.g., going to the kitchen and setting the table).

Given food at fixed intervals, organisms will first engage in

interim behaviors when food is not likely and then in terminal
behaviors when the time for food approaches.

Partial Reinforcement
The experiment by Hammond (Figure 3.12) used a partial reinforcement sched-
ule; that is, only some of the responses were rewarded. It is sometimes hard to
discern that the partial reinforcement rate for a response is greater than the back-
ground rate of reward. Suppose that the probability of getting a reward in 1 sec is
5 percent if an animal presses a bar, but 4 percent if the animal does not press the
bar. The animal might fail to detect the contingency and not display conditioning.
When organisms are being maintained on partial reinforcement schedules,
especially schedules with low rates of reinforcement, they also have a problem
discriminating when extinction begins. It is easy to discriminate 0 percent rein-
forcement in extinction from 100 percent during conditioning, harder to discrim-
inate 0 percent from 25 percent, and much more difficult to discriminate 0 per-
cent from 1 percent. Organisms are found to take longer to extinguish after train-
ing on a partial reinforcement schedule, and their resistance to extinction
increases as the reinforcement rate is lowered. This phenomenon is called the
partial reinforcement extinction effect. It is a bit paradoxical because it implies
that the less reinforcement received in the past, the slower the organism is to give
up on an activity. This effect has interesting implications for molding the behav-

102
 Association: Contiguity or Contingency?

ior of people. For instance, if parents want their children to be persistent in pur-
suing a goal in the face of adversity, it suggests that they should only occasional-
ly reinforce their children’s goal-seeking activities. Eisenberger, Heerdt, Hamdi,
Zimet, and Bruckmeir (1979) demonstrated that children completed more work
in handwriting and mathematics if they had been partially reinforced in the past.
Partial reinforcement increases resistance to extinction because the condi-
tions under which the animal learns are similar to the conditions of extinction.
Basically, the animal learns to respond to the features that occur during extinc-
tion. Several researchers have proposed what these features might be. Capaldi
(1967) suggested that during learning organisms come to associate sequences of
nonreinforced responses with eventual reinforcement. Thus, in extinction, when
the organism encounters a sequence of nonreinforced trials, it expects rein-
forcement. Amsel (1967) proposed that during initial training, the organism
becomes frustrated when it does not receive reinforcement and has associated
its frustration with reinforcement. Thus, when frustrated in extinction, it also
expects reinforcement. Both theories have in common the idea that the partial-
ly reinforced organism learns to associate reinforcement to the kinds of features
encountered in extinction.

Conditioning is more difficult in partial reinforcement sched-

ules, but such schedules result in greater resistance to extinction.
cd

Learned Helplessness
Perhaps the most dramatic evidence that organisms can be aware of the contin-
gency (or lack thereof) between their behavior and reinforcement is found in the
learned helplessn at a scone experiment by
re given painful shocks at unpredictable

whereas the experimental group could do nothing to escape the shock. Thus,
one group of dogs learned a behavior that would eliminate shock, whereas the
other did not.
Both groups were then placed in the same escape avoidance condition:
they could avoid the shock if they jumped over a barrier after hearing a tone.
Dogs in the control group, which could control their shock in the first phase,
readily learned to jump over the barrier. In contrast, the experimental dogs
whined and yelped but made no attempt to escape the shock. After many trials,
the animals simply lay down and hardly moved at all. They had learned that
nothing they could do would prevent shock—that there was no contingency
between their behavior - 1s nt nl ois

Shocredictns of Whetheritwillreceiveperoce orsso the organ-

ior has petia poor
ism continues to assume its behavior will have no effect in a situation where it

103
 CHAPTER 3 Instrumental Conditioning

could learn to escape shock. This situation is like latent inhibition in classical con-
ditioning, where an organism comes to ignore a certain CS (see Chapter 2), or
like dimensional bapeae? in instrumental ea where bie Se oat

vforc t oeg.,Tob,1989).Ifa Tones ae ongtiomi not been aenoeeied

with food, they will continue to ignore it when it acquires such a contingency.
Similar effects occur in many situations with many species, including
humans. ome argue that this may beas is behind such phenomena as math

- a 8: a
noe and Seligman (1975) showed that humans mee to a long series of
unsolvable anagram problems failed to learn other easy-to-learn experimental
tasks. Seligman (1975) also suggested that clinical depression may be a variety

To deal with these clinical problems, Seligman has suggested a number of

measures, based on analogy to research with dogs. If a helpless dog is forced to
cross the barrier enough times with success, it will eventually cross on its own.
By analogy, depressed patients might be helped by exposure to success experi-
ences. Dogs can also be immunized by initial exposure to situations where they
can escape from shock; they are then less likely to learn helplessness when later
exposed to inescapable shock. By analogy, early successes in mathematics for
children, earned by their hard work and efforts, may inoculate them against later
math difficulties, developing in them the tendency to persist in the face of diffi-
culties or failures. However, given what we know about partial reinforcement
(previous subsection), a schedule of “partial success” would probably be more
effective than a schedule of“success only” in promoting persistence in the pres-
ence of temporary future (Dweck, 1975; Kennelly, Dietz, & Benson, 1985).

Granea pn veponteany received unavoidable aversive

stimuli come to ignore therelationship between their behavior
ay
dicracaronincntat euicoutes,
a Bitty

Associative Bias
Although organisms may be capable of learning many response—reinforcer
associations, they are biologically predisposed tolearncertain associations, just
as they are predisposed to learn certain stimulus-stimulus associations in clas-
sical conditioning (e.g., taste-poisoning discussed in Chapter 2). A pigeon can
more readily learn to peck to receive food than to avoid shock (Hineline &
Rachlin, 1969; MacPhail, 1968; Schwartz, 1973), but it can quite readily learn to
flap its wings to escape shock (Bedford & Anger, 1968). These outcomes make

104
 Association: Contiguity or Contingency?

sense because pecking is part of the pigeon’s eating repertoire and wing flap-
pageis tt of itseli ho of escape behaviors.
hettlewor 75) did an interesting analysis
of the effects of reinforce-
msters. She ma that they tended to engage
in1 certain behaviors when hungry, such as standing on their hind legs (which
she called open rear), scraping at walls (scrabbling), and digging in the ground.
Other activities, such as washing their faces, scratching, and marking (pressing
a scent gland), did not increase when they were hungry. Different hamsters
were reinforced by food for each of these six behaviors. Figure 3.13 shows the
results. Subjects learned to increase the oe behaviors but not the noneating

Scrabble

500

400

Open rear

WwoOiS

Mean
sec

FIGURE 3.13 Mean time

spent performing the rein-
forced response per 1200-sec 200
session. Source: From S. J.
Shettleworth. Reinforcement
and the organization of
behavior in golden hamsters.
Hunger, environment, and
food reinforcement. Journal of 100} Prine
Experimental Psychology. Ani-
mal Behavior Processes, Volume iPad iy eels
104. Copyright © 1975 by the
American Psychological Asso- .cx Aa
ciation. Reprinted by permis- 0 Mark
sion. Reinforcement sessions

105
 CHAPTER 3 Instrumental Conditioning

behavior. For instance, rats find it easy to learn to flee to avoid a shock but hard
to learn to press a bar to escape shock. The relative ease of these two responses
is reversed if the reinforcer is food.
Humans face difficulties in skill learning when the skills entail learning
responses that are antagonistic to human predispositions. For example, in
downhill skiing the skier leans forward to control speed and should lean for-
ward more the steeper the hill. Most beginners have difficulty because of their
natural tendency to lean backwards. As another example, when a car is skidding
on an icy road the driver needs to turn into the skid and not slam on the
brakes—drivers have great difficulty learning the appropriate response and
inhibiting the incorrect response.

Organisms are biologically prepared to learn certain

response—outcome combinations.
ELM

Instrumental Conditioning
and Causal Inference
We have focused on instrumental conditioning experiments from the animal’s
perspective. But human subjects can be placed in similar situations. Imagine what
it would be like if you were put in a room to explore and discovered that some-
times when you flipped a switch on the wall, money came forth. If you thought
you would be able to keep any money you found, you might find yourself flipping
that switch as fast as a rat pushes a lever or a pigeon pecks a key. Your perfor-
mance could be plotted in cumulative response records, and we could speak of
you as learning an association between the switch and money. To speak of it as an
association, though accurate, would probably not fully express your mental state.
You probably also would have formed the belief that flipping the switch caused
the money to come forth. It is unclear to which other organisms such causal
beliefs may be ascribed, but it is appropriate to ascribe them to humans.
Wasserman (1990) studied the development of humans’ causal beliefs in
instrumental conditioning paradigms and found that these causal beliefs devel-
op much as associations do in lower organisms. Subjects were given a key,
which they were encouraged to press. Sometimes when the subject pressed the
key a light went on, and sometimes when the subject did not press the key the
light went on. The light was like a reinforcer (or in this case more like a neutral
stimulus) that followed the response. Wasserman varied the probability that the
key press would be followed by the light. He broke the experiment into 1-sec
intervals. If a subject pressed the key in the interval, the interval would end with
a light flash with different probabilities in different experimental conditions. He
used probabilities of 0.00, 0.25, 0.50, 0.75, and 1.00. These probabilities were
referred to as P(O|R) for probability of outcome given response. Wasserman also

106
 Instrumental Conditioning and Causal Inference

100 ~~ P(O|R) = 1.00

“> P(O|R)=0.75
- P(O|R) = 0.50
50
“= P(O|R) = 0.25
~- P(O|R) = 0.00

ro)

subject
Average
rating
20)

FIGURE 3.14 Causal inference as

a function of the probability of a
light given a key press and the -100
probability given no key press. 0.00 0.25 0.50 0.75 1,00
(Data from Wasserman et al., 1993.) P(O|-R)

manipulated the probabilities that a 1-sec interval without a key press would
result in a light. These probabilities were referred to as P(O|-R), for probability
of outcome given no response, and they similarly took on the same values of
0.00, 0.25, 0.50, 0.75, and 1.00. Wasserman looked at all combinations of P(O|R)
and P(O|-R) for 5 x 5 = 25 conditions.
Wasserman asked subjects to rate the causal relationship between the
press and light on a scale that varied from —100 (prevents light) to +100 (causes
light). Figure 3.14 illustrates the results. As in the animal conditioning experi-
ments, subjects’ ratings of causal strength was a function of the difference
between P(O|R) and P(O|-R). The particular level of causal strength for a value
of P(O|R) depended on the value of P(O|-R). This is the same sort of relationship
Rescorla illustrated in his experiment on classical conditioning (see Figure 2.9).
Chapter 10 examines human causal inference further, but the research described
here indicates that causal inference may be closely related to conditioning.

Human judgments of causality are affected by the same con-

tingency variables that influence animal conditioning.

Application of the Rescorla-Wagner Theory

Wasserman, Elek, Chatlosh, and Baker (1993) showed that the behavior of their
human subjects could be predicted by the Rescorla—Wagner theory. First, let’s
consider how the theory would apply to instrumental conditioning in general.
Recall that in classical conditioning this theory assumes that the strength of asso-
ciation between the CS and the US changes according to the following equation:
AV=a(A
- V)

107
 CHAPTER 3 Instrumental Conditioning

where a is the learning rate; 4 is the maximum strength of association; and V is

the sum of the existing associative strengths from the CSs presented on that
trial. This theory can be mapped onto instrumental conditioning by letting the
experimental context and the response be two cues (i.e., the CSs) that are asso-
ciated to the reinforcement (i.e., the US). Then 4 represents the strength of asso-
ciation that can be conditioned to the outcome or reinforcement. When the out-
come occurs after a response, there are two cues for conditioning: the response
and the stimuli of the experimental context. If the outcome occurs without the
response, then only the contextual stimuli are present. This is a competitive
learning situation in which the response and the context are competing for
association to the reinforcement. This way of applying the Rescorla-Wagner
theory to instrumental conditioning predicts many features of instrumental con-
ditioning, just as it predicts the features of classical conditioning.
One outcome that the Rescorla—-Wagner theory predicts is the subject’s
sensitivity to the difference in reinforcer rates in the presence versus the absence
of the response. This sensitivity is seen in Hammond’s experiment on bar press-
ing with rats (Figure 3.12) and in Wasserman’s human analogue (Figure 3.14).
Chapman and Robbins (1990) showed mathematically that, according to the
Rescorla—Wagner theory, the competition between context and response results
in a strength of association to the response that is proportional to the difference
in reinforcement rates. Thus, the process by which people form causal inferences
corresponds closely to the predictions of an associative learning theory.

The Rescorla—Wagner theory can predict behavior in an

instrumental conditioning paradigm by assuming competitive
learning between context and response.

Interpretations
Two rather different conclusions are possible from this research on causal infer-
ence and the Rescorla-Wagner theory. One conclusion is that the simple associa-
tive learning processes of the Rescorla—Wagner theory are responsible for human
causal inference. As noted, Chapman and Robbins showed that the theory results
in strengths of association between response and outcome that are exactly equal
to the difference P(O|R) — P(O|-R). The theory in no way explicitly estimates prob-
abilities P(O|R) and P(O|-R), let alone takes their differences. Nonetheless, it esti-
mates this quantity, supporting the point made in the previous chapter that sim-
ple associative learning judgments can mimic sophisticated statistical inference.
A dramatically opposite conclusion can also be drawn. Subjects in these
experiments were not in conditioning experiments; that is, they were not in sit-
uations in which experimental contingencies reinforced their responses. Rather,
they were asked to make judgments of causal relatedness between response and
outcome. The fact that their causal inferences were like conditioning suggests
that causal inference, and not simple associative learning processes, underlies

108
 The Hippocampus and Conditioning

conditioning. That is, what organisms are learning in an instrumental condition-

ing experiment might be a causal model of the environment, and they act con-
sciously according to it. As discussed in Chapter 10, this view is the appropriate
interpretation of the human situation, and it may be the appropriate interpreta-
tion of the conditioning behavior of higher nonhuman organisms as well.
Holyoak, Koh, and Nisbett (1989) showed that many conditioning phenomena
in classical and instrumental conditioning can be explained by assuming that
organisms learn causal rules to predict the structure of their environment.
A wide range of conditioning phenomena can be explained by assuming
either simple associative learning or conscious cognitive judgment. To reiterate
a theme of this book, this is not an either—or situation. Some instances of con-
ditioning in some organisms may be due to unconscious, associative processes,
and other instances of conditioning in other organisms may be due to develop-
ment of causal models. There may be subtle differences between conditioning
behavior produced by simple associative learning versus conscious inference,
but by and large they look similar behaviorally because both reflect learning
adaptations of the organism to the structure of its environment. We will return
to this issue in greater detail when we discuss human causal inference in
Chapter 10.

Conditioning phenomena can be explained by assuming either

acquisition of simple associations or development of causal
models.

The Hippocampus and Conditioning

The hippocampal formation is a relatively small structure surrounded by the
temporal cortex. The hippocampus has been strongly implicated in learning and
memory in many organisms. In Chapters 7 and 8 we will discuss its important
role in human memory and how damage to it can result in profound memory
deficits. Here we will simply discuss the research that has been done involving
rats. Figure 3.15 compares the hippocampal formation of a rat, a monkey, and a
human. Note their differences in size and anatomy. It is not obvious that it
serves the same function in all species, but it is the hope of the field that it serves
similar functions. If this is true, then research done on the rat will shed light on
the nature of human memory and its deficits.
Figure 3.16 presents an outline of the relevant components of the rat hip-
pocampus and a schematic of their relevant connections. There are connections
from the cortex itself to what is called the parahippocampal region (which is
adjacent to the hippocampus) and connections from this to the hippocampus
itself. Many experiments have been performed studying the impact of lesions
(removal) of the hippocampus.on learning in rats. These lesion studies initially
involved removal of both the parahippocampal areas and the hippocampus.

109
 CHAPTER 3. Instrumental Conditioning

5mm
(a) rat >

5mm

(b) monkey (c) human

FIGURE 3.15 Comparative hippocampal gross anatomy: rat (a), monkey (b), and
human (c). (From Rosene & Van Hoesen, 1987.)

More recent research, using more refined techniques, has tried to separate out
the contribution of the parahippocampal areas to learning from the contribution
of the hippocampus (and we will describe these studies momentarily). Rats with
lesions to the parahippocampal area and the hippocampus perform poorly in a
wide range of instrumental and classical conditioning paradigms. They show a
particular deficit in tasks involving a substantial spatial component, such as
maze learning. An example that illustrates the deficit involves the Morris water-
escape task (Morris, 1981). Rats are placed in a circular pool of water and must
swim to an escape platform. If they climb onto the escape platform, the experi-
menter removes them from the pool; otherwise they are left to swim around.
The water is murky, and so the rats are unable to see anything below the sur-
face. In some conditions, the escape platform is above the water’s surface and
the rats can see it; in other conditions, it is just below the surface and they can-
not see it. In the original experiment Morris contrasted these four conditions:

110
 The Hippocampus and Conditioning

Cortical areas

Parahippocampal
region

Sle
Hippocampus

(a) (b)

FIGURE 3.16 (A) Simple schematic diagram of cortical-hippocampal connection.

(B) Outline of a horizontal rat brain section illustrating the locations and flow of
information between components of the hippocampus, parahippocampal region, and
adjacent cortical areas. DG, dentate gyrus; EC, enthorhinal cortex; FF, fimbria-fornix;
Hipp, hippocampus proper; OF, orbitofrontal cortex; Pir, piriform cortex; PR, perirhi-
nal cortex; Sub, subiculum. Source: From H. Eichenbaum, Declarative Memory: Insights
from Cognitive Neurobiology, Vol. 48, p. 559. Reprinted with permission.

1. Cue + place. The escape platform is always visible and always in the same
location.
2. Place. The platform is submerged but always in the same location.
3. Cue only. The escape platform is always visible but in different locations
on different trials.
4. Place random. The platform is submerged and in different locations.
Rats learned to swim quickly to the escape platform in all conditions but the
last. Figure 3.17 shows the tracks taken by a rat in each group on the last four
trials. Only the rats in the last group wandered much in the pool. This task is sig-
nificant because it shows that rats are excellent in using a spatial representation
to navigate through their environment. As Figure 3.17 illustrates, although rats
in the place condition started from a different part of the pool on each trial, they
knew where the submerged platform was and swam to it.
This experimental paradigm has become important for understanding the
role of the hippocampus. Rats with hippocampal lesions perform poorly in the
place condition—no better than normal rats perform in the place-random con-
dition. In contrast, normal and lesioned rats behave similarly when the platform
is visible (Morris, Garrud, Rawlins, & O’Keefe, 1982). Results such as this have
been used to argue that the hippocampus is significant in spatial learning. Some
other kinds of learning are not impaired by hippocampal lesions. For instance,
lesioned rats can still learn taste aversions and how to make simple visual dis-
criminations.

111
 CHAPTER 3 Instrumental Conditioning

Trial
17 18 19 20

Cue + place

Place

Cue only

Place random

BOWL BO
BDO
VOL
FIGURE 3.17 A vertical view of the tracks taken by rats in each group. Source: From
R. G. M. Morris. Learning and Motivation, Volume 12. Copyright © 1981 by Academic
Press. Reprinted by permission.

SeoEEE a%

Rats with hippocampal lesions perform poorly in many tasks

that require spatial learning.

The Nature of Hippocampal Learning

The field has been struggling to characterize the kinds of learning impaired by
hippocampal damage. O’Keefe and Nadel (1978) proposed that the hippocam-
pus, at least in the rat, is especially designed for learning spatial information. In
effect, it encodes Tolman’s spatial map (see Chapter 1). They reported that many
neurons in the hippocampus fire only when the animal is in a certain location
in space.
Olton, Becker, and Handelmann (1979) argued for a different interpreta-
tion of hippocampal deficits. Noting that many deficits occur in nonspatial tasks
and that some spatial tasks fail to show a deficit, they argued that the deficit is
amore general inability to hold information in working memory (a concept dis-
cussed at length in Chapter 5) over short periods of time. An example of the dis-
tinction to which they refer can be illustrated with respect to the radial maze

112
 The Hippocampus and Conditioning

(Figure 3.10). Olton et al. reported a study that used a 17-arm version of this
maze in which 8 of the arms were baited with food and the other 9 were not.
With experience with this maze, normal rats learned two things:
1. Never to enter the 9 arms that were never baited with food.
2. To efficiently explore the baited arms to avoid repeat visits, as discussed
with respect to Figure 3.10.
Rats with hippocampal lesions learned 1 but not 2. Both sorts of information are
spatial, but lesioned rats can learn one and not the other. In Olton’s terms what
they cannot do is rapidly update their working memory to avoid repeated visits
(2). Given enough experience, however, they can learn permanent properties of
their spatial environment (1).
Sutherland and Rudy (1991), taking a more traditional conditioning per-
spective, argued that the deficit is in the ability to form configural associations.
(See the discussion in Chapter 2 of the distinction between associations to stim-
ulus configurations versus stimulus elements.) They argued that to solve the
Morris water-escape task when the platform was submerged, the animal had to
respond to a configuration of spatial cues, whereas when the platform was vis-
ible the animal could simply respond to the visible platform. Eichenbaum,
Stewart, and Morris (1990) ran a variation of the submerged condition in which
rats always started from the same location. In this case, hippocampal lesioned
rats learned the task. In this condition, they did not have to respond to the con-
figural cues but could just swim in the same direction.
Sutherland and Rudy (1991) performed the following experiment, which
showed that rats with hippocampal lesions had difficulty learning a nonspatial
task that involved forming configural associations. Animals were rewarded with
food for pressing a bar when a light alone or a tone alone appeared. However,
they were not reinforced for responding when the light and tone were present-
ed simultaneously. As discussed in Chapter 2, normal animals can perform this
task, which requires learning associations to stimulus configurations of light +
no tone and tone + no light. Rats with hippocampal lesions are unable to learn
these associations, although they can learn to respond to the simple single stim-
uli. Thus, this is a nonspatial task in which lesioned rats show a deficit.
In recent years, the field has been developing elaborations of this config-
ural cue proposal. The basic idea is that an organism without a hippocampus can
only respond to single stimulus dimensions, but with a hippocampus, the
organism can respond to stimulus combinations. It has been shown that hip-
pocampal cells will fire selectively to various combinations of cues (for instance,
odors in rats—Otto & Eichenbaum, 1992) just as O’Keefe and Nadel found cells
that responded to combinations of various spatial cues.
Both Eichenbaum and Bunsey (1995) and Gluck and Myers (1995) have
made a similar distinction between two ways in which the hippocampus can
join different elements into a whole. In one case, the elements are fused into a
single whole in which the identity of the elements are lost. Eichenbaum and
Bunsey (1995) suggest the analogy of combining two words like hell and o into

113
 CHAPTER 3 Instrumental Conditioning

the word “hello.” The other way is to join the elements into an association in
which the individual element identity is preserved. The analogy suggested by
Eichenbaum and Bunsey is joining the same words in English, army and table,
into a paired-associate army-table. There is evidence that the parahippocampal
region performs stimulus fusion, while the hippocampal region performs the
element combination (see Figure 3.16).
Gluck and Myers use this distinction to explain latent inhibition and the
effect of hippocampal lesions on latent inhibition. Recall from Chapter 2 that
latent inhibition refers tg the phenomenon that if a CS is presented a number
of times before a US, it becomes harder to condition to a US. According to Gluck
and Myers, this is because the CS becomes fused with the context and is hard
to separate from the context. Lesions that include both the parahippocampal
region and the hippocampus abolish latent inhibition and make it easier for the
lesioned animals to condition in a latent inhibition paradigm. However, latent
inhibition is maintained if just the hippocampus is affected. According to Gluck
and Myers, this is because the parahippocampal region performs the fusion that
is responsible for latent inhibition. Recall that latent inhibition was one of the
problems with the Rescorla—Wagner theory. Gluck (1997) has argued that the
Rescorla-Wagner learning rule describes cortical learning but that different
learning rules are required to characterize hippocampal learning.
PSS SLSMON Rate ay Naat oe reopen EERE ESR UNIS REE EOE EE SNE ES NT CEI SS RPI RET

The apparent role of the hippocampus is to bind stimulus ele-

ments into combinations.
REEL IR LEST EB LO LOE TO RIL LL LIE TELE ETS GENESEE
LLY BRET STOEESTELOIE EE SLE NISELEIEE DLN IEDRELEN ARE NTE OS CEN Rt

Long-Term Potentiation (LTP)

Another reason for the interest in the hippocampus is that it is one region of the
brain where a particular type of neural learning has been displayed. When brief,
high-frequency electrical stimulation is administered to some neural areas of
the hippocampus, there is a long-term increase in the magnitude of the
response of the cells to further stimulation (e.g., Bliss & Lomo, 1973). This
change, called long-term potentiation (LTP), occurs immediately and lasts for
weeks. LIP involves increasing the synaptic connections among neurons. For
LTP to take place, the presynaptic and postsynaptic neurons must be simulta-
neously active. Because it is a permanent change and depends on joint activa-
tion of two neurons, it is thought to be involved in at least some kinds of asso-
ciative learning. Although LIP in the hippocampus has been studied most, it
occurs in many other regions of the brain as well.
A great deal of research has been done on the physical basis for LTP in the
hippocampus (for a review, see Bliss & Lynch, 1988, or Swanson, Teyler, &
Thompson, 1982). The LIP procedure results in structural changes in the den-
drites onto which axons synapse. The dendrites grow new spines at points
where the axons synapse, and existing receptors on the dendrites become
rounder. The change in the shape of the receptors appears temporary, but the

114
 The Hippocampus and Conditioning

increase in the number of spines is more long lasting. In addition to these post-
synaptic changes there are presynaptic changes involving an increase in the
release of neurotransmitters. Recall that the neural basis of learning in Aplysia
also involved an increase in the presynaptic release of neurotransmitters.
Considerable work has been done on the biochemistry of these changes
in the spines. Certain receptors in the postsynaptic membrane on the dendrite
(NMDA receptors) are normally blocked and become unblocked only if the
postsynaptic cell has fired. If the presynaptic cell fires and sends a neurotrans-
mitter to the postsynaptic membrane at the same time that the postsynaptic cell
is firing, then these unblocked receptors can receive the neurotransmitter. It is
thought that the unblocking of these NMDA receptors is the critical step in the
production of LTP. This unblocking in turn enables calcium to enter into the
postsynaptic neuron, resulting in an increase in both the receptors in the post-
synaptic neuron and the presynaptic release of the neurotransmitter. Kandel
and Hawkins (1992) speculated that the postsynaptic influx of calcium may
cause chemical messengers to be transmitted to the presynaptic axon, resulting
in the increased release of neurotransmitter.
ERR reeennRRessaRe sscayentennesvons

Simultaneous activity of presynaptic and postsynaptic cells in

the hippocampus can produce long-term facilitation of the
synaptic connection.
LEAR OEE EES AEN

Long-Term Potentiation and Hippocampal Learning

Much of the interest in LTP has arisen because LTP has been well documented
in the hippocampus and hippocampal damage is known to produce learning
deficits in a wide range of tasks. Thus, it has been conjectured that, when intact
animals learn these tasks, LTP is the neural process that underlies their learn-
ing. An effort has been made to bolster this connection by showing that phar-
macological interventions that interfere with LTP produce learning deficits sim-
ilar to those produced by hippocampal lesions.
Morris, Anderson, Lynch, and Baudry (1986) examined the effects of
blocking LIP by injecting a drug that prevents activation of NMDA receptors
involved in LIP. They looked at the performance of injected rats in the Morris
water tasks and found significant impairment, similar to that of lesioned rats.
The same injected rats were not impaired in tasks such as visual discrimination,
which are also not impaired by hippocampal lesions. Similar drug-induced
learning deficits that mimic those of rats with hippocampal lesions have been
reported by Staubli, Thibault, DiLorenzo, and Lynch (1989) and Robinson,
Crooks, Stinkman, and Gallagher (1989).
Some doubt has been expressed about whether LIP is really involved in
the kind of learning observed in tasks such as the Morris water-escape task.
Keith and Rudy (1990) noted that drug-injected rats, though impaired, showed
more learning than hippocampal-lesioned rats. Thus, while LTP may play some

115
 CHAPTER 3 Instrumental Conditioning

role in learning, it does not seem to be all that there is to hippocampal involve-
ment. More recent research (Bannerman, Good, Butcher, Ramsay, & Morris,
1995; Saucier & Cain, 1995) has found totally normal learning of the water maze
in drug-injected rats that have had some general training in this kind of task,
although not the specific water maze. Thus, it seems that the LIP component
may only block general learning of how to do the task and not the actual spa-
tial properties of the water maze. The spatial structure of the maze appears to be
learned by some other hippocampal process not involving LTP.
Sana
sr een meee sen

LTP is only part of the neural changes that underlie learning

in hippocampal-dependent tasks.

Final Reflections on Conditioning

In both instrumental and classical conditioning experiments, animals and
humans are capable of learning about their environments and responding adap-
tively. In classical conditioning they learn that one stimulus predicts another,
and they respond in anticipation of that fact. In instrumental conditioning they
learn that a stimulus signals that a certain class of responses will lead to some
outcome, and they respond according to whether or not that outcome is rein-
forcing. This research fits the adaptive function of learning identified in the first
chapter.
In keeping with the language of the field, this chapter has referred to
organisms forming associations among stimuli and responses. However, the
meaning of the term association does not capture all that is going on. The organ-
isms are not just connecting these stimuli and responses; rather, they are learn-
ing that certain elements predict other elements. In the case of instrumental
conditioning, they are learning about the causal structure of their environ-
ment—for instance, that a bar press causes food to be delivered. This learning
need not involve an explicit causal model. This chapter showed how the
Rescorla—Wagner theory is capable of implicitly encoding this causal structure in
simple associations. We will return to the issue of causal inference in Chapter 10,
where we will learn about other mechanisms for inferring causal structure in
humans.
In other paradigms, such as maze learning, organisms are learning some-
thing more specific than just what predicts what. They are learning about the
spatial layout of their environment and what objects are located where. This
cognitive map can be used flexibly to achieve goals. The nature of spatial mem-
ory in the human case is investigated further in Chapter 6.
A general characterization of conditioning is that it involves learning use-
ful information that allows the organism to respond adaptively to the reinforce-
ment contingencies of the experiment. The next chapter focuses on the role of
reinforcement in conditioning.

116
 Further Reading

In a conditioning experiment, organisms are learning things

about their environment and using this information to achieve
their needs.
a een

Further Readings
The textbooks and journals cited at the end of Chapter 2 are also excellent
sources for research on instrumental conditioning. In addition, many research
articles on instrumental conditioning are found in Learning and Motivation and
Journal of the Experimental Analysis of Behavior. Balsam (1988) reviews the
research relevant to stimulus generalizations and discriminations. Staddon and
Ettinger’s (1989) text on learning emphasizes its adaptive function. Gluck and
Meyers (1997) and Eichenbaum (1997) are two reviews of research relevant to
hippocampal function, and Landfield and Deadwyler (1988) edited a series of
articles on LTP. The November 1996 edition of the journal Hippocampus was
devoted to theories of the role of the hippocampus in learning.

slslys
 Reinforcement and Learning

Some Basic Concepts and Principles

The idea that organisms seek what is good for them and avoid what is bad for
them is as old as antiquity (and philosophical arguments about what is“ good”
and“bad” are equally ancient). Clearly, the reinforcement contingencies associ-
ated with a behavior have a lot to do with whether the organism actually per-
forms the behavior. One long-standing question is, What is the relationship of
learning to reinforcement? Thorndike (see Chapter 1) proposed a particularly
intimate relationship in his law of effect: learning would only occur if there was
reinforcement. This idea was maintained by many of the behaviorists and
proved a dividing issue between Hull and Tolman. Over time it has become
apparent that too much learning takes place without any reinforcement for the
law of effect to be viable. However, the question about the relationship of rein-
forcement to learning still stands. The answer was roughly outlined in Chapter
1—learning provides the knowledge, and reinforcers provide the goals to cause
the organism to act on that knowledge. This chapter is concerned with how
reinforcers provide those goals.
The basic thesis of this chapter is that organisms tend to behave rationally.
Using the contingencies they have learned in the environment, they select the
behavior that creates the best state of affairs for them. Suppose that four respons-
es are available to an organism: R1, which increases the amount of food available;
R2, which increases the rate at which the organism is shocked; R3, which
decreases the rate at which it gets food; and R4, which decreases the rate at
which it is shocked. The organism would not tend to produce R2, or R3, because
nothing good comes of them; it would alternate between R1 and R4 as a func-
tion of how important getting food is relative to avoiding shock. This is rational
behavior. This chapter further defines rational behavior and presents evidence
relevant to assessing how rational organisms are. At the outset, this chapter
states a disclaimer repeated elsewhere in the book: Behavior that appears ratio-
nal or optimal need not imply conscious deliberation on the organism’s part.
Simple associative mechanisms often can produce highly adaptive behavior.

118
 Some Basic Concepts and Principles

Although organisms tend to do the right thing, this chapter reviews situ-
ations in which they produce behavior that is far from optimal. This situation
can be viewed as a glass half full or half empty. Historically, psychology has
taken a half-empty perspective and emphasized deviations from optimality.
More recently, psychologists have become impressed with how well even sim-
ple organisms do at behaving in near optimal ways. Often cases of nonoptimal-
ity can be understood as generally adaptive behavioral tendencies that go astray
in situations for which they did not evolve. For instance, human affection for
sweet food reflects a tendency that selected food of high nutritional value at one
time in our evolutionary history. However, in modern society with its capacity to
create almost arbitrary food products, this tendency often selects the least nutri-
tious of the food alternatives.
NSN TET ENE SESE ENS EOE NTE RELAIS SOR BE

Learning provides a knowledge of the reinforcement contin-

gencies of actions, and organisms generally select the most
beneficial action given their knowledge.
RHR EE LORE tO EAN SEER SEINE NNN ONY EE US NAY LEE EAA

Rational Behavior
What is meant by rational behavior? Consider a situation that a rat might
encounter in a laboratory experiment. Suppose that three significant actions are
available to the rat: It can press a bar, play in an activity wheel, or do nothing
(or, at least, do neither of the first two activities). Suppose there are four possi-
ble consequences of its actions: it will receive food; it will be shocked; it will
receive exercise; or nothing will happen.
The experimenter has arranged contingencies between each activity and
each outcome, as shown in Table 4.1. If the rat presses the bar, there is a 67 per-
cent chance of getting food and a 33 percent chance of being shocked. If it
enters the activity wheel, there is a certainty of exercise. If it does nothing, there
is a 90 percent chance of nothing happening and a 10 percent chance of getting
food. The rat has learned these behavioral contingencies from its exploration of

TABLE 4.1 Probabilities of Outcomes Given Behaviors

Outcomes Behaviors

Food

Shock

Exercise

Nothing
 CHAPTER 4 Reinforcement and Learning

the experimental situation. The knowledge in Table 4.1 reflects the product of its
learning.
Simply knowing the behavioral contingencies in Table 4.1 does not tell us
what is optimal behavior for the rat; we also need to know the value it places on
various outcomes. Assume that the outcome’of nothing has a value of 0, food
has a large positive value of 10, shock has a large negative value of —25, and
exercise has a mild positive value of 1. Now it is possible to predict what the rat
will do if it is behaving rationally. Rational theory states that the rat should select
the behavior with the highest expected value. The expected value of an action
is calculated by multiplying the probability of each possible outcome by its value
and taking the sum of these products. This result reflects the average value that
can be expected from that action. In the case of the bar press, there are two pos-
sible outcomes—food and shock. Performing this calculation for these two
yields
Probability(food) x Value(food) + Probability(shock) x Value(shock)
= .67 x 10.0 + .33 x —25.0 = -1.55

In the case of entering the activity wheel, there is only one possible outcome. Its
value is calculated as
Probability(exercise) x Value(exercise) = 1 x 1.0 = 1.00
Finally, in the case of doing nothing, there are two possible outcomes:
Probability(nothing) x Value(nothing) + Probability(food) x Value(food)
= .90 x 0.0 + .10 x 10.0 = 1.00

Thus the exercise wheel and doing nothing are of equal value, and the rat would
be predicted to alternate between them. If the rat became satiated so that food
lost its value, the rat would be predicted to select the exercise wheel exclusive-
ly. If the rat became hungrier and food increased its value, the rat would select
nothing; if it became hungry enough (and food approached a value of 15 or
more), the rat would select to press the bar despite the shocks.
If the hunger of the rat were manipulated, the rat would probably shift
between the activity wheel, doing nothing, and pressing the bar as implied by
this rational analysis. This behavior would not mean that the animal was explic-
itly representing probabilities, values, and calculating expected values, which is
extremely implausible in the case of the rat. Rather, the rat would probably be
doing something much simpler that allowed it to behave as if it were engaged
in rational calculation. This chapter discusses several such simple mechanisms
for selecting appropriate behavior.

Rational behavior implies combining the probabilities of the

outcomes of actions with their values and selecting the action
with the highest expected value.
LR EES ERR SE Sart ae

120
 Some Basic Concepts and Principles

Effects of Reinforcement on Learning

Implicit in the analysis of Table 4.1 is that learning the contingencies or proba-
bilities in the table anes not depend on reinforcement. Reinforcement deter-
mines how the animal acts given knowledge of these probabilities. The claim
that Pers — = seer on reinforcement is quite remarkable. Certain
things are more worthwhile to an organism, and thus itis to its advantage to
jeam these things rather than other pineThe a aa it ac ooRtive advantage of

s. ee it seems einer mee a connection weioes not exist. iti,4

reviewed Tolman’s research on latent learning in the rat, but some of the best
research onn the cole ofpoegrcetent in
4n learning has been same on human sub-

Numerous experiments involved telling subjects that they will be reward-

ed more for learning some items than for learning others. Such experiments fre-
quently have subjects learn lists of words or other verbal stimuli. Subjects
respond by learning the more valuable items more rapidly. On the other hand,
if the manipulation is done between subjects so that some subjects are told that
all items are worth more than other subjects are told, the reward has no effect
(e.g., Harley, 1965). Thus, one line of research (when reward is manipulated
within subjects) seems to indicate that learning depends on reinforcement,
whereas the other line of research (when reward is manipulated between sub-
jects) seems to indicate that it does not.
The explanation of these apparently contradictory results comes from
studies of how subjects allocate their time as a function of reinforcement. A typ-
ical experiment is that of G. R. Loftus (1972). He presented subjects with pairs
of naturalistic pictures to study for 3 sec. The left member of a pair was assigned
1, 5, or 9 points, and the right member of a pair was independently assigned 1,
5, or 9 points. Subjects were later asked to identify which pictures they had stud-
ied when these were mixed in with pictures they had not studied. Subjects were
paid bonus points in proportion to the value of the pictures they could recog-
nize. Figure 4.1a shows the probability of recognizing the target picture as a
function of its value and the value of the picture with which it was paired.
Subjects showed better recognition memory for a picture the more points.
assigned to it and the fewer points assigned to the other picture of the pair. This
experiment is like the studies mentioned earlier that show the effects of reward
when the reward varies within a set of items.
Loftus also monitored how often subjects fixated on each picture during
the 3 sec of exposure. These data are presented in Figure 4.1b. Subjects fixated
on the picture more if it was worth more and if the other picture was worth less.
This result raises the question of whether memory performance is a function of
the value of the picture or the number of fixations. Loftus did the relevant analy-

121
 CHAPTER 4 Reinforcement and Learning

Value of other picture

—e 1 point 9.0
| —= 5 points
= | -- 9 points 8.0 Value of other picture
[e)

= 70 : 1 point
op 2 7.0
rs) 2
e |
5 = 6.0 5 points
= 5
a 5 5.0
6 60 E
As Zz
4.0 9 points

3.0

2.0
1 3) 2) 1 5 9
Value of picture (points) Value of picture (points)
(a) (b)

FIGURE 4.1 (a) Probability of recognition and (b) mean number of fixations for pic-
tures worth 1,5, and 9 points. Separate curves are plotted for each of the three values
of the paired-with picture. (From G. R. Loftus, 1972.)

sis in Figure 4.2, where memory performance is plotted as a function of the

number of fixations for pictures of different values. As can be seen, memory per-
formance was a function of how often subjects looked at a picture and not how
much it was worth. As Figure 4.1 illustrates, subjects tend to look more at more
valuable pictures and so show better memory for these pictures. However, as

0.8

O57

% i=
Ss
5, 0.6
5
2
x)
£05
=
FIGURE 4.2 Probability of recog- 4
nition as a function of number of & “o>1 point
fixations. A separate curve is plot- — “> 5 points
ted for pictures worth 1, 5, and 9 ~~9 points
points. Source: G. R. Loftus. “Eye
fixations and recognition memory
for pictures,” Cognitive Psychology,
Vol. 3 (1972). Reprinted by permis- 0-1 2-3 4-5 6-7 8+
sion of the author. Number of fixations

122
 Some Basic Concepts and Principles

Figure 4.2 confirms, when Loftus controlled for the number of fixations received
bya picture, value had no effect. These results sous the general understanding

eines peaieehsine so wr the apparent poriiadietionain the sailtee ieneeeaD

When different items in one list had more value, subjects tended to allocate more
time to them and remember them better. When all the items in a list had the
same value, subjects could not differentially allocate time to them as a function
of reward. In this case, the value assigned to the items had no effect on learning.

; Organisms pay more attention to material associated with

greater reward, but, controlling for attention, there is no effect
of amount of reward on learning.

Reward and Punishment

isms oredues the response in1 question, the contingency can be one such that the
stimulus is given or not given. Table 4.2 illustrates the four logical possibilities
obtained by crossing desirable or aversive stimuli with different contingencies
between the stimulus and the response. In the first case, reward is made contin-
gent on a behavior. For instance, a child may be ee a sum of money for mow-
einforc ement and should

sayoaergon the eeu This is the favorite response of many parents to misbe-
es “You'reeed. situation is referred to as omission training and

spanking, and again shot a fior.The final

eersine is for dis-
appearance of an aversiveDee desttobeparce on the behavior. This situa-

TABLE 4.2 Type of Stimulus and Contingency on Response

Stimulus Given If Behavior Stimulus Removed If Behavior

Is Performed Is Performed

Desirable Positive Reinforcement Omission

stimulus (reward training) Training

Aversive Punishment Negative Reinforcement

stimulus (escape or avoidance)

123
 CHAPTER 4 Reinforcement and Learning

The recatee REM is that these contingencies control the behav-

ior at hand. For a long time, learning theorists were reluctant to accept such a pro-
posal because it seemed to imply that something in the future (the reinforcement)
was causing the response. Causes only work forward in time, and so future rein-
forcement cannot cause present behavior. As Chapter 1 reviewed, Tolman was crit-
icized on these grounds for his proposal that animals performed certain behaviors
because they expected that these would lead to certain desired results. Chapter 1
also reviewed the major contribution of simulation models such as Newell and
simon‘ s GPS, which showed how knowledge of CONS learned from expe-

1e future. Not all organisms that

andgaplay fet thentdl
learning behave like GPS, but GPS demonstrates that there are mechanical ways
in which knowledge of contingencies can control behavior. Many other mecha-
nisms have since been proposed, some of which may be more plausible for lower
organisms. In many cases, knowledge of contingencies is not explicit or conscious;
rather, it is knowledge implicit in the processing of the organism.
ARRAN ASTIN mete RTE ENE TENELOL TLE CN A ERE TNT RE RTE

rteiiaels ee in jini wees eer ets

in such a way as to obtain desirable stimuli and avoid aver-
sive ones.

A chimpanzee trained to exchange

tokens for food.

124
 Aversive Control of Behavior

Aversive Control of Behavior

Tables 4.1 and 4.2 imply that aversive stimuli, such as shock, are effective“in con-
trolling behavior and that their effects are symmetrical with the effects of desirable
stimuli, such as food. As Chapter 1 noted with respect to Thorndike’s attitudes
about punishment, there has been a long tradition in popular psychology of believ-
ing that punishment is not effective. In brief, these beliefs are wrong. This section
reviews the evidence that aversive stimuli are quite gHtcotive, and it discusses bow

Punishment
Sometimes punishment can be so effective that a single learning experience
eliminates a behavior. A child who touches a hot stove is unlikely to do so again.
In one experimental paradigm (Jarvik & Essman, 1960), a rat is placed on a plat-
form above a grid floor. When it steps off the platform it receives a painful shock.
After a single experience, the rat will not step down again. It learns to com-
pletely suppress a natural response in a sing

(just as
as . delay of BDA aneeat Tes,aetane effooo THonePeeence
illustrating the effects of delay, Camp, Raymond, and Church (1967) contrasted
several groups of rats. Each group was first trained to press a lever in response
to a clicking sound; they were reinforced with food. After this training, half the
presses resulted in a shock at varying delays. For one group, the shock came
immediately after pressing the lever, and for other groups it came 7.5 or 30 sec
after pressing the lever. A control group of rats received as many shocks, but the
shocks were unrelated to when they pressed the lever. Figure 4.3 plots the per-
centage of clicks to which the rats pressed the bar. Note that the group with a
30-sec delay showed only a little more suppression of bar pressing than the con-
trol group for which there was no contingency. (That is, shocks were delivered
on a schedule completely unrelated to when the bar was pressed.) Much more
suppression of bar pressing occurred in the rats with immediate shock.
It is easy to extrapolate from such a result with rats and leap to the con-
clusion that punishment should be immediate with humans, particularly with
children. However, because children can be told the contingency that exists
between the behavior and the punishment, the immediacy of punishment is
probably not so critical.
The erity of the puni shmer nt ca n also have
a strong influence on response
sup ‘igure 4.4 displays data from Church (1969) on the amount of sup-
pression (see the discussion of response suppression in CER with respect to
Figure 2.9) in lever pressing for different levels of severity of shock (including no
shock). There was only a little suppression with .15 mA (milliampere) of shock, a

125
 CHAPTER 4 Reinforcement and Learning

80

+
o(@)
jo)

“1
‘\

Mean
of NOS) T
percent
responses
FIGURE 4.3 Mean percentage of
responses as a function of sessions for
groups with 0.0-, 7.5-, and 30.0-sec.
delay of punishment and for a noncon-
tingent-shock control group. (From aN ES 30 Control

Camp et al., 1967.) Delay of punishment, sec

good deal more with .50 mA, and still more with 2.0 mA. Extrapolating this result
to human beings raises some of the ethical issues in using punishment. Certain
degrees of oS Dine are simie too extreme to be used.

introduced and then econ in severity, the organism became less sensitive to
punishment and the severest level of punishment was then not as effective as it
would have been if introduced immediately (an example of habituation described
in Chapter 2). Azrin, Holz, and Hake (1963) found that the effectiveness of the
punishment was reduced if only some responses were followed by punishment.

0.8

SoS py
oO)

%
suppression,
of
Degree
o iN)
FIGURE 4.4 Median suppression
ratio as a function of intensity of non-
contingent shock. Lower values reflect
0.0
increased suppression of responses. 0.15 0.5
(From Church, 1969.) Strength of shock, mA

126
 Aversive Control of Behavior

fect. noncontine
inget punishmenton later
Poy ee oie

ats were given ten 30-min training sessions in which

they eeea to press a lever for food. In sessions 11 through 15, an experimen-
tal group received random 105-V shocks, independent of responses, while a
control group continued to receive just reinforcement during those sessions.
Both groups were retrained without shock for sessions 16 through 20. Finally,
during sessions 21 through 25, both groups received 145-V shocks contingent
on pressing. Figure 4.5 shows the results in terms of rates of responding relative
to the rates during the initial 10 sessions. During the initial noncontingent shock
the experimental rats pressed somewhat less, showing a CER (see Chapters 1
and 2). They recovered during retraining and continued a high level of respond-
ing during the final phase, when the shock was contingent. In contrast, the con-
trol rats showed a nearly complete suppression in the final phase, when shock
was made contingent on ae response. The implications
i ofthis epenment are

vonageso that they a not have to steal, punishments for stealing (e.g., impris-
onment) will be more effective in deterring the behavior.
An experiment by Azrin and Holz (1966) illustrates the importance of
offering an alternative behavior if punishment is to be effective. Pigeons were
first trained to peck at a key to receive food. Then they were shocked for peck-
ing at the key. There were two conditions: in one there was another key that they

Prior Retraining | Punishment

exposure
120

100

leo))

oO[e)
-e No shock
-o- Shock (110 V)
B (e)

baseline
Response
of
rate
percent
as
NO(s)
FIGURE4.5 Median response rate to
a punishment of 145 V as a function of
prior exposure to noncontingent 10 15 me’ 25
shock of 105 V. (From Church, 1969.) Session

127
 CHAPTER 4 Reinforcement and Learning

300

--c--r--r-—ft

pe)fo)fo)
| \.
Cc

§ Alternative
D response
2 available |
2 \
100 \ No alternative
. \ response
available

;
(a | rere
0 20v 40v 60v 80v
Punishment intensity, V

FIGURE 4.6 The rate of punished responses as a function of the punishment inten-
sity. (From Azrin & Holtz, unpublished data.) Source: Figure 4.6 from N. H. Azrin and
W. C. Holz. Punishment, in Operant Behavior: Areas of Research and Application, Honig
ed., Copyright © 1966, p. 405. Reprinted by permission of Prentice-Hall, Englewood
Cliffs, New Jersey.

could peck at, and in the other there was only the single key. Figure 4.6 shows
response to the shocked key as a function of the intensity of the shock. Up to
about 40 V, the shock intensity was not severe enough to affect responding to
the key. However, at 50V it was intense enough to produce a complete cessa-
tion of response and a shift to the alternative key in the condition that had an
alternative key. In the condition without an alternative key, the pigeons persist-
ed in pecking when the shock was much more intense.
This review of punishment should not be read as encouragement to use it
as a major mechanism for controlling the behavior of children or others.
Punishment can have a number of serious negative side effects. Azrin and Holz
(1966) argued that punishment can lead to a general suppression of all behav-
iors, good and bad; it can lead to anger in the punished person, and it can moti-
vate deception to avoid punishment. Children sometimes lie to their parents to
get even for past punishment and to avoid future punishment. Also, there is evi-
dence that punishment leads to more aggressive behavior in the punished
(Ulrich & Azrin, 1962). Finally, children may inappropriately use punishment in
interactions with their peers, modeling their parents’ behavior toward them
(Eron, Walder, Toigo, & Lefkowitz, 1963).

Punishment is effective to the degree that it is administered

immediately, severely, and consistently and to the degree that
the organism is offered alternative behaviors.
ARN UNRATE RR SENDER

128
 Aversive Control of Behavior

Negative Reinforcement
mail L achieves | timuli, behav-
ain ecal muli. Solomon and Wynne
(1953) site arae ina compartment with a steel grid floor. At the beginning
of a trial, the light went off; 10 sec later a severe shock was sent through the grid,
causing the dog to run about trying to escape. The dog could leap over a barri-
er to escape the shock, and eventually it jumped over this barrier into another
compartment that was shock free. Within a few trials, the dog learned to jump
the barrier on the signal and thus to avoid shock completely.
A curious feature of such avoidance behavior is that it can be much more
difficult to extinguish than behavior maintained by positive reinforcement. If food
is no longer given conditional on some behavior, such as jumping over a barrier
in response to a stimulus, a dog soon stops the behavior. On the other hand, if the
shock is removed, the dog will continue aria eat without any ae of extinction

shorten aatari ne barrier that neal the rats from escaping. After
about 5 min of forced exposure to the formerly aversive situation, the rats seemed
to learn that there no longer was a eas atedbetween the tone ne the shock.

For instance, an 11-year-old child who was Pre of loud

noises was stents to break a set of balloons (Yule, Sacks, & Hersov, 1974).
After a few sessions of balloon breaking, the child lost the phobia and came to
enjoy breaking balloons as most children do. This therapy worked because the
child had been made to realize that nothing terrible happens just because there
is a loud noise.
The classic theory of avoidance learning is the two-process theory pro-
posed by Mowrer he at!and Semen by Miller (1951). ene to the two-
h !

This division of avoidance learning into a classical conditioning compo-

nent and an instrumental conditioning component remains generally accepted.
What is problematical about the dual-process theory is its conception of both
classical conditioning and instrumental conditioning. More modern views of
classical conditioning and instrumental conditioning appear more appropriate.

129
 CHAPTER 4 Reinforcement and Learning

First, it does not appear that it is conditioned response of fear that is learned to
the CS. As we noted in Chapter 2 on classical conditioning, rather it is the US
that is typically conditioned to the CS. That is, the animal comes to expect that
the US will follow the CS. The avoidance response is made in anticipation of the
US. Although the CS will often evoke fear, the animal will make the avoidance
response even when the CS no longer evokes fear. For instance, Kamin, Brimer,
and Black (1963) showed that animals continued to avoid the response even
after the CS had lost its ability to evoke a conditioned emotional response (sup-
pression of bar pressing—see Chapters 1 and 2).
Second, elimination of the CS is apparently not necessary to learn the
avoidance response. Kamin (1956) showed that animals would learn an avoid-
ance response even when the CS remained after the avoidance response. Thus,
animals were learning even though they were not eliminating the CS. Another
demonstration that CS elimination is not necessary to learning is provided by
the Sidman shock-postponement procedure (Sidman, 1966) in which there is
no overt CS. This procedure involves presenting an aversive stimulus like shock
every so often without warning. However, the animal can avoid the stimuli if it
performs some response. For instance, by pressing a lever, an animal might be
able to postpone shock for 30 sec. If it presses the lever in that 30-sec period, it
gets another reprieve of 30 sec from the point of this new press. If the animal
presses the lever at least once every 30 sec, it permanently avoids shock. Dogs
master this task well, responding only a few times a minute and avoiding virtu-
ally all shock. The experimental context serves as the CS for this behavior, but
the response does not get rid of this CS, only the US.
In Chapter 3 we discussed the evidence that, in the case of positive rein-
forcement, the organism has learned an association between the CS, the
response, and positive reinforcement. The corresponding analysis in the case of
negative reinforcement (e.g., Seligman & Johnson, 1973) is that what animals
have learned is an association between the CS, the response, and avoidance of
the aversion stimulus. Just as organisms use what they learn to select behavior
in the case of positive reinforcement, they use knowledge of this association in
the case of negative reinforcement.

Stable patterns of behavior can be maintained if they avoid

‘

aversive consequences. ;

The Nature of Reinforcement

Drive-Reduction Theory
What makes a reinforcer reinforcing? An obvious idea from biology is that pos-
itive reinforcers are good for the organism and negative reinforcers are bad,
with” good” and“bad” defined in evolutionary terms of survival of the organism

130
 The Nature of Reinforcement

and maximization of the number of its offspring. The problem with this view is
that an erating RE not really know coe is good foritin such abstract

Ranees and thirst, this proposal is particularly intuitive. Almost all of us ‘have felt
hunger, found it aversive,and more So as de privation SOR AAS: 1

ree instance, Butler (1953) found that monkeys learn to perform a behavior just
for the opportunity to look around the laboratory for a few moments. Rats learn
behaviors for the opportunity to run in an exercise wheel. One could postulate
curiosity drives and exercise drives (perhaps with boredom as the aversive state)
and speculate about their potential biological value (for instance, the values of
learning about the environment and keeping fit), but this has struck many peo-
ple as creating a rather hollow theory. Any behavior could be explained by pos-
tulating a drive for it and proposing some fanciful biological function. Also, such
hypothetical drives do not fit well with experiences of deprivation with more
basic biological needs. Many people live a complete life without a strong desire
to exercise e similar to the desire they have to eat after a day without food.
ven more the t is that beha r can be reir

why ae rats ran a maze ie the Sorority to copulate with a female rat, even
though they were not allowed to ejaculate (Sheffield, Wulff, & Backer, 1951). The
male rats were being reinforced for a behavior that left them with increased
drive. Ina less extreme mode, humans find the company of attractive members

asec nes EEE RNIN MN te

Drive-reduction theory aapted pat Fu (iraeners satwiaicd

of reduction of various eee shige
SREB ANBAR AU SERENE ERSN RES AESEI MONEE LINER ICRC OO REE

1 However, there often seems to be a limit to the increase in aversiveness as the peri-
od of deprivation continues.
2 Some athletes and other people do report such desires, however.

131
 CHAPTER 4 Reinforcement and Learning

Premack’s Theory of Reinforcement

Such difficulties led to an alternative conception developed by Premack
1965). Premack’s seat which jes infnenced many subsequent ate is
that responses, not stimuli, are reir g
cal reinforcements, atic as ted, aretee g
ued behaviors TE gaiie,

hus pt ey Perioeeanrunning inan bt rete o a oe rat

because eating is more valued than seal ns relative yale of two activities

each activity
when the organism has
ponemore timeseeae than running i
in anTaN wheel. Another met
he ani y to get either rcement—for instance, pressing

result chosen more ac ahanesone that is preferred.

The basic predictions of Premack’s theory have been well supported. For
example, a thirsty rat can be shown to prefer drinking to running in an activity
wheel, and it will increase its rate of running in an activity wheel if that behav-
ior gets it access to water. On the other hand, a nonthirsty rat can be shown to
prefer running in an activity wheel to drinking, and it will increase its drinking
if that gets it access to an activity wheel (Premack, 1962). Premack (1959) found
similar results with children. Some children preferred eating candy to playing a
pinball machine. If access to candy was made contingent on playing the pinball
machine, their rate of playing the pinball machine went up. However, if playing
the pinball machine was made contingent on eating candy, their rate of eating
candy did not change or it went down. The converse relationships were
observed in those children who preferred inball machine to eating

ehavior. Thus, eens children ane ser tae candy to ai the pinball
machine if they ate candy would reduce their rate of eating candy.
Premack (1971) described an experiment by Weisman and Premack (1966)
that illustrates the relativity in the concepts of reinforcement and punishment.
They compared rats that were deprived of water and rats that were not deprived.
When offered simultaneous access to an activity wheel and a drinking tube,
deprived rats spent more time licking from the tube than running, whereas non-
deprived rats spent more time running. Figure 4.7 shows the amount of time
spent in the two activities in the two conditions when rats could choose to do
either. Premack used such free-choice information to establish the relative value
of the two activities. For the deprived rats drinking water was more valued,
whereas for the nondeprived rats running the activity wheel was more valued.
Then Weisman and Premack introduced a contingency such that if a rat
licked the tube 15 times it had to run for 5 sec and could run for no more than
5 sec. What should this contingency do to their licking? For the nondeprived

132
 The Nature of Reinforcement

400

g 300
2) + .
e | Licking
=
8
= 200-
i=
o
oe L
®

= 100 Running

FIGURE 4.7 Comparison of base

rates of running and licking for rats
deprived of water and those not 0
deprived of water. Deprived Not deprived

rats, the activity wheel was more valued, and this contingency should reinforce
licking. For the deprived rats, the wheel was less valued, and so having to run in
it should punish licking. Figure 4.8 compares the rates of licking before and after
introduction of the contingency. As predicted, the contingency reinforced lick-
ing for the nondeprived rats and increased their rate of licking. On the other
hand, it Ngaio the ie for the deprived rats and decreased their rate.
Sat aera SSE URINE EN seo ence tea OLR SNARE TREES OPER ULES ELE ETERS ELITES ETE SESE NENG BE

eee Leper ae eee ne A ennmay on

behavior B would reinforce B if A was more valued and pun-
ish B if A was less valued.
PER ea
I TENT
RAT ETN HE NNR TIS A ISSR IIL IEEE EIEN SE INNER NEI,

400

Ww[e)oO

Water-deprived rats

3 Oo ]

Time
licking,
spent
sec Nondeprived rats
100

FIGURE 4.8 Impact of a contin-

gency between licking and running 0
on rate of licking for rats deprived of No Running contingent
water and those not deprived. contingency on licking

133
 CHAPTER 4 Reinforcement and Learning

Neural Basis for Reinforcement

Insight into reinforcement (and the difficulties for both the drive-reduction the-
ory and Premack’s theory) can be esas from studies of the brain mecha-
nisms involvedi Mu

ypotha amus is phylogenetically a very

old part of the brain. If different parts of the hypothalamus are removed, animals
overeat, fail to eat or drink, or show loss of sexual behavior. Electrical stimula-
tion of different regions"ean produce eating and sexual behavior (for a review
see Stein, 1978).
Olds and Milner (1954) found that electrical stimulation of the hypothal-
amic area of the brains of rats could also serve as a reinforcer. Rats learned to
press bars or perform other activities in order to receive such stimulation. In a
few studies of human patients who had had such implants as part of treatment
for severe neurological problems, such as epilepsy, the patients reported a num-
ber of feelings associated with self-stimulation, including feelings of being
drunk and sexually aroused (e.g., Heath, 1963).
Stein (1978) argued that special neurotransmitters in these regions of the
brain are biochemically distinct from other neurotransmitters. There is evidence
that the effects of such drugs as opiates and cocaine take place in part on these
neural areas, affecting the rate of synaptic transmission. The administration
of drugs that attenuate the effects of opiates and cocaine also attenuates the
effects of brain stimulation (see Vaccarino, Schiff, & Glickman, 1989, for a
review).

| (they reduce no
drive) and Premack’s Denner teary (they volves no behaviors).
Se a SRS NEE La RON soa ANP HST IEE TACRT ROTTER

The ah aisseemsee oan with i a

ment, and animals find pleasurable both electrical and phar-
macological stimulation of the hypothalamus.
eS see create oe eres a

Equilibrium Theory and Bliss Points

yer ites
Premade sealswas a major conceptual advance, it sonta BUINDeE -

ECC itt

co (1989) foe that telatisa Shay: a electrical brain stimula-

tion is a reward and neither can really be conceived of as a behavior. It seems
probable that animals can be reinforced by many things, including drive-reduc-

134
 The Nature of Reinforcement

ing stimuli (like food), by behaviors (like running in an activity wheel), and by
things that are neither (like brain stimulation).

given the enone between two saccharin solutions, spent more time on the
sweeter of the two when drinking freely. Premack would predict that drinking
the less sweet solution would punish drinking the sweeter solution. For
instance, suppose rats had to lick the less sweet solution once for every time
they licked the sweeter solution (a 1-to-1 response ratio). Rats should decrease
their rate of licking the sweet solution because a less desirable solution fol-
lowed. This is probably what Allison and Timberlake would have found with a
1-to-1 ratio, but in their study the rat had to lick the sweeter solution 10 times
to get access to the less sweet solution. Since the 10-to-1 ratio was greater than
the natural distribution between the solutions, if the rats licked the sweeter
solution as much as they did in the free drinking situation, they would get less
of the less sweet solution than in the free situation. In this experiment, the rats
increased their rate of drinking the sweeter solution iin order to gainmore access

te experiment just described, the reinforcement schedule mueed the rats to

move from their bliss point for distributing their drinking over the two solutions.
They had to increase their drinking of the sweeter solution to above its base
level and to reduce their drinking of the less sweet solution to below its base
level. They did this to achieve a compromise that was as close to their ideal bliss
point as possible given the reinforcement schedule.
Figure 4.9 illustrates Allison’s (1989) general demonstration of the opera-
tion of bliss points. The figure represents the various amounts of activity that are
possible for behaviors A and B. It shows the animal’s bliss point for the optimal
combination of these two behaviors. The animal might want to spend 150
min/day in activity A (perhaps eating) and 50 min/day in activity B (perhaps
running in an exercise wheel). A schedule is introduced in which the animal
must spend 1 min in activity A for each minute in activity B. The straight diag-
onal line reflects this schedule. The animal finds the point on this schedule clos-
est to its bliss point—in this case the point at which it spends 100 min on each
activity. In the example in which activity A is eating and activity B is running,
food could be viewed as reinforcing running in the wheel because it increases
running. (Alternatively, running in the wheel might be seen as punishing eat-

135
 CHAPTER 4 Reinforcement and Learning

Bliss
point
150

<x
ro} .
= 100°
=

a
FIGURE 4.9 Behavior B as a function of 50
Behavior A under a condition that con-
strains the two behaviors to be equal. The
arrow points from the place on the curve i:
closest to the bliss point. (From Allison, 0% 50 100 150 200
1989.) Behavior B

ing, because it decreases eating.) Suppose that a schedule is created in which

the animal must spend 10 min in activity A for each minute in activity B. Then
the animal would increase activity A to get more B. In the example of food and
exercise wheel, running would reinforce eating.
Konarski (1979) and Konarski, Johnson, Crowell, and Whitman (1980)—
both reported in Timberlake (1980)—provide an interesting demonstration of
the application of equilibrium theory to the education of children. Most young
children can easily be made to do more math if an opportunity to color is made
contingent on math. According to Timberlake, this happens because children
normally have less opportunity than they want to color and more opportunity
than they want to do math. However, if deprived of an opportunity to do math,
children can be made to increase the amount of coloring over what they would
pounaly do if this activity results in access to
o the e/] RES to do math.

{ liss po in si Al api experi-

ments such as that just Hestribed are impressive demonstrations of the predic-
tive power of equilibrium theory, the theory is still somewhat incomplete
because it does not fully explain how the bliss points are determined.
Equilibrium theory proposes that features such as the deprivation state of an
organism and the quality of food in the feeder combine to determine the bliss
point or optimal distribution of responses such as eating and exercising. Once
that bliss point is known, the results of constraining these behaviors and mak-
ing one behavior contingent on the other can be predicted. The theory does not
provide an analysis of how this bliss point is set in the first place; ultimately, bio-
logical explanations for the setting of these bliss points are needed.
LNT TEL EE AR LLG R I LITT TANT

Organisms choose their behavior so as to move as close as pos-

sible to some pista! i acne ks KihasbeicyoMot babies

136
 Studies of Choice Behavior

Studies of Choice Behavior

According to the current conception of reinforcement, an organism balances
competing needs or goals in order to achieve the combination closest to its bliss
point. Research on reinforcement seeks to determine how organisms make
choices, given their experience with the constraints of their environment. Recent
research has focused on choice behavior in animals. Before discussing this
research, it is important to review some of the basic effects of different sched-
ules of reinforcement, since many of these recent studies offer animals choices
among schedules of reinforcement. It turns out that understanding behavior
even under a single schedule also requires conceiving of the organism as mak-
ing choices among alternative behaviors.

Schedules of Reinforcement
The publication of Ferster and Skinner's Schedules of Reinforcement in 1957
marked a sharp increase in interest in the relationship between the schedule
with which reinforcements are delivered and the resulting behavior. Four basic
schedules have been studied, akthough there are many exotic variations. In the
reinforcement
is given after every so many
‘s. For example, a schedule with a reinforce-

sehediile the organism receives a reinforcement for its next ieeoanse after 15 sec
have elapsed; the organism then waits another 15 sec before its next response
produces a Sibi 8 and so on. Finally, inav
i e

wait ann average of 30 sec before a response produces a reinforcement.

It is important to appreciate a subtlety in the interval schedules. In a FI 15
schedule, for instance, the delay between reinforcements is not 15 sec; it is
greater. Fifteen seconds must pass before a response from the organism pro-
duces the reward; the total time between rewards is 15 sec plus however long
the organism then waits to respond.
Each schedule of reinforcement produces its own characteristic behavior.
The behavior is typically measured in terms of cumulative response records, as
discussed in Chapter 1 (see Figure 1.9), which are graphs of how the total num-
ber of responses so fariincreases with the passage of time. nae 4.10 sNews typ-

137
 CHAPTER 4 Reinforcement and Learning

Fixed Variable Fixed Variable

ratio ratio interval interval

>
Cumulative
responses
———

ne

FIGURE 4.10 Stylized cumulative response records obtained under four common
schedules of reinforcement. The ticks on the functions denote delivery of reinforce-
ments.

variable rates.In the fixed-interval schedule, the organism appears to have come
close to figuring out what the interval is and does its responding at about the end
of that interval. In the fixed-ratio schedule, the organism pauses after each rein-
forcement, as if it is taking a rest before starting the next set of responses.
Response rates are generally higher in the ratio schedules than in interval
schedules, an adaptive behavior, since the rate of reward in such schedules is
directly related to the rate of response. Animals will respond to extreme ratios,
as high as 1000 responses to one reinforcement; however, they have to be
shaped to do so, starting with much lower ratios and slowly working up. The
rate of nesponding. is little related to the reinforcement rate in a ratio schedule.
poetnigher the ratio ofresponses to reinforcers,the longerthe organism pauses —
after a reinforcement. There is a big burst of effort followed by a rest.
The organisms’ pattern of behavior under these various schedules appears
quite adaptive. It need not be the case, however, that they caieaianale know
what the schedule i
is andRC Ree Bate it is thoug th
es re ey are adapted rf)
the CORRES emadon & renen 1989). For instance, in a fixed-inter-
val schedule, the animal is indirectly reinforced for interspersing other behaviors
(such as grooming) between the bar presses, since such composite behavior
produces the food plus other benefits (i.e., closer to bliss point). Animals can
also learn to respond according to a differential low-rate schedule, in which they
are reinforced only if they do not respond too rapidly. Under such a schedule,
intervening behaviors not only produce their own rewards but also serve a tim-

138
 Studies of Choice Behavior

ing function to enable the animal to wait long enough to receive the primary
reinforcement (Hemmes, Eckerman, & Rubinsky, 1979).

Organisms adaptively adjust their pattern of responding given

various reinforcement schedules.

Variable-Interval Schedules and the Matching Law

In the situations presented thus far, the experimenter has offered the organism a
single response-reinforcer contingency. However, organisms usually have a
choice of a number of behaviors, as seen in Table 4.1. A great deal of research has
been devoted to situations in which the experimenter provides multiple (usually
two) responses with two different reinforcement contingencies. The most widely
studied situation is one in which the two responses (typically, two bars that can
be pressed by rats or two keys that can be pecked by pigeons) are reinforced
according to two different variable-interval schedules. Thus, a pigeon may be
presented with two keys, one of which is reinforced according to a VI 15-sec
schedule and the other of which is reinforced according to a VI 30-sec schedule,
so that a reinforcer is available on average every 15 sec or every 30 sec after a
response. The time to the next reinforcer in either schedule only weakly depends
on how often the animal pecks at the key.° Still, it is in the animal’s best interest
to peck at each key some of the time in order to obtain the reinforcers from both.
The question of interest is how it divides its pecking time between the two bars.
With practice the pigeon comes to respond in a reliable way. Let B1 (B
stands for Behavior) be the number of pecks to key 1, and let B2 be the number
of pecks to key 2. Let R1 (R stands for Reinforcement) be the rate at which the
pigeon is reinforced for pecking at key 1, and let R2 be the rate at which it is
reinforced for pecking at key 2. In the example given, the pigeon receives an
average of four reinforcements per minute for pecking at key 1 and two rein-
forcements per minute for pecking at key 2. Thus, R1 is 4 and R2 is 2. The ani-
mal divides its behavior between the two response alternatives in a proportion
that matches the reinforcement proportion. That is,
Sehatvenaital
Bln? kt Re
If R1 = 4 and R2= 2, R1/(R1 + R2)= 2/3; hence, B1 oak be 10 aba B2
es
be 5 leeebvalieee such that B1/ ae+ le his ec ‘

ge but also most other organisms, including humans. The law also holds

3 The weak dependence is produced because once the time for the reinforcer has
come up, the food remains idle until the pigeon pecks. Only after the pigeon pecks
does the timing begin for the next reinforcer.

139
 CHAPTER 4 Reinforcement and Learning

09

0.8

0.7 +
0.6
0.5

0.4
Responses
A%
Key
on
0.3

OFZ

0.1
0
FIGURE 4.11 Choice and reward ©) 101, O2200)30.4..0'5 0160057 10:20.
proportions from Herrnstein (1961). % Rewards on Key A

true, at least approximately, if different magnitudes rather than different rates of

reinforcement are used. For example, if one lever offers two pellets of food on a
VI 30-sec schedule and the other lever offers one pellet of food on a VI 30-sec
schedule, a rat spends twice as long pressing the lever that offers two pellets as
pressing the one that offers one pellet.
Figure 4.11 provides some of the data from Herrnstein (1961) illustrating
the matching law. It plots for various pigeons their proportion of their pecks on
key 1, B1/(B1 + B2), as a function of the proportion of reinforcers they received
on key 1, R1/(R1 + R2). As can be seen, each individual pigeon responds in a
proportion that closely mirrors the proportion of rewards.
eoverncusnnsstctens a ee . - -

Faced with two variable-interval schedules, an organism

divides its responses between them in proportion to their two
rates of reinforcement.
IS CRSP IRENATEN TIN
N EEEESTESE SESE INAS SE ST SHO

Momentary Maximizing
It can be shown mathematically that for an organism to optimize its rate of rein-
forcement in a situation in which it is choosing between two VI schedules, its
behavior should correspond closely to the matching law.* Thus, the organism’s

* This is an elaboration of footnote 3: Under these VI schedules, food becomes avail-

able but is not delivered until the next response. In order to minimize the time that
the food is available but not delivered, the organism engages in behavior that results
in matching. Such behavior maximizes total food by minimizing the time when food
is waiting to be consumed and a VI schedule is not operative. There has been some
dispute about whether matching produces the actual optimal rate of reinforcement
or only a close approximation (e.g., Heyman & Luce, 1979).

140
 Studies of Choice Behavior

behavior in matching can be nearly optimal. It could be argued that the organ-
isn guring out what pattern of behavior will achieve a global optimum, or
me im, in terms of overall intake of food. Such behavior is referréd to as
maximization. It is unlikely, however, that the animal has consciously
figured the contingencies and calculated the behavior that will result in optimal
reward. Herrnstein and Vaughan proposed the melioration theory of choice,
and others (Shimp, 1969; bees ne Hamilton, Ziriax, & Casey, 1978) have pro-
eon e Sem de
italy
diod ch loose

Pr orider an application of melioration to the aang ers avi 1-min

schedule and a VI 2-min schedule. Suppose that a pigeon makes some 3600
pecks in an hour (not an unusual number for a pigeon) and has to divide the
pecks up between two keys. It starts out dividing equally, giving 1800 pecks to
each key. Since it receives 60 reinforcers over the hour on the VI 1 schedule, the
rate of reinforcement on the key is 60/1800 = 1/30 of the pecks. Since it receives
30 reinforcements on the VI 2 key, its rate of reinforcement on that key is
30/1800 = 1/60 of the pecks. According to melioration, since the rate of rein-
forcement is greater for theVI 1 schedule, the pigeon should shift to giving more
pecks to the key with the higher reinforcement rate, namely, theVI 1 key.
Suppose that the pigeon then shifts to 3000 pecks forVI1 and 600 pecks for
VI 2. Its reinforcement rate forVI 1 is 60/3000 = 1/50, and its reinforcement rate for
VI 2 is 30/600 = 1/20. Now VI 2 has the higher rate, and the pigeon should shift to
giving a greater proportion of its pecks to that key. The pigeon reaches a stable rate
of responding when it gives twice as many pecks toVI 1. Assuming a total of 3600
pecks, we see that VI 1 would receive 2400 and VI 2, 1200; the resulting rate of
reinforcement on VI 1 is 60/2400 = 1/40, and on VI 2 it is 30/1200 = 1/40.
As a contrast, consider the application of momentary maximizing to a vari-
able-ratio schedule. Suppose that an animal is choosing between aVR5 and aVR
10 schedule, which means that one-fifth of the responses are rewarded on aver-
age for the first key and one-tenth on average for the second key. No matter how
often the first key is chosen, the reward for that key remains higher. In this situ-
ation momentary maximizing predicts that the animal will settle down to select-
ing the first key exclusively; this is, in fact, what happens (Staddon, 1983, chap.
8). This result is in sharp contrast with what happens with two VI schedules,
when the animal divides its responses proportionately between the two keys.

Organisms choose the alternative that is currently offering the

higher rate of return.

5 The two theories are subtly different. Momentary maximizing claims that the organ-
ism makes the best response now, whereas melioration claims that the response dis-
tribution shifts toward the current best. They differ in the time window over which
the local optimization takes place.

141
 CHAPTER 4 Reinforcement and Learning

oi lees Petia:
Probab 7 mate hing i sg afie on Ot an the ~ Ww It has

more Sen been studied 1in ene aerate the ee law has been more
often studied in lower organisms. A typical experiment (e.g., Friedman, Burke,
Cole, Keller, Millward, Estes, 1964) presents the subject with a pair of buttons
with two associated lights. On each trial the subject must predict which light
will come on by pressing the associated button. Different conditions are defined
by the probability with which each light comes on. In the Friedman et al. exper-
iment, one light came on,with probabilities of 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8,
and 0.9, and the other light came on the remaining time. kick choose ane

Ee in a wide variety of circumstances. People will probability match in many

betting situations where they get paid according to how often they predict the
correct button (e.g., Myers, Fort, Katz, & Suydam, 1963). It has been found not
only in simple choice experiments but also in complex problem-solving experi-
ments where people have to choose different strategies to solve a problem
(Lovett, 1998).
Probability matching has been characterized as irrational because it does
not maximize the potential number of successes. People would maximize their
number of successful choices if they always chose the button that was more suc-
cessful. To see that this is true, compare the expected number of successful pre-
dictions when the probability of one alternative is 0.8. If people just choose that
alternative all of the time, they will be correct 80 percent of the time. However,
consider what happens if they probability match. They will choose that alterna-
tive 0.8 of the time and be successful 0.8 of these times, while they will choose
the other 0.2 of the time and be correct only 0.2 of the time. Their expected prob-
ability of success then is 0.8*0.8 + 0.2*0.2 = 0.68. Thus, they will make the cor-
rect prediction only 68 percent of the time—considerably less than what would
happen if they chose the more probable option exclusively.
Gallistel (1990) describes an undergraduate demonstration in which rats
seemed more rational than humans. He brought a T maze to his introductory
psychology class programmed so that one arm of the maze was rewarded a ran-
dom 75 percent of the time and the other arm 25 percent. The rat ran down the
maze and got rewarded if it made the choice that was correct on that trial. A
light also lit up over the correct arm so that the class could see what was the cor-
rect choice. Before each trial Gallistel asked the class to predict which arm would
be the correct one. His class probability matched and picked the more success-
ful arm 75 percent of the time. On the other hand, the rat came to choose the
more successful arm exclusively.
The behavior was different because the rat and the class had different per-
spectives on the matter. From the rat’s perspective, it is choosing between two
arms that are rewarded on different VR schedules. Down one arm it gets reward-

142
 Studies of Choice Behavior

ed according to a VR 4/3 (it gets three rewards for every four trials on average)
schedule, while on the other arm it gets rewarded according to a VR 4 schedule.
As we noted, in situations that offer a choice between two VR schedules, organ-
isms will choose the schedule with a smaller ratio of trials to rewards—that is,
in this case theVR 4/3 arm. The rat does not see what happens in the other arm.
It does not know that when it failed on one arm it would have gotten rewarded
on the other arm. In contrast, the class has access to this information. When ani-
mals have access to the outcome of the other side, they also probability match
(Graf, Bullock, & Bitterman, 1964; Sutherland & Mackintosh, 1971).

When faced with two alternatives, only one of which is suc-

cessful on any trial, organisms choose the more successful
Bea ee apre tay re its rate oesuccess.
LALOR LRN LID LIE

Optimal Foraging Theory

The choice behavior of animals in the laboratory is not that different from such
behavior in the wild, where animals must make choices about where and how
to seek food. ied have es a theory of how animals make their

n ce between two renee in whieh

to seek food (e.g., a bird looking for food in one of two fields). The animal
appears sensitive to the rate at which food is found in the patch, just as labora-
tory animals are sensitive to the momentary rate of reinforcement from two
keys. The animal chooses the patch with the highest rate. Frequently, as it
searches that patch, it depletes the patch. The animal chooses to move on to the
other patch when the expected increase in food rate justifies the energy cost
involved in making the journey.
Some of the decisions that animals face when foraging in the wild have
been investigated in the laboratory, for example, the effect of travel time
between patches of food. In the studies of the matching law described so far, the
animal faced no cost in shifting from one key (patch of food) to another key
(another patch of food). However, in the wild it costs time to change patches of
food. Thus, animals should be more reluctant to leave an inferior patch when it
takes a significant amount of time to get to another patch. This situation was
investigated in the laboratory by introducing a delay when the animal switched
keys, so that after switching from key 1 to key 2, both keys turned dark for a
fixed interval, after which only key 2 functioned (Fantino & Abarca, 1985). In
one experiment, Fantino and Abarca gave pigeons a choice between a VI 30-sec
schedule and aVI 60-sec schedule. At any point in time, only one key was lit and

143
 CHAPTER 4 Reinforcement and Learning

0.6

0.5

0.44

0.3

% 0.2
Probability
changing
of
keys
FIGURE 4.12 Probability of remain-
ing in the presence of the stimulus 0.1
leading to the less preferred outcome
as a function of the duration of the 60
changeover requirement (travel time). . fo) 4 8 12 16
(From Fantino & Abarca, 1985.) Changeover delay, sec

could be pecked. If the pigeon started out on the VI 60 key, it could switch to the
superiorVI 30 key by pecking a changeover key. However, a delay of 0, 4, or 16
sec was introduced before the other key became operative. Figure 4. 12 shows

Humans can be similarly observed to forage and shift patches for foraging.
A good analogy to foraging in the human case is working. People switch jobs if
a new job offers a high enough increase in income. People are also sensitive to
the travel time involved in making the switch. They are much more willing to
change jobs if the new job is in the same city and quite reluctant to change if it
is in a different country. If the economic disparity is great enough, however,
humans move great distances, as evidenced by the waves of immigration to the
United States. Recently, Pirolli and Card (in press) have applied optimal forag-
ing theory to how people seek information from sources like the World Wide
Web. They show that humans looking for information behave in many ways like
animals looking for food.
Another complication of foraging is that repeated foraging of some patches
tends to deplete them. This situation was explored in the laboratory by giving blue
jays a choice between a depleting and a nondepleting key (Kamil, Yoerg, &
Clements, 1988). The probability of a peck being reinforced with a moth for the
nondepleting key stayed constant at 25 percent. The probability of a peck being
reinforced on the depleting key started at 50 percent, but only a fixed number of
prey were available. When the last prey was taken, the probability dropped from 50

° The blue jays were shown pictures of moths, but they were actually given pieces of
mealworms.

144
 Studies of Choice Behavior

A Bohemian Waxwing foraging for

berries.

percent to zero. Blue jays learned to be adaptive in their strategy, starting with the
depleting key but switching over when that key began to reach depletion. Kamil et
al. found that blue jays showed a strong tendency to switch over after experiencing
a string of three failed pecks at the depleting key. After three such failed pecks, the
odds were in fact fairly high that they had depleted the food source.
EA Re ONS ss

Animals choose among patches for foraging as a function of

travel time to the patches and the current states of depletion of
these patches.
LAER CLE LO LEO NAL AS

Effects of Delay of Reinforcement

Problems that humans, particularly children, have with delayed gratification are
well known. For example, my two boys had been campaigning for a puppy for
months. We had finally agreed that we would get one when we returned from a
vacation in Australia. However, while on vacation, we came upon a trashy (in my
opinion) store, with a theme called’The Lost Forest,”that sold stuffed animals. The
boys were so enamored of those animals that they offered to give up their future
prospect of a real puppy if they could have a few stuffed animals right away. They
eventually prevailed upon us, and we made a decision that we all came to regret.
Lower organisms appear to be even more sensitive to delay of reward.
Rachlin and Green (1972) showed that pigeons would peck at a red light that
gave them an immediate reward of a small amount of food rather than at a

145
 CHAPTER 4 Reinforcement and Learning

2.0 =

-o- Large reward

=—o- Small reward

1 versus 4 sec

ON

value
Subjective 4 versus 8 sec

10 versus 14 sec

FIGURE 4.13 A hypothetical com- 0

parison of the value of different Die iS BO (BPO ee eres
rewards at different delays. Delay, sec

ereen light that gave them a large reward at a 4-sec delay. It is difficult to know
how to judge this issue. It could be argued that in the real world of pigeons
things in the future are so uncertain that it makes sense for the pigeon to take
what it can get right away.’ The situation with our children turned out to illus-
trate the uncertainty of calculations about the future. Eventually, they were able
to cajole their parents to get a puppy despite the bargain to the contrary. In this
case a potential future loss (no puppy) turned out to be unreal.

ture. One an ysis 3 dis-

counting the future turns on the cnpredicebily of the future. Figure 4.13 illus-
trates the economic ct ah and how it extends to ee such as that of

is ee twice the value of the small reward, but the subjective values drop off
quite rapidly. Thus, when a small reward at 1 sec is compared with a large
reward at 4 sec, the small reward has greater subjective value. This explains why
pigeons might choose the small reward at the short delay.
These curves decay in an interesting way. Suppose that a pigeon is offered a
choice between a small reward at a 10-sec delay and a large reward at a 14-sec
delay. The pigeon should choose the large reward at a 14-sec delay, which is of
greater value; Ainslie and Herrnstein (1981) confirmed this prediction. Figure 4.13
implies that pigeons should also be relatively indifferent to a choice between a
small reward at 4 sec and a large reward at 8 sec; this prediction was also con-
firmed (Ainslie & Herrnstein, 1981; Rachlin & Green, 1972). The way these curves

7 T could not resist making a pun about a bird in the hand being worth two in the
bush.

146
 Studies of Choice Behavior

Small reward
Red key

10-sec delay ;
Second choice

Green ke
Right key ‘i 4-sec delay
Large reward

First
choice

Left key

Green key ————— Large reward

10-sec delay 4-sec delay

FIGURE 4.14 The procedure in the experiment by Rachlin and Green (1972).

decay with time determines how sensitive one is to delay. Humans (e.g., King &
Logue, 1987) and other primates (e.g., Tobin, Logue, Chelonis, Ackerman, & May,
1996) are less sensitive to delay than pigeons and rats. On the other hand, young
children respond more like pigeons (Sonuga-Barke, Lea, & Webley, 1989).
Rachlin and Green studied this issue in an interesting paradigm illustrat-
ed in Figure 4.14. At the beginning of a trial, pigeons pecked at either a right key
or a left key. If the right key was chosen, after 10 sec of darkness they were
exposed to a choice between a red key that gave the small reinforcement imme-
diately or a green key that gave the large reinforcement after another 4-sec
delay. They manifested an inability to delay gratification and almost always
chose the red key. Thus, if they chose the right key initially, they were effective-
ly choosing a small reward at a 10-sec delay. If they chose the left key, 10 sec of
blackout was followed by a green key that, if pecked, gave them a large reward
in 4 sec. Thus, if they chose the left key they wereicaiagial choosing a large
Hees at resec. peal BEER the leftheThey wer making a choice early

The notion of ipetorntnement nae been extended to human decision

making. For instance, dieters often have trouble choosing a healthy, low-calorie
dish over a rich, high-calorie dish when both choices are presented to them. The
immediate rewards of the high-calorie dish outweigh the long-term benefits of
the healthy dish. However, if the dieters can order ahead, they choose the
healthy dish. At a delay, the rich dish does not seem so tempting.

Guaee erate ae costhmeeinee a wayine ie will

prefer a sil bio siik eames
toaPa ge cud at oe
a
ees RNR EON eA

147
 CHAPTER 4 Reinforcement and Learning

Mechanisms of Choice
Animals have thus far been described as being quite rational in their decision
making. They strive to achieve the optimal tradeoff among their competing
needs, displaying rates of responding under different schedules that comes as
close as possible to their bliss point. They make appropriate choices foraging in
the wild (and between alternative schedules in the laboratory), behaving in a
way that can be viewed as maximizing their net energy intake. Their discount-
ing of time corresponds to economists’ prescriptions for rational behavior.
This discussion raisés the question of how animals achieve the near opti-
mality of their choice behavior. It is unlikely that they are engaging in anything
like explicitly calculating the prescription for rational choice set forth at the
beginning of the chapter. It also is implausible that they are always correct in the
choices they make. Rather, animals can be viewed as being governed by what
amounts to rules of thumb, which serve them relatively well in most situations.
For instance, an animal could produce the matching law by always choosing the
alternative with the higher momentary rate of reward. Similarly, the blue jays in
the Kamil et al. (1988) experiment chose to leave a depleting patch after a string
of three failures (which happened to be quite predictive of patch depletion).
Animal behavior is probably controlled by such simple near-term rules that tend
to approximate global optimal behavior.

Animals seem to make their choices by simple rules of thumb

that approximate global optimality.
HERRERA EAS EBS NaS ENR

Human Decision Making

Insight into animal decision making can be gained by looking at human deci-
sion making. Humans are presumably more deliberative and rational than other
creatures and therefore should define the high point for rationality. It turns out,
however, that humans also tend to make their choices using such short-term
tules of thumb. Much research has involved making choices among sets of alter-
natives, such as the four apartments described in Table 4.3. One model for ratio-
nal choice is to rate each apartment on each dimension, add up the ratings, and
choose the apartment with the highest ratings. The table illustrates a set of rated
values for each attribute for each apartment. In terms of the bliss-point model
of reinforcement, the rated values in Table 4.3 could be viewed as reflecting how
close each apartment is to the bliss point on each dimension (a high rating is
closer to the bliss point). At the bottom of the table, these numbers are summed
to get the total overall value for the apartments. The best choice is revealed to
be B, which is cheap and scores reasonably on the other dimensions. Typical
proposals for optimal behavior have the subject consider each apartment on
each dimension and calculate some aggregate measure of goodness.

148
 Mechanisms of Choice

TABLE 4.3 Attributes of Various Apartments and Their Rated Values

(in parentheses)
Apartment A Apartment B Apartment C Apartment D
Cost

Number of
rooms

Distance
from work

Parking

Condition

Apartment
A: =2 +8 +0+3+48=17
Apartment B:10+4+4+3+4+5= 26
Apartment C:9+4+6+0+5=24
Apartment D:6+8+6+0+2=22

People have been observed to do many things in making decisions such as

choosing among apartments, but seldom have they been observed to corre-
spond to what would typically be recognized as rational choice behavior. The
following are two of the nonoptimal strategies that have been observed in
human subjects.

Elimination by aspects (Tversky, 1972). Focus on just the most impor-

tant dimension first, for example, price. Eliminate from consideration any
choices that are not close to the best on this dimension. Using Table 4.3,
this practice might leave just B and C in the choice set. Then consider the
next most important dimension. Continue until a single choice has been
identified. The problem with elimination by aspects is that it is possible,
by focusing on the initial dimension, to eliminate a choice that might be
so good on other dimensions as to be best overall.
Satisficing (Simon, 1955). Consider the alternatives one at a time in the
order in which they occur. Set a cutoff for the value of an alternative of
each dimension. Reject an alternative if any of its values are worse than
the cutoff. Accept the first alternative whose values on all dimensions are
above the threshold. Using Table 4.3, we might set thresholds on cost at
$400, number of rooms at 3, distance at 1 mile, available parking, and good
condition. Apartment A can be rejected immediately because it is too
expensive; Apartment B is immediately accepted because it passes the
threshold on all dimensions. C and D would not even be considered.

149
 CHAPTER 4 Reinforcement and Learning

Because it does not consider all the alternatives, satisficing may miss the
best choice.

Although these strategies do not always yield the best results, they often pro-
duce the best result or a result close to the best and they give results quickly
because they allow the decision maker to focus on a subset of the information.
Payne, Bettman, and Johnson (1988) showed that under time pressure, the elim-
ination-by-aspects strategy more often yields the correct decision than the com-
plete rational decision strategy. When the time is up, the rational decision strat-
egy may not yet have considered the correct alternative, whereas the elimina-
tion-by-aspects strategy may have already identified the correct alternative.
They also showed that subjects are sensitive to time pressure and tend to switch
from an exhaustive strategy to an elimination-by-aspects strategy when time
pressure increases.
Thus, humans, like lower animals, use rules of thumb for making deci-
sions, and these rules of thumb do not always yield the optimal choice. The
Payne et al. study identified another important aspect of the process of making
rational choice: different strategies take different amounts of time to execute. An
individual cannot exhaustively consider all the apartments listed in a typical
urban newspaper; nor can an animal in the wild spend inordinate amounts of
time making decisions. It might therefore be rational to use rules of thumb that
only approximate the best decision but to do so quickly. Economists refer to this
consideration as the cost of information. In judging what the best choice is, we
need consider not only the value of the choice but also the cost of seeking the
information that went into making that choice.
a eSSRNR a aT Nd

Humans often use rules of thumb to approximate best deci-

sions, particularly when they are under time pressure.
ea i : PEARSALL RIEONSET NESSIE
ITESEEDS EIEN TEE I IEE EE:

Final Reflections
This chapter concludes the discussions of research focused on animal learning,
although later chapters will return many times to issues of animal cognition and
memory. In general, animals learn the contingencies in the environment and
behave in a way so as to nearly maximize what they desire. Issues of reinforce-
ment necessarily play a major role in-research on animal learning because
appropriate reinforcement conditions are needed to get the animal to display
what it has learned. In contrast, humans can be directly queried, and experi-
menters can count on their general social cooperativeness to say what they have
learned. .
Much of human learning is available to consciousness, and humans often
consciously decide what knowledge to display and how. For instance, students
can tell what they know of a history lesson and why they chose to display this

150
 Further Readings

knowledge in a certain way in response to an essay question. However, as dis-

cussed in Chapters 8 and 9, not all human knowledge is available to conscious-
ness, and people are not always aware of the knowledge they are displaying.
Few of us can tell what we know about riding a bike, nor are we aware of how
our knowledge comes into play in bike riding. The suspicion has always been
that more (or all) of animal knowledge is of this unconscious, behavioral variety
than is human knowledge.
Human behavior, which involves unconscious use of experience, can be as
adaptive as behavior that involves deliberate choice based on consciously
recalled experiences. The fact that animals tend to respond in near optimal ways
to experience does not necessarily imply any conscious deliberation on their part.

Humans display what they have learned in both conscious

and nonconscious ways. Both displays of learning can be
equally adaptive.
AAAS

Further Readings
In addition to the sources suggested in Chapters 2 and 3, much of the research
on schedules of reinforcement can be found in the Journal of the Experimental
Analysis of Behavior. B. Williams (1988) provides an overview of research of
choice behavior under different schedules. Herrnstein (1990) offers a readable
discussion of the matching law and other issues about choice behavior. Staddon
(1983) and Staddon and Ettinger (1989) present an adaptive analysis of behav-
ior. The book edited by Klein and Mowrer (1989) contains a number of recent
articles on the nature of reinforcement. The May 1993 edition of Psychological
Science contains a series of articles on comparative cognition—relating animal
cognition to human cognition.

151
 Transient Memories

Conditioning Research
Versus Memory Research
This chapter marks a transition from research focused on animal conditioning to
research focused on human memory. The critical distinction is not so much the
species of the subject as it is the methodology used—the conditioning experiment
versus the memory experiment. The typical conditioning experiment involves a
rather complex situation from the point of view of the organism. The organism is
placed in a novel environment and is given some experiences that typically include
strong manipulations of motivation. The experimenter is interested in the behavior
that arises as a function of these experiences and motivational manipulations.
Figure 5.1 illustrates one way to conceive of what is happening in a con-
tioning experiment Theorganism explo imply hastofigureoutthe
structure of the environment, including reinforcement contingencie
experiences. The process by which it does so is Sometimes calle
input to this process is its experience, possibly combined with biological predis-
positions and, in the case of humans, with some instructions. The result of this
process is some knowledge of the environment, and this knowledge is deposit-
ed in memory. A process called Fi verts_this
knowledge into behavior. Figure 5.1 illustrates that this process of motivation is
also affected by the organism's current goals. Memory is just one part of a larg-
er system that is involved in a conditioning experiment; memory experiments
try to focus on that one part.
Figure 5.1 also embodies the important discovery that learning processes
are separate from motivational processes. Chapters 1 and 4 reviewed the evi-
dence that motivation does not influence what is learned but how organisms
display this learning. For instance, Tolman’s rats learned about mazes in the
absence of reinforcement, but displayed that knowledge only when food was
put in the goal box. Thus, Figure 5.1 illustrates that motivation does not control
what goes into memory, but rather how memory maps onto behavior.

152
 Conditioning Research Versus Memory Research

Environment
Instructions — Experience Behavior

Biological
Organism
predispositions

Knowledge of environment
Memory

FIGURE 5.1 A conception of systems in a conditioning experiment.

€

Figure 5.1 is a rational reconstruction of the factors that must be involved

in a conditioning experiment. It is not meant to imply that there is always an
induction stage, a memory stage, and a motivation stage. For instance, in Aplysia
(Chapter 2) learning involved directly associating sensory neurons to motor
neurons. This learning process implicitly induced the relationship between the
CS and the US, resulted in a memory stored at the synapse, and produced an
adaptive reflex, but there were no separate induction, memory, and motivation
stages. As noted throughout this book, organisms can behave as if they went
r through a rational process without really doing so. The higher the organism is
on the phylogenetic scale, the more likely that such rational processes actually
occur at least in some cases. However, even in humans there is undoubtedly
simpler “as if” learning as well.
UI ES RAE SUH PENNIES SE ET AENNSE NIEENE CIES SEI BE URAL EUDORA BELO NE MESES

. .

achieve its goals. 3

ath ha NL REI SSN a UE SIRES NA RENAME ERIC TET SCR RRUESTER ANI ROSEN

Animal Research Versus Human Research

With this picture in mind, let us reflect again on the relative advantages of ani-
mal versus human experimentation, a topic mentioned in Chapter 1. Because of
the greater complexity of human behavior, it is hard to study the full system
shown in Figure 5.1 in one experiment, whereas a prototypical experiment with
a rat in a Skinner box taps all aspects of the system. Animals are therefore often

153
 Chapter 5 Transient Memories

more suited for the study of the complete system. Animals are also often more
suited for the study of a single aspect of the system as well because there are
fewer ethical constraints regarding their treatment. Many experiments on the
motivation and physiology of learning involve extreme manipulations that can
only be done with animals. -
Humans often offer advantages, however, because they can be more care-
fully directed through instructions and experimenters can count on their social
cooperativeness to obtain appropriate information. The advantage of human
subjects
DN1eCC is
| strongest
Noes in the e behavioral study of the me ystem. In thetypi-
cal conditioning experiment, the memory system is only indirectly connected to
the environment. The induction phase intervenes between the input from the
environment and the memory system. The induction system figures out what
goes with what and hence what should be stored in memo

ereas a rat must figure out

that pressing a lever causes food to be delivered, a human subject can simply be
told. Because it is so easy to inform humans about what they should remember,

A simple human
memory experiment might require subjects to learn 20 stimulus—-response
paired associations, such as dog—6, where they must respond with the second
term when shown the first. Twenty such pairs are a herculean number of stim-
ulus—response associations to train in a conditioning experiment. With human —

Just as induction causes difficulty on the input side in a conditioning

experiment, motivational processes on the output side also create difficulty in
making inferences about memory.experimenters
With humans, do not usually

These experimental advantages have made humans the subjects of choice

in behavioral studies of memory. These studies have yielded a great deal of
knowledge, which is explored over the next four chapters. Researchers have‘also~

Research on human memory has another advantage: it is more obviously

relevant to us. Subsequent chapters frequently explore the implications of
research for our own need to learn and remember information. This memory
research is nowhere more relevant than in education. The typical memory
experiment is often a miniature of the learning processes involved in mastery of
a subject domain (such as the one you are studying in this text). The last chap-

154
 Sensory Memory

ter of this book is devoted to developing the implications of research on learn-

ing and memory for education.
This chapter is concerned with in contrast to the next
three chapters which will be concerned with more permanent memory. Our sys-
tem has numerous ways of temporari ly holding information for further process-
ine. These behave like a television screen.

permanent memory. In the television

screen analogy, if one is not recording the program, one will not be able to recre-
ate the contents of the screen. Many of these transient memories are quite sen-
sory in character, and we will begin our discussion with storage of information
in the systems that do processing of the sensory signals we receive from the
environment.
4

Research on human memory focuses on a subset of the larger

system that is involved in research on conditioning.

Sensory Memory
How does information from _ GSS ORs ou Peper iL Some perma”
nent memory? t

é uch research has been devoted to the cieettties of

_ the temporary stares that exist to hold visual and auditory information.

Visual Sensory Memory

"One of the temporary memories isivistal sensory memory. which holdsinfor:
mation perceived bytheHisualsystem. ct’consider one line of experiments
that has been used to study visual sensory memory. Take a brief glance at Figure
5.2, turn your eyes away, and try to recall what letters were there. Most people,
if they have just one brief look (less than 1 sec), report that they are able to recall
only four or five letters. This procedure seems to tell us about the capacity of
. These infor-

X M R J
C N K 1?
Ww F L B
FIGURE 5.2 Type of visual display used in visual report procedures.

205
 Chapter 5 Transient Memories

mal observations have been confirmed by more careful laboratory studies in

which subjects were given a very brief (e.g., 50 msec) exposure to an array of let-
ters such as that shown in Figure 5.2. Such results might lead to the conclusion
t only a few items can be stored in visual sensory memory.
Many subjects, however, indicated that they felt they actually saw more
than the few letters they reported, but that the other letters faded before they
could report them. Perhaps you have also had this experience. Sperling (1960)
conducted a study that confirmed what the subjects were saying. As in the other
experiments, he presented the visual array for a short period (50 msec). Rather
than ask the subjects to report the whole visual array, he asked them to report
one row from the array. Immediately after the array was removed, Sperling pre-
sented his subjects with a high tone, a middle tone, or a low tone. If the tone was
igh, subjects were supposed to report the top row; a middle tone corresponded
o the middle row; and a low tone corresponded to the bottom row. This proce-
dure is called the in contrast to the whole report proce-
dure, in which subjects must report everything. Sperling found that the subjects
e able to recall a little over three of the four items that appeared in a row.

chasiscissanidaadedieogenemem He studied the time course ofaz:

d ecay by delaying presentation of the tone after removal of the array. Figure 5.3
shows how the number of letters that the subject could report decayed as the
tone was delayed up to 1 sec: The number quickly dropped off. This result indi-
cates that information in_the visualsensorymemory hasa_veryshartlie
Probably much of what the subjects were able to recall after 1 sec reflected what
they had identified in that second before they knew which row they would be
asked to report.
7e8) calleduhus :

<< feed some system to hold the information until it i

encoded into a more permanent form

the retina. She showed that many of the timing and sensitivity properties of the
iconic image mirror those of the rods, which are the photoreceptors in the eye
responsible for night vision. According to this analysis, the icon is very much like
the afterimage of a bright light at night. However, there has to be more to icons

rod-based afterimages (Coltheart, 1983).

'That their reports were not always perfect probably reflected difficulties in execut-
ing the reporting and the fact that some of that row might have faded before it could
be reported.

156 i
po
 Sensory Memory

4.0

2 cola

recalled
letters
of
number
Mean

FIGURE 5.3. Results from Sperling’s

partial report procedure. As delay of 0.0
signal is increased, number of items oe gO Se.30 1.00
reported decreases. Delay of tone, sec

‘Haber (1983) questioned the relevance of the icon for normal visual per-
ception because humans do not usually perceive the world in such brief flashes;
64 OOD) tee

Hiidt DIOCESSITL?E 15 OL

us, it may not matter that there was only a 50-

ms. flash in Sperling’s experiment.
The evidence is quite persuasive that visual information exists in the sys-
tem at many levels after we see something. A great deal of information is main-
tained in such a form but only briefly. However, there is not agreement as to the
forms in which this information is maintained or how this information is used
in higher-level cognition.

Sensory information is maintained in the visual system for

brief periods of time.
Leh RI AI gA NARMS ANN NN EAS A EIR LODE AUN ESSN A LE SER

Auditory Sensory Memory

‘information is temporarily stored in sensory memories. Early demonstrations of

an i
( nted subjects
with simultaneous recordings of three lists of three items over two headphones.
The items were letters and numbers; thus, one list might be 4 L 6. By stereo-
_ phonic mixing of the signals from the two headphones, it appeared that one list
was coming from the subject's left, another list from the subject’s right, and a

157
 Chapter 5 Transient Memories

third from directly above the subject. As in the visual report procedures, subjects
were not able to report all nine items (3 lists x 3 items) when asked to recall
everything. However, the researchers also used a partial report condition simi-
lar to that used by Sperling. They asked the subjects to report just one of the
three lists. They cued subjects as to which listto report by presenting a visual
indicator on the right, middle, or left of a screen.
Subje i centage oO
they were cued than when they were told to recall the total list. Like Sperling,
Darwin et al. interpreted this to mean that there were items in an auditory buffer
that could not be recalled’in the whole report procedure because they had faded
from memory before the subject was able to report them. In the partial report
procedure, subjects had to report fewer items and so got to reporting the criti-
cal items sooner. Darwin et al. also showed that the amount that could be
reported in the partial report procedure decreased as the delay between the end
of the lists and the cue increased) After 4 sec there was very little advantage for
the partial report procedure.Asin the case of iconic memory, this result indicat-
ed that items were decaying and that subjects were only able to remember
something from the critical list if they encoded it before it had faded.
Neisser called the auditory sensory memory that held these items echoic
memory. As in the case of iconic memory, he argued that people need some
‘memory to hold the sensory experience so that it can be analyzed and encoded
into a more permanent form.
Glucksberg and Cowan (1970) conducted a rather different experiment,
which resulted in a similar estimate of the duration of the echoic store. They pre-
sented subjects with two spoken messages, one in each ear, and they required
subjects to repeat the message that was being said in one ear. This task, called a
shadowing task isvery demanding, and subjects typically remember nothing of
what is said in the other ear. From time to time, the experimenters said a digit
to the ear that was not being shadowed. They stopped the subjects and asked
whether a digit had occurred. They found that if they asked subjects right after
the digit, subjects could still detect it with some success. This performance
dropped off dramatically over the first 2 sec, and after 5 sec the subjects showed
no ability to detect the digit. Thexesearchers concluded that in less than 5 sec
echoic memory was completely lost for the unattended digit.
Conrad (1960) and Crowder and Morton (1969) showed not only that
ecays from echoic memory with the passage of time but also that
additional auditory information can interfere with it. orton pre-
ented thewordzerojustbefore asking subjecis Torecall a list of digits. Subjects
were told to ignore the last zero and to recall the preceding digits. This final digit
seriously impaired memory for the list. This phenomenon is called the suffix

Impairment occurs because the zero enters auditory memory and interferes with
the target list.
The interference associated with the suffix effect seems to be verbal in
character. Crowder and Morton found that if they used a buzzer rather than a

158
 Sensory Memory

word as the suffix, there was no interference iimnilggiesaeye eRe,

! eath, Surprenant, and Crowder (1993)

ound that a“baa” sound produced a suffix effect if subjects were told it was pro-
duced by a human but not if they were told it was produced by a sheep.
SNE NNID EDR REASIR REM EOOLUN SEROUS SE ESM TERN THCY SENS TERED ELEN LOE SPDR scala sumsanamudiiaias

Oral or spoken information is maintained in the auditory sys-

tem for brief periods of time.
S OR STREETER SATAN
ZINAH OIG MNS ARETE IOUT
ae

Conclusions about Sensory Memory

Sensory memory iscapable ofstoring more orless complete records ofwhat has

before it decays, itislost. What subjects encode depends on what they are pay-
ing attention to. Typically, the environment offers much more information at one
time than we can attend to and encode. Therefore
beeconrs fae ear MRR SEte Similarly, instrumental conditioning
research (see the discussion of dimensional learning in Chapter 3) has shown
that animals often do not initially attend to certain aspects of the stimuli pre-
sented to them. Presumably, nonhumans are also overwhelmed by the richness
of the stimuli they experience and can only pay attention to certain things.
Although the auditory and visual systems are perhaps the most important,
they are not the only sensory systems that display transient memories for infor-
mation. a pe rd
performance of a skill. Information,
etas it is by many of ourersensory
systems s. In some cir-
cumstances,
The remainder of the chapter describes
‘the overall information processing, including their roleincreating permanent _
ies. At one time it was thought that there was a sequence of memories,
SF ction passed from sensory memory to short-term memory and then
to long-term memory. The next section discusses the problems associated with
the concept of an intermediate short-term memory. The remainder of the chap-
ter describes the more current idea of a working memory.
ESERIES IRON ELIE OLE NREL
SLL LLANELLI LELAND PILLLELEL INE HERA IRAE IOUSLESNON NUNES REPRESS

Sensory re cana Featbai iets

seeds of ue information
sie are oN alah and Wegcan serve as pled ag
memories.
aaa NG NEE BNO RSTNUPEESSERE RON NSE NELERED
RST DARE ENON PEERSSSRN AHA ARETE NEI AE NEE IEE EIR RRO

159
 Chapter 5 Transient Memories

The Rise and Fall of the

Theory of Short-Term Memory
ory (see Figure 1.11 from Chapter 1)pro=

e classic example is a telephone number that we will

repeat over and over agaih until it is memorized. The key feature of this theory is
the proposal that short-term memc a necessary halfway station between se
sor memonuand—long=term memory. Many psychology texts still include this
proposal of two separate memories, and it is worthwhile reviewing why this view
was held. The distinction between short-term and long-term memory was predi-
cated on a number of claims; particularly significant were the following three:

Each of these claims was based on some empirical data. However, it has become
apparent that all the data cannot be properly understood if there is assumed to
be a short-term memory between sensory memory and long-term memory. This
section considers each claim, the evidence for it, and the problems with it.

The theory of short-term memory was based on claims about

the effects of rehearsal, coding differences, and retention of
information. »
stubs emcee emma eins

Effects of Rehearsal
As discussed in Chapter 1,
He asked subjects to rehearse out loud and
counted the number of times they r ehearsed each wor
at \ ehearsed more were better recalled (see Figure 1.12).
This result
was just as predicted by the Atkinson and Shiffrin theory, which proposed that
information got into long-term memory by being rehearsed in short-term mem-
ory. Sometimes, however, rehearsal does little to improve long-term memory.
977) had subjects study a four-digit number for 2
sec, then rehearse a word for 2, 6, or 18 sec, and finally recall the four digits.
Subjects participated in 64 trials. Subjects thought that the experimenter’s inter-
est was in digit recall and that the words were only being used to fill up the reten-
tion interval. After the experiment, subjects were asked to recall the words they

160
 The Rise and Fall of the Theory of Short-Term Memory

had been rehearsing. Their recall averaged an abysmal 11, 7, and 12 percent,
respectively, in the 2-, 6-, and 18-sec rehearsal intervals. Thus, subjects showed
little recall and no relationship between the amount of rehearsal and the amount
of recall. Glenberg et al. also tried a recognition test for the words and found only
a weak effect of amount of rehearsal on memory performance.
Craik and Watkins (1973) used another paradigm to show the lack of an
effect of passive rehearsal‘on memory. Their subjects heard a list of 21 words and
were supposed to recall the last word that began with a certain letter. Thus, if the
critical letter was G, subjects might hear daughter, rifle, garden, grain, table, foot-
ball, anchor, giraffe, fish, tooth, book, heart, mouse, gold, can, ball, paper, fire, glass,
house, shoe, and should recall glass. When the subjects heard the first G-word,
they did not know whether it was the word they would have to recall; therefore,
they had to rehearse it until they heard the next G-word. Different words had to
be rehearsed for different lengths of time. In the example given, garden would
be rehearsed for zero words, grain three words, giraffe five words, gold four
words, and glass two words. After studying 27 such lists, subjects were given a
surprise recall of all the words. Craik and Watkins found no relationship
between how long the word was rehearsed and its final probability of recall.
Perhaps the most dramatic evidence for the lack of relationship between
the amount of rehearsal and long-term recall is the report cited by Neisser (1982)
of a Professor Sanford, who estimated that he had read the family prayers at
meals some 5000 times over 25 years. Despite all this rehearsal, Professor
Sanford found that when he tested his memory, he had very little memory for the
prayers. | ee good long-term memory.
If it is not sheer repetition, what does determine how much we remem-
ber? Rundus (refer to Figure 1.12) found a relationship between number of
rehearsals and recall, but, unlike the experiments just cited, his subjects were
actively processing the material with an eye to remembering it. In an influential
article, Craik and Lockhart (1972) argued that what _was critical was the depth
to which information was processed. According to this theory, called the depth.
r i
of processing arsal_ improves
theory, rehearsal improves memory only if
memory only if the
the material
material is
_is
rehearsed in a deep and meaningful way; passive rehearsal does not result in
better memory. This point of view was nicely illustrated in a series of experi-
ments by Craik and Tulving (1975). In one experiment they showed subjects a
word, such as table, and asked them to make three types of judgments. The shal-
low-level judgment, about case, was whether the word was in capitals. The
intermediate-level judgment was whether it rhymed with another word, for
example, cable. The deep-level judgment was whether it fit in a sentence, for
example, “He put the plate on the .” They later asked subjects to rec-
ognize the words. Figure 5.4 shows the proportion of words recognized as a
function of the type of processing of the words. The more deeply processed
words were better remembered. Similar effects of depth of processing can be
shown in memory for faces. Subjects show better memory for photos of faces if
they are asked to judge whether the person is honest rather than whether the
person is male or female (Bloom & Mudd, 1991; Sporer, 1991).

161
 Chapter 5 Transient Memories

.80

Ss
aS)oO

recognized
Proportion
Nm fo)

FIGURE 5.4 Proportion of words recognized oo

as a function of type of initial processing. (From ; 1 2 3
Craik & Tulving, 1975.) Case Rhyme — Sentence

ere was no improvement with rehearsal time in the Glenberg et zy or the

“Craik and Watkins studies because subjects continued to rehearse the words at
_a shallow level. The depth of processing explanation has been criticized because
the concept of depth is somewhat vague (Nelson, 1977) and because, as
reviewed in Chapter 8, there are interactions between theBayne ofbpracessing at
stad and at test. Ng

es automaticall ory. This result discon-

firmed the original Atkinson and in theory of short-term memory, which
_ proposed that information was transferred to long-term memory as a function
_of verbal rehearsal. 2
SA SRE EERO ES UNL RIT RISEN DUAR ERIE ETS ee een GSC LMA RUS nN

Passive rehearsal of material will not increase its recall, but

deeper processing does. *
fora
ev ACRE DANILO SS SER ET IER TT

Coding Differences

-te é . re People fies

rehearse material, aah as word lists in experiments or telephone numbers in
everyday life, by saying the items over and over again to themselves. This means
that the information tends to have articulatory and acoustic features. It was

162
 The Rise and Fall of the Theory of Short-Term Memory

A classic experiment showing this distinction between short-term and

long-term memory was performed byaintschand Buschike%(1969) They had
subjects study lists of 16 words presented visually at a rate of one word every 2
sec. They then provided subjects with one word from the list and asked them to
recall the word that followed it in the list. It is worth reviewing how the
Atkinson and Shiffrin theory is supposed to apply to memory for such a list of
words (see the discussion surrounding Figure 1.12). The subject was assumed to
be rehearsing in short-term memory the last few words read. Therefore, when
asked to recall the words after study, the subject should display particularly good
memory for the last few words in the list because these were rehearsed. This
result il occur beadvantage fortheasa wordsfs
referredto
astherecen--
eietiectMemory or the rest of the list depended on retrieving the words
from
O ng-term memory.

Og bjs were asked what followed

see, they might recall the word that followed sea. Such acoustical confusion
would be a particular problem for words at the end of the list, which were sup-
posed to be in short-term memory. On the other hand, words from the begin-
ning of the list might produce semantic confusion if they were synonyms, such
as sea and ocean, because these words were in long-term memory, where the
coding was supposed to be semantic.
In three conditions, subjects saw words that were unrelated, synonyms
(similar meaning), or homophones (similar sounding). Kintsch and Bushke found
that recall for the last few words in the list was impaired if the words were homo-
phones and that memory for the first words of the list was impaired if the words
were synonyms. It seemed that short-term memory was acoustic (sound-orient-
ed) and long-term memory was semantic (meaning-oriented), as hypothesized.
Yet, serious problems arise when one tries to advance coding differences
as a fundamental distinction between short-term memory and long-term mem-
ory. For instance, tasks that are supposed to rely on short-term memory also
take advantage of meaningful codes..Bower and Springston (1970) showed that
subjects had much larger memory spans for sequences of letters when they
formed meaningful acronyms, as in,

IBM FBI ABC USA

‘Memory for these letters was supported by semantic information which, accord-
. ing to the short-term memory theory, is a long-term memory coding. In addi-
tion, the fact that people can remember rhymes and sounds for long periods of
time provides evidence of acoustic codings in long-term memory.
People can also make semantic confusions in a short-term memory task.
Potter and Lombardi (1990) presented subjects with sentences such as

The knight rounded the palace searching for a place to enter.

163
 Chapter 5 Transient Memories

Subjects then heard a list that included the word castle. After hearing the list,
they were asked to recall the original sentence. About a quarter of the subjects
recalled

The knight rounded the castle searching for a place to enter.

and thus intruded the semantically related word. This semantic confusion indi-
cates that our immediate memory for a list of words (in this case a sentence)
involves semantic information as well as acoustic information.
An alternative proposal, based on the concept of depth of processing
2 (Craik & Lockhart, 1972; Wickelgren, 1974), proposes that different types of
information are forgotten at different rates and that sensory information may be
more shallowly processed and may be forgotten more rapidly. Initially, memo-
ries may show a preponderance of sensory traces, but as time passes, semantic
traces remain. (This issue of differential forgetting for different types of memo-
ry is addressed in more detail in Chapter 6.) Thus, acoustic confusion predomi-
nates early in the Kintsch and Bushke experiment because the acoustic infor-
mation has not been forgotten. As time passes the acoustic information is for-
gotten, leaving primarily semantic information, which was there all the time,
and resulting in mainly semantic confusions.
ORSON

Both acoustic and semantic information can serve as the basis

of memory performance at short and long delays.
SRNL RULE ‘she RAMA RST SPA RT

The Retention Function

The ability to recall information decays rapidly in the short interval after it has
been studied. According to the hypothesis of a separate short-term memory, the
superior memory for recent material indicates that it is held in a special short-term
memory store and the rapid decay reflects loss from short-term memory. Several
other paradigms have been used to show that there is an initial rapid loss of infor-
ree oe the retention of
simple nonsense trigrams, such as CH], ove ~Subjects might not be expect-
ed to exhibit much forgetting of three letters over such a short period. However,
the experimenters distracted subjects by having them count backwards by threes
from a large number, such as 418. With this sort ofdistracting task there was rapid»
«forgettingof the trigrams. Figure 5.5 shows thesetention functions» obtained./A»

recall. Note that retention decreased to about 20 percent, not to zero, supposedly
reflecting what the subjects were able to store in long-term memory. Given longer
initial study, subjects would have shown a higher level of performance after 18 sec,
reflecting the transfer of more information to long-term memory.
There are several problems with such demonstrations. Such steeply
decreasing retention functions are not obtained for subjects’ recall of their first

164
 The Rise and Fall of the Theory of Short-Term Memory

100

80

60

40
recall
Percent

20

0 S 6 9 ee 15 18
Time, sec

FIGURE 5.5 Decrease in recall as a function of duration of the distracting task.

(From Murdock, 1961.)

nonsense trigram (Keppel & Underwood, 1962); in this case they show relative-
ly little forgetting.
i Thus,
animportant
Chapter 7 examines these
studies and retention effects at greater length, but for now it should be noted
that the forgetting in Figure 5.5

committed
to long-term memory.
The retention curve in Figure 5.5 appears to show rapid initial forgetting
followed by no further forgetting. This apparent discontinuity of the forgetting
curve was used to argue for two stores (e.g., Waugh & Norman, 1965). The ini-

whereas the flat portion of the curve was supposed to represent what got into
long-term memory. However, note that performance is still dropping off from 9
to 18 sec (and this curve would continue to drop off beyond 18 sec).
pointed out that all retention curves for any period of time show negative accel-.

Consider the retention study of Ebbinghaus (Figure 1.1 from Chapter 1). He
looked at retention of material that was much better learned than that repre-
sented in Figure 5.5 over intervals of up to 30 days. Figure 1.2 shows a negatively
accelerated retention curve with an apparent discontinuity at around two days,
where the curve is flattening. Clearly, there is not a short-term memory of two
days! The forms of retention curves are not evidence for a discontinuity between
short-term and long-term memory; the curves are really continuous, and their
apparent discontinuity is merely an artifact of the scales on which they are
graphed. Chapter 7 discusses the nature of these retention functions more fully.

165
 Chapter 5 Transient Memories

The shape of the early part of the retention function and the
factors that influence it are the same as for the later part of the
retention function.

Conclusions about Short-Term Memory

The two-memory proposal claimed that there are two separate memory stores
with their own distinct Gharacteristics and that information passes from the
short-term store to the long-term store. The alternative proposal is that there is
just one general memory system into which information from our sensory sys-
tems is encoded (e.g., Melton, 1963; Wickelgren, 1974). Different memories in
that system can have different attributes. The way we process that information
can influence how well we remember it. In particular, more deeply processed
information tends to be forgotten less rapidly. There is no discontinuity in the
retention functions for any kind of material. All forgetting functions are nega-
tively accelerated through any retention interval.
The data on the effects of rehearsal, types of codes, and retention effects
can be accommodated without postulating a short-term memory as a halfway
station between sensory memory and long-term memory. However, people do
engage in rehearsal processes, and these processes can affect memory perfor-
mance. Baddeley (e.g., 1986) proposed a concept of slave rehearsal s s to
help explain these effects. These rehearsal systems are considered next.

Rehearsal Systems
A previous section reviewed the evidence for the existence of sensory memories
that maintain transient records of our experience. For instance, we have an
acoustic memory that holds information for a few seconds. What happens if we
say a set of numbers to ourselves? Presumably, we are filling our echoic mem-
ory with a sensory record of our own speech. If we do this over and over again,
we can keep a record of those digits in echoic memory. By rehearsal, we are
making our echoic memory a system for holding the digits. There are other ways
‘in which we can use our bodies as systems for holding information. We can
remember a digit by holding out the appropriate number of fingers; we can
remember where something is in space by staring at that location; we can
remember the width of a board by placing our two hands at its sides and then
keeping our hands that far apart. People can be extremely imaginative in how
they use their bodies as transient memories to hold information. The verbal sys-
tem is particularly important, and a great deal of research has been devoted to
how this system may be used to hold information. Baddeley used the term
phonological loop to refer to our use of the verbal system as a transient mem-
ory.

166
 Rehearsal Systems

The Phonological Loop

According phonological
to Baddeley, the loop is composed of two systems—astore

speech (speaking to oneself). The a loop S107 not require ine

aloud. Since no auditory signal is needed, th ological loop is not quite the
same thing as the echoic store\ Baddeley and Lewis (1981) referred to the system
inner ear and the system for speaking.
ee they are not the same, the inner ear and
the inner voice are closely tied to the outer ear and the outer voice. Baddeley pro-
posed that the phonological loop iscapable of holding about 2 sec worth of infor-
mation. Some of the best evidence for such a loop involves memory-span tests for
various kinds ofinformation that people tend torehearse verbally. In a memory-
span test, the subjects hear a series of words and try to repeat them back perfectly.
Baddeley, Thomson, and Buchanan (1975) had subjects try to repeat five words.
They varied the number of syllables in the words from one syllable,

wit, sum, harm, bag, top

to five syllables,

university, opportunity, expository, participation, auditorium

Figure 5.6 shows the recall results. Recall decreased as the number of syllables
increased, falling from about 4.5 words for the one-syllable words to 2.6 words
for the five-syllable words. Figure 5.6 also plots the reading rate—the number
of words of each syllable length that can be said in 1 sec. The reading rate also
decreased as number of syllables increased. Dividing the number recalled by the
reading rate produces a value of about 2 for all syllable lengths. For instance,

> Reading rate ERY)

-e- Span recall

a8 2.00 $
fs
© 3

=
2)
5
CJ

iS) 1.75 &

-
E
:
=
Zz
FIGURE 5.6 Number of words 1.50
recalled (left-hand scale) and mean
reading rate (right-hand scale) for
sequences of five words as a function 25 1.25
of the number of syllables in the ay ila id Dae) Na
words. (From Baddeley, 1986.) Number of syllables

167
 Chapter 5 Transient Memories

subjects could recall about 3.5 of 5 three-syllable words and their reading rate
was about 1.8 words/sec—giving 3.5/1.8, or approximately 2-sec worth of
words. This result implies that subjects can recall what they can rehearse in 2
sec. The capacity of the phonological loop is limited by how far back the inner
ear can remember hearing a word. The research of Darwin, Turvey, and Crowder
(1972) and Glucksberg and Cowan (1970), reviewed earlier, indicated that
echoic memory had a span of 4 or 5 sec. This span is a good bit longer than that
estimated for Baddeley’s phonological loop, but these studies involved rather
different procedures and measures.
The crucial variable is spoken duration, not number of syllables. Words with
long vowels, such as Friday and harpoon, show shorter spans than words with short
vowels, such as wicked and bishop (Baddeley, 1997). There is an interesting rela-
tionship between speech rate and digit span. Digit span is afairly standard test of
memory that appears in intelligence tests. ILinvolves seeing how many single-digit
numbers aperson can repeat back perfectly. Adult speakers ofEnglish have spans
of about seven or eight digits. There is a correlation across languages between span
and articulation length for digits. The articulation rate for Chinese is 265 msec/digit
(Hoosain & Salili, 1988), compared with 321 msec/digit for English and 385
msec/digit for Welsh (Ellis & Hennelly, 1980). Correspondingly, spans are longest
for Chinese (9.9), intermediate for English (6.6), and shortest for Welsh (5.8).
Trying to maintain information in the phonological loop is analogous to
the circus act that involves spinning plates on reeds. The circus performer starts
one plate spinning on a reed, then another on another reed, then another, and
so on. The performer must return to the first before it slows down and falls off,
respin it, and then respin the rest of the plates. There are only so many plates
that the performer can keep spinning. Baddeley proposed the same situation
with respect to working memory. If we try to keep too many items in working
memory, by the time we get back to rehearse the first one, its representation is
no longer available in the phonological store.
This phonological loop seems to involve speech. Conrad (1964) performed
some of the original research establishing this point. He showed that when sub-
jects misremembered something from a memory-span task, they tended to
recall something that sounded similar. In his experiment, subjects were asked to
recall a string of letters, such as HBKLMW. They were much more likely to mis-
recall B as its soundalikeV than as an S, which does not sound similar. Conrad
also found that subjects had a harder time recalling a string of letters that con-
tained a high proportion of rhyming letters (such as BCTHVZ) than a string that
did not (such as HBKLMW). He speculated that this problem arose because of
confusion among the similar-sounding letters.
Further evidence that the phonological loop involves speech is provided
by articulatory suppression techniques, which require the subject to say repeat-
edly an irrelevant word, such as the (Baddeley, Lewis, & Vallar, 1984). Subjects
repeat the word while they listen to a list of words and while they try to write
down the list in recall. Requiring the subject to say the word prevents the
rehearsing of anything else in the phonological loop. When subjects are required

168
 Rehearsal Systems

to engage in such articulatory suppression, their memory spans are shorter. In

contrast, concurrent nonspeech tasks, such as tapping, do not affect span.
Articulatory suppression also reduces the phonological confusions améng let-
ters that Conrad found. When the original list is presented visually and the sub-
ject is engaged in articulatory suppression, the effect of phonological similarity
is eliminated; it is reduced, but not eliminated, when the original presentation
is auditory. With auditory presentation, the subject still has a phonological
encoding of the original presentation, which can be confused on a phonological
basis (Baddeley et al., 1984); with visual presentation, the subjects have neither
the auditory presentation nor the results of their inner speech.
What is the difference between Baddeley’s phonological loop and the
short-term memory of Atkinson and Shiffrin? Although both are transient and
rehearse verbal information, the-presolagical loopisnotaNaltway station 1,
long-term memory. Information does not have to go through the loop to get
into permanent memory, and while it is being rehearsed, no buildup of its per-
manent representation need occur, There is as little relationship between the
phonological loop and what happens to permanent memory as there is between
notes we take on a sheet of paper and things we store in our permanent mem-
ory. Like the paper, the phonological loop can be a valuable system for storing
information; unlike the paper, however, it is a transient representation, and all
records can be lost if rehearsal ceases.
GAENGMAENG Ze i

The phonological loop can maintain about 2 seconds of speech

by implicit verbal rehearsal.

The Visuo-spatial Sketch Pad

Among the other slave rehearsal systems proposed by Baddeley is the visuo-
spatial sketch pad, a system for rehearsing visual or spatial information. He
proposed that people can rehearse material by creating mental images that are
in some ways like the sensory experiences they have when seeing. For instance,
Baddeley (an Englishman) reported that he had difficulty following an American
football game on the radio while driving. He had developed a complex image of
the game, which was interfering with his ability to process the visual informa-
tion required for driving. His imagery process was interfering with his visual
process, suggesting that they are part of the same system.
Figure 5.7 illustrates some material that Baddeley adapted from Brooks
(1967) to study the use of the visuo-spatial sketch pad to store visual informa-
tion. In the spatial condition, subjects heard a series of sentences that they were
to remember. To help them remember the sentences, they were instructed to
imagine placing the objects in a 4 x 4 matrix. Figure 5.7 illustrates how the
matrix could code the information in the spatial sentences. In the nonsense con-
dition, they heard sentences that were similar but could not be coded in a
matrix. Subjects were able to remember about eight of the spatial sentences and

169
 Chapter 5 Transient Memories

Spatial material Nonsense material

In the starting square puta l. In the starting square puta 1.
In the next square to the right puta 2. —_In the next square to the quick put a 2.
In the next square up put a 3. In the next square to the good put a 3.
In the next square to the right puta 4. _In the next square to the quick put a 4.
In the next square down put a 5. In the next square to the bad put a 5.
In the next square down put a 6. In the next square to the bad puta 6.
In the next square to the left put a 7. In the next square to the slow put a 7.
In the next square down put an 8. In the next square to the bad put an 8.

FIGURE 5.7 Example of material used by Baddeley in his study of the visuo-spatial
sketch pad. Source: From A. D. Baddeley, S..Grant, E. Wight, and N. Thomson.
Attention and Performance V, Volume 5. Imagery and Visual Working Memory.
Copyright © 1975 by Academic Press. Reprinted by permission.

only five of the nonsense sentences; this suggests that they were able to use the
image of the 4 x 4 matrix to supplement their memory for the sentences.
In an important elaboration on this basic study, Baddeley, Grant, Wight,
and Thomson (1975) looked at the effect of a concurrent spatial tracking task on
memory for the sentences. The concurrent tracking task involved keeping a sty-
lus in contact with a spot of light that followed a circular track. Subjects had to
remember eight spatial sentences or five nonsense sentences like those in
Figure 5.7. Figure 5.8 shows the total number of errors made in remembering
these sentences when the subjects were and were not simultaneously perform-

Nonsense material
3.0
Spatial material

S©
S 2.0
&
=
=
5
= 1.0

FIGURE 5.8 The influence of concur-

rent tracking on memory span for spa-
tial and nonspatial sequences. (From ee oe
Baddeley et al., 1975.) ; No tracking Tracking
 Rehearsal Systems

ing a spatial tracking task. The error rate was approximately the same for spatial
and nonsense material without a spatial tracking task and was not impaired for
nonsense sentences given a spatial tracking task. The error rate rose dramiatical-
ly when subjects had to perform a spatial tracking task concurrently with mem-
orizing the spatial sentences. This result indicates that the visuo-spatial sketch
pad that supports the memory of the spatial sentences is tapping the same sys-
tem that supports the representation of the tracking task.

The visuo-spatial sketch pad can maintain transient informa-

tion in a spatial organization.

Working Memory and the Central Executive

Figure 5.9 illustrates Baddeley’s overall conception of how these various slave
systems interact. A central executive controls the use of various slave systems,
like the visuo-spatial sketch pad and the phonological loop. The central execu-
tive can put information into any of these slave systems or retrieve information
from the systems. It can also translate information from one system to the other.
Baddeley claimed that the central executive needs its own transient store of
information to make decisions about how to control the slave systems.
Baddeley calls the overall system displayed in Figure 5.9 working memo-
ry. By this he means to denote that it is the system that is holding all the infor-
mation that is currently being operated upon. Baddeley believes that the capac-
ity of all these systems, particularly the central executive, is critical to mental
performance. The different memories in Figure 5.9 are independent. Thus, if one
is using the articulatory loop to rehearse a set of words, it will have little effect
on the performance of a more cognitive task that requires use of the central
executive or spatial sketchpad.
Consider the involvement of these systems in performing a mental multi-
plication task, such as 37 x 28. Try to figure out this product in your head and

Visuo-spatial Central executive Phonological

sketch pad loop

FIGURE 5.9 Baddeley’s theory of working memory where a central executive coor-
dinates a set of slave systems. Source: From A. D. Baddeley. Working memory: Oxford
psychology series No. 11. Copyright © 1986. Reprinted by permission of Oxford
University Press.

171
 ‘Chapter 5 Transient Memories

observe what you do. You might try to hold an image of the multiplication,
which looks like
37
x
28
296
_740
1036
You might verbally rehearse information to help you retain it. Thus, you might
well use both your phonological loop and your visuo-spatial sketch pad to help
you perform the task. But you need to access information that is in neither store.
You have to remember that your task is multiplication; where you are in the
multiplication; and temporary carries, such as 5 from 56. All this information is
held by the central executive and is used to determine the course of solving the
problem and the use of the slave systems. In addition, you would have to
remember information such as 7 x 8 = 56 from permanent memory. Another
function of the central executive is to coordinate all of these memories.

In Baddeley’s theory the central executive and the slave

rehearsal systems constitute working memory.

The Sternberg Paradigm

Just because someone is rehearsing information in working memory does not
mean that they have instantaneous access to the information| Sternbsrg\1969)
introduced a paradigm that became popular for studying speed of access to such
information. His paradigm is now referred to as thd Sternberg paradigm] He
gave subjects a set of digits to hold in memory, such as 4 1 8 5, and then he
asked them whether a particular probe digit was in the set. For this example a
positive probe digit would be 8, and a negative probe would be 6. Sternberg was
interested in the speed with which his subjects could make the judgment as a
function of the size of the set they were holding in memory. Figure 5.10 shows
his results. The straight line is the best-fitting linear function relating judgment
time to the size of the memory set. As the size of the memory set got larger, sub-
jects took longer to make the judgment in the case of both positive probes and
negative probes. He found that each digit added an extra 38 msec to the judg-
ment time. This 38 msec is the slope of the linear function in Figure 5.10.
Sternberg proposed an influential theory to account for these results:
Subjects serially searched through the list of digits they held in memory. If they
found the digit in the list, they responded yes to the probe; otherwise, they
responded no. The more digits in the list, the longer it took to search the list. The
38 msec measured the time to consider one item in the list. The positive probe
might be expected to result in a shallower slope, because the subject could stop
as soon as the target digit was encountered, whereas in the case of a negative

172
 Rehearsal Systems

600— @ Positive
o Negative
oO
eo — Mean
&
oO
£
c
2 500
oO

8
=

®
=

400

il 2 3 4 5 6
Number of items

FIGURE 5.10 Judgment time as a function of number of items in a memory set.

(From Sternberg, 1969.) Source: From J. Antrobus. Cognition and affect. Copyright ©
1970. Published by Little, Brown and Company. Reprinted by permission.

probe the subject would have to exhaust the list. However, Sternberg proposed
that, even in the case of a positive probe, subjects exhausted the list before
responding. He argued that the scanning was taking place very rapidly and that
checking whether to stop (because the probe had been found) would take
longer than just going on to the end. Subjects certainly are not aware of doing
any of this scanning; it is all happening too fast.
Many studies have involved variations of Sternberg’s paradigm, and many
alternative theories have been advanced for his data (see Glass, 1984, for one
review). The functions are frequently not as linear as reported by Sternberg and
have a decided curvilinear trend such that the rate of increase slows down for high
memory set sizes. Even Sternberg’s data contain a hint of this trend, since the
increase for positive probes is largest in going from memory set sizes of one to two.
Some researchers (e.g., James Anderson, 1973) have argued that 38 msec is
too rapid a comparison process to implement neurally. They have therefore advo-
cated a parallel comparison process in which all the digits in the memory set are
compared simultaneously against the probe. A single neural firing takes about 10
msec (see review in Chapter 1); thus, there can be a sequence of only about four
neurons involved in checking an element for a match, which some people con-
sider too few to perform a comparison. Parallel processing models have been
proposed (e.g., Jones & Anderson, 1987) in which all the items in the memory set
are processed at once, but the time to process the items is a function of how
active they are. These models assume that there is a limit on how much activa-
tion can be allocated to the items. The more elements there are in this memory
set, the less active any one item is, and the slower it will be processed.

173
 Chapter 5 Transient Memories

80
e Nonsense syllables
e@ Random forms
60

Geometric shapes @ =
Words
40
® Digits

20
msec/item
time,
comparison
Memory

3.0 4.0 5.0 6.0 7.0 8.0

Memory span

FIGURE 5.11 Memory comparison rate and memory span covary. (From
Cavanagh, 1972). Source: From J. P. Cavanagh. Relation between the immediate mem-
ory span and the memory search rate. Psychological Review, Volume 79. Copyright ©
1972 by the American Psychological Association. Reprinted by permission.

In terms of Baddeley’s theory, the digits are presumably being held in the
phonological loop. Thus, Sternberg’s results are apparently related to the
rehearsal process. Cavanagh (1972) looked at the relationship between memo-
ry span (the maximum number of elements that can be recalled perfectly) for
various types of materials and the slope in a Sternberg task (e.g., the 38-msec
increase shown in Figure 5.10). Figure 5.11 displays his results. Some of the
material is verbal (nonsense syllables, words, letters, digits) and presumably is
rehearsed in the phonological loop, whereas other information is visual (ran-
dom forms, geometric shapes, colors) and presumably is rehearsed in the visuo-
spatial sketch pad. The abscissa (memory span) in Figure 5.11 reveals a wide
variation in how many items can be held in a span, and the ordinate (memory
comparison time) reveals a wide variation in slope in a Sternberg task. The
straight-line function in Figure 5.12 reveals that the two are closely related such
that shorter spans are associated with larger slopes.
One explanation of the results of Cavanagh goes as follows: As Baddeley
argued (see the discussion relating to Figure 5.6), memory span for these items
varies because they vary in the speed at which they can be rehearsed. Compare
digits (memory span of eight) and words (memory span of five) and assume a
phonological loop of 2.0 sec. The digits would be rehearsed at the rate of 2.0 +
8 = 0.25 sec/digit, and the words would be rehearsed at the rate of 2.0 + 5 = 0.40
sec/word. Suppose that the subject was trying to hold a four-item memory set
in memory. In the case of digits, the subject would be able to rehearse the dig-
its every 0.25 x 4 = 1 sec, but the words would be rehearsed every 0.40 x 4 = 1.6
sec. On average, the digits would have been rehearsed more recently and would

174
 Rehearsal Systems

be more active. If time to access these items is a function of how active they are
in the rehearsal system, Cavanagh’s function would be expected, because slow-
er rehearsal would result in less active elements and in shorter memory spans.

As subjects have to hold more items in a rehearsal system,

their rate of access to any item decreases.
SLATED ROEDER ME

Rehearsal Processes in Lower Organisms

Humans apparently are not the only organisms that use rehearsal processes to
maintain information. Considerable research has been performed studying
rehearsal processes in pigeons in what is called delayed match-to-sample_task
(e.g., Grant, 1981; Honig, 1981; Maki, 1984). The typical trial might begin with a
key lit red. After it has been lit for a period of time, say, 5 sec, the key is turned
off, and the pigeon must wait for a period of time. After this retention interval,
two keys are turned on—one red and one green. The pigeon must peck the key
with the same color as the earlier lit key in order to obtain reinforcement. The
initial key that the pigeon must match (the sample) varies from trial to trial.
Pigeons are quite capable of solving this problem, but their accuracy in match-
ing depends on the interval between the sample color and the test. Figure 5.12
presents some data from Grant (1976), showing how the probability of pecking
the correct color falls off with delay. At all points, performance is quite a bit bet-
ter than chance performance, which would be 50 percent.
This retention curve is like the retention curves displayed by humans (e.g,,
Figure 5.5). Humans try to bridge these intervals by rehearsing the material to

100 ;-—

80 |-

ZO

Percent
correct
60

50

ed 0 20 40 60
Retention interval, sec

FIGURE 5.12 The percentage of correct choices (which key to peck) made by
pigeons as a function of delay between the presentation of the information and
opportunity to respond. Source: From D. S. Grant. Learning and Motivation. Volume 7.
Copyright © 1976 by Academic Press. Reprinted by permission.

175
 Chapter 5 Transient Memories

keep it active in working memory. There is evidence that pigeons also engage in
rehearsal during this interval. For instance-humans show-more-rapid forgetting
when they are distracted from rehearsal, and pigeons appear to do the same.
Pigeons show better retention when the factors that might distract them from
rehearsing are reduced, such as when the lights are turned off during the reten-
tion interval (e.g., Grant & Roberts, 1976; Roberts & Grant, 1978). With the
lights out, the pigeons can no longer look around the experimental chamber
and process distracting stimuli.
One mechanism that pigeons use to rehearse this information is body
position; they remain oriented toward the critical key during the interval. This is
a particularly clear case of how an organism can use its body as a transient
memory for information. Postural rehearsal is not the only way in which pigeons
can perform this task, however; they can also perform the task when the posi-
tion of the colored key shifts from study to test and they must remember the
color and not the location.
It further appears that pigeons can be “instructed” whether or not to
rehearse the material (Maki & Hegvik, 1980). This phenomenon was demon-
strated in experiments that were identical to match-to-sample experiments
except that on some trials the pigeons did not get the opportunity to peck the
key after the delay. They were given a cue to tell them whether or not they could
peck. That is, after the pigeons had seen the original sample, a signal told them
whether this was a trial in which they could peck or not peck. These cues are
called remember cues and forget cues. For some pigeons, turning the house
light on served as the remember cue, and its absence as the forget cue; for oth-
ers, the cues were reversed. The pigeons remembered which key to peck quite
well after a remember cue. Occasionally, they were given a surprise test after a
forget cue; on such surprise trials, they were observed to show rapid forgetting.
These results are illustrated in Figure 5.13, which shows some data from
Maki and Hegvik (1980). The percentage of recall in a delayed match-to-sample
experiment is plotted for forget cues and remember cues at short delays (2 sec) and
long delays (7 sec). Right after seeing the sample there was relatively little differ-
ence in performance with forget and remember cues. However, without the moti-
vation to rehearse, the ability to remember where to peck dropped off dramatical-
ly over the 7-sec retention interval (forget cue). Apparently, the pigeons’ ability to
remember what to do was maintained by an active rehearsal process over the inter-
val and they engaged in this rehearsal process only if they expected to be tested.
There is also evidence for a role for rehearsal in classical conditioning.
Wagner’s (1981) SOP theory of conditioning (mentioned in Chapter 2) claims
that immediately after a conditioning event the organism is actively rehearsing
the CS-US pairing and that the success of conditioning depends on this
rehearsal successfully taking place. Evidence for this assertion was demonstrat-
ed in an experiment by Wagner, Rudy, and Whitlow (1973) on eyelid condition-
ing in rabbits. They followed a pairing of a CS and a puff of air by a surprising
event (an unexpected pairing of two other stimuli). They predicted that this
would cause the rabbit to cease rehearsing the CS—US pairing and start rehears-

176
 Rehearsal Systems

1.0

0.8

8
+ 0.6
e
ae Forget cue

FIGURE 5.13 Retention in a delayed

match-to-sample experiment as a function
of whether the pigeon anticipated being
tested or not. (From Maki & Hegvik, 1980.) Delay, sec

ing the new event. Figure 5.14 displays the degree of conditioning as a function
of the interval between the CS-US pairing and the surprising event. The longer
the interval before the surprising event, the more time the rabbits should have
to rehearse the CS—US pairing. As shown, conditioning did increase as the rab-
bits had longer to rehearse the pairing.
Wagener (1978) also suggested that such rehearsal processes might be part
of the explanation of latent inhibition, which causes difficulty for the

80

70

Median
CRs
of
percent
60

310 80 300
Trial-posttrial episode interval, sec

FIGURE 5.14 Amount of conditioning in four different groups of rabbits as a func-

tion of the interval between the conditioning trial and a surprising event. Source:
From A. R. Wagner et al. Rehearsal in animal conditioning. Journal of Experimental
Psychology. Volume 15. Copyright © 1989 by the American Psychological Association.
Reprinted by permission.

£77
 Chapter 5 Transient Memories

Rescorla—Wagner theory. This is the phenomenon that preexposure to a CS with-

out a US makes it more difficult to later condition the CS to the US. Wagner sug-
gested that the preexposure might make the CS so expected that the animal does
not encode the CS when the animal is presented with a US. If it is not encoded
it cannot be rehearsed, and if it is not rehearsed it cannot be conditioned.
i escemaumopeoncccteans oaatrseearsernete

It appears that lower organisms rehearse stimuli to help them

bridge retention intervals and to form associations.
es ea past = ts LMS SESE RR NEES EE IE

The Neural Basis of Working Memory

The fact that nonhuman organisms engage in rehearsal processes has enabled
considerable progress to be made in understanding the neural bases of working
memory. The evidence shows that the frontal cortex plays a major role in work-
ing memory, at least in primates. The frontal cortex shows a major enlargement
from lower mammals, such as the rat, to the monkey, and it shows a propor-
tionately greater development from the monkey to the human. The frontal cor-
tex plays an important role in tasks that can be considered working memory
tasks. The task that has been most studied in this respect is a version of the
delayed-match-to-sample task used with primates, which is illustrated in
Figure 5.15. A monkey is shown an item of food, which is placed in one of two
identical wells (Figure 5.15a). Then the wells are covered, and the monkey is
prevented from looking at the scene for a delay period, typically 10 sec (Figure
5.150). Finally, the monkey is given an opportunity to retrieve the food and must
remember in which well it was hidden (Figure 5.15c). Monkeys with lesions to
the frontal cortex are unable to perform this task (Jacobsen, 1935, 1936). Human
infants are unable to perform successfully in similar tasks until their frontal cor-
tices have matured at about one year of age (Diamond, 1991).
A particular small region of the frontal cortex is involved when the mon-
key must remember where in space the object was placed (Goldman-Rakic,
1988). This area, called area 46, is found on the side of the frontal cortex (see
Figure 5.16). Lesions to this specific area produce deficits in this task. It has been
shown that neurons in this region fire only during the delay period of the task,
as if they are keeping information active during that interval. They are inactive
before and after the delay. Moreover, different neurons in that region seem
tuned to remembering objects in different portions of the visual field
(Funahashi, Bruce, & Goldman-Rakic, 1991).
Goldman-Rakic (1992) examined monkey performance on other tasks
that required maintaining different types of information over the delay interval.
For instance, in one task the monkey had to remember to select a red circle and
not a green square after an interval. A different region of the prefrontal cortex
appeared to be involved in this task. Different neurons in this area fired when
the red circle was being remembered rather than the green square. Goldman-

178
 The Neural Basis of Working Memory

Response

FIGURE 5.15 An example of a delayed memory task. (a) Food is placed in well on
the right and covered; (b) curtain is drawn for delay period; and (c) curtain is lifted
and monkey can lift cover from one of the wells. Source: From P. S. Goldman-Rakic.
Child Development, Volume 58. Development of cortical circuitry and cognitive func-
tion. Copyright © 1987 by the Society for Research in Child Development. Reprinted
by permission.

Rakic speculated that the prefrontal cortex is parceled into many small regions,
each responsible for remembering a different kind of information.
The prefrontal cortex has strong connections with the hippocampus, a
subcortical structure that has a major role in learning (see Chapters 3 and 8). The
different regions of the cortex have appropriate connections to the other more

179
 Chapter 5 Transient Memories

Sey Ses,
7

y- hs
h7
#Lt 4 es

FIGURE 5.16 Lateral view of human and monkey cerebral cortex, with area 46
shaded. Source: From P. S. Goldman-Rakic. Handbook of physiology: the nervous system,
higher functions of the brain. Copyright © 1987 by the American Physiological Society.
Reprinted by permission.

sensory parts of the cortex. For instance, area 46, which serves as spatial work-
ing memory, has connections to the region of the parietal cortex that is respon-
sible for processing spatial information in the world.
serene a ee ASN N

In primates, different small regions of the frontal cortex serve

as eblasies ee for ee types of information.
AenTEEN N EAT SESE
IS SA ECRSRSEN RE RER NESE RENEE
SME LESTIL SEYRET
IIE EIEN IT RITERETENTION

Neural Imaging of Working Memory in Humans

It appears that the prefrontal cortex serves similar working memory roles in
humans as it does for other primates. Although it is not clear that the exact same
regions of prefrontal cortex serve the exact same functions in humans, it does
seem that different areas of the prefrontal cortex serve to maintain different
kinds of information. Research with humans does not use lesion studies or sin-
gle-cell recordings but rather various neural imaging techniques that track
blood flow. Two of these imaging techniques are positron emission tomogra-
phy (PET) and functional magnetic resonance imaging (fMRI). Both tech-
niques count on the fact that there is greater metabolic activity in regions of the
brain that are more active. PET measures a radioactive tracer that will be con-

180
 The Neural Basis of Working Memory

centrated in regions of the brain with greater blood flow. {MRI measures the dif-
ferent magnetic field caused by the fact that there is greater oxygen-rich blood
(which has greater magnetic properties because of the iron in blood) in regions
of greater activity.
Smith and Jonides (1995) used PET to compare the working memory
required for spatial memory and object memory. Figure 5.17a illustrates the
design they used to test spatial memory. In the memory condition, subjects saw
a set of three dots that they had to remember for 3 seconds. Then a circle
appeared, and they had to press a key once or twice to indicate whether the cir-
cle included the dot. This was contrasted with a control”perception” condition
illustrated in Figure 5.17b. In this condition, subjects saw the dots and circle
together and simply had to indicate whether the circle included the dot. Thus,
they did not have to remember the spatial location of the dots. Smith and
Jonides were interested in the differential activation produced by these two
tasks. They found that the memory condition produced greater activation in the

500 msec

200 msec +

3000 msec +

(b)
@
+ 1500 msec

500 msec +

3000 msec

"Perception" Condition
1500 msec

FIGURE 5.17 Schematic drawings of the events on each trial of the spatial memory
and spatial perception conditions. Source: E. Smith and J. Jonides, The Cognitive
Neurosciences, MIT Press. Copyright © 1995. Reprinted by permission.

181
 Chapter 5 Transient Memories

right prefrontal cortex in a region close to area 46 (actually area 47—consult the
human brain in Figure 5.16). Humans differ from other primates in that their
brains are strongly lateralized, with the right hemisphere more involved in spa-
tial and perceptual processing and the left hemisphere involved in more lin-
guistic and symbolic tasks. Thus, it is not surprising that Smith and Jonides
found activation concentration in the right frontal cortex.
Figure 5.18 illustrates the procedure Smith and Jonides used in the object
memory task. In part (a), the experimental memory condition subjects saw two
objects on either side of a cross. After a delay of 3 seconds, they saw one of these
objects and had to press a key once or twice to indicate the object was identical
to one of the original objects. In the “perception” control condition, they were
shown three objects and had to press once or twice to indicate whether the cen-
tral object matched either of the two peripheral objects. In a contrast between the
memory condition and the control condition, Smith and Jonides found that the
memory condition produced more prefrontal activation in the left hemisphere in
area 6 (again consult the human brain in Figure 5.16). This is a region of the pre-

(a)

500 msec

a aa

oa ie
1500 msec

500 msec +

3000 msec P| + [x]

400 msec

"Perception" Condition

1500 msec
FIGURE 5.18 Schematic drawings of the events on each trial of the object memory
and object perception conditions. Source: E. Smith and J. Jonides, The Cognitive
Neurosciences, MIT Press. Copyright © 1995. Reprinted by permission.

182
 Final Reflections

frontal cortex in a region that has been associated with linguistic processing
(Petrides, Alivisatos, Evans, & Meyer, 1993). Smith and Jonides suggest that their
subjects may have been verbally labeling the shapes to help remember them.

When pine sabe Goals et overTE tee rey

activation in different prefrontal areas for different types of
material.

Final Reflections
This chapter has described some of the systems that hold temporary informa-
tion. It is remarkable how many there are. Sensory systems briefly hold the
information they are receiving so that the organism has a chance to encode it
into a permanent memory. Information can also be maintained in various sen-
sory-like buffers by slave rehearsal processes. The information in these various
transient memories is used to guide our information processing. For this reason
they can be collectively referred to as working memory.
It might seem a bit peculiar to speak of lower organisms, such as pigeons
and rabbits, as having expectations and engaging in active rehearsal processes.
However, it is to the advantage of lower organisms as well as humans to keep cer-
tain information available even when it is no longer present in the environment.
Although nonverbal organisms do not have phonological loops for rehearsal,
which often seem the preferred means of rehearsal for humans, they probably do
have other rehearsal systems for maintaining information in an active state when
it is no longer present in the environment. For instance, pigeons orient to the key
they will have to peck and maintain that orientation over a delay. In this case, they
are using their body posture as a slave system to rehearse the information they
have to remember. Although these animals are rehearsing information, as do
humans, this similarity to human behavior does not imply that they are acting in
the same conscious ways in which humans can act.
In all of these working memory tasks, there is some change in neural acti-
vation that encodes the information. The location of this neural activation can
vary from the peripheral sensory systems to the prefrontal cortex. In all cases,
however, when the activation returns to its base level the memory is lost.
Therefore, we called this chapter a study of transient memory. However, pat-
terns of activation can also produce changes in synaptic efficacy which will
result in relatively permanent memories. The next three chapters are concerned
with these relatively permanent memories.
NARS RORRIOR NL VE LRG LLLLLL LODE LEE LN LOL INTELNSSELLE SLE MERTENS IE LIE EEE BERETA TEST TI LL INTELL NEE ELE

Ore. can ee greens in Dranainaa memoriesae

Suite their LUDO BE PEO S
REG SBD YASEIIERAT SORE HES DESERTS EOIN nN NHTSA RN INS YE RN RAND SRE
EEL SIT R TE IES

183
 Chapter 5 Transient Memories

Further Readings
Crowder (1976) is a classic text surveying memory. Baddeley’s textbook (1997)
on human memory provides a good exposition of his theory of working memo-
ry, and his research monograph (1986) also presents his theory. Journals devot-
ed to human memory include the Journal of Language and Memory (formerly the
Journal of Verbal Learning and Verbal Behavior), Memory and Cognition, and the
Journal of Experimental Psychology: Learning, Memory and Cognition. The Scientific
American article by Goldman-Rakic (1992) discusses the role of the prefrontal
cortex in the working memory of primates. Roitblat, Bever, and Terrace (1984)
review research on animal memory and cognition.

184
 Acquisition of Memories

Stages of Memory
The previous chapter dealt with how information comes in and is maintained
in an active state within working memory. These next three chapters discuss the
subsequent course of information through memory.

ocuses on retention—
ormation is maintained in memory. Chapter 8 examines retrieval—
how information is brought out of memory when needed. This organization fol-
lows Melton’s (1963) classic division of memory into these three processes.
It is not possible to study just one of these processes in isolation. Any
experiment involves initial acquisition of material, followed by some minimal
retention interval, followed by a test that requires retrieval of information. Each
chapter focuses on experiments that reflect mainly on one of the processes. As
the chapters review, however, in some cases the interest is in the interaction
between processes—for example, how the way we encode material at study
determines the best retrieval conditions, how different types of memory records
decay at different rates, how learning one type of material can cause us to for-
get another, how different retrieval conditions display different amounts of for-
getting, and so on.
This chapter focuses on two principal issues. First,

ASR eC RR LARGER 2SERS EAE LRA REE 8S OS ANA HU EN RG OS

The memory proees can bedivided ciatean eesti singe,

a retention stage, and a retrieval stage.
acne net eee ANREP TLS MARAE EE SESE IN
SER ANI RNB ORIG ENROLL
IIE OLATER NETIC

185
 CHAPTER 6 Acquisition of Memories

Practice and Trace Strength

According to the proverb, practice makes perfect eae
shasibesn comceengsaitir en cnoeaAe PREAMPS COnsi CeraStn
ple paired-associate task. In one experiment of mine (Anderson, 1981) subjects
were presented with 20 paired associates, such as the pair dog—3. Subjects were
asked to learn the pairs so that they could recall 3 when prompted with dog.
They were given seven opportunities to study the list of paired associates. Figure
6. fa Shows laguratlunegg eclhdcease cheree tase ae
‘tice. Subjects started out failing to recall about 47 percent of the items and
ended up failing to recall only about 5 percent of the items. Figure 6.1b shows
the time it took subjects to recall the correct responses;
Thus, even after subjects had reached the point of recall-
ing the paired associates successfully, ice i i

ere are other ways in which subjects show improved memory with fur-
ther practice after they are able to recall the memory. As Chapter 1 discussed,

nani
enn ona
ings measure, which involved looking at how much
S>binghaus used a sav-
faster the list was relearned
the next day after varying amounts of practice. As Figure 1.2 showed,

It is often assumed thatlearning curves like those shown in Figure 1.2 or

6.1 reflect the growth in the hat encodes the memory. Earlier
chapters used a concept of strength in explaining conditioning phenomena. For

0.5 jeg

0.4
2.0

8o 0.3 8
_
3S 32 1.8
a £
[e)
a 0.2 2
rai =

} 1.6
OM

0 1.4
2 4 6 2 4 6
Practice trials Practice trials ;
(a) (b)
FIGURE 6.1 (a) Probability of recall and (b) time to recall paired associates as a
function of amount of practice. (From Anderson, 1981.)

186
 Practice and Trace Strength

instance,
Those earlier chapters noted that researchers often
thought of strength of conditioning in terms of the strength
ofsynaptic connec-
tions. For instance, Chapter 2 discussed thelearning
delta rule, a neural embod-
—imentvof*the Rescorla-Wagner theory. The concept of strength is often given

A fair amount of research has studied how memory improves with mas-
sive amounts of practice.

ensitive-+ ce none eer a

& Anderson, 1985) 2 ubjects mentee memory for sentences for 25 days for
a day. During this time, they practiced just 15 sentences, such as

The doctor hated the lawyer.

The radical touched the debutante.
The sailor shot the barber.

A recognition memory test was used to test their memory for these sentences.
After having memorized the sentences, subjects were required to discriminate
them from sentences that they had not studied but that were made up of the
same words. Examples of such foil sentences are

The doctor touched the barber.

The radical shot the lawyer.

Subjects had to press one button if a test sentence had been studied and anoth-
er if it had not; the speed with which they were able to make this recognition
judgment was measured. Subjects spent 25 days practicing these judgments and
hence practicing the sentences.
improvement was rapid over the ini-
tial days, but the rate of improvement slowed down with the amount ofpractice.
asserts eee ames cAS CEN SRS YS SSPE RO
EONS EEECO SELDEN SOUL SSSA eB A RE ARSE CRN SHALES HCN EOOI SPIN ESN

Memories continue to increase in strength with practice even

after recall is perfect.

The Power Law of Learning

Interestingly, learning curves like those depicted in Figure 6.2a all have a simi-
lar mathematical form. This form is revealed if, rather than plotting time to
respond as a function of days of practice,
atural logarithms
are used throughout this section, although other types of logarithms could be

187
 CHAPTER 6 Acquisition of Memories

0.5 Log sec = .34 — .24 log days

time,
Recognition
sec

Days of practice (1) (5) ae

(a) A Logdays |OQ — Sta \
Tower Funchon (b) ene)
FIGURE 6.2 (a) Time to recognize sentences as a function of number of days of
practice; (b) log-log transformation of data in (a) to reveal power function. (From
Pirolli & Anderson, 1985.) <<

used just as well.) The logarithm compresses the differences among larger num-
bers. Some examples of log days for certain of the numbers of days in Figure 6.2
are as follows:

Days Loc Days

i 0.00
5 1.61
10 2.30
15 pagl
20 3.00
25 3.22

The difference between 25 days and 20 days is much smaller on the logarithmic
scale than is the difference between 5 days and 1 day.
Figure 6.2b shows the data from Figure 6.2a replotted on a log-log scale.
Consider how the data point for day 1-is plotted. The latency on day 1 is 1.61
sec. Log 1 = 0 and log 1.61 = .47. Therefore this point is plotted with coordinates
of 0 and .47. Similarly, all the other points are replotted in Figure 6.2b on the
transformed scale. Since the logarithm of a value less than 1 is negative, many
of the log latencies are negative. The untransformed times and days that corre-
spond to the transformed logarithm values are given in parentheses.
Figure 6.2b Fexeaiasainesr Vincente onsis Se,
»
Practice. That is, the points fall very close to the straight line plotted in the fig-

188
 Practice and Trace Strength

ure. If T is used to denote the time in seconds and P the amount of practice in
days, this linear relationship is described by the following function:
log T = .34 — .24 log P
The value .34 is where the function crosses the y-axis (i.e., when log P = 0), and
-.24 is the slope of the line. When this equation is transformed back into the
original scales of time and practice, it becomes a power function,!
a1 .40.P 28

—— this aes is - SRR curve drawn in Figure 6.2a. The

(measured in terms of response time
and a number of other measures
. The straight-line function in the Pere of metre
6.2b becomes a curvilinear function in the original scale of Figure 6.2a. This

aiiontpmartise . The fact that almost all learning functions are power functions
has been called th ewell osenbloom, 1981).
Newell and Rosenbloom (1981), following up on the work of Lewisene
brou ght this ubiquitous law of learning to the attention os:the field.

igure 6.3 showssome data

from} Blackbur (1936), who studied the effect of practicing addition problems
for 10,000 trials
On this graph and some others in this book, the original numbers
(those given in parentheses in Figure 6.2b), rather than the logarithms of these
numbers, are plotted on the logarithmic scale. (When original numbers are plot-
ted on a log scale, it is so noted on the figure.)Blackburn’s data show that the

‘shown in-Figure 6.2. Since its identification by Newell and Rosenbloom, the
power law has attracted a great deal of attention in psychology, and researchers
have tried to understand why learning should take the same form in all experi-
ments (e.g., Anderson, 1982; Lewis, 1978; Logan, 1988; MacKay, 1982; Shrager,
Hogg, & Huberman, 1988).
Figures 6.2 and 6.3 look at the decrease in time with practice. Often error
rates also improve according to power functions. Figure 6.4 replots the data
from Figure 6.1 in terms of log errors against log trials of practice. The approxi-
mately linear relationship implies that error rates decrease as a power function
of time. Not all dependent measures show this power function relationship, but
many, like performance time and error rate, do. Such power functions are dis-
tinguished by the property of negative acceleration; that is, each unit of prac-

1 Taking the exponential of both sides of the equation yields

log P= 94=-24logP or T=e*P-* or T=140P™.

189
 CHAPTER 6 Acquisition of Memories

(log
Time,
scale)
sec

1 2 5 TORO) 50° 100 200 500 1000 10,000

Problem number (log scale)

FIGURE 6.3 The improvement in adding two numbers as a function of practice.

Data are plotted separately for two subjects. (Plot by Crossman, 1959, using data
from Blackburn, 1936.) Source: Figure 6.4 from E.R.RW. Crossman. Ergonomics,
Volume 2. A theory of the acquisition of speed-skill. Copyright © 1976 by Taylor &
Francis Ltd. Reprinted by permission.

tice produces a smaller and smaller improvement in performance. Thus, the

power law is a law of diminishing returns. If these measures are interpreted to
reflect the underlying strength of the record, then these power functions imply
that practice will always increase the strength of the record, though by smaller
and smaller amounts. -

Memory strength increases as a power function of practice and is

Re in oh eh measures elHoestime or error rates.
SLUMSit a AR EEE YA DESIR IM ESTAREETA SISOS IE EIN OEE SESH INS RA
PCE Dn RIO INSP SEB EER RE ROSAS

Log error = —0.44 — 1.34 log trial

0.50

0:10

of
Probability
scale)
(log
error

FIGURE 6.4 Data from Figure

6.2a replotted as log errors against 1 2 5 10
log trials. Trial (log scale)

190
 Practice and Trace Strength

Repetition and Conditioning

What is the relationship between the practice curves in these human memory
experiments and the conditioning curves given in earlier chapters? It is likely that
practice in a conditioning experiment is controlling more than just the strength
of a memory trace. That is, the more often the US follows the CS, the greater the
evidence for a causal contingency between the two events. As we noted in
Chapters 2 through 4, conditioning seems to reflect an inference of causal rela-
tionship or contingency. This contrasts with practice curves that reflect frequen-
cy or contiguity. Despite this difference, conditioning curves sometimes have
approximately the same shape as they do in human learning research. Figure 6.5
shows the time it took rats deprived of food for 2 or 22 hr to run a T-maze for
food (Hillman, Hunter, & Kimble, 1953). Although the rats deprived for 22 hr ran
the maze more rapidly, both groups showed improvements approximating a lin-
ear function on a log-log scale. Thus, the speed of responding in rats is also
approximately a power function of the amount of practice.
In other cases, the learning curves in the conditioning literature show a
substantial difference from those of human memory functions. As discussed
earlier with respect to classical conditioning (see Figure 1.4), animals often show
little increase in conditioning, then a large increase, and then little. Figure 6.6
plots an average learning function from Brogden (1949). He was concerned with
the conditioning of an avoidance response in dogs. Figure 6.6 plots the percent-
age of avoidance in each unit of training. There was relatively little learning from
the first time period to the second, then rapid learning for a number of periods,
and finally slower learning. These conditioning functions differ from human
memory curves in this initial phase of slow learning. These functions are often
referred to as S-shaped curves (see Culler & Girden, 1951, for a review of such
conditioning functions).

e 2 hr of food deprivation
Log sec = 2.22 — 0.22 log trials

= 22 hr of food deprivation
Log sec = 1.96 — 0.26 log trials

8.0

C)oO
2 6.0
2g
°°
D

FIGURE 6.5 Time to run a maze as 4.0

a function of the number of prior tri-
als of training and the number of
hours of water deprivation. (From 1 2 arr 6 10 14
Hillman et al., 1953.) Training trials (log scale)

191 .
 CHAPTER 6 Acquisition of Memories

100

90

» 80
(ni
Z RID
°
; ©
o 60
S
2 50 i
FIGURE 6.6 Acquisition § 40
of an avoidance response. 5
Source: From W. J. Brogden. ae
Journal of Comparative and 20
Physiological — Psychology,
Volume 42. Copyright © i”
1949 by the American 0
Psychological Association. Thetis 2 3 a4 Ota Gas! 1S Sun BLO! ctl
Reprinted by permission. 11ths to criterion of acquisition

To understand why conditioning curves are different from learning curves,

see Figure 5.1, which shows a schematic of a conditioning experiment. The con-
ditioning experiment involves an additional induction phase in which the
organism must figure out the causal relationship. The association cannot begin
strengthening until it is identified. Conditioning curves for individual organisms
often reveal some discrete point in time at which the organisms catch on. Figure
6.7 illustrates this situation with respect to the conditioning of four dogs in an
experiment described by Culler and Girden (1951). Animals learned to retract
their paws when they heard a sound that warned that a shock would come in 2
sec. The figure reveals that for long periods of time the individual animals did
not recognize the connection. After some trials, the response began to strength-
en. Different animals began to show conditioning after different trials: in Figure
6.7a after trial 50, in Figures 6.7b and 6.7c after trial 100, and in Figure 6.7d after
trial 75. Although the individual curves are quite variable after conditioning
begins, at least some of them then show the property of power functions—rapid
initial gains followed by slower gains.

Conditioning functions are often S-shaped, because condition-

ing requires an induction process before associative learning
can begin.

Long-Term Potentiation and the Environment

Chapter 3 discussed long-term potentiation (LTP), which occurs in the hippocam-
pus and cortical areas. LTP is a form of neural learning that seems to be related to
behavioral measures of learning. When pathways are stimulated with high-fre-

192
 Practice and Trace Strength

100 100

80 80

© 60 & 60
8 8
& 40 a 40

20 20

) 0
25 50975 1007 125 150 25 SON 75) LOO 25) 150)
Trials Trials

100 100

80 80

© 60 & 60
S 8
& 40 © 40

20 20

(6) 0
50 100 150 200 SOS 1000 150\.6200)5 250
Trials Trials

FIGURE 6.7 Conditioning of a sound-shock association in four dogs. (From Culler

& Girden, 1953.)

quency electrical current, the sensitivity of cells along that pathway to further stim-
ulation is increased. Barnes (1979) studied this phenomenon in rats, measuring the
percentage of increase in excitatory postsynaptic potential (EPSP) over its initial
value.* She stimulated the hippocampus of the rats each day for 11 successive days
and measured the growth in LIP in terms of the percentage of increase. Figure 6.8a
displays the results, plotting percent of change against day of practice. There
appears to be a diminishing increase with the amount of practice. To determine
whether there is a power function, Figure 6.8b plots log percentage of change
against log practice. The relationship appears approximately linear; thus, it appears
that neural activation changes with practice just as do behavioral measures.

? As discussed in Chapter 1, as the dendrite and cell body of a neuron become more
excited, the difference in electric potential between the outside and inside of the cell
decreases. EPSP refers to the size of this change.

193
 CHAPTER 6 Acquisition of Memories

3.8
50)j-

3.6

WwWro)

change
Percent Log oS fo)
change
percent

20 ee
2.8

2 4 6 8 10 0
Days of practice Log days of practice
(b)
(a)

FIGURE 6.8 Growth in LTP as a function of number of days of practice: (a) in nor-
mal scale; (b) in log-log scale. (From Barnes, 1979.)

Note that the activation measure in Figure 6.8a increases more and more
slowly, whereas the performance measures, such as errors (Figure 6.1a) and time
(Figure 6.1b), decrease more and more slowly. These performance measures are
assumed to be inverse reflections of the growth in strength that is happening
internally. As the strength of the record increases (reflected in Figure 6.8), the
performance measures get better (which means shorter times and fewer errors).
It has been suggested (Anderson & Schooler, 1991) that memory (both its
behavioral and its neural expressions) displays properties such as the power law
of learning because these properties reflect an optimal response to the environ-
ment. A very general characterization of the learning functions reviewed in this
chapter is that the more something is encountered, the more available it is in the
future. A conjecture about the environment is that the more an organism has
needed to remember something, the more likely it is that it will need to remem-
ber that thing again. Thus, memory can be viewed as making more available
information that is more likely to be needed.
This viewpoint raises the question of whether the power function dis-
played in the learning behavior of subjects mirrors a similar power function in
the environment. Anderson and Schooler (1991) studied the patterns by which
information tends to repeat in a number of different environments, including
parental speech to children, electronic mail messages, and newspaper head-
lines. In the case of parental speech to children, the frequency of various words
in parental utterances was examined. For example, if the word ball occurred 8
times in the last 100 utterances, what is the probability that it will occur on the
101st utterance? Figure 6.9 shows the relationship between the log frequency

194
 Practice and Trace Strength

35
Log probability = e
-4.96 + 1.05 log frequency

scale)
(log
probability
Future
FIGURE 6.9 Relationship between log
frequency with which a word has
occurred in the last 100 utterances and
log probability of it occurring in the
101st utterance. (From Anderson & 1 5 20 50
Schooler, 1991.) Past frequency (log scale)

that a word has appeared in the last 100 utterances and the log probability that
it will appear in the 101st utterance. The linear relationship is striking, implying
a power function. The functional form of memory seems to mirror the function-
al form in the environment. Similar patterns were found in mail messages and
newspaper articles. The more often one has had to retrieve the meaning of a
word, the more likely one will have to do so again.
In summary, three types of dependent measures change as power func-
tions of practice: behavioral measures, such as error rate and latency; percent-
age of change in LIP; and probability of repetition in the environment. One pro-
posal (Anderson, 1993) is that neural activation, which controls behavior,
reflects the probability of an item occurring in the environment; thus, the neur-
al processes are designed to adapt behavior to the statistical properties of the
environment.

Long-term potentiation and probability of repetition in the envi-

ronment are both power functions offrequency of exposure.
#8

Significance of a Power Function

What is the significance of the fact that the behavioral measures are power func-
tions of practice? These power functions capture the result that learning is neg-
atively accelerated and that each learning experience produces less of a benefit
in terms of a performance measure. However, many potential mathematical
functions have this property. The fact that learning is best captured by a power
function is an important and hard-won generalization that emerged in psychol-
ogy only after the extensive analysis of Newell and Rosenbloom.

195
 CHAPTER 6 Acquisition of Memories

1!

1.0

0.85

0.6

Performance

0:4 Power function

0.2

FIGURE 6.10 Comparison of expo- 0 20 40, 60 80 100

nential and power functions. Practice

The natural hypothesis that formerly dominated psychology was that per-
formance was an exponential function of practice. If a time were an exponen-
functionof practice (P), it would take the form of a fraction raised to the
tial
amount of practice. For instance, one possible exponential function is:
T=10x.8?
(where the 10 and the .8 are just arbitrary choices chosen for illustration). This
exponential hypothesis was natural because it implied that with each trial, per-
formance improved by a constant fraction. In the example given, each trial takes
80 percent, or .8, of the time of the previous trial. If the previous trial took 10 sec,
a .8 reduction would yield 8 sec, for a 2-sec gain for the current trial. When the
time is 1 sec, a .8 reduction results in a .2-sec gain. Thus, the function was neg-
atively accelerated (showed less and less decrease) because the base time
(which was being reduced by .8 each trial) was going down.?
Figure 6.10 compares an exponential function with a power function. Both
functions are negatively accelerated, but the power function is much more so.
Only after collection and analysis of a great deal of data was it established that
learning functions are better described by power functions than by exponential
functions (Newell & Rosenbloom, 1981). The rate of improvement with practice
turns out to be even slower than formerly believed.
Psychologists find it exciting to be able to find simple functions, like the
power function, that fit such a wide range of data. It is similar to discovering
laws of the sort that physicists find (and psychologists have always had physics
envy). However, the power function that describes the learning curve does not
have the same status as the equations in physics books. The data from learning

3 An exponential learning function is predicted by the Rescorla-Wagner theory,

which was introduced in Chapter 2 and will be discussed at many points in this book.

196
 Elaborateness of Processing

experiments are not perfectly accommodated by power functions, although they

are accommodated much better by power functions than by some other forms,
such as exponential functions. There are systematic deviations, and the nature
of these deviations changes from experiment to experiment. The power function
is only a good approximation to the learning functions.
Psychologists are not content with the discovery that the learning curves are
approximated by power functions; they also try to understand why the data have
this approximate form. As noted earlier, a number of theories of the mechanisms
behind this power function have been proposed; frequently, these theories predict
only that the learning curve should be a good approximation to a power function.*
These competing theories have not been resolved. My own view is that the expla-
nation of the power function is to be found in the neural processes that underlie
associative learning and that these neural learning processes have evolved to their
current form because of the statistical properties of the environment. This phe-
nomenon is just one token of the adaptiveness of learning.
By way of summary, the following Strength Equation describes how the
strength of a memory trace increases as a function of how frequently the mem-
ory record has been practiced.
Strength = Practice? (Strength Equation)
where b is the exponent that controls how fast strength increases.°
LLANE TOE IE NILES SELENE LIES EID,

The growth of strength with practice better approximates a

power function than other common negatively accelerated
functions.
DEEL LIESLEE LIES LEEDS SENSEI ERNIE I ENE TEE COE NELSON EISELE I EDO EE EIR TLE IITLEDER ENE MEMES SELES,

Elaborateness of Processing
The preceding analysis implied that items accrue strength at a rate that depends
only on how much the item has been studied. However, a great deal of evidence
also indicates that how an item is studied can have an enormous impact on how
much the subject remembers of the item. In some cases people can process
items many times without much benefit. Chapter 5 discussed Neisser’s descrip-
tion of Professor Sanford, who remembered little from thousands of repetitions
of his prayers.

4 Heathcote and Mewhort (1995, 1998), for instance, have argued that the learning
function is really an exponential with an asymptote I: 1 + B f? where f is a fraction less
than 1. However, a simple power function, AP~’ gives a good approximation to such
a function.
5 The exponent b is positive because strength is assumed to increase, producing
decreases in measures of performance difficulty, such as time and errors.

197
 CHAPTER 6 Acquisition of Memories

Bekerian and Baddeley (1980) reported a study of the efforts BBC radio
made to get its listeners to learn the new wavelength at which it was going to
broadcast. The BBC saturated the airwaves with announcements of the new
wavelength. Bekerian and Baddeley tested subjects who had been exposed to
the information for over 25 times a day for many weeks. Less than a quarter of
the people they surveyed had learned the new wavelength. Sheer exposure to
material is not enough to guarantee learning. Chapter 5 also discussed Craik
and Lockhart’s (1972) depth-of-processing theory, which held that memory
improves only if information is processed at a deep level. This chapter describes
a variant of the depth-of-processing proposal, which emphasizes that what is
important is how elaborately we process the to-be-remembered information.

Memory for material is improved the more elaborately it is

processed.

The Generation Effect

Although the way in which information is processed is critical to what is recalled,
there has been some question (even by Craik) as to whether the concept of depth

For instance, if we
be-remembered word, we recall less than
if we think about what words ar . The latter activity forces us to get at
the . Although the depth of processing seems to have an
effect, several studies have also shown that superior memory is achieved by active-
ly processing material in a way that does not involve its meaning.

f a target word t eg e€ . For example, the subjects might

have been asked to generate a synonym of sea that began with 0 (i.e., ocean) or a
rhyme of save that began with c (e., cave). In the read condition, subjects read pairs
of words that exemplified these relationships, for example, sea—ocean and save-—cave.
Then subjects were tested for their recognition of the second word in these pairs
(e.g., ocean and cave). Figure 6.11 displays the results. Subjects showed an advan-
_tagethsynonymsoveshynes.Ths
effect is similar to other depth of processing
effects in that it shows the advantage of semantic processing. However,

° There was a flurry of debate about whether the original Slamecka and Graf result
was due to the fact that it involved a within-list design and subjects gave differential
practice to the generate items. Slamecka and Katsaiti (1987) claimed it was an artifact,
but others (e.g., Begg, Snider, Foley, & Goddard, 1989; Burns, 1992; Hirshman & Bjork,
1988; McDaniel, Waddill, & Einstein, 1988) have found effects in between-list designs.

198
 Elaborateness of Processing

1.0
Generate

Read
0.8
s
a 0.6
2
6
2
= 0.4
S
2
oO

~ 0.2
FIGURE 6.11 Probability of recognition as a
function of type of elaboration and whether it
was generated or read. (From Slamecka & Graf, 0.0
1978, experiment 2.) Synonym Rhyme

generate
for themselves.
@) Burns (1992) and Hirshman and Bjork (1988) have provided an explanation
of the advantage of generative processing in experiments in which subjects must
learn paired associates, such as the Slamecka and Graf experiment. According to
these authors,
or instance, generating cave as a rhyme of save
guarantees that subjects encode that both stimulus and response in the save—cave
pair are in fact rhymes, whereas they might not attend to this common feature if
they just read the pair of words. If the rhyme relationship is encoded at study, the
G) subject can use it to retrieve the response at recall.

shaaco ihlegen The looked at subjects’ ability to remember

simple paired associates, such as dog—bike. In one condition, subjects were
shown sentences that linked the two words such as,“The dog chased the bike.”
Such subjects did better than subjects who just studied the paired associates. A
— third group of subjects, who had to generate such sentences, did even better.

These activities provide

subjects with additional retrieval routes. With respect to the Bobrow and Bower
experiment, if the subjects are given a cue of dog for recall and have studied the
sentence, they can recall bike directly, but if this fails, they also can recall chase

at gives the added advantage when the subject has to generate the
verb? The subject is likely to generate a verb that is easy for that subject to recall
and from which it is easy for that subject to recall bike. When subjects can choose
their own verbs, they generate items that make for particularly good retrieval

199
 CHAPTER 6 Acquisition of Memories

routes. To summarize, elaborations help because they can provide additional

retrieval routes. Generation helps further because the retrieval routes generated
are particularly effective for the subject. The Hirshman and Bjork explanation of
the generation effect in the Slamecka and Graf paradigm similarly emphasizes
facilitating the connection between stimulus and response.
SRR He ASP OEIC ELH SSSLLTOE

Elaborative processing and self-generation improve memory

through the enhancement of retrieval routes.
sean
DG SSS LLANES

Differences Between Elaboration and Strength

In the context of this chapter, elaborative processing might be viewed as
increasing the strength of the memory trace, but this is not the best interpreta-
tion.
f they cannot remember the original
memory trace, they may be able to use the other traces to retrieve what they
want. Thus, if a subject has studied dog—bike without elaborations, the subject
can only remember the bike given dog by retrieving the trace of this paired asso-
ciate. If that trace cannot be retrieved, the subject is out of luck. However, if the
subject has generated the elaboration, “The dog chased the boy on the bike,”
there is an additional route from which to retrieve bike given the cue of dog.

get memory and do not increase the strength ofaparticular memory record. The
experiment looked at subjects’ ability to learn some little-known information
about some famous people. In one condition subjects studied just a single fact:

Mozart made a long journey from Munich to Paris.

In another condition subjects learned two additional facts that were causally
related to the target fact:

Mozart made a long journey from Munich to Paris.

plus
Mozart wanted to leave Munich to avoid a romantic entanglement.
Mozart was intrigued by musical developments coming out of Paris.

The additional sentences were experimenter-provided elaborations designed to

boost memory for the target sentence. :
Subjects were tested for their ability to remember the target facts at a
week’s delay. They were presented with names such as Mozart and were asked
to recall the target sentence. The results are displayed in Table 6.1 in terms of per-
centage of recall of the target sentence and time to recognize the target sentence.

200
 Elaborateness of Processing

TABLE 6.1 Effect of Elaborating Sentences on Percentage of Recall Versus

Recognition Time

Single Sentence Multiple Elaborating

Alone Sentences

-orating sentences were presented at study. However, when subjects were tested
for their ability to recognize the target sentences, there was no latency advantage.
for the elaborated sentences. If anything, they were slower, Earlier, this chapter
noted th The target sentences were
not any stronger when studied with elaborations; however, these elaborations
offered subjects another way to retrieve the target sentence if they were unable
to recall the sentences directly.

BSCR HIER a

rv Strength involves the encoding of a specific memory record,

. whereas elaboration creates additional records to help retrieve
the original record.

Incidental Versus Intentional Learning

One principle of memory implicit in the discussion to this point is that it does not
really matter whether or not the person intends to learn the material. What is crit-
ical for memory is how the material is processed. This surprising truth about
human memory has been demonstrated numerous times. For instance
looked at memory for individual words. Subjects performed one of
two processing tasks: they rated the words on the basis of how pleasant they were
or whether they had an e or a g.The former task required the subject to think about
the meaning of the words and should have led to more elaborative processing and
better memory than the latter task. Half the subjects in each processing condition
were informed that they would be tested on their memory for the words; the other
subjects were led to believe that the rating task was the prim ose of the
experiment.
All groups were asked to recall the list of 24 words. Table 6.2 displays the
tion recalled by the four groups of subjects

other experiment that controlled processing and examined the effect of

intention to learn was performed byiMiandler (1967). He asked two groups of

201
 CHAPTER 6 Acquisition of Memories

TABLE 6.2 Percent Recall as a Function of Orienting Task and Intention to Learn
Orienting Task

Learning-purpose Rate Check

Condition Pleasantness Letters

Incidental - 68 39

Intentional 69 43

subjects to sort a set of words on cards (one word per card) into categories. One
group of subjects was told that they would be tested later for their memory of
the words; a second group was not so warned.

oth studies illustrate a gen-

eral result:
requently, subjects intending to
learn are in fact able to recall more material, but only because they engage in
processing that is more conducive to learning the material. For reviews of the
lack of effect of intention to learn, see Nelson (1976) and Postman (1974).
These results add to the research reviewed in Chapter 4 tha

The failure of motivation to affect learning can be seen from the viewpoint that
people cannot learn what is important to them or from the viewpoint that they
cannot avoid learning things that are unimportant. The latter interpretation
seems more appropriate. People are best viewed as storing in memory every-
thing they attend to, whether or not they want to remember it. To understand
why there are memory failures, the next two chapters examine the processes of
forgetting and retrieval.
ete penceneonemesicemsas aes

The memory system encodes a person’s experiences whether or

not there is any intention to learn.

Implications for Education

The educational implications of the research thus far reviewed in the chapter are
both clear and important. The learning functions establish the obvious: practice
Sige cane The results on depth of processing, elaborative processing, and
self-generation establish something that is not so obvious:

These mode-of-processing results can be directly applied to trying to

remember factual information, such as what is communicated in this textbook, A
number of successful study skill programs have been built on this insight, includ-

202
 The Structure of Memory

ing the SQ3R method (Robinson, 1961) and the PQ4R method (Thomas &
Robinson, 1972). PQ4R stands for preview, question, read, reflect, recite, and
review. The rea osed to preview the text to be read (e.g., a chapter of a
book) and identify the main’sections, make up questions relevant to each section,
and read the section trying to answer the questions and reflect on the text. After
each section, the reader is supposed to recite the material from that section. At the
end of the text, the reader should review the main points of the text. This method
and methods like it require the reader to attack a text aggressively, making up
questions about it and thinking about its implications. This is just the sort of elab-
orative and generative processing that has proved effective in the laboratory.
One experimental study (Frase, 1975) on the effectiveness of such pro-
cessing techniques for study involved a collaborative learning technique. The
text was divided into sections, and two people read it together. For each section,
one person read with a mind to making up questions about that section, and the
other read and then had to answer those questions. Then roles were switched
for each section. A control group of subjects who just read the material without
doing anything special got 50 percent correct on the posttest. For the experi-
mental subjects, the posttest questions could be divided roughly in half accord-
ing to whether they were anticipated by the subjects’ questions. The experimen-
tal subjects also got 50 percent correct of the unanticipated posttest questions
but 69 percent correct of the anticipated posttest questions.
Chapter 11 further reviews research establishing the importance of such
elaborative study skills. That chapter argues that this may be the most important
educational application of research on human memory.

Question making and answering are effective ways to elabora-

tively process textbook material.

The Structure of Memory

Having now discussed the factors that determine how well a memory is retained,
in the remainder of this chapter we focus on the issue of how these memories
are encoded. We begin with a discussion of the current understanding of how
these memories are encoded in the brain. Then we discuss more abstract infor-
mation-processing ideas about how knowledge is represented in memory.

The Brain and Memory

Previous chapters have already discussed evidence Be two of the critical
neural structures for memory. First there is the hippocampal formatic
3 discussed the key role that the hippocampal formation in adie con-
figural memories in animals. (Chapter 8 will describe how damage to the hip-
pocampal area and related areas can produce amnesias in humans.)

203
 CHAPTER 6 Acquisition of Memories

Second, there are the various areas of prefrontal cortex that serve as work-
ing memory to hold information over delays, as discussed in Chapter 5.
However, it is generally believed that neither the hippocampus nor these pre-
frontal areas are where memories are actually stored. The OC peem
Se
2 AN QMA®

frontal we g ry areas hold transient 1 ries)Rather, it is believed

that permanent memories are stored cortically but in more posterior (toward the
back of the head) regi ese three areas (hippocampus, pre-
frontal cortex, posterior cortex) have many paths of neural interconnection. For
instance, area 46, which»was critical in primate spatial memory, has connections
both to the hippocampal area and to areas of the parietal cortex that is respon-
sible for processing the spatial aspects of perceptual information.
It appears that different types of memories are stored in the different cor-
tical areas responsible for processing that type of information. Visual memories
are stored in regions responsible for visual processing, and linguistic memories
are stored in areas responsible for language processing. For instance, Sakai and

The inferior temporal cortex is a region of the brain

involved in shape recognition. Sakai and Miyashita (1991) had monkeys learn to
recognize paired associates of various shapes such as those displayed in Figure
6.12. The monkeys would be shown one shape and then had to recognize
whether a second shape was the one paired with it. Recording from single neu-
rons in the inferior temporal cortex, Sakai and Miyashita found neurons that
would respond when particular visual shapes were presented. These cells did not
fire before training but now had come to encode the particular paired associate.

FIGURE 6.12 Examples of the shape paired associates used in the experiment of
Sakai and Miyashita (1991). Source: [Redrawn from “Neural Organization for the
Long-Term Memory of Paired Associates” by Sakai, K. & Miyashita, Y., Nature,
354,
pages 152-155, (1991). Reproduced with permission from Nature, 1991, Macmillan
Magazines Limited. ]

204
 The Structure of Memory

“Two types of learning are required in creating memories.

One type

arr (1971) originated a proposal

about these two kinds of learning which has been elaborated quite a bit recent-
ly (e.g., Gluck & Myers, 1993; McClelland, McNaughton, & O’Reilly, 1995). This
proposal is that the hippocampus is responsible for creating new configurations
of elements in the cortex and that the cortex can then gradually strengthen
these memories. McClelland et al. (1995) propose that the hippocampus replays
these new configurations of elements over and over again and gradually trains
the cortex on the new associations. They suggest that this replaying may take
place during sleep, and they cite evidence that bursts of neural activation arise
in the hippocampus and propagate to the cortex.

Permanent memories are stored in the different cortical areas

that process the different types of information.

An Abstract Representation of Permanent Memory

illustrates some memory records encoding some of

ual might know. These records include the arithmetical fact that 4 * 7 = 28, the
note that the person is currently engaged in a mental multiplication, the loca-
tion of the person’s car in aRe lot, and the fact that cars ae four
wheels. é Ss com 5 a configuration¢ nen c

| am multiplying My car is in the Cars have 4

numbers parking lot wheels

Memory records

FIGURE 6.13 Memory records in permanent memory and their connections to var-
ious cues.

205
 CHAPTER 6 Acquisition of Memories

a multiplication fact. As we just noted, the hippocampus is hypothesized to be

critical in creating such configurations.
CIiCilel

o prompt recall for that memory. When cues are attended to, either because
they are in the environment or because the person is thinking about them, they
activate their associated records in permanent memory. A person has access to
these memory records to the degree to which they are active. The state of being
active
AC or available
C QO] Ta IDLE is a transient property of memory records: Therefore, the cur-
rently active information in permanent memory is, in effect, another transient
memory (the topic of Chapter 5) and an important part of the person’s overall
working memory. By focusing on different information in the environment or by
rehearsing different information, a person can make different parts of perma-
nent memory active. Thus, on hearing the word mother, information about a per-
son’s mother becomes active; similarly, on viewing a picture of the Eiffel Tower,
information about Paris and France becomes active. In these examples mother
and the Eiffel Tower are the cues, and the retrieved memories are the records.
A number of different theories in cognitive psychology embody this basic
analysis of memory as records and cues, although they use different language to
describe it. The language used in this chapter is similar to that of my ACT theory
(Anderson, 1983a, 1993). Another well-known theory, SAM (Gillund & Shiffrin,
1984; Raaijmakers & Shiffrin, 1981), speaks of images (memory records) becom-
ing more or less familiar (active) as a function of the cues in the context.
Connectionist theories, reviewed in Chapter 2, are theories of neural information
processing and have similar assumptions, except that the cues and records are
referred to as neural units. Thus, the system in Figure 6.13 reflects an emerging
consensus in the field, although there is not yet consensus on the language that
describes it. Much of the evidence for this system is presented in subsequent
chapters. The following section reviews some of the data indicating how the pre-
sentation of appropriate cues can facilitate access to memory records.
Ce ee

Permanent memory consists of a set of records that can be acti-

vated when their associated cues are in the environment or in
rehearsal systems.

Priming

only some memory records. For instance, the word cow is associated to memo-
ty records concerning cow-like things, such as giving milk. A fair amount of
work has been done documenting the effects of such associations in what is
called a lexical decision task. These are experiments in which subjects are given
strings ofletters, such aS milk or milc, and have to decide which are words
and
which are not. The experiments manipulate whether subjects also see an asso-
ciate of a target word, for example, cow. In the experimental condition, subjects

206
 The Structure of Memory

might first see cow and then have to judge whether milk is a word, whereas in
the control condition they might first see cot and then milk. Using the analysis
of record activation just given, we might expect the word cow to activate the rep-
resentation of the memory record that encodes the spelling of milk (since cow is
associated to milk), and so the lexical decision could be made faster in that case.
There have been numerous experiments showing just this (e.g., Meyer &
Schvaneveldt, 1971; Neely, 1977). In one cxperiment(Balatarand pieh11986)

The research of Balota and Lorch is just one

example of the many lines of research showing that subjects’ access to informa-
tion is primed when associates of that information are present. In a priming
experiment of a very different scale, Kaplan (1989) gave subjects a set of puzzles .
to take home and solve. One problem was

What goes up a chimney down but cannot go down a chimney up?

One of his subjects was stuck on this particular problem and could not solve it
for days. Kaplan arranged to have the subject receive what appeared to be a mis-
taken phone call during which the caller asked if she had left her umbrella in
the subject’s office. Shortly after that, the subject reported the solution to this
problem—umbrella—even though he was unaware that als ils phone ae was
james to his eeitie the answer. Over

enn cone pa eaemorefaites ae associated

UAC is in cusPOC

Chunking
_

Research indicates that relatively little information can be stored in any one
memory record, and therefore many records have to be used to store large
amounts of information. People tend to store about three elements in any
record. Much of the research demonstrating limited record size has involved
subjects trying to remember a series of letters, for example, DRQNSLWCF.
Subjects break oe series into a nisms of pale amencce of— Miller secs

7 The 48-msec difference might seem small, but it reflects about a 10-percent differ-
ence, and these experiments achieve very accurate measurements.

207
 CHAPTER 6 Acquisition of Memories

Subjects rehearse DRQ-pause-NSL-pause-WCF, with the

pauses reflecting where they have broken up the series into chunks. When they
recall the series later, they are observed to break it up into the same units.
The inference is that subjects store the letters in terms of these small
chunks—D, R, and Q together; N, S, and L together; W, C, and F together. How
can experimenters determine that these elements are actually stored together?
Maybe subjects’ rehearsal behavior just reflects habits of their speech and noth-
ing about underlying memory representation

sigs wena aah ACRE ag AHe presented his subjects with random
strings of letters but encouraged them to use a particular chunk structure by plac-
ing spaces between the letters. So, he might have presented his subjects with

DY JHOQ GW
&

ored together wit ubjects tended to recall a chunk, like JHQ, in an

If subjects could recall J, there was only a(10 percentychance
that they would fail to recall the subsequent H. In contrast, if the subjects recalled
the Y, there was still a\80 percent change of failing to recall the J in the next
chunk. Thus,

Such all-or-none performance would be expected if the letters were stored

together in a memory record. If the record could be accessed, all the elements in
the record could be retrieved. Thus, if the subject could retrieve one element
from the record, the others should be retrievable. On the other hand, retrieving
elements from one record implies nothing about whether elements from anoth-
er record can be retrieved.
It seems that society implicitly recognizes the limited size of memory
records in how it breaks up numbers. In the United States, phone numbers are
divided into an area code of three digits, followed by a prefix of three digits and
a final group of four digits. Similarly, U.S. social security numbers are grouped
into chunks of three digits, two digits, and four digits. Almost everywhere num-
ber strings are conventionally broken into chunk sizes of about three.

For instance, Qcsparen Hardy, and Hirtle (1989) looked at memory for the 28
objects illustrated in Figure 6.14. Subjects saw these objects laid out in a 20 x 22 ft
area. By looking at their patterns of recall, McNamara et al. found evidence that
subjects organized these objects into groups. Specifically, subjects tended to recall
certain groups of objects always together. Figure 6.14 illustrates the groups used
by one subject. These objects can be chunked in many different ways, and differ-
ent subjects came up with different organizations. Subjects gave evidence by other
behavioral measures that el from within a chunk were organized together
in memory. For instance M@leNamaig sal.(1985)used a priming paradigm such

208
 The Structure of Memory

scissors

ring
e button £

rubber band
e

eraser
e

envelope
e

flashlight
e

cufflink

thimble
e
light bulb
e

e cigarettes

FIGURE 6.14 A layout used in the experiment by McNamara et al. (1989). The cir-
cles indicate hierarchical organization imposed by one subject on this array. Circles
enclose objects in the same chunk or enclose chunks in the same higher order units.
Source: From T. P. McNamara, J. K. Hardy, and S. C. Hirtle. Journal of Experimental
Psychology: Learning, Memory and Cognition, Volume 15. Copyright © 1989 by the
American Psychological Association. Reprinted by permission.

as the paradigms discussed in the previous subsection. Subjects saw the name of
one object (prime) from the array, followed 150 msec later by the name of anoth-
er object (target). Subjects had to judge whether the second object was from the
array. They compared target objects that either were or were not from the same
chunk as the prime. In both cases, the targets were the same physical distance
from the primes .
Agetne

In addition to letter strings and scenes, similar effects have been

strated in memory for prose (e.g., Rumelhart, 1975; Thorndyke, 1977) and mem-
ory for chess positions (Chase & Simon, 1973). In all instances, a subject takes a
complex stimulus and breaks it into a set of smaller units. Each unit has about
three elements in it.

209
 CHAPTER 6 Acquisition of Memories
25/3 IEEE TEP EN TNL ELE SLL SEERA LOLS OL EULSERED LEELA LEE LE LLLILIES DL LLLLELVERL LLL

Memories are stored in records or chunks of about three ele-

ments.
PORNO CAAA ERLE ESE LLL EL OER OME ES ELEMIS ELAR LES AEEL ANE LD DEL LEED.

Representation of Knowledge
As we have noted, different types of information appear to be stored in differ-
ent cortical areas. It has been pror hat these di : vforr
stinctive forms of encoding referred to as (remor Tes One popular
theory of memory is thé dual- theory
code (Bower, 1972; Paivio, 1971), which»

red in_picturelike
Cil memory records. An experiment by Santa (1977) provides a
nice illustration of the different properties of these two memory codes.
Subjects studied an array of three geometric figures. Then the subjects had
to judge whether a test array contained the same elements as the original stim-
ulus. Some of the possible test arrays are illustrated in Figure 6.15a. In many of

Test
arrays

Identical Same elements, Different elements, Different elements,

linear configuration same configuration linear configuration
(a)

Study | Triangle Circle

Square

Test | Triangle Circle Triangle Circle

arrays Triangle Circle Square. Triangle Circle Arrow
Square Arrow

Identical Same words, Different words, Different words,

linear configuration same configuration linear configuration
(b)
FIGURE 6.15 Procedure in Santa’s experiment (1977): Subjects studied an array of ele-
ments and then had to judge whether a test array had the same elements. (a) Geometric
condition; (b) verbal condition. Source: From J. R. Anderson. The architecture ofcognition,
Copyright © 1983 by Harvard University Press. Reprinted by permission.

aa Ve
 The Structure of Memory

FIGURE 6.16 Time to judge an Verbal

array in Santa’s experiment as a
function of type of material and
the relationship between study
array and test array. Source: From time,
Reaction
sec Geometric
J. R. Anderson. The architecture of
cognition. Copyright © 1983 by
Harvard University Press. Re- Same Linear
printed by permission. ~ configuration configuration

the test arrays, the same elements were arranged in a different order from that
of the study array. Subjects were to give affirmative responses to the array even
though the order was different. Santa was interested in how Eas it took sub-

The stimuli in Figure 6.15b offer a striking contrast. These are the same
stimuli as in Figure 6.15a except that the subjects studied words instead of geo-
metric figures. Santa reasoned that subjects would encode these stimuli verbal-
ly rather than visually. In this case, subjects might actually do better when test-
ed with the linear array because this encodes the items in the order in which
they would occur if read left to right and top down, which is the standard read-
ing order. As Figure 6.16 shows, Santa.was zs ht.(hihiensianedackinalssnen gem

esemera! caret tae had mauler

consequences for how sub ects could later access the information.

Teesis pat
a autrade ae eta en, ie asneer ee Mer
verbal ees
SSE RNASE RES EN AEONLR NERO
BNI ETS LE ELLE SEES LEE LEED LEE SLL LMLESAER SIMA

Memory for Visual Information

Different encodings of information appear to be remembered with_different
degrees of success.
ialisparticularly wellremembered. Shephard. (1967)-compared recognition mem-
ory for magazine pictures with recognition memory for sentences. After studying
a series of pictures or sentences, subjects were asked to identify the picture or sen-
tence cob had seh when it wasDepts Hal a plete oh or sen-

211
 CHAPTER 6 Acquisition of Memories

FIGURE 6.17 Examples of the snow crystal photographs used as stimuli.

It is not the case, however, that people have photographic memories and
can remember whatever they see Goldstein and Chance 0 970) studied memo-
ry for two types of visual materia faces and snowflakes)Figure 6.17 shows
examples of the snowflakes. The individual snowflakes are quite distinct and
seem at least as discriminable perceptually as individual faces. Subjects studied
either 14 faces or 14 snowflakes and for their recognition 48 hr later.
, (Subjects were able to recognizd74 percent of the facesYbut sriradimereenreah
‘Qe snoMATERES Thusgood memory for pictorial material reflects not only how
/\) distinctive the stimuli are but also how well the subject can encode the materi-
\ al. Subjects are capable of attributing more significance to the features that sep-
'arate faces than those —— a
Memory for pictorial information seems to be determined by the ability of
subjects to place a meaningful interpretation on the picture. An amusing
demonstration of this fact was performed by Bo Karlin, and Dueck (1975).
They had_subjects study what are called droodles, such as those illustrated in
Figure 6.18. One group of subjects saw both just the droodles, without tae
ex lanatory labels. Another jects saw the droodles and the explana-
tory labels, Subjects displayed better recognition memory for the pictures when
they wereaccompanied
by the labels, presumably because the labels enabled
the subjects to elaborate. the pictures.

.4
“FIGURE 6.18 Droodles: A midget
playing a trombone in a telephone
booth. Panel b—An early bird who
caught a very strong worm. Source:
From G. H. Bower, M. B. Karlin, and
A. Dueck. Memory & Cognition, -—
Volume 3, pp. 216-220. Copyright ©
1975. Reprinted by permission of Hones)
Psychonomic Society, Inc. (a) (b)

212
 The Structure of Memory

Target Token distractor Type distractor

*K FIGURE 6.19 Pictures used by Mandler and Ritchey (1977). Source: From J. M.
Mandler and N. S. Johnson. Journal of Experimental Psychology: Human Learning and
Memory, Volume 2. Copyright © 1976 by the American Psychological Association.
Reprinted by permission.

S UT eittatl p C C 5 UL C me
U :&

identifywhich picture the had originally sta Ted TE e distractors sould Beeither
token_distractors or type distractors. A token d orthanged some semanti-
ea (relevant detail for example, in Figure 6 heteacher’s dress is changed.
shanged a detail that might bejaan an interpretation of
ne picture; in Figure 6.19, the material on the board is changed (trom a geog-
raphy lesson to perhaps an art lesson). Subjects falsely recognized 40 percent of

_distractors. The Its indicate that subjects remembered a meanin inter-

pretation of the picture rather than the physical details of the picture. When a~
i REE eons setnienniensivddemianl i a
 CHAPTER 6 Acquisition of Memories

LUA ALLL RATE ANRC AEN SHEE REE ELTA LOL IER MIELE LEE NEILL EDEL LLL LD ELE LENE

Poni fe Tnieanety ar Poe Faetheir interpretations

ofeee Hise
SESSCL EEDA EELeeDIELS SEMEL ie LSE LEBEL
LEE DV e a
ELBNE ve
MDL EE
SES LLLELLES LED ad
ALS Ize
SLOOP SCRE

Effects of Imagery
As a reflection of our high level of memory for visual material, we can improve
visual
memory for verbal material by constru images of the material to be
cting
memorized. In one experiment, Bower (1972) had subjects try to commit to

As indicated in this example, the visual images can be bizarre (a dog riding
a bicycle) or more common (a dog chasing a child on a bicycle). Another dimen-
sion of difference is that images can involve objects interacting (as is true of both
the previous examples) or objects not interacting (e.g., a dog standing beside a
bicycle). Wollen, Weber, and Lowry (1972) studied the relative contribution of
bizarreness and interactive quality to memory. They gave their subjects pictures
to help them learn paired associates, such as piano-cigar. Figure 6.20 illustrates
the four kinds of pictures they used to realize the four possible combinations of
bizarreness and interaction. Table 6.3 shows the levels of recall in the four condi-
tions. There is a large effect of interaction but not of bizarreness; the effect of
interaction is probably related to the effects of elaboration reviewed earlier. The
interactive images promoted elaborative encoding that helped later recall.
Studies of whether the bizarreness of the imagery improves memory have
had varied results. McDaniel and Einstein (1986) and Hirshman, Whelley, and
Palu (1989) found cases in which subjects displayed better memory for bizarre
images. McDaniel and Einstein related this finding to whether a within-subject
or between-subject design was used. The original Wollen et al. study used a
between-subject design, in which some subjects studied all bizarre images and
other subjects studied all nonbizarre images. In this kind of design, there is usu-
ally not an effect of bizarreness. In a within-subject design, half the items the
subjects study involve bizarre images and half do not. In these designs bizarre
images usually have advantages. MeDaniel/and Einstein argued thatinawith:

Interactionjnluiged
esthicly cones
W wenersontehieeswivtehrte
BE are more ape pes mak are aad
SESS ESSIEN TA IONS SAE SE Rte re
ee sssseenusreNty

214
 The Structure of Memory

Piano Cigar Piano

Noninteracting, nonbizarre Noninteracting, bizarre

Piano Cigar Piano Cigar

Interacting, nonbizarre Interacting, bizarre

FIGURE 6.20 Examples of pictures used to associate piano and cigar in the Wollen
et al. (1972) study of image bizarreness. Source: From K. A. Wollen, A. Weber, and D.
H. Lowry. Cognitive Psychology, Volume 3. ae i:
1972
ae Academic Press. ol
Reprinted by permission. Bikes! ‘
mone) ie
Meaningful a eae
Qo p ovdS
As reviewed earlier, when subjects remember a os eytend to> Femiember
See. rerpretanon of the picture-Mlila o-evidence- n pe

rN 1967). Ore of my experiments

TABLE 6.3 Mean Percentage of Recall in Wollen, Weber, and Lowry (1972)

Bizarre Nonbizarre

Interacting

Noninteracting

215
 CHAPTER 6 Acquisition of Memories

(Anderson, 1974b) illustrates this abstract memory for sentences. Subjects were
asked to remember sentences such as:

The missionary shot the painter.

This sentence is in the active voice. Other sentences were in the passive voice,
for example:

The lawyer was kicked by the doctor.

Later, subjects were asked to recognize which sentences they had heard. If they
had studied the first sentence in the example, they were tested with any of the
following sentences:

1. The missionary shot the painter.

2. The painter was shot by the missionary.
3. The painter shot the missionary.
4. The missionary was shot by the painter.

Subjects had no difficulty in rejecting sentences such as 3 and 4 that had a dif-
Ae ae nee
ferent meaning.
ad ce

t a certain level this phenomenon seems

to be highly adaptive. What we need to remember from a linguistic message
(such as this textbook) is seldom the exact words but rather the meaning of the
text. Similarly, when we come upon a scene involving a set of people, it is much
more likely that we will later need to know what they were doing rather than
what they were wearing.
PORERL SIS RIS EDEL LEENA PEERED ELOISE LOIRE LESLIE DEL SLED LESLIE RGAE IESE TIN ETDS
SE LT RS EDEL LESS NONE

Subjects tend to remember the meaning of a text rather than

its exact wording.
AERO A ORES SEA ENLARGE TELLUS RL IIS BOT
BI ROE BR OA TSHR ERSTEPUSHES
BREE OES OESTRONE ET RRS SA RM DIEU ACTON SIRS

Differential Decay of Sensory and Semantic Information

the latter explanation. Forgetting is discussed more thoroughly in the next chap-
ter, but this section reviews two studies showing that different types of materi-
al have different rates of forgetting.
*

216
 The Structure of Memory

FIGURE 6.21 Pictures displayed in one orientation and the reverse. Source: From M.
A. Gernsbacher. Cognitive Psychology, Vol. 17. Copyright © 1985. Reproduced by per-
mission of Academic Press.

Germsbacher 2985) had subjects study pictures such as those shown in

Figure 6.21. Subjects studied one picture and then had to identify it when pre-
sented with a forced choice between that picture and the other. Because these
pictures are mirror images of each other, this is an example of making a discrim-
ination that is not-critical to the meaning of the pi At a 10-sec delay, sub-.

The experiment described in the preceding section (Anderson, 1974b) also

looked at subjects’ ability to remember which of the following two, sentences
SS hey had studied ettmbwed
\ 2sThe missionary shot the painter.
x 2. The missionary was shot by the painter.

centat
a2-mindela:Thus,subjectsmaintained nearlyperfectmemory.Subjects
ability to discriminate between the following pair of sentences was also tested:
‘ 1. The missionary shot the painter.
é 2. The painter was shot by the missionary. x

217
 CHAPTER 6 Acquisition of Memories
AEE NSN RAYS LOLA ENROI ELEN AN MERE
meme

Ree for semantic pent en issine reine than is

saggigUe er sensory yo
SRNL EILEEN ELON PENSE ESERIES A ITT Ee SR EUSP ISOS ISSN ELSI TS AILMENTS LEIE LIER IEE EME AS EIDE

Kintsch’s Propositional Theory of Text Memory

This apparent abstraction from detail to meaning raises an interesting question.
If the information is not represented in terms of the detail of the original sen-
sory experience, what isthe form of the more meaningful structure in which it
is represented? Many researchers believe that memory for meaning is repre-
aesTae developed one propositional theory of memory for sen-
HEINCES! rteSeo aa meaning Hes is stored in memo

because he has extended it to Ree eee (e.g., Kintsch &

van Dijk, 1978). This application is important because it indicates that principles
studied with simple laboratory materials may extend to the complex material we
have to remember in our everyday lives.
Consider how a propositional analysis applies to the following sentence:
1. Lincoln, who was president during a bitter war, freed the slaves.
The information conveyed in this sentence can be communicated by the follow-
ing simpler sentences.
2. Lincoln was president during a war.
3. The war was bitter.
4. Lincoln freed the slaves.
Each of these simpler sentences corresponds to a primitive proposition. This —

ee any of these smaller sentences were false,

‘the larger sentence would be false. However, ass sentences ences are not
eu shea rather, they-reflect-the A iona essence,

for making this distinction between the sentence and the proposition is the evi-
dence that subjects tend to remember thegeneral meaning_of even such simple
sentences rat the exact wording of the sentences.
Kintsch (1974) advanced a proposal for how to represent the propositions
that convey the meaning of these sentences. According to Kintsch, the proposi-
tions underlying sentences 2 through 4 are represented by the following list
structures:

218
 The Structure of Memory

2a. (president, Lincoln, war)

3a. (bitter, war)
4a. (free, Lincoln, slaves) :
Each of these lists begins with tne elation.preside iterfren—tha orga-
nizes the proposition. The relation isfollowed by the other key terms from the
proposition. These terms, which are typically mouns, are calle
Thus, the arguments for president are the president (Lincoln) and the ti
The same propositional structure would be produced no matter how the sen-
tence was stated. For instance, 2a could be the propositional representation for
any of the following sentences as well as for 2:

During the war, Lincoln was president.

The president during the war was Lincoln.
It was Lincoln who was president during the war.

Critical in the Kintsch representation is the issue of what the relations are and
what the arguments are. The list notation in 2a through 4a is only a convenient
way of denoting that. Thepropositional lists in2athrough 4acan beregarded.
One line of evidence = i ro oe analysis comes from looking at
patterns of recall for sentences. Gaal IRS! 97) had subjects study
sentences such as:

The doctor who hated the lawyer liked the captain.

The two underlying propositions are:

(hate, doctor, lawyer)

(like, doctor, captain)

Subjects who could recall the noun lawyer were more likely also to recall the
verb hated, which appeared in the same proposition, than the verb liked, whic

Lo LES ESSENSERR ISLET RRR BNE ADO EI

Kintsch proposed that semantic information is stored in propo-

sitional records.
eR

The Bransford and Franks Study

The fundamental claim of a propositional analysis is that when we hear a sen-
tence, such as 1, the propositions, s 2a through 4a, are an important part
of what we commit to memory. (1971) performed an

219
 CHAPTER 6 Acquisition of Memories

set of sen-
interesting demonstration of this fact. They had subjects study a
tences, including:

The ants ate the sweet jelly which was on the table.
The rock rolled down the mountain and crushed the tiny hut.
The ants in the kitchen ate the jelly.
The rock rolled down the mountain and crushed the hut beside the
woods.
The ants in the kitchen ate the jelly which was on the table.
The tiny hut was beside the woods.
The jelly was sweet.

These sentences are all composed fronytwo sets of four propositions) One set of
four propositions can be represented as follows:

(eat, ants, jelly)

(sweet, jelly)
(on, jelly, table)
(in, ants, kitchen)

The other set of four propositions can be represented as follows:

(roll down, rock, mountain)

(crush, rock, hut)
(beside, hut, wood)
(tiny, hut)

Bransford and Franks then presented subjects with various sentences and asked
them to judge whether that exact sentence had been studied. The following
illustrates the three types of sentences they used:

OLD: The ants in the kitchen ate the jelly.

NEW: The ants ate the sweet jelly.
NONCASE: The ants ate the jelly beside the woods.

The first sentence has been studied, whereas the other two had not. The second
sentence, however, was composed of the same propositions that the subject had
studied. Subjects showed almostnoabilityto discriminate theNEW sentences)
from the OLD. These sentences seemed equally familiar, and the subjects sim-
ply could not remember which sentences they had studied. eee
: hicl Lf iene

220
 Memory Representation in Other Species

When the subjects heard the sentences, they abstracted the propositions
and remembered them. They did not keep track of what propositions they had
seen in what sentences. Indeed

The ants in the kitchen ate the sweet jelly that was on the table.

When we hear a complex linguistic communication, we tend to

store it in terms of its primitive propositions.

Memory Representation
in Other Species
It is interesting to ask whether other species think about the world in the same
way that we do. A way to formulate this question more specifically is to ask
whether their memory representations are similar to that of humans. Therevis
evidence that other animals encode information into serial and spatial patterns _
(e.g., Roberts, 1984; Terrace, 1984). Pigeons are capable of learning specific
sequences of pecks just as humans can remember sequences of letters (Terrace,
1984), and, as reviewed in Chapter 3, rats can learn the location of objects in
space as well as humans can (Roberts, 1984). In addition, evidence
there is that »
qiinalsaneeapeblsnehis
enlidyingdhom caning etpicnessitependentinpitiog
exact physical details. For instance, Chapter 3 reviewed the evidence that pigeons
prerearaeae’
can instances of natural categories, such as“ tree.” In this section, we will
discuss memory representations in two common laboratory species—one far
from us phylogenetically (pigeons) and the other close (other primates).

Sequential Memory of Pigeons

A number of paradigms have been used to study memory for serial order in
pigeons. C s pigeons to discrin ers
ofstimuli fromvothers. Forinstance clencenetalcsscinnt Doddvanaaanes®
(1980) trained ae recognize that two sequences of visual stimuli sig-
naled that a peck would be reinforced with food. These two sequences might be
a green color, followed by ared, followed by a horizontal bar and the sequence
red—green—vertical. Any other combination of colors and bars indicated that a
peck would not be-reinforced.The-pigeons
ee
learned to peck only- at the
sequences, indicating that they
hey had discriminated t ese{roy al othe
Anotl . a ]

@ g., Straub, smn Bever, & Terrace, 1979),‘Theyss

be shown a set of four keys (e.g., red, white, blue, green) and be reinforced for

221
 CHAPTER 6 Acquisition of Memories

pecking in a particular sequence of colors no matter what the spatial arrangement

of the keys. Again,

sequences. More impressive are the experiments that ae the pigeons to

learn a different sequence each trial. For instance 76) presented a
sequence of X-shaped forms occurring either to the Teff or to the right of the
pigeon. The first form might be presented to the left of the pigeon and the next
two to the right. The pigeon was then presented with a color. Red signaled that
the pigeon would be reinforced for pecking where the first form occurred (in the
example given this would be pecking left), a blue indicated to peck where the
second form appeared (right in the example), and a white signaled to peck
where the third form appeared (right in the example). Pigeons learned to per-
form this task with considerable accuracy.
Although both humans and pigeons show an ability to remember serial
order, it does not follow that they remember it in the same way. For one thing,
much of human sequential memory is verbally mediated, for example, through
rehearsing a sequence of letters. Even human memory for nonverbal stimuli,
such as movement, shows properties that are not true of pigeon memory.
Humans show an accuracy gradient in reproducing a sequence of three ele-
ments such that they are most accurate for the first, next most accurate for the
last, and least accurate for the middle. In contrast, invall'the pigeon»paradigms

Pigeons as well as humans must be able to encode the sequential struc-

ture of the environment in order to adapt to it. Therefore, it should come as no
surprise that pigeons can perform well at these tasks. However, pigeon brains
are different from human brains; hence, it is also not surprising that the sequen-
tial structure is encoded differently.

Pigeons can remember the sequential structure of events and

act on those memories.

Representational Structures in Primates

Primates are much closer to humans on the phylogenic scale, and there is evi-
dence that they have quite similar representationsTerrace (e.g., Terrace, 1998;
Terrace, Jaswal, Brannon & Chen, 1996; Terrace & McGonigle, 1994) has com-
i e finds mon-
keys and humans to be both quite similar and different from pigeons. He uses a
paradigm in which monkeys have to touch pictures in the order that they are pre-
sente

to chunk long lists into sublists, aswe noted earlier for humans. Monkeys’also
show a capacity for transitive inference that pigeons do not. Transitive inference

222.
 Final Reflections

can be tested in a paradigm in which pigeons have been trained to peck a

sequence of five keys or monkeys have been trained to press a series of five pic-
tures. Let’s refer to the elements in the sequence abstractly as A, B, C, D, and E.
What happens when just the B and D are presented as possible responses (the
others missing)? Monkeys or humans can apply the serial order they have learned
and will produce B and then D. In contrast, pigeons are unable to generalize their
experience to this new test and make their pecks on B and D in random order.
In general, it seems that primates are capable of many of the sophistications
in human representation of knowledge. This is particularly the case for chim-
panzees who are the closest to humans.
(e.g., Gardner & Gardner, 1969; Premack, 1976;
Terrace, Pettito, Sanders, & Bever, 1979). Chapter 10 discusses these experiments
in detail. This section examines one of the consequences of such language train-
ing for representational capacity. Premack (1976) taught chimpanzees to order

igure 6.22 shows some of the test

material used with the chimps. The animals were to choose the instrument that
transformed the first object into the second. So, in the first case shown in Figure
6.22, a knife transforms a whole apple into pieces of an apple. In the second
case, water converts a dry sponge into a wet ares Solving such problems
ancadino

} S arr eee i ion. There

is no
common element to the answers to these problems at a superficial level. The»

ti isti remack and

Premack noted that only language-trained chimpanzees were able to solve such
problems. This observation raises an interesting question about the role of lan-
guage in enabling propositional representations.
Perhaps it is the use of lan-

stant cE EERE NENTS IRAE AEM AREER sien sane ae

Price appear to ee Entiat ae Pipesitional repre-

re
sentation similar to bare of gies
ase S2 YEON OSS EROS TREE BESIDES LE SEAS ENN RIE REE RELI LENS SALES IO NILES ER LEONE DEEL SELLE DEE ILA BEES LEE LE EET

Final Reflections
This chapter has addressed the topic of acquisition—that is, how information is
stored in memory. One view of memory is that everything we attend to is orga-
nized in terms of small, chunk-like records and stored away in memory. It does

223
 CHAPTER 6 Acquisition of Memories

FIGURE 6.22 The tests are for cutting, wetting, and marking, respectively. The
missing item is the instrument. In the lower right is the chimpanzee Elizabeth enthu-
siastically cutting an apple.

224
 Further Readings

not matter what the memory is about or whether we want to remember it at all;
we are always storing away the things we attend to. The animal learning litera-
ture often creates a picture of rather little learned, but this is because it is so dif-
ficult to get nonhumans to display what they have learned. With human sub-
jects, there is evidence of relatively rapid rates of learning.

till, people clearly fail to

recall much of what they experience.
e topics of the following two
chapters.
Whether it is because of differential success at initial encoding or differ-
ential success at retention and retrieval, what we do at the time of study can
have a large impact on how much we remember. reviewed
This chapter three
classes
zn ofencoding factors: amount ofstudy, which results indifferential strength
sqnishombioumaisisianesnitsctonaeicitioneimmntorn
iin enim cr< is something
generally adaptive in how memory responds. We are more likely to need to
remember things the more frequently we encounter them and the more elabo-
rately we process them. We are more likely to need to remember the gist of our
experiences than the details. It might even be argued that we are more likely to
need to recall visual experiences than verbal ones, since visual information nec-
essarily comes from our direct experience, whereas linguistic information can
communicate experiences we may never encounter.

We may store everything we attend to, and memory failure

may be due to forgetting and retrieval factors rather than to
acquisition factors.

Further Readings
In addition to the sources mentioned in the previous chapter, several works pre-
sent the topics of this chapter in greater detail. The power law of learning is
examined by Anderson and Schooler (1991) and Newell and Rosenbloom
(1981). The book edited by Cermak and Craik (1978) presents a number of dis-
cussions on depth and elaborativeness of encoding. My cognitive psychology
text (Anderson, 2000) discusses the issue of knowledge representation at greater
length. Kintsch (1998) is a good source of ideas about representation and lin-
-guistic processing. Roitblatt (1987) presents extensive discussion of knowledge
‘representation in nonhuman animals. Finally, Premack and Premack (1983)
offer an interesting description of the cognitive capabilities of chimpanzees.

225
 Retention ofMemories

Overview
I don’t know how many times I have had the same conversation with people on
planes, on ski slopes, at baseball games, and wherever strangers meet. The con-
versation follows the same script with slight variations:

Stranger: What do you do?

Me: I ama research psychologist.
Stranger: What do you research?
Me: Human memory.
Stranger: I have a terrible memory. I am always forgetting things. What
can I do about that?

This dialogue reflects the fact that many people’s most salient experience of their
memories is that they are aware of having known many things that they can no
longer remember. This chapter is concerned with what underlies forgetting.
Along the way the chapter notes two possibilities that might surprise people:

1. Forgetting may not be such a bad thing.

2. People may not really forget anything.

However, there is some territory to cover before expanding on these two curiosi-
ties.
There are three basic hypotheses about what causes forgetting. T

- bloc - 1
. eR . est
pri
- a ae ee :
cue-hy pothesis asserts that@

ditionally, research on forgetting

has tried to distinguish among these theories, but each factor contributes to the

226
 The Retention Function

overall forgetting process. This chapter examines the first two hypotheses in
some detail, postponing full discussion of the third hypothesis until Sees 8.
CUMS RE Nena REN SS MPN SSW seis anit

Three shame offorgetting are the es ae eae the inter-

ference hypothesis, and the Seilate-cueSEU EH EatE
cmeeaaRRIN SASS LESSEE NTL RENN RARITIES UT RY I LOI i Re

The Retention Function

Memories seem to fade with the passage of time. Many experiments have stud-
ied memory loss as a function of time. Chapter 1 discussed some ofthe earliest
studies hemeno by Ebbinghaus on the ret

LLL

All retention functions show this same basic form. Initially, forgetting is
etention functions dif-
fer with respect to the scales on which they display these basic phenomena.
Other measures, in addition to Ebbinghaus’s savings measure, include proba-
bility of recall or time to recall memories that can be recalled. These measures

60
4.2
Log P = 3.862 - 0.126 log H

4.0
50

2 E§
2)
ny oo 3.8

8n ~
z 40 5 3.6
3) ®
oD Qa
a 00
5 3.4
30
ae

20 3.0
0 200 400 600 800 -2 0 p> 4 6 8
Delay, hr Log delay, hr
(a) (b)

FIGURE 7.1 Ebbinghaus’s classic retention data (a) showing percentage of savings
as a function of retention interval; (b) with both scales log transformed to reveal a
power relationship.

227
 CHAPTER 7 Retention.of Memories

show rapid initial deterioration, followed by continued, ever-slowing deteriora-

tion. Retention functions also vary in the time spans over which they occur.
Ebbinghaus’s functions are over 30 days. a 5.5 displayed similar forgetting
over 18 sec. 1) I gis 1 1

wlll BSS Later in this chapter, we will expand on the relationship

between degree of learning and retention.

enka insti edt ere initial bah Towed as

ever-slower forgetting.
LLANE LE NTL LL ITLL OLSEN LENE IEE
OCT NE OI I ETT ELEM TORR

Decay: The Power Law of ie

-ed, as is the forgetting function. In the case of the learning TOIT,

more practice produces smaller and smaller gains whereas in the case of the for>

eared in the preceding chapter, taking the logarithm of both the perfor-
mance measure and the practice measure reveals a regularity with respect to the
learning function. Figure 7.1b shows the regularity that is revealed by taking log-
arithms of the performance scale Be the time scale for retention functions.
Again, the a(inear tel 1
Ebbinghaus’s data, this functionis
Log savings = 3.86 — .126 Log delay
where 3.86 is the sabe ofthe line inMaes. 7.1b and — .126 is the slope. As »

As noted in Chapter 6, a
as noted, power functions are not the only negative-
lyaccelerated functions that could be fitted to the data. The more obvious neg-
atively accelerated functions are exponential functions, which have been partic-
ularly popular for theories of forgetting because they describe many decay
processes in nature, including radioactive decay. Only recently Rubi &
ee 1996; Wixted & Ebbesen, aide as it b
 The Retention Function

Although Ebene used percentage of savings, the more common

emory performance are probability of recall and retrieval time.
igure 7.2 shows
some data (Anderson & Paulson, 1977) obtained from looking at the speed of
pee E ENS a sentence on delays si aca from 5 sec to 30 min. a shown in

igure 7.2b displays these ERwith both scales trans-

formed accordi logarithm function; the functions in the log-log scales are
ote that the functions in
igure 7.2 are imcreasing, whereas those in Figure 7.1 are decreasing, since
longer latency reflects worse performance (Figure 7.2) and smaller savings
reflect worse performance (Figure 7.1).

Squire (1989) documented such retention functions on a very different

time scale. He looked at people’s ability to recognize the name of a TV show for
varying numbers of years after it had been canceled. Figure 7.4a shows percent-

1.30
L [ Log T = 0.64 + .08 log$
3,50 Lao

oO

o i B
d o 1.10
a £
£ 3.00;- e
fe}
a
iS = 1.00

= - 30
2 8
® bp 0.90
9 5) ES
0.80

2.00 (on ee eee eelee i 0.70 N SLE ieee lle

500 1000 1500 il 3 5 i
Delay, sec Log delay, sec
(a) (b)

FIGURE 7.2 (a) Success at sentence recognition as measured by reaction time as a

function of delay; (b) replotted as a log-log plot. (Adapted from Anderson & Paulson,
1977.)

229.
 CHAPTER 7 Retention of Memories

6.0 .
Log N = 2.03 - 0.82 log D

5.0

Pray
4.0 Boag!
3 8
ral o
25 3.0 2
5
a =]
Ee ic
=] y felt)
FDO £0

ie

-l
one 10 20 30 0 1 2 3 4
Days Log days
(a) (b)

FIGURE 7.3 (a) Krueger’s (1929) data showing number of paired associates recalled
as a function of retention interval; (b) log transformation of the data.

age of recognition as a function of number of years since the show last aired.
Recognition dropped more rapidly initially and then slowed down. Figure 7.4b
regraphs this data on log-log coordinates, revealing a linear relationship indica-
tive of a power function in the original scale.

80 4.4
Log P = 4.36 - 0.13 log Y

4.3

70 a2)
@
w @o

a
5 4.2
oO
s)
we =
wo

S cu a
60 S

4.0

50 3.9
5 10 15 0.0 1.0 2.0 3.0
Years since TV show Log years
(a)
(b)
FIGURE 7.4 (a) Probability of recognizing a TV show as a function of time since
cancellation; (b) the same data on a log-log scale. (From Squire, 1989.)

230
 The Retention Function

How should this systematic decrease in performance with time be inter-

preted? Wickelgren (1975) argued that the strength of a memory record system-
atically decays over time. Such a decay theory of forgetting is obvious and has
been around since the beginning of psychology. People with no background in
psychology often believe that their memories spontaneously decay with time.
Despite its obviousness, decay theory has had a controversial history in psy-
chology, which will be described after the discussion of interference. To fore-
shadow the conclusion, however: there is more to forgetting than decay, but
decay is part of the story.

Memories decay as a power function of the time over which

the memories are being retained.
25m RRR eR RN ENN EYES SSE SEEGER EE

Degree of Learning and Forgetting

How does amount of practice affect the retention curve? Can forgetting be pre-
vented by initially practicing the material enough? This question has been
investigated in a number of experiments, and a systematic relationship has
emerged between the degree of practice and the r ion function_over-a wide
variety of material. Figure 7.5a shows data iron Slameck>and McBlres 2983)
looking at retention_of sentences—over i to five days. Figure 7.5b
shows data from :
delays of up €c._In both cases, subjects received more or less study of the

a0
iA

2.5 5 sec study

3 4 study trials 2 i
Ss 3
o f
®
5 50 e
= E 1.0
= =

g
oO

a 3 study trials
15) 0.8

e
1.0 ri aI ‘teh ae eal lalla om
0.0 2.0 4.0 6.0 8.0 10.0 0.0 1.0 2.0 Sh) 4.0
Log min Log sec
(a) (b)

FIGURE 7.5 (a) Data from Slamecka and McElree (1983) showing the effect of an
extra trial practice in the retention function. (b) Data from Wixted and Ebbesen (1991)
showing the effect of 1 versus 5 sec of study in the retention function.

7 al
 CHAPTER 7 Retention of Memories

100 ;—

90 |

So 80
3
eo 70
8 14-sec presentation
ae 60 - 8-sec presentation
4-sec presentation

50 ie 1-sec presentation

40 a
0 20 40 60
Retention interval, sec

FIGURE 7.6 Pigeon’s accuracy of matching as a function of retention interval and

time of exposure to the sample. Source: From D. S. Grant. Learning and motivation,
Volume 7. Copyright © 1976 by Academic Press. Reprinted by permission.

material. The figures plot the data in log—log form. Increased practice resulted in
increased retention, but performance fell off linearly in the log-log scale for all
degrees of practice. In both cases the underlying functions are not only approx-
Tete near buttheyare approximately parallel, indicating that materials at
different degrees of learning were being forgotten at the same rate.
Similar retention curves are found in studies of animal memory. Chapter
5 described the matching-to-sample paradigm used to study pigeon memory
(see Figure 5.16). Pigeons are shown a color and must remember it for some
period of time so that they can later peck a key of that color.

mask dichpata eanaelaieamtaasieessal 21° 7.6 shows their performance

as a function of both the amount of practice and the retention interval. There are
different retention curves for different amounts of practice. Again we see that
the functions are approximately parallel.

Retention functions for different degrees of practice are

approximately parallel.

Environmental and Neural Bases

for the Power Law of Forgetting
The same retention function appears to describe the memory performance of
many species in a wide variety of situations using all sorts of measures. Why
should there be this apparently universal property of memory? Chapter 6 noted
things which have been more frequent in the past are more likely to eas and
that this situation may underlie the power law of learning. nd

Zon
 The Retention Function

To appreciate this analysis, it isnecessary to realize that things fee with

different patterns in the environment. Think about the need to know information
about various countries. If Libya has recently been in the headlines, it is much
more likely to appear again in tomorrow" s news than if it has not. Therefore, our

igure 7.7 shows the probability that a

word will appear on the current day as a function of the recency of its last occur-
rence and how many times it occurred in the last 100 days. These data are plot-
ted on a log-log plot for comparison with the forgetting functions in Figures 7.1
through 7.5. As shown, these data from the environment mirror_data_from
human memory. In particular,

Chapter ais enewed that the power aw increase in be avioral measures

with practice was mirrored by a power law increase in long-term potentiation
(LTP) with practice. Does this measure of neural learning similarly mirror the
behavioral retention function? Figure 7.8 presents some data from Barnes

13 past occurrences

8 past occurrences

3 past occurrences

FIGURE 7.7 Log probability of a, probability

day
critical
Log
on
word occurring in a New York Times
headline as a function of time since
last occurrence for three levels of fre- BA an sna daa Yah
quency of past occurrence. (From 1.0 2.0 3.0
Anderson & Schooler, 1991.) Log days since last occurrence

233
 CHAPTER 7 Retention.of Memories

4.5

e 4 stimulations
Log change = 4.20 — 0.10 log min
= iS)

BaD)

baseline
above
percent
Log

FIGURE 7.8 Percentage of LTP asa 25

function of log delay for two levels of
practice. (Data from Barnes, 1979.) Log min

(1979), who examined how LTP in the hippocampus decreased with time. Recall
that the LIP procedure involves administering a high-frequency stimulation to
a neural path. Afterward this path shows increased response to further stimula-
tion. As discussed with respect to Figure 6.9, LTP is frequently measured in
terms of how much more responsive the neural pathway is to stimulation over
its baseline responsiveness.
Barnes investigated the decrease in LIP for periods from 2 min up to 14
days, and she also looked at retention after one or four high-frequency stimula-
tions. As Figure 6.9 showed, LTP increases with frequency of stimulation. Figure
7.8 plots log percentage above baseline against log delay. Individual data points
are somewhat noisy, but Figure 7.8 shows the best-fitting linear functions for
one and four stimulations. The rates of decay are approximately linear and par-
allel. Thus, the behavioral retention functions shown in Figures 7.1 through 7.5
may reflect changes in the strength of neural association. To reiterate a theme
from the previous chapter, the neural learning function may have this form
because it mirrors the structure of the environment (Figure 7.7).

The power law offorgetting also characterizes changes in LTP

and the pattern of repetition of information in the environment.
AAR RRR ARR 88 CARED i see NSTI

Spacing Effects
Scientists love parsimonious theories of empirical phenomena. From a scien-
tist’s perspective, it would be nice if memory were just a matter of practice and
retention interval. However, things are a good bit more complicated in many

234
 Spacing Effects

ways. Chapter 6 discussed the compl

created
icatiby elaborative
ons
processing.
tudi elegant a of these effects is found in the research of
enberg) 976 HO sed ontinuo yNaired ociate S
10.-2AdMEeENnCeLC Aa SSA ET
PE eT

Part of such a sequence might be:

bank-tail
fish-home
fish—??
bank-tail
pail—nose
frog—girl
pail—??
snow-ball
bank—??

defined in terms of how many trials (study or test) were be

events. In the bank-tail example there are two events between the two studies
and four events from the second study to the test. Glenberg used delays of 2, 8,

aGimple forgetting>However, the effect of the interval between the two study
presentations changed with the test interval. At long test intervals, there was an.

curvilinear. The study interval that gave the best performance for each test inter-
val is starred in Fi

There are other demonstrations of the effect of

pou spagng: For instance,

q iti “al ae studied

the
50 items and underwent a series of fiveseem sme In each cycle they were

235
 CHAPTER 7 Retention of Memories

* Retention
interval
2 events

8 events
eX

32 events
64 events
*
FIGURE 7.9 Results from
Glenberg (1976). The effect of dif- Proportion
recalled
3
ferent spacings between two
studies at different retention
intervals. Source: From A. M. 2
Glenberg. Journal of Verbal Learn-
ing and Verbal Behavior, Volume
15. Copyright© 1976 by Academic o1°4 8 20 40
Press. Reprinted by permission. Number of events between two presentations

first tested for the Spanish equivalent of the vocabulary item, and,
if they
In
condition anes pas he delay pewccn. these cles. In'the@ne)

vies. guslesih at is, ach day subjects were tested for the 50 vocabulary items
and then had an climes to study them.

Howeyer, all groups were administered one final testat aa 30-day delay from
their last test-study cycle. In this case the ord erof the results was reversed; the.
Fe ELE ETS Se

TABLE 7.1 Percentage of Recall of Spanish Vocabulary Items for Various Delays
Between Studies
Intersession Test
Interval Final
(days) 1 2 3 4 5 30-Day Test
0 82 92 96 96 98 68
1 53 86 94 96 98 86
30 Bill 51 ie: 79 82 95
Source: Data from Bahrick, 1984.

236
 Spacing Effects

Memory is best when the study intervals match the retention

interval.

Spacing Effects on the Retention Function

If forgetting is more rapid

learning superior at short retention intervals?
ho

Distributed practice

Mean
recalled
number
Massed practice

FIGURE 7.10 Retention following

learning by distributed versus massed
practice. (From Keppel, 1964, by permis- EC AiG Brae 5 3 POG YF
sion.) Retention interval, days

237
 CHAPTER 7 Retention of Memories

Ped icreeitn
the Pe saysoci sel had aes eadied once or twice, and
pace the pOrauaes in ave they had been one Semen,

entices a word that is first apes cisally and ar then auditorily (presenta-
tions were counterbalanced; this discussion refers to one specific case). If the
subject remembers only one presentation, will it be the auditory or the visual
presentation! At short delays between the presentations, subjects tended to

If an sien is piece ceSaree ataaia Laeerya previous

study, the effect of the second presentation decays more rapidly.

Spacing Effects in the Environment

Data from Anderson and Schooler presented earlier in this chapter (Figure 7.5)
suggested that the retention function for memory mirrors a retention function
in the environment. Do retention functions in the environment show effects of
spacing similar to those of memory retention functions? Figure 7.11 presents an
analysis of items from the New York Times database of Anderson and Schooler
that appeared just twice in the a 100eel The figure presents separate
eten ion. fun One

Lek) ion a ) LON

relatively close together, Items that 6< ar :

D1ICTV C
 Interference

Short interval
between
occurrences

-4.2
Long interval
FIGURE 7.11 Probability that a word between
that has appeared twice in the New York occurrences
Times will appear again. The curves are 101
probability
day
Log
on
a function of the amount of time since
the last occurrence. The two curves 4.6
reflect cases when the two presenta-
tions were either close together or far
apart. (From Anderson & Schooler, 1 2 3 4
1991.) Log days since last occurrence

spacing effect in (Ae memory does appear to mirror a spacing effect in the
environment. Memory identifies massed items and makes them unavailable _

lati
show that there
Both Figures 7.7 and 7.11 a:re ipthe
between
is onsh
emory seems to be adapted to these patterns and
adjusts the strengths of records accordingly.
the most likely to be needed. Recall from Chapters 2 and 3 that animal condi-
tioning appears to reflect near-optimal statistical inference about the causal
structure of the environment.
. In both cases, the apparent opti-
mality of these basic phenomena has come as something of a surprise to psy-
chologists. However, the caveat from previous chapters needs to be repeated:
the apparent optimality does not imply that the system is engaged in explicit
statistical inference. Simple associative learning processes can display these
properties of statistical optimality.
BLN SI RT niu tei RO COSINE NEL SRN ENR MESES EHR REN EEE ESN SOAS EOIN NTE

RABY uses theaera ofwaaroccurrences Ho infer which

items are most ee un
to1a casnow.
LOR ASN NTR EER RCAOA ES AR St PERANTEAU BE PN SRR LE MESRAINE N NIE SEEN

Interference
Thus far, this chapter has described forgetting as if it were just a function of time.
However, the amount and rate of forgetting can vary dramatically with what is
learned before and after the critical material. Consider again the

239
 CHAPTER 7 Retention of Memories

iin L i aR ra GF
BE ae Naeat 2

FIGURE 7.12 Retention as a " is

function of length of interval and rt Trial 3
T
number of prior syllables.
Compare with Figure 5.5. Source:
From G. Keppel and B. J.
Underwood. Journal of Verbal
Learning and Verbal .Behavior, of
Proportion
correct
responses
Volume 1. Copyright © 1962 by 0
Academic Press. Reprinted by 3 9 18
permission. Retention interval, sec

Brown-Peterson paradigm (Figure 5.5), which showed substantial forgetting

over 18 sec. Data in those figures were averaged over many trials. Keppel and
Underwood (1962) were interested in what happened on the very first trial of
such an experiment. Figure 7.12 shows the retention functions for the first, sec-
Gna and third trials ofom an Epeene Thctadiaccnhanneiiiadmmenaat

rae Other research (e.

exe Houston, 1965; Noyd, 1965; see Crowder, 1989, for a
fener has either eae none or oula uteee on the first trial. Thus,

great deal of research has studied such interference effects.

SainealaaenenS uchSSE Euan can bee manifested in three ways.

known 2as eanrts illustrated inn the ea of the Brown

and Peterson pee ig in Figure 7.12. TERRE PRIS ae REL EE

hang rena example iis parking one’s car in a lot. Some people find
it eta and harder to remember where they parked the car. Is this because
they learn less well the new location (negative transfer) ornrforget it more rapid-
ly kproachne antemenence)

time,itwould beproactive inference. SinceFigure 7.12 does not have a reten-

tion test at zero delay, whether it is negative transfer or proactive interference
cannot be determined.!

Other methods have been used to show that the poorer performance in the Brown
and Peterson task is due to proactive interference and not to negative transfer (Loftus
& Patterson, 1975; Watkins & Watkins, 1975).

240
 Interference

ite ice. Retroactiveimtérference is an obvious wantlidare for a cause

e forgetting functions documented earlier (see Figures 7.1—-7.5). As*more.
time passes following the original learning, there is more opportunity to learn —
new material, which will interfere with retention of old material. It could be
argued that the forgetting seen in these earlier figures is entirely a function of
such retroactive interference. However, there has to be more to forgetting than
just retroactive interference. For instance, the forgetting shown in Figure 7.12 is
affected by proactive interference, not by retroactive interference.

The learning of one set of materials often interferes with the

learning and retention of another set of materials.

Item-Based Interference
uly into the natureat EMSS comes— nae OU bees the

is phenomenon has tradi ly

been studied in a
paired-associate paradigm in which subjects learn to respond with one item to
another. For instance, subjects learn to say a response, for example, dog, to a
stimulus, such as vanilla.
Table 7.2 illustrates some of the interference paradigms for paired associ-
ates. (For reviews of research using such paradigms see Postman, 1974, and
Wickelgren, 1976.) The experimenter can focus either on the effect of an earlier
experience on a later list or on the effect of a later experience on retention of an
earlier list. The first is a proactive paradigm, and the latter is a retroactive para-
digm. In the proactive paradigm in Table 7.2, the experimenter is interested in
the learning and retention of the second list, which is designated A-D. A stands
for the stimuli, and D for the responses. An A stimulus might be frog, and a D
response tire. That is, the subject has to learn to say tire to frog. Before learning
the A-D list, the subject may learn another list or be in a rest condition. If there
is a preceding list, it can share the same stimuli or not share them. The A-B con-
dition reflects shared stimuli (A) with different responses (B). Thus, the subject
might learn frog—door. The C-B condition denotes new stimuli (C) as well as new
responses (B). Thus, a subject might learn coat—ball.
Part (a) of Table 7.2 also includes information about whether learning and
retention are better or worse than the rest control ..In
the proactive paradigm,

2There is a fourth logical possibility that the learning of the second material can
impede the learning of the first material, but this situation would be a scientific con-
tradiction inasmuch as it would require causality to work back in time such that a
later event would affect an earlier event.

241
 CHAPTER 7 Retention of Memories

TABLE 7.2 Interference Paradigm for Paired Associates

(a) Proactive Paradigm
A-B, A-D C-B, A-D Rest, A-D
Experimental Experimental Control

Manipulation Learn A-B Se canine—B Rest

Target list Learn A-D Learn A-D Learn A-D
(worse) (better)
Retention Test A-D Test A-D Test A-D
(worse) (worse)
Te
==

(b) Retroactive Paradigm

A-B, A-D A-B, C-D A-B, Rest
Experimental Experimental Control
Target list Learn A-B Learn A-B Learn A-B
Manipulation Learn A-D Learn C-D Rest
Retention Test A-B Test A-B Test A-B
(much worse) (worse)

ist. (Perhaps they have learned to use

the elaborative techniques discussed in Chapter 6.) This learning-to-learn must
also be occurring in the A-B, A-D condition, so the net negative transfer in that
condition is all the more remarkable.
In the typical experiment, the A-D lists are brought to the same level of
learning in all three conditions by giving more learning trials in the A-B, A-D
condition and fewer learning trials in the C-B, A-D condition. All groups are
then given a retention test for A-D at some delay. At this point both groups that
learned prior lists are worse than the rest condition, even though they were
brought to the same level of learning. This result is said to reflect proactive inter-
ference.
Part (b) of Table 7.2 displays the retroactive paradigm. Interest is in the
retention of an original A-B list as a function of whether there is an interpolat-
ed list with the same stimuli (A-D), a list with different stimuli (C-D), or a rest
period. The standard result is worst retention in the A-B, A-D condition and
best retention in the A-B, rest condition, with the A-B, C-D condition falling in
the middle. Thus, a subsequent paired-associate list can interfere considerably
with the retention of the first list, particularly when the two lists share the same
stimuli.

242
 Interference

~waisnsaesncspecialivestenneuinciyiniel ams LaiasemmeNeima A promt

emphasized that such interference is specific to using the common A items as
stimuli. For instanceyin-an-A=B, A-D retroactive paradigm, although stbjects..

Similar effects of inference can be shown on retrieval time.

VIPS SO RE LS AOE TEE I a

associate paradigms, phcieagtiese described

inTable. 7.2:

In the A-B, A-D retroactive paradigm, for example, it _ mi sec to

_ a A-B associate after learning A—D, but it took only 1.4 sec in the con-
trol condition.

» facts have to be learned of the form

Only about 100
2+ 3=5,4+3=7,and7 x3 = 21, but chil-
dren spend years of intensive practice trying to master them. In the meantime
they are learning dozens of other facts each day (such as the names of the
newest cartoon pegtacters):Naasareathe es facts sos hanato niearry? Be

Sve eSATA ELE EOE TNE ESOTEE a

There ist aer eat of se RTOHER me we ve to RE

puahple &fee aee same items.
iN

‘is the same basic explanation

found in the theory
ACT (Anderson, 1983a, 1993) and in the ‘SAM theory
ilund & iti, 1984; ale &a pel)Uenemesreneresenc:

aaa Both ACT and SAM offermore complex Roreneace enemas

than are presented here, but the following discussion captures their gist.
Figure 7.13 illustrates the underlying concept in the case of the A-B, A-D
interference paradigm versus an A-B, C-D condition. From list 1 the subject
learns the associations frog—tire and hair—fence; then the subject learns the asso-
ciations frog—door and coat-ball from list 2. The frog-tire and frog-door combina-
tions define an A-B, A-D condition, and the hair-fence and coat—ball combina-
tions define an A-B, C-D condition. These are stored as memory records along

243
 CHAPTER 7 Retention of Memories

frog—tire hai r-fence frog—door coat—ball

(list 1) (Iist 1) (list 2) (list 2)

A-B 3 A-B A-D C-D

Memory Records

FIGURE 7.13 A representation of the associations to memory records in an A-B,

A-D paradigm (frog-tire, frog—door) and in an A-B, C-D paradigm (hair—fence,
coat-ball).

with information about the relevant list. Figure 7.13 shows the stimuli, respons-
es, and lists as different elements that can serve to activate the target memory.?
The assumption is that any item (like the stimulus frog) has a fixed capac-
ity for activating memories. Thus, if an item is part of two memories, it cannot
activate either as well as it could if it were associated with only one. It takes
longer to learn a new response to the same stimulus because the new response
must compete with the other record for activation, producing negative transfer.
Once the new response has been learned, the existence of this new association
takes activation away from the old response, producing retroactive interference.
The basic theory can be described with the following two equations. The»
h ‘a

~ Record activation = Record strength + Association strength

(Activation Equation)
A further assumption is that there is an upper bound A on the total strength of
associations to a stimulus; that is, if there are n associations to a stimulus, each
has a fraction 1/n of the total strength A:
Association strength = A/n
; (Association Equation)
Thus, there is a fixed capacity for sending activation from any stimulus, and the
association between that stimulus and any particular record must compete with
all other associations from the stimulus.

’Figure 7.13 has compressed the A-B, A-D and A-B, C-D paradigms together, where-
as more typically they are between subjects; thus, it illustrates a mixed-list design,
where a single subject sees both conditions.

244
 Interference

The two equations can be combined to predict the basic interference phe-
nomenon. Consider the situation in which a subject learns one paired-associate
list and then learns a second associate list involving the same stimuli. Since each
stimulus must be associated to two responses, the association strength will be
divided by 2 and hence will be 4/2. Therefore, there is less association strength
than when each stimulus has a unique response. Thus, there will be more for-
getting of the first list, resulting in retroactive interference. The lower associative
strength will also lead to poorer learning of the second list, resulting in negative
transfer or proactive interference.

According to the theory of associative interference, item-based

interference occurs because the record gets less activation from
its association to the item.
RE ENE AD ESTES EER SEE EEA 2 aE nee MELON SNE NNER SRN tee partt

Relationship to the Rescorla-Wagner Theory

The association equation has an interesting relationship to the Rescorla-Wagner
learning theory described in Chapters 2 and 3. Recall from Chapter 3 that, in
applying the Rescorla—Wagner theory to instrumental conditioning, the strength
of association between the stimulus and the response changes according to the
following equation:
AV =a(A — XV)
where © is the learning rate; 4 is the maximum strength of association; and XV
isthe sum of the
Pend associative strengths from the stimuli present on that

This
one would seem to imply
nply that the Rescorla—Wagner cheowy and the
Association Equation are diametrically opposed in terms of where the competi-
tion lies. However, upon closer inspection it turns out that the two rules are con-
sistent and, in fact, thatthe
Rescorla-Wagner
theory provides a mechanism for

Consider the implications of the Rescorla—Wagner rule for the learning of the
paired associates shown in Figure 7.13. Since the stimulus frog is associated to
two different responses (tire and door) in two different lists, each response is
being reinforced only some of the time. Consider the association between frog
and tire. On trials in which fire occurs as a response, the strengthening rule
according to Rescorla—Wagner is
AV = a(A
- V)

245
 CHAPTER 7 Retention of Memories

is the current strength of association between frog and tire. On trials in

where V
which tire does not occur (and door does), the strengthening rule for frog-tire is
AV=a (0-V)

where 0 is the strength of association that can be maintained when the response
does not occur. The Rescorla-Wagner role will learn the value of V that will
result in the minimal AV. In the case of an equal mixture of list 1 and list 2, con-
tinued learning will result in an asymptotic value of V = 4/2. More generally, if
there are n responses equally practiced to a stimulus, the resulting strength of
association is A/n. This result is exactly the Association Equation, which asserts
that there is a fixed value for the sum of associative strengths to one stimulus.
ASSAD REN ECMO TESTOR,

The Rescorla—Wagner theory produces associative interference

because interference manipulations result in less reinforcement
of any one response.
“Se RSE NALS
BU ESTEE ESMOND REDE REELED TONERS ES ALES EELS E IIE LES LIEN GERL EIBE L ET DELLE LEA LLL NOSOLE LLL

Recognition Memory and Multiple Cues

Thus far in this chapter, interference has been considered with respect to recall
memory. What happens in tests of recognition memory? That is, what happens
if, rather than presenting a word and asking subjects to recall the associated
memory, the experimenter presents subjects with what they studied and asks
them whether they recognize it. Consider the following experiment with sen-
tence recognition (Anderson, 1974a). Subjects memorized facts of the form A
_person-isin-the location. Examples of the material subjects studied are

(1-1) A doctor is in the bank.

(1-2) A firefighter is in the park.
(2-1) A lawyer is in the church.
(2-2) A lawyer is in the park.

As can be seen, the same profession could occur in multiple sentences, and the
same location could occur in multiple sentences. The sentences here are prefixed
by two digits. Thefirst digit indicates how many sentences the profession occurs _
in anc in. In the experi-
ment, some professions and locations occurred in three sentences as well as in
the one- and two-sentence cases shown here. Figure 7.14 illustrates the mem-
ory records that are being created and their associations. terms:such
Note that
mais Hae ore ords.and.se-shouldsbedeass.
a : 7% anecueses ada Aa el linia
Subjects practiced this material until they could recall all the material cor-
rectly. Then they were asked to perform a fact-recognition task. They were sup-

246
 Interference

doctor in bank firefighter in park lawyer in church lawyer in park

Memory Records

FIGURE 7.14 Memory records being created in an experiment by J. R. Anderson

(19742).

posed to recognize sentences they had seen when those sentences were mixed
in with combinations of professions and locations that they had not seen, for
example,”The doctor is in the park.”Table 7.3 displays the speed with which sub-
jects made these judgments as a function of the number of facts they had stud-

associations emanating from the cus ration like Figure 7.14

This experiment differs in some impo from the interference par-
adigms previously considered_Fi is remory tes} ar
recall
test. Second, thereis no contro ome of
i e Ss not.
ularly important s=

TABLE 7.3 Mean Times to Recognize Sentences in Person—Location Experiment

Number of Sentences per Profession

1 ?4 3

Number of 1 1.11 sec 1.17 sec 1.22 sec

sentences per 2 1.17 sec 1.20 sec 1.22 sec
location 8 1.15 sec 1.23 sec 1.36 sec

Note: hypothetical activation values in parentheses.

247
 CHAPTER 7 Retention.of Memories

(e.g., Postman, Stark, & Fraser, 1968). For instance, when subjects try to recall
Asya
7.13). consequ
door to frog in list 1, tire from list ‘ 2 intrudes (see ; Figure iat ence,
: al oni

e 1
This theory also explains why performance is usually better on recognition
memory tests than on recall memory tests. A recognition probe typically pro-
vides more stimuli from which to probe memory. For instance, the recognition
question, “True or False: Harding was the president after Wilson,” provides one
more item to probe memory than the recall question, “Who was the president
after Wilson?” This extra item is the term Harding. The Association Equation
implies that activation sums from this additional cue and increases the avail-
ability of this memory record. Chapter 8 expands on the difference between
recall and recognition.

Activation flows from the various terms in a memory probe in

inverse proportion to their number of associations.
oR
Ee NSE UY SESE eS COLERAINE DERELICT PAE SNET RI SLSR

Item Strength and Interference

This analysis implies that the strength of a memory record and the strengths of
‘associations sum to produce an overall activation. Thus, even highly overprac-
ticed memories should show effects of associative interference. Consider the
experiment described in Chapter 6 in which subjects practiced the recognition
of sentences such as“The doctor hated the lawyer” for 25 days. Figure 6.3 shows
data for sentences in which each concept occurred uniquely. However, there
was also an interference condition in which each of the main words (doctor,
hated, and lawyer) occurred in two sentences. Figure 7.15 compares the data for
interference and noninterference conditions of this experiment. The speedup in
both conditions follows a power law improvement rather closely. The lines are
the best-fitting power functions. Even after 25 days there is a substantial disad-
vantage for the interference material.
Note, however, that the reaction time disadvantage of the interference
condition decreases with practice. On day 1, it is more than .4 sec, but by day 25
it is less than .2 sec. This occurs because the judgment times are getting closer
to zero and so differences among conditions are being compressed. In psychol-
ogy such compression of data is often referred to as a floor effect.

Even strongly encoded records show interference when mea-

sured by recognition time.
SA RN PNET SLT ETEVSL SLNEE NTOE LOENE
O RTES EN SNES TE
PISO

248
 Interference

Interference

be pe)io)

time,
Judgment
sec

fe)
: oO
Noninterference

1 5 10 15 20 25
Days of practice

FIGURE 7.15 Recognition times for interference and noninterference sentences as a

function of practice. The solid lines represent the predictions of the best-fitting power
functions. (From Pirolli & Anderson, 1985.) Source: From J. R. Anderson. The architec-
ture of cognition. Copyright © 1983 by Harvard University Press. Reprinted by per-
mission.

Interference with Preexperimental Memories

This discussion has been assuming that the only memories associated with a
term like doctor are those learned in the experiment. However, subjects have
learned many prior associations. Suppose that a subject has m prior associations
and learns n new associations in the experiment. Then the amount of activation
to an experimental fact from a term, such as doctor, should be approximately
A/(m + n). One of the interesting implications of this analysis is that learning
information in the laboratory should interfere with memories that the subjects
had before the experiment. These memories may be so strong that there is no
effect on the probability of recalling them, but an effect should still be detectable
with latency measures.
Peterson and Potts (1982; see also Lewis & Anderson, 1976) performed an
experiment to determine whether there was inter e for material known
Defore theexpenment Table 7.4 shows the material they used. Subjects studied
one or four facts that they did not previously know about famous historical fig-
ures, such asJulius Caesar and Beethoven. They were then tested on memory for_
facts the did know before the riment, for facts they had learned as part of
the experiment, and for false facts. They had to recognize as true the first two
Ae he ee Subjects were
tested two weeks later about people for whom they had not learned any exper-
imental facts, about people for whom they had learned one fact, and about peo-
ple for whom they had learned four facts. The speed with which they could

249
 CHAPTER 7 Retention of Memories

TABLE 7.4 Examples from the Peterson and Potts Materials

Examples of Learned Facts
1 fact studied Julius Caesar was left-handed
4 facts studied Beethoven never married
Beethoven suffered from syphilis.
Beethoven was a very poor student.
Beethoven died of pneumonia.
Examples of Test Items
Known facts ‘
0 facts studied Thomas Edison was an inventor.
1 fact studied Julius Caesar was murdered.
4 facts studied Beethoven was a musician.
Learned facts
1 fact studied Julius Caesar was left-handed.
4 facts studied Beethoven never married.
False facts
0 facts studied Thomas Edison was a congressman.
1 fact studied Julius Caesar was a printer.
4 facts studied Beethoven was an exceptional athlete.

judge the known facts and the studied facts is shown in Figure 7.16 at the two-
week delay. (Similar results were found in the immediate test.) Since subjects
knew the studied facts less well than the known facts, they recognized them less
well. However, recognition memory for both studied and known facts was influ-
enced by the number of experimental facts learned. These interference effects
were weak—in no case much more than 0.1 sec. A careful experimental design
is required to detect such weak effects. It is particularly interesting that these
interfering influences remained two weeks after the initial learning.

1.9

1.8 Studied facts

3
2.1.7
=
S
& 1.6
oO
i Known facts
FIGURE 7.16 Reaction times from
Peterson and Potts (1982). The task was pa oO
to recognize known and experimental-
ly learned facts about public figures.
Data shown are a function of number 14
of experimental facts learned and %0 1 2 = 4
delay for testing. Number of experimental facts

250
 Interference

This last result brings up what has been called the paradox of the expert
(Smith, Adams, & Schorr, 1978). The results thus far indicate that the more a
person knows about any particular topic, such as Julius Caesar, the harder it is
to remember anything more about that topic. This implication paints a pretty
dismal picture of memory in everyday life. The paradox of the expert is so called
because it implies that the more expert a person is on a topic and the more that
is known about the topic, the greater the interference and the poorer memory
will be for that material. However, even though there are times when new infor-
mation is hard to retain, such as when children try to remember all their addi-
tion and multiplication facts, most of the time people do not feel hard-pressed
to learn new material.
There are boundary conditions on these interference effects. A boundary
condition means that the result does not apply in some situations. For instance,
interference occurs only when the memories people are trying to associate to a
term are unrelated. When the memories are related, memory does not worsen,
and it often improves as additional facts are learned. For example, Chapter 6
described an experiment by Bradshaw and Anderson (1982) in which subjects
learned some little-known information about famous people. In one condition,
subjects studied just a single fact:

Mozart made a long journey from Munich to Paris.

In another condition, subjects learned two additional facts that were causally
related to the target fact:

Mozart made a long journey from Munich to Paris.

plus
Mozart wanted to leave Munich to avoid a romantic entanglement.
Mozart was intrigued by musical developments coming out of Paris.

The additional sentences were experimenter-provided elaborations designed to

boost memory of the target sentence. As reviewed in Chapter 6, subjects’ mem-
ory for the target sentence was improved by having to learn these redundant
sentences. The experiment also examined subjects’ memory when they studied
additional sentences unrelated to the target sentences:

Mozart made a long journey from Munich to Paris.

plus
Mozart wrote an important composition when he was 14 years old.
Mozart’s father was critical of his marriage.

Subjects who learned two such unrelated facts showed worse memory for the
target facts than subjects who studied just the target fact. This experiment shows
that whether additional facts are facilitating or interfering depends on whether

251
 CHAPTER 7 Retention of Memories

they are consistent with the target material. Since an expert’s knowledge about
a topic is usually consistent, they do not suffer serious interference.

Subjects’ tmemory
fag th
ict- knowledge ccan ye peer mate
by unrelated material learned in the laboratory.

Context-Based Interference
The previous sections documented that item-based interference is a robust phe-
nomenon that happens for all sorts of materials in all sorts of conditions.
However, interference can be obtained among materials that do not explicitly
overlap in any component items. For instance, in an A-B, C—-D condition, there
is interference in retention of the A-B items, even though their stimuli do not
overlap with the C—D items.
On deeper analysis it turns out that the lists A-B and C-D may overlap in
items that might cue their memory. For instance, they are learned in the same
laboratory. Rather than committing to memory “B is the response for A,” the
subject must actually commit to memory something like,“B is the response for
A in the laboratory.” Then laboratory becomes an element that could cue the
recall of the paired associate. This is an example of a context cue. Context cues
are elements of the general learning situation that can become associated to the
memory record. There are many potential context cues, including things like the
temperature of the room or the sound of a bird chirping outside. Context cues
also include internal items, such as the subject’s mood and hunger pangs. In our
discussion of animal conditioning (Chapters 2 and 3), we also saw that contex-
tual cues were important stimuli for association. The basic idea is that the exter-
nal and internal environment of the subject provides items that might become
associated to the memory record. In the A-B, C-D paradigm, memory records
from both lists are likely to be associated to some of the same contextual cues
and so interfere with each other for association to these cues. The interference
in such a paradigm may be a case of item-based interference, where the items
are parts of the context.
Anderson, (1983b) performed an experiment that found evidence for inter-
fererice as a result of shared context. The experiment contrasted two groups’ of
subjects who learned three lists of paired associates on three successiverda
Two experimental contexts were used. In one context, subjects learned the rd
from a computer in a windowless cubicle. In the other context, they learned the
lists from a human experimenter in a windowed seminar room. They learned a
list on one day and were retested on the next day. There were two conditions for
learning lists 2 and 3; they could be learned in the same context as the previous
lists or in a different context. (The assignment of context to conditions was
counterbalanced over subjects.) Figure 7.17 plots the retention performance of
subjects with the chee context versus those with the nonchanging context.
The subjects with the constant context showed less retention of each successive

252
 Interference

90

Changing
context

jee)ol
Constant
context

e
80

FIGURE 7.17 Percent retention of successive

lists as a function of whether the context in vn
which they were learned changed or not. (From 1 2 3
J. R. Anderson, 1983b.) List

list, displaying cumulative proactive interference. Subjects with a changing con-

text showed little loss in their retention. Thus, contextual interference may be
one cause of memory loss in conditions where material does not explicitly over-
lap in the items being memorized. Even though the items do not overlap, the
contextual cues do and the memory records are interfering with each other's
association to the cues.
As described, memory can be impaired when the physical contexts for two
sets of memories overlap, producing interference. Memory can also be impaired
if the context changes over time, since the context cues at test may not be the
cues associated to the memories at study. Certainly a person’s internal state
(mood, boredom, hunger) changes over time. In addition, what a person
encodes from the external context may change over time. For instance, a subject
may focus initially on the experimenter, encode his or her features, and later
attend to features in the room. The elements that serve as context gradually drift.
Many theorists (Estes, 1955; Gillund & Shiffrin, 1984; Landauer, 1975) have
speculated that this drift may be the true source of the gradual decay in the for-
getting curves. As time passes, the overlap between the context cues at study
and those at test gradually decreases. For example, the material that students
learn in a classroom is associated to that classroom, their mood at the time, the
people with whom they associate, the season of the year, and so on. As time
passes, these context cues tend to change, and access to these memories is lost.
Chapter 8 reviews evidence that memory is impaired when the context changes
from study to test.

Memory can be impaired when there is interference with asso-

ciations to context cues or when the context cues shift.
SMELT ASE LENSE LITOR ELD

253
 CHAPTER 7 Retention. of Memories

Is All Forgetting a Matter of Interference?

There is evidence that prior learning can have a massive effect in producing for-
getting. Keppel, Postman, and Zavortnik (1968) had five subjects study some 36
lists of paired associates. The lists, consisting of 10 pairs of common words, were
learned at two-day intervals. Just before learning a new list, subjects were test-
ed on the previous list. Figure 7.18 plots performance on the retest averaged
across successive sets of three lists. Subjects averaged more than 50 percent on
the first three lists, but by the last lists they averaged less than 10 percent. Thus,
subjects showed massive cumulative proactive interference. This deterioration
in performance is an extended version of the phenomenon displayed in Figure
7.17 and may be similarly due to contextual inference.
Results like these show that interference can produce a great deal of forget-
ting, and so they encourage the hypothesis that all forgetting might be a function
of interference from prior materia addition, decay, jthe
alternative forgetting mechanism, has been questioned as a respectable scientific
hypothesis. In an influential paper, McGeoch (1932) argued that time per se is an
unsatisfactory theoretical variable because it cannot cause forgetting. Rather, some
process, such as interference, which is correlated with time, must be the cause.
That is, the more time that has passed, the more opportunity there is for material
to retroactively interfere with the memory. Although it is true that time per se can-
not cause forgetting, it is possible that some neural change may occur, similar to
the atrophy of muscles with disuse and that this change may not be affected by
material learned earlier or later. Decay theory is best understood as the proposal
that forgetting is produced by neural processes that progress at a steady rate inde-
pendent of what other material has been learned. As such, it is a perfectly
respectable scientific theory, though not necessarily a correct theory.*
The interference results reviewed in this section make the case that decay
cannot be the only cause of forgetting, since different amounts of forgetting occur
over the same delay depending on interference conditions. The question is
whether any forgetting can be attributed to decay. The ideal way to see if there is
any forgetting due to decay would be to eliminate all interference by preventing
the subject from learning any new material over the retention interval. Although
this is impossible, efforts have been made to minimize the amount of other
material that is learned between study of the critical material and the retention
test. Thus, such experiments try to eliminate retroactive interference but do noth-
ing about proactive interference. One manipulation has been to have subjects
sleep or not sleep during the retention interval. As early as 1924, Jenkins and
Dallenbach conducted this type of study. Subjects learned a list of 10 nonsense

4McGeoch (1942) considered the possibility that decay should be given a neural inter-
pretation, but he believed such a view was unsupported by neural evidence. As he
wrote, “No one has ever published experimental evidence that synaptic junctions
decrease in intimacy, or in anything else, when one forgets” (p. 24). On this score, the
experimental evidence has changed dramatically in the last 50 years (e.g., Figure 7.8).

254
 Interference

60

50 i

40

~€ 30
8
&
20

10
FIGURE 7.18 Mean percent recall after
48 hr, as a function of lists and succes- 0
sive blocks of three lists. (From Keppel 2 a TRS epee yor 12
et al., 1968.) Successive blocks of 3 lists

syllables and were tested after 1, 2, 4, or 8 hr, during which they were awake or
asleep. Much less forgetting occurred during sleep, during which there should be
less interference. Ekstrand (1972) reviewed a great deal of research consistent
with the conclusion that less is forgotten during the period of sleep.
, Unfortunately, these results are not conclusive evidence against the decay
theory. Decay theorists (Wickelgren, 1977) have argued that the rate of decay is
slower during sleep. Particularly compelling for this point of view is the study of
Hockey, Davies, and Gray (1972). They observed that previous studies of sleep
compared retention during the night (sleep condition) with retention during the
day (waking condition). When they kept subjects awake during the night and
had them sleep during the day, they found that night versus day. and not sleep,
was.the critical factor. This result suggests that forgetting may vary with the daily
thythms of the body. A later section of this chapter discusses the effect of time
of day on retention.
Although the sleep studies do not support the interference hypothesis, it
is nonetheless possible that forgetting is entirely a function of interference.
Sleep studies test only the hypothesis that forgetting is due to retroactive inter-
ference—that is, that forgetting is influenced by other materials learned in the
retention interval. As can be seen in Figures 7.12 and 7.18, a more potent factor
might be proactive interference—that is, influence of material prior to retention.
There is not a good theoretical understanding of proactive interference.
Proa e when it is properly demonstrated by experiments such as
that of Keppel et al. (1968), involves accelerated Joss of material that was
brou the same level of initial learning. It is extremely difficult to explain
why materials brought to the same level of learning then display different for-
getting functions over intervals in which the subjects’ activities are identical.°

> For an attempt at an explanation, see Anderson, 1983.

299
 CHAPTER 7 Retention of Memories

How much offorgetting can be attributed to interference

remains very much an open issue.

Retention of Emotionally
Charged Material
Before concluding the topic of retention, it is worth considering whether the
retention of materials might be affected by their emotional content. As it turns
out, there does appear to be an effect, but it is rather different from what most
scholars had initially expected.

Freud’s Repression Hypothesis

An influential hypothesis about forgetting, advanced by Freud (1901), is that
i ries. There is little doubt that terrible
experiences can produce disturbances of memory, although by their very nature
they defy careful experimental analysis. People who have experienced traumat-
ic events, such as the murder of a loved one, are often unable to recall many
details and sometimes go into what are called fugue states, in which they tem-
e st_of their memories.
Freud thought that repression was much more common and did not
require such extreme negative situations. In his view, repression was a major
factor in promoting forgetting. Several laboratory studies have attempted to
determine whether negative memories are repressed. Loftus and Burns (1982)
looked at how well bank employees remembered a training film about a holdup
in which a boy was brutally murdered. Memory for detail in the film was poor-
er than memory for a comparable film that did not involve the murder. Peters
(1988) asked subjects to try to recognize a nurse who had given them an inoc-
ulation. Memory for the nurse was poorer than memory for a neutral helper.
One problem with these studies is that subjects’ encoding of the events may
change as a function of the negative experience; for instance, subjects may
choose not to look at the nurse’s face. Thus, the results may reflect poorer
encoding rather than more rapid forgetting.
To show accelerated forgetting, the retention curve must be traced out.
Meltzer (1930) had college students describe their experiences over the December
break immediately after returning to school from the holidays. Six weeks later he
asked them to recall their experiences again. He found that more of the unpleas-
ant memories had been forgotten in the interval. Such an experiment is open to
other interpretations besides Freud’s repression hypothesis. For instance, it is pos-
sible that the subjects chose not to rehearse the unpleasant memories.
Parkin, Lewinsohn, and Folkard_(1982), in an extension of an experiment
by Levinger and Clark (1961), looked _at retention of associations to negative- _

256
 Retention of Emotionally Charged Material

TABLE 7.5 Mean Number of Associations Recalled out of 30 as a Function of

Retention Interval and Emotionality.
Immediate Recall Delayed Recall

Emotional Neutral Emotional Neutral

24.1 27.6 2 eal! 18.3

charged words, such as quarrel, angry, and fear, in contrast to neutral words, such
as window, cow, and tree. They looked at the number of associations recalled in
the two categories immediately and at a seven-day delay. Table 7.5 shows their
results. Consistent with Loftus and Burns and Peters, they found superior mem-
ory neutral
for the words in an immediate test. However, at adelay the results
were reversed and memory was better for the emotional words. These results are
the
ees opposite
8 of what Freud’s repression hypothesiswould have predicted.

Although subjects may show worse encoding of negative mem-

ories, there is not a well-established relationship between
valence of memory and its retention.

Arousal and Retention

The critical variable for retention may not be the negative emotions associated
with the material but the arousal that the material produces in the subject.
Kleinsmith and Kaplan (1963) had subjects learn neutral paired associations
and monitored the subjects’ galvanic skin response (GSR)° to identify the level
of arousal while they were learning each paired associate. For each subject,
Kleinsmith and Kaplan classified half the paired associates as learned with rel-
atively high arousal and the other half as learned with low arousal. They then
looked at retention of these paired associates at 2 min, 20 min, and 1 week.
Figure 7.19 shows the results. Subjects displayed better memory for the low
arousal items initially, but this result reversed at a delay, showing reminiscence
(improvement with time) in the high arousal condition. Although the general
interaction reported by Kleinsmith and Kaplan has been replicated, reminis-
cence in the high arousal condition is unusual. There is usually less forgetting of
high arousal items rather than actual improvement. Levonian (1972) reviewed a
series of studies that showed greater retention but often worse initial perfor-
mance for high arousal items. One reason for the initially poor performance is
that subjects may encode less of what they are presented in a high arousal state;
however, they retain more of what they do encode.

6 GSR measures the capacity of the skin to conduce electrical current by passing a
small electrical current through the skin. Because one component of arousal is per-
spiration, this measure of skin conductance increases at points of arousal.

257
 CHAPTER 7 Retention of Memories

recalled
Percent

1 10 10,800
Log time, min (1 week)

FIGURE 7.19 Differential recall of paired associates as a function of arousal level.

Source: From L. J. Kleinsmith and S. Kaplan. Interaction of arousal and recall interval
in nonsense syllable paired-associate learning. Journal of Experimental Psychology,
Volume 67. Copyright © 1964 by the American Psychological Association. Reprinted
by permission.

The effect of arousal on memory appears to go beyond specific items.

Memory performance is enhanced when learning takes place after administra-
tion of drugs that are stimulants, such as amphetamines, caffeine, nicotine,
picrotoxin, and strychnine (McGaugh & Dawson, 1971), and memory perfor-
mance decreases after administration of depressants such as alcohol, marijuana,
chlorpromazine, ether, or nitrous oxide (e.g., Steinberg & Summerfield, 1957).
Learning tends to be best at times of day associated with highest arousal in the
daily cycle (Eysenck, 1982). Exactly why this relationship between learning and
arousal exists is something of a mystery. However, it seems basically adaptive to
remember better material acquired in high arousal states since high arousal is
evidence that the material is important to remember.
. Folkard, Monk, Bradbury, and Rosenthal (1977) reported a demonstration
of the effects of time of day. The average young person is at the highest level of
arousal somewhere between noon and 8 pM., depending on the measure of
arousal (Folkard, 1983). In this study 13-year-olds memorized a story at either
9:00 in the morning or 3:00 in the afternoon. Folkard et al. looked at recall on an
immediate test or a test at a one-week delay. Figure 7.20 shows the results.
Immediately, students showed better memory when they had studied at 9:00 in
the morning, a time of relatively low arousal. In the retention test a week later,
performance was better when children had studied at 3:00 in the afternoon, a
time of relatively higher arousal. Interestingly, older adults have their highest
arousal in the morning and they show best retention for material learned at that
time of day (Lynne Hasher, personal communication).

258
 The False Memory Syndrome

17

16 Study at 9:00 a.m.

a |
N

eS
o= |. Study at 3:00 p.m.
8
S14
5
=

13

FIGURE 7.20 Effect of time of day of

learning on retention of a story. (From Immediate 1 week
Folkard et al., 1977.) test delay

ET

Retention is better for material learned in high arousal states.

SCRE Oa NESS HERONS Sarena eta cae LAT

The False Memory Syndrome

Related to Freud’s repression hypothesis is the recent controversy over the issue
of the“false memory syndrome.” psychotherapists
Many believe that repressed
traumatic events from childhood are a common cause of psychological distress
-inadults. For instance, Steele (1994) writes that repression is the re

That we forget events because they are too horrible to contemplate;

that we cannot remember these forgotten events by any normal
process of casting our minds back but can reliably retrieve them by
special techniques; that these forgotten events banished from con-
sciousness, strive to enter it in disguised forms; that forgotten events
have the power to cause apparently unrelated problems in our lives,
which can be cured by excavating and reliving the forgotten events.
(p. 41)
It is particularly believed that repressed memories of sexual abuse in childhood
can be a cause of dysfunction in adult women. Many therapists have told
patients presenting various symptoms that they were probably abused in child-
hood and then have proceeded in a program of probing for those hidden mem-
ories. Sometimes these memories are“ uncovered” and become the basis for the
breakup of families and legal actions by child against parent.
‘

259
 CHAPTER 7 Retention of Memories

A number of memory_re i her these “recoy-

ered memories” were actually recovered or whether they were created by the
strong suggestions of the therapists. In one study, by using methods similar to
those in therapy, Loftus and Pickrell (1995) succeeded in convincing about 25
percent of their adult subjects that they had. been lost as children in a mall
(which was not true). In another study Hyman, Husband, and Billings (1995)
convinced about 25 percent of their subjects that nonevents had happened in
their childhood—such as spilling a punch bowl on their parents at a wedding
reception. In a study with young children ages 3 to 6, Ceci, Loftus, Leichtman,
and Bruck (1994) succeeded in creating such false memories in as many as 50
percent of their subjects.
Such research cast great doubt on memories recovered through therapy.
There is now a backlash of lawsuits against therapists. An organization has been
created called the Halse Memory Syndrome Foundation consisting of family
members, including former clients who have recanted their stories. It would be
wrong to conclude that all memories recovered in therapy are false, but this
esearch shou aware uman memory can_be confused
it does not always provide us with an accurate picture of the past.

Through suggestive methods, people can be made to remember

things that never happened.

Eyewitness Memory and Flashbulb Memories

The preceding results have some interesting implications for eyewitness testi-
mony. A witness who sees a crime is given very high weighting in the delibera-
tions of a jury. The results on the false memory syndrome would make us sus-
picious that police or lawyers might be able to plant memories. Also, if the
Freudian hypothesis had any validity, eyewitness reports might be suspect at
least in cases of awful crimes. On the other hand, the results reviewed on
arousal might lead to just the opposite conjecture. Apparently, lawyers have
conflicting views as to whether high arousal leads to accurate testimony. Most
defense lawyers think high arousal impairs face recognition, whereas most
prosecutors do not (Brigham, 1981). Perhaps their beliefs just reflect what is best
for the positions they argue in court.
In fact, eyewitness testimony is often quite inaccurate. People who swear
earnestly and honestly that they saw a person at a crime scene are often wrong.
One of the more peculiar cases (described in Baddeley, 1998) involved a psy-
chologist, Donald Thomson, who had appeared on a television program dis-
cussing eyewitness testimony. A few weeks later he was picked up and identi-
fied by a woman as having raped her at the exact time he had been on televi-
sion. The woman had indeed been raped at that time, but the television had
been on while she was being raped so she had confused his face with that of the
rapist.

260
 The False Memory Syndrome

Mistaken identity: The men on the left and right were arrested for the crimes per-
formed by the man in the middle.

Easterbrook (1959))wrote an analysis of the effects of high levels of arousal

on eyewitness testimony. He reviewed some ofthe research on arousal inanj-
mals showing that_arousal_resul » For example E: B.
ohnson (1952) had found that water-deprived animals displayed less latent
learning of a maze, and Bruner, Matter, and Papanek (1955) had found that
food-deprived animals learned less about various features in maze running. On
the basis of these studies, Easterbrook hypothesized that humans in-conditions
of extreme arousal tend_to_ focus on just_a few detai sand do not encode the,
entire situation. This view has been called thef/weapon focus hypothesis beca
Victims of violent crimes may z00m in on the weapon and not even encode the
crim J
. Research on eye movements in simulated crime situations shows
a greater tendency to fixate on the crime-relevant aspects (like the gun) and
ignore irrelevant details (Loftus, Loftus, & Messo, 1987). Generally, victims do
show better memory for crime-relevant details and poorer memory for periph-
eral information (Christianson, 1992).
A general theory in psychology known as the PEK EC 1SOT

(Yerkes & Dodson, 1908) Claims thatthereisan-aplumal levelofarouse ore

ce_on. sk and that performance is poor at low or high levels o
arousal and best atintermediate levels of arousal. st
This seneral law may apply to
memory. Higher degrees of arousal lead to better retention. On the other hand,
as arousal increases, a_person_becomes_more narrowly focused and_ only
encodes a smaller amount of the available information. Thus, an intermediate
level of arousal might be best because a person can encode most of the mater-

Yerkes—Dodson law has interesting implications for test anxiety: a little test anx-
iety may improve performance on a test, but too much is detrimental.

261
 CHAPTER 7 Retention. of Memories

A related phenomenon concerns what are called _flash ies

(Brown & Kulik, 1977)—memoxies—for_sudden, significant events that people
_fgel have been burned intotheir minds forever. Many people report flashbulb
memories for where they were when they learned of or saw the Challenger~
explosion in 1986. People of my generation have this feeling about when they
learned of the Kennedy assassination in 1963, and people of an earlier genera-
tion have the same feeling about hearing of the bombing of Pearl Harbor in
be — memories are extremely emotionally- charged, and people can
many details of their recei t_of th
aren of = high emotional o
racy in these details. It has been proposed that some special memory mecha-
nism may be responsible for the extreme vividness of these memories. Whether
these memories are particularly accurate is unclear (Neisser, 1982). McCloskey,
Wible, and Cohen (1988) interviewed 29 subjects one week after the Challenger
accident and again nine months later. They found that, although subjects still
reported vivid memories nine months later, they had forgotten and distorted
information. Somewhat different results were obtained in a study of subjects’
memories of the 1989 San Francisco earthquake (Palmer, Schreiber, & Fox,
1991). People who experienced the earthquake firsthand showed enhanced
retention of many details of their experience, whereas subjects who only
watched it on TV did not. The critical factor may have been the firsthand nature
of the earthquake experience. Flashbulb memon

RAOULT SNR LEER ATEN

bien may
ts Pen eaieee erspat begsexperience in a eh
eae state,eehe re CESTretaineeea Ey do nee?
RHR v6 ANNAN AGRON

Final Reflections
Laypeople typically view forgetting as one of the most frustrating aspects of
their minds. However, numerous memory theorists have pointed out that for-
getting can be quite adaptive. Maintaining a memory has costs, both because of
the physical cost of storing it and because it can intrude when it is not wanted.
Every other system that stores records (e.g., libraries, computer file systems, per-
sonal records of bills and payments) eventually reaches its capacity for storage
and must throw something out. It is in the interest of our memories, too, to for-
get (throw out) those things that are not useful.
The factors that influence retention can be seen as examples of memory
throwing out the less useful information. The retention and spacing functions in
memory mirror similar functions in the environment. These environmental
functions measure how likely the information is to be useful. Item-based inter-
ference can be understood in terms of a similar tendency to favor likely memo-
ries. When an item like frog appears, it is likely that we will need to retrieve some

262
 Final Reflections

memory involving it. As an item is associated with more memories, any partic-
ular memory is less likely to be the one that is needed when the item appears.
An interesting question, considered in detail in Chapter 8, is whether for-
getting amounts to the actual loss of the information from memory or whether we
just no longer can retrieve the information. Often, retrieval failure is the better
interpretation of forgetting. Less useful memories may not be truly lost—they may
just become less accessible. There are many analogies in information storage sys-
tems. Less useful books in libraries are often moved to auxiliary storage buildings
where they are not so readily accessible. Data not currently used in computer sys-
tems are stored on tapes, which can be a bother to retrieve and read.
Nelson (1971, 1978) conducted a series of experiments to show that mate-
rial that subjects can no longer recall or recognize is still there. Subjects learned
paired associates and were then retested at delays varying from two to four
weeks. Nelson identified items that subjects could no longer recall or recognize
in the retention test. A subject may have learned a pair like 43—dog but could no
longer recall or recognize it. Subjects then learned a new list that involved either
the old paired associate, 43-dog, or a repairing, such as 43—house. He found that
subjects were able to better learn the unchanged 43-dog, even though they
could not recall or recognize having seen it from their earlier learning experi-
ences. They showed savings in relearning for old material. Although this result

People with firsthand experiences

have flashbulb memories of the San
Francisco earthquake.
 CHAPTER 7 Retention of Memories

does not establish that everything that has been forgotten is still there, it does
show that there are some remnants of some memories that we no longer
retrieve. It may make more sense to think of forgetting as resulting in memories
becoming less and less available rather than being deleted.

The memory system makes more available those memories

that are more likely to be useful.
*

Further Readings
Two reviews of the classic research on retention are found in works by Postman
(1974) and Wickelgren (1977). Rubin and Wenzel (1996) review the literature on
the retention and the fit of various mathematical functions. Bahrick (1984) pre-
sents some data on retention functions over 50 years, in which he characterizes
their asymptotic levels as being in a”perma store.” Neisser (1982) presents a
series of articles concerned with the relationship between memory and every-
day life. Christianson (1992) presents a review of research on the effects of emo-
tional arousal, particularly on eyewitness testimony. The April 1996 issue of the
Journal of Memory and Language was devoted to the issue of memory illusions,
including the false memory syndrome.

264
 Retrieval of Memories

Overview
This chapter considers issues about the retrieval of information from memory.
The retrieval process logically follows the acquisition and retention processes
addressed in the two previous chapters. Retrieval is perhaps the most critical
process in that often information can_be in memory and _yet_not retrieved.
Chapter 7 ended with the research of Nelson which demonstrated that subjects
enjoyed savings in relearning for memories they could no longer recall or rec-
ognize. That research raised the tantalizing possibility that people may never
truly forget memories but rather may just lose access to them. Unfortunately,
the question of whether this haunting possibility is really true has no resolution,
but this chapter discusses how memories can be unavailable for recall in one
situation and yet show their influences in another situation. It reviews the three
main approaches to this issue:
1. The Relationship among Explicit Measures of Recall. Everyone has-had
the experience of being unable to recall something on one occasion but
able to recall the same thing on some other occasion. Although memory
is inherently variable, some ways of testing it are more sensitive than oth-
ers. The most common example of this situation is the different perfor-
mance people display on recognition versus recall tests. For instance, stu-
dents almost always claim that multiple-choice questions are easier than
fill-in-the-blank questions.
2. Interactions between Study and Test. How well people perform on a test
of memory depends not only on the conditions of test but also on the
relationship of these conditions to the conditions of learning. adenaaas
have had the_experience of ret
years and being flooded with memories that we had enbinets we sae
Or we have gone to see a movie that we seem to have forgotten and find
ourselves remembering everything once we begin to watch. Apparently,
being placed back in the context in which these memories were learned

265
 CHAPTER 8 Retrieval of Memories

makes them available again, A great deal of research has been done on
Such effects. Such interactions may underlie some of forgetting in that
with thepassage oftime people may loseaccess tocues that hadallowed
them to recall their memories.
3. Implicit Measures of Memory. People know many things of which they
are quite unaware. If explicitly asked about these things they may draw a
blank, but in an appropriate circumstance they give evidence of what they
know. For instance, students often claim that they have completely forgot-
ten what they have learned in abstract mathematics courses, but they are
nonetheless able td relearn the material faster (just as Nelson’s subjects
were able to relearn the paired associates faster). This chapter reviews
some of the many ways in which people give evidence of things they can-
not consciously remember.

These three topics reflect the shift of interests in the psychology of memory.
Research on the relationship between explicit measures of memory was an
important topic in the 1960s and 1970s, to be replaced by research on interac-
tions between study and test in the 1970s and 1980s, and in turn to be replaced
by research on implicit memories in the 1980s and 1990s. As understanding of
one topic was reached, attention shifted to the next topic.

People’s memory performance depends on the type of memory

test and its relationship to the conditions at study.
SBE ERIS

The Relationship Between Various

Explicit Measures of Memory
There is ample evidence that information can be stored away in our long-term
memory and yet cannot be retrieved in some circumstances. As already noted,
the most common demonstration of this phenomenon involves contrasts of
recognition memory tests and recall tests. People typically do better on recogni-
tion tests (although cases in which this outcome is reversed will be discussed).
Using a history test as an example, a student who is unable to recall which U.S.
president followed Wilson, might well-be able to recognize that Harding was
that president. How much we can remember is in part a function of the condi-

Chapter 7 discussed why recognition memory might be superior to recall

memory. The Association Equation assumed that the activation of a memory
record increased with the number of associated cues in the environment. Thus,
a recall question, such as “Who was the president after Wilson?” presents one
relevant cue, namely, Wilson. A recognition question, such as“Was Harding the

266
 The Relationship Between Various Explicit Measures of Memory

20
Cued recall

3
8eee)
3
S
S Free recall
= 10}

FIGURE 8.1 Number of words

recalled as a function of number of sub-
sequent lists studied. (From Tulving & 6 1 2 3 4 5
Psotka, 1971.) Number of subsequent lists

president after Wilson?” presents two relevant cues, namely, Harding and Wilson.
With two cues, the memory record is more active and more likely to be recalled.
How well we can remember something depends in part on how well we
can regenerate the cues to which the memory is associated. An experiment by
Tulving and Psotka (1971) showed that what might appear to be recall failure
may really be loss of access to appropriate retrieval cues. Subjects studied from
one to six lists of 24 words. Each list consisted of four members of each of six
categories, for example, dog, cat, horse, and cow from the mammal category. After
they had studied all the lists, subjects were tested for their memory of the first
list under two conditions:
1. Free recall. They were to recall the words from the list in any order.
2. Cued recall. They were shown the six category names and were asked to
recall the words in any order.
Figure 8.1 shows the number of words successfully recalled from list 1 as a func-
tion of the number of subsequent lists learned. The free-recall data show a stan-
dard result of retroactive interference in that recall goes down as a function of
the number of subsequent lists. When subjects were given the category labels as
cues, there was relatively little forgetting. Tulving and Psotka argued that forget-
ting is largely loss of access to retrieval cues, such as category labels.

Much of memory failure can be attributed to loss of access to

appropriate retrieval cues.
~y)
Actually, president could also be counted as a cue, in which case the comparison
would be of two versus three cues.

267
 CHAPTER 8 Retrieval.of Memories

Recognition Versus Recall of Word Lists

Experimental psychologists have studied extensively the relationship between
recognition and recall. Reviewing this research offers an opportunity to test
whether the difference between the two memory measures simply indicates
that recognition provides more retrieval cues: Much of this research has been
concerned with understanding memory for a list of words. In a typical experi-
ment, subjects might be shown 30 words at the rate of one every 2 sec and then
be asked to recall as many of the words as they can in any order (a free-recall
test) or to recognize the 30 words when they are mixed in with 30 distractors.
Such experiments often show that subjects have near-perfect recognition mem-
ory for the 30 words but may be able to recall fewer than 10 words.
The issue of the difference between recall and recognition is much larger than
the issue of how people recall versus recognize such lists of words. Much more com-
plex memories can be tapped by recall or recognition tests, as any student can tes-
tify on the basis of exam experiences. Still, learning lists of words has been the focus
of much of the research, and this section concentrates on this paradigm.
As discussed in Chapter 7, subjects appear to be learning associations
between items they have to remember and the experimental context, which
includes information about the external environment and the subject’s internal
state, List learning can be viewed as paired-associate learning in which subjects
form associations between words and some representation of the experimental
context. This representation of the list is sometimes referred to as the list con-
text. Figure 8.2 shows a representation of the memory records that might be
formed. A separate record encodes the appearance of each word in the list con-
text. The list context is associated to all these records. Each word is also associ-
ated to the record encoding was studied in the list context.

Dog

[dog in list 7 ball in list rock in list hat in list gun In list

Memory Records
ie
FIGURE 8.2 Memory records encoding some of the words in a list and their con-
nections with the list context.

268
 The Relationship Between Various Explicit Measures of Memory

FIGURE 8.3 Recognition memory 1.0

for words as a function of the number ,
of nontarget lists in which they
appeared. (From Anderson & Bower, ae 1 2 3 4
1974.) Number of nontarget lists

In a recall task, subjects are informed of the list they are to recall and must /
retrieve the words. Thus, they are given the list context as the cue and must
retrieve memories of words seen in that context. Because the list context is asso-
ciated to all the records, this is a massive interference paradigm; it is not sur- c
prising that performance is usually poor in a recall test-In contrast, subjectsina __)
recognition test are given basically two cues to memory—the list context and
the word to be recognized. The word is a much better cue than the list context .
because there is no experimental interference involving the word. It is not sur- y,
prising, then, that recognition memory is much better.
Anderson and Bower (1974) conducted an experiment in which subjects
studied a number of lists of words, with certain words reappearing in varying
numbers of lists. They found that as the same word appeared in other lists,
recognition memory for whether the word appeared in a target list deteriorated.
Figure 8.3 shows how recognition performance, measured by a d-prime (d’)
measure,” declined with the greater number of additional lists, just as this asso-
ciative analysis predicted. There would be a different list context element for
each list. Thus, not only would there be multiple associations to the list context,
but as seen in Figure 8.2, there would also be multiple associations to the words.
As a word appears in more lists, it acquires more a ist con-
texts,
Lt Bi
andeR
these interfere
Se
with
ee
one another.
Tae

In list memory, both the word and the list context are avail-
able as retrieval cues in a recognition test, whereas only the
list context is available in a free recall test.

2Later this section discusses the d-prime measure, which has been advanced as a
superior measure of recognition memory.

269
 CHAPTER 8 Retrieval. of Memories

Retrieval Strategies and Free Recall

In free-recall situations, many subjects take special actions to help themselves
remember the words, such as forming special associations among the words.
One subject (described in Anderson, 1972), studying a list of words for the sec-
ond time, generated a narrative to connect.the words. Portions of that narrative
follow (the number beside each word indicates where it occurred in a list of 40;
capitalized words indicate words the subject was supposed to recall):
1. garrison—GARRISON, LIEUTENANT, DIGNITARY.
eh vulture—VULTURE ... bird, there was a bird PRESENT ... VULTURE,
bird ... GARRISON.
13. lieutenant—LIEUTENANT is in the GARRISON ... and he is being
attacked by aVULTURE that came through the window.
21. scorpion—SCORPION, remember VULTURE with SCORPION, the
GARRISON is loaded with kooky animals.
28. mercenary—the LIEUTENANT was the MERCENARY, right.
31. officer—the LIEUTENANT is an OFFICER in the ... oh ... he didn’t obey
the duties.
32. destroyer—the LIEUTENANT is an OFFICER, DESTROYER, MERCE-
NARY ... the LIEUTENANT is too much ... he’s a DESTROYER.
37. sideburns—SIDEBURNS, the LIEUTENANT has SIDEBURNS, the DIG-
NITARY has a BEARD.
The subject was building up a set of associations among the words. Then at time of
recall the subject used these interword associations as an aid to recalling the words:

The LIEUTENANT...lieu-ten-ant...is a MERCENARY with SIDE-

BURNS...DESTROYER...OFFICER...who’s in the GARRISON...
and is being attacked byVULTURES and SCORPIONS ...and a...

As the subject said each capitalized word, she wrote it down as part of the recall.
Subjects often use interword associations to avoid having to cue all their recall
from just the list context. If they can retrieve one word, they can use it to cue
recall of associated words, and these to cue recall of associated words, and so on.
Much of the behavior of subjects in a free-recall experiment can be under-
stood in terms of their attempts to come up with additional retrieval cues to help
recall. The subject just quoted was spontaneously using a story-making strategy
to help retrieve the words. Bower and Clark (1969) performed an experiment that
looked explicitly at the effect of story making on memory for a list of words. They
told their subjects to commit to memory lists of 10 unrelated nouns by making
up a story involving the words. One subject made up the following story:

A LUMBERJACK DARTed out of a forest, SKATEd around a HEDGE

past a COLONY of DUCKS. He tripped on some FURNITURE, tear-

270
 The Relationship Between Various Explicit Measures of Memory

ing his STOCKING, while hastening toward the PILLOW where his
MISTRESS lay.

The control group was given equal time to just study the words. Subjects in the
two groups studied 12 lists of 10 words. At the end of the experiment they were
asked to recall all 120 words. The experimental group was able to recall 94 per-
cent of the words, whereas the control group could only recall 14 percent. This

improved in a tree-recall experiment. Another method is to organize the words

to help the formation of associations among them. Consider the following list:
Dog, cat, mouse, chair, sofa, table, milk, eggs, butter. The list is organized into fare]
\orie—three animals, three pieces of furniture, and three food items. If subjects
detect such a categorical organization, they take advantage of it to improve their
recall. They can recall many more words when a list is explicitly organized into
categories, as in this example, than when the same words are randomly spread
throughout the list (Dallett, 1964). If subjects recall one word from a category,
they tend to recall the rest, and then move on to the next category, Their mem-
ories are further improved if at the time of test they are cued with the category
names, for_example, animal and food (Tulving & Osler, 1968; Tulving &
Pearlstone, 1966). Even though such words do not appear in the list, subjects can
use them to organize recall by generating various members of the category and
then trying to recognize which ones they saw in the list.
One theory of how subjects recall items in a free-recall test is that they
have some strategy for generating words that might be in the list. They may con-
sider words that pop into their minds, recall stories they made up, or think of
instances of categories they noticed. Whenever they think of a word, they
engage in a recognition judgment to see if it is a word they studied. They recall
the word if they can recognize it. This theory of recall is called the generate-rec-
ognize theory (Anderson & Bower, 1972b; Kintsch, 1970b) because it assumes
that subjects first generate candidate words and then try to recognize them.

The generate-recognize theory offree recall assumes that sub-

jects use various strategies for generating words and then try
to recognize words that they generate.

Mnemonic Strategies for Recall

Everyday life presents situations similar to the free-recall situation. We might
want to make a series of points in a speech that has to be delivered without
notes, or we might want to remember a grocery list without writing it down.
Waiters are often expected to take orders without notes. Memory can be great-
ly enhanced in such situations by the use of some method to systematically cue

271
 CHAPTER 8 Retrieval of Memories

memories for the information to be recalled. There are several such mnemonic
s. section describes two of the more famous techniques, the peg-
techniqueThis
word method and the method of loci, and shows how their effectiveness can be
understood in terms of the generate-recognize theory.

Pegword Method. The pegword method involves learning a set of associa-

tions between numbers and words, as in the following frequently used set:

One is a bun
Two is a shoe »
Three is a tree
Four is a door
Five is a hive
Six is sticks
Seven is heaven
Eight is a gate
Nine is wine
Ten is a hen

Suppose you want to remember the following grocery list: milk, hot dogs, dog
food, tomatoes, bananas, and bread. You would take the first item and try to asso-
ciate it to the element that corresponds to one—bun. Perhaps you would develop
an image of a bun floating in milk. Similarly, you would develop images for the rest
of the list: hot dogs sticking out of a shoe like toes, a tree bearing cans of dog food
as fruit, a door with a tomato for a handle, a hive with bananas flying in and out of
it, and sticks that when broken turn out to be bread (i.e., breadsticks). These images
are bizarre, but as reviewed in Chapter 6, they are effective ways of associating
items. When you wanted to recall the list, you could retrieve the word that corre-
‘sponded to one, namely, bun, and then retrieve the item associated with it, name-
ly, milk, and then continue through the rest of the list. The pegwords, such as bun,
can be used over and over again to learn new lists (Bower & Reitman, 1972).
This technique is very successful and confers on the user near perfect
memory for the items to be remembered. The basic technique capitalizes on two
things. First, memorizing a sequence of items, such as bun, shoe, tree, ahead of
time provides an orderly way of going through the material to prompt recall of
each item. Second, the concrete pegwords provide excellent cues to memory
when combined with learning by imagery. Both of these advantages have their
effect by helping the person generate items for recognition.

The Method of Loci. Another classic mnemonic technique, the method of

loci, also has its effect by promoting good organization in recall situations. This
method involves using some familiar path in life and associating to-be-remem-
bered items to locations on that path. For instance, you might know a path that

272
 The Relationship Between Various Explicit Measures of Memory

goes from a service station past a police station, a department store, a movie
theater, and a restaurant, to a beach. Suppose you want to use this path to mem-
orize the same list of six items: milk, hot dogs, dog food, tomatoes, bananas, and
bread. You would mentally walk along the path forming visual images that link
the locations and the items. Thus, you might imagine the service station atten-
dant pumping milk from the gas pump, a police officer at the station smoking a
hot dog, a mannequin holding dog food in the department store window, a
movie theater advertising The Attack of the Killer Tomatoes, the restaurant’s menu
written on a banana, and loaves of bread washed ashore by waves at the beach.
To recall these items at a later date, you would walk down this path in your
mind, reviving the images associated with each location. Like the pegword
method, this method has proved an effective way of learning multiple lists
(Christen & Bjork, 1976; Ross & Lawrence, 1968).
Both the method of loci and the pegword method combine the same two
principles to achieve high levels of recall. They start with a fixed sequence of ele-
ments that the memorizer already knows. Then they use vivid, interactive visu-
al images to ensure that the new items get associated to these elements. Their
effectiveness can be understood in terms of the generate-recognize theory. They
are designed to try to guarantee success in the difficult generation phase. The
assumption is that once the items are generated, memory will be able to recog-
nize them. The next section considers situations (different from those created by
these mnemonic techniques) in which the assumption of successful recognition
memory is not valid.
russe Scns

The pegword method and the method of loci facilitate recall by

.

helping to generate candidates for recognition.

RANTES son ate te ee SRE

Evaluation of the Generate-Recognize Theory

Much evidence suggests that in many situations subjects try to recall by gener-
ating possible candidates and seeing which they can recognize. As in the exam-
ple given earlier, subjects can sometimes be observed to do this. Manipulations
that affect the organization of lists (like storytelling, categorization, or
mnemonic strategy) have much stronger ef ecognition
(Kintsch, 1970b; Mandler, 1967). Such results make sense because organization
should help subjects generate items for recognition but should do little to help
them recognize the words. Subjects who are instructed that there will be a
memory test do better than incidental learning subjects on a free-recall test but
not on a recognition test (Eagle & Leiter, 1964). This result makes sense because
the intentional learning subjects would know to engage in appropriate organi-
zational strategies.°

3s we discussed in Chapter 6, intention to learn has an effect when it causes subjects

to process the material in different ways.

273
 CHAPTER 8 Retrieval'of Memories

The generate-recognize theory seems to imply that recognition memory

would always be better than recall memory because recall involves both gener-
ating the words and recognizing them. This assumption came in for some criti-
cal evaluation in a series of experimental investigations reported by Tulving and
Thomson (1973) and Watkins and Tulving’ (1975). Subjects studied pairs of
words, such as train—black, and were told they would be tested on their memo-
ty for the second word (e.g., black). The pairs of words were chosen because they
were weak associates; that is, people will occasionally generate black as an asso-
ciate to train in a free association test.
Subjects were testéd in two critical conditions:

Recall condition. Subjects were presented with cues, such as train, and
were asked to recall the target words, here black. Note that this is not the
free-recall condition for which the generate-recognize theory was devel-
oped; this condition provides a much better cue for recall (namely, train)
than does the typical free-recall experiment in which the subject only has
the list context.
Recognition condition. Subjects were presented with a high associate of
the target word, for example, white (people frequently generate black as an
associate of white), and asked to generate four free associates to the word.
Typically, one of these free associates was the target word, black. The sub-
jects were asked to judge if any of the words generated was the target
word. Thus, the subjects were put in a situation in which they would have
a high probability of generating the word, and their only difficulty should
be recognizing the word.

The results from such an experiment can be classified both according to whether
a word is recalled and according to whether the word can be recognized. Table
8.1 shows some data from Tulving and Wiseman (1975) classified according to
these factors. The table reports the proportion of words in each of the four states
obtained by crossing these factors. Two results from this paradigm are thought
to challenge seriously the generate-recognize theory. One is that memory per-
formance is sometimes higher in the recall condition than
in the recognize con-
dition. Table 8.1 shows that subjects can display a higher probability of recalling
black to train (60 percent) than of recognizing black (40 percent) when they gen-
erate it as an associate to white. This result is surprising because it seems to vio-
late the common wisdom that recognition is easier than recall.
The second result involves a comparison of the conditional probability of
recogniofaword,given thatitis recalled, with the unconditional probabil-
tion
ity of recogni the word. The unconditional
oftion probability of recognition is
calculated by dividing the number of words recognized by the number tested.
The unconditional probability is 40 percent in Table 8.1. The conditional proba-
bility is the number of recalled and recognized words divided by the total num-
ber recalled. The conditional probability might be expected to be much higher
than the unconditional probability and close to 1.0 on the view that any word

274
 The Relationship Between Various Explicit Measures of Memory

TABLE 8.1 Proportion of Words in Various Conditions of Tulving and Wiseman

(1975)
Recognized Not Recognized Totals

Recalled .60

Not Recalled 40

Totals 1.00

that can be recalled should be able to pass the easier recognition test. In fact, the
conditional probability is only slightly higher than the unconditional probabili-
ty. In Table 8.1, it is 30/60 = 50 percent, which is only slightly higher than 40 per-
cent, the unconditional probability. Many words can be recalled but not recog-
nized when they are generated in the free association test. Failure to recognize
recallable words is called recognition failure. Although these results do not
irectly a question of what is happening in a free-recall experiment,
they do call into question the view that recognition is easier than recall—one of
the basic assumptions of the generate-recognize theory of free recall.
These results can be understood by considering the retrieval cues available
for the subjects to access their memory in the two cases. In the recall case the
cue was train; in the recognition case it was black. In each case there was just one
cue. In cases in which recognition is superior to recall, the recognition test has
provided more cues to memory. These words were not chosen randomly—train
was chosen because it has a low but nonzero probability of evoking black in a
free association test, not vice versa. Thus, the reason for better performance in
the recall situation may be that train is a better cue for the memory than black.
Subjects were also instructed to study the words so that they could recall black
given train. Rabinowitz, Mandler, and Barsalou (1977) turned the typical exper-
iment around. They looked at the relationship between recognition of black (as
before) and recall of train given black as a prompt (turned around). They found
that recall was much poorer in the reverse direction (black as a prompt for train),
confirming that target words (black) are poorer cues to memory than cue words
(train). Moreover, recognition failure was much lower when conditional on
recall in the reverse direction. That is, the probability was very high that the sub-
ject could recognize black in a recognition test conditional on being able to recall
train to black. Tulving and his associates were able to get recall to be better than
recognition because they created a situation in which the recall test provided
better cues for memory than did the recognition test.
ers Soe UNH EN SEPA EASE MESA aASO RSE ES ESA SHEE SNS a

Recall tests can produce b memo C

Someanenineneeneterma ST USS AE TSUN REESE EERENA BSH POR

HERSSO EATERS GET ERE SEPALS NREL SY SRCAT LO NT

20
 CHAPTER 8 Retrieval of Memories

Measuring Recognition Memory:

The High-Threshold Model
The discussion of recognition memory so far has ignored the issue of just how
to conceive of and measure recognition memory. Suppose that a subject recog-
nizes all 25 words in a list. That might seem to be good memory, but what if the
subject also claims to recognize all 25 distractors? Then the subject is obviously
guessing and should not be given credit for high recognition memory. Of
course, subjects do not typically behave in this way. Typical subjects might say
that they recognize 20 of the words they saw and that they do not recognize the
other 5. They might also say that they recognize 5 of the distractors and correct-
ly reject the other 20. Such a false acceptance is often called afalse alarm. How
can psychologists assign a measure to how good the memory of a subject is?
They need some way of combining the probability of accepting a target—
P(YES|Target) = 20/25 = .80—and the probability of accepting a distractor—
P(YES|Distractor) = 5/25 = .20—to get a single measure of recognition memory.
One model for measuring recognition memory, the high-threshold model
(Murdock, 1974), views false acceptances by subjects as reflecting guesses. In
this
example,
with five false acceptances, the subject is guessing one-fifth of the
time. The high-threshold model assumes that the subject says that the item is a
target if it is actually recognized or if it is not actually recognized and the sub-
ject guesses. Thus, if p is the probability of actually recognizing the item and g is
the probability of guessing, the probability of saying yes to target is

P (YES|Target) =p + (1—p) g
A little algebra reveals the following correction for guessing to obtain the true
probability:
p = P(YES|Target) — P(YES|Distractor)
1 — P(YES|Distractor)
substituting P(YES|Distractor) for g. In the example, where P(YES|Target) = .8
and P(YES|Distractor) = .2, the actual probability, p, of recognizing a target, can
be estimated to be p = .75.

In measuring recognition memory it is necessary to correct for

the subject’s tendency to false alarm to items not studied.

Signal Detectability Theory

Psychology has developed a more sophisticated and useful way of measuring
recognition memory than this simple correction for guessing. This better
method turns on a deeper understanding of what is happening when the sub-
ject commits a false alarm. Sometimes a false alarm reflects a wild guess on the
subject’s part (as is assumed in the analysis of the preceding section), but other

276
 The Relationship Between Various Explicit Measures of Memory

times it reflects a deeply held belief. For instance, subjects can be asked to assign
confidences to their recognition judgments, for example, on a 1 to 7 scale, with
1 indicating a guess and 7 indicating high confidence. Subjects identify some of
their false alarms (and some of their correct recognitions) as guesses, but assign
considerable confidence to others. More than once I have had heated arguments
with subjects who insisted that Iwas wrong when I informed them that a word
did not occur on a list.
How can a subject hallucinate that a word occurred on the list? It is impor-
tant to appreciate what a recognition experiment is from a subject’s perspective.
A distractor word has occurred in many contexts, and the subject may confuse
some other context with the list context. Anderson and Bower (1974; see Figure
8.3) presented words in multiple lists. Subjects frequently thought a word
occurred in a target list if it occurred in the preceding list, consistent with the
notion that subjects were somewhat confused about just what context defined
the list context. Subjects decided that a word was studied if it occurred in a con-
text similar to the study context.
Researchers have suggested other bases for deciding whether a word has
occurred in a target list. As discussed later, a common idea is that subjects use the
raw sense of familiarity that they have about the word; a word that occurred in
the most recent list might seem particularly familiar, and subjects use this feeling
of familiarity to infer they have seen the word. Words not in the target list might
be familiar for other reasons and so might be the source of false alarms.
There are probably other bases for making recognition judgments besides
similarity of context and familiarity. Abstracting over these various possible
bases, a word can be considered as offering some evidence for being in the tar-
get list. A word that is in the target list usually offers greater evidence than a
word that is not, but sometimes a word not in the list offers more evidence than
a word in the list.
A methodology see aycece eer peer been developed to
help psychologists model how subjects make decisions when faced with the
need to discriminate between two stimuli of this kind. In the case of recognition
memory, the assumption is that there is a distribution of evidence for list mem-
bership for those words that are in the list and another distribution of evidence
for distractor words. Figure 8.4 illustrates these two distributions as normal dis-
tributions, which is what they are usually assumed to be. These distributions
reflect the probability that a particular word has a particular degree of evidence.
As shown, most target words have higher evidence than most distractor words,
but there is some overlap in the distributions, and some distractor words show
more evidence than some target words.
What subjects do is to select some criterion of evidence such that if the
word is above this criterion they accept it and if it is below this criterion they
reject it. The target words above the criterion point correspond to those words
that are correctly recognized. The distractor words above the criterion point cor-
respond to the false alarms. The proportions of these two types of words can be
used to estimate how far apart the two distributions are in terms of distance

210
 CHAPTER 8 Retrieval.of Memories

Criterion

Distractors

Correct Correct
rejections recognitions

Frequency
—>
False
rejections alarms

Degree of evidence —>

FIGURE 8.4 Distribution of evidence for targets and distractors (foils) in a recogni-
tion memory experiment.

from the center of the target distribution to the center of the distractor distrib-
ution. This distance is measured in terms of standard deviations, often referred
to as a d’ (d-prime) measure.*
Signal detectability theory is not an esoteric model that applies only to
deciding whether a word has been seen in a list memory experiment. Judgments
of this sort are constantly involved in memory decision judgments. When we
decide whether we have met someone before, we ate judging some sense of
familiarity in the person’s face and trying to decide whether it is the high famil-
iarity we would associate with a face we have seen before or whether it reflects
the low familiarity associated with a novel face. When we try to remember
whether we have been in a particular location, we are judging how similar that
location is to other locations where we have been. Signal detectability theory
provides a helpful way to model these decisions. It has also been used to
describe sensory judgments, such as deciding whether a faint tone is heard.
Indeed, the signal detectability methodology was originally developed to
describe sensory judgments.
This analysis of recognition memory implies that a subject’s performance
on a recognition memory test is a function of how difficult it is to discriminate
distractors from targets. Presumably, if the targets were words and the distrac-
tors were numbers, subjects would display very good recognition memory. In
this case, the two distributions would be very far apart in terms of degree of evi-
dence. If the distractors were very similar, recognition memory would be poor.
For instance, subjects fare worse in recognition memory tests in which the dis-
tractors are semantically similar to the targets (Underwood & Freund, 1968).

4Massaro (1989) is one source for the details of how to compute these quantities.

278
 Interactions Between Study and Test
EMM SRE ERA O RRL NA

Signal detectability theory measures recognition memory in

terms of how far the average evidence for targets is from the
average evidence for distractors.
on iP VOLES IAN RAEN

Conclusions about Recognition Versus Recall

This section started with the general observation that recognition is better than
recall. Although this phenomenon could be attributed to the greater number of
cues a recognition memory test usually provides, there are complications. For
instance, subjects can use mnemonic strategies to generate additional cues and
so improve their performance in free recall. Just how well a subject does on a
recognition test depends on the context (cues) in which the test is given and the
difficulty of the distractors. Thus, the exact level of performance in recall and
recognition tests can depend on many factors.

Interactions Between Study and Test

The preceding section treated recognition and recall as measures that generally
differ in their sensitivity to memory. Sometimes, however, one testing procedure
does not uniformly reveal more memory than another; rather, different test pro-
cedures are more appropriate for material learned in different ways. The next
section considers manipulations of context at study and at test.

Context Dependency of Memory

Chapter 7 introduced the idea of context-dependent memory—that is, mem-

\ at items get associated to some representation of the

list context. A person’s ability to recall an item depends on the person’s ability
to reproduce the list context. This ability might well be a function of the similar-
ity between the context at study and the context at test,There is evidence that
Subjects have diiticulty recafing Mems when thecontext changes between study
and test. Perhaps the most dramatic demonstrat is fact was provided by
Godden and Baddeley (1975). They had divers learn a list of 40 words either on
land or underwater, and they had the divers recall the words either on land or
underwater. Figure 8.5 displays the results of this experiment. Subjects displayed
much better memory when the context of the recall test matched the context in
which the list was studied. The interpretation is that some of the cues that the
divers had associated with the words were the contextual elements of water or
land, and it was difficult to retrieve these items in the other context. This out-
come portends a serious problem for diver education since much of it occurs on
dry land but must be retrieved underwater.

279
 CHAPTER 8 Retrieval. of Memories

52) —

ss 13 Dry recall
3 12 i environment
2)
2
tuilillal is
So
B10F
5 gL Wet recall
< environment
®
=> 8 “Y

| scaly deen
Dry Wet
Recall environment

FIGURE 8.5 Mean number of words recalled as a function of study and test envi-
ronments. (From Godden & Baddeley, 1975.)

The effect displayed in the Godden and Baddeley study is much larger than
most other context effects reported (e.g., Smith, Glenberg, & Bjork, 1978), which
have used less substantial manipulations of context. Several researchers have failed
to find context effects at all (e.g., Fernandez & Glenberg, 1985; Saufley, Otaka, &
Bavaresco, 1985). Eich (1985) argued that the magnitude of these effects depends
on the degree to which the context is integrated into the memories. He contrasted
two conditions in which subjects learned a list of nouns by means of imagery. In
one condition subjects were to imagine the nouns alone, and in the other case they
were to imagine the nouns integrated into the context. Eich found larger effects of
context variation when subjects imagined the words integrated with the context. In
terms of the cue-record representation (e.g., Figure 8.2), such integration manipu-
lations can be thought of as affecting whether contextual elements, such as the
experimental room, get associated as cues to the memory record.
Context-dependency effects have interesting implications with respect to
tasks such as exam taking. Such effects imply that people will do best on an exam
if they study in the same context in which they will take the exam, and that test
performance will be further enhanced if students try to integrate what they are
studying with the test context. Unfortunately, it is not always easy to gain access
to a test room or to get a match on many of the internal components of context.

When people integrate the context with their memories, they

show enhanced recall if they are put back in that context.
cee St AL ELSIE ND SSNS RES HRN

State-Dependent Memory
The concept of context can be extended to the internal state of the subject,
which can vary depending on whether the subject is happy or sad, hungry or
sated, excited or calm, and so on. In some cases subjects show better recall when

280
 Interactions Between Study and Test

Test while
intoxicated

o1

Test
recalled
number
Mean
while sober

FIGURE 8.6 Mean number of errors of

associative recall as a function of study Ww
and test states. (From Goodwin et al., Sober Intoxicated
1969.) State at learning

their state at test matches their state at learning. This phenomenon is referred
to as paledependent meniony One dimension of state dependency on which
there has been considerable research involves various drug-induced states. With
drugs like alcohol and marijuana, there is some evidence that subjects show bet-
ter recall if they study and are tested while sober or if they study and are tested
while intoxicated than if they study in one state and are tested in another state
(Eich, Weingartner, Stillman, & Gillin, 1975; Goodwin, Powell, Bremer, Hoine, &
Stern, 1969). A representative experiment (Goodwin et al., 1969) from this area,
illustrated in Figure 8.6, looked at the effect of being tested while sober or intox-
icated with alcohol after studying while sober or intoxicated. Subjects on the
first day (learning) were asked to make up eight paired associates and then on
the second day (recall) to recall them. There is an interaction in this figure such
that subjects who learn sober remember better when tested sober and subjects
who learned intoxicated remember better when tested intoxicated. Figure 8.6
also reflects another effect frequently found in this research: subjects performed
worse when they studied while intoxicated. This result is particularly evident in
the poor performance of subjects who studied while intoxicated and were test-
ed while sober. Depressant drugs, such as alcohol, tend to lower the amount
learned, and this effect often overwhelms any effects of state dependency.
Subjects tend to show poor memory for material they learned while in an intox-
icated state, independent of how they are tested. This outcome may in part
reflect the effect of lack of arousal on retention. As reviewed in Chapter 7, there
is better retention for material learned in a high arousal state.
SRROEDILLEDN LANE RTE NERA TR NILE LAE LN LE METI SOND SELLE NEL LTE ELT LI ELE ELLE DENA LOLA ROLL

x bjects c
study and at test matc.
\aerereeinieeza ast RSE I MN SAAN LINERS INNER IRA RIO LEONE
SRRIRAINE NELLIE SST UNL NEE NIRA

281
 CHAPTER 8 Retrieval.of Memories

Mood-Dependency and Mood-Congruence Effects

Similar state-dependent effects occur when internal state is defined in terms of
mood. Figure 8.7 shows data from Eich and Metcalfe (1989) on the interaction
between mood at study and at test. Subjects studied and recalled in happy or
sad moods induced by listening to happy or sad music. Subjects learned words
in a generate condition or a read condition, similar to the experiment of
Slamecka and Graf (1978) described in Chapter 6; that is, subjects either read a
to-be-remembered word (vanilla) or generated it to a cue that had very high
probability of evoking it (e.g., milkshake flavors: chocolate—). Three effects are
apparent in these data:
1. Replicating Slamecka and Graf, there was much higher recall when sub-
jects generated the words.
2. There was a state-dependent effect, vit better recall when the test mood
matched the study mood.
3. The state dependency was much greater inNhe generate condition.
There have been frequent findings of weak or no state-dependent effects of mood.
This is basically the result found in the read condition shown in Figure 8.7. As with
the effects of external context, tale Cependent chcetsatelargerwhen themoasis
integrated into the memories. The generate condition of Eich and Metcalfe can be
viewed as achieving an integration of study mood with memory record. Study
mood influences retrieval only if it is associated to the memory record.
In contrast to the mood- Baer effects reviewed above, mood con

0.4

0.3

a)
£ > Generate condition
oO

6 0.2
ie
a L Test sad
(e}
o

OL Test happy q

Se > Read condition

Test sad /
0.0
Happy Sad
State at learning
FIGURE 8.7 Mean proportion of generate and read items recalled as a function of
encoding and retrieval moods. (From Eich and Metcalfe, 1989, Experiment 1.)

282
 Interactions Between Study and Test

the fact that people find it easier to remember happy memories when ha

between state dependency and mood congruency. The state-dependent effect

concerns the effect of the mood the subject was in during study on memory for
all elements, including emotionally neutral items. The mood-congruent effect
concerns memory for happy or sad material even if acquired in an emotionally
neutral state. Both cases involve a match to test mood, but in one case the match

tional content of the memo

Blaney (1986) reported a review of such research. A typical study was con-
ducted by Teasdale and Russell (1983). They had subjects learn a word list con-
taining neutral, negative, or positive trait words. Before recall, an elated or
depressed mood state was induced. Figure 8.8 shows the result of mood induc-
tion on mean recall of trait words. Subjects recalled many more words that
matched their mood at test. In another study, Laird, Wagner, Halal, and Szegda
(1982) looked at memory for anger-provoking editorials or humorous Woody
Allen stories. Mood at test was induced by asking subjects either to frown or to
smile. Smiling subjects recalled more of the Woody Allen material, whereas
frowning subjects recalled more of the editorial material.
The results of mood-congruence effects can snowball for depressed.
patients. Once depressed, patients tend to remember unhappy events, which
increases the depression, which increases the retrieval of unhappy events, and
so on. At high levels of depression there is also an overall decrement in memo-
ry performance, not just for pleasant memories. Depressed subjects show lower
memory performance on standard memory tests (e.g., Watts, Morris, &
MacLeod, 1987; Watts & Sharrock, 1987). Baddeley (1997) argued that
depressed people put less effort_into elaborative learning strategies. Watts,
MacLeod, and Morris (1988) found that depressed patients show improved

12

il

1.0,

3 0.9 Positive words

iy)

FIGURE
.
8.8 Recall of
:
positive,; neg- ©ov 0.8
ative, and neutral trait words in elat- co
ed and depressed mood states. Source: =
From J. D. Teasdale and M. L. Russell. ail Negative words
Differential effects of induced mood
on the recall of positive, negative, and ag
neutral words. British Journal of ae Neutral words
Clinical Psychology, Volume 22, pp.
163-171, Figure 1. Copyright © 1983 5
by the British Psychological Society. Elation Depression
Reprinted by permission. Mood state at test

283
 CHAPTER 8 Retrieval. of Memories

memory performance if they are encouraged to use memory strategies, s

interactive mental imagery.
Although mood-congruence and state-dependent mood effects differ in
the experimental conditions that produce them, they probably reflect the same
underlying mechanism. The mood the subjectis in at test serves as one element
to help cue memory. As a consequence, the subject shows better memory for
things associated with that mood element. Mood congruence is produced
because happy and sad memories are associated to the corresponding mood
elements. State-dependent mood effects occur because in elaborating at study
the subject associates thood at study as a cue to the memory records. In both
cases, the effect is produced by overlap between the mood at test and the ele-
ments associated to the memory.

Subjects show better memory when their mood at test matches

the mood elements they have integrated into their memories.
a Ze

Encoding-Specificity Principle and

Transfer-Appropriate Processing
This chapter has reviewed some special cases of context-dependent learning
that manipulate the match between the cues at study and test. Tulving (1975)
articulated a general principle of memory that captures such interactions. This
encoding-specificity principle says that memory performance is best when
the cues present_at test_match those that were encoded with the memory at
study. A good illustration of the encoding-specificity principle is the difficulty
people have recognizing someone they normally see dressed informally when
they encounter that person dressed formally (or vice versa). Part of our recogni-
tion of such individuals is tied to the clothes they wear.
Bransford articulated a variant of this principle, known as transfer-appro-
_-ptiate processing. This principle focuses on the processes (rather than the cues)
involved in the original encoding and at test. Bransford’s principle claims that
memory is best when subjects process the memory probe at test in the same
way in which they process the material at study. A representative experiment
showing such effects was performed by Morris, Bransford, and Franks (1977).
Subjects processed words with reference to either their semantic properties or
their phonetic properties. For example, for the word hail, semantic processing
was induced by having the subject study the word with the associate snow,
whereas phonetic processing was induced by having the subject study the word
with the rhyme pail. At test subjects were cued for their recall of the words by
being tested with either a different associate (e.g., sleet) or a different rhyme
(e.g., bail).
Figure 8.9 illustrates two basic results from the Morris et al. study. First,
replicating the results about depth of processing, semantic processing at encod-
ing produced higher levels of recall. However, there was also an interaction such

284
 Interactions Between Study and Test

0.4

Associate test
3 0.3
=
2
25
Sally
2)
a 0.2
Rhyme test

FIGURE 8.9 The interaction between

the encoding condition and the recall Associate Rhyme
condition (From Morris et al., 1977.) Encoding condition

that a semantic associate was a better cue if the processing at encoding was
semantic, whereas a rhyme was a better cue if the processing at encoding was
phonetic. Since the cues were always changed from study to test, the results are
not a matter of simple overlap in the cues. What is critical is the processing that
these cues induced. Transfer-appropriate processing is further discussed in the
section on implicit memory.

Memory is better when the cue at test is processed in the same

a in which the memory was gece) at study.

formal vs. informal

Reconstructive and Inferential Memory
One type of semantic processing that has frequently been investigated is inferen-
tial, or r st, People often cannot retrieve the memo-
ry they have studied but can retrieve other memories that allow them to recon-
struct or infer what the target memory must have been. A.good deal of everyday.
recall depends on reconstructive memory. For example, if you saw the Star Wars
trilogy some time ago, try to recall the plot. You will quickly find that you cannot
remember many of the events and are inferring what happened. You will also find
yourself unsure of whether you are actually remembering things or just inferring
that they must have happened. Similar inferential processes can be shown in
response to more direct questions. Try to answer the question, “Was Princess Lea
related to Darth Vader?” You may not remember whether this relationship was
ever directly asserted, but you may recall that Luke Skywalker was Princess Lea’s
brother and Darth Vader’s son. Combining these two facts, you might infer that
Princess Lea and Darth Vader were related. Or consider the question,”Was Darth

285
 CHAPTER 8 Retrieval.of Memories

Vader evil?” Again, you might not remember whether this trait was ever asserted
in the movie series, but you can recall various events that allow you to answer this
question in the affirmative. Thus, people can use memories that they can retrieve
to infer what must be true. This ability to extend our knowledge inferentially is an
important additional attribute of our memory-system.
The British psychologist E_C, Bartlett wrote an important treatise on mem-
ory in 1932.and is famous for emphasizing the reconstructive character of
uman memory. Neisser, an American psychologist, reemphasized the recon-
i on-
structing a memory from what could beretrieved as similar to the process a
paleontologist follows to reconstruct a dinosaur from bone chips:

The traces are not simply “revived” or“reactivated” in recall; instead,

the stored fragments are used as information to support a new con-
struction. It is as if the bone fragments used by the paleontologist did
not appear in the model he builds at all—as indeed they need not, if
it is to represent a fully fleshed-out, skin-covered dinosaur. The
bones can be thought of, somewhat loosely, as remnants of the
structure which created and supported the original dinosaur, and
thus as sources of information about how to reconstruct it. (Neisser,
1967, pp. 285-286)

The basic idea is that people retrieve whatever they can from memory and then
infer what the experiences must have been that gave rise to these memory frag-
ments. Reconstructive memory is the term used to refer to the processes b
which people try to inferentially recreate their memories from what they_can
recall,—
How would a psychologist go about documenting that people actually
engage in such inferential processes when trying to recall information? One way
is to contrast conditions that facilitate or inhibit such inferences. Bransford and
Johnson (1972) looked at the effect of enabling or not enabling inferential elab-
orations. They had two groups of subjects study the following passage, which
you should try to read and then recall:

The procedure is actually quite simple. First arrange items into dif-
ferent groups. Of course one pile may be sufficient depending on
how much there is to do. If you have to go somewhere else due to
lack of facilities that is the next step; otherwise, you are pretty well
set. It is important not to overdo things. That is, it is better to do too
few things at once than too many. In the short run this may not seem
important but complications can easily arise. A mistake can be
expensive as well. At first, the whole procedure will seem complicat-
ed. Soon, however, it will become just another facet of life. It is diffi-
cult to foresee any end to necessity for this task in the immediate
future, but then, one never can tell. After the procedure is complet-

286
 Interactions Between Study and Test

ed one arranges the material into different groups again. Then they
can be put into their appropriate places. Eventually they will be used
once more and the whole cycle will then have to be repeated.
However, that is part of life. (p. 322)

Before reading this passage, some subjects were told that the passage involved
washing clothes. Given this information, they found (and presumably you would,
too) that it was easier to elaborate on this material with inferences. For instance,
the beginning of the passage could be elaborated with information about sorting
clothes by colors, and the middle of the passage, with information about costly
mistakes in washing clothes. Subjects who were told that the passage was about
doing laundry before they read it were able to recall more of the story than were
two control groups. One control group was not given this information at all. The
other control group was given this information only after reading the story. So,
knowing that the passage was about doing laundry only at test was not ade-
quate; the material had to be encoded in this way at study. This experiment pro-
vides a nice example of Bransford’s transfer-appropriate processing. By studying
the story with the knowledge that it involved washing clothes, subjects enabled
themselves to take advantage of that information at recall.
Chapter 6 discussed how memory for information is better if it is processed
more elaborately at study. One explanation is that this practice allows for recon-
structive retrieval at the time of recall. The elaborations generated at study can be
used at test to infer what the actual studied material was. There can be a benefi-
cial interaction between elaborative processing at study and test, as in the
Bransford and Johnson experiment. These findings have implications for reading
a text such as this: by placing as much meaning as possible on the text while
studying it, the reader is optimally positioned for meaningful reconstruction later.

People’s ability to reconstruct what they have studied is facili-

tated if they have processed the material in an appropriate,
meaningful way.
oe ANE TENE RETA NEE ANCE GNSS EEL IEEE RNIN TSE ISL LLE DERE LESSEN ELAS

Inferential Intrusions in Recall

Another way to show inferential processing in recall is to demonstrate that sub-
jects recall things they did not study but that follow inferentially from what they
did study. For instance, Sulin and Dooling (1974) had subjects study the follow-
ing passage:

Carol Harris’s Need for Professional Help

Carol Harris was a problem child from birth. She was wild, stubborn,
and violent. By the time Carol turned eight, she was still unmanage-
able. Her parents were very concerned about her mental health.

287
 CHAPTER 8 Retrieval of Memories

There was no good institution for her problem in her state. Her par-
ents finally decided to take some action. They hired a private teacher
for Carol. (p. 256)

One group of subjects studied this paragraph, but another group of subjects
read a paragraph that substituted”Helen Keller” for“Carol Harris.”° Later, sub-
jects were asked whether they had read the following sentence:

She was deaf, dumb, and blind.

Subjects are much more likely to think that they had studied this sentence if
they had read the Helen Keller passage than if they had read the Carol Harris
passage. From the point of view of a laboratory memory experiment, such a
recognition is often classified as an error. However, from the point of view of
adapting to the world at large, such inferences can be seen as quite appropriate.
For instance, in taking an exam, a student is expected to include plausible infer-
ences from the study materials as part of the answer to a question.
Researchers have been interested in how subjects come to recall such sen-
tences that are not part of the original passage. One possibility is that subjects
make the inference while reading the passage, and the other possibility is that
they make the inference only at the time of recall. Dooling and Christiansen
(1977) tested these possibilities by having subjects study the Carol Harris pas-
sage and then presenting them, just before test, with the information that Carol
Harris was really Helen Keller. Subjects were much more inclined to believe
they had studied the”deaf, dumb, and blind” sentence when informed about the
identity of Helen Keller just before test than when not informed at all. Since
they could not have made the inference when they studied the paragraph, they
must have made the inference when tested with the sentence.
An experiment by Owens, Bower, and Black (1979) showed that when
subjects engaged in inferential processing, there was an increase not only in
their ability to retrieve the information that they had read, but also in their intru-
sion of information that they had not read. They had subjects read a story about
a typical day in the life of a college student. Included in the story was the fol-
lowing paragraph:

Nancy went to see the doctor. She arrived at the office and checked
in with the receptionist. She went to see the nurse, who went
through the usual procedures. Then Nancy stepped on the scale and
the nurse recorded her weight. The doctor entered the room and
examined the results. He smiled at Nancy and said, “Well, it seems
my expectations have been confirmed.” When the examination was
finished, Nancy left the office. (p. 186)

°Helen Keller is famous to most Americans as someone who overcame being both
blind and deaf.

288
 Interactions Between Study and Test

TABLE 8.2 Number of Facts Recalled in Theme Versus Neutral Condition

Theme Condition Neutral Condition

Studied facts 29.2 202:

Inferred facts 15.2 Oar

Source: Adapted from Owens et al., 1979.

Two groups of subjects studied the story. The only difference between the groups
was that the theme group had read the following information before reading
any of the story:

Nancy woke up feeling sick again and she wondered if she really
were pregnant. How would she tell the professor she had been see-
ing? And the money was another problem. (p. 185)

Much like telling subjects that Carol Harris was Helen Keller, this additional
information made the passage much more interesting and enabled the subjects
to make many inferences that they might not otherwise have been able to make.
Owens et al. asked subjects to recall the story 24 hr later. They looked at facts
recalled from the story that were either actually stated in the story or could be
inferred from the story—for example, “The doctor told Nancy she was preg-
nant.”Table 8.2 displays the number of facts of each kind that were recalled as a
function of whether or not subjects were given the additional thematic passage.
Given the thematic passage, subjects recalled many additional facts that were
studied as well as many that were inferred. By increasing the subjects’ ability to
make inferences, the experimenters enabled them to remember a much richer
version of the story.

As part of memory reconstruction, subjects infer and recall

information that was not actually studied.

Conclusions about Study-Test Interactions

Many of the results about study-test interactions are captured by Tulving’s
encoding-specificity theory and Bransford’s transfer-appropriate processing
theory. The encoding-specificity theory emphasizes the overlap among the ele-
ments at study and at test. Transfer-appropriate processing emphasizes the
overlap in processes. An additional dimension of complexity concerns semantic
processing. Generally, focusing on meaningful elements or meaningful process-
ing produces more potent results, partly because subjects can better reconstruct
their memories at test from meaningfully elaborated memory fragments.

289
 CHAPTER 8 Retrieval.of Memories

Explicit Versus Implicit Memories®

The discussion in the preceding three chapters has focused on explicit memo-
ries—memories that subjects are consciously aware of when they retrieve them.
Much r has been con i ing that subjects ca
show evide emories for experiences they canno i ieve.
Such memories are called implicit memories, in contrast with the explicit
memori i le are conscious.

’ Feeling of Knowing
Sometimes memories can be just on the verge of consciousness. When people
can almost recall an item but not quite, they are said to be in a tip-of-the-tongue
state. An example is almost remembering someone’s name but not quite being
able to recall it. This phenomenon was investigated experimentally by Brown and
McNeill (1966), who presented subjects with dictionary definitions, for example,
“an instrument used by navigators to measure the angle between a heavenly
body and a horizon” or“a flat-bottomed Chinese boat usually propelled by two
oars.”” Sometimes the subjects were able to recall the word being defined or
could confidently report that they had no idea of what the word was. However,
other times the subjects reported that they felt the word was on the tip of their
tongues. If the target word was sampan and the subjects were not quite able to
recall the word, they reported that it sounded like saipan, Siam, Cheyenne, and
sarong. For words that subjects identified as being in tip-of-the-tongue states,
Brown and McNeill asked subjects questions like,“What is the first letter?”” How
many syllables does it have?” and”Can you tell me what the word sounds like?”
Subjects were able to answer such questions quite accurately.
Subjects are fairly accurate in judging whether they know something. In
one of the original studies of the feeling of knowing, Hart (1967) presented sub-
jects with questions like,“Who wrote The Tempest?” and“What is the capital of
Colombia?”Ifsubjects were unable to recall the answer, they were asked to rate
whether they would be able to recognize the answer. Subjects were able to pre-
dict quite well whether they would be able to recognize the answer. Other
research has demonstrated the accuracy of such feeling-of-knowing judgments
in other ways. Freedman and Landauer (1966) and Gruneberg and Monks
(1974) showed that subjects who thought they knew the answer were better
able to recall the answer when cued with the first letter. Nelson, Gerber, and
Narens (1984) showed that subjects who reported high feelings of knowing
were better able to perceive the answer when it was presented in a brief visual
flash. All these experiments converge in demonstrating that subjects can quite
accurately judge that they know facts they cannot consciously recall.

61 would like to thank Lynne Reder for her assistance in pointing out the studies
reviewed in this section.
7Sextant, sampan.

290
 Explicit Versus Implicit Memories

A related phenomenon is the quick judgment of knowing that occurs in

game shows. The announcer may begin to ask a question of a contestant, and
before the question is finished the contestant presses a buzzer and claims to
know the answer to the question. Reder (1987) demonstrated that subjects can
judge they know an answer to a question before they have retrieved the answer.
She asked subjects to judge as quickly as possible whether they could answer
questions like”Where did the Greek gods live?” by pressing a button. She found
that subjects could judge that they knew the answer (Mount Olympus) much
faster than they could recall the answer. Subjects averaged 2.5 sec to begin
recalling the answer, but only 1.7 sec to judge that they knew the answer. Their
judgments of knowing were also quite accurate. Ninety percent of the time
when they said they knew the answer they could in fact recall the answer.
These are all examples of subjects being aware that they know something
without being aware (yet) of what it is they know. Subjects’ implicit knowledge
is manifested in the accuracy of their answers to such questions as how many
syllables the item has or whether they will be able to recall the answer later. The
next section considers situations in which subjects are aware that they have
some familiarity with the material but do not really know the basis for that
familiarity.

People can be aware that they know something without being

able to recall what they know.

Familiarity
The earlier discussion of recognition memory spoke of subjects judging whether
they had seen an item in terms of the degree of evidence they had for its list
membership. Two types of evidence were suggested: an explicit memory that the
word was seen in the list context and a sense that the word just seemed more
familiar. Subjects sometimes are not sure why the word is familiar but judge that
they have seen the word because of its familiarity.
Tulving (1985) developed a paradigm for studying these two bases for
making a recognition memory judgment. He asked subjects to indicate whether
they explicitly’remembered’a prior presentation of the items or only”knew” the
item was in the list. This remember/know distinction has been extensively
investigated by Gardiner (e.g., Gardiner & Java, 1993). They present evidence
that subjects can accurately discriminate between these two bases for a recog-
nition judgment.
Other evidence for this distinction came from earlier research of Atkinson
and Juola (1974), who had subjects study a list of words and then looked at sub-
jects’ recognition for these words when they were mixed in with distractor
words. Subjects underwent a series of four tests in which they had to discrimi-
nate the targets from the distractors. Atkinson and Juola were interested in the
speed with which subjects could make these recognition judgments. Figure 8.10

291
 CHAPTER 8 Retrieval of Memories

850 |—

; fe Targets

ee Distractors
3 g00 +
2
®
a
5

3 750}

FIGURE 8.10 Time to recognize targets

and distractors as a function of the number
: 5 700
of times they were presented in the recog- 1 2 3 4
nition test. (From Atkinson & Juola, 1974.) Presentation number

shows that the speed of these recognition judgments varied with the number of
times subjects had been tested on the target or distractor. With repeated testing,
subjects got faster on the targets but slower on the distractors. Atkinson and
Juola argued that in the first test subjects could reject the distractors quickly
because they were unfamiliar, but with repeated testing the distractors became
more familiar and subjects had to consciously decide whether they occurred in
the list. The targets, with repeated testing, became so familiar that subjects could
quickly recognize them.
Jacoby (1991) used a paradigm in which subjects read a list of 15 words
and then heard a list of 15 different words. Then subjects were presented with a
recognition test in which they saw these 30 words plus 15 more new ones.
Subjects were instructed to recognize only the last 15 words they had heard and
not the earlier ones they had seen. They were tested under two conditions. In a
divided-attention condition, subjects had to monitor a sequence of digits spo-
ken on a tape recorder, looking for a sequence of three odd digits in a row (e.¢.,
9, 3, 7); in the full-attention condition, they could devote their full attention to
the primary task. Figure 8.11 shows the results. First, subjects falsely recognized
many of the words they had seen. Thus, having read the words created a sense
of familiarity, which led them to believe they had heard the words. Second, this
tendency was enhanced under conditions of divided attention. Subjects were
less able to engage in a process of conscious recollection and so had to count
more on their sense of familiarity.
Reder (Reder & Gordon, 1997; Reder, Nhouyvanisvong, Schunn, Ayers,
Angstadt, & Hiraki, 1997; Reder & Schunn, 1996) developed a theory that
explains this result and a great many other phenomena in implicit memory. She
proposed that in judging the familiarity of items subjects may simply be
responding to the strength of the memory records that underlie these items.
Subjects can more rapidly and more easily judge how strong a memory record

292
 Explicit Versus Implicit Memories

Full attention
0.60
[7] Divided attention
a=!

8
50 0.40
oO
£
s=
fo}
S
= 0.20

FIGURE 8.11 Proportion of words

recognized as heard under full and
divided attention. (From Jacoby, P22 Lieard
1991.) (targets) (distractors) (distractors)

is than what its actual contents are. Thus, strength serves as a basis for rapid
judgments of familiarity in the Atkinson and Juola experiment or as a basis for
judgment under divided attention in experiments like that of Jacoby.
Jacoby, Woloshyn, and Kelley (1989) showed that the sense of familiarity
can lead subjects to make a number of memory misattributions. First, they had
subjects read a series of names, for example, Sebastian Weisdorf. Subjects stud-
ied this material in a divided-attention condition or in a full-attention condition.
Then subjects were presented with these names mixed in with names of famous
people, such as Wayne Gretzky, as well as names of other, nonfamous people.
Subjects were to judge who was famous and who was nonfamous. An impor-
tant aspect of this experiment was that subjects were explicitly told that names
from the earlier, study phase were not famous. Figure 8.12 shows the results.
Subjects who were in the full-attention condition were better able to reject stud-

0.4

3
E03 Full
ss attention
2
8
FIGURE 8.12 Probability of judg- >
ing a nonfamous name famous 4%
after reading a list on which the B02 Divided
name appeared. Source: From L. L. a attention
Jacoby, “Subjective hierarchies in
spatial memory,” Current Directions
in Psychological Science, Volume 1.
Copyright © 1992, Blackwell New Old
Publishers. Type of name

293
 CHAPTER 8 Retrieval of Memories

ied names than other new names as nonfamous. They were able to use their
explicit recall of studying these names in the experimental context as a basis for
rejecting them. On the other hand, subjects in the divided-attention condition
tended to false alarm to names they had studied. Reder has explained this result
by assuming that when subjects studied the.names under divided attention,
they increased the strength of the memory records encoding these names but
did not explicitly associate the experimental context with the names.
Note that the experiment depicted in Figure 8.11 manipulated attention at
test, whereas the experiment depicted in Figure 8.12 manipulated attention at
study. Divided attention at test produces greater reliance on record strength
because the subject cannot process the test material so carefully. Divided atten-
tion at study makes it harder for the subject to encode the source of strength and
so makes it harder to filter out records that are strong for spurious reasons.
Arkes, Hackett, and Boehm (1989) and Hasher, Goldstein, and Toppino
(1977) showed that this sort of familiarity can lead subjects to come to believe
various assertions. They had subjects study sentences such as“The largest dam
in the world is in Pakistan,” and then the subjects were asked whether they
believed these assertions when mixed in with others. The previously studied
sentences received increased credibility. This is a potentially frightening result in
that it implies that propaganda does work. Merely exposing people to assertions
increases the credibility of these assertions.

People sometimes respond to the raw familiarity of an item

without determining the source of that familiarity.

Retrieval Facilitation
People can also show implicit memory for material by showing facilitation in
their processing of material as a function of exposure to the material. They can
sometimes show improved processing of material when they do not even
remember the material. Jacoby, Toth, and Yonelinas (1993) had subjects study
words under full- or divided-attention conditions as in the earlier Jacoby stud-
ies (Figures 8.11 and 8.12). They then tested the subjects in a stem-completion
task in which subjects were given a word stem and asked to complete it. For
example, the word might be motel and the stem, mot—. Some of the subjects
were explicitly instructed not to complete the stem with a word that they had
studied, whereas others were told that they could complete the stem with any
word that came to mind. Figure 8.13 shows the results in terms of how fre-
quently subjects completed the stem with the target (i.e., motel in this example).
When subjects studied the target and were told they could give it as a response,
they generated it much more frequently than when they did not have prior
exposure (inclusion instructions versus no prior exposure). Thus, they were facil-
itated in their retrieval of the target. The more interesting contrast involves sub-
jects’ performance under exclusion instructions. Particularly when they had

294
 Explicit Versus Implicit Memories

S0y=

m 40 L
£ Full
S attention
3
®
9 Divided
@ 30 attention
jou

FIGURE 8.13 Probability of generating

a word ina stem-completion task with ~ Inclusion Exclusion No prior
full and divided attention. instructions instructions exposure

studied the words under divided attention, they were more likely to recall the
target word, even though they had been explicitly instructed not to do so. The
word was more available because of its prior exposure, but they did not remem-
ber having experienced it.
Jacoby argued that the bases for implicit and explicit memory are inde-
pendent. Implicit memory is relatively unaffected by divided attention, whereas
explicit memory is seriously impaired. In the experiment depicted in Figure 8.13,
Jacoby argued that only implicit memories were formed when subjects studied
under divided attention. Since no explicit memories were formed, subjects did
not have the advantage of them to boost recall in the inclusion condition or to
filter recall in the exclusion condition.

Access to information can be facilitated by experiences that do

not result in explicit memories.

Interactions with Study Conditions

How subjects study the material appears to have different effects on implicit
versus explicit memory. For example, in another experiment, Jacoby (1983) had
subjects study information about a word in one of three conditions. Using the
word”“woman” as the example:
1. No context. Subjects just studied woman alone.
2. Context. Subjects studied the word in the presence of an opposite,
man—woman.
3. Generate. Subjects saw the word man and had to generate the opposite,
woman.

295
 CHAPTER 8 Retrieval of Memories

These three conditions manipulate the degree to which the subject engaged in
elaborative processing of the material. As indicated in Chapter 6, more elabora-
tive processing results in better memory in a standard memory test. Jacoby then
tested his subjects’ memories in one of two ways:
1. Explicit. Subjects were given a standard recognition test—they saw a list
of words and had to recognize which they had studied.
2. Implicit. Subjects were presented with the word for a brief period of time
(40 msec) and had to simply say what the word was. This was a test of their
ability to perceivesthe word when presented briefly.
The results from these two tests are displayed in Figure 8.14. The explicit condi-
tion showed the classic generation effect, with best memory in the condition
that involved the greatest semantic engagement by the subject. The results were
just the opposite in the implicit condition. Identifications were best in the No
Context condition that involved the least semantic processing. In all study con-
ditions, word identification was better than in a condition of no prior exposure.
In this control condition, subjects were able to perceive only 60 percent of the
words. Jacoby interpreted these results in terms of the match between the pro-
cessing required at study and at test. In the no-context condition, when subjects
originally encountered the word they had to rely mostly on perceptual process-
ing to identify it, whereas in the generate condition there was not even a word
to read. The result that perceptual identification is better in the no-context con-
dition than the generate condition has not always been found (e.g., Masson &
MacLeod, 1992); in some experiments, there is no difference. However, there is
always the interaction between type of processing and type of test.
Schacter, Cooper, Delaney, Peterson, and Tharan (1991) demonstrated
another example of perceptual facilitation. They presented their subjects with
drawings similar to those shown in Figure 8.15. Some were possible figures, and

90

Perceptual identification
80
(implicit memory)

(3)
~S

Percent
correct

FIGURE 8.14 Ability to rec- a Recognition judgment

ognize a word increases with (explicit memory)
depth of processing while
ability to perceive the word 50
decreases. (From Jacoby, 1983.) No context Context Generate

296
 Explicit Versus Implicit Memories

BIe
OO©4||81&aS
|S
FIGURE 8.15 Representative examples of target objects. The figures in the upper two
rows depict possible objects, and the figures in the lower two rows depict impossible
objects. Source: From D. L. Schacter, L. A. Cooper, S. M. Delaney, M. A. Peterson, and M.
Tharan. Journal of Experimental Psychology: Learning, Memory, and Cognition, Volume 17.
Copyright © 1991 by the American Psychological Association. Reprinted by permission.

some were impossible figures. Subjects were asked to judge whether these
objects faced primarily to the left or to the right. Some subjects were also asked
to make a conceptual decision—whether the object best fit the category type of
furniture, household object, or type of building. Thus, Schacter et al. manipulat-
ed the depth at which their subjects processed the material, with the perceptu-
al judgment being shallow and the conceptual judgment deep. At test the sub-
jects were presented with figures they had studied and figures they had not
studied and were asked to make one of two decisions about these objects:
1. Perceptual decision. The object appeared for just 100 msec, and the sub-
ject had to decide whether or not it was a possible object. This is an implic-
it memory test in which the experimenters were interested in how much
better subjects judged studied versus nonstudied objects.
2. Object recognition. Subjects were given unlimited time to view the
objects and had to decide whether they were objects that had been stud-
ied. This is an explicit memory test.

297,
 CHAPTER 8 Retrieval of Memories

LOim

0.9 |—
; Object recognition
io i

ie
7 0.8 |—

S
‘ rm Perceptual decision

0.7

FIGURE 8.16 Performance on percep- ote

tual-decision and object-recognition Not Ponceons
tasks as a function of type of processing conceptual
at study. (From Schacter et al., 1991.) Type of processing

The results are displayed in Figure 8.16. Typical of other explicit memory tasks, sub-
jects showed a large advantage of a conceptual or semantic processing. In sharp
contrast, there was no significant effect in the implicit perceptual-decision task.
Both the Jacoby and the Schacter et al. studies involved an interaction
between the mode of processing at study and the type of test. Elaborative or
conceptual processing led to enhanced performance on a test of subjects’ abili-
ty to consciously recognize what they had studied. Subjects showed no advan-
tage of such processing in a task that only implicitly tapped their memory for
such a task. Roediger and Blaxton (1987) interpreted such results in terms of
Bransford’s notion of transfer-appropriate processing, discussed earlier in this
chapter. They argued that tests of implicit memory, such as stem completion,
word identification, or object recognition, involve perceptual processes, where-
as tasks such as explicit recall and recognition memory are more conceptual in
nature. Therefore, only the explicit memory tests should be facilitated by study
tasks that involve conceptual processing. They argued that high performance is
obtained when the type of test matches the type of processing at study.

Elaborative processing facilitates explicit memory but not

implicit memory.

Amnesia in Humans
The distinction between implicit and explicit memory is important in under-
standing the data on amnesia in humans. Amnesia refers to the loss of memo-
ry. It can be caused by many insults to the brain, such as a blow to the head;

298
 Explicit Versus Implicit Memories

electrical convulsive shock; brain infections, such as encephalitis; a stroke; aging

phenomena, such as senile dementia; chronic alcoholism; or surgical removal of
part of the brain. Two types of amnesia have been observed—retrograde amne-
sia, which refers to lo items and experiences that occurred
before the brain insult, and anterograde esia, which refers to loss of abili-
to remember things that occurred aftes the injury. Damage to thecemporaty
Clebe appears to be particularly prone to produce such memory losses, especial-
ly when the damage includes the hippocampal formation, a structure embedded
within the temporal lobes (see Chapter 3).
Sometimes the neural damage is permanent and the resulting amnesia is
permanent; sometimes, too, there is neural recovery and partial recovery from
the amnesia. The pattern of partial recovery is particularly interesting. Figure
8.17 shows a typical pattern of recovery for a patient who suffered a closed head
injury. The patient was in a coma for seven weeks after the injury and was test-
ed 5 months, 8 months, and 16 months after the injury. Five months after the
injury the patient was in the midst of profound anterograde amnesia and was
not able to learn anything new. Moreover, the patient had severe retrograde
amnesia—total loss of memory for all events two years prior to the injury and
poor memory for events going back to infancy. When tested at 8 months the

Trauma Examination
a ag

Gross disturbance of RA, total: 2 years AA, total:

memory back to infancy not fixed

A Trauma Examination

RA, partial: RA, total: 1 year AA, total: A few

4 years’ patchy memory 3 months memories
B recalled
Trauma Examination

Memory normal RA, total: AA, total: Memory precise

2 weeks 3.5 months

23 weeks

Residual permanent memory loss

FIGURE 8.17 Recovery of a patient from amnesia (a) after 5 months; (b) after 8
months, (c) after 16 months. Source: J. Barbizet, Human memory and its pathology. W. H.
Freeman and Company, New York, Copyright 1970. Reprinted by permission.

299
 CHAPTER 8 Retrieval of Memories

total period of retrograde amnesia had shrunk to 1 year and the period of par-
tial amnesia only went back to 4 years. Also, the anterograde amnesia was
beginning to diminish, and the patient was able to remember some of the things
that had happened in the three months since the previous testing. When tested
16 months after injury, the retrograde amnesia had shrunk to just the two weeks
prior to the injury, which the patient never recovered. The deficit in learning new
things had totally disappeared—while the patient still had no memory for the
events in the 3.5 months after the coma, memory appeared normal for events
after that period. This patient illustrates how retrograde amnesia and antero-
erade amnesia often go hand in hand. As the retrograde amnesia shrinks, to
include only events just before the injury, the anterograde amnesia also dimin-
ishes and the patient is more successful in forming new memories. However,
anterograde amnesia and retrograde amnesia are not always perfectly correlat-
ed. Patients with damage to the anterior temporal lobe can_show relatively
strong retrograde amnesia with weak anterograde amnesia. The severity of the
two types Of amnesias is often reversed in patients whose damage is confined
more to the hippocampal formation.
The most famous amnesic patient is H.M., who showed no recovery. He
had large parts of his temporal lobes and related subcortical areas removed to
relieve intractable epilepsy. Included in the subcortical areas removed was the
hippocampus. H.M. has poor memory for events before his surgery (retrograde
amnesia) but apparently suffered no loss of memories from his early childhood.
Most dramatically, he appears to have lost all ability to learn new information
(anterograde amnesia). Because of the removal of his temporal lobes, his
amnesia is permanent and has lasted 40 years. He quickly forgets people he has
met and has virtually no memory for what has happened in the 40 years since
his surgery. A number of other patients with hippocampal and temporal lobe
damage show severe memory loss, though usually not as complete as that of
H.M.
There are also patients who suffered severe damages to the hippocampal
area because of a history of severe alcoholism coupled with nutritional deficits.
They show memory loss, known as Korsakoff’s syndrome, with a pattern sim;
ila a atients like H.M. Such patient roximatel mal
i information but_show severe deficits in tests of long-
t ired_after developin toms.
Korsakoff’s patients and other patients suffering hippocampal damage show
some loss of memories acquired before the onset of their condition. However,
the loss before the onset is not as dramatic as the near total inability to form new
memories after the onset.
Nonhuman primates with hippocampal lesions also show relatively pre-
served memory for information learned prior to hippocampal damage. In the
case of humans and primates, the hippocampus cannot be the site of permanent
memory storage, or there would be greater loss of memories acquired before the
injury. Rather, it seems that the hippocampus must be critical in the creation of
permanent memories, which are stored elsewhere, probably in the cortex.

300
 Explicit Versus Implicit Memories
OVP SENDESSAA ESSE SEE MERI RNASE ALUM CN CORES AIOE ROS EN SUNS EECA OS sericea megs anaes a Neo

Damage to the temporal lobe and related structures can result

in both retrograde amnesia and anterograde amnesia.
SiN EIS Ait tina tei I

Selective Amnesia
As noted in Chapter 3, lower organisms do not show complete loss of ability to
learn after removal of the hippocampus; however, there is some controversy
about how to characterize their selective learning difficulty. Such selective loss
also occurs in humans, and the amnesia (both retrograde and anterograde)
appears to be restricted to what are called explicit memories or declarative
memories. De clarative memories are_memories for factual knowledge and
ex ware.
GrafSquire, and Mandler (1984) performed an experiment that illustrates
one of the ways in which amnesiacs have preserved memory. Subjects were
shown a list of common words, such as cheese, and then later tested for their
memory of these words in one of three conditions:
1. Free recall. They were simply asked to recall all the words they had studied.
2. Cued recall. They were shown the three-letter stem of the word (e.g., che
for cheese) and asked to recall the word they had studied that began with
that stem.
3. Completion. They were shown the stem and asked to say any word (not
necessarily from the list) that began with that stem.
Figure 8.18 compares the performance of normal and amnesic subjects in these
three conditions. Normal subjects did better in the free-recall condition. This
advantage was much reduced in the cued recall condition and was actually
reversed for the completion task. In the completion task, the baseline probabil-

100

Control

80 1 Amnesic

FIGURE 8.18 Memory for

words displayed by amnesic and
normal subjects in three kinds of
tests. Source: From P. Graf, L. R. Mean
percent
Squire, and G. Mandler. Journal
of Experimental Psychology:
Learning, Memory and Cognition,
Volume 10. Copyright © 1984 by
the American Psychological : a
Association. Reprinted by per- Free Cued Completion
mission. recall recall

301
 CHAPTER 8 Retrieval of Memories

ity of completing the stem with the target word was only 9 percent when the
target word had not been studied. So, both normal and amnesic subjects
showed large effects of their exposure to the words, but amnesiacs were only
able to make this information available when they were not explicitly trying to
recall it. This experiment is an instance of a priming paradigm. Amnesiacs show
normal levels of priming in most paradigms.
Priming is only one of the paradigms in which amnesiacs show preserved
learning. Amnesiacs also show preserved ability to learn new skills. They have
been shown capable of learning many skills, such as rotary pursuit tasks, mirror
reading, or finger maze tasks. They show normal learning curves on such tasks,
even though they claim on the next day of training not to have seen the task.
Phelps (1989) argued that amnesic subjects are capable of learning any skill that
does not require explicitly retrieving information from long-term memory. Under
appropriate circumstances patients even appear capable of learning a new lan-
guage (Hirst, Phelps, Johnson, & Volpe, 1988) or a new mathematical algorithm
(Milberg, Alexander, Charness, McGlinchey-Berroth, & Barrett, 1988). The
patient H.M. (Cohen, Eichenbaum, Deacedo, & Corkin, 1985) has been shown
capable of learning a complex problem-solving skill over days, even though each
day when he was shown the task he protested that he had never seen it before.
Thus, skill learning is another major type of learning left intact in such patients.
It appears that it is a very select kind of knowledge that cannot be remem-
bered by amnesiacs with hippocampal damage: they seem unable to create new
declarative memory records. They can strengthen existing memory records and
thus show priming, and they can learn new skills. Chapter 3, in discussing the
effects of hippocampal damage in rats, reviewed the theory that the effect of hip-
pocampal lesions was to prevent learning of configural associations. Configural
associations link a number of elements together in a conditioning experiment. A
memory record is essentially a configuration of several cues. For instance, mem-
ory for the chunk RXL involves associating R, X, and L together in one configu-
ration. Since humans with hippocampal lesions have difficulty with just such
tasks, it may be that the nature of the deficit is similar in humans and in rats.

Humans with hippocampal lesions have selective deficits in

learning new declarative information.

Final Reflections
One way to review the research presented in this chapter is to consider its impli-
cations for good memory performance. Suppose you are trying to remember
some past memories. Given that they are in the past, there is nothing you can
do to better encode these memories or retain them—the topics of the previous
two chapters. Worrying about these factors would be worrying about spilled
milk. What can you do to help retrieve those old memories?

302
 Further Readings

This chapter demonstrated that people enjoy better memory if they can
recreate the elements that were associated with the memory. If you are trying to
retrieve a former acquaintance’s name, it might help to recreate in your mind
past experiences and contexts in which you used that name. For example, you
might think of names of people associated with the person whose name you are
trying to recall. It would also help if you could convert the task to a recognition
task, such as going through.an old class list.
The chapter also reviewed the importance of inferential memory for
reconstructing what can no longer be recalled. Suppose that you are trying to
remember where you placed an object. You might try to reconstruct where you
might have put it, perhaps retracing your steps, and so on.
The last part of the chapter was devoted to the notion that people have
implicit memories of which they are not consciously aware. This implies that we
should try to engage in some task that might involve the information and see if
our task performance does not have the critical knowledge embedded in it. A
classic example is knowledge of the positions of the keys on a standard type-
writer keyboard. Many people are not able to recall this information but are
nonetheless successful touch typists. They can remember where a letter is by
imagining themselves typing a word that involves the letter and seeing where
their finger goes.

It is sometimes possible to recall additional information by

utilizing knowledge about different conditions of retrieval.
neue espana

Further Readings
Massaro (1989) provides a review of the high-threshold and signal detectability
theories of recognition memory. Tulving’s (1983) book is an extensive develop-
ment of his theory of memory. Hintzman (1992) and Tulving and Flexser (1992)
engaged in an exchange on recognition failure. Squire (1987) reviews the phys-
iology of memory, including a thorough discussion of amnesic dissociations.
Schacter (1987) provides a classic article reviewing the research on implicit
memory. Squire (1992) reviews the research on the role of the hippocampus in
human memory. Roediger (1990) also provides a review of the distinction
between implicit and explicit memory and discusses this distinction in terms of
the concept of transfer-appropriate processing. Reder (1996) contains a set of
recent papers on implicit memory. Tulving and Schacter (1990) discuss research
on priming and its relationship for various memory systems.

303
 Skill Acquisition

Overview
The last four chapters, which have concentrated on the human memory system,
have ignored how our memories participate in a full functioning system. The
goal of the last three chapters of this book is to put the full system back togeth-
er. Figure 9.1 (a variation of Figure 5.1 cast more in the language of cognitive
psychology) shows how memory functions in a more general system to produce
adaptive cognitive function. Memory stores our knowledge of the world, but
there is the issue of how this knowledge is shaped in the first place and how that
knowledge is organized to produce adaptive behavior. This chapter will focus on
how the knowledge we learn is organized to provide skilled performance in var-
ious situations. The next chapter will look at how human inductive processes
play a major role in interpreting experiences to create useful memories. The final
chapter will consider how the whole system functions within an important

Environment
Instructions Experience Adaptive
cognitive functions

Human
Innate Mind Goal
constraints structures

FIGURE 9.1 A schematic represen-

Knowledge stored
tation of how the knowledge in mem-
in memory
ory participates in a full functioning
cognitive system.

304
 Overview

learning environment—schools, which are social institutions intended to create

useful memories.
Since the time of Tolman (see discussion in Chapter 1), an important issue
has been how knowledge is converted into behavior. Figure 9.1 shows the cur-
rent conception, which is basically Tolman’s proposal. Some process, often called
the” central executive” (Norman & Shallice, 1986), takes the knowledge we have
in memory and the goals we are trying to achieve and produces adaptive behav-
ior. Sometimes the process by which this is achieved involves deliberate and
conscious reasoning, but other times it is much more automatic and uncon-
scious. As we will see, one dimension of learning is this conversion of the delib-
erate into the automatic. The process of acquiring fluency in the use of knowl-
edge is called skill acquisition, which is the topic of this chapter.
We all acquire many skills to varying degrees of proficiency, and each of us
learns a few of these skills to a high degree of proficiency. For most of us, these
high-proficiency skills include speaking our native language, reading, basic
mathematical skills, interacting with other people, and driving a car. As we spe-
cialize, we tend to develop our own unique skills. Some of us become excellent
chess players, tennis players, physicists, computer programmers, Nintendo play-
ers, carpenters, pianists, teachers, and so on. The amount of time it takes to
become highly proficient at a skill is great, often measured in the hundreds and
sometimes thousands of hours. Over that period of practice, the nature of the
skill may change dramatically.
Skills such as those mentioned here are much more complex than the
behavior that is typically studied in a conditioning experiment or a memory
experiment. One of the major issues this chapter addresses is how people cope
with such complexity. One dimension of learning is the acquisition of better and
better strategies for dealing with complexity, and one way of coping with this
complexity is to automate more and more of the skill. When part of a skill is
automated, it no longer requires cognitive involvement, freeing the cognitive
system to focus on the most problematic aspects of the skill.
An example of a fairly complex skill is editing text on a word processor.
When people are first introduced to such a system, they use it in a painfully slow
and self-conscious way. With time, the situation changes dramatically. My sec-
retary is quite capable of carrying on a telephone conversation while whipping
through edits on a manuscript such as the one that led to this book. This chap-
ter traces how such a skill develops from its initial awkward performance to a
state of high automatization.
A good deal of research has focused on the development of text-editing
skill. In one experiment, Singley and Anderson (1989) studied the beginning
stages of the acquisition of text-editing skills. The subjects were secretarial stu-
dents who were proficient as typists but had not yet used word processors. Over
six days they were given practice at text editing 3 hr/day. They received a page
of a manuscript marked with six changes to be made (see Figure 9.2). This page
appeared on their computer screens, and they had to make changes to reflect
the edits.

305
 CHAPTER 9 Skill Acquisition

not only will the unit nodes in these traces

accrue strength with days of practice, but also

the element nodes will accrue strength. As will

be seen, this power function prediction

corresponds to the data about practice. A set of

experiments was conducted Pee the prediction

about a power-law increase in’strength with

extensive practice. In one experiment subjects

»
studied subject-verb-object sentences of the form

(Th¢e lawyer hated the doctor). After studying

these sentences cat!were transferred to a

sentence recognition paradigm in which they had to

discriminate these sentences from foil by-the-rAHté

sentences made of the same words as thoes:

sentence but in new combinations. There were 25 days of

tests and hence practice. Each day subjects were tested

on each sentence 12 times (in one group) or 24

times in the other group. There was no difference

FIGURE 9.2 Sample page of corrections. Source: From M. K. Singley and J. R.

Anderson. The transfer of cognitive skill. Copyright © 1989 by the President and
Fellows of Harvard College. Reprinted by permission of Harvard University Press.

Figure 9.3 shows how the time to make the edits to each page decreased
as a function of the number of days of practice. Subjects took an average of
almost 8 min per page on the first day, which became about 2 min per page by
the sixth day. This total time was divided into two categories. There were peri-
ods of time during which the subjects were not typing; any period of time when
more than 2 sec elapsed between keystrokes was classified as thinking time. The
remaining time, when keystrokes were produced at a rate of more than one
every 2 sec, was classified as keystroking time. Most of the reduction in time
resulted from a reduction in thinking time. Keystroking time was somewhat
reduced but not as a result of an improved rate of typing; rather, because the
subjects were making fewer errors and more efficient edits, they were produc-
ing fewer keystrokes. The subjects’ rates of keystroking remained constant
throughout the experiment at about two and a half keystrokes per second.
Figure 9.3 reflects the basic characteristics of many examples of skill acqui-
sition. Skill acquisition starts out with a large cognitive component. With prac-
tice, that cognitive component decreases. By day 6 shown in Figure 9.3, the
thinking component was reduced to taking the same amount of time as a motor
component (keystroking). As this chapter documents, with continued practice

306
 Overview

8.0

6.0
o
oy
ion
5 /

7 4.0
£ Thinking
= time
=
2.0

FIGURE 9.3 Improvement in text- Keystroking time

editing skill over six consecutive days of
practice. (From Singley & Anderson, 08 1 2 3 4 5 6
1989.) Days

the thinking component continues to decrease. Eventually, all cognitive involve-

ment is squeezed out, and there is only an automated motor routine.
Sy RNS ENR CS EO EESTI AUS erm vdoeminnepeomatercntseeeeN

As a skill cries more practiced, the skilluiniderghes aie

see arr ie IS ga ineee Oe
SS ENEMA RRN A ME EN RR RN ROAST ES OGL TE SENDS RNNENERIIS sansa erpemeatesihene

Power Law Learning

The learning function shown in Figure 9.3 is fit well by a power function, like
the functions of Chapter 6 that describe simple associative learning. In general,
power law learning curves fit skill-acquisition functions well. Figure 9.4 presents
data collected by Neves and Anderson (1981), who looked at improvement in
doing proofs in a logic system. Figure 9.4a displays the data on the original
scales, and 9.4b presents the data on log-log scales, where the power function
appears as a linear function. Figure 9.5 presents some of the most famous skill-
learning data in the literature (Crossman, 1959), obtained from monitoring a
factory worker's improvement in making cigars over a 10-year period. The rate
of improvement followed a power function until the worker reached the cycle
time of the equipment she was using. This situation is generally true of skill
learning—the only limitation on sl is the cycle time of the equip-
ment being used. The“equipment” in this statement includes the physical struc-
ture of the person—it takes a certain amount of time for nerve impulses to reach
the brain from receptors, such as the eye, and to go from the brain to effectors,
such as the hand. In addition, the hand can only move through space so fast.
Skilled performance continues to speed up until it reaches the minimum time
implied by these physical limitations.

307
 CHAPTER 9 Skill Acquisition

solution,
to
Time
sec

20 40 60 80 100
Number of problems
(a)

2000

1000 ;—

iBfe}(oe)
a
Log,
sec
200 ;-—

100

2 4 10 20 40 100
Log problems
(b)

FIGURE 9.4 Time to generate proofs in a geometry-like proof system as a function

of the number of proofs already done. (a) Function on a normal scale; (b) function on
a log-log scale. (From Neves & Anderson, 1981.) Source: Figure 9.3 from Cognitive
psychology and its implications by John R. Anderson. Copyright © 1990 by W. H.
Freeman and Company. Reprinted with permission.

The fact that skill learning tends to display this continuous power law
learning may seem surprising. Over the course of many years of practice mak-
ing cigars, the skill itself undergoes rather dramatic shifts in the nature of its
performance, which might be expected to be mirrored by shifts in the learning
function. Anderson (1982) argued that the reason for the uniformity in the
learning function is that all the changes, including the qualitative changes,
depend on simple associative learning, which obeys a power law, as discussed
in Chapter 6. The complex skill obeys a power law because each of its compo-
nents does.

308
 Overview

30

20
i)
oO
Pa
00
x)
°
d
£10
@
oO
oO Minimum machine
cycle time

5 (1 year) (7 year)

ob, 10,000 100,000 1,000,000 100,000,000

Number of items produced (log scale)

FIGURE 9.5 Time to produce a cigar as a function of amount of experience. Source:

From P. A. Koler and P. N. Perkins. Cognitive psychology. Copyright © 1975 by
Academic Press. Reprinted by permission...

An interesting case study of skill acquisition was reported by Ohlsson

(1992), who looked at the development of Isaac Asimov’s writing skill. Asimov
was one of the most prolific authors of our time, writing approximately 500
books in a career that spanned 40 years. He sat down at his keyboard every day
at 7:30 A.M. and wrote until 10:00 P.M. Figure 9.6 shows the average number of
months he took to write a book as a function of practice on a log-log scale. It
corresponds closely to a power function.

2.50

ov
o
2g
=(e}
§
2 1.00
@Qa
E
8
2
é
§ 0.50
FIGURE 9.6 Time to complete a book
as a function of practice, plotted with
logarithmic coordinates on both axes. 100 200 300 500
(From Ohlsson, 1992.) Number of books (log scale)

309
 CHAPTER 9 Skill Acquisition
ee LALO LEER ENROLL MEN LL OLET LER INE EEE,

The speed of performing a complex skill improves according to

a power function.
OO LEEN EEERSI SASL ILE EEE
UES SES ERI PRES EER EOI ENO TEEN ERLE LEE LL
BEL ANSNES NETS

Stages of Skill Acquisition

Fitts (1964) and Anderson (1982) proposed that skills go through three charac-
teristic stages as they develop. The following sections consider each of these
stages. Fitts called the first stage the cognitive stage.Inthisstage. the“came
oF2 Sx2M0P12 OfnO neta ee
oftenworksfrominstructionshow
or example, when I learned to shift gears with a standard transmission, I
was told the principles and in addition, my teacher demonstrated how to shift
the gears. The learner often represents the knowledge verbally and can often be
observed to rehearse the instructions in the cognitive phase—“Second is direct-
ly below first,”I would say to myself.
The_second stage is called the-associatstage. ive In this stage the skill
makes a transition from a slow and deliberat e use of the knowledge to amore

or instance, I slowly learned to coordinate releasing the clutch in first gear with
applying the gas so as not to kill the engine. Verbalization of the skill drops out
in this phase. I no longer rehearsed where second was, and I went to it much
more rapidly.
The third stage isthe autonomous stage,The skill becomes continuously
more automated and rapid, and cognitive involvement is gradually eliminated.
Sometimes a person even loses the ability to verbally describe the skill. In such
a case, the skill becomes totally a matter of implicit memory (see Chapter 8). An
interesting example involves my wife, who was teaching me how to shift gears.
She had completely forgotten whether the gas should be released when engag-
ing the clutch; that is, she could not say what she did, though her foot knew per-
fectly well what to do. When she wanted to find out what to tell me, she had to
assume the driver’s seat and see what she did.
This chapter is organized according to these three stages. These are not
discrete stages, but they characterize approximate points in the qualitative evo-
lution of a skill. The continuous nature of the power law improvement reviewed
in the previous section seems somewhat at odds with the fact that a skill under-
goes what amounts to a dramatic qualitative evolution. The apparent power law
improvement is probably only approximate; it is as good an approximation as it
is because associative learning (which also approximates a power function) gov-
erns the qualitative changes.
RETNA STANLEY
NIA TELLIN TT TIS

A skill develops from the cognitive stage to the associative

KG stage and then to the autonomous stage.
5S RNS ERR EN EIRENE
EEN RAN INNER ED — oe

310
 The Cognitive Stage

The Cognitive Stage

Most people might not think about the intimate relationship between skill
acquisition and problem solving. When we think of a skill, we tend to think of a
smooth behavioral performance. When we think of problem solving, we tend to
think of something that is performed in fits and starts as a person works out a
solution. However, every smooth, skilled performance has its origins as a solu-
tion to a novel problem. For instance, the secretarial students in the Singley and
Anderson experiment were faced with making edits in a novel word-processing
system—something they had never done before. This task eventually became
facile for them, but they started out with all the awkwardness associated with
novel problem solving. The first thing a learner must do when faced with a new
task is to organize some solution to the problem. The learner therefore starts
with some factual information about the problem. The secretarial students, for
instance, were told about the basic commands of the word processor and were
given examples of each command. The cognitive stage of skill acquisition
involves these initial problem-solving efforts. The field of problem-solving
research is concerned with how people go from some initial factual knowledge
about a problem domain, such as text editing, to their first solutions of problems
in that domain.
Research on problem solving has been strongly influenced by develop-
ments in artificial intelligence, a branch of computer science concerned with
creating intelligent computers. As reviewed in Chapter 1, Newell and Simon
(1972) synthesized ideas from artificial intelligence and experimental research
on human problem solving to produce an extremely influential framework for
understanding human problem solving. Their theory of problem solving is basi-
cally an elaboration of what is meant by the central executive in Figure 9.1. They
assume that we use our knowledge to form what are called o, rs to achieve
the Z0al ures for changing the current sit-

o L First, we may not have knowledge of the nec

sary_operators. For instance, one might not be able to play chess because one
does not know the rules of chess—that is, what the legal moves are. Each legal
move would be considered an operator. However, just knowing the operators is
not enough to guarantee success as any chess duffer can attest. It is also neces-
sary to know when to apply particular operators to solve the problem. In the
case of chess, there are many moves to make and there is a question of how to
choose the right one. The second reason that problem solving is problematic is
that we do not know how to select the right operator.
In our discussion of memory and conditioning we have already discussed
at great length how a problem solver acquires the knowledge that forms the
basis for problem-solving operators. We will elaborate more on this subject in
the next chapter on induction. Here we will focus on the issue of how one
selects operators to achieve one’s goals. The Newell and Simon analysis reveals

311
 CHAPTER 9 Skill Acquisition

the existence of two principal mechanisms by which people and other organ-
isms select operators to perform tasks:
1. Difference reduction. People select operators that will eliminate differ-
ences between their current states and their goals. A simple case is when
a single operator transforms the current state into the goal state, such as a
bar press delivering food. Often the problem solver must settle for an
operator that removes a single difference between the current state and
the goal state. For instance, a subject in the text-editing experiment, faced
with the markedsup page shown in Figure 9.2, might choose to delete
“illustrates,” because this action would move the page one step closer to
the target state but leave more changes to be performed.
. Operator subgoaling. In the process of trying to achieve a goal, people set
subgoals when operators do not work because some precondition is not
satisfied. For example, the subject may want to delete a word but first must
find where that word is in the manuscript. Locating the word in the man-
uscript file becomes a subgoal to deleting it. A subgoal is a goal pursued
in service of a higher goal. This is the means—ends step in the Newell and
Simon theory of problem solving, which was discussed in Chaptert—
The next sections describe difference reduction and operator subgoaling in
human problem solving. These mechanisms are used to convert what one has
learned into adaptive behavior. Basically, we are addressing one aspect of the
motivation issues discussed in Chapter 4, but from a more cognitive perspective.

Difference reduction and operator subgoaling are two mecha-

Se nisms for guiding the selection of problem-solving operators.
SRE Sea acai Dai

Difference Reduction
Difference reduction is a guiding force in many domains. When people try to get
ion to another, they choose moves that reduce their distance from
the goal. When I need to tidy up my office, I choose to tidy up part of it at a time,
confident that by eliminating differences one at a time between the current
office and a tidy office I will finally arrive at a tidy office.! More often than not,
problem solving that focuses on difference reduction is successful because we
can usually get from where we are to where we want to be by reducing the dif-
ferences. However, puzzles can be created that violate this general rule of
thumb. Sometimes the only way to solve a problem is to temporarily increase
the differences between the current state and the goal. Some of the best evi-

‘Actually, as my secretary points out, this leaves a somewhat exaggerated impression

of how much I am responsible for tidying my office. It appears that little elves do a
lot of the work while I am on business trips.

a2
 The Cognitive Stage

dence that distance reduction is important in human problem solving comes

from the difficulty subjects have in solving such puzzles. A good example is the
hobbits and orcs problem: ’

On one side of a river are three hobbits and three orcs. There is a
rowboat on their side, but only two creatures can row across at a
time. All of them want to get to the other side of the river. At no
point can orcs outnumber hobbits on either side of the river (or the
orcs would eat the outnumbered hobbits). The problem, then, is for
the creatures to find a method of rowing back and forth in the boat
such that they all eventually get across and the hobbits are never
outnumbered by the orcs.

Figure 9.7 illustrates a solution to this problem. It represents where the hobbits
(H) are, where the orcs (O) are, and where the boat (b) is relative to the river,
which is the line. The transition between state 6 and state 7 is critical. On the far
side of the river in state 6 there are two hobbits and two orcs, whereas in state
7 there is only one hobbit and one orc. This transition goes against the grain of
difference reduction, but it is absolutely critical to solving the problem. Subjects
have particular difficulty with this move and often give up finding a solution at
this point (e.g., see Greeno, 1974; and Jeffries, Polson, Razran, & Atwood, 1977).

y PHHHOOO
Kane

HHOO
b HO
i
bHHHOO
O
Y
HHH
bOOO

‘i
bHHHO

HO
b HHOO
bHHOO
HO
00
bHHHO
g b_000
HHH
g
ni bOOHHH

FIGURE 9.7 A diagram of the successive states in a solution to the 4, b 00

hobbits and orcs problem. Source: From Cognitive psychology and its OHHH
implications by John R. Anderson. Copyright © 1990 by W. H. '
Freeman and Company. Reprinted with permission. bHHHOOO

313
 CHAPTER 9 Skill Acquisition

Frequently, what makes a puzzle a puzzle is that it requires the problem solver
to temporarily abandon difference reduction.’
Difference reduction describes the approach to solving problems used by
almost all species. Even the simplest organisms have tropisms, which are ten-
dencies to approach various desired states. For instance, the wood louse (Gunn,
1937) continuously moves in the direction of moister areas because it will dehy-
drate if the air is too dry, and cockroaches flee light as a general defense mech-
anism. Organisms generally behave in a way that reduces the difference
between their current state and their goal state (“bliss point” in the language of
Chapter 4). To reiterate the theme from that earlier chapter, this tendency does
not mean that organisms consciously choose such operators (e.g., move from
light); it means only that they act as if they are choosing them.
Humans and other primates are capable of organizing their behavior in
ways that are more complex than difference reduction. This more complex
behavior is produced by operator subgoaling, the topic of the next section.

Organisms have a strong tendency to behave in a way that

reduces the differences between the current state and the goal
state.
EE URENRa RAR A APR BORED RGIS NS 2 SPE TOIT OR CSR

Operator Subgoaling
When humans set an operator subgoal, they suspend the attempt to achieve
their main goal and pursue the subgoal, which has no intrinsic value; its pursuit
is justified by the belief that it will help achieve the main goal. Most of the goals
people try to achieve, for example, good grades, are really subgoals in service of
higher goals, such as graduation from college, which in turn are in service of yet
higher goals. Tool building, a trait associated primarily with humans and to a
much lesser degree with higher primates, such as chimpanzees, is an exercise in
subgoaling. Creating a tool means creating an object whose justification is the
higher goals it helps achieve.
Tool building is almost unique to the human species. The only other species
that have been observed to engage in novel tool building to any significant extent
are the apes, particularly the chimpanzees. Chimpanzees have been observed to
make novel objects to serve as weapons, to shelter them from rain, and to reach
food (Beck, 1980). A clever episode (Kéhler, 1927) concerned a chimpanzee that
was trying to reach food outside its cage with two poles, each of which was too
short. Finally, the chimp fitted one pole inside the other and so made a compos-
ite pole long enough to reach the food. Figure 9.8 shows the chimpanzee at the
critical moment of insight. The chimpanzee already knew about sticks and reach-
ing for food from past experience and now put all its experiences together into a

*However, illustrating the general theme of skill acquisition, Greeno did find that
with repeated exposure to this problem students no longer found this move difficult.

314
 The Cognitive Stage

PR
ee a ak ie ie ee
5 OI EERGINONeee

FIGURE 9.8 Kohler’s

chimpanzee solving
the two-stick problem.

solution of a novel problem. Without the insight produced by this operator sub-
goaling, all of that past learning would have been useless.
It is significant that only the species closest to humans have been observed
to engage in novel tool building. This fact indicates that the capacity to learn to
handle goals and subgoals is a relatively species-specific skill, unlike many of the
phenomena reviewed in this book. The ability to engage in operator subgoaling
‘does not reflect, strictly speaking, a greater ability to learn; rather, it reflects a
greater ability to use what we have learned. Despite the accomplishments of
chimpanzees and other apes, humans far exceed them in the ability to manage
goal structures. It has been speculated that the process of handling goals is per-
formed by the frontal cortex of the brain (Anderson, 1993), a structure that is much
expanded in primates over most mammals and much expanded in humans over
primates. (For evidence of the critical role of the frontal cortex in working memo-
ry, see Chapter 5.) Memory for goals is a special kind of working memory.
A fair amount of research on the moment-to-moment dynamics of human
subgoal creation has been conducted using the Tower of Hanoi problem. (A
simple version of this problem is illustrated in Figure 9.9.) There are three pegs
and four disks of differing sizes. The disks have holes in them, so they can be
stacked on the pegs. The disks can be moved from any peg to any other peg.
Only the top disk on a peg can be moved, and it can never be placed on a small-
er disk. All the disks start out on peg A, but the goal is to move them all to peg
C, one disk at a time, by means of transferring disks among pegs.

315
 CHAPTER 9 Skill Acquisition

A B C A B C
Disk Disk
oes » <—
eae eae
nag = 4

FIGURE 9.9 The four-disk version of the Tower of Hanoi problem.

This problem can be mimicked with paper and coins. Draw three circles in
a row on a sheet of paper and place four coins (a quarter, a nickel, a penny, and
a dime) in order of sizein one circle. Your task is to move all of the coins to anoth-
er circle one at a time. The constraint is that you can never place a larger coin on
a smaller coin. This is an analogue of the Tower of Hanoi problem in which the
circles are the pegs and the coins are the disks. Try to solve this problem.
Table 9.1 attempts to illustrate the goals that have to be created to solve the
Tower of Hanoi problem using a subgoaling approach. At the beginning the
focus is on the biggest difference, which is to move disk 4 from peg A to peg C.
Disk 4 is blocked by disk 3, which is on top of it. A subgoal is created to move
disk 3 out of the way to peg B. But movement of disk 3 is blocked by disk 2, and
a subgoal is created to move it out of the way to peg C. To move disk 2, a subgoal
is created to move disk 1 out of the way to peg B. This goal can be achieved
directly by a move, and this is the first move made in Table 9.1. Before this move
could be made, four subgoals had to be created. After each move in Table 9.1, the
number of goals necessary before that move could occur is given in parentheses.
Subjects are often quite explicit about their subgoaling. Consider the fol-
lowing protocol of a subject (Neves, 1977) who was faced with the Tower of
Hanoi problem shown in Figure 9.10. This is the problem in an intermediate
state, with disks 1 and 2 moved off peg 1. The subject chose to move disk 1 to
peg 3, but before doing so gave the following justification of the choice.?

The 4 has to go to the 3. But the 3 is in the way. So you have to move
the 3 to the 2 post. The 1 is in the way there. So you move the 1 to
thers.

The subject began by setting the subgoal to remove the largest difference
between the goal and the current state—“The 4 has to go to the 3.”The opera-
tor to move this disk was blocked by the precondition that there could be noth-
ing on disk 4. The subject then saw a subgoal of getting disk 3 out of the way;
this was the second subgoal. To achieve this subgoal, the subject had to set a
third subgoal—to get disk 1 off peg 2. This third subgoal was actually governing
the move that the subject made. The second and third subgoals were operator
subgoals that would enable moves; in contrast, the first goal was a difference
reduction goal designed to get a disk to the target peg.

3 In this protocol the subject is using digits to refer to pegs as well as disks; see Figure
9:10;

316
 The Cognitive Stage

TABLE 9.1 Goals, Subgoals, and Moves in Solving the Tower of Hanoi Problem
(Moves are numbered)
Difference Reduction Goal: Move disk 4 to Peg C
Operator Subgoal: Move disk 3 out of the way to Peg B
Operator Subgoal: Move disk 2 out of the way to Peg C
Operator Subgoal: Move disk 1 out of the way to Peg B
Move disk 1 to Peg B (4 goals)
Move disk 2 to Peg C (0 goals)
Operator Subgoal: Move disk 1 out of the way to Peg C
Move disk 1 to Peg C (1 goal)
Move disk 3 to Peg B (0 goals)
Operator Subgoal: Move disk 2 out of the way to Peg B
Operator Subgoal: Move disk 1 out of the way to Peg A
Move disk 1 to Peg A (2 goals)
Move disk 2 to Peg B (0 goals)
Operator Subgoal: Move disk 1 out of the way to Peg B
Move disk 1 to Peg B (1 goal)
Move disk 4 to Peg C (0 goals)
Difference Reduction Goal: Move disk 3 to Peg C
Operator Subgoal: Move disk 2 out of the way to Peg A
Operator Subgoal: Move disk 1 out of the way to Peg C
Move disk 1 to Peg C (3 goals)
10. Move disk 2 to Peg A (0 goals)
Operator Subgoal: Move disk 1 out of way to Peg A
11, Move disk 1 to Peg A (1 goal)
12. Move disk 3 to Peg C (0 goals)
Difference Reduction Goal: Move disk 2 to Peg C
Operator Subgoal: Move disk 1 out of the way to Peg B
{keh Move disk 1 to Peg B (2 goals)
14. Move disk 2 to Peg C (0 goals)
Difference Reduction Goal: Move disk 1 to Peg C
a5: Move disk 1 to Peg C (0 goals)

Although Neves’s experiment gave verbal evidence that one subject might
have used subgoaling once, it did not address the issue of how prevalent this
strategy was in solving the Tower of Hanoi problem. Anderson, Kushmerick, and
Lebiere (1993) examined the problem solving of a large number of subjects and

Disk 3

FIGURE 9.10 The state of the Tower of Hanoi problem facing the subject whose
protocol was reported in Neves (1977).

317
 CHAPTER 9 Skill Acquisition

—e Goals
—= Sec

of
Number
goals

2 4 6 8 TO S12 4
Move
FIGURE 9.11 A comparison of the number of goals and latencies associated with
the steps of solving the Tower of Hanoi problem illustrated in Figure 9.8. The left axis
gives the number of goals, and the right axis gives latencies in seconds. (From
Anderson et al., 1993.)

determined how many subgoals subjects had to set before each move if they
were engaged in operator subgoaling. This number was inferred by assuming
that the subjects were using an optimal subgoaling strategy to solve the prob-
lem. Fifteen moves were required to solve the problem. For each move the num-
ber of subgoals was determined from Table 9.1, and the average amount of time
the subject took to make the move was calculated. Figure 9.11 shows the rela-
tionship between these two measures. The time to make a move strongly mir-
rored the number of subgoals that had to be set. It appears that goal setting is a
major determinant of problem-solving time.
The evidence for such goal structure involvement occurs only when we are
solving novel problems. As we practice the same problem over and over again,
we learn the solution and can simply retrieve it without going through the sub-
goaling to discover it. For instance, Ruiz.(1987) found that the latency peaks like
those in Figure 9.11 disappeared as subjects practiced solving the problem. The
next section, on the associative stage, is concerned with how we come to mem-
orize solutions to problems.
NOY SNE

Subjects set subgoals to enable them to apply operators that

will achieve the main goal.

318
 The Associative Stage

The Associative Stage

The second stage of skill acquisition is the associative stage, in which- people
stop using general problem-solving methods and start using methods specific
to the problem domain. The learning of domain-specific procedures is referred
to as proceduralization. The changes in skill performance as learners move to
the associative stage can-be quite dramatic. Neves and Anderson (Anderson,
1982) looked at the changes in application of knowledge of geometry, such as
that involved in the side-angle-side postulate. This postulate states that if two
sides and the included angle of one triangle are congruent to the corresponding
parts of another triangle, the triangles are congruent. Figure 9.12 illustrates the
first problem one student had to solve using this postulate. Below is the proto-
col of that student as he reasoned through how to apply the postulate.

If you looked at the side-angle-side postulate (long pause) well RK

and RJ could almost be (long pause) what the missing (long pause)
the missing side. I think somehow the side-angle-side postulate
works its way into here (long pause). Let’s see what it says:“Two sides
and the included angle.” What would I have to have to have two
sides? JS and KS are one of them. Then you could go back to RS = RS.
So that would bring up the side-angle-side postulate (long pause).
But where would Angle 1 and Angle 2 are right angles fit in (long
pause) wait I see how they work (long pause). JS is congruent to KS
(long pause) and with Angle 1 and Angle 2 are right angles that’s a
little problem (long pause). OK, what does it say—check it one more
time:”If two sides and the included angle of one triangle are congru-
ent to the corresponding parts.” So I have got to find the two sides
and the included angle. With the included angle you get Angle 1 and
Angle 2. I suppose (long pause) they are both right angles, which
means they are congruent to each other. My first side is JS is to KS.
And the next one is RS to RS. So these are the two sides. Yes, I think
it is the side-angle-side postulate. (Anderson, 1982, pp. 381-382)

Given: 21 and
22 are right angles
R S JS =Ks
Prove: ARSJ = ARSK

FIGURE 9.12 The first geometry proof problem that a student encounters that
requires the side-angle-side postulate. (From J. R. Anderson, 1982.)

319
 CHAPTER 9 Skill Acquisition

by, Given: 21 = 22
ip athe! 45E 56
BK =CK
2» Prove: AABK = ADCK
c

D be

FIGURE 9.13. The sixth geometry proof problem encountered after studying the
side-side-side and side-angle-side postulates. (From J. R. Anderson, 1982.)

After solving two more problems by means of side-angle-side (and two by side-
side-side), the student faced the more difficult problem illustrated in Figure
9.13. The method-recognition portion of the protocol follows:

Right off the top of my head I am going to take a guess at what Iam
supposed to do: Angle DCK is congruent to Angle ABK. There is only
one of two and the side-angle-side postulate is what they are getting
to. (Anderson, 1982, p. 382)

The contrast between these protocols is striking. The student no longer had to
verbally rehearse the postulate and search for correspondence between it and
specific pieces of the problem. Rather, the student simply recognized the applic-
ability of the rule. Because the application of the protocol switched to a pattern
recognition, there was no longer a need to hold information in a rehearsal
buffer, and so there were not the frequent failures of working memory when the
student lost track of what he was trying to do. This is part of what is involved in
reducing the cognitive component in a skill. Much of the effort that went into
recognizing an appropriate problem-solving operator (in this case side-angle-
side) has disappeared.

As students become more practiced in a skill, they come to rec-

ognize directly what they formerly had to think through.

The Conversion of Problem Solving into Retrieval

Logan (1988) has proposed that what is happening in skill acquisition is that
people are learning solutions to specific problems or parts of problems and no
longer have to go through the problem-solving process. He has argued that this
is the basic mechanism of skill acquisition, and he has shown that, under some
assumptions, it predicts the power law learning curves like those in Figures 9.4

320
 The Associative Stage

through 9.6. A study by Zbrodoff (1995—see also Rabinowitz and Goldberg,

1995) illustrates his ideas. She used an experimental task called alphabet arith-
metic introduced by Logan (Compton & Logan, 1991; Logan & Klapp, 1991). In
this task, subjects are asked to answer questions like what F + 2 is or what C +
4 is. Subjects are instructed to count that many letters forward in the alphabet.
Thus, F + 2 is H and C + 4 is G. Zbrodoff manipulated the size of the number
(the addend) from 2 to 4 and the amount of practice subjects had on these prob-
lems. Figure 9.14 shows her results. Initially, subjects take much longer to count
forward more letters. Thus, they take more than a second longer to answer a
problem like C + 4 than a problem like F + 2. This makes sense because they
have to count through more letters. They practiced a small number of these
problems over and over again. With practice on specific problems, they not only
get faster but they cease to show any effect of the addend. According to Logan’s
theory, subjects have come to memorize answers like F+ 2=HandC+4=G
and now are just retrieving them. They are faster because they do not count, and
they show no effect of the number of intervening letters because they are not
counting through them.
This general process of converting problem solving into retrieval seems a
very general characteristic of skill acquisition. Recently, researchers have
obtained evidence for the brain changes associated with this transition. Posner
(1994) has proposed that th late gyrus—a portion of the cor
that is buried wit
central executive in Figure 9.1. It has been shown to have greater activi
novel problem solving but to become inactive after practice. An experiment by
Raichle, Fiez, Vidden, MacLeod, Pardo, Fox, and Petersen (1994) illustrates this

—- 4 Addend
== 3 Addend
—e 2 Addend

(sec.)
Latency

FIGURE 9.14 Speedup of subjects in

alphabet arithmetic with practice. | L l |
Different functions are for different 1 2 2 2 3
addends. Block of Practice

321
 CHAPTER 9 Skill Acquisition

involvement. They asked subjects to repeatedly generate associates to a word

like apple, and they used brain imaging techniques to find out what areas of the
brain are active. Initially, there was greater activity in the anterior cingulate and
other frontal areas. This reflects the process of trying to seek out the associate.
However, as the word is repeated and the subject repeatedly gives the same
associate (perhaps peel), they can just retrieve their past answer. Now activation
is found in the more posterior parts of the brain.
Fincham, Schneider, & Anderson have recently found a similar pattern of
change in a more complex rule induction task. Subjects memorized facts like
“The two football games were played on Saturday at 3 and Sunday at 1.” Later,
we told subjects that these facts reflected’rules” for the timing of two games of
the sport. In the example above, the rule was that the second game had to be one
day later and two hours earlier. They were then asked questions such as, “If the
first game was Tuesday at 7 when is the second game?” The answer is Wednesday
at 5. Initially, subjects took about 25 seconds because they had to retrieve the
example, induce the relationship it reflected, and apply it to find the answer.
However, with practice they learned that football followed the”+ one day, — 2
hour” rule, and they just retrieved the rule and applied it without having to fig-
ure out what the rule was. Now they took under 10 seconds. Early on, there was
considerable anterior cingulate activation as they induced the relationship. As
they became practiced, the activation decreased in the anterior cingulate and
increased in the hippocampal regions of the brain. As we have noted in earlier
chapters, the hippocampus seems involved in the formation of new memories.

Activation moves from frontal to posterior regions of the brain

as practice changes skill performance from problem solving to
retrieval.

Production Rules
Numerous researchers (e.g., Anderson, 1983; Bovair, Kieras, & Polson, 1990;
Newell, 1991) have postulated that much of the knowledge underlying a cogni-
tive skill takes the form of what are called production rules. The following pro-
duction rule corresponds to the recognition of the applicability of the side-
angle-side postulate:
IF the goal is to prove that triangle 1 is congruent to triangle 2,
and triangle 1 has two sides and an included angle
that appear congruent to the two sides and included angle of
triangle 2
THEN set as subgoals to prove that the corresponding sides and angles
are congruent
and then to use the side-angle-side postulate to prove that
triangle 1 is congruent to triangle 2.

322
 The Associative Stage

A production rule contains a condition in its IF part, which specifies when the
rule is to apply. In its THEN part is an action, which specifies what to do in that
situation. Rules like this correspond to the basic steps of solving a problem, and
a complex cognitive skill involves many of these rules. As described in Chapter
11, the competence needed to do proofs in high school geometry involves many
hundreds of such rules.
One consequence of transforming knowledge into a production rule form
is that use of the knowledge becomes asymmetrical. Singley and Anderson
(1989) studied the development of asymmetry in rule application by looking at
the relationship between differentiation and integration in calculus. The exper-
iment involved calculus rules, such as the rule for powers:

y = ax" <—> a = pit

In differentiation, an equation such as that on the left-hand side is transformed

into an equation such as that on the right-hand side; integration moves from
the right-hand side to the left-hand side. Initially, subjects had equal difficulty
going in either direction. However, they were given practice in going in just one
direction and so created production rules to perform that transformation. This
practice improved the subjects’ performance in the practiced direction but did
not affect their performance in the unpracticed dimensions. Subjects had
acquired one-direction rules, such as:
IF the goal is to differentiate y = ax”
THEN write dy/dx = anx"-1.
which could not be reversed. DeKeyser (1997) has looked at a similar asymme-
try that develops in the learning of foreign languages. He has shown that if sub-
jects practice generating syntactic constructions, they gain expertise in generating
but not in understanding these constructions. On the other hand, if they prac-
tice understanding, they get good at that and not at generation.
Another case of learning production rules involves the formation of new
rules to capture repeated sequences of steps in the performance of a skill. For
instance, whenever I use the bank machine, I put in my card, type in my code,
hit Enter, and then hit Withdrawal. At one point these were four separate steps,
but they have been collapsed into a production rule of the form:
IF the goal is to get money from the bank machine
and I have put in my card
THEN type the code, hit Enter, and hit Withdrawal.
In the section on the autonomous stage of problem solving, we will discuss how
such action ee become ane into a motor ea
Ai ENRON

satel leasareAurideiiti bapaciioriypairsaha, arepomtalated

to tieealy intial sia
ee eee LS SSERREEA OSI ASE CLE IE RNP ERE RED

323
 CHAPTER 9 Skill Acquisition

The Knowledge-Intensive Nature of Skill

Underlying a complex skill are hundreds or thousands or even tens of thou-
sands of such rules. Simon and Gilmartin (1973) estimated that chess masters
have acquired on the order of 50,000 rules for playing chess. It has been esti-
mated that students who successfully master high school mathematics material
have to acquire somewhere between 1000 and 10,000 rules (Anderson, 1992). It
takes a great deal of time to achieve expertise in demanding domains. Hayes
(1985) studied geniuses in fields varying from music to science and found that
no one produced work*teflecting genius until after 10 years of work in a partic-
ular field.4 Contrary to popular opinion, genius is 90 percent perspiration and 10
percent inspiration. A great deal of knowledge has to be mastered to display
genius-level work, and mastering that knowledge takes a great deal of time.
Ericsson, Krampe, and Tesch-Romer (1993) have taken an extreme posi-
tion and have argued that there is nearly no contribution of innate talent to
expert performance and that expert performance is almost entirely a matter of
practice. To support their position, they quoted studies (Bloom, 1985a, 1985b) of
the histories of people who eventually became great in such fields as swimming
and music. Bloom found that at some point in childhood the parents of these
experts and geniuses decided that their children had special talents, even
though objective evidence would suggest that this was just wishful thinking on
the parents’ parts. The parents then expended enormous resources in time and
money on their children’s training. It is this effort in training that produced the
results, Ericsson and Krampe argued, not any initial talent. Ericsson (1994)
points out that what enabled Mozart to have such a dominant place in music
was the intensive education he received from his father, which was an unusual
practice at that time.
Ericsson and Krampe studied violinists at the Music Academy of West
Berlin and found that what determined the relative rankings of these violinists
was the amount of practice in their past. These researchers also found that the
best violinists were characterized by high-quality practice, such as practicing
when they were well rested. Ericsson and Krampe’s research does not really
establish the case that a great deal of practice is sufficient for great talent. It
seems more accurate to claim that extensive practice is a necessary but not a suf-
ficient condition for developing a great ability. There probably is an innate con-
tribution as well.
One of the interesting differences between experts and novices in a
domain is their ability to remember information about problems they solve in
that domain. De Groot (1965) found few differences between grand masters and
other chess players besides the quality of their game. Grand masters are not

“Frequently cited as an exception to this generalization is Mozart, who wrote his first
symphony when he was 8 years old. However, his early works are not of genius cal-
iber and are largely of historical value only. Schonberg (1970) claimed that Mozart’s
great works were produced after the twentieth year of his career.

324
 The Autonomous Stage

more generally intelligent by conventional measures of intelligence, do not con-

sider more moves, and do not appear to see further ahead in the game. The one
difference he did find concerned their ability to reproduce a chessboard after a
single glance. Grand masters are able to reconstruct more than 20 pieces from a
chessboard after viewing it for only 5 seconds, whereas novices can only recon-
struct 4 or 5 pieces. This result holds only when experts are shown actual board
positions from chess games. If they are shown random configurations of pieces,
their ability to reconstruct the board is no better than that of novices. Similar
abilities of experts to remember meaningful problem situations have been
shown in a large number of domains, including the game of Go (Reitman,
1976), electronic circuit diagrams (Egan & Schwartz, 1979), bridge hands
(Charness, 1979; Engle & Bukstel, 1978), and computer programming
(McKeithen, Reitman, Rueter, & Hirtle, 1981; Schneiderman, 1976).
It is speculated that the high memory performance of experts reflects the
acquisition of relevant patterns that they have encountered during their pro-
longed efforts in the domain. Chase and Simon (1973) showed that subjects
represented a chessboard in terms of game-relevant configurations, such as
pawn chains. Newell and Simon (1972) speculated that these patterns are effec-
tively the condition sides (the IF part) of productions. Experts recognize certain
patterns in a problem and have rules for responding to these patterns. In
essence, the good memory for problem states is reflecting the many production
rules that experts have acquired for behaving in these domains.
A similar conclusion was reached in efforts to develop artificial intelligence
programs that would reproduce various types of human expertise. Computer sys-
tems can perform medical diagnosis, decide where to drill oil wells, and config-
ure computers (Hayes-Roth, Waterman, & Lenat, 1983). These systems perform
their specialized tasks with the facility of the best practitioners in the fields. In
each case, a great deal of knowledge was learned from human experts and coded
into the computer systems to match human performance. Still, these systems
only match human performance in a narrow domain. To create a computer sys-
tem as widely intelligent as a human seems a truly overwhelming task. Millions
of facts and rules would have to be identified and coded into the computer.

People who develop a great talent have to invest an enormous

amount of time to acquire and perfect a great amount of
knowledge.

The Autonomous Stage

As people continue to practice a skill, its nature changes beyond the addition of
new rules and facts. These further changes belong to the autonomous stage of
skill acquisition, so named because the skill becomes more automatic, requiring
less attention and interfering less with other ongoing tasks. When people learn

325
 CHAPTER 9 Skill Acquisition

to drive, initially they need to pay all their attention to driving and they are
unable to maintain a conversation. As they become more practiced, they are able
to maintain a conversation while driving. The driving skill becomes so automat-
ic that it seems to require no attention at all, at least when driving conditions are
not demanding. People report driving miles on a highway without recalling any-
thing of what they did. )
Spelke, Hirst, and Neisser (1976) reported an interesting experimental
demonstration of the growth of automaticity. Their subjects performed an odd pair
of tasks simultaneously. The first task was to read a text for comprehension. The
second was to transcribé\(without looking at what they were writing) material that
was being spoken. At first, subjects found this a difficult combination of tasks. Their
comprehension of the text suffered dramatically. The rate at which they read also
slowed down dramatically. Over a six-week period they gradually improved their
speed of reading until they were reading at preexperimental speeds without any
loss of comprehension. However, they showed almost no ability to remember what
they were transcribing, just as someone loses memory for what happens on the
road if engrossed in conversation. Spelke et al. were able to show that, with even
further practice, subjects could also remember what they were transcribing.
Two key features characterize a skill when it becomes totally automatized:
it can be performed without engaging the cognitive system, freeing the person
to pursue other cognitive goals in a dual-task situation; and it becomes less
interruptible. When learning to shift gears, for instance, a driver can choose to
stop at any point in the process, but once the driver is proficient, the shifts are
produced in one fluid motion and it is difficult to stop at a particular point. Skills
develop these two features because more and more of the skill becomes imple-
mented as a motor program and less and less is performed at a cognitive level.
The next section discusses motor programs.
Shin PARTE LETTS

As a skill develops in the autonomous stage, it requires less

attention but is harder to interrupt.

The Motor Program

A motor program is a prepackaged sequence of actions. Good examples are a
person’s signing his or her own name or a skilled typist’s typing of “the.”A
novice typing“ the” will be observed to find the t, strike it, check that the correct
result was produced, find the h, strike, check, and then do the same for the e.
None of this sequentiality is observed in the skilled typist. While the left index
figure is going to the t, the right index finger is already going to the h. The exe-
cution of one keystroke is not waiting for feedback on the completion of the
execution of the other keystroke. If the execution of the ¢is blocked by a stuck
key, the h follows without alteration.
A distinction is made within the motor performance literature between
open-loop performance and closed-loop performance. A closed-loop system

326
 The Autonomous Stage

waits for feedback from one action before taking the next action. A thermostat
that controls a furnace is an example. After turning the furnace on, it waits until
the temperature reaches the target level before turning the furnace off. An open-
loop system executes a fixed sequence of actions without checking to see that
the earlier actions achieved their intended effects. Old-fashioned copiers
worked this way; they continued to feed paper after a jam had occurred. Modern
copiers are closed loop and sense whether there is a paper jam.
Motor programs are open-loop segments of behavior. They are typically
embedded in a larger, closed-loop structure, as in the case of a typist who may
encode a familiar sequence of letters, type them in a closed-loop fashion, check,
encode, execute, and so on. Also, what might be open-loop at a higher neural
level is often closed-loop at a lower neural level. For instance, this section
reviews the evidence that the cortex issues open-loop instructions to the effect
that a hand should push and does not provide further monitoring. Closed-loop
circuits at the level of the spinal cord monitor the execution of these commands.
Schmidt (1988b) cited three lines of evidence for the existence of open-
loop motor programs at the cortical level. One is the slowness of closed-loop
behavior. It takes about 200 msec for information to be perceived from the envi-
ronment and registered in the cortex and for an appropriate reaction to be
taken. (This estimate comes from the fact that the shortest simple reaction time
is about 200 msec.) People are capable of executing actions much faster. Skilled
pianists can perform as many as 16 finger movements per sec. There is no time
for the person to sense the result of one movement before executing the next.
Schmidt’s second argument is that movements appear to be planned in
advance. As the complexity of the movement increases, it takes longer to initi-
ate the movement. If a typist is shown a single word and has to type it, the time
from the presentation of the word to the first keystroke increases with the
length of the word (Salthouse, 1985, 1986). The delay that increases with pro-
gram complexity reflects the time needed to prepare the open-loop program.
The third argument comes from the results of deafferentation studies of
monkeys (e.g., Taub & Berman, 1968). The deafferentation procedure eliminates
sensory input by cutting through the dorsal roots of the spinal cords. It does not
affect motor signals to the effectors. This procedure creates an organism that can
move its limbs but has lost all sensory feedback from them. Such animals are
still capable of learning to perform complex actions and then performing them
in the dark (lights out) so that they cannot receive any visual input to guide the
limbs. These animals receive no sensory feedback, and yet their limbs execute
learned sequences, such as moving a hand to a lever.
There is evidence that at a certain point the instructions in the motor pro-
gram are sent to the effectors (muscles) and the response cannot be stopped.
Skilled typists type the next few characters after being told to stop (Salthouse,
1985, 1986). In a study by Slater-Hammel (1960), subjects watched the hand of
a sweep timer that made one revolution per second. They were supposed to stop
the movement by lifting a finger from a key when the hand reached a certain
position. To do this successfully, they had to send a signal to their hand in antic-

327
 CHAPTER 9 Skill Acquisition

ipation of it reaching the target location. Occasionally, the timer hand stopped
before it reached the target position, and the subjects were to inhibit lifting their
fingers. If the timer hand stopped at least 250 msec before reaching the target
position, the subjects were able to stop their movement, but when it stopped
less than 150 msec before the target, the subjects were unable to stop lifting
their fingers. Subjects reported that they saw the clock stop and their hands
responded anyway, as if they had no control over their hands.
Global control of behavior is a closed-loop routine that calls many open-loop
motor programs. A tennis player responds to where the ball is by positioning the
body and choosing the ténnis stroke. Once the stroke begins its execution, it large-
ly runs in an open loop. A person composing at a typewriter may deliberately
choose a word and then type it in an open-loop manner. The closed-loop phase
involves the more deliberative processes of the cognitive and associative stages.
What has been totally routinized gets packaged into open-loop motor programs.
Successful performance depends on being able to assign much of the behavior to
these open-loop segments that do not require cognitive monitoring. For instance,
successful tennis is very much a game of strategy in terms of positioning on the
court and choosing where to place the shots. Professional players can focus on this
strategy because the actual process of executing a tennis shot has been automated.

Motor programs are open-loop segments of behavior that are

performed without cognitive control.

Noncognitive Control
There can be nonconscious control over the execution of a behavior. As a skill
develops, more and more of the control shifts to this nonconscious level. A good
fraction (but hardly all) of this nonconscious control is performed by neural
structures that are below the cortex. In one experiment, Dewhurst (1967) had
subjects hold a light weight at a particular angle. When the weight ddenly
changed, compensating activity could be rec in the muscles j Sgc
atter the change. [his activity was initiated in the spinal cord where the senso-
ry neurons synapse onto the motor neurons.
— = : 4
The spinal cord”knew”
e to be held in a certain position and began to take compensato
action as soon as a change occurred.
Other motor controf takes € above the spinal cord but still at a sub-
conscious level. This control occurs in-both the cortex and the cerebellum. The
cerebellum in the brain stem is particularly important in motor control (see
Figure 1.15). The Dewhurst study demonstrated another compensating reflex to
pressure changes that took about 80 msec. Unlike the spinal reflex, this reflex
was to some degree instructable. If the person was told to let go when there was
increased pressure, there was no compensatory response at 80 msec, but the
spinal 30 msec reflex still occurred. This 80 msec is still much quicker than the
200 msec for conscious reaction time.

328
 The Autonomous Stage

Schmidt called the 80-msec response the long-loop response. It is respon-

sible for performing much of the microstructure of intentional action. As an
interesting example, some objects, such as wineglasses, tend to slip through the
hand and require compensatory increases of pressure to hold them in place.
Johansson and Westling (1984) showed that when an object began to slip, there
was a compensating increase in pressure 80 msec later. Subjects were quite
unaware of the fact that their hands were making these adjustments.
Subcortical mechanisms control many complex aspects of behavior. This
control has been displayed in research (Shik, Severin, & Orlovskii, 1966) on cats
whose midbrains were severed such that the cerebellum was still connected to the
spinal cord. Although the cats could no longer exercise cortical control over their
behavior (since the cortex was severed from the spinal cord), they were nonethe-
less capable of displaying coordinated walking patterns on a treadmill. As the
treadmill sped up, they shifted their pattern from a walk to a trot to a gallop. These
cats also shook a paw violently to get rid of a piece of tape placed on it. If placed
on the treadmill with tape on a paw, they displayed a coordinated sequence of
shaking the paw and walking. These experiments show that complex and coordi-
nated pieces of behavior can be performed without any cortical involvement.

The detailed execution of motor programs can be guided by

short-latency control processes in the spinal cord and cerebel-
lum and by nonconscious cortical structures.

Generality of Motor Programs

Schmidt also argued that motor programs are general, not specific, sequences of
behavior. Consider the writing examples in Figure 9.15 from Raibert (1977). In
the first case, the writing is normal with the right hand; in the second case, with
the wrist immobilized; in the third case, with the left hand; in the fourth case,
with the pen gripped in the teeth; and in the last case, with the pen taped to the
foot. Not only was Raibert able to write in each case, but the writing preserved
certain invariant features of his style, such as the curl on the top of the capital E.
It appears that the same motor program is being executed in each case. Thus,
when one learns a motor skill such as handwriting, one is not learning a specif-
ic set of motor actions, but a general motor program that can be executed by
many effectors.
Schmidt suggested that the general program can be executed by different
effectors (e.g., hands, mouth, feet) and with different parameters. Among the
critical parameters that can be varied are the force and timing of the behavior. A
person can write in large strokes by increasing the force of the movements, and
a person can slow down the rate of writing. Schmidt noted that these changes
tend to be proportionate: if writing a signature is slowed by 50 percent, all com-
ponents of the signature are slowed by about the same amount; if the force is
increased, all letters show the same magnification.

320
 CHAPTER 9 Skill Acquisition

Na n 0S Saeco end 6fae

yoo Winevaded

» Ob word ved row CU

l A sew Ee,
c AWe worom
oO CORE
Coe -siarammeebunae r
© Oke wand a0 vaur he,
FIGURE 9.15 Raibert’s attempts at writing: (A) with right hand; (B) with wrist
immobilized; (C) with left hand; (D) with teeth; (E) with foot. Source: From Motor con-
trol and learning by the state-space model by M. H. Raibert, 1977, Technical Report,
Artificial Intelligence Laboratory, MIT (AI-TR-439), p. 50. Copyright © 1977 by M. H.
Raibert. Reprinted by permission.

Rosenbaum, Inhoff, and Gordon (1984) reported an interesting example of

the generality of motor programs. College students were asked to perform a fin-
ger-tapping sequence. For instance, subjects might have to tap twice the index
finger on their left hand and then their left middle finger. They were responsi-
ble for two sequences—one for the left hand and one for the right hand. They
were faster if both sequences involved the corresponding fingers on both
hands—for example, they were faster if they had to tap index, index, middle on
both hands than if they had to tap index, index, middle on the left and index, mid-
dle, index on the right. If both sequences were alike, the students had to hold in
mind only one motor program, which they could send to either hand. This result
illustrates how different effectors (in this case, the two hands) can execute the
same motor program.

A motor program has certain parameters associated with it

that allow it to be executed with different speed, with different
force, and by different effectors.

Learning of Motor Programs

How are such motor programs acquired? Keele (discussed in Schmidt, 1988b)
suggested that new motor programs are generated by stringing together small-
er units of behavior. Figure 9.16 illustrates his proposal for the development of

330
 The Autonomous Stage

feo]
i
Ss
00
RD
x n
eh
§ s 2 s&
Seb dy tee,oS 153
~ x x

ee
8 oe Ee a he
O.) Gave ue) oo
cayactce—>
(1)(2)(2)(4)(5)(6)
Ac

~)
Celery,
ator
dow,

Middle practice —> 3 4

Late practice —> 1 2 3 4 5 6 7

FIGURE 9.16 Keele’s proposal for the process by which individual components of
the gearshift change become composed into a single production. Source: From R. A.
Schmidt, motor control and learning: a behavioral emphasis, Third Edition (p. 477)
Champaign, IL: Human Kinetics, copyright’© 1999. Reprinted by permission of
Richard A. Schmidt and T. D. Lee.

the skill of shifting gears. The process starts out as individual actions, such as
lifting the foot, which the person presumably already knows. Eventually, these
actions become packaged into an overall behavior, which is shifting gears.
It seems that different regions of the brain are involved as the performance
of a sequence of actions becomes organized into a motor plan. Jenkins, Brooks,
Nixon, Frackowiak, and Passingham (1994) studied subjects learning to execute
a sequence of eight button presses and the researchers imaged which regions of
the brain were active. Figure 9.17 illustrates the results. Early in learning the lat-
eral prefrontal area, associated with planning in general, and the posterior parietal
cortex, associated with motor planning (Andersen, 1995) in particular, were
highly active. However, with practice the more active areas became the supple-
mental motor area, which is responsible for guidance of action, and the hip-
pocampus, which is responsible for retrieval of memories. It appears that the
sequence no longer has to be planned but rather can be retrieved and directly
executed.

With practice, sequences of actions become bundled into motor

programs that can be executed without planning.

Tuning of Motor Program: Schema Theory

A critical issue in developing a motor program is learning to properly tune it.
How do we learn just the right force and angle to use in making a basketball
shot? How can we generalize that knowledge to different locations on the bas-
ketball court? Adams (1971) developed a theory of such learning, which was

331
 CHAPTER 9 Skill Acquisition

Premotor
Supplemental cortex Parietal
motor area cortex

Occipital
regions

Lateral
prefrontal
cortex

Hippocampus

Cerebellum

New > Old a

old > New fi
FIGURE 9.17 Results of positron emission tomography (PET) scans obtained when
the performance of the sequence was novel (new) versus when it was practiced (old).
Source: From M. S. Gazzaniga et al., “Cognitive neuroscience: The biology of the
mind,” Journal of Neuroscience. Copyright © 1994 Journal of Neuroscience. Reprinted by
permission.

elaborated by Schmidt (1988b) in his schema theory. Schmidt holds that the
learner develops two representations of the skill. One, called the recall memory,
is the motor program itself—a prepackaged sequence of actions. The second,
called the recognition memory, is a representation of the desired outcome of the
action in terms of both the response-produced feedback and the external sen-
sory consequences. A player taking a basketball shot can compare the outcome
with the ideal (recognition memory) and adjust the motor program (recall
memory) appropriately.
Schmidt emphasizes that neither the recall memory nor the recognition
memory is for a specific action but rather is for a class of actions. Different
actions can be achieved from the same motor program by evoking it with dif-
ferent parameters. A person can throw a ball a novel distance, having been
trained on specific distances, by extrapolating the forces used for the training
distances to the force needed for the new distance. In numerous studies, sub-
jects have been trained to perform a skill that involves different positions or to
react to objects at different speeds (see Shapiro & Schmidt, 1982, for a review).
People show considerably greater success in extrapolating to new values if they
have practiced with a variety of values.

332
 The Autonomous Stage

In a study on variability of practice, Catalano and Kleiner (1984) had sub-

jects press a button when a moving target reached a certain point. The moving
object could be traveling at one of a number of speeds (5, 7, 9, or 11 mph). The
constant group was trained on just one of these speeds (different subjects, dif-
ferent speeds), and the variable group had practiced on all four speeds. Then
subjects were transferred to novel speeds outside the previous range (1, 3, 13, or
15 mph). During original training, subjects were less accurate in the variable
training condition (52 vs. 38 msec error), but in the transfer task they were more
accurate (49 vs. 60 msec error). The constant training conditioning was easier
because it involved adapting to only one value, but it did not prepare subjects as
well for dealing with novel values. As Schmidt and Bjork (1992) noted, these
results generalize to verbal tasks. For instance, it is easier to learn the meaning
of a new word if it is used in a number of contexts.
Koh and Meyer (1991) conducted one study of how subjects extrapolate
response values. At the beginning of each trial, subjects saw two vertical bars sep-
arated by various distances. Subjects were to make two taps separated by a pause,
the duration of which was determined by the distance. Figure 9.18 illustrates the
true function that Koh and Meyer wanted subjects to learn. Subjects were tested
with 12 different stimulus distances, but they received information on the correct
duration for only the outer 8 stimuli. It was up to the subjects to assign durations
to the middle four distances; they did not receive feedback as to what durations
were correct. Subjects were given five 1-hr training sessions. Figure 9.18 presents
data from the first and fifth sessions, including subjects’ responses for the middle
four values, on which they received no feedback. Subjects were somewhat more
accurate in the fifth session than in the first, but in all sessions they were fairly
accurate at extrapolating responses to the untrained values.

1200

1000

FIGURE 9.18 Duration of respon-

ses to stimuli in session 1 and ses-
sion 5 of the experiment. The solid x/ — True function
line represents the true function. Time,
msec
600 |— x -o- Session 1
The x’s denote values for which
-o Session 5
subjects received no feedback. L
Source: From G. Wulf, R. A.
Schmidt, and H. Deubel. Journal of 400 |—
Experimental Psychology: Learning,
Memory, and Cognition. Copyright
© 1993 by the American Psycho- 200 PC Cioey Nee Dh Sat
logical Association, pp. 1136 and 0 20 40 60 80
1143. Reprinted by permission. Length, mm
 CHAPTER 9 Skill Acquisition

Schmidt’s theory holds that the recall memory is improved by comparing

the action produced with an internal standard in the recognition memory of
what the,gition should be like. This position implies that subjects should be
capable of detecting errors in their actions without any external feedback. For
instance, Schmidt and White (1972) looked-at a ballistic timing task in which
subjects were to move a slide 23 cm, taking as close to 150 msec as possible.
After each movement and before being told what their actual error was, subjects
were asked to estimate how far off the timing of the move was. Subjects’ esti-
mates of their errors were quite accurate. The correlations between the actual
errors and the estimated errors approached 1 (perfect) after two days of practice.

We learn what thé appropriate behavior of a motor program

should be, and we use this knowledge to correct the program.

The Role of Feedback

A critical issue is how the learner takes advantage of feedback to tune the motor
programs. Some feedback is necessary for learning, but is more feedback always
better? Detailed and immediate feedback after every attempt might be expect-
ed to produce the best learning results, but the research indicates otherwise.
Bilodeau and Bilodeau (1958) looked at subjects learning to turn a knob to a tar-
get position. They varied the probability that a trial would be followed by feed-
back on magnitude of error from 10 percent to 100 percent. There was no dif-
ference in amount learned as a function of percentage of feedback in training.
Other studies (e.g., Ho & Shea, 1978; Schmidt & Shapiro, 1986) have looked at
what happens when subjects are retested at a delay but with no feedback pro-
vided during the retest. The group that received the lowest frequency feedback
during training often performed best at a delay. Salmoni, Schmidt, and Walter
(1984) speculated that this result occurs because subjects with constant feed-
back come to rely too much on it and cannot perform without the feedback.
Also, processing the feedback may disrupt learning.
Schmidt, Shapiro, Winstein, Young, and Swinnen (1987) looked at a task
that involved intercepting a pattern of moving lights by an arm movement
(something like hitting a ball with a bat). They gave subjects feedback about the
error in their movement after every trial, or information about average error
after 5, 10, or 15 trials. Subjects who received constant feedback did best during
the training. However, when tested in a situation with no feedback, subjects
who had received summaries after every five trials did best. Thus, intermittent
feedback proved better than continuous feedback.
According to Schmidt, learning a skill does not depend on correcting the
motor program with some external result. Rather, the motor program must be
corrected with respect to an internal representation of the skill—Schmidt’s
recognition memory. Whether or not there is external feedback, there is always

334
 The Autonomous Stage

an internal representation, and the motor program is corrected after each trial.
It is not necessary to have feedback after every trial to build up this internal rep-
resentation of what the skill is like. Occasional information about how the per-
formance is progressing is sufficient to update the representation of the desired
behavior.
Is it better to give only intermittent feedback for other learning tasks
besides motor skills? What about academic learning tasks? As Schmidt and
Bjork (1992) lamented, there has not been a great deal of research on this topic.
Schooler and Anderson (1990) did one study of the acquisition of computer pro-
gramming skill, which showed that at least sometimes feedback disrupts the
learning. While students try to understand the feedback, they lose track of
where they are in the problem.
Wulf, Schmidt, and Deubel (1993) pointed out that learning a motor pro-
gram really involves two components: learning the general structure of the pro-
gram and learning how to parameterize the program. They looked at subjects
making the sinusoidal movements illustrated in Figure 9.19. All the movements
in Figure 9.19a illustrate the same general program. The only difference is the
timing of the up and down movements. Similarly, all the movements in Figure
9.19b reflect the same program and vary only in the force with which the up and
down movements are made. Wulf et al. contrasted intermittent feedback (63

70 80

)
oD
oO
no}

oO
no)
ee
=

27
Amplitude,
degrees
&y7®

12 10

0 908
¢)
P bese, Time, msec
Time, msec

(a) (b)

FIGURE 9.19 (a) A set of actions reflecting the same motor routine but varying in
timing; (b) a set of actions reflecting the same motor routine but varying in force.
Source: From G. Wulf, R. A. Schmidt, and H. Deubel. Reduced feedback frequency
enhances generalized motor program learning but not parameterization learning.
Journal of Experimental Psychology: Learning, Memory, and Cognition, Volume 19.
Copyright © 1993 by the American Psychological Association. Reprinted by permis-
sion.

335
 CHAPTER 9 Skill Acquisition

A subject tracking a signal like that in Figure 9.19.

percent of the time) with constant feedback (100 percent of the time). They
developed separate measures of whether the subjects had learned the general
pattern of the movement and whether they had learned the precise timing and
force of one of these movements. They found that learning the general move-
ment was better with intermittent feedback. However, learning the timing and
the force was as good, and even a little better, with constant feedback.

Only occasional feedback is sufficient to tune the internal rep-

resentation of the skill, and too frequent feedback can be dis-
ruptive.

Final Reflections
This chapter has followed the process of skill learning from its initial organiza-
tion in the first performances to the point at which the detail of the performance
of a task is lost to cognition and becomes embedded in motor programs. The
production rule is the key construct unifying the course of skill acquisition.
Production rules embody the organization placed on the cognitive skills by the
problem-solving processes during the cognitive stage. The development of the

336
 Further Readings

skill in the associative stage can be decomposed into the learning histories of
the many component production rules. A motor program is essentially the
action (or THEN) part of a production rule, and its learning in the autonomous
stage amounts to the fine-tuning of a production rule.
Several studies of transfer among skills indicate that the degree of trans-
fer is a function of the overlap between the skills in terms of production rules.
As a person becomes more advanced in a domain, the person acquires addi-
tional production rules that take advantage of the special characteristics of that
domain. One consequence of this specialization of skills is that there is less and
less transfer among skills as the skills become more advanced (Henry, 1968).
The picture of learning from the skill-acquisition literature is much more
complex than the picture painted in the early chapters of animal learning and
human memory. Skill acquisition is concerned with behaviors of true signifi-
cance that develop over scales of time more typical of learning outside the lab-
oratory. Problem-solving organization has a role in human (and probably pri-
mate) skill acquisition that was not apparent in the simpler laboratory studies of
learning. However, the processes of learning and strengthening apply to the
component production rules much as they apply to the learning seen in the sim-
pler laboratory studies.

The course of skill acquisition is determined by the learning

history of the individual productions that make up the skill.

Further Readings
The book by Newell and Simon (1972) remains a classic on problem solving. The
artificial intelligence perspective on problem solving can be found in Russell &
Norvig (1995). Research on cognitive skill acquisition is reviewed in Anderson
(1990) and Van Lehn (1989). Singley and Anderson (1989) review production
rule theories of skill acquisition and transfer with a particular discussion of text
editing. Klahr, Langley, and Neches (1987) review production system theories of
skill acquisition. Newell (1991) describes his SOAR theory, and the ACT theory
is described in Anderson and Lebiere (1998). Rosenbaum (1991) and Schmidt
(1988b, 1991) discuss motor performance and motor learning, and Schmidt
includes an exposition of his schema theory. A comparison of motor learning
and verbal learning is found in the work of Schmidt and Bjork (1992). The sec-
ond half of the book edited by Osherson, Kosslyn, and Hollerbach (1990) con-
sists of a series of review chapters on motor behavior and motor programs.

337
 Inductive Learning

Overview
Most laboratory research on human learning has studied learning in a setting
that is much like a traditional classroom. Subjects are supposed to learn basic
facts or skills and are trained until these facts or skills are mastered. This sort of
paradigm is useful for understanding the principles of memory, but it ignores a
critical component of learning—determining just what it is that should be
learned. This component looms large in the conditioning experiments where the
organism has to determine what controls the appearance of a reinforcer. Figures
5.1 and 9.1 show that a critical inductive component determines how experience
gets represented in memory. Induction refers to the process by which the sys-
tem makes probable inferences about the environment on the basis of experi-
ence. For instance, in a typical conditioning experiment the inductive compo-
nent figures out what caused what. In contrast, the need for this process is large-
ly bypassed by the use of instructions in a memory experiment.
Much of human learning avoids the need for induction because many fea-
tures of our environment are understood and we can be directly instructed on
them. Still, a significant fraction of human learning involves induction. We have
to figure out what things annoy or please an acquaintance. We have to figure out
how to operate many appliances without direct instruction. Children manage a
great deal of learning without instruction. For instance, they figure out which
animals are dogs and which are cats without being told what makes an animal
a dog or a cat. Perhaps most impressive, they learn to speak their first language
without direct instruction.
Learning the structure and rules of a particular domain without direct
instruction is referred to as inductive learning. Inductive learning involves
making uncertain inferences from experience. Suppose that you come upon a
microwave oven and press 1 followed by start on its button panel. You observe
that the oven runs for exactly 60 sec. You might make the inference that press-

338
 Overview

ing the 1 caused it to run for 1 min and that pressing the 2 would cause it to run
for 2 min. This conclusion is an inductive inference. It may not be correct—the
1 may control the power level at which the oven operates—but supposé your
inference is correct. Now you want to run the oven for 10 min, but there are only
buttons with the digits 0 through 9. What do you do? You might infer that press-
ing a 1 followed by a 0 would achieve the goal. This is another inductive infer-
ence. What if you want the device to operate for only 30 sec? You might notice
a button labeled seconds and infer that pressing seconds before pressing 3 and
0 would yield the desired effect. What if you want to have it run 2 min and 20
sec? What if you want it to run at half intensity? By a process of hypothesis and
test, you would probably come to an understanding of how the device functions.
The total episode is an example of inductive learning.
Philosophers make a contrast between induction and deduction. In
deduction, inferences are logically certain, whereas in induction they are not. If
told that an animal is a poodle, a person can infer with certainty that the animal
is a dog. This type of inference is a deductive inference because all poodles are
dogs. If an animal is heard barking, a person might also infer that the animal is
a dog. This is an inductive inference, because it is conceivable that some other
animal might bark. As this example makes clear, inductive inference really adds
something to knowledge and so counts as a kind of learning. In contrast, there
is a sense in which deductive inference adds nothing but only makes explicit
what is already known.
An element of inductive learning is involved in nearly every learning situ-
ation—even direct instruction. A tennis coach illustrates how to hold the racket
for a two-handed backhand but leaves the player to figure out which aspects of
the demonstration are the critical aspects. A geometry teacher works through
the steps of a proof and usually does not tell the students why an inference is
made at one point rather than another—the student must figure it out. In
English class a teacher marks a sentence as awkward—but what aspect of the
sentence is awkward?
Much of what the earlier chapters reviewed about conditioning was con-
cerned with inductive learning. Here we return to these earlier issues but focus
more on human inductive learning. The chapter addresses the topics of concept
formation, causal inference, and language acquisition, three types of human
inductive learning that have received much research. It reveals that the induc-
tive learning required of people can be very tricky indeed. This chapter also
reviews the arguments that it would be impossible for children to learn a lan-
guage unless they were born knowing a great deal about language already and
that language acquisition is a uniquely human ability.

_
EREDAR NIRA ER IES AMAL
ESEAN REN ELLIE wets

Inductive lectin sopines Ln prereninHnrereriees ae

go Beyorte0our direct Eee
RALH RAS RC RE LR RANEY IIE ENT AROSE INN

339
 CHAPTER 10 Inductive Learning

Concept Acquisition
Concept acquisition research is concerned with how we learn natural cate-
gories, such as“ dog,” “chair,” “car,” and “tree.” It is not particularly obvious how
children learn what separates a dog from a cat or a chair from a table. Adults
have difficulty articulating the difference between these concepts—how do chil-
dren figure it out? The mystery of this learning process is intensified by the fact
that most of it happens in childhood when the learners are not particularly
articulate. However, concept learning does extend to adulthood. When I visited
Australia, I learned new animal concepts, such as “echidna” and “kookaburra,”
and I learned to recognize a number of categories of birds for which I never did
learn the verbal labels. It is no more clear to me how I learned to recognize these
animals than it is to a child how he or she comes to recognize a dog. However,
researchers have used a number of different approaches to shed light on what
is involved in learning a new concept. Much of this research has been done with
adult subjects.
A major issue in this research concerns the degree to which concept acqui-
sition is like the associative learning characteristic of conditioning experiments.
An early experiment by Hull (1920) suggested that concept learning is like con-
ditioning. His subjects learned to categorize different Chinese alphabet charac-
ters, such as those shown in Figure 10.1; each row reflects a different concept

Radical
(concept) List 1 Liste2 List 3 List 4 List 5 List 6

SRVRS &ES
B
UdOF
Dw
Say
Ws SynN SS
Qe
FY
ch
Sh&Se
OR
ny
SS Pe Ba
RY
Vt
eo
me
FIGURE 10.1 Example of stimulus material used by Hull (1920). Each row repre-
ey a category defined by the presence of the Chinese radical. Source: From C. L.
ull. Quantitative aspects of the evolution of concepts: An experimental study. Copyright
© 1920 in the Public Domain. ‘ 4 ee

340
 Concept Acquisition

defined by the presence of a Chinese radical. Subjects were not informed about
the critical feature. Gradually, they learned how to classify the stimuli, but they
were quite incapable of saying what they did to classify these stimuli. Hull con-
cluded that concepts are learned by simple associative learning.
Since Hull’s research, the field has vacillated in terms of how to think about
the nature of human concept learning, moving full circle from Hull’s position to
regarding human concept learning as dramatically opposed to associative learn-
ing, back to seeing much in common between the two types of learning.

A key issue has been whether human concept learning can be

understood in terms of simple associative learning.

Concept-Identification Studies
A new view about human concept learning began with a classic series of exper-
iments by Bruner, Goodnow, and Austin (1956). Figure 10.2 illustrates the kind
of material they used. Eighty-one stimuli varied along four possible dimensions:
number of objects (one, two, or three); number of borders (one, two, or three);
shape of objects (cross, circle, or square); and color of objects (green, black, or
red—shown as white, black and grey). There were three possible values on each
of the four dimensions and thus 3 x 3 x 3 x 3 = 81 possible ob =ects. Subjects were

Hig
8®@
3+] [sll°
ap oP oP ©©O

EEG
=

AIC
FEERIEEEE
lelSET (EEE
eteielel
oP
EIEIEIE)
8)0)
60/8] oO

ot ot at
HiB/o
QEIERLEEE
OOO O OO ap ap ¢F ©@@

FIGURE 10.2 Material used by Bruner et al. (1956) to study concept identification.
Source: From J. S. Bruner, J. J. Goodnow, and G. A. Austin. A study of thinking.
Copyright © 1956 by J. S. Bruner et al., p. 42. Reprinted by permission...

341
 CHAPTER 10 Inductive Learning

told that there was some concept that referred to a specific subset of the objects,
for instance, all green squares. Subjects were instructed to discover the concept
and were shown various stimuli identified as members or not members of the
concept. When they thought they knew what the concept was, they could
announce the concept.
Given the research of Hull and others, psychologists had thought of con-
cept learning as involving the simple associative learning processes discussed in
the conditioning chapters. Thus, for the concept of “green squares,” “green” and
“square” would gradually get associated to the category. Bruner et al. and sub-
sequent researchers fottnd evidence that subjects engaged in conscious hypoth-
esis testing. This research helped fuel the cognitive revolution against the pre-
vailing behaviorist paradigm. Before describing the results of this research, it is
important to identify its methodological features.
The Bruner et al. paradigm is somewhat different from concept learning in
real life. For instance, people normally do not suddenly announce that they have
a new concept. A person displays knowledge of the concept “dog” by success-
fully classifying new instances as dogs. Research subsequent to that of Bruner et
al. has used successful classification behavior as evidence for knowing the con-
cept; that is, if the subjects could correctly categorize new instances, they were
credited with having the concept.
Another dimension of the experimental design is whether subjects select
instances to get information about or whether they receive a series of instances
classified for them. The former paradigm is called the selection paradigm, and
the latter the reception paradigm. Subjects in a selection paradigm can behave
more like a scientist and select instances to test their current hypothesis about
what the concept is. Learning in the reception paradigm is more like learning
concepts in the real world, where we encounter instances and noninstances of
categories, with little control over which instances we encounter. Although
Bruner et al. studied both paradigms, subsequent research has tended to focus
on the reception paradigm.
Figure 10.3 contains three examples of what subjects might see. Each col-
umn contains a sequence of objects associated with a different category. A plus
sign (+) beside the object means it is a member of the category, and a minus sign
(-) means it is not. Subjects are presented with these instances one at a time.
You should try to figure out the category represented by each column.
The category for the first column is“two crosses,” for the second column,
“two borders or circles,” and for the third column,”“number of borders equals
number of objects.” The first category probably seems the most natural. It is
referred to as a conjunctive concept because it requires that all of a set of features
be present. The second is called a disjunctive concept because it only requires that
at least one member of a set of features be present. The third is called a relation-
al concept because it involves a relationship among the dimensions. Subjects find
conjunctive concepts much easier than disjunctive or relational concepts and
learn them after seeing fewer instances (Bourne, 1974; Bourne, Ekstrand, &
Dominowski, 1971).

342
 Concept Acquisition

(2)So S a oO ao}= = oOfo) =] 2)@ UC= ine) 2)fo} =] ° @ ae}ee Ww

eS oP |

FIGURE 10.3 Examples of sequences of +

instances from which a subject is to identify

concepts. Each column gives a sequence of
instances and noninstances for a different
concept: A plus (+) signals a positive
instance and a minus (-) a noninstance.
Source: From Cognitive psychology and its
implications by John R. Anderson.
Copyright © 1990 by W. H. Freeman and
Company. Reprinted with permission.

TAPESSC IEORISA I LNVEEA EON

NG SEI NOES EBAIEEE

Cenren seater has es studied in i ee in ini

iAE ee RNS EER ERIE USDA RE SUELO emeci OH

subjects see instances identified as members or nonmembers of

a concept they are supposed to identify.
OLESEN RSELAE DLE ILI TREE DEEL NEE UIE IMDM NLS LITE ROB

Hypothesis Testing
These tasks are not all that different from conditioning experiments (Chapters 2
through 4) in which the organisms must figure out what features in the envi-
ronment are controlling reinforcement. The earlier chapters reviewed the evi-
dence that organisms tend to strengthen associations between various features
in the environment and responses. One feature of their learning is that it is
gradual. In contrast, human learning in these experiments seems to be anything
but gradual. Bruner et al. characterized their subjects as engaging in hypothe-
sis testing. Subjects have specific hypotheses, such as“I think it is three cross-
es,”and they may completely change their hypothesis from one trial to the next.
Bruner et al. characterized hypothesis testing as following roughly these steps:
1. Pick some hypothesis consistent with the instances that have been
encountered. (In some experiments the instances are arrayed in front of

343
 CHAPTER 10 Inductive Learning

the subjects, whereas in other experiments subjects must try to remember

the instances.)
2. Classify new instances according to this hypothesis.
3. If the classification of new instances is correct, stick with the hypothesis.
4. If the classification is wrong, go back to step 1.
This procedure, which is dramatically different from associative learning, pro-
duces discontinuities in performance such that subjects behaving according to
one hypothesis will change their behavior to conform to a new hypothesis given
a single disconfirming episode in step 4.
As an example of hypothesis-testing behavior, suppose that a subject is
entertaining the hypothesis that the concept is simply “black objects.” The sub-
ject sees three black squares in two borders and identifies it as an instance of the
concept. Suppose this identification is correct. Next the subject sees three white
squares in three borders and identifies it as not an instance of the concept. If the
subject is told that this is wrong, the subject completely abandons the old
hypothesis of “black” and may then entertain the hypothesis that the concept is
“square objects.” This one-trial switch from “black” to “squares” is quite unlike
associative learning about the relevant features of a stimulus, in which organisms
eradually strengthen or weaken associations to features such as squareness.
If subjects are asked to say what concept they are considering, they will
often announce that they are making such switches in their hypotheses.
However, experimenters have often worried that asking subjects to describe
their hypotheses may influence what they are doing. Therefore, researchers
more often simply observe which instances subjects classify as members of the
category and which they do not. From this classification data researchers infer
what concept a subject is entertaining.
Some early research on human concept formation that used this method-
ology yielded data that seemed to favor the gradual learning rather than the all-
or-none-learning assumption of hypothesis-testing models. In the typical para-
digm, during one trial a subject is shown a stimulus, is asked to indicate if it is
in the category, and is then given feedback as to whether the classification is
correct. Such trials are repeated over and over again. Figure 10.4 shows a typi-
cal plot of probability of a correct response as a function of trial number. The
percentage of correct classifications is averaged over all the subjects in the
experiment. The figure shows a continuous approach to perfect performance,
suggesting that subjects gradually develop associations to the correct stimulus
features. The apparent gradualness-.of Figure 10.4 seems to contradict the
hypothesis-testing explanation, which claims that subjects identify the correct
hypothesis or concept on a single trial. According to that view, subjects would
perform at a chance level (50 percent) for a while and then abruptly jump to 100
percent on the trial when they identified the correct hypothesis.
Bower and Trabasso (1964) wondered whether the apparent gradualness
of concept discovery might be an artifact of averaging over subjects. Perhaps one
subject might have selected the correct concept after an error on trial 10 and

344
 Concept Acquisition

100

ol fo)
FIGURE 10.4 Probability of
classifying an instance as a func-
Percent
successes
tion of trial in the typical concept-
identification experiment. The
probabilities in this hypothetical P | | | |
curve have been averaged over 5 10 15 20 25 30
subjects. Trials

shown an abrupt jump to perfect categorization; another subject might have

selected the correct concept on trial 6; and still another on trial 20; and so on.
Averaging different subjects together would give the illusion of gradual
improvement as more and more subjects identified the correct concept. The
increase in average probability of correct classification would just reflect the
growing percentage of subjects who had identified the concept and who were
responding perfectly. To test for this possibility, Bower and Trabasso developed a
new method for plotting the data. They identified the last trial on which each
subject made an error and then plotted probability of correct categorization back
from that trial. Figure 10.5 shows those data. Suppose the last error one subject
made was on trial 9. For this subject, trial 1 in Figure 10.5 would come from trial

100

90

®72)
2)

n
o

rs} eve | °
phy
Ss
we
A
e
a eae
e
ir ey
po
Oo
o
oO
{o)
te

Single trials

1 5 10 16-20 36-40 56-60

Trials before last error

FIGURE 10.5 A backwards learning curve: Probability of correctly classifying an

instance as a function of number of trials before last error. (From Bower & Trabasso,
1963.) Source: Figure 10.5 from R. C. Atkinson, Ed. Studies in mathematical psychology.
Copyright © 1964 by Stanford University Press. Reprinted by permission.

345
 CHAPTER 10 Inductive Learning

8, trial 2would come from trial 7, and so on; finally, trial 8 would come from trial
1. For another subject whose last error was on trial 21, trial 1 in Figure 10.5
would come from trial 20, trial 2 would come from trial 19, and so on; finally,
trial 20 would come from trial 1.1 Thus, trial 1 is the trial just before the last error
for all subjects, trial 2 is the second trial before the last error for all subjects, and
so on. The curve illustrated in Figure 10.5 is called a backwards learning curve.
If each subject had been gradually learning the concept, this backwards
learning curve would gradually improve as it approached trial 1 (the trial before
the last error for all subjects). However, probability correct hovers around the
chance level of 50 percent right up to the trial just before the last error (trial 1 in
Figure 10.5). These data are good evidence for all-or-none learning. That is, on
the last error subjects made a complete switch from a wrong hypothesis, which
was yielding chance performance, to the correct one, which yielded perfect per-
formance. This analysis contains a significant lesson: an average learning curve
(Figure 10.4) that apparently displays gradual learning can actually be hiding
all-or-none learning, which can be uncovered by a backwards learning curve
(Figure 10.5).

Concept learning can involve abrupt changes in hypotheses

when the current hypothesis is disconfirmed.
aS EARS NTEEON

Natural Concepts
Research on hypothesis testing was in its heyday during the 1960s. After that
time there was increased questioning of what these laboratory experiments on
concept acquisition revealed about human learning of natural categories, such
as “dog” or “tree.” Subjects approach such experimentswith a problem-solving
orientation, which does not seem to be how natural categories are learned. This
approach certainly appears too sophisticated and conscious for children learn-
ing in a natural environment. Moreover, it has been argued (Rosch, 1973, 1975,
1977) that natural categories are not the sorts of things that have the all-or-none
logical structure of these laboratory categories. One of the major characteristics
of natural categories is that they are not defined by the presence of a few fea-
tures. Rather, many features tend to be associated with the category, and an
instance is a member of a category to the degree that it possesses these charac-
teristic features. For example, birds are characteristically of a certain size and can
fly. However, an ostrich, which cannot fly and is very large, is recognized as a
bird because it has other birdlike features, such as feathers, wings, and a beak.
Even though an ostrich can be recognized as a bird, a person seeing one for the
first time might certainly hesitate in making the classification.

1One consequence of this process is that fewer and fewer subjects contribute to trials
that are more and more removed from the trial of last error.

346
 Concept Acquisition

TABLE 10.1 Ratings of Category Goodness

Fruit Vehicle
Apple 1.3 Car 1.0
Strawberry 29 Boat PAU
Fig 4.7 Tricycle 3.5
Sport Crime
Football ike Murder 1.0
Hockey 1.8 Embezzling 1.8
Wrestling 3.0 Vagrancy 5.3
Bird Vegetable
Robin tei Carrot 11
Wren 1.4 Onion 2a,
Ostrich he) Parsley 3.8

Source: From Rosch, 1973.

Rosch performed several experiments examining the structure of natural

categories. In one experiment (1973), subjects were asked to rate how typical
various members of categories were on a 1 to 7 scale, where 1 meant very typi-
cal and 7 meant very atypical. Table 10.1 reproduces some of the ratings
obtained. Different members of a category received rather different ratings as to
how typical they were. Rosch argued that various instances are members of a
category to the degree that they are typical of that category. In another experi-
ment, Rosch (1975) had subjects categorize pictures of animals and plants, such
as robins and chickens (both birds) and apples and watermelons (both fruits).
She found that subjects were more rapid at identifying the category member-
ship of the pictures of the more typical objects. For instance, they more quickly
recognized a robin as a bird than an ostrich as a bird, and they more quickly rec-
ognized an apple as a fruit than a watermelon as a fruit. Rosch argued that
instances are members of a category to the degree to which they possess fea-
tures associated with that category. Thus, robins are seen as more typical of birds
than are ostriches because robins can fly and their size and shape are more
commonly associated with birds.
McCloskey and Glucksberg (1978) provided one of the more convincing
demonstrations of the fact that natural categories do not have fixed boundaries.
They found that there were various items about which subjects could not agree
as to category membership. For instance, is stroke a disease? Sixteen of 30 sub-
jects thought so, whereas the other 14 disagreed. Is a leech an insect? Thirteen
subjects said yes and 17 said no. Is a pumpkin a fruit? Sixteen subjects said yes
and 14 said no. In a retest a month later, 11 subjects reversed themselves on
stroke, 3 reversed themselves on the leech, and 8 reversed themselves on pump-
kin. Not only do subjects disagree among themselves as to category member-
ship, but individual subjects are inconsistent in assignment to category.
A basic feature of natural categories is that they have these vague bound-
aries; as a consequence, instances are members of the categories to the degree

347
 CHAPTER 10 Inductive Learning

TABLE 10.2 An Example of the Experimental Material Used in Medin and

Schaffer (1978)
1. One large red triangle is in category A.
2. Two small red triangles is in category A.
3. One large blue circle is in.category A.
4. Two small blue circles is in category B.
5. One large red circle is in category B.
6. Two small blue triangles is in category B.

to which they are typical. The hypothesis-testing behavior reviewed earlier

seems appropriate only for learning categories with crisp, discrete boundaries. A
number of experiments have studied the acquisition of artificial concepts that
have a structure more like that of natural categories. A simple example of such
an experiment was performed by Medin and Schaffer (1978). They had subjects
study the six items in Table 10.2. In categoryA the majority of the stimuli are one
object, large, red, and triangular, whereas in category B the majority of the stim-
uli are two objects, small, blue, and circular. There are exceptions, and no feature
is sufficient for category membership, nor is there any apparent rule for classifi-
cation. After studying these items, subjects were asked to rate how typical vari-
ous stimuli were of their category. Medin and Schaffer found that stimuli were
judged typical to the extent that they had the features associated with that cat-
egory. In Table 10.2 stimulus 1 was judged most typical of category A, and stim-
ulus 4 was judged most typical of category B.
Two types of theories based on such experiments have been advanced to
account for how people learn the structure of natural categories. Schema theo-
ries hold that people represent the various features that define the category. The
way this knowledge is represented depends on the specific schema theory.
Some theories (e.g., Nosofsky, Palmeri, & McKinley, 1994) hold that subjects
form explicit rules for using features to categorize objects. Other theories (e.g.,
Reed, 1972) propose that subjects form a prototype of what a typical member of
the category is like. The next section will describe a schema theory that propos-
es subjects form associations from features to categories like those learned in
conditioning experiments. Exemplar theories (e.g., Medin & Schaffer, 1978;
Nosofsky, 1986) hold that people classify instances as members of a category to
the extent to which they are similar to other instances of a category. This theo-
ty holds that we do not really create categories but rather judge category mem-
bership on the basis of similarity to specific instances.
Schema theories and exemplar theories are two very different conceptions
of categorization. The first states that people form abstract representations of
categories, and the second says that people do not really form categories at all.
It would seem that it should be easy to tell which is right, but it has proven dif-
ficult to discriminate between the two types of theories. Part of the problem is
that the general characterization of a schema theory or an exemplar theory is
underspecified and it is necessary to be more precise to have a testable theory.

348
 Concept Acquisition

The next two sections describe examples of these two kinds of theories and how
they can be tested.

Natural concepts have vague boundaries in which items are

members of the category to varying degrees.

A Schema Theory: Gluck and Bower

Chapter 2 described Gluck and Bower’s (1988) application of the
Rescorla—Wagner theory to describe the learning of two disease categories. Their
model is an example of a schema theory. It has attracted particular attention
because it has been presented as a connectionist model related to other data on
neural learning (Gluck & Thompson, 1987). Gluck and Bower’s model assumes
that strengths of association are formed between stimulus features and cate-
gories according to the Rescorla~Wagner rule. (The stimulus features are treat-
ed as CSs and the category is treated as the US.) Suppose that a stimulus con-
sisting of two small red triangles is presented and is said to be in category A.
Then the strengths of association from the stimulus features (two objects, small,
red, and triangular) to category A are changed according to the rule:
AV = aA, — XV)
where AV is the amount of change; © is the learning rate; A, is the strength of
association that category A can support; and XV, is the total strength of the
existing associations of these features to category A. If there is an alternative cat-
egory B, strengths of association from these features to that category are
decreased according to the rule:
AV = a(0 - XVj)
where XV, is the sum of the strengths of existing associations to B. As discussed
in the conditioning chapters, the strengths of association among individual
stimulus features and a category are set according to how well these stimulus
features predict that category. With respect to the example in Table 10.2, these
equations imply that the features associated with category A (one object, large,
red, triangular) eventually acquire strengths of association to category A of .25
As, and the other features have strengths of association of zero. With respect to
category B, these other features acquire strengths of .25 A,, whereas the A fea-
tures have no association. The Gluck and Bower theory nicely accounts for the
fact that instances are seen as members of categories to the extent that they dis-
play features associated with the category. Another advantage of the Gluck and
Bower theory is that it accurately predicts the learning curves that describe how
subjects gradually develop their ability to categorize with exposure to more and
more instances (as reviewed in the next section).
Consider how the Gluck and Bower model would apply to the stimuli in
Table 10.2. It would predict that the features (one object, large, red, and triangu-

349
 CHAPTER 10 Inductive Learning

lar) would be associated with category A and the features (two objects, small,
blue, and circular) would be associated with category B. Consider how it would
respond to the stimuli:
1. One large red triangle (in category A): 4A features, 0 B features
2. Two small red triangles (in category A): 2A features, 2 B features
3. One large blue circle (in category A): 2A features, 2 B features
4. Two small blue circles (in category B): 0A features, 4 B features
5. One large red circle (in category B): 3A features, 1 B feature
6. Two small blue ciréles (in category B): 1A feature, 3 B features
If the subjects adopted the rule to classify in category A all stimuli with two or
more A features, they would correctly classify all but stimulus 5. The
Rescorla-Wagner rule cannot correctly categorize stimulus 5 because it contains
more A features than B features. To deal with such problems, Gluck and Bower
proposed that subjects use configural stimuli. For instance, subjects might asso-
ciate the feature combination of red plus circle with category B. Chapter 2 dis-
cussed how it was necessary to augment the Rescorla-Wagner theory with con-
figural stimuli to account for results in the conditioning literature. The same his-
tory of theories that appeared in the conditioning literature is being played out
in this categorization research.
The Gluck and Bower theory illustrates the apparent cyclic character of
theories in psychology. Psychologists such as Hull originally proposed using
strength-of-association theories to account for human concept formation.
Subsequent research on hypothesis testing in concept formation indicated that
categories were learned in a more all-or-none manner than envisioned in those
theories. Later researchers questioned whether the concepts learned in the
experiments were like natural concepts. When the learning of natural concepts
was studied, the process of learning appeared to be much more similar to that
proposed by Hull.

Gluck and Bower proposed that strengths of association

between features and concepts are strengthened according to
the Rescorla—Wagner learning rule.

An Exemplar Theory: Medin and Schaffer

The other kind of theory, the exemplar theory, proposes that the subject forms
no categories at all; rather, the subject simply remembers some or all the
instances of various categories. When asked to categorize an instance, the sub-
ject determines what past instance is similar to this test instance and infers that
the test instance is in the same category as the past instance. A particularly suc-
cessful version of this kind of theory is the exemplar theory of Medin and
Schaffer (1978; see also Nosofsky, 1988).

350
 Concept Acquisition

To formally develop the Medin and Schaffer theory, it is necessary to spec-

ify the probability of retrieving a past study instance given a particular test
instance. This is a function of the similarity of the study instance to the test
instance relative to the similarity of other studied instances to the test instance.
The similarity between a study instance and a test instance is calculated in terms
of the similarity of their component features. Thus, a study stimulus of two large
blue triangles is similar to a test stimulus of one large blue circle to the extent
that one and two are similar, large and large are similar, blue and blue are sim-
ilar, and triangle and circle are similar. Medin and Schaffer proposed to measure
the features on a 0-to-1 scale, with 0 meaning totally dissimilar and 1 meaning
identical. They proposed that the similarity of the two stimuli was a product of
the similarities of the component features. In one of their applications, they pro-
posed that the similarity would be 1 if two features matched and .2 if the fea-
tures mismatched. In the example given here, where there are two matches and
two mismatches, the overall similarity is 1 x 1 x .2 x .2 = .04.
As a full illustration of the Medin and Schaffer theory, suppose that the
subject has studied the six stimuli in Table 10.2 and musi classify the test stim-
ulus one large blue triangle. To apply the theory, it is necessary to calculate the
similarity of this test stimulus to each study stimulus in Table 10.2. These calcu-
lations are performed in Table 10.3. The probability of categorizing the test stim-
ulus in categoryA is its total similarity to the categoryA stimuli (first three in the
table) relative to its similarity to all stimuli:
200 + .008 + .200.
200 +008 +200+ .008 +.040+.040 ~ >
Thus, the chances are high that the subject would place this particular stimulus
in category A.

TABLE 10.3 Calculation of the Similarity of One Large, Blue Triangle to Each
Study Stimulus in Table 10.2
Number Size Color Shape Similarity
Category A
Study Item 1 1 x 1 x 12. x 1 = .200
Category A
Study Item 2 se x ea. x 2 x it = .008
Category A
Study Item 3 il x 1 x 1 x 2 = .200
Category B
Study Item 4 p72 x 2 x 1 x 2 = .008

Category B
Study Item 5 1 x 1 x wa x 2 = .040
Category B
Study Item 6 2 x 2. x il x 1 = .040
NE... eee

351
 CHAPTER 10 Inductive Learning

The exemplar theory is able to predict certain results that simple versions
of schema theory cannot. Consider classification of the problematical item 5, one
large red circle, given the items in Table 10.2. Since the features one, large, and red
are all associated with category A, the schema theory would predict that the
item would be classified in that category. To account for the correct classification
of this item, Gluck and Bower assumed that subjects responded to configural
cues, which are combinations of features. The Medin and Schaffer theory has no
problems accounting for the successful classification of this item. Since the item
was studied as being in category B and is maximally similar to itself, subjects are
likely to retrieve and usé@it for its own classification. Put another way, the Medin
and Schaffer theory proposes that subjects classify this item by remembering it
specifically and the category it came from.
Estes, Campbell, Hatsopoulos, and Hurwitz (1989) compared the Gluck
and Bower network model with the exemplar model. Their experiment involved
subjects categorizing patients’ symptoms into a rare versus a common disease
(the same sort of experiment as described with respect to Figure 2.15). Subjects
practiced categorizing the symptoms of 240 patients. Various symptom combi-
nations occurred with various probabilities with the two diseases. Figure 10.6
compares the subjects’ success in predicting blocks of 10 patients with the suc-
cess of the two models. Since symptom combinations occurred with the dis-
eases with only certain probabilities, subjects could not be perfect. The fluctua-
tions in Figure 10.6 from block to block reflect how difficult each block was.
Subjects showed some tendency to improve (chance is 50 percent). What is
remarkable is how well the two theories do at predicting the ups and downs in
subject accuracy. Both theories seem to do a good job in predicting all the ups
and downs in the data. The Gluck and Bower model does a little better, but the
real message of this figure is that two very different theories can yield such sim-
ilar predictions.

Exemplar theories claim that subjects classify a test instance

A Pluralistic View of Concept Acquisition

There has been a long history of concern in psychology with how people form
categories. Research has progressed as if all categories were learned a single
way, but in retrospect this seems unlikely. Some categories are probably learned
by direct instruction, as you, the reader of this book, learned about the category
of classical conditioning experiments. Other categories, if defined by rigid rules,
may be learned by explicit hypothesis testing. For instance, an observer of base-
ball for the first time might use this method to determine what defines a hit.
Categories that have less rigid definitions may be learned by simple associative
learning (such as in the Gluck and Bower theory) if they have a core set of fea-
tures. If members of the category are more scattered, as is the case for the items

352
 Concept Acquisition

100

80

Data

Percent
correct
60

Schema

40 Tact TT es sae ae ie ge ee
6 12 18 24
10-trial block

(a)
=

Data

80 |-

Exemplar
Percent
correct
60 |-

6 12 18 24
10-trial block

(b)
FIGURE 10.6 Comparison of schema and exemplar models in accounting for block-
by-block learning data. Source: From W. K. Estes, J. A.Campbell, N. Hatsopoulos, and
J. Hurwitz. Journal of Experimental Psychology: Learning, Memory and Cognition, Volume
4. Copyright © 1989 by American Psychological Association, p. 561. Reprinted with
permission.

in Table 10.2, an exemplar theory, like that of Medin and Schaffer, may be appro-
priate. There is probably not one correct theory of category learning. It is more
likely that the different theories reviewed in this section are correct in different
situations.

353
 CHAPTER 10 Inductive Learning

Recently, the field has become attracted to the idea that there might not
be a single categorization process and that different subjects might be doing cat-
egorization by different mechanisms. Indeed, the same subject might be doing
categorization by different processes at different times. For instance, Erickson &
Krushke (1998) show that subjects categorize some stimuli according to rules
and that they categorize other stimuli according to exemplars. Similarly, Smith,
Patalano, Jonides, and Kleppe (1998) find different patterns of brain activation
depending on whether subjects are categorizing examples by rules or exem-
plars. In particular, when using rules there was activation in the frontal cortex
which was not present When using exemplars. As we have noted elsewhere in
this book (Chapters 5 and 9), the frontal cortex tends to be active in tasks that
make high demands on the working memory or that involve substantial goal
manipulation. This would seem to imply that using rules to categorize is a more
demanding endeavor.

People categorize objects by multiple means.

Causal Inference
Another critical kind of inductive learning involves figuring out what causes
what in our environment. People frequently engage in such causal inference.
Every time we come across a new device, we have to figure out how it works. A
person entering a new room may need to determine what causes a light to go
on. Children try to figure out what gets their parents angry, and parents try to
figure out what gets their children to obey. Police try to find out who committed
crimes, and physicians try to identify the causes of symptoms. It is important to
understand the causal structure of our environment because knowing that
allows us to use it to achieve our purposes. Indeed, as suggested in Chapters 2
and 3, much of animal conditioning was really concerned with how animals, in
effect, inferred the causal structure of their environments. Research with
humans sheds further light on causal inference.
It is useful to appreciate how both categorization and causal attribution
are instances of inductive inference but also how they are different. Categorical
inference involves noting that a set of features cluster together. In forming the
category of“bird,”a person is responding to the fact that the same kinds of ani-
mals tend to have feathers, to have beaks, and to lay eggs. Causal inference
involves noting that one set of events tends to predict another. For instance, a
person may determine that flipping a switch causes the light to come on or that
pressing a bar causes food to appear. In both cases, the person infers a predic-
tive relationship in the world—in one case among features of an object and in
the other case among events. Causal inference is inherently directional. If a bar
is pressed, food is expected to appear in the feeder, but if food is put in the feed-
er, the bar is not expected to depress. Categorical inference is symmetrical. An

354
 Causal Inference

animal that has feathers is expected to have a beak, and an animal that has a
beak is expected to have feathers.
The general approach to understanding human causal inference has been
to study how people use various cues to causality (e.g., Einhorn & Hogarth,
1986). When two events (such as flipping a switch and turning a light on) are in
a cause-and-effect relationship, there are certain telltale cues as to that relation-
ship. The following sections review some of the cues people use to determine
causality.

Statistical Cues
Perhaps the most obvious cue is that of statistical contingency. Recall from the
discussion of conditioning that contingency refers to whether one event predicts
another. If whenever event A occurs, event B follows, and B never occurs unless
A has occurred, then there is strong evidence for the proposition that A causes
B. However, things are often not that certain. Consider the proposal that smok-
ing causes heart disease. Suppose that we observe some people who smoke and
others who do not. Some people who smoke will develop heart disease, some
who smoke will not develop heart disease, others who do not smoke will devel-
op heart disease, and yet others who do not smoke will not develop heart dis-
ease. The data can be organized according to a 2 x 2 table, as in Table 10.4, which
gives the number of observations of the four kinds.
The data in Table 10.4 (totally hypothetical) appear to provide evidence for
a causal relationship: a person who smokes has a 75 percent chance of devel-
oping heart disease, and a person who does not smoke has only a 40 percent
chance. Table 10.4 uses the variables a to d to stand for the frequencies in vari-
ous cells. The greater a (co-occurrence of cause and effect) and d (occurrence of
neither) are, the stronger evidence there is for an effect. The greater b (occur-
rence of cause but not effect) and c (no occurrence of cause but effect) are, the
less evidence there is for a relationship. Researchers have studied how sensitive
subjects are to variations in these four quantities—a, b, c, and d. The animal
research considered earlier (e.g., Figures 2.9 and 3.12) indicated a general sen-
sitivity to these factors. The human research (e.g., Crocker, 1981; Jenkins &

TABLE 10.4 Hypothetical Relationship Between Cholesterol and Heart Disease

Number of Patients in Each Cell

Effect Present: Effect Absent:

Heart Disease No Heart Disease

Cause Present:
Smoking a=75 B= 25

Cause Absent:
No smoking c= 40 d = 60

355
 CHAPTER 10 Inductive Learning

Ward, 1965; Schustack & Sternberg, 1981; Shaklee & Tucker, 1980) has been
specifically concerned with how people respond to variations in the four indi-
vidual quantities. In general, people behave in a rational way by increasing their
belief in a causal relationship as a or d increases and decreasing their belief as b
or c increases. Subjects appear to be most sensitive to changes in a, about equal-
ly sensitive to changes in b or c, and least sensitive to changes in d.
In a typical experiment performed by Anderson and Sheu (1995), subjects
tried to judge whether or not a drug had a side effect. Three of the variables a
through d were held constant, and the remaining variable was manipulated
from values of 1 to 15. Subjects had to judge on a scale from 0 to 100 how like-
ly it was that the drug caused the side effect. Figure 10.7 shows how these judg-
ments varied with changes in these variables. Changing a (cause and effect pre-
sent) from 1 to 15 increased judged causal effectiveness by 40 points; changing
b (cause but not effect) decreased judged effectiveness by 30 points; changing c
(no cause and effect) decreased judged effectiveness by 20 points; and changing
d (neither cause nor effect) increased judged effectiveness by only 5 points.
Subjects were sensitive to all variables but certainly differed in how sensitive
they were.
Chapter 3 noted that, according to the Rescorla-Wagner theory inferences
(see the discussion pertaining to Figure 3.14), the strength of association learned
between cause and effect is proportional to the differences between the proba-
bilities of the effect in the presence of the purported cause and the probability of
the effect in the absence of the purported cause. That is, the strength of associa-
tion between possible cause C and effect E is proportional to P(E|C) — P(E|-O)
where

PEIChe a 7 b

PEE|-O)i= C :d
This model predicts a general sensitivity to the variables a through d. One prob-
lem with this model is that it predicts that subjects should be as sensitive to
a change in a or b affecting P(E|C) as they are to a change inc or d affecting
P(E| — C). However, Figure 10.7 shows that subjects are not equally sensitive.
It seems unlikely that human subjects are simply forming strengths of
association as the Rescorla-Wagner theory suggests. When queried by
Anderson and Sheu, the majority of subjects reported that they were explicitly
trying to calculate P(E|C) and P(E|-C) and compare them. However, a sizable

2P(E|C) is to be read as probability of effect given cause, and P(E |-C) is to be read as
probability of effect given absence of cause.
Wasserman et al. (1993) showed that having different learning rates for the four
cases can yield different effects of a through d.

356
 Causal Inference

0a

L
= (cause
60 and effect) d
(no cause
and no effect)

Rating aij Cc

(no cause
and effect)

b
(cause and
20 no effect)

FIGURE 10.7 Effects of variables a

through d on causal inference. From 5 10 15
Anderson & Sheu (1995). Value

minority of subjects reported calculating only P(E|C) and basing their judgments
on this. The majority showed relatively equal effects of a through d, whereas the
minority showed only effects of a and b. It seems that subjects followed con-
scious strategies of probability estimation and that the variables had different-
sized effects because some subjects ignored some of the information. Subjects
behaved in a rather conscious hypothesis-testing manner, which happened to
be mimicked by the Rescorla—Wagner theory. It is quite plausible that lower
organisms are incapable of such conscious calculations but do behave in accord
with the Rescorla—Wagner theory. This is another instance of rather different
mechanisms producing similar results.

Subjects base their causal attributions on the probability of the

effect in the presence versus absence of the cause.

Cues of Spatial and Temporal Contiguity

The British philosopher David Hume noted that when one event occurs just
before a second event and close in space to the second event, the first appears
to be the cause of the second. Thus, if lightning is followed immediately by
thunder, we tend to think of the lightning as causing the thunder. The earlier
chapters reviewed the evidence that animals are also more likely to display con-
ditioning when two events to be conditioned are close together in time.
Proximity in space is important, too—when I found spilled milk near one of my
children and not the other, I attributed the spill to the child it was close to.
Researchers have studied how people combine spatial and temporal contiguity
to make attributions of causality.

357
 CHAPTER 10 Inductive Learning

|
SAE

FIGURE 10.8 The apparatus used by Bullock et al. (1982). Source: From M. Bullock,
R. Gelman, and R. Baillargeon in W. J. Friedman, Ed. The developmental psychology of
time. Copyright © 1982 by Academic Press. Reprinted by permission.

Figure 10.8 shows the device used by Bullock, Gelman, and Baillargeon
(1982) in a study of causal attribution. Subjects saw two balls drop into tubes at
the end of a box. Then a jack-in-the-box appeared in the center of the box. There
were four conditions that varied in terms of the temporal proximity and spatial
proximity of the dropping of the two balls. In condition 1, the two balls were
equally distant from the jack-in-the-box, but one was dropped before the other.
In this condition, 65 percent of the subjects attributed the appearance of the jack-
in-the-box to the dropping of the second ball, which was closer in time. In condi-
tion 2, the two balls were dropped at the same time, but one was closer to where
the jack-in-the-box appeared. In this circumstance, 100 percent of the subjects
attributed the cause to the closer ball. In a third condition, one ball was close in
both time and space to the appearance of the jack-in-the-box. In this condition,
100 percent of the subjects also chose the closer ball. The fourth condition
involved a conflict: one ball was closer in time, but the other was closer in space.
In this condition, 70 percent of the subjects chose the ball that was closer in space.
Subjects in the Bullock et al. experiment tended to prefer the cue that was
closer in space over the cue that was closer in time. Other researchers (e.g.,
Shultz, Fischer, Pratt, & Rulf, 1986) found that subjects preferred as a cause what
was closer in time. Thus, one cue is not always dominant. Both cues are effec-
tive, and which is dominant depends on the particular situation.
Figure 10.9a shows a computer display used to study the role of temporal
and spatial contiguity in causal attribution (Anderson, 1991). In this experiment,
a hand dropped a weight on a beam and a trapdoor opened up, releasing a ball.
Subjects were asked to rate how compelling was the perception of a causal rela-
tionship between the dropping of the weight and the popping out of the ball.
The distance was varied between the weight and the door, and the delay was
varied between the dropping of the weight and the opening of the door.
Subjects rated the causal link between the two events on a 1 to 7 scale, where 7
meant definitely causally related and 1 meant no causal relationship. Figure
10.9b shows how the strength of causal attribution varied with these two fac-
tors. In this experiment, time was the dominant variable, and distance only
entered into strength of attributions when the delay was short.
Researchers (e.g., Shultz, 1982) have argued that subjects do not blindly
use spatial and temporal contiguity to infer a causal relationship but that they

358
 Causal Inference

es
Click mouse for next trial
o

A
Distance
5 —o- 15 units
=>7 units
— 3 units
25 —= 1 unit
Cc
oO
52
&O4

2 | | | |
0.1 0.3 0.9 2.7 8.1
Time, sec

(b)
FIGURE 10.9 The vibratory wave model: (a) the computer display; (b) strength of
causal perception as a function of distance in space and time. Source: From J. R.
Anderson. Is human cognition adaptive? Behavioral and Brain Sciences, Volume 14.
Copyright © 1991. Reprinted by permission of Cambridge University Press.

show some appreciation of the possible underlying mechanisms. Figure 10.9a

can be interpreted in these terms. The probable causal mechanism is one in
which the dropping of the weight on the beam sets up a vibratory wave, which
propagates down the beam and releases some mechanism (e.g., a catch). Such
mechanical waves should propagate down the beam almost instantaneously,
and so any significant delay between the dropping of the weight and the open-

359
 CHAPTER 10 Inductive Learning

ing of the door would appear causally inconsistent. The closer the weight to the
door, the more force the wave has, and so the more likely it is that the wave will
jar loose the mechanism. Thus, at short delays, subjects should prefer the cause
when it is closer in space, which is just what is shown in Figure 10.90.
Figure 10.10a shows an interesting contrast condition (Anderson, 1991) to
the one in Figure 10.9a. A hand dropped a ball into a hole in the beam.

Click mouse for next trial

Distance
: -o 15 units
— 7 units
6 —> 3 units
—= 1 unit

Confidence

On 0.3 0.9 PM 8.1

Time, sec

(b)
FIGURE 10.10 The ball and projectile model: (2) the computer display; (b) strength
of causal perception as a function of distance in space and time. Source: From LR
Anderson. Is human cognition adaptive? Behavioral and Brain Sciences, Volume 14.
Copyright © 1991. Reprinted by permission of Cambridge University Press.

360
 Causal Inference

Sometime later a trapdoor opened and a ball appeared, just as in Figure 10.9a.
As with Figure 10.9a, the distances in time and space between the first event
and the second were manipulated. Figure 10.10b shows how subjects’ causal
attributions varied in this condition. The results are in considerable contrast to
the data in Figure 10.9b. In Figure 10.10b there is no favored time or distance.
Rather, as the distance increased, subjects favored longer and longer times.
Subjects reported looking for a match between distance and time. Their model
was one in which the ball traveled through the beam to appear at the trapdoor,
and they were looking for a situation where the time was appropriate (not too
long or too short) for the distance traveled.
Human subjects are quite sophisticated in their interpretation of temporal
and spatial contiguity. Earlier chapters reviewed the evidence for a similar
degree of sophistication in animals in how much they rely on temporal conti-
guity. For instance, rats connect taste with poisoning after much longer delays
than usually work in conditioning. A possible reason is that poisoning is the
kind of effect that often appears at a considerable delay after ingestion. It seems
unlikely, however, that rats behave with as explicit and conscious a model as
human subjects sometimes do.

Humans use temporal and spatial cues to judge causality in

accord with how well these cues fit a prior causal model.

Kinematic Cues
Subjects often display great sophistication in their interpretation of kinematic
cues. Kinematic cues refer to properties that would be expected of events
causally related according to the laws of physics. When the conditions are right,
kinematic cues can give rise to extraordinarily compelling perceptions of causal-
ity. Some of the original research on this topic was performed by Michotte
(1946). Subjects observed a black circle move across a screen and touch a sec-
ond circle; then the second circle moved off. When the second circle moved
immediately after it was touched, subjects had a compelling impression of a col-
lision in which the first object set the second in motion, as when one billiard ball
hits another. When there was any delay between the two events, the perception
of a causal connection dissolved.
In variations on Michotte’s experiment, Kaiser and Proffitt (1984) manip-
ulated the velocity and angle at which the two objects parted after the collision.
Subjects’ perceptions of causality were sensitive to the laws of physics govern-
ing such collisions, and they judged as causally anomalous collisions that
involved impossible angles or rates of acceleration. Subjects could also judge the
relative mass of the two objects from the velocity and angle at which the objects
separated.
In some situations, subjects’ judgments are not so in tune with the correct
scientific model. Consider the situation in which an object moves off a surface,

361
 CHAPTER 10 Inductive Learning

FIGURE 10.11 C-shaped tube problem used by

Kaiser, McCloskey, and Proffitt (1986). Source: From
M. K. Kaiser, M. McCloskey, and D. R. Profit.
Developmental Psychology, Volume 22. Copyright ©
1986 by the American Psychological Association.
Reprinted by permission.

such as when a ball rolls off a table. The correct scientific model is one in which
the trajectory of the object after it leaves the table is a curve reflecting a combi-
nation of the original horizontal velocity and the downward negative force
caused by gravity. Some people believe that the object will go directly down, and
others predict an L-shaped trajectory in which the object goes straight forward
for a while and then falls down. Judgments in this domain show a definite
developmental trend, with older subjects showing fewer misconceptions.
Apparently, people come to tune their models with experience and education
(Kaiser, Proffitt, & McCloskey, 1985).
Even more curious are judgments about the trajectory of an object after it
leaves a curve-shaped tube, such as the one in Figure 10.11 (Kaiser, McCloskey,
& Proffitt, 1986; McCloskey, 1983). A common misconception is that it will show
a curved trajectory rather than a straight one. This belief shows a U-shaped
developmental trend, with children around the sixth grade showing the most
misconceptions and preschoolers and college students about equivalent and
somewhat better.
People possess models for how physical events should take place and use
these models to make judgments of causality. Sometimes their physical models
are correct, and sometimes they are not. People are referred to as having naive
physics models that are partially correct. One goal in modern physics education
is to better train these naive physics models (e.g., Champagne, Gunstone, &
Klopfer, 1985).

People have naive physics models, which they use to judge

Understanding Complex Systems

Shrager, Klahr, and Dunbar performed an interesting series of studies. In the orig-
inal experiment by Shrager (1985), subjects were shown a toy tank with a keypad
similar to that shown in Figure 10.12 and were told to figure out how the machine
worked. The tank could be instructed to go forward or backward for a number of
feet, to rotate for a number of clock ticks (minutes on a clock), to pause for vari-
ous amounts of time, and to fire its gun. Given the command sequence:

362
 Causal Inference

ihe: By
<— 7,
ies
— 15,
Hold 5,
Fire 2,
8
the tank would move forward 5 ft, rotate counterclockwise 7 clock ticks, move
forward 3 ft, rotate clockwise 15 ticks, pause 5 sec, fire twice, and move back 8
ft. Most of Shrager’s subjects, who were undergraduates, were able to figure out
within an hour how to program the tank by simply experimenting with various
key combinations. He noted that a key to their success was that subjects made
constant reference to past experiences they had with similar objects that had to
be programmed (e.g., a microwave) and to their prior knowledge about the
kinds of behavior one might expect to see from a toy tank.
Klahr and Dunbar (1988) studied in detail how subjects learned about one
aspect of the device—the RPT key. When this key is followed by a number, it
repeats that number of previous moves. For example, RPT 3 repeats the last
three moves. However, subjects had many different ideas about what the key
did. A favorite hypothesis was that RPT 3 would repeat the whole sequence of
actions three times. Klahr and Dunbar found that subjects behaved basically like
scientists in determining what the key did, designing experiments and formu-
lating hypotheses on the basis of the results, and then designing new experi-
ments to test these hypotheses. However, their hypotheses were strongly biased
by their past experiences with what a key labeled RPT might do.
Learning can be greatly facilitated if we can learn something new as analo-
gous to something else. Blessing (1996) showed that key to students learning a
particular, new formal mathematical system was seeing how it was similar to stan-
dard algebra. Kieras and Bovair (1984) had subjects learn to operate a control panel
for a novel device. They gave subjects instructions about how to perform all of the

)
(7) (8) (2) er)
FIGURE 10.12 Key pad for programming a toy tank.
(4)(5)(s)
Source: From J. C. Shrager. Instructionless learning: Discovery bai} 2 (3) (cx )
of the mental model of a complex device. Copyright © 1985 by
Jeffrey C. Shrager. Reprinted by permission. (0)

363
 CHAPTER 10 Inductive Learning

procedures. This was all one group was told. However, a second group was told
that the device they were operating was the control panel for the phaser bank on
the Starship Enterprise.They were provided with a totally made-up story about how
the panel controlled the phaser bank. Nonetheless, given this model, the subjects
learned the procedures faster, retained them more accurately, executed them faster,
and found efficient shortcuts more often. Usually, when we learn to deal with
something new, we can recall experiences of dealing with something similar.
Success in learning the new often depends on making successful bridges to the old.

In learning about complex systems, people make reference to

their past experiences with similar systems.

Conclusions about Causal Inference

As discussed in the early chapters on animal conditioning, causal inference is par-
ticularly important to adapting to the environment. Knowing what causes what
enables an organism to achieve its goals. Humans are often quite sophisticated and
deliberate in how they go about inducing the causal structure of their environment.
They entertain sophisticated hypotheses for what mechanisms might produce the
effects in their environment, and they test these hypotheses against the available
data. Throughout, this chapter has noted that simpler associative learning mecha-
nisms often mimic the results of humans’ more conscious hypothesis-testing
approach. Why is it, then, that humans engage in this more sophisticated approach
when simpler methods yield the same results? Simple associative learning only
works in simple situations. The advantages of the more deliberate approach
become apparent when looking at learning about complex devices, such as the toy
tank shown in Figure 10.12. It may be the propensity to use tools that has moved
the causal learning mechanisms of humans in the direction of conscious hypothe-
sis testing. Hypothesis testing may well be overlaid on more automatic inductive
learning mechanisms, such as that captured by the Rescorla—Wagner theory.

Conscious hypothesis testing is more successful than simple

associative learning when the situation is complex.

Language Acquisition
Some researchers believe that acquisition of a natural language is the most
impressive inductive learning feat of the human species. Many argue that only
humans can learn a language and that our language facility reflects something
unique about the human mind (e.g., Chomsky, 1965, 1975). In a few short years
young children figure out what generations of Ph.D. linguists have not—the
rules of language. This contrast needs to be emphasized—scientists have not

364
 Language Acquisition

been able to characterize what the rules of language are, and yet children figure
it out with relative ease. Of course, children cannot say what the rules are that
they have learned. In the terms of Chapter 8, this is an instance of implicitdearn-
ing. The fact that people are so successful in such implicit learning of a first lan-
guage has been used to argue that we must have special innate and unconscious
knowledge as to the structure of language.
It is important to recognize what makes language learning such a difficult
task. Most people are quite aware that language contains many words—tens of
thousands in fact. Studies of young children have suggested that they learn
more than five new words each day (Carey, 1978; Clark, 1983). Learning the
meanings of all of these words defines an enormous concept-acquisition task.
Unlike learning a second language, a person learning a first language cannot
rely on the assistance of definitions stated in another language.
Although vocabulary is the most obvious aspect of language learning, it is
generally not considered the most daunting aspect. Learning the mor-
phophonology and syntax of language is more demanding. Morphophonology
refers to how the sound system determines meaning (for example, adding an s
in English to indicate possession—*Fred’s sister”), and syntax is concerned with
how word order determines meaning (for example, “sister of Fred”). Languages
may possess tens of thousands of such rules, many of which are rather subtle.
Long after second-language learners have mastered the vocabulary of the lan-
guage, they continue to make errors in pronunciation and grammar.
Not only are there many phonological and syntactic rules to be learned,
but the conditions in which children learn them seem far from ideal. No one
explicitly instructs children as to what the rules of language are. Children have
to induce these rules by hearing language spoken to them. Any particular sen-
tence involves many rules acting together to determine the sentence. The many
components of the sentence must be unraveled. Parents and other caregivers
might be thought to teach children by explicitly correcting their speech.
However, many children learn a language just fine without receiving such cor-
rection, and the available evidence suggests that such correction does not help
those children who receive it (e.g., Braine, 1971; McNeill, 1966; for a recent dis-
cussion, see MacWhinney, 1993). McNeill (1966) cited a famous example of how
impervious children can be to correction:
Child: | Nobody don’t like me.
Mother: No, say,” Nobody likes me.”
Child: | Nobody don’t like me.
Mother: No, say,“ Nobody likes me.”
Mother: No, say,“Nobody likes me.”
Child: | Nobody don’t like me.
[dialogue repeated 7 more times]
Mother: — Now listen carefully, say,“ Nobody likes me.”
Child: | Oh! Nobody don’t likes me.

365
 CHAPTER 10 Inductive Learning

Children accomplish an enormous task in acquiring language, and they put a lot
of time into it. Children do not master the subtleties of language until age 10
(Chomsky, 1970). By that time, they have put thousands of days and presumably
tens of thousands of hours into language acquisition. Recall from the previous
chapter the evidence that mastery of any complex skill involves an enormous
investment of time in which the various rules of that skill are mastered one by
one. Language is no exception to this principle. It may, however, be the most
complex system that people have to learn.
Not only do children often ignore explicit instruction about language but
also they are capable oflearning a language even if they are not exposed to any
language at all. Evidence for this fact comes from studies of deaf children of
speaking parents (approximately 90 percent of deaf children are born to speak-
ing parents). Goldin-Meadow, Butcher, Mylander, and Dodge (1994) studied a
deaf child whose parents chose to teach him by the oral method. Nonetheless,
as many deaf children do, he started to invent his own sign language to com-
municate with his parents. His sign language made a distinction between nouns
and verbs just as standard sign languages or natural languages do. Moreover,
some of his signs (e.g., hammer, brush, comb) could serve as either nouns or
verbs just like words in other languages. To distinguish noun uses from verb
uses he invented a series of syntactic markers. For instance, he tended to put
verbs at the end of his signing sequences. Thus, we see that to a certain extent
humans are born with an instinct to learn a language with certain features and
will construct such a language no matter what. Typically, the language they con-
struct corresponds to the one they hear, but much of their language learning
comes from the drive of this instinct and not from what they actually hear.
Similarly, it has been shown that birds have a strong instinct to learn birdsong,
but they will learn different dialects depending on the social context in which
they sing (Marler & Peters, 1982; West & King, 1980).

Natural languages are very complex rule systems that humans

have special propensities for learning.

Character of Language Acquisition

Children show a characteristic way of approximating adult speech (for more
detail, see Pinker, 1989). It takes a long time before they speak in what adults
recognize as sentences. Almost all children, starting at about 1 year of age, go
through a one-word utterance stage in which all their utterances are single
words, such as“Mommy,” “jump,” and “bird.” Starting at about 18 months, chil-
dren tend to go through a distinct two-word stage, in which their utterances are
either single words or pairs of words, such as“ doggie bark,” “shoe off,” or”there
cow.” These two-word utterances appear to be communicating meanings that
adults would communicate in more complete sentences.

366
 Language Acquisition

TABLE 10.5 Multiword Utterances

Put truck window My balloon pop
Want more grape juice Doggie bit me mine boot
Sit Adam chair That Mommy nose right there
Mommy put sock She’s wear that hat :
No I see truck I like pick dirt up fire truck
Adam fall toy No pictures in there

Source: From Brown, 1973.

Even when children graduate from the two-word stage and start speaking
in longer utterances, their utterances preserve what might be regarded as a tele-
graphic property in that they tend to omit some of the less important words.
Table 10.5 shows some examples of such multiword utterances. Gradually, the
utterances begin to fill out, so that by the age of about 4 children speak essen-
tially in sentences. The sentences may be simple and limited by adult standards
and still contain grammatical errors, but they are recognizable as sentences.
This developmental sequence is unique to children. Adults learning a sec-
ond language try to speak in more complete sentences right from the begin-
ning—even if their sentences are limited and often not grammatical. It has been
conjectured that limited memory capacity may be a reason for the shortness of
children’s utterances (Anderson, 1983). Young children are not able to keep in
mind and plan longer utterances. It has been shown, for instance, that children
have severe limitations in their ability to repeat longer utterances (Brown &
Fraser, 1963). Chapter 5 noted that memory span is related to speed of articula-
tion. Since young children are still learning to speak, their articulatory rate is
much slower (Gathercole & Hitch, 1993), and so they are able to encode less of
the sentences spoken to them and can only plan shorter utterances. Newport
(1990) has argued that this limited processing capacity actually makes language
learning easier. Because young children do not process all the words of a com-
plex sentence, they avoid dealing with many of the most complex aspects of lan-
guage. They can concentrate on getting the core of the language right. This is
called the“less is more” hypothesis.
Another feature of language is that it contains many rules that often apply
to only some of the words in the language, for example, rules for past tense in
English. Most verbs are made into the past tense by adding ed or one of its
phonological variants. There are, however, clusters of exceptions—ring—rang,
sing-sang, and so on. Some words follow their own unique rules, such as eat—ate.
As children learn the complex rules and exceptions of past tense, they go through
a series of stages. First, they do not try to indicate past tense; then, they over-
generalize the dominant rules (e.g., singed); finally, they achieve basic mastery.
Children (and adults) are capable of applying such rules to novel words. So, for
instance, on being told that there is a verb gring (meaning, perhaps,“to splash in
the waves”), children spontaneously use either gringed or grang as the past tense.

367
 CHAPTER 10 Inductive Learning

Children start speaking in short nonsentences and gradually

increase their length and approximation to grammatical sen-
tences.

Theories of Past-Tense Acquisition

In the domains of both categorization and causal inference, there have been
contests between the simple associative explanations of inductive learning and
the more rule-based, hypothesis-testing explanations. The same debate has
taken place in the case of language acquisition. This debate has been particular-
ly detailed concerning the acquisition of past tense in English. The prevailing
rule-based viewpoint had assumed that children learn the basic rules and the
exceptions. Learning may not involve conscious hypothesis testing and rule for-
mation on the children’s part, but the rule-based account claimed that uncon-
sciously children speak according to rules, such as“add -ed” or’ change -ing in
the verb to an -ang.”
Rumelhart and McClelland (1986) offered a major challenge to this prevail-
ing viewpoint and argued that past-tense generation could be explained by a con-
nectionist net, such as that illustrated in Figure 10.13. The root form of the word
(e.g., kick, sing) enters the input nodes and, after passing through a couple of lay-
ers of network, the past-tense form (e.g., kicked, sang) comes forth. The system must
learn associations between the features of roots and the features of past tenses.
These associations are learned according to a variant of the Rescorla—Wagner asso-
ciative rule introduced in Chapter 2 (see Figures 2.15 and 2.16).

Fixed encoding Pattern associator Decoding/binding

network modifiable connections network

Phonological Feature Feature Phonological

representation representation representation representation
of root form of root form of past tense of past tense
FIGURE 10.13 A network for past tense. The phonological representation of the
root is converted into a distributed feature representation. This is converted into the
distributed feature representation of the past tense, which is then mapped onto a
phonological representation of the past tense. Source: From J. L. McClelland and D. E.
Rumelhart. Parallel distributed processing: explorations in the microstructure of cog-
nition. Copyright © 1986 by the MIT PRESS. Reprinted by permission.

368
 Language Acquisition

A computer simulation of their model was trained on 420 root—past-tense

pairs. It successfully learned to generate past tenses for roots on which it was
trained; moreover, it was able after learning to generate past tenses for roots on
which it had not been trained. It also mirrored a particular sequence of general-
ization stages noted of children. Initially, it tended to produce the irregular past
tenses, like came for come; then it generated overregularized forms (comed for
come); and finally it returned to the original, correct irregular forms. Children
appear to have precisely this pattern of development. Rumelhart and
McClelland proclaimed:

We have, we believe, provided a distinct alternative to the view that

children learn the rules of English past-tense formation in any
explicit sense. We have shown that a reasonable account of the
acquisition of past tense can be provided without recourse to the
notion of a“rule”as anything more than a description of the language.
We have shown that, for this case, there is no induction problem. The
child need not figure out what the rules are, nor even that there are
rules. (p. 267; emphasis in the original)

Their bold claim has not gone unchallenged, however. Pinker and Prince (1988)
published an extensive criticism of the Rumelhart and McClelland study, focus-
ing on the details of the model. They pointed out that the ability to account for
the generalization stages depended on initially presenting the network with
many irregular verbs, whereas children do not encounter a special abundance of
irregular verbs early on. Pinker and Prince also pointed out that some of the
errors made by the model were quite unlike errors made by children. For
instance, the model generated membled as the past tense of mail.
Pinker and Prince argued that the English past-tense system is more rule-
governed than can be accounted for by a simple associative network. They used
the example that the same root word can have different past-tense realizations
depending on context. The word ring has two realizations as a verb—to make a
sound or to encircle. Although the root is the same, the past tense for the for-
mer meaning is rang, and for the latter meaning the past tense is ringed, as in:

He rang the bell.

They ringed the fort with soldiers.

Without further elaboration, it is impossible for the Rumelhart and McClelland

network to capture this subtlety of the language.
A substantial debate has been ongoing in psychology about whether the
deficits in the Rumelhart and McClelland model reflect fundamental flaws in
the associative network approach. More adequate connectionist models have
subsequently appeared (e.g., Daugherty, MacDonald, Petersen, & Seidenberg,
1993; MacWhinney & Leinbach, 1991) and been criticized (Marcus, Brinkman,
Clahsen, Wiese, Woest, & Pinker 1995). It is fair to say that the issue is far from

369
 CHAPTER 10 Inductive Learning

resolved. The debate once again illustrates the tension between associative
models of inductive learning and rule-based, hypothesis-testing approaches.

Connectionist models have been proposed that claim that

English-past tense formations can be accounted for by simple
associative learning.

A Critical Period for Language Acquisition

One of the curious features of language acquisition in children is that there
appears to be a critical period during which children are best able to learn at
least some of the features of language. An analogy can be made to critical peri-
ods for learning in other species. For instance, some songbirds (Nottebohm,
1970) can learn the song for their species only if they are exposed to it at a cer-
tain critical:period in their development. For humans, the critical period appears
to stop at puberty, or about age 12. It seems much easier for a child to learn a
first or second language before that period (Lenneberg, 1967). From a certain
perspective, this is a rather amazing claim, since older children are generally
able to learn most subjects faster than their younger siblings—presumably
reflecting the advantage of their greater intelligence.
There are two main sources of data supporting the claim of a critical peri-
od. One source is the ability to reacquire language after a severe brain injury has
resulted in aphasia (loss of language function). Lenneberg (1967) reported that
all children who suffered such an injury before the age of 10 or so were able to
recover full language function, whereas, at best, 60 percent of those sustaining
such injuries after the age of 12 were able to recover language function. It has
subsequently become clear that Lenneberg’s characterization was far too sim-
ple. In fact, it has been argued that there may be no difference between success
of recovery for children versus adult when one controls for cause of the aphasia
(Dennis, 1997). For example, many children suffer aphasia from closed head
injury, while many adults suffer aphasia because of stroke.
The second source of evidence is observations of children when they move
into a new linguistic community, as in the case of immigrants to the United
States or children whose parents move to a new country in response to a cor-
porate assignment. It is often claimed that younger children learn the language
faster than older children or their parents. Such observations are confounded
with factors such as motivation and opportunity. For instance, children are often
forced to become immersed in the other linguistic community at school, are less
self-conscious about trying to learn, and are less resistant to changing their pri-
mary language. Younger children’s utterances are also simpler and judged by a
more lenient standard. In more controlled studies, it appears that older children
are able to learn the same aspects of the language more quickly than their
younger siblings (Ervin-Tripp, 1974).

370
 Language Acquisition

Although older children are able to learn more rapidly initially, they tend
never to master certain fine points of the language as well as do their younger
siblings (Lieberman, 1984; Newport, 1986). One of the most characteristic fine
points is the ability to speak a second language without an accent, which is dif-
ficult if the language is learned after about age 12.
Newport (1990) reported a rare study on the effect of age of acquisition on
first language learning. She studied deaf children of speaking parents who often
did not learn sign language until quite late in their development and did not
master any language during their early years. She found a strong relationship
between the age at which these deaf people started to learn sign language and
their eventual level of mastery of sign language in adulthood. There appeared to
be a particularly large deficit if their learning was postponed until after the age
of 12. The existence of a critical period for language acquisition has been used to
argue for the special character of human language learning. However, as noted
earlier, Newport has argued that children learn better not because they have
special knowledge of language but precisely because they are less capable. This
is her less-is-more hypothesis—that the reduced information-processing
capacity of children actually reduces the complexity of the language hypotheses
that children have to consider.

It is difficult to come to full mastery of a language if it is

learned after the age of about 12.

Innate Language-Learning Abilities?

As we noted in the previous section, several researchers have proposed that
children are born with special innate abilities to learn a natural language but
that these abilities atrophy at about the age of 12. It has also been argued that
learning a language is so difficult that it would be impossible unless children
were born with special knowledge of what a natural language might be like
(Chomsky, 1965, 1986). This position sees human language learning as a
species-specific ability similar to the species-specific learning tendencies
reviewed in Chapters 2 and 3.
The linguist Chomsky (1965) proposed that children are born knowing
language universals, which are features that are true of all languages. Children
do not have to learn these facts; they know them from birth. For instance, the
research we reviewed on deaf children inventing a language (Goldin-Meadow
et al., 1994) seemed to imply that they are born knowing about the difference
between nouns and verbs. Chapter 1 argued that it is more adaptive to code
innately into the organism things that do not vary and that learning should be
reserved for things that vary. The natural languages of the world certainly vary
enormously in vocabulary, phonology, and syntax. However, Chomsky and oth-
ers have argued that they all have in common some deep properties, such as the
distinction between nouns and verbs, that do not need to be learned.

371
 CHAPTER 10 Inductive Learning

Mathematical analyses show that in some sense Chomsky is right—chil-

dren could not learn every possible language; therefore, they must come into the
world biased to learn certain languages, and these languages are the natural
languages (Pinker, 1989). The significance of this argument has generated a lot
of controversy. Humans similarly enter the world only able to use certain kinds
of vehicles (e.g., they could not use a vehicle that required three hands), but no
one would want to claim that humans have innate vehicle-acquisition devices.
Rather, human capacities shape the artifacts they build, and similarly human
capacities may shape the languages they have created. The question is whether
the constraints on language reflect anything more than the general cognitive
and physical constraints that are part of being a human. These constraints may
have shaped the form of languages invented by humans, but they may not be
special to language. For instance, perhaps the noun-verb distinction reflects our
perceptual abilities to identify objects and motor abilities to perform actions.
Chomsky and others have argued that there is something special about
the innate contribution to human language acquisition that goes beyond the
obvious. They tend to emphasize the syntax of all natural language. For instance,
a correlation exists across languages between what is called pronoun dropping
and the existence of expletive pronouns. Some languages, such as Spanish,
allow for optional dropping of pronouns. Whereas English speakers say “I go to
the cinema tonight” with an obligatory “1,”Spanish speakers can say “Voy al cine
esta noche” without a first-person pronoun. Certain languages, such as English,
have expletive pronouns, for example, it and there in the sentences “It is raining”
and“There is no money.”Itis a universal of natural languages that any language
that has optional pronoun dropping does not have expletive pronouns. Hyams
(1986) argued that children are born knowing this rule, and if they notice that
pronouns are being dropped, they know they do not have to learn expletive pro-
nouns. Language induction would be easier if humans were born knowing
these details of language rather than having to figure them out.
There is a second sense in which Chomsky and others have argued that
innate linguistic abilities are special and specific to the human species: No other
organism has innate knowledge about natural languages, and therefore no
other organism can learn the kinds of languages that humans can learn. The
argument is sometimes made that it is linguistic ability more than anything else
that distinguishes humans intellectually as a species.

It is argued that only humans have special innate knowledge

about natural languages.

Animal Language Learning

One way to address the question of whether there is something special about
how humans learn language is to see if other animals can learn a language. The
natural targets for these experiments are the higher apes, which are quite intel-

372
 Language Acquisition

ligent and phylogenetically most similar to humans. Several early attempts to

teach chimpanzees to speak failed completely (Hayes, 1951; Kellogg & Kellogg,
1933). It is now clear (Lenneberg, 1967) that the human’s vocal apparatusis spe-
cially designed to permit speech, whereas the ape’s is not.
Although the vocal ability of apes is limited, their manual dexterity is con-
siderable. Several attempts have been made to teach apes languages using a
manual system. Some studies have used American Sign Language (Ameslan),
which many deaf people use. If apes could become proficient in Ameslan, their
capacity for acquiring a language would be firmly established. One of the best-
known research efforts was started by the Gardners in 1966 on a 1-year-old
female chimpanzee named Washoe (Gardner & Gardner, 1969). Washoe was
raised like a human child, following regimens of play, bathing, eating, and toilet
training, all of which provided ample opportunities for sign learning. After four
years, she had a vocabulary of 132 signs and was able to sign utterances up to
five words in length. Washoe showed some mastery of word order, using utter-
ances such as“ You tickle me” and“I tickle you”to distinguish subject from object.
This was the first study of many establishing that apes have more linguistic
capability than was previously thought.
One of the more impressive demonstrations of ape linguistic capacity was
performed by Premack (1971, 1976; Premack & Premack, 1983). He developed
an artificial language in which the words were colored plastic shapes that could
be attached to a magnetic board. A chimp named Sarah, raised in a laboratory
situation, was trained to use the symbols to make up sentences. Sarah showed
considerable mastery of a number of aspects of language: yes—no interrogatives;
negatives; class concepts of color, size, and shape; compound and coordinate
sentences; quantifiers (all, none, one, several); logical connectives (if...then);
linking verbs (is); metalinguistic utterances (e.g., name of); and who interroga-
tives (what, where, when, etc.). Chapter 6 discussed Premack’s conjecture that
linguistic training, such as that given to Sarah, enabled chimps to develop more
complex propositional representations.
Despite such successes, it has been argued that chimpanzees cannot
approach the proficiency in language that humans can. A particularly negative
report in this regard was published byTerrace et al. (1979), who taught American
Sign Language to a chimpanzee named Nim. They noted that there were sub-
stantial differences between the utterances of their subject and those of chil-
dren. Nim’s utterances tended to be more repetitive, more imitative, and more
stereotypical.
More recent research has produced linguistic competences in great apes
that are not subject to the Terrace et al. criticisms. Perhaps, the most impressive
results have come from a bonobo great ape called Kanzi (e.g., Savage-
Rumbaugh, Murphy, Sevik, Brakke, Williams, & Rumbaugh, 1993; see Figure
10.14). Bonobos are considered to be more like to humans than chimpanzees
but are quite rare. Kanzi’s situation is rather unusual because he was not origi-
nally a subject of a language training procedure but rather from the age of 6
months observed efforts to train a language to his mother. The language his

373
 CHAPTER 10 Inductive Learning

Peony, another of Premack’s chimpanzees, interpreting plastic instructions on the

board.

mother was trained with involved artificial tokens called lexigrams. At age 3,
when he started spontaneously using lexigrams, investigators began to train
him as a subject. His language generations are not of the same repetitive vari-
ety as Nim’s. Also, it was discovered that he had acquired a considerable ability
to understand spoken language. When he was 5.5 years old, his ability to follow
spoken language was compared with that of a 2-year-old child. He was able to
outperform the 2-year-old child.
The accomplishment of outperforming a 2-year-old reflects the glass- halt-
full and glass-half-empty state of research on the language capacity of apes. It
remains unclear what should be the reasonable aspirations for these efforts to

374
 Final Reflections

FIGURE 10.14 The bonobo, Kanzi, listening to English.

train apes. The expectation certainly should not be that they come to process
language identically to adult humans. It is not really a matter of their being infe-
rior to humans; they are different in many ways that have nothing to do with
linguistic ability. Therefore, one should not expect their linguistic learning (or,
indeed, any other learning) to be identical to human learning. One lesson from
the early chapters on animal learning is that learning manifests itself according
to the unique characteristics of each species. Still, the current state of the
research should make us cautious in our assumptions about how much of lan-
guage is unique to humans.

Chimpanzees and other apes can learn languages that are

more limited than human languages.

Final Reflections
Induction is the process by which we make inferences that go beyond our expe-
riences to make predictions about new situations—about whether a creature we
encounter is a dog and whether it is likely to bite, about whether having a high-
cholesterol diet will cause a patient to suffer heart disease, or about whether a
new linguistic utterance will be deemed grammatical. It is a critical aspect of

375
 CHAPTER 10 Inductive Learning

learning, since the function of learning is to allow us to use our experiences in

the past to adapt to new situations in the future.
Any situation in which we manifest learned behavior involves an induc-
tive component, although it is easy to overlook this inductive component in
many human situations. When children spell words as they have been taught,
they are making the inference that what has been accepted in the past will con-
tinue to be accepted in the future. We depend on an implicit social contract in
our lives that the rules will not change. Indeed, much of human society can be
seen as organized to diminish the inductive component in learning (to take the
guesswork out of learning). Scientists study the world and codify its principles,
and educational institutions communicate this knowledge to the next genera-
tion. Products, such as appliances, are made with an eye to quality control so
that they will behave reliably and as expected. They come with manuals that try
with varying degrees of success to make explicit their principles of operation.
Our next chapter, Chapter 11, deals with what the research on learning has to
say about facilitating the process of explicit instruction.
ERT SAON REN SRN A HAN EEN NEM ESTE SEINE TNE BEEN NLT SEE LLL MER ESSEA EIR ENCES SEEN HERES

Modern society is structured to reduce the inductive compo-

nent in learning.
PEERS
LEE INES TERRE TLE DS NM AER ELENA ES LEPELLITER ELLIE LEDS INL LELE NDNA ELELELEE ELLIE PEGA LICL AONE SR

Further Readings
Levine (1975) presents several reports on the hypothesis-testing approach to
concept learning. Smith (1989) provides a survey of research on induction and
concept learning. Shultz (1982) offers an overview of-causal learning, and
Wasserman (1990b) is particularly interested in the connections between causal
inference and animal conditioning. Klahr, Fay, and Dunbar (1993) report an
assessment of children’s skills at scientific experimentation. Pinker (1989, 1994)
provides reviews of research on language acquisition. Holland, Holyoak,
Nisbett, and Thagard (1986) wrote an influential monograph on the many vari-
eties of inductive learning.

376
 Applications to Education

The Goals of Education

An underlying message in the previous chapters was that learning is an adap-
tation of the organism to its environment. To understand an organism’s learn-
ing, it is necessary to understand the learning tasks that the organism faces in
its environment. Chapters 2 and 3 reviewed how animals show species-specif-
ic associative biases; pigeons, for instance have the bias to peck at objects to
receive food and to flap their wings to escape harm. Chapters 6 and 7 reviewed
how human memory is tuned to the statistical patterns by which things reap-
pear in the environment.
When we attempt to understand the situation in which human learning
takes place, it is important to recognize that a major portion of human learning
takes place in situations in which there is some explicit effort at instruction. The
process of teaching the young is hardly unique to the human species, but it
takes on a uniquely large scale with humans. One of the ways in which humans
differ from other species is in the length of childhood—both in absolute years
and as a proportion of the total life span. It has been argued (de Beer, 1959) that
the reason for prolonged childhood is to enable the task of education to be
completed. This argument suggests that to fully understand human learning it
is important to understand its relationship to education.
Although heavy investment in the education of youth has been found in
all cultures throughout the history of the human species, the prominence of
formal educational institutions is a relatively recent phenomenon. The inven-
tion of writing systems a few thousand years ago was the precipitating event for
many changes in human society, including the advent of formal education. A
writing system created the need to learn demanding new skills—reading and
writing—and allowed for the accumulation of a great repository of knowledge,
some of which was deemed worthy of communicating to the next generation.
The beginnings of modern Western schooling can perhaps be traced to early

377
 CHAPTER 11 Applications to Education

schools in Greece (sixth century B.C.) that taught music, reading, and gymnas-
tics (Boyd & King, 1975).
At least since ancient Greek times, some privileged children have received
formal education, but the concept of public education has only taken shape over
the last few hundred years. The current system of public elementary schools,
high schools, and universities in the United States was established in the nine-
teenth century (Good, 1962). The goal of universal education through high
school was articulated in the United States between the two world wars and in
many other modern countries only after World War II. The current curriculum
that occupies public education has been in place less than 100 years.
Higher education has always carried with it a sense of class privilege.
Much of education in previous centuries, such as training in the classics, was
conceived of as placing on students the mark of the upper class—that is, teach-
ing students to be”gentlemen.” Only in the last 100 years has education for all
been justified in terms of its utilitarian content—that is, teaching students
things that are useful to being good citizens (e.g., being able to read and judge
an article about public policy in a magazine) and productive workers (e.g., being
able to run an accounting system for a company). Even today there are serious
questions about how much of public education really is useful and how much
of it is actually devoted to establishing tokens of class privilege (e.g., Lave, 1988).
This chapter discusses some of these current criticisms in the section on math-
ematics education.
There is much debate over the success of the educational system, partic-
ularly in the United States. At one level it is a resounding success: almost all the
citizenry achieves at least a modest level of literacy. The United States has a
technologically sophisticated society that simply would not be possible without
modern education. On the other hand, the U.S. educational system is succeed-
ing much less well than the nation’s citizens would like. As Resnick and
Resnick (1977) documented with respect to literacy, this lack of success results
largely because people have ever higher aspirations as to what the education-
al system should achieve. Although there has always been a very literate elite,
in this century Americans have demanded that more and more citizens be able
to read more and more difficult material. An interesting question is whether or
not these expectations are realistic. One way to answer this question is to com-
pare the achievement of American children with that of children in other edu-
cational systems. If other countries are able to produce higher achievement, the
U.S. system also should be able to do so. This chapter focuses on what psy-
chology has to say about teaching (or learning) reading and mathematics, two
of the core skills targeted by modern education. It is worthwhile considering
the relative international standing of American students with respect to these
two topics.

Formal education has evolved to teach useful knowledge. wit

378
 The Goals of Education

Reading
It is difficult to compare the reading achievement levels of American students
to those of students of other nations because different nations have different
languages and orthographies. Differences in societies also pose a problem in
judging international performance. The United States aspires to keep all stu-
dents in school for 12 years and succeeds with over 80 percent (McKnight,
Crosswhite, Dossey, Kifer, Swafford, Travers, & Cooney, 1990). In other nations,
many students leave the school system earlier. It would not be fair to compare
the top 20 percent of one nation with the greater mix represented by the 80
percent of graduating American students. Societies also vary in terms of their
cultural and economic heterogeneity; the United States has a culturally diverse
society and increasing disparity in the economic standings of its citizenry.
Educational goals are generally easier to achieve in a culturally homogeneous
society, where education can be tailored to the specifics of the one group, than
in a diverse society, where different cultural groups may require different edu-
cational situations to optimize their learning. There is also a very strong corre-
lation between economic status and educational achievement (California
Assessment Program, 1980; Gamoran, 1987). This correlation suggests that the
growing underclass in the United States will be an increasing source of educa-
tional underachievement.
Despite these problems, there is reason to believe that U.S. schools suc-
ceed relatively well at teaching reading. In most international comparisons,
American students do as well as students of most other countries (Stevenson &
Stigler, 1992). When the reading achievements of non-Hispanic, white, middle-
class Americans are compared with those of the ethnically dominant, middle-
class children of other societies, Americans often outperform the other children.
This result suggests that a major mechanism for improving the reading scores of
American children as a group would be to improve the economic standing of the
underclass and to learn how to tailor reading education to minority ethnic
groups. In summary, the U.S. educational establishment does a relatively good
job at teaching reading; the problems that exist are as much societal as educa-
tional.
Figure 11.1 compares American and Chinese students and is representa-
tive of the international comparisons. It shows the number of students graduat-
ing from the first grade with reading scores at various grade levels (1:1 stands
for the first half of first grade and 1:2 for the second half). Note that a significant
number of first graders in the United States read one and two grade levels above
their age and a few read at even higher grade levels, whereas very few Chinese
students read above their grade level. On the other hand, almost 40 percent of
American first graders have not even mastered the lowest level of first-grade
reading, whereas almost all Chinese students have. This chart illustrates the
ereat disparity among reading performances within the United States. The stu-
dents with very low reading scores come disproportionately from economically
disadvantaged families.

B79
 CHAPTER 11 Applications to Education

100

80

ES00
FIGURE 11.1 Histogram show- <5
ing relative proportion of Chinese §&
students (from Beijing) and 6 40
American students (from*Chicago) ie
at various reading levels. Source:
From A. S. Palinscsar and A. L. a6
Brown, Cognition and Instruction.
Copyright © 1984 by Lawrence
Erlbaum Associates, Inc., Mahwah, 6
NJ, pp. 138-139. Reprinted with 2 Td 1:2°0 211 9212 Sa Nae gaiieas2
permission. Reading level

PURER

There
States, with the better students doing at least as well as the top
students of most nations.

Mathematics
The international standing of U.S. students in terms of mathematics achieve-
ment is much worse. In international comparisons, American children enter the
first grade knowing a comparable amount of mathematics (Stevenson & Stigler,
1992) but steadily fall behind until they are one of the poorest-achieving groups
of students (McKnight et al., 1990). Figure 11.2 compares the performance of
U.S. twelfth graders with respect to algebra, geometry, and calculus with the
performance of students at comparable levels in other countries. For each topic,
~ the Americans are near the bottom of all countries surveyed, and Japan and
Hong Kong hold the top two positions. Unlike some of the other countries,
Japan aspires to give all its citizens 12 years of education and succeeds with
more than 90 percent. So this comparison is not an artifact of a relatively select
Japanese student population. Also, Japan and the United States are not very dif-
ferent in the level of reading achievement found in their schools. Several com-
parisons of Japanese and U.S. mathematics education have tried to identify the
cause of the difference in mathematics achievement. In one careful comparison
of fifth graders in Minneapolis and Sendai, Japan, two comparable cities, it was
found that no school in Minneapolis had an average mathematics score higher
than any school in Sendai (Stevenson & Stigler, 1992). This sample included
rich, suburban Minneapolis schools, indicating that the problems with mathe-

380
 The Goals of Education

Algebra

URaI|
a109S
l=

100;—

Calculus
80 }-
Uka|\| 9409S

+ fe) Al

Geometry

rl L

URD|\|
3109S oe

FIGURE 11 2 International comparison of 12th grade achievement scores in (a)

algebra, (b) calculus, and (c) geometry. Source: From C. C. McKnight, F. J. Crosswhite,
J. A. Dossey, E. Kifer, J. O. Swafford, K. J. Travers, and T. J. Cooney. The underachiev-
ing curriculum: assessing U.S. school mathematics from an international perspective.
Copyright © 1987 by C. C. McKnight et al., pp. 22 and 24. Reprinted by permission.

381
 CHAPTER 11 Applications to Education

matics achievement in the United States are not restricted to cultural minorities
or economically disadvantaged children.
McKnight et al. (1990) reviewed a number of popular explanations of the
difference in learning achievement in mathematics. One explanation involves
classroom size, but Japanese classrooms average 41 students and U.S. class-
rooms average only 26. A second explanation is the quality of the teachers.
Teachers in the two countries have relatively comparable training and back-
grounds, but Japanese teachers have more time for class preparation and do less
teaching (an average 23 class hours in the United States versus 17 in Japan).
Japanese teachers also teceive more mentoring and more state direction and
guidance on how to prepare their class presentations. American teachers often
complain that the curriculum is constantly changing, so that they never can
develop expertise in teaching it. Later, this chapter explores the reasons for fre-
quent changes in the U.S. mathematics curriculum.
Probably the major reason for the differences in achievement is the
amount of time actually spent on mathematics education. Large differences
exist at almost all grade levels. In Japan, elementary students spend twice as
much class time on mathematics as do elementary students in the United States
(White, 1987). Time is also used much more efficiently in the Japanese class-
room, and there are fewer distractions (such as announcements, special class
elections, or outings; Stevenson & Stigler, 1992). Many Japanese students go to
after-school Juku classes in which mathematics is the prime subject, especially
in the later grades. Juku classes (White, 1987) are special classes to help students
improve their performance in school and prepare for national exams. It is esti-
mated that the average Japanese parent spends $2000 per year on Juku.! Rohlen
(1998) calculated that the average Japanese student spends 3.2 times more time
on studying between grades 5 and 12 than the average U.S. student.
Interestingly, styles of instruction vary substantially among East Asian countries
like Taiwan and Japan (Stevenson & Lee, 1998) but all produce higher achieve-
ment. The common denominator to Asian success is that much more time is
spent on mathematics.
Societal attitudes probably magnify the basic differences (Stevenson &
Stigler, 1992). Japanese parents and children tend to believe that mathematics
achievement is primarily a result of effort, whereas American parents and chil-
dren tend to believe it is a talent that a person either has or does not have.
American parents, although they may deplore their nation’s achievement in
mathematics, tend to be satisfied with the achievement levels of their own chil-
dren. In contrast, Japanese parents, who may be proud of their nation’s achieve-
ment, tend to think that their own children could do better (Stevenson & Stigler,

“More than 50% of the 5 million lower secondary (7th to 9th grades) school students
attend a juku. In the last seven years, for instance, the amount spent by Japanese par- ~
ents on cram-schools and tutoring has doubled to $10.9 Billion ($109 oku).” From
“For Japanese, Cramming for Exams Starts Where the Cradle Leaves OFS
International Herald Tribune, April 28, 1992, pull!

382
 Psychology and Education

1992). Japanese parents are inclined to help their children with mathematics at
home, whereas American parents think that such help is the province of the
school system. ;
There is always a suspicion that the differences in mathematics achieve-
ment between the United States and Asian countries may reflect some innate
racial differences. Although it cannot be proven conclusively that this is not the
case, there is good evidence that it is not a major factor. Stevenson and Stigler
(1992) reported no differences in general intelligence scores of the population.
The best American students typically do very well in international Math
Olympiad competitions. Invariably, such students are the few American children
who have devoted themselves to mathematics achievement and received exten-
sive support outside the classroom. Although there may well be innate mathe-
matical talent, the more critical variable is effort, as discussed in Chapter 9.
It surely should come as no surprise in the context of this book that time
spent learning has a major impact on learning outcome. In this regard, it should
also be noted that the amount of time spent on reading education is much more
comparable between Japan and the United States, and the outcomes are corre-
spondingly more comparable (Stevenson & Stigler, 1992; White, 1987).
However, although time spent learning has an important impact on learning
outcome, most of this book has been devoted to secondary variables, which
have an impact on how effective learning time is. This chapter focuses on how
learning time can be made more effective.
pieenti

American students do less well on mathematics than students

from many Asian countries because they spend less time on
mathematics.

Psychology and Education

The psychology of learning can suggest ways to make the instructional process
more effective. Corresponding to the two themes in the research on learning,
two approaches to education can be identified: a behaviorist approach and a
cognitive approach.

The Behaviorist Program

Starting with Thorndike and Skinner's attempts to apply their work to educa-
tion, there was a considerable tradition of behaviorist applications to education.
Despite its acknowledged weaknesses, the behaviorist approach remains the
most coherent psychology-based approach that has been applied to education.
Probably the major contribution of the approach was task analysis. Just as a
Skinnerian might decompose a conditioning task into a set of subtasks, so task

383
 CHAPTER 11 Applications to Education

analysis takes a complex skill, such as multicolumn subtraction, and decompos-

es it into a set of what are called behavioral objectives. Consider what is
involved in solving a subtraction problem, such as:

4203
—728.-

Behavioral objectives include knowing simple subtraction facts (13 — 8 = 5), bor-
rowing, the special case of borrowing across zero, and subtracting when there is
only a top number. Each of these behavioral objectives was taught separately.
Gagné (1962) succinctly described what is involved in task analysis:

The basic principles of training design consist of: (a) identifying the
component tasks of a final performance; (b) insuring that each of
these component tasks is fully achieved; and (c) arranging the total
learning situation in a sequence that will insure optimal mediation-
al effects from one component to another. (1962, p. 88)

Each of the components identified by Gagné was essential to the success of the
behaviorist program in education:

a. The emphasis on task analysis is absolutely critical. If teachers know what

they want to teach, they are in a much better position to achieve that goal.
The problem with the behaviorist approach, as this chapter discusses, is
that it sometimes does a poor job of analyzing the components of a com-
plex cognitive skill.
b. The behaviorist methodology is associated with the concept of mastery
learning—ensuring that each component is brought to a level of achieve-
ment. This concept is probably the most profound and controversial aspect
of the behaviorist program. The next section discusses mastery learning in
detail.
c. Gagné is best known for his behavioral learning hierarchies—the idea that
some skills are prerequisite to others. For instance, Gagné would argue
that basic arithmetic skills need to be mastered before a student can
progress to algebra. Many algebra teachers can attest to the difficulty stu-
dents have who do not know their fractional arithmetic. However, recent
technological changes are beginning to challenge this wisdom (Anderson,
Corbett, Koedinger, & Pelletier, 1995). Students can now use hand-held
calculators to do the fractional arithmetic.

Gagné’s statement contained no reference to principles of reinforcement.

Although these principles were emphasized in Skinner’s and Thorndike’s orig-
inal educational proposals, most behavioral applications in the classroom paid
little attention to actually manipulating reinforcement. Rather, they emphasized
immediate feedback in terms of knowledge of results. As discussed in Chapters

384
 Psychology and Education

4 and 6, learning is largely independent of the reinforcement contingencies,

provided that the student processes the material appropriately. This is not to
deny that reinforcement contingencies can often be an effective way to get stu-
dents to process the material appropriately.
A way of designing and delivering instruction, often called instructional
technology, was built around these ideas. Sometimes each student was individu-
ally taught by computer, an approach frequently referred to as programmed
instruction or computer-assisted instruction. Many textbooks for a wide variety of
subjects were written according to the principles of programmed instruction,
and mastery learning principles were articulated in ways to manage whole
classes so that every student reached mastery. The key idea unifying these vari-
ous approaches was that a task can be broken down into a set of behavioral
objectives, each of which could be taught individually.
Some large gains have been reported for these efforts. One spelling pro-
gram got students to mastery in one-third of the time required by conventional
instruction (Porter, 1961). On average the level of gain associated with behav-
iorist programs was probably much less than behaviorist theories would sug-
gest. For advanced topics such as high school geometry, application of the pro-
gram might even lead to lower achievement. A behaviorist program applied to
geometry tended to emphasize instructional objectives, such as knowing the
reflexive law, whereas a cognitive analysis reveals the many different mental
rules that are needed to successfully use this law (see the later section on geom-
etry). Behaviorist analyses had these shortcomings because they ignored the
covert problem-solving steps involved in some problems, and so the units they
identified for instruction only correlated somewhat with what students really
had to learn. To the extent that the correlation was poor, the instruction was less
effective than might be hoped.
Behaviorist educational programs have all but disappeared from the mod-
ern educational scene in the United States. Modern educational writing fre-
quently assumes that the behaviorist approach to education was a failure,
although little hard evidence has been cited. In fact, the reason for their aban-
donment is more a matter of educational fashion. Recent writings have tended
to generalize the perceived failures of the behaviorist program to the conclusion
that any program that attempts to analyze a skill into components will fail. In a
eross misreading of the cognitive psychology literature, it has been claimed that
modern cognitive research has proved that such componential analysis is in
error (e.g., Shepard, 1991). However, modern cognitive psychology only quar-
rels with what the units of analysis are. Given the correct units, there is every
reason to believe that Gagné’s basic program for designing instruction would be
successful.

Behaviorists developed a powerful approach to education but

applied it to a weak analysis of the knowledge to be taught.

385
 CHAPTER 11 Applications to Education

Mastery Learning
One of the most significant outcomes of the behaviorist tradition is a set of
instructional strategies generally called mastery learning. Mastery learning is
based on two assumptions:
1. Almost all students should be able to learn almost all the material in a
standard school curriculum.
2. If students have not learned material early in a curriculum, they will have
more difficulty with material later in the curriculum.
With mastery learning techniques, students are given as much time as they need
to master early material before moving on to later material. This approach guar-
antees learning and makes the learning of later material easier. There are two
types of mastery curricula. One (Glaser, 1972; Keller, 1968; Suppes, 1964)
involves having each student follow an individual course of learning. The sec-
ond, associated with Bloom (1968, 1976), is more appropriate to standard
schoolroom classes and gets the whole class to mastery of a topic before the stu-
dents move on. The latter approach is much easier to manage than the individ-
ualized method, which requires tracking each student separately. The individu-
alized method has been used primarily with college or adult populations, where
individual students are more mature and more capable of managing their own
learning. Many psychology departments (including my own at Carnegie
Mellon) have self-paced introductory courses in which students have to achieve
a particular grade level on a chapter before they can go on to the next. It is up
to the student to know how to study and when to schedule the tests. Often
these are lectureless courses.
In general, mastery programs lead to higher achievement (Guskey &
Gates, 1986; Kulik, Kulik, & Bangert-Downs, 1986). However, it was also
claimed that mastery programs would result in reduced individual differences,
since all students master the prerequisite material (Bloom, 1976). That is, since
all students master the same material in the earlier units, they should take about
the same time to master the material in the later units that depend on these ear-
lier units. Evidence for this claim is weaker than that for the higher achievement
claim (Resnick, 1977), and large individual differences remain in the time to
learn material to reach the mastery level. Mastery learning’s failure to eliminate
individual differences should not obscure the fact that it results in higher edu-
cational achievement for all students. Despite the positive evidence, the educa-
tional establishment tends to view these efforts as failures. This is one of many
examples of how the educational establishment ignores empirical evidence in its
desire to follow fashion.
Many mastery programs introduced in schools have been dropped.
Maintaining a successful mastery program requires a lot of teacher commitment
and energy. The standard classroom is much easier to manage—this issue of
ease of classroom management is a significant issue that has caused many edu-
cational reform effects to fail (Grittner, 1975).

386
 Reading Instruction

Significant instructional gains can be achieved by ensuring

that students have mastered earlier material before progressing
to later material.

The Cognitive Approach

The field of cognitive psychology has not produced anything as well organized
or coherent as the instructional design approach of behavioral psychology. Which
principles reviewed in the previous chapters could form a basis for ideas to make
instruction more effective? A great deal of psychological research has focused on
human memory for declarative facts, which seems a logical starting point.
Perhaps the most powerful idea in this area for education is the importance of
elaborative strategies and organizational strategies for good memory. Chapter 6
reviewed some of the educational programs based on these strategies, for exam-
ple, the PQ4R method, and this chapter reviews some elaborations of these ideas
under the topic of reading. Such techniques are important for learning declara-
tive information, such as that presented in this book, and certainly much of mod-
ern education involves trying to communicate such facts. These study techniques
are important skills, and they need to be taught every bit as much as the more
basic skills involved in the three Rs (reading, writing, and arithmetic).
The research on skill acquisition reviewed in Chapter 9 provides ideas for
instruction on skills like reading and mathematics. Chapter 9 discussed how
complex skills can be decomposed into a large number of production rules and
how the learning of the skills can be analyzed as the learning of the component
rules. This approach implies that the key step in teaching a particular skill is a
cognitive analysis into the component rules and that a Gagné-like program
should be applied to the instruction of these target rules. Componential analy-
sis is the term used for the analysis of instructional material into its underlying
components, which are the basic facts and rules. There is some controversy in
the field of education about what the components should be, but when the skills
are agreed upon, there can be good success in instructing them.

Analysis of instructional material into its cognitive compo-

nents enables more effective instruction.

Reading Instruction
Compared to the situation in mathematics, there is relative agreement about what
the target of reading instruction should be. Society wants its citizens to be able to
read at a level that will allow them to process policy information so that they can
vote intelligently, to process information about commodities so that they can be
good consumers, to process technical information so that they can be good work-

387
 CHAPTER 11 Applications to Education

ers, and in other ways to take in written information so that they can be effective
citizens. As already noted, the educational establishment can claim some success
at achieving these goals, at least for those who are not socially disadvantaged.
However, there has been one major controversy in the teaching of read-
ing. This is the conflict between the phonics method and the whole-word
method of instruction. The phonics method emphasizes training children in
how to go from letters and letter combinations to sound, and from sound to the
words and their meanings. The whole-word method emphasizes direct recogni-
tion of words and phrases and going directly from words to meaning.
In the United States reading instruction has been subject to many swings
of fashion. Until the 1800s, the principal emphasis was on the phonics method.
With the appearance of the McGuffey readers in the 1800s, emphasis switched
to the whole-word method. Emphasis has fluctuated back and forth since then.
Children learn to read under both methods, although studies tend to find that
the phonics approach is somewhat superior (Adams, 1990; Beck, 1981; Chall,
1967; Johnson & Baumann, 1984). Williams (1979) and Perfetti (1985) pondered
why greater use is not made of phonics-based instruction, given the positive
evidence. As Williams lamented,”Today as in the past, data do not carry a great
deal of weight in determining educational practice.” (p. 921). Both Williams and
Perfetti concluded that the whole-word approach, with its emphasis on mean-
ing, is more appealing in that it appears superficially to be in keeping with cog-
nitive trends in psychology, to be more fun, and to be a higher-status skill to
teach than low-level phonetic decoding skills.
The popular press has given a great deal of attention to dyslexia, which is
best defined as underachievement in reading performance. The most common
definition of a dyslexic is a person of normal or high IQ who reads at least two
grade levels below the expected reading level. In a review of dyslexia, Just and
Carpenter (1987) noted that in the United States and Great Britain 0.5 percent of
gitls and 3 percent of boys could be classified by this criterion as dyslexic. The
general public tends to believe that dyslexia is a visual problem involving confu-
sion of letters, but as Just and Carpenter noted, the major source of the deficit is
in making correspondences from symbols to sounds. The phonics method, which
focuses on this aspect of reading, may be particularly appropriate for dealing with
the problems of the dyslexic population (Lundberg, 1985; Perfetti, 1985).

There is somewhat greater success in teaching reading accord-

ing to the phonics method than according to the whole-word
method.

Nature of the Reading Skill

To determine what children should learn, it is useful to look at successful adult
readers and determine what it is they do. Some people believe that reading is
the process of moving the eyes smoothly across the page, but this is not so.

388
 Reading Instruction

Flywheels are one of the oldest mechanical devices known to man. Every

@ @ Ew@ @&
internal-combustion engine contains a small flywheel that converts the jerky

ss) G7)@ss)
motion of the piston into the smooth flow of energy that powers the drive shaft.
FIGURE 11.3 Eye fixation while reading a passage about flywheels. Reading is left-
to-right except for the one regression indicated with the arrows. The fixation times in
msec are given in circles. (Adapted from Just & Carpenter, 1980.) Source: From L.
Resnick et al. Addition and subtraction: a cognitive perspective. Copyright © 1982 by
Lawrence Erlbaum Associates, Inc., Mahwah, NJ, p. 140. Reprinted by permission.

Readers engage in a large number of jumps of eye fixation, called saccades. The
fixations between these quick jumps usually last at least a quarter of a second.
Figure 11.3 illustrates the eye movements of one college student while reading
a text. Readers tend to make one fixation on each word, sometimes skipping less
important words and sometimes making more than one fixation on long or dif-
ficult words. When a reader fixates on a word, that word is centered so that its
image falls on the fovea, which is the most sensitive part of the retina of the eye.
Normally, people are successful only in detecting letters that are close to the
fovea, and they perceive no more than 10 characters to the left or right of the
fovea (McConkie & Rayner, 1974). Thus, people can read at most a few words in
a particular fixation.
As the difficulty of the text increases, the length of fixations, number of fix-
ations per line, and number of regressions also increase. A regression is return-
ing to a previously read word. One minor regression is illustrated in Figure 11.3.
The average adult can read material of average difficulty at a rate between 200
and 400 words per minute; more adults read nearer the 200-word rate than the
400-word rate. The reading rate for adults is limited not by perceptual or ortho-
graphic skills, but by comprehension skills. Adults also cannot follow spoken
text at much faster than 200 to 400 words per minute. For adults, individual dif-
ferences in the comprehension of spoken material are the best predictors of
individual differences in the comprehension of written materials (Jackson &
McClelland, 1979; Sticht, 1972).
There are three logical steps to reading skill. One step is the perceptual
skill of identifying the individual graphemes (letters). For a language with an
alphabet, this skill is relatively easy; for example, English has only 26 graphemes
that need to be recognized. For a language with thousands of characters, such
as Japanese or Chinese, this identification is much more challenging, and chil-
dren spend many years mastering it. The second step of reading is the ortho-
graphic step. Orthography is concerned with going from symbol combinations

389
 CHAPTER 11 Applications to Education

to sound, whereas the perceptual component is concerned with identifying the

individual symbols. Orthography is the major component of reading skill in
English and involves a great many complicated rules and special cases, as all of
us can attest who have struggled with spelling. It is a relatively smaller compo-
nent in languages such as Chinese or Japanese, where symbols can map onto
whole words. The task in these languages is more perceptual and less ortho-
eraphic. The two reading systems can be viewed as involving different design
decisions about how to divide up the task between perceptual work and ortho-
eraphic work. The third stage of reading involves going from the words to the
meaning. This component is not unique to reading but is also a part of listening
to language. These steps may not always be discrete. In particular, skilled read-
ers may go directly from perceptual patterns to meaning, bypassing the need for
an orthographic stage.
The whole-word method of teaching reading is an attempt to minimize
the intermediate orthographic stage. It is a step in the direction of treating
English like a nonalphabetic system. The fact that people learn to read quite well
in languages without alphabetic orthographies should be proof that the whole-
word method can succeed. However, the method does not appear to do as well
for English, which may indicate that it is more efficient to learn to read English
by including an intermediate letter-to-sound stage. Adams (1990) argued that
high-frequency words in English may be processed by the whole-word method
but that most low-frequency words are better dealt with by the phonics method,
because low-frequency words do not receive enough practice to become auto-
matically recognized at the whole-word level.
Logically, it would seem that a skilled reader must be capable of both the
whole-word method because this is the only way to read exception words like
one or sugar and the phonics method because they can read nonwords like wog.
Thus, it seems unlikely to be an either/or matter. Neurological data (for a review
read Coslett, 1997) support the idea that words can be processed in two differ-
ent ways—one that goes directly from the printed word to the meaning, and the
other that goes from the printed word to the sound and then to the meaning.
Brain damage can result in two rather distinct losses to reading competence.
Patients with phonological dyslexia show little change in their ability to read fre-
quent words in the language but suffer considerable difficulty in their ability to
read infrequent regular words or regular nonwords. It appears that such patients
have disturbances to their phonics method of reading but an intact whole-word
method which enables them to read most common words. In surface dyslexia,
patients show preserved ability to read regular words and nonwords but an
inability to read irregular words. They are thus unable to pronounce words like
colonel or yacht but can pronounce words like hand and mint as well as nonwords
like blape. It would appear that such patients have their phonic method intact
but have suffered losses to their whole-word method, which is the only way to
deal with irregular words. It is also the case the children who suffer develop-
mental dyslexia have also been identified who can read common words but not
pronounce regular rare words or nonwords and other children have been iden-

390
 Reading Instruction

tified who can pronounce regular words and nonwords but have trouble with
exception words (Castles & Coltheart, 1993; Manis, Seidenberg, Doi, McBride-
Chang, and Peterson). .
An average successful adult reader has put an enormous amount of time
into mastering this skill—students may have spent 10,000 hours reading by the
time they reach college. The effort put into reading is commensurate with the
complexity of the skill. Over this period of learning, there is a pattern of skill
development such that measures of letter recognition best predict reading skill
in the first grade, measures of orthographic knowledge best predict reading per-
formance in the later elementary grades (Lesgold, Resnick, & Hammond, 1985),
and measures of comprehension of spoken text best predict reading skill in
adults (Sticht & James, 1984). This situation suggests that, as children grow up,
perceptual and then orthographic components become less important, leaving
the most difficult comprehension skills as the critical factor. Even in adulthood
phonetic decoding skills contribute to the prediction of reading level, although
general language comprehension skills contribute more Jackson & McClelland,
1979). Some adults appear to have reading difficulties at the orthographic level.
As noted, dyslexia is a condition associated with impaired ability to perform the
character-to-speech transition, although dyslexics form only a small fraction of
the overall population.
This pattern of development would be predicted from a componential
analysis of reading—letter recognition must be mastered before sound can be
assigned to letter combinations, and assigning sound to letter combinations
must be mastered before the words can be comprehended. The following sec-
tions focus on phonetic decoding skills and language comprehension skills,
which are the two critical components of reading English.

Reading English involves letter recognition, letter-to-sound

conversion, and language comprehension.

Phonetic Decoding Skills

The study by Ehri and Wilce (1983) is typical of research illustrating the impor-
tance of orthographic or phonetic decoding skills in the early grades. They mea-
sured how fast children in the first through fourth grades could read common
words, like hat and boy. They looked at reading speed separately for skilled read-_
ers and unskilled readers. The results are shown in Figure 11.4. In the first grade,
the gap between the ability of skilled and unskilled readers to perform such
simple word identifications was wider, but the differences almost completely
disappeared by the fourth grade. This research implies that, at least in the early
erades, reading ability is strongly related to the speed with which simple decod-
ing can be performed.
Other, more subtle measures indicate that decoding differences between
good and poor readers remain into the later grades. Frederiksen (1981) looked

391
 CHAPTER 11 Applications to Education

te

@ Less skilled
i
2
FIGURE 11.4 Timetoreadawordas £2
a function of grade level. (From Ehri & ie
Wilce, 1983). Source: Figure 11.4 from ‘ 1.0
D. A. Hinsley, J. R. Hayes, and H. A.
Simon in M. A. Just and P. A. More skilled
Carpenter, Eds. Cognitive processes in
comprehension. Copyright © 1977 by
Lawrence Erlbaum Associates, Inc., 0 | | | _
Mahwah, NJ, p. 140. Reprinted by per- 1 2 3 4
mission. Grade

at the speed with which high school students could read pronounceable non-
sense words, such as noke or pight. Figure 11.5 shows the results as a function of
the reading level: students with higher reading abilities were faster at identify-
ing the nonsense words.
Results such as those shown in Figures 11.4 and 11.5 only indicate a cor-
relation between phonetic decoding skills and reading ability; they do not actu-
ally establish that the reading differences are caused by these phonetic skill dif-
ferences. It could be the other way around; that is, better readers may read more
and so develop their decoding skills further. Lesgold et al. (1985) showed that a
student’s phonetic decoding skills in the previous grade predict current reading
comprehension better than reading comprehension ability in the previous grade

1200

Pseudoword
5 1100 decoding
o

E
>
S 1000

s
FIGURE 11.5 Mean speed to reada pro- § 900
nounceable nonsense word as a function &
of reading level. Source: From J. R. +=
Fredriksen in A. M. Lesgold and C. A. ee
Perfetti, Eds. Interactive processes in reading.
Copyright © 1981 by Lawrence Erlbaum re | Ff
Associates, Inc., Mahwah, NJ, p. 369. Re- Lo RES STAY 35 4
printed by permission. Reading ability level

392
 Reading Instruction

predicts current decoding skills. This study implies that the decoding skills
developed in one year lay the groundwork for reading gains in the next year.
To strongly establish the direction of causality, it must be shown that
improving phonetic decoding skills directly improves reading skills. The fre-
quent success of phonics-based reading programs (Adams, 1990; Beck, 1981;
Chall, 1967; Johnson & Baumann, 1984) is one piece of evidence. More focused
studies show that phonemic training does help reading skills, at least in the
early grades. A study by Lundberg, Frost, and Petersen (1988) looked at the
effect of teaching Danish kindergarten children phonological skills to identify
the sounds that make words such as tom. This sound training involved no prac-
tice reading. Children were trained to identify beginning sounds, end sounds,
vowels, and the like. The training proved useful when the children entered a
regular reading program in the first grade. They were able to read more words
correctly, and they maintained this advantage into the second grade. Another
study, by Ehri and Wilce (1987), looked at the effect of spelling training on read-
ing. Spelling is a major exercise in understanding the orthography of the lan-
guage. Kindergarten children who were given spelling training showed better
ability to read new words.
Adams (1990) argued that successful reading depends on practicing these
phonetic decoding skills to the point at which they become automatic. Recall
from the discussion of automaticity in Chapter 9 that people are better able to
concentrate on high-level skills when the low-level components are automated.
In the case of reading, people can concentrate on comprehending the text when
they no longer have to worry about decoding the words.

Improving phonetic deccding skills in the early grades will

improve later reading performance.

Comprehension Skills
Comprehension skills play an increasingly important role in the later grades.
These skills can be analyzed and trained. One of the more successful training
programs was Palinscar and Brown’s (1984) reciprocal teaching program. They
identified four basic strategies that foster comprehension: summarizing, clarify-
ing of difficulties, asking a question that could appear on a test, and making a
prediction about upcoming content. They worked on a one-on-one basis with
seventh-grade students who were performing at the 20th percentile of their
class in reading scores. The researchers modeled how to perform these activities
and slowly encouraged the students to do the same. Table 11.1 illustrates the
modeling with one seventh grader of question-asking skills over a 15-day peri-
od. After this training program, the students were scoring at the 56th percentile,
and they maintained their improved achievement two months later.
The Palinscar and Brown method is simply a more extensive version of the
kinds of study-skill programs discussed in Chapter 6, which work by creating

393
 CHAPTER 11 Applications to Education

TABLE 11.1 The Acquisition of Question Asking Skill

(Note: T = teacher; S = student)
Day 1:
TEXT: The water moccasin, somewhat longer than the copperhead, is found in the southeastern
states. It lives in swampy regions. It belongs, as dovalso the copperhead and the rat-
tlesnakes, to a group of poisonous snakes called pit vipers. They have pits between their
eyes and their nostrils which, because they are sensitive to heat, help the snakes tell when
they are near a warm-blooded animal. Another name for the water moccasin is” cotton-
mouth.” This name comes from the white lining of the snake’s mouth.
1. S: What is found in the southeastern snakes, also the copperhead, rattlesnakes, vipers—they
have. I’m not doing thi right.
: All right. Do you want to know about the pit vipers?
: Yeah.
: What would be a good question about the pit vipers that starts with the word“why?”
: (No response)
: How about,”Why are the snakes called pit vipers?”
: Why do they want to know that they are called pit vipers?
: Try it again.
Cia
Opa
Sy
OD
SO
eo : Why do they, pit vipers in a pit?
: How about,“Why do they call the snakes pit vipers?”
a
aS) : Why do they call the snakes pit vipers?
i Ia nAH
4
DM
HH
NH
HANH : There you go! Good for you.

5 < “NI
TEXT: Perhaps you are wondering where the lava and other volcanic products come from. Deep
within our earth there are pockets of molten rock called magma. Forced upward in part by
gas pressure, this molten rock continually tries to reach the surface. Eventually—by means
of cracks in the crustal rocks or some similar zone of weakness—the magma may break
out of the ground. It then flows from the vent as lava, or spews skyward as dense clouds
of lava particles.
22. S: How does the pressure from below push the mass of hot rock against the opening?
Is that it?
23. T: Not quite. Start your question with,“What happens when?”
24. S: What happens when the pressure from below pushes the mass of hot rock against the
opening?
25. T: Good for you! Good job.

Day 15:
TEXT: Scientists also come to the South Pole to study the strange lights that glow overhead dur-
ing the Antarctic night. (It’s a cold and lonely world for the few hardy people who“ winter
over” the polar night). These“southern lights” are caused by the Earth acting like a magnet
on electrical particles in the air. They are clues that may help understand the Earth’s core
and the upper edges of its blanket of air.
28. 5S: Why do scientists comes to the South Pole to study?
29. T: Excellent question! That is what this paragraph is all about.
LL
:
SSS

Source: From A. S. Palinscar and A. L. Brown, Cognition and instruction. Copyright © 1984 by
Lawrence Erlbaum Associates, Inc., Mahwah, NJ. Reprinted by permission.

394
 Reading Instruction

more elaborative representations of the text. Palinscar and Brown’s effort was
more successful than most programs because it involved 15 days of training,
which is a good deal more than in most experimental studies. Their research
emphasizes two critical points about reading comprehension. First, an impor-
tant measure of reading comprehension is memory, and successful readers are
those who can remember more from what they read. Second, the kinds of skills
that go into achieving good memory performance are hardly automatic and
require extensive training, as does any other kind of skill.
When reading, it is important to be able to appreciate the main points of
paragraphs and how other points relate to the main points. So, for instance, the
points in this paragraph are all organized as evidence for the main point in the
preceding sentence (and what was that point?). Meyer, Brandt, and Bluth (1978)
found that many ninth graders were poor at recognizing the relationship of var-
ious points in a paragraph to one another. Poorer readers tended to be less
skilled in identifying the main points of a piece of text. In a companion study,
Bartlett (1978) found that a program to train students to identify the main points
of paragraphs and their relationships more than doubled their recall perfor-
mance. Once the students identified the main points, they could organize the
rest of the text with respect to these points.
Dansereau and his colleagues (Dansereau, 1978; Dansereau, Collins,
McDonald, Holley, Garland, Diekhoff, & Evans, 1979; Holley, Dansereau,
McDonald, Garland, & Collins, 1979) taught a networking strategy for identify-
ing the main points of a passage. This strategy involved identifying all the ideas
in a text and the relationships among them and then drawing a network show-
ing the relationships. The types of relationships included part of, type of, leads
to (causal), and characteristic of. Figure 11.6 shows a network representation of
a passage on wounds from a nursing textbook. The strategy led to about a 50
percent improvement in the recall of low-GPA students but did not benefit
high-GPA students. Apparently, high-GPA students already had effective strate-
gies for organizing text material.
The studies reviewed here were concerned with reading to extract factual
information from a text. Such a reading strategy is appropriate to texts such as
this, but extracting such information is just one purpose of reading. The purpose
of reading a mathematics or physics text can be different; for example, the goal
might be to extract information about problem-solving procedures. Chi,
Bassock, Lewis, Riemann, and Glaser (1989) showed that successful readers of
such texts spend a great deal of time trying to understand examples of problem-
solving procedures. When reading these examples, they try to imagine them-
selves going through the steps of the problem and compare what they do with
the example.

Reading comprehension can be improved by programs that try

395
 CHAPTER 11 Applications to Education

Scar tissue
is weak
Discussion
of wounds

Take steps to
p p c Aahe
minimize scars

Process of
ee wound healing
t t p p
t t p

: : Lag Fibroplasia Contradiction

c Cc c ql 7
peer ee

t Break No break Only for : .

in skin in skin therapeutic
purposes

-Incision (sharp cutting instrument)

-Abrasion (scraping or rubbing) Blood and Small blood
Granulation vessels
-Puncture/stab (nail, bullet) Serunnfonn tissues
-Laceration (blunt instrument) Fibrin disappear
(fibroblasts
and scar
network and small
shrinks
(scab) blood vessels)
grow along
May occur in
fibrin network
any combination
and gradually
Epithelial absorb it
cells grow
in from
edges
Tissue
continuity

Soft, pink
and
friable

FIGURE 11.6 A network of a chapter from a nursing textbook, from Holley et al.,
1979). Note that p = part of, t = type of, 1 = leads to, and c = characteristic of. Source:
From C. D. Holley, D. F. Dansereau, B. A.McDonald, J. C. Garland, and K. W. Collins.
Contemporary Educational Psychology, Volume 4. Evaluation of a hierarchical mapping
technique as an aid to prose processing. Copyright © 1976 by Academic Press. Reprinted
by permission.

Conclusions about Reading Instruction

Although most students seem to learn to read in most reading programs, the
more successful reading programs are those that identify the critical compo-
nents of reading skill and try to find ways to train these components. Under any

396
 Mathematics Instruction

teaching method, reading skill requires learning and automating a lot of specif-
ic orthographic knowledge, which requires a great deal of time. In addition, a
major part of adult reading skill overlaps with adult listening skills; by practic-
ing listening to complex communications, people are also practicing skills rele-
vant to reading.
Not everything relevant to reading is being taught. The large effects pro-
duced by the training programs of Palinscar and Brown and of Bartlett are evi-
dence that schools are not doing everything they can to instruct reading. In par-
ticular, it seems that the goal of schools is to create students who have mastered
the orthographic component of reading. More effort needs to be given to read-
ing for special purposes, such as retention of concepts and facts and acquisition
of problem-solving procedures; more attention also needs to be given to train-
ing the cognitive components involved in reading for these special purposes.
Perhaps the lack of adequate training of reading skills for particular purposes
comes from an unfortunate tendency to align reading instruction with literature
instruction. The goal in reading a poem or a novel is literary appreciation.
Although undoubtedly an important ability, this is only one reason for reading.
A major flaw in most educational programs is the belief that reading for literary
appreciation prepares a student to read for other purposes. This belief reflects a
lack of task analysis, which would readily reveal the different goals of different
reading activities.

Reading ability can be improved by teaching how to read for

particular purposes.

Mathematics Instruction
Controversy surrounds the subject of what the target of mathematics instruction
should be, and this controversy is reflected in the waves of curriculum reform
that have occurred in U.S. mathematics education. In the 1960s, the new math
movement attempted to conform to modern conceptions of mathematics and to
develop mathematics in school as a mathematician would. This movement was
followed by the back-to-basics movement in the 1970s and 1980s, which
emphasized developing perfection of traditional mathematical skills, such as
addition facts or the solution of linear equations. This approach has now large-
ly been replaced by what is termed the constructivist mathematics (called” fuzzy
math’ by its critics—e.g., Gardner, 1998), which focuses on having children dis-
cover mathematics for themselves and relate mathematics to their experiences
in everyday life. This newest movement is motivated in part by the observation
that many students do not value mathematics and do not see it as having any
practical role in their lives. Each of these movements has some intrinsic merit.
However, there was nothing about society in the 1960s that made formal math-
ematics more essential then, or about the 1980s that especially required basic

397.
 CHAPTER 11 Applications to Education

skills, or about today that especially requires making mathematics practical and
relevant. The goals of formal appreciation, proficiency, and practicality are
always desirable. Each has its constituency, and the different fashions reflect
which constituency momentarily has the attention of mathematics education.
As noted earlier, the result for teachers is disaster. Teachers no sooner learn how
to teach to promote one goal than they find that the curriculum goals have
changed. Few nations undergo such rapid changes in the goals of their mathe-
matics curriculum as the United States.
Fad is not the only reason for curriculum change. The nature of mathe-
matics is changing. Mathematics is a developing subject, and there are new
domains that may be important to teach to children. The computer revolution is
requiring a major rethinking of the kind of skills that should be expected of chil-
dren. As much as a year of a child’s mathematics education may be spent learn-
ing the long-division algorithm. This seems rather wasteful in the era of calcu-
lators. Schools are abandoning teaching the long-division algorithm just as they
long ago gave up teaching children the algorithm for calculating square roots.
Algebraic skills that were formerly the domain of high school mathematics
classes and college calculus classes can also be embedded in a hand-held calcu-
lator. Calculators are available that can solve an equation with the push of a but-
ton, differentiate it, or integrate it. Everywhere educators must ask themselves
what is still important to teach. Also changing are the practical needs of the
nation. There is relatively little need to perform routine arithmetic calculations
and greater need to be able to use mathematically based computer software,
such as spreadsheets. The need for computer literacy has increased the need for
students to understand discrete mathematics. Developments in mathematics
and the social sciences require a citizenry that is much more sophisticated in
statistics. For instance, almost everyday the news contains claims about the
effectiveness of some social program; these claims cannot be evaluated without
a considerable understanding of the statistical basis for that claim.
Another problem in achieving consensus with respect to mathematics
education lies in the different conceptions of the nature of mathematical talent.
As mentioned earlier, Americans hold an unhealthy belief that mathematical
talent is a gift and not something developed through extensive practice. It is part
of the general Western illusion that genius is something that should come
effortlessly and in great flashes of insight. Since mathematical talent is so close-
ly tied to the Western conception of intelligence, it bears an unfortunately heavy
load of this misconception about the nature of talent. The worst part of this mis-
conception is a resistance to believing that mathematical talent can be analyzed
into its component skills and that these skills can be taught.
This section considers several mathematical skills that have been success-
fully analyzed into their components, including knowledge of basic arithmetic
facts, multicolumn subtraction skills, solution of algebraic word problems, and
proof skills in geometry. This section ends with a review of the debate about the
value of such mathematical skills.

398
 Mathematics Instruction

Mathematics instruction has been hampered by the rapidly

changing goals for the mathematics curriculum.

Basic Arithmetic Facts

Strange as it may sound, there are not that many basic arithmetic facts to be
learned in a base-10 number system. Exactly how many there are depends on
just how they are counted. Is 3 + 4 = 7 separate from 4 + 3 = 7? Is the addition
fact 3 + 4 = 7 separate from the subtraction fact 7 — 4 = 3? The same questions
can be asked about multiplication and division. Is 3 x 4 = 12 different from
4x 3 = 12 and 12 + 4 = 3? By any way of counting, there are not that many
facts—not more than 500 and perhaps as few as 100. Still, learning these facts
poses a considerable challenge for children. As discussed in Chapter 7, these
facts define a horrendous interference paradigm in which a number, for exam-
ple, 3, is being associated with a large number of interfering facts: 4 x 3 = 12;
2—42 3:34 3 = 6:3 %o = 9: 6—3 =.3) and.so on.
Siegler (Siegler, 1988; Siegler & Shrager, 1984) completed a successful
analysis of how these facts are learned, focusing on the more basic addition and
multiplication facts. He noted that most children know a backup strategy for
solving these problems, which they can use if they cannot remember the facts.
In the case of addition, this backup strategy is counting. Thus, to add 4 and 3,
some children can be observed to count out 4 fingers, then count 3 more, and
then count that they now have 7 fingers. Other children count silently to them-
selves. There are a number of variations on this counting strategy, and some are
more efficient than others. The strategies are all sound mathematically, although
young children often make slips in trying to execute the counting strategies and
come up with wrong answers. The backup strategy for multiplication is repeat-
ed addition. To multiply 3 x 4, the child just adds 4 three times. If addition has
been mastered, this backup strategy for multiplication is also a mathematically
sound strategy, but one in which the students make frequent errors, such as 6 x
4 = 18 (forgetting to add one 6).
At the same time that they are using these backup strategies, children try
to memorize the facts, since recall is a much faster and ultimately less error-
prone way of solving these problems. Once they can automatically recall the
facts, they free up working-memory capacity for higher-order problem solving.
As a simple example, it is very difficult for children to execute the repeated addi-
tion algorithm for multiplication if they also have to execute the counting algo-
rithm for addition. The degree of learning of specific facts is a function of how
often a particular fact is encountered. Children learn simple facts, such as 2 + 2
= 4, faster than facts such as 4 + 7 = 11, because they encounter them more
often—another testimony to the effects of practice reviewed in Chapter 6. A
major complication is that the errors children make in their backup computa-

399
 CHAPTER 11 Applications to Education

tions create false facts, which can interfere with memory. Also, similar facts will
interfere with one another. Occasionally, children display evidence of these
interference problems; given a problem such as 3 x 4 = ?, they recall answers
such as 15 (giving a different answer from the multiplication table) or 7 (con-
fusing multiplication and addition). Siegler showed that children are sensitive to
their state of knowledge of specific facts and only begin to recall them when
they are fairly sure of producing a correct recall.
Siegler’s analysis is a triumph of the componential approach. Children’s
ability to reproduce the addition and multiplication facts rests on mastering spe-
cific strategies and memorizing specific facts, one at a time. Final mastery is
achieved when the child has built up enough strength for each correct fact to
overcome the interference inherent in the addition and multiplication tables.
Further practice brings automaticity, which facilitates using this knowledge in
more complex algorithms. For instance, Haverty (1999) has found that by
increasing students’ fluency in basic mathematics facts, she would improve their
ability to solve induction problems.

There are relatively few arithmetic facts to be learned, but they

suffer from high interference.

Multicolumn Subtraction
Multicolumn subtraction involves solving problems, such as:

3206
—1147

Production rule theories, such as those considered in Chapter 9, can be used to

model such skills (Van Lehn, 1990; Young & O’Shea, 1981). Much of subtraction
skill can be modeled by the seven rules given in Table 11.2. Van Lehn (1990)
studied typical mathematics texts and found that each lesson tends to introduce
one new tule. Children at different levels of competence can be modeled as
knowing different numbers of these rules.
What do children do when they come upon a problem that has a step that
they cannot perform? Often they do not just fail to do the problem, but they try to
invent some answer to fill the gap in their knowledge. Sometimes they come to
believe their inventions and display what Brown and Van Lehn (1980) called bugs.
A bug is a wrong rule that leads the child to make systematic errors. Burton and
Brown tabulated over 100 such bugs. Table 11.3 displays some of the most fre-
quently encountered bugs. The most common error is to always subtract the
smaller number from the larger in order to avoid the need for borrowing. Many of
the errors involve inventions to deal with the problem of borrowing across zero.
Burton (1982) developed a diagnostic program called BUGGY, which can
take a student’s performance on subtraction test problems and automatically

400
 Mathematics Instruction

TABLE 11.2 Production Rules for Multicolumn Subtraction

IF the goal is to solve a subtraction problem
THEN make the subgoal to process the right-most column.

IF_ there is answer in the current column

and there is a column to the left
THEN make the subgoal to process the column to the left.

IF the goal is to process a column

and there is no bottom digit
THEN write the top digit as the answer.

IF the goal is to process a column

and the top digit is not smaller than the bottom digit
THEN write the difference between the digits as the answer.

IF the goal is to process a column

and the top digit is smaller than the bottom digit
THEN add 10 to the top digit
and set as a subgoal to borrow from the column to the left.

IF the goal is to borrow from a column

and the top digit in that column is not zero
THEN decrease the digit by 1.

IF the goal is to borrow from a column

and the top digit in that column is zero
THEN replace the zero by 9
and set as a subgoal to borrow from the column to the left.

identify the student’s bugs. In one experiment, BUGGY processed the solutions
of 1300 students and found that 40 percent had systematic bugs. The program
has also been used to help train teachers to diagnose various bugs.

Students’ subtraction errors can be explained by what correct

production rules they are missing and what incorrect rules
they have.
sponses

Algebraic Word Problems

Children’s ability to perform the algorithmic mathematical skills taught in
school is often divorced from their ability to use these skills in solving real-world
problems. Carraher, Carraher, and Schliemann (1985) reported a study of
Brazilian schoolchildren who also worked as street vendors. These children were
capable of solving addition and subtraction problems in the marketplace (What
%

401
 CHAPTER 11 Applications to Education

TABLE 11.3 Some of the More Common Bugs in Subtraction

1. Smaller-from-larger. The student subtracts the smaller digit in a column from the larger digit
regardless of which one is on top.
326 542
-/17 Senlndaisicadcded
all 247
2. Borrow-from-zero. When borrowing from a column whose top digit is 0, the student writes 9 but
does not continue borrowing from the column to the left of the 0.
& Z2 8 D2
oat, ST -396
265 soo
3. Borrow-across-zero. When the student needs to borrow from a column whose top digit is 0, he
or she skips that column and borrows from the next one.

5g02 "foe
-3a7 -4586
aZas 306
4. Stop-borrow-at-zero. The student fails to decrement 0, although he or she adds 10 correctly to
the top digit of the active column.
70,3 60,4
-678 -~367
179 307
5. Don’t decrement-zero. When borrowing from a column in which the top digit is 0, the student
rewrites the 0 at 10 but does not change the 10 to 9 when incrementing the active column.

For 'L,0,5
- 368 9
344+ 1/06
6. Zero-instead-of-borrow. The student writes 0 as the answer in any column in which the bottom
digit is larger than the top.
326 S42
= ist “389
zIO zoo
7. Borrow-from-bottom-instead-of-zero. If the top digit in the column being borrowed from is 0,
the student borrows from the bottom digit instead.
TOr 50,6
-3 £8 -489
4354 03
Source: From L. Resnick in T. P. Carpenter, J. M. Moser, and T. A. Romberg, Eds. Addition and sub-
traction: a cognitive perspective. Copyright © 1982-by Lawrence Erlbaum Associates, Inc., Mahwah,
NJ. Reprinted by permission.

is the cost of five lemons at 35 cruzeros apiece?) when they could not solve the
equivalent classroom problem (5 x 35 = ?). Perhaps more disturbing to mathe-
matics educators is evidence of children who can solve the formal mathematics

402
 Mathematics Instruction

TABLE 11.4 Examples of Algebra Problem Types

Category Example
Triangle Maria walks one block east along a vacant lot and then two blocks north to
a friend’s house. Phil starts at the same point and walks diagonally through
the vacant lot coming out at the same point as Maria. If Maria walked 217
feet east and 400 feet north, how far did Phil walk?
Distance-rate-time _In a sports-car race, a Panther starts the course at 9:00 A.M. and averages 75
miles per hour. A Mallotti starts 4 minutes later and averages 85 miles per
hour. If a lap is 15 miles, on which lap will the Panther be overtaken?
Interest A certain savings bank pays 3% interest compounded semiannually. How
much will $2500 amount to if left on deposit for 20 years?
Area A box containing 180 cubic inches is constructed by cutting from each cor-
ner of a cardboard square a small square with sides of 5 inches and then
turning up the sides. Find the area of the original piece of cardboard.
River current A river steamer travels 36 miles downstream in the same time that it travels
24 miles upstream. The steamer’s engines drive in still water at a rate that is
12 miles an hour more than the rate of the current. Find the rate of the cur-
rent.
Number The units digit is 1 more than 3 times the tens digit. The number represent-
ed when the digits are interchanged is 8 times the sum of the digits.
Work Mr. Russo takes 3 minutes less than Ms. Lloyd to pack a case when each
works alone. One day, after Mr. Russo spent 6 minutes packing a case, the
boss called him away, and Ms. Lloyd finished packing in 4 more minutes.
How many minutes does it take Mr. Russo alone to pack a case?

Source: D. A. Hinsley, J. R. Hayes, and H. A. Simon (1977). From words to equations: Meaning and
representation in algebra word problems. In M. A. Just and P. A. Carpenter (Eds.) Cognitive process-
es in comprehension. Copyright © 1977 by Lawrence Erlbaum Associates, Inc., Mahwah, NJ, pages
93-94. Reprinted by permission.

problems but who cannot apply mathematics outside the classroom. This issue
has been studied using algebraic word problems, where students must use
knowledge of algebra to solve problems stated verbally.
Table 11.4 gives some of the examples of the algebraic word problems fre-
quently used in high school geometry texts. Many students who have mastered
the mechanics of algebra find such problems difficult. Mayer (1987) and Singley,
Anderson, Givens, and Hoffman (1989) conducted task analyses of what is
involved in solving such problems. These analyses identified four major stages:
comprehension, equation embellishment, combination of information, and
algebraic symbol manipulation.
1. Comprehension. Although their language comprehension abilities are gen-
erally adequate, many high school students lack the ability to process appropri-
ately the kinds of linguistic expressions that are used to communicate mathe-
matical relationships. A particularly notorious example was studied by Soloway,
Lochhead, and Clement (1982), who asked subjects to translate the following
assertion:”There are six times as many students as professors at this university.”
Many students translated this as 6S = P rather than 6P = S.

403
 CHAPTER 11 Applications to Education

The output of this comprehension phase should be a set of equations that

summarize the information in the problem. Consider the following problem:

A picture frame measures 20 cm by 14 cm; 160 square cm of the pic-

ture shows. What is the width of the frame around the picture?

The reader might be helped by Figure 11.7, but such aids are typically not given
to the student. The output of comprehending this problem should include
assignments, such as:
a. Total length = 20
b. Total width = 14
c. Picture area = 160
This list represents some of the equations required to solve the problem, but, as
shown in the next stage, it does not include all the equations.
2. Equation embellishment. Problems like this picture-frame problem require
recognizing the type of situation and stating a set of equations to describe that
situation. For this problem, the additional equations include:*
d. Total area = Picture area + Frame area
e. Total area = Total width x Total length
f. Picture area = Picture width x Picture length
g. Total length = Picture length + 2 x Frame width
h. Total width = Picture width + 2 x Frame width
Since this information is not contained in the problem statement, students must
embellish the problem with an appropriate mathematical model, as shown.
Students often find it easier to come up with these relationships if they draw a
diagram such as that shown in Figure 11.7. The example given is a case of the pic-
ture-frame problem, but there are many other types of problems, and students
appear to master each separately. Table 11.4 includes some of the other types.
Mayer (1987) found approximately 100 problem types in typical algebra texts.
3. Combination of information. The information has to be combined to find
the desired quantity. If x is the width of the picture frame, these steps of combi-
nation might be
i. 20 = Picture length+ 2x (combining a and g)
j. 14 = Picture width + 2x (combining b and h)
k. 160 = (20 — 2x)(14 — 2x) (combining c, i, and j)
Students have great difficulty in seeing the appropriate combinations of equa-
tions to achieve the quantity they want.

* All these equations are required to solve this problem, but they do come up in other
picture-frame problems.

404
 Mathematics Instruction

/————— Total width (14) ————>|

|e Picture width (?) >

Frame
Picture Z width (x)
area (160)

(20)
length
Total

FIGURE 11.7. A diagrammatic repre- ——>

/k-——Picture
(?)
length
sentation of the situation described by Total
area(?)
the word problem. EE

4, Algebraic symbol manipulation. Finally, various symbol manipulation skills are

needed to solve the equations. In this case, the student has to do some algebraic
rearrangement and then solve a quadratic equation. Thus, equation k becomes
4x? — 68x + 120=0
which can be simplified to
oor y= 0
which can be factored into
(x -—15) (x-2) =0
which shows that the one acceptable solution is x = 2. Some students have dif-
ficulty with this stage, but, as noted, others have difficulties with the other com-
ponents.
Singley et al. (1989) showed that students have great difficulty with these
problems partly because not all of these steps are explained to them. The com-
ponents of algebraic symbol manipulation are sometimes well taught, but the
other components of solving word problems are not explicitly taught. Singley et
al. were able to substantially enhance students’ performance by explicitly teach-
ing these steps.

Solving algebraic word problems involves four stages, and stu-

dents do not receive instruction on all these stages.

Geometric Proof Skills

A traditional geometry course with a focus on doing formal proofs in a
Euclidean system is a frustrating course for most high school students. Students
often rate it as their least favorite course (Hoffer, 1981). A typical geometry class

405
 CHAPTER 11 Applications to Education

(a) X

W iu GIVEN: XY = XZ, ZWMY = ZTMZ

M midpoint of YZ

4 M Z Prove YT= ZW

STATEMENT REASON

M is midpoint of YZ Given
YM = MZ Definition of midpoint
OY Given
LXV LON EXE Base angles of isosceles triangles
ZWMY = ZTMZ Given
AWMY = ATMZ Angle-side-angle (ASA)
WY =TZ Corresponding parts
WL 2 Reflexive
AWYZ = ATZY Side-angle-side (SAS)
YT =ZW Corresponding parts

(b) YT =ZW
Y
CORRESPONDING
PARTS

AWYZ = ATZY
Y
SAS
fies
ieee BING
VZ=VZ WY =TZ

REFLEXIVE CORRESPONDING
PARTS

AWMY = ATMZ

YM = MZ LXYZ = LXZY

Midpoint lsosceles

ZLWMY = ZTMZ ___-M Midpoint of XY = XY

Ve
FIGURE 11.8 (a) A proof problem; (b) a flow-proof representation of the logical
structure of inferential support.

is characterized by low class morale and low levels of achievement. In an inves-

tigation of what is involved in the skill-of doing proofs in geometry, Anderson,
Bellezza, and Boyle (1993) discovered that much of what is involved in the skill
of proof construction is not being taught to students. First, the typical two-col-
umn proof format hides from students the overall structure of a proof problem.
Therefore, a tree structure was introduced to illustrate the proof. Figure 11.8
shows the contrast between the typical, linear, two-column proof structure and
the graphic, or flow proof. The flow proof connects the statements below that
support a conclusion above.

406
 Mathematics Instruction

Second, to create a proof in geometry students need to know a lot of

strategic information about when to make various rules of inference. For
instance, geometry has the reflexive rule, which states that any segment is con-
gruent to itself. This is usually a rather useless rule of inference, but on some
occasions it is critical to a proof. One such occasion is when a student needs to
prove that two triangles are congruent and the two triangles share a side. In this
instance, it is useful to establish that the two triangles have one pair of congru-
ent sides, because the shared side is congruent to itself. This rule is used in this
way in Figure 11.8, where the inference is made that YZ is congruent to YZ so
that AWYZ can be proved congruent to ATZY. The researchers found that much
of geometry competence could be modeled by means of special production rules
that contained such strategic information, for instance:

IF the goal is to prove AABC congruent to ADBC

THEN conclude that BC is congruent to BC by the reflexive rule.°
Anderson et al. developed an intelligent tutoring system (such systems are
described in the next section) for instructing such geometry skills, which result-
ed in a large improvement in the achievement level of students. Moreover, as
Schofield and Evans-Rhodes (1989) described in their studies of these geome-
try classrooms, there was a dramatic change in the attitude and motivation of
the students. Able to succeed at the task, they found geometry fun.

Geometric proof skills are better learned when a student is

given more direct instruction on the underlying components.

Intelligent Tutoring Systems

The major lessons in each of the case studies just reviewed are that mathemat-
ical competence can be decomposed into a number of underlying rules or facts
and that the course of mathematical learning is the course of acquisition of
these individual components. The Pittsburgh Area Cognitive Tutoring Center at
Carnegie Mellon University (Anderson et al., 1995) has developed an approach
to instruction based on such a componential analysis. It involves having a com-
puter deliver instruction with the goal of providing individualized instruction for
each student and optimizing that student’s learning of the component rules. The
approach is an attempt to take the instructional design approach of the behav-
iorist program and apply it to a more accurate model of the underlying cogni-
tive skills to be taught. These systems are often called intelligent tutoring sys-
tems (Polson & Richardson, 1988; Wenger, 1987) because they combine cogni-
tive models with techniques from artificial intelligence to achieve computer-
based instructional interactions with students.

3 Where the letters A, B, C, and D are variables that can match to any letters in the
problem.

407
 CHAPTER 11 Applications to Education

The first step in creating such a tutor is to work with educators to develop
a computer interface for the instruction of the skill. This interface will be a pow-
erful system for doing mathematical problem solving. Figure 11.9 illustrates a
system for solving problems in algebra. It consists of a spreadsheet, a graphing
facility, and a symbol manipulation facility. The tutor teaches students to use
these three types of mathematics software to solve real-world problems, such as
the one illustrated in Figure 11.9.
After designing the interface, the next step is to develop a cognitive model
of the knowledge that students must have to solve the problems posed in that
interface. Such a modelis created as a production system and is quite capable
of solving the problems given to students. This model serves two major instruc-
tional goals. First, the production rules are the instructional objectives. Second,
because the model is created as a working program, it can run along with the
student to help the computer tutor understand what the student is doing to
solve the problem.
Then the instructional component is prepared. This includes instruction
on the individual productions and a system of hints and corrections that the
computer tutor can provide during the problem solving. The tutor is able to
point out to students where they are making errors and steer them in the cor-
rect direction, avoiding much of the confusion and wasted time that normally
occurs as students practice a problem-solving skill.
Finally, a curriculum plan and a set of mastery criteria are prepared. These
are used to guide the students through the material in a way that ensures that
they reach mastery on the underlying production rules. Figure 11.9 shows a

Student working with a computer tutor.

408
 Mathematics Instruction

?
4o3n{ aAniuba) | euqabry

slaquunu peuoyedabaqueuou Busy

adojs aAyebau ‘uolssaidxa ayn,

adojs aAnisod “uolssaudxa aun),

suaquiny abse) Busy)

Sahessayy
sytun BuKyquapy
k

) Busyjeus
quunu
sua
Buse)
suafajui puly
aayebau’)
adojs‘x
aAyebau
adojs
puly adojs
puly
“y%
aanisod
Wd
Buvaquy
edie

puly
“A
aanisod
adojs

40:9anf
juapnis

G2@S
Penqns

=X1¢0+S¢S
0S8 SZE
X10
=

FEUM
Aq
éj12S)
Op
0}
yI0q
SepIs
0]
386
X
Ue?
NOA

SIQeLIeA
jeu! jsoo
JOO)

LUOY
WO
‘Slay|IIM USdSn U4]

©} UWS S|Y] “YON. MOY AUELU SO|ILUUBD 8M

B £Z

[EIU
00S 1! BulALUp 2q IIIA 2MJIZH Wow) 49N.Q Oy] JUS4 0} S091 PINOM YON MOH] “E

SIL
|IIM UAL

£ZHOH
ebseyo

Sa ASN
Ue

JO
saludo
‘ZLWaH

JSOD PUB
WY
“AuBduiod
Jeyjouy ‘Auedwios

SY] USALIP
YONA
FEUY
POU OS SEI/LU
/RQUaH
oMbUBDS
dja

ANY
JO
smopuLM

“ojiuied
“USALIPsug

S}4] JOqUUNU
SY]
HuAUSA
JSSYSHAOAA
“SSM

SALIP
94]
DSBS
0] sn Joy ¢7¢$ snid 120gsed “aij

Jo O]
YO!WS SjILU

JO)
BIUI9I She} JON.QJOY SUO

“SHE}OP PSSU
ydeig saalos

sues zis yongJos ¢ag$ snid Od€

S[qeeA
WO) &ZUPH
e [Jo]JO

SM
payebpnq
coe

Ul
PIMOM
}S09Jo}

YSOo
peuoRIPpe

B
SULJSp
jEAIU! SBIILU sd¥7100
SH¥71100
SAA
Usd
I

aSIAY Wold {Sell

“BINUUOS Z2143H
SATIN
N3AldC
“| J} ara eaey

Uy AUBLUMOH
31pz

SAPJY aM
PaoU
0]

SIIEUA,
Sy]
snjdue
YONA
aly

104
SA\

'Z “p
éSlay pinom
3q
élenbs
yoes
Aueduios84]
ZS}
0}
SLIM
SEIN
PUYON
S4y
ISO9
BuQUS!
Sy]
ISOO
Jo
BUNUSI
ou]
Yong
WOY
10) LUOY
“SLAY
PB

ajdwexy
jo
ay}
eigasye
prom
wia,qoid
AUNOL
6IL
‘Ioyn}

409
 CHAPTER 11 Applications to Education

window that displays for students how well they are doing on the component
skills (production rules). Skills are associated with identifying the units of mea-
surement, determining appropriate bounds for the graph, and identifying the
points of intersection.
Table 11.5 presents a short interaction with the LISP tutor, which has been
teaching a course in LISP at Carnegie Mellon University since 1984. LISP has
been the main programming language of artificial intelligence. The student in
this example is writing a function to convert temperature from Celsius to
Fahrenheit. The table shows the tutor and the student interacting in a series of
cycles in which the student types some code, the tutor responds with a com-
ment, and the student writes some more code. Note that the tutor can monitor
what the student is doing on a symbol-by-symbol basis, judging each symbol as
it comes in. It can provide such articulate instruction because, behind the scenes,
it is solving the problem in a symbol-by-symbol basis just as the student is. The
tutor can provide cogent help and correction as needed. Students working with
the LISP tutor are able to reach the same level of achievement as students in a
conventional classroom in one-third the time (Corbett & Anderson, 1990).
A number of tutors, such as those for algebra and geometry, have achieved
similar successes in instruction. In each case, success depends on a careful
analysis that identifies the cognitive components (production rules) underlying
successful performance. With these components identified, the tutor is then able
to provide intelligent instruction directed to the underlying skill. As Table 11.5

TABLE 11.5 The LISP Tutor Interacting with a Student

The student is working on writing a piece of code to satisfy the problem description below. Before
each point where the tutor and student interact, the partial code that the student has written is
printed out.
The Problem Description
Define a function called ctof that takes a degree reading in Celcius degrees and converts it to
Fahrenheit degrees. First, you need to change the scale by multiplying the argument by 1.8. Then
you need to translate the scale by adding 32, since 0 Celcius = 32 Fahrenheit. For example, (ctof 20)
returns 68.
Code: (defun ctof (cdeg) (*))
Tutor: Yes, you will need to call“*” to get the degree on the enlarged scale, but you are ultimately
trying to get the Fahrenheit of the Celsius degree. So you need to code a different func-
tion here that takes a call to“*”as an argument.
Code: (defun ftoc (cdeg) (+ <NUM> <NUM2>))
Student: Help
Tutor: You need to multiply together the Celsius degree reading and the size of Fahrenheit
degrees using the function *. When you multiply these numbers you will get the degree
on the enlarged scale, which is 36 in our example.
Code: (defun ctof (cdeg) (+ (* 1.8 cdge) <NUM2>)))
Tutor: Did you mean cdeg?
Student: Yes
Code: (defun ctof (cdeg) (+ (* 1.8 cdeg) 32 )))
Tutor: You are done. Type next to go on after testing the function you have defined.

410
 Mathematics Instruction

illustrates, this instruction can be tailored to the particular student. The success
of the tutors is evidence that the learning of a complex skill is produced by the
learning processes reviewed in this book taking place on the individual compo-
nents of the skill. Today thousands of high school students throughout the
United States are learning their mathematics from such tutors (Koedinger,
Anderson, Hadley, & Mark, 1997). Evidence indicates that they are performing
about one standard deviation or a letter grade above students in comparable
control classrooms. I expect that some of the graduates of this program will go
on to college and read this text.
The approach taken by the tutors has a lot in common with the mastery
learning approaches reviewed earlier. Although the tutors typically result in
somewhat higher achievement gains, other mastery approaches also result in
achievement gains. Nonetheless, traditional mastery classrooms largely disap-
peared in part because such classrooms were difficult to manage. On this score
the tutoring approach has an advantage. Each computer tutor is in effect a
teaching assistant to the teacher, helping the teacher manage the learning of a
particular student.
Teachers in tutored classrooms report that their experiences are fulfilling
(e.g., Wertheimer, 1990). When students are learning with these tutors, the
teachers circulate around the class, providing instruction to the students for
whom the tutor’s explanations are not adequate. Teachers are the ultimate
domain experts, focusing on the difficult learning problems and leaving the sim-
ple learning problems to the computer.

Intelligent tutoring systems can be built around a production-

rule analysis of the components of the skill to be learned.

The Role of Mathematics in Life

Discussion has proceeded on the assumption that the mathematics that should
be taught in U.S. schools is more or less the mathematics that has been taught
in the schools of the United States and the rest of the world during the twenti-
eth century and that the only issue is how to achieve competence in this math-
ematics. However, there are controversies about what should be taught as well
as how it should be taught, and it is only honest to expose some of this discus-
sion here. Much of the controversy surrounds the purpose of mathematics edu-
cation. Three general purposes for mathematics education have been cited:
1. Students should learn to do mathematics because it makes them much
better thinkers generally. There is virtually no evidence for such general
transfer, and much of this book suggests that this is an unlikely possibility.
Transfer depends on two tasks having specific rules and facts in common.
2. Students should learn mathematics to appreciate the intellectual beauty of
that discipline. The public has extremely varied opinions on the worth of

411
 CHAPTER 11 Applications to Education

mathematics appreciation, and few students achieve deep mathematics

appreciation.
3. Students should learn mathematics to make them better citizens and bet-
ter workers. This is undoubtedly the major reason for public support of
mathematics education and the major reason for the crisis attitude about
the poor mathematical achievement of American students.

Unfortunately, people have a poor understanding of how mathematics is used

in everyday life. Employers are forever complaining about the mathematical
preparation of American workers, but when asked what they are looking for,
they report a need either for the basic computational skills that are taught in the
early grades or for skills so specific to the particular job that it would be unrea-
sonable to expect a general education to teach them (Secretary’s Commission
on Achieving Necessary Skills, 1991). Little mention is made of the academic
mathematics that occupies high school or college. This mathematics undoubt-
edly plays a major foundational role in many engineering applications, but com-
puter software can now perform many of the mathematical calculations that
engineers once had to perform by hand. At the high-technology end, employ-
ers are looking more for workers who can use this software intelligently and cre-
atively rather than for workers who deeply understand the underlying mathe-
matical foundations. Scientists and mathematicians still need to practice acade-
mic mathematics, and such people play a key role in society. However, it is
unclear whether all students should be taught academic mathematics to prepare
such a select minority of society.
Again, there are serious questions about the relationship between school-
taught mathematics and the mathematics used in everyday life, as illustrated by
the example of the Brazilian children who could perform mathematics in the
marketplace but not in school. As another example, Lave (1988) reported a
study of Orange County, California, shoppers making best-buy calculations.
Some of the problems were rather simple, such as, “What is the better buy, an
8-oz yogurt at 35 cents or a 6-0z yogurt at 43 cents?” Others were more difficult,
for example, “What is the better buy, a 20.5-oz can of refried beans at 57 cents
or a 17-0z can of refried beans at 49 cents?” The more difficult problems required
some form of fractional arithmetic. Lave found that these experienced shoppers
were able to make 98 percent of their choices correctly. In contrast, in a study of
their ability to solve standard mathematics problems, such as 2/5 x .75, they only
averaged 70 percent correct.
Lave found an interesting relationship among various measures of indi-
vidual differences. She found a strong correlation between the income of her
subjects, their performance on academic mathematics, and their schooling, but
none of these measures was related to how well they did at best-buy calcula-
tions. From this finding Lave drew a powerful conclusion: standard academic
mathematics is used to help define the class structure in our society (and hence
the correlation with income), but it is in fact as arbitrary as Latin and Greek were
in past generations (hence the lack of relationship to best-buy calculations).

412
 Final Reflections

Lave went far beyond the evidence, although there may be some truth to
her conclusions. Since all her participants were expert at best-buy calculations,
there was no room for a relationship to be found between that skill and any-
thing else. Not all people perform all real-world tasks at uniformly high levels
of excellence. As any employer can testify, different workers vary widely in the
quality of their job performance. Numerous studies show modest to large cor-
relations between school achievement and work performance (e.g., Boissiere,
Knight & Sabot, 1985; Hunter & Hunter, 1984), even after partialing out the
effects of general ability measures (which are sometimes larger).
Whatever the final verdict on Lave’s arguments, they highlight the difficult
issues that society faces about the role of mathematics in everyday life and the role
it should have in the classroom. The lack of answers and consensus is part of what
fuels the never-ending reforms of the mathematics curriculum. As noted earlier,
the consequence of the reform-crazy movement in the United States is that math-
ematics educators never settle on a curriculum long enough to teach it well.

There are serious questions about whether academic mathe-

matics helps in the performance of real-world tasks.

Final Reflections
The news media in the United States is full of proclamations on the failure of
the educational system. As discussed previously, these claims of failure are a bit
overstated. Part of the failure to reach levels of high achievement in the United
States reflects the lack of economic equity in the society. It is a serious mistake
to burden the school systems with trying to patch up a problem whose source
is elsewhere. Even with its diverse population, the United States does reason-
ably well in the instruction of reading but is failing dismally in international
comparisons of achievement in mathematics. The explanation for the disparity
in reading and mathematics achievement is largely in the amount of time spent
in mathematics instruction and in the pernicious belief in the United States that
mathematics achievement is a matter of talent rather than effort. Poor achieve-
ment is further exacerbated by the failure of the mathematics education com-
munity to stay committed to a specific curriculum.
Success in learning a subject, such as mathematics, is not just a function of
the amount of time spent in the classroom and doing homework. As this book
has shown in many ways, how the time is spent is critical. Some ways of repre-
senting the skill are more effective than others. In the case of reading instruction,
for example, although students can learn to read by both the phonics method
and the whole-word method, the phonics method appears to be more success-
ful, probably because it teaches a more efficient method for reading English.
Frequently, however, the problem is not that there are competing ways of
teaching a skill, but that there is no way of teaching the skill. With respect to

413
 CHAPTER 11 Applications to Education

many skills of reading comprehension and such topics in mathematics as alge-

braic word problems and geometric proof skills, the usual problem is that stu-
dents are not told how to perform the task and instead are left to find some
method for themselves.
Cognitive task analysis is the critical prerequisite to effective instruction. It
identifies precisely the skills to be taught, and it allows effective programs of
instruction to be pursued. In particular, it serves as the foundation for a com-
puter-based method of individualized instruction that can speed up the rate of
learning by as much as a factor of three.
This is the appropriate topic on which to end a book on learning and
memory. Education is the obvious application for the research reviewed in the
book, and education is sorely in need of a science to understand how educa-
tional manipulations map onto outcomes. In the future, as psychology generates
more and better analyses of educationally relevant outcomes and as education
becomes more knowledgeable about these analyses, education, like other
applied fields, can be expected to have a more scientific basis for its applications.
It would be a sign that both fields have finally matured if the psychology of
learning and cognition could play a cogent role in educational applications.

Instruction can be improved by a cognitive task analysis that

identifies the components to be learned.

Further Readings
Several texts describe the application of the psychology of learning and cognition
to instruction, including those by Farnham-Diggory (1992), Gagné, Yekovich, and
Yekovich (1993), and Mayer (1987). Stevenson and Stigler (1992) wrote a popu-
lar and psychologically informed comparison of educational achievement in
Japan and the United States. Gagné, Briggs, and Wager (1988) present a series of
papers on instructional technology. Just and Carpenter (1987) provide a thorough
analysis of the reading process and apply that analysis to educational issues.
Adams (1990) offers a good discussion of reading and the phonics method.
Anderson, Corbett, Koedinger, and Pelletier (1995) describe their intelligent
tutoring work and its application to mathematics education. Two other major
efforts to bring cognitive psychology to the classroom are described in Brown &
Campione (1996) and Cognition and Technology Group and Vanderbilt (1997).
Westbury (1992) and Baker (1993) engage in an informed debate about the sig-
nificance of the differences in mathematical achievement between Japan and the
United States. Bruer (1993) describes a number of cognitively inspired applica-
tions for education. Some of the best informed general discussions of education
are found in a book edited by Ravitch (1998).

414
 Glossary

acquisition: The process by which new arguments: The elements of a proposi-

memories are encoded into long-term tional representation that are organized
memory. by the relation; frequently, they are nouns.
ACT: J. R. Anderson’s theory of how artificial intelligence: A field of com-
declarative and procedural representa- puter science that tries to get computers
tions underlie human information pro- to behave intelligently.
cessing. Association Equation: The equation
action potential: The sudden change in stating that the strength of association
electrical potential that travels down the between a cue and a memory record
axon of a neuron. decreases with the number of records
activation: An abstract concept in cog- associated to that cue.
nitive psychology used to refer to the associative bias: The predisposition to
availability of information; sometimes associate certain stimuli to certain other
thought of in terms of neural excitation. stimuli or responses.
activation equation: The equation stat- associative stage: The second stage in
ing that the activation of a memory Fitts’s stages of skill acquisition which
record is the sum of the record strength involves developing production rules to
plus its strengths of association to the perform the skill.
cues in the memory probe. auditory sensory memory: A system
all-or-none learning: Learning that that holds about the last 4 seconds of
takes place in a single trial rather than auditory information.
gradually. autonomous stage: ‘The third stage in
amnesia: Loss of memory, frequently Fitts’s stages of skill acquisition in which
occurring as a result of brain injury. performance of the skill becomes auto-
anterograde amnesia: Inability to learn matic.
new information after a brain insult. autoshaping: The experimental phe-
aphasia: Loss of language function nomenon that animals spontaneously
resulting from brain injury. produce species-specific consummatory
Aplysia: Aseaslug with a simple nervous responses to stimuli that precede rein-
system that has been studied extensively. forcers.

415
 Glossary

avoidance: Behavior in a negative rein- chunk: A term coined by Miller (1956) to

forcement situation that prevents an refer to the units or memory records that
aversive stimulus from occurring. encode a small number of elements.
axoaxonic synapse: A synapse of an classical conditioning: The procedure
axon onto another axon. in which an organism comes to display a
axon: The portion of a neuron that car- conditioned response (CR) to a neutral
ries information from one region of the conditioned stimulus (CS) that has been
brain to another. paired with a biologically significant
unconditioned stimulus (US) _ that
evoked an unconditioned response (UR).
backwards learning curve: A curve in
which the probability of an error is plot- closed-loop performance: A sequence
ted backward from the trial of the last of actions in which execution of later
error. actions waits for feedback from the
results of earlier actions.
behavior systems analysis: An
approach to learning that emphasizes codes: See memory codes.
the natural, unlearned organization of cognitive map: A mental representation
behavior for a species. of the layout of objects and routes in
behavioral objectives: A set of goals for space that an organism can use to guide
instruction derived from a behavioral its locomotion.
task analysis. cognitive stage: The first stage in Fitts’s
behaviorism: An approach to psycholo- stages of skill acquisition, which involves
gy that emphasizes casting theories in working from a declarative representa-
terms of external behavior rather than tion of the skill.
discussing the internal mechanisms cognitivism: An approach to psycholo-
responsible for the behavior.
gy that involves abstract descriptions of
bliss point: The organism’s ideal distri- the information-processing mechanisms
bution of time spent on various events. responsible for behavior.
blocking: The tendency for one stimu- competitive learning: The proposal in
lus to overshadow another in condition- such theories as the Rescorla—-Wagner
ing. theory that a US can support only so
much associative strength and that the
causal inference: Induction of what CSs compete for their share of this max-
causes what in our environment. imum possible strength.
central executive: The part of componential analysis: The analysis of
Baddeley’s working memory that con- instructional material into its underlying
trols the various slave rehearsal systems. cognitive components.
Also used to refer to the process that computer simulation: A methodology
guides cognition. for deriving predictions from a complex
CER: An abbreviation of conditioned theory by simulating on a computer the
emotional response. processes assumed by the theory.
cerebellum: The subcortical structure concept acquisition: The learning of
involved in motor coordination. categories, such as“dog” and“ chair.”
cerebral cortex: The highest area of the conditioned emotional response: A
brain. In the human it is a sheet of neur- characteristic response pattern emitted
al tissue folded around various subcorti- by an organism in anticipation of an
cal areas. aversive stimulus, such as a shock.

416
 Glossary

conditioned inhibition: The expecta- cumulative response record: A record

tion conditioned to a CS that it is associ- of the total number of responses emitted
ated with the absence of the US. as a function of time.
conditioned reinforcer: See secondary
reinforcer. decay hypothesis: The theory of forget-
conditioned response: The response ting that asserts that memories weaken
that the organism learns to make in a with the passage of time.
classical or instrumental conditioning declarative knowledge: Knowledge of
paradigm. factual information.
conditioned stimulus: The stimulus declarative memories: See explicit
that signals the unconditioned stimulus memories.
in a classical conditioning paradigm. deductive inference: An inference that
conditioning curve: The function definitely follows from what is known
showing the increase in the conditioned about the world.
response as a function of the number of delayed match-to-sample task: A para-
conditioning trials. digm in which the organism is shown a
configural cues: Stimulus combinations correct response alternative and must
that become associated as single ele- remember that response over a delay
ments to stimuli and responses. period.
connectionism: The attempt to account delta rule: An application of the
for behavior by the computations of large Rescorla-Wagner theory to learning in
numbers of neural elements connected neural networks.
to one another. dendrites: The branching portion of a
context cue: An element in the experi- neuron that receives synapses from the
ment context that can serve as a cue fora axons of other neurons.
memory record. depth of processing theory: The theory
context-dependent memory: The phe- that memory for information is a func-
nomenon that memory performance is tion of the depth to which it is processed.
often better when the context of test desensitization: A technique for treat-
matches the context of study. ing phobias by having the patient learn
contiguity: The occurrence of two items to relax in situations that gradually
close together in time and space, which approximate the fear-evoking situation.
some theories claim is sufficient to con- devaluation paradigm: A paradigm in
dition an association. which a US or a reinforcer is devalued.
contingency: The occurrence of one difference reduction: Selecting prob-
item increasing the probability that lem-solving operators to reduce the dif-
another item will occur; some theories ference between the current state and
claim that this is sufficient to form an the goal state.
association. digit span: A memory-span test in
CR: An abbreviation of conditioned which subjects must reproduce a series
response. of digits.
CS: An abbreviation of conditioned discounting the future: Valuing a
stimulus. future gain or loss less than a current
cue: An element that is associated to a gain or loss.
memory record and that can help retrieve discrimination: Differential responding
it. to stimuli.

417
 Glossary

discrimination learning: The process excitatory postsynaptic potential: A

by which an organism learns which stim- measure of the decrease in the difference
uli are associated with the experimental in electrical potential between the out-
contingency. side and inside of the neuron; used as a
d-prime measure: Measure of the dis- measure in studies of long-term potenti-
tance between signal and distractor dis- ation.
tributions of evidence in_ signal excitatory synapses: Synapses where
detectability theory. the neurotransmitters decrease the
drive-reduction theory: The theory that potential difference across the mem-
reinforcement depends on the reduction brane of a neuron.
of biological drives. : exemplar theories: Theories that hold
drives: States of deprivation which sup- that subjects categorize a test stimulus
posedly energize behavior in an organism. according to past stimuli that are similar
dual-code theory: The theory of Paivio to the test stimulus.
that information is stored in long-term expected value: The sum of the values
memory in terms of verbal and visual of the consequences of an action weight-
representations. ed by the probabilities of these conse-
dyslexia: A condition whereby children quences.
or adults of normal or above-normal explicit memories: Memories of which
intelligence are substantially subnormal a person is consciously aware during
in their reading ability. retrieval.
exponential function: A mathematical
echoic memory: Neisser’s term for function of the form y = ab“, where the
auditory sensory memory. independent variable, x, is in the expo-
elimination by aspects: A theory pro- nent.
posed by Tversky according to which extinction: The procedure in a condi-
people make choices among alternatives tioning experiment where the uncondi-
by first focusing on the most important tioned stimulus or the reinforcer is no
aspects of the alternatives. longer presented.
encoding: The process of creating a extinction function: The reduction in
long-term memory record to store an the conditioned response as a function of
experience. the number of extinction trials.
encoding-specificity principle: The
idea that memory performance is better false alarm: The tendency for subjects to
when tested in the presence of the same say that they have studied an item that
cues that were present when the memo- they have not studied.
ry was formed. fan effect: Increase in time to retrieve a
EPSP: An abbreviation of excitatory memory from a cue as more memories
postsynaptic potential. are associated to the cue.
equilibrium theory: The theory that an FI: An abbreviation of fixed-interval
organism finds reinforcing anything that schedule.
moves it toward its bliss point and pun- fixed-interval schedule: An organism
ishing anything that moves it away from is given a reinforcement for the first
that point. response after a fixed interval.
escape: Behavior in a negative reinforce- fixed-ratio schedule: An organism is
ment situation that terminates an aver- given a reinforcement after a fixed num-
sive stimulus. ber of responses.

418
 Glossary

flashbulb memories: Memories for hypothalamus: The subcortical area

extremely significant and emotion-laden that regulates expression of basic drives
events; such memories often seem par- and is involved in motivation.
ticularly vivid and detailed. hypothesis testing: A deliberate
forgetting function: See retention func- approach to inductive learning in which
tion. particular hypotheses are consciously
FR: An abbreviation of fixed-ratio considered and tested against the data.
schedule. Fy = as,

free recall: A memory paradigm in which’ | iconic memory: Neisser’s term for visu-
items are presented one at a time and, al sensory memory.
then subjects can recall them in any order. © implicit-memories:—Memories that a
frontal cortex: The region at the front of person is not consciously aware of
the cerebral cortex that includes the retrieving.
motor cortex and the prefrontal cortex. induction: The process by which a sys-
frontal lobes: Frontal cortex. tem makes inferences about the structure
functional magnetic resonance imaging of the environment from its experience
({MRI): Measurement of metabolic with that environment.
activity by measuring the magnetic field inductive inference: An _ uncertain
produced by the iron in oxygenated blood. inference about the state of the world
based on experience with that world.
generalization: When a behavior is inductive learning: Learning by means
evoked by a stimulus other than the one of inductive inferences.
it was conditioned to. information-processing approach: “An
generalization gradient: Representa- approach in cognitive psychology that
tion of the tendency of various stimuli to theorizes about information in the
evoke a conditioned response. abstract and how it progresses through
generate-recognize theory: A theory of the cognitive system.
free recall that claims that subjects gen- inhibition: A response suppression
erate candidate items and then recognize caused by factors such as fatigue and
which ones they have studied. extinction.
generation effect: People tend to dis- inhibitory synapses: Synapses where
play better memory for material they the neurotransmitters decrease the
generate for themselves. potential difference across the mem-
global maximization: The theory that brane of a neuron.
an organism will choose a pattern of innerear: The part of Baddeley’s phono-
responding that will lead to optimal out- logical loop responsible for perceiving
come overall. inner speech.
goal: The desired state in solving a prob- inner voice: The part of Baddeley’s
lem. phonological loop responsible for gener-
ating inner speech.
habituation: Repeated presentation of instinctive drift: The tendency for ani-
the US can result in a reduced magnitude mals to revert to innate, species-specific
in the UR evoked by that US. response patterns in a learning experi-
hippocampus: The subcortical area that ment.
plays a critical role in the formation of instrumental conditioning: The proce-
permanent memories. dure in which a reinforcement is made

419
 Glossary

conditional on emitting a response in a long-term memory: A rather perma-

particular stimulus situation. nent memory system that stores most of
intelligent tutoring systems: Com- our knowledge about the world.
puter systems that combine cognitive long-term potentiation: A long-term
models with techniques from artificial increase in the magnitude of the response
intelligence to create instructional inter- of the neurons to stimulation that occurs
actions with students. when a brief, high-frequency electrical
interference: A negative relationship stimulation is administered to some areas
between the learning of two sets of of the brain, including the hippocampus.
material. LTP: An abbreviation of long-term
‘“
potentiation.
interference hypothesis: The theory of
forgetting that asserts that competing
memories block retrieval of the target mastery learning: An_ instructional
memory. strategy in which earlier material is
brought to mastery before instruction
begins on later material.
Korsakoff’s syndrome: Amnesia that
matching law: Given the choice between
occurs after a long history of alcoholism
two variable-interval schedules, organisms
coupled with nutritional deficits.
distribute their responses between the two
schedules in proportion to the rates of
language universals: Features that are reinforcement from the two schedules.
true of all natural languages. means-ends analysis: A method of
latent inhibition: Slowing of the rate of problem solving that sets subgoals as the
conditioning to the stimulus when a means to obtaining some larger goal.
stimulus is given preexposures before melioration theory: The theory that an
conditioning. organism will shift its behavior toward
latent learning: Learning that takes the alternative that is currently offering
place in the absence of any reinforcer and the highest rate of return.
is manifest only when a reinforcement is memory: The relatively permanent trace
introduced into the situation. of the experience that underlies learning.
law of effect: The claim that reinforce- memory codes: Distinctive ways of
ment is necessary for learning. encoding information in a memory
learned helplessness: When an aver- record. Memory codes include verbal,
sive stimulus, such as shock, is given spatial, and propositional.
independent of an organism’s behavior, memory-span test: A task in which sub-
the organism comes to behave as if it jects are presented with a series of items
believed that it has no control over the and must reproduce them, usually
environment. immediately.
learning: The process by which relative- mnemonic techniques: Techniques for
ly permanent changes occur in behay- enhancing memory performance.
ioral potential as a result of experience.
momentary maximizing: The theory
learning curve: A function showing that at any point in time an organism will
increase in learning as a function of choose the response alternative that is cur-
amount of practice. rently offering the highest rate of return.
list context: Representation of the list to mood congruency: The phenomenon
which items are associated in a list learn- that recall of material may be higher if
ing experiment. the subject’s mood at recall matches the

420
 Glossary

emotional tone of the material the sub- changes the frequency of a response type
ject is trying to recall. in the environment.
motor program: A __ prepackaged operator: An action that transforms one
sequence of actions that can be executed problem-solving state into another prob-
according to different parameters with- lem-solving state.
out central control. operator subgoaling: If an operator
cannot be applied to achieve a goal, the
natural categories: Categories of problem solver sets a subgoal to trans-
objects that are found in the real world, form the state so that the operator can be
such as“ dog” or“ tree.” applied.
negative acceleration: A property of opponent process: A mechanism that is
functions, such as learning curves or evoked when a stimulus evokes a strong
retention curves, whereby the rate of response in one direction; this mecha-
change becomes smaller and smaller. nism produces a compensatory response
in the opposite direction.
negative reinforcement: An _instru-
mental conditioning procedure in which optimal-foraging theory: The theory
an aversive stimulus is made contingent that organisms forage for food so as to
on omission of a response. maximize their net energy gain (food
intake minus energy spent foraging).
negative transfer: The phenomenon
that learning of earlier material impairs
the learning of later material. paired-associate learning: A memory
procedure in which the subject learns to
nerve impulse: Action potentials that
give a response when presented with a
move down axons.
stimulus.
neurons: The cells in the brain that are
parietal lobe: The region at the top of
most directly responsible for neural
the cerebral cortex that is involved in
information processing. higher level sensory functions.
neurotransmitters: Chemicals that partial-reinforcement extinction effect:
cross the synapse from the axon of one The phenomenon that animals show
neuron to alter the electrical potential of greater resistance to extinction when
the membrane of another neuron. trained under a partial reinforcement
schedule.
occipital lobe: The region at the back of partial-reinforcement schedule: A
the cerebral cortex that is devoted main- reinforcement schedule that reinforces
ly to vision. only some of the organism’s responses.
omission training: An instrumental peak shift: The phenomenon in discrim-
conditioning procedure in which a desir- ination learning that maximal response is
able stimulus is made contingent on gotten to stimuli shifted away from the
omission of a response. positive stimulus in a direction that is also
open-loop performance: A sequence of away from the negative stimulus.
actions performed without waiting for phonics method: A method of reading
feedback from the results of earlier instruction that emphasizes going from
actions before performing later actions. letter combinations to sound.
operant: A term Skinner used to describe phonological loop: The system pro-
an action, such as a lever press, that pro- posed by Baddeley for rehearsing verbal
duced some change in the environment. information by silently saying it over and
operant conditioning: Learning that over again.

421
 Glossary

positive reinforcement: An instrumen- proposition: A type of code in memory

tal conditioning procedure in which a in which the record abstractly represents
desirable stimulus is made contingent on the smallest meaningful unit of informa-
a response. tion. Kintsch proposed a propositional
positron emission tomography (PET): representation in which relations orga-
Measurement of metabolic activity on nized arguments.
different regions of the brain using a punishment: An instrumental condition-
radioactive tracer. ing procedure in which an aversive stimu-
power function: A mathematical func- lus is made contingent on a response.
tion of the form y = ax”, where the inde-
pendent variable, x, is raised to a power rate of firing: The rate at which nerve
to get the dependent variable, y. impulses are generated along axons.
power law of forgetting: The observa-
recency effect: The phenomenon that
tion that memory performance decreases
the last items in a list are better remem-
as a power function of the delay since
bered.
training.
recognition failure: Failure to recognize
power law of learning: The observation
items in one context when these items
that performance increases as a power
can be recalled in another context.
function of the amount of practice.
reconstructive memory: The phenome-
prefrontal cortex: The region at the front
non that people will try to inferentially
of the frontal cortex that is involved in
recreate their memories from what they
planning and other higher-level cognition.
can recall.
presynaptic facilitation: An enhance-
ment of a sypaptic connection by record: Abstract conception of the unit
increasing the neurotransmitter release in which memories are encoded.
from the axon. rehearsal: The process of repeating
primacy effect: The phenomenon that information to oneself to help remember
the early items in a list are better remem- the information.
bered. rehearsal systems: Systems for main-
priming: The process by which a prior taining transient sensory records of
exposure makes a memory more avail- information.
able or facilitates the perceptual process- reinforcer: A stimulus that changes the
ing of an item. probability of a response in an instru-
proactive interference: The phenome- mental conditioning paradigm.
non that the learning of earlier material relation: The element in a propositional
accelerates the forgetting of later material. representation that organizes the argu-
probability matching: The tendency to ments.
choose one alternative with a probability reminiscence: The occasional result that
that matches its probability of success. memories improve with time.
proceduralization: The process of con- repression: A forgetting mechanism
verting declarative knowledge about a proposed by Freud that actively represses
domain into domain-specific procedures. unpleasant memories.
procedural knowledge: Knowledge of Rescorla-Wagner theory: Theory that
how to perform various tasks. the rate of growth of the strength
production rules: Condition—action between a CS and a US is proportional to
pairs that are postulated to represent the difference between the sum of cur-
procedural knowledge. rent associative strengths and the maxi-

422
 Glossary

mum associative strength that the US secondary reinforcer: A neutral stimu-

permits. lus that has become associated with rein-
response-prevention paradigm: A par- forcement. Money is sometimes thought
adigm in classical conditioning where of as a secondary reinforcer.
the organism is prevented from emitting second-order conditioning paradigm:
the UR. A classical conditioning paradigm in
retention: The maintenance of memo- which an association is first learned from a
ries after their initial encoding. neutral CS1 to a US, and then an associa-
retention function: Function showing tion is learned from a neutral CS2 to CS1.
amount remembered as a function of sensitization: Presentation of the US
time. alone makes the animal more likely to
retrieval: The process of getting access respond to the presentation of a neutral
to memories. stimuli.
retrieval-cue hypothesis: The theory of sensory preconditioning paradigm: A
forgetting that asserts that people lose classical conditioning paradigm in which
access to memories because they lose an association is first learned from a neu-
access to the cues that can retrieve them. tral CS2 to a neutral CS1, and then an
retroactive interference: The phenom- association is learned from CS1 to a US.
enon that the learning of later material serial position curve: A function show-
causes the forgetting of earlier material. ing probability of recall of an item as a
retrograde amnesia: Inability to function of its position in the input order.
remember information that was learned short-term memory: A purported stor-
before a brain insult. age system in human memory capable of
holding a small amount of information
for a short period of time.
SAM: Shiffrin’s theory of memory,
which holds that memories are retrieved signal detectability theory: The theory
as a function of their strengths of associ- that subjects judge whether they have
ation to cues. seen an item according to how much evi-
dence there is for having seen an item. It
satisficing: A theory proposed by Simon
is assumed that this evidence continu-
that people make choices among alter-
ously varies and that subjects must set
natives by selecting the first item that
some threshold for recognition.
meets a certain threshold of acceptabili-
ty. skill: Procedural knowledge of how to
scalloped function: Functions, such as
perform a task.
some cumulative response records, Skinner box: An apparatus for studying
which display cycles of rapid increases instrumental conditioning containing a
followed by slow rates of increase. lever that the animal can press for rein-
schema theories: Theories according to forcement.
which subjects categorize a test stimulus SOP: Stands for “sometimes opponent
into the category that shares the most processes,” which is Wagner's theory of
features in common with the test stimu- the conditioning of opponent processes;
lus. also sometimes stands for “standard
schema theory: Schmidt's theory of operating procedure.”
motor learning, which holds that people spacing effect: Performance on a reten-
learn a representation of the desired out- tion test is usually best when the spacing
come of an action which they use to tune of the studies of the material matches the
the motor program. retention interval.

423
 Glossary

species-specific defense reactions: behaviors even when there is no contin-

Innate behaviors that are evoked in dif- gency between that behavior and rein-
ferent species when they are in danger. forcement. Skinner described the ani-
spontaneous recovery: ‘The recovery of mals as developing the” superstition” that
a conditioned response after a period of the behavior was instrumental in getting
time has intervened since an extinction the reinforcement.
procedure. synapse: The location at which the axon
S-R association: An association of one neuron almost makes contact with
between a stimulus and a response, a another neuron.
hypothetical CS—CR association in classi-
cal conditioning. task analysis: An attempt to identify the
S-S association: An association between components of a task that need to be
two stimuli, a hypothetical CS—US associ- taught.
ation in classical conditioning. temporal lobe: The region at the side of
_ S-shaped curves: Functions, such as the the cerebral cortex that has the primary
conditioning function, that begin with auditory areas and is involved in the
little change, then show rapid change, long-term memory.
and then have little changéagain. trace: The record that encodes a memo-
state-dependent memory: The phenom- ry experience.
enon that memory performance is often transfer-appropriate processing: The
better when the subject’s internal state at idea that memory performance is better
“test matches the internal state at study. when the subject processes the material
Sternberg paradigm: A procedure for at study in the same way as the material
studying retrieval from memory in which will be used at test.
subjects are given a small set of items transposition: The term Kohler used to
and then asked whether a probe, item . indicate that the organism had transferred
: iP
occurred in that set. the relationship between one pair of stim-
‘ stimulus-response bonds: Direct asso- uli to choosing between a different pair.
ciations between stimuli and responses
that many behaviorists believed underlie unconditioned response: The response
all behavior. that the unconditioned stimulus natural-
strength: An attribute of memory ly evokes.
records and their associations that deter- unconditioned stimulus: The biologi-
mines how active they can become. cally significant stimulus that follows a
strength equation: The equation stating conditioned stimulus in a classical condi-
that the strength of a memory record tioning paradigm.
varies as a product of a power function of UR: An abbreviation of unconditioned
the amount of practice. response.
subgoal: A goal pursued in service of a _US: An abbreviation of unconditioned
higher goal. stimulus.
subgoaling: The process in problem
solving by which one goal is created as a variable-interval schedule: An organ-
subgoal in service of another. ism is given a reinforcement for the first
subjective value: The value that an response after a variable interval. The
organism places on an alternative. intervals average to a certain specified
superstitious learning: The observation length.
that animals will spontaneously produce variable-ratio schedule: An organism

424
 Glossary

is given a reinforcement after a variable VR: An abbreviation for variable-ratio

number of responses. The number of schedule.
responses average to certain specified
value. whole-word method: A method of
VI: An abbreviation for variable-interval reading instruction that emphasizes
schedule. direct recognition of words and phrases.
visual sensory memory: A system that working memory: ‘The information that
holds the last 1 or 2 seconds of visual is currently available in memory for
information. working on a problem.
visuo-spatial sketch pad: Baddeley’s Yerkes—Dodson law: The proposal that
system for rehearsing visual or spatial performance is optimal at intermediate
information. levels of arousal.
\

425
 Bibliography

Adams, J. A. (1971). A closed-loop theory of motor learning. Journal of Motor Behavior, 3,

111-150.
Adams, M. J. (1990). Beginning to read: Thinking and learning about print. Cambridge,
MA: MIT Press.
Ainslie, G., & Herrnstein, R. J. (1981). Preference reversal and delayed reinforcement.
Animal Learning and Behavior, 9, 476-482.
Akins, C. K., Domjam, M. & Guitierez, G. (1994). Topography of sexually conditioned
behavior in male Japanese quail (Coturnix japonica) depends on the CS-US interval.
Journal of Experimental Psychology: Animal Behavior Processes, 20, 199-209.
Alkon, D. L. (1984). Calcium-mediated reduction of ionic currents: A biophysical memo-
ry trace. Science, 226, 1037-1045.
Allison, J. (1983). Behavioral economics. New York: Praeger.
Allison, J. (1989). The nature of reinforcement. In S. B. Klein & R. R. Mowrer (Eds.),
Contemporary learning theories: Instrumental conditioning and the impact of biological con-
straints on learning (pp. 13-39). Hillsdale, NJ: Erlbaum.
Allison, J., & Timberlake, W. (1974). Instrumental and contingent saccharin-licking in
rats: Response deprivation and reinforcement. Learning and Motivation, 5, 231-247.
Amiro, T. W., & Bitterman, M. E. (1980). Second-order appetitive conditioning in gold-
fish. Journal of Experimental Psychology: Animal Behavior Processes, 6, 41-48.
Amsel, A. (1967). Partial reinforcement effects on vigor and persistence. In K. W. Spence
& J. T. Spence (Eds.), The psychology of learning and motivation (Vol. 1). New York:
Academic Press.
Andersen, R. A. (1995). Coordinate transformations and motor planning in posterior
parietal cortex. In M. S. Gazzaniga (Ed.), The cognitive neurosciences. Cambridge, MA: MIT
Press.
Anderson, J. A. (1973). A theory for the recognition of items from short memorized lists.
Psychological Review, 80, 417-438.
Anderson, J. R. (1972). FRAN: A simulation model of free recall. In G. H. Bower (Ed.), The
psychology of learning and motivation (Vol. 5). New York: Academic Press.

426
 Bibliography

Anderson, J. R. (1974a). Retrieval of propositional information from long-term memory.

Cognitive Psychology, 6, 451-474.
Anderson, J. R. (1974b). Verbatim and propositional representation of sentences in imme-
diate and long-term memory. Journal of Verbal Learning and Verbal Behavior, 13, 149-162.
Anderson, J. R. (1976). Language, memory, and thought. Hillsdale, NJ: Erlbaum.
Anderson, J. R. (1981). Interference: The relationship between response latency and
response accuracy. Journal of Experimental Psychology: Human Learning and Memory, 7,
311-325.
Anderson, J. R. (1982). Acquisition of cognitive skill. Psychological Review, 89, 369-406.
Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard University
Press.
Anderson, J. R. (1983b). A spreading activation theory of memory. Journal of Verbal
Learning and Verbal Behavior, 22, 261-295.
Anderson, J. R. (1991). Is human cognition adaptive? Behavioral and Brain Sciences, 14,
471-484.
Anderson, J. R. (1992). Intelligent tutoring and high school mathematics. Proceedings of
the Second International Conference on Intelligent Tutoring Systems. Montreal.
Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Erlbaum.
Anderson, J. R. (2000). Cognitive psychology and its implications (5th ed.). New York: Worth.
Anderson, J. R., & Bower, G. H. (1972a). Configural properties in sentence memory.
Journal of Verbal Learning and Verbal Behavior, 11, 594-605.
Anderson, J. R., & Bower, G. H. (1972b). Recognition and retrieval processes in free recall.
Psychological Review, 79, 97-123.
Anderson, J. R., & Bower, G. H. (1973). Human associative memory. Washington, DC:
Winston.
Anderson, J. R., & Bower, G. H. (1974). A propositional theory of recognition memory.
Memory and Cognition, 2, 406-412.
Anderson, J. R., & Lebiere, C. (1998). Atomic components of thought. Mahwah, NJ: Erlbaum.
Anderson, J. R., & Paulson, R. (1977). Representation and retention of verbatim informa-
tion. Journal of Verbal Learning and Verbal Behavior, 16, 439-451.
Anderson, J. R., & Reder, L. M. (1979). An elaborative processing explanation of depth of
processing. In L. S. Cermak and F. I. M. Craik (Eds.), Levels of processing in human memo-
ry. Hillsdale, NJ: Erlbaum.
Anderson, J. R., & Schooler, L. J. (1991). Reflections of the environment in memory.
Psychological Science, 2, 396-408.
Anderson, J. R., & Sheu, C. F. (1994). Causal inferences based on contingency informa-
tion.
Anderson, J. R., Bellezza, F. S., & Boyle, C. F. (1993). The geometry tutor and skill acqui-
sition. In J. R. Anderson (Ed.), Rules of the mind. Hillsdale, NJ: Erlbaum.
Anderson, J. R., Corbett, A. T., Koedinger, K., & Pelletier, R. (1995) Cognitive tutors:
Lessons learned. The Journal of Learning Sciences, 4, 167-207.
Anderson, J. R., Kushmerick, N., & Lebiere, C. (1993). The Tower of Hanoi and goal struc-
tures. In J. R. Anderson (Ed.), Rules of the mind. Hillsdale, NJ: Erlbaum.

427
 Bibliography

Arkes, H. R., Hackett, C., & Boehm, L. (1989). The generality of the relation between
familiarity and judged validity. Journal of Behavioral Decision Making, 2, 81-94.
Atkinson, R. C., & Juola, J. F. (1973). Factors influencing speed and accuracy of word
recognition. In S. Kornblum (Ed.), Attention and performance (Vol. IV, pp. 583-612). New
York: Academic Press. :
Atkinson, R. C., & Juola, J. F. (1974). Search and decision processes in recognition
memory. In D. H. Krantz, R. C. Atkinson, R. D. Luce, & P. Suppes (Eds.), Contemporary
developments in mathematical psychology (Vol. 1, pp. 242-293). San Francisco: W. H.
Freeman.
Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its
control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and
motivation (Vol. 2). New York: Academic Press.
Atkinson, R. C., Bower, G. H., & Crothers, E. J. (1965). Introduction to mathematical learn-
ing theory. New York: Wiley.
Atkinson, R. L., Atkinson, R. C., & Hilgard, E. R. (1983). Introduction to psychology (8th
ed.). San Diego: Harcourt Brace Jovanovich.
Ayres, T. J., Jonides, J., Reitman, J. S., Egan, J. C., & Howard, D. A. (1979). Differing suffix
effects for the same physical suffix. Journal of Experimental Psychology: Human Learning
and Memory, 5, 315-321.
Azrin, N. H., & Holz, W. C. (1966). Punishment. In W. K. Honig (Ed.), Operant behavior:
Areas of research and application. New York: Appleton-Century-Crofts.
Azrin, N. H., Holz, W. C., & Hake, D. F. (1963). Fixed-ratio punishment. Journal of the
Experimental Analysis of Behavior, 6, 141-148.
Baddeley, A. D. (1986). Working memory. Oxford: Oxford University Press.
Baddeley, A. D. (1997). Human memory: Theory and practice. Boston: Allyn & Bacon.
Baddeley, A. D., & Lewis, V. J. (1981). Inner active processes in reading: The inner voice,
the inner ear, and the inner eye. In A. M. Lesgold and C. A. Perfetti (Eds.), Interactive
processes in reading (pp. 107-129). Hillsdale, NJ: Erlbaum.
Baddeley, A. D., Grant, S., Wight, E., & Thomson, N. (1975). Imagery and visual working
memory. In P. M. A. Rabbit & S. Dornic (Eds.), Attention and performance, (Vol. V, pp.
205-217). London: Academic Press.
Baddeley, A. D., Lewis,V.J., & Vallar, G. (1984). Exploring the articulatory loop. Quarterly
Journal of Experimental Psychology, 36, 233-252.
Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of
short-term memory. Journal of Verbal Learning and Verbal Behavior, 14, 575-589.
Bahrick, H. P. (1979). Maintenance of knowledge: Questions about memory we forget to
ask. Journal of Experimental Psychology: General, 108, 296-308.
Bahrick. H. P. (1984). Semantic memory content in permastore: Fifty years of memory for
Spanish learned in school. Journal of Experimental Psychology: General, 113, 1-24.
Bailey, C. H., & Chen, M. (1983). Morphological basis of long-term habituation and sen-
sitization in Aplysia. Science, 220, 91-93.
Baker, D. P. (1993). Compared to Japan, the U.S. is a low achiever...Really: New evidence
and comment on Westbury. Educational Researcher, 22, 18-21.
Balota, D., & Lorch, R. (1986). Depth of automatic spreading activation: Mediated prim-

428
 Bibliography

ing effects in pronunciation but not in lexical decision. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 12, 336-345.
Balsam, P. D. (1988). Selection, representation and equivalence of controlling stimuli. In
R. C. Atkinson, R. J. Herrnstein, G. Lindzey, and R. D. Luce (Eds.), Stevens’ handbook of
experimental psychology, 2nd ed., (Vol. 2, pp. 111-166). New York: Wiley.
Banich, M. T. (1997). Neuropsychology: The neural bases of mental function. Boston, MA:
Houghton Mifflin Company.
Bannerman, D. M., Good, M. A., Butcher, S. P., Ramsay, M., & Morris, R. G. M. (1995).
Distinct components of spatial learning revealed by prior training and NMDA receptor
blockade. Nature, 378, 182-186.
Barbizet, J. (1970). Human memory and its pathology. San Francisco, CA: Freeman.
Barnes, C. A. (1979). Memory deficits associated with senescence: A neurophysiological
and behavioral study in the rat. Journal of Comparative Physiology, 43, 74-104.
Bartlett, B. J. (1978). Top-level structure as an organizational strategy for recall of class-
room text. Unpublished doctoral dissertation, Arizona State University.
Bartlett, F. C. (1932). Remembering. Cambridge: Cambridge University Press.
Baum, W. M. (1969). Extinction of avoidance response following response prevention:
Some parametric investigations. Canadian Journal of Psychology, 23, 1-10.
Beck, B. B. (1980). Animal tool behavior: The use and manufacture of tools by animals. New
York: Garland STPM Press.
Beck, I. L. (1981). Reading problems and instructional practices. In T. S. Waller & G. E.
MacKinnon (Eds.), Reading research: Advances in theory and practice (Vol. 2). New York:
Academic Press.
Bedford, J., & Anger, D. (1968). Flight as an avoidance response in pigeons. Paper presented
to the Psychonomic Society, St. Louis.
Beecroft, R. S. (1966). Classical conditioning. Goleta, CA: Psychonomic Press.
Begg, I., Snider, A., Foley, F, & Goddard, R. (1989). The generation effect is no artifact:
Generation makes words distinctive. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 15, 977-989.
Bekerian, D. A., & Baddeley, A. D. (1980). Saturation advertising and the repetition effect.
Journal of Verbal Learning and Verbal Behavior, 19, 17-25.
Bernstein, I. L., & Borson, S. (1986). Learned food aversion: A component of anorexia
syndromes. Psychological Review, 93, 462-472.
Bernstein, I. L., Webster, M. M., & Bernstein, I. D. (1982). Food aversions in children
receiving chemotherapy for cancer. Cancer, 50, 2961-2963.
Best, M. R., Dunn, D. P, Batson, J. D, Meachum, C. L., & Nash, S. M. (1985).
Extinguishing conditioned inhibition in flavour-aversion learning: Effects of repeated
testing and extinction of the excitatory element. Quarterly Journal of Experimental
Psychology, 37B, 359-378.
Bilodeau, E. A., & Bilodeau, I. M. (1958). Variable frequency knowledge of results and the
learning of simple skill. Journal of Experimental Psychology, 55, 379-383.
Blackburn, J. M. (1936). Acquisition of skill: An analysis of learning curves (HRB Report No.
Tay
Blaney, P. H. (1986). Affect and memory: A review. Psychological Bulletin, 99, 229-246.

429
 Bibliography

Blessing, S. B. (1996). The use of prior knowledge in learning from examples. Doctoral
dissertation, Carnegie Mellon University, Pittsburgh, PA.
Bliss, T.V.P,, & Lomo, T. (1973). Long-lasting potentiation of synaptic transmission in the
dentate area of the anesthetized rabbit following stimulation of the preforant path.
Journal of Physiology, 232, 331-356.
Bliss, T.V.P., & Lynch, M. A. (1988). Long-term potentiation of synaptic transmission in
the hippocampus: Properties and mechanisms. In P. W. Landfield and S. A. Deadwyler
(Eds.), Neurology and neurobiology: Vol.35. Long-term potentiation: From biophysics to behav-
ior. New York: Alan R. Liss.
Bloom, B. S. (1968). Learning for mastery. Evaluation Comment, 1, 2.
Bloom, B. S. (1976). Human characteristics and school learning. New York: McGraw-Hill.
Bloom, B. S. (Ed.). (1985a). Developing talent in young people. New York: Ballantine Books.
Bloom, B. S. (1985b). Generalizations about talent development. In B. S. Bloom (Ed.),
Developing talent in young people (pp. 507-549). New York: Ballantine Books.
Bloom, L. C. & Mudd, S. A. (1991). Depth of processing approach to face recognition: A
test of two theories. Journal of Experimental Psychology: Learning, Memory and Cognition,
17, 556-565.
Blough, D. S. (1959). Delayed matching in the pigeon. Journal of the Experimental Analysis
of Behavior, 2, 151-160.
Blough, D. S. (1975). Steady state data and a quantitative model of operant generalization
and discrimination. Journal of Experimental Psychology: Animal Behavior Processes, 1, 3-21.
Bobrow, D. G., & Bower, G. H. (1969). Comprehension and recall of sentences. Journal of
Experimental Psychology, 80, 455-461.
Boissiere, M., Knight, J. B., & Sabot, R. H. (1985). Earnings, schooling, ability, and cogni-
tive skills. American Economic Review, 75, 1016-1030.
Bolles, R. C. (1970). Species-specific defense reactions and avoidance learning.
Psychological Review, 77, 32-48.
Boring, E. G. (1950). A history of experimental psychology. New York: Appleton-Century.
Boring, E. G. (1953). A history of introspection. Psychological Bulletin, 50, 169-189.
Bourne, L. E., Ekstrand, B. R., & Dominowski, R. L. (1971). The psychology of thinking.
Englewood Cliffs, NJ: Prentice-Hall.
Bourne, L. E., Jr. (1974).An inference model of conceptual rule learning. In R. Solso (Ed.),
Theories in cognitive psychology. Washington, DC: Erlbaum.
Bovair, S., Kieras, D. E., & Polson, P. G. (1990). The acquisition and performance of text-
editing skill: A cognitive complexity analysis. Human Computer Interaction, 5, 1-48.
Bower, G. H. (1972). Mental imagery and associative learning. In L. Gregg (Ed.), Cognition
in learning and memory. New York: Wiley.
Bower, G. H., & Clark, M. C. (1969). Narrative stories as mediators for serial learning.
Psychonomic Science, 14, 181-182.
Bower, G. H., & Hilgard, E. R. (1981). Theories of learning (5th ed.). Englewood Cliffs, NJ:
Prentice-Hall.
Bower, G. H., & Reitman, J. S. (1972). Mnemonic elaboration in multilist learning. Journal
of Learning and Verbal Behavior, 11, 478-485.

430
 Bibliography

Bower, G. H., & Springston, F. (1970). Pauses as recoding points in letter series. Journal of
Experimental Psychology, 83, 421-430.
Bower, G. H., & Trabasso, T. R. (1963). Reversals prior to solution in concept identifica-
tion. Journal of Experimental Psychology, 66, 409-418.
Bower, G. H., Karlin, M. B., & Dueck, A. (1975). Comprehension and memory for pic-
tures. Memory and Cognition, 3, 216-220.
Boyd, W., & King, E. J. (1975). The history of Western education (11th ed.). Totowa, NJ:
Barnes & Noble Books.
Bradshaw, G. L., & Anderson, J. R. (1982). Elaborative encoding as an explanation of lev-
els of processing. Journal of Verbal Learning and Verbal Behavior, 21, 165-174.
Braine, M.D.S. (1971). On two types of models of the internalization of grammars. In D.
I. Slobin (Ed.), The ontogenesis of grammar: A theoretical symposium. New York: Academic
Press.
Bransford, J. D., & Franks, J. J. (1971). The abstraction of linguistic ideas. Cognitive
Psychology, 2, 331-380.
Bransford, J. D., & Johnson, M. K. (1972). Contextual prerequisites for understanding:
Some investigations of comprehension and recall. Journal of Verbal Learning and Verbal
Behavior, 11, 717-726.
Bransford, J. D., Franks, J. J., Morris, C. D., & Stein, B. S. (1979). Some general constraints
on learning and memory research. In L. S. Cermak & FILM. Craik (Eds.), Levels of pro-
cessing in human memory (pp. 331-354). Hillsdale, NJ: Erlbaum.
Breland, K., & Breland, M. (1951). A field of applied animal psychology. American
Psychologist, 6, 202-204.
Breland, K., & Breland, M. (1961). The misbehavior of organisms. American Psychologist,
16, 681-684.
Brigham, J. C. (1981, November). The accuracy of eyewitness evidence: How do attorneys
see it? The Florida Bar Journal, pp. 714-721.
Broadbent, D. E. (1957). A mechanical model for human attention and immediate mem-
ory. Psychological Review, 64, 205-215.
Brodgen, W. J. (1949). Acquisition and extinction of a conditioned avoidance response in
dogs. Journal of Comparative Physiological Psychology, 42, 296-302.
Brooks, L. R. (1967). The suppression of visualization by reading. Quarterly Journal of
Experimental Psychology, 19, 289-299.
Brown, A. L., & Campione, J. C. (1996). Psychology theory and the design of innovative
learning environments: On procedures, principles, and systems. In L. Schauble & R.
Glaser (Eds.) Innovations in learning: New environments for education (pp. 289-325).
Mahwah, NJ: Erlbaum.
Brown, J. (1958). Some tests of decay theory of immediate memory. Quarterly Journal of
Experimental Psychology, 10, 12-21.
Brown, J. S., & Van Lehn, K. (1980). Repair theory: A generative theory of bugs in proce-
dural skills. Cognitive Science, 4, 397-426.
Brown, P. L., & Jenkins, H. M. (1968). Auto-shaping of the pigeon’s key-peck. Journal of
the Experimental Analysis of Behavior, 11, 1-8.
Brown, R. (1973). A first language. Cambridge, MA: Harvard University Press.

431
 Bibliography

Brown, R., & Fraser, C. (1963). The acquisition of syntax. In C. N. Cofer and B. Musgrave
(Eds.), Verbal behavior and learning: Problems and processes (pp. 158-201). New York:
McGraw-Hill.
Brown, R., & Kulik, J. (1977). Flashbulb memories. Cognition, 5, 73-99.
Brown, R., & McNeill, D. (1966). The “tip of the tongue” phenomenon. Journal of Verbal
Learning and Verbal Behavior, 5, 325-337.
Bruer, J. T. (1993). Schools for thought: A science of learning the classroom. Cambridge, MA:
MIT Press.
Bruner, J. S., Goodnow, J. I, & Austin, G. A. (1956). A study of thinking. New York: Wiley.
Bruner, J. S., Matter, J., & Papanek, M. L. (1955). Breadth of learning as a function of drive
level and mechanization. Psychological Review, 62, 1-10.
Bullock, M., Gelman, R., & Baillargeon, R. (1982). The development of causal reasoning.
In W. Friedman (Ed.), The developmental psychology of time (pp. 209-254). New York:
Academic Press.
Burns, D. J. (1990). The generation effect: A test between single- and multifactor theories.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 1060-1067.
Burns, D. J. (1992). The consequences of generation. Journal of Memory and Language, 31,
615-633.
Burton, R. R. (1982). Diagnosing bugs in a simple procedural skill. In D. Sleeman & J. S.
Brown (Eds.), Intelligent tutoring systems. New York: Academic Press.
Butler, R. A. (1953). Discrimination learning by rhesus monkeys to visual-exploration
motivation. Journal of Comparative and Physiological Psychology, 46, 95-98.
California Assessment Program. (1980). Student achievement in California schools: 1979-80
annual report. Sacramento: California State Department of Education.
Camp, D. S., Raymond, G. A., & Church, R. M. (1967). Temporal relationship between
response and punishment. Journal of Experimental Psychology, 74, 114-123.
Capaldi, E. J. (1967). A sequential hypothesis of instrumental learning. In K. W. Spence &
J.T. Spence (Eds.), The psychology of learning and motivation (Vol. 1). New York: Academic
Press.
Carew, T. J., Hawkins, R. D., & Kandel, E. R. (1983). Differential classical conditioning of
a defensive withdrawal reflex in Aplysia californica. Science, 219, 397-400.
Carey, S. (1978). The child as word learner. In M. Halle, J. Bresnan, & G. Miller (Eds.),
Linguistic theory and psychological reality., Cambridge, MA: MIT Press.
Carraher, T. N., Carraher, D. W., & Schliemann, A. D. (1985). Mathematics in the streets
and in the schools. British Journal of Developmental Psychology, 3, 21-29.
Castles, A., & Coltheart, M. (1993). Varieties of developmental dyslexia. Cognition, 47,
149-180.
Catalano, J. F,, & Kleiner, B. M.(1984). Distant transfer and practice variability. Perceptual
and Motor Skills, 58, 851-856.
Catania, A. C. (1992). Learning (3rd ed.). Englewood Cliffs, NJ: Prentice-Hall.
Cavanagh, J. P. (1972). Relation between the immediate memory span and the memory
search rate. Psychological Review, 79, 525-530.
Ceci, S. J., Loftus, E. F., Leichtman, M. D., & Bruck, M. (1994). The possible role of source

432
 Bibliography

misattributions in the creation of false beliefs among preschoolers. International Journal of

Clinical and Experimental Hypnosis, 42, 304-320.
Cermak, L. S., & Craik, FIM. (Eds.). (1978). Levels of processing in human,memory.
Hillsdale, NJ: Erlbaum.
Chall, J. S. (1967). Learning to read: The great debate. New York: McGraw-Hill.
Champagne, A. B., Gunstone, R. F., & Klopfer, L. E. (1985). Effecting changes in cognitive
structures among physics students. In H. T. West & A. L. Pines (Eds.), Cognitive structure
and conceptual change. Orlando, FL: Academic Press.
Chapman, G. B., & Robbins, S. J. (1990). Cue interaction in human contingency judg-
ment. Memory and Cognition, 18, 537-545.
Charness, N. (1979). Components of skill in bridge. Canadian Journal of Psychology, 33,
1-16.
Chase, W. G., & Simon, H. A. (1973). The mind’s eye in chess. In W. G. Chase (Ed.), Visual
information processing. New York: Academic Press.
Chi, M.T.H., Bassock, M., Lewis, M., Reimann, P., & Glaser, R. (1989). Self-explanations:
How students study and use examples in learning to solve problems. Cognitive Science,
13, 145-182.
Chomsky, C. (1970). The acquisition of syntax in children from 5 to 10. Cambridge, MA: MIT
Press.
Chomsky, N. (1959). Review of Skinner's Verbal Behavior. Language, 35, 26-58.
Chomsky, N. (1965). Aspects of the theory of syntax. Cambridge, MA: MIT Press.
Chomsky, N. (1975). Reflections on language. New York: Random House.
Chomsky, N. (1986). Knowledge of language. New York: Fontana.
Christen, F., & Bjork, R. A. (1976). On updating the loci in the method of loci. Paper pre-
sented at the seventeenth annual meeting of the Psychonomic Society, St. Louis.
Christianson, S-A. (1992). Emotional stress and eyewitness memory: A critical review.
Psychological Bulletin, 112, 284-309.
Church, R. M. (1969). Response suppression. In B. A.Campbell & R. M. Church (Eds.),
Punishment and aversive behavior. New York: Appleton-Century- Crofts.
Clark, E.V. (1983). Meanings and concepts. In P. H. Mussen (Ed.), Handbook of child psy-
chology. New York: Wiley.
Cohen, N. J., Eichenbaum, H., Deacedo, B. S., & Corkin, S. (1985). Different memory sys-
tems underlying acquisition of procedural and declarative knowledge. In D. S. Olton, E.
Gamzu, & S. Corkin (Eds.), Memory dysfunctions: An integration of animal and human
research from preclinical and clinical perspectives. Annals New York Academy of Sciences,
444, 54-71.
Coltheart, M. (1983). Iconic memory. Philosophical Transactions of the Royal Society,
London B, 302, 283-294.
Colwill, R. M., & Delamater, B. A. (1995). An associative analysis of instrumental bicon-
ditional discrimination learning. Animal Learning & Behavior, 23, 218-233.
Colwill, R. M., & Rescorla, R. A. (1985a). Instrumental responding remains sensitive to
reinforcer devaluation after extensive training. Journal of Experimental Psychology: Animal
Behavior Processes, 11, 520-536.
Colwill, R. M., & Rescorla, R. A. (1985b). Postconditioning devaluation of a reinforcer

433
 Bibliography

affects instrumental responding. Journal of Experimental Psychology: Animal Behavior

Processes, 11, 120-132.
Colwill, R. M., & Rescorla, R. A. (1986). Associative structure in instrumental learning. In G.
H. Bower (Ed.), The psychology of learning and motivation (Vol. 20). New York: Academic Press.
Colwill, R. M., & Rescorla, R. A. (1988). Associations between the discriminative stimu-
lus and the reinforcer in instrumental learning. Journal of Experimental Psychology: Animal
Behavior Processes, 14, 155-164.
Compton, B. J., & Logan, G. D. (1991). The transition from algorithm to retrieval in mem-
ory-based theories of automaticity. Memory & Cognition, 19, 151-158.
Conrad, R. (1960). Very brief delay of immediate recall. Quarterly Journal of Experimental
Psychology, 12, 45-47.
Conrad, R. (1964). Acoustic confusion in immediate memory. British Journal of Psychology,
55, 75-84.
Corbett, A. T., & Anderson, J. R. (1990). The effect of feedback control on learning to pro-
gram with the LISP tutor. Proceedings of the 12th Annual Conference of the Cognitive Science
Society (pp. 796-803).
Coslett, H. B. (1997). Acquired dyslexia. In T. E. Feinberg, and M. J. Farah (Eds.) Behavioral
neurology and neuropsychology, 197-208. New York: McGraw-Hill.
Cowles, J. T. (1937). Food-tokens as incentive for learning by chimpanzees. Comparative
Psychology Monographs, 14 (No. 5).
Craik, F.IL.M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory
research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684.
Craik, F.LM., & Tulving, E. (1975). Depth of processing and the retention of words in
episodic memory. Journal of Experimental Psychology: General, 104, 268-294.
Craik, F.I.M., & Watkins, M. J. (1973). The role of rehearsal in short-term memory. Journal
of Verbal Learning and Verbal Behavior, 12, 599-607.
Crocker, J. (1981). Judgment of covariation by social perceivers. Psychological Bulletin, 90,
272-292.
Crossman, E. R. F. W. (1959). A theory of the acquisition of speed skill. Ergonomics, 2,
153-166.
Crowder, R. G. (1976). Principles of learning and memory. Hillsdale, NJ: Erlbaum.
Crowder, R. G. (1982). The demise of short-term memory. Acta Pychologica, 50, 291-323.
Crowder, R. G. (1989). Modularity and dissociations in memory systems. In H. L.
Roediger, & F. I. M. Craik (Eds.), Varieties of memory and consciousness: Essays in honour of
Endel Tulving. Hillsdale, NJ: Erlbaum.
Crowder, R. G., & Morton, J. (1969). Precategorical acoustic storage (PAS). Perception and
Psychophysics, 5, 365-373.
Culler, E., & Girden, E. (1951). The learning curve in relation to other psychometric func-
tions. American Journal of Psychology, 64, 327-349.
Dallett, K. M. (1964). Number of categories and category information in free recall.
Journal of Experimental Psychology, 68, 1-12.
Dansereau, D. F. (1978). The development of a learning strategies curriculum. In H. F.
O'Neill, Jr. (Ed.), Learning strategies. New York: Academic Press.

434
 Bibliography

Dansereau, D. F., Collins, K. W., McDonald, B. A., Holley, C. D., Garland, J. C., Diekhoff,
G., & Evans, S. H. (1979). Development and evaluation of an effective learning strategy
program. Journal of Educational Psychology, 71, 64-73. :
Darwin, C. (1975). The origin of the species (P. Appleman, Ed.). New York: W. W. Norton
(original work published 1859).
Darwin, C. J., Turvey, M.T., & Crowder, R. G. (1972). The auditory analogue of the Sperling
partial report procedure: Evidence for brief auditory storage. Cognitive Psychology, 3, 255-267.
Daugherty, K. G., MacDonald, M. C., Petersen, A. S., & Seidenberg, M. S. (1993). Why no
mere mortal has ever flown out to center field but people often say they do. Proceedings
of the 15th Annual Conference of the Cognitive Science Society (pp. 383-388).
Davenport, D. G., & Olson, R. D. (1968). A reinterpretation of extinction in discriminat-
ed avoidance. Psychonomic Science, 13, 5-6.
de Beer, G. R. (1959). Paedomorphesis. Proceedings of the 15th International Congress of
Zoology (pp. 927-930).
de Groot, A. D. (1965). Thought and choice in chess. The Hague, Netherlands: Mouton.
DeKeyser, R. M. (1997). Beyond explicit rule learning: Automatizing second language
morphosyntax. SSLA, 19 pp. 195-221. (Studies in Second Language Acquisition).
Dempster, F. N. (1992). The rise and fall of the inhibitory mechanism: Toward a unified
theory of cognitive development and aging. Developmental Review, 12, 45-75.
Dennis, M. (1997). Acquired disorders of language in children. In T. E. Feinberg, and M.
J. Farah (Eds.), Behavioral neurology and neuropsychology. New York: McGraw-Hill.
Dewhurst, D. J. (1967). Neuromuscular control system. IEEE Transactions on Biomedical
Engineering, 14, 167-171.
Diamond, A. (1989). Developmental progression in human infants and infant monkeys,
and the neural bases of inhibitory control of reaching. In A. Diamond (Ed.), The develop-
ment and neural bases of higher cognitive functions. New York: Academy of Science Press.
Diamond, A. (1991). Frontal lobe involvement in cognitive changes during the first year
of life. In K. R. Gibson & A. C. Peterson (eds.), Brain maturation and cognitive development:
Comparative and cross-cultual perspectives (pp. 127-180). New York: Aldine de Gruyter.
Domjan, M. (1993). The principles of learning and behavior (3rd ed.). Pacific Grove, CA:
Brooks/Cole.
Dooling, D. J., & Christiansen, R. E. (1977). Episodic and semantic aspects of memory for
prose. Journal of Experimental Psychology: Human Learning and Memory, 3, 428-436.
Dweck, C. (1975). The role of expectations and attributions in the alleviation of learned
helplessness. Journal of Personality and Social Psychology, 31, 674-685.
Eagle, M., & Leiter, E. (1964). Recall and recognition in intentional and incidental learn-
ing. Journal of Experimental Psychology, 68, 58-63.
Easterbrook, J. A. (1959). The effect of emotion on cue utilization and the organization of
behavior. Psychological Review, 66, 183-201.
Ebbinghaus, H. (1913). Memory: A contribution to experimental psychology (H. A. Ruger &
C. E. Bussenues, Trans.). New York: Teachers College, Columbia University. (Original
work published 1885)
Egan, D. E., & Schwartz, B. J. (1979). Chunking in recall of symbolic drawings. Memory
and Cognition, 7, 149-158.

435
 Bibliography

Ehri, L. C., & Wilce, L. C. (1983). Development of word identification speed in skilled and
less skilled beginning readers. Journal of Educational Psychology, 75, 3-18.
Ehri, L. C., & Wilce, L. C. (1987, winter). Does learning to spell help beginners learn to
read words? Reading Research Quarterly, 22.
Eich, E. (1985). Context, memory, and integrated item/context imagery. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 11, 764-770.
Eich, E., & Metcalfe, J. (1989). Mood dependent memory for internal versus external
events. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 443-455.
Eich, J. E., Weingartner, H., Stillman, R. C., & Gillin, J. C. (1975). State-dependent acces-
sibility of retrieval cues in the retention of a categorized list. Journal ofVerbal Learning and
Verbal Behavior, 14, 408-417.
Eichenbaum, H. (1997). Declarative memory: Insights from cognitive neurobiology.
Annual Review of Psychology, 48, 547-572.
Eichenbaum, H. & Bunsey, M. (1995). On the binding of associations in memory: clues
from studies on the role of the hippocampal region in paired-associate learning. Current
Directions in Psychological Science, 4, 19-23.
Eichenbaum, H., Stewart, C. & Morris, R.G.M. (1990). Hippocampal representation in
spatial learning. Journal of Neuroscience, 10, 331-339.
Einhorn, H. J., & Hogarth, R. M. (1986). Judging probable cause. Psychological Bulletin, 99,
mls
Eisenberger, R., Heerdt, W. A., Hamdi, M., Zimet, S. & Bruckmeir, G. (1979). Transfer of
persistence across behaviors. Journal of Experimental Psychology: Human Learning and
Memory, 5, 522-530.
Ekstrand, B. R. (1972). To sleep, perchance to dream. In C. P. Duncan, L. Sechrest, & A. W.
Melton (Eds.), Human memory: Festschrift in honor of Benton J. Underwood (pp. 59-82). New
York: Appleton-Century-Crofts.
Ellis, N. C.,, & Hennelly, R. A. (1980). A bilingual word-length effect: Implications for
intelligence testing and the relative ease of mental calculation in Welsh and English.
British Journal of Psychology, 71, 43-52.
Engle, R. W., & Bukstel, L. (1978). Memory processes among bridge players of differing
expertise. American Journal of Psychology, 91, 673-689.
Epstein, S. (1967). Toward a unified theory of anxiety. In B. A. Maher (Ed.), Progress in
experimental personality research (Vol. 4). New York: Academic Press.
Erickson, M. A., & Kruschke, J. K. (1998). Rules and exemplars in category learning.
Journal of Experimental Psychology: General, 127, 107-140.
Ericsson, K. A., & Lehmann, A. C. (1994). Marks of genius? Re-interpretation of excep-
tional feats by great musicians. In The Proceedings of the 35'* Annual Meeting of the
Psychonomics Society.
Ericsson, K. A., Krampe, R.T., & Tesch-Romer, C. (1993). The role of deliberate practice in
the acquisition of expert performance. Psychological Review, 100, 363-406.
Eron, L. D., Walder, L. O., Toigo, R., & Lefkowitz, M. M. (1963). Social class, parental pun-
ishment for aggression, and child aggression. Child Development, 34, 849-867.
Ervin-Tripp, S. M. (1974). Is second language learning like the first? TESOL Quarterly, 8,
111127;

436
 Bibliography

Estes, W. K. (1955). Statistical theory of spontaneous recovery and regression.

Psychological Review, 62, 145-154.
Estes, W. K., Campbell, J. A., Hatsopoulos, N., & Hurwitz, J. B. (1989). Base-rate effects in
category learning: A comparison of parallel network and memory storage-retrieval mod-
els. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 556-571.
Etscorn, F., & Stephens, R. (1973). Establishment of conditioned taste aversions with a
24-hour CS-US interval. Physiological Psychology, 1, 251-253.
Eysenck, M. W. (1982). Attention and arousal: Cognition and performance. Berlin: Springer-
Verlag.
Fantino, E., & Abarca, N. (1985). Choice, optimal foraging, and the delay-reduction
hypothesis. Behavioral and Brain Sciences, 8, 315-330.
Farnham-Diggory, S. (1992). Cognitive processes in education (2nd ed.). New York: Harper
Collins.
Fernandez, A., & Glenberg, A. M. (1985). Changing environmental context does not reli-
ably affect memory. Memory and Cognition, 13, 333-345.
Ferster, C. S., & Skinner, B. F. (1957). Schedules of reinforcement. New York: Appleton-
Century-Crofts.
Fitts, P. M. (1964). Perceptual-motor skill learning. In A. W. Melton (Ed.), Categories of
human learning. New York: Academic Press.
Fitzgerald, R. D., Martin, G. K., & O’Brien, J. H. (1973). Influence of vagal activity on classi-
cally conditioned heart rate in rats. Journal of Comparative Physiological Psychology, 83, 485-491.
Flaherty, C. F. (1985). Animal learning and cognition. New York: McGraw-Hill.
Folkard, S. (1983). Diurnal variation. In G.R.J. Hockey (Ed.), Stress and fatigue in human
performance (Chap. 9, pp. 245-272). New York: Wiley.
Folkard, S., Monk, T. H., Bradbury, R., & Rosenthal, J. (1977). Time of day effects in school-
children’s immediate and delayed recall of meaningful material. British Journal of
Psychology, 68, 45-50.
Foree, D. D., & LoLordo, V. M. (1973). Attention in the pigeon: The differential effects of
food-getting vs. shock avoidance procedures. Journal of Comparative and Physiological
Psychology, 85, 551-558.
Frase, L. T. (1975). Prose processing. In G. H. Bower (Ed.), The psychology of learning and
motivation (Vol. 9). New York: Academic Press.
Frederiksen, J. R. (1981). Sources of process interactions in reading. In A. M. Lesgold &
C. A. Perfetti (Eds.), Interactive processes in reading. Hillsdale, NJ:
Freedman, J. L., & Landauer, T. K. (1966). Retrieval of long-term memory: “Tip-of-the-
tongue” phenomenon. Psychonomic Science, 4, 309-310.
Freud, S. (1971). The psychopathology of everyday life. (A. Tyson, Trans.). New York: W. W.
Norton. (Original work published in 1901)
Friedman, M. P., Burke, C. J., Cole, M., Keller, L., Millward, R. B., Estes, W. K. (1964). Two-
choice behavior under extended training with shifting probabilities of reinforcement. In
R. C. Atkinson (Ed.), Studies in mathematical psychology. Stanford, CA.
Funahashi, S., Bruce, C. J., & Goldman-Rakic, P. S. (1991). Neural activity related to sac-
cadic eye movements in the monkey's dorsolateral prefrontal cortex. Journal of
Neurophysiology, 65, 1464-1483.

437
 Bibliography

Gagné, R. (1962). The acquisition of knowledge. Psychological Review, 69, 355-365.

Gagné, R., Briggs, L. J., & Wager, W. W. (1989). Principles of instructional design. New York:
Holt, Rinehart, & Winston.
Gaené, E., Yekovich, C. W., & Yekovich, F. R. ee The cognitive psychology of school learn-
ing. New York: HarperCollins.
Gallistel, C. R. (1990). The organization of learning. Cambridge, MA: MIT Press.
Gamoran, A. (1987). The stratification of high school learning opportunities. Sociology of
Education, 60, 135-155.
Garcia, J., & Koelling, R. A. (1966). Relation of cue to consequence in avoidance learning.
Psychonomic Science, 4, 123— 124,
Gardiner, J. M., & Java, R. I. (1993). Recognizing and remembering. In A. Collins, M. A.
Conway, S. E., Gathercole, & P. E. Morris (Eds.). Theories of memory (pp. 163-188).
Hillsdale, NJ: Erlbaum.
Gardner, M. (1998). The new New Math. The New York Review of Books, September 24,
Qa,
Gardner, R. A., & Gardner, B.T. (1969). Teaching sign language to a chimpanzee. Science,
165, 664-672.
Gathercole, S. E., & Hitch, G. J. (1993). Developmental changes in short-term memory:
A revised working memory perspective. In A. F. Collins, S. E. Gathercole, M. A. Conway,
& P. Morris (Eds.), Theories of memory. Hillsdale, NJ: Erlbaum.
Gazzaniga, M. S., Ivry, R. B. & Mangun, G. R. (1998). Cognitive neuroscience: The biology of
the mind. New York: W. W. Norton.
Gernsbacher, M. A. (1985). Surface information loss in comprehension. Cognitive
Psychology, 17, 324-363.
Gibbon, J. & Balsam, P. (1981). Spreading association in time. In C. M. Locurto, H. S.
Terrace, & J. Gibbon (Eds.), Autoshaping and conditioning theory (pp. 219-253). New York:
Academic Press.
Gillund, G., & Shiffrin, R. M. (1984). A retrieval model for both recognition and recall.
Psychological Review, 91, 1-67.
Glaser, R. (1972). Individuals and learning: The new aptitudes. Educational Researcher, 1,
53:
Glass, A. L. (1984). Effect of memory set on reaction time. In J. R. Anderson & S. M.
Kosslyn (Eds.), Tutorials in learning and memory (pp. 119-136). New York: W. H. Freeman.
Glenberg, A. M. (1976). Monotonic and nonmonotonic lag effects in paired-associate and
recognition memory paradigms. Journal of Verbal Learning and Verbal Behavior, 15, 1-16.
Glenberg, A. M., Smith, S. M., & Green, C. (1977). Type I rehearsal: Maintenance and
more. Journal of Verbal Learning and Verbal Behavior, 16, 339-352.
Gluck, M. A. (1997). The Rescorla-Wagner model isn’t dead, just brain damaged. In
Proceedings of the 38th Annual Meeting of the Psychonomics Society. Symposium I: The
Rescorla-Wagner Model: 25 Years Later, 181.
Gluck, M. A., & Bower, G. H. (1988). From conditioning to category learning: An adap-
tive network model. Journal of Experimental Psychology: General, 8, 37-50.
Gluck, M. A., & Myers, C. E. (1993). Hippocampal mediation of stimulus representation:
A computational theory. Hippocampus, 3, 491-516.

438
 Bibliography

Gluck, M. A., & Myers, C. E. (1995). Representation and association in memory: a neu-
rocomputational view of hippocampal function. Current Directions in Psychological Science,
4, 23-29.
Gluck, M. A., & Myers, C. E. (1997). Psychobiological models of hippocampal function in
learning and memory. Annual Review of Psychology, 48, 481-514.
Gluck, M. A., & Thompson, R. F. (1987). Modeling the neural substrates of associative
learning and memory: A computational approach. Psychological Review, 94, 176-191.
Glucksberg, S., & Cowan, G. N,, Jr. (1970). Memory for nonattended auditory material.
Cognitive Psychology, 1, 149-156.
Godden, D. R., & Baddeley, A. D. (1975). Context-dependent memory in two natural
environments: On land and under water. British Journal of Psychology, 66, 325-331.
Goldin-Meadow, S. Butcher, C., Mylander, C., & Dodge, M. (1994). Nouns and verbs in
a self-styled gesture system: What's in a name? Cognitive Psychology, 27, 259-319.
Goldman-Rakic, P. S. (1987). Circuitry of primate prefrontal cortex and regulation of
behavior by representational memory. In Handbook of physiology. The nervous system.
Higher functions of the brain (Vol. 5, pp. 373-417). Bethesda, MD: American Physiology
Society.
Goldman-Rakic, P. S. (1988). Topography of cognition: Parallel distributed networks in
primate association cortex. Annual Review of Neuroscience, 11, 137-156.
Goldman-Rakic, P. S. (1992). Working memory and mind. Scientific American, 267,
111-117.
Goldstein, A. G., & Chance, J. E. (1970). Visual recognition memory for complex config-
urations. Perception and Psychophysics, 9, 237-241.
Good, H. G. (1962). A history of American education. New York: Macmillan.
Goodwin, D. W., Powell, B., Bremer, D., Hoine, H., & Stern, J. (1969). Alcohol and recall:
State-dependent effects in man. Science, 163, 1358-1360.
Gordon, W. C. (1989). Learning and memory. Pacific Grove, CA: Brooks/Cole.
Gormezano, I. (1965). Yoked comparisons of classical and instrumental conditioning of
the eyelid response; and an addendum on“Voluntary Responders.”InW. F. Prokasy (Ed.),
Classical conditioning (pp. 48-70). New York: Appleton-Century-
Crofts.
Gormezano, I., Kehoe, E. J., & Marshall, B. S. (1983). Twenty years of classical condition-
ing research with the rabbit. In J. M. Prague & A. N. Epstein (Eds.), Progress in psychobi-
ology and physiological psychology (Vol. 10). New York: Academic Press.
Gormezano, I., Prokasy, W. F., & Thompson, R. F. (Eds.). (1987). Classical conditioning (3rd
ed.). Hillsdale, NJ: Erlbaum.
Graf, P., Squire, L. R., & Mandler, G. (1984). The information that amnesic patients do not
forget. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 164-178.
Graf, V., Bullock, D. H., & Bitterman, M. E. (1964). Further experiments on probability
matching in the pigeon. Journal of the Experimental Analysis of Behavior, 7, 151-157.
Grant, D. A. (1939a). The influence of attitude on the conditioned eyelid response. Journal
of Experimental Psychology, 25, 333-346.
Grant, D. A. (1939b). A study of patterning in the conditioned eyelid response. Journal of
Experimental Psychology, 25, 455-461.
Grant, D. A. (1973). Cognitive factors in eyelid conditioning. Psychophysiology, 10, 75-81.

439
 Bibliography

Grant, D. A., Hunter, H. G. & Patel, A. S. (1958). Spontaneous recovery of the conditioned
eyelid response. Journal of General Psychology, 59, 135-141.
Grant, D. A., & Norris, E. B. (1947). Eyelid conditioning as influenced by the presence of
sensitized beta-responses. Journal of Experimental Psychology, 37, 423-434.
Grant, D. S. (1976). Effect of sample presentation -time on long delay matching in the
pigeon. Learning and Motivation, 7, 580-590.
Grant, D. S. (1981). Short-term memory in the pigeon. In N. E. Spear & R. Miller (Eds.),
Information processing in animals: Memory mechanisms (pp. 227-256). Hillsdale, NJ:
Erlbaum.
Grant, D. S., & Roberts, W.’A. (1976). Sources of retroactive inhibition in pigeon short-
term memory. Journal of Experimental Psychology: Animal Behavior Processes, 2, 1-16.
Greeno, J. G. (1974). Hobbits and orcs: Acquisition of a sequential concept. Cognitive
Psychology, 6, 270-292.
Grittner, F. M. (1975). Individualized instruction: An historical perspective. The Modern
Language Journal, 323-333.
Groves, P. M. & Thompson, R. F. (1970). Habituation: A dual-process theory. Psychological
Review, 77, 419-450.
Gruneberg, M. M., & Monks, J. (1974). Feeling of knowing and cued recall. Acta
Psychologica, 38, 257-265.
Gunn, D. L. (1937). The humidity reactions of the wood louse, Porcellio scaber (Latreille).
Journal of Experimental Biology, 14, 178-186.
Guskey, T. R., & Gates, S. (1986). Synthesis of research on the effects of mastery learning
in elementary and secondary classrooms. Educational Leadership, 43, 73-80.
Guthrie, E. R. (1952). The psychology of learning (rev. ed.). New York: Harper & Row.
Guttman, N., & Kalish, H. I. (1956). Discriminability and stimulus generalization. Journal
of Experimental Psychology, 51, 79-88.
Gynther, M. D. (1957). Differential eyelid conditioning as a function of stimulus similar-
ity and strength of response to the CS. Journal of Experimental Psychology, 53, 408-416.
Haber, R. N. (1983). The impending demise of the icon: A critique of the concept of icon-
ic storage in visual information processing. Behavioral and Brain Sciences, 6, 1-11.
Haig, K. A., Rawlins, J.N.P., Olton, D. S., Mead, A., & Taylor, B. (1983). Food searching
strategies of rats: Variables affecting the relative strength of stay and shift strategies.
Journal of Experimental Psychology: Animal Behavior Processes, 9, 337-348.
Hammond, L. J. (1980). The effect contingency upon the appetitive conditioning of free
operant behavior. Journal of the Experimental Analysis of Behavior, 34, 297-304.
Harley, W. F., Jr. (1965). The effect of monetary incentive in paired-associate learning
using a differential method. Psychonomic Science, 2, 377-378.
Hart, J.T. (1967). Memory and the memory-monitoring process. Journal ofVerbal Learning
and Verbal Behavior, 6, 685-691.
Hasher, L., Goldstein, D., & Toppino,T. (1977). Frequency and the conference of referen-
tial validity. Journal of Verbal Learning and Verbal Behavior, 16, 107-112.
Haverty, L. (1999). Mathematical problem solving requires the basics: Skill with sary
number knowledge improves inductive reasoning performance. Doctoral dissertation.
Carnegie Mellon University, Pittsburgh, PA.

440
 Bibliography

Hawkins, R. D., & Bower, G. H. (Eds.). (1989). The Psychology of Learning and Motivation:
Vol. 23. Computational models of learning in simple neural systems. San Diego: Academic
Press.
Hawkins, R. D., Abrams, T. W., Carew, T.J., & Kandel, E. R. (1983). A cellular mechanism
of classical conditioning in Aplysia: Activity-dependent amplification of presynaptic
facilitation. Science, 219, 400-404.
Hayes, C. (1951). The ape in our house. New York: Harper.
Hayes, J. R. (1985). Three problems in teaching general skills. In J. Segal, S. Chipman, &
R. Glaser (Eds.), Thinking and learning (Vol. 2). Hillsdale, NJ: Erlbaum.
Hayes-Roth, F., Waterman, D. A., & Lenat, D. B. (1983). Building expert systems. Reading,
MA: Addison-Wesley.
Hearst, E. (1988). Fundamentals of learning and conditioning. In R. C. Atkinson, R. J.
Herrnstein, G. Lindzey, and R. D. Luce (Eds.), Stevens’ handbook of experimental psycholo-
gy: Vol. 2. Learning and cognition (pp. 3-110). New York: Wiley.
Heath, R. G. (1963). Electrical self-stimultion of the brain in man. American Journal of
Psychitry, 120, 571-577.
Heathcote, A. & Mewhort, D.J.K. (1995). The law of practice. Poster presented at the 36th
Annual Meeting of the Psychonomic Society, Los Angeles, CA.
Heathcote, A. & Mewhort, D.J.K. (1998). A survey of the law of practice. Poster presented
at the 39th Annual Meeting of the Psychonomic Society, Dallas, TX.
Hemmes, N. S., Eckerman, D. A., & Rubinsky, H. J. (1979). A functional analysis of col-
lateral behavior under differential-reinforcement-of-low-rate schedules. Journal of
Animal Learning and Behavior, 7, 328-332.
Henry, F. M. (1968). Specificity vs. generality in learning motor skill. In R. C. Brown & G.
S. Kenyon (Eds.), Classical studies on physical activity (pp. 331-340). Englewood Cliffs, NJ:
Prentice-Hall. (Original work published 1958)
Herrnstein, R. J. (1961). Relative and absolute strength of response as a function of fre-
quency of reinforcement. Journal of Experimental Analysis of Behavior, 4, 267-272.
Herrnstein, R. J. (1990). Rational choice theory. American Psychologist, 45, 356-367.
Herrnstein, R. J., & Vaughan, W. (1980). Melioration and behavioral allocation. In J.E.R.
Staddon (Ed.), Limits to action: The allocation of individual behavior. New York: Academic
Press,
Herrnstein, R. J., Loveland, D. H., & Cable, C. (1976). Natural concepts in pigeons. Journal
of Experimental Psychology: Animal Behavior Processes, 2, 285-302.
Heyman, G. M., & Luce, R. D. (1979). Operant matching is not a logical consequence of
maximizing reinforcement rate. Animal Learning & Behavior, 7, 133-140.
Hillman, B., Hunter, W. S., & Kimble, G. A. (1953). The effect of drive level on the maze
performance of the white rat. Journal of Comparative Physiological Psychology, 46, 87-89.
Hineline, P. N., & Rachlin, H. (1969). Escape and avoidance of shock by pigeons pecking
a key. Journal of the Experimental Analysis of Behavior, 12, 533-538.
Hinton, G. E. (1992). How neural networks learn from experience. Scientific American,
267, 145-151.
Hintzman, D. L. (1992). Mathematical constraints and the Tulving-Wiseman law.
Psychological Review, 99, 536-542.

441
 Bibliography

Hintzman, D. L., Block, R. A., & Summers, J. J. (1973). Modality tags and memory for repe-
titions: Locus of the spacing effect. Journal ofVerbal Learning and Verbal Behavior, 12, 229-238.
Hiroto, D. S., & Seligman, M.E.P. (1975). Generality of learned helplessness in man.
Journal of Personality and Social Psychology, 31, 311-327.
Hirshman, E., & Bjork, R. A. (1988). The generation effect: Support for a two-factor the-
ory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 484-494.
Hirshman, E., Whelley, M. M., & Palu, M. (1989). An investigation of paradoxical memo-
ry effects. Journal of Memory and Language, 28, 594-609.
Hirst, W., Phelps, E. A., Johnson, M. K., & Volpe, B. T. (1988). Amnesia and second lan-
guage learning. Brain and Cognition, 8, 105-116.
Ho, L., & Shea, J. B. (1978). Effects of relative frequency of knowledge of results on reten-
tion of a motor skill. Perceptual and Motor Skills, 46, 859-866.
Hockey, G. R. J., Davies, S., & Gray, M. M. (1972). Forgetting a function of sleep at differ-
ent times of day. Experimental Psychology, 24, 386-393.
Hoffer, A. (1981). Geometry is more than proof. Mathematics Teacher, 11-18.
Holland, J. H. Holyoak, K., Nisbett, R. E., & Thagard, P. R. (1986). Induction: Processes of
inference, learning, and discovery. Cambridge, MA: MIT Press.
Holland, P. C. (1985a). Element pretraining influences the content of appetitive serial
compound conditioning in rats. Journal of Experimental Psychology: Animal Learning and
Behavior, 14, 111-120.
Holland, P. C. (1985b). The nature of conditioned inhibition in serial and simultaneous
feature negative discriminations. In R. R. Miller & N. E. Spear (Eds.), Information process-
ing in animals: Conditioned inhibition (pp. 267-297). Hillsdale, NJ: Erlbaum.
Holland, P. C., & Rescorla, R. A. (1975). The effects of two ways of devaluing the uncon-
ditioned stimulus after first- and second-order appetitive conditioning. Journal of
Experimental Psychology: Animal Behavior Processes, 1, 355-363.
Holley, C. D., Dansereau, D. F., McDonald, B. A., Garland, J. C., & Collins, K. W. (1979).
Evaluation of hierarchical mapping technique as an aid to prose processing.
Contemporary Educational Psychology, 4, 227-237.
Holyoak, K. J., Koh, K., & Nisbett, R. E. (1989). A theory of conditioning: Inductive learn-
ing within rule-based default hierarchies. Psychological Review, 96, 315-340.
Honig, W. K. (1981). Working memory and the temporal map. In N. E. Spear & R. Miller
(Eds.), Information processing in animals: Memory mechanisms (pp. 167-197). Hillsdale, NJ:
Erlbaum.
Hoosain, R., & Salili, F. (1988). Language differences, working memory, and mathemati-
cal ability. In M. M. Gruneberg, P. E. Morris, & R. N. Sykes (Eds.), Practical aspects of mem-
ory: Current research and issues: Vol. 2. Clinical and educational implications (pp. 512-517).
Chichester, England: Wiley.
Houston, J. P. (1965). Short-term retention of verbal units with equated degrees of learn-
ing. Journal of Experimental Psychology, 70, 75-78.
Houston, J. P. (1991). Fundamentals of learning and memory (4th ed.). Orlando, FL:
Harcourt Brace Jovanovich.
Hull, C. L. (1920). Quantitative aspects of the evolution of concepts. Psychological
Monographs (Whole No. 123).

442
 Bibliography

Hull, C. L. (1952a). Autobiography. In C. A. Murchinson (Ed.), A history of psychology in

autobiography (Vol. 4). New York: Russell & Russell Press.
Hull, C. L. (1952b).Abehavior system: An introduction to behavior theory concerning the indi-
vidual organism. New Haven, CT: Yale University Press.
Hunter, J. E. & Hunter, R. F. (1984). Validity and utility of alternative predictors of job per-
formance. Psychological Bulletin, 96, 72-98.
Hyams, N. M. (1986). Language acquisition and the theory of parameters. Dordrecht,
Netherlands: D. Reidel.
Hyde,T.S., & Jenkins, J. J. (1973). Recall for words as a function of semantic, graphic, and
syntactic orienting tasks. Journal of Verbal Learning and Verbal Behavior, 12, 471-480.
Hyman, I. E., Husband, T. H., & Billings, F. J. (1995). False memories of childhood expe-
riences. Applied Cognitive Psychology, 9, 181-197.
Jackson, M. D., & McClelland, J. L. (1979). Processing determinants of reading speed.
Journal of Experimental Psychology: General, 108, 151-181.
Jacobsen, C. F. (1935). Functions of frontal association areas in primates. Archives of
Neurology and Psychiatry, 33, 558-560.
Jacobsen, C. F. (1936). Studies of cerebral functions in primates: I. The function of the
frontal association areas in monkeys. Comparative Psychology Monographs, 13, 1-60.
Jacoby, L. L. (1983). Remembering the data: Analyzing interactive processes in reading.
Journal of Verbal Learning and Verbal Behavior, 22, 485-508.
Jacoby, L. L. (1991). A process dissociation framework: Separating automatic from inten-
tional uses of memory. Journal of Memory and Language, 30, 513-541.
Jacoby, L. L., Toth, J. P, & Yonelinas, A. (1993). Separating conscious and unconscious
influences of memory: Measuring recollection. Journal of Experimental Psychology: General,
122, 139-154.
Jacoby, L. L., Woloshyn, V., & Kelley, C. (1989). Becoming famous without being recog-
nized: Unconscious influences of memory produced by dividing attention. Journal of
Experimental Psychology: General, 118, 115-125.
Jarvik, M. E., & Essman, W. B. (1960). A simple one-trial learning situation for mice.
Psychological Reports, 6, 290.
Jeffries, R. P., Polson, P. G., Razran, L., & Atwood, M. (1977). A process model for mis-
sionaries-cannibals and other river-crossing problems. Cognitive Psychology, 9, 412-440.
Jenkins, H. M., & Harrison, R. H. (1960). Effects of discrimination training on auditory
generalization. Journal of Experimental Psychology, 59, 246-253.
Jenkins, H. M., & Harrison, R. H. (1962). Generalization gradients of inhibition following
auditory discrimination learning. Journal of the Experimental Analysis of Behavior, 5,
435-441.
Jenkins, H. M., & Moore, B. R. (1973). The form of the auto-shaped response with food
or water reinforcers. Journal of the Experimental Analysis of Behavior, 5, 435-441.
Jenkins, H. M., & Ward, W. C. (1965). Judgment of contingency between responses and
outcomes. Psychological Monographs: General and Applied, 79 (1, Whole No. 594), 1-17.
Jenkins, I. H., Brooks, D. J., Nixon, P. D. Frackowiak, R. S. J., & Passingham, R. E. (1994).
Motor sequence learning: A study with positron emission tomography. Journal of
Neuroscience, 14, 3775-3790.

443
 Bibliography

Jenkins, J. G., & Dallenbach, K. M. (1924). Obliviscence during sleep and waking.
American Journal of Psychology, 35, 605-612.
Job, R.E.S. (1989). A test of proposed mechanisms underlying the interference effect pro-
duced by noncontingent food presentations. Learning and Motivation, 20, 153-177.
Johansson, R. S., & Westling, G. (1984). Roles of glabrous skin receptors and sensorimo-
tor memory in automatic control of precision grip when lifting rougher or more slippery
objects. Experimental Brain Research, 56, 560-564.
Johnson, D. D., & Baumann, J. F. (1984). Word identification. In P. D. Pearson, R. Barr, M.
Kamil, & P. Mosenthal (Eds.), Handbook of reading research. New York: Longman.
Johnson, D. M. (1972). A systematic introduction to the psychology of thinking. New York:
Harper & Row.
Johnson, E. E. (1952). The role of motivational strength in latent learning. Journal of
Comparative Physiological Psychology, 45, 526-530.
Johnson, N. F. (1970). The role of chunking and organization in process recall. In G. H.
Bower (Ed.), Psychology of language and motivation (Vol. 4). New York: Academic Press.
Jones, W. P,, & Anderson, J. R. (1987). Short- and long-term memory retrieval: A com-
parison of the effects of information load and relatedness. Journal of Experimental
Psychology: General, 116, 137-153.
Just, M. A., & Carpenter, P. A. (1987). The psychology of reading and language comprehen-
sion. Newton, MA: Allyn & Bacon.
Kaiser, M. K., McCloskey, M., & Proffitt, D. R. (1986). Development of intuitive theories
of motion: Curvilinear motion in the absence of external forces. Developmental Psychology,
ESM TIN
Kaiser, M. K., & Proffitt, D. R. (1984). The development of sensitivity to causally relevant
dynamic information. Child Development, 55, 1614-1624.
Kaiser, M. K., Proffitt, D. R., & McCloskey, M. (1985). The development of beliefs about
falling objects. Perception and Psychophysics, 38, 533-539.
Kamil, A. C. (1978). Systematic foraging by a nectar-feeding bird, the Amakihi (Loxops
virens). Journal of Comparative Physiological Psychology, 92, 388-396.
Kamil, A. C., Yoerg, S. I, & Clements, K. C. (1988). Rules to leave by: Patch departure in
foraging blue jays. Animal Behavior, 36, 843-853.
Kamin, L. J. (1968). “Attention-like” processes in classical conditioning. In M. R. Jones
(Ed.), Miami Symposium on the Prediction of Behavior: Aversive stimulation (pp. 9-31).
Miami, FL: University of Miami Press.
Kamin, L. J. (1969). Predictability, surprise, attention, and conditioning. In B. A. Campbell
& R. M Church (Eds.), Punishment and aversive behavior (pp. 279-296). New York:
Appleton-Century-
Crofts.
Kamin, L. J. (1956). The effects of termination of the CS and avoidance of the US on
avoidance learning. Journal of Comparative and Physiological Psychology, 49 420-421.
Kamin, L. J., Brimer, C. J., & Black, A. H. (1963). Conditioned suppression as a monitor of
fear of the CS in the course of avoidance-training. Journal of Comparative and Physiological
Psychology, 56, 497-501.
Kandel, E. R. (1976). Cellular basis of behavior: An introduction to behavioral neurobiology.
New York: Freeman.

444
 Bibliography

Kandel, E. R., & Hawkins, R. D. (1992). The biological basis of learning and individuality.
Scientific American, 267, 78-87.
Kandel, E. R., & Schwartz, J. H. (Eds.). (1985). Principles of neural science (2nd ed.). New
York: Elsevier.
Kandel, E. R., Schwartz, J. H., & Jessell, T. M. (Eds.). (1991). Principles of neural science.
New York: Elsevier.
Kaplan, C. A. (1989). Hatching a theory of incubation: Does putting a problem aside real-
ly help? If so, why? Unpublished doctoral dissertation, Carnegie Mellon University.
Pittsburgh, PA.
Katz, B. (1952). The nerve impulse. Scientific American, 187, 55-64.
Keele, S. W. (1987). Sequencing and timing in skilled perception and action: An overview.
In A. Allport, D. G. McKay, & W. Prinz (Eds.), Language perception and production:
Relationships between listening, speaking, reading, and writing (pp. 463-487). London:
Academic Press.
Keeton, W. T. (1980). Biological science. New York: W. W. Norton.
Kehoe, E. J., & Gormezano, I. (1980). Configuration and combination laws in condition-
ing with compound stimuli. Psychological Bulletin, 87, 351-387.
Keith, J. R., & Rudy, J. W. (1990). Why NMDA-receptor-dependent long-term potentia-
tion may not be a mechanism of learning and memory: Reappraisal of the NMDA-recep-
tor blockade strategy. Psychobiology, 18, 251-257.
Keller, F. S. (1968).“Good-bye teacher....” Journal of Applied Behavior Analysis, 1, 78-89.
Kellogg, W. N., & Kellogg, L. A. (1933). The ape and the child. New York: McGraw-Hill.
Kennelly, K. J., Dietz, D., & Benson, P. (1985). Reinforcement schedules, effort vs. ability
attributions, and persistence. Psychology in the Schools, 22, 459-464.
Keppel, G. (1964). Facilitation in short- and long-term retention of paired associates follow-
ing distributed practice in learning. Journal of Verbal Learning and Verbal Behavior, 3, 91-111.
Keppel, G., & Underwood, B. J. (1962). Proactive inhibition in short-term retention of
single items. Journal of Verbal Learning and Verbal Behavior, 1, 153-161.
Keppel, G., Postman, L., & Zavortnik, B. (1968). Studies of learning to learn: VIII. The
influence of massive amounts of training upon the learning and retention of paired-asso-
ciate lists. Journal of Verbal Learning and Verbal Behavior, 7, 790-796.
Kieras, D. E., & Bovair, S. (1984). The role of mental model in learning to operate a device.
Cognitive Science, 8, 255-273.
Kimble, G. A. (1961). Conditioning and learning (2nd ed.). New York: Appleton-Century-
Crofts.
King, G. R., & Logue, A. W. (1987). Choice in a self-control paradigm with human sub-
jects: Effects of changeover delay duration. Learning and Motivation, 18, 421-438.
Kintsch, W. (1970a). Learning, memory, and conceptual processes. New York: Wiley.
Kintsch, W. (1970b). Models for free recall and recognition. In D. A. Norman (Ed.), Models
of human memory. New York: Academic Press.
Kintsch, W. (1974). The representation of meaning in memory. Hillsdale, NJ: Erlbaum.
Kintsch, W. (1998). Comprehension: A paradigm for cognition. Cambridge, MA: Cambridge
University Press.

445
 Bibliography

Kintsch, W., & Buschke, H. (1969). Homophones and synonyms in short-term memory.
Journal of Experimental Psychology, 80, 403-407.
Kintsch, W., & van Dijk, T. A. (1978). Toward a model of text comprehension and repro-
duction. Psychological Review, 85, 363-394.
Klahr, D., & Dunbar, K. (1988). Dual space search.during scientific reasoning. Cognitive
Science, 12, 1-55. :
Klahr, D., Fay, A. L., & Dunbar, K. (1993). Heuristics for scientific experimentation: A
developmental study. Cognitive Psychology, 25, 111-146.
Klahr, D., Langley, P., & Neches, R. (Eds.). (1987). Production system models of learning and
development, Cambridge, MA: MIT Press.
Klein, S. B., & Mowrer, R. R. (Eds.). (1989). Contemporary learning theories: Instrumental con-
ditioning theory and the impact of biological constraints on learning. Hillsdale, NJ: Erlbaum.
Kleinsmith, L. J., & Kaplan, S. (1963). Paired associate learning as a function of arousal
and interpolated interval. Journal of Experimental Psychology, 65, 190-193.
Koedinger, K. R., Anderson, J. R., Hadley, W. H., & Mark, M. (1997). Intelligent tutoring
goes to school in the big city. International Journal of Artificial Intelligence in Education, 8,
30-43.
Koh, K., & Meyer, D. E. (1991). Function learning: Induction of continuous stimulus-
response relations. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17,
811-836.
KGhler, W. (1927). The mentality of apes. New York: Harcourt, Brace.
Kohler, W. (1955). Simple structural functions in the chimpanzee and in the chicken. In
W. D. Ellis (Ed.), A source book on Gestalt psychology. London: Routledge & Kegan Paul.
Konarski, E. A., Jr. (1979). The necessary and sufficient conditions for increasing instru-
mental responding in the classroom: Response deprivation vs. probability differential.
Unpublished doctoral dissertation, University of Notre Dame, Notre Dame, IN.
Konarski, E. A., Jr., Johnson, M. R., Crowell, C., & Whitman, T. L. (1980). Response depri-
vation, reinforcement, and instrumental academic performance in an EMR classroom. Paper
presented at the 13th annual Gatlinburg Conference on Research in Mental Retardation
and Developmental Disabilities, Gatlinburg, TN.
Krueger, W.C.F. (1929). The effects of overlearning on retention. Journal of Experimental
Psychology, 12, 71-78.
Kulik, C., Kulik, J., & Bangert-Downs, R. (1986). Effects of testing for mastery on student
learning. Paper presented at the annual meeting of the American Educational Research
Association, San Francisco.
Laird, J. D., Wagner, J. J., Halal, M., & Szegda, M. (1982). Remembering what you feel:
Effects of emotion on memory. Journal of Personality of Social Psychology, 42, 646-657.
Landauer, T. K. (1975). Memory without organization: Properties of a model with random
storage and undirected retrieval. Cognitive Psychology, 7, 495-531.
Landfield, P. W., & Deadwyler, S. A. (Eds.) (1988). Neurology and neurobiology: Vol. 35.
Long-term potentiation: From biophysics to behavior. New York: Alan R. Liss, Inc.
Lashley, K. S. (1924). Studies of cerebral function in learning: V. The retention of motor
habit after destruction of the so-called motor area in primates. Arch, neuro. Psychiatry, 12,
249-276.

446
 Bibliography

Lave, J. (1988). Cognition in practice: Mind, mathematics, and culture in everyday life. New
York: Cambridge University Press.
Lawrence, D. H., & DeRivera, J. (1954). Evidence for relational transposition. Journal of
Comparative and Physiological Psychology, 47, 465-471.
Leaf, R. C. (1964). Avoidance response evocation as a function of prior discriminative fear
conditioning under curare. Journal of Comparative Physiological Psychology, 58, 446-449.
Leahey, T. H. (1992). A history of psychology. Englewood Cliffs, NJ: Prentice-Hall.
Lenneberg, E. H. (1967). Biological foundations of language. New York: Wiley.
Lesgold, A. M., Resnick, L. B., & Hammond, K. (1985). Learning to read: A longitudinal
study of word skill development in two curricula. In G. Waller, & E. MacKinnon (Eds.),
Reading research: Advances in theory and practice (Vol. 4). New York: Academic Press.
Levine, M. (1975). A cognitive theory of learning. Hillsdale, NJ: Erlbaum.
Levinger, G., & Clark, J. (1961). Emotional factors in the forgetting of word associations.
Journal of Abnormal and Social Psychology, 62, 99-105.
Levonian, E. (1972). Retention over time in relation to arousal during learning: An expla-
nation of discrepant results. Acta Psychologica, 36, 290-321.
Lewis, C. H. (1978). Production system models of practice effects. Unpublished doctoral
dissertation, University of Michigan, Ann Arbor.
Lewis, C. H., & Anderson, J. R. (1976). Interference with real world knowledge. Cognitive
Psychology, 7, 311-335.
Lieberman, P. (1984). The biology and evolution of language. Cambridge, MA: Harvard
University Press.
Light, J. S., & Gantt, W. H. (1936). Essential part of reflex arc for establishment of condi-
tioned reflex. Formation of conditioned reflex after exclusion of motor peripheral end.
Journal of Comparative Psychology, 21, 19-36.
Loftus, E. F., & Burns, T. E. (1982). Mental shock can produce retrograde amnesia.
Memory and Cognition, 10, 318-323.
Loftus, E. E., & Pickrell, J. (1995). The formation of false memories. Psychiatric Annals, 25,
720-725.
Loftus, E. F., Loftus, G. R. & Messo, J. (1987). Some facts about“weapon focus.” Law and
Human Behavior, 11, 55-62.
Loftus, G. R. (1972). Eye fixations and recognition memory for pictures. Cognitive
Psychology, 3, 525-551.
Loftus, G. R., & Patterson, K. K. (1975). Components of short-term proactive interference.
Journal of Verbal Learning and Verbal Behavior, 14, 105-121.
Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review,
96, 492-527.
Logan, G. D., & Klapp, S.T. (1991). Automatizing alphabet arithmetic: I. Is extended prac-
tice necessary to produce automaticity? Journal of Experimental Psychology: Learning,
Memory, and Cognition, 17, 179-195.
Lovett, M. (1998). Choice. In J. R. Anderson & C. Lebiere (Eds.), Atomic components of
thought. Mahwah, NJ: Erlbaum.
Lundberg, I. (1985). Longitudinal studies of reading and reading difficulties in Sweden.

447
 Bibliography

In G. E. MacKinnon & T. G. Waller (Eds.), Reading research: Advances in theory and prac-
tice (Vol. 4, pp. 65-105). Orlando, FL: Academic Press.
Lundberg, I., Frost, J., & Petersen, O. P. (1988). Effects of an extensive program for stim-
ulating phonological awareness in preschool children. Reading Research Quarterly, 23,
263-284. ;
MacCorquodale, K. O. (1970). On Chomsky’s review of Skinner's Verbal Behavior. Journal
of the Experimental Analysis of Behavior, 13, 83-99.
Macfarlane, D. A. (1930). The role of kinesthesis in maze learning. California University
Publication Psychology, 4, 277-305.
MacKay, D. G. (1982). The problem of flexibility, fluency, and speed-accuracy trade-off in
skilled behavior. Psychological Review, 89, 483-506.
Mackintosh, N. J. (1974). The psychology of animal learning. New York: Academic Press.
Mackintosh, N. J. (1975). A theory of attention: Variations in the associability of stimuli
with reinforcement. Psychological Review, 82, 276-298.
Mackintosh, N. J., & Little, L. (1969). Intradimensional and extradimensional shift learn-
ing by pigeons. Psychonomic Science, 14, 5-6.
MacPhail, E. M. (1968). Avoidance responding in pigeons. Journal of the Experimental
Analysis of Behavior, 11, 625-632.
MacWhinney, B. (1993). The (il)logical problem of language acquisition. Proceedings of the
15th Annual Conference of the Cognitive Science Society (pp. 61-70). University of Boulder,
CO. Hillsdale, NJ: Erlbaum.
MacWhinney, B., & Leinbach, J. (1991). Implementations are not conceptualizations:
Revising the verb learning model. Cognition, 29, 121-157.
Maier, S. F., Jackson, R. L., & Tomie, A. (1987). Potentiation, overshadowing, and prior
exposure to inescapable shock. Journal of Experimental Psychology: Animal Behavior
Processes, 13, 260-270.
Maki, W. S. (1984). Some problems for a theory of working memory. In H. L. Roitblat, T.
G. Bever, & H. S.Terrace (Eds.), Animal cognition (pp. 117-133). Hillsdale, NJ: Erlbaum.
Maki, W. S., & Hegvik, D. K. (1980). Directed forgetting in pigeons. Animal Learning and
Behavior, 8, 567-574.
Mandler, G. (1967). Organization and memory. In K. W. Spence & J.T. Spence (Eds.), The
psychology of learning and motivation (Vol. 1, pp. 327-372). New York: Academic Press.
Mandler, J. M., & Ritchey, G. H. (1977). Long-term memory for pictures. Journal of
Experimental Psychology: Human Learning and Memory, 3, 386-396.
Manis, F. R., Seidenberg, M. S., Doi, L. M., McBride-Chang, C. & Peterson, A. (1996). On
the bases of two subtypes of developmental dyslexia. Cognition, 58, 157-196.
Marcus, G. F., Brinkman, U., Clahsen, H., Wiese, R., Woest, A. & Pinker, S. (1995). German
inflection: The exception that proves the rule. Cognitive Psychology, 29, 189-256.
Marler, P, & Peters, S. (1982). Long-term storage of learned birdsongs prior to produc-
tion. Animal Behaviour, 30, 479-482.
Marr, D. (1971). Simple memory: A theory for archicortex. Philosophical Transactions of the
Royal Society of London, Series B, 262, 23-81.
Massaro, D. W. (1989). Experimental psychology: An information processing approach. San
Diego: Harcourt Brace Jovanovich.

448
 Bibliography

Masson, M.EJ., & MacLeod, C. M. (1992). Reenacting the route to interpretation:

Enhanced identification without prior perception. Journal of Experimental Psychology:
General, 121, 145-176.
Mayer, R. E. (1987). Educational psychology: A cognitive approach. Boston: Little, Brown.
Mazur, J. E. (1990). Learning and behavior (2nd ed.). Englewood Cliffs, NJ: Prentice-Hall.
Mazur, J. E. (1994). Learning and behavior (3rd ed.). Englewood Cliffs, NJ: Prentice-Hall.
McAllister, W. R. (1953). Eyelid conditioning as a function of the CS-UCS interval. Journal
of Experimental Psychology, 45, 417-422.
McClelland, J. L., McNaughton, B. L., & O'Reilly, R. C. (1995). Why there are comple-
mentary learning systems in the hippocampus and neocortex: Insights from the success-
es and failures of connectionist models of learning and memory. Psychological Review,
102, 419-457.
McCloskey, M. E. (1983). Intuitive physics. Scientific American, 248, 122-130.
McCloskey, M. E., & Glucksberg, S. (1978). Natural categories. Well-defined or fuzzy
sets? Memory and Cognition, 6, 462-472.
McCloskey, M. E., Wible, C. G., & Cohen, N. J. (1988). Is there a special flashbulb mem-
ory mechanism? Journal of Experimental Psychology: General, 117, 171-181.
McConkie, G. W., & Rayner, K. (1974). Identifying the span of the effective stimulus in read-
ing. (Final Report OEG 2-71-0531). U. S. Office of Education.
McDaniel, M. A., & Einstein, G. O. (1986). Bizarre imagery as an effective memory aid:
The importance of distinctiveness. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 12, 54-65.
McDaniel, M. A., Waddill, P. J., & Einstein, G. O. (1988).A contextual account of the gen-
eration effect: A three factor theory. Journal of Memory and Language, 27, 521-536.
McGaugh, J. L., & Dawson, R. G. (1971). Modification of memory storage processes.
Behavioral Science, 16, 45-63.
McGeoch, J.A. (1932). Forgetting and the law of disuse. Psychological Review, 39, 352-370.
McGeoch, J. A. (1942). The psychology of human learning. New York: Longmans, Green.
McKeithen, K. B., Reitman, J. S., Rueter, H. H., & Hirtle, S. C. (1981). Knowledge organi-
zation and skill differences in computer programmers. Cognitive Psychology, 13, 307-325.
McKnight, C. C., Crosswhite, F. J., Dossey, J. A., Kifer, E., Swafford, J.O., Travers, K. J., &
Cooney, T. J. (1990). The underachieving curriculum: Assessing U. S. school mathematics from
an international perspective. Champaign, IL: Stipes Publishing Company.
McNamara, T. P., Hardy, J. K., & Hirtle, S. C. (1989). Subjective hierarchies in spatial
memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 211-227.
McNeill, D. (1966). Developmental psycholinguistics. In F. Smith & G. A. Miller (Eds.),
The genesis of language: A psycholinguistic approach. Cambridge, MA: MIT Press.
Medin, D. L., & Schaffer, M. M. (1978). Context theory of classification learning.
Psychological Review, 85, 207-238.
Melton, A. W. (1963). Implications of short-term memory for a general theory of memo-
ry. Journal of Verbal Learning and Verbal Behavior, 2, 1-21.
Meltzer, H. (1930). Individual differences in forgetting pleasant and unpleasant experi-
ences. Journal of Educational Psychology, 21, 399-409.

449
 Bibliography

Meyer, D. E., & Schvaneveldt, R. W. (1971). Facilitation in recognizing pairs of words:

Evidence of a dependence between retrieval operations. Journal of Experimental
Psychology, 90, 227-234.
Meyer, B.J.F., Brandt, D. M., & Bluth, G. J. (1978). Use of author's textual schema: Key for
ninth-grader’s comprehension. Paper presented at the annual conference of the American
Educational Research Association, Toronto.
Michotte, A. (1946). La perception de la causalité. Paris: Vrin.
Milberg, W., Alexander, M. P., Charness, N., McGlinchey-Berroth, R., & Barrett, A. (1988).
Complex arithmetic skill in amnesia: Evidence for a dissociation between compilation
and production. Brain and Cognition, 8, 77-90.
Miller, G. A. (1956). The magical number seven plus or minus two: Some limits on our
capacity for processing information. Psychological Review, 63, 81-97.
Miller, N. E. (1951). Learnable drives and rewards. In S. S. Stevens (Ed.), Handbook of
experimental psychology. New York: Wiley.
Miller, N. E. (1960). The value of behavioral research on animals. American Psychologist,
40, 423-440.
Miller, R. R., & Matzel, L. D. (1989). Contingency and relative associative strength. In S.
B. Klein & R. R. Mowrer (Eds.), Contemporary learning theories: Pavlovian conditioning and
the status of traditional learning theory (pp. 61-84). Hillsdale, NJ: Erlbaum.
Miller, R. R., Barnet, R. C., & Grahame, N. J. (1992). Responding to a conditioned stimu-
lus depends on the current associative status of other cues present during training of that
specific stimulus. Journal of Experimental Psychology: Animal Behavior Processes, 18, 251-264.
Miller, R. R., & Spear, N. E. (1985). Information processing in animals: Conditioned inhibi-
tion. Hillsdale, NJ: Erlbaum.
Moore, J. W., & Gormezano, I. (1961). Yoked comparisons of instrumental and classical
eyelid conditioning. Journal of Experimental Psychology, 62, 552-559.
Moray, N., Bates, A., & Barnett, T. (1965). Experiments on the four-eared man. Journal of
the Acoustical Society of America, 38, 196-201.
Morris, C. D., Bransford, J. D., & Franks, J. J. (1977). Levels of processing versus transfer
appropriate processing. Journal of Verbal Learning and Verbal Behavior, 16, 519-533.
Morris, R.G.M. (1981). Spatial localization does not require the presence of local cues.
Learning and Motivation, 12, 239-260.
Morris, R.G.M. (1990). It’s heads they win, tails I lose. Psychobiology, 18, 261-266.
Morris, R.G.M., Anderson, E., Lynch, G. S., & Baudry, M. (1986). Selective impairment of
learning and blockade of long-term potentiation by an N-methyl-D-aspartate receptor
antagonist, AP5. Nature, 319, 774-776.
Morris, R.G.M., Garrud, P., Rawlins, J.N.P,& O'Keefe, W. (1982). Place navigation is
impaired in animals with hippocampal lesions. Nature (London), 297, 681-683.
Mowrer, O. H. (1947). On the dual nature of learning: A reinterpretation of “condition-
ing” and“problem-solving.” Harvard Educational Review, 17, 102-150.
Muenzinger, K. F. (1928). Plasticity and mechanization of the problem box habit in
guinea pigs. Journal of Comparative Psychology, 8, 45-69.
Murdock, B. B., Jr. (1961). The retention of individual items. Journal of Experimental
Psychology, 62, 618-625.

450
 Bibliography

Murdock, B. B., Jr. (1974). Human memory: Theory and data. Hillsdale, NJ: Erlbaum.
Myers, J. L., Fort, J. G., Katz, L., & Suydam, M. M. (1963). Differential monetary gains and
losses and event probability in a two-choice situation. Journal of Experimental Psychology,
66, 521-522.
Nathan, P. E. (1976). Alcoholism. In H. Leitenberg (Ed.), Handbook of behavior modification
and behavior therapy. Englewood Cliffs, NJ: Prentice-Hall.
Neath, I., Surprenant, A. M. & Crowder, R. G. (1993). The context-dependnent stimulus-
suffix effect. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19,
698-703:
Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibi-
tionless spreading activation and limited-capacity attention. Journal of Experimental
Psychology: General, 106, 226-254.
Neisser, U. (1967). Cognitive psychology. New York: Appleton.
Neisser, U. (1982). Memory observed. San Francisco: W. H. Freeman.
Nelson, T. O. (1971). Savings and forgetting from long-term memory. Journal of Verbal
Learning and Verbal Behavior, 10, 568-576.
Nelson, T. O. (1976). Reinforcement and human memory. In W. K. Estes (Ed.), Handbook
of learning and cognitive processes (Vol. 3). Hillsdale, NJ: Erlbaum.
Nelson, T. O. (1977). Repetition and depth of processing. Journal of Verbal Learning and
Verbal Behavior, 16, 151-172.
Nelson, T. O. (1978). Detecting small amounts of information in memory: Savings for
nonrecognized items. Journal of Experimental Psychology: Human Learning and Memory, 4,
453-468.
Nelson, T. O., Gerber, D., & Narens, L. (1984). Accuracy of feeling-of-knowing judgments
for predicting perceptual identification and relearning. Journal of Experimental Psychology:
General, 113, 282-300.
Neves, D. M. (1977). An experimental analysis of strategies of the Tower of Hanoi puzzle (C.LP.
Working Paper No. 362). Pittsburgh, PA: Carnegie Mellon University.
Neves, D. M., & Anderson, J. R. (1981). Knowledge compilation: Mechanisms for the
automatization of cognitive skills. In J. R. Anderson (Ed.), Cognitive skills and their acqui-
sition. Hillsdale, NJ: Erlbaum.
Newell, A. (1991). Unified theories of cognition. Cambridge, MA: Harvard University Press.
Newell, A., & Rosenbloom, P. S. (1981). Mechanisms of skill acquisition and the law of
practice. In J. R. Anderson (Ed.), Cognitive skills and their acquisition. Hillsdale, NJ:
Erlbaum.
Newell, A., & Simon, H. A. (1961). GPS, a program that simulates human thought. In H.
Billing (Ed.), Lernende Automaten (pp. 109-124). Munich: R. Oldenbourg.
Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ:
Prentice-Hall.
Newport, E. L. (1986, October 17-19). The effect of maturational state on the acquisition of
language. Paper presented at the 11th Annual Boston University Conference on Language
Development.
Newport, E. L. (1990). Maturational constraints on language learning. Cognitive Science,
14, 11-28.

451
 Bibliography

Norman, D. A., & Shallice, T. (1986). Attention to action: Willed and automatic control of
behavior. In R. J. Davidson, G. E. Schwartz, and D. Shapiro (Eds.), Consciousness and self-
regulation. Vol. 4 (pp. 1-18). New York: Plenum Press.
Nosofsky, R. (1988). Similarity, frequency, and category representation. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 10, 104-114.
Nosofsky, R. M. (1986). Attention, similarity, and the identification-categorization rela-
tionship. Journal of Experimental Psychology: General, 115, 39-57.
Nosofsky, R. M., Palmeri, T. J., & McKinley, S. C. (1994). Rule-plus-exception model of
classification learning. Psychological Review, 101, 53-79.
Nottebohm, F. (1970). The ohtogeny of birdsong. Science, 167, 950-956.
Noyd, D. E. (1965). Proactive and intra-stimulus interference in short-term memory for two-,
three- and five-word stimuli. Paper presented at the meeting of the Western Psychological
Association, Honolulu, HI.
O'Keefe, J. A., & Nadel, L. (1978). The hippocampus as a cognitive map. London: Oxford
University Press.
Obrist, P. A., Sutterer, J. R., & Howard, J. L. (1972). Preparatory cardiac changes: A psy-
chobiological approach. In A. H. Black & W. F. Prokasy (Eds.), Classical conditioning II.
New York: Appleton-Century- Crofts.
Ohlsson, S. (1992). The learning curve for writing books: Evidence from Professor
Asimov. Psychological Science, 3, 380-382.
Oldfield, R. C. (1937). Some recent experiments bearing on “internal inhibition.” British
Journal of Psychology, 28, 28-42.
Olds, J., & Milner, P. (1954). Positive reinforcement produced by electrical stimulation of
septal area and other regions of rat brain. Journal of Comparative Physiology and
Psychology, 47, 419-427.
Olton, D. S. (1978). Characteristics of spatial memory. In S. H. Hulse, H. Fowler, & W. K.
Honig (Eds.), Cognitive processes in animal behavior. Hillsdale, NJ: Erlbaum.
Olton, D. S., & Samuelson, R. J. (1976). Remembrance of places passed: Spatial memory
in rats. Journal of Experimental Psychology: Animal Behavior Processes, 2, 97-116.
Olton, D. S., Becker, J. T., & Handelmann, G. E. (1979). Hippocampus, space and memo-
ry. Behavioral and Brain Sciences, 2, 313-365.
Osherson, D. N., Kosslyn, S. M., & Hollerbach, J. M. (Eds.). (1990). Visual cognition and
action: An invitation to cognitive science (Vol. 2). Cambridge, MA: MIT Press.
Otto, T. & Eichenbaum, H. (1992). Dissociable roles of the hippocampus and
orbitofrontal cortex in an odor-guided delayed nonmatch to sample task. Behavioral
Neuroscience, 105, 111-119.
Owen, A. M., Roberts, A. C., Hodges, J. R., Summers, B. A., Polkey, C. E., & Robbins, T. W.
(1993). Contrasting mechanisms of impaired attentional set-shifting in patients with
frontal lobe damage or Parkinson’s disease. Brain, 116, 1159-1175.
Owens, J., Bower, G. H., & Black, J. B. (1979). The “soap opera” effect in story recall.
Memory and Cognition, 7, 185-191.
Paivio, A. (1971). Imagery and verbal processes. New York: Holt, Rinehart, & Winston.
Palinscar, A. S., & Brown, A. L. (1984). Reciprocal teaching of comprehension-foster
ing
and comprehension-monitoring activities. Cognition and Instruction, 1, 117-175.

452
 Bibliography

Palmer, S., Schreiber, G., & Fox, C. (1991, November 22-24). Remembering the earthquake:
“Flashbulb” memory of experienced versus reported events. Paper presented at the 32nd annu-
al meeting of the Psychonomic Society, San Francisco.
Parkin, A. J., Lewinsohn, J., & Folkard, S. (1982). The influence of emotion on serenbdlats
and delayed retention: Levinger and Clark reconsidered. British Journal of Psychology, 73,
389-393,
Pavlov, I. P. (1927). Conditoned reflexes. Oxford: Oxford University Press.
Payne, J. W., Bettman, J. R., & Johnson, E. J. (1988). Adaptive strategy selection in decision
making. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 534-552.
Pearce, J. M. (1994). Similarity and discrimination: A selective review and a connection-
istic model. Psychological Review, 101, 587-607.
Pearce, J. M., & Hall, G. (1980). A model for Pavlovian learning: Variations in the effec-
tiveness of conditioned but not unconditioned stimuli. Psychological Review, 87, 532-552.
Perfetti, C. A. (1985). Reading ability. New York: Oxford University Press.
Peters, D. P. (1988). Eyewitness memory and arousal in a natural setting. In M. M.
Gruneberg, P. E. Morris, & R. N. Sykes (Eds.), Practical aspects of memory: Current research
and issues: Vol. 1. Memory in everyday life (pp. 89-94). Chichester, England: Wiley.
Peterson, L. R., & Peterson, M. (1959). Short-term retention of individual items. Journal of
Experimental Psychology, 58, 193-198.
Peterson, S. B., & Potts, G. R. (1982). Global and specific components of information inte-
gration. Journal of Verbal Learning and Verbal Behavior, 21, 403-420.
Petrides, M., Alivisatos, B, Evans, A. C., & Meyer, E. (1993). Dissociation of human mid-
dorsolateral from posterior dorsolateral frontal cortex in memory processing. Proceedings
of the National Academy of Science, USA, 90, 873-877.
Phelps, E. A. (1989). Cognitive skill learning in amnesics. Unpublished doctoral disserta-
tion, Princeton University, Princeton, NJ.
Pinker, S. (1989). Language acquisition. In M. I. Posner (Ed.), Foundations of cognitive sci-
ence. Cambridge, MA: MIT Press.
Pinker, S. (1994). The language instinct. New York: William Morrow and Co.
Pinker, S., & Prince, A. (1988). On language and connectionism: Analysis of a parallel dis-
tributed processing model of language acquisition. Cognition, 28, 73-193.
Pirolli, P., & Card, S. (in press). Information foraging. Psychological Review.
Pirolli, P. L., & Anderson, J. R. (1985). The role of practice in fact retrieval. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 11, 136-153.
Polson, M., & Richardson, J. (Eds.). (1988). Handbook of intelligent training systems.
Hillsdale, NJ: Erlbaum
Porter, D. (1961). An application of reinforcement principles to classroom teaching.
Cambridge, MA: Harvard University, Graduate School of Education, Laboratory for
Research in Instruction.
Posner, M. (1994). Attention: The mechanisms of consciousness. Proceeding of the National
Academy of Science, USA, 91, 7398-7403.
Postman, L. (1964). Short-term memory and incidental learning. In A. W. Melton (Ed.),
Categories of human learning. New York: Academic Press.

453
 Bibliography

Postman, L. (1974). Transfer, interference, and forgetting. In L. W. Kling & L. A. Riggs

(Eds.), Experimental psychology. New York: Holt, Rinehart, & Winston.
Postman, L., Stark, K., & Fraser, J. (1968). Temporal changes in interference. Journal of
Verbal Learning and Verbal Behavior, 7, 672-694.
Potter, M. C., & Lombardi, L. (1990). Regeneration in the short-term recall of sentences.
Journal of Memory and Language, 29, 633-654.
Premack, D. (1959). Toward empirical behavioral laws: I. Positive reinforcement.
Psychological Review, 66, 219-233.
Premack, D. (1962). Reversibility of the reinforcement relation. Science, 136, 235-237.
Premack, D. (1965). Reinforcement theory. In D. Levine (Ed.), Nebraska Symposium on
Motivation. Lincoln: University of Nebraska Press.
Premack, D. (1971). Catching up with common sense or two sides of a generalization:
Reinforcement and punishment. In R. Glaser (Ed.), The nature of reinforcement. New York:
Academic Press.
Premack, D. (1976). Intelligence in ape and man. Hillsdale, NJ: Erlbaum.
Premack, D., & Premack, A. J. (1983). The mind of an ape. New York: W. W. Norton.
Prokasy, W. F. (Ed.). (1965). Classical conditioning: A symposium. New York: Appleton-
Century-Crofts.
Prokasy, W. F., Grant, D. A., & Myers, N. A. (1958). Eyelid conditioning as a function of
UCS intensity and intertrial interval. Journal of Experimental Psychology, 55, 242-246.
Raaijmakers, J.G.W., & Shiffrin, R. M. (1981). Search of associative memory. Psychological
Review, 88, 93-134.
Rabinowitz, J. C., Mandler, G., & Barsalou, L. W. (1977). Recognition failure: Another case
of retrieval failure. Journal of Verbal Learning and Verbal Behavior, 16, 639-663.
Rachlin, H. C., & Green, L. (1972). Commitment, choice, and self-control. Journal of the
Experimental Analysis of Behavior, 17, 15-22.
Raibert, M. H. (1977). Motor control and learning by the state-space model (Tech. Rep. No.
AI-TR-439). Cambridge, MA: MIT, AI Laboratory.
Raichle, M. E., Fiez, J. A., Vidden, T. O., MacLeod, A. K., Pardo, J.V., Fox, P.T., & Petersen,
S. E. (1994). Practice-related changes in human brain function anatomy during nonmo-
tor learning. Cerebral Cortex, 4, 8-26.
Rashotte, M. E., Griffin, R. W., & Sisk, C. L. (1977). Second-order conditioning of the
pigeon’s keypeck. Animal Learning and Behavior, 5, 25-38.
Ratcliff, R. (1990). Connectionist models of recognition memory: Constraints imposed by
learning and forgetting functions. Psychological Review, 97, 285-308.
Ravitch, D. (1998, Ed.). Brookings papers on Education Policy: 1998. Washington, DC:
Brookings Institution Press.
Raymond, M. J. (1964). The treatment of addiction by aversion conditioning with apo-
morphine. Behavior Research and Therapy, 1, 287-291.
Razran, G. (1971). Mind in evolution: An East/West synthesis of learned behavior and cogni-
tion. Boston: Houghton Mifflin.
Reder, L. M. (1987). Strategy selection in question answering. Cognitive Psychology, 19,
90-138:

454
 Bibliography

Reder, L. M. (Ed.) (1996). Implicit Memory and Metacognition. Mahwah, NJ: Erlbaum.
Reder, L. M., & Gordon, J. S. (1997) Subliminal Perception: Nothing Special, Cognitively
Speaking. In J. Cohen and J. Schooler (Eds.), Cognitive and Neuropsychological approaches
to the study of Consciousness (pp. 125-134). Mahwah, NJ: Erlbaum.
Reder, L. M., & Schunn, C. D. (1996). Metacognition does not imply awareness: Strategy
choice is governed by implicit learning and memory. In L. M. Reder, (Ed.) Implicit Memory
and Metacognition. (pp. 45-77). Mahwah, NJ: Erlbaum, pp. 45-77.
Reder, L. M., Nhouyvansivong, A. Schunn, C. D., Ayers, M. S., Angstadt, P., & Hiraki, K.
(1997). Modeling the Mirror effect in a Continuous Remember/Know Paradigm.
Proceedings of the Cognitive Science Conference, 644-649.
Reed, S. K. (1972). Pattern recognition and categorization. Cognitive Psychology, 3,
382-407.
Reiss, S., & Wagner, A. R. (1972). CS habituation produces a“latent inhibition” effect but
no active conditioned inhibition. Learning and Motivation, 3, 237-245.
Reitman, J. (1976). Skilled perception in GO: Deducing memory structures from interre-
sponse times. Cognitive Psychology, 8, 336-356.
Reitman, W. (1965). Cognition and thought. New York: Wiley.
Rescorla, R. A. (1968a). Pavlovian conditioned fear in Sidman avoidance learning. Journal
of Comparative and Physiological Psychology, 65, 55-60.
Rescorla, R. A. (1968b). Probability of shock in the presence and absence of CS in fear
conditioning. Journal of Comparative and Physiological Psychology, 66, 1-5.
Rescorla, R. A. (1973). Effect of US habituation following conditioning. Journal of
Comparative Physiological Psychology, 82, 137-143.
Rescorla, R. A. (1988a). Facilitation based on inhibition. Animal Learning and Behavior, 16,
169-176.
Rescorla, R. A. (1988b). Pavlovian conditioning: It’s not what you think it is. American
Psychologist, 43, 151-160.
Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations on
_ the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy
(Eds.), Classical conditioning: II. Current research and theory (pp. 64-99). New York:
Appleton-Century- Crofts.
Resnick, D. P,, & Resnick, L. B. (1977). The nature of literacy: An historical exploration.
Harvard Educational Review, 47, 370-385.
Resnick, L. B. (1977). Assuming that everyone can learn everything, will some learn less?
School Review, 85, 445-452.
Restle, F. (1957). Discrimination of cues in mazes: A resolution of the “place-vs.-
response” question. Psychological Review, 64, 217-228.
Rilling, M. (1977). Stimulus control and inhibitory processes. In W. K. Honig & J.E.R.
Staddon (Eds.), Handbook of operant behavior. Englewood Cliffs, NJ: Prentice-Hall.
Rizley, R. C., & Rescorla, R. A. (1972). Associations in higher order conditioning and sen-
sory preconditioning. Journal of Comparative Physiological Psychology, 81, 1-11.
Roberts, R. J., Hager, L. D., & Heron, C. (1994). Prefrontal cognitive processes: Working
memory and inhibition in the antisaccade task. Journal of Experimental Psychology:
General, 123, 374-393.

455
 Bibliography

Roberts, W. A. (1984). Some issues in animal spatial memory. In H. L. Roitblat, T. G. Bever,

& H.S. Terrace (Eds.), Animal cognition (pp. 425-443). Hillsdale, NJ: Erlbaum.
Roberts, W. A., & Grant, D. S. (1978). An analysis of light-induced retroactive inhibition
in pigeon short-term memory. Journal of Experimental Psychology: Animal Behavior
Processes, 4, 219-236.
Robinson, F. P. (1961). Effective study. New York: Harper & Row.
Robinson, G. S., Jr., Crooks, G. B., Jr., Stinkman, P. G., & Gallagher, M. (1989). Behavioral
effects of MK-901 mimic deficits associated with hippocampal damage. Psychobiology, 17,
156-164.
Roediger, H. L. (1990). Implicit memory: Retention without remembering. American
Psychologist, 45, 1043-1056.
Roediger, H. L., & Blaxton, T. A. (1987). Retrieval modes produce dissociations in mem-
ory for surface information. In D. Gorfein & R. R. Hoffman (Eds.), Memory and cognitive
processes: The Ebbinghaus Centennial Conference (pp. 349-379). Hillsdale, NJ: Erlbaum.
Rohlen, T. (1998). Comment on Stevenson, H. W., & Lee, S. An examination of American
student achievement from an international perspective.
Roitblat, H. L. (1987). Introduction to comparative cognition. New York: W. H. Freeman.
Roitblat, H. L., Bever, T. G., & Terrace, H. S. (1984). Animal cognition. Hillsdale, NJ:
Erlbaum.
Rosch, E. (1973). On the internal structure of perceptual and semantic categories. In T. E.
Moore (Ed.), Cognitive development and the acquisition of language. New York: Academic Press.
Rosch, E. (1975). Cognitive representations of semantic categories. Journal of Experimental
Psychology: General, 104, 192-223.
Rosch, E. (1977). Human categorization. In N. Warren (Ed.), Advances in cross-cultural psy-
chology (Vol. 1). London: Academic Press.
Rosenbaum, D. A. (1991). Human motor control. San Diego: Academic Press.
Rosenbaum, D. A., Inhoff, A. W., & Gordon, A. M. (1984). Choosing between movement
sequences: A hierarchical editor model. Journal of Experimental Psychology: General, 113,
8725993.
Ross, J., & Lawrence, K. A. (1968). Some observations on memory artifice. Psychonomic
Science, 13, 107-108.
Ross, L. E. (1965). Eyelid conditioning as a tool in psychological research: Some problems
and prospects. In W. F. Prokasy (Ed.), Classical conditioning (pp. 249-268). New York:
Appleton-Century-Crofts.
Rubin, D. C. & Wenzel, A. E. (1996). One hundred years of forgetting: A quantitative
description of retention. Psychological Review, 103, 734-760
Rumelhart, D. E. (1975). Notes on a schema for stories. In D. G. Bobrow & A. M. Collins
(Eds.), Representation and understanding. New York: Academic Press.
Rumelhart, D. E., & McClelland, J. L. (1986). Parallel distributed processing: Explorations in
the microstructure of cognition (Vol. 1). Cambridge, MA: MIT Press/Bradford Books.
Rundus, D. J. (1971). Analysis of rehearsal processes in free recall. iceiion of Experimental
Psychology, 89, 63-77.
Russell, S., & Norvig, P. (1995). Artificial intelligence: A modern approach. Upper Saddle
River, NJ: Prentice Hall.

456
 Bibliography

Sachs, J. S. (1967). Recognition memory for syntactic and semantic aspects of connected
discourse. Perception & Psychophysics, 2, 437-442.
Sakai, K. & Miyashita,Y.(1991). Neural organization for the long-term memory-of paired
associates. Nature, 354, 152-155.
Sakitt, B. (1976). Iconic memory. Psychological Review, 83, 257-276.
Salmoni, A. W., Schmidt, R. A., & Walter, C. B. (1984). Knowledge of results and motor
learning: A review and critical reappraisal. Psychological Bulletin, 95, 355-386.
Salthouse, T. A. (1985). Anticipatory processes in transcription typing. Journal of Applied
Psychology, 70, 264-271.
Salthouse, T. A. (1986). Perceptual, cognitive, and motoric aspects of transcription typing.
Psychological Bulletin, 99, 303-319.
Saltzman, I. J. (1949). Maze learning in the absence of primary reinforcement: A study of
secondary reinforcement. Journal of Comparative and Physiological Psychology, 42, 161-173.
Santa, J. L. (1977). Spatial transformations of words and pictures. Journal of Experimental
Psychology: Human Learning and Memory, 3, 418-427.
Saucier, D., & Cain, D. P. (1995). Spatial learning without NMDA receptor-dependent
long-term potentiation. Nature, 378, 186-189.
Saufley, W. H., Otaka, S. R., & Bavaresco, J. L. (1985). Context effects: Classroom tests and
context independence. Memory and Cognition, 13, 522-528.
Savage-Rumbaugh, E. S., Murphy, J., Sevik, R. A., Brakke, K. E., Williams, S. L., &
Rumbaugh, D. M. (1993). Language comprehension in ape and child. Monographs of the
Society for Research in Child Development, 58, Serial No. 233.
Schacter, D. L. (1987). Implicit memory: History and current status. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 13, 501-518.
Schacter, D. L., Cooper, L. A., Delaney, S. M., Peterson, M. A., & Tharan, M. (1991).
Implicit memory for possible and impossible objects: Constraints on the construction of
structural descriptions. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 17, 3-19.
Schmidt, R. A. (1988a). Motor and action perspectives on motor behavior. In O. G. Meijer
& K. Rother (Eds.), Complex movement behavior: The motor—action controversy (pp. 3-44).
Amsterdam: Elsevier.
Schmidt, R. A. (1988b). Motor control and learning: A behavioral emphasis. Champaign, IL:
Human Kinetics Publishers.
Schmidt, R. A. (1991). Motor skills acquisition. In R. Dulbecco (Ed.), Encyclopedia of human
biology (Vol. 5, pp. 121-129). Orlando, FL: Academic Press.
Schmidt, R. A., & Bjork, R. A. (1992). New conceptualizations of practice: Common prin-
ciples in three paradigms suggest new concepts for training. Psychological Science, 3,
207-217.
Schmidt, R. A., & Shapiro, D. C. (1986). Optimizing feedback utilization in motor skill train-
ing (Tech. Rep. No. 1/86). Motor Control Laboratory, UCLA. (ARI Contract No. MDA-
903-85-K-0225).
Schmidt, R. A., & White, J. L. (1972). Evidence for an error detection mechanism in motor
skills: A test of Adams’ closed-loop theory. Journal of Motor Behavior, 4, 143-154.
Schmidt, R. A., Shapiro, D. C., Winstein, C. J., Young, D. E., & Swinnen, S. (1987). Feedback

457
 Bibliography

and motor skill training: Relative frequency of KR and summary KR (Tech. Rep. No 1/87).
Motor Control Laboratory, UCLA. (ARI Contract No. MDA903-85-K-0225).
Schneiderman, B. (1976). Exploratory experiments in programmer behavior. International
Journal of Computer and Information Sciences, 5, 123-143.
Schneiderman, N. (1973). Classical (Pavlovian) conditioning. Morristown, NJ: General
Learning Press.
Schofield, J. W., & Evans-Rhodes, D. (1989). Artificial intelligence in the classroom: The
impact of a computer-based tutor on teachers and students. Proceedings of the Fourth
International Conference on AI and Education (pp. 238-243). Amsterdam.
Schonberg, H. C. (1970). The lives of great composers. New York: W. W. Norton.
Schooler, L. J., & Anderson, J. R. (1990). The disruptive potential of immediate feedback.
Proceedings of the 12th Annual Conference of the Cognitive Science Society (pp. 702-708).
Cambridge, MA.
Schustack, M. W., & Sternberg, R. J. (1981). Evaluation of evidence in causal inference.
Journal of Experimental Psychology: General, 110, 101-120.
Schwartz, B. (1973). Maintenance of keypecking in pigeons by a food avoidance but not
a shock avoidance contingency. Animal Learning and Behavior, 1, 164-165.
Schwartz, B. (1989). Psychology of learning and behavior (3rd ed.). New York: W. W. Norton.
Secretary's Commission on Achieving Necessary Skills. (1991, November). Scales for com-
petencies and foundation skills (Draft). Washington, DC: United States Department of
Labor.
Seligman, M. E. P. (1975). Helplessness: On depression, development and death. San
Francisco: W. H. Freeman.
Seligman, M.E.P,, & Johnson, J. C. (1973).A cognitive theory of avoidance learning. In F.
J. McGuigan & D. B. Lumsden (Eds.), Contemporary approaches to conditioning and learn-
ing (pp. 69-110). Washington, DC: V. H. Winston.
Seligman, M.E.P, & Maier, S. F. (1967). Failure to escape traumatic shock. Journal of
Experimental Psychology, 74, 1-9.
Shaklee, H., & Tucker, D. (1980). A rule analysis of judgments of covariation between
events. Memory and Cognition, 8, 459-467.
Shapiro, D. C., & Schmidt, R. A. (1982). The schema theory: Recent evidence and devel-
opmental implications. In J.A.S. Kelso & J. E. Clark (Eds.), The development of movement
control and coordination (pp. 113-150). New York: Wiley.
Sheffield, F. D., Wulff, J. J., & Backer, R. (1951). Reward value of copulation without sex
drive reduction. Journal of Comparative and Physiological Psychology, 44, 3-8.
Shepard, L. A. (1991). Psychometricians’ beliefs about learning. Educational Researcher, 20,
22163
Shephard, R. N. (1967). Recognition memory for words, sentence, and pictures. Journal of
Verbal Learning and Verbal Behavior, 6, 156-163.
Shettleworth, S. J. (1975). Reinforcement and the organization of behavior in golden
hamsters: Hunger, environment, and food reinforcement. Journal of Experimental
Psychology: Animal Behavior Processes, 104, 56-87.
Shik, M. L., Severin, F.V., & Orlovskii, G. N. (1966). Control of walking and running by
means of electrical stimulation of the mid-brain. Biophysics, 11, 756-765.

458
 Bibliography

Shimp, C. P. (1969). Optimal behavior in free-operant experiments. Psychological Review,

19, 311-330.
Shimp, C. P. (1976). Short-term memory in the pigeon: Relative recency. Journal of the
Experimental Analysis of Behavior, 25, 55-61.
Shrager, J. (1985). Instructionless learning: Discovery of the mental model of a complex
device. Unpublished doctoral dissertation, Carnegie Mellon University, Pittsburgh, PA.
Shrager, J. C., Hogg, T., & Huberman, B. A. (1988). A dynamical theory of the power-law
learning in problem-solving. Proceedings of the 10th Annual Conference of the Cognitive
Science Society (pp. 468-474). Hillsdale, NJ: Cognitive Science Society.
Shultz, T. R. (1982). Rules for causal attribution. Monographs of the Society for Research in
Child Development, 47 (1, Serial No. 194).
Shultz, T. R., Fischer, G. W., Pratt, C. C., & Rulf, S. (1986). Selection of causal rules. Child
Development, 57, 143-152.
Sidman, M. (1966). Two temporal parameters of the maintenance of avoidance behavior
in the white rat. Journal of Comparative and Physiological Psychology, 46, 253-261.
Siegel, S. (1983). Classical conditioning, drug tolerance, and drug dependence. InY. Israel,
F. B. Glaser, H. Kalant, R. E. Popham, W. Schmidt, & R. G. Smart (Eds.), Research advances
in alcohol and drug problems (Vol. 7). New York: Plenum.
Siegel, S., Hearst, E., George, N., & O’Neil, E. (1968). Generalization gradients obtained
from individual subjects following classical conditioning. Journal of Experimental
Psychology, 78, 171-174.
Siegler, R. S. (1988). Strategy choice procedures and the development of multiplication
skill. Journal of Experimental Psychology: General, 117, 258-275.
Siegler, R. S., & Shrager, J. C. (1984). A model of strategy choice. In C. Sophian (Ed.),
Origin of cognitive skills (pp. 229-293). Hillsdale, NJ: Erlbaum.
Silberberg, A., Hamilton, B., Ziriax, J. M., & Casey, J. (1978). The structure of choice.
Journal of Experimental Psychology: Animal Behavior Processes, 4, 368-398.
Simon, H. A. (1955). A behavior model of rational choice. Quarterly Journal of Economics,
69, 99-118.
Simon, H. A., & Gilmartin, K. (1973). A simulation of memory for chess positions.
Cognitive Psychology, 5, 29-46.
Singley, M. K., & Anderson, J. R. (1989). The transfer of cognitive skill. Cambridge, MA:
Harvard University Press.
Singley, M. K., Anderson, J. R., Givens, J. S., & Hoffman, D. (1989). The algebra word
problem tutor. Proceedings of the Fourth international Conference on AI and Education
(pp. 267-275). Amsterdam.
Skinner, B. F. (1938). The behavior of organisms. New York: Appleton-Century-Crofts.
Skinner, B. F. (1948). Walden two. New York: Macmillan.
Skinner, B. F. (1957). Verbal behavior. Englewood Cliffs, NJ: Prentice-Hall.
Skinner, B. F. (1971). Beyond freedom and dignity. New York: Knopf.
Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 4, 592-604.
Slamecka, N. J., & Katsaiti, L. T. (1987). The generation effect as an artifact of selective
displaced rehearsal. Journal of Memory and Language, 26, 589-607.

459
 Bibliography

Slamecka, N. J., & McElree, B. (1983). Normal forgetting of verbal lists as a function of
their degree of learning. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 9, 384-397.
Slater-Hammel, A. T. (1960). Reliability, accuracy and refractoriness of a transit reaction.
Research Quarterly, 31, 217-228.
Smith, E. E. (1989). Concepts and induction. In M. I. Posner (Ed.), Foundations of cognitive
science (pp. 501-526). Cambridge, MA: MIT Press.
Smith, E. E., Adams, N., & Schorr, D. (1978). Fact retrieval and the paradox of interfer-
ence. Cognitive Psychology, 10, 438-464.
Smith, E. E., & Jonides, J. (T995). Working memory in humans: Neurophychological evi-
dence. In M. S. Gazzinga (Ed.) The cognitive neurosciences, 1009-1020. Cambridge, MA
MIT Press.
Smith, E. E., Patalano, A. L., & Jonides, J. (1998). Alternative strategies of categorization.
Cognition, 65, 167-196.
Smith, S. M., Glenberg, A., & Bjork, R. A. (1978). Environmental context and human
memory. Memory and Cognition, 6, 342-353.
Solomon, R. L., & Corbit, J. D. (1974). An opponent-process theory of motivation: I. The
temporal dynamics of affect. Psychological Review, 81, 119-145.
Solomon, R. L., & Wynne, L. C. (1953). Traumatic avoidance learning: Acquisition in nor-
mal dogs. Psychological Monographs, 67 (Whole No. 354).
Solomon, R. L., Kamin, L. J., & Wynne, L. C. (1953). Traumatic avoidance learning: The
outcomes of several extinction procedures with dogs. Journal of Abnormal and Social
Psychology, 48, 291-302.
Soloway, E., Lochhead, J., & Clement, J. (1982). Does computer programming enhance
problem solving ability? Some positive evidence on algebra word problems. In J. Seidel,
R. E. Anderson, & B. Hunter (Eds.), Computer literacy. New York: Academic Press.
Sonuga-Barke, E.J.S., Lea, S.E.G., & Webley, P. (1989). The development of adaptive
choice in a self-control paradigm. Journal of the Experimental Analysis of Behavior, 51,
77-85.
Spear, N. E., & Miller, R. R. (1981). Information processing in animals: Memory mechanisms.
Hillsdale, NJ: Erlbaum.
Spelke, E., Hirst, W., & Neisser, U. (1976). Skills of divided attention. Cognition, 4,
215-230.
Spence, K. W. (1937). The differential response in animals to stimuli varying within a sin-
gle dimension. Psychological Review, 44, 430-444.
Spence, K. W. (1952). The nature of the response in discrimination learning. Psychological
Review, 59, 89-93.
Spence, K. W., & Ross, L. E. (1959). A methodological study of the form and latency of
eyelid responses in conditioning. Journal of Experimental Psychology, 58, 376-381.
Sperling, G. A. (1960). The information available in brief visual presentation. Psychological
Monographs, 74 (Whole No. 498).
Sporer, S. L. (1991). Deep—deeper—deepest? Encoding strategies and the recognition of
human faces. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17,
323-333.

460
 Bibliography

Squire, L. R. (1987). Memory and brain. New York: Oxford University Press.
Squire, L. R. (1989). On the course of forgetting in very long-term memory. eles of
Experimental Psychology: Learning, Memory and Cognition, 15, 241-245.
Squire, L. R. (1992). Memory and the hippocampus: A synthesis from findings with rats,
monkeys, and humans. Psychological Review, 99, 195-232.
St. Claire-Smith, R., & MacLaren, D. (1983). Response preconditioning effects. Journal of
Experimental Psychology, 1, 41-48.
Staddon, J.E.R. (1983) Adaptive behavior and learning. Cambridge: Cambridge University
Press.
Staddon, J.E.R., & Ettinger, R. H. (1989). Learning: An introduction to the principles of adap-
tive behavior. San Diego: Harcourt Brace Jovanovich.
Staddon, J.E.R., & Simmelhag, V. L. (1971). The“superstition” experiment: A reexamina-
tion of its implications for the principles of adaptive behavior. Psychological Review, 78,
3-43.
Standing, L. (1973). Learning 10,000 pictures. Quarterly Journal of Experimental Psychology,
25, 207-222.
Staubli, U., Thibault, O., DiLorenzo, M., & Lynch, G. (1989). Antagonism of NMDA
receptors impairs acquisition but not retention of olfactory memory. Behavioral
Neuroscience, 103, 54-60.
Steele, D. R. (1994). Partial recall. Liberty, 37-47.
Stein, L. (1978). Reward transmitters: Catecholamines and opioid peptides. In M. A.
Lipton, A. DiMascio, & K. F. Killam (Eds.), Psychopharmacology: A generation of progress
(pp. 569-581). New York: Raven Press.
Steinberg, H., & Summerfield, A. (1957). Influence of a depressant drug on acquisition in
rote learning. Quarterly Journal of Experimental Psychology, 9, 138-145.
Stephens, D. W., & Krebs, J. R. (1986). Foraging theory. Princeton, NJ: Princeton University
Press.
Sternberg, S. (1969). Memory scanning: Mental processes revealed by reaction time
experiments. American Scientist, 57, 421-457.
Stevenson, H. W., & Lee, S. (198). An examination of American student achievement
from an international perspective. In D. Ravitch (Ed.) Brookings papers on Education Policy:
1998, pp. 7-52. Washington, DC: Brookings Institution Press.
Stevenson, H. W., & Stigler, J. W. (1992). The learning gap: Why our schools are failing and
what we can learn from Japanese and Chinese education. New York: Summit Books.
Sticht, T. G. (1972). Learning by listening. In R. O. Freedle & J. B. Carroll (Eds.), Language
comprehension and the acquisition of knowledge. Washington, DC: Winston.
Sticht, T. G., & James, J. H. (1984). Listening and reading. In P. D. Pearson, R. Barr, M. L.
Kamil, & P. Mosenthal (Eds.), Handbook of reading research. New York: Longman.
Straub, R. O., Seidenberg, M. S., Bever, T. G., & Terrace, H. S. (1979). Serial learning in the
pigeon. Journal of the Experimental Analysis of Behavior, 32, 137-148.
Sulin, R. A., & Dooling, D. J. (1974). Intrusion of a thematic idea in retention of prose.
Journal of Experimental Psychology, 103, 255-262.
Suppes, P. (1964). Modern learning theory and the elementary school curriculum.
American Educational Research Journal, 2, 79-93.

461
 Bibliography

Sutherland, N. S., & Mackintosh, N. J. (1971). Mechanisms of animal discrimination learn-

ing. New York: Academic.
Sutherland, R. J., & Rudy, J. W. (1991). Configural association theory: The role of the hip-
pocampal formation in learning, memory, and amnesia. Psychobiology, 17, 129-144.
Swanson, L. W., Teyler, T. J., & Thompson, R. F. (Eds.). (1982). Mechanisms and functional
implications of hippocampal LTP (Neurosciences Research Program 20). Boston: MIT Press.
Taub, E., & Berman, A. J. (1968). Movement and learning in the absence of sensory feed-
back. In S. J. Freedman (Ed.), The neuropsychology of spatially oriented behavior.
Homewood, IL: Dorsey Press.
Teasdale, J. D., & Russell, M L. (1983). Differential effects of induced mood on the recall
of positive, negative and neutral words. British Journal of Clinical Psychology, 22, 163-171.
Terrace, H. S. (1963). Errorless transfer of a discrimination across two continua. Journal of
Experimental Analysis of Behavior, 6, 223-232.
Terrace, H. S. (1972). By-products of discrimination learning. In G. H. Bower (Ed.), The
psychology of learning and motivation (Vol. 5). New York: Academic Press.
Terrace, H. S. (1984). Animal cognition. In H. L. Roitblat, T. G. Bever, & H. S. Terrace (Eds.),
Animal cognition. Hillsdale, NJ: Erlbaum.
Terrace, H. S. (1998). The comparative psychology of serially organized behavior. In S.
Fountain, J. H. Danks, & M. K. McBeth (Eds.), Biomedical Implications of Model Systems of
Complex Cognitive Capacities. New York: Sage Publishing Company.
Terrace, H.S., & McGonigle, B. (1994). Memory and representation of serial order by chil-
dren, monkeys, and pigeons. Current Directions in Psychological Science, 3, 180-185.
Terrace, H. S., Jaswal,V.,Brannon, E., & Chen, S. (1996). What is a chunk? Ask a monkey.
Abstracts of the Psychonomic Society, 1, 35.
Terrace, H. S., Pettito, L.A., Sanders, R. J., & Bever, T. G. (1979). Can an ape create a sen-
tence? Science, 206, 891-902.
Thomas, E. L., & Robinson, H. A. (1972). Improving reading in every class: A sourcebook for
teachers. Boston: Allyn & Bacon.
Thompson, R. F. (1986). The neurobiology of learning and memory. Science, 233, 941-947.
Thompson, R. F., Donegon, N. H., & Lavond, D. G. (1988). The psychobiology of learn-
ing and memory. In R. C. Atkinson, R. J. Herrnstein, G. Lindzey, and R. D. Luce (Eds.),
Stevens’ handbook of experimental psychology: Vol. 2. Learning and cognition (pp. 245-350).
New York: Wiley.
Thorndike, E. L. (1898). Animal intelligence: An experimental study of the associative
processes in animals. Psychological Monographs, 2 (Whole No. 8).
Thorndyke, P. W. (1977). Cognitive structures in comprehension and memory in narra-
tive discourse. Cognitive Psychology, 9, 77-110.
Timberlake, W. (1980). A molar equilibrium theory of learned performance. In G. H.
Bower (Ed.), The psychology of learning and motivation (Vol. 14). NewYork: Academic Press.
Timberlake, W. (1983). Rats’ responses to a moving object related to food or water: A
behavior-systems analysis. Animal Learning and Behavior, 11, 309-320.
Timberlake, W. (1984). Behavior regulation and learned performance: Some misappre-
hensions and disagreements. Journal of the Experimental Analysis of Behavior, 41, 355-375.

462
 Bibliography

Timberlake, W., & Grant, D. S. (1975). Auto-shaping in rats to the presentation of anoth-
er rat predicting food. Science, 190, 690-692.
Tinklepaugh, O. L. (1928). An experimental study of representative factors in monkeys.
Journal of Comparative Psychology, 8, 197-236.
Tobin, H., Logue, A. W., Chelonis, J. J. Ackerman, K.T., & May, J. G., III (1996). Self-con-
trol in the monkey Macaca fascicularis. Animal Learning and Behavior, 24, 168-174.
Tolman, E. C., & Honzik, C. H. (1930a). “Insight” in rats. University of California Publ.
Psychology, 4, 215-232.
Tolman, E. C., & Honzik, C. H. (1930b). Introduction and removal of reward, and maze
performance in rats. University of California Publ. Psychology, 36, 221-229.
Tolman, E. C., Ritchie, B. F., & Kalish, D. (1946). Studies in spatial learning: II. Place learn-
ing versus response learning. Journal of Experimental Psychology, 36, 221-229.
Tolman, E. C., Ritchie, B. F., & Kalish, D. (1947). Studies in spatial learning: VResponse
learning vs. place learning by the non-correction method. Journal of Experimental
Psychology, 37, 285-292.
Trowbridge, M. H., & Cason, H. (1932). An experimental study of Thorndike’s theory of
learning. Journal of General Psychology, 7, 245-258.
Tulving, E. (1975). Ecphoric processing in recall and recognition. In J. Brown (Ed.), Recall
and recognition. London: Wiley.
Tulving, E. (1983). Elements of episodic memory. Oxford: Oxford University Press.
Tulving, E., & Flexser, A. J. (1992). On the nature of the Tulving-Wiseman function.
Psychological Review, 99, 543-546.
Tulving, E., & Osler, S. (1968). Effectiveness of retrieval cues in memory for words. Journal
of Experimental Psychology, 77, 593-601.
Tulving, E., & Pearlstone, Z. (1966). Availability versus accessibility of information in
memory for words. Journal of Verbal Learning and Verbal Behavior, 5, 381-391.
Tulving, E., & Psotka, J. (1971). Retroactive inhibition in free-recall: Inaccessibility of
information available in the memory store. Journal of Experimental Psychology, 87, 1-8.
Tulving, E., & Schacter, D. L. (1990). Priming and human memory systems. Science, 247,
301-306.
Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in
episodic memory. Psychological Review, 80, 352-373.
Tulving, E., & Wiseman, S. (1975). Relation between recognition and recognition failure
of recallable words. Bulletin of the Psychonomic Society, 6, 79-82.
Tversky, A. (1972). Elimination by aspects: A theory of choice. Psychological Review, 79,
281=299.
Ulrich, R. E., & Azrin, N. H. (1962). Reflexive fighting in response to aversive stimulation.
Journal of the Experimental Analysis of Behavior, 5, 511-520.
Underwood, B. J., & Freund, J. S. (1968). Errors in recognition learning and retention.
Journal of Experimental Psychology, 78, 55-63.
Underwood, B. J., & Freund, J. S. (1970). Relative frequency judgments and verbal dis-
crimination learning. Journal of Experimental Psychology, 83, 279-285.
Vaccarino, F. J., Schiff, B. B., & Glickman, S. E. (1989). Biological view of reinforcement. In

463
 Bibliography

S. B. Klein & R. R. Mowrer (Eds.), Contemporary learning theories: Instrumental condition-

ing and the impact of biological constraints on learning (pp. 111-142). Hillsdale, NJ: Erlbaum.
Van Lehn, K. (1989). Problem-solving and cognitive skill acquisition. In M. Posner (Ed.),
Foundations of cognitive science. Cambridge, MA: MIT Press.
Van Lehn, K. (1990). Mind bugs: The origins of procedural misconceptions. Cambridge, MA:
MIT Press. ;
Vaughan, W., Jr. (1981). Melioration, matching, and maximizing. Journal of the
Experimental Analysis of Behavior, 36, 141-149.
Venezky, R. L. (1970). The structure of English orthography. The Hague, Netherlands:
Mouton. 3
Vye, N. J., Schwartz, D. L., Bransford, J. D., Barron, B. J., Zech, L., & Cognition and
Technology Group at Vanderbilt (1998). SMART environments that support monitoring,
reflection, and revision. In D. Hacker, J. Dunlosky, & A. Graesser (Eds.), Metacognition in
educational theory and practice. Mahwah, NJ: Erlbaum.
Wagner, A. R. (1969). Stimulus validity and stimulus selection in associative learning. In
N. J. Mackintosh & W. K. Honig (Eds.), Fundamental issues in associative learning (pp.
90-122). Halifax, Nova Scotia, Canada: Dalhousie University Press.
Wagner, A. R. (1978). Expectancies and the priming of STM. In S. H. Hulse, H. Fowler, &
W. K. Honig (Eds.), Cognitive aspects of animal behavior. Hillsdale, NJ: Erlbaum.
Wagner, A. R. (1981). SOP: A model of automatic memory processing in animal behav-
ior. In N. E. Spear & R. R. Miller (Eds.), Information processing in animals: Memory mecha-
nisms (pp. 5-47). Hillsdale, NJ: Erlbaum.
Wagner, A. R., Rudy, J. W., & Whitlow, J. W. (1973). Rehearsal in animal conditioning.
Journal of Experimental Psychology, 97, 407-426.
Wasserman, E. A. (1990a). Attribution of causality to common and distinctive elements of
compound stimuli. Psychological Science, 1, 298-302.
Wasserman, E. A. (1990b). Detecting response-outcome relations: Toward an under-
standing of the causal texture of the environment. In G. H. Bower (Ed.), The psychology of
learning and motivation (Vol. 26, pp. 27-82). San Diego: Academic Press.
Wasserman, E. A., & Miller, R. (1997). What's elementary about associative learning? The
Annual Review of Psychology, 48, 573-607.
Wasserman, E. A., Elek, S. M., Chatlosh, D. C., & Baker, A. G. (1993). Rating causal rela-
tions: Role of probability in judgments of response-outcome contingency. Journal of
Experimental Psychology: Learning Memory, and Cognition, 19, 174-188.
Wasserman, E. A., Franklin, S. R., & Hearst, E. (1974). Pavlovian appetitive contingencies
and approach versus withdrawal to conditioned stimuli in pigeons. Journal of Comparative
and Physiological Psychology, 86, 616-627.
Wasserman, E. A., Kiedinger, R. E., & Bhatt, R. S. (1988). Conceptual behavior in pigeons:
Categories, subcategories, and pseudocategories. Journal of Experimental Psychology:
Animal Behavior Processes, 14, 235-246.
Watkins, M. J., & Tulving, E. (1975). Episodic memory: When recognition fails. Journal of
Experimental Psychology, General , 1, 5-29. ;
Watkins, O. C., & Watkins, M. J. (1975). Build-up of proactive inhibition as a cue-overload
effect. Journal of Experimental Psychology: Human Learning and Memory, 104, 442-452.

464
 Bibliography

Watts, F. N., & Sharrock, R. (1987). Cued recall in depression. British Journal of Clinical
Psychology, 61, 1-12.
Watts, F. N., MacLeod, A., & Morris, L. (1988). A remedial strategy for memory and con-
centration problems in depressed patients. Cognitive Therapy and Research, 12, 185-193.
Watts, F. N., Morris, L., & MacLeod,A. (1987). Recognition memory in depression. Journal
of Abnormal Psychology, 96, 273-275.
Waugh, N. C., & Norman, D. A. (1965). Primary memory. Psychological Review, 72, 89-104.
Weisberg, P., & Waldrop, P. B. (1972). Fixed-interval work habits of congress. Journal of
Applied Behavior Analysis, 5, 93-97.
Weisman, R. G., & Premack, D. (1966). Reinforcement and punishment produced by the
same response depending upon the probability relation between the instrumental and
contingent responses. Paper presented at the Psychonomic Society Meeting, St. Louis.
Weisman, R. G., Wasserman, E. A., Dodd, P. W., & Lunew, M. B. (1980). Representation
and retention of two-event sequences in pigeons. Journal of Experimental Psychology:
Animal Behavior Processes, 6, 312-325.
Wenger, E. (1987). Artificial intelligence and tutoring systems: Computational and cognitive
approaches to the communication of knowledge. Los Altos, CA: Morgan Kaufmann.
Wertheimer, M. (1979). A brief history of psychology (2nd ed.). New York: Holt, Rinehart, &
Winston.
Wertheimer, R. (1990). The Geometry Proof Tutor: An“Intelligent” Computer-based tutor
in the classroom. Mathematics Teacher, 308-313.
West, M. J., & King, A. P. (1980). Enriching cowbird song by social deprivation. Journal of
Comparative Physiological Psychology, 94, 263-270.
Westbury, I. (1992). Comparing American and Japanese achievement: Is the United
States really a low achiever? Educational Researcher, 21, 18-24.
Westbury, I. (1993). American and Japanese achievement...again: A response to Baker.
Educational Researcher, 22, 21-25.
White, M. (1987). The Japanese education challenge: A commitment to children. New York:
Free Press.
Wickelgren, W. A. (1974). How to solve problems. New York: W. H. Freeman.
Wickelgren, W. A. (1975). Alcoholic intoxication and memory storage dynamics. Memory
& Cognition, 3, 385-389.
Wickelgren, W. A. (1976). Memory storage dynamics. In W. K. Estes (Ed.), Handbook of
learning and cognitive processes (Vol. 4). Hillsdale, NJ: Erlbaum.
Wickelgren, W. A. (1977). Learning and memory. Englewood Cliffs, NJ: Prentice-Hall.
Wilcoxon, H. C., Dragoin, W. B., & Kral, P. A. (1971). IlIness-induced aversions in rat and
quail: Relative salience of visual and gustatory cues. Science, 171, 826-828.
Williams, B. (1988). Reinforcement, choice, and response strength. In R. C. Atkinson, R. J.
Herrnstein, G. Lindzey, and R. D. Luce (Eds.), Stevens’ handbook of experimental psycholo-
gy: Vol. 2. Learning and cognition (pp. 167-244). New York: Wiley.
Williams, J. P. (1979). Reading instruction today. American Psychologist, 34, 917-922.
Wixted, J. T., & Ebbesen, E. B. (1991). On the form of forgetting. Psychological Science, 2,
409-415.

465
 Bibliography

Wolfe, J. B. (1936). Effectiveness of token-rewards for chimpanzees. Comparative

Psychology Monographs, 12 (5, Serial No. 60).
Wollen, K. A., Weber, A., & Lowry, D. H. (1972). Bizarreness versus interaction of images
as determinants of learning. Cognitive Psychology, 3, 518-523.
Wolpe, J. (1958). Psychotherapy by reciprocal inhibition: Stanford, CA: Stanford University
Press.
Wolpe, J. (1982). The practice of behavior therapy (3rd ed.). New York: Pergamon.
Wulf, G., Schmidt, R. A., & Deubel, H. (1993). Reduced feedback frequency enhances
generalized motor program learning but not parameterization learning. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 19, 1134-1150.
Yerkes, R. M., & Dodson, J. D. (1908). The relation of strength of stimulus to rapidity of
habit formation. Journal of Comparative Neural Psychology, 18, 459-482.
Young, R., & O’Shea, T. (1981). Errors in children’s subtraction. Cognitive Science, 5,
153-177.
Yule, W., Sacks, B., & Hersov, L. (1974). Successful flooding treatment of a noise phobia
in an eleven-year-old. Journal of Behavior Therapy and Experimental Psychiatry, 5, 209-211.
Zbrodoff, N. J. (1995). Why is 9 + 7 harder than 2 + 3? Strength and interference as expla-
nations of the problem-size effect. Memory & Cognition, 23, 689-700.
Zimmer-Hart, C. L., & Rescorla, R. A. (1974). Extinction of Pavlovian conditioned inhibi-
tion. Journal of Comparative and Physiological Psychology, 86, 837-845.

466
 Photo Credits

Chapter 1
Page 10: Sovfoto/Eastfoto. Page 13: Courtesy Robert Yerkes Papers, Manuscripts and
Archives, Yale University Library Page 21 (left): Courtesy Pfizer, Inc. Page 21 (right):
Courtesy The B. F. Skinner Foundation. Page 31: From Comparative Psychology; A
Modern Survey by Donald A. Dewsbury, 1973 by McGraw-Hill, Inc.

Chapter 3
Page 93: From R.J. Herrnstein, D.H. Loveland and C. Cable“Natural Concepts in Pigeons,”
Journal of Experimental Psychology: Animal Behavior and Processes, Vol. 2, No. 4 (1976).
Page 99: From”The Form of the Auto-shaped Response with Food or H,0 Reinforcers” by
Jenkins, H.M. and Moore B.R., Journal of the Experimental Analysis of Behavior, 1973, vol. 20,
page 175, fig. 2. ©1973 by the Society for the Experimental Analysis of Behavior, Inc.
Page 110: From “The Hippocampal Formation of the Primate Brain: A Review of Some
Comparative Aspects of Cytoarchitecture and Connections” in Cerebral Cortex, Vol. 6, p.
348, by D. L. Rosene and G. W. Van Hoesen, ©1987 by Plenum Press.

Chapter4
Page 124: Courtesy Yerkes Primate Research Center. Page 145: Thomas D.
Mangelsen/Images of Nature.

Chapter 6
Page 212: Courtesy Psychonomic Society. Page 224: From The Mind of an Ape by David
Premack and Ann James Premack, ©1983 by Ann J. Premack and David Premack.
Reprinted by permission of W. W. Norton & Company, Inc.

Chapter 7
Page 263: Forden/Sygma Photo News. Page 269: AP/Wide World Photos.

Chapter 9
Page 315: From The Mentality of Apes by W. Kohler, 1926 © by Harcourt, Brace & Co,,
Routeledge. Page 336: Courtesy Gabriele Wulf, Max Planck Institute of Psychology
Research.

467
 Photo Credits

Chapter 10
Page 374: From The Mind of an Ape by David Premack and Ann James Premack, ©1983
by Ann J. Premack and David Premack. Reprinted by permission of W. W. Norton &
Company, Inc. Page 375: Courtesy Duane Rumbaugh, Language Research Center,
Georgia State University.

Chapter 11
Page 408: © John Heller.

468
 Author Index

A Ayers, T. J., 158-159

Abarca, N., 143, 144 Azrin, N. H., 126, 127, 128
Abrams, T. W., 46
Ackerman, K. T., 147 B
Adams, J.A., 331 Backer, R., 131
Adams, M. J., 388, 390, 393, 414 Baddeley, A. D., 166, 167, 168, 169, 170, 171, 172,
Adams, N., 251 174, 184, 198, 260, 279, 280, 283
Ainslie, G., 146 Bahrick, H. P., 235, 236
Akins, C. K., 57 Bailey, C. H., 46 ,
Alexander, M. P., 302 Baillargeon, R., 358
Alivisatos, B., 183 Baker, A. G., 107
Alkon, D. L., 46 Baker, D. P,, 414
Allison, J., 135, 136 Balota, D., 207
Amiro, T. W., 53 Balsam, P., 72, 117
Amsel, A., 16, 103 Bangert-Downs, R., 386
Andersen, R. A., 331 Banich, M.T., 38
Anderson, E., 115 Bannerman, D. M., 116
Anderson, J. R., 173, 186, 187, 188, 189, 194, 195, 199, Barbizet, J., 299
200, 206, 210, 211, 216, 217, 219, 225, 229, 232- Barnes, C. A., 193, 194, 233-234
233, 238, 239, 243, 246, 247, 249, 251, 252, 253, Barnet, R. C., 77
255n5, 269, 270, 271, 277, 305, 306, 307, 308, 310, Barnett, T., 157
313, 315, 317, 318, 319, 320, 324, 335, 337, 343, Barrett, A., 302
356, 357, 358, 359, 360, 367, 384, 403, 406, 407, Barsalou, L. W., 275
410, 411, 414 Bartlett, B. J., 395, 397
Anger, D., 104 Bartlett, F. C., 286
Angstadt, P,, 292 Bassock, M., 395
Antrobus, J., 173 Bates, A., 157
Aristotle, 58 Batson, J. D., 60
Arkes, H. R., 294 Baudry, M., 115
Asimoy, I., 309 Baum, W. M., 129
Atkinson, R. C., 11, 24, 27-29, 160, 163, 169, 291, Baumann, J. F., 388, 393
292, 345 Bavaresco, J. L., 280
Atkinson, R. L., 11 Beck, B. B., 314
Atwood, M., 313 Beck, I. L., 388, 393
Austin, G. A., 341 Becker, J.T., 112
Ayers, M. S., 292 Bedford, J., 104

469
 Author Index

Beecroft, R. S., 42 Brigham, J. C., 260

Begg, I., 198n6 Brimer, C. J., 130
Bekerian, D. A., 198 Brinkman, U., 369
Bellezza, F. S., 406 Broadbent, D. E., 27
Bensadoun, P. D., 32, 33 Brogden, W. J., 191, 192
Benson, P,, 104 Brooks, D. J., 331
Berman, A. J., 327 Brooks, L. R., 169
Bernstein, I. D., 62 Brown, A. L., 393, 394, 395, 397, 414
Bernstein, I. L., 62, 77 Brown, J., 164, 240
Best, M. R., 60 Brown, J. S., 400
Bettman, J. R., 150 Brown, P.L., 98
Bever, T. G., 184, 221, 223 ‘“ Brown, R., 262, 290, 367
Bhatt, R. S., 94 Bruce, C. J., 178
Billings, F. J., 260 Bruck, M., 260
Bilodeau, E. A., 334 Bruckmeir, G., 103
Bilodeau, I. M., 334 Bruer, J.T., 414
Bitterman, M. E., 53, 143 Bruner, J. S., 261, 341, 342, 343
Bjork, R. A., 198n6, 199, 273, 280, 333, 335, 337 Buchanan, M., 167
Black, A. H., 130 Bukstel, L., 325
Black, J. B., 288 Bullock, D. H., 143
Blackburn, J. M., 189, 190 Bullock, M., 358
Blaney, P. H., 283 Bunsey, M., 113, 114
Blaxton, T. A., 298 Burke, C. J., 142
Blessing, S. B., 363 Burns, D. J., 198n6, 199
Bliss, T.V. P., 114 Burns, T. E., 256, 257
Block, R. A., 238 Burton, R. R., 400
Bloom, B. S., 324, 386 Buschke, H., 163, 164
Bloom, L. C., 161 Butcher, C., 366
Blough, D. 5., 88 Butcher, S. P,, 116
Bluth, G. J., 395 Butler, R. A., 131
Bobrow, D. G., 199
Boehm, L., 294 C
Boissiere, M., 413 Cable, C., 93
Bolles, R. C., 105 Cain, D. P., 116
Boring, E. G., 38 California Assessment Program, 379
Borson, S., 77 Camp, D. S., 125, 126
Bourne, L. E., Jr., 342 Campbell, J.A., 352, 353
Bovair, S., 322, 363 Campione, J. C., 414
Bower, G. H., 24, 38, 71, 73, 74, 75, 163, 199, 210, Capaldi, E. J., 103
212, 214, 219, 269, 270, 271, 272, 277, 288, 344, Card, S., 144
345, 349, 350, 352 Carew, T. J., 45, 46
Boyd, W., 378 Carey, S., 365
Boyle, C. F., 406 Carpenter, P. A., 388, 389, 392, 403, 414
Bradbury, R., 258 Carpenter, T. P, 402
Bradshaw, G. L., 200, 251 Carraher, D. W., 401
Braine, M. D.S., 365 Carraher, T. N., 401
Brakke, K. E., 373 Casey, J., 141
Brandt, D. M., 395 Cason, H., 15
Brannon, E., 222 Castles, A., 391
Bransford, J. D,, 219, 284, 286, 289, 298 Catalano, J. F., 332
Breland, K., 23, 97, 98 Cavanagh, J. P,, 174, 175
Breland, M., 23, 97, 98 Ceci, S. J., 260
Bremer, D., 281 Cermak, L. S., 225
Briggs, L. J., 414 Chall, J. S., 388, 393

470
 Author Index

Champagne, A. B., 362 Davenport, D. G., 129

Chance, J. E., 212 Davies, S., 255
Chapman, G. B., 108 Dawson, R. G., 258
Charness, N., 302, 325 Deacedo, B. S., 302
Chase, W. G., 209, 325 Deadwyler, 5. A., 117
Chatlosh, D. C., 107 de Beer, G. R., 377
Chelonis, J. J., 147 de Groot, A. D., 324
Chen, M., 46 DeKeyser, R. M., 323
Chen, S., 222 Delamater, B. A., 81
Chi, M.T. H., 395 Delaney, 5. M., 296, 297
Chomsky, N., 23-24, 364, 366, 371, 372 Dempster, F. N., 92
Christen, F., 273 Dennis, M., 370
Christiansen, R. E., 288 DeRivera, J., 89, 90
Christianson, S.-A., 261, 264 Deubel, H., 335
Church, R. M., 125, 126, 127 Dewhurst, D. J., 327
Clahsen, H., 369 Diamond, A., 92, 178
Clark, E.V., 365 Dietz, D., 104
Clark, J., 256 DiLorenzo, M., 115
Clark, M. C., 270 Dodd, P. W., 221
Clement, J., 403 Dodge, M., 366
Clements, K. C., 144 Dodson, J. D., 261
Cognition and Technology Group, 414 Doi, L. M., 391
Cohen, N. J., 262, 302 Dominowski, R. L., 342
Cole, M., 142 Domjan, M., 57, 77
Collins, K. W., 395, 396 Donegon, N. H., 48, 77
Coltheart, M., 156, 157, 391 Dooling, D. J., 287, 288
Colwill, R. M., 81 Dossey, J.A., 379, 381
Compton, B. J., 321 Dragoin, W. B., 61
Conrad, R., 158, 168, 169 Dueck, A., 212
Cooney, T. J., 379, 381 Dunbar, K., 362, 363, 376
Cooper, L. A., 296, 297 Dunn, D. P., 60
Corbett, A. T., 384, 410, 414 Dweck, C., 104
Corbit, J. D., 56
Corkin, S., 302 E
Coslett, H. B., 390 Eagle, M., 273
Cowan, G. N., Jr., 158, 168 Easterbrook, J.A., 261
Cowles, J.T., 83 Ebbesen, E. B., 228, 231
Craik, F. I. M., 161, 162, 164, 198, 225 Ebbinghaus, H., 6, 7-8, 11, 29, 165, 186, 227, 228, 229
Crocker, J., 355 Eckerman, D. A., 139
Crooks, G. B., Jr., 115 Egan, D. E., 325
Crossman, E. R. F. W., 190, 307, 309 Egan, J. C., 158-159
Crosswhite, F. J., 379, 381 Ehri, L. C., 391, 393
Crothers, E. J., 24 Eich, E., 280, 282
Crowder, R. G., 157, 158, 159, 168, 184, 240 Eich, J. E., 281
Crowell, C., 136 Eichenbaum, H., 111, 113, 114, 117, 302
Culler, E., 191, 192, 193 Einhorn, H. J., 355
Einstein, G. O., 198n6, 214
D Eisenberger, R., 103
Dallenbach, K. M., 254 Ekstrand, B. R., 255, 342
Dallett, K. M., 271 Elek, S. M., 107
Dansereau, D. F., 395, 396 Ellis, N. C., 168
Darwin, C., 3, 6 Engle, R. W., 325
Darwin, C. J., 157, 158, 168 Epstein, 5., 56
Daugherty, K. G., 369 Erickson, M. A., 354

471
 Author Index

Ericsson, K. A., 324 Gardner, R. A., 223, 373

Eron, L. D., 128 Garland, J. C., 395, 396
Ervin-Tripp, S. M., 370 Garrud, P,, 111
Essman, W. B., 125 Gates, S., 386
Estes, W. K., 142, 253, 352, 353 Gathercole, S. E., 367
Etscorn, F., 61 Gazzaniga, M. S., 38
Ettinger, R. H., 117, 134, 138, 151 Gelman, R., 358
Evans, A. C., 183 Gerber, D., 290
Evans-Rhodes, D., 407 Gernsbacher, M. A., 217
Eysenck, M. W., 258 Gibbon, J., 72
Gillin, J. C., 281
F ‘“\ Gillund, G., 29, 206, 243, 253
Fantino, E., 143, 144 Gilmartin, K., 324
Farnham-Diggory, S., 414 Girden, E., 191, 192, 193
Fay, A. L., 376 Givens, J. S., 403
Fernandez, A., 280 Glaser, R., 386, 395
Ferster, C. S., 20, 137 Glass, A. L., 173
Fiez, J.A., 321 Glenberg, A., 160, 161, 235, 236, 280
Fincham, J. M., Schneider, W. & Anderson, J. R., 322 Glickman, S. E., 134
Fischer, G. W., 358 Gluck, M. A., 71, 72, 73, 74, 75, 113, 114, 117, 205,
Fitts, P. M., 310 349, 350, 352
Fitzgerald, R. D., 50 Glucksberg, S., 158, 168, 347
Flexser, A. J., 303 Goddard, R., 198n6
Foley, F., 198n6 Godden, D. R., 279, 280
Folkard, S., 256, 258, 259 Goldin-Meadow, S., 366, 371
Foree, D. D., 86 Goldman-Rakic, P. S., 178, 179, 180, 184
Fort, J. G., 142 Goldstein, A. G., 212
Fox, C., 262 Goldstein, D., 294
Fox, P.T., 321 Good, H. G., 378
Frackowiak, R. S. J., 331 Good, M. A., 116
Franklin, S. R., 60 Goodnow, J. J., 341
Franks, J.J., 219, 284 Goodwin, D. W., 281
Frase, L. T., 203 Gordon, A. M., 330
Fraser, C., 367 Gordon, J. S., 292
Fraser, J., 248 Gormezano, I., 40, 43, 47, 64
Frederiksen, J. R., 391, 392 Graf, P, 198, 199, 200, 282, 301
Freedman, J. L., 290 Graf,V., 143
Freud, S., 256-257, 259, 260 Grahame, N. J., 77
Freund, J. S., 278 Grant, D. A., 41, 42, 43
Friedman, M. P,, 142 Grant, D. S., 98, 175, 176, 232
Friedman, W. J., 358 Grant, S., 170
Frost, J., 393 Gray, M. M., 255
Funahashi, S., 178 Green, C., 160
Green, L., 145, 146, 147
G Greeno, J. G., 313, 314n2
Gagné, E,, 414 Griffin, R. W., 53
Gagné, R., 384, 385, 387, 414 Grittner, F. M., 386
Gallagher, M., 115 Groves, P. M., 42
Gallistel, C. R., 142 Gruneberg, M. M., 290
Gamoran, A., 379 Guitierez, G., 57
Gantt, W. H., 50 Gunn, D. L., 314
Garcia, J., 61 Gunstone, R. F,, 362
Gardner, B. T., 223, 373 Guskey, T. R., 386
Gardner, M., 397 Guthrie, E., 19

472
 Author Index

Guttman, N., 84 Hoffman, D., 403

Gynther, M. D., 54 Hogarth, R. M., 355
Hogg, T., 189
H Hoine, H., 281
Haber, R. N., 157 Holland, J. H., 376
Hackett, C., 294 Holland, P. C., 51, 52, 53
Hadley, W. H., 411 Hollerbach, J. M., 337
Hager, L. D., 92 Holley, C. D., 395, 396
Haig, K. A., 96, 97 Holyoak, K. J., 109, 376
Hake, D. F., 126 Holz, W. C., 126, 127, 128
Halal, M., 283 Honig, W. K., 175
Hall, G., 71 Honzik, C. H., 18
Hamdi, M., 103 Hoosain, R., 168
Hamilton, B., 141 Houston, J. P,, 240
Hammond, K., 391 Howard, D. A., 158-159
Hammond, L. J., 100, 101, 102 Howard, J. L., 56
Handelmann, G. E., 112 Huberman, B. A., 189
Hardy, J. K., 208, 209 Hull, C. L., 16-17, 24, 25, 88, 118, 340, 341, 342, 350
Harley, W. F,, Jr., 121 Hume, D., 357
Harrison, R. H., 84, 85, 86, 87, 89 Hunter, H. G., 41
Hart, J.T., 290 Hunter, J. E., 413
Hasher, L., 258, 294 Hunter, R. F., 413
Hatsopoulos, N., 352, 353 Hunter, W. S., 191
Haverty, L., 400 Hurwitz, J. B., 352, 353
Hawkins, R. D., 45, 46, 77, 115 Husband, T. H., 260
Hayes, C., 373 Hyams, N. M., 372
Hayes, J. R., 324, 392, 403 Hyde, T. S., 201
Hayes-Roth, F., 325 Hyman, I. E., 260
Hearst, E., 54, 55, 60
Heath, R. G., 134 I
Heathcote, A., 197n4 Inhoff, A. W,, 330
Heerdt, W. A., 103 Ivry, R.B., 38
Hegvik, D. K., 176, 177
Hemmes, N. S., 139 J
Hennelly, R. A., 168 Jackson, M. D., 389, 391
Heron, C., 92 Jackson, R. L., 103
Herrnstein, R. J., 93, 140, 146, 151 Jacobsen, C. F., 178
Hersoy, L., 129 Jacoby, L. L., 292, 293, 294, 295, 296, 298
Heyman, G. M., 140n4 James, J. H., 391
Hilgard, E. R., 11, 38 Jarvik, M. E., 125
Hillman, B., 191 Jaswal, V.,222
Hineline, P. N., 104 Jeffries, R. P,, 313
Hinsley, D. A., 392, 403 Jenkins, H. M., 84, 85, 86, 87, 89, 98, 355-356
Hintzman, D. L., 238 Jenkins, I. H., 331, 332, 335
Hiraki, K., 292 Jenkins, J. G., 254
Hiroto, D. S., 104 Jenkins, J. J., 201
Hirshman, E., 198, 199, 214 Jessell, T. M., 32, 38
Hirst, W., 302, 326 Job, R. F. S., 104
Hirtle, S. C., 208, 209, 325 Johansson, R. S., 329
Hitch, G. J., 367 Johnson, D. D., 388, 393
Ho, L., 334 Johnson, E. E., 261
Hockey, G. R. J., 255 Johnson, E. J., 150
Hodges, J. R., 91 Johnson, J. C., 130
Hoffer, A., 405 Johnson, M. K., 286, 302

473
 Author Index

Johnson, M. R., 136 Krebs, J. R., 143

Johnson, N. F., 208 Krueger, W. C. F., 229
Johnson, N. S., 213 Krushke, J. K., 354
Jones, W. P., 173 Kulik, C., 386
Jonides, J., 158-159, 181, 182, 354 Kulik, J., 262, 386
Juola, J. F.,, 291, 292 Kushmerick, N., 317
Just, M. A., 388, 389, 392, 403, 414 ,
L
K Laird, J. D., 283
Kaiser, M. K., 361, 362 ; Landauer, T. K., 253, 290
Kalish, D., 18, 19 Landfield, P. W., 117
Kalish, H. I., 84 ia: Langley, P, 337
Kamil, A. C., 97, 144, 148 Lashley, K. S., 95
Kamin, L. J., 63, 64, 68, 69, 129, 130 Lave, J., 378, 412, 413
Kandel, E. R., 32, 38, 45, 46, 77, 115 Lavond, D. G., 48, 77
Kaplan, C. A., 207 Lawrence, D. H., 89, 90
Kaplan, S., 257, 258 Lawrence, K. A., 273
Karlin, M. B., 212 Lea, S. E. G., 147
Katsaiti, L. T., 198n6 Leaf, R.'C.,50
Katz, B., 34 Leahey, T. H., 38
Katz, L., 142 F Lebiere, C., 317, 337
Keele, S. W., 330, 331 Lee, S., 382
Keeton, W.T., 32, 33 Lefkowitz, M. M., 128
Kehoe, E. J., 47, 64 Leichtman, M. D., 260
Keith, J. R., 115 Leinbach, J., 369
Keller, F. S., 386 Leiter, E., 273
Keller, L., 142 Lenat, D. B., 325
Kelley, C., 293 Lenneberg, E. H., 370, 373
Kellogg, L. A., 373 Lesgold, A. M., 391, 392
Kellogg, W. N., 373 Levine, M., 376
Kennelly, K. J., 104 Levinger, G., 256
Keppel, G., 165, 237, 240, 254, 255 Levonian, E., 257
Kiedlinger, R. E., 94 Lewinsohn, J., 256
Kieras, D. E., 322, 363 Lewis, C. H., 189, 249
Kifer, E., 379, 381 Lewis, M., 395
Kimble, G. A., 191 Lewis, V. J., 167, 168
King, A. P, 366 Lieberman, P., 371
King, E. J., 378 Light, J. S.,50
Kintsch, W., 163, 164, 218-219, 225, 271, 273 Little, L., 91
Klahr, D., 337, 362, 363, 376 Lochhead, J., 403
Klapp, S.T., 321 Lockhart, R. S., 161, 162, 164, 198
Klein, S. B., 151 Loftus, E. F., 256, 257, 260, 261
Kleiner, B. M., 332 Loftus, G. R., 121, 122, 123, 240n1, 261
Kleinsmith, L. J., 257, 258 Logan, F., 16
Kopfer, L. E., 362 Logan, G. D., 189, 320, 321
Knight, J. B., 413 Logue, A. W., 147
Koedinger, K., 384, 411, 414 LoLordo,V. M., 86
Koelling, R. A. 61 Lombardi, L., 163
Koh, K., 109, 333 Lomo, T,, 114
Kohler, W., 89, 314, 315 Lorch, R., 207
Konarski, E. A., Jr., 136 Loveland, D. H., 93
Kosslyn, S. M., 337 Lovett, M., 142
Kral, P. A., 61 Lowry, D. H., 214, 215
Krampe, R. T., 324 Luce, R. D., 140n4

474
 Author Index

Lundberg, I., 388, 393 Mead, A., 96, 97

Lunew, M. B., 221 Medin, D. L., 348, 350-352, 353
Lynch, G. S., 115 Melton, A. W., 166, 185
Lynch, M. A., 114 Meltzer, H., 256
Messo, J., 261
M Metcalfe, J., 282
MacCorquodale, K. O., 24 Mewhort, D. J. K., 197n4
MacDonald, M. C., 369 Meyer, B. J. F., 395
Macfarlane, D. A., 95 Meyer, D. E., 207, 333
MacKay, D. G., 189 Meyer, E., 183
Mackintosh, N. J., 71, 91, 143 Michotte, A., 361
MacLaren, D., 81 Milberg, W., 302
MacLeod, A. K., 321 Miller, G. A., 207
MacLeod, C. M., 296 Miller, N. E., 16, 126, 129
MacLeod, L., 283 Miller, R., 72, 77
MacPhail, E. M., 104 Millward, R. B., 142
MacWhinney, B., 365, 369 Milner, P,, 134
Maier, S. F., 103 Miyashita, Y., 204
Maki, W. S., 175, 176, 177 Monk, T. H., 258
Mandler, G., 201, 273, 275, 301 Monks, J., 290
Mandler, J. M., 213 Moore, B. R., 98
Mangun, G. R., 38 Moore, J. W., 40
Manis, F. R., 391 Moray, N., 157
Marcus, G. F., 369 Morris, C. D., 284, 285
Mark, M., 411 Morris, L., 283
Marler, P., 366 Morris, R. G. M., 110, 111, 112, 113, 115, 116
Marr, D., 205 Morton, J., 158
Marshall, B. S., 47 Moser, J. M., 402
Martin, G. K., 50 Mowrer, O. H., 16, 129
Massaro, D. W., 278n4, 303 Mowrer, R. R., 151
Masson, M. E. J., 296 Mudd, S. A., 161
Matter, J., 261 Muenzinger, K. F., 94, 95
Matzel, L. D., 72 Murdock, B. B., Jr., 165, 276
May, J. G., III, 147 Murphy, J., 373
Mayer, R. E., 403, 404, 414 Myers, C. E., 75, 113, 114, 117, 205
McAllister, W. R., 41 Myers, J. L., 142
McBride-Chang, C., 391 Mylander, C., 366
McClelland, J. L., 25, 205, 368, 369, 389, 391
McCloskey, M., 262, 347, 362 N
McConkie, G. W., 389 Nadel, L., 112, 113
McDaniel, M. A., 198n6, 214 Narens, L., 290
McDonald, B. A., 395, 396 Nash, S. M., 60
McElree, B., 231 Neath, I., 159
McGaugh, J. L., 258 Neches, R., 337
McGeoch, J.A., 254 Neely, J. H., 207
McGlinchey-Berroth, R., 302 Neisser, U., 156, 158, 161, 197, 262, 264, 286, 326
McGonigle, B., 222 Nelson, T. O., 162, 202, 263, 265, 290
McKeithen, K. B., 325 Neves, D. M., 307, 308, 316, 317, 319
McKinley, S. C., 348 Newell, A., 19, 24-26, 124, 189, 195, 196, 225, 311,
McKnight, C. C., 379, 380, 381, 382 022)325,387
McNamara, T. P., 208, 209 Newport, E. L., 367, 371
McNaughton, B. L., 205 Nhouyvanisvong, A., 292
McNeill, D., 290, 365 Nisbett, R. E., 109, 376
Meachum, C. L., 60 Nixon, P. D., 331

475
 Author Index

Norman, D. A., 165, 305 Peterson, L. R., 164, 240

Norris, E. B., 42 Peterson, M., 164, 240
Norvig, P., 337 Peterson, M. A., 296, 297
Nosofsky, R. M., 348, 350 Peterson, S. B., 249, 250
Nottebohm, F., 370 Petrides, M., 183
Noyd, D. E., 240 Pettito,L.A., 223
Phelps, E. A., 302
O Pickrell, J., 260
O’Brien, H. J., 50 Pinker, S., 366, 369, 372, 376
Obrist, P. A., 56 Pirolli, P. L., 144, 187, 188, 249
Ohlsson, S., 309 Polkey, C. E., 91
O’*Keefe, J.A., 112, 113 “ Polson, M., 407
O'Keefe, W., 111 Polson, P. G., 313, 322
Oldfield, R. C., 42 Porter, D., 385
Olds, J., 134 Posner, M., 321
Olson, R. D., 129 Postman, L., 202, 241, 248, 254, 264
Olton, D. S., 95, 96, 97, 112, 113 Potter, M. C., 163
O'Neil, E., 54, 55 Potts, G. R., 249, 250
O'Reilly, R. C., 205 Powell, B., 281
Orlovskii, G. N., 329 Pratt, Cis 358,
O'Shea, T., 400 Premack, A. J., 223, 225, 373, 374
Osherson, D. N., 337 Premack, D., 132-133, 134, 223, 225, 373, 374
Osler, S., 271 Prince, A., 369
Otaka, S. R., 280 Proffitt, D. R., 361, 362
Otto, T., 113 Psotka, J., 267
Owen, A. M., 91
Owens, J., 288, 289 R
Raaijmakers, J. G. W., 206, 243
P Rabinowitz, J. C., 275
Paivio, A., 210 Rabinowitz, M. & Goldberg, N., 321
Palinscar, A. S., 393, 394, 395, 397 Rachlin, H., 104, 145, 146, 147
Palmer, S., 262 Raibert, M. H., 329, 330
Palmeri, T. J., 348 Raichle, M. E., 321
Palu, M., 214 Ramsay, M., 116
Papanek, M. L., 261 Rashotte, M. E., 53
Pardo, J.V., 321 Ravitch, D., 414
Parkin, A. J., 256 Rawlins, J. N. P, 97, 111
Passingham, R. E., 331 Raymond, G. A., 125
Patalano, A. L., 354 Rayner, K., 389
Patel, A. S., 41 Razran, G., 91
Patterson, K. K., 240n1 Razran, L., 313
Paulson, R., 229 Reder, L., 199, 290n6, 291, 292, 294, 303
Pavloy, I. P,, 6, 9-12, 13, 15, 17, 40, 44, 60, 78 Reed, S. K., 348
Payne, J.W., 150 Reiss, S., 70
Pearce, J. M., 71, 93 Reitman, J., 158-159, 272, 325
Pearlstone, Z., 271 Rescorla, R. A., 44, 50, 51, 52, 56, 58, 59, 60, 65-75,
Pelletier, R., 384, 414 77, 81, 91, 107-108, 114, 116, 178, 187, 196n3,
Perfetti, C. A., 388, 392 245-246, 349, 350, 356, 357, 368
Peters, D. P,, 256, 257 Resnick, D. P., 378
Peters, S., 366 Resnick, L., 378, 386, 391, 402
Petersen, A. S., 369 Restle, F., 19
Petersen, O. P., 393 Richardson, J., 407
Petersen, S. E., 321 Riemann, P., 395
Peterson, A., 391 Ritchey, G. H., 213

476
 Author Index

Ritchie, B. F., 18, 19 Schneider, W., 73

Rizley, R.C., 51 Schneiderman, B., 56, 325
Robbins, S. J., 108 Schofield, J. W., 407
Robbins, T. W., 91 Schonberg, H. C., 324
Roberts, A. C., 91 Schooler, L. J., 194, 195, 225, 232-233, 238, 239, 335
Roberts, R. J., 92 Schorr, D., 251
Roberts, W. A., 176, 221 Schreiber, G., 262
Robinson, G. S., Jr. 115 Schunn, C. D., 292
Robinson, H. A., 203 Schustack, M. W., 356
Roediger, H. L., 298, 303 Schvaneveldt, R. W., 207
Rohlen, T., 382 Schwartz, B., 104
Roitblat, H. L., 184, 225 Schwartz, B. J., 325
Romberg, T. A., 402 Schwartz, J. H., 32, 38
Rosch, E., 346, 347 Secretary's Commission on Achieving Necessary
Rosenbaum, D. A., 330, 337 Skills, 412
Rosenbloom, P. S., 189, 195, 196, 225 Seidenberg, M. S., 221, 369, 391
Rosene, P. L. & Van Hoesen, G. R., 110 Seligman, M. E. P,, 103, 104, 130
Rosenthal, J., 258 Severin, F, V., 329
Ross, J., 273 Sevik, R. A., 373
Ross, L. E., 43 Shaklee, H., 356
Rubin, D. C., 228, 264 Shallice, T., 305
Rubinsky, H. J., 139 Shapiro, D. C., 332, 334
Rudy, J. W., 113, 115, 176 Sharrock, R., 283
Rueter, H. H., 325 Shea, J. B., 334
Ruiz, D., 318 Sheffield, F. D., 131
Rulf, S., 358 Shepard, L. A., 385
Rumbaugh, D. M., 373 Shephard, R. N., 211
Rumelhart, D. E., 25, 209, 368, 369 Shettleworth, S. J., 105
Rundus, D. J., 28, 29, 160, 161 Sheu, C. F., 356, 357
Russell, M. L., 283 Shiffrin, R. M., 27-29, 160, 163, 169, 206, 243, 253
Russell, S., 337 Shik, M. L., 329
Shimp, C. P, 141, 222
S Shrager, J. C., 189, 362, 363, 399, 400
Sabot, R. H., 413 Shultz, T. R., 358, 376
Sachs, J. S., 215 Sidman, M., 130
Sacks, B., 129 Siegel, S., 54, 55, 56
St. Claire-Smith, R., 81 Siegler, R. S., 399, 400
Sakai, K., 204 Silberberg, A., 141
Sakitt, B., 156 Simmelhag,V.L., 102
Salili, F,, 168 Simon, H. A., 19, 24-26, 124, 149, 209, 225, 311, 324,
Salmoni, A. W., 334 325, 337, 392, 403
Salthouse, T. A., 327 Singley, M. K., 305, 306, 307, 337, 403, 405
Saltzman, I. J., 83 Sisk) Galanoc
Samuelson, R. J., 96 Skinner, B. F., 20-24, 25, 26, 79, 81, 83, 95, 97, 101,
Sanders, R. J., 223 137, 153, 383, 384
Santa, J. L., 210, 211 Slamecka, N. J., 198, 199, 200, 231, 282
Saucier, D., 116 Slater-Hammel, A. T., 327
Saufley, W. H., 280 Smith, E. E., 181, 182, 251, 354, 376
Savage-Rumbaugh, E. S., 373 Smith, S. M., 160, 280
Schacter, D. L., 296, 297, 298, 303 Snider, A., 198n6
Schaffer, M. M., 348, 350-352, 353 Solomon, R. L., 56, 129
Schiff, B. B., 134 Soloway, E., 403
Schliemann, A. D., 401 Sonuga-Barke, E. J. S., 147
Schmidt, R. A., 327, 329, 330, 331, 333, 334, 335, 337 Spelke, E., 326

477
 Author Index

Spence, K, W., 16, 88-89 Thorndike, E. L., 6, 12-16, 17, 18, 26, 78, 80, 118, 125,
Sperling, G. A., 156, 157, 158 209, 383, 384
Sporer, S. L., 161 Timberlake, W., 98, 99, 135, 136
Springston, F., 163 Tinklepaugh, O. L., 81
Squire, L. R., 229, 230, 301, 303 Tobin, H., 147
Staddon, J. E. R., 83, 102, 117, 134, 138, 141, 151 Toigo, R., 128
Standing, L., 211 Tolman, E. C., 15, 17-20, 26, 82, 95, 118, 121, 124, 305
Stark, K., 248 Tomie, A., 103
Staubli, U., 115 Toppino, T., 294
Steele, D. R., 259 Toth, J. P, 294
Stein, L., 134 Trabasso, T. R., 344, 345
Steinberg, H., 258 ~ Travers, K. J., 379, 381
Stephens, D. W., 143 Trowbridge, M. H., 15
Stephens, R., 61 Tucker, D., 356
Stern, J., 281 Tulving, E., 161, 162, 267, 271, 274, 275, 283, 291, 303
Sternberg, R. J., 356 Turvey, M.T., 157, 168
Sternberg, S., 172, 173, 174 Tversky, A., 149
Stevenson, H. W., 379, 380, 382, 383, 414
Stewart, C., 113 U
Sticht, T. G., 389, 391 Ulrich, R. E., 128
Stigler, J.W., 379, 380, 382, 383, 414 Underwood, B. J., 165, 240, 278
Stillman, R. C., 281
Stinkman, P. G., 115 Vv
Straub, R, O., 221 Vaccarino, F. J., 134
Sulin, R. A., 287 Vallar, G., 168
Summerfield, A., 258 van Dijk, T. A., 218
Summers, B. A., 91 Van Lehn, K., 337, 400
Summers, J. J., 238 Vidden, T. O., 321
Suppes, P, 386 Volpe, B.T., 302
Surprenant, A. M., 159
Sutherland, N. S., 143 Ww
Sutherland, R. J., 113 Waddill, P. J., 198n6
Sutterer, J. R., 56 Wager, W. W., 414
Suydam, M. M., 142 Wagner, A., 16, 44, 50, 56, 64, 65-75, 70, 77, 91, 107-
Swafford, J. O., 379, 381 108, 114, 116, 176, 177, 178, 187, 196n3, 245-246,
Swanson, L. W., 114 349, 350, 356, 357, 368
Swinnen, S., 334 Wagner, J. J., 283
Szegda, M., 283 Walder, L. O., 128
Waldrop, P. B., 22
T Walter, C. B., 334
Taub, E., 327 Ward, W. C., 355-356
Taylor, B., 96, 97 Wasserman, E. A., 60, 77, 94, 106, 107, 108, 221,
Teasdale, J. D., 283 356n3, 376
Terrace, H. S., 85, 184, 221, 222, 223, 373 Waterman, D. A., 325
Tesch-Romer, C,, 324 Watkins, M. J., 161, 240n1
Teyler, T. J., 114 Watkins, O. C., 240n1, 274
Thagard, P. R., 376 Watson, J., 5, 15, 16
Tharan, M., 296, 297 Watts, F. N., 283
Thibault, O,, 115 Waugh, N. C., 165
Thomas, E. L., 203 Weber, A., 214, 215
Thompson, R. F., 42, 47, 48, 56n2, 77, 114 Webley, P., 147
Thomson, D, M., 274 Webster, M. M., 62
Thomson, N., 167, 170 Weingartner, H., 281

478
 Author Index

Weisberg, P.,, 22 Wixted, J.T., 228, 231

Weisman, R. G., 132, 221 Woest, A., 369
Wenger, E., 407 Wolfe, J. B., 83
Wenzel, A. E., 228, 264 Wollen, K. A., 214, 215
Wertheimer, M., 38 Woloshyn, V., 293
Wertheimer, R., 411 Wulf, G., 335
West, M. J., 366 Wulff, J.J., 131
Westbury, I., 414 Wynne, L. C., 129
Westling, G., 329
Whelley, M. M., 214 Y
White, J. L., 334 Yekovich, C. W., 414
White, M., 382, 383 Yekovich, F. R., 414
Whitlow, J. W., 176 Yerkes, R. M., 261
Whitman, T. L., 136 Yoerg, S. I., 144
Wible, C. G., 262 Yonelinas, A., 294
Wickelgren, W. A., 164, 165, 166, 231, 241, 255, 264 Young, D. E., 334
Wiese, R., 369 Young, R., 400
Wight, E., 170 Yule, W., 129
Wilce, L. C., 391, 393
Wilcoxon, H. C., 61 Z
Williams, B., 151 Zavortnik, B., 254
Williams, J. P., 388 Zbrodoff, N. J., 321
Williams, S. L., 373 Zimet, S., 103
Winstein, C. J., 334 Zimmer-Hart, C. L., 60
Wiseman, S., 274, 275 Ziriax, J. M., 141

479
 - tt si wale
~ a ort2 3
a, ee
(
: OF, TF ae
‘i a L1H 2 Siow a 2 ‘"

A ee 1 sme
eemuy *
iy ~ oe PRLe ae
b bab AD abr 7 apt 6, ae aS
ship 20i3
thoy i eo b.. V™_ 2

} A NP eet “ho
five } af 1S geee

pciecaerae
*, é ’ } rad 7.) yattaur? iegt. wi ee TLE
lie Ti i. Du ’ £2.16 grow en LET AN atAW
sheep fy iy The The Ob Syrian
bhewmmnaain, 1 AP PIS. “, SER DRA ais
eos C afi3 - u ay ard
anent, 4 Let 5 ovionantell
Seiden, |W TN, WA, 2 339, aye a MS
“1p prdatiy hf on TA ree BrAdostS
pee Fa. | : £Q) .® penis
pata FA. am, 02 t Dea’
Pdi, RA 38) . > fei ph
Conewien A, Se lk
daoters WA, i = e
senna, |ha :
feakiae By:
Mae rant 4a sy i
‘Ratinlatel, hy Fae F es
bitter’ Ko 3 S
Satins, ’ x. an P
—_ hana 4, Me ia): ‘ ee
nh iw z 7? «iy Tk ASSn rea
Serarcees.1.W. 1 le 9, 6 OT OR si
i <a 7
vagrviny, 3 ls = Wages. |. eH
28 i
Ses 1D Wile 1 Ue ten e
vo
vere
if aan 7 er
Wabtiigs Edis
Wola (Sh
ae
GeibeR, 33%, of tak © oti
Tayler, fh,OA REY? hg me. 4" ane: ee
Tesh’ 0. 2 ee : ial Ye i
- Tasrand, 62 aL be SIL Oe a F thieving, LAS as
ir ior Lt =-70 5 ‘. Pothics BL an
p> be a it ’ © a¢
Thapant # ais
Thi. Uo, De HRS no
ry aul, oO INS
em 7 J
LM oy - 4 we ' Pea ~m

ine
Ld *
he,
’ 4
37. 48,
j
BERS oh aM
Sy ‘Se és

tty Ta) Om ; =
Tt. «e<.'6

be * 7

7%
 Subject Index

A retrieval, 320-322, 331-332

Acquisition Attentional learning, 90-92
in conditioning, 10-11, 14, 40-41, 65-70 Auditory sensory memory, 157-159
of memories, 185-225 Autonomous stage of skill acquisition, 310, 325-336
See also Conditioning curves, Elaborateness of See also Motor programs
processing, Practice, Representation of Autoshaping, 98-99, 102
knowledge Axoaxonic synapse, 45
ACT theory, 206, 243 Axon, 33-35
Action potential, 34-35
Activation, 173-175 B
Activation equation, 244 Backwards learning curve, 344-346
Adaptation to environments, 1-2, 194-195, 232-233, Behavioral objectives in education, 383-384
238-239 Behaviorism, 3-4, 5-6, 15-16
See also Biological predispositions, Hull’s theory, 16-17
Rationality Skinner's theory, 20-24
All-or-none learning, 344-346 Biological predispositions, 84-86, 97-99, 152-153
Amnesia, 298-302 See also Associative bias
Animal research versus human research, 3-4, 30, 37- Bliss points, 134-136, 314
38, 106-108, 153-155 Blocking, 63-64, 64-65, 68-70
Anterograde amnesia, 299-300
Aphasia, 370 Cc
Aplysia, conditioning in, 45-47 Category learning
Applications in pigeons, 93-94
psychotherapy, 23, 104 See also Concept acquisition
See also Education Causal inference, 106-109, 354-364
Arousal and memory, 257-262 complex devices, 362-364
Artificial intelligence, 24, 311-312 contiguity, 357-361
intelligent tutoring systems, 406-411 contingency, 355-357
Association equation, 244 kinematic cues, 361-362
Associative bias, 61-62, 71, 86, 104-106 prior models, 371-376,363-364
See also Biological predispositions Rescorla-Wagner theory, 107-108, 356-357
Associative stage of skill acquisition, 310, 319-325 See also Classical conditioning, Instrumental
expertise, 324-325 conditioning
loss of verbalization, 319-320 Central executive, 171-172, 304-305, 311, 321
problem memory, 324-325 Cerebellum, 30, 32, 328-329
production rules, 322-323 Cerebral cortex, 30-32

481
 Subject Index

Chess learning, 324-325 schema theories, 348-350

Choice behavior, 137-150 See also Category learning
foraging, 143-145 Conditioned emotional response (CER), 10, 51, 59,
human decision making, 148-150 61, 63-64
matching law, 139-143 Conditioned inhibition, 60-61, 69-70
mechanisms, 148-150 Conditioned reinforcer, see Secondary reinforcement
Chunk, 207-210 Conditioned response (CR), 9, 54-58
Classical conditioning, 39-77 Conditioned stimulus (CS), 9, 53-55
acquisition, 10-11, 40-41, 65-70 Conditioning
Aplysia, 45-47 Hullian theory, 161-7
associative bias, 61-62, 71 versus memory experiment, 152-153
autoshaping, 98-99 om See also Causal inference, Classical conditioning,
awareness, 42-43 Instrumental conditioning
blocking, 63-64, 64-65, 68-70 Conditioning curves, 10-11, 65-70
conditioned inhibition, 60-61, 69-70 versus learning curves, 11, 191-193
conditioned response, 54-58 See also Acquisition
conditioned stimulus, 53-55 Configural cues, 64-65, 71
configural cues, 64-65, 71 hippocampus, 75, 113-114
contingency, 58-62 past tense, 368-370
devaluation paradigm, 51 See also Delta rule
extinction, 11, 40-41 Connectionism, 72-75
eye blink, 39-43, 47-49 Context
instrumental conditioning comparison, 79-80 cue, 252-253, 279-289
latent inhibition, 70-71, 177-178 interference, 252-253 ‘
neural basis, 44-49 Context-dependent memory, 279-285
Pavlov’s research, 9-12 drugs, 280-281
Rescorla-Wagner theory, 65-75 encoding specificity, 284-285
response prevention paradigm, 50 mood, 282-284
second-order conditioning, 52-53, 82 Contingency, 58-62, 99-106, 355-357
sensory preconditioning paradigm, 51-52, 82 Contiguity, 12, 41, 47-48, 57-58, 61, 99-106, 125-126,
spontaneous recovery, 11 145-147, 357-361
stimulus combinations, 63-65, 66-72 Critical period for learning, 370-371
S-S versus S-R associations, 49-53 Cue retrieval hypothesis, 226, 253, 267
temporal relation of CS and US, 12, 41, 47-48, See also Context-dependent memory
57-58,61 Cumulative response record, 22
Closed-loop performance, 326-328
Cognitive map, 18-19 D
See also Maze learning Decay hypothesis, 228-239
Cognitive stage of skill acquisition, 310-318 versus interference, 254-255
See also Problem solving Deductive inference, 339
Cognitivism, 3-4 Delayed match-to-sample task
Competitive learning, 66-70 pigeons, 175-176
Componential analysis, 387, 411 primates, 178-180
intelligent tutoring systems, 407-411 Delta rule, 72-75, 349-350, 368-370
mathematics, 399-411 See also Rescorla-Wagner theory
reading, 389-390 Dendrite, 33
See also Task analysis for education Depth of processing, 161-164
Computer simulation, 25-27 See also Elaborateness of processing
Concept acquisition, 340-354 Devaluation paradigm, 51
concept-identification studies, 341-346 Difference reduction, 26, 312-314
exemplar theories, 348-349, 350-353 Digit span, 168
Hull’s research, 340-341 Discounting the future, 145-147
hypothesis testing, 341-346 Discovery learning, see Conditioning, Induction
natural concepts, 346-349 Discrimination learning, 54-55, 86-92

482
 Subject Index

Spence’s theory, 88-89 in the human, 39-43, 54

d’ measure, 277-278 in the rabbit, 47-49, 56, 64
Drive Eyewitness memory, 260-261
drive reduction theory, 130-131, 134
Hullian theory, 16-17 id
Skinner's opinion, 20 False alarm, 276
Dual-code theory, 210-211 False memory syndrome, 259-260
Dual-process theory of avoidance learning, 129-130 Familiarity, 277, 291-294
Dyslexia, 388, 390-391 Feeling of knowing, 290-291
Fixed interval reinforcement (FI), 21-23, 137-138
E Fixed ratio reinforcement (FR), 137-138
Echoic memory, 157-159 Flashbulb memories, 262
See also Phonological loop Foraging, 96-97, 143-145
Economic view of choice, 136, 145-147, 150 Forgetting, see Retention
Education, 154-155, 202-203, 377-414 Feedback, 333-336
behaviorist program, 383-387 See also Reinforcement
class structure, 378, 411-413 Free recall paradigm, 28-29
cognitive approach, 387 generate-recognize theory, 270-275
goals, 377-378 retrieval strategies, 270-273
history, 377-378 versus recognition, 266-275
intelligent tutoring systems, 407-411 Frontal cortex, 31-32, 331-332, 354
international comparisons, 378-383 anterior cingulate gyrus, 321-322
mastery learning, 384-387, 411 reversal shifts, 91-92
Skinner, 383-385 tool building, 314-315
study skills, 202-203, 393-395 working memory, 178-183, 204
Thorndike, 14-15, 383-385 Functional Magnetic Resonance Imagery (fMRI)
See also Mathematics learning, Reading learning, 180-181
Skill acquisition
Elaborateness of processing, 197-203 G
generation effect, 198-200 General problem solver (GPS), 25-26, 124
interactive images, 214-215 Generate-recognize theory of free recall, 271-275
interference, 251-252 Generation effect, 198-200
reconstructive retrieval, 285-289 Generalization gradients, 53-55, 84-86
versus strength, 200-201 Gestalt psychology, 15, 89
test interactions, 279-289, 295-298 Global maximization, 140-141
See also Depth of processing Goals and learning, 17-20
Elimination by aspects, 149-150 See also Means-ends analysis, Problem solving
Encoding specificity principle, 288-289
Environment, 1-2, 194-195, 232-233, 238-239 H
See also Education Habituation, 41-42, 51
Evolution and learning, 1-2, 6 Hippocampus, 31
Excitatory neurotransmitters, 34 amnesia, 298-302
Excitatory postsynaptic potential (EPSP), 192-194, conditioning, 75, 109-116
233-234 connections to frontal cortex, 179
Exemplar theories of concept learning, 348-349, 350- long-term potentiation, 114-116, 192-194, 233-
353 234
Expected value, 119-120 memory, 203-205, 331-332
See also Subjective value parahippocampus, 109-111, 114
Expertise, 324-325 Hypothalamus, 30, 134
Explicit memories, see Implicit versus explicit memo- Hypothesis testing, 341-346
ty, Declarative knowledge
Exponential functions, 196-197 I
Extinction, 11, 40-41 Iconic Memory, 155-157
Eye blink conditioning, 10 Imagery, see Visual imagery

483
 Subject Index

Imaging, see Neural Imaging K

Implicit versus explicit memory, 290-303 Korsakoff’s syndrome, 300
amnesia, 298-302
familiarity, 291-294 L
feeling of knowing, 290-291 Language acquisition, 364-375
language, 365 apes, 223-224, 372-375
priming effects, 294-298, 301-302 child language learning, 364-372
retrieval facilitation, 294-295 critical period, 370-371
Incidental versus intentional learning, 201-202 innate abilities, 371-372
See also Law of effect past tense, 367-370
Induction, 152-153 second-language learning, 365, 370-371
Inductive inference, 339 “ Skinner's theory, 23-24
Inductive learning, 338-376 Language universals, 371-372
See also Causal inference, Concept acquisition, Latent inhibition, 70-71, 114, 177-178
Conditioning, Language acquisition Latent learning, 15, 18, 121, 152
Information-processing theories, 35-36 Law of effect, 14-15, 118, 121-123
Inhibition, 17 See also Incidental versus intentional learning
See also Conditioned inhibition, Latent inhibition Learned helplessness, 103-104
Inhibitory neurotransmitters, 34 Learning curves, 186-197
Inner ear and voice, 167 Ebbinghaus’s research, 7-8
Instinctive behavior, see Biological predispositions power law of learning, 186-197
Instinctive drift, 97-98 practice, 231-232
Instrumental conditioning, 78-117 versus conditioning curves, 11, 191-193
associative bias, 104-106 Learning definition, 4-5
category learning, 93-94 Learning to learn, 242
causal inference, 106-109 Literacy, see Reading instruction
classical conditioning comparison, 79-80 Long-term memory, 27-29
conditioned response, 13, 94-99 Long-term potentiation, 114-116, 192-194, 233-234
conditioned stimulus, 83-94
contingency, 199-106, 125-128 M
dimensional learning, 90-92, 159 Mastery learning, 384-387, 411
hippocampus, 109-116 Matching law, 139-143
Hull's research, 16-17 Mathematics learning, 397-413
maze learning, 18-19, 95-97, 110-113 algebraic word problems, 401-405, 408-411
neutral outcomes, 81-82 arithmetic facts, 243, 399-400
operant conditioning and Skinnerian research, geometry learning, 319-320, 405-407
20-24, 95 international comparisons, 380-383
Rescorla-Wagner theory, 107-108 multicolumn subtraction, 400-401
secondary reinforcement, 82-83 utility, 411-413
temporal relationship of response and reinforce- Maze learning, 18-19, 95-97, 110-113
ment, 83, 99-100, 125-126, 143-144, 145-147 Means-ends analysis
Thorndike’s research, 12-16, 78 general problem solver (GPS), 25-26
Tolman’s research, 17-20 relationship to Tolman’s means-ends readiness-
three-term contingencies, 80-81 es, 19, 26
See also Reinforcement See also Operator subgoaling
Intelligent tutoring systems, 406-411 Meaning memory, see Semantic information
Interference, 165, 226, 239-256 Melioration theory, 140-141
elaborations, 251-252 Memory codes, 210-213
item-based, 241-256 Memory definition, 5-6
list recall, 268-269 Memory record, see Record structure of memory
preexperimental memories, 249-251 Memory span tests, 163, 167-169
recognition memory, 246-251 See also Sternberg paradigm
Rescorla-Wagner theory, 245-246 Mnemonic strategies, 271-273
versus decay, 239-240, 254-256 Momentary maximizing, 140-141

484
 Subject Index

Mood dependency and congruency, 282-284 Power law of forgetting, 227-239

Motivation, see Reinforcement environment, 232-233
Motor programs, 326-336 long-term potentiation, 233-234
learning, 330-336 Power law of learning, 187-197
noncognitive control, 328-329 environmental repetition, 194-195
open-loop versus closed-loop, 326-328 long-term potentiation, 192-194
schema theory, 329-336 skill acquisition, 305-310
versus exponential function, 195-197, 228
N Practice, 186-203
Nativism law of exercise, 15
language acquisition, 370-372 See also Learning curves, Power law of learning,
See also Evolution and learning Rehearsal
Natural categories, 346-349 Prefrontal cortex, 31-32
Negative acceleration, 7 See also Frontal cortex
See also Power law of forgetting, Power law of Presynaptic facilitation, 45-46, 114-115
learning Primacy effect, 28-29
Negative reinforcement, 123-124, 129-130 Priming, 206-207, 208-209, 294-298, 301-302
Negative transfer, 239-243 Proactive interference, 239-243
Neural basis of learning and memory, 6, 30-36 cumulative, 255
and the computer metaphor, 24 Probability matching, 142-143
classical conditioning, 44-49 Problem solving, 311-318
memory, 203-205 See also Difference reduction, Means-ends analysis
Pavloy’s speculations, 12 Proceduralization, 319-320
reinforcement, 134 Production rules, 322-323, 336-337
See also Connectionism, Frontal cortex, intelligent tutoring systems, 407-411
Hippocampus, Long-term potentiation geometry, 322, 407
Neural imaging 5, 36, 180-183, 331-332, 354 multicolumn subtraction, 400-401
Neuron, 33-35 Propositional representations, 218-221
Neurotransmitter, 34-35 primates, 223-224
Nerve impulse, 35 See also Semantic information
Nervous system, 30-35 Punishment, 123-124, 125-129
Thorndike’s view, 14-15, 125
O
Occipital lobe, 31-32 R
Omission training, 123-124 Rate of firing, 35
Open-loop performance, 326-328 Rationality, 119-120, 148-150
Operant conditioning, see Instrumental conditioning See also Adaptation to environment, Economic
Operator, 25, 311-312 view of choice
Operator subgoaling, 312, 314-318 Reading instruction, 387-397
See also Means-ends analysis comprehension skills, 389-390, 393-395
Opponent process, 56-57 international comparisons, 379-380
Optimal foraging theory, 143-145 phonetic decoding skills, 389-393
phonics versus whole-word method, 388-391
P, Reading skill, 388-391
Parietal lobe, 31-32 Recency effect, 28-29, 164-166
Partial reinforcement, 102-103 See also Retention
See also Schedules of reinforcement Recognition failure, 274-275
Peak shift, 86-89 Recognition memory
Phonics method of reading instruction, 388-393 familiarity, 291-294
Phonological loop, 166-169, 171-172 high threshold model, 276
See also Echoic memory interference, 246-252
Pictorial material, see Visual imagery remember-know distinction, 291
Positron emission tomography (PET) 180-183 relationship to recall, 268-275
Positive reinforcement, 123-124 signal detectability theory, 276-279

485
 Subject Index

Reconstructive memory, 285-289 delayed matching to sample, 175-176, 231-232

Record structure of memory, 5, 205-206, 246-247, Ebbinghaus’s research, 7
268-269, 302 emotionally charged material, 256-262
See also Activation, Representation of knowledge Freud’s repression hypothesis, 256-257
Rehearsal, 27-29, 160-162, 197-198,207-208 practice, 231-232
classical conditioning, 176-178 sensory versus semantic information, 216-218
frontal cortex, 178-183 short-term memory, 164-166
phonological loop, 166-169, 171-172 sleep, 254-255
pigeons in delayed match to sample, 175-176 spacing effect, 237-239
Sternberg paradigm, 172-175 time of day, 254-255, 258
visuo-spatial sketch pad, 169-171 See also Cue retrieval hypothesis, Decay hypoth-
See also Sensory memory, Short-term memory, esis, Interference
Working memory Retention function, 7, 227-239
Reinforcement, 14, 118-151 Retrieval, 265-303
aversive stimuli, 125-130 explicit versus implicit memories, 290-302
bliss points, 134-136 recall versus recognition, 268-275
choice behavior, 137-150 reconstructive and inferential memory, 285-289
delay, 83, 99-100, 125-126, 143-144, 145-147 skill acquistion, 320-322
drive reduction theory, 130-131, 134 strategies for recall, 270-273
economic view, 136, 138-139, 145-147, 150 study-test interactions, 279-289, 295-298
effects on learning, 15, 16-18, 118, 121-123, 201- Retroactive interference, 239-243
202 Retrograde amnesia, 299
equilibrium theory, 134-136 Reward, see Reinforcement
Hull’s incentive motivation, 17
matching law, 139-143 S
neural basis, 134 SAM theory, 206, 243
optimal foraging theory, 143-145 Satisficing, 149-150
partial reinforcement, 102-103 Scalloped function, 22
Premack’s theory, 132-133 Schedules of reinforcement, 20-23, 136-143
rational behavior, 119-120, 148-150 Partial reinforcement, 102-103
reward and punishment, 123-124 Schema theories of concept learning, 348-350
schedules, 20-23, 136-143 Schema theory of motor learning, 331-336
Tolman’s goals 19 Secondary reinforcement, 82-83
See also Feedback, Goals, and Learning Second-order conditioning, 52-53, 82
Relational responding, 88-90 Semantic information, 162-164, 212-214, 215-220
Reminiscence, 257-258 propositional representation, 218-220, 223-224
Representation of knowledge, 205-225 retention, 216-218
chunking, 207-210 Sensory memory, 155-159
meaningful information, 212-214, 215-220 auditory, 157-159
in other species 204-205, 221-224 visual, 155-157
propositional representations, 218-220, 223-224 See also Rehearsal systems
sequential memory, 207-208, Sensory preconditioning paradigm, 51-52, 82
verbal information, 207-208, 210-211 Sensitization, 41-42
visual information, 204-205, 210-215 Sentence memory, see Semantic information
See also Category learning, Spatial memory Serial position curve, 28-29
Repression, 256-257 Short-term memory, 28-29, 160-166
Rescorla-Wagner theory, 65-75, 91, 114, 177-178 Atkinson & Shiffrin theory, 28-29, 160,
associative interference, 245-246 coding, 162-164
causal inference, 107-108, 356-357 rehearsal, 160-162
delta rule, 72-75 retention, 164-166 :
Gluck and Bower model, 73-75, 349-350 See also Rehearsal systems, Sternberg paradigm,
Response-prevention paradigm, 50 Working memory
Retention, 226-264 Signal detectability theory, 276-279
arousal, 257-262, 281 Skill acquisition, 304-337

486
 Subject Index

power law of learning, 305-310 Temporal lobe, 31-32, 204, 298-301

Logan’s theory, 320-321 Text editing, 305-307
stages, 310 Tip-of-the-tongue state, 290-291
See also Associative stage, Autonomous stage, Tool building, 314-315
Cognitive stage, Education, Skill acquisition Tower of Hanoi problem, 315-318
Skinner box, 20 Transfer, 323, 337, 402-403, 411-413
Sleep and forgetting, 254-255 Transfer-appropriate processing, 284-285, 289, 298
Spacing effects, 234-239 Transient memories, 155-184
Spatial memory, 95-97 See also Activation, Sensory memory, Short-term
frontal cortex, 178-182 memory, Rehearsal, Working memory
hippocampus, 110-114
See also Cognitive maps, Visual imagery U
Species-specific defense reactions, 105-106 Unconditioned response (UR), 9
SOP theory, 56, 176-178 Unconditioned stimulus (US), 9
Spontaneous recovery, 11 Universals, see Language universals
State-dependent memory, 280-281
Sternberg paradigm, 172-175 Vv
Stimulus-response (S-R) associations, 15, 18, 20, 49- Variable-interval schedule (VI), 137-141, 143-145
53 Variable-ratio schedule (VR), 137-138, 141-143
Stimulus-stimulus (S-S) associations, 49-53 Verbal code, 210-211
Strength See also Echoic memory, Phonological loop
Hullian theory, 16-17 Visual memory, 204-205, 208-217
interference, 248-249 imagery, 169-171, 214-215
law of exercise, 15 See also Spatial memory
practice, 186-197 Visual sensory memory, 155-157
Skinner's opinion, 20 Visuo-spatial sketch pad, 169-171
versus elaborativeness of processing, 200-201
See also Acquisition, Extinction, Learning curves, Ww
Practice, Retention, Rescorla-Wagner theory Whole-word method of reading instruction, 388-393
Strength equation, 197 Word processing, 305-307
See also Activation equation Working memory
Subjective value, 119-120, 145-150 Baddeley’s concept, 171-172
See also Expected value neural Imaging, 180-183
Subgoaling, 26, 312, 314-318 Olton’s concept, 112-113
Superstitious learning, 101-102 primate frontal cortex, 178-183, 204
Synapse, 34-35 See also Activation, Rehearsal systems, Short-
term memory
il;
Task analysis for education, 383-384 XYZ
See also Componential analysis Yerkes-Dodson law, 261
Taste aversion, 39, 61-62, 71, 76

487
 maAnee 20 wht hee?
sa ce Dials er -
4 ot) 19S tet seqgnotetl- gt ee
‘ on We npetthl gnitltcing peat ewncsiitas
i vitck, METfesere
720, Sy ener wdepae
iat 5ote hg te AMM °F LER seeanif oh shay alr
awed ba nia hatalaro>
RED ES Bes MET Gitkeroung sere rryqe ter
akceai « @ WORLR2 err ieee deay, TS
inept idcieget ened sesh Nea Me wag lin, rom Wat ove
yr ll
“phe amedig e Tarot arene aie Ofcael SS54-055,
vistaced Bet (ud 1 Pee werote Cn: wired eh,
ae an. iiey ary, Sa oe ’ Murr 0 ~ ingen pe a
Reine ye er Oat) sacar Sew eititonrnn tee io PR DR ered 2
heen. )) TRE} sulci beeebibral? Dati, 965-28 bbenaldan:
bi aesdien of er Vitnasi dose sleerncinld - rae: Rncaegea’
ome. ~ esis 7.
| be LM
pave. 4
— ix . vo , a “As
he bdedsaast

elt pedacnedaiente 0-2=o

7 =
ak SHOR AVG dl iete beentalab ade/
7 Ui EdtdPha RESTV) Sea
aiororac eve,14 4 -S 165-012,
oltn-widetey
dot
ann im
hb
poat st
teeam eI)
=

Coda: basiggatony oS,qeoTeparl ats. eh wee- eee thebipetiecerice, RIAA

24 VIL <@OS, GOS-DOS atereet eel Pes geeks reel, 2 Tet Woe
a Jequlbrash Down),
tae
ypinverive Die apap: long
SOE TL eee
ode
eed
ie
or Meee
’ wt
rare Soeur vow dle
4 ae seen yer, bed nt = i
(apt, Dal, ati OM sas;Honey hele Ou Si SeepDae f1 Of aoligo
option: \apyigneny, GF 145 % Wied yic=enn lo atieevitn 4 Las
payTle ou eeerd. AY <i) wt neegeageates. OHS AONE ee
EXEDU svthaminn seYa botion brey-stcl vn empeNlnetsemndh sighed’
me euteh oie 61) RE Green 7m bistiv# Pecagil hei meees
reve! aonlpp wher Tlaegceam jhe! Boas \oeey ete
edie; 3) BRAY Spncns vetoed nfora ery of
Tebnyts goes 39» CALUAL getyerlaensn micrshety ersten, 2
Sa ow ie GS Bhi Garten (ret%> “wwe at racliion‘ a e
Gigs s| dn peaionaseartndWeatiae latinos otelihg ‘oreo 7phconsiny nt tae! 3
or
et Le dwiriver
r enc] ioeFl nalteri Au nf me:
ij . gr1s onal in

-§ i al hh pol agg, “pean om 7 PA (ee, “343 :

a

»
Vaan, 29g 1h. ™ ake i hel iy
megs ehcp. po ee YK mitts te BOERNE,
7 ‘ aa
in dee me SS. aw erule/] entra 2k)” etevigna ed
Pryvietia c, Gey ceelperm, 22 wee EACH he we Laer oo i904
EAST UY Sieh, “ d “Sereiey eerie ty plea.
vere iosiaae, ler 21-29) bares icc, ae 2 me
WA KoA De TL) Bxb yt eye a aeny = a
Sie de |dhe phetiwe Dette yecsicey . rh pa i) eral ie, ae
Cc Geen. if ; y Sa eh 7
ae Ga Vs ial Aty MEET oh 7A
Soe sins Dos Ma Se :
Carat! ieee: [NTE Sep Hee
é a ° ‘ — al .

aye o'r ‘ 7

HM 240 Dievey fie, rr Fay Rea)

R eee tron woge aI vw

VeHentwr. Tn wd ‘
= o *. ab
JOHN WILEY & SONS, INC.
New York = Chichester = Wei
Brisbane = Singapore = Toront
http://www.wiley.com/college