Cognitive Science Perspectives on Personality and Emotion

COGNITIVE SCIENCE PERSPECTIVES ON PERSONALITY AND EMOTION

ADVANCES IN PSYCHOLOGY 124 Editors:

G. E. STELMACH E A. VROON

ELSEVIER A m s t e r d a m - Lausanne - New Y o r k - O x f o r d - Shannon - S i n g a p o r e - Tokyo

COGNITIVE SCIENCE PERSPECTIVES ONPERSONALITY AND EMOTION

editedby Gerald MATTHEWS University of Dundee Dundee, Scotland

1997

ELSEVIER Amsterdam - Lausanne- New Y o r k - O x f o r d - Shannon- Singapore- Tokyo

NORTH-HOLLAND ELSEVIER SCIENCE B.V. Sara Burgerhartstraat 25 p.o. Box 211, 1000 AE Amsterdam, The Netherlands

ISBN: 0 444 82450 2 9 1997 Elsevier Science B.V. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior written permission of the publisher, Elsevier Science B.V., Copyright & Permissions Department, EO. Box 52 l, 1000 AM Amsterdam, The Netherlands. Special regulations for readers in the U . S . A . - This publication has been registered with the Copyright Clearance Center Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923. Information can be obtained from the CCC about conditions under which photocopies of parts of this publication may be made in the U.S.A. All other copyright questions, including photocopying outside of the U.S.A., should be referred to the copyright owner, Elsevier Science B.V., unless otherwise specified. No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. This book is printed on acid-free paper. Transferred to digital printing 2005

List of Contributors Jean P. Banquet*. Neuroscience et Modrlisation, lnstitut des Neurosciences,

UPMC, 9 quai St Bernard, 75252 Paris cedex, France. Anthony Beech*. Department of Forensic Psychology, Fair Mile Hospital,

Wallingford, Oxfordshire OX 10 9H, England. Jean Claude Dreher. Equipe de Traitement des Images et du Signal (ETIS),

ENSEA/UCP, Umversit6 de Cergy-Pontoise, 6 Avenue du Ponceau, 95014 Cergy-Pontoise cedex, France. Kevin M. Carlsmith. Department of Psychology, Princeton University,

Princeton, NJ 08544, U.S.A. Gerald L. Clore. Deparment of Psychology, University of Illinois at Urbana-

Champaign, 603 East Daniel Street, Urbana-Champaign, IL 61820, U.S.A. Douglas Derryberry*. Department of Psychology, Oregon State University,

Corvallis, OR 97331, U.S.A. Heather Frasier Chabot. Department of Psychology, University of New

Hampshire, Durham, NH 03824, U.S.A. Philippe Gaussier. Equipe de Traitement des Images et du Signal (ETIS),

ENSEMUCP, Universit6 de Cergy-Pontoise, 6 Avenue du Ponceau, 95014 Cergy-Pontoise cedex, France. Wilfried Gtinther. Neuroklinik Bamberg, St Getreu Strasse 14-18, 8600

Bamberg, Germany. Rick E. Ingram. Department of Psychology, San Diego State University, San

Diego, CA 92182-0551, U.S.A. C~dric Joulain. Equipe de Traitement des Images et du Signal (ETIS),

ENSEA/UCP, Universit6 de Cergy-Pontoise, 6 Avenue du Ponceau, 95014 Cergy-Pontoise eedex, France. Timothy Ketelaar*. Center for Adaptive Behavior and Cognition, Max

Planck Institute for Psychological Research, Leopoldstrasse 24, 80802 Munich, Germany.

Contributors

vi

Shinobu Kitayama*. Faculty of Integrated Human Studies, Kyoto University,

Kyoto 606-01, Japan. GeraM Matthews*. Department of Psychology, University of Dundee,

Dundee DD 1 4HN, Scotland. John D. Mayer*. Department of Psychology, University of New Hampshire,

Durham, NH 03824, U.S.A. Edward Necka*. Instytut Psychologii, Uniwersytet Jagiellonski, ul. Golebia

13, 31-007 Krak6w, Poland. Marjorie A. Reed. Department of Psychology, Oregon State University,

Corvallis, OR 97331, U.S.A. Carien M. van Reekum. Department of Psychology, Universit6de Gen6ve, 9,

route de Drize, CH- 1227 Carouge-Geneva, Switzerland. Equipe de Traitement des Images et du Signal (ETIS), ENSEA/UCP, Universit6 de Cergy-Pontoise, 6 Avenue du Ponceau, 95014 Cergy-Pontoise cedex, France.

Arnaud Revel

Klaus R. Scherer*. F.P.S.E. Section Psychologie, Universit6 de Gen6ve, 9,

route de Drize, CH- 1227 Carouge-Geneva, Switzerland. Greg Siegle*. Doctoral Training Facility, San Diego State University, 6363

Alvarado Court, San Diego, CA 92120, U.S.A. W.W. Tryon*. Department of Psychology, Fordham University, Rose Hill

Campus, 441 East Fordham Road, Bronx, New York, NY 10458-5198, U.S.A. Leanne Williams. Psychology Department, University of New England,

Armidale NSW 2351, Australia.

* Corresponding author

Preface We are all cognitive scientists now. Researchers routinely use the language of cognition in developing models of personality and emotion. Constructs such as automatic processing, schemas, working memory, attentional resources and the like are now part of the essential fabric of theory. The popularity of information-processing models offers both a promise and a threat. The promise is that of a true understanding of how the different psychological faculties of perception, attention, memory and so forth are inter-woven to create the whole person, and to create the mtegrat~ adaptive reactions we call emotions. Contemporary cognitive science is at ease with multiple levels of description and explanation, and so is especially well-suited to explaining the origins and expressions of emotion and personality. But do we really speak a common language, or are we heading for a new Babel? Constructs such as schemas and strategies sometimes seem plastic enough to fit almost any theoretical conception, so that the verbal labels become private rather than shared. As subjects of inquiry, emotion and personality are particularly vulnerable to the use of language as artifice rather than as scientific discourse. The decline of psychoanalysis as a scientific enterprise illustrates the nature of the threat. In contemporary research, there is an evident risk of "cognitivism", dressing up untestable ideas in cognitive jargon. The differing perspectives provided by different strands of cognitive research are a strength, not a weakness, but communication between different perspectives requires us to work from common scientific bases. This book aims to highlight the vigour, diversity and insight of the various cognitive science perspectives on personality and emotion. It aims also to emphasise the rigorous scientific basis for research to be found in the integration of experimental psychology with neuroscience, connectionism and the new evolutionary psychology. Collectively, the contributors to this book provide a wide-ranging survey of leading-edge research topics. It is, a little arbitrarily, divided into three parts, on general frameworks for cognitive science, on perspectives from emotion research, and on perspectives from studies of personality traits. In the first, introductory chapter, I begin Part I with a personal view of the impact of the cognitive revolution, and apply the "classical theory" of cognitive science to issues in personality and emotion. As the book took shape, I came to appreciate how much a cognitive science of personality and emotion is necessarily a science of motivation too. In

Preface

viii

Chapter 2, Mayer, Frasicr Chabot and Carlsmith inter-relate these three constructs in the context of the traditional "trilogy of mind": conation, affect and cognition. They procr~ to outline a new "quatcrnity of mind", encompassing consciousness also. One of the most radical and exciting innovations of cognitive science is the use of connectionist models, and the remaining two contributors to Part I provide two different perspectives on their application. Tryon's Bidirectional Associative Memory (BAM) uses the conncctionist metaphor of memory as wells in an energy surface as a source of insight into normal emotion and pathological conditions (Chapter 3). He also outlines how psychotherapy may be directed towards re-landscaping the energy surface, by shrinking memory wells whose diameter gives them too much power over the person's experiences, for example. In Chapter 4, Banquet, Gaussier, Drehcr, Joulain, Revel and G0nthcr describe a more ncurologically-orientod conncctionist perspective on personality. They discuss how the person's sense of identity in space and time derives from circuits in hippocampus and prefrontal cortex, supporting spatio-tcmporal processing, working memory, planning and goal propagation. Part II reviews perspectives derived primarily from emotion research, which explore the interplay between emotion as a common human characteristic and individual difference factors. One of the flaws in an overly cognitivistic conception of emotion is neglect of unconscious, prcattcntivc processes which guide later, attentive processing. Kitayama (Chapter 4) presents the amplification model of affect-cognition interaction in early perceptual processing. The model describes how the emotional content of stimuli may either enhance or impede subsequent conscious rccognition, explaining phenomena such as "perceptual dcfencc". Van Rcckum and Schcrcr (Chapter 5) also address distinctions between different levels of processing, in the context of appraisal, which may be supported by sensorymotor, schematic or conceptual processing routincs. They review ncuroscicncr bases for appraisal, and link personality to different appraisal characteristics. In Chapter 6, Sicgle and Ingram explore conncctionist modelling of the negative biases in cognition characteristic of depression and other emotional disorders, expressed in appraisal, attention and memory. They focus especially on lcxical decision and valence identification as tasks which bring to thc surface the abnormalities of processing underlying pathology. The pcrspcctivc from evolutionary psychology is presented in Chapter 7 (Kctelaar and Clorc), which discusses the long-term adaptive significance of emotions, as informative and motivational signals. The authors review evidence suggesting that analysis of the evolved functions of

Preface

ix

emotions helps us to understand their more immediate effects on cognition in experimental studies. Part III is oriented towards research on personality traits, within a loosely Eysenckian framework, with contributions relating to the three superfactors of extraversion-introversion, neuroticism (anxiety) and psychotir (schizotypy). Perhaps a future volume of this kind will be able also to cover additional dimensions from the five factor model; conscientiousness, agreeableness and openness to experience. In Chapter 9, I present a cognitiveadaptive model of extraversion, which reviews information-processing correlates of the trait in the context of adaptive specialisation. Extraverts may be superior in verbal facilities such as short-term recall, retrieval and multitasking because these cognitive characteristics contribute to coping with their preferred environments. Derryberry and Read (Chapter 10) discuss the relationship between motivational and attentional aspects of anxiety, from the standpoint of cognitive neuroscience. Experimental data illustrate anxietyrelated biasing of specific attentional functions which may contribute to shaping the higher-level cognitions and motivations of anxious individuals. Beech and Williams (Chapter 11) assess the cognitive bases for schizophrema and schizotypal personality. They develop a model of activation and inhibition processes which explains priming data obtained experimentally, and the positive symptomatology of schizophrenia such as delusions and hallucinations. Finally, contemporary trait researchers are increasingly engaged with exploring the relationships between personality and ability traits. In Chapter 12, Necka links intelligence, extraversion and neurotir to an attentional resource model. Both personality and ability traits are related to arousal processes, whose impact on cognition is shown in experimental studies of dual-task performance and memory scanning. I am grateful to the Medical Research Council for their support for my research while this book was in preparation. I would also like to thank the contributing authors. I have enjoyed reading and re-reading the chapters, and my schemas and networks are greatly enriched. This is the book I would have liked to have read when I first began researching personality and emotion as a doctoral student in the early 1980s. I hope it will serve as an inspiration and a guide to all those with an interest in this exciting new research area.

Gerald Matthcws

This Page Intentionally Left Blank

Contents

P A R T I. F R A M E W O R K S F O R C O G N I T I V E S C I E N C E

Chapter 1. An Introduction to the Cognitive Science of Personality and Emotion ....................................................................... 3 Gerald Matthews Landmarks of the Cognitive Revolution .............................................. 3 A Cognitive Science Framework ...................................................... 7 Towards a Cognitive Neuroscicnce of Personality and Emotion? ......... 13 Developing Adaptive Explanations .................................................. 15 An Example: Explaining Anxiety and Cognition .............................. 20 Conclusions .................................................................................... 24

Chapter 2. Conation, Affect, and Cognition in Personality ................... 31 John D. Mayer, Heather Frasier Chabot and Kevin M. Carlsmith The Relational Model of Personality .................................................. 32 Understanding Conation, Affect, and Cognition ................................. 39 The Quaternity of Mind and Personality Dynamics ............................ 52 Conclusions and Other Considerations .............................................. 60

Chapter 3. Introduction to the Bidirectional Associative Memory Model: Implications for Psychopathology, Treatment, and Research .......................................................................65 Warren W. Tryon Bidirectional Associative Memory (BAM) ......................................... 67 Encoding Emotion ............................................................................. 70 Implications for DSM-IV Disorders .................................................. 75 Treatment ......................................................................................... 92 Research Strategies ........................................................................... 99 Conclusions .................................................................................... 101 Appendix: Description of the Bidirectional Associative Memory ....... 109

Contents

xii

Chapter 4. Space-Time, Order, and Hierarchy in FrontoHippocampal System: A Neural Basis of Personality .......................... 123

dean P. Banquet, Philippe Gaussier, Jean Claude Dreher, Cddric Joulam, Arnaud Revel and Wilfried G~tnther Hippocampal Function: An Extended View ..................................... Working Memory as Both a Cortical and a Hippocampal System ..................................................................................... Neuropsychology, Brain Imaging and Working Memory ................... Neurophysiology: Human Versus Animal Working Memory ............ Spatio-Temporal Processing in Hippocampus and Prefrontal Cortex ...................................................................................... Functional Model ............................................................................ Fronto-Hippocampal Function and Personality ................................ Conclusion .....................................................................................

126

129 135 148

151 159 176

179

PART II. PERSPECTIVES FROM EMOTION RESEARCH Chapter 5. Affective Influence in Perception: Some Implications of the Amplification Model ..................................... 193

Shinobu Kitayama The Amplification Model of Affect-Cognition Interaction ................. 196 Evaluation Criteria of the Amplification Model ................................ 202 Experiment 1 .................................................................................. 212 Experiment 2 .................................................................................. 221 The Amplification Model Evaluated ................................................ 230 Relations with Extant Theories of Attention ..................................... 232 Amplification of Attention in Other Domains ................................... 235 Perceptual Defense and Vigilance? .................................................. 238 Future Research Directions ............................................................. 240 Concluding Remarks ....................................................................... 242

Contents

xiii

Chapter 6. Levels of Processing in Emotion-Antecedent Appraisal .............................................................................................. 259

Carien M. van Reekum and Klaus R. Scherer Critique of Appraisal Notions ......................................................... Levels of Processing in Appraisal ........................... : ........................ Hierarchical Process Notions in Related Traditions .......................... Issues in Rewriting Appraisal Theory .............................................. Individual Differences in Appraisal Processes .................................. Conclusions ....................................................................................

260 263 266 277 280 289

Chapter 7. Modeling Individual Differences in Negative Information Processing Biases .............................................................. 301

Greg.1. Siegle and Rick E. Ingram Personality Research and Vulnerability to Depression: A History ...... 302 Simulating Aspects of Depression and Personality on a Computer ....................................................................................... 304 Simulating Personality Factors ........................................................ 320 A Brief Conclusion ......................................................................... 348

Chapter 8. Emotion and Reason" The Proximate Effects and Ultimate Functions of Emotions ........................................................... 355

Timothy Ketelaar and GeraM L. Clore Why Does Emotion Affect Cognition? ............................................. 356 Specific Aims of this Chapter .......................................................... 358 Consequences of Mood ................................................................... 360 Consequences of Emotions .............................................................. 365 Emotion-as-motivation and Frank's (1988) Commitment Model ....... 371 Affect-as-lnformation and Behavior ................................................ 378 The Future of Affect and Information Processing ............................. 387 Conclusion: Deficits, Biases, and Functions ..................................... 388

Contents

xiv

P A R T IIl. P E R S P E C T I V E S F R O M P E R S O N A L I T Y TRAIT RESEARCH

Chapter 9. Extraversion, Emotion and Performance: A Cognitive-Adaptive Model ................................................................ 399

Gerald Matthews ..

Extravorsion and Affect .................................................................. Extraversion and Performance ......................................................... Extraversion, Arousal and Attontion: Empirical Studies ................... An Adaptivr Framowork for Cognitive Correlates of Extraversion-lntroversion ................................................................ Conclusions ....................................................................................

400 405 409 426 434

Chapter 10. Motivational and Attentional Components of Personality ............................................................................................ 443

Douglas Derryberry and Marjorie A. Reed Biological Approachos to Personality .............................................. Assessing Attcntional Processes in Anxiety ..................................... Extensions to Complex Cognitive Processing ................................... Conclusions ....................................................................................

444 450 462 466

Chapter 11. Investigating Cognitive Processes in Schizotypal Personality and Schizophrenia .......................................... 475

Anthony Beech and Leanne Williams Mechanisms of Selective Attention .................................................. Experimental Investigations of Inhibitory Processes ......................... Inhibitory Processes in Schizophrenia .............................................. Towards a "Roducexl Cognitive Inhibition" Model of Schizophrenic Symptomatology ...................................................... Revising the Model ......................................................................... Conclusion .....................................................................................

477 478 485 490 494 497

Contents

xv

Chapter 12. Attention, Working Memory and Arousal: Concepts Apt to Account for the "Process of Intelligence" ................. 503 Edward Necka Theoretical Notions ......................................................................... Assumptions ................................................................................... "The Process o f Intelligence" ........................................................... Preliminary Empirical Data ............................................................. Cognitive Science Perspectives ........................................................

504 512 519 525 542

Subject index ......................................................................................... 555

This Page Intentionally Left Blank

PART I FRAMEWORKS FOR COGNITIVE SCIENCE

This Page Intentionally Left Blank

Cognitive Science Perspectives on Personality and Emotion-G. Matthews (Editor) 1997 Elsevier Science B.V. CHAPTER 1

An Introduction to the Cognitive Science of Personality and Emotion GeraM Matthews

The cognitive revolution has transformed the face of research on personality and emotion. Information-processing theories spring up like poppies in a cornfield, and often wither just as quickly, crowded out by more recent growth. With cognitive approaches so firmly established, it is timely to stand back a little from the intellectual ferment, and take stock of the achievements and limitations of the research area. This book represents the leading edge of research on the cognitive science of personality and emotion, and so the contributions concern a variety of specific topics. But for personality and emotion research to mean anything at all, it must above all be integrative. Both constructs hang on a multi-layered web of data and hypothesis, spanning the gamut of psychological phenomena from neuronal firing to social interaction. The aim of this introductory chapter is to outline the overall framework provided by cognitive science, and its place in personality and emotion research. In this chapter, I will sketch the progress so far of the cognitive revolution in personality and emotion (PE) research. I will then describe the classical model of cognitive science, and its three levels of explanation: the biological, the symbol-processing and the knowledge levels. Cognitive science emphasises that information-processing models are necessary but not sufficient for understanding, l will show that cognitive science explanations provide new perspectives on some old problems, and demonstrate its integrative potential by outlining its application to anxiety.

Landmarks of the Cognitive Revolution Emotion and cognition

The cognitive science of emotion has disparate roots, which demonstrate the diversity of "cognitive" approaches. The information-processing approach is based on empirical, performance-based studies of emotion, addressing problems such as the deleterious effects of anxiety on attention. It

4

Chapter 1

accommodates the various conccptualisations of emotion: as a universal but situationally-contingent human response, as an individual difference factor, and as a property of stimuli (valence). Emotion may be conccptualiscd as a dependent variable influcnce~ by processes such as appraisal, or as an independent variable which itself influences information-processing. The more sophisticated applications of the approach (e.g. Ingrain, 1984) build in feedback from appraisals of performance back into emotion. In linking emotion to bchaviour, the basic research tactic is to demonstrate moderation of effects of emotion on performance by task factors, the standard technique of experimental cognitive psychology. Emotion • task parameter interactions inspire processing models which may then be subjected to further test. The approach scores highly on scientific rigour, but like much cognitive psychology, risks degenerating into an account of the minutiae of a specific experimental paradigm with little wider relevance (cf. Neisscr, 1976). An alternative approach is design-oriented: what might be the purpose of emotion within the cognitive system as a whole? Simon (1967) linked emotion to an interrupt function, contributing to people's capacity to adapt to unpredictable environments by switching back and forth between different goals. This analysis begs the question of why the interrupt function requires all the various concomitants of emotion such as physiological arousal, biases in thinking, action tendencies and the like. Research in the Artificial Intelligence (AI) tradition simulates complex, goal-directed systems to discover basic design principles, indicating, for example, what other features arc required for interrupts to work properly. This approach generates rich and thought-provoking data, but its scientific rigour is open to question. Argument tends to proce~ by analogy and comparison of features of artificial and human systems, and it is unclear that the parallels drawn arc open to falsification or to formal test against alternative explanations. A third tradition derives from stress and clinical research, and the observation that negative emotions derive from the way people interpret and manage events, rather than from fixed properties of the events themselves. It is exemplified by the work of Lazarus (1991; Lazarus & Folkman, 1984) on the transactional model of stress, and the roles of appraisal and coping within specific, potentially stressful encounters. As theory, it has some of the characteristics of both the design and information-processing approaches. Like the design approach, it is explicitly systems-based, with emotion conceptualiseA as a "core relational theme" charactcrising the personenvironment system as a whole. However, like the information-processing tradition, transactional theory attempts to establish local cause-and-effect

G. Matthews

5

relationships open to direct empirical test, such as the effect of appraisals on emotion. Similarly, clinical accounts of anxiety and depression which emphasise the role of the person's self-knowledge and reasoning processes in generating negative emotion as an overall indicator of system functioning (e.g. Beck, 1967; Ellis, 1962). The advantages of such approaches are depth of insight obtained into the experiences of people in real environments, and practical applications to stress management and cognitive behaviour therapy (Matthews & Wells, 1996). Their shortcomings relate, first, to emphasis on self-report data, which may present a partial and distorted view of underlying processing, and, second, as with the design tradition, to difficulties in rigorous theory testing. Finally, the neuroscience of emotion has become increasing cognitive in orientation, as traditional arousal theory has fallen from favour (e.g. Robbins, 1986). Increasingly, it has becomes possible to align specific neural circuits with information-processing and behavioural function (e.g., Gray, 1982). Emotion is notoriously diffmult to localise, but advances in brain scanning technology, and in simulation of neural function are a source of optimism for the future. Pessimists focus on the extent to which feelings are intertwined with thinking, and consequent difficulties in discriminating neural and cognitive influences. There remain fundamental disagreements over the extent to which psychological phenomena are reducible to neural processes (see Lazarus, 1984, 1991, and Gazzaniga, 1992, for the end-points of the continuum of views). At the least, though, computational theories permit testable predictions concerning neural influences on behaviour, contributing to the development of cognitive neuroscience models of emotion. Personality and cognition

Much of personality research is structure- rather than process-oriented, and so unaccommodating to cognitive perspectives. The current popularity of the Five Factor Model owes much to the prodigious empirical programmes of researchers such as Costa and McCrae (1992) in deriving the Big Five as a structural description of various data sets. Personality trait theories have often been based on somewhat naive biological or conditioning models, inspired by Pavlov and J.B. Watson rather than by contemporary research. Arousal theory, in particular, has proved to be a mixed blessing. The concept undoubtedly has integrative value (K.A. Anderson, 1990), and the basic principle that personality reflects biology is becoming increasingly securely supported by behaviour and molecular genetic studies (Loehlm, 1992: Lesch

6

Chapter 1

ct al., 1996). Eyscnck and Eyscnck's (1985) application of arousal theory to personality has scored some notable empirical successes in predicting cxtraversion-introvcrsion effects on sensory thresholds and simple conditioning tasks. Unfortunately, psychophysiological data on personality arc confusing and inconclusive, and arousal theory has proved to be a poor basis for predicting personality effects on cognitive tasks (e.g. Matthews, 1985; Matthews & Deary, in press). Despite the conservatism of much personality research, there arc increasing signs that the cognitive revolution is taking root in this area also. As in the case of emotion, its expressions arc diverse. Information-processing analyses of personality effects on performance arc becoming increasingly common. The trail has been blazexl by research on anxiety traits, driven by the observation that cognitive worry is more predictive of performance than emotional and physiological tension. Detrimental effects of anxiety arc now routinely explained in terms of constructs such as attentional capacity (Sarason, Sarason, & Pierce, 1995) and working memory (Eyscnck, 1992). Humphmys and Rcvr162 (1984) have proposed an ambitious integration of individual differences research which links achievement motivation, anxiety and impulsivity to arousal and effort, which in turn influence availability of multiple resources for performing attcntional and working memory tasks. There is also a rather separate tradition with a basis in social-cognitive psychology, concerned with the knowledge structures which support personality, such as the self-schema (Cantor & Zirkcl, 1990). This approach supports some information-processing work, such as studies of self-referent processing (Klein & Loftus, 1988) and priming (Bargh, Chaikcn, Govcndcr, & Pratto, 1992), but also leans heavily on qualitative and self-report data. Hence, it resembles the transactional approach to emotion: its allegiance is to cognition but not necessarily to cognitive science. On the other hand, it is sufficiently flexible to br applied to both nomothctic and idiographic aspects of personality, and engages with individuals' actual life experiences.

Integration of personality and emotion research The distinction made between personality and emotion is artificial to the extent that much personality research has an explicit trait-state orientation, within which personality effects arc mediated by emotional states (e.g. Spiclbcrgcr's, 1966, anxiety theory). We cannot do personality research without consideration of emotion, but the converse also applies. Some studies of mood make a strong equation between positive and negative affect on the

G. Matthews

7

one hand, and extraversion and neuroticism on the other. Individual differences in mood may substantially reflect individual differences in reward and punishment systems said to be the basis for extraversion and neuroticism (Watson & Clark, 1992). Unfortunately, taken to the extreme, this approach leads to a dreary tautology, such that some unfortunates have negative genes, negative brains, negative emotions and negative personalities, and little more can be said. More promising are interactionist approaches which emphasise that individual differences in emotional response are not mechanically linked to personality, but depend on a more complex interplay between person and environment. Within the transactional model, personality is seen as biasing the appraisal and coping processes which are perhaps more direct influences on emotion (Matthews & Deary, in press). Interactionism can easily degenerate into an unfalsifiable everything-affects-everything position, but computational models can potentially supply much needed precision to theory in this area. Information-processing analyses of performance frequently attempt to discriminate trait and state effects on different processing components. Eysenck's (1992) review of the area suggests that simple traitstate models, within which trait effects are entirely mediated by gross state constructs, are not viable: trait anxiety may sometimes influence cognition and behaviour even with state anxiety controlled. Integration of trait and state research requires a more sophisticated view, such that traits affect stable parameters of processing systems which moderate their reactions to stimuli. We might link traits to knowledge structures in long-term memory (LTM) which feed into appraisal and coping (Wells & Matthews, 1994), or, from a connectionist perspective, to parameters of networks which govern the spread of activation (Matthews & Harley, 1993). In either case, moderating effects of traits are apt to be subtle, and require careful modelling.

A Cognitive Science Framework The brief overview above demonstrates the vigour of the cognitive approach to P E. It also shows that progress has been uneven, and the diversity of differing "cognitive" approaches. We require a general framework for examining where cognitive research has been most successful, and where its impact has so far been limited. Fortunately, the "classical theory" of cognitive science provides a ready made framework, discriminating different levels of explanation. Next, I will outline these levels, and discuss their application to PE research.

8

Chapter I

Pylyshyn (1984) presents a detailed analysis of knowledge, symbolprocessing and biological levels of explanation, from which the following account is derived (see Figure 1). The central point is that psychological events are open to qualitatively different explanations. Suppose we observe an extraverted man at a party, engaging in cheerful social interaction. How do we explain this bohaviour? One approach is to refer to his motives and goals. Perhaps he is a newcomer, and wishes to make new friendships from which he will benefit. This level of explanation is the knowledge-based or semantic level. It is concerned especially with the way the cognitive system is designed for adaptive interaction with the external environment, in pursuit of its goals. It has been developed in PE research through AI approaches to understanding emotion, through work on the adaptive functions of PE, and through social knowledge approaches to personality. Alternatively, we might present an account based on the formal cognitive architecture: a computational description of the processing structures and operations linking inputs to social behaviours. We may then identify individual differences in specific computations, such as spee~ of accessing items of social knowledge, which explain the individual's social behaviour at the processing level. Explanations of this kind are concerned with the formal characteristics of processing, rather than with the adaptive significance of processing routines. They provide the basis for much of the extensive research on information-processing models of emotion and personality previously described. Classical theory requires the architecture to be based on discrete symbols, expressing propositions. Pylyshyn distinguishes sub-levels of algorithm and functional architecture, which differentiate the logical operations performed on symbols form the cognitive structures implementing symbol processing. The centrality of symbols is a controversial area. Some authors place symbol-based accounts of processing centre stage, due to identifiability problems of modelling functional architecture (J.R. Anderson, 1990). Conversely, connectionist models see network implementations as a more powerful method for modelling behavioural data than symbolic accounts, and may even reject symbolic representations as irrelevant to theory (Smolensky, 1988). I will take the view that, in the light of the successes of connectionism, an a priori commitment to symbolic accounts may be too constraining for PE research. I will use the term "architectural explanation" to refer to explanation in terms of the formal properties of the processing machinery, irrespective of whether or not it is symbolic in nature.

G. Matthews

Knowledge

=

Goals, intentions and personal meaning, supporting adaptation to external environments

Algorithm Symbol

9

=

Formal specification of program for symbol manipulation

Functional _ Architecture

Real-time processing operations supporting symbol manipulation

/

processing

Biology

=

Physical, neuronal representationof processing

Figure 1. Levels of explanation in cognitive science.

Finally, we may look to the functioning of the neural hardware for explanation. We might use brain scanning techniques to investigate which neural structures and circuits are active during social interaction, and develop a theory linking the individual's social behaviour to the activity of the circuits concerned. We must then tackle transducaon problems; the conversion of analogue physical events into symbolic codes (Pylyshyn, 1984), or other abstract codes. In the next section of this chapter, I develop the position that information-processing models of PE are necessary but not sufficient for understanding. Processing models possess the rigour provided by computational specification, and, if adequately formulated, are readily testable against empirical data. However, a processing description of PE phenomena requires supplementation with explanations which look both downwards, to architecture and cognitive neuroscience, and upwards to knowledge-level explanations. Information-processing models." Strengths and limitations

Information-processing models of personality and emotion have an impressive track record in characterising empirical phenomena in terms of constructs such as resources, processing stages and activation of network units. The application of such models is demonstrated throughout this

10

Chapter I

volume. Multi-level models, distinguishing qualitatively different types of processing, such as stimuhs-drivcn and strategic processing, have been particularly successful in explaining empirical data (see van Rcekum & Scherer, this volume). Processing models are essential for predicting and understanding the correlates of P E, and they arc increasingly finding applications in the clinical domain (see Beech & Williams, Sicgle & Ingrain, Tryon, this vohmc). However, it is important to be clear about what such models provide and do not provide. Most models provide a snapshot description of processing at a single time epoch, although there is growing interest in learning models (e.g. Kanfcr & Ackerman, 1989). Such a description leaves open alternative types of explanation. The first question is whether effects of PE factors on processing reflect genuine differences in cognitive architecture, or differences in strategy, i.e. how the same architecture is used to support different processing sequences within a given context. It is unlikely that PE has dramatic effects on architecture; we would not expect syntactic deep structure to vary across individuals, for example (cf. Pinker, 1994). Perhaps more likely are quantitative cross-individual or cross-occasion differences in system components such as resource availabilities, short-term memory (STM) slots and speed of execution of key processes (e.g. Nccka, this vohmc). The architecture may also handle emotional stimuli differently to neutral stimuli (Kitayama, this volume). Architecture as a source o f variation

Care is needezl in showing that variance in processing reflects variation in architecture, as opposed to variation in strategy and intention (Pylyshyn, 1984). A strategy may be defined as a goal-dircctexi, voluntarily-imtiatcd processing routine. Typically, a strategy is implcmcnteA and regulated through executive processes which bias involuntary processing (see Norman & Shallice, 1985). There are rather few instances of attempts to establish systematically whether PE phenomena are strategy-dependent, although effects of emotion on strategy-insensitive processes such as early stimulus analysis (Kitayama, this volume) and procedural learning (Corr, Picketing, & Gray, 1995) are suggestive of architectural differences. More generally, processes of interest depend on both the fixed architecture and strategy, and it is difficult to disentangle the two types of influence. For example, extraverts tend to show greater STM capacity than introverts (Matthews, 1992), but this effect might reflect either individual differences in cognitive architecture, perhaps derived from physiological processes (Eysenck & Eysenck, 1985), or

G. Matthews

11

from extraverts' choice of coding strategies which tend to enhance short-term recall at the expense of long-term recall (Schwartz, 1975). If a PE effect on architecture is established, explanatory questions remain. One possibility is that PE variance in architecture reflects relatively straightforward properties of the brain. The neural substrate for emotional states may influence the formal properties of processing over short timescales. Given the heritability of personality traits, including traits related to emotionality, it is plausible that genes code for individual differences in architecture. Alternatively, the architectural difference may be more readily conceptualised as a learning effect, such as changes in control structure associated with "proceduralization" of knowledge (Anderson, 1982). We may also ask if individual differences result from biological bases for learning, or from socially-influenced exposure to learning opportunities: each level of explanation poses further questions. Strategy choice and adaptation PE effects on processing may derive not from architecture but from strategy choice. Architectural accounts of strategy implementation which describe specific executive functions (e.g. Shallice, 1988) are important but incomplete. We need also to address knowledge level questions concerning the person's goals, and choice of strategy to meet those goals. Again, answers generate new questions. How has the person acquired the goals concerned? How does the person's knowledge of strategies, such as strategy efficacy in the current context, feed into strategy choice? At one level we can answer such questions through addressing the cognitive and social factors which influence motivations and associated learning (e.g. Bandura, 1977). Understanding strategy choice may requires understanding of the Shaping of cognition within the wider social matrix, through the person's attempts to meet social norms, negotiate shared identities with others, and generally adapt to social demands (Hampson, 1988). Mayer, Frasier Chabot and Carlsmith (this volume) provide a detailed discussion of the inter-relationship between motivation, emotion and cognition. A radically different perspective is provided by evolutionary psychology (Tooby & Cosmides, 1992). The person's most important life goals are influenced by the set of genetically programmed mechanisms for solving specific evolutionary problems. Some proximate goals such as "stay warm" may be directly coded. More generally, the individual's goals are indirectly influenced by the structuring of experience imposed by the set of adaptive

12

Chapter 1

mechanisms, which, at the least, is likely to signal that certain types of stimuli and encounters are of special significance. In particular, the motivations which tend to accompany emotional states (e.g. avoidance as a correlate of anxiety) are likely to reflect adaptive pressures. There is an argument too that specific strategies, such as the decision rules used in "Prisoner's Dilemma" social encounters (Ketelaar & Clore, this volume), may be directly encoded (Cosmides & Tooby, 1992). However, the evolutionary psychologist's description of a "strategy" carries no commitment to a particular informationprocessing mechanism. The strategy might be implemented through architecture, or, alternatively, through cxxling for motivational factors. Evolutionary psychologists have perhaps shown insufficient interest in whether strategies in the evolutionary sense are contingent upon implementation of strategies in the information-processing sense previously defined. Strategies for processing reflect voluntary control and potentially complex, contingent decisions which may not be related to geneticallyprogrammed adaptations in any simple way. Evolution is an essential part of the backdrop to understanding the inter-relationship of PE and cognition, but it is simplistic to imagine that every such relationship may be traced back to the operation of an adaptive mechanism (of. Lazarus, 1991). Two qualifications are required here. First, definitions of "adaptation" differ confusingly. To evolutionary psychologists the term refers to genetically-programmed mechanisms. I prefer Lazarus' (1991) broader usage of the term to refer to any attempt to manage the demands and opportunities of an environment, which leaves open the utility of an evolutionary analysis. I will use "adaptation" subsequently in this broad sense, unless otherwise indicated. Second, in emotion research especially, it is important to distinguish explanations for emotion as a human characteristic from individual differences in emotion and associated behaviour. Adaptive explanations at the species level do not necessarily generalise to explanations for individual differences. In summary, processing models are only the beginning of the cognitive science enterprise. For further explanation, we may look either towards a reductionist approach of seeking PE effects on the cognitive architecture, which may be supported by neural mechanisms. Alternatively, we may adopt a more systems-orientexl holistie approach of establishing strategy effects, and their role in the person's adaptation to the physical and social environment. We may also nee~ to consider how neural systems, processing and motivations have been shaped by evolution. The new evolutionary psychology provides a different kind of adaptive, knowledge-level explanation to that

G. Matthews

13

afforded by motives for personal strategy choice. Next, the prospects for developing these complementary levels of explanation are discussed further. Towards a Cognitive Neuroscience of Personality and Emotion?

Investigations of the neuroscience of PE have been dogged by two fundamental problems: the use of over-generalised constructs, exemplified by general arousal theory, and nagging doubts about the causal status of physiological constructs. Criticisms of arousal theory are familiar. In brief, there are four sources of difficulty (Matthews & Amelang, 1993). Empirical criticisms focus on the failure of arousal theory predictions: the supposed inverted-U relationship between arousal and performance is simply not robust (Matthews, 1985; Neiss, 1988). Methodological criticisms relate to weaknesses in inference from empirical data, such as the difficulty in falsifying arousal theory within typical stressor-interaction designs (Hockey, 1984). Psychometric criticisms point to the failure of alternative arousal measures to intercorrelate, implying that the construct cannot be operationalised (Lacey, 1967). Conceptual criticisms concern the construct validity of "arousal" and "performance", both of which are multi-faceted (Hockey, 1984; Robbins, 1986). Hockey's cognitive critique of arousal theory is especially important: "arousal" effects vary across stressors and processing functions, and are often associated with subtle strategic effects rather than changes in parameters of the architecture. None of these considerations rule out the possibility of a better arousal theory. Such a theory would require the discrimination of different circuits whose overall activity might influence processing, a description of the specific information-processing functions sensitive to each circuit, and satisfactory methods for manipulating and measuring these multiple arousal dimensions independently. Various multi-dimensional arousal theories have been proposed (e.g. Sanders, 1990), but none have succeeded in explaining more than a small part of the empirical data. A severe barrier to theory development is the sheer complexity and interactivity of neural systems. In the personality context, Zuckerman (1991) points out that there is no one-toone mapping between neural systems and personality traits. He sees each trait as supported by several systems, and, conversely, each system feeds into several traits. Hence, even if neurological reductionism is correct in principle, it may be difficult to establish in research practice. The other basic criticism of the psychobiological enterprise may be traced back to peripheralist views of emotion and the Jamesian view that

14

Chapter I

emotion derives from perceptions of physiological reactions, perhaps through the appraisal and evaluation of autonomic nervous system activity (Schachter & Singer, 1962). The logic of this approach may be extended by denying physiological reactions any special status. Emotions may be constructed from appraisal of a variety of cues, from the external physical and social environment, as well as from physiological reactions. In contemporary research, this position has been expressed most forcefully by Lazarus (1984, 1991) who argues that the influence of physiology is always shaped by appraisal and cognition. Lazarus (1991) does suggest that there may be qualitatively different types of appraisal, trading off speed of processing against depth and complexity, which might be loosely associated with different brain structures. However, explaining how different modes of appraisal influence emotion is a cognitive- rather than a brain-level question: the distinction is between two different cognitive modules. In terms of the current framework, the explanatory questions are how the architecture supports different types of appraisal, and how appraisal and emotion are driven by adaptation to the environment. Van Reehan and Scherer (this volume) argue that multiple levels of processing must be distinguished in relating appraisal to emotion and brain mechanisms. Despite the difficulties outlined, there are several promising lines of research which elucidate mappings between brain and cognitive processes. One approach is the fine-grained analysis of neural pathways (e.g. Banquet et al., this volume; Gray, 1982; LeDoux, 1995). Gray's (1982) account of the septo-hippocampal system (SHS) as the basis for anxiety and behavioural inhibition demonstrates the potential of this approach. He explicitly describes the SHS as performing processing functions, such as calculation of the mismatch between current sensory events and expectancy. Processing is mapped onto brain circuitry to an impressive degree. However, as with other animal models, fundamental questions concerning the coding of information are left open. Some system components are clearly non-propositional, such as the "enabling signal" which gates output from the SHS, and biasing effects of ascending afferents associated with arousal. The system must also make and verify predictions about the world, a process which, in humans, we might imagine to be prepositionally coded. Rats and people may process information differently, of course, but, in any case, it is difficult to develop the theory as a cognitive account of human emotion when the computational basis for comparator function is uncertain. Predictions from Gray's theory have met with mixed success, in part, because of difficulties in operationalising its constructs in human subjects (Pickering, Diaz & Gray,

G. Matthews

15

1995). It has been most powerful when modified through integration with human cognitive neuroscience (see Derryberry & Reed, this volume). Perhaps the most promising solution to the coding problem is the use of connectionist models (see Banquet et al., this volume). It is emphasised that connectionist models do not necessarily correspond to actual neural net processes: Smolensky (1988) describes a variety of important differences between neural functioning and the connectionist architectures typically applied to psychological problems. However, connectionist models do possess some of the key formal properties of nerve cell assemblies. They comprise linked elementary processing units representing analogue information only ("activation"), which is transmitted through associative pathways. There is no direct representation of symbols, which is assumed to be distributed across units, and learning is a direct consequence of the formal properties of the net. Modelling allows testable predictions to be derived concerning behavioural consequence of neural function, as illustrated by Cohen and ServanSchreiber's (1992) work on the consequences of abnormality in dopamine function for attention in schizophrenics. However, connectionism is scientifically valuable irrespective of whether activation corresponds directly to neural functions such as rate of firing; typically, activation is best treated as a formal attribute of the cognitive architecture (of. J.R. Anderson, 1990).

Developing Adaptive Explanations Knowledge level explanations in P E research address questions of adaptation. I will take an "adaptive explanation" as a demonstration that expressions of emotion or personality arc functionally useful in achieving personal goals or dealing with environmental demands. Traditionally, personality theory has been much concerned with the challenges posed by the interplay between basic drives such as sex and power-seeking in an often threatening and unaccommodating world, as expressed in various psychodynamic theories. Although "adaptive", such explanations are unsatisfactory because of their failure to specify mechanisms in testable form (e.g. Popper, 1957). In contemporary emotion research, Lazarus (1991) places adaptation at the heart of emotion processes" emotions map onto "core relational themes" describing the adaptational relationship between person and environment. Somewhat similarly, social-cognitive approaches to personality are concerned with the person's strivings to implement "personal projects" through interaction with the social environment (Cantor & Zirkel, 1990). Such theories are testable, by and large, but rarely computational.

16

Chapter 1

Contemporary research on adaptive models of PE is open to the criticism that it is poorly integrated with information-processing models. How can such an integration be effected in future research? The processing construct bridging architectural and knowledge levels of explanation is strategy. We can describe strategies in terms of processing constructs such as selection of processing codes, criterion-setting and so forth. Understanding of strategy use also requires understanding of why the person chooses one strategy over another; the motivational guidance of strategy choice. Thus, adaptation understood cognitively refers most straightforwardly to the acquisition, selection and implementation of computationally-specified strategies which aim to facilitate the person's goals within a given environment. The primary source of data is then experimental and simulation studies which allow computional models of strategy use to be developed.

Transient adaptation and strategy selection Explanations for experimental data require an understanding of how and why PE factors are related to strategy selection. For example, much recent research on distressing environmental stressors such as loud noise suggests that their effects on strategy are often more pronounced than effects on basic structural parameters of the processing system (Hockey, 1986). Noise appears to enhance use of the dominant strategy for performance, whereas fatigue is associated with a switch to low-effort strategies. Ecological theories of stress (Hancock & Warm, 1989) see behaviour in performance contexts as driven both by strivings to perform well and strivings to maintain a comfortable task load. Negative emotion and performance degradation are influence~ by the success or failure of the strategies which implement such motivations (see Kluger & DoNisi, 1996). Strategy choice under environmental stress reflects the subject's immediate motivations, beliefs about the personal significance of the stressor, and beliefs about the efficacy of strategy use in meeting salient goals. Beliefs vary dynamically, and perhaps even on a trial-to-trial basis, as the person modifies strategy in response to error feedback (cf. Rabbitt, 1979). These "state" variables are influenced by "trait" representations in LTM of the person's goals and general beliefs relevant to the particular situation (Matthows & Wells, 1996). For example, detrimental alter-effects of noise on performance may derive from reduced use of active coping strategies, resulting from appraisals of the stressor/task environment as uncontrollable and the limited relevance of the laboratory situation to personal goals (see Cohen, 1980). Individual

G. Matthews

17

differences in susceptibility to noise may reflect the individual's beliefs about the threat and controllability of noise stimuli (Jones, 1984).

Stabilities of adaptation There are different timescales for adaptation (Revelle, 1993). In addition to "single-occasion" instances of strategy-driven behaviour, there are stabilities of adaptation associated with PE evident over periods up to a single life time (see Mayer et al., this volume). The key question here is the nature of the representation which maintains stability, and there are several options. Emotion effects on performance may often be somewhat context-specific, and contingent upon context-bound appraisals and motivations (Matthews, Sparkes, & Bygrave, 1996). At the same time, data from widely diverse contexts suggests that emotions such as anxiety and depression may have some cognitive correlates which are intrinsic to the emotional state (Martin & Jones, 1995), or at least prototypical of the emotion. Oatley and JohnsonLaird's (1987) hypothesis that emotions signal the status of current action plans implies a degree of context-independence. Sadness indicates failure of a major plan (a description of adaptive status), which in turn constrains cognitions and action. As Lazarus (1991) states, sadness is associated with appraisals of irrevocable loss, and an action tendency for withdrawal from the environment, so that a given emotion entails a given representation of adaptive status. The basis of emotions in adaptive status forces at least some consistency in emotion-cognition relationships across individuals and occasions, despite the influence of contextual factors. Similarly, personality traits may reflect stabilities of adaptation. Matthews and Dora (1995) present an adaptive account of the diversity of independent information-processing functions associated with traits such as extraversion and neuroticism. They argue that personality traits represent fitnesses for adapting to certain kinds of environment, defined in terms of their informational properties. Cognitive correlates of traits provide the building blocks for acquisition of the skills necessary for success in the environments concerned. For example, extraverts are adapted to environments characterised by high information flows, including social environments (see Matthews, this volume). Correlates of extraversion such as high STM capacity, low response criterion and efficient dual-task performance facilitate the development of skills and strategies for handling rapidly-changing inputs. Viewed in terms of information-processing alone, extraversion is associated with an arbitrary collection of cognitive correlates. The link between

18

Chapter I

processing and the central characteristics of extraversion, such as impulsivity and sociability, requires the adaptive perspective. Representation of the adaptive potentials associated with traits is distributed over a number of distract processing characteristics.

Genetic basesfor adaptation The final element of adaptive explanations is the evolutionary perspective, operating over a time scale of many lifetimes. At the species level Darwin recognised that emotional responses may be understood at the species level in terms of their functional properties in aiding survival and reproduction. Evolutionary psychologists argue that emotions solve the regulatory problems posed by the need to coordinate multiple processing modules to handle imperative situations (Tooby & Cosmides, 1992). Instructions for building modules during development are represented within the genes. Modules may then be characterised both eomputationally and in terms of their design for solving adaptive problems. Perhaps a more contentious question is how individual differences in genotype are expressed as individual differences in module functioning. Despite the controversial nature of the research, there is now convincing evidence from structural modelling of behaviour genetic data to suggest that major personality traits such as negative emotionality are partially inherited (Loehlin, 1992), and the beginnings of a molecular genetics of personality are emerging (e.g. Leseh et al., 1996). The thinking of psychobiological researchers often seems unduly linear: the implicit model seems to be that the random outcomes of the genetic dice feed forward powerfully into personality, with perhaps a little modification by gene-environment interaction. This model leads naturally to the naive good genes/bad genes perspective previously criticised. It is hardly possible to estimate the selection pressures on the various traits. However, even traits which are socially devalued, such as neuroticism and psychoticism presumably have adaptive value in some circumstances, or the genes coding for them would have been selected out. Matthews and Dora (1995) argue that neuroticism is adaptive when the environment is characterised by disguised or subtle threats, especially social threats. Similarly, psychoticism may facilitate creativity (Eysenck, 1995), perhaps through attentional mechanisms such as those described by Beech and Williams (this volume). Thus, while genes may feed forward into the cognitive correlates of traits at the level of the individual, the cognitive

G. Matthews

19

components of traits represent feedback from the environment over many generations. If a person is to function as an extravert, by relying on social interaction to promote survival, for example, then those cognitive characteristics supporting social interaction skills will be selected for. This process m turn entails selection for the neural net parameters associated with the cognitive characteristics. The patterning of cognitive/neural functions associated with traits represents, in part, the toolkit of functions required for adapting to the environments associated with the trait (Matthews, 1997). Thus, natural selection links the adaptive and biological levels of explanation: individual differences in brain functioning support individual differences in choice of environment. We can reconceptualise the ladder of explanation as a loop, as shown m Figure 2, with connectionist networks, strategies and natural selection as the key constructs bridging the levels of explanation. Adaptation, in the broad sense, is not solely driven by natural selection, of course. Learned adaptations may be equally or more important, although it is uncertain how much learning influences basic parameters of neural net functioning. The present account emphasises the importance of skills rather than processing components in determining adaptation. Good STM for words does not necessarily assist a person to function as an extravert, but being able to remember ongoing conversations most likely does. Skills must be learnt, a process which reflects the interaction between the person's choice of strategies for acquiring knowledge (knowledge level) and the processing routines which implement learning (architectural level).

Adaptation (environmental fitness)

I

Strategies (performance and/earning)

Information processing

Knowledge

Natural selection

Architecture

Connectionism

~--

Biology

t Neuroscience

Figure 2. Levels of explanation reconccptualised as a loop.

20

Chapter 1

An Example: Explaining Anxiety and Cognition In discussing levels of explanation for PE phenomena, quite a lot of ground has been covered, and the scope for confusion and over-complexity in explanation will be evident. What the researcher must do, of course, is to select levels of explanation appropriate for the research problem at hand. In this section, I illustrate the application of the cognitive science approach to explaining relationships between anxiety and attention. The phenomena to be explained are well-known: impairment of attention, bias of selective attention to threat stimuli, and the relationship between abnormality in attentional function and clinical anxiety disorder (see Wells & Matthews, 1994, for a review). There are a variety of well-regarded information-processing models in this area (e.g. Bower, 1981; Ingram, 1984; Williams, Watts, MacLeod, & Mathews, 1988). Matthews and Wells (in press a) address the question of how we can go beyond the information-processing description of phenomena to explain associations between emotion and attentional functioning, and their implications for clinical disorder.

Anxiety and information-processing The first step is to decide what kind of explanation is sought. There is a psychobiology of anxiety-related bias but it has proved difficult to integrate with studies of selective attention in humans (Wells & Matthews, 1994, pp. 325-332). For example, Gray's (1982) SHS influences attention to threat stimuli (punishment cues), but anxiolytic drugs which act on the SHS fail to influence attcntional bias on the emotional Stroop test (Golombok, Stavrou, & Bonn, 1991). Thus, while acknowledging that biological (and evolutionary) factors may be important, the most straightforward approach is to focus on the architectural and knowledge levels. The next step is to characterise the performance correlates of anxiety in processing terms. The central architectural issue here is the extent to which anxiety influences strategic and/or automatic processing. The distinction between plan-driven strategic control and stimulus-driven "automatic" control of processing has been developed in considerable detail (Norman & Shallice; 1985). Anxiety might influence both the processing routines implementing strategic or executive control, and parameters of involuntary processing. Matthews and Wells (m press b) review the evidence on the automaticity of attentional bias, and conclude that bias is predominantly strategic. There is considerable evidence for context-sensitivity of bias (e.g. Calvo & Castillo,

G. Matthews

21

1997), even with subliminal stimuli (Fox, 1996). Similarly, deficits evident on tasks with neutral stimuli, demonstrated in test anxiety research (Samson et al., 1995), appear to be associated with loss of attentional resources or working memory (Eysenck, 1992), constructs associated with strategic rather than automatic processing. The clinical literature too tends to emphasise the strategies that anxiety patients develop for interpreting and coping with a world appraised as threatening (Beck, Emery, & Greenberg, 1985; Wells, 1995). Put differently, people with anxiety traits have developed "skills" for handling threat, which are sometimes maladaptive. One effect of state anxiety may be to bias retrieval of the processing routines controlling these skills. Matthews and Harley (1996) investigated the computational basis for attentional bias using a connectionist simulation of the emotional Stroop. The network was trained to discriminate colour and semantic inputs using the backpropagation algorithm. Bias towards negative emotion semantic content was introduced through various mechanisms, and the performance of the network compared with real data. The most satisfactory mechanism was a strategic one: low-level activation of a "threat-monitoring" task demand unit during colour-naming and word reading. In other words, strategic processes modulate the spread of activation from input to output units. "Automatic" mechanisms, such as sensitivity of input units to negative stimuli, and overlearning of response to negative stimuli, generated patterns of performance incompatible with real data. Siegle and Ingram (this volume) and Tryon (this volume) discuss alternative connectionist architectures for modelling phenomena relating to negative emotion. Investigation of the underlying architecture through experiment and simulation suggests that bias is more than just an "accidental" over-sensitivity of automatic threat-processing mechanisms. However, various explanatory questions are left open. It is conceivable that the primary consequences of anxiety are architectural, such as loss of resources, and anxiety effects on strategy are an attempt to "work around" these limitations. Alternatively, anxiety may not affect the architecture at all, but, instead, it influences personal goals and motivations which directly impinge on strategy choice and acquisition of threat-management skills. Questions also remain about the inter-relationship of the various performance correlates of anxiety, which are sufficiently diverse that multiple processing mechanisms are likely to be involved (Eysenck, 1992). Diversity in component processes may be associated with unity at the knowledge level (Matthews & Dom, 1995). For example, the various processing characteristics of anxiety may all subserve an overall orientation towards hypervigilance (Eysenck, 1992).

22

Chapter 1

A multi-level explanatory model

The Wells and Matthews (1994, 1996) Self-Referent Executive Function (S-REF) model of attention and negative emotion integrates architectural and knowledge levels of explanation within a multi-level framework. Three main components of the architecture are distinguished: stable self-knowledge encoded in LTM in procedural form, stimulus-driven automatic processing networks, and a supervisory executive. In response to various internal and external threat stimuli, the executive retrieves generic procedures for coping with threat from LTM, and tailors them on-line to the specific demands of the situation. As in most models of this kind (e.g. Norman & Shallice, 1985), routines under executive control influence behaviour indirectly, though biasing automatic processing. In the S-REF configuration, operation of the executive is characterised by self-focus of attention, cognitive interference generated by worry, and the pursuit of self-regulative goals, such as maintaining self-esteem. The S-REF model also emphasises the dynamic interplay of components: self-knowledge drives processing of threat stimuli, but is itself often modified by self-appraisal. Clinical disorder is generally associated with dynamic disturbances, such as perseverative cycles of rumination which fail to modify self-beliefs adaptively (of. Siegle & Ingram, this volume; Tryon, this volume). Within the model, architectural and knowledge levels of understanding are linked through coping strategies (Matthews & Wells, 1996). The knowledge level specifies the personal goals and beliefs about goal attainment which influence strategy choice. For example, generalised anxiety patients are motivated to protect themselves against various (often unrealistic) threats, and they hold the metaeognitive belief that worry is a successful strategy for so doing (Wells, 1995). The architectural level delineates the specific processing routines which implement coping. The S-REF model makes two general statements about processing in distress states, consistent with empirical evidence reviewed by Wells and Matthews (1994). First, processing activities associated with worry tend to interfere with both the internal operations of the executive system, such as formulating coping strategies, and with implementing and regulating the strategies themselves, if they are attentionally demanding. Second, although there is considerable variability in coping, distressed individuals often choose the task-focused strategy of monitoring for threats congruent with personal concerns. Threat monitoring (which is voluntarily initiated but not necessarily fully conscious) is responsible for emotional Stroop effects. It remains for future research to

G. Matthews

23

determine the specific processing routines involved: the Matthews and Harley (1996) simulations illustrate how this might be done computationally. Figure 3 summarises levels of explanation for inter-relationships between cognition and anxiety (and other negative emotions). The three classical levels of explanation provide alternative ways of describing anxiety phenomena. At the knowledge level, anxiety relates to self-knowledge and goals, as in Beck et al.'s (1985) schema theory. There may also be anxiety effects on processing specified at the architectural level (de-emphasised within the S-REF model). A full account must accommodate the neuroscience of anxiety, which is becoming increasingly integrated with architectural descriptions (see Derryberry & Reed, this volume; Kitayama, this volume). In this section, we have argued that deeper understanding is obtained through use of constructs which bridge the levels, especially strategies which control the use of the architecture to serve personal goals, and neural nets which describe processing phenomena using constructs broadly compatible with neurophysiology. To the extent that anxiety is genetically-influenced, we need also to consider how the neural basis of anxiety has developed through natural selection. Anxious individuals are sensitive to threat stimuli, but often they are conspicuously poor at handling the demands of threatening

Threat-driven self-regulation

Coping strategies - threat monitoring rumination

I ~-

Knowledge

Genetic adaptation to environments characterised by subtle threats

Attentional processes - r e s o u r c e loss

-

Architecture

- bias, etc.

Neural net parameters - e.g. activation of threat monitoring units

Biological

I Cortical and s u b c o r t i c a l circuits a c t i v a t e d b y t h r e a t stimuli

Figure 3. Lcvds of explanation for associations between anxiety and cognition.

24

Chapter 1

environments. Matthews and Dom (1995) argue that the processing correlates of anxiety serve the adaptive goal of maintaining vigilance for subtle or disguised threats (especially social threats), and neural correlates of anxiety may have evolved for this purpose. Conclusions

I have suggested that the multi-faceted emerging cognitive science of personality and emotion may be clarified by distinguishing informationprocessing models from explanations of the phenomena those models describe. Sperry (1993) has claimed that cognitive science introduces a new model of causal determinism, combining traditional microdeterminism with the top-down influences of emergent, macro mental state variables. Consistent with this view, reductionistic and holistic explanatory strategies may be distinguished in PE research. Reductionism requires a focus on the transient (state) or fixed (trait) differences in cognitive architecture which may be associated with emotion and personality factors. Architectural differences may in turn be traced to properties of neural circuits. For reductionism to be scientifically valid, the mappings between these different levels must be sufficiently simple that novel, testable predictions of behaviour may be derived from theory. Predictions include those derived from connectionist models, which may provide an important bridge between neural and architectural levels of explanation. The range of phenomena open to cognitive neuroscience explanation remains to be determined. The alternative, holistic approach to explanation seeks to explore the adaptive basis of emotion and personality, in the broad sense proposed by Lazarus. We require an understanding of how state and trait characteristics subserve the goals associated with emotions and personality. That is, the functional design of the processing system for implementing and acquiring contextualised skills may vary across individuals and across occasions. Cognitive science requires that adaptive explanations are linked to computational accounts of phenomena. Over short time-scales, the link may be achieved through specifying the strategies which allow goals to be met through implementing specific processing routines. Over longer time scales, there are several approaches to explaining stabilities of adaptation. First, representations of genetic strategies in LTM may drive consistency in computation. Second, representations of adaptive status may be intrinsic to emotional states. Third, personality traits may be associated with bundles of

G. Matthews

25

relatively stable, functionally independent computational characteristics which support successful adaptation to specified environments. Finally, both reduetionist and holistic explanations may feed into evolutionary explanations. To the extent that reductionism results in neural accounts of personality and emotion, evolutionary psychology may explain how the brain systems concerned have been shaped by the pressures of natural selection. In addition, the person's goals, and/or the strategies available for satisfying those goals, may be directly or indirectly related to genetically-programmed adaptive mechanisms, it is likely that the adaptive characteristics of personality and emotion reflect some complex interplay between social learning and genetics. However, as the example of anxiety research shows, it is wise to be selective in choosing levels of explanation. Different levels within the overall cognitive science framework are appropriate to different problems in personality and emotion research. References

Anderson, J. R. (1982). Acquisition of cognitive skill. Psychological Review, 89, 369-406. Anderson, J. R. (1990). The adaptive character of thought. Hillsdale, NJ: Erlbaum. Anderson, K. J. (1990). Arousal and the inverted-U hypothesis: A critique of Neiss's "Reconceptualizing Arousal". Psychological Bulletin, 107, 96100. Bandura, A. (1977). Social learning theory. Englewood Cliffs, NJ: PrenticeHall. Bargh, J. A., Chaiken, S., Govender, R., & Pratto, F. (1992). The generality of the automatic attitude activation effect. Journal of Personality and Social Psychology, 62, 893-912. Beck, A. T. (1967). Depression: Causes and treatment. Philadelphia: University of Pennsylvania Press. Beck, A. T., Emery, G., & Greenberg, R. L. (1985). Anxiety disorders and phobias: A cognitive perspective. New York: Basic Books. Bower, G. H. (1981). Mood and memory. American Psychologist, 36, 129148. Calvo, M. G., & Castillo, M. D. (1997). Mood-congruent bias in interpretation of ambiguity: Strategic processes and temporary activation. Quarterly Journal of Experimental Psychology, 50A, 163182.

26

Chapter I

Cantor, N., & Zirkel, S. (1990). Personality, cognition, and purposive behavior. In L. A. Porvin (Ed.), Handbook of personality: Theory and research. New York: Guilford. Cohen, J. D., & Sorvan-Sehreibor, D. (1992). Context, cortex and dopamine: A conne~ionist approach to behavior and biology in schizophrenia. Psychological Review, 99, 45-77. Cohen, S. (1980). After effects of stress on human performance and social behavior: A review of research and theory. ,Psychological Bulletin, 88, 82-108. Cosmides, L., & Tooby, J. (1992). Cognitive adaptations for social exchange. In J. H. Barkow, L. Cosmides & J. Tooby (Eds.), The adapted mind: Evolutionary psychology and the generation of culture. Oxford: Oxford University Press. Costa, P. T., Jr., & MeCrae, R. R. (1992). Four ways five factors are basic. Personality and IndivMual Differences, 13, 653-665. Ellis, A. (1962). Reason and emotion in psychotherapy. New York: Lyle Smart. Eysenck, H. J. (I 995). Creativity as a product of intelligence and personality. In D. H. Saklofske & M. Zeidner (Eds.), International handbook of personality and intelligence. New York: Plenum. Eysenck, H. J., & Eysenck, M. W. (1985). Personality and individual differences: A natural science approach. New York: Plenum. Eysenck, M. W. (1992). Anxiety: The cognitive perspective. Hillsdale, NJ: Erlbaum. Fox, E. (1996). Selective processing of threatening words in anxiety: The role of awareness. Cognition and Emotion, 10, 449-480. Gazzaniga, M. S. (1994). Nature's mind: The biological roots of thinlang, emotions, sexuality, language and intelligence. Harmondsworth: Penguin. Golombok, S., Stavrou, A., & Bonn, J. (1991). The effects of diazepam on anxiety-related cognition. Cognitive Therapy and Research, 15, 459467. Gray, J. A. (1982). The neuropsychology of anxiety: An enquiry into the functions of the septo-hippocampal system. Oxford: Oxford University Press. Hampson, S. E. (1988). The construction of personality (2nd ed.). London: Routledge. Hancock, P. A., & Warm, J. S. (1989), A dynamic model of stress and sustained attention. Human Factors, 31, 519-537.

G. Mattnews

zI

Hockey, G. R. J. (1984). Varieties of attentional state: The effects of the environment. In R. Parasuraman & D. R. Davies (Eds.), Varieties of attention. New York: Academic. Hockey, G. R. J. (1986). A state control theory of adaptation to stress and individual differences in stress management. In G. R. J. Hockey, A. W. K. Gaillard, & M. G. H. Coles (Eds.), Energetics and human information processing. Dordrecht: Martmus Nijhoff. Humphreys, M. S., & Revelle, W. (1984). Personality, motivation and performance: A theory of the relationship between individual differences and information processing. Psychological Review, 91, 153-184. Ingram, R. E. (1984). Toward an information-processing analysis of depression. Cognitive Therapy and Research, 8, 443-478. Jones, D. M. (1984). Individual and group differences in the response to noise. In D. M. Jones & A. J. Chapman (Eds.), Noise and society. New York: Wiley. Kanfer, R., & Ackerman, P. L. (1989). Motivation and cognitive abilities: An integrative/aptitude-treatment interaction approach to skill acquisition. Journal of Applied Psychology, 74, 657-690. Klein, S. B., & Loflus, J. (1988). The nature of self-referent encoding: The contributions of elaborative and organizational processes. Journal of Personality and Social Psychology, 55, 5-11. Kluger, A. N., & DeNisi, A. (1996). The effects of feedback interventions on performance: A historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin, 119, 254-284. Lacey, J. I. (1967). Somatic response patterning and stress: Some revisions of activation theory. In M. H. Appleby & R. Tumbull (Eds.), Psychological stress. New York: Appleton-Century-Crofts. Lazarus, R. S. (1984). On the primacy of cognition. American Psychologist, 37, 1019-1024. Lazarus, R. S. (1991). Emotion and adaptation. Oxford: Oxford University Press. Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal and coping. New York: Springer. LeDoux, J. E. (1995). Emotion: Clues from the brain. Annual Review of Psychology, 46, 209-235. Lesch, K. -P., Bengel, D., Heils, A., Sabol, S. Z., Greenberg, B. D., Petri, S., Benjamin, J., Miiller, C. R., Hamer, D. H. & Murphy, D. L. (1996). Association of anxiety-related traits with a polymorphism in the serotonin transporter gene regulatory region. Science, 274, 1527-1531.

28

Chapter 1

Loehlin, J. C. (1992). Genes and environment in personality development. Newbury Park, CA: Sage. Martin, M., & Jones, G. V. (1995). Integral bias in the cognitive processing of emotionally linked pictures. British Journal of Psychology, 86, 419436. Matthews, G. (1985). The effects of extraversion and arousal on intelligence test performance. British Journal of Psychology, 76, 479-493. Matthews, G. (1992). Extraversion. In A. P. Smith & D. M. Jones (Eds.), Handbook of human performance. Vol. 3: State and trait. London: Academic. Matthews, G. (1997). Intelligence, personality and information-processing: An adaptive perspective. In W. Tomic & J. Kingsma (Eds.), Advances in cognition and educational practice (Vol. 4), pp. 475-492. Greenwich, CT: JAI Press. Matthews, G., & Amelang, M. (1993). Extraversion, arousal theory and performance: A study of individual differences in the EEG. Personality and Indi~dual Differences, 14, 347-364. Matthews, G., & Deary, I. J. (in press). Personality traits. Cambridge: Cambridge University Press. Matthews, G., & Dorn, L. (1995). Cognitive and attentional processes in personality and intelligence. In D. H. Saklofske & M. Zeidner (Eds.), International handbook of personality and intelligence. New York: Plenum. Matthews, G., & Harley, T. A. (1993). Effects of extraversion and self-report arousal on semantic priming: A connectionist approach. Journal of Personality and Social Psychology, 65, 735-756. Matthews, G., & Harley, T. A. (1996). Connectionist models of emotional distress and attentional bias. Cognition and Emotion, 10, 561-600. Matthews, G., Sparkes, T. J., & Bygrave, H. M. (1996). Stress, attentional overload and simulated driving performance. Human Performance, 9, 77-101. Matthews, G., & Wells, A. (1996). Attentional processes, coping strategies and clinical intervention. In M. Zeidner & N. S. Endler (Eds.), Handbook of coping: Theory, research, applications. New York: Wiley. Matthews, G., & Wells, A. (in press a). The cognitive science of attention and emotion. In T. Dalgleish & M. Power (Eds.), Handbook of cognition and emotion. New York: Wiley.

G. Matthews

29

Matthews, G., & Wells, A. (in press b). Attention, automaticity and affective disorder. Behavior Modification. Neiss, R. (1988). Reconceptualizing arousal: Psychobiological states in motor performance. Psychological Bullean, 103, 345-366. Neisser, U. (1976). Cognition and reality. San Francisco: Freeman. Norman, D. A., & Shallice, T. (1985). Attention to action: Willed and automatic control of behaviour. In R. J. Davidson, G. E. Schwartz & D. Shapiro (Eds.), Consciousness and self-regulation: Advances in research (Vol. 4). New York: Plenum. Oatley, K., & Johnson-Laird, P. (1987). Towards a cognitive theory of emotions. Cognition and Emotion, 1, 29-50. Popper, K. (1957). The poverty of historicism. London: Routledge & Kegan Paul. Pickering, A. D., Diaz, A., & Gray, J. A. (1995). Personal@ and reinforcement: An exploration using a maze-learning task. Personality and Individual Differences, 18, 541-558. Pinker, S. (1994). The language instinct. Harmondsworth: Penguin. Pylyshyn, Z. W. (1984). Computation and cognition: Toward a foundation for cognitive science. Cambridge, MA: MIT Press. Rabbitt, P. M. A. (1979). Current paradigms and models in human information processing. In V. Hamilton & D. M. Warburton (Eds.), Human stress and cognition: An information processing approach. London: Wiley. Revelle, W. (1993). Individual differences in personality and motivation: "Non-cognitive" determinants of cognitive performance. In A. Baddeley & L. Weiskrantz (Eds.), Attention: Selection, awareness and control Oxford: Oxford University Press. Robbins, T. W. (1986). Psychopharmacological and neurobiological aspects of the energetics of information processing. In G. R. J. Hockey, A. W. K. Gaillard, & M. G. H. Coles (Eds.), Energetics and human information processing. Dordrecht: Martinus Nijhoff. Sanders, A. F. (1990). Issues and trends in the debate on discrete versus continuous processing of information. Acta Psychologica, 74, 123-167. Sarason, I. G., Sarason, B. R., & Pierce, G. R. (1995). Cognitive interference: At the inteUigenee-personality crossroads. In D. H. Saklofske, D. H., & M. Zeidner, M. (Eds.), International handbook of personality and intelligence. New York: Plenum. Schachter, S., & Singer, J. E. (1962). Cognitive, social, and physiological determinants of emotional state. Psychological Review, 69, 379-399.

30

Chapter I

Schwartz, S. (1975). Individual differences in cognition. Journal of Research in Personality, 9, 217-225. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press. Simon, H. A. (1967). Motivational and emotional controls of cognition. Psychological Review, 74, 29-39. Smolensky, P. (1988). On the proper treatment of eonnectionism. Behavioral and Brain Sciences, 11, 1-74. Sperry, R. W. (1993). The impact and promise of the cognitive revolution. American Psychologist, 48, 878-885. Spielberger, C. D. (1966). The effects of anxiety on complex learning and academic achievement. In C. D. Spielberger (Ed.), Anxiety and behavior. London: Academic Press. Tooby, J., & Cosmides, L. (1992). The psychological foundations of culture. In J. H. Barkow, L. Cosmides & J. Tooby (Eds.), The adapted mind: Evolutionary psychology and the generation of culture. Oxford: Oxford University Press. Watson, D., & Clark, L. A. (1992). On traits and temperament: General and specific factors of emotional experience and their relation to the fivefactor model. Journal of Personality, 60, 441-476. Wells, A. (1995). Meta-eognition and worry: A cognitive model of generalised anxiety disorder. Behavioural and Cognitive Psychotherapy, 23, 301-320. Wells, A., & Matthews, G. (1994). Attention and emotion: A clinical perspective. Hove: Edbaum. Wells, A., & Matthews, G. (1996). Modelling cognition in emotional disorder: The S-REF model. Behaviour Research and Therapy, 34, 881888. Williams, J. M. G., Watts, F. N., MacLeod, C., & Mathews, A. (1988). Cogniave psychology and emotional disorders. Chiehester: Wiley. Zuckerman, M. (1991). Psychobiology of personality. Cambridge: Cambridge University Press.

Cognitive Science Perspectives on Personality and Emotion -G. Matthews (Editor) 9 1997 Elsevier Science B.V. All fights reserved. CHAPTER 2

Conation, Affect, and Cognition in Personality John D. Mayer, Heather Frasier Chabot and Kevin M. Carlsmith

During much of the 20th Century, personality psychology has been a field divided into competing schools of psychodynamic, trait, humanistic, and other perspectives, with little communication among perspectives, and no common language. Recently, however, a consensus view of the field has been developing which considers personality from a systems perspective and attends to (a) the location of personality, (b) its parts, (c) its organization, and (d) its development (Mayer, 1993; 1995a, b, Pcrvin, 1980; Sears, 1960). For instance, pcrsonality's location is defined in relation to such neighboring systems as biology and sociology. Personality's parts include components that arc relatively basic such as hunger, happiness, and working memory, and more complex components as well, including extraversion, the self and the ego. Thousands of parts of personality have been proposed (Allport, 1958), and of these thousands, at least 400 parts are regularly discussed (Mayer, 1995b). Keeping 400 parts of personality in mind is a near impossibility, so one alternative strategy is to consider them in groups or classes (e.g., Barratt, 1985, Buss & Finn, 1987; Mayer, 1995a,b). Most classification systems for these components employ one or more of three categories of mind that have a centuries-old tradition: the conaave, affective, and cogmtive - what Hilgard (1980) has referred to as the trilogy of mind. According to this division, c~nation (or motivation) includes components that propel or move the organism such as the hunger drive, and the need for achievement. The affect group, principally containing emotion, includes such basic feelings as anger and happiness, along with related parts such as the mental programs for emotional facial expressmns. The cognition group, containing thought-related processes and mechanisms, includes such elements as worlang memory, judgment, and reasoning. The division of the mind into r affect, and cognition is so embedded in our discipline that many of our journals are named a~er those parts: Cogmtion, Motivation and Emotion, Cogmtion and Emotion, and so on. Despite this, many of us would be hard-pressed to recall the origin of this classification system, or to describe the differences among the three

32

Chapter 2

categories. Along these lines, Henle (cited in Hilgard, 1980, p. 115) remarked: as we become absorbed in our own specialties we often become cryptosystematists, that is, our beliefs are embedded in larger systems of thought that are not explicit but may serve to perpetuate errors. Indeed, the differences among motivation, affect, and cognition can become paper thin. A person's associations to the word "success" may reveal her need for achievement (conative), while also being influenced by her mood (affect), and memory (cognition). To accommodate such blended areas of performance, there exist blended areas of study such as "cognition and affect", and "motivation and emotion." Still, in what sense is one such class of mental process to be distinguished from the others? In this chapter we clarify the meaning of this tripartite division. We will begin by examining a general systems model of personality (already introduced at the outset). This model's further development relies in part on the distinction among classes of conative, affective, and cognitive components. The systems model illustrates how the three spheres of conation, affect, and cognition, can be used to classify aspects of personality psychology. The usefulness of the three spheres, however, relies on a clear understanding of each one's meaning. Following description of the systems model, we focus on conation, affect, and cognition, including (a) their historical origins, (b) their changing description across time, (c) their conceptualization, and (d) a recommended update of their meaning. Finally, we return to questions of conation, affect, and cognition in personality and in contemporary research, and discuss how the trilogy may be integrated into a picture of the person as a whole.

The Relational Model of Personality Several contemporary models of personality employ one or more classes

of conation, affect, and cognition in their construction (e.g., Barratt, 1985; Buss & Finn, 1987; Mayer, 1995a,b). Examination of one such model demonstrates one way the trilogy of mind is used today, and highlights some of the issues surrounding its use. The specific model employed here is the relational model of personality, so-called because personality and its parts are all described in relation to one another and their neighboring

e /

I

-

'

GROUPS INCLUDING OR INTERACTING WITH

PERSONALITY

INTERNAL PERSONALITY

NERVOUS SYSTEM

EXTERNAL SITUATION

SITUATIONAL ELEMENTS

[

Figure 1 . An view of the personality system amidst its neighboring systems, includmg biology, sociology, and situations. A molecular-molar dunension is represented vertically, an internal-external dimension horizontally, and an organismic dependent-constructed dunension depthwise.

,I.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

L

I

33

34

Chapter 2

systems (Mayer, 1995a,b). This relational model is typically developed according to four systems-oriented topics: that is, according to personality's location, components, organization, and development. One aspect of the relational model that makes it particularly worth discussing is its highly integrative aspects; it contains or subsumes several models developed by others (e.g., Buss & Finn, 1987). Certain conceptual dimensions can be employed to distinguish personality from its neighboring fields of scientific study. The most important of these include a molecular-molar dimension, that distinguishes more molecular brain sciences which underlie personality from personality itself, and also distinguishes personality from more molar social structures that "contain" it such as the family and society. A second, internal-external dimension, distinguishes inside mental processes from outside observable behavior. To this, a third, organismic-constructed dimension can be added, which distinguishes between those parts of personality that are most constrained by the biological organism (i.e. basic motivations) from those that are most independent (i.e., formal reasoning). The use of three dimensions makes possible a three dimensional pictorial representation of personality and its component parts (see Figure 1). The purpose of this initial picture is to orient personality amidst its neighboring system in the three-dimensional space. Internal personality is contained in a box labelled "personality" on the left-hand side of the figure, mid-way between nervous system substrates beneath it, and family and social systems above it. In the picture, this vertical dimension represents the molecular-mOlar continuum in the sense that the lower brain sciences are more molecular than personality whereas the family and other social groups above personality are more molar. The second, horizontal dimension, represents the internal-external continuum with internal personality to the left, and personality's external manifestation (i.e., its interaction with the environment) to the right. Finally, the third, depth dimension, distinguishes more organismic parts of personality (to be added momentarily) in the foreground from more constructed parts (also to be added) in the background. The empty personality box can now be filled with classes of personality components in a manner that is consistent with each of the three dimensions. For example, in Figure 2, conation, affect, and cognition are placed along the floor of the cube, near the biological level, with a slight rise toward the back indicating the greater molarity of cognition relative to conation. This particular placement implies that conation, affect, and cognition refer to

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

GROUPSlNCLUDlNG OR INTERACTING WITH PERSONALITY

Figure 2. A second view of the personality system including the enablers: Conation, affect, cognition, and consciousness (modified from Mayer, 1995a, Figure 2).

35

36

Chapter 2

internal, more molecular components of mind - that is, close to the biological level, or, only minimally influenced by learning. Notice that toward the innermost part of personality a fourth category has been added, consciousness. The placement of consciousness near conation, affect, and cognition suggests that consciousness, like them, is a more molecular, biological phenomenon, which may interact with the other three. Too little is known about consciousness to place it definitively anywhere, of course. One very respectable and influential tradition views consciousness as analogous to an imago in a hologram, in that it emerges from layered information within the cerebral cortex (Pribram, 1971, p. 171). This view would place consciousness at the ceiling of the personality box. But the rdational model puts it to the left bottom for reasons to be developed later. Within this relational model, conation, affect, cognition, and consciousness arc subgroups of a class of personality components containing them, termed enablers. Enablcrs are mechanisms that carry out, or enable, the basic functions of personality. The r arc one of four broad classifications that collectively contain all the parts of personality. The other three classes arc establishments, themes, and agencies. Establishments arc so-called because they are established (or leamod, or constructed) models of the self, the world, and the self in the world. Examples of establishments include the self-conc~t, self-esteem, attachment patterns, and expert knowledge. Establishments develop from experience and learning, and utilize the cnablcrs' functions to operate. For example, the self concept's self-love or self-hatred will be generated and intcrprcteA by emotional enablers, its self assessment will require cognitive cnablers. The connection between cnablcrs and establishments is often limited, however, to the fact that cnablers support establishments. At the establishment level, for example, expert knowledge can be fairly independent of a good or bad memory at the enabler level. That is, children may construct expert knowledge about dinosaurs independent of whether they possess an impoverished or superior memory. Thus, the establishment can be dcfine~ primarily according to its specific content. Establishment models arc illustrated in Figure 3, as the three floating cubes of internal personality. They are more molar than the enablcrs, and arc more independent of the organism as they proceed back toward models of the world. Note that all parts of personality arc vicwod as connected to all others; no arrows or connections are drawn in, however, as such a thicket of connections would obscure the rest of the depiction.

,I.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

N E R V O U S S Y S T E M

37

Figure 3. A more complete view of the personality system now including all four major classes of personality components. The enablers (wnation, affect, cogmtion, and consciousness) are on the floor of the personality box. The establishments (models of the self, world, and self-in-world) are represented as boxes floating in the inside of the cube. The themes combine features of enablers and establishments; one theme, extroversion, is illustrated toward the back center of the Figure. Finally, agencies are larger supercomposites of individual components that collectively act as sub-personalities; one such agency, James' self-as-knower, is represented, as a cloud that intersects with the "Models of the Self' box (modified from Mayer, 1995a, Figure 2).

38

Chapter 2

The third class of components, the themes, represent thematic connections between establishments and enablers. Themes combine features from enablers and from establishments so as to form conceptually related mixtures that reveal themselves to observers in a coherent fashion. Whereas establishments are focussed on contents, themes are focussed on common or integrated features across enablers, across establishments, or across the two combined. Thus, a need for stimulation alone is an enabler; a model of "joining friends for a party," is an establishment. But the two can be viewed as thematically related. Thus, extroversion, according to Eysenck (1982), involves both a need for stimulation, (the eonative enabler), and establishment models of things such as how to throw a party. Extroversion is illustrated as elliptical features found in both eonation and in models of the world; these features are labelled ("extroversion features") to the right of the internal personality cube. The fourth class of components, the agencies, refer to large subdivisions of personality that carry out much of a personality's activities, but in partial independence of the whole; these include the id, ego, or superego. Another example of an agency is James' concept of the self-as-knower, which comes close to a self-conscious free spirit or free will. The self-as-knower is represented as a cloudlike column that runs through the Models of the Self. A more comprehensive discussion of the classification of personality components into enablers, establishments, themes, and agencies, and their twenty-one subcategories can be found elsewhere (Mayer, 1995a,b). Here, we are particularly interested in conation, affect, cognition, and consciousness, the subgroups of enablers. Enablers, as already noted, are viewed as close to the biological level in the relational model. For that reason, there must be plausible biological bases for the operation of these parts, and their division. Moreover, these parts form a larger class that describe mechanisms that carry out the functions of personality. Hence, the enablers must be divided and understood foremost according to what they enable, that is, what functions they perform. Because enablers are so basic, and perform basic functions of personality, almost all other parts of personality rely on them and are influenced by them. Better defining conation, affect, and cognition, and understanding the rationale underlying these concepts, can clarify understanding of personality as a whole.

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

39

Understanding Conation, Affect, and Cognition Conation, affect, and cognition through recent history Hilgard's (1980) classic article The Trilogy of Mind... recounts the rise and fall of these three concepts from the early 1700's to early 1900's, and offers a rationale and recommendation for their resurrection. Surprisingly, Hilgard's work omits virtually any discussion of the meanings of conation, affect, or cognition, aside from their special status as a three-fold classification for the overall mind. Nonetheless, his article provides a basis for such an exploration by tracing the major figures who developed the trilogy over its history.

Faculty psychology and the trilogy of mind Hilgard (1980, p. 108) starts with the German faculty psychologists of the 18th century. He credits, in particular, Moses Mendelssohn's Letters on Sensation for bringing together the three concepts for the first time. Mendelssohn distinguished conation, affect, and cognition according to the fact that they operated differently from one another and that they might even interfere with one another. For example, when reason (cognition) "laboriously investigates the origin of pleasure," he wrote, "pleasure may be destroyed" (Mendelssohn, 1755/1971, p. 66) 1. There is both a phenomenological quality to this statement, indicating a sensitivity to the inner conscious experience of cognition and affect, and also a functional notion, identifying that cognition "investigates" pleasure. Mendelssohn also noted the independent behavior of the three components, writing that "convictions...belong in the realm of man's cognitive psychology," and that "by their very nature, [convictions] cannot be influenced by coercion or bribe" (Mendelssohn, 1983/1969, p. 44). On the other hand, will or motivation could be encouraged or discouraged by "reward and punishment" (Mendelssohn, 1983/1969, p. 44). Mendelssohn's approach is a partly functional one in the sense that he is specifying the conditions under which operations of the three spheres can be teased apart. The faculty psychology of late 18th century Germany gradually spread 1 Mendelssohn'swork is not yet translated in English. Hans G. Hirsch was kind enoughto translate fragments of the work which at least suggest some flavor of the original writings (see also Mayer, 1995b).

40

Chapter 2

to England and Scotland in the early 19th century. A number of psychologists contributed to classifying aspects of the mind during this period. For example, Thomas Reid, the great Scottish faculty psychologist, divided mental faculties into the intellectual (cognitive) and the active (motivational), dropping out emotion. By the late 19th century a summary of these British works was published in Alexander Bain's two-volume English textbook on psychology. Bain was fairly critical of attempts such as Reid's to reduce the trilogy to only two categories. He wrote that Reid's "submerged department of Emotion," could not be made to disappear but rather that its parts, such as emotions, feelings, and so on, "will be found partly taken in among the Intellectual Powers...and partly treated among the Active Powers," (Bain, 1855/1977, pp. 6-7), where they did not plainly fit. "Mind," wrote Bain (1855/1977, p. 1) at the outset of The Senses and the Intellect, ...possesses three attributes or capacities. I. It has Feeling, in which term I include what is commonly called Sensation and Emotion. II. It can Act according to Feeling. IIl. It can Think. Bain's trilogy, however, differs from the contemporary. For Bain, Feeling included sensation, whereas today's mental divisions typically group sensation with perception, outside the trilogy. Additional information concerning Bain's views on each member of the trilogy appear in the top portion of Table 1, which has three columns. Table 1 indicates the views of several central theorists, beginning with Bain. The three columns are divided so as to represent that theorist's view of conation, affect, and cognition. For example, in Table l's affect column, Bain says feeling and consciousness are "one and the same;" a statement which appears untenable today given contemporary research on unfelt, unexpressed, or unconscious emotions (e.g., Taylor, 1984). At the same time, Bain successfully develops a contemporary understanding of conation as he distinguishes between mental actions, which are part of the mental sphere, and those external actions that are not (Table 1, under "conation").

Chapter 2

41

Table 1. Historical and Contemporary Views of Conation, Affect, and Cognition: Direct Quotations and Brief Summaries from Key Figures. .

.

Conation .

.

.

.

.

.

.

.

.

.

.

.

.

Emotion .

.

.

.

.

.

.

Cognition

BAIN (1855/1977) "Action is...The putting forth of power to execute some work or perform some operation...in speaking of Action, however, as a characteristic of mind, we must render explicit the distinction between mental actions and such as are not mental...mental actions [are]... under the prompting and guidance of Feeling." (pp 2-3) "...There are in the human system movements and tendencies to movement 9..The eyes may open of themselves, the voice may break forth into utterance ...Yet those movements belong to the sphere of mind. The term Volition applies...to the entire range of mental or feelingprompted actions ." (p. 5)

"The three terms, Feeling, Emotion, and Consciousness, will, I think be found in reality to express one and the same fact or attribute of mind..." (p. 1) "...for a notion of what feeling is, I must refer each person to their own experience. The warmth felt in sunshine, the fragrance of flowers, the sweetness of honey..." (p. 2)

"...discriminating with preference, and the performance of intermediate actions to attain an end, are the most universal aspects of intelligence, inasmuch as they pervade the whole of the animal kingdom." (p. 6) "...the intellect [is]...a distinct endowment following laws of its own, being sometimes well developed and sometimes feeble without regard to the force or degree of the other two attributes." (p. 6) Intellect is distinct from emotion and volition because it allows for sensations and ideas to be relived without the stimulus (pp. 315-316) "Reason without affect would be impotent, affect without reason would be blind." (p. 112)

"In the evolutionary transition from reptiles to mammals, three cardinal behavioral developments were (1) nursing in

"The neocortex [can be described as]...ballooning out progressively in evolution and reaching its greatest proportions in the

MACLEAN (1990) "The protoreptilian formation is represented by a particular group of ganglionic structures located at the base of the

42

Chapter 2

Table 1 continued. forebrain in reptiles, birds, and mammals...these ganglia must be of 'enormous significance'for otherwise they would not be found as a constant feature in the vertebrate forebrain...[It is involved in] such basic behavior as the struggle for power, adherence to routine, 'imitation,' obeisance to precedent, and deception." (pp. 15-16)

conjunction with maternal care, (2)audiovocal communication for maintaining maternal-offspring contact, and (3) play...The limbic system plays a basic role in thymogenic functions reflected as emotional behavior...Two evolutionarily older subdivisions...have proved to be involved, respectively, in oral and genital functions...The third subdivision, for which there appears to be no

human brain...[it] has afforded a progressive capacity for problem solving, learning, and memory of details... linguistic translation and communication of subjective states..." (p. 17)

rudimentary counterpartin reptiles...[involves] parental care,audiovocal communication, and play behavior" (pp. 16-17) TOMKINS (1962) "In the human being the drive system plays a central role in... self-maintenance and reproduction." (p. 29) The system's primary function is to provide "motivating information" "information that drives and a drive that informs"specific to survival. (pp. 3031) It communicates "...where and when to do what- when the body does not know otherwise how to help itself." (p. 31 )

"The affective system [possesses]...numerous invariant instigators of any particular affect... [and] numerous invariant reducers of the same affect...It is this differentiated coupling and uncoupling characteristic which permits the affect system to assume a central position in the motivation of man." (p. 23) "Affects are sets of muscle and glandular responses located in the face and also widely distributed through the

[Not compared]

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith Table 1 continued. "The drive system with its relatively primitive signal and feedback mechanisms will work well enough [signalling internal changes] because of this predictable and small variability of the internal environment." (p. 124) "...a variety of materials must be regularly transported in and out of the body and thus drive signals wax and wane." (p. 125)

body, which generate sensory feedback which is either inherently 'acceptable' or kmacceptable'." (p. 243) Affects (associated with the reticular activating system, p. 90) such as interest, enjoyment, surprise, fear, shame, arise in response to learned or unlearned triggers (p. 22, p. 337). There is a partly invariant trigger-affect relation (p. 23). Affect is partially independent of the motivational system; it can mask motivation, or amplify the drive system so as to motivate the individual (p. 22). "This [affect] system is the primary provider of blueprints for cognition..." (p. 22) "There is here no essential rhythm as there is with respect to the drive system." ([- 125)

PLUTCHIK (1984) Aroused by changing internal states of the organism" (p. 214) "Aroused by the absence of homeostatically significant stimuli" (p. 214) "There are specific 'natural' objects toward which motives direct the organism (e.g., food, water)" (p. 214)

"Aroused by external stimuli" (p. 214) "Aroused by the presence of a survival-related event" (p. 214) "There are few 'natural' objects in the environment toward which emotions are automatically directed" (p. 214)

[Not compared]

43

44

Chapter 2

Table 1 continued. "Induced before the process of search is begun" (p. 214) "Tend to have a rhytlunic character" (p. 214)

"Induced aRer an object is seen or evaluated" (p. 214) "Depend on events in environment which may occur on a random basis" (p. 214)

IZARD (1993) "Drives such as hunger, thirst, sex...are cyclical in nature." (p. 72) "[Drives are] dependent upon peripheral physiological processes" (e.g., stomach growling; p. 73) "Drives provide specific information regarding the time and place that something needs to be done..." (p. 73) Drives, "cue a relatively specific set of responses..." (p. 73)

An emotion has no temporal cycle (p. 73) "...an emotion...is not dependent on peripheral physiological processes" (e.g., stomach growling) (p. 73) "...can be associated with a virtually limitless variety of phenomena" (p. 73) Emotions "can motivate an equally wide range of cognitions and actions" (p. 73) "the emotions system preceded the cognitive system in evolution and outpaces it in ontogeny" (p. 73)

"Clearly, information processing consists of several types or levels... ranging from that which leads to the color of an eye to that which produces a Mona Lisa or a theory of relativity" (p. 73) "I propose four differentiable sorts of information processing: cellular, organismic, biopsychological, and cognitive...the first three of the forgoing categories involve types of noncognitive information processing" (p. 70) Cognition is about knowledge- learning, memory, symbol manipulation, thinking, and language (p. 73) Emotion-cognition interactions occur in all the many coping activities that require stimulus appraisal and judgment before action (p. 73)

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

45

Hilgard (1980, pp. 113-114) concludes his survey of the trilogy of mind shortly after his discussion of Bain, with the psychologists of the 1920's and 1930's. He comments: Those in America who were proposing a new experimental or laboratory psychology rejected faculty psychology and along with it the classification of mental activity into three categories.., with [the American psychologist] McDougall the history of the trilogy of mind appears to have ended, nearly two centuries after it began in Germany and Scotland. In part, the fading of such a "generally accepted" view may have coincided with the decline of a felt need for such a comprehensive classification of mental processes. To be sure, Hilgard (1980, p. 113) wrote, "the trilogy of mind was still familiar in the vocabulary of psychology," but psychologists of the time were more interested in experimental advances than in the classification systems of the past. We believe that Hilgard's own interest in the trilogy suggests that its history was - and is - not over, although it may no longer occupy so central a place in the field. For that reason we proceed to more recent developments.

MacLean and the influence of psychiatry on the trilogy of mind By the mid-20th century enough had been learned about the brain structure and function that some initial statements could be made regarding its relation to mental faculties. Of course, this had been attempted earlier. Phrenologists had attempted to connect mental faculties such as learning or feeling to specific brain areas, for the purpose of charting personality according to a shape of an individuars cranium. Thus, someone with a cranial indentation alongside the presumed brain-site for imagination would be regarded as having a stodgy, uncreative personality. But phrenology was based on pure speculation, and as a consequence, was discredited. Brain localization became a reality, however, with the identification of some language abilities in Broca's area. And it was shortly thereafter, with the writings of Paul MacLean (e.g., 1949, 1973, 1990), that the trilogy of mind found a possible home in brain science. MacLean inferred from the structure of the human brain the existence of three partially independent subbrains, or brain divisions, which reflected three distract epochs in the human brain's evolutionary development. The first such brain, which was structurally innermost, was shared in all its essentials with the complete brain

46

Chapter 2

of reptiles. The second brain, which corresponded to the limbic system, was shared in common with most mammals. The third brain, which corresponded to the cerebral cortex, was most highly developed in humans. MacLean (1990, p. 9) wrote: In popular terms the three evolutionary formations might be imagined as three interconnected biological computers, with each having its own special intelligence, its own subjectivity, its own sense of time and space, and its own memory, motor, and other functions. Although MacLean never emphasized the point, parallels exist between conation and the reptilian brain, affect and the old-mammalian brain, and cognition and the neo-mammalian brain. For example, the reptilian brain had associated with it, "such genetically constituted forms of behaviour as selecting homesites, establishing territory, engaging in various types of display, hunting, homing, mating, bree~ing, imprinting, forming social hierarchies, and selecting leaders." (MacLean, 1973, pp. 9-10; 1990; see also Table 1). The old mammalian brain, "plays an important role in elaborating emotional feelings that guide behaviour with respect to the two basic life principles of self-preservation and the preservation of the species..." (MacLean, 1973, pp. 12-13). The third, neomammalian brain, is concerned with higher cognitive processes. MacLean suggests a number of innovative comparisons among the three brains. He notes that "the limbic system might be imagined as particularly designed to amplify or lower the intensity of feelings involved in guiding behavior required for self-preservation and preservation of the species." (1991, p. 17). He further notes that the different brains vary as to their external orientation, with the neomammalian (cognitive) brain most external in that it receives its information through signals conducted from the eyes, ears, and somatic receptors (MacLean, 1991, p. 19). MacLean's writings were influential in the 1950's and it is not surprising that they turned up, shortly thereafter, in psychological writings more explicitly identified with the mental trilogy.

Modern psychologists and the trilogy of mind Silvan Tomkins, an evolutionary emotions psychologist, focussed on the function of psychological processes and may have been influenced by

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

47

MaeLcan's writings. Recall that MacLean saw the limbic system, which was largely emotional, as amplifying survival-relatod feelings; Tomkins raised this idea again, arguing that the emotion system's role was to amplify motivation. Recall also that MacLcan described the nco-mammalian brain as more closely connected to the outside world than were the palco-mammalian or reptilian brains. Tomkins was perhaps influenced by this comparison when he notexi that the emotion system was directed toward the outside world whereas the conativc system was directed to the internal world. Finally, Tomkins shared with MacLcan and others of the time the use of an informationprocessing metaphor, describing r for example, as providing "readouts" of the organism's internal states. For Tomldns, conation has evolutionary significance in that it "plays a central role in...self maintenance and reproduction" (Tomkins, 1962, p. 29) as well as an information-processing aspect in which "primitive signal and feedback mechanisms" provide a readout of the internal homeostatic rhythms of the organism (Tomkins, 1962, p. 124). Tomkins went on to earcfuUy detail some of the characteristics that distinguished the conativc system from the affectivc. For example, Tomkins noted that "internal states" trigger conation, and that conation is typically rhythmic. In contrast, "external stimuli" trigger emotion, and emotion follows no particular set timclinc. These ideas have become generally accepted. For example, Robert Phtchik's (1980) side-byside comparisons of conation and affect included those and other distinctions that had been outlined by Tomkins. Plutchik's comparisons can also be found in Table 1. Tomkins and Plutchik both distinguish conation from emotion, with less attention paid to cognition (the cognition columns of Table l arc essentially empty for these theorists). The conation-affcct distinction was likely viewed as requiring more theoretical attention because motivation and emotion are so inextricably intertwined in behavior. There is something so different between conation and affect, on the one hand, and cognition, on the other, that the difference was often unattended to (Bain, 1855, p. 6, made this same point). Nonetheless, there arc some difficulties involved in distinguishing conation and affect from cognition. A central problem is caused by the frequent use of an information-processing metaphor to describe both the functions of conation and affect. If both conation and affect arc processing information, what is unique about cognition? Tomkins' former student, Cal Izard, recently addressed this problem by distinguishing between non-cognitive and cognitive information processing. Non-cognitive information processing inchdes that accomplished by genetic

48

Chapter 2

codes, chemical reactions, and "reflective instinctive, and biologically prepared or genetically disposed behavior" (Izard, 1993, p. 70). Cognitive processing, in contrast, "involves more general and flexible processes that operate on experience based learning and memory. Cognitive activities involves judgment, planning, problem-solving and understanding." Trends in thinking on the trilogy across time Considerable shifts in meaning of the tfilogy's categories have taken place, even from Alexander Bain's writings in the late 19th century to the present. This progression reflects (to us) a cumulative understanding of the utility of the trilogy, and of the differences among the tripartite areas. Several trends appear to best describe this progression: a trend toward identifying the trilogy as taking place exclusively internal to personality, a trend toward localizing each member of the trilogy in one or more brain areas, a trend toward an information-processing metaphor to describe them, and a reformulation of each class so as to create a more meaningful trilogy. The trend toward distinguishing the internal from the external. There has been a more or less constant recognition that conation, affect, and cognition are internal mental events, i.e., associated with brain function rather than with external events. Mendelssohn's comments that pleasure and pain change a person's will but not their cognition suggests that cognition is something intrinsically private, hidden and autonomous (Mendelssohn, 1755/1971, p. 66). A century later, Alexander Bain struggled to define will's internal location. Bain (1855, p. 2) referred to will as conative action that required the "putting forth of power to execute some work." Bain (1855, pp. 2-3) noted that, "In speaking of Action...as a characteristic of mind, we must render explicit the distinction between mental actions and such as are not mental." Bain's clarification that action was "a characteristic of mind," and therefore internal, was probably necessitated by his description of mental action as "putting forth power," which could readily be mis-understood as taking place externally. This metaphorical difficulty evaporated with MacLean's switch to the use of information processing metaphors for brain function, which suggested an internal computer. The trend toward brain localization and informaaon processing. Consistent with the internalization of these three processes was the attempt to find serious associations between the three classes and brain function. Although a non-scientific beginning to this pursuit originated with the phrenologists, serious connections awaited the works of MacLean, in

J D. Mayer, H. Frasier Chabot and K.M. Carlsmith

49

biological psychiatry, and Tomkins, in psychology. Although MacLean's work focussed on brain localization, Tomkins' work provided an interesting supplemental conception by extending localization to the larger nervous system. For example, "affects" were "sets of muscle and glandular responses closely associated with the brain's reticular activating system" (Tomkins, 1962, p. 243). Along with the increased focus on the brain and nervous system was the aforementioned shift in metaphor from industrial machines to an information processing paradigm. Bain's view of conative action as the "putting forth of power to execute some work" seems embedded in his own era of mechanical engines, whereas Tomkins' (1962, p. 124) view that conation provides "signal and feedback mechanisms" of internal organismic information, seems embedded in an era of computers. Although the information processing metaphor is today dominant it is still possible that multiple metaphors can best describe the phenomenon, just as in physics, light is both described as a wave and a particle (Bohr, 1963). For example, conation seems best described by combining Bain's and Tomkins' descriptions, so that conation is said to provide "a primitive readout of the internal, more or less homeostatic rhythm of the organism", and generates "power to execute some work." The trend toward finding more homogeneous categories at a common level offunction. There has also been an important narrowing of the trilogy's members such that each category is individually more circumscribed, and so that they operate collectively at a common level of function. For example, Bain's category of affect originally included the three concepts of feeling, consciousness and sensation, whereas contemporary views have essentially restricted the category to emotions and closely related feeling states such as calmness and arousal. This narrowing of focus represented a growing recognition that consciousness, sensation, and affect are incommensurate processes that perform different functions, are localized separately, and therefore are best treated separately. In today's Introductory Psychology books, sensation has been paired off with perception, and consciousness is treated, if at all, in its own chapters. The remaining affect category retains only emotion and closely related feelings. This narrowed version of the affect category seems more parallel to the similarly narrowed categories of conation and cognition. A similar and no less important transition occurred for conation, which originally referred to will, but with the transition from Mendelssohn to Tomkins has come to refer to more-or-less basic, unlearned motivations. The conation category now includes only basic motivations, which are, once

50

Chapter 2

again, both more homogeneous, and easier to compare to the similarly revised category of affect. The concepts of "will" and "consciousness", although excluded from the trilogy, were not plainly grouped with other parts of the mind. "Will" is perhaps covered in personality in discussing self-control and selfmanagement. Consciousness, however, could perhaps form a fourth category added to the trilogy of conation, affect, and cognition - a possibility we examine shortly. The trend toward emphasizing unlearned or innate qualiaes. As the categories of eonation, affect, and cognition have been more narrowly focussed, the focus has been directed toward their unlearned or innate qualities. The effort to distinguish these three mental categories has almost always best suece,exted when descriptions of them focus on their developmentally early, unlearned states. Thus, to say that motivations are "rhythmic"~ whereas emotions are not, is to emphasize such motivations as hunger, thirst, and sex, rather than more learned, less rhythmic motivations such as a desire for education or achievement. Similarly, to focus on the fact that emotions are triggered by external events is to emphasize their basic nature rather than more complex, learned emotions that might be triggered by reminiscence. This lower level, more mechanical conception was yet another reason to homogenize the categories and dispense with those parts, such as consciousness and will, that did not fit well. What remains in each category is a set of mechanisms, or basic functions of personality. Recall that it was their basic mechanical qualities that led to the label of enablers for conation, affect, cognition, because they help personality get the job done. The reason this emphasis on innate, or minimally learned qualities of the enablers is so important, is that as learning increases, more complex structures are created that are less plainly divisible into the three categories. For, as the enablers engage together in more complex functions it is clear that they become inexorably combined and intertwined. There exist a relatively few pure psychological enablers: pure conative urges for food and water, or pure affective joy or sadness, and pure memory networks. Soon atter these enablers begin work, they construct a much larger set of established thoughts that combine them. For example, a person develops models of the self, or a self concept, that includes conation (what I want), affect (what I feel about myself), and cognition (what I know about myself). But the general selfconcept, which includes all three, by necessity integrates the enablers. It was sensitivity to this point that led McDougall (1923, p. 266) to say that the trilogy work cooperatively rather than individually:

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

51

We often speak of an intellectual or cognitive activity; or of an act of willing or of resolving, choosing, striving, purposing; or again of a state of feeling. But it is generally admitted that all mental activity has these three aspects, cognitive, conative, and affective; and when we apply one of these three adjectives to any phase of mental process, we mean merely that the aspect named is the most prominent of the three at that moment. Each cycle of activity has this triple aspect; though each tends to pass through these phases in which cognition, conation, and affection are in turn most prominent; as when the naturalist, catching sight of a specimen, recognizes it, captures it, and gloats over its capture. The trend toward more limited inclusiveness. Through the time of Bain, some claim was made that the trilogy encompassed all mental function. With the increasingly focussed meaning of the three classes of mentation, it became easier to eject some concepts outside the trilogy. As has already been noted, sensation and perception were paired outside the trilogy. Similarly, will and consciousness were moved outside. The trilogy is no longer a trilogy of the entire mind, perhaps, but remains a critical trilogy operative within the more molecular, basic aspects of personality - and remains of considerable research importance. Caveat emptor This particular reading of the history of the trilogy of mind is, of course, our own, and alternatives are possible. The relational model of personality was constructed in part according to this reading of the evolution of the categories and employs those categories according to their outline here; alternative models are possible. Still, the relational model has very evident strengths in relation to classification models that have been developed before (see Mayer, 1995b), and it is worth, therefore, further considering how the trilogy of mind can be developed within it. Clarifying the trilogy m an expanded quaternity of mind

Although conation, affect, and motivation have been narrowed and clarified across time, many of the original distinctions among them still apply, even more clearly. The above discussion, atter all, has distinguished the three realms in several important ways. Phenomenological distinctions focus on different conscious experiences of the trilogy - that conation, affect, and motivation all "feel" differently from one another. Structural brain distinctions focus on differences in brain localization of the trilogy.

52

Chapter 2

Functional distinctions focus on the different actions of the three systems, and so on. These distinctions, as well as a number of others, can be summarized across theorists in a new, enlarged format. To create this summary, we chose the clearest statements from the Table 1, edited them, and supplemented them where necessary, in Table 2. Although Table 2 was constructed on the basis of the above discussion of the trilogy of mind, the table denotes a quatcmity - consciousness has been added. Some comment is necessary on this. As noted, Bain joined consciousness to feeling, but consciousness nowadays is just as likely to be joined to cognition (e.g., Bower, 1981), or denoted as a blackboard to represent all three (e.g., Bower & Cohen, 1982). In fact, consciousness is implicated whenever any of the three systems reach a high enough level of activation. For these reasons, it seems useful to separate consciousness from any single one of the other three and provide it with a place of its own. Because one interpretation of consciousness is that it is basic and elemental, a place among the enablers seems one possibility. Such a classification is useful from a systemic viewpoint because, just as the conativc-cnabler class includes urges, instincts, and mental energy, so a conscious-enabler class could include such components as the stream of consciousness, the phenomenal field, and so on. This provides a strong classificatory rationale, if nothing else, for provisionally converting the trilogy into a quatcmity, with the addition of consciousness.

The Quaternity of Mind and Personality Dynamics If the discussion until now seems removed from contemporary concerns that is one of the problems frequently encountered with discussions of classification. Contemporary research is concerned with dynamics - causal or mutual influences among different parts of personality. Another difference between the classification thus far and contemporary research is the sheer generality of the discussion. So far, we have talked of all affect as if it were a single entity, when in fact, it is divisible into many parts. The contemporary researcher, in contrast, typically is interested in more specific personality parts and their dynamics. So, whereas up-to-now we have discussed the interaction between affect and cognition, the researcher might be more interested in the influence of happiness on memory. Discussion at the global level has indisputable value, however, because it can make clear the conceptual background within which more specific research is conducted.

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

53

T a b l e 2. Conation, Affect, Cognition, and Consciousness Compared. Characteristic

Conation

Affect

Cognition

Consciousness

FUNCTION

To direct the organism to carry out basic acts so as to satisfy survival and reproductive needs

To organize a limited number of basic responses quickly, adaptively, and in an organized fashion; to link those responses to complex situational environments

To learn from the environment and to problem solve so as to assist with motives and emotions

To assign mental activity where needed; to intervene flexibly in conation, affect, or cognition, where new responses are called for

CONSCIOUS MANIFESTATIONS

If conscious, specific urges, e.g., toe.at, to drink

If conscious, the pleasure and pain of objects and stimuli; also, specific emotions such as happiness, fear, anger, etc.

Conscious and unconscious parts; conscious examination of problem

Direct consciousness itself; also reflective awareness of existence

AGENCY

Involuntary

Partly involuntary; partly voluntary

Mostly voluntary Partly voluntary; partly involuntary

DEVELOPMENTAL ONSET

Basic urges present immediately, including hunger, thirst, comfort..

Two or more basic emotions (e.g., pleasure, pain) present immediately; later development includes more complex emotions

Concrete reasoning early on, later the ability to reason with abstract information

Unknown; selfawareness from 18 months; continuous conscious identity from around age 3 with the end of infantile amnesia

Chapter 2

54 Table 2 continued.

i,

Characteristic

Conation

Affect

Cognition

Consciousness

INrITATION OF Predominantly RESPONSE responsive to internal bodily states

Predominantly responsive to external environment

Responsive either to internal or external environment

Responsive to non-habituated, i.e., novel, or unusually intense, internal or external events

TEMPORAL CHARACTERISTICS

Motivations precede action; rise and fall rhythmically or cyclically

Emotions often respond to events; they possess no set timeline

Occurs any time; Alternates no set timeline according to the sleep-wake cycle.

INFORMATIONAL SPECIFICITY

Specific as to what is lacking and what must be done

Identifies a class of possible events that must be addressed, without necessarily being specific

Either specific or general depending upon problem requirements, work accomplished, and mental capacity

Can incorporate and become aware of a wide variety of information; is very plastic in how it interprets information and proceeds

BRAIN LOCALIZ-

The limbic system is a subcortical structure, near the center of the cerebral hemispheres. It encircles the top of the brainstem. It is commonly divided into three tracts, or circuits, composed of different

Emotion is commonly associated with the limbic system, particularly with the amygdala, and secondarily with the hypothalamus. There is also recent evidence that the frontal cortex of the left hemisphere may

Information processing can be distinguished from higher level cognition. Although the entire brain processes information, we reserve cognition to encompass flexible processing based on learning and memory; this

May be located in the reticular activating system, or may be an emergent property of the mind as a whole

ATIONS

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

55

Table 2 continued. Characteristic

Conation

Affect

Cognition

structures. One mechanism of importance involves the hypothalamus which controls hormones that target various parts of the body and may regulate drives, e.g., of hunger and sex (Reeve, 1992).

specialize in processing positive emotion, the right hemisphere in negative,

includes judgment, planning, problem solving, and understanding. These are commonly viewed as dependent upon the association cortex and the cerebral cortex.

DESCRIPTION Unmotivated OF QUANTITY Motivated

Unemotional Emotional

Unthinking Thinking

Unconscious Conscious

SOCIALLY Constructive vs. DESIRED AIMS Destructive Motivations

Pleasant vs. Unpleasant Emotions

Intelligent vs. Unintelligent Thinking

Spiritually conscious vs. self-conscious

OPEN VERSUS Accepting vs. CLOSED/INAC- Repressed CESSIBLE

In Contact vs. Out of Contact with Feelings.

Flexible vs. Rigid

Receptive versus Unreceptive

JOINT MOLECULARMOLAR DEVELOPMENTAL CONTINUUM

*Basic emotions; e.g. happiness, anger, fear **Complex emotions, e.g., shame, guilt, mixed emotions ***Sentiments (emotions attached to objects) e.g., loving one's country,

*Basic cognition: sensory motor operations, learning **Middle cognition: concrete operations, symbol learning ***Complex cognitions: formal operations, abstract thought.

*Basic consciousness **Reflective consciousness ***Higher consciousness (e.g., reflective, spiritual, etc.).

*Basic urges, e.g., hunger, thirst, physical contact; **Learned motivations: e.g., pleasing others, achievement ***Functionally autonomous motives, e.g., doing a good job, helping others,

Consciousness

56

Chapter 2

For example, such a general discussion can provide hints as to where the more important enabler-to-enabler interactions will take place. Treating conation, affect, cognition, and consciousness as equals would suggest there exist 6, i.e., (4 • 3)/2, equivalently important sets of interactions to cover. An interesting alternative view, however, suggests that the central interactive areas among the classical trilogy will be more limited. Recall MacLean's triune brain that emerges in stages from conation to affect to cognition. If we assume adjoining areas (in terms of brain localization) have more interactions, greater interactions should occur between the adjoining areas of conation and affect, and affect and cognition, than between conation and cognition. This seems borne out by (our admittedly subjective impression of) today's research literature, which focusses on the former two interactions. Limitations of time and space have encouraged us to focus on the central conative-affective, and affective-cognitive interactions. The interactions between consciousness and the trilogy will be considered briefly at the end.

Conation and affect To recap, conative phenomena concern include hunger, thirst, and reproduction. Conative functions chart homeostasis in the body and alert the organism about needs for survival and reproduction. Thus, hunger tells us we should eat; thirst tells us we should drink, and so forth. In contrast, affect is concerned with such feeling states as happiness, joy, and alertness. Its primary concern is to provide us with signals about our relations with external individuals and objects. Thus, happiness tells us we are in harmony with others, and anger that we are treated unjustly. It is plain that conation and affect must serve the same master to some extent (e.g., overall personality). Thus, basic-level motivations provide constraints on emotions that ensure survival. Say you agree to eat your bagged lunch with someone late in the day. Then, during a walk in the woods you become hungry and think of the bagged lunch you brought along. You are likely to feel frustrateA, but you won't eat immediately because you know it will make you feel guilty later. Should the motivation to cat become stronger, however, most people will cat, so as to promote their energy and clear-headedness - their likelihood for survival. In the above instance, motivation (conation) and emotion work together, assessing different necessities, and balancing one against another. In that example, whether motivation or emotion "wins" is a matter of which signal (i.e., hunger or guilt) is the strongest. Often, however, more sophisticated

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

57

interactions may take place. For example, the emotion system (which is the more flexible) may "filter" motivations by allowing expression of those that are adaptive in a given situation, and by (at least temporarily) disallowing or suppressing those needs that are inappropriate. For example, if one is hungry, and there are people around who are eating, but none offer food, the original sense of hunger may be replaced by a feeling of injustice. An angry injustice might be a motivator for requesting food even though the act could be viewed as impolite or even improper (making a request might be suppressed by guilt). Say that, in this instance, the anger does replace the original hunger motivation and redirects the individual to ask for food. This is in part what Tomkms (1962, p. 22) meant when he wrote that "Affect...can mask motivation, or amplify the drive system so as to motivate the individual." Similarly, Oatley and Johnson-Laird (1987) view emotions as coordinating motivational urges and plans. Finally, motivation and emotion may contribute to one another more directly. Say you become happy because you have accomplished an important goal. You may need companionship as a consequence, and the motivational system may provide urges - phenomenological bursts of energy - to assist you to pursue social companionship. As another example, you may suddenly become sad; motivationally you may need to return to your own territory, or as the present idiom has it, you "need space." Helpful or harmful though this motivational accompaniment may be that moment, it is hard to change its directional quality. Research on the interaction between motivation and affect often reflects explorations in physiological, non-verbal communication, and evolutionary psychology. A review of such literature can be found in the chapter, "Motivation and Emotion," in Mook's (1996) textbook, Motivation. Because this area has been reviewed so recently, and because a large portion of it lies outside our own areas of expertise, we will move ahead to the relation between affect and cognition.

Affect and cogmtion We have already recapped the affect system, focussing on its depiction of relationships between oneself and the external world. The cognitive system, on the other hand, is useful for more flexible understandings of the world and the events in it. One of affeet's most important contributions to cognition is to prioritize it (Mandler, 1984). Thus, when working on a project, a fear of something going on at home, although distracting at first,

58

Chapter 2

may turn one's attention to what is, ultimately, a higher priority to one's survival. Not only do affects interrupt cognitions, but they can also change them in ways that may promote better judgment and creativity. One of the major influences of affect on cognition is through that of the mood-congruent cognition effect. Modified slightly from Mayer, Gaschke, Braverman, & Evans (1992, p. 129), the mood-congruent cognition effect: ...states that people's cognitions are sensitive to the correspondence between the pleasant-unpleasant quality of their mood and the pleasant-unpleasant connotations of their ideas. An affective match between a person's moods and ideas increases both the memorability and the judged merit, broadly defined, of those ideas. For example, mood-congruent concepts will be more readily learned and recalled. In addition, mood-congruent ideas will be judged richer in their associations, mood-congruent attributes will be judged as more applicable, mood-congruent examples of categories will be judged as more typical, and mood-congruent causes and outcomes will be judged more plausible. It is possible to read into this effect another way mood facilitates cognition: As a person's moods shift, the shift will force changes in a person's perspective on the surrounding world. Changing perspectives, in turn, allows for creative thinking about a problem, and the construction of a greater number of alternative courses of action. Such mood shifts drag the cognitive system along with them, forcing alterations in thinking and motivating changes in perception, and potentially enhancing planning and creativity (see Mayer, 1986, or discussion in Mayer, McCormick, & Strong, 1995). At a still broader level, cognitions seem to keep affects tolerable. That is, much thinking involves doing something for the emotion system, and consequently, for the motives those feelings relate to. This is what Tomkins (1962, p. 22) meant when he wrote that, "...this [affect] system is the primary provider of blueprints for cognition..." It is also at least loosely related to Freud's notion that the ego derives its energy from the id. The more one's emotions are satisfied, the less directive they are and the more chance the cognitive system has to operate well according to its own rules of logic, propositions, and formalism. Although cognition follows the blueprint of affect, it can also turn around and change affect where affect (or motivation) seems

d.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

59

counterproductive. For example, cognitions can help manage affects when they get out of hand, and separate good or useful affects, from misleading ones. So-called meta-or reflective experiences of mood (e.g., "This mood is clear to me," "This feeling is unacceptable," etc.) involve cognitive attempts to evaluate and regulate moods so as to improve their responsiveness beyond a simple reflexive attempt at survival (e.g., Mayer & Gaschke, 1988; Mayer & Stevens, 1994; Salovey et al., 1995). The recently developed concept of emotional intelligence (e.g., Mayer & Geher, 1996; Mayer & Salovey, in press; 1993; Salovey & Mayer, 1990) is basically a compendium of the areas in which emotion facilitates thought, and thought improves emotion. One recent definition of emotional intelligence (Mayer & Salovey, in press) describes it as including four broad classes of abilities: ...the ability to perceive accurately, appraise, and express emotion: the ability to access and/or generate feelings when they facilitate thought; the ability to understand emotion and emotional knowledge; and the ability to regulate emotions to promote emotional and intellectual growth. The role of consciousness

It is hard to resist a mention of consciousness' function at this point. To us, consciousness plays a role similar to that of a family bulletin board upon which messages are placed (cf., Bower & Cohen, 1982, pp. 309-310). The consciousness "bulletin board," more specifically, receives messages from conation, affect, and motivation: urges, such as "need water," emotions, such as "anxiety", and thoughts, such as "l should talk more at my upcoming meeting to appear more assertive." Just as in a family, each member has different handwriting, so too, conation, affect, and cognition, have their own individually recognizable modalities, their signature phenomenology. An integrated personality recognizes messages from each source because it experiences each differently, and evaluates each system on its own terms, much as one evaluates messages from family members on the basis of their recognizable styles. That is, an adult personality uses consciousness to recognize that an urge is an urge, and as such, has a different status than a logical proposition. Ideally, it weighs the urge ("l am increasingly hungry") with the thought ("This project would best be finished before l eat") and wisely chooses which to follow depending on circumstances.

60

Chapter 2 Conclusions and Other Considerations

The four cnablers of conation, affect, cognition, and consciousness represent only the lower level portions of personality. Emerging from them arc establishments, including models of the self, the world, and the self-inthe-world, and themes, coherent collections of features drawn from cnablcrs and establishments that arc expressed as behavioral traits. Conation, affect, and cognition work closely together to support these more complex structures. For example, research on cognition and affect as they extend into a person's models of the self and world (i.e., establishments) are being conducted by Fiskc and her colleagues on affect-triggered schemata (Fiskc, 1982); by Higgins and his colleagues on self-schema and affect (Higgins, 1987), and by Petty and his colleagues on attitudes (e.g., Pricstcr & Petty, 1996). Summary. Researchers in the area of cognition and affect are, by virtue of their interdisciplinary interest, unusually broad in the problems they pursue. Successful research across affect and cognition may be facilitated by better understanding the scope of affect and cognition, the distinctions between them, and their relationship to personality. To better understand cognition and affect, their original grouping: conaaon, affect, and cognition the so-called trilogy of mind - was examined in considerable detail. We provided a historical review of the trilogy of min~! and attempted to discover some trends in their evolving meaning. The dofufitions of conation, affect, cognition, were refined and updated. An alteration of the trilogy to a quaternity was recommended so as to include consciousness. This quatemity/trilogy was located within one possible contemporary model of personality, the relational model. Finally, the relevance of the quatemity and the interactions among its members were briefly applied to a discussion of some contemporary research in cognition and affect. References

Allport, G. W. (1958). What units shall we employ7 In G. Lindzey (Ed.), Assessment of Human Motives (pp. 239-260). New York: Rinehart & Company, Inc. Bain, A. (1855/1977). The senses and the intellect. London: John W. Parker & Son. [Roprintexi in D. N. Robinson (Ed.), Significant contributions to the history of psychology: 1750-1920 [Series A: Orientations; Vol. 4]. Washington, DC: University Publications of America.

J.D. Mayer, H. Frasier Chabot and K.M. Carlsmith

61

Barratt, E. S. (1985). Impulsiveness defined within a systems model of personality. In C. D. Spiclbcrgcr & J. N. Butcher (Eds.), Advances in personality assessment (Vol. 5, pp. 113-132). HiUsdalc, NJ: Lawrence Erlbaum. Bohr, N. (1963). Essays, 1958-1962, on atomic physics and human knowledge. New York: Wiley. Bower, G. H. (1981). Mood and memory. American Psychologist, 36, 129148. Bower, G. H., & Cohen, P. R. (1982). Emotional influences in memory and thinking: Data and theory. In M. S. Clark & S. T. Fiskc (Eds.), Affect and cognition. Hillsdalc, NJ: Lawrence Erlbaum. Buss, A. H., & Finn, S. E. (1987). Classification of personality traits. Journal of Personality and Social Psychology, 52, 432-444. Clark, M. S., & Fiskc, S. T. (1982). Affect and cognition: The seventeenth annual Carnegie Symposium on cognition. Hillsdalc, NJ: Lawrence Erlbaum. Eyscnck, H. J. (1982). Personality, genetics, and behavior. New York: Pracgcr. Fiskc, S. T. (1982). Schema-triggered affect: Applications to social perception. In M. S. Clark & S. T. Fiskc (Eds.), Affect and cognition. Hillsdalc, NJ: Lawrence Erlbaum. Higgins, E. T. (1987). Self-discrepancy: A theory relating self and affect. Psychological Review, 94, 319-340. Hilgard, E. R. (1980). The trilogy of mind: Cognition, affection, and conation. Journal of the History of the Behavioral Sciences, 16, 107117. Izard, C. E. (1993). Four systems for emotion activation: Cognitive and noncognitivc processes. Psychological Review, 100, 68-90. MacLcan, P. D. (1949). Psychosomatic disease and the 'visceral brain'. Recent developments bearing on the Papcz theory of emotion. Psychosomatic Medicine, 11, 338-353. MacLean, P. D. (1973). A triune concept of the brain and behaviour. Toronto: University of Toronto Press. MacLcan, P. D. (1990). The triune brain m evolution: Role in paleocerebralfunctions. New York: Plenum Press. Mandlcr, G. (1984). Mind and body: Psychology of emotion and stress. New York: W. W. Norton & Co. Mayer, J. D. (1993). A system-topics framework for the study of personality. Imagination, Cognition, and Personality, 13, 99-123.

62

Chapter 2

Mayer, J. D. (1995a). The system-topics framework and the structural arrangement of systems within and around personality. Journal of Personality, 63, 459-493. Mayer, J. D. (1995b). A framework for the classification of personality components. Journal of Personality, 63, 819-877. Mayer, J. D., & Gaschke, Y. N. (1988). The experience and recta-experience of mood. Journal of Personality and Social Psychology, .55, 102-111. Mayer, J. D., & Geher, G. (1996). Emotional intelligence and the identification of emotion. Intelligence, 22, 89-113. Mayer, J. D., McCormick, L. J., & Strong, S. E. (1995). Mood-congruent recall and natural mood: New evidence. Personality and Social Psychology Bulletin, 21,736-746. Mayer, J. D., & Salovey, P. (1993). The intelligence of emotional intelligence. Intelligence, 17, 433-442. Mayer, J. D., & Salovey, P. (in press). What is emotional intelligence? In P. Salovey & D. Sluyter (Eds.), Emotional development and emotional intelligence: Implications for educators. New York: Basic Books. Mayer, J. D., & Stevens, A. (1994). An emerging understanding of the reflective (meta-) experience of mood. Journal of Research in Personality, 28, 351-373. Mendelssohn, M (1971). Moses Mendelssohn: Gesammelte Schrifien Jubilaumsausgabe (Band 1: Schriflen zur Philosophie und Astheak). Stuttgart: Friedrieh Frommann Verlag (Gunther Holzboog). (Original work published 1755). Mendelssohn, M. (1969). Jerusalem (A. Jospe, Trans. & Ed.). New York: Schocken. (Original work published 1783). Mook, D. G. (1996). Motivation: The organization of action (2nd ed.). New York: W. W. Norton. Oatley, K., & Johnson-Laird, P. N. (1987). Towards a cognitive theory of emotion. Cogniaon and Emoaon, 1, 29-50. Pervin, L. A. (1990). A brief history of modem personality theory. In L. A. Pervin (Ed.), Handbook of personality theory and research (pp. 3-8). New York: Guilford. Plutchik, R. (1984). Emotions: A general psychoevolutionary theory. In K. R. Scherer & P. Ekman (Eds.), Approaches to emotion. Hillsdale, NJ: Lawrence Erlbaum. Priester, J. R., & Petty, R. E. (1996). Gradual threshold model of ambivalence: Relating the positive and negative bases of attitudes to subjective ambivalence. Journal of Personality and Social Psychology,

J.D Mayer, H. Frasier Chabot and K.M Carlsmith

63

71, 431-449. Pribram, K. H. (1971). Languages of the brain: Experimental paradoxes and principles in neuropsychology. Englewood Cliffs, NJ: Prentice Hall. Reeve, J. (1992). Understanding motivation and emotion. Fort Worth, TX: Harcourt, Brace, Jovanovich. Salovey, P., Mayer, J. D., Goldman, S., Turvey, C, & Palfai, T. (1995). Emotional attention, clarity, and repair: Exploring emotional intelligence using the Trait Meta-Mood Scale. In J. W. Pennebaker (Ed.), Emotion, disclosure, and health (pp. 125-154). Washington, DC: American Psychological Association. Salovey, P. & Mayer, J. D. (1990). Emotional intelligence. Imagination, Cognition, and Personality, 9, 185-211. Sears, R. R. (1950). Personality. Annual Review of Psychology, 1, 105-118. Taylor, G. J. (1984). Alexithymia: Concept, measurement, and implications for treatment. American Journal of Psychiatry, 141,725-732. Tomkins, S. S. (1962). Affect, imagery, consciousness. Vol. 1: The positive affects. New York: Springer. Author Notes

Paul Presson was instrumental in developing the graphics for the relational model of personality; his patience during design sessions enabled us to develop a far clearer picture than we would have otherwise, and we are grateful for his assistance.

This Page Intentionally Left Blank

Cognitive Science Perspectives on Personality and Emotion - G. Matthews (Editor) 9 1997 Elsevier Science B.V. All rights reserved. CHAPTER 3

Introduction to the Bidirectional Associative Memory Model: Implications for Psychopathology, Treatment, and Research Warren W. Tryon

Learning and memory are arguably the two most fundamental psychological processes. Without learning, infants would not acquire the skills that make them children and adults. Without memory, cumulative learning could not occur; we would continuously relearn everything. All connectionistic neural networks (CNNs) both learn and remember; they entail a learning and memory mechanism. It is therefore impossible to discuss learning in the absence of memory or memory in the absence of learning. An important advantage of CNNs is that they are also compatible with biological and genetic explanations. The possibility that the synaptic network comes preset at birth with sensitivities to, and biases for, processing information in certain ways was addressed by Seligman (1970) and Seligrnan and Hager (1972) in terms of biological preparedness. It is also possible that not all aspects of the CNN are equally modifiable by experience. It may be that certain networks function essentially unchanged throughout the subject's lifetime. These possibilities do not detract from the fact that many organisms, especially humans, learn a great deal during their lifetime and that some of what is learned plays an important role in developmental changes. Personality is heavily dependent upon memory. Persons with Alzheimer's Disease provide empirical support for this assertion. Their personalities gradually dissolve as they forget their life experiences including where they have been, what they have done, and who their children and parents are or were. Psychopathology and psychotherapy are also highly dependent upon memory. A phobic person is afraid only because they have anxious memories about certain stimuli. If the anxious memories of a car phobic can be replaced with memories of positive experiences, then the person will no longer fear automobiles. Other feelings not generated by immediate environmental stimuli are also memories. This includes feelings of depression, insecurity, and low self-image. Lotius (1980, p. xiv) described a hypothetical future memory doctor as being able to cure psychological disorders by modifying the memories giving rise to the associated feelings. Schafer's (1978) hermeneutic psychotherapy seeks relief in just such a way; by recalling and altering

66

Chapter 3

memory for past events through reinterpreting them so that they are experienced more positively and in an integrated fashion. Wachtel's (1977) description of psychoanalysis includes recapturing disassociated memories and diffusing their emotional impact through catharsis. Psychoanalysis can be accurately summarized as a theory of conscious and unconscious memory formation and recall especially under stressful conditions. The Bidirectional Associative Memory (BAM), one variety of CNN, enables psychologists to address many of the same topics as psychoanalysts but with much more testable models since CNNs, including the BAM, can be implemented on a computer and are therefore fully open to analysis and experimentation. An added benefit of CNNs is their fundamental compatibility with neuroscience and biological psychiatry. Contemporary behavior therapy is dominated by cognitive and cognitivebehavioral models. Reference to emotion or affect is conspicuously absent; Ellis excepted (1962, 1980). Hollon and Beck's (1994) description of cognitive and cognitive-behavioral therapies discusses thinking, beliefs, and interpretations as important elements but does not include emotionalmotivational variables. Blatt and Bers (1993, p. 165) observe that "The role of affect is not only ignored in most cognitive behavioral considerations of self-schemas, but it is often considered an impediment to the assessment of them. Rather than viewing the self-schema as a cognitive-affective structure, research from a cognitive-behavioral orientation often attempts to eliminate or control current mood as possibly confounding the assessment of schemas". The authors subsequently noted that cognitive-behavioral theorists are generally reluctant to explore motivational, affective, and developmental issues. Cognitive and information processing models of normal and abnormal behavior stress intellectual control. Contemporary behavior therapies for children and adults emphasize corrective thinking for emotional as well as behavioral disorders. Put otherwise, psychologists have over intellectualized emotional disorders. Any comprehensive explanation of normal and abnormal behavior must address emotion as well as cognition and behavior. The main purpose of the purpose of this chapter is to augment interest in modeling mechanisms underlying normal and pathological phenomena using connectionistic neural networks by applying one particular CNN, the BAM, to several areas of interest. The fact that CNNs in general and the BAM in particular are new to many psychologists means that little empirical work has been conducted to date. Hence, this chapter cannot review and evaluate the BAM in terms of quantitative empirical data. The scope of this chapter is

W. W. Tryon

67

therefore restricted to discussing the heuristic value of the BAM for understanding a wide range of phenomena related to cognition, emotion, and psychopathology. This chapter extends the BAM so that it learns emotions of varying intensities in specific contexts thereby forming affective memories. It is important to note that the same learning mechanism used to form intellectual memories is also capable of forming affective memories. In order to fully understand this approach to "hot cognition", we must review how the BAM stores and recalls memories.

Bidirectional Associative Memory (BAM) The BAM model was selected for the following masons. First, it is designed to form and recall memories. Second, because the BAM is equally able to associate among stimuli, emotions, and behaviors, it is applicable to the full spectrum of psychological and behavior disorder. Third, the concepts of memory well and basin of attraction associated with the BAM provide new ways to conceptualize psychopathology and treatment; both psychological and biological. Fourth, the BAM is a relatively simple system and consequently is a good point of departure. The BAM entails symmetric interconnections that the brain does not have and is therefore less biologically plausible than some other neural networks. However, the BAM is not intended to be an exact brain copy of an actual brain structure but rather to simulate memory formation and recall using selected brain functions such as parallel distributed processing and local processing at each node. The present discussion derives mainly from Kosko (1987a, 1987b, 1988) and Wasserman (1989). The Appendix provides details regarding how the BAM works. Because it is not entirely necessary to understand every detail of how the BAM functions to appreciate its heuristic value in understanding psychopathology, a succinct overview of the most important elements is given next. The stimuli and responses that the BAM learns to associate are represented as vectors, a sequence of numbers, of l's and O's defining the presence or absence of a set of characteristics. The attributes coded for can be cognitive, affective, and/or behavioral which makes the BAM a highly general model of memory formation. Any level of detail can be modeled. At a very low level of abstraction, vector entries can represent the state of individual sensory neurons and motor fibers. At a high level of abstraction, vector entries can represent the results of other neural networks dedicated to

Chapter 3

68

recognizing perceptual features (red hair), affeetive states (see below), and/or behavioral dispositions (response vs. no response or flight vs. fight). If the elements of one vector (A) index rows and the dements of the second vector (B) index columns, the resulting square matrix (outer product) constitutes a memory matrix (M) for the AB association. For example, if vector A = 1, -3, 5, 7 and vector B = 2, 4, -6, 8, then memory matrix M is defined as follows:

2 Vector A

1

2

-3 5 7

-6 10 14

.

.

Vector B 4 -6 4 -6 -12 18 20 -3O 28 -42 .,

8 ....... 8 -24 40 56 ,,,,

The numerical values of the matrix dements simulate functional synaptic properties of excitation (positive values) and inhibition (negative values). Multiple memories, up to a computable limit, can be accurately encoded into a single memory matrix by summing corresponding cells over all individual memory matrices. Memory recall is accomplished by multiplying a stimulus vector by the composite memory matrix. If the result of multiplying vector A times memory matrix M is not exactly vector B (correct recall), then the obtained result is fed back through the memory matrix by multiplying the obtained result by the transpose of M. The result of this calculation is used as a modified stimulus and therefore multiplied by M, as was vector A. The result will either be vector B or something closer to it This active reverberating and reconstructive process of memory recall, continues until vector B is fully recalled or no further improvement can be obtained in which case the memory recalled is, as with people, the best approximation that can be generated. This process enables pattern completion where a whole memory can often be reconstructed from a partial stimulus. Neural networks are good at Gestalt psychology. This pattern completion property will be emphasized in our discussions of psychopathology. Because of parallels with physics, an "energy" value can be calculated for each memory. This calculation provides the two dimensional memory matrix with a third dimension; height in this case, that enables one to visualize memory formation as the creation of memory wells in an otherwise

W. W. Tryon

69

fiat surface. This is because the state of minimum energy is the state of best fit between the AB vector pair. Memory recall occurs whenever this energy state recurs. Imagine a fiat rubber sheet upon which a ball bearing has been placed causing a vertical indentation. Since the ball bearing comes to rest at a point below the surface, it is associated with a negative, and therefore minimum, energy state (see Figure 1). That memory formation is associated with a minimum energy state can be understood as similar to how "best fit" occurs when the deviation of data points about a regression line is minimized. Both are measures of fit.

I Iglll IBm

60

LU

X

ILl

4}1t

2il

IOlD IBm ....J<J aim

Y

f.<J" III

aoe

6e

\

II

2o

X

Figure 1. Example memory field containing five memories; one near each comer plus one in the middle.

The process of memory recall can be visualized as placing a small frictionless ball on the memory surface and letting it roll down into a memory well. The memory is recalled when this locus of memory recall reaches the

70

Chapter 3

bottom of the well because that is the point associated with the best fit between the AB vector pair. It is important to note that although vectors A and B have been represented in a distributed way across several or many elements and entail many synapses, the entire memory has a single energy value. Each memory formation creates a local minimum in the memory field as illustrated schematically by Figure 1. All points on the memory surface leading downward to the memory well are called the basin of attraction for that memory. Memory wells vary in their depth (intensity) and breadth of their basins of attraction. Broader and deeper memory wells are more potent organizers of cognition, affect, and behavior than are narrower and shallower ones. The possibility that trauma creates a superbasin in the midst of existing memories is discussed below. One effect of such an event is to incorporate prior basins of attraction within a larger one, tilting those basins so that the flow of memory recall might pass by the normal memories and recall the traumatic one. It should be noted that memory wells are not pure metaphor but the geometric consequences of the mathematics associated with memory formation. Memory wells are visual representations of the mathematics of memory formation and therefore are explicit consequences of the BAM model. The BAM generalizes across the traditional distinction of semantic and episodic memory. The same memory mechanism is postulated for both types of memories.

Encoding Emotion This section draws heavily from Tryon (1996a). The first section briefly reviews previous efforts to encode emotions into CNNs. Subsequent sections recommend more direct solutions.

Previous efforts That the brain mediates emotion makes brain-inspired neural networks logical candidates for incorporating emotional factors into and integrating them with cognitive processes. Levine and Leven (1992) discuss "motivation, emotion, and goal direction in neural networks". Part II of their book contains articles on "Top-down processes, attention, and motivation in cognitive tasks" by Banquet, Smith and Giinther (1992), "A neural network theory of manicdepressive illness" by Hestenes (1992), "Learned helplessness, memory, and the dynamics of hope" by Leven (1992), "Integration, disintegration, and the

W. W. Tryon

71

frontal lobes" by Levine, Leven, and Prueitt (1992) plus "Familiarity and novelty: The contributions of the limbic forebrain to valuation and the processing of relevance" by Pribram (1992). Generalized affect, drive, has already been incorporated into some neural networks capable of classical conditioning (e.g., Grossberg & Levine, 1987; Grossberg & Schmajuk, 1987). However, the ability to generally represent specific emotions in a CNN has not yet been accomplished. Because CNNs can learn anything they can encode, our problem reduces to the question of how to represent emotions as a vector, the required BAM inputs. Color and emotion

The following sections draw heavily upon Plutchik (1980, in press) and Tryon (1996a). Plutchik (in press) recognizes McDougall (1921) as the first author to comment on the parallel between emotions and colors. Schlosberg (1941) analyzed emotions in response to the 72 Frois-Wittman pictures of facial expression and found that they could be arranged in a two dimensional circumplex. Schlosberg (1954) created a cone shaped model by adding an intensity dimension. The fundamental idea being that some emotions are primary, like primary colors, while all others derive from combinations of basic emotions. Plutchik (1994, pp. 53-64) reports complete agreement across investigators that at least 3 and no more than 11 primary emotions exist and that all other emotions are combinations of these primary ones. Most theorists identify between 5 and 9 primary emotions. Plutchik (1958, 1980) proposed 8 basic emotions based on Conte (1975) who asked subjects to rate 146 emotional words on an 1 l-point bipolar scale ranging from -5 = opposite, through 0 = no relation, to +5 = the same relative to three reference words: accepting, angry, and sad. The correlations among ratings over subjects were calculated. NunnaUy (1967, p. 299) discusses how the correlation between two variables can be expressed as the cosine of an angle between two unit vectors originating from the same point. Using each of three words as a referent, all other words were plotted on a circle using angular displacements calculated from obtained correlations. The final angular placement was the average of the three methods; each using a separate referent. The resulting circumplex has the following structure: Items with high positive correlations are placed close to one another. Items that are uncorrelated with one another are placed at right angles. Items that are negatively correlated, polar opposites, are placed opposite one other.

72

Chapter 3

Beginning at the top and moving clockwise, the eight Basic Emotions are: Acceptance, Fear, Surprise, Sadness, Disgust, Anger, Anticipation, and Joy. They form four bipolar pairs: Acceptance - Disgust, Fear - Anger, Surprise Anticipation, Sadness - Joy. Beginning with any emotion on the circumplex and skipping three consecutive emotions identifies the polar opposite emotion. Using the color analogy, Plutchik (1980) defined Primary Emotional Dyads as equal mixtures of adjacent pairs of Basic Emotions which resulted in the following 8 pairs of primary emotional dyads: Love - Joy + Acceptance, Submission = Acceptance + Fear, Awe = Fear + Surprise, Embarrassment = Surprise + Sadness, Misery = Sadness + Disgust, Scorn = Disgust + Anger, Aggression = Anger + Anticipation, Optimism = Anticipation + Joy. These emotions also form four bipolar pairs: Love Remorse, Submission - Contempt, Awe - Aggressiveness, Disappointment Optimism. Secondary Emotional Dyads are formed from equal mixtures of two Basic Emotions once removed, separated by one circumplex sector. Tertiary Emotional Dyads are formed from equal mixtures of two Basic Emotions twice removed, separated by two eircumplex sectors. Plutchik (1994) noted that different words represent the same emotion at various intensities. For example, annoyance, irritation, anger, rage, and fury differ primarily in intensity. Adding an intensity dimension orthogonal to the circumplex represents these related emotions. Hence, every emotion requires a circumplex and an intensity code.

Proposed emotional codes The following five coding schemes for representing the three dimensional extension of the emotional circumplex are offered. First, 8 vector elements are sufficient to represent one Basic Emotion as an 8-position 1-of-N code I . Another 8 vector elements are required to represent its intensity using a thermometer code2. This approach is economical in that only 16 vector elements are required but limited in that only a single emotion and its intensity are represented.

1 A l-of-N code selects from among N = 8 choices as follows: 10000000 selects the first item, 00010000 selects the fourth item, and 00000001 selects the eighth item. 2 A thermometer code represents intensity by the number of elements, from left to right, that are in the "on" position. If a thermometer code contains 8 elements, then the code 00000000 indicates none, 11110000 indicates half, and 11111111 indicates the maximum amount.

W. W. Tryon

73

The second approach encodes one Primary, Secondary, or Tertiary Emotional Dyad by using 16 vector elements to encode two 8-position 1-of-N codes representing the appropriate pair of Basic Emotions. Eight additional vector elements are required to represent the intensity of the composite emotion using a thermometer code. This is limited in that it assumes that mixtures of Basic Emotions are achieved using components of equal intensity. The third approach allows varying amounts of Basic Emotions to be encoded. Gradations of pairs of Basic Emotions can be accomplished by using 32 vector elements to represent each of the Basic Emotions using an 8position l-of-N code and their intensities using an 8-position thermometer code. The fourth method allows for the simultaneous representation of multiple Primary, Secondary, and Tertiary Emotional Dyads by encoding multiple pairs of Basic Emotions by doubling the number of vector elements described above in the second and third methods. This would require 32 or 64 vector elements respectively. The two variants of the fitth method are general and allow one to store from 1 to 8 Basic Emotions and their intensities. The first variant stores emotions in the vector array in any order by specifically encoding each emotion using 16 vector elements. The first eight elements use a 1-of-N code to select the Basic Emotion and the next eight elements to represent the corresponding intensity using an 8-position thermometer code. All 8 Basic Emotions at 8 different intensities can be represented using 8 • 16 = 128 vector elements. Emotions not represented in this method are encoded 00000000 as are their intensities. The second variant of the fifth method requires only 64 vector elements to store all 8 Basic Emotions by presuming that these emotions are represented in a fixed order beginning with a fixed referent emotion. An 8position thermometer code is used to indicate the intensity of each of the8 Basic Emotions. Emotions not present are coded 00000000. Emotions occur in a context. Current behaviors, other persons, consequences, sights, sounds, smell, taste, and touch provide important contextual information. This context information can be encoded using additional vector elements.

74

Chapter 3

Application Emotion can be encoded using the same vector approach used to by Anderson (1983) to encode cognitive components regarding three Greek mortals and three Greek gods using 50-clement vectors where elements 1-16 coded name information (Socrates, Alcibiades, Plato, Zeus, Apollo, Diana), elements 17-32 coded supernatural status (man or god), and elements 33-48 coded life span (immortal or mortal). Elements 49 and 50 were not used and were set to zero. All elements wore either +1 or -1. Multiple elements were used to represent discrete categorical information to implement distributed representation; i.e., distributing information across multiple nodes or neurons. This approach can also be used to learn about emotions and the contexts in which they occur. Anderson's (1983) approach could be used to associate emotions and contexts thereby forming emotional memories. This would result in purely emotional content like Anderson's purely cognitive content. Because emotion and cognition are highly interdependent, I (cf. Tryon, 1996a) propose extending cognitive vectors, like those used by Anderson (1983), to include emotional and contextual information. The memories formexi by these vectors will be a cognitive-affective composite; they will entail "hot cognition". Prior research on the pattern completion, content addressablr properties of these networks (i.e., Lcvine, 1991; Wasserman, 1989) indicates that this approach will integrate cognitive and affectivr information. Presenting cognitive stimuli will recall affoetive memories. Presenting affectivr stimuli will recall cognitive memories. If context stimuli are included in the vectors, then presenting context stimuli will recall both cognitive and affectivr memories.

Additional considerations The approach just taken representeA emotion and cognition using a common vector. A consequence of this choice is that affectivr and intellectual content are highly integrated, fused, into a single memory and therefore jointly influence behavior and the retrieval of associated fused memories. This approach assumes that emotions have no special status in connectionistic systems and consequently do not need to be treated separately in terms of how they are stored and/or processed. This approach is parsimonious in that no second memory system is required to store emotions. Nor is a second emotion processor required. Nor is any method required to

W. IV. Tryon

75

integrate emotion and cognition. These theoretical advantages strongly argue for coding emotion and cognition in the same vector. An alternative approach is to represent emotions using a separate vector. This approach assumes that using a single vector to represent both emotion and cognition is inappropriate. One possible reason for using separate vectors is that emotion and cognition may be considered to be different experiential domains. This approach does not solve the problem of how emotion and cognition are fused into unified memories. It does not address how emotion is integrated with cognition. Is a second network required to integrate these two aspects of experience? At least, some additional processing at some level is required but the nature of such processing remains unspecified. It should be noted that all of the above mentioned CNN options presume the existence of other networks functioning as feature detectors and that they selectively turn individual vector elements on or off in both the cognitive and affective fields. These feature detection networks are responsible for perceiving cognitive and affective elements from sensory experience.

Implications for DSM-IV Disorders The purpose of this section is to show that the memory concepts described above are generally applicable to a broad range of psychopathology found in DSM-IV (APA, 1994). Of special relevance are the concepts of energy well and basin of attraction (Tryon, 1995a). General reference to neural networks will occasionally be made when appropriate. Hypotheses about DSM-IV (APA, 1994) disorders are presented, implications for treatment are considered, and recommendations are made for future research. What follows are suggestions regarding new ways to think about clinical disorders based on a few neural network principles rather than definitive resolutions and proofs supported by empirical research. Treatment implications and directions for future research follow are subsequently considered. One approach to this section is to select one or two specific DSM-IV diagnoses within a single DSM-IV category to illustrate the theoretical relevance of neural networks. Since the underlying logical structure of this section is induction, critics could rightly question whether the BAM applies to any other disorders. Hence, I chose to address the question of generality by discussing multiple diagnostic entries within each of the three major DSM-IV categories: Dissociative, Anxiety, and Mood Disorders.

76

Chapter 3

Dissociative disorders Carson, Butcher, and Coleman (1988) define dissociation as the "separation or 'isolation' of mental processes in such a way that they become split off from the main personality or lose their normal thought-affect relationship" (p. G-5). Hence, dissociation is a disruption of normal associative processes. Such a dysfunction could be due to problems with memory formation and/or recall. It is important to determine if a person with a thought-affect isolation: (1) ever had appropriate emotions, (2) had appropriate emotions but lost them at some point, or (3) retains appropriate thought-affect associations for some topics but not others. A neural network explanation of the first case implicates abnormal learning experience, abnormal neural architecture, and/or abnormal memory formation process. Excessive axonal pruning such as Hoffman (1987, 1992) and Hoffman and Dobscha (1989) have reported in adolescent schizophrenics can explain why they had normal cognition as children but form "loose" associations as adults. Traumatic or toxic insult to the brain of adults can explain why normal thought-affect associations are lost at some point. Having appropriate thought-affect associations for some topics but not others implicates specific learning experiences. Spiegel (1990) discusses three theoretical advantages of parallel distributed processing (PDP) models of dissociation. First, the autoassociative pattern completion property of neural networks causes them to recall a complete memory given partial information. This means that one need not be fully conscious of all aspects of a stimulus situation before reacting to the situation. Second, neural networks entail local learning without governance from a central processing unit. Learning can therefore take place at different levels with varying degrees of consciousness. Third, "The concept of dissociation implies some kind of parallel access to awareness" (p. 123); hence, PDP models intrinsically reflect a fundamental property of dissociation. Hilgard (1977) explained hypnosis using a horizontal view of conscious states, as did Janet (1920), versus Freud's vertical model. Conscious states are seen as existing side by side like rooms in a one story ranch house. Dissociation entails access to some rooms but not others. Hilgard hypothesizes that hypnosis activates two or more of these distributed conscious states. Dissociative Identity Disorder (multiple personality). An extraordinary consequence of massive and systematic dissociation can be the formation of two or more personalities each with their own distinct set of memories and

W. W. Tryon

77

associations. One neural network explanation of two personalities entails the creation of two large crater-like basins of attraction in an otherwise fiat memory surface. While the locus of memory formation/recall remains in the first crater, normal size memory wells form in the craters' floor. The crater walls usually prevent the locus of memory recall from flowing into the other crater, super basin of attraction. However, a sufficient temporary energy increase might jump the locus of memory formation/recall out of the first crater and into the second one thereby causing a personality shift. An indirect transfer could occur by first boosting the locus of memory formation/recall onto the plateau dividing the sunken areas and "rolling", converging, into the other basin thereby accomplishing a personality shift. This argument assumes that momentum from the ejection is sufficient to traverse the fiat middle section. While in the second crater, normal size memory wells would form in its floor until the locus of memory formation/recall was again ejected and returned to the first associative area. An alternative means of isolating memory formation sites into two functionally separate regions would be the formation of a wall or mountain range like structure such that the locus of memory formation/recall would normally be contained on one side; in one valley or the other. This formulation differs from the one above in that a thinner barrier rises up from a fiat plane rather than requiring two large depressed areas to be formed. As before, unusual circumstances may push the locus of memory formation/retrieval over the top of this elevated structure allowing it to roll (flow) into the other side, valley. Alternatively, it is possible that the dividing structure does not extend completely from one end of the memory field to the other. Perhaps it occupies the middle 90 percent leaving a 5% region at each end where it is possible to move from one side, valley, to the other without crossing the barrier. Or perhaps the dividing structure begins at one end of associative memory and extends 90 percent of the way across the memory field leaving a 10% transfer section at one end. A third possibility is that the barrier extends from one end of the memory field to the other but with one or more significant breaks along the way providing one or more paths to the other valley. A variant of all three options is that the barrier might extend from one end of the memory field to the other but have variable height such that at one or more points it becomes low enough that extraordinary energy increases are not needed to cross the barrier at these places; like crossing a mountain pass. A fourth possibility is that the barrier might have variable thickness such that the locus of memory formation/recall might tunnel through, penetrate, the barrier at its thinnest points analogous to tunneling in

78

Chapter 3

quantum mechanics. Additional alternatives can be created using different combinations of these possibilities. The nature of memories/associations occupying all transfer regions is especially important because activity in these areas positions the locus of memory recall/formation where the probability of crossing to the other side is substantially greater than with other topics. Future investigators should determine the topography of associative structures to comment upon the above hypotheses. Dissociative Amnesia. DSM-IV (APA, 1994) cites the "... inability to recall important personal information, usually of a traumatic or stressful nature, that is too extensive to be explained by ordinary forgetfulness" as the primary inclusion criterion. To explain Dissociative Amnesia via the BAM model is to explain how to prevent memory retrieval from occurring. We consider two possibilities. Energy hills. The formation of energy wells in one area of the memory surface entail the formation of energy hills in one or more other areas of the same surface. Memory encoding for every association also encodes its complement. The complement of the vector 1, 0, 1, 0 is the vector 0, 1, 0, 1. The energy values for both memories are equal but of opposite sign. Memories have negative energy values and their complements have positive energy values resulting in energy hills. Memory hills can extend the basin of attraction of a memory well if they are adjacent to the well. A locus of memory retrieval located anywhere on the side of the energy hill adjoining the memory well will descend into the memory well. Since memory recall entails seeking energy minima, an energy hill prevents memory recall. If a memory hill were to be created in a path normally taken toward a memory well, then the associative process would be blocked in direct proportion to the diameter of the base of the memory hill. Perhaps a series of adjacent energy hills could wall off a memory well thereby isolating it. EEG. If the polarity of the BAM remained constant, then the contents of memory hills would not ordinarily be accessed. Additional energy would be required to move the locus of memory recall up the hill. This problem is solved by temporarily reversing the sign of all BAM vectors which temporarily converts memory hills into memory wells thereby allowing retrieval of memory opposites consistent with Ryehlak's (1981) dialectical emphasis. This process is especially efficient when memory hills border directly on memory wells such that the surface of the well wall and the hillside are contiguous or nearly so.

IV. IV.. Tryon

79

Memory access takes time. If the memory well is deeper than the product of memory retrieval speed and repolarization time, then the memory well will be repolarized into a memory hill before the locus of recall reaches the bottom and the memory is retrieved. This could account for the inability to retrieve traumatic memories since they are hypothesized to be associated with deep memory wells. Time is reciprocally related to frequency. Faster cycling results in less time per cycle. More time per memory cycle would be available in a relaxed state where polarity reversals occur more slowly thereby providing more time to access deeper wells and consequently allowing access to information not available during a more aroused state. Martin (1991) labels the 0.5 - 4 Hz EEG band as delta, the 4 - 7 Hz band as theta, the 8 - 13 Hz band as alpha, and the 13 - 30 Hz band as beta. Martin (1991) indicates that "Beta waves are normally seen over the frontal regions and over other regions during intense mental activity" (pp. 778-779). Slower "alpha waves are generally associated with a state of relaxed wakefulness" (p. 778). Martin further indicates that delta and theta waves are associated with sleep and have the largest amplitudes. These results are in exact agreement with BAM model expectations. Rapid polarity changes would quickly give alternate access to memory hills and wells but would prevent memory access to deep memory wells. This hypothesis is consistent with rapid beta waves being associated with intense mental activity. It also explains why anxiety impairs memory (e.g., test anxiety). The rapid EEG oscillations associated with hyper arousal provide little time for memory access. A slower rate of oscillation provides more time to reach deeper memory wells. This hypothesis is consistent with better memory for traumatic events under conditions of relaxed wakefulness where slower alpha waves predominate such as during hypnosis. The slowest EEG waves are associated with sleep and give the longest time to probe deep memory wells and consequently access more traumatic memories, perhaps in the form of dreams since we are not conscious when these memories are being retrieved. Such slow alternate access to memory locations may further impair our ability to effectively process the recalled information. Still deeper memory wells may not be accessed at all and thereby are fully dissociated from consciousness. Perhaps these deepest recesses could be probed if ways were found to further slow the EEG or to increase the proportion of slower EEG waves. A corollary to the above argument is that all stimulants defend against memory retrieval by limiting repolarization time. This could explain why some people chronically seek stimulation.

80

Chapter 3

The possibility that the energy values associated with memory wells could oscillate is entirely a theoretical conjecture at this point. While the impression may have been given that memories are stable in all respects it may be that while their structure, shape, remains constant that their polarity oscillates. Just as there is no empirical evidence that memory wells can oscillate with regard to polarity, no empirical evidence indicates that they cannot do so. The possibility of oscillation is theoretically attractive in the ways described above and deserves further consideration. These hypothetical memory polarity changes are not necessarily explained by the same mechanisms that govern EEG signals; they could have an entirely different physical basis but c,ovary with EEG. Dissocmtive Fugue. The same dissociative mechanisms discussed in connection with Dissociative Amnesia above may also operate here. Avoidance sometimes entails removing ones' self to a different location. Elements of Dissociative Identity Disorder (multiple personality) are present to the extent that the person assumes a new identity elsewhere. Depersonalization Disorder. This disorder appears to involve a mild form of the dissociative mechanisms discussed above in that memory is not lost for specific events. Anxiety disorders Posttraumatic Stress Disorder (PTSD). Although not explicitly connectionistic in nature, Chemtob r al. (1988), Creamer, Burgess, and Pattison (1992), Foa and Kozak (1986), Foa and Riggs (1993), Foa, Steketee, and Rothbaum (1989), Foa, Zinbarg, and Rothbaum (1992), and Lang (1979, 1985) theoretically implicate an emotion-memory network in the etiology of PTSD. Their work is important here because it emphasizes the concept of network, and by implication, the parallel distributed processing approach to PTSD advocated below. Both Jones and Barlow (1990) and Litz (1992) cite Lang's (1985) emotion-memory network (cf. Lang, 1979). Leventhars (1984) perceptual-motor theory of emotional response entails associative memory structures which neural networks clearly are. Li and Spiegel (1992) explain both PTSD and Multiple Personality Disorder in terms of traumatic constraints placed on a neural network which alter the topology of its "goodness-of-fit surface" which conforms to the memory energy field discussed above. Jones and Barlow (1990) require the following characteristics of a comprehensive PTSD theory: 1) symptom constellation of the disorder, 2)

W. W. Tryon

81

differential symptom severity, 3) reexperiencing the trauma through memory, and 4) why only some people develop PTSD given a traumatic experience. The DSM-IV (APA, 1994) symptom constellation includes: 1) persistently reexperieneing the traumatic event, 2) persistent avoidance of stimuli associated with the trauma and a general emotional numbing, and 3) persistent symptoms of arousal. Brewin, Dalgleish, and Joseph (1996) specified five points that a complete PTSD theory must address. Two of these issues are the same as Jones and Barlow's criteria and three are different. The two that are the same are: a) to account for the clinical characteristics of PTSD and b) to explain individual differences in symptom severity. The three additional criteria are: a) to indicate whether PTSD symptoms are indicative of an abnormal process and if not how they differ from normal processes, b) why comorbidity is found with depression, generalized anxiety, substance abuse, and somatization disorder, and c) doing a better explanatory job than other theories plus making novel predictions. This third requirement contains two distinct parts. The first concerns comprehensiveness of explanation and the second entails new and unique predictions. Tryon (1996b) provides a BAM explanation of all required aspects of PTSD. Because the DSM-IV (APA, 1994, pp. 428-429) diagnosis of Posttraumatic Stress Disorder requires symptoms to exist for at least 1 month, and because the diagnosis of Acute Stress Disorder (APA, 1994, pp. 431-432) requires symptoms to last between 2 days and 4 weeks, the following comments pertain to both disorders. Combat. War experiences are hypothesized to warp existing memory energy fields by creating new deep memory wells with broad basins of attraction. Stimuli which previously flowed to normal memories and associations now reside within the basin of attraction of a war related memory and therefore retrieve war memories. The energy minima seeking nature of cognition and the steepness of the war-related energy well walls causes the locus of memory retrieval to pass through or by a previous terminus toward the deeper, more compelling, war-related memory/association. Unable to prevent this associative process from reaching energy minima, the PTSD veteran copes by minimizing the frequency of war related associations through avoidance of all stimuli associated with the broad basin of attraction resulting in emotional numbing in direct proportion to the scope of the attractor basin. The clinical appearance is that many of their memories work in unison. To reduce the frequency of war-related feelings, PTSD patients find it helpful to inhibit all feelings resulting in emotional numbing.

82

Chapter 3

Since a separate memory well is associated with every memory, many deep memory wells may exist, each with its own broad basin of attraction. PTSD severity would therefore be a function of the depth and breadth of all attractor basins plus their number and distribution across the memory field. Perhaps these memory wells arc separated by broad relatively fiat areas or they may be sufficiently close to one another that their basins of attraction intersect yielding a complex super basin of attraction. Either of these eases could explain the broadly generalized nature of EN. Future research should attempt to map the basins of attraction to determine the frequency with which each of these possibilities exists. The present view of EN is consistent with Keane et al.'s (1985) view of emotional numbing as avoidance motivated and Litz (1992) who argued that PTSD veterans retain the ability to fed normally. Neural networks can be cascaded to form a series of associations thereby introducing the possibility of connecting associations. If an association is connected to a war-related memory, then any stimulus associated with the connecting memory will evoke the war-related memory. Future investigators should examine the topography of memory fields containing connecting associations to learn more about how such systems work. Rape~incest. Emotional numbing is also characteristic of rape and incest victims. In the ease of adult rape, sexual behavior that may have been previously associated with love and affection is traumatieaUy associated with fear, anger, and other strong negative emotions. Synaptic weights undergo important changes resulting in observed symptoms. Because it is generally much easier to destroy than to build up, and because learning about life threatening events require one-trial learning to minimize fatalities, larger changes in synaptie weights may result from aversive than positive experience; a point for future research to clarify. For example, a woman develops into a normal adult over say 25 years as the result of many constructive experiences. Yet a single rape experience can compromise so much of what took so long to develop. One possible neural network mechanism is that memory formation is a relatively rapid process and that traumatic memories produce broad deep memory wells much as an intense explosion rapidly creates a large crater. This makes evolutionary sense in that memory for nearly fatal behaviors must be formed quickly and retained across the organisms life span so that this behavior is not repeated. Efforts to avoid recalling memories associated with rape, and the associated negative emotions, require avoidance of all relevant cues including normal sexual

W. W. Tryon

83

situations with husband/lover which unfortunately strains an important source of social support sometimes to the point of separation or divorce and that occasions the onset of more problems thereby further complicating her readjustment. In the case of child sex abuse and incest, the young person, usually female, is much smaller and weaker than the full grown, usually male, offending adult. When the adult is also the child's parent, then the victim is dependent upon the abuser for emotional support as well as food, clothing, and shelter. Further trauma is inflicted when a normally protective mother ignores her husband's child abuse. Parents can enlist police and other agencies to find and return their child should she or he run away. Lack of education and financial resources further curtail the child's running away. These conditions probably create large deep aversive memory wells with broad basins of attraction that produce enduring psychopathology which warps the memory surface associating thoughts, feelings, and actions. Given that physical escape from ongoing abuse is not possible and that the related aversive associations are unavoidable, dissociation becomes likely. One might try to forget during traumatic memory forming experiences. One method is to concentrate on an external stimulus such as a ceiling light to reduce awareness of the traumatic events in progress thereby minimizing their present impact and impairing memory formation regarding these events. More complete dissociation affords greater psychological protection. However, the reconstructive aspect of associative processes can create the full memory from a portion of it. Partial cues, can sometimes evoke the entire traumatic memory. Hence, dissociative strategies are only partially effective unless extreme. DSM-IV (APA, 1994) treats anxiety disorders as entirely separate from dissociative disorders. The above considerations indicate an important overlap between the two classes of psychological disorder. Obsessive-Compulsive Disorders. In the absence of a DSM-IV (APA, 1994) definition, we return to DSM-III-R (APA, 1987) which defines obsessions as "... persistent ideas, thoughts, impulses or images ..." (p. 245). All of these symptoms are associative in nature and can therefore be addressed from the neural network perspective. A prominent feature of obsession is that certain associations are highly repetitive. Neural networks can be autoassociative which means that Stimulus A evokes Stimulus B which can elicit Stimulus A and repeat the cycle or may recall Stimuli C, D. E, etc. before recalling Stimulus A and repeating once again.

84

Chapter 3

Kosko (1988) indicates that the continuous BAM digs its own energy (memory) wells. Excessive worry, while perhaps initially reality based, might create an association which dominates a region of memory surface by virtue of an excessively broad basin of attraction and deep energy well. The broad basin of attraction will cause many stimuli to evoke the same memory, association. Each obsession may have its own memory well and attractor basin. These memory wells may be widely spread across the memory surface and therefore be independent of each other or they may reside sufficiently close to one another that their basins of attraction intersect forming a complex super basin of attraction. Future investigators should attempt to determine the topography of these attractor basins. An obsessive episode refers to a period of time during which associative processes operate within an abnormal attractor basin. The trigger stimuli locate the associative process within the relevant attractor basin. How one escapes from the influence of the attractor basin is less obvious. One possibility is to suspend the associative process. This could be done through meditation aimed at clearing one's mind of all thoughts. Or it could be done by initiating behaviors incompatible with thought such as reading aloud or singing. Another possibility is to remove oneself from the cliciting stimulus by leaving the situation. A third possibility is to initiate another line of association such as engaging a cross word puzzle or mathematical or logic proof. This is the primary rational behind thought stopping techniques. Compulsions arc defined as "... repetitive, purposeful, and intentional behaviors that arc performed in response to an obsession ..." (APA, 1987, p. 245). Obsessions set the occasion for compulsions. Treatments directed at compulsions arc hypothcsize~ to work because of their effect on associative processes. Compulsive disorders can be explained similarly to obsessive disorders by substituting Behavior A for Stimulus B such that Stimulus A sets the occasion for Behavior A which elicits Stimulus A which again sets the occasion for Behavior A, etc.. Hence, the autoassociativc nature of the BAM can account for compulsions in addition to obsessions. The Hcbbian nature of the BAlM allows memory processes to create ever deeper, and perhaps broader basins of attraction, through repeated association. This would allow normal worry to escalate into obsession through excessive rcassociation that could set the occasion for acting consistently with the obsession. Pamc Disorder. Panic attacks entail the sudden onset of at least 4 of 13 somatic or cognitive symptoms (APA, 1994, p. 395). Jones and Barlow (1990) report that "Almost all patients presenting with Panic Disorder have a

IV.. IV. To, on

85

high incidence of negative life events preceding the first panic attack" (p. 309). "Situationally bound (cued) Panic Attacks" and "situationally predisposed Panic Attacks" (APA, 1994, p. 395) can be explained on the basis that elements of the present situation recall memories and emotions associated with the initial trauma. An important feature of neural network memory systems is that they can retrieve a complete memory from a partial stimulus. While this property facilitates normal perception given degraded stimuli, i.e., partial views, it also provides a basis for psychopathology. The immune system sometimes misperceives allergens as disease and mounts an inappropriate attack. Hence, memory systems may misrecall certain memory content especially when given only partial information. An abnormal variant of this process might include inappropriately recalling fearful material on the basis of a few stimulus elements. Perhaps this is what sometimes occurs in "situationally bound (cued) Panic Attacks" and in "situationally predisposed Panic Attacks" (APA 1994, p. 395). Maybe one or a few elements erroneously elicits fearful emotion. The pattern completion property of the BAM in combination with its well documented ability to associate responses with stimuli provides for the possibility that stimuli with only a partial similarity to those of the traumatic incident may mistakenly elicit unprovoked aggressive behaviors in traumatized combat veterans. Panic disorder may be explained by a theoretical connection with obsessions. We know that Hebbian learning mechanisms, and their BAM equivalent, generate, synthesize, increasingly deep memory wells and associated basins of attraction as the associative process repeats. This allows an obsessive associative process to dig a large energy well which then functions as a memory. Obsessions about traumatic events may therefore create the functional equivalent of traumatic memories. Such abnormal memories may elicit strong anxiety when accessed. Alternatively, there may be nothing unpleasant about the synthetic memory basin but its development might intersect with one or more basins of attraction associated with legitimately fearful memories thereby providing passage for the locus of memory recall from the synthetic basin to one or more another basins associated with legitimately anxiety provoking memories. Specific Phobia. Formerly called Simple Phobia, Specific Phobias entail fear cued by a specific object or situation (APA, 1994). If their phobia was created by a traumatic event, then memory for this event occupies the energy minimum, bottom, of the memory well. If no such event can be recalled, then memory for it may have been dissociated.

86

Chapter 3

The breadth of the basin of attraction determines the generality of the disorder. A single highly circumscribed simple phobia is modeled by a single deep memory well with a relatively small basin of attraction. Multiple phobias involve multiple wells which may either be in separate areas of memory or they may be sufficiently close to one another that their basins of attraction intersect to create a complex super basin. Social Phobia. The same comments apply here as to Specific Phobia with the qualification that the feared object is a social situation. Generalized Anxiety Disorder. The DSM-IV inclusion criteria for Generalized Anxiety Disorder (GAD) include excessive worry for at least six months over concerns that are not limited to a particular object, setting, event or medical condition that the person finds difficult to control. It seems unlikely that such diffuse anxiety can be explained on the basis of multiple traumatizations. Partial pattern completion is one possible explanation of GAD. Perhaps persons with GAD have been traumatizexl but the memory completion process is both partial and limited to affcctive components. This would explain why a variety of ordinary stimuli would elicit anxiety without the ability to describe the basis for feeling anxious. The nonrecall of cognitive components would preclude being able to say why they felt anxious. The conditions under which partial pattern completion can occur, if it can occur at all, are presently unclear. Another explanation is based on the fact that CNNs can form spurious memories for events that were never experienced. These are composite memories derived from two or more nearby memory wells. If a spontaneous memory develops in an area populated by unpleasant and fearful memories, then a spurious nonspecific fearful memory might result. Hoffman (1987, 1992), H o ~ and Dobscha (1989) and Hopfield, Feinstein, and Palmer (1983) describe the formation of parasitic, spurious, memories as a consequence of normal memory formation. Uncertainty remains regarding the clarity such memories may have. These memories may be clear or they may be diffuse. The haphazard manner of their formation suggests that it is more likely that memory formation is fuzzy than clear. If these memories entail anxious content, then stimuli able to recall such memories can explain the presence of diffuse anxiety. It is entirely possible that the so-called False Memory Syndrome is based on spurious memories meaning that the memories derive from the interaction of memory wells and not from actual experience. Hopfield et al. (1983) have demonstrated that unlearning spurious memories improves the normal memory function of Hopfield CNNs. Crick

W. W. Tryon

87

and Mitchison (1983) hypothesize that dreams entail unlearning. If a problem exists with this purgative process then spurious memories would be retained despite repeated dreams about their content thereby explaining the persistence of diffuse anxiety over time. Neuropsychology of Anxiety Disordersl Current Cognitive-behavioral theories maintain that emotions, such as anxiety, are a consequence of thought (frontal lobes). Emotions are believed to occur because of attributions we make; what we tell ourselves (Ellis, 1962, 1980). Cognitive psychotherapy of emotional disorders largely entails altering irrational attributions and beliefs. Segal and Blatt's (1993) contributors consistently recommended that affect should be better integrated into cognitive/models of psychopathology. Neuropsychological evidence regarding the formation of emotional memories, like those associated with anxiety, phobias, panic attack, and posttraumatic stress disorder, indicate that major changes to cognitivebehavioral theories need to be made. Subcortical pathways play an important role in the formation of emotional memories (LeDoux, 1994) that are not considered by contemporary cognitive theories. Neural network models, such as the BAM, readily lend themselves to accounting for both cortical and subcortical associations. LeDoux (1994) reviews neuropsychological research, spanning at least the last decade, that clarifies the phYsiological basis of conditioned fear. Aversive stimuli inform the thalamus which jointly informs the lateral nucleus of the amygdala and the cortex. The lateral nucleus of the amygdala directly communicates with the central nucleus of the amygdala which initiates physiological and behavioral changes via the brain stem. The lateral nucleus communicates indirectly with the central nucleus through the accessory basal nucleus and the basolateral nucleus of the amygdala. This subcortical system is shorter than the cortical route and provides a more immediate response. The cortex informs the lateral nucleus of the amygdala as does the thalamus but with higher resolution, more fully processed, information. The cortex also informs the hippocampus which communicates with the lateral nucleus of the amygdala. The only site where lesions can be made without interfering with the ability to learn conditioned emotional responses is the cortex. If subjects are conditioned while intact and the cortex lesioned subsequently, the conditioned emotional response is disrupted. However, this effect appears to be because the lesions interfere with long-term memory retrieval since the conditioned emotional response partly recurs when reminder cues are presented. The same brain pathways appear to be involved

88

Chapter 3

in all mammals and possibly in all vertebrates. Many of these species are not known for their higher cognitive processes and all of them do not have language nor do they make attributions. LeDoux (1994) reports that emotional memories are long lasting and the result of long-term potentiation (LTP) of NMDA receptors to the neurotransmitter glutamate. This synaptic alteration causes larger postsynaptie responses to the same neural signals. The above comments are consistent with the BAM in two important ways. First, the BAM, like all connectionist neural networks, learns by changing synaptie weights which is what LTP entails. Second, the BAM is equally capable of associating emotions with stimuli and emotions with emotions as it is in associating stimuli with responses and stimuli with other stimuli; the BAM is a general associative mechanism. Hence, it is possible for stimuli to directly cue emotions. Because neural network models derive their functional properties by interconnecting simple neuron-like elements and changing connection (synaptic) weights as a result of experience, they can be directly informed by advances in the neurosciences. Traditional cognitivebehavioral theories have largely ignored neuroscientific evidence because they focus exclusively on psychological processes; they do not have the bridging properties inherent in neural network models. Mood disorders Depression involves repetitive self-deprecating associations of hopelessness, uncontrollability, and worthlessness (excessive or inappropriate guilt), among other symptoms including psychomotor retardation or agitation, insomnia or hypersomnia, difficulty concentrating, thinking, making decisions, and reAucexi participation in occupational and social events. At least three causal hypotheses exist regarding the relationship between cognition and depression. The first possibility is that cognitive style causes depression. A second possibility is that cognitive change is part of the depressive disorder and emerges simultaneously with other depressive features. A third possibility is that cognitive changes occur as a consequence of being depressed. I argue for the first, and especially the second, and against the third possibility in the remainder of this section. The first etiological possibility is represented by Abramson, Seligman and Tcasdale (1978) and Seligman, Abramson, Scmmcl, and von Bacycr (1979) who describe a dcpressogcnir attributional style (DAS) where persons invoke internal, stable, and global explanations of negative events, and to a

W. W. Tryon

89

lesser extent, external, specific, and unstable explanations of positive events as causally relevant to the onset of depression. Causal relevance does not imply either a necessary or sufficient condition but only that a risk factor exists that can range from small to large. Beck (1967, 1976) and Beck, Rush, Shaw, and Emery (1979) argue that psychosocial stress precipitates depression in persons who already emphasize negative outcomes, over generalize, magnify the importance of negative events, and think in absolute "all-or-none" terms. A CNN view of this hypothesis is that the pattern of synaptic weights that instantiates the depressive attributional style is consistent with, and partially implements, the pattern associated with depression but is insufficient to give rise to a diagnosable mood disorder. However, the more intense the attributional style, the closer the synaptic weight pattern is to a depressed state. That depressive attributional style is a risk factor for and not a necessary condition of depression is demonstrated by the fact that physical illness, alcohol and cocaine dependencies, death of a loved one, marital separation and/or divorce, and childbirth can elicit a Major Depressive Episode (APA, 1987, p. 221). Hence, a much broader range of premorbid personalities than those characterized by a DAS can become depressed. These events may also change synaptic weights so that they converge towards a depressive state. The second etiological hypothesis is that cognitive and emotional changes that characterize depression emerge simultaneously as a consequence of a particular pattern of synaptic weights. I base this hypothesis on the theoretical position that the functional attributes of a CNN are jointly dependent upon its architecture and pattern of synaptic states (connection weights) across the network. I conclude that architecture is not critical to depression because virtually all depressed persons have a history of normal emotion and cognition as children and perhaps as adolescents and some portion of adulthood prior to the onset of depression. Moreover, depression remits either naturally or as a consequence of treatments. Several investigators (Dobson & Shaw, 1986; Eaves & Rush, 1984; Hollon, Kendall, & Lumry, 1986) have reported an increase in negative thoughts during depression plus a return to normal when depression remits. Changes in neural architecture probably do not occur these cases. Hence, depression probably results from the pattern of synaptic weights. It is these connection weights that determine the graphs showing the structure of the basins of attraction associated with memories.

90

Chapter 3

The present CNN hypothesis about the etiology of depression assumes that cognition and emotion are encoded as a single vector thereby integrating intellectual and affective content as described above. It is hypothesized here that depression warps the memory field such that basins of attraction associated with normal memories that entail both positive and negative affect now lie within a super basin that leads to a global minimum associated with negative affect. This condition causes stimuli associated with normal memories to consistently return negative affect. Put otherwise, the part-whole pattern completion CNN property now returns negative emotion most of the time. This memory effect has the following hypothesized causal consequences. That many different stimuli (situations) return negative affect leads to a global attribution. That the pattern completion process is highly replicable leads to a stable attribution. The absence of precipitating external events, or the over reaction to events such as personal loss, leads to an internal attribution. The uncontrollable nature of the pattern completion process leads to hopelessness. Hence, the cognitive distortions associated with depression are a consequence of a distorted memory process associated with a broad and deep basin of attraction. A consistent preexisting DAS means that the network is already partially trained toward a depressive configuration and therefore requires less change to reach a state sufficient to give rise to cognitive and affeetive characteristics associated with a clinical diagnosis of Major Depression. DAS is therefore a risk factor as acknowledged above. However, these changes can occur in anyone following a depressogenic event therefore explaining why DAS is not a necessary condition. Storing emotion and cognition in the same vector fuses these two aspects of experience into the same memory. The part-whole pattern completion property of CNNs means that an emotional stimulus, or partial stimulus, can recall other emotional elements and all cognitive elements associated with that memory thereby explaining state-dependent (mood-congruent) learning and recall (Matt, Vazquez, & Campbell, 1992). Depressive mood may also steepen memory well walls thereby reducing the time taken to recall such material (el. Blaney, 1986; Bower & Cohen, 1982; Isen, 1984; Teasdale & Fogarty, 1979; Williams, Watts, MacLeod, & Mathews, 1988). The third etiological assumption is that negative affect is a consequence of, comes afLer, depression. Lewinsohn, Steinmetz, Larson, and Franklin (1981) provide longitudinal data showing that cognitive distortions are a consequence, rather than a cause, of depression. The absence of accurate data regarding the delay between the onset of depression and onset of cognitive

145.W. Tryon

91

distortions allows for the possibility of simultaneous emergence and therefore compatibility with the second alternative discussed above. A major problem with the sequential hypothesis is that it presumes two separate processes where the first (emotion, depression) influences the second (cognition). This raises the problem of how one process influences the other. This problem is reminiscent of the mind-body problem where the distinction between mind and brain created the problem of how they interacted. Mind as emergent from brain "solves" the problem by not making the initial distinction. I suggest that the same lesson applies to the cognitive and affective aspects of depression. All treatments are hypothesized here to exert their effects by altering synaptic weights. Pharmacological treatments attempt to directly alter synaptic function. Kandel (1991) has shown that learning entails long term synaptic change due to the synthesis of new proteins under genetic control. Hence, psychological treatments entail biological changes. That both treatments have a common effect makes it understandable why combined intervention often works best. This view should make psychologists more understanding of and respectful toward drug treatments and psychiatrists more understand of and respectful toward psychological treatments. For example, the effectiveness of Beck's cognitive-behavioral treatment can be explained on the basis that experience alters synaptic weights. Therapeutic experiences, both in the office and during homework, may alter the synaptic weights in ways that normalize memory processes. As another example, Bellack (1985) reported that pharmacologic therapy normalizes depressive cognitions as effectively as cognitive therapy. Long term success in either case depends upon the durability of synaptic change. Segal and Blatt's (1993) contributors called for the integration of affect into cognitive/models of psychopathology. Encoding emotion and cognition in the same vector seamlessly integrates these two important aspects of human experience. It was previously observed that traumatic events, such as rape, exert a large negative effect in a small amount of time. Perhaps affective intensity exerts strong influence over the formation of memory wells thereby creating a dominant, controlling, influence for a long time. It makes evolutionary sense that memory formation would be much more potent for potentially lethal events than for positive ones. Forgetting a potentially lethal experience could result in death if repeated even once whereas forgetting a positive encounter would not have this effect.

92

Chapter 3 Treatment

Learning therapy Psychotherapy and behavior therapy share a common assumption that therapeutic change entails new learning; theorists differ mainly in what they believe is learned. Some therapists emphasize the role of reinforcement contingencies while others accentuate learning now attributional styles or other information processing strategies, but all agree that therapeutic changes are learned; otherwise all but purely biological treatments would be ineffective and pointless. Neural networks are learning mechanisms and consequently provide proximal causal explanations for how infrahuman and human learning can occur (Tryon, 1995b). This scientific base provides a unified perspective capable of organizing the disparate treatment techniques associated with behavior therapy. I have previously termed this general connectionist approach to learning as Neural Network Learning Theory (NNLT: Tryon, 1993b). Computer models of memory formation and alteration, such as the BAM, can serve the same heuristic function as animal models of psychopathology, they provide a well controlled context in which to study therapeutic principles. This is not to say that animal research will be replaced by computer simulation but rather that computer simulation can be used as a productive new tool for evaluating hypotheses. Eysenck (1964) maintained that "Bchaviour therapy may be defined as the attempt to alter human bchaviour and emotion in a beneficial manner according to the laws of modern learning theory" (p. 1). Wolpe and Lazarus (1966) agreed that behavior therapy entailed "... the application of experimentally established principles of learning" (p. l). However, Kazdin (1978) noted that "The definition of behavior therapy has been broadened, and the role of learning theory has been reduced substantially to the point that the precise role of learning theory in actual practice of behavior therapy has been questioned" (p. 195). Kazdin (1979) criticized the claim that behavior therapy was based on modem learning theory was a fiction. Spiegler and Guevremont (1993) argued that the statement, "Behavior therapy is the application of well-established laws of learning," is currently "predominantly false" (pp. 4-5). Connectionism in the form of NNLT provides a learningmemory base for all psychological therapies, including behavior therapy. The full spectrum of behavior therapies can once again be said to rest on CNNs as modem learning theory. Whereas traditional animal models of

W. W. Tryon

93

psychopathology are seen as too limited to form a general framework for behavior therapy, neural networks have the required broad relevance to both animal and human perception, cognition, memory, and behavior. Three corollaries of learning can be distinguished from our neural network perspective. First is that new learning, information, is linked with previous knowledge. Developmental processes, such as cumulative hierarchical learning (of. Staats, 1986), are good examples. The BAM mechanism discussed above illustrates one mechanism for how these changes can occur. Connectionist systems blend knowledge sources because multiple associations are encoded by modifying the same set of synaptic weights. This aspect of therapy entails new learning to provide missing or inefficient interpersonal and other skills required for "normal" performance. A second corollary of learning therapy is unlearning; attempting to undo the effects of prior learning that has caused the person problems. Systematic desensitization is a good example. A new, and more normal, response is learned to stimuli and events that previously produced fear and avoidance. The BAM model provides a possible mechanism for unlearning (forgetting). Any specific memory can be deleted by storing its complement. For example, if stimulus complex 1,1,0,0,1,1 has been stored then all memory for it can be erased by storing stimulus 0,0,1,1,0,0. Put more informally, learning that dogs are kind affectionate house pets removes the association that they are cruel vicious wild animals. The relevant substitutions are kind-cruel, affectionate-vicious, house pets-wild animals. Personal experience, verbal association, and/or observation all function to store inverse memories thereby ameliorating pathological memories/associations. Alternatively, unlearning can be accomplished by storing a new memory formed by associating the opposite characteristics. It follows that therapeutic efficacy is directly proportional to the extent to which all elements of a stimulus complex have been addressed and is directly proportional to the degree to which each element can be fully inverted. It may only rarely be possible to completely erase clinically relevant memories in this way but may well suffice to bring much appreciated relief. Since memory formation entails generating a well of some depth with a basin of attraction of some breadth, it follows that memory removal by storing the exact opposite association reduces both the depth and breadth of the memory well to zero. It also follows that storing intermediate associations should reduce both the depth and breadth of memory wells. Perhaps the reversal effect is a nonlinear function of similarity to an exact complement. Normal forgetting could be seen as the result of incidental complementarity

94

Chapter 3

as a result of new memory formation. This makes forgetting an active rather than passive process. The third corollary of learning theory is that the pattern completion property of the BAM allows one to facilitate memory recall in at least two ways. The first method is based on the fact that content addressable connectionist neural networks, like the BAM, are capable of recalling a entire memory given only a memory fragment. Hence, by systematically focusing on partial sights, sounds, tactile and olfactory sensations, and kinesthetic cues, more complete memories often form. The second proeexture for facilitating memory entails mood-congruent (state-dependent) recall (Matt et al., 1992). This method is based on the finding that affective state influences memory recall. Happy memories are more available when one feels cheerful and sad memories are more available when one feels depressed. Affect appears to be stored with content and consequently, content which may not be readily available for recall may be remembered in the presence of affective cues. Both methods can be combined by first inducing an emotional state, or capitalizing on a naturally produced one, and presenting specific visual, auditory, olfactory, tactile, and/or kinesthetic cues. Old photographs taken during childhood and/or other memorabilia may be used. Unconscious processes. Experimental evidence has established that unconscious (implicit) learning (Seger, 1994) occurs, though in a much more limited way that Freud suggested (Greenwald, 1992; Jacoby, Lindsay, & Toth, 1992; Kihlstrom, 1987; Kihlstrom, Barnhardt, & Tataryn, 1992; Schacter, 1987). For example, implicit memory is demonstrated through the "savings" technique where material once learned, but forgotten, can be relearned in fewer trials than it initially took to learn. Savings result when the connection (synaptic) weights associated with the hidden middle units retain values close to what they were when the behavior was fully leamed. Fewer weight adjustments are required to satisfy the performance criterion used to demonstrate learning thereby resulting in savings. Serial information processing models have never satisfactorily explained how semantic analysis is possible for information only partially perceived. Kihlstrom (1987) indicates that parallel distributed neural networks solve this problem because: 1) unconscious processes can occur within the "hidden" network layer (el. Greenwald, 1992), 2); no central processing unit is involved, 3) there are no rules to be aware of, 4) partial effects can be exerted, and 5) neural networks merge perception and cognition thereby allowing unconscious processes to exert their influence (of. Greenwald, 1992). Learning occurs because the connection (synaptic) weights change.

IV.. IV.. Tryon

95

The same mechanism accounts for unconscious learning. The absence of a ruleogovcrned central pro~ssing unit (scl0 makes it unnecessary to explain unconscious learning differently from normal learning. That learning takes place in the same way regardless of whether the subject is aware or not makes the distinction pointless thereby rendering the conscious vs. unconscious schism a nonissuc. This is another example of how neural networks produce theoretical synthesis by making problematic schisms like (cf. Tryon, 1993a, b), previous problematic schisms become nonissucs 3. Grccnwald (1988) explains self-deception as avoidance based on partial perception. Perceiving a subset of cues may bc sufficient to occasion avoidance behavior even in the absence of full awareness of the underlying associative process. Grccnwald (1988) explains repression as self-deception based on memory versus current perception; avoidance behavior in response to a subset of recalled associations. Neural networks can implement both of these functions. The memory energy function (cf. equation 5 in the Appendix) suggests another view of repression. Positive energy states result in energy hills with basins of repulsion just as negative energy wells have basins of attraction. Repulsion results because memory recall entails seeking an energy minimum. Just as water does not flow up hill on its own, so also do associative

3 Tryon (1993a,b) discusses other instances where problematic schisms become nonissues. One is the mind vs. body issue. Terms like psychosomatic and somatopsychic attempt to explain how the mind can effect the body and how the body can effect the mind. CNNs provide existence proofs that mind emerges from bodily networks, that two separate entities do not exist, and therefore it is pointless to talk about how they interact. The behavioral vs. cognitive debate is also resolved by CNNs. CNNs are cognitive models developed to study the microstructure of cognition. Donahoe and Palmer (1989) and Tryon (1993b, 1995c, 1996c) have shown that connectionism is completely consistent with operant behaviorism. Consequently, connectionism is theoretically synthetic of the cognitive behavioral debate in the Hegelian sense of combining thesis and antithesis into synthesis. It is a single perspective consistent with two seemingly contradictory perspectives. Consequently this debate is now moot. A third schism is between human and animal research. These areas of research employ different vocabulary, concepts, and do not cite each others work. Tryon (1995b) shows that connectionism applies equally well to animals and humans. One vocabulary and set of concepts is applied across the phylogenetic scale. This list of theoretical syntheses is not exhaustive.

96

Chapter 3

processes not flow up hill. Recalling such memories would require an energy increase to reach the p e ~ of this inverted memory well. Unremembered traumatic events can influence both memory and behavior. Janet (1893, 1904) was perhaps the first to describe cases of hysterical (traumatic) amnesia in which frightening experiences were both unremembered and influential. He explain~ these disorders as dissociation of memories by emotion (see above comments on dissociative disorders). New learning. Unlike F rcudian theory which gives extraordinary influence to early experience, neural networks suggest that new learning continues to change synaptir weights throughout the life span. However, the accumulation of prior learning and the normal developmental decrease in neurotransmitters probably reduce plasticity over time. Psychotherapy equates to methods for producing new learning to correct and/or compensate for the effects of prior learning. Individual therapy. Learning can occur through experience with the therapist. This includes both the informational content of what the therapist says, the therapist's calm reassuring voice, and the therapist's relaxed posture plus other accepting/reassuring nonverbal cues. Thought stopping procedures can be prescribed on the basis that they inhibit the memory well enlargement described in connection with generalized anxiety and obsessive compulsive disorders above. New associative strategies may be found to shrink memory wells. Specialized therapeutics may be necessary for clients who have dissociated traumatic experiences. The phenomenon of mood-congruent recall (Matt et al., 1992), also known as state-dependent memory, suggests a possible approach. Whereas normal memory is biased toward recalling positive events, depression alters this bias toward negative events, in direct proportion to the severity of depression. Because experimentally induced dysphoria also negatively biases memory recall, the same or similar procedures might be used to temporarily augment depression to facilitate recollection of traumatic experiences. Care should be taken to avoid recalling too much during a single session and thereby further traumatizing the person. This would include terminating a memory search after one or two items were recalled in order not to overwhelm the subject. Each recollection would need to be examined in the session during which it was rctrieve~. Group therapy. ~ r n i n g can occur in group settings where clients interact with other people experiencing similar problems. The redundant credible information provided by group members should facilitate the learning process.

W. W. Tryon

97

Token economies. Bellack (1986) documented the effectiveness of token economies for schizophrenic patients. The therapeutic efficacy for this population can be explained by Hoffman's (1987, 1992) observation that memory energy flow over a significant Hamming distance (see Appendix) is vulnerable to attraction by large deep energy wells associated with the disorder. The clarity and consistency of contingent stimulus-consequence sequences associated with token economies might help patients with schizophrenia retain sufficient reality focus to function more normally. This formulation explains why they remain dependent upon highly structured environments; a behavioral prosthesis. Life experience. Learning can occur though personal experience outside of therapy settings. This experience can be in the form of carefully crafted exposure therapy guided by prior hierarchy construction or it can occur spontaneously. Pharmacotherapy Basins of attraction are calculated on the basis of synaptic weights and consequently anything that modifies synaptic weights also modifies basins of attraction. Three arguments implicate synaptic change as the basis of therapeutic improvement. First, cognitive change is a form of learning and learning has been shown to entail synaptic change (Donahoe & Palmer, 1994; Thompson, 1986, 1990). Second, pharmacotherapy influences neurotransmitters such as dopamine which are known to influence synaptic function. Third, Bellack (1985) reported that pharmacologic therapy alone normalizes, or changes, depressive cognition as much as does cognitive therapy. Because the two primary variables affecting the BAM network are depth of the memory well and breadth of its basin of attraction, it follows that the psychological and behavioral effects of pharmacotherapy result from altered synaptic function causing changes in either the depth of the memory well and/or breadth of its basin of attraction. Because shallower and narrower are associated with normal, I hypothesize that pharmacotherapy reduces the depth of BAM memory wells and diminishes their basins of attraction. The primary theoretical means of decreasing the energy values associated with stored memories is to work backwards, using Equation (5) from the Appendix (E = -SMR T) from energy (E) to memory matrix elements (M). The S and R vectors remain as they are because they describe external stimuli and responses. After reading the Appendix, the reader will realize this

98

Chapter 3

entails minimizing the SM multiplication which is done by minimizing the elements in the M matrix. This partly depends upon how many of the component products are of the same sign. More same signed products will create a larger sum whereas products of different signs will tend to cancel out leaving a smaller sum.

Electroconvulsive therapy (ECT) The therapeutic effects of ECT can be understood from a neural network perspective in one of two ways. The first explanation entails the assumption that depression results from a deep memory well with a large basin of attraction and that ECT produces a temporary energy increase like that utilized when training Boltzman neural networks using "simulated annealing". Annealing is a metallurgic process used to remove internal stresses and toughen metal and glass. The process begins by heating, energizing, the substance to a point above its melting temperature such that the atoms are in violent random motion. As the temperature is slowly lowered, the atoms gradually form a crystalline structure which forms a collective energy minimum. Simulated annealing is a method of escaping unproductive local energy minima to promote convergence on the global energy minimum corresponding to problem solution or optimal performance (cf. Wasserman, 1989, pp. 7783). Neural networks trained by annealing contain a "temperature" parameter in their learning function which controls the probability of the network being in a particular energy state. In the beginning, all energy states, including high ones, are essentially equally probable. A preference develops for lower energy states as temperature decreases but the possibility of temporarily jumping to a higher energy state remains. That the system sometimes increases its energy state before continuing its gradient descent toward an energy minimum often enables it to escape from a local energy minimum and move toward a global energy minimum. ECT may provide the temporary energy increase needed to escape a local energy minimum associated with a depressive memory well. ECT may propel the locus of memory recall/formation to another area of the memory field. The number of sessions required for clinical effectiveness may depend upon the depth of the memory well and the diameter of its basin of attraction. The need for repeated ECT sessions may be due to the nondirective nature the treatment. Increasing energy state does not necessarily project the locus of memory recall/formation in a predictable direction; its trajectory is likely to

W. W. Tryon

99

be random. The energy jump may relocate the locus of memory recall/formation to a point still within the current basin of attraction but through repeated efforts may escape this basin. Perhaps a more focal and/or phased application of energy would produce more specific and controllable effects. A limitation of this hypothesis is that it assumes that the energy well remains just as deep and wide as before and could subsequently reattract memory given prior stimulus conditions. This problem can be solved by postulating a second therapeutic effect. Perhaps ECT directly modifies the structure of the memory well; making it more shallow and narrow. Memory wells are deep by virtue of large negative energy values. ECT may make all memory energy levels more positive, in which case the well retains its relative depth. This process would effect all memories equally. A more therapeutic assumption is that energy is absorbed in direct proportion to the pre-existing negative energy level. Consequently, the most negative regions, deepest wells, absorb the greatest amount of energy thereby reducing them the most. The same process may also reduce the diameter of the attractor basin thereby explaining the possibility of a relatively permanent cure. ECT produces amnesia for events just prior to treatment. This phenomenon can be understood from the neural network perspective as disrupting the outer product matrix multiplication and subsequent matrix addition involved in creating long term memory storage and representation across the network (see Appendix). In so far as the patient is experiencing depressive symptoms at the time of treatment, the underlying associative process will be disorganized. Repeated disorganizations of a depressive complex may normalize the associative process. It may therefore be beneficial to have the person focus on their most depressing associations immediately prior to administering ECT.

Research Strategies Assessment I am unaware of any current methods for mapping the breadth and depth of memory wells. Such a technology needs to be developed because these two theoretical constructs are critically important and empirical tests of this model are dependent upon such assessments. Perhaps the first step is to exhaustively survey all existing methods of memory assessment to determine their suitability for mapping basins of attraction and depth of memory wells.

1O0

Chapter 3

All methods of measuring associative strength are potentially relevant to assessing the basin of attraction and possibly depth of memory wells. Clinical methods for ascertaining which events trigger which memories should also be considered. Treatment

The hypothesis that phobics have large basins of attraction, compared to normals, can be evaluated, in principle, by applying the assessment techniques to be developed to phobic and control subjects. The hypothesis that successful treatment shrinks the basin of attraction can be evaluated by comparing phobic subjects before and after treatment with empirically validated procedures and comparing the prc and post assessments with untreated control subjects. Consideration was given above under the heading of pharmacothcrapy as to how the depth of memory wells could be reduced by altering the memory matrix so that smaller energy values resulted. The brevity of this discussion partly reflects a lack of study of this issue. No theoretical conjecture or empirical evidence could be found relating learning based or any other psychological intervention to changes in the BAM or any other CNN. Perhaps readers of this chapter will have, and publish, additional ideas on this topic. The most that can be said at this juncture is that the mechanism of therapeutic action for learning based therapies will very likely be the same as for pharmacotherapy. This conclusion has an important implication for psychologists and psychiatrists (pharmacologists) especially when working together in medical settings. Connectionism emphasizes the compatibility of learning and pharmacological therapies; both of them are directed at synaptic change. Pharmacological agents rapidly change synaptic function but these changes may be temporary in which case relapse occurs when medication is removed. Learning based approaches to synaptic change often take longer but can produce more lasting change. The value of using both therapeutic methods concurrently is obvious. This view should make psychologists more understanding and supportive of psychopharmacology. Neuroleptic medications are making similar psychological changes to those produced by learning based therapies. I cite the changes in cognitive style that accompany remission (of. Dobson & Shaw, 1986; Eaves & Rush, 1984; Hollon et al., 1986) or successful antidepressive pharmaeotherapy (el. Bellack, 1985). Likewise, connectionism should make physicians more accepting of learning based therapies since they are altering the same synaptic functions as

W. W. Tryon

101

pharmacologists target. Hence, the theoretical synthesis provided by connectionism also provides a basis for more cooperative and respectful collegial relationships between psychology and psychiatry. That psychologists are creating brain changes through learning therapies has positive implications for reimbursement by third parties and inclusion within health care legislation. Connectionism makes it considerably more difficult for medical organizations to exclude psychological interventions. Conclusions

The bidirectional associative memory and its resulting memory field and funnel-shaped memory wells has many heuristic properties for understanding both normal and abnormal psychological processes and the behaviors they mediate. Like all neural networks, the BAM can be implemented on a digital computer and its functional properties studied in detail. The combination of compelling theoretical hypotheses and openness to complete experimental investigation set the occasion for many research opportunities. References

Abramson, L. Y., Seligman, M. E. P., & Teasdale, J. D. (1978). Learned helplessness in humans: Critique and reformulation. Journal of Abnormal Psychology, 87, 49-74. American Psychiatric Association (1987). Diagnostic and staastical manual of mental disorders (3rd ed. - Revised). Washington, DC: American Psychiatric Association. American Psychiatric Association (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: American Psychiatric Association. Anderson, J. A. (1983). Cognitive and psychological computation with neural models. IEEE Transactmns on Systems, Man, and Cybernetics, SMC13, 799-815. Banquet, J. P., Smith, M., & Giinther, W. (1992). Top-down processes, attention, and motivation in cognitive tasks. In D. S. Levine & S. J. Leven (Eds.), Moavation, emotion, and goal direction in neural networks (pp. 169-208). Hillsdale, NJ: Lawrence Erlbaum. Beck, A. T. (1967). Depression: Causes and treatment. Philadelphia, PA: University of Pennsylvania Press.

102

Chapter 3

Beck, A. T. (1976). Cognitive therapy and depressive disorders. New York: International Universities Press. Beck, A. T., Rush, A. J., Shaw, B., & Emery, G. (1979). Cogniave therapy of depression: A treatment manual. New York: Guilford Press. Bellack, A. S. (1985). Psychotherapy research in depression: An overview. In E. Beckham & W. Leber (Eds.), Handbook of depression: Treatment, assessment and research (pp. 204-219). Homewood, IL: Dorsey Press. Bellack, A. S. (1986). Schizophrenia: Behavior therapy's forgotten child. Behavior Therapy, 17, 199-214. Blaney, P. H. (1986). Affect and memory: A review. Psychological Bulletin, 99, 229-246. Blatt, S. J., & Bers, S. A. (1993). Commentary on "A cognitive perspective on self-representation in depression". In Z. V. Segal & S. J. Blatt, (Eds.), The self in emotional distress: Cognitive and psychodynamic perspectives (pp. 131-170). New York: Guilford Press. Bower, G. H., & Cohen, P. R. (1982). Emotional influences in memory and thinking: Data and theory. In M. S. Clark & S. T. Fiske (Eds.), Affect and cognition (pp. 241-331). Hillsdale, NJ: Lawrence Erlbaum. Carson, R. C., Butcher, J. N., & Coleman, J. C. (1988). Abnormal psychology and modern life (Sth ed.). Glenview, IL: Scott, Foresman & Co. Chemtob, C., Roitblat, H. L., Hamada, R. S., Carlson, J. G., & Twentyman, C. T. (1988). A cognitive action theory of post-traumatic stress disorder. Journal of Anxiety Disorder, 2, 253-275. Conte, H. R. (1975). A circumplex model for personality traits. Unpublished doctoral dissertation, New York University. (Dissertation Abstracts lnternaaonal: University Microfilms No. 2601731). Conte, H. R., & Plutchik, R. (1981). A circumplex model for interpersonal personality traits. Journal of Personality and Social Psychology, 40, 701-711. Creamer, M., Burgess, P., & Pattison, P. (1992). Reaction to trauma: A cognitive processing model. Journal of Abnormal Psychology, 101, 452-459. Crick, F., & Mitchison, G. (1983). The function of dream sleep. Nature, 304, 111-114. Dobson, K. S., & Shaw, B. F. (1986). Cognitive assessment with major depressive disorders. Cognitive Therapy and Research, 10, 13-29. Donahoe, J. W., & Palmer, D. C. (1989). The interpretation of complex human behavior: Some reactions to Parallel Distributed Processing,

W. W. Tryon

103

edited by J. L. McClelland, D. E. Rumelhart, and the PDP research group. Journal of the Experimental Analysis of Behavior, 51,399-416. Donahoe, J. W., & Palmer, D. C. (1994). Learning and complex behavior (pp. 66-67, Note 70). New York: Allyn and Bacon. Eaves, G., & Rush, A. J. (1984). Cognitive patterns in symptomatic and remitted unipolar major depression. Journal of Abnormal Psychology, 93, 31-40. Ellis, A. (1962). Reason and emotion in psychotherapy. New York: Lyle Stuart. Ellis, A. (1980). Rational-emotive therapy and cognitive behavior therapy: Similarities and differences. Cognitive Therapy and Research, 4, 325340. Eysenck, H. J. (1964). Experiments in behaviour therapy: Readings in modern methods of treatment of mental disorders derived from learning theory. New York: Macmillan. Foa, E. B., & Kozak, M. J. (1986). Emotional processing of fear: Exposure to corrective information. Psychological Bulletin, 99, 20-35. Foa, E. B., & Riggs, D. S. (1993). Post-traumatic stress disorder in rape victims. In J. Oldham, M. B. Riba, & A. Tasman (Eds.), American Psychiatric Press review of psychiatry (Vol. 12, pp. 273-303). Washington, DC: American Psychiatric Press. Foa, E. B., Steketee, G., & Rothbaum, B. O. (1989). Behavioral/cognitive conceptualization of post-traumatic stress disorder. Behavior Therapy, 20, 155-176. Foa, E. B., Zinbarg, R., & Rothbaum, B. O. (1992). UncontroUability and unpredictability in posttraumatic stress disorder: An animal model. Psychological Bulletin, 112, 218-238. Gilgen, A., & Abraham, F. (1995). Chaos theory in psychology. Westport, CT: Greenwood Publishing. Greenwald, A. G. (1988). Self-knowledge and self-deception. In J. S. Lockard & D. L. Paulhus (Eds.), Self-decepaon: An adaptive mechanism? (pp. 113-131). Englewood Cliffs, NJ: Prentice Hall. Greenwald, A. G. (1992). New look 3' Unconscious cognition reclaimed. American Psychologist, 47, 766-779. Grossberg, S., & Levine, D. S. (1987). Neural dynamics of attentionaUy modulated Pavlovian conditioning: blocking, interstimulus interval, and secondary reinforcement. Applied Optics, 26, 5015-5030.

104

Chapter 3

Grossberg, S., & Schmajuk, N. A. (1987). Neural dynamics of attentionaUy modulated Pavlovian conditioning: Conditioned reinforcement, inhibition, and opponent processing. Psychobiology, 15, 195-240. Hestenes, D. (1992). A neural network theory of manic-depressive illness. In D. S. Levine & S. J. Leven (Eds.), Motivation, emotion, and goal direction in neural networks (pp. 209-257). Hillsdale, NJ: Lawrence Erlbaum. Hilgard, E. R. (1977). Divided consciousness: Multiple controls in human thought and action. New York: Wiley. Hoffman, R. E. (1987). Computer simulations of neural information processing and the schizophrenia-mania dichotomy. Archives of General Psychiatry, 44, 178-188. Hoffman, R. E. (1992). Attractor neural networks and psychotic disorders. Psychiatric Annals, 22, 119-124. Hoffman, R. E., & Dobscha, S. K. (1989). Cortical pruning and the development of schizophrenia: A computer model. Schizophrenia Bulletin, 15, 477-490. Hollon, S. D., & Beck, A. T. (1994). Cognitive and cognitive-behavioral therapies. In A. E. Bergin & S. L. Garfield (Eds.), Handbook of psychotherapy and behavior change. (4th ed., pp. 428-466). New York: Wiley. Hollon, S. D., Kendall, P. C., & Lumry, A. (1986). Specificity of depressotypie cognitions in clinical depression. Journal of Abnormal Psychology, 95, 29-52. Hopfield, J. J. (1982). Neural networks and physical systems with emergent collective computational abilities. Proceedings of the National Academy of Sciences, 79, 2554-2558. Hopfield, J. J., Feinstein, D. I., & Palmer, R. G. (1983). 'Unlearning' has a stabilizing effect in collective memories. Nature, 304, 158-159. Hopfield, J. J., & Tank, D. W. (1987). Computing with neural circuits: A model. Science, 233, 625-632. Isen, A. M. (1984). Affect, cognition, and social behavior. In R. S. Wyer & T. R. Srull (Eds.), Handbook of social cognition (Vol. 3, pp. 179-236). Hillsdale, NJ: Lawrence Erlbaum. Jacoby, L. L., Lindsay, D. S., & Toth, J. P. (1992). Unconscious influences revealed: Attention, awareness, and control. American Psychologist, 47, 802-809. Janet, P. (1893). L'amnesie continue [Continuous amnesia]. Revue General Des Sciences, 4, 167-179.

W. W. Tryon

105

Janet, P. (1904). L'amnesie et la dissociation des souvenirs par l'emotion [Amnesia and the dissociation of memories by emotion]. Journal de Psychologie Normale et Pathologique, 1, 417-453. Janet, P. (1920). The major symptoms of hysteria. New York: Macmillan. Jones, J. C., & Barlow, D. H. (1990). The etiology of post-traumatic stress disorder. Clinical Psychology Review, 10, 299-328. Kandel, E. R. (1991). Cellular mechanisms of learning and the biological basis of individuality. In E. R. Kandel, J. H. Schwartz, & T. M. Jessell (Eds.), Principles of neural science (pp. 1009-1031). Norwalk, CT: Appleton & Lange. Kazdin, A. E. (1978). History of behavior modification: Experimental foundations of contemporary research. Baltimore, MD: University Park Press. Kazdin, A. E. (1979). Fictions, factions, and functions of behavior therapy. Behavior Therapy, 10, 629-654. Keane, T. M., Fairbank, J. A., Caddell, J. M., Zimering, R. T., & Bender, M. E. (1985). A behavioral approach to assessing and treating posttraumatic stress disorder in Vietnam veterans. In C. R. Figley (Ed.), Trauma and its wake (pp. 257-294). New York: Brunner/Mazel. Kihlstrom, J. F. (1987). The cognitive unconscious. Science, 237, 1445-237. Kihlstrom, J. F., Barnhardt, T. M., & Tataryn, D. J. (1992). The psychological unconscious: Found, lost, and regained. American Psychologist, 47, 788-791. Kosko, B. (1987a). Constructing an associative memory. Byte, 12, September, 137-144. Kosko, B. (1987b). Adaptive bidirectional associative memories. Applied Optics, 26, 4947-4960. Kosko, B. (1988). Bi-directional associative memories. IEEE Transactions on Systems, Man and Cybernetics, 18, 49-60. Lang, P. J. (1979). A bio-information theory of emotional imagery. Psychophysiology, 16, 495-512. Lang, P. J. (1985). The cognitive psychophysiology of emotion: Fear and anxiety. In A. H. Turner & J. Maser (Eds.), Anxiety and the anxiety disorders (pp. 131-170). Hillsdale, NJ: Lawrence Erlbaum. LeDoux, J. E. (1994). Emotion, memory and the brain. Scientific American, 270, 50-57. Leven, S. J. (1992). Learned helplessness, memory, and the dynamics of hope. In D. S. Levine & S. J. Leven (Eds.), Motivation, emotion, and

106

Chapter 3

goal direction in neural networks (pp. 259-299). Hillsdale, NJ: Lawrence Erlbaum. Leventhal, H. (1984). A perceptual-motor theory of emotion. In K. R. Scherer & P. Ekman (Eds.), Approaches to emotion (pp. 271-291). New York: Academic Press. Levine, D. S. (1991). Introduction to neural and cognitive modeling (pp. 41-92). Hillsdale, NJ: Lawrence Erlbaum. Levine, D. S., & Leven, S. J. (Eds.) (1992). Motivation, emotion, and goal direction in neural networks (Part II, pp. 169-365). Hillsdale, NJ: Lawrence Erlbaum. Levine, D. S., Leven, S. J., & Prueitt, P. S. (1992). Integration, disintegration, and the frontal lobes. In D. S. Levine & S. J. Leven (Eds.), Motivation, emotion, and goal direction in neural networks (pp. 301-335). Hillsdale, NJ: Lawrence Erlbaum. Lewinsohn, P. M., Steinmetz, J. L., Larson, D. W., & Franklin, J. (1981). Depression-related cognitions: Antecedent or consequences? Journal of Abnormal Psychology, 90, 213-219. Li, D., & Spiegel, D. (1992). A neural network model of dissociative disorders. Psychiatric Annals, 22, 144-147. Litz, B. T. (1992). Emotional numbing in combat-related post-traumatic stress disorder: A critical review and reformulation. Clinical Psychology Review, 12, 417-432. LoRus, E. (1980). Memory: Surprising new insights into how we remember and why we forget. Reading MA: Addison-Wesley. Martin, J. H. (1991). The collective electrical behavior of cortical neurons: The electroencephalogram and the mechanisms of epilepsy. In E. R. Kandel, J. H. Schwartz, & T. M. Jessell (Eds.), Principles of neural science (3rd ed., pp. 777-791). New York: Elsevier. Matt, G. E., Vazquez, C., & Campbell, W. K. (1992). Mood-congruent recall of affectively toned stimuli: A meta-analytic review. Clinical Psychology Review, 12, 227-255. McDougall, W. (1921). An introduction to social psychology. Boston: Lute. NunnaUy, J. C. (1967). Psychometric theory. New York: McGraw-Hill. Plutchik, R. (1958). Outlines of a new theory of emotion. Transactions of the New York Academy of Sciences, 20, 394-403. Plutchik, R. (1980). Emotion: A psychoevolutionary synthesis. New York: Harper & Row. Plutchik, R. (1994). The psychology and biology of emotion. New York: HarperCollins.

W. W. Tryon

107

Plutchik, R. (in press). The circumplex as a general model of the structure of emotions and personality. In R. Plutchik & H. R. Conte (Eds.), Circumplex models of personality and emotions. Washington, DC: American Psychological Association. Pribram, K. (1992). Familiarity and novelty: The contributions of the limbic forebrain to valuation and the processing of relevance. In D. S. Levine & S. J. Leven (Eds.), Motivation, emotion, and goal direction in neural networks (pp. 337-365). Hillsdale, NJ: Lawrence Erlbaum. Rychlak, J. F. (1981). A philosophy of science for personality theory (2nd ed.). Malabar, FL: Robert E. Krieger Publishing Co. Schacter, D. L. (1987). Implicit memory: History and current status. Journal

of Experimental Psychology: Learning, Memory, and Cognition, 13, 501-518. Schafer, R. (1978). Language and insight. New Haven: Yale University Press. Schlosberg, H. (1941). A scale for the judgment of facial expressions. Journal of Experimental Psychology, 29, 497-510. Schlosberg, H. (1954). Three dimensions of emotion. Psychological Review, 61, 81-88. Segal, Z. V., & Blatt, S. J. (Eds.) (1993). The self in emotional distress: Cognitive and psychodynamic perspectives. New York: Guilford Press. Seger, C. A. (1994). Implicit learning. Psychological Bulletin, 115, 163-196. Seligman, M. E. P. (1970). On the generality of the laws of learning. Psychological Review, 77, 406-418. Seligman, M. E. P., Abramson, L. Y., Semmel, A., &von Baeyer, C. (1979). Depressive attributional style. Journal of Abnormal Psychology, 88, 242-247. Seligman, M. E. P., & Hager, J. L. (Eds.) (1972). Biological boundaries of learning. Englewood Cliffs, NJ: Prentice-Hall. Spiegel, D. (1990). Hypnosis, dissociation, and trauma: Hidden and overt observers. In J. L. Singer (Ed.), Repression and dissociation: Implications for personality theory, psychopathology, and health (pp. 121-142). Chicago, IL: The University of Chicago Press. Spiegler, M. D., & Guevremont, D. C. (1993). Contemporary behavior therapy (2nd ed.). Pacific Grove, CA: Brooks/Cole Publishing Co. Staats, A. W. (1986). Behaviorism with a personality: The paradigmatic behavioral assessment approach. In R. O. Nelson & S. C. Hayes (Eds.), Conceptual foundations of behavioral assessment (pp. 242-296). New York: The Guilford Press.

108

Chapter 3

Teasdale, J. D., & Fogarty, S. J. (1979). Differential effects of induced mood on retrieval of pleasant and unpleasant events from episodic memory. Journal of Abnormal Psychology, 88, 248-257. Thompson, R. F. (1986). The neurobiology of learning and memory. Science, 233, 941-947. Thompson, R. F. (1990). Neural mechanisms of classical conditioning in mammals. Philosophical Transactions of the Royal Society, 306, London, 161-170. Tryon, W. W. (1993a). Neural Networks: I. Theoretical Unification Through Connectionism. Clinical Psychology Review, 13, 341-352. Tryon, W. W. (1993b). Neural Networks: II. Unified Learning Theory and Behavioral Psychotherapy. Clinical Psychology Review, 13, 353-371. Tryon, W. W. (1995a). Neural networks for behavior therapists: What they are and why they are important. Behavior Therapy, 26, 295-318. Tryon, W. W. (1995b). Synthesizing animal and human research via neural network learning theory. Journal of Behavior Therapy and Experimental Psychiatry, 26, 303-312. Tryon, W. W. (1995e). Resolving the cognitive behavioral controversy, the Behavior Therapist, 18, 83-86. Tryon, W. W. (1996a). Encoding emotion in eonneetionistic neural networks. Manuscript submitted for publication. Tryon, W. W. (1996b). A neural network model of posttraumatic stress disorder. Manuscript submitted for publication. Tryon, W. W. (1996c). Yes - Neural network learning theory can resolve the behavioral-cognitive controversy. The Behavior Therapist, 19, 72-73. Wachtel, P. L. (1977). Psychoanalysis and behavior therapy: Toward an integration. New York: Basic Books. Wasserman, P. D. (1989). Neural compuang: Theory andpractice (pp. 113125). New York: Van Nostrand Reinhold. Williams, J. M. G., Watts, F.. N., MacLeod, C., & Mathews, A. (1988). Cognitive psychology and emoaonal disorders. New York: Wiley. Wolpe, J., & lazarus, A. A. (1966). Behavior therapy techniques: A guide to the treatment of neuroses. New York: Pergamon. Author Notes

I wish to thank Scott Badger for his careful reading and helpful comments during an earlier stage of manuscript preparation and for testing the readability of the Appendix material.

IV. 14(. Tryon

109

Appendix: Description of the Bidirectional Associative Memory Network architecture Figure 2 illustrates a simple two layer neural network which forms the basis of the BAM mechanism we will explore. The S column of stimulus nodes is completely interconnected with the R column of response nodes. The connecting synaptic weights are drawn generically as lines but can also be represented as a memory matrix (M).

S

M

()

0 0 0 (Z)

C) (3 S--

;R

MT

R

Figure 2. Two layer neural network underlying the bidirectional associative memory (BAM) mechanism.

Stimulus and response definitions Stimulus and response vectors (number strings vs. matrices) can be defined as a binary (0,1) pattern. Elements in the stimulus vectors can refer to the presence vs. absence of specific characteristics such as red hair, bald, and

110

Chapter 3

tall. Elements in the response vectors can reflect the presence vs. absence of particular acts such as saying thank-you. Different levels of specificity can be chosen. At a micro level, each stimulus node can refer to a afferent neuron and each response node can refer to a efferent neuron. More typically, each stimulus node corresponds to a perceptual attribute and each response node to an entire behavioral response. Stimuli can include the physical consequences of one's own behavior and that of others as well as inanimate stimuli. Responses, in this article, primarily refer to the results of memory retrieval, including pertinent associations, but can also represent actions taken in response to stimuli which set the occasion for behaving. Stimulus and response vectors can also be given bipolar codes o f - 1 and 1 instead of 0 and 1 thereby directly representing polar opposite characteristics such as large vs. small, heavy vs. light, sharp vs. smooth or responses such as approach vs. avoid, dominant vs. submissive. Long stimulus and response vectors can be folded to form matrices. For example, the 100 elements of a stimulus vector can be folded to represent elements in a 10 by 10 matrix on which a visual pattern can be imposed by coloring the pixels (picture elements) according to 0 = white and 1 = black. For example, the letter A could be encoded as illustrated in Figure 3. Vector elements 0 - 9 form the first row, elements 10 - 19 form the second row, through elements 90 - 99 which form the final row. Another folded matrix based on the response vector might represent the subject's perception of the letter A. Response vectors, and folded matrices, can represent memories of stimuli which in turn can function as stimuli for other associations. We limit our discussion to a single S-R configuration but many sequences can be cascaded. Response vectors can also represent a sequence of actions to be taken or a code for complex behavior.

Memory (S-R) encoding The steps necessary to encode three S-R associations, memories, into our neural network are described in Table 1. The architecture of this network is fully interconnected meaning that all stimulus nodes are connected to every response node. The first step is to define stimulus-response pairs; items to be associated in memory. The three pairs are specified in binary form as S l-R1, $2-R2, and $3-R3. Notice that each stimulus is characterized by 8 digits and each response by 5 digits. An equal number could have been chosen, or more

~ W. Tryon

111

Ill //

Start

//

// // m

m

m

~

i

!

l..- ~.-.

End

i m

m

Figure 3. Example of encoding the letter A into a 100 element stimulus vector that has been folded into a 10 by 10 matrix.

digits allocated to responses. The 8 stimulus digits could represent the presence vs. absence of 8 physical attributes of the stimulus or they could represent a binary code referencing one of 28 = 256 situations. Similarly, the 5 response digits could represent the presence vs. absence of 5 physical characteristics of the subject's response or it could represent a binary code referencing one of 25 = 32 different responses. The Hopfield variant of the BAM (Hopfield, 1982; Hopfield & Tank, 1987) sets the R vector equal to the S vector. The second step is to redefine these pairs in bipolar form to avoid subsequently introducing asymmetrical effects when the threshold function is implemented. This is done by converting every 0 to -1 and leaving the + l's unchanged. The distance between -1 and 0 is now equal to the distance between 0 and + 1. Next we construct a memory matrix using steps three and four. The third step creates a I • J distributed memory matrix (M) in accordance with equation (1) where T refers to transposition meaning that numbers previously written as a horizontal row are now written as a vertical column or vice

Chapter 3

112

Table 1. Example Calculations for M e m o r y Encoding and Decoding in a Bidirectional Associative M e m o r y (BAM).

Step I" Define I Binary Values for S, J Binary Values for R SI=11000011 $2=01000010 $3= 1 0 0 0 0 0 0 1

RI=10101 R2=II011 R3=10110

Step 2: Transform Binary Values for S and R into Bipolar Values S I = 1 1-1-1-1-1 1 1 $2=-1 1-1-1-1-1 1-1 $3= 1 - 1 - 1 - 1 - 1 - 1 - 1 1

RI=I-1

1-1

1

R 2 = I 1-1 1 1 R3 = 1-1 1 1-1

Step 3: Create a Memory Matrix for Each Stimulus-Response Pair First Pair

Second Pair

SI T

1 -1 1 1 -1 -1 =1 -1 1 1

R1 1 -1

-1 -1 1 1 1 1 -1 -1

1 1 -1 -1 -1 -1 1 1

-1 -1 -1 1 1 1 1 -1 -1

1 1 1 -1 -1 -1 -1 1 1

=1 -1 1 1 1 1 -1 -1

R2 1

$2 T

1

1 -1

1

1

1 1 =1 -1 -1 -1 1 1

-1 1 -1 -1 -1 -1 1 -1

-1 1 =1 -1 =1 -1 1 -1

=1 1 1 -1 -1 1 -1 1 -1 1 -1 1 1 -1 -1 1

-1 1 =1 -1 =1 -1 1 -1

-1 1 =1 -1 -1 -1 1 -1

Third Pair R3 S3 T

1 1 1 -1 =1 -1 -1 1 1

1 1 1 -1 -1 -1 -1 1 1

-1 -1 -1 1 1 1 1 -1 -1

W. W. Tryon

113

Table 1 continued.

Step 4: Create a Composite Memory Matrix Through Adding Corresponding Elements Across the Above Three Matrices

1 1 -3 -3 -3 -3 1 1

-3 1 1 1 1 1 1 -3

3 -1 -1 -1 -1 -1 -1 3

-1 -1 -1 -1 -1 -1 -1 -1

-1 3 -1 -1 -1 -1 3 -1

Step 5: Select a New Stimulus for Presentation We choose the following variant of S 1" S = (1 1 0 1 1 0 1 1)

Step 6: Vector Multiply M by S The stimulus vector is written horizontally as follows: S=(11011011) The memory matrix is written as follows:

1 1 -3 -3 -3 -3 1 1

-3 1 1 1 1 1 1 -3

3 -1 -1 -1 -1 -1 -1 3

-1 -1 -1 -1 -1 -1 -1 -1

-1 3 -1 -1 -1 -1 3 -1

Matrix multiplication requires combining the one stimulus row with each of the columns in the M matrix resulting in a single entry for each column of the M matrix as follows.

Chapter 3

114 Table 1 continued.

The first term of each numerical pair refers to the S vector and the second term refers to a column in the M matrix: [(IX1) + (IX1) + (ox-3) + (IX-3) + (I X-3) + (ox-3) + (IX1) + (I)(I)1 = -2, which is the first entry in the first memory response vector. The second entry in the R vector is obtained by multiplying the S1 vector by the second column of the M matrix. The third, fourth, and fifth entries in the R vector are obtained by combining the S vector with the third, fourth, and fifth columns of the M matrix. R=(-2-22-62) Multiplying a matrix by a binary vector, such as S, equates to adding the column entries in the matrix by the rows associated with l's. In our case, this means that we add the first, second, fourth, fifth, seventh, and eighth row entries in each M column.

Step 7: Apply the Threshold Function (Eq. 3) The first entry of R in Step 6 is -2 which is less than 0, therefore the threshold function replaces the -2 entry with a 0 as indicated below. The same is true for the second entry of2. The third entry of +2 exceeds 0 and is therefore replaced by + 1. The fourth entry of-6 is less than 0 and is replaced by 0 whereas the fifth entry of 2 is replaced by +l because it exceeds 0. An value of 0 remains 0. R-(00101) This response differs by I bit from the target memory of I 0 l 0 1 and therefore is said to have a Hamming distance of 1 from the target memory.

Step 8: Use R to Associate to S The first response vector is written as follows: R=(00101) Since the transpose of a matrix involves interchanging rows and columns, we can operate on the transpose of the M matrix by treating its rows as columns.

W. W. Tryon

115

Table 1 continued. The R vector is combined with the ROWS of the M matrix as follows: [(0)(1) + (0)(-3) + (1X3) + (0)(-1) + (1 X-I)] - +2, which is the first entry in the modified S vector. Multiplying the transpose of the memory matrix by a binary vector reduces to adding the row entries corresponding to a 1 in the R vector. Given the R vector (0 0 1 0 1), we add the third and fifth row entries in the M matrix. The second through eighth entries of the S vector are obtained by combining the R vector with the second through eighth rows of the M matrix resulting in the following modified S vector. S =(2 2-2-2-2-2 2 2)

Step 9: Redefine S by Applying the Threshold Function (Eq. 3) S = ( I 1 0 0 0 0 1 1)

Step I 0: Multiply M by S and Apply the Threshold Function (Eq. 3) R = ( I 0 l 0 1) R is the correctly recalled memory. If this response is used to associate to a new stimulus S = (1 1 0 0 0 0 1 1) will be obtained which will again produce R = (1 0 1 0 1). The recall process is said to have stabilized.

versa. Notice that Table 1, Step 3, First Pair contains a matrix where R1 values are entered horizontally just as written in Step 2. The S 1 values have been transposed to a vertical column. The matrix entries equal the product of the marginal (row • column) values. Another such matrix is constructed for each of the remaining S-R pairs as illustrated under Step 3 in Table 1. Stated technically, we have formed the outer product of the transpose of the S matrix times the R matrix for each associated pair. Eq. (1)

M =

ST R

The fourth step creates a composite memory matrix by adding the corresponding elements o f all memory matrices. Hence, the entry in Row 1,

116

Chapter 3

Column 1 of the first pair matrix (+ 1) is added to the entry in Row 1, Column 1 of the second pair matrix (-1) which is added to the entry in Row 1, Column 1 of the third pair matrix (+ 1) to obtain the Row 1, Column 1 entry of the M matrix of +1. The Row 1, Column 2 entry of-3 in the M matrix is derived from the corresponding entries of-1 in the first, second, and third pair matrices. This fmal matrix is sometimes described as a "correlation matrix" because it associates all stimuli and responses. Learning entails encoding/representation. Neural networks can learn anything they can encode/represent. The memory matrix is composed of the weights which link each stimulus node with every response node. We refer to this as the M rather than W matrix to emphasize the long term memory function of these weights. Since W is the flip of M, others may wish to think of it as the weight matrix. Two observations are addressed to psychologists interested in dialectical psychological processes (el'. Rychlak, 1981). First, encoding memory for the association of S to R simultaneously encodes memory for the association between S c to R c where c represents "complement of" meaning that l's and O's have been exchanged (e.g., the complement of 1 0 0 1 1 1 is 0 1 1 0 0 0). Second, memory removal can be accomplished by adding the complement of the memory to be deleted; for example, adding S R c or S c R. Alternately, memories for specific S-R pairs can be deleted from memory by subtracting the corresponding "correlation matrix" or the equivalent operation of adding -X T Y to M. Forgetting is hereby modeled as an active process of altering memory content either incidental to new learning or by therapeutic design. We have associated one stimulus with one response. Many stimuli can be associated with a given response and many responses can be associated with a single stimulus. Given I stimuli and J responses an I • J memory matrix results.

Memory recall A stimulus is selected in Step 5 for presentation to the BAM to evaluate its memory ability. It can be one of the original three stimuli encoded into the memory or it can be a new stimulus which can be seen as a corrupted version of one of the encoded stimuli; an environmental stimulus which is similar to, but not exactly like, S 1. Choosing one of the stimuli selected for memory encoding will readily yield the appropriate response associated with it. We have chosen a slight variation of the first stimulus to make recall more demanding and to illustrate the important property that the BAM is a flexible

W.. l,g.. Tryon

117

memory device that creatively recalls the best fitting memory to a novel stimulus. Because we have chosen a novel stimulus, the BAM will not converge to a stable response in a single cycle. It will be necessary to generate a first memory response, feed it back through the BAM to modify the stimulus representation, which will then be used in a second successful attempt at memory retrieval. Step 6 involves applying a stimulus to the memory matrix and calculating the first memory response by multiplying the memory matrix by the selected input vector as per Equation (2). Table 1 describes how this matrix multiplication is accomplished. Eq. (2)

R=SM

Neurons only fire when their input exceeds a threshold value. Accordingly, in Step 7 we apply a threshold function (Eq. 3) to the results obtained from Eq. 2. Eq. (3)

If R > 0 then R = 1 If R <= 0 then R = 0

The first memory response of R = (0 0 1 0 1) differs by 1 bit from the target memory of R I to S 1 of 1 0 1 0 1 and therefore is said to have a Hamming distance 4 of 1 from the target memory. Had we used exactly S 1 we would now have exactly R1. Since, we do not, memory retrieval continues until convergence on the smallest Hamming value (including zero) has been achieved resulting in an energy minimum. Hence, we continue with Step 8 where the response is used to modify the stimulus trace and associate to a new stimulus by multiplying the response vector by the transpose of the memory matrix as per Equation (4). The transpose of matrix M simply uses the rows as columns as illustrated in Step 8 of Table 1. The resulting modified stimulus trace (representation) is S = (2 2 -2 -2 -2 -2 2 2). Eq. (4)

S = R MT

4 Hamming distance refers to the number of unequal entries in two vectors or matrices of equal size. VectorA = [1 1 0 0 1 1] and vector B = [1 0 1 0 1 1] differ in their second and third positions and are therefore a Hamming distance of 2 apart.

118

Chapter 3

T a b l e 2. Illustration o f M e m o r y E n e r g y C a l c u l a t i o n for S l - R 1 .

The stimulus vector S 1 from Step 2, Table I is S 1 = (1 1 -1 -1 -1 -1 1 1) Negating S results in S 1 = (-1 -1 1 1 1 1 -1 -1)

The memory matrix is written as before.

Step 1 Combine SI with each of the columns of M. The calculations for the first column of M is as follows: [(-1Xl) + (-lXl) + (1X-3) + (lX-3) + (1X-3) + (1X-3) + (-1X1) + (-1X1)] = 16 The calculations for all 5 columns are: [-16 8 -8 0 -8]

Step 2 The transpose of the response vector RI is written as follows: RI = 1 -1 1 -1 1 It is necessary to rewrite R1 as a column because matrix multiplication requires that when multiplying vector X times vector Y (e.g., XY) that one move left to right in the X vector and from top to bottom in the Y vector. Since -SM produced a horizontal vector, R must be transposed to create a column vector. The results of the -SM multiplication are combined with the transpose of R as follows for the first memory: [(-16XI) + (8X-l) + (-8X1) + (ox-1) + (-8)(1)] = -40. This value would be plotted on a graph like Figure 1 at the decimal equivalent of the stimulus and response vector.

W. W. Tryon

119

In Step 9 (Table 1) we reapply the threshold function to the modified stimulus trace obtained from Step 8 and find S = (1 1 0 0 0 0 1 1) which is exactly equal to S 1. We now apply the modified stimulus vector as we did the original stimulus. Step 10 multiplies the memory matrix by the new stimulus, reapplies the threshold function and computes a new response which is now equal to R1 resulting in a correct memory recall despite a slightly corrupted stimulus cue. Had the Hamming distance associated with the first association been larger than 1, perhaps 5, then additional memory cycles may have ensued until either a completely correct memory, Hamming = 0, was obtained or until no further reductions in Hamming distance were achieved. Convergence is especially rapid in small networks such as the one demonstrated here.

BAM energy function Neural networks are described by an energy function. If one imagines the energy field prior to memory creation as a fiat rubber sheet, then memory encoding is like placing a ball bearing on that sheet. Memory encoding creates a memory well whose depth is given by Equation 5. S represents a stimulus vector as before. M is the memory matrix as before. R T is the transpose of the response vector which means that its entries are arranged vertically down the page rather than horizontally across the page. Because memory recall involves finding energy minima, memory energy is taken as the negative of SMR T. However, positive energy memory hills form when SMR T is negative. Table 2 illustrates the energy calculation for the first memory. Eq. (5)

E =-SMR T

Figure 1 depicts what five memories might look like; one near each comer plus a middle entry. The coordinates of a memory well are found by translating the binary stimulus and response patterns into decimal X (stimulus) and Y (response) coordinates and plotting the energy value associated with that memory along a vertical Z axis labeled energy. For example, S = 0 1 0 0 0 0 1 1 is a binary number whose decimal equivalent is, working right to left:

120

Chapter 3 1(20) + 1(21) + 0(22) + 0(23) + 0(24) + 0(25) + 1(26) + 0(27) = 1 +2 +0+0+0+0+64+0 =67

Background energy values are the energy values associated with stimulus and response codes not previously associated. They are represented by all remaining permutations of the binary digits within the stimulus and response vectors. Background energy values arc obtained by substituting all possible remaining binary stimulus and response patterns into Equation 5 and plotting the result at the decimal equivalent coordinate of the S-R pair. For example, S = 1 1 1 1 1 1 1 1 and R = 0 0 0 0 0, two binary patterns not part of the three pairs we used in Step 1 of Table 1, constitute a background pair because neither this S vector nor this R vector were purposely stored in memory matrix M; yet when applied to Equation 5 will yield an energy value that can be plotted at the decimal equivalent coordinates of S and R. Each memory well is a local energy minimum called an attractor 5. Its field of influence is specified by the shape of its walls; steeper equals stronger. The extent of a memory wcU's attraction is termed its basin of attraction and is graphically rcprcscntexi by the mouth of its funnel-like structure (see Figure 1). If this were a physical structure, then a ball bearing placed at all points defining the basin of attraction would roll toward it. Like water running down hill, memory and associative processes located anywhere within the rim of a basin of attraction proceed (flow) downhill until a local energy minimum is encountered where upon the corresponding memory or association is retrieved. Cognition entails seeking energy minima. The shape, or topography, of the energy fidd completely determines all aspects of the associative process including its direction at all points in time and its final end state. Figure 4 was created from Figure 1 by increasing the basin of attraction of the central memory well. The memory associated with its local minimum is what will be recalled for many more starting points than was previously the case. Stimuli that may previously have flowed into one of the comer memories may now be drawn into the abnormally large central well.Why should energy minima be associated with memories, associations, and neural network solutions to a wide variety of problems such as perception and

5 The attractor concept is also found in chaos theoryto which neural networksare related. Interestedreaders may wish to consult Gilgen and Abraham (1995).

121

W. W. Tryon

IIDI

I10

>0 Z W

60

40

211

Ill O0 60

I011 410

Y

2a 21

X

Figure 4. Figure 1 modified such that the central memory has a much larger basin of attraction.

language skills? For the same reason that a regression line is the best fitting line; because it minimizes the sum of squared deviations between observed and predicted points. The correct or desired solution to many problems involves minimizing a discrepancy quantity.

Summary We have formed a simple three memory BAM connecting three pairs of 8-bit stimuli and 5-bit response patterns. The three separate associations have been combined into a single memory matrix whose elements can be interpreted as synaptic weights between a set of stimulus "neurons" and a set of response "neurons". It is crucial to note that all elements of the memory matrix participate in storing each of the three S-R associations (memories). This is the essence of a parallel distributed memory system and differs

122

Chapter 3

fundamentally from the serial computer analogy of storing information in discrete places. An energy value is associated with every memory resulting in a funnel shaped memory well with a basin of attraction. Memory retrieval entails descending to the bottom of the memory well through a process known as gradient descent. Our example took two cycles to reach an energy minimum and correctly recall R1 when given S 1. We now apply these fundamental concepts to psychopathology.

Cognitive Science Perspectives on Personality and Emotion - G. Matthews (Editor) 1997 Elsevier Science B.V. CHAPTER 4

Space-Time, Order, and Hierarchy in Fronto-Hippocampai System: A Neural Basis of Personality Jean P. Banquet, Philippe Gaussier, Jean Claude Dreher, Cddric Joulain, Arnaud Revel and Wilfried G~nther

Personality is to a large extent a characteristic of human primates, in the same way that language is. It results indeed from a unique combination of higher order brain functions including an explicit historical memory of the past, and at the same time, the ability to make plans, to change goals, to adapt strategies and therefore to project into the future. Both functions are based on the perception of a temporal dimension, or at least of an ordering of sequences of past or future events, in the field of consciousness. This field of consciousness itself, or its extension in working memory (WM), can be considered, from a neurophysiological point of view, as the embodiment of the present. In the past decades, the anatomical and structural bases of these functions have been more and more precisely delineated. In particular, hippocampal system (Hs) has been established as the support for the acquisition of explicit or declarative historical memory. Prefrontal cortex (Pc) could bring this temporal function one step further. Indeed, it is not so much the specific repository of learned past sequences of events, a function it could share with other cortical structures such as temporal and parietal cortices. More specifically, it emerges as the elective site for planification, intentionality, motivation and goal propagation, and therefore the structure where the hierarchical ordering of the sequences of cognitive or motor events takes place. More importantly, this planification is not limited to extemaUyelicited behaviors, but includes self-initiated endeavours founded on internal motivations or "willed" actions. These higher order brain functions can be considered as the highest expression of creativity and personality, in so far as they truly express field independence in behavior, the neurophysiological counterpart of the controversial philosophical construct of freedom. In "physical" time, present is an elusive reality caught between time-gone and time-to-be. Yet, in the time experienced by the mind, this present supports the field of consciousness, and forms the building blocks of our personality. The more elaborated the building blocks, the more sophisticated will be the whole

124

Chapter 4

construction. The brain has evolved a solution to this elusiveness of the present, that we could name working memory (WM). WM can be defined as an unique capacity to maintain information for an extended duration, and at the same time to process this information at any order of complexity during task performance. As such, it can be considered as an extension of the present. This WM is the brain function which makes the junction between past and future. Indeed, by its capacity to extend the duration of instantaneous events, it operates to link sequences of events together. Most investigations have emphasized the controlled aspect of WM. Not surprisingly, the volitional support of this function has been located in the prefrontal and temporal cortices. In this paper we provide ncuropsychological and physiological evidence for a twin aspect of this function, which has been overlooked, namely an automatic uncontrolled face of WM, presented as one of the slave systems of the frontal executive. The main support of this automatic WM could be the hippocampal system. The integration and novelty detection capacities of the hippocampus (Dcnham & Boitano, 1996) arc twin functions which help both to create a stable state and to delineate transitions between states. They allow the construction of a stable representation of the environment and also the linking together of sequences of events. Symmetrically, by its capacity to reactivate and reenact memories, Hs provides the cortex with the capacity to maintain together in the field of consciousness, clusters of logically related events, and therefore, gives to Pc in particular the possibility for an hierarchical ordering and linking of full sequences of motor or cognitive actions to come. The importance of prefrontal cortex as a support of personality is evident from the symptomatology of patients with a prefrontal syndrome. It is largely a consequence of perturbation of most of the higher integrative cortical functions such as judgement, reasoning and motivation, and degradation of some part of the LTM stores and the controlled aspect of working memory. These perturbations together put familial, professional and social life in jeopardy. Conversely, it is not obvious on the basis of superficial observation, that hippocampal lesions induce a deep disturbance of personality. In particular, there is no dramatic extended retrograde amnesia. Nevertheless, three decades of ncuropsychological observation have made patent that personality disturbance is hardly loss than after a frontal lesion, at least in cases presenting a total loss of Hs functions. Antcrogradc amnesia (the incapacity to acquire new declarative or explicit memories) was emphasized in the exploration of these patients. From these results was inferred the fundamental function of the hippocampus in the process of LTM

d.P. Banquet et al.

125

consolidation, but not in LTM storage. These last functions were instead devoted to the cortices. From there, a logical step was to explore the hippocampus-cortical relations. Hippocampus receives information from and sends it back to the totality of the cortical mantle. Nevertheless, a specially close relationship is established with the prefrontal cortex. The cooperation of the two systems in such diverse functions as working memory, rule formation, tcmporo-spatial processing is so intricate that it is often difficult to disentangle the specific contributions of the two systems. In this article we make the assumption that some of the most elaborated functions classically attributed to the prefrontal cortex, such as working memory, sequence ordering, timing, and novelty detection, are in fact based on a strong memory and the computational support of the hippocampus, and then further claboratod in the prefrontal cortex. From this perspective, we do not consider the relations between the two systems as symmetrical. We rather think that the prefrontal cortex function is more dependent upon the integrity of the hippocampal system rather than the reverse. Accordingly, we first explore the hippocampal contribution to WM. We exclude the function of LTM consolidation. We consider this WM function of the hippocampus, on the basis of ncuropsychological evidence, as the foundation of some of the higher order functions attributed to the prefrontal cortex. In both Pc and Hs WM capacity is supposed to be responsible for the processing of order and time, and some aspects of spatial processing such as the building of maps and plans. In spite of this apparent communality of function, in particular for time and space processing, we emphasize the functional specialization and complementarity of the two structures: 1) Transient intermediate memory registers in Hs compute auto- and hetero-correlation and are suited for fusion of data on an intermediate range (of the order of minutes or more), and therefore allow for match-mismatch comparison on a window of time of this order of magnitude. Further, fast learning takes place in the hippocampal synapses, at the limit in one trial. These capacities together make the Hs ideal for the fast and transient learning of temporal sequences (Banquet & Contreras-Vidal, 1992a, b; 1993a, b; 1994) or the multiple views of a global scene (Gaussier & Zrehen, 1994a, 1995; Gaussier, Joulain, Revel & Banquet, 1996; Gaussier et al., 1997a,b). 2) Complementarily, the activation and the propagation of goals at Pc level, under the influence of sub-cortical structures like the amygdala, in particular, endows Pc with the control power to design goal-oriented, hierarchically ordered sequences of actions (Zrehen & Gaussier, 1997). The

126

Chapter 4

links of the Pc structures with different subcortieal structures give Pc the possibility for both changing goals as a function of the context (link to Hs) and selecting the appropriate programs for these goal implementations (links to the basal ganglia). Learning there is slower than in I-Is, and therefore needs either repetition, or the mediation of Hs. Synunetrieally, just as Hs inputoutput is in direct relation with perception and cognition, the Pc input-output is directly connected to cognition and action. Due to this orientation to action, and also to its LTM capacity for storing sequences, Pc is the most plausible site for recombination of novel sequences of events on the basis of previous learning and present context, and, therefore, the seat of creativity.

Hippocampal Function: An Extended View Combined ncuropsyehological and biological evidence suggests a fundamental but selective role of the hippocampal system (Hs) in some forms of learning. Hs is necessary for rapidly (one exposure at the limit) forming declarative, explicit long-term memories (Banquet ct al., 1997), but it appears to be unnecessary for the progressive acquisition of procedural, implicit memories. Recall of previously acquired declarative memories becomes gradually independent of Hs itself, suggesting a graded process of consolidation of traces which would be stored in another structure, plausibly cerebral and/or eerebellar cortices. In the study of declarative explicit memory and WM, and particularly of the role of Hs in this type of memory, human neuropsychology leads animal neurophysiology. This unusual state of affairs raises the problem of the transposition:to the animal domain of concepts specifically coined for human cognitive functions. Integrating in one single model concepts from neuropsyehology and animal neurophysiology implies the implicit assumption of a continuity or even similarity (but certainly not identity) between memory processes taking place both in human and nonhuman primates or even lower order species. Anterograde/retrograde amnesia. With this caveat, most authors agree that a major milestone in memory research was the report by Scoville and Milner (1957) of a dramatic but selective impairment in memory consecutive to a bilateral ablation of hippocampus and related structures in medial temporal lobe for the sake of an otherwise untraetable epilepsy. This type of memory impairment in the patient HM contrasted with the preservation of skill learning and priming effects, and more broadly all the capacities labelled as procedural, implicit memory (Schacter, Chu & Ochsner, 1993). Most

J.P. Banquet et al.

127

authors draw two conclusions from those results. First, some (declarative) forms of memory are initially stored in Hs, and then gradually transferred from there to other more permanent sites of storage, as neocortex. Second, other (procedural) forms of memory are directly stored in cortical areas. These theories grossly capture the gist of neuropsychological results. But in some sense they are partially wrong. Progress mostly in neurophysiology, but also in neural imaging and modelling, allows for more precise statements about the locus and the nature of stored information, in spite of the fact that the precise modalities of consolidation of these types of memory still remain a mystery. Hippocampus: A link-operator store and a multirange buffer. In particular, despite the contention of many neurophysiologists (Horel, 1994; O'Keefr & Nadr 1978; Burgess, Rr162 & O'Kccfr 1994) and also some modellers, we will argue that Hs does not" play a central role in the primitive storage and recall of the content of specific episodes and events...". Only the connectivity or "link operators" between compressed hippocampal representation of conical activation patterns are transiently stored in Hs, and secondarily transferred and developed in the cortex under Hs control. Place cells in the rat and view cells in the monkey are a typical illustration of these processes of temporo-spatial correlation-integration that take place in Hs and lead to a highly symbolic representation of the surrounding world. Yet, the full-fleAged memory traces are initiated and finally stored at the cortical level. Cortex is the alpha and the omega of our souvenirs. There is no need and no capacity for transfer and storage of the full traces in Hs. Preliminary evidence also suggests that even if procedural memory does not rely on Hs for its consolidation, it nevertheless implies complex conico-subcortical circuits involving in particular basal ganglia whenever motor responses are implicated in the learned processes. Whatever the cortico-hippocampal mechanisms involved in long-term consolidation of declarative memories, experimental results suggest that the role of the Hs extends far beyond that of a transient LTM store, for the time required for cortical trace consolidation. It seems to be implicated even during the very first wave of cortical processing triggered by stimulus input. Even in the simple role of witholding and buffeting information there is neuropsychological evidence from amnesic patients that Hs is already necessary in the short-term range, as far as information excee.As the STM span. This should not be surprising since the activation lag of the Hs compared to primary cortical areas in response to an external input does not exceed one or a few hundred milliseconds. A strong conclusion, in

Chapter 4

128

consequence, is that both systems process in parallel and interactively. Similar type of evidence argues in favor of a crucial role of Hs in the intermexiiate range as a WM buffer. A unitary mechanism supporang WM and LTM consolidation. Its multirange temporal capacity makes the Hs act as a buffer that can reenact activation patterns of information at cortical levels, not only during the lengthy process of LTM consolidation, but also when functioning as an intermediate (minutes time-scale) register supporting WM, during the moment to moment operation of current tasks. This last hypothetical function is conceptualised in the model as an automatic working memory. These two proposed functions are: - first, complementary, the process of LTM consolidation being engaged only if WM processing has transformed a short-lived trace to a transient LTM trace; second, they are based on identical, or at least similar physiological processes, namely reinstatement or reenactment of electrical patterns of cortical activity either spontaneously or in reaction to a cortical cue. This unique process allows the network activation, according to its locus of initiation, either to trigger buffeting and rapid learning of information in hippocampal subsystems (if the focus of activation is first cortical) or to reactivate cortical patterns, and therefore reinstate recent memories (if the initial focus of activation is endogenous in the hippocampus). These processes of reactivation of cortical activity patterns, or reenactment of recent memories, directly derive from the capacity of the Hs to function according to two distinct, complementary modes: first in a read mode, when it registers and processes external, cortical information; then in a print mode, when it "endogeneously" or reactively reinstates the corresponding patterns of activation either locally in the hippocampal subsystems, or in both the Hs and the cortex. These two modes correspond to two clearly defined electrophysiological patterns in some animal species, theta and sharp waves (Buzsacki, 1989). In primates only sharp waves have been consistently individualised. It is claimed that this peculiarity corresponds in fact to an extension of the read mode which takes place not only as a consequence of the physical exploration of the environment (theta phase), but also as a follow up to any endogenously generated "cognitive processing". The reactivation function is related in the model to the internal "bursting" capacity of the CA3 pyramidal neurons, and their collective capacity to synchronize under the modulation of septal inputs. -

-

-

J.P. Banquet et al.

129

Hippocampus as a temporo-spatial processor In relation plausibly with this transient buffer function, behavioral and neurophysiological results point to a more cognitive role of the Hs in learning the temporal order of serial sequences, at least at a low level of processing. This function could be illustrated in the process of place recognition in lower species. Yet, spatial learning is only one example (possibly the best in some species) of this multidimensional correlational processing. Further, conditioning literature points to an important role of some parts of the hippocampus in learning temporal intervals between events, and more generally durations or timing. Some of these functions seem to be based on very basic processes of differential synaptic plasticity (Granger, Whitson, Larson & Lynch, 1994) as a result of the correlational learning capacity attributed to most of the hippocampal subsystems, applied to successive events. Some others, such as timing, could result from population coding by cells endowed with different dynamics that we locate in the dentate gyrus in our model. In this view of the hippocampal function as combining and contrasting fusion-integration and match-discrimination, place recognition appears in many respects as a degraded by-product of the Hs capacity to register sequences of events. The long-term declarative memory consolidation function attributed to LTP is the current focus of much investigation. But, in the present paper, we compare and contrast the functions, in particular WM, of Hs and Pc. By neuropsychological and physiological arguments, we make a case for the automatic working memory function supported by ITM (intermediate-term memory), operating in parallel to and in relation with the cortical WM, as a particular slave system of the frontal executive. We sketch a comparison and a unification of the experimental results in both human and nonhuman primates. We contrast and relate this WM function with the more classical LTM consolidation function attributed to Hs. We also provide physiological and behavioral evidence for the implication of the Hs in temporal order sequence processing and timing, which are the basis for our neural network model implementation.

Working Memory as Both a Cortical and a Hippocampal System The existence of a graded retrograde amnesia is a strong argument in favor of a transient LTM, probably based on hippocampal LTP and involved in LTM consolidation. We make here a case for a shorter time constant type of memory, WM being based on both a cortical system or systems, and a hippocampal ITM. Neuropsychological and brain imagery arguments in favor

130

Chapter 4

of STM and ITM functions of hippocampus are important for our model, which is illustrated in Figure 1. Since there is a general agreement on the role of the associative cortex, either prefrontal or temporal, in WM function, we will mainly emphasize the arguments in favor of an hippocampal contribution to the WM function.

Psychological arguments for a slave amodal automatic Working Memory Baddeley (1986) proposed a multicomponent model of WM as a substitute for the shortcomings of the modal model of Atkinson and Shiffrin (1968), which implied a unique STM store as a necessary passage to LTM. A single STM store could not simultaneously function as an adequate WM, and was therefore evolved into a multicomponent model. Still, WM was viewed by Baddeley as a single common resource, with a limited capacity. Worlang Memory as control and slave systems. The definition of WM as a "temporary storage of information in connection with performing other, more complex tasks" is vague enough to allow for any possible extensions or modifications of the model. Baddeley assumes a limited-capacity attentional controller, the central executive, that supervises two slave modality-specific systems, the visuo-spatial sketchpad to hold and manipulate visual and spatial images, and the articulatory loop, to rehearse speech-based information. The articulatory loop manipulates memory for sounds. It comprises a memory store for holding phonological information for a period of one or two seconds, coupled with an articulatory control process (Baddeley, 1986). Overt or covert subvocal articulation allows both refreshing the auditory memory traces and also feeding the phonological store with phonologically translated visual information. The temporary storage of visual information in a visuospatial sketchpad, would imply an occipital system involved in the visual aspects and a parietal system involved in spatial coding, and also possibly a frontal lobe participation (Goldman-Rakic, 1988). A related but more comprehensive construct proposed by Fuster (1995) is that of active memory. This is a state rather than a system of memory, and includes a widely distributed and changing representational network in the awake organism. Active memory therefore includes WM, but does not presuppose any mental or cognitive operation. Until recently, only vague reference was made by Baddeley to the underlying brain structures supporting either executive or slave systems, but implicitly WM function was under the control of conscious awareness and therefore had plausibly a cortical location. The same is also assumed for

Motor Areas Actions Prefrontal

-

"Where"

J.P. Banquet et al.

Parietal Areas

Primary Visual Areas

131

Figure 1 . A global view of cortico-hppocampal relations. In the model, the hippocampus is used as a plastic fusion operator and Workmg Memory (WM). Its connection with the limbic system allows regulation of vigilance and learning levels.

132

Chapter 4

animal WM (Goldman-Rakic, 1988). Yet, a fundamental ambiguity concerning the precise definition of WM, and the explanation of the results derived from the paradigms designed to probe WM theory, remains unresolved if one does not refer to the brain structures that support WM: "There is clearly a danger that a concept such as the central executive may reflect nothing more than a convenient homunculus..." (Baddeley, 1995). Furthermore, the sole reference to cortical structures is not enough to acx,ount for all the experimental data. More recently, results of brain imagery and animal neurophysiology were incorporated into the theory (Baddeley, 1995). Unfortunately, most of the experiments to date, either with metabolic or electrical brain imaging with event-related potentials (ERPs), use paradigms that are specifically designed for the study of the components of the model of Baddeley (1986), rehearsal systems in particular (Paulescu, Frith & Frackowiak, 1993). Furthermore, the ERP approach (Ruchkin, Johnson, Canoune & Ritter, 1991; Ruchkin et al., 1992) is more adapted to the exploration of the cortical mantle, than of the deep structures. Therefore, these studies confirm the involvement of cortical structures in different WM paradigms. Yet, more classical recall paradigms, not formally requiring rehearsal, show joint cortical and hippocampal activation (Squire et al., 1992). Working Memory as a hippocampal automatic slave system. Neuropsychology and imagery suggest an automatic component of WM, supported by an intermediate register located in Hs, along with the controlled, supposedly cortical, component of the WM system explored both in humans (Baddeley, 1986) and animals (Fuster & Alexander, 1971; Olton, Becker & Handelmann, 1979). Like the cortical controlled WM, this system is endowed with both storing and processing capacity. It is supposedly based upon several systems: First, a complementary set of intrahippocampal or hippocampo-cortical loops; - Second, a battery of memory registers covering a large temporal spectrum; Finally, rapidly instantiated but transient synaptic facilitation demonstrated in different hippocampal subsystems (Buzsaki, 1988; Jones, 1993). Moreover, the operation mode of this intermediate system supporting WM is supposed to be just a restriction to the intermediate range (minutes) of the more comprehensive process of LTM consolidation generally attributed to Hs. This general process consists of maintenance and/or reenactment of -

-

d.P. Banquet et al.

133

cortical patterns of activation by reverberant activity between reciprocally connected systems. This hypothesis extends to WM the dichotomy that already exists in LTM between declarative explicit memory and procedural implicit memory. Psychological arguments in favor of this hippocampal automatic component of WM are provided below. Refreshing either the visuo-spatial sketchpad or the phonological store by rehearsal every second or so is a controlled process. So too is performance of the concurrent cognitive task, usually verbal, required from subjects in WM paradigms. Supposedly, the central executive responsible for planning, strategy selection, and coordination of information is monitoring both of these tasks. Thus, several controlled processes work in parallel. Shiffrin and Schneider (1977) demonstrated the very limited capacity for controlled processing in the human brain. They made a good case that only several automatic processes, or at best one controlled and one automatic process, could be performed in parallel. Yet, surprisingly, the usual paradigms testing WM are tractable without overwhelming difficulty even by patients or aging subjects (Br6bion, 1994). There could be several reasons for that, not necessarily exclusive, and more or less implicated according to tasks and subjects: - The so-called complex cognitive task may be largely automatized in spite of its complexity, in particular when it involves verbal comprehension. Thus, attention would be relatively free to focus on active rehearsal of the tobe-memorized material. Alternatively, the material to be memorized may be more or less related to the task to be performed, so there is not really competition and interference between withholding information and processing it, but a nice interleaving and integration between the two tasks. This type of paradigm (where the information to be remembered is related to the cognitive process in progress) is certainly close to actual WM operation in natural conditions. Certainly, one cannot deny the reality of the rehearsal process. But we contend that this low level and rote strategy is expensive in terms of limited controlled processing capacity, and unnecessary in most everyday situations. Whenever possible, subjects resort to cognitive strategies aiming to create supraordinate chunks of items or events, in order to increase the limited capacity of STM. However, this is still a controlled process. We therefore suppose the existence of an automatic support to WM, an ITM, which does not need rehearsal. This register is in such a functional relationship with the various cortical areas, that it may readily refresh recent memories relevant for the task in progress, by simple maintenance and/or reenactment of the -

134

Chapter 4

corresponding activation patterns in the cortical populations. This reactivation does not preclude a state of priming or subliminal activation of the cortical areas to-be-reactivated due to their recent activity. This function is exactly what I-Is supposedly performs, in the different processes of longterm consolidation of the declarative episodic and factual semantic memory traces. Both WM and transient LTM refer in some sense to an episodic, context dependent, or a factual type of information to be memorized. Nevertheless, the only important distinction between the two is that the trace already "long-term potentiated" will certainly be transferred to LTM, while the trace (and eventually the activity pattern) in WM is just a candidate for LTP and undergoes a test of eligibility to permanent LTM store. It is therefore further hypothesized that the two processes of LTM consolidation on one hand, and of WM refreshing on the other, are roughly similar and complementary. Only the duration during which the two types of memory traces keep relevance for the subject differs. This is of the order of minutes for working memories, but weeks or months for memories that must be "permanently" consolidated in the long-term stores. They are also complementary because transformation into transient LTM can only be considered if the long-term relevance of the information to be stored has been confirmed by the processing performed either in the intermediate hippocampal store or in the cortical system of WM, or in both. The neurophysiological counterpart of this information selection for long-term storage would correspond to the transition from loop iterative and punctual activation or even more important short-term synaptic facilitation (based on short-term potentiation: STP) to a transient long-term synaptie facilitation (based on LTP). Both types of learning are documented at different levels of hippocampal subsystems (Jones, 1993; Buzsaki, 1988).

Personality correlates of Working Memory and transient LTM One of the main implications of our model is the augmentation of WM beyond the classical, cortical, rehearsal-expandable STM range, by a rehearsal-independent register mostly based on ITM in the hippocampus. This ITM itself is distinct from transient LTM based on LTP. It is noteworthy that personality traits seem to correlate differentially with these memory components. Both extraversion and neuroticism relate, in experimental studies, to performance on retention tasks. Howarth and Eysenck (1968) showed that, in retention of paired-associates, extraverts show superior recall over retention intervals up to 5 minutes or so (in the ITM range), but

J.P. Banquet et al.

135

thereat~er introverts show reminiscence and increasingly better performance than extraverts. These results are difficult for current WM theory to explain, because the advantage of extraverts persists beyond the typical durations associated with the short-term components of WM. The present analysis suggests extraverts may have superior STM/ITM. Plausibly, on the basis of different performance equilibria in neuromodulators (in particular noradrenergic, dopaminergic, or serotonergic), cortical STM is more stable, and the Hs tends to operate so as to refresh these memories in extraverts. Conversely, the LTM consolidation processes would be more powerful in introverts. It is already admitted that emotional charge associated with events (and supposed to be larger in introverts than extraverts) favours LTM consolidation. Could it be that this same emotional component is detrimental to the automatic (non-rehearsal based) maintenance of information in STM/ITM? Consistent with this hypothesis, extraverts tend to show superior recency in free recall, a function normally attributed to STM, but effects of extraversion on WM tasks requiring active processing are unreliable (Matthews, 1992). It is well-established that trait anxiety and neuroticism are associated with impairment of STM (Eysenck, 1982). In this case, effects are stronger for active WM tasks than for passive storage (e.g. Darke, 1988), implying that this anxiety effect may relate preferentially to the controlled, cortical component of WM. This psychopathological result is also consistent with our interpretation for the effects relating to the introvert-extravert dimension.

Neuropsychology, Brain Imaging and Working Memory Neuropsychological evidence in favor of an automatic component of WM based on hippocampal structures remain important. Indeed, brain imaging experiments performed during WM paradigms are still scarce, and for most of them oriented to testing the controlled aspect of WM according to Baddeley's theory of the executive controller. First, Brown (1958) and Peterson and Peterson (1959) showed that witholding information in STM is dependent on rehearsal, and rapidly lost if active rehearsal was prevented. This fact lies at the basis of the distinction between immediate STM and primary memory (James, 1890) which results from STM extension due to the rehearsal process. Second, responses of normal subjects to different recall or recognition tests, show in the absence of rehearsal, a residual memory which the STM decay tends to asymptotically. This residual memory is suppressed after bilateral Hs lesion. Therefore, we

Chapter 4

136

attribute it to an hippocampal ITM component. Classically, in amnesia consecutive to a mextial temporal lobe lesion, STM in the sense of immediate memory is fully intact, in contrast with the loss of new acquisitions in declarative LTM (Baddeley & Warrington, 1970; Cave & Squire, 1991). The picture sketched from the study of amnesic patients is much more subtle than this black and white portrayal, in particular when one takes into account the role of the extent of the lesions in determining the depth of the deficit. The following neuropsychological arguments derive from recall and forgetting curves, or visual recognition obtained by classical tests of STM and LTM both in normals and amnesic patients. In spite of the thorough analysis of these results, some repeatedly confirmed r has been either overlooked or even not accounted for. The patients show a deficit of learning and recall for verbal as well as nonverbal material. In terms of our model, these tests explore in fact both STM and intermexfiate term memory (ITM), according to the variable delay of recall. Since the depth and nature of the memory deficit depends on the extent and location of the Hs lesions, these parameters are taken into account in the interpretation of the results. Bilateral extensive lesions with complete loss of hippocampal function will therefore be treated separately from unilateral lesions and/or partial loss of hippocampal function. But, for the purpose of separating hippocampal from cortical components of WM, the most important parameter is whether or not the subjects are allowed controlled rehearsal, as supposed by the articulatory loop in the WM model of Baddeley (1986). Rehearsal allowed. The results from subjects with complete loss of Hs function (like patient HM) will be emphasized since a normal or close to normal performance on tests of STM or ITM in this case would imply all the more a normal performance when hippocampal function is partially preserved. In this case, the only possibility will be to attribute the corresponding performanc, to STM or to the controlled, cortical component of WM, since rehearsal is allowed and Hs function is lost. This complete loss of Hs function results either from bilateral and extensive resection of the medial temporal lobe, as in HM's case, or from unilateral resection associated either with a severe degeneration (post-mortem diagnosis) or with a severe dysfunction (EEG recording) of the spared medial temporal hemilobe (respectively cases PB and HF). First, rehearsal is spontaneous, as with verbal material well within the memory span and made of consonant trigrams presented in a variable delay matching-to-sample task (Sidman, Stoddard & Mohr, 1968)..Then STM range can be extended theoretically at will in what William James (1890) has -

J.P. Banquet et al.

137

named a primary memory (see Figure 2d), even for patients with total loss of hippocampal function. Second, the material is easily verbalizable but at the limit of the memory span as in the short version of the visual maze (Milner, Corkin & Teuber, 1968). This test requires the memorization of a sequence of turns (8 choice points). The patients cannot learn the task even after many trials. There is complete disruption of the memory process. Thus these patients face an actual cognitive defect, related to the incapacity to simultaneously maintain and organize accessible information (type of turns to make), and to implement the task (topographical translation of the turns on the maze). This is a genuine definition of a WM task. In this situation, rehearsal is not sufficient due to the memory load at the limit of the STM span. Furthermore, practice does not improve performance. There is no evidence of learning over 125 trials, as if the subject was unable to devise a leaming strategy in order to split a too difficult task into accessible subgoals. - Third, if the material to be memorized is not naturally or easily verbalizable (elliptic geometric forms with one variable radius to compare to a sample after various delays), the extension of the STM range is not possible in patients (Figure 2d). Even with very sensitive measures, a limited residual control of the sample stimulus on the performance is restricted to the 16-24 see range, i.e. the classical decremental STM range. Remarkably, this is not the case with normal subjects or even 9-12 year old children who demonstrate no evidence of performance deterioration at delays up to 40 msecs. Similar results of poor performance in the case of extensive bilateral lesions have been found in tests involving other types of geometrical items (Milner et al., 1968). Finally, these defects are further corroborated and extended to the immediate STM by the classical test of digit span. Digit sequences of various lengths are presented at a typical rate of one digit per second. Subjects speak out recalled digits at their own pace. The percentage of strings correctly recalled (Figure 2c) demonstrates an inverse linear relation between the recall performance and the length of the string, similar in amnesic and controls. But, consistently (Drachman & Arbib, 1966; Baddeley & Warrington, 1970) amnesic subjects perform worse (in fact, are completely unable to recall) on sequences exceeding the normal memory span (7 items). These last results point to another critical factor in recall performance, besides the delay of retention, namely the memory load. As for the delay duration, medial temporal lobe seems to play an important role with regard to memory load. Furthermore, this very last result shows clearly that medial temporal lobe -

138

loo;

-

--O-

AilNESICS

CONTROLS

Chapter 4

5040.

ITEM NUMBERS

DELAY (secs)

Figure 2. A.Two component free recall task. Mean percentage correct recall (immediate recall) as a function of order of presentation for subjects presenting partial lesions of the hippocampus or Korsakoff syndrome (amnesic) and controls. B. Short-term forgetting. Short-term retention of word triads for amnesic and normal controls.

--O-

90. V)

0

5

80-

CONTROLS

DELAYED MATCHING ADJUSTING DELAY

9

70-

I-

V)

lV

60-

W

500

v

40-

5L

30. 20 -

H.M.

10. 0

5

6

7

8

-

3.P. Banquet et al.

+ z

W

TRIALS

LENGTH OF STRING

Figure 2. C. Digit span: Immediate memory for Qgits in amnesic patients and normal controls. Mean percentage of sequences correct as a function of sequence length. (Adapted from Baddeley & Warrington, 1970, with permission). D. Adjusting delay procedure of the delaycd matching-to-sample task. Delay periods on consecutive trials of the adjusting delay series allow a quick check on the limits of the subject's ability to remember the sample. On combination of three consonants trigrams, HM had no difficulty in remembering the trigrams for as long as 40 secs. On the set of geometrical stimuli (eight lfferent ellipses) progression was reduced to steps of 1 sec with best performance at 5 sec delay, achieved once.

CL

w

V3

140

Chapter 4

helps to buffer overflow of information out of the very limited capacity attcndefl STM store. It can be argued in this case that rehearsal is not efficient anymore as a consequence of the overflow of the immediate memory span. Still, normal subjects demonstrate a residual mnemonic performance in spite of this overflow. Taken together, these results confirm the integrity of immediate STM for material within the immediate STM span in amnesics. They also confirm the efficiency of the articulatory loop as a rehearsal device allowing the extension of the immediate STM span into a primary memory, either for verbal material or for material easily verbalizable. This artieulatory loop is intact and efficient in patients. Nevertheless, the results also point to the fact that the possibility of rehearsal is not enough to guarantee recall or learning in patients. If the material is difficult to verbalize and/or exceeds or even is close to the immediate memory span, as in the short version of the visual maze (Milner et al., 1968), there is a complete disruption of the memory process. Therefore, for patients with Hs lesions, there is a clear difficulty in fraetioning a too difficult task into accessible subgoals. Conversely, the results do not give a strong support for a corresponding rehearsal system which refreshes the visual sketchpad when material is not naturally verbalizable. Otherwise, the performance in geometric tasks should have been close to normal in patients. Rehearsal prevented The tests where rehearsal is prevented during the delay between learning and recall are similar to experiments designed by Baddeley to test WM. Indeed, they combine withholding information with an interfering task. However, they differ from genuine WM paradigms, in that the items to be remembered are not at all related to the interfering task, and also not related to each other. Conversely, in a WM paradigm, the task is not designed to interfere with item memorization. The items may even be related to the task, and task performance is quantitatively evaluated. Also, in a classical WM task, STM and ITM components cannot be dissociated while they can in the memory tests by monotonic variation of the delay of recall. The results of these tests critically depend on the extent of the lesion. -Bilateral lesions of Hs. Subjects with bilateral extensive lesion of the medial temporal lobe like patient HM are clinically evaluated as being in a severe, even dramatic, condition. They depend on continuous rehearsal of the information to be retained. Catastrophic forgetting is induced by distraction (Milner, 1966; Milner et al., 1968). The most dramatic illustration of this psychological condition is reported by Milner: the patient HM was able, by devising an elaborate mnemonic scheme, to remember a three-digit number

J.P. Banquet et al.

141

for 15 minutes, but forgot it as soon as he was distracted. The question is then why a normal subject aider maintaining for 15 minutes a three digit number in his field of awareness has no difficulty in remembering it, even after distraction. Even hypothesising that 15 minutes could represent enough exposure for a "permanent" LTM storage, this type of material does not represent any long-term relevance for the subject. There is little chance that the transient storage takes place at cortical level; otherwise patient HM could have learned this information, as he did learn from procedural memory paradigms. Since the main difference between normals and patients, in this case, is the presence or absence of the hippocampal system, it is logical to attribute the primary cause of the recall defect to the Hs and not the cortex. This does not provide a clue for the actual mechanisms responsible for the retrieval of information in normals. In normal subjects, continuous rehearsal or refreshing memories cannot be directly responsible for this retrieval which looks to us so natural. Indeed, the very definition of distraction implies an interruption of the controlled rehearsal process. The retrieval of information in normals (compared with the non-retrieval in patients with extensive Hs lesions) strongly suggests that the continuous process of rehearsal, or reactivation of memory has somehow succeeded in rapidly laying down a trace, not at the cortical level (otherwise both patients and normals would have been able to recall), but at the Hs level. The ITM hypothesis explains the retention of this information in normal subjects, in spite of its irrelevance for long-term storage (LTM consolidation was not involved), and in spite of the interruption of the rehearsal process (STM was not prolonged any more by rehearsal). Fifteen minutes is well within the WM range. This is therefore an argument in favor of an automatic component to WM. We will see how recent findings on the multiple-range memory registers operating in the hippocampal subsystems, in conjunction with the complex closed-loop system of the hippocampus, may provide a clue to the actual mechanisms responsible for these transient intermediate-range remembrances (so useful for our moment-to-moment processing of the continuous flow of information). A more formal assessment of this type of patient was performed in a visual memory task, based on face recognition. Subjects had to select from an array of 25 faces the 12 faces that had been shown 90 sees earlier. Performance fell to chance level in patients when the test was performed with a distracting task interpolated between the presentation of the two sets of photographs (Milner et al., 1968). We have seen previously how excess in memory load, or difficult verbalization of the material to remember, is equivalent to a prevention of rehearsal.

142

Chapter 4 - Partial lesion o f Hs. The picture is apparently different when there is a

partial preservation of the hippoeampal function. This is the case for patients after unilateral temporal lobectomy, or alcoholic patients suffering a Korsakoff syndrome which present an identical pattern of results to the latter (Baddeley & Warrington, 1970). As we have seen, delayed recall tests with interfering tasks present some similarities with the WM paradigms even if they are not identical. The variable delay between item presentation and recall allows a quantitative separation between deficits in STM (20-30 sees), and intermediate memory, beyond this range. This is not the case in the WM paradigms, which represent a more natural situation, in the sense that the material to be memorized may be related to the associated processing task. In the short-term forgetting task subjects are presented for a few seconds (e.g. 3 sees) with item sequences (3 words) well within the memory span. They are required to recall the item sequences, atter delays varying according to trials (0, 5, 10, 15, 30, 60 sees). During these delays they perform a tightly controlled intervening task designed to prevent rehearsal. Manifestly, the forgetting curves present an exponential decay (Figure 2b), and reach asymptote within 30 msecs, which is the maximal range of STM. For this reason, the corresponding decay can be fully explained by a STM decay. Conversely, the residual information withheld beyond this short-term range is logically attributable to an intermediate store independent of either the articulatory loop, or any other aware controlled cognitive process, which are both prevented by the experimental design. Therefore the process responsible for this memory is automatic. That the two curves are almost similar in controls and unilateral temporal lobectomy or Korsakoff patients can be attributed in both cases to a close to normal cortical function sustaining the STM capacity, combined with a residual hippoeampal function attributed to the spared side of the Hs, or to the diffuse and partial lesions of the Korsakoff syndrome, in a situation where item lists to be remembered are well within the STM span (3 items). This interpretation is strongly suggested by the catastrophic forgetting of patients with bilateral lesions, when faced with an interference situation. Further, there is also electrical brain imaging evidence that unilateral hippocampectomy alters little or not at all the cognitive ERP patterns (in particular P300) recorded on the scalp, with the possible exception of very limited areas of the temporal lobes where a trend towards asymmetry shows up (Johnson, 1995). In the classical free recall two component task, lists of unrelated items exceeding the memory span (e.g. 10 words) are presented to the subjects at a pace similar to that of the previous experiment. At the end of the presentation,

J.P. Banquet et al.

143

with a variable delay, during which subjects have to perform an intervening task, free recall in any order is performed. The main difference with the previous task is that the memory load here exceeds the memory span. In the zero delay, immediate recall condition, in spite of the absence of an intervening task rehearsal after each item presentation is prevented by the continuing presentation of to-be-recalled new stimuli. Very robustly, the recall curves (Figure 2a) show in controls both a primacy and a recency effect (Glanzer & Amitz, 1966). The items delivered first and l&st in the list are better remembered than the ones in the middle. The recency effect is classically attributed to a persistence of items in STM. This interpretation is consistent with our model which differentiates STM (present both at cortical and hippocampal level) from ITM (which would be a more specific attribute of Hs). As such, the recency effect disappears both in patients and controls when a 30 see delay with performance of an intervening task is interpolated between presentation and recall. The primacy effect is also classically attributed to long-term memory. But in the context of our model, it is supposedly dependent on intermediate-term memory (ITM). This primacy effect greatly deteriorates at zero delay even in these patients with partial preservation of medial temporal lobe functions (Figure 2a). It becomes completely abolished in the same patients after 30 see delays, while it still persist in normal controls. Thus, when the memory load exceeds STM span, partial preservation of the medial temporal lobe function is not anymore sufficient to prevent intermediate memory deficiency, compared to controls, as was the case in the short-term forgetting task with a memory load within the STM span. In summary, both patients with bilateral and unilateral lesions of the Hs present close to normal immediate STM for information well within the STM span (typically 3 items), and can further extend this temporal range when rehearsal is possible and natural (verbal material). Both types of patients present catastrophic loss of information and forgetting when there is overflow, beyond the strict STM span. This loss of information is much less systematic and dramatic with normal subjects. There are some indications that the STM span may be reduced in patients. When information is well within the memory span (and only then), patients with unilateral temporal lesions or Korsakoff syndrome, preserve close to normal intermediate range memory capacity, independent of rehearsal. This is not the case for bilateral temporal lesions. It must be further emphasized that the side of the unilateral temporal lobectomy is not neutral for the type of performance which is the most spared (verbal or visual). Thus, both the forgetting curves and the

144

Chapter 4

primacy-recency effect in partial lesions support the distinction b ~ e c n an immediate STM and an ITM, rather than an LTM, as was classically stated. Further, the distinctive behavior of patients with a bilateral extensive ablation or lesion of Hs corroborates this ITM as a possible automatic support (non rehearsal-based) of WM. The development of brain imaging techniques based either on EEG, MEG, PET or functional MRI, raises the possibility of a simultaneous and direct investigation of anatomy and function of memory, both in normal subjects and patients. The few results obtaineA by these techniques vary according to the experimental paradigms. They lend support to both cortical and hippocampal components of WM. Electrical brain imaging. Electrical brain imaging can be used to record event-rdated potentials (ERPs). These ERPs can be defined as scalp potentials reflecting different cognitive processing operations or steps performed by the brain. There are different ERP responses discriminated on the basis of their latency and topography. Some reflect automatic identification of the stimulus, like N200 (negativity at 200 msccs) Mismatch Negativity; others like Processing Negativity (N~tanen, 1982) reflect attended stimulus processing (Banquet, Smith & Renault, 1990) and still others like P300 (positivity at 300 msees) reflect context processing or updating. The ERP investigation of WM confirms the involvement of various modality specific or associative cortical regions, for components later than P300, like P600 and over (Ruchkin r al., 1991, 1992). Surprisingly, P300 related to context updating and in particular probability processing (Banquet, Renault & Les6vrr 1981; Johnson & Donchin, 1982) does not reflect the cortical "explicit" component of WM. These results do not allow the exclusion of the participation in a typical WM paradigm, or in other types of paradigms, of a deeper structure like the Hs. Indeed, scalp electrical recording as in EEG is known to explore predominantly, if not exclusively, the cortical mantle of the brain. Furthermore, a P300-1ikr activity has been recorded in the hippocampus (Halgren r al., 1980). But this P300 hippocampal source is not the generator of cortical P300s. Further, it has little influence on cortical P300 generators, since right or left hippocampectomy does not induce any significant cortical asymmetry. There are however two notable exceptions. First, far-lateral temporal electrodes (T5-T6) in oddball paradigms present a reduced P300 on the side of Hs removal. Second, left hippocampectomy induces a change in P300 behavior rather than P300 asymmetry, along with a performance deficit in the number of correctly recognized items in recognition paradigms involving stimulus familiarity (Johnson, 1995).

J.P. Banquet et al.

145

Positron Emission Tomography. PET or functional MRI studies still contribute very little to the solution of the problem of hippocampal involvement in intermediate term memory, despite their long-term promise. Neural correlates of verbal WM involving the articulatory loop were explored by PET measures of regional cerebral blood flow in a task engaging both components of the articulatory loop, the phonological store and the subvocal rehearsal system, compared to a simpler condition engaging only the subvocal rehearsal system (Paulescu et al., 1993). This paradigm allowed localization of the phonological store to the left supramarginal gyms whereas the subvocal rehearsal system was associated with Broca's area. These results are a strong support for the cortical component of WM, and more specifically of the articulatory loop. Nevertheless, since subjects were explicitly instructed to rehearse the consonants to be recognized after a 2 minutes delay period, the absence of significant engagement of Hs cannot be considered as an argument against the implication of Hs in WM, in as far as a specific memorisation strategy was forced upon subjects. It must also be noticed that the task involved recognition and not recall. One of the most salient items of evidence in favor of a combined involvement of cortex and hippocampus in Intermediate Transient Memory (ITM) comes from Squire et al. (1992). In a delayed (3 minutes) cued recall paradigm without any interfering task subjects learned visually presented word lists (15 words). During cued recall, PET scan found significant activation, compared to baseline conditions, of fight hippocampus and parahippocampal gyms, plus right (and to a lesser degree leit) frontal lobes (Figure 3). Left hippocampal region and amygdala did not change their activation level during cued recall. In our model this task corresponds to a test of ITM supporting WM, because of both the delay of recall and the memory load, which far exceeds the STM span. The absence of any interfering task during the delay may have favored rehearsal strategies and a bias towards cortical activation. But this bias may have been limited by the length of the item list. Nevertheless, the activation of Hs along with the frontal lobes provides strong support for a WM with two components, cortical and hippocampal, combined in a closed-loop system. The selective activation of the fight Hs can be explained as a processing bias introduced by the cued recall (visually presented stem completion) towards processing the visual characteristics of the word-forms. The dual involvement of Hs in both priming condition and cued recall (Figure 3) contrasts with the selective involvement of the frontal lobes in cued recall. This suggests that the declarative aspect of memory involved in recall

Chapter 4

146

A

Right Hippocampal 30 .=o

t 201-

0

10 _o

.

'E

-

o ~ W

='~

~

~ o Z

// //

. . . . . .

O-

-10 ~. ~ ~

"20

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

No response

'

Baseline Priming Memory

B Right Occipital 70 60 50 40 30 20 10 No Baseline Priming Memory response

147

J.P. Banquet et al.

C

Right Prefrontal 2O

,

,,

,

,

,

,

,

,

10

El

-10

-20

" No

Base'iine Priming

Memory"

response

Figure 3. Behavior of the fight hippocampal (A), right occipital (B), and fight prefrontal (C) regions, in comparison to the fixation-point control condition, for four task conditions (i: no response; ii: baseline stem completion by first word to come to mind, no stem could form words presented; iii: priming: stem completion by first word to come to mind, half the stems could form words already presented; iv: stems had to be used to recall words from the list presented, half of the stems could be completed to form these words). The right hippocampal response observed in memory minus baseline subtraction (Figure 2A, Table 1, Squire et al. 1992) did not arise simply as a result of a reduced hippocampal activity in the baseline condition. A right hippocampal response was observed both in the priming-minus baseline subtraction and in the memory-minus priming subtraction. Conversely, the right prefrontal response was more specific to the memory task. (Adapted from Squire et al. 1992, with permission).

(in contrast to the automaticity of stem-completion priming) is a cortical, possibly frontal cortical characteristic. At the same time, it justifies the term of automatic component of WM for ITM. The interpretation of these results is biased by Squire in order to support his theory of the role of Hs in the declarative aspect of memory. Thus Hs activation in the simple priming task

148

Chapter 4

is accounted for by a covert implication of explicit memory during the priming condition (Squire et al., 1992). But this explanation is not parsimonious and is to some extent tailored to fit his hypothesis. Conversely, we attribute the involvement of Hs in stem-completion priming to the implicitprocedural characteristic of this task which matches the automatic aspect of ITM support to WM. Further, if we make the plausible assumptions, -that the declarative character of memory is independent of the process of consolidation itself, and that the fmal repository of memories is cortical, in particular frontal, the interpretation of Squire is not supported by the decrease of prefrontal activation to a level close to that of the baseline condition during the priming task.

Neurophysiology: Human Versus Animal Working Memory At variance with some neurophysiologists (Goldman-Rakic, 1994), we do not believe that the criterion "relevant only transiently" is the unique characteristic of WM. As previously stated, WM, at least in humans, could be characterized by combining temporary storage of information and a capacity for more complex cognitive processing. Among tests designed to explore WM in animals, these two functions are obvious in radial maze performance, but not so much in delayed response tasks, where the emphasis is on temporary information storage. The emergence of the construct of WM occurred almost simultaneously in animal and human research. In parallel with human neuropsychological studies of WM (Baddeley & Hitch, 1974), the same concept was developed independently in animal learning to refer to the capacity to retain information across trials within a test session (Olton et al., 1979). These authors were among the first (and perhaps the only ones so explicitly) to propose that the hippocampal system could be necessary for animal WM (memory for recent information of current and specific relevance). At variance with this passive witholding function of relevant information, we think that hippocampus is also equipped for an active information selection and processing, more in the line with human models of WM. In Olton's radial maze, each arm is baited with food. Typically, on a series of trials, in the same session, the animal will avoid revisiting an arm from which it has already taken food, suggesting a retention of behavioral events associated with entering each arm. This WM could involve the encoding of specific episodes associated with specific maze arms. As such it could represent the ITM equivalent of episodic memory. But

J.P. Banquet et al.

149

performance could also be guided by the stronger relative familiarity of cues related to arms more recently visited. In the seventies, recordings performed on awake monkeys trained on delayed response tasks allowed for an extension of the concept of WM. These recordings showed that some prefrontal cortex neurons were activated during the delay period between stimulus and response (Fuster & Alexander, 1971; Fuster, 1980; Niki, 1974). From the very beginning, these activities were supposed to reflect the cellular expression of mnemonic processes. The evidence in favor of a mnemonic process rather than a motor set or any other activity has grown more convincing by the discovery of some specificity in the response. In particular, in an oculomotor delayed response paradigm, neurons alter their discharge rate only for one or a few target locations, thereby demonstrating a kind of memory field (Funahashi, Bruce & GoldmanRakic, 1989). This result has been extended to other brain areas, inferotemporal cortex in particular during DMS tasks (Fuster & Jervey, 1981; Miyashita & Chang, 1988). This form of activity has been qualified as active by some authors, in contrast to a passive form (Eichenbaum, Otto & Cohen, 1994). Passive memory is characterized by a reduced response to familiar or repeated stimuli. Some neurons in inferotemporal cortex fired much less in response to an immediately repeated stimulus in a serial recognition task (Baylis & Rolls, 1987; Rolls, Baylis, Hasselmo & Nalwa, 1989). Item specificity of the neuronal response was also demonstrated (Baylis & Rolls, 1987; Miller, Li & Desimone, 1993). This paradoxical response has been interpreted as a rapid form of habituation. The decrement in stimulus-elicited firing would reflect a decreased responsiveness of cortical neurons to familiar stimuli. At variance with the active memory this passive response could be interference-resistant and may persist through the presentation of intervening mismatch choice cues within the same trial (Miller et al., 1993). Yet, both active and passive memory representations disappear between trials, suggesting a system reset when the information is no longer relevant (Miller et al., 1993). Nevertheless, a gradual cumulative decrement of response across testing sessions to multiple repetitions of items confirms our hypothesis of a graded transition between ITM supporting WM, and transient LTM based on LTP supporting permanent LTM consolidation. It should be noted that this decrement in response to stimulus repetition is dependent on an automatic, passive type of processing. When the sample stimulus has to be actively maintained in awareness for comparison with several test stimuli, stimulus repetition does

150

Chapter 4

not induce a decreased response, but an increased one (Miller & Desimone, 1994). There are striking similarities between electrophysiological recordings in monkey and brain imagery in man, either electrical recording of ERPs on the scalp, or motabolic PET imaging. We have previously mentioned the functional significance of specific ERP components. In particular, N200 or Mismatch Negativity reflects automatic identification of the stimulus in an iconic or echoic memory; others like Processing Negativity represent attended stimulus processing; and still others like P300 reflect context processing or updating. When two types of stimuli are sequentially presented with different probabilities (rare and frequent), the P300 and N200 response to rare and frequent stimuli is at first very similar, but becomes smaller for frequent and larger for rare types of stimuli. This evolution in response amplitude takes place after only a few stimuli (less than 10) for P300, but needs more stimuli presentations for N200 (Banquet & Grossborg, 1987). This kind of probability processing is fully automatic, and in fact better reflected by P300 amplitude when the subject is not aware of it (Johnson & Donchin, 1982). The decrease in amplitude of response to frequent stimuli can therefore be compared to the decrease in neuronal response to stimulus repetition in monkey. Conversely, the Processing Negativity, corresponding to an attended, selective filtering of only some preselected type of stimulus, remains insensitive to event probability and presents an increased amplitude in response to a match (i.e. repetition) condition. The same results hold for PET imaging. A decreased activation was found in the occipital cortical areas in response to items that had been recently presentexl (Squire et al., 1992). This result was interpreted by Squire as a reduction in neural computations required for the processing of recently presented information. The similarities between human and animal elctrophysiological activities extend to preparatory set. A contingent negative variation (CNV) paradigm in humans is a formal analog to experiments used in monkey to explore WM (Fuster, 1980; Niki, 1974). After a warning stimulus S 1 and a delay, an imperative stimulus $2 commands the subject to emit an (usually motor) response. Cellular recording experiments in the monkey show that in the S 1-$2 interval the CNV presents two distinct components, an early one which has been related to the processing of S 1, and a late f r o n t ~ t r a l component with a ramp-like activity preceding $2. This late component does not just correspond to a motor-s~, but reflects also timing and perceptual-cognitive set (Ruchkin, Sutton, Mahaffey & Glaser, 1986).

J.P. Banquet et al.

151

Spatio-Temporal Processing in Hippocampus and Prefrontal Cortex Most of the specific functions attributed to Hs and Pc, mainly in the domain of temporospatial processing, can be related if not reduced to the global WM capacity proper to these two systems. In fact, some of the paradigms used to probe the WM capacities of these two systems, and already mentioned in the previous paragraph, presented strong explicit or implicit spatial or temporal factors. Nevertheless, in this section we aim at a specific parametric exploration of spatial or temporal functions, and also other related functions such as rule formation, planification etc.. Here also we make a distinction between neuropsychological or brain imaging arguments and neurophysiology of the monkey in particular, since both approaches make a specific contribution to the problem.

Neuropsychology and imaging of spatio-temporal functwns in prefrontal cortex and hippocampus Neuropsychology has provided evidence that important lesions of prefrontal cortex (Pc) are compatible with the performance of conventional linguistic, memory or even intelligence tests, which can be sensitive indicators of damage to the temporal lobe. The portraying of the Pc as the "seat of intelligence" was therefore discarded. At the same time, less conventional paradigms brought to light less basic, but no less critical, perturbations of cognitive and relational adaptations following important lesions of the Pc. These patients completely lost the ability to make choices, although sensory and motor sequences remained intact. Reduced flexibility and inventiveness on the approach to new problems, poor adjustment to everyday life and disturbances of personality were among the most common signs of prefrontal dysfunction. These disturbances of personality are diverse, but could find some of their common denominators in a disinhibition of basic drives, combined with lack of motivation and purposeful behavior. For these reasons these personality changes are most obvious in family, professional and social life, all spheres which require a high degree of instinctual sublimation, not to speak of motivation. Since these familial, professional and social components of a person are the more distinctive ones, the patient loses his singularity to become common if not vulgar: "Gage is no more Gage". In spite of their limitations, these neuropsychological explorations uncovered the structural

152

Chapter 4

and functional diversity of Pc, and sketched a "functional topology" that is presently dcmonstratexl by brain imagery. Since Hs seems to contribute to the computation of the leaming of places and sequences and amygdala participates in the integration of external and internal drive signals, one can wonder why motivation, based on this drives, seems to be intcgrate~ not at this subcortical level, but in the Pc. Several reflections along these directions can be proposed: - First, goal formation supposes a precise representation of the target form and location, which can only be precisely dr at cortical level, - Second, goal implementation supposes comparison with LTM, precisely located in cortex and particularly in Pc, but not in Hs; - Third, goal enactment supposes a direct and bidirectional relation with premotor and motor structures which is a property of Pc, but not of Hs, - Finally, goal execution ne~s control and willed action, which depends on the executive function of Pc. All those reasons seem to make the prefrontal stage necessary. But the most compelling argument in favor of the necessity of a prefrontal stage for behavior integration comes perhaps from computational constraints. If drives were integrated at subcortical level such as Hs for subsequent planning, there would be confusion between perceptual and "executive" representations of sequences of events (Gaussier & Zrehen, 1994b; Zrehen 1995). Veridical reproductions of learned sequences could still be possible. Indeed, short cut relations between Hs and motor programming systems such as basal ganglia do exist through subiculum and nucleus accumbens. This anatomical connection permits the possibility of short-circuiting prefrontal cortex through the subcortical implementation of overleamed automatic programs. But there would be lacking the extra degree of freedom provided by a representation which is directly dependent neither on perceptual units nor from motor units, but lies in fact at the junction of the two (see Figure 11, below). This extra degree of freedom results in a giant evolutionary step because it allows for adaptation and creation of new behavioral sequences as needed by specific circumstances.

Spatial Working Memory, principal sulcus and hippocampus. The brain imaging results from tasks inspired by delayed spatial tasks in monkey locate spatial working memory in man either in Brodmann's area 46 which corresponds to principal sulcus in monkey (McCarthy et al., 1994) or in area 47 of the inferior convexity (Jonides et al., 1993). Recent results from PET paradigms manipulating spatial working memory factors (Owen et al., 1996; Petrides, Alivisatos, Evans & Meyer, 1993) or simply simulating this spatial

J.P. Banquet et al.

153

information (Burgess ct al., 1994) have shown the involvement of the hippocampal formation in man during these tasks. Temporal ordering and dorsolateral prefrontal cortex. Frontal lobe lesions present a specific memory deficit. In delayed comparison tasks, they induce an impairment of temporal ordering of recent events (Milner, 1982). The subjects have no problem in making the distinction between new and previously presented items. This could be a function of the hippocampus, or even modality specific cortical areas. But they present a specific impairment in judging the relative recency of items suggesting an incapacity to keep successive, trials apart. Considering the pace of presentation of the item sequences this recency judgement cannot be based on the rehearsal of temporal order or the organization of material, but rather on the relative salience of items in memory. This result is congruent with a defect in STM (maximum 30 sees) whereas the distinction between new and old items is better accounted for by a dysfunction of Hs-basext ITM. A symmetrical pattern of deficit in medial temporal lobe lesions, with no impairment on recency discrimination but a deficit in recognition memory confirms this interpretation. This distinction could be related to the dichotomy in experimental psychology between recency effects (attributed to STM) and primacy effects related to LTM (Atkinson & Shiffrin, 1968). In the framework of our hypothesis this LTM would correspond to ITM. STM would be cortical, and prefrontal in particular, whereas ITM would have an hippocampal support. The left frontal lobes make a special contribution to the organization and planning of responses a few moves ahead. Therefore, the poor performance of patients with unilateral lesion of dorsolatcral (DSL) Pc on memory tasks could rather result from a failure in control processes of memory rather than a deficit in retention per se. The perturbation in the temporal ordering of events aiter DSL lesions is congruent with several interpretations. Pc is either the site of processing of such information, or the site of storage, or both. The processing of this order information in other structures such as Hs in particular, and the subsequent storage in Pc cannot be excluded. We will discuss in the model how this ordering or sequencing function could be dependent on the WM function of both Pc and Hs. Other cognitive functions: - Feedback regulation of behavior and superior Pc. Patients with a superior Pc lesion present an incapacity to extract information from environmental cues elicited by their own responses in order to regulate behavior, in the sense of formulating or altering rules of their behavior. This

154

Chapter 4

inability to overcome previously established response tendencies is manifested by the generation of fewer hypotheses, and perseveration. While superior, mostly left, frontal cortex lesions produce persistent impairment, unilateral, orbital, or inferior frontal lesions produce no deficit of this type. This defect is independent of the valence of the feedback signal, of the nature, verbal or spatial, of the material. This "rule breaking" behavior does not only concern the identification or formulation of the rule. It extends to a specific failure to comply with task instructions and a tendency to error perseveration. As such, it could be viewed as a special "cognitive" instance of loss of inhibition otherwise encountered in the everyday life of these patients. These attitudes of perseveration contrast with the hyperflexibility in adaptation to rule changes of the patients presenting mediotemporal lesions. In this latter case a defect in ITM and the consequent impairment in extracting consistent perceptual or motor schemes for a sufficiently protracted period of time could be responsible for this inconsistent behavior. - Conditional associative learning and periarcuate cortex. In contrast with patients presenting lesions of the anterior temporal cortex, who have normal learning, there is an impairment in learning conditional tasks, either spatial or not, aiter lesions of the frontal lobes. Lesions producing such deficits are more specifically located in the posterior part of the DSL frontal cortex and the periarcuate cortex (areas 6, 8). Even at a plain neuropsychological level, prefrontal cortex demonstrates neither structural nor functional homogeneity. Nevertheless, the diversity of these functions is more an expression of the nature of its cortical or subcortical connections, than the result of structural variations. In parallel with this compartmentalization of prefrontal cortex according to sensory modality other authors isolate regions according to processing hierarchies. According to this criterion, upper and mid-dorsolateral parts of Pc (Brodmann's areas 9) could be devoted to the most executive, high monitoring level ofnonspatial WM. Conversely, low-monitoring requirements (such as those of delayed matching-to-sample) would be processed by inferior convexity. Neurophysiology of spatio-temporal function in animals.

Rats and monkeys have been the main subjects of investigation, especially for the learning of space rather than for the learning of sequences. Hippocampus: Rat place cells versus monkey view cells. It is not the point of this article to make a review of the vast experimental literature

J.P. Banquet et al.

155

concerning place cells and more generally the learning of space in living beings. We will just point to some inter or intra-species experimental disparities that do not find a satisfying issue in experimental accounts, and yet can be integrated in a single unified mechanism by our model. Experimental results on spatial integration seem, at least at first look, to differ significantly from one species to the other, or even in the same species from one experimental condition to the other. One of the most striking interspecies difference, between the rat and the monkey in particular comes from the quasi-absence of place cells in the monkey. Instead, electrophysiological recordings (Rolls & O'Mara, 1995) demonstrate the existence of "view cells" that react not so much to the position of the animal in space, but to the direction of gaze. From such facts some authors have inferred that the hippocampus, at least for some species, could be the site of the construction of an allocentric frame of reference for space representation. Our model and its robotic implementation (Gaussier et al., 1996, 1997a, b) demonstrate that a unique mechanism based on a realistic simulation of the CA3 network explains both the emergence of CA1 place cell-like activity when the robot camera pans 270 ~ around the room (see Figures 4 and 5). Conversely, when the "visual field" of the robot is restricted to less than 180 ~ (similar restriction of the visual field takes place in the monkey compared to rats), the robot becomes unable to develop cells that generalize a place and rather develops analogs of directional view cells. Another important intraspecies critical factor for the development of place cells concerns the position of the landmarks in the arena. In their usual standard position the landmarks are located at the periphery of the arena, and more or less uniformly disposed around the circle. Nevertheless, a few experiments have placed the same landmarks at the center of the arena (Cressant, Muller & Poucet, 1997). In this new experimental paradigm it becomes very difficult to record a place cell type of activity. In most of the cases, place cell coding was not stable. In a very few cases where stability could be found, this happened after a prolonged stay in the arena. In the same vein, the model gives a mathematical account of the absence or at least difficulty of constitution of stable spatial attractor basins corresponding to the learning of a place when the referent is a landmark and not an absolute direction (Gaussier & Zrehen 1995). Spa#o-temporal function of the frontal lobe in monkey: Location versus form in DSL. In monkey, several functionally distinct regions have been delineated. There is some agreement with neuropsychological results in human. In particular, a dissociation exists between the representation of

Chapter 4

156

%"%..~,.\

.~h~.If

.~hr

I i chnit~ talkie

i,_.7,.,......,

/

.,""

L__q.,

_,.,._.....,...,,f.?--'"",,,.

.'i

r ........ i LL.J ,,

~.. ..,"

!.

I'T"}

{.2U

..........a)

9

...:

I 09~

:

:. . . . .

~ ":"..i:...:. ..... .: . . \

0.g~.t

".

...;:~'\

-Ip-"

0S

0.2~"~'.,,~..~ 0

~'"" "

-~

,,<,/""".Sr..

"~.,> .~/" 0 -0,2 "~'~.~ fl" -0.2 -0.4 "--.,., ,./" .-0.4

Figure 4. Place cell-like segregation of space in a robotic experiment, a) Four views are learned (circles), others are associated with one of those learned (boxes). As we can see, if the robot learn to reach the cross from each learned view it can reach the cross from all the views (generalization). The set of arrows represents a possible path. The views are in a 1.2 m • 1.2 m area; the learned views are at 30 cm from the center. The scale is not respected for the position of the different furniture (in fact they are about 1.5 m from the center), b) The different learned view cell activities for several positions in the same environment (unit is meter).

9~:

9

~ i

!

' ,.

....

.

.....

,~,

.~"

.

~L."." .....

J.P. Banquet et al.

~ .,~.. " ::~: ' ] " :::::::::: [ ~ i

- ......

..... .!..,..:!~~

: ~ ~,~..-~,x.:~.

..: ~

.... i~'~

...:~,,:,.

....

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

i

:2 " '~

~

t

:........ .9.............

i

.......

-"-.

'

'

i....,' :

'

:

".,

;''

.

",~'~..

9

i

............. !

157

9 l,-,q

r

0

0

158

Chapter 4

external space, for a stimulus which has been removed from the field of vision (spatial form of delayed matching to sample or delayed alternation), and representation of the form of the visual objects. Elcctrophysiological recordings and lesions studies have localiscd the memory for space in dorsolatcral prefrontal region, principal sulcus in particular. The inferior prefrontal convexity vcntrolatcral to the principal sulcus processes color and form, since lesions in those areas induce deficits in tasks requiring memory for the identity of objects (Goldman-Rakic, 1994). Specific stimulus identity neurons there, are particularly responsive to pattern delayed response tasks. The reverse is true for neurons in the dorsolateral prefrontal cortex, and sulcus principalis coding for location. Neurons hold "on line" these distinct types of information when the stimulus is no longer present, and arc clearly distinct from neurons coding for the direction of limb or eye movements that can also be found in the same areas. Degeneration studies have confirmexl the anatomical links between sulcus principalis and posterior parivtal cortex (area 7), and between inferior convexity and infcrotcmporal cortex (area TE) which respectively process spatial and pattern visual information. As yet, it is not clear if spatial information recording takes place in either alloccntrie or egocentric coordinates or in both. The possibility remains of both, since at least in primates hippocampus could be the locus of processing for space in an alloccntric (not necessarily cartesian) frame of reference. Response pattern and motor control in inferior and arcuate convexity. Inferior convexity has also been attributed diverse other functions. According to the unlearning hypothesis, these may include inhibition or unlearning of response patterns which do not maximize the probability of reinforcement, and therefore induce a change of a predominant response mode. Finally, inferior convexity could be involved in motor control for interrelating the stimulus-reinforcement associations with behavior, in order to promote motivated behaviors and to prevent indiscriminate motor choices. Autonomic and motivational control m orbital frontal cortex. Another, uncontrovcrsial functional focus is located in the modial orbital cortex and endowed with autonomic and emotional control. It could be involved in motivational evaluation of reward and continuous modulation of ongoing behavior by competing drives. Nevertheless, in monkeys as in human, there arc arguments against a functional compartmentalization that would be uniqudy based on perceptual modality. Reversible lesions of the dorsolatcral Pc during a cross-modal version of delayed matching-to-sample task produce deficits in somasthcsic,

J.P. Banquet et al.

159

spatial or nonspatial WM demonstrating the amodal or supramodal processing taking place in these areas. This could result from a partial convergence of inputs from posterior cortical areas to the Pc. As in man, at this neurophysiological level, there is evidence both for functional uniformity of the prefrontal cortex, in particular with regard to WM function and neurotransmitter modulation such as dopamine, and also for diversity, as shown by lesion and degeneration studies in animal. Functional Model

We present first a global model of Pc-Hs functional relations which specifies the contribution of Hs in LTM consolidation and transient WM operation. Then we compare and contrast the respective roles of both structures in spatio-temporal processing. In particular we give an integrated view of the different experimental results concerning place cells, view cells and the experimental conditions leading to the emergence (or not) of such abstract representation of space as place cells, and also the plausibility of more elaborated cognitive maps. We suggest one plausible mechanism (retropropagation of goals) and its cortical implementation which controls selection and hierarchical ordering of a sequence of elementary step-actions in order to lead to subgoals and final goal. In this endeavor we not only take into account cognitive processes but also drives and motivations. The mathematical model has been simulated. It is also the object of an algorithmic implementation for the control of robot navigation either in open space, or maze environment. Co rti co-hippocampal relations

Figure 6 shows some of the key pathways of the Hs. There are many common points between functions of Hs and Pc, and one of the major problems of research on the functional relations between these two structures is to find what are the specific contributions of each of the two systems to this apparent functional overlap. In particular both systems seem to participate in different aspects of WM, and the capacity for a WM function can be expected to contribute a major step to encephalization. On anatomical grounds, the localization of these functions in Hs and mostly in Pc (the development of which is the most pre-eminent anatomical correlate of primate evolution), supports the functional importance of a "protracted present" for humanization and personification. WM provides the animal with the

160

pattern recognition and time spectrum expression

Associative and pnfrontai Cortical areas

prcfrontal conex

refusion configurntonal associations

Associative cortical (nucleus accumbe

panem completion

bl

A

plastic fusion of multimodal ECZ

OR operation ?

EC3

-

I

1

1

*-_,-Y

,u

:

.

,

- - - - - - - - - - - - - - - 4-:---, ~eptum

(ACh)

1I

Novelty detection. learning modulation

Figure 6. Schematic representation of the hippocampus. Entorhinal Cortex (EC) receives information from the associative cortical areas. Highly filtered information from the EC layer 2 is transmitted to the Dentate Gyrus (DG) and to the CA3 pyramidal cells. In the model, the DG recognizes activity pattern from EC2 and develops a temporal activity spectrum (factorization of time and pattern). CA3 recurrent links allow pattern completion and association between an incoming pattern and a previous pattern recognized and maintained in DG granular cells. CA3 allows a "plastic" fbsion of multi-modality information. Integrated recognition of the CA3 representation is performed on CA1 pyramidal cells. It can be either place recognition (i.e. rat experiments) or view recognition (i.e. primates) and more generally sequence recognition. Information from CAI is then treated by the Subiculum (Sub). This information could become invariant to animal orientation due to this structure (i.e. head direction cells).

Chapter 4

u

I

J.P. Banquet et al.

161

possibility to escape from a purely reflexive behavior, by taking into account its more or less immediate past and simulating the consequences of planned actions into the future, so the animal's behavior can be grounded not only upon external stimuli, but also on internal representations. A direct consequence of this WM capacity is the possibility to learn and to plan long temporo-spatial sequences of events or actions hierarchically subordinated to a global but not physically accessible goal through intermediate subgoals. This capacity seems to constitute one of the distinctive properties of Pc. This temporo-spatial property is crucial for the orientation and situation of the animal in its environment. Nevertheless, even in this domain of sequence learning and spatio-temporal integration, several levels of analysis must be delmeatexl. At the level of the hippocampus in particular, in rodents, there is evidence of them and gamma rhythms possibly generated by inhibitory feedback circuits. These rhythms can be related to a basic oscillation of postsynaptic potentials under the influence of this inhibitory modulation. The gamma rhythms in particular, overriding the them activity, have been viewed as the encoding support of discrete event sequences during exploratory behavior. These sequences would be partially repeated, proceeding further as a travelling window of activity, at each theta cycle. The hippocampal NMDA receptors are endowed with slower time constants (circa 200 msecs) than the short time constants of the cortical NMDA receptors (circa 20 msecs). Therefore, they will allow for the learning of correlations between successive events occurring less than 200 msecs apart, that is about a full single them cycle, and approximately 7 event-coding gamma cycles. This possibility remains quite plausible for rodents that would thus be able to separate events about 20-30 msecs apart. In confirmation that learning this type of sequences of elementary events could take place at Hs level, a primacy gradient of LTP has been found at the synapses between Schaffer collaterals and CA I cells during the presentation of a sequence of stimuli (Granger et al., 1994). By virtue of this primacy gradient the earlier the event in the sequence, the larger the corresponding synaptic facilitation. Therefore at the presentation of a stimulus of the sequence the playback of the sequence in the correct order is favored. Despite the interest of this mechanism, our model does not operate at this low elementary level of resolution. It rather tries to emulate events reaching awareness in the human field of attention and consciousness. At this level individual mental states have an approximate duration of the order of the second. This does not certainly preclude the possibility, even in humans, of automatic subconscious temporal discriminations of sensory events in

162

Chapter 4

particular, of the order of a few tenths of millise~s. Increasing in duration, our activation states gain also in flexibility and control, and become closer to the brain states that actually control behavior. In the course of the chapter, arguments from neuropsychology and brain imaging have been provided in favor of an involvement of Hs at every phase of information processing, from the very short-term to the transient long-term. There are also some arguments against the implication of Hs as a "permanent" LTM store, although opinions diverge on this issue. Thus, we have been led to contrast hippocampal function devoted more specifically to information selection, consolidation, and basic temporo-spatial correlation from cortical function supporting a direct dialectic confrontation between STM and LTM during recall and recognition. Cortex-hippocampus complementarity. The first claim concerns the complementarity between cortical and hippocampal memory systems. Cortex is mostly endowed with short-term and long-term permanent memory capacity, even though there exist range variations from primary, to secondary, and association areas (Lii, Williamson & Kaufinan, 1992). It has more specifically, but probably not solely, a capacity for slow learning. Hippocampus is richly endowed with a full spectrum of temporal ranges, from the short-term to the transient long-term, possibly to the exclusion of the permanent long-term. More specifically, this combination of a large variety of memory registers, with just as much diversity of close loops of different sizes, including one to at least five synaptic relays, seems unique in the brain. It is probably responsible for the property of one-exposure learning proper to hippocampus. On the basis of logical arguments yet to be experimentally confirmed, we suggest that fast, transient learning should take place not only inside the hippocampus itself but also at the interface of the convergent cortico-hippocampal pathway (either r cortex or dentate gyms), and similarly at the interface of the divergent hippocampo-cortical pathway (possibly in the cortex superficial layers). Conversely, experimentally confirmexi slow-permanent learning in the neocortex could be restricted to the level of polysynaptic cortico-cortical connections. Two types of transient memories. The second claim concerns the complementarity of the two aspects of hippocampal transient memory, Intermediate Transient Memory (ITM) supporting a WM function and transient LTM supporting the process of WM consolidation. These complementary functions cooperate to fulfill the contradictory constraints of storing as much relevant information as possible in a large but limited capacity system. The WM function thought to operate both at cortical and hippocampal level is devoted to this segregation between information worthy

J.P. Banquet et aL

163

to be "permanently" stored because of its relevance for survival, or its human social relevance for the personal history of the subject, and information which can be forgotten without major damage. For this purpose several hardwired devices implement different functions in various subsystems of the hippocampus: Severe filtering on the basis of stimulus intensity, duration, or repetition, mostly at the level of entorhinal cortex; Orthogonalization of noncorrelated patterns of reformation, along with lumping correlated information and/or suppression of redundant patterns, possibly in the dentate gyrus; Correlation of these orthogonal patterns on the basis of spatio-temporal criteria of either co-occurrence or sequential ordering. The CA3 subsystem is specifically equipped for these processes of autocorrelation. There is some evidence of temporal order processing of the events at the level of CA1 (Granger et al., 1994). All these functions argue in favor of a specific processing capacity devoted to the Hs, based on ITM. This may be possibly more devoted to temporal aspects of information (like temporal order and timing), than spatial learning, which is the focus of current investigation. More exactly, spatial learning itself would include a strong temporal dimension (in the sequential recording of snapshots), where temporal order is not relevant. Correlational processing could be the hallmark of Hs, and could account for many aspects of its cognitive processing, including temporal processing. But it is certainly not the only form of processing performed in the hippocampus. On the basis of CA3 autoassociative architecture, artificial systems can be designed such that simultaneous pattern correlation processing becomes a special case of successive pattern temporal order processing. The main difference is that events occur simultaneously instead of taking place in sequence. The other class of transient memory is transient LTM based on LTP, and a support for permanent LTM consolidation. The most prominent sites of LTP in the hippocampus are at CA I level, but also in the entorhinal cortex, dentate gyrus and CA3. LTP is not specific to the hippocampus. Different cortical areas, in particular, are also prone to LTP. What seems more specific, if not exclusive, to the Hs is the conjunction of systems susceptible to LTP along with shorter versions of synaptic facilitation, in the framework of an architecture adapted to facilitate a smooth transition between short-term and long-term facilitation, without the need to resort to externally dependent repetitions of activation. These two functions of ITM and transient LTM are therefore complementary. Indeed, information or events have to proceed -

-

-

164

Chapter 4

through the cortical and/or hippocampal WM process in order to get a "certification" as being worth passing to the history of the individual. This certificate could correspond to the transition from ITM supported by short duration synaptie facilitation (STP), to transient LTM supported by LTP. A unique mechanism for LTM and W~. The third claim concerns the unique mechanism supporting WM operation and LTM consolidation. This unique process takes advantage of the reciprocal, topographically organized connections between cortex and hippocampus, combined with the endogenous property of pyramidal neurons, mostly in the CA3 region, to discharge by bursts (Buzsaki, 1989), and synchronise within populations connected by previously facilitated synapses. Endogenously generated bursts of activity have been documented in the CA3 region, and are manifest at CA1 level as sharp waves. In contrast with theta rhythm which have been consistently recorded mostly in rodents, sharp waves have been also found in primates and humans. They could form the basis for reactivation of recently facilitated neuronal populations either at hippocampal level or at cortical level or both. They would correspond to what we call the print mode of the Hs, by contrast to the read mode. In episodic learning, this print mode would subserve the function, attributed to the repetition or practice process during the formation of procedural memories. Two general mechanisms will cooperate in the reactivation of cortical patterns, starting either from these endogenous hippocampal bursting and synchronization capacities, or from hippocampal reactions to cortical activation of cue patterns. First, there is resonant or reverberating activity between reciprocally connected networks. It has been used in neural network modelling as a mechanism for synaptic weight modification leading to class learning (Grossberg, 1976a,b). Second, neural population synchrony, a consequence of the reverberant activity, has been presented as a plausible temporal coding used locally by the brain for the coalescence of features into a unified percept (Singer, 1983; Gray et al., 1989). The main contention of the model is that the central location of the hippocampus as a site for both convergence from, and divergence towards the different cortical areas allows the Hs to act as an information selective pacemaker (Banquet, 1983; Banquet & Contreras-Vidal, 1994). As such, it can selectively synchronize distant cortical areas that have been previously coaetivated. This mechanism provides for both the refreshing of memories during WM operation, and the reactivation of cortical patterns during the more lengthy process of LTM consolidation. In this last case it will create the conditions for the slow facilitation of distant polysynaptic cortieo-cortieal connections, which

J.P. Banquet et al.

165

eventually will make recall or recognition independent of hippocampal function. Transient storage by the hippocampus. The last claim is largely but not universally accepted. The information processed through the hippocampus is not bound to be permanently stored there, at least in humans. Beside anatomical arguments, like the relatively limited capacity of the system, the strongest support for this contention comes from the limited range of the retrograde amnesia in hippocampal lesions. Most of the memories from the distant past are spared. This raises the issue of the necessary transfer of the information to cortical and/or cerebellar areas, which are supposed to be among the most pre-eminent sites of permanent storage.

Model of dentate-CA 3 function The experimental sources for the inspiration of our mathematical model are of two different types. First, the learning of space by animals as demonstrated by rat place cells or monkey view cells, and eventually the formation of cognitive maps in an allocentric frame of reference, more plausibly in primates. In our model this structuring of space is considered as a degraded form of sequence learning. Second, in trace conditioning with a motor response there is evidence of adaptive timing of the behavioral response, as we shall see in the dentate gyrus model. In our model these two subsystems of Hs, dentate gyrus (DG) and CA3, are endowed with specific functions which go beyond the classical attributes of pattern orthogonalization (dentate gyms) and autocorrelation (CA3) which are the usual characteristics of these subsystems in neural network modeling. These two classical functions are in fact implemented in such a way as to add a temporal dimension to both of them, on the basis of experimental evidence. The addition of this temporal characteristic to these networks is globally justified by the temporal properties of the hippocampal registers which are capable of fast transient short-term or long-term learning thanks to the synaptic properties of short-term and long-term potentiation particularly developed in Hs. To this propensity for rapid transient learning must be added the loop-like organization of the different circuits which allows for a few repetitions or iterations of patterns of activation even though they have been externally triggered only once (Buzsaki, 1989). Thus, these loops endow Hs with the potential to maintain for some limited duration, or rather to punctually reactivate, a recently activated pattern that has been transiently learned in the synaptic connections of the recruited populations.

166

Modality 1 Modality 2

!

.r

Modality 3 DG 0

No ACh

I

1.

11

t -4

-

intermediate connections Proximal

Ach modulation

Septum

"L_J Figure 7. Detailed model of the CA3 connections.

Chapter 4

ACh

Distal connections

J.P. Banquet et al.

167

As a consequence of these temporal properties, the model implements two different temporal functions: First, a function of timing, i.e. a flexible evaluation of time intervals or durations, at the level of granule cells of the dentate gyms; - Second, the corresponding temporal or order function of the CA3 network consists of linking together successive similar or different events of a sequence (see Figure 7). This chaining process results from a combination of autocorrelation and crosscorrelation as we will see in the paragraph on the CA3 model. Factorization o f time and pattern in the dentate gyrus. The model presented here is based on the same principles of population learning and coding of time by a limited assembly of neurons whose dynamic range of activation varies along a biologically plausible continuum. This timing function has been attributed to the dentate gyrus on the basis of results derived from the paradigm of trace conditioning of the nictitating membrane response in the rabbit (Berger & Thompson, 1978; Solomon, 1980). The evidence of first monitoring and then anticipation of the behavioral response, by firing patterns of pyramidal cells shows up first at the level of CA3 neurons. This lets us suppose that processing the duration of time intervals takes place upstream to the CA3 system. Dentate gyrus is the immediately preceding stage (see Figure 6). Granule cell population there is sufficiently important to present variations in size and time constants which could support different dynamics of activation. There are several lines of evidence for neural dynamics in the brain varying according to a continuum, in particular in spinal motoneurons and there is also evidence of spectral decomposition of space in the visual cortex. This hypothesis forms the basis for a population coding of time by cells endowed with these different dynamics. Similar neural mechanisms underlying the hippocampal adaptive timing function during conditioning were proposed by Grossberg and Merrill (1992), among others. We have already proposed a model including a timing system which gates or modulates the flow of information in a categorization system (Banquet & Contreras-Vidal, 1993a,b; 1994). The main difference between this last model and the present one, is, first, that the onset of the input stimulus is not held on by an hypothetical external device but the STM is directly implemented through the positive feedback from the GC to MC and their return to the GC (see Figure 8 for the DG architecture and Figure 9 for a simulation of its temporal dynamics). A second difference is that the same populations of cells process, or register, .both pattern and time, thus realizing a "factorization" of the two parameters. This basic computational competence -

Chapter 4

168

may also be shared by cerebellum which is more suited for the motor response (Ivry & Keele, 1989; Bullock, Fiala & Grossberg, 1994). There is at present no decisive argument against the fact that the cerebellum could support the upstream system forwarding timing information to CA3.

time axis

Pattern

GC "

...-,--r . -,,. ,--

,r.I .I,,

i l

I

;,

i

-,--~ ~

i

~ ,

.t,

!

-,--

---r-

!

:l ( ( C

,L

/

I _

BC ---- ~ I "~

~.

( 17

EC2 -

!

',/,,

c,c ~ BC,

I

L-O t~

" .....,.

BC

~

.-

~

.

.

"

.

.

MC

.:

"

. . - . . . 9

inhibitory link excitatory link

.

.

.

. . . . .

. ,

I to C A 3

Figure 8. Detailed representation of the different neurons and connections in DG used in our simulations. Instantaneous information coming from EC2 is learned by the Granule Cells (GC). We suppose activity spectrum of GC depends on their size. GC are linked together in a functional cluster through a particular Mossy Cell (MC). This recurrent connection allows GC to develop time spectrum activity while EC2 stimulation is over. Basket Cells (BC) allow choice of a winner cluster of GC: pattern recognition of EC2 information. Another MC and BC subset allow the reset of the active GC cluster at the onset of new significant input from EC2 to avoid pattern recognition problems.

activ~tyof the associared MC

J.P. Banquet et al.

0

time

169

Figure 9. Simulation of a cluster of "spectral" cells (GC) randomly selected in the Dentate Gyrus (DG). A pattern is presented at time 0. As a result, the GC develop a time spectrum activity depending on their size. Activity of the GC is maintaining due to recurrent connection through a MC cell. The decrease of GC activity is due to an habituation term at the synaptic level (between MC and GC).

L

%

170

Chapter 4

Nevertheless, the plausibility of multiple local timing systems dedicated either to perceptual, motor, or even cognitive processes argues in favor of an hippocampally based local system for the timing of perceptual events. Figure 9 shows how a particular punctual event can be extended in time by a battery of cells endowed with randomly varying time constants, until the subsequent significant event takes place. The two events could be the CS and the US of a trace conditioning sequence. But more generally, our system deals with any arbitrary sequence of either distinct or repetitive events or both, occurring at variable time intervals. Thus, it can account for the learning of any pair, and therefore sequence, of events independently of the presence or not of a reinforcer. Reinforcement learning and conditioning becomes a particular case of this more general associative learning. It cannot be excluded, on the basis of the presently available experimental evidence, that the function of maintaining a pattern of information in STM or ITM could not be performed by CA3 itself. There is indeed evidence of such a maintenance of information under a "dormant" form of short-term or long-term synaptic facilitation. In order to be used by the system for purposes of information processing (match-mismatch, filtering, correlation...) this dormant form needs to be reactivated or reenacted as a coherent pattern of activation. In a sequence of dissimilar events, this reenactment is not possible from one event to the next. Therefore the pattern of activation must be maintained in an active form in order to be compared or associated to the next one. The circulation of information in a loop could in theory also perform this function of maintaining a pattern of information in an active state. Also, event-locked them activity could be another mean used by Hs to protract in time an active pattern. Our system of population coding by a battery of cells seems parsimonious, flexible and therefore plausible, but it does not exclude these other possibilities. In our model, relations of Hs with other structures in particular frontal cortex for temporal order processing, and cerebellum for timing are viewed as special cases, in the temporal domain, of the more global function of Hs as a rehearsal system for the rapid acquisition of any kind of information. This interpretation seems to be confirmed by the fact that learning this type of information, time intervals in particular, is not really suppressed by hippocampal lesion. It just requires more trials to be learned. ,4 temporal order network in CA3-CA/. Different models of temporal order processing have been designed. The most recent models aim in particular to emulate temporal processing attributed to prefrontal cortex. Guigon, Dorizzi, Bumod and Schultz (1995) have designed a model directly

J.P. Banquet et al.

171

inspired from the electrophysiological studies performed in monkeys. The implementation of processing units which can commute between two stable states of activity (bistable units) in response to synaptic activations allows the learning of temporal sequences. Atter learning, the sustained activation of a given neuron represents the selective memorisation of a past event, selective anticipation of a future event, or the prediction of a reinforcement. Thus, the model reproduces the functions of delay neurons encountered in frontal cortex. Bapi and Levine (1990, 1994) and Levine and Park (1992) designed network models of frontal cortex function. In these models, sequence learning is secured by storing in LTM (synaptic weights) the transitions between the events of a sequence. The different types of learned sequence are encoded in a compressed form, and then categorized. It presents some similarities with our own model of WM for temporal order and probability coding (Banquet & Contreras-Vidal, 1992a,b; 1993a,b; 1994). Both are in particular inspired from the same design principles of Grossberg (1978). Dehaene, Changeux and Nadal (1987) implemented a network that can learn temporal sequences based on biological properties of allosteric receptors. In our model, the dentate gyms performs simultaneously the double coding of time and pattern. This output of the dentate gyms can then be used by the CA3 stage. This stage represents a multimodal event or state as a correlation pattern of activation. It performs then for each event a double correlation: -First, a zero-delay "auto-correlation" of the event with itself thanks to the temporal conjunction of the direct pathway input to CA3 and the fast spectral component of the indirect input transiting by the trisynaptic loop of the dentate gyms (Figure 6). -Second, a delayed auto- (stimulus repetition) or "cross-correlation" (successive stimuli different) between the present direct input to CA3 and the previous indirect input which has been maintained in an active state thanks to the slow spectral components of the dentate gyms granule cell population (Figure 9). CA3 learns this cross-correlation between successive events. This last process results in the recording of transition states between events. These transition states allow at any points in the sequence prediction, priming or playback of the subsequent event according to the functional mode of the system (i.e. learning, recognition or reenactment of a sequence). The transition from a functional mode to the next is endogenously determined by the type of neuromodulation (ACh in particular; Hasselmo & Schnell, 1994) and the biological rhythms.

172

Chapter 4

For the purpose of linking a specific behavior to a particular view of the environment, as in a PerAe (Perception-Action) architecture (Gaussier & Zrehen, 1995), linking of thesetransitional states to elementary motor actions has been found to better support the planning and implementation of specific sequences, rather than the mere relation of perceptual states to actions (Figure 10; Gaussier et al., 1997a,b). These transition states do not differ in their coding from simple states since they are also encoded as correlation patterns in CA3 network. Their meaning is different since they correspond to the correlation between patterns which do not occur simultaneously in time. The timing system of the dentate gyms is set in derivation on the main flow of information that transits through CA3, and thus modulates or even gates and controls the activity of CA3. This property makes the system very flexible with respect to the nature and variable timing of successive significant events. The learning not only of autocorrelations between simultaneous patterns of activation, but also of "cross-correlations" between successive patterns can subserve several functions. First, during the learning process of a repetitive sequence, the predictive activity of the system results in priming the next event in the sequence when the present pattern of activation acts as a trigger for the transition pattern. This priming is critical for the implementation of the match-mismatch process that operates to recognize the successive events of a sequence at CA3 level. There, the previous pattern maintained in DG and reaching CA3 by the mossy fibers is matched with the present pattern of activation forwarded by the direct connections from EC to CA3 pyramides distal dendrites. Therefore, the learning process is modulated according to the degree of predietedness or novelty of the events. Hasselmo and Schnell. (1994) have shown how ACh septal modulation of the Hs activity could provide for an automatic control of the learning process, as a function of novelty or familiarity of the events. One can suppose that a match inducing familiarity preempts the learning of a new transition and sequence. The resulting "resonance" between learned and input patterns provides for both the transient stability of a state and the reinforcement of previous learning. Conversely, mismatch between the expected or primed pattern and the input pattern fosters, by means of a high Ach modulation, the prevalence of input over the stored pattern and therefore the recording of a new transition pattern. Second, in sequences of events that occur only once or a few times, the Hs provides the unique facility for fast transient learning, thanks to its capacity for reenactment of recent patterns of coherent activation. This reenactment correspond to a synchronous discharge of CA3 neurons which activates CA I pyramidal cells. The activity of CA1 cells manifests as sharp waves (Buzsaki,

173

J.P. Banquet et al.

,%. ~o'~ r

q

#7/.~ ~,

..~o"/

/,"

/

l

,~~5'~'0-~

l

)

__~

~

~

~"

T//~

.~S~"/

'~W-'----",jta~ ~l),~'- ~ C

'.,,

', AE

/

Limbic System

Max operator

. . . . . . . . . . /. . . . . "'.

~,,*V 0A I t - . . .

//I\\

Motivation

Activation diffusion

/

: , ..... ~..... / efferent copy of CA! pyramidal cells

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

-"

/__. /'~

.... .K'-->" / ....

". . . .

direction of movement

/

/.__~ ~

Motor Area

--~"

Associative motor area

Figure 10. Goal backpropagation: the model is used to learn transitions between stable states such as to go from A to B (AB) and next from B to C (BC)... on CA1 neurons. CA3 neurons are used to merge the recognition of the current state coming directly from the entorhinal cortex EC2 and delayed information about the previous state that we suppose comes from the Dentate Gyrus and more precisely from the Granular Cells (the GC developing a time spectrum expression of the information coming from EC2). The CA3 representation of the state transition is then learned on CA1 neurons and copied on prefrontal cortical neurons which learn the "anti-causal" (backpropagated) links between pairs of transition states. These prefrontal neurons also learn associative links with motivation nodes if they are activated simultaneously with them. Assuming the activation of a motivation coming from the limbic system (the will to eat for instance), the model explains the propagation of the goal to subsequent subgoals on the prefrontal cortex. When a subgoal transition can be achieved because the robot state corresponds to the first part of the transition, the associated learned action is the trigger and allows the robot to reach a new state which is the starting point for triggering the recognition of a new transition. When several transitions are possible from a current state (like BC or BD) the level of activation of the prefrontal neurons associated with the different possible transitions allows the choice of taking the shorter or the most interesting pathway from the current state to the final goal.

174

Chapter 4

1989). These sharp waves present some of the characteristics, in particular in amplitude and frequency, of the stimuli capable of inducing LTP. They could therefore participate in the reactivation of cortical patterns necessary for the LTM storage at cortical level. The linkage between successive events constituting transitional states would result in a playback not of single events but of sub-sequences of events.

From hippocampus to prefrontal cortex This memory of states and transitions forming a sequence of events is, in the model, ~rther integrated at CA1 level, with a possible partial restitution there of the "cortical" topology thanks to the direct pathways from entorhinal cortex third layer to CA1. From CA1 there are direct as well as indirect connections to prefrontal cortex. At this prefrontal site the long-term registration of a sequence can be considerd as a result of the successive activation of nodes which can be assimilated to cortical columns. By the very orderly nature of a sequence encoded by a spatial pattern of activation, one could assume that the best neurophysiological support for such a storage would be a unidirectional facilitation of the synaptic weights of a specific path in a network architecture, as it has been implemented in Bapi and Levine (1994). It is plausible that such an oriented unidirectional facilitation takes place in primary or secondary cortical areas. Nevertheless, the prefrontal cortex is the most plausible site for the linkage between sensory and motor sequences at least at a high level of controlled processes (Figure 11). This does not preclude the possibility of sensory-motor links at subcortical or even lower levels as schematized by Figure 11. At cortical level, the execution of a sensory-motor sequence is necessarily linked, at least implicitly, to the completion of a goal selected by motivation. A goal in the model corresponds in fact to a secondary goal, i.e. a situation which allows the satisfaction of a basic drive or of a sublimation of this basic drive. At the executive controlled level of the prefrontal cortex there is clear, even if only subjective, evidence that the goal is usually present and therefore activated at the very onset of the sequence. Thus it can influence the choices of subgoals and the hierarchical unfolding of specific endeavors to reach them. The most parsimonious implementation of this psychophysical reality requires the instantiation of a bidirectional facilitation of the different pathways leading from the starting point to the goal of a sensory-motor sequence. In this way, the activation of a goal induces a retropropagation of activity, similar to a priming by top-down activation from the categorial

J P. Banquet et al.

,2

Posterior association cortices

175

Premotor cortex

e~

// cL

Primary motor cortex

Primary sensory cortex

,// Perception

Action

I I I

External environment I I I

Figure 11. Flowchart of multilevel information transfers: 1) between sensory and motor areas (horizontal connections) 2) between sensory (or motor) and associative areas of increasing complexity (vertical connections) 3) between internal drives and planification capabilities (prefrontal cortex).

nodes in an ART architecture (Grossberg, 1976a, b). Yet, here the priming process concerns an entire sequence of events, and accordingly, is implemented according to a gradient. This subliminal backward priming of a sequence in conjunction with a bottom-up activation from subcortical structures such as the hippocampus helps the selection of the best sequence of actions to reach a specific goal. This process of goal retropropagation is not solely efficient for selecting the optimal way for goal attainment. It also operates in deciding the order of goal satisfaction, when several goals are

176

Chapter 4

simultaneously active or in competition. As such, it implements aspects of hierarchy setting performed by Pc. Similar disambiguation between several possible sequences takes place at Hs level during the playback of the sequence if the prediction of the future event is based on several previous events instead of just the immediate preceding one. When such a neural network is used for the control of a mobile robot, different types of behavior in order to select and reach a goal are exhibited by the robot. These choices are dependent on three parameters of the system: the relative strength of the different drives, the weights of the connections implicated in the different paths, and the required match level between perceived and memorized steps towards the goal. Any node of the cognitive map leams transitions between pairs of learned places. The level of activation of these nodes results from the addition of bottom-up (match-related) and topdown (drive and path length-related) activations. These combined activations of the nodes can lead to a variety of behaviors of the system which have counterparts in real life. If the top-down influences are too weak, the system is unable to follow a specific path for the attainment of a specific goal. It is susceptible to distraction by any new input previously associated with a different behavior. Conversely, if the saliency of the top-down input is too strong, the recognition of a situation could be biased in the direction of a situation corresponding to the satisfaction of its goal. The initiation of such erroneous recognitions can be self-reinforcing. All these situations correspond to pathologies of frontal lobe.

Fronto-Hippocampal Function and Personality Personality in humans presents three key "primitives": 1) A temporal function which seems to obey some principle of symmetry of past and future, memory and prospective, with respect to the present. This temporal memory function is mostly based on the capacity to evaluate and record the order of occurrence of event sequences (a kind of segmentation lost in frontal patients), and eomplementarily the capacity to recognize new from familiar events (a kind of fusion of events lost in hippocampal patients). We have seen the importance of the Hs in the recording of "one exposure" events. The consequence of the suppression of this memory function is illustrated by anterograde amnesia, i.e. the incapacity starting at a period of life, corresponding to some time prior to the lesion, to build up a continuing history, just as if the factual life of a person had stopped at this moment. Yet, the Pc is also involved in this historical function, as one of the favoured

J.P. Banquet et al.

177

cortical sites for the permanent recording of these event memories. Symmetrically, the prospective function supposes the capacity to project in the future an orderly sequence of planned events in order to either actually perform them or merely simulate them. This planning capacity is also an hallmark of personality. This function is as important as the previous one, and in fact intimately linked to it. Our capacity to make plans, i.e. to project our actions in the future, is narrowly dependent on a library of past behavioral schemes and of their consequences. As our personal history goes back in time as far as our early childhood, our ability to project our life in the future concerns more or less remote time. The range of this prospective capability is closely linked to the strength and integration of our personality, and supports our motivation. Mostly from neuropsyehological studies, the role of the Pc in this function is well documented. Pc is essential not only for the strict and logical ordering of events or actions. Furthermore, it operates in the determination of an hierarchy of subgoals and actions to reach a predetermined goal. The incapacity to forecast the consequences of actions could be responsible, along with the neutralisation of affective life, for the incoherent and self-destructive behavior eventually encountered in prefrontal patients. 2) Working memory can still be considered as a part of the temporal function. Nevertheless its unifying role, and its implication in practically every other function related to personality deserves a separate account. The historical and prospective function, in particular, could not exist without the support of an "extended present", i.e. a working memory. The capacity to link successive, logically related events oriented towards the performance of a task, or the accomplishment of a goal, is essential to the development of personality. WM is not present in early childhood and this absence explains the non-permanence of hidden objects in the field of consciousness as internal representation, and therefore the incapacity to perform delayed tasks. This capacity progressively develops during infancy, and probably supports the unfolding of logical reasoning. This logical function is a prototypical illustration of the characteristic of WM defined as both maintenance and manipulation of information over an extended period of time. Classically, Pc is endowed with WM capacity, but so also are temporal and plausibly other cortical or subcortical structures. Our contention is that Hs also partakes of an automatic aspect of WM, even though delay neurons have not been recorded there, as in different cortices. We have proposed at least three subsidiary mechanisms that could support this function at Hs level.

178

Chapter 4

3) Emotions and motivations arc another important facet of personality. One could hardly contend that a robot is a person even though by its previous experience it can have a semblance of history. Emotions could be viewed as resulting from sublimation, but certainly not a suppression, of basic drives including the instinct for survival and even the instinct of death. Motivation could result from an integration and a trade-off between the need for drive satisfaction, emotions, and social constraints. Motivation is usually dependent on the degree of satisfaction of these needs. Hero also limbic system and Pc act conjointly. Amygdala is as important for emotional lifo as hippocampus proper is important for correlational and WM functions. Similarly, orbitofrontal and meAial Pc are essential for the integration of drives, emotions and motivations while more cognitive information processed in dorso-lateral Pc. The suppression of any type of affr162 colour, positive or negative, in the life of severely damaged prefrontal patients, as after lobotomy, induces a disengagement from real life. This underlies the importance of the integrative function of Pc, in particular, between cognition and emotion. Those different functions are not compartmentalized. In particular emotional charge of events, as previously mentioned, modulates the process of memory consolidation in the hippocampo-cortical system. The cooperation of these different functions is perhaps best perceived in the mechanism of attainment of goals. 4) Attainment of goals can be considered as the uttermost expression of the cooperation between Pc and limbic system. This function presents sensory aspects which consist of recognition of goals and evaluation of the outcomes of action, and a motor aspect made up of the setting and execution of motor programs. In the classical learning theory, such as proposed by Skinner (1953), the necessary chaining of sequences of sensory-motor events results from associative (or operant) conditioning of a neutral stimulus by a reinforcer. Cascades of secondary, and higher order, conditioning could account for linking sequences of events together. Obviously, this process can be and has been accounted for without the extensive implication of prefrontal cortex as in our model (Gray ct al., 1991). The main structures concerned are hippocampus, amygdala and basal ganglia. They certainly correspond to a kind of automatic operation mode for the attainment of goals. Nevertheless, several problems arise if the basic components of sequences, plans, or chained actions remain limited to stimulus-response conditional associations. In particular, latent learning (Tolman, 1948) does not obey any clearly defined drive satisfaction, motivation or goal attainment. The conditioning process seems to work correctly for simple sequences of actions. But, taking

J.P. Banquet et al.

179

into account simply the present states in the recognition or reproduction of long sequences of actions leads to a combinatorial explosion of possible paths, which can only be avoided by an active representation of more than one pair of events. Furthermore, the strength of secondary or higher order reinforcers seems to sharply decrease with the distance to the unconditional stimulus. Sigmficant progress has been achieved by the identification of latent learning useful to build cognitive maps, even if these maps are used independently of any prespecified goal. We think that the definition of goals based on the satisfaction of basic (or not so basic) needs conjointly with the learning of more or less complex maps is a further step required to account for complex behaviors. The possibility for diffusion, and in particular retropropagation of goals allows the discovery of solutions that have never been experienced during learning, and thus are created from new by the system. This is an actual illustration of creativity. This functioning mode requires a supplementary degree of freedom in the system, independent from both sensory and motor processing, but still bridging the two systems. This extra degree of freedom is provided by Pc. The efficacy of an algorithmic version of the model for the parsimonious solution of several problems of robotic learning and navigation either in free space or maze constraint does not automatically deliver a certificate of biological plausibility for the system. Nevertheless, the fact that this efficacy has been obtained thanks to a stringent taking into account of essential neurobiological constraints, makes us confident that the model is oriented in a relevant direction. Conclusion

Two different forms of memory, "active" and "dormant", supported respectively by post synaptic potentials (PSPs) and synaptic potentiation are present everywhere in the brain. The interplay between the two forms of memory and in particular the transition and/or the modulation of one form by the other are at the bases of the different processing modes and memory capacities of the brain. Variations in the implementation modalities and in the ranges of these two types of memory along with variation of connectivity give functional specificities to the different systems. This is specially true for Hs and Pc. Cortical processing depends essentially on two memory registers, STM and permanent LTM. The transition from the long-term to the shortterm seems to be direct and normally encounters few problems. Plausibly, at a gross level of analysis the anatomical substrates are topographically identical (STM representing active forms of LTM). Nevertheless, at a fine

180

Chapter 4

grainexi level, the neurophysiological supporting mechanisms must be necessarily different as previously mcntionexi, involving respectively either electrical or durable structural-chemical changes. Conversely, the transition from STM to LTM store follows a more intricate path, probably for the sake of optimizing the amount of information stored, but also for securing the storage of unique events which built up the unique history of each living being. Between these two extreme ranges (STM and LTM), only minor variations from primary to associative areas can be recorded at the sole cortical level, with a tendency to an increase in the temporal range of memory with the increasing level of complexity in processing performed by these areas (Lii ~ al., 1992). Prefrontal and temporal cortices arc endowed with delay neurons that can bridge a gap between two sensory or sensori-motor events. Nevertheless, the tg~nporal range in the usual experimental tests of this property remains largely in the domain attributed to STM, i.e. less than 30 SCCS.

The specificity and vantage point of Hs concerns both topographical and temporal facets. The topographical aspect of Hs specificity as a unique compact site of input convergence and output divergence has been extensively emphasized. It has been credited with the correlational function of Hs which implies some loss of the cortical topology. This functional characteristic is corrected and complemented by a loose topological correspondence between cortical and hippoeampal system in the longitudinal direction. This loose correspondence could be transformed into a dynamic learning-dependent precise mapping between hippocampal and cortical neuronal populations in order to implement the topologically specific consolidation function. This function could be implemented thanks to the fast-transient learning capacities present both within the hippocampus itself, and also at the interfaces between cortex and hippoeampus. The emphasis placed on the spatial aspects of Hs function was detrimental to the exploration of the no less important temporal function. This function results from the capacity of Hs to interact very flexibly with a whole spectrum of registers from the short-term to the longterm, and also possibly to be detached from permanent LTM. The unique characteristic of Hs would be the conjunction of this array of registers with a wide variety of loops of various sizes providing for an easy transition between dormant-inactive and re-activated forms of memory. Beyond these range differences between cortex and Hs memory registers, some more subtle differences could exist in the implementation modalities of active memory. Extensive research has been conducted on delay cells in Pe or temporal cortex, as a support for WM. Indeed, this type of activity can bridge

J.P. Banquet et aL

181

the gap between two sensory or sensory-motor events. Up to now, delay cells have not been located at the level of the Hs. Nevertheless, the equivalent function in Hs could be performed by different mechanisms subserving slightly different functions. First, loop iterative activation could operate the punctual reenactment of recorded patterns of activation either during information processing in WM, or during the more lengthy process of LTM consolidation. Second, spectral timing as performed in our model by DG could also operate the function equivalent to that of the cortical delay neurons. This function consists of maintaining significant information in an active state, while waiting for correlation with a new significant event. This process creates the chaining of basic components of the sequence. The Hs functional specificity would be in multimodal fusion and correlations. Finally, event locked and modulated theta activity could constitute, at least for some species, a basic mechanism for the maintenance of a pattern in an active state, thus making possible a crosS correlation with forthcoming significant patterns. These types of complementary "hardware" constraints in the implementation of active memory and in the range of "dormant" registers determine the type of cooperation established between the two structures Hs and cortex. Further complementarity results from the direct contact of the cortex with environment, favoring externally triggered activation. Conversely, Hs is the only brain structure so easily prone to autoactivation. That property leads, in the pathological domain, to seizure activity. The specific import of Pc to this processing chain seems to result from its unique position at the top of the hierarchy of sensory-motor and motivational streams of information (Figure 11). Both, its independence from and its close contact with multisensory and complex motor representations or codes provides the entire system with an extra degree of freedom. This feature gives to the brain the capacity for: Recording and simulating both sensory and motor sequences independently of their actual implementation, in relation with planning and adaptation; Motivated hierarchical selection of goals and subgoals, and goal attainment; Finally, invention and creativity. These properties can be considered as the highest expression of all these capacities. -

-

-

182

Chapter 4 References

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" Advances in research and theory (Vol. 2, pp. 89-185). New York: Academic Press. Baddeley, A. D. (1986). Working memory. Oxford: Oxford University Press. Baddeley, A. D. (1995). Working memory. In M. S. Gazzaniga (Ed.), The cognitive neurosciences (pp. 755-764). Cambridge, MA: MIT Press. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. A. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 8, pp. 47-90). New York: Academic Press. Baddeley, A. D., & Warrington, E. K. (1970). Amnesia and the distinction between long-and short-term memory. Journal of Verbal Learning and Verbal Behavior, 9, 176-189. Banquet, J. P. (1983). Inter and intra-hemispheric relationships of the EEG during sleep in m a n . Electroencephalography and Clinical Neurophysiology, .55, 51-59. Banquet, J. P., & Grossberg, S. (1987). Probing cognitive processes through the structure of event-related potentials during learning: An experimental and theoretical analysis. Applied Optics, 26, 4931-4946. Banquet, J. P., Renault, B., & Les~vre, N. (1981). Effect of task and stimulus probability on evoked potentials. Biological Psychology, 13, 203-214. Banquet, J. P., Smith, M. J., & Renault, B. (1990). Bottom-up versus topdown: An alternative to the automatic attended dilemma? Behavioral and Brain Sciences, 13, 233-234. Banquet, J. P., & Contreras-Vidal, J. L. (1992a). Temporal order, timing, and probability context effects on pattern recognition and categorization. In I. Aleksander & J. Taylor (Eds.), Artificial neural networks 2 (pp. 1885-1890). Amsterdam: Elsevier. Banquet, J. P., & Contreras-Vidal, J. L. (1992b). An integrated neural network-event related potentials model of temporal and probability context effects on categorization. In Proceedings of the International Joint Conference on Neural Networks (pp. 541-546). Hillsdale, NJ: Lawrence Erlbaum. Banquet, J. P., & Contreras-Vidal, J. L. (1993a). Learning temporal contexts and priming preparation modes for pattern recognition. In Proceedings of the International World Conference on Neural Networks (pp. 126131). Hillsdale, NJ" Lawrence Erlbaum.

JP. Banquet et al.

183

Banquet, J. P., & Contreras-Vidal, J. L. (1993b). Spectral timing and integration of multimodal systemic processes. In J. Taylor (Ed.), Artificial neural networks (pp. 350-354). Amsterdam: Elsevier. Banquet, J. P., & Contreras-Vidal, J. L. (1994). Medium and long-term memory in context processing: A network model of cortex-hippocampus relations. Proceedings of the International Conference on Neural Networks, 4, 647-654. Banquet, J. P., Contreras-Vidal, JL., Gaussier, P., Gissler, A., & Bumod, Y. (1997). The cortico-hippocampal system as a multirange temporal processor: A neural model. In R. Park & D. Levin (Eds.), Fundamentals of neural network modelling for neuropsychologists. Boston: MIT Press. Bapi, R. S., & Levine, D. S. (1990). Networks modelling the involvement of the frontal lobes in learning and performance of flexible movement sequences. In Proceedings of Twelth Annual Conference of the Cognitive Science Society (pp. 915-922). Bapi, R. S., & Levine, D. S. (1994). Modelling the role of the frontal lobes in sequential task performance. I: Basic structure and primacy effect. Neural Networks, 7, 1167-1180. Baylis, G. C., & Rolls, E. T. (1987). Responses of neurons in the inferior temporal cortex in short-term and serial recognition memory tasks. Experimental Brain Research, 6.5, 614-622. Berger, T. W., & Thompson, R. F. (1978). Neuronal plasticity in the limbic system during classical conditioning of the rabbit nictitating membrane response. I: The hippocampus. Brain Research, 14.5, 323-346. Br6bion, J. (1994). Memoire de travail, comprehension de texte et vieillissement. Unpublished doctoral dissertation, University Ren6 Descarte, Paris. Brown, J. (1958). Some tests of the decay theory of immediate memory. Quarterly Journal of Experimental Psychology, 1O, 12-21. Bullock, D., Fiala, J. C., & Grossberg, S. (1994). A neural model of timed response learning in the cerebellum. Neural Networks, 7, 1101-1114. Burgess, N., Reece, M., & O'Keefe, I. (1994). A model of hippocampal function. Neural Networks, 7, 1065-1081. Buzsaki, G. (1988). Polysynaptic long-term potentiation: A physiological role of the perforant path - CA3/CA1 pyramidal cell synapse. Brain Research, 4.55, 192-195. Buzsaki, G. (1989). Two-stage model of memory trace formation: A role for 'noisy' brain states. Neuroscience, 31, 551-570.

184

Chapter 4

Cave, C. B., & Squire, L. R. (1992). Intact and long-lasting visual object priming in amnesic patients. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 509-520. Cressant A., Muller R., & Poucet B. (1997). Failure of centrally placed objects to control the firing fields of hippocampal place cells. Journal of Neuroscience, 17. Darke, S. (1988). Anxiety and working memory capacity. Cognition and Emotion, 2, 145-154. Dehaene, S., Changeux, J. P., & Nadal, J. P. (1987). Neural networks that learn temporal sequences by selection. Proceedings of the Natwnal Academy of Science of the USA, 84, 2727-2731. Denham, M. J., & Boitano, J. (1996)A model of the interaction between prefrontal cortex, septum and the hippocampal system in the learning and recall of goal-directed sensory-motor behaviours. Technical Report NRG-96-01, University of Plymouth School of Computing. Drachman, D. A., & Arbib, M. (1966). Memory and the hippocampal complex. Archives of Neurology, 15, 52-61. Eichenbaum, H., Otto, T., & Cohen, N. J. (1994). Two functional components of the hippocampal memory system. Behavioral and Brain Science, 17, 449-518. Eysenck, M. W. (1982) Attention and arousal: Cognition and performance. New York: Springer-Verlag. Funahashi, S., Bruce, C., & Goldman-Rakic, P. W. (1989). Mnemonic coding of visual space by neurons in the monkey's dorsolateral prefrontal cortex revealed by an oculomotor delayed-response task. Journal of Neurophysiology, 61, 331-349. Fuster, J. M., & Alexander, G. E. (1971). Neuron activity relatext to shortterm memory. Science, 173, 652-654. Fuster, J. M. (1980). The prefrontal cortex: Anatomy, physiology and neuropsychology of the frontal lobe. New York: Raven Press. Fuster, J. M. (1995). Memory in the cerebral cortex. Cambridge, MA: MIT Press. Fuster, J. M., & Jervey, J. P. (1981). Inferotemporal neurons distinguish and retain behaviorally relevant features of visual stimuli. Science, 212, 5255. Gaussier, P., & Zrehen, S. (1994a). Complex architectures for autonomous agents. In P. Gaussier & J. D. Nicoud (Eds.), PerAc (pp. 278-290). Lausanne: IEEE Press.

J.P. Banquet et al.

185

Gaussier, P., & Zrehen, S. (1994b). Navigating with an animal brain : A neural network for landmark identification and navigation. In Proceedings of Intelligent Vehicles (pp. 399-404), Paris. Gaussier, P., & Zmhen, S. (1995). PerAc: A neural architecture to control artificial animals. Robotics and Autonomous Systems, 16, 291-320. Gaussier, P., Joulain, C., Revel, A., & Banquet, J. P. (1996). Are shaping techniques the correct answer for the control of an autonomous robot ? In Proceedings of UKACC International Conference on Control '96, IEEE. Gaussier, P., Joulain, C., Revel, A. Zrehen, S, & Banquet, J. P. (1997a). Autonomous robot learning: What can we take for free? To appear in Proceedings of IEEE conference on ISIE, IEEE. Gaussier, P., Joulain, C., Zrehen, S. Banquet, J. P, & Revel, A. (1997b). Visual navigation in an open environment without map. To appear in Proceedings of IROS Conference. Glanzer, M., & Amitz, A. R. (1966). Two storage mechanisms in free recall. Journal of Verbal Learning and Verbal Behavior, 5, 351-360. Goldman-Rakic, P. W. (1988). Topography of cognition: Parallel distributed networks in primate association cortex. Annual Review of Neuroscience, 11, 137-156. Goldman-Rakic, P. W. (1994). Working memory dysfunction in schizophrenia. Journal of Neuropsychmtry and Clinical Neuroscience, 6, 348-357. Granger, R., Whitson, J., Larson, J., & Lynch, G. (1994). Non-Hebbian properties of long-term potentiation enable high-capacity encoding of temporal sequences. Proceedings of the Nanonal Academy of Science of the USA, 91, 10104-10108. Gray, C. M., KOnig, P., Engel, A. K., & Singer, W. (1989). Oscillatory responses in cat visual cortex exhibit intercolumnar synchronization which reflect global stimulus properties. Nature, 338, 334-337. Gray, J. A., Feldon, J., Rawlins, J. N. P., Hemsley, D. R., & Smith, A. D. (1991). The neuropsychology of schizophrenia. Behavioral and Brain. Sciences, 14, 1-20. Grossberg, S. (1976a). Adaptive pattern classification and universal recoding. I: Parallel development and coding of neural feature detectors. Biological Cybernetics, 23, 121-134. Grossberg, S. (1976b). Adaptive pattern classification and universal recoding. II: Feed-back, expectation, olfaction, and illusions. Biological Cybernetics, 23, 187-202.

186

Chapter 4

Grossberg, S. (1978). A theory of human memory: Self-organization and performance of sensory-motor codes, maps, and plans. In R. Rosen & F. Snell (Eds.), Progress in theoreacal biology (Vol. 5, pp. 233-374). New York: Academic Press. Grossberg, S., & MerriU, J. (1992). A neural network model of adaptively timed reinforcement learning and hippocampal dynamics. Cognitive Brain Research, 1, 3-38. Guigon, E., Dorizzi, B., Bumod, Y., & Schultz, W. (1995). Neural correlates of learning in the prefrontal cortex of the monkey: A predictive model. Cerebral Cortex, 2, 135-147. Halgrcn, E., Squires, N. K., Wilson, C. L. Rohrbaugh, J. W., Babb, T. L., & CrandaU, P. H. (1980). Endogenous potentials generated in the human hippocampal formation and amygdala by infrequent events. Science, 210, 803-805. Hasselmo, M. E., & SchneU, E. (1994). Laminar selectivity of the cholinergic suppression of synaptic transmission in rat hippocampal region CA l: Computational modeling and brain slice physiology. Journal of Neuroscience, 14, 3898-3914. Horel, J. A. (1994). The localization of general memory functions. Behavioral and Brain Sciences, 17, 482. Howarth, E., & Eysenck, H. J. (1968). Extraversion, arousal, and pairedassociate recall. Journal of Experimental Research in Personality, 3, ll4-116. Ivry, R. B., & Keele, S. W. (1989). Timing functions of the cerebellum. Journal of Cognitive Neurosciences, 1, 134-150. James, W. (1890). The principles of psychology. New York: Holt. Johnson, R., Jr. (1995). On the neural generators of the P300: Evidence from temporal lobeetomy patients. Electroencephalography and Clinical Neurophysiology, 44 (Supplement), 110-129. Johnson, R., Jr., & Donchin E. (1982). Sequential expectancies and decision making in a changing environment: An electrophysiological approach, Psychophysiology, 19, 183-199. Jones, R. S. G. (1993). Entorhinal-hippocampal connections: A speculative view of their function. TINS, 16, 58-64. Jonides, J., Smith, E. E., Koeppe, R. A., Awh, E., Minoshima, S., & Mintun, M. A. (1993). Spatial working memory in humans as revealed by PET. Nature, 363, 623-625.

J.P. Banquet et al.

187

Levine, D. S., & Park, R. (1992). Frontal lesion effects on verbal fluency in a network model. Proceedings of the International Joint Conference on Neural Networks, 2, 39-44. Lii, Z. L., Williamson, S. J., & Kaufman, L. (1992). Human auditory primary and association cortex have differing lifetimes for activation traces. Brain Research, 572, 236-241. Matthews, G. (1992). Extraversion. In A. P. Smith & D. M. Jones (Eds.), Handbook of human performance. Vol. 3: State and trait (pp. 95-126). London: Academic Press. McCarthy, G., Blamire, A. M., Puce, A., Nobre, A. C. Bloch, G., Hyder, F., Goldman-Rakic, P. W., & Shulman, R. (1994). Functional magnetic imaging of human prefrontal cortex activation during a spatial working memory task. Proceedings of the National Academy of Science of the USA, 91, 8690-8694. Miller, E. K., Li, L., & Desimone, R. (1993). Activity of neurons in anterior inferior temporal cortex during a short-term memory task. Journal of Neuroscience, 13, 1460-1478. Miller, E. K., & Desimone, R. (1994). Parallel neuronal mechanisms for short-term memory. Science, 263, 520-522. Milner, B. (1966). Amnesia following operation on the temporal lobes. In C. W. M. Whitty & O. L. Zangwill (Eds.), Amnesia. London: Butterworths. Milner, B., Corkin, S., & Teuber, H.-L. (1968). Further analysis of the hippocampal amnesia syndrome: 14-year follow-up study of H. M. Neuropsychologia, 6, 215-234. Milner, B. (1982). Some cognitive effects of frontal lobe lesions in man. Philosophical Transactions of the Royal Society, London, B, 298, 211226. Miyashita, Y., & Chang, H. S. (1988). Neuronal correlate of pictorial shortterm memory in the primate temporal cortex. Nature, 331, 68-70. Niki, H. (1974). Differential activity of prefrontal units during right and left delayed response trials. Brain Research, 70, 346-349. N~i/it~en, R. (1982). Processing Negativity: An evoked-potential reflection of selective attention. Psychological Bulletin, 92, 605-640. O'Keefe, J., & Nadel, L. (1978). The hippocampus as a cognitive map. Oxford: Oxford University Press. Olton, D. S., Becker, J. T., & Handelmann, G. E. (1979). Hippocampus, space and memory. Behavioral and Brain Science, 2, 313-365.

188

Chapter 4

Owen, A. M., Morns, R. G., Sahakian, B. J., Polkey, C. E., & Robbins, T. W. (1996). Double dissociations of memory and executive functions in working memory tasks following lobe excisions, temporal lobe excisions or amygdalo-hippocampectomy in man. Brain, 119, 15971615. Paulescu, E., Frith, C. D., & Frankoviak, R. S. J. (1993). The neural correlates of the verbal component of working memory. Nature, 362, 342-345. Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58, 193-198. 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 of the USA, 90, 873-877. Rolls, E. T., Baylis, G. C., Hasselmo, M. E., & Nalwa, V. (1989). The effect of learning on the face selective responses of neurons in the cortex in the superior temporal sulcus in the monkey. Experimental Brain Research, 76, 153-164. Rolls, E. T., & O'Mara, S. M., (1995) View-responsive neurons in the primate hippoeampal complex. Hippoeampus, 5, 409-424. Ruchkin, D. S., Sutton, S., Mahaffey, D., & Glaser, J. (1986). Terminal CNV in the absence of a motor response. Electroencephalography and Clinical Neurophysiology, 63, 445-463. Ruehkin, D. S., Johnson, R., Jr., Canoune, H., & Ritter, W. (1991). Eventrelated potentials during arithmetic and mental rotation. Electroencephalography and Clinical Neurophysiology, 79, 473-487. Ruehkin, D. S., Johnson, R., Jr., Grafman, J., Canoune, H., & Ritter, W. (1992). Distinctions and similarities among working memory processes: An event-related potentials study. Cognitive Brain Research, 1, 53-66. Sehacter, D. L., Chu, C. -Y., & Oehsner, K. N. (1993). Implicit memory: A selective review. Annual Review ofNeuroscience, 16, 159-82. Seoville, W. B., & MiMer, B. (1957). Loss of recent memory after bilateral hippoeampal lesions. Journal of Neurology, Neurosurgery, and Neuropsyehiatry, 20, 11-21. Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic information processing. II: Perceptual learning, automatic attending and a general theory. Psychological Review, 84, 127-90.

J.P. Banquet et al.

189

Sidman, M., Stoddard, L. T., & Mohr, J. P. (1968). Some additional quantitative observations of immediate memory in a patient with bilateral hippocampal lesion. Neuropsychologia, 6, 245-54. Singer, W. (1983). Neuronal activity as a shaping factor in the selforganization of neuron assemblies. In E. Basar, H. Flohr, H. Haken, & A. J. Mandell (Eds.), Synergetics of the brain. New York: SpringerVerlag. Skinner, F. (1953). Science and human behavior. New York: McMillan. Solomon, P. R. (1980). A time and a place for everything? Temporal processing views of hippocampal function with special reference to attention. Physiological Psychology, 8, 254-61. Squire, L. R., Ojemann, J. G., Miezin, F. M., Petersen, S. E., Vdeen, T. O., & Raichle, M. E. (1992). Activation of the hippocampus in normal humans: A functional anatomical study of memory. Proceedings of the National Academy of Science of the USA, 89, 1837-41. Tolman, E. C. (1948). Cognitive maps in rats and men. The Psychological Review, 55, 189-208 Zrehen., S. (1995). Elements of brain design for autonomous agents. Unpublished PhD thesis, Swiss Federal Institute of Technology, Lausanne. Zrehen, S., & Gaussier P. (1997). A neural architecture for motivated landmark-based navigation. ETIS internal report (submitted for publication). Author Note

This research was supported by INSERM, NATO and DGA/DRET Grant # 911470/A000/DRET/DS/DR.

This Page Intentionally Left Blank

P A R T II PERSPECTIVES FROM EMOTION RESEARCH

This Page Intentionally Left Blank

Cognitive Science Perspectives on Personality and Emotion - G. Matthews (Editor) 1997 Elsevier Science B.V. CHAPTER 5

Affective Influence in Perception: Some Implications of the Amplification Model Shinobu Ki tayama

Will affectively charged stimuli be perceived any differently from affectively neutral ones? Will affect inherent in a focal stimulus disrupt perceptual processing? Or will it facilitate the latter so that affective stimuli stand out in the perceptual field? After years of waxing and waning (e.g., Dixon, 1980; Erdelyi, 1974, 1985), the influence of stimulus affect on immediate perception remains as a topic of considerable significance. Mechanisms underlying immediate conscious perception are both logically (e.g., Helmholtz, 1884) and empirically (e.g., Marcel, 1983a) preconscious. Thus, if it can be shown that the immediate perception of a stimulus is indeext influenced by affect inherent in the stimulus itself, we will have identified a window through which to observe what Kihlstrom (1990) has called the psychological unconscious. Championed by Freud and his successors (e.g., Freud, 1895/i966), the functional structure of the unconscious, especially the one involving affect, has turned out to be one of the most formidable problems in psychology, often evading scientific scrutiny. However, with rigorous experimental methodologies and theoretical tools now available at hand, recent investigations on detectionless processing (e.g., Bargh, Bond, Lombardi, & Tota, 1986; Carr & Dagenbach, 1990; Greenwald, Klinger, & Lui, 1989; Marcel, 1983a,b; Niedenthal, 1990; Shevrin, 1990), automatic processing (e.g., Uleman & Bargh, 1989; Shiffrin & Schneider, 1977), and implicit memory (e.g., Schacter, 1989) have taken significant steps toward more comprehensive and accurate understanding of the unconscious. And a new theoretical framework has begun to emerge (e.g., Erdelyi, 1985; Lewicki, 1986; Kihlstrom, 1990; Marcel, 1983b; Rumelhart, 1989; Zajonc, 1980). The present paper seeks to contribute to this literature. We will examine whether and how the perceptibility of a faintly shown stimulus can vary with the affective significance of the stimulus itself. The goal is to identify distinctly affective phenomena in a perceptual identification task, and integrate them with current theories of cognition, affect, and attention, thereby laying a solid foundation for understanding other forms of

194

Chapter 5

"hot cognitions" (Abelson, 1963). Should perceptibility depend on affective significance, that would suggest that affect can be preconsciously elicited by an impinging stimulus and, moreover, that the elicited affect influences subsequent processing required to develop a conscious percept. Accordingly, the present work has the potential of revealing the nature of a preconscious interaction between affect and cognition. The current approach emphasizes a non-associative, energizing consequence of affect (of. Osgood, 1962) and analyzes how stimulus affect influences the perception of the stimulus itself. Specifically, it will be proposed that affect evoked through preattentive processing amplifies attentive processing, thereby either enhancing or impairing the emerging conscious percept. This work therefore will supplement a presently dominant, largely associative approach to affect (e.g., Bargh et al., 1986; Bower, 1981; Fazio, Sanbonmatsu, Powell, & Kardes, 1986; Greenwald et al., 1989; Isen, Shalker, Clark, & Karp, 1978; Johnson & Tversky, 1984; Lang, 1084; Niedenthal, 1990; Zajone, Murphy, & Ingelhart, 1989), which has proved powerful in analyzing how affect of one stimulus (prime) can bias the perception of another (target). It is typically assumed in this literature that the activation of affective information, caused by the prime, can spread to related information within a network of associative memory, thus biasing the perception of the target. A historicalperspective: The "New Look" and its aftermath The general issue of affect-cognition interaction in perception can be traced back to the literature of "New Look" in perception in the 1950s (Bruner, 1957). It was then proposed that perception depends not only on exogenous factors, but also on endogenous factors including perceptual set, expectation, motivation, personality, and affect (see e.g., Allport, 1955, for a review). In a pioneering experiment on affect and perception, MeGinnies (1949) examined the perception of affcctivdy charged words. He briefly flashed either a taboo word or a neutral word, and found recognition threshold to be considerably higher for the taboo word than for the neutral word. He maintained that this resulted from perceptual processes; affect ("anxiety") evoked by a taboo word recruited the process of psychological defense, which blocked further perceptual processing, thus diminishing the conscious percept (see also Blum, 1954). Other studies observed that affect sometimes enhanced perception (e.g., Postman, Bronson, & Gropper, 1953).

S. Kitayama

195

The proponents of perceptual defense then suggested the operation of perceptual vigilance. Critics quickly pointed out an apparent paradox involved in McGinnies's assertion that one can feel "anxiety" without knowing the identity of the stimulus. They maintained that the finding could be explained most parsimoniously by post-perceptual response biases such as reluctance or readiness to report a taboo word (Eriksen, 1963; Goldiamond, 1958: see Erdelyi, 1974; Dixon, 1980 for reviews). There are some methods, however, that allow one to examine perceptual accuracy independent of post-perceptual response bias. For instance, subjects may be asked to choose the item shown from a pair of equivalently valenced words. Research employing this and other similar procedures has demonstrated that the affective tone of a stimulus does influence accuracy in perception (e.g., Bootzin & Natsoulas, 1965; Dorfman, 1967). Thus, McGinnies was correct in this regard. Further, to be reviewed below, recent cognitive research has strongly suggested that conscious perception is the end product of a number of preconscious operations. Hence, McGinnies's notion that affect can be induced by an impinging stimulus before the stimulus is consciously identified is no longer considered paradoxical (Erdelyi, 1974). Nevertheless, his theory that processes of defense/vigilance mediate the effects of affect in perception has faced serious challenges. Neither he nor his successors articulated the mechanisms of defense or vigilance. Hence, no prediction about the direction of the influence of affect is possible. Further, evidence suggests that affect can influence the perceptibility of a stimulus whether its valence is positive or negative (Broadbent & Gregory, 1967; Kitayama, 1990, 1991). In retrospect, then, it would seem that McGinnies was correct in that the effect he observed was, at least in part, perceptual. However, his hypothesis of defense/vigilance as an underlying mechanism is increasingly suspect.

The present approach The current paper presents a model of affect-cognition interaction designed to account for perceptual influence of affect. We reconsider this old problem, traditionally studied under the rubric of defense and vigilance, from a new perspective afforded by a number of theoretical and methodological innovations accomplished in the interim. Informed by current theories of cognition, affect, and attention, the model hypothesizes that affect induced through preattentive processing of an impinging stimulus amplifies

196

Chapter 5

subsequent attentive processing. We will review past studies on preattentive processing, attention, and affect, which together form the empirical and theoretical basis for the proposed model. It will be shown that the model provides a coherent account for an anomalous pattern of past findings in the defense/vigilance literature. Further, the model has guided more recent empirical investigations on the topic. Evidence for the model will be reviewed, and two new experiments will be reported. Finally, broader implications of the proposed model will be discussed and directions for future research explored. The present attempt to test implications of the model strategically focuses on just one form of perception: word perception. Studying this relatively simple case should make it possible to exert precise experimental control and, thus, to test the model in a more rigorous fashion. Further, in the domain of word perception, there is considerable overlap between research on affect and research on cognition. Much of past research on affect and perception, dating back to McGinnies's original contribution, was done with words as stimuli, and word perception has been extensively studied in current cognitive psychology (see e.g., Posner, 1989). The Amplification Model of Affect-Cognition Interaction A model

A model of affect-cognition interaction in early perceptual processing (Kitayama, 1990, 1991; Kitayama & Howard, 1994) is illustrated in Figure 1. The basic tenet of the model is that affect induceA through preattentive processing of an impinging stimulus amplifies subsequent attentive processing, thereby either enhancing or impairing the conscious percept of the stimulus. Unlike the defense/vigilance hypothesis, this model assumes that an influence of affect in perception results from interaction among three component processes commonly implicateA in ordinary processes of perception, i.e., preattentive processing, attentive processing, and affect. Preattentive processing. According to current cognitive theories of perceptual processing of lexieal materials such as words (e.g., McClelland & Rumelhart, 1981; Posner, 1978) and certain graphic stimuli such as faces (Bauer, 1984; Damasio, Damasio, & Van Hoesen, 1982; Tranel & Damasio, 1985), an impinging stimulus is initially processed automatically without any involvement of attention. Through preattentive processing, the graphic and possibly the semantic perceptual codes that correspond to the stimulus can be

S. Kitayama

197

ii

Preattentive processing

,11111

i

)" i|1 ii iii

I Engagement of attention in a relevant

i

i

|l

Activation of affective circuits and subsequent amplification of attentive processing

Attentive

processing

Conscious

percept

Response (i.e., reading out relevant features, selecting responses, etc.)

Figure 1. A schematic illustration of the amplification model of affectcognition interaction in early perceptual processing. The model is composed of, as its major components, preattentive processing, attentive processing, and activation of affect and subsequent amplification of attentive processing, and response. These components are highlighted in bold squares.

activated before the conscious percept of the stimulus is developed. A number of recent studies with a semantic priming paradigm have shown that semantic information (and, by implication, graphic information as well) can be activated by a word that is pattern masked and thus made undcteetable in consciousness (e.g., Allport, 1977; Balota, 1983; Carr, McCaulcy, Spcrber, & Parmclee, 1982; Carr & Dagcnbach, 1990; Grccnwald ct al., 1989; Fowler, Wolford, Sladc, Tassinary, 1981, Marcel, 1983a, 1983b). This

198

Chapter 5

literature lends strong support to the hypothesis that attention, in the sense of a set of processing mechanisms selectively and serially applied to a confined sp~/tial or semantic region, is not necessary for the activation of meanings of an impinging lexieal stimulus, let alone the activation of its shape. The foregoing conclusion might seem at odds with cognitive research on vision, which has traditionally assumed that preattentive vision is very crude (Egeth, 1977; Neisser, 1967). For example, Sagi and colleagues (Braun & Sagi, 1990; Sagi & Julesz, 1985) have proposed that preattentive vision allows one to note that a field contains a discrepancy, but does not enable one to identify any object. Similarly, for Treisman (1988), preattentive vision is sufficient to identify separate features of a field (such as "green" or "square"), but not any object defined by a conjunction of more than two features (such as "green square"). However, the vision literature does not necessarily contradict the above evidence for the preattentive activation of shape and meaning of a perceptual object. The two lines of research typically use conspicuously different stimulus materials. On the one hand, the vision literature has focused primarily on simple and to a large extent, arbitrary graphic stimuli (e.g., colors, lines, simple geometric figures) that have no obvious, unique meanings. On the other hand, the studies attesting to the presence of preattentive activation of shape and meaning employ meaningful stimuli that are routinely encountered in everyday life, viz., mostly lexical materials such as words, but occasionally certain complex and realistic graphic materials such as faces. It goes without saying that some kind of preexisting processing structures such as the ones exemplified in connectionist networks are required for preattentive activation of shape or meaning to take place (McClelland & Rumelhart, 1981). These structures will develop gradually from everyday encounter with relevant stimuli (e.g., LaBerge & Samuels, 1973; Shiffrin & Schneider, 1977) although those for certain phylogenically significant stimuli such as faces may be hard-wired through evolution (Field, 1985). Thus, the extent of preattentive processing can vary from very crude (as in the case of arbitrary and/or meaningless stimuli for which no ready-made processing structure is available) to very thorough and sophisticated (as in the case of meaningful/lexical materials for which elaborate processing structures have been established and, thus are readily available). A series of studies by Shevrin and his colleagues provided some evidence (Shevrin & Fritzler, 1968; Shevrin, Smith, & Fritzler, 1971). They found that evoked potential to a subliminally shown picture is significantly more intense if the picture is meaningful than if it is meaningless. Further, the meaningfulness of the

S Kitayama

199

picture systematically influenced subsequent free associations as well, suggesting that semantic activation was actually caused by the subliminally shown picture if it was meaningful. All in all, then, once a meaningful stimulus commonly encountered in daily life such as a word is presented, it will be processed automatically, and corresponding perceptual codes are activated. Although this activation is preconscious and quite weak especially if the stimulus is impoverished, it has been shown to be sufficient for affect associated with the stimulus to be elicited covertly (Ohman, 1985). Perhaps, the activated codes summon affective circuits of the brain located in the limbic or subcortical regions. LeDoux (1987, 1989) has reviewed neuroanatomical evidence suggesting numerous neuronal connections between sensory processing areas of the brain and the limbic regions. Furthermore, several studies (e.g., Corteen & Wood, 1972; Lazarus & McCleary, 1951; Zajonc, 1962) have shown a reliable autonomic response to a subliminal affective stimulus. Although this literature has been criticized on methodological grounds (e.g., Merikle, 1982; Holender, 1986), more recent research with a strict criterion for awareness has also shown that affect can be elicited by undetectable stimuli (e.g., Dawson & Schell, 1982; Greenwald et al., 1989; Niedenthal, 1990; Tassinary et al., 1984; see also Kunst-Wilson & Zajonc, 1980). Attentive processing. It is reasonable, then, to postulate that affect elicited via preattentive processing in turn influences subsequent attentive processing, which is generally believed necessary for conscious perception of the stimulus (Neely, 1977; Posner & Snyder, 1975). Unlike preattentive processing, attentive processing is selective, limited solely to a perceptual code to which attention has been directed. Thus, once a relevant perceptual code has been automatically and preconsciously activated by an impinging stimulus, two operations need be performed. First, attention is shifted and directed to the relevant code and, second, once so directed, attention furthers the code's processing (cf. Posner, 1980). Through attentive processing, a more elaborate perceptual and, perhaps, semantic image of the stimulus is developed, which corresponds to the immediate conscious percept of the stimulus. Finally, the conscious percept may be scanned and its more specific features may be read out to control subsequent action (Allport, 1989). In order to grasp the proposed relationship between preattentive processing and attentive processing, a metaphor of attention as a spotlight is useful (Crick, 1984; LaBerge, 1983; Moser, 1988; Posner, 1980). According to this metaphor, preattentive processing activates the relevant one of numerous perceptual codes in long term memory. This activation itself,

200

Chapter 5

however, is not enough to produce conscious perception. For the latter to occur, an attentional spotlight must be deployed. The spotlight must first be shifted to the relevant code and then used to illuminate the code. This illuminating of the relevant information amounts to additional processing performed on the latter and, as such, is thought to enable conscious perception of the stimulus. In this formulation, attentive processing is distinct from preattentive processing in its selective nature. Whereas preattentive activation can occur simultaneously at multiple loci (e.g., letter, graphic, semantic codes that are valid as well as, especially when the stimulus is impoverished, those that are invalid; see below), attentive processing can be focused on only one of them. Another implication of the current formulation is that attention and consciousness are distinct even though there is substantial overlap between them. Generally, preattentive (nonselective) processing takes place without conscious awareness, whereas attentive (selective) processing is mostly conscious. However, attentive (selective) processing required to produce conscious awareness is necessarily preconscious. Amplification by affect. One widely postulated property of affect is arousal, or its ability to amplify a variety of psychological functions. In his pioneering work, Tomkins (1962, 1980) has proposed that various basic emotions such as joy and anger can be described in terms of differential patterns of amplification of a nervous system. Although Tomkins's analysis may no longer seem feasible, arousal or an intensity dimension of affect has been shown to be essential in defining a variety of everyday vocabularies of emotion and concepts in general, and suggested to be universal across cultures (Osgood, 1962; Russell, 1980). Another major dimension of affect identified in this literature is pleasantness. From the very beginning, it has been widely recognized that an amplifying property of affect can have a variety of consequences on psychological processes. Under the guise of drive, this assumption is central to a behavioral theory of learning proposed by Hull, Spence, and Taylor in the 1950s (e.g., Spence, 1956). It also is at the core of the Yerkes-Dodson law (Yerkes & Dodson, 1908), as well as its modem extensions by H. Eysenck (1967) to analysis of personality dimensions of extraversion/ introversion and impulsivity. It has also proved applicable to social facilitation (Zajone, 1965). More recently, Revelle, Humphreys, and their colleagues (Humphreys & Revelle, 1984; Revelle & Loftus, 1990; see also M. Eysenck, 1976) have elaborated on some specific consequences of arousal on different stages of memory processes.

S. Ki tayama

201

As has been pointed out by a number of researchers, the notion of generic, uni-dimensional arousal involving all aspects of the sympathetic nervous system and those of cortical processes seems too simplistic (e.g., Lacey & Lacey, 1968). Nevertheless, the hypothesis that affect amplifies some aspects of psychological processes remains both reasonable (Lindsley, 1951) and empirically viable (Stembach, 1968). And, as such, it has the potential of clarifying ways in which affect influences cognition. Extrapolating from this literature, it may be hypothesized that affect elicited through preattentive processing amplifies attentive processing. This simple hypothesis suggests both (i) the conditions in which affect associated with an impinging word is most likely to enhance the perception of the word itself and (ii) those in which the affect is most likely to impair the perception. Enhancement and impairment of perception by affect. First and most obvious, if attention has accurately been directed to a relevant perceptual code (i.e., the one corresponding to an impinging word), affect and ensuing amplification of attention should enhance the veridical perception of the impinging stimulus. In this case, affective stimuli will be more accurately perceived than neutral stimuli (affective enhancement). Suppose, however, that a stimulus is presented in an extremely impoverished manner, as is often the case in perceptual identification experiments. Under these conditions, the relevant perceptual code will not receive strong activation. As we have reviewed earlier, this weak activation seems sufficient to produce a degree of affect, thus amplifying subsequent attentive processing. Nevertheless, the weak activation will cause considerable difficulty in computing exactly which perceptual code corresponds to the impinging stimulus, especially because residual activations caused by past experience are likely to remain for many other irrelevant codes (e.g., Jacoby, 1983; Higgins & Bargh, 1987). Because of this difficulty in locating the relevant code, attention may be misdirected to an irrelevant code. Under these conditions, affect produced through preattentive processing will amplify attentive processing that has been directed, accidentally, to invalid perceptual information and, as a consequence, it will impair an emerging conscious percept. In this case, affective stimuli will be less accurately perceived than neutral stimuli (affective impairment). In terms of the spotlight metaphor introduced earlier, preattentive processing of an affective stimulus activates the corresponding code and, as a consequence, evokes associated affect, which in turn increases the illumination of the attentional spotlight. However, because the activation of the relevant code is weak, perhaps no stronger than residual activations

202

Chapter 5

remaining in irrelevant codes, the spotlight is likely to be locked on to one of the irrelevant codes, accidentally illuminating the latter and, thus, causing an impairment of valid perception: In sum, the present model (the amplification model hereafter) predicts that affective impairment should be most likely to occur when the presentation of a target word is extremely impoverished. It further implies that the impairment effect should disappear or even reverse itself once the difficulty in locating the relevant code is alleviated, that is, when the activation of the relevant code is increased relative to the activation of other irrelevant codes. Under these conditions, attention will be successfully directed to the relevant perceptual code and, as a consequence, affect and subsequent amplification of attentive processing should enhance the emerging percept. The general prediction tested, therefore, can be stated in terms of affective enhancement (higher accuracy for affective than for neutral stimuli within a given experimental condition) or affective impairment (lower accuracy for affective than for neutral stimuli within a given experimental condition): Any variable that increases the activation of a relevant perceptual code relative to the activation of other, irrelevant codes will increase the likelihood of affective enhancement and~or decrease the likelihood of affective impairment. Evaluation Criteria of the Amplification Model Two points must be made explicit before setting out to test implications of the amplification model. First, the model predicts that stimulus affect should influence perceptual accuracy independently of response bias either for or against reporting an affective stimulus. Earlier studies in the defense and vigilance literature were criticized largely because they used recognition threshold as a dependent variable. With this measure it is extremely difficult to separate perceptual accuracy from response bias (Eriksen, 1963; Goldiamond, 1958). As noted above, however, there are some methods, most notably forced choice between two affectively equivalent stimuli, that allow one to control for response bias (Natsoulas, 1965). In the following, we will draw primarily on those studies that have adequately controlled for response bias. Second, the engagement of attentive processing in a preattentively activated perceptual code is only one of several distinct operations that can

S. Kitayama

203

contribute to overall perceptual identification (see Figure 1). The model therefore assumes that any variable that can enhance attentional engagement may also either improve or impair other operations, and thus either increase or decrease the overall perceptual identification independently of stimulus

affect. To illustrate, consider stimulus complexity, which is likely to have opposing effects on preattentive processing and response selection. To begin with, the more complex a stimulus is, the greater the number of features it contains. Because each of these features will serve as an additional constraint in preattentive processing, as stimulus complexity increases, the corresponding perceptual code may be more unequivocally activated. According to the amplification model, under these conditions the percept will be more accurate for affective stimuli than for neutral stimuli (affective enhancement). In addition, however, once the percept has been developed, the respondent will subsequently have to make an overt response. Most of the studies to be reviewed or reported in the current paper examine accuracy in a forced choice, whereby the respondent supposedly scans and compares the percept with available alternatives. Because complex stimuli contain more features to be compared and matched in the choice, stimulus complexity should make response selection more difficult, thereby leading to poorer overall performance. In short, stimulus complexity is likely to increase the likelihood of affective enhancement, while simultaneously decreasing choice performance. Once these two effects of stimulus complexity are super-imposed on each other, one will observe performance for affectively neutral stimuli to decline with stimulus complexity. Relative to this base line defined by the neutral stimuli, performance for comparable affective stimuli should improve. Yet, this improvement due to stimulus affect may or may not compensate for the decline of overall performance due to choice difficulty. This means that performance for affective stimuli may or may not actually improve with stimulus complexity. The crucial prediction of the amplification model in this case then, is that a decline of performance as a function of stimulus complexity is less for affective stimuli than for neutral stimuli. In general, it is safe to assume that any variable that can enhance attentional engagement (e.g., stimulus complexity) may also either improve or impair other operations (e.g., response selection), and thus either increase or decrease overall performance independently of stimulus affect. Accordingly, the amplification model must be evaluated in terms of its ability to predict either affective enhancement or impairment, rather than its ability to predict

204

Chapter 5

an absolute increase or decrease of performance for affective or neutral stimuli with the manipulated variable. With these evaluation criteria at hand, we now turn to a review of extant studies pertinent to some predictions of the amplification model.

Empirical evidence Expectation and word frequency. Research on priming has amply demonstrated that when individuals are led to expect the identity of a target, the corresponding perceptual code receives extra activation (e.g., Higgins & Bargh, 1987; Neely, 1977; Posner & Snyder, 1975). It can be hypothesized, therefore, that in a perceptual identification task, the difficulty of locating a relevant perceptual code is relieved by a correct expectation about the target word. Thus, affective enhancement will be more likely and affective impairment less likely in the presence of a valid expectation than in its absence (Kitayama, 1990). Initial support for the prediction was uncovered in a review of the literature of perceptual defense and vigilance. Because most studies in this literature examined recognition threshold and failed to control for response bias, their status as evidence for the current analysis is uncertain (Eriksen, 1963; Goldiamond, 1958; see e.g., Dixon, 1980; Erdelyi, 1974, for reviews). Nevertheless, two experiments which manipulated expectation tended to support the amplification model (Freeman, 1954; Lacey, Lewinger, & Adamson, 1953). In these studies, when there was no expectation, recognition threshold was higher for affective words than for neutral words (affective impairment); but when an expectation about the identity of a target was provided, recognition threshold for affective words was lower than that for neutral ones (affective enhancement; see also Postman et al., 1953, for a similar result). Additional evidence for the present analysis can be found in more recent, methodologically more sophisticated studies that assess perceptual accuracy independently of response bias (either readiness or reluctance to report affective rather than neutral stimuli). Kitayama (1990) located nine such studies (see Table 1). In most of these studies (Bootzin & Natsoulas, 1965; Broadbent & Gregory, 1967; Dorfman, 1967; Dorfman, Grossberg, & Kroeker, 1965), the dependent variable was correct response rate, with appropriate adjustments made for response bias. Two additional studies used different methods to minimize response bias. Chapman and Feather (1972) examined the ability to detect (rather than identify) a novel graphic stimulus using a signal detection procedure. They assigned an affective tone to the

S. Kitayama

205

stimulus by classically conditioning an electric shock to it. Sales and Haber (1968) minimized response bias by having subjects report individual letters of a flashed word rather than the word itself, and analyzed the number of letters correctly reported.

Table 1. Comparison of nine experiments in which effects of response bias were minimized (adapted from Kitayama, 1990). Experiment

Expectation

Outcome

Chapman & Feather (1972) Dorfman (1967) Dorfman et al. ( 1965) Mathews & Wertheimer(1958) Minard (1965) Van Egeren (1968) Bootzin & Natsoulas (1965) Broadbent & Gregory(1967) . Sales & Harber (1968)

Clear Clear Clear Vague Vague Vague Absent Absent Absent

Enhancement Enhancement Enhancement Inconsistent1 Inconsistent2 No effect Impairment Impairment Impairment

1 Significant impairment effect was found for "high-hysteria" subjects, but no effect was obtained for "high-psychasthenia"subjects. 2 Significant impairment effect was found for males, whereas significant enhancement effect was found for females.

Among these nine studies, three obtained affective enhancement (Chapman & Heather, 1972; Dorfman, 1967; Dorfman et al., 1965). Interestingly, all the three studies inadvertently used a procedure that assured that the subjects had a clear expectation about the target stimulus. In two experiments by Dorfman subjects were shown the target word plus a nontarget word immediately before the target was actually flashed. They were told that one of the pre-target words would be flashed on that trial. Chapman and Feather (1972) had subjects keep in mind the target stimulus while seeing a visual display. Thus, both methods provided subjects with a clear expectation. Some other experiments implanted subjects with vague expectations by familiarizing them with experimental stimuli at the beginning of the session. In these studies there was no systematic pattern. Mathews and Wertheimer (1958) and Minard (1965) found the influence of affect to depend

206

Chapter 5

on certain individual difference variables, and Van Egeren (1968) failed to find any influence of affect on perceptual accuracy. The remaining did not use any of the above procedures and reported affective impairment (Bootzin & Natsoulas, 1965; Broadbent & Gregory, 1967; Sales & Haber, 1968). Kitayama (1990) subsequently conducted an experiment in which expectation was systematically manipulated. In this experiment word frequency was also varied. It was hypothesized that as word frequency increased, the valid perceptual code would be more strongly activated and, as a consequence, affective impairment would become less likely and affective enhancement more likely. Subjects were exposed to a 25 ms flash of a target word. They then chose the target word from a word pair. In half the trials, this word pair was given before the flash to create an expectation. Further, on some trials no target was presented although subjects were led to believe that it was actually shown. Analysis of the data from these trials revealed no response bias for or against reporting affective stimuli, so choice hit rate was used as a measure of perceptual accuracy. Consistent with the amplification model, both expectation and word frequency increased the likelihood of affective enhancement and decreased the likelihood of affective impairment. As can be seen in Figure 2, when words were low in frequency (10-50 occurrences per million) and an expectation was absent, affective words were identified significantly less accurately than neutral ones (affective impairment). This pattern, however, was reversed to show a reliable enhancement effect when high-frequency words (more than 100 occurrences per million) were examined and an expectation was present. Finally, the influence of affect in the remaining two conditions (high frequency/unexpected and low frequency/expected) was no greater than that in the former two conditions. The Kitayama (1990) study thus generally confirmed the predictions of the amplification model. Nevertheless, it was not totally conclusive. First, it tested only a small number of words (12 in total). Second, it found the predicted effect of expectation only for high-frequency words. There was no such effect for low-frequency words: as can be seen in Figure 2, affeetive impairment of evidently equal strength was observed regardless of expectation. Stimulus contrast. The amplification model states that affective impairment is most likely when the presentation of a target is extremely impoverished. Another recent set of experiments with a larger number of stimulus words (126 in total, ranging from 8 to 65 per million in frequency of occurrence) has provided support (Kitayama, 1991). In Study 1, a target was

S. Kitayama

207

0.68 0.66

"

0.64

--

['7

Affeetivq

1

Neutral

0.62 0.60

--

0.58 0.56 Im qD

0.54 0.52 0.50

i

.......

I

i

..0

Word Frequency/Expectation

Figure 2. Perceptual accuracy (hit rate) as a function of target affectivity, word frequency, and expectation (adapted from Kitayama, 1990). presented in dark gray (luminance = .72 if-L) in a black background (.67 ilL), so the contrast between the target and the background was extremely low. As predicted, a significant affective impairment effect was obtained - that is, the identification of affective words was less accurate than the identification of neutral words. In Study 2, however, the contrast was increased so that the target was shown in lighter gray (.75 if-L). Under the latter condition, there was no influence of affect. Exposure time. According to the amplification model, a relevant perceptual code needs to be located quite early in the processing before attention is directed. It then follows that effects of exposure times should depend crucially on the range in which they are manipulated. When relatively long exposure times are manipulated, it will be only late in the processing that these variations begin to increase the activation of the relevant code. Thus, under the conditions of extremely impoverished stimulus contrast, attention should be misdirected to an irrelevant code regardless of the exposure times. In Study 1 of Kitayama (1991) described above, three relatively long exposure times (100, 150, and 200 ms) were tested. As predicted, affective impairment of evidently equal magnitude was observed in all the three

208

Chapter 5

exposure time conditions. By contrast, when relatively short exposure times are tested, an increase in exposure times should increase the activation of a relevant code early in processing, thus mitigating the difficulty in locating the code. Thus, an increase in relatively short exposure times should result in a lesser likelihood of affective impairment and a greater likelihood of affective enhancement. Kitayama (1989) showed that affective impairment observed with a 25 ms exposure (Kitayama, 1990) could disappear with a 40 ms exposure. Word length. Virtually every past study in this area has examinexl only relatively short words (less than 6 letters long). However, current cognitive models of word recognition (e.g., McCleUand & Rumelhart, 1981) suggest that word length may systematically change the likelihood of affectivr enhancement and impairment. According to these models, the initial, preattentivr processing of a visual input process in parallel, leading to simultaneous activation of parts of the entire input. Currently, there is no consensus about exactly what defines functional parts of a word. Drawing on some prominent models of word recognition (e.g., McCleUand & Rumelhart, 1981), we assume here that "word-parts" correspond fairly closely to individual letters, although, for the purposes of the present argument, however, it makes little difference whether the units are letters or something else. Once individual letters have been activated, they in turn impose significant constraints on the likely identity of the input, permitting only a limited number of English words as reasonable candidates for the input. All else being equal, as word length increases, a greater number of letters should be activated and the letter-level information should more strongly constrain the word-level identity. To illustrate, imagine that the processing of a fourletter word successfully activated half of the constituent letters, say, "LxxE." There are several 4-letter candidate words that meet these constraints, say, "LIVE," "LIKE," "LOVE," "LAKE," and so on. In contrast, if half of the constituent letters are activated in a word that is 10 letters long, say, "AxTRAxxlxx," there will be very few 10-letter words other than "ATTRACTIVE" that fully meet the constraints. Thus, as word length increases, the perceptual code corresponding to an impinging word will be more unequivocally and uniquely activated and attention will be more likely to be directed to the valid perceptual code. As word length increases, there should be a greater chance of affective enhancement as opposed to affective impairment.

S. Ki tayama

209

In a recent experiment, both word affect and word length were systematically manipulated (Study 1 in Kitayama, 1991). Words comparable to Kitayama's (1990) low-frequency words were presented with an extremely diminished stimulus contrast and no expectation was provided - a condition the amplification model predicts to be highly conducive to affective impairment. A reliable affective impairment effect was observed. However, this impairment effect was also observed for longer words, thus failing to confirm the model's prediction. Perhaps, with the extremely diminished stimulus contrast examined in this study, there was only marginal activation of a relevant perceptual code regardless of word length. It would seem reasonable that an increase in word length could contribute to the unique and unequivocal activation of a relevant perceptual code only when there was enough stimulus input. In the experiments to be reported below, therefore, targets were presented with greater stimulus contrast. The perceptibility of the target was then reduced by presenting a masking stimulus immediately after the disappearance of the target. Under these conditions of backward pattern masking, an increase in word length was predicted to decrease the likelihood of affective impairment and to increase the likelihood of affective enhancement. Valence of affect. One potential divergence between the defense/vigilance hypothesis and the amplification model concerns the effect of the valence (positive or negative) of affect. Unlike the amplification model, the defense/vigilance hypothesis has never been explicit enough to advance clear-cut predictions for affective enhancement and impairment. Yet, it would seem to predict that the processing is either prohibited (the defense) or enhanced (the vigilance) if and only if "anxiety" (or, equivalently, "psychodynamic conflict") is evoked. Since "anxiety" is more closely linked with negative than positive affect, the perceptual influence of affect should be obtained primarily with negative affective words. In contrast, the amplification model is non-committal in this regard. It is possible that attention is amplified once the significance or the interest value of an impinging stimulus has been detected. If this is the case, the perceptual influence of affect need not depend on the valence (positive or negative) of the affect; for the significance or interest value can be signalled by any affect either positive or negative, In virtually all the past studies that examined accuracy in perception independently of response bias for or against affective stimuli, only taboo (mostly affectively negative) words were used (see Kitayama, 1990, for a review). The exclusive use of taboo words was justified on the supposition

210

Chapter 5

that Freudian defense should mediate the perceptual influence of affect (e.g., Blum, 1954, Erdelyi, 1974; McGinnies, 1949), but ironically made it impossible to test the supposition itself. A few studies, however, examined both positive and negative affective words while controlling for response bias. It was found that the perceptual influence of affect was mostly identical whether affect was positive or negative (Broadbent & Gregory, 1968; Kitayama, 1990, 1991). This evidence is consistent with the amplification model, but raises some doubt on the defense/vigilance hypothesis.

The present experiments In an attempt to further test the implications of the amplification model, two experiments were conducted. As noted above, initial support for the model was obtained in a review of past studies that differed in the extent to which a valid expectation was available to subjects. It was hypothesized that a valid expectation should activate the relevant code prior to the presentation of a target stimulus, thus alleviating the difficulty in locating the code in perceptual identification. The expectation, therefore, should increase the likelihood of affective enhancement and decrease the likelihood of affective impairment. So far, however, only a few studies have actually manipulated expectation (Freeman, 1954; Laccy ct al., 1953; Kitayama, 1990; Postman ct al., 1953). Although these studies supported the predictions of the amplification model, they were not conclusive. Freeman (1954), Laccy ct al. (1953), and Postman et al. (1953) measured recognition threshold, so response bias may in part account for their fmdings. Although response bias was controlled in the Kitayama (1990) experiment, only a small number of words (12 in total) were tested, leaving open the generality of the fmdings. Thus, the effects of expectation was further examined in the present experiments. Another variable tested was word length. According to current models of word processing (e.g., McClelland & Rumclhart, 1981), an increase in word length should impose more constraints on the identity of the word and, thus, conduce to unequivocal activation of the relevant perceptual code. Thus, affcctive enhancement should be more likely and impairment less likely with an increase of word length. Only Kitayama (1991) studied the effect of this variable, and failed to find any evidence. To test the conjecture that this failure was due to the highly degraded input, the current series of experiments employed a pattern-masking procedure, whereby a target stimulus was presented with a relatively high stimulus contrast, but was immediately

S. Kitayama

211

followed by a pattern mask. In these experiments word frequency was held constant at the level comparable to the low-frequency condition of the Kitayama (1990) experiment (the level also examined in the Kitayama (1991) study) to maximize the comparability across different studies. Finally, both positive and negative affective words were tested to determine whether the perceptual influence of affect depends on its valence. Predictions. We predicted that both expectation and word length would increase the likelihood of affective enhancement and decrease the likelihood of affective impairment. Thus, our first two predictions were: (1) Affective impairment will be most likely when short words are used and no expectation is provided. (2) Affective enhancement will be most likely when long words are tested and an expectation is provided. It was not certain exactly how expectation and word length would jointly operate. According to the amplification model, in order for these variables to have additive impacts on the likelihood of affective enhancement or impairment, two conditions must be met. First, expectation and word length must additively increase the activation of a target perceptual code relative to the activation of other irrelevant codes. Second, the relative increase of the activation of the target code must linearly increase the likelihood of affective enhancement (or decrease the likelihood of affective impairment). Neither assumption has been explicitly tested in the literature. Thus, no a priori prediction could be made regarding whether the two variables would interact or have additive effects. Thus, our third prediction was: (3) The influence of affect in the remaining two conditions (i.e., short words/expected, long words/unexpected) will fall somewhere between the above two extremes (short words/unexpected, and long words/expected); in other words, the influence of affect in the former conditions will be no greater than that in the latter. In addition to their hypothesized role to improve attentional engagement and thereby to increase the likelihood of affective enhancement and to reduce that of affective impairment, there are some suggestions in the literature that both expectation and word length may have some extraneous effects on forced choice performance. To begin with, it has been demonstrated that explicit

212

Chapter 5

formulation of a "hypothesis" or expectation can impair perceptual identification (e.g., Bruner & Potter, 1964; Lawrence & Coles, 1954). Perhaps, a clear expectation may direct one's attention to only one type of information that is potentially available (e.g., orthographic, phonemic, semantic, etc., in perceptual identification). Information on the ignored dimensions may then be unavailable in response selection, leading to poorer performance in the presence of an expectation than in its absence. Note that this restrictive effect of expectation should occur equally regardless of stimulus affect. Similarly, word length is also likely to depreciate overall performance independently of stimulus affect. In a forced choice task tested in the present research, the respondent scans the percept and compares it with available alternatives. The longer the word, the greater the number of features (e.g., individual letters) that must be matched and, therefore, the more difficult response selection should be. Furthermore, as we shall show below (p. 219), this difficulty in response selection for longer words may be exacerbated by the fact that any given pair of long words tend to share a greater number of common letters than a pair of short words. All in all, as word length increases, response selection will be more difficult and, further, this effect of word length on response selection will occur regardless of word affect. In sum, we hypothesized that both expectation and word length would depreciate overall performance in perceptual identification, while simultaneously increasing the likelihood of affeetive enhancement and decrease that of affeetive impairment. Taken together, we predicted a general decline of perceptual identification with expectation and word length, and further expected this decline of performance to be significantly less for affective words than for neutral words. Notice that this latter prediction amounts to the three predictions stated above.

Experiment 1 Method Overview and subjects. There were 128 trials, divided into two blocks, differing in the length of the words (long versus short). The order of the two blocks was counter-balanced over subjects. On each trial subjects were exposed to a 33 ms flash of either an affectively positive, negative, or neutral target word, immediately followed by a pattern mask (a string of "&"s of the same length as the word). They were then presented with the target word and an equivalently valenced word of the same length, and asked to choose the

S. Kitayama

213

one presented. Whereas word length was a within-subject variable, expectation was a between-subject variable. Thus, for half the subjects the two choices were presented immediately before the target was flashed to create an expectation, but for the other half there was no such expectation. Seventy undergraduates at the University of Oregon (both males and females) participated in the experiment to partially fulfill their introductory psychology course requirements. All the subjects claimed to be a native English speaker. Materials. The one-hundred twenty-eight words used in the present experiment are listed in Appendix A. There were an approximately equal number of affectively positive, negative, and neutral words. Thirty three undergraduates who did not participate in the present experiment were asked to judge the affective quality of each word (1 = unpleasant, 5 = pleasant). The positive words were rated as more positive (M = 4.3, s = .23) than the neutral words (M = 3.1, s = .21), which in turn were rated to be more positive than the negative words (M = 1.8, s = .30). About half of the words of each affect type were long (more than nine letters long) and half were short (fewer than six letters long). Frequency of occurrence ranged from 8 to 65 appearances per million words, as determined by Kucera and Francis's (1967) norms. The mean frequencies of occurrence for the six word categories (3 affective types • 2 length types) were practically identical (varying from 27 to 37 occurrences per million). This frequency range roughly corresponded to the low-frequency condition of the Kitayama (1990) experiment. Sixty-four pairs were formed between words with equivalent valences and lengths, as shown in Appendix A. Each word served as target once and as nontarget once, resulting in 128 experimental trials. The order of these trials were randomized within each block for each subject. All stimulus words were shown in uppercase letters. Equipment. The experiment was controlled by an AMDEK-286 personal computer with an AMDEK-132 VGA adapter. Stimuli were presented on an AMDEK-732 color graphic monitor. Both the brightness and contrast of the screen were kept maximal. To the monitor was attached a translucent tube. The inner contour of the tube was 17 cm • 23.5 cm, and the length, 60 cm. One end of the tube was fitted to the monitor, and subjects watched the screen through the other end. Target words were presented at the center of the monitor. The height of the words was approximately 4 mm. The length of a five-letter word was 12 mm, and that of a nine-letter word was 23 mm, resulting in the visual angle of approximately 1.15 ~ and 2.06 ~ for the short and the long words, respectively. The experiment was controlled by the MEL (Micro Experimental Laboratory) system developed by Schneider (1988).

214

Chapter 5

This system synchronizes the presentation of stimuli with the refresh cycle of the monitor, enabling accurate control over the exposure durations. The experiment was conducted under normal room illumination. With the VGA facility, it is possible to create 62 shades of gray in addition to white and black. In a black background (luminance = .67 ft-L [foot-Lambert]), a target was shown in the 23rd shade"of gray (1.02 ft-L). This level of stimulus contrast was considerably higher than the level employed in the Kitayama (1991) study (15th shade of gray; .72 It-L). A fixation point, a mask, an expectation, and response choices were shown in the 32nd shade of gray (2.65 i~-L). Procedure. Subjects were tested individually. Upon arrival, they were randomly assigned to one of the four conditions representing the presence or the absence of an expectation and the order of the short-word and the longword blocks. They were instructed to look into the tube, and to place their left and fight index fingers respectively on the Z and the M keys of the computer keyboard. The subjects were told that the experiment was concerned with perception of briefly shown words. After the procedure was described (see below), they were given the following instruction: "As I mentioned to you, words are presented very briefly. We want to know how accurately people can recognize a word under such impoverished viewing conditions. So, the p r ~ u r e is set up so that you cannot perfectly see the word, yet you can still recognize some fragments or parts of the word. For instance, you may be able to recognize a letter or two, or even a small part of some letter. Or you may be able to recognize the contour of the word. Such partial information has proved very useful in performing this task. I will explain to you exactly what can be le~meA from this sort of experiments later. For the time being, even though you might occasionally feel that you are merely guessing, don't be discouraged or disturbed by this. Instead, try to pick up as many physical cues from the flash as possible. In this way your responses will be most accurate. Even when you don't think that you have enough information to make a choice, give us your very best guess. Please never use any intuition or gut-feeling in making judgment; once you do this, your performance cannot be better than chance. From past research we know that it is essential that you try to pick up physical features such as letter segments and overall contour if you are to perform this task at a better-thanchance level."

S. Kitayama

215

In the expectation-present condition, each trial started with the presentation of a pair of words with one to the fight of the center of the screen and the other to the left. After 2 see, the pair was replaced by a fixation point at the center of the screen. The subjects in this condition had been told that one of the two words would be a target in the upcoming trial and that it Was important to keep the pair in mind in order to perform well in the task. In the expectation-absent condition the trial began with the presentation of a fixation point. The two expectation conditions were otherwise identical. Thus, in both conditions, when the subject simultaneously pressed the two response keys, the fixation point disappeared and, 200 ms later, a target word was presented for 33 ms, immediately followed by a pattern mask. A sequence of "&"s of the same length as the target word served as a mask. The mask was presented for 1500 ms. Immediately after the disappearance of the mask, the corresponding word pair (the one shown at the beginning of the trial in the expectation-present condition) was shown. The subjects had been instructed to press either the Z key (with the left index finger) if the word on the left side of the fixation point had been presented, or the M key (with the right index finger) if the word on the right side had been presented. Response time was measured from the onset of the choice pair in milliseconds. The next trial automatically started after an interval whose length varied randomly between 2 and 7 see with the average of 4.5 sec. At the completion of the first block, the subjects were given a short break prior to the second block. The experimental trials were preceded by 26 practice trials. On the practice trials both affective and neutral words not used in the experimental trials were presented. On the first practice trial the target word was shown for 200 ms. On each of the subsequent practice trials, the exposure time was gradually reduced so that by the 21st trial, it was set at the exposure level used in the experimental trials, i.e., 33 ms.

Results and discussion Two dependent variables were examined. First, the percentage of correct choices (hit rate or the accuracy score) was computed for each condition. In the present procedure the subjects chose between two affectively equivalent stimuli. Hence, there was no room for response bias for or against affective stimuli to come into play, and the accuracy score can be safely taken as an unbiased index of the perceptibility of the target (Natsoulas, 1965). Second, the time required to make a choice was analyzed. Because few existing studies in this area reported response time, it was not clear whether the

216

Chapter 5

perceptibility under the present conditions could be reliably and adequately indexed by response time. It was considered reasonable, however, that as the perceptibility of a target word increased and the accuracy score increased, response time would decrease. Accuracy. Two separate analyses were pefformexl on the accuracy scores. First, mean accuracy scores were calculated separately for each subject and then submitted to. an Analysis of Variance (ANOVA) with subjects as a random variable. A significant F statistic obtained in this analysis (designated as F1) indicates the gcneralizability of the effect across subjects. Second, mean accuracy scores were computed separately for each word pair and thon submitted to an ANOVA with word pairs as a random variable. A significant F statistic from this analysis (designated as F2) indicates the generalizability of the effect across word pairs. Clark (1973) has recommended the use of mmF', which indicates the extent to which the effect can be generalizeA simultaneously over both subjects and word pairs. A mmF' can be computed from the corresponding F 1 and F 2 according to the formula, F 1,,F2/(FI+F2). Some have argued, however, that this statistic is too conservative, prone to Type II errors (Wike & Church, 1976); both F 1 and F 2 will have to be quite large before the corresponding mmF' can approach statistical significance. In the present paper, thus, F 1 and F 2 are used to gauge the reliability of effects. MinF' will be reported only when it attains statistical significance. A preliminary analysis involving two between-subject variables (expectation and the order of long- versus short-word blocks) and two withinsubject variables (word affect and word length) showed no effect of block order (Fs < 1). This variable, thus, was dropped in subsequent analysis. The relevant mean accuracy scores are given in the first half of Table 2. Note, first, that in all the conditions the accuracies for positive and negative affective words were virtually identical, lending support to the prediction that the influence of stimulus affect on immediate conscious perception should not depend on the valence of the affect. Consistent with the amplification model, yet somewhat contrary to theories based on the notion of Freudian defense (e.g., Blum, 1954; Erdelyi, 1974; McGinnies, 1949), this apparent irrelevance of the valence (positive or negative) indicates that attention was amplified once the significance or the interest value of an impinging stimulus had been detected. Because the effect of the valence was negligible, the two valence categories were subsequently combined.

S. Ki tayama

217

Table 2. Means (and standard deviations) of accuracy and response time in perceptual identification as a function of word affect, word length, and expectation (Experiment 1). Expectation Word

Present (N=34)

Length

Absent (N=36) Word Affect

Positive

Negative

Neutral

Positive

Negative

Neutral

Accuracy Long Words

.63 (.11)

.63 (.14)

.57 (.12)

.60 (.13)

.57 (.14)

.59 (.09)

Short Words

.61 (.16)

.61 (.14)

.67 (.10)

.63 (.15)

.57 (.15)

.67 (.15)

Response Time (ms) Long Words

827 (363)

799 (326)

890 (431)

1760 (571)

1792 (740)

1684 (571)

Short Words

762 (397)

862 (458)

719 (402)

1651 (481)

1645 (490)

1571 (431)

A 2 x 2 x 2 (expectation, word affect, and word length) ANOVA showed a significant main effect for word length (/71(1,68)=22.5 , p<.001, F2(1,60)=19.2, p<.001, and minF'(1,125)=lO.4, p<.001). This effect, however, was qualified by the interaction between word affect and word length (F1(1,68)=12.9, p<.001, F2(1,60)=11.1, p<.005, and minF'(1,126)=5.97, p<.05) as well as by the interaction among word affect, word length, and expectation (F1(1,68)=4.04 , p<.05, and F2(1,60)=5.70 , p=.02). Thus, as predicted by the amplification model, the influence of affect on perceptual accuracy depended on both expectation and word length. The relevant means are shown in Figure 3. Based on the amplification model, we hypothesized that both expectation and an increase in word length would improve the accurate engagement of attentive processing in the relevant perceptual code. From this hypothesis, three predictions were advanced. First, affective enhancement should be most

Chapter 5

218

~176 i 0.68

['-]

Affective

m

0.66 0.64 0.62 "~

0.60

gh

0.~8 0.56 0.$4

/ O'' / / Word Length/Expectation

Figure 3. Perceptual sensitivity (hit rate) as a function of word affect, word length, and expectation (Experiment 1).

likely when both of these two factors are present. In support of this prediction, accuracy was higher for affective words than for neutral words, (t1(136)=2.95 , p<.01, t2(113)=2.98, p<.01, and minF'(246)=4.37, p<.05), when (i) the words were long and (ii) an expectation was present. Second, it was preAictexl that affective impairment should be most likely when (i) the words were short and (ii) an expectation was absent. Indeed, in this condition accuracy was lower for affective words than for neutral words (q (136)= 1.93, p<.06, and t2(113)=2.02, p<.05). Finally, the current analysis also predicted that the influence of affect in the remaining two conditions should be no greater than in the above two conditions. Consistent with this prediction, there was no influence of affect when (i) words were long and (ii) an expectation was absent, ts < 1. Further, there was an affoctive impairment effect when (i) words were short and (ii) an expectation was present (t1(136)=2.61, p<.01, and t2(113)=2.80, p<.01), but the magnitude of this effect was virtually identical to the comparable effect in the short/expectation-absent condition, ts < 1. When comparisons were made across the experimental conditions

S. Kitayama

219

separately for affective and neutral words, three pairs were reliable: the expectation present versus absent conditions for long, affective words (p<.06), and the long versus short neutral words at both of the levels of expectation (ps<. 01). We also hypothesized that an increase in word length would lead to a greater difficulty in response selection in a forced choice task. Further, we reasoned that this increase of difficulty in response selection due to word length should be exacerbated by the fact that a pair of any arbitrarily chosen words tend to share a greater number of letters as they become longer. To check this possibility, we counted the number of letters shared by the words in each pair. An ANOVA showed that regardless of word affect, the long word pairs shared a considerably greater number of letters (5.0) than the short word pairs (1.7) (F(1,58)=83.2, p<.0001). The same pattern emerged even when we controlled for the total number of letters in a word by testing the ratio of the number of shared letters to the total number of letters. Regardless of word affect, this ratio was significantly higher for long word pairs (.50) than for short word pairs (.33) (F(1,58)=18.9, p<.0001). In both measures, there was virtually no overlap between the short word and the long word conditions; so it was not possible to statistically test the relation between word length and performance independently of the number of shared letters. As predicted by the current analysis, word length generally depreciated performance in perceptual identification, but this detrimental effect of word length was significantly less for affective than for neutral words. It was also predicted that expectation would tend to impair response selection, causing a general decline of performance, especially for neutral words. Some hint of such an effect was found for long/neutral words; it, however, was far from significant. We will discuss this latter effect after Experiment 2 is reported. Response time. To the extent that response time decreases as the perceptibility of a target improves, we should find response time to vary systematically with word affect, word length, and expectation. Affective enhancement would be indicated by a shorter response time for affective than for neutral words and affective impairment, by a longer response time for the former than for the latter. The mean response times are given in the lower half of Table 2. A preliminary analysis indicated that there was no systematic difference between positive and negative words with one exception. In the short-word/expectation-present condition, response time was longer for negative words (862 ms) than for positive words (762 ms). Note, however, that both response times were longer than that for neutral words (719 ms).

220

Chapter 5

Hence, they are both demonstrating affective impairment. Furthermore, the accuracy measure has suggested that affective impairment is no stronger for negative than for positive words in this as well as in the remaining conditions. At this point, then, neither the replieability nor the theoretical significance of the above anomaly is entirely clear. In what follows, as in the analysis of the accuracy score, the two valence categories (positive and negative) of affective words were combined. An ANOVA performed on the mean response times showed a significant main effect for expectation (F1(1,68)=76.0 , p<.001, F2(1,60)=12.79 , p<. O01, and minF'(1,76)=71.7, p<. 001). Also significant were the two main effects for word length (F1(1,68)=4.12 , p<.05, and F2(1,60)=26.1, p<.001) and word affect (F1(1,68)=5.24, p<.05, F2(1,60)=3.40 , p<.08). Further, these main effects were qualified by the interaction between word length and word affect (F1(1,68)=4.11, p<.05, F2(1,60)=5.18, p<.03), as well as by the second order interaction among expectation, word length, and word affect (F1(1,68)=4.55 , p<.04, F2(1,60)=2.76, p=. 10). It must be noted first that response time was shorter when an expectation was present (810 ms) than when it was absent (1677 ms). This main effect of expectation simply indicates that response time was much briefer when choices had been given in advance. Further, consistent with the corresponding effect of word length on accuracy, response time was significantly briefer for long words (1214 ms) than for short words (1274 ms). These two main effects, however, interacted with word affect, as predicted by the amplification model. In support of the amplification model, affeetive enhancement was found in the long-word/expectation-present condition; response time was shorter for affeetive words than for neutral words. Further, affeetive impairment was evident in the short-word/expectation-absent condition; response time was longer for affeetive words than for neutral words. Finally, the influence of affect in the remaining two conditions was predieted to fall somewhere between the above two extremes. Both in the long-word/expectation-absent condition and in the short-word/expectation-present condition, affeetive impairment was apparent. But, consistent with the current analysis, the magnitude of this impairment effect was no greater than the comparable effect observed in the short-word/expectation-absent condition (ts < 1). Overall, the pattern for response time paralleled that for accuracy. First, in both measures affeetive enhancement was found when long words were expected. Second, also in both measures affeetive impairment was observed when short words were unexpected. Finally, the influence of affect in the

S. Ki tayama

221

remaining two conditions was no stronger than in the above two conditions. It seems clear, then, that under the present condition of backward pattern masking, both expectation and word length are importantly implicated in determining the precise effect (i.e., enhancement or impairment) of stimulus affect on perceptual identification. It appears that the failure by Kitayama (1991) to obtain an interaction between word length and word affect can be attributed to the fact that in the Kitayama (1991) study, the perceptibility of a target was reduced by an energy mask, which diminished stimulus contrast. The assumption is that in order for an increase of word length to contribute to the unique and unequivocal activation of a relevant perceptual code, there must be enough stimulus input. In the Kitayama (1991) experiment, stimulus input was marginal to begin with, whereas in the present procexture of backward pattern masking, stimulus input was presumably great enough for the influence of word length to show up.

Experiment 2 Experiment 2 was designed to determine the generality of the results of the first experiment. Specifically, one can always suspect that allegedly perceptual effects might be due to artifacts operating at the post-perceptual, response end (Eriksen, 1963; Goldiamond, 1958). It would seem important, therefore, to ensure that different methods of assessing perceptual sensitivity produce similar results. In the first experiment each word pair consisted of equivalently valenced words to exclude any response bias for or against affective stimuli (Natsoulas, 1965). In Experiment 2 we used an alternative approach to control for bias. Following Dorfman (1967; Dorfinan et al., 1965), each pair was now composed of an affective (either positive or negative) word and a neutral word. Thus, response bias for or against an affective choice was allowed. A signal detection framework (Green & Swets, 1965) was employed to measure this response bias and to yield an unbiased index of perceptual sensitivity. In addition, in Experiment 2 expectation was employed as a within-subject variable and word length as a between-subject variable, reversing their roles in Experiment 1. Method Overview and subjects. The experiment consisted of a large number of trials (240 and 252 trials in the long and short word conditions, respectively). On two thirds of the trials, subjects were exposed to a brief flash of either an

222

Chapter 5

affectivr or a neutral target word. On the remaining trials, no word was flashed although the subjects were led to believe that one had been flashed. The data from the latter trials were used to estimate response bias for or against an affective choice. After the flash, the subjects were presented with the target and nontarget words, and asked to choose the one that had been presented. On half the trials the pair was given prior to the flash as well, resulting in an expectation about the target. In the other half no such expectation was provided. For half the subjects the words were relatively long and for the other half they were relatively short. Twenty seven undergraduates at the University of Oregon (both males and females) participated in the experiment to partially fulfill their introductory psychology course requirements. All the subjects claimed to be native English speakers. Materials and equipment. The same equipment as in Experiment 1 was used. The stimulus words were virtually identical to those used in Experiment 1. Pairs were formed between an affective word and a neutral word of the same length. There resulted 20 positive-neutral and 20 negative-neutral word pairs for the long words (40 pairs in total), and 18 positive-neutral and 24 negative-neutral word pairs for the short words (42 pairs in total), as shown in Appendix B. Three kinds of trials were constructed: In the first kind, an affective word in each pair was a target; in the second, a neutral word in the pair was a target; and in the third no word was flashed as a target. The third kind of trials were used to assess response bias for or against an affective choice. There were the three corresponding trials for each of the pairs. The resulting 120 trials (40 x 3) in the long word condition and the 126 trials (42 x 3) in the short word condition were tested under each of the two expectation conditions. Two hundred forty (120 x 2) trials and 252 (126 x 2) trials were thus constructed in the long and the short word conditions, respectively. The order of trials was randomized for each subject. All stimulus words were printed in upper-case letters. Procedure. Subjects were tested individually. Upon arrival, they were randomly assigned to one of the two word length conditions (long versus short). These conditions were identical but for the stimuli used. The long word condition consisted of 240 trials, while the short word condition consisted of 252 trials (see above). The instructions and the proczxture were the same as in Experiment 1, except for the following changes. A trial in which an expectation was available started with the presentation of a pair of words with one to the fight of the center of the screen and the other to the lefL On a trial in which the expectation was not available, a pair of strings of eight "-"s appeared instead of the two words. After 1.5

S. Kitayama

223

sec, the pair (of either words or "-" strings) was replaced by a central fixation point. The subjects had been told that when the pair was given in advance, one of them would be presented on that trial and that it was important to keep the pair in mind in order to perform well in this task. When the subject simultaneously pressed the two response keys, the fixation point disappeared and, 200 ms later, on two thirds of the trials a target word was presented for 33 ms, immediately followed by a pattern mask, which lasted for 1500 ms. On the remaining one third of the trials, no stimulus was presented before the mask was shown for the same duration. The subjects had been told that a target was always presented "behind" the mask. A post-experimental interview revealed that no one had any suspicions about this manipulation. As in Experiment 1, the mask was a string of "&"s of the same length as the target. Immediately after the disappearance of the mask, the corresponding word pair (the one shown at the beginning of the trial in the expectation-present condition) was shown. In a randomly chosen half of the cases an affective word appeared on the right side and a neutral word, on the left, whereas in the other half the location of the affective and the neutral words was switched. Response time was measured from the onset of the choice pair, and the next trial automatically started after an interval whose length varied randomly between 1 and 3 see with the average of 2 see. At the completion of the first half of the entire set of the trials (120 and 126 trials in the long and the short word conditions, respectively), the subjects were given a short break before the second half. As in Experiment 1, the experimental trials were preceded by 26 practice trials. Finally, two minor procedural points may be noted. Experiment 2 consisted of about twice as many trials as Experiment 1. To complete the experiment within an hour, the duration with which an expectation was shown was reduced from 2 see to 1.5 see, and the intertrial interval was reduced from the average of 4.5 see to the average of 2 see. Results and discussion As in Experiment 1 both accuracy and response time were analyzed. Unlike in Experiment 1, perceptual accuracy was estimated by adjusting the probability that a given word was chosen given that the word had been presented (hit rate) by the probability that the word was chosen given that no word had been presented (false alarm rate) within a signal detection framework. In what follows, hit rate and false alarm rate will be briefly

224

Chapter 5

described before an adjusted accuracy measure is examined in greater detail. Response time will be reported at the end. Hit rates. Hits rates, shown in the upper part of Table 3, were submitted to an ANOVA with one between-subject variable (word length) and three within-subject variables (the presence/absence of an expectation, affective versus neutral words, and the valence (positive versus negative) of an affective word). The main effect for expectation proved significant (/71(1,25)=7.82 , p<.01, F2(1,78)=9.69 , p<.005, and minE'(l,68)=4.33, p<.05) indicating overall that the hit rate was lower when the expectation was present (.70) than when it was absent (.74). This main effect, however, was qualified by a significant interaction between expectation and target affeetivity (F1(1,25)=4.25, p<.05, and F2(1,78)=4.27, p<.05). The detrimental effect of expectation was much more pronouncexl for the neutral words (.68 versus .75) than for the affective words (.72 versus .74). False alarm rates. False alarm rates are given in the center of Table 3. In the present procexture, a false alarm rate for a neutral choice was equal to one minus the corresponding false alarm rate for an affective choice within any given condition. Hence, only false alarm rates for affective choices were analyzeA in an ANOVA with one between subject variable (word length) and two within subject variables (the presence/absence of an expectation and the valence of an affective word in a pair). Only word length turned out to be significant (F1(1,25)=7.24, p<.01, F2(1,78)=9.15, p<.005, and minF'(1,67)=4.04, p<.05). There was a bias to choose a neutral word when words were long (the false alarm rate for an affeetive choice=.45); when words were short, however, this trend was reversed (the false alarm rate for an affeetive choice=.53). These biases did not depend on the valence (positive versus negative) of the affect involved, somewhat contrary to what might be predicted by the notion of Freudian defense. Accuracy. The results for the hit rates, described above, can be seen as a combination of the accuracy with which a target stimulus was perceived and the bias for or against an affeetive choice (Green & Swets, 1966). Hence, to obtain an unbiased estimate of perceptual accuracy, hit rate must be adjusted for the corresponding response bias, especially when, as in the present case, response bias varied systematically across conditions. Following Pollack and Norman (1964), Grier (1971) has suggested that the average area under a hypothetical operating characteristic curve (nmning through the points [0, 0], [x, y], and [1, 1] within a unit square, with false alarm rates on the x axis and hit rates on the y axis) can be used as an unbiased estimate of the perceptual sensitivity (,4 9. As recommended by Grier, the following formula was used to

225

S. Kitayama

Table 3. Mean accuracies (and standard deviations) in perceptual identification as a function of word affect, word length, and expectation (Experiment 2). Expectation Word

Present

Length

Target Affectivity Valence

Long Words (N= 14)

Short Words (N= 13)

Absent

Affective

Neutral

Affective

Neutral

Positive

.73 (.15)

Hit Rate .68 (.15) .73 (.17)

Negative

.71 (.18)

.69 (.14)

.71 (.20)

.74 (.17)

Combined

.72

.69

.72

.74

Positive

.74 (. 16)

.67 (. 19)

.78 (.21)

.72 (.21)

Negative

.72 (.20)

.67 (.21)

.74 (.21)

.78 (.22)

Combined

.73

.67

.76

.75

.75 (.16)

False Alarm Rate Long Words (N= 14)

Short Words (N= 13)

Positive

.44 (.14)

.56 (.14)

.44 (.09)

.56 (.09)

Negative

.47 (. 11)

.53 (.11)

.45 (.09)

.55 (.09)

Combined

.46

.54

.44

.56

Positive

.52 (. 15)

.48 (.15)

.57 (.14)

.43 (.14)

Negative

.50 (.10)

.50 (.10)

.51 (.12)

.49 (.12)

Combined

.51

.49

.54

.46

Sensitivity (,4') Long Words (N= 14)

Short Words (N= 13)

Positive

.72 (.14)

.60(.16)

.72(.13)

.67(.14)

Negative

.69 (. 14)

.63 (. 16)

.69 (. 16)

.66(.15)

Combined

.70

.62

.71

.66

Positive

.68 (. 15)

.64 (.19)

.68 (.19)

.71 (.18)

Negative

.68 (. 15)

.64 (.19)

.68 (.17)

.72(.17)

Combined

.68

.64

.68

.71

Chapter 5

226

compute this index of perceptual sensitivity (A ~ when, as should theoretically be the case, a hit rate (y) was equal to or greater than the corresponding false alarm rate (x): 1

+

2

(y - x)(1

+y-x)

4y(1 -x)

Practically, however, a hit rate can be smaller than the corresponding false alarm rate due to the variability inherent in any empirical data especially when sensitivity is quite low. In this case, the comparable average area under an imaginary "operating characteristic curve" that runs through the three points, [0, 0], [x, y], and [ 1, 1], was computed by the formula:

A'=--

1

2

+

(x-y)(1 + x - y )

4x(1 -y)

A sensitivity index (A) was computed in two different ways. First, it was computed over relevant word pairs for each subject. Analysis on these A's would suggest the reliability of a given effect over subjects (as indexed by F1). A ' w a s also calculated over subjects separately for each word pair. Analysis on the latter would suggest the reliability of the effect over word pairs (as indexed by F2). In the lower part of Table 3 are presented the pertinent means. A' is not a mere arithmetic mean of individual choice data. Hence, even though F 1 and F 2 can be computed in the ways described in the text, the corresponding minF' cannot be defined based on the corresponding F 1 and F 2. Therefore, no attempt was made to compute minF's. For the same reason, the condition means of A's differed slightly depending on whether A's were originally computed over word pairs or over subjects. As may be expected, however, the difference was quite negligible. An ANOVA performed on the sensitivity index (A') showed significant main effects for expectation (F 1(1,25)=7.03, p<.02, and F2(1,78)=8.18, p<.005), and the affectivity of a target (F1(1,25)=4.49, p<.05, and F2(1,78)=5.25, p<.03). These main effects, however, must be evaluated in view of an interaction between word length and target affectivity (/71(1,25)=3.59, p<.07, and F2(1,78)=8.06, p<.O1), and an interaction between expectation and target affectivity (F1(1,25)=3.91, p<.06, and F2(1,78)=3.51, p<.07). No other effect approached statistical significance. Among others, there was no significant effect involving the valence (positive

227

S. Kitayama 0.72 [']

Affectiv

1

Neutral

0.70 -

0.68 -

IO

0.66

--

0.64

"

I,m IO

0.62 -

0.60 -

Word Length/Expectation

Figure 4. Perceptual accuracy (the sensitivity index A ) as a function of word affect, word length, and expectation (Experiment 2).

or negative) of an affective word in each pair. We thus found again that the perceptual influence of affect does not depend on the valence of the affect. The relevant means are displayed in Figure 4. The most crucial prediction of the amplification model is that affective enhancement would be most likely in the long-word/expectation-present condition whereas affectivr impairment would be most likely in the short-word/expectation-absent condition. In support of this analysis, a significant affectivr enhancement effect was obtained in the long-word/expectation-present condition (t1(49)=2.84, p<.01, and t2(156)=2.80, p<.01). This enhancement effect, however, was reversed, albeit non-significantly, in the short-word/expectation-absent condition. The simple interaction representing this pattern proved significant (t1(25)=2.95, p<.01, and t2(78)=2.55, p<.05). In the remaining two conditions, a marginal effect of affectiw enhancement was evident (ps > .10). Consistent with the amplification model, however, these affective enhancement effects were no stronger than the comparable effect in the long-word/expectation-present condition.

228

Chapter 5

In addition, as in Experiment 1, an increase in word length diminished the general performance level. This effect can be attributed to a greater choice difficulty in the long word condition than in the short word condition. As predicted, however, this decline was less for affective words than for neutral words. Further, in Experiment 2 expectation decreased the performance for neutral words. No such effect was found in Experiment 1. In another recent experiment that used only relatively short words (Kitayama, 1990), a similar detrimental effect of expectation was found only for neutral words of an extremely high frequency: it was not found for word frequencies comparable those used in the present studies (see Figure 2). All in all, while such a detrimental effect of expectation seems to occur occasionally, the exact condition for its occurrence is currently unclear. This detrimental effect is consistent with the possibility, suggested by the previous research, that explicit formulation of a "hypothesis" or expectation can impair perceptual identification (e.g., Bruner & Potter, 1964; Lawrence & Coles, 1954). Such a "hypothesis" may direct one's attention to only one type of information that is potentially available (e.g., orthographic, phonemic, semantic, etc., in perceptual identification). Information on the ignored dimensions may then be unavailable, leading to poorer performance in the presence of an expectation than in its absence. Thus, as predicted, performance is generally worse in the presence of an expectation than in its absence, but this detrimental effect of expectation was significantly less for affcctivc words than for neutral words. Response time. Mean response times arc given in Table 4. Data from the trials in which a target was actually prcscntexi were analyzed. Unlike in the first experiment, no significant effect involving affect was observed. Nevertheless, it is noteworthy that a pattern that closely resembles that of Experiment 1 (Table 2) is discernible in Table 4. Relevant to the present hypothesis arc the response times when a target stimulus was shown (i.e., affcctivc and neutral target types). When combined over the two valence categories (positive and negative) of affcctivc words, response time was shorter for affcctivc than for neutral words in the long-word/expectationpresent condition (affcctivr enhancement; 585 ms versus 615 ms). This effect was reversed both in the long-word/cxpoctation-absent condition (affe~ivc impairment; 1115 ms versus 1100 ms) and in the short-word/expectationpresent condition (638 versus 613 ms); fmaUy, when there was no expectation and the words were short, the reversal was somewhat stronger (1047 ms versus 985 ms). It is not clear why this pattern, which was statistically significant in Experiment 1, failed to attain statistical significance in Experiment 2. It seems to fair to suggest, consistent with Kitayama (1991),

S Kitayama

229

Table 4. Mean response time in milliseconds (and standard deviations) as a function of word length, expectation, and target type (Experiment 2). Expectation Word

Present

Absent Target Type

Length Valence

Affective Neutral

N o n e Affective Neutral

None

Long Words Positive

562 (212)

627 (263)

676 (212)

1107 (351)

1122 (325)

1318 (351)

Negative

607 (240)

603 (287)

689 (239)

1123 (394)

1077 (317)

1.325 (420)

Combined

585

615

683

1115

1100

1321

Short Words Positive

651 (281)

556 (213)

627 (211)

1068 (384)

994 (337)

1318 (333)

Negative

628 (302)

656 (380)

831 (299)

1031 (256)

979 (343)

1401 (378)

Combined

638

613

743

1047

985

1366

that although perceptibility can be indexed by both accuracy and response time, the former is a more reliable and sensitive measure than the latter under the current conditions of impoverished stimulus presentation. Comparison between Experiments 1 and 2. The above results replicated the general pattern obtained in Experiment 1. In both experiments, affective enhancement was observed when long words were expected; whereas affective impairment was observed when short words were unexpected. Further, in both experiments the influence of affect in the remaining two conditions (i.e., when long words were unexpected, and when short words were expected) was no stronger than in the above two conditions (i.e., when long words were expected, and when short words were unexpected). Thus, these two experiments converge with the past evidence (e.g., Kitayama, 1990) to suggest that both word length and expectation are crucial in determining the influence of stimulus affect on perception. Both an increase in word length and the presence of an expectation render affective enhancement more likely, and impairment less likely.

230

Chapter 5 The Amplification Model Evaluated

Summary oft he present research It was hypothesized that affect elicited by preattentive processing amplifies attentive processing. Accordingly, affect inherent in a stimulus may either impair or enhance the veridical perception of the stimulus itself depending on the accuracy with which attentive processing is directed to the relevant perceptual code. When it is difficult to locate a perceptual code corresponding to an impinging stimulus so that attentive processing is misdirected to an irrelevant perceptual code, affect magnifies invalid perceptual information, leading to affective impairment (i.e., poorer perception of affective than neutral stimuli). If, however, the difficulty of locating the relevant code is alleviated by some means so that attentive processing is correctly directed to the relevant code, then the affect-caused amplification of attentive processing should lead to affective enhancement (i.e., better perception for affective than for neutral stimuli). In perceptual identification experiments like the present ones, an extremely impoverished target word is presented. Under these conditions, it should be difficult to determine exactly which perceptual code corresponds to an impinging stimulus. Hence, unless the difficulty in locating the relevant perceptual code is alleviated by some means, affective impairment should be found. The two experiments reported here examined two variables that should aid in determining the relevant code. First, a valid expectation, or priming, should activate the relevant code prior to the presentation of the stimulus. Second, an increase in word length should conduce to unequivocal activation of the relevant perceptual code, especially when high stimulus contrast is used. Under these conditions, initial, letter-level activations should impose more constraints on the identity of the word. In general congruence with this analysis, in both experiments perception was better for affective than neutral words (affective enhancement) when long words were used and an expectation was present, but it was worse for the former than the latter (affective impairment) when short words were used and there was no expectation. In the remaining two conditions where only one of the two factors aids in locating the relevant code (long words/unexpected, and short words/expected), the influence of affect was no stronger than in the above two conditions where these two factors were either both present or both absent. Response times tended to corroborate the pattern obtained for the

S. Ki tayama

231

accuracy measures. Finally, virtually the identical effects were observed for both positive and negative affective words.

Methodological comments Confounding variables. These findings were documented here using a large number of stimulus words. Further, the frequency of occurrence of these words was carefully controlled at a level comparable to previous studies (the low-frequency condition of the Kitayama, 1990, 1991, studies). This methodological point is not trivial, because the generalizability of past findings was otten suspect due to the small number of stimuli tested. It is now difficult to dismiss the findings as mere artifacts due to certain idiosyncratic stimuli or a confounding between affect and word frequency. Response bias. There is always room to suspect that allegedly perceptual effects might be due to methodological artifacts operating at the post-perceptual, response end. For example, recognition threshold often used in the perceptual defense literature in the 1950s did not eliminate the possibility of such response bias (Eriksen, 1963; Goldiamond, 1958). In view of this, it should be emphasized that the present experiments used two different forced choice formats both designed to control for response bias. Also, unlike in the past studies, response time was analyzed along with accuracy measures. We found a similar pattern of results across the two forced choice methods, with both accuracy and response time, making it unlikely that the observed influence of affect is due to response artifacts. Evaluation All in all, along with the studies reviewed earlier, which demonstrate that other variables such as stimulus contrast and word frequency can interact with word affect to determine the perceptibility of the word, the present research has provided clear evidence for the amplification model of affectcognition interaction. Affective enhancement becomes more likely and impairment, less likely when the difficulty in locating a relevant perceptual code is alleviated or, more specifically, when (i) stimulus contrast increases, (ii) exposure time increases within a range of relatively short times, (iii) word frequency increases, (iv) word length increases, or (v) a valid expectation is made available. Finally, the affective influence is mostly the same for both positive and negative affect.

232

Chapter 5

The research guided by the amplification model, described so far, has raised a number of important issues. The rest of this paper will be devoted to a discussion of four such topics. First, the key idea in the amplification model is that attention me~atos an interaction between affect and cognition. It is important, then, to examine this idea within a broader context of attention research. We will review major theories in the past three decades and explicate the current model's relation with each of them. Second, although the present research has focused exclusively on immediate perception, the amplification model may prove applicable to other, non-perceptual tasks. Relevant empirical evidence will be reviewed. Third, we are now in a far better position than was possible before to critically evaluate the notion of psychological defense both in perceptual identification and in real life. We will thus examine the current theoretical status of Freudian defense in view of the current findings. Finally, some directions for future research will be discussed. Relations with Extant Theories of Attention

Early versus late selection Since Broadbont (1958), theories of attention have been classified according to the location of "information filter", whereby sensory input is selected for further processing. Placing the filter early in the processing, Broadbcnt argued that attcntional selection is based on crude sensory or external cues such as voic~ tone or location. Consistent with this early selection model, selection is quite easy when distinct physical, external cues are readily available. Nevertheless, the model did not seem to account for some other, quite common observations. For example, attention is oi~en captured by a significant stimulus (e.g., one's own names) presented in a channel that is being ignored (the "cocktail party" effect; e.g., Moray, 1959; Treisman, 1964). Several researchers therefore proposed late selection models (e.g., Deutsch & Deutsch, 1963; Marcel, 1983b, Norman, 1968; Shallic~, 1987), whereby all sensory inputs are thought to receive full processing before one of them is selected for attention and thus for conscious awareness. The amplification model is congruent with the late selection models. In fact, if it is only the early selection model that wore valid, the phenomena of affective enhancement and impairment would seem quite puzzling, for they imply selection of sensory data based on affective or semantic information. Indeed, Broadbent himself supplied one of the best early demonstrations of

S. Kitayama

233

affective impairment (Broadbent & Gregory, 1967; see also Broadbent, 1975). Having unexpectedly found a reliable affective impairment effect, he remarked: "...we have to suppose that emotional words do not behave like words of low probability...there appears actually to be a failure of the stimulus reformation to reach the perceptual mechanism...Frankly we were not expecting these results, and have described [the findings] at an informal level as showing that 'Freud was right after all' (Broadbent & Gregory, 1967,

p. 583)." It would seem more reasonable, then, to suppose that the early and the late selection models of attention specify quite different and separate aspects of attentional selection. Whereas the early selection model is concerned primarily with selection based on external or physical cues, the late selection models focus on selection based on internally generated features such as the activation strength of relevant or irrelevant internal representations (Norman, 1968). Both selective functions may co-exist in the human brain and supplement each other to coordinate an action (such as conscious thought or overt action) with respect to an input (Allport, 1989). As proposed by the early selection model, information can be selected for conscious awareness in terms of physical or external cues. Yet, as proposed by the late selection models, information unselected on a physical or external basis may still receive considerable preattentive processing and can provide a basis for subsequent attentional selection. A recent development in cognitive neuroscience supports this view. Drawing on data from positron emission tomography (PET) as well as from patients with brain injuries, Posner and his colleagues have suggested distinct anatomical structures for each of these selection operations (Posner, Peterson, Fox, & Raichle, 1988). Specifically, tasks that require attentional selection based on spatial information (e.g., orienting to an external cue) seem greatly impaired in patients with parietal damage, whereas in tasks that require attentional selection based on internal information (e.g., generating uses [pound] for an object [hammer], and a Stroop color naming task), more anterior areas including the anterior cingulate gyrus and the surrounding regions display the strongest PET activation. Thus, a system subserving selection based on external information (called the posterior attention system) appears quite distinct in both anatomical structure and function from a system serving selection based on internal information (called the anterior attention system; see Posner & Peterson, 1990, for a review).

234

Chapter 5

Preattentive and attentive processing Along with prior theories of late selection (e.g., Deutsch & Deutsch, 1963; Marcel, 1983a; Norman, 1968), the amplification model elaborates on the mechanisms implicated in selection based on internally generated information. Specifically, it rests on a formulation in which the development of a conscious percept is decomposed into two relatively separate component processes (preattentive processing and attentive processing), first made explicit by Posner and Snyder (1975) and subsequently elaborated on by a number of researchers (e.g., Neely, 1977). To this formulation the amplification model has added a possibility that affect, caused by preattentive processing, influences subsequent processing. Because considerable evidence now suggests that preattentive processing is sufficient to produce some affect, this addition seems logical. Indeed, Posner (1978) had already made a similar suggestion over ten years ago. In his seminal Chronometric Exploration of Mind, he remarked (p. 238): 9..it may be possible that feedback from the emotional arousal of a given word can influence the conscious processing of that word itself. This provides a mechanism whereby an emotional word can influence the direction of our attention to itself. Around the same time, Broadbent (1973, p. 79) also made a similar comment: Consider then a different kind of perceptual system, in which... ...the output of (a perceptual) mechanism feeds back to the mechanism itself in such a way as to change its parameters .... the (preconscious) identification of a word as nasty then produces an emotional state which changes the strategy of (further) intake of information... This in turn might change the (conscious) identification of the stimulus. [words in parentheses were added by the present author to clarify the point.] According to Posner, the effect of preconscious affect is limited to conscious processing. The present research, however, has broadened the range of such an effect, providing evidence that preattentive affect can also amplify attentive (yet preconscious) processing required to develop a

S. Kitayama

235

conscious percept, thus confirming Broadbent's conjecture. The entire mechanism suggested by Posner can be preconscious.

Resource theory of attention Another key assumption of the amplification model is that attention is a set of mental capacities that can be increased (or amplified). This idea can be traced back to F reudian notion of attentional cathexis, which was further developed by more recent ego-psychologists, most notably by Rapaport (e.g., 1967). Subsequently the idea has provided impetus to the resource theory of attention by Kahneman (1973) and to a number of experiments on attention (see e.g., Johnson & Dark, 1985, for a review). Nevertheless, there are some obvious difficulties in the resource theory (Allport, 1989; Navon, 1984). Specifically, the notion of a single set of generic processing resources is very likely a too-simplistic, albeit a useful and heuristic, fiction. Future theoretical work in this area must specify the exact, mechanism or computation involved in attentive processing (see e.g., LaBerge & Brown, 1989, Treisman, 1988, for analyses for spatial attention) and delineate what is to happen to this computation when attentive processing is "amplified". The present data should constrain such an endeavor of further theorizing.

Amplification of Attention in Other Domains

Other tasks With the amplification model, both affeetive enhancement and impairment in perceptual identification can be seen as a variant of the general capacity of affect to amplify perceptual and cognitive processes. This perceptual influence of affect was first proposed by Easterbrook (1959), who suggested that affect limits the range of cues that can be utilized by the perceiver. Thus, affect may enhance ("amplify") the perception of central, focal features. Recent studies by Christianson and colleagues (Christianson & Loflus, 1991; Christianson, Loftus, Hoffman, & Loftus, 1991) have demonstrated that detailed information at the center of a briefly shown emotional scene is significantly better recalled than the same detail from a comparable neutral scene. Furthermore, there is some, albeit mixed, evidence that enhanced perception of a central emotional stimulus is accompanied by an impairment of the perception of peripheral, nonfoeal features. Although Christianson et al. (1991) failed to find any evidence, Erdelyi and Appelbaum

236

Chapter 5

(1973) did show that a highly emotional symbol located at the center of a tachistoscopic display decreases incidental memory of peripheral stimuli. Likewise, Kitayama (1996) finds that emotional intonation of a speech can disrupt the attentive processing and, thus, subsequent memory, of the attendant verbal content. The same theme is evident in the literature on arousal and memory. Reviewing this literature, M. Eyscnck (1976) has concluded that "high levels of arousal have the effect of biasing the subject's search process toward readily accessible sources of stored information more than is the case with lower levels of arousal." A similar point has been raised more recently by Paulhus and Levitt (1987), who showed that a para-fovcal affcctive stimulus intensifies "ego-enhancing," presumably personally dominant, thinking tendencies. A more elaborate model of arousal and memory has been proposed by Humphmys and Rovcllo (1984). Consistent with previous theorizing by Walker (1958), they have suggested that arousal produced by stimulus affect interferes with short-term retention of the stimulus (because arousal amplifies competing responses along with a correct response) while improving long-term memory trace (because it helps sustain the "transfer" of the correct response to long term memory). In support of the model, several studies have found that in a paired associate learning task, recall of affective materials is poorer than recall of neutral materials immediately after learning; but after a long delay the opposite pattern emerges (e.g., Kleinsmith & Kaplan, 1963; Walker & Tarte, 1963). The exact relationship between the amplification model and the Humphreys-Revelle model has yet to be specified, although both have in common the assumption that affect amplifies some relevant cognitive or perceptual operation(s). Amplification due to affect has been demonstrated in still another measure of perceptual sensitivity. Worthington (1969) presented two words in succession, one at a time, below a recognition threshold, and had subjects judge which was brighter. It.was found that taboo words were judged as brighter than neutral words even though the subjects were not aware of the identities of the words. According to the amplification model, preattentively induced affect amplified the attentive processing of the word itself, producing a more vivid image for taboo words than for neutral words. Further support for this analysis comes from a second experiment, in which an affective and a neutral stimulus were shown simultaneously. Under this condition, the foregoing effect of affect entirely disappeared. Perhaps, attentive processing amplified by an affective word was applied to either the word itself, another (neutral) word, or both of them, erasing any difference in the vividness of the

S. Kitayama

237

two images. Finally, the literature of "sensory interaction" in the 1940s and 1950s tended to demonstrate that a simultaneously presented extraneous stimulus lowers the detection threshold of a focal target (Zajonc & Dorfman, 1964; see Dixon, 1980, for a review). This literature is also consistent with the present analysis, insofar as the presentation of an extraneous stimulus, by itself, may evoke some arousal. All in all, the amplification of perceptual and cognitive processes by affect seems quite pervasive across a wide variety of experimental tasks. Affect can amplify both behavioral and cognitive responses that are "dominant." According to the present model, stimulus affect can amplify attentive processing, which is directed to a strongly activated code, that is, "dominant" perceptual information. Thus there is clear correspondence between the current model and prior (mostly behavioral) theories on amplifying functions of affect. The amplification model adds to the previous literature the possibility that amplification by affect may occur even in a very early, preconscious stage of processing. It would seem that further specifying the nature of this amplification may be a key to understanding the interaction between affect and cognition that occurs preconsciously.

Valence of affect Although the present research has found no evidence for any effect of the valence of affective stimuli, this by no means implies that valence is not activated preconsciously. In fact, existing evidence suggests that positive and negative affect can be evoked by undetectable stimuli (Greenwald et al., 1989; Niedenthal, 1990; Zajonc et al., 1989). The present study does suggest, however, that preattentively elicited affect is likely to amplify subsequent processing regardless of its valence. The present finding also leaves open the possibility that positive and negative affect may have divergent effects on attention in tasks other than immediate perception. Affect examined in the present research is neither differentiated nor consciously recognized. Affect of this kind is shown to amplify attention regardless of valence. More elaborate and differentiated forms of affect, however, may entail quite different processing consequences. As proposed by appraisal theories of emotion (Frijda, 1986; Smith & Ellsworth, 1986; Roseman, 1984; Scherer, 1984), the processing of affect is likely to proceed from the detection of global valence or significance to the identification of more differentiated affective states in terms of discrete emotion categories. Thus, consistent with an earlier formulation by Schachter

238

Chapter 5

and Singer (1962), initially undifferentiated affect is combined with a variety of cognitions about the surrounding situation ("appraisals") to form a fullfledged emotional state. It may well be the case that once valence is clearly differentiated, perhaps, in conscious awareness, it recruits characteristic response patterns (Davidson, 1984; Lang, Bradley, & Cuthbert, 1990). Specifically, negative affect may mobilize coping responses, causing additional attention to be allocated to the stimulus event, whereas positive affect may halt any such active attentional and motor processes (Tucker & Williamson, 1984; see Taylor, 1991, for an integrative review of this literature). Hence, once an emotional event has become conscious, more attention may be allocated to the focal stimulus only if the event is negative. Several studies have examined allocation of attention to clearly perceptible stimuli by measuring gaze time (e.g., Fiske, 1980), performance in a secondary task (e.g., MacLeod, Mathews, & Tara, 1986), speed in lexical judgment task (Matthews, Pitcaithly, & Mann, 1995), and interference in a Stroop color naming task (e.g., Pratto & John, 1991). These studies have consistently found that negative stimuli summon more attention than either positive or neutral stimuli. Analogous effects have also been found for a negative mood state. Derryberry and his colleagues (Derryberry, Bran&, & Reed, 1991) used a probe reaction time task and found that reaction time to a central probe is briefer after failure feedbacks in a concurrent tasks (designed to induce a negative mood) than after success feedbacks (designed to induce a positive mood). Interestingly, however, there was no comparable effect of feedback on reaction time to a peripheral probe. Thus, consistent with the studies by Christianson et al. (1991) reviewed earlier, a negative mood amplifies the processing of central stimuli, but not peripheral ones 1.

Perceptual Defense and Vigilance? Defense in perception

Traditionally, the notion of Freudian defense was postulated to account for the influence of affect on perception (e.g., Blum 1954, Brunet, 1957, 1 Evidence indicates that in a positive mood, attention tends to be more broadly distributed in either space or semantic region (Derryberryet al., 1991; Isen, 1987; Isen & Daubman, 1984). In terms of the spotlight metaphor, a positive mood appears to "broaden" the scope of attention. The influence of affect on this more structural aspect of attention deserves further research (Tucker & Williamson, 1984).

S. Kitayama

239

Dixon, 1980; Erdelyi, 1974; McGinnies, 1947). Unfortunately, however, exact mechanisms of defense have never been specified enough to advance clear predictions. In contrast, the amplification model describes these effects as a result of an interaction between affective and cognitive pathways commonly involved in any ordinary processes of perceiving and thinking. As noted above, the amplification model provides a coherent explanation for an otherwise anomalous pattern of the past studies conducted under the Freudian tradition. Furthermore, it successfully predicts the key outcomes of the present experiments. The model can also comfortably accommodate the evidence that the influence of stimulus affect on immediate conscious perception is mostly the same for positive and negative affect (Broadbent & Gregory, 1967; Kitayama, 1990, 1991) - evidence incongruous with the notion of Freudian defense. It would seem that attentive processing can be amplified once the significance or interest value of an impinging stimulus has been detected through preattentive processing. Future research must now be directed toward the precise nature of this preattentive computation of the significance of the stimulus (cf. LeDoux, 1989). All in all, although effects of affect in a perceptual identification task have traditionally been studied under the rubric of perceptual defense/vigilance, the present research suggests that they may actually have nothing to do with Freudian defense. Instead, the significance of these effects lies primarily in the fact that they provide valuable clues about the functional organization of unconscious processes commonly involved in various forms of perceiving and thinking.

Defense in real life Although the notion of defense is perhaps irrelevant in understanding the influence of affect on perceptual processes in the present experimental paradigm, we also acknowledge that a variety of psychological defenses can be identified in real life, especially in many clinical settings (e.g., Schwartz, 1990; Weinberger, 1990). Yet, possible mechanisms by which these defensive functions are implemented have remained as ill-specified as they were over 100 years ago when Freud (1895/1966, p. 370) observed the following: Everything that I call a biological acquisition of the nervous system is in my opinion represented by a threat of unpleasure..., the effect of which consists in the fact that those neurones which lead to a release of unpleasure are not cathected. This is primary defense. How primary defense, non-cathexis owing to a threat of unpleasure,

240

Chapter 5 is to be represented mechanically - this, I confess, I am unable to say.

From the present theoretical perspective, however, we may suggest that the interactive process between affect and cognition captured in the amplification model is one mechanism that can serve a defensive funetion under certain circumstances in real life, although this by no means precludes any other processes that might be implicated especially when affect involved is very intense. Specifically, an idea said to be "conflictual" in a psychodynamic sense (of. Shevrin, i 990) may often be characterized by two features. First, such an idea may be assoeiateA with an extremely intense affective response. Second, the idea may be relatively isolated from more innocuous scripts or schemas commonly activated in everyday life. Under these conditions, the amplification model would predict that even when a "conflictual" idea is preconsciously activated and a strong affect evoked, attentive processing is likely to be allocated to a more accessible script or schema. As a result, the perception or recognition of the innocuous script or schema may be magnified; but the original "conflictuar' idea will stay nonconscious. The mechanism specified by the present model, then, seems to serve the purpose of psychological defense in situations in which the above two features validly characterize a "eonflietual" idea, even though the mechanism itself is not specifically designed for defense. Future Research Directions

The amplification model has suggested a reasonable answer to the specific agenda set for the current research, but it has also generated an array of open issues and questions. In particular, assessing the generality of the model must receive first priority in future research. To begin with, we must pose questions about different forms of cognitive performance: Does the nature of the cognitive task determine the influence of affect? For example, will affect have any influence on lexical judgment and, if so, can the nature of the influence be described by the amplification model? What about categorization or some other tasks requiring more deliberate thinking or explicit memory retrieval? Many similar questions can be asked for affect as well: Does the influence of affect depend on the exact form of affect being induced? Is the amplification model generalizable to other forms of affect? For example, will

S. Kitayama

241

affect influence cognition differently depending on whether it is directly attached to a stimulus object (e.g., the negative affect usually attached to representations of snakes) or whether, on the other hand, it is induced by a certain "irrelevant," "incidental" stimulus (e.g., the negative affect commonly elicited by a scratching sound)? What about more enduring affective states, or the moods, of the perceiver? Alternatively, would affect implicit in cognitive representations (e.g., the affect associated with the notion of ANGER) behave in some different ways from, say, the affect more directly expressed by patterns of sensory signals (e.g., vocal expression of hostility; of. Scherer, 1986)? Before further generality of the amplification model is claimed, each one of these empirical issues must be addressed. In the current research we have strategically focused on word perception, because it seemed the best task for the purpose of formulating a processoriented theory of an interaction between affect and cognition. Future research may apply the theoretical processes captured in the amplification model to perception in social domains, such as facial perception, in which preattentive processing of a stimulus is known to play critical roles (Damasio et al., 1982; Hansen & Hansen, 1989; Niedenthal, 1990). To illustrate one possible extension along this line, faces of one's own race tend to be more familiar and their representations, more differentiated than foreign faces. Accordingly, faces of one's own race should yield a greater number of constraints on preconscious processing, hence making for more unequivocal activation of the relevant representation. From the current results on word length, one could predict that one's ability to recognize a faintly seen face would depend both on the emotional expression of the face and its race (one's own race or a foreign race) - that is, emotional faces should be more perceptible and more easily identified than neutral faces if these faces are of one's own race, but not if they are foreign faces. Future work along this line can contribute to basic social perception research (e.g., Schneider, Hastorf, & Ellsworth, 1979), as well as more applied domains such as eye-witness testimony (Brigham, Maass, Martinez, & Whittenberger, 1983; Loftus, 1979). The amplification model also has a more general implication for the current social cognition literature. With a few notable exceptions (e.g., Isen, 1987), this literature has examined an interaction between affect and cognition within the confine of associative network models of memory (e.g., Bower, 1981; Lang, 1984). Yet, it may be argued that evocation of affect changes not only the contents of thought (as predicted by these models), but also the structure or organization of thought (e.g., Isen, 1987; Tucker &

242

Chapter 5

Williamson, 1984). Thus, it seems likely that thinking about a negatively stereotyped ethnic group not only raises negative connotations, but also leads to more rigid and narrower thought processes (Kitayama & Bumstein, 1989). The structure of thought may in turn depend on certain functional characteristics of attentive processing. The present analysis of the interdependence between affect and attention may serve as a useful point of departure for further elucidating some important questions and issues in, for example, stereotyping and prejudice.

Concluding Remarks In sum, the present paper has described a model of affect-cognition interaction in perception, presented experimental evidence, and discussed a variety of empirical and theoretical issues raised by the model. We believe that the current research revives the possibility that affect can participate in very early, preconscious stages of processing. With sophisticated cognitive (e.g., Bower & Clapper, 1989), neuroanatomical (e.g., Posner & Peterson, 1990), and electrophysiologieal (e.g., Cacioppo, Martzke, Petty, & Tassinary, 1988; Davidson et al., 1990) techniques, experimental analysis of nonconscious processes is now quite feasible (Kihlstrom, 1987, 1990). This analysis emphasizes that the role of nonconscious processes may be far greater than is routinely supposed in cognitive and social psychology. The current work suggests that affect participates in these nonconscious processes. By elucidating the role of affect in early perceptual processing, the present research has supplied a new theoretical perspective to some old problems such as perceptual defense and vigilance or subception. We showed that these phenomena cannot be dismissed solely on methodological grounds as was often done in the past. However, we also raised a serious doubt on previous theorizing that tended to link these phenomena to Freudian defense. We argued, instead, that the theoretical significance of the phenomena of defense and vigilance actually lies primarily in their potential to reveal the functional structure of the nonconscious processes commonly involved in virtually every form of perceiving and thinking. We believe that the present attempt to integrate affective phenomena within the current cognitive framework has the potential of enriching this framework itself, insofar as the latter is presently essentially "affect-free." Furthermore, granted that affective involvement or affective charge is one defining feature of social cognition (of. Fiske & Taylor, 1984; Markus &

S. Ki tayama

243

Zajonc, 1984), a process-oriented model of affect-cognition interaction, such as the one presented in the current work, will be a necessary theoretical element with which social cognitive processes are analyzed with a greater rigor and precision than is possible today. The present approach can be distinguished from a more traditional view of affect as a final product of what Lazarus (e.g., 1982; Smith & Lazarus, 1990) has called the cognitive appraisal. Echoing this view of affect as a consequence of cognition, many social psychological studies have described a variety of affective states as a function of prior cognition (e.g., Anderson, 1981; Smith & Ellsworth, 1986; Hastie et al., 1981; Higgms, 1987; Weiner, 1982). Although this program of research has revealed a great deal about an interaction between affect and cognition, with an important exception of the literature on the effects of mood on information processing (e.g., Blaney, 1986; Bower, 1981; Clark, 1982; Isen, 1984; Salovey & Bimbaum, 1990), a causal influence of affect on cognitive processes is frequently ignored (see Zajonc, 1980). In contrast, the current empirical findings and theoretical analysis seem to invite a more dynamic notion of affect-cognition interaction as recursive, mutual, and beginning very early in stimulus processing. The view of dynamic interaction between affect and cognition developed and defended in the present paper is quite congenial to recent connectionist theorizing in cognitive science (e.g., Rumelhart, 1989). More generally, it can be seen as one instantiation of the modularity hypothesis of mind (e.g., Fodor, 1983; Marr, 1976; ShaUice, 1988; see also Buck, 1987; LeDoux, 1987, for similar views as applied to emotion), which postulates any given psychological function, say, conscious perception, as an emergent property of a lawfial organization of elementary component processes (or "processing modules") such as preattentive processing, activation of affective circuits, and attentive processing. Further clarifying this mutual, dynamic interaction between affect and cognition, and then specifying its phylogenic precursors as well as patterns of ontogenic development in different social, cultural environments, presents a challenge for the future. References

Abdson, R. P. (1963). Computer simulation of "hot cognitions." In S. Tomkins & S. Mesick (Eds.), Computer simulation of personality. New York: Wiley.

244

Chapter .5

Allport, A. (1977). On knowing the meaning of words we are unable to report: The effects of visual masking. In S. Dornic (Ed.), Attention and performance FI. Hillsdale, NJ: Erlbaum. Allport, A. (1989). Visual attention. In M. I. Posner (Ed.), Foundaaons of cognitive science. Cambridge, MA: MIT Press. Anderson, N. (1980). Foundations of information integration theory. New York: Academic Press. Balota, D. A. (1983). Automatic semantic-activation and episodic memory encoding. Journal of Verbal Learning and Verbal Behavior, 22, 88104. Bargh, J. A. (1989). Conditional automatieity: Varieties of automatic influence in social perception and cognition. In J. S. Uleman & J. A. Bargh (Eds.), Unintended thought. New York: Guilford Press. Bargh, J. A., Bond, R. N., Lombardi, W. J., & Tota, M. E. (1986). The additive nature of chronic and temporary sources of construct accessibility. Journal of Personality and Social Psychology, .50, 869878. Bauer, R. M. (1984). Autonomic recognition of names and faces in prosopagnosia: A neuropsychological application of the guilty knowledge test. Neurophysiologia, 22, 456-469. Blaney, P. H. (1986). Affect and memory: A review. Psychological Bulletin, 99, 229-246. Blum, G. S. (1954). An experimental reunion of psychoanalytic theory with perceptual vigilance and defense. Journal of Abnormal and Social Psychology, 49, 94-98. Bootzin, R. R., & Natsoulas, T. (1965). Evidence for perceptual defense uncontaminated by response bias. Journal of Personality and Social Psychology, 5, 461-468. Bower, G. H. (1981). Mood and memory. American Psychologist, 36, 129148. Bower, G. H. (1990). Awareness, the unconscious, and repression: An experimental psychologist's perspective. In J. L Singer (Ed.), Repression

and dissociation: Implications for personality theory, psychopathology, and health. Chicago: The University of Chicago Press. Bower, G. H., & Clapper, J. P. (1989). Experimental methods in cognitive science. In M. I. Posner (Ed.), Foundations of cognitive science. Cambridge, MA: MIT Press. Braun, J., & Sagi, D. (1990). Vision outside the focus of attention. Perception and Psychophysics, 48, 45-58.

S. Kitayama

245

Brigham, J. C., Maass, A., Martinez, D., & Whittenberger, G. (1983). The effect of arousal on facial recognition. Basic and Applied Social Psychology, 4, 279-293. Broadbent, D. E. (1958). Perception and communication. London: Pergamon. Broadbent, D. E. (1973). In defense of empirical psychology. London: The Camelot Press. Broadbent, D. E., & Gregory, M. (1967). Perception of emotionally toned words. Nature, 215, 581-584. Bruner, J. (1957). On perceptual readiness. Psychological Review, 64, 123152. Bruner, J., & Potter (1964). Science, 144, 424-425. Buck, R. (1985). Prime theory: An integrated view of motivation and emotion. Psychological Review, 92, 389-413. Buck, R. (1988). Human motivation and emotion (2nd ed.). New York: Wiley. Cacioppo, J. T., Martzke, J. S., Petty, R. E., & Tassinary, L. G. (1988). Specific forms of facial EMG index emotions during an interview: From Darwin to the continuous flow hypothesis of affect-laden processing. Journal of Personality and Social Psychology, 54, 592-604. Carr, T. H., & Dagenbach, D. (1990). Semantic priming and repetition priming from masked words: Evidence for a center-surround attentional mechanism in perceptual recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 341-350. Carr, T. H., McCauley, C., Sperber, R. D., & Parmelee, C. M. (1982). Words, pictures, and priming: On semantic activation, conscious identification, and the automaticity of information processing. Journal of Experimental Psychology: Human Perception and Performance, 8, 757-777. Chapman, C. R., & Feather, B. W. (1972). Modification of perception by classical conditioning procedure. Journal of Experimental Psychology, 93, 338-342. Christianson, S.-A., & Loftus, E. F. (1991). Remembering emotional events: The fate of detailed information. Cognition and Emotion, 5, 81-108. Christianson, S.-A., Lottus, E. F., H o g a n , H., & Loftus, G. R. (1991). Eye fixations and memory for emotional events. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 693-701. Clark, H. H. (1973). The language-as-fixed-effect fallacy. Journal of Verbal Learning and Verbal Behavior, 12, 335-359.

246

Chapter 5

Clark, M. S., & Fiske, S. T. (1982). Affect and cognition: The 17th annual Carnegie symposium on cognition. Hillsdale, NJ: Erlbaum. Corteen, R. S., & Wood, B. (1972). Autonomic responses to shockassociated words in an unattended channel. Journal of Experimental Psychology, 94, 308-313. Crick, F. (1984). The function of the thalamic reticular complex: The search light hypothesis. The Proceedings of the National Academy of Sciences, 81, 4586-4590. Damasio, A. R., & Damasio, H., & Van Hoesen, G. W. (1982). Prosopagnosia: Anatomical basis and behavioral mechanisms. Neurology, 32, 331-341. Davidson, R. J. (1984). Affect, cognition, and hemispheric specialization. In C. E. Izard, J. Kagan, & R. Zajone (Eds.), Emotion, cognition, and behavior. New York: Cambridge University Press. Davidson, R. J., Ekman, P., Saron, C. D., Senulis, J. A., & Friesen, W. V. (1990). Approach-withdrawal and cerebral asymmetry: Emotional expression and brain physiology. Journal of Personality and Social Psychology, 58, 330-341. Dawson, M. E., & Schell, A. M. (1982). Electrodermal responses to attended and nonattended significant stimuli during dichotic listening. Journal of

Experimental Psychology: Human Perception and Performance, 8, 315-324. Derryberry, D., Brandt, J., & Reed, M. A. (1991). Motivational focusing of attention. Unpublished manuscript, Oregon State University, Corvalis, OR. Deutsch, J. A., & Deutsch, D. (1963). Attention: Some theoretical considerations. Psychological Review, 70, 80-90. Dixon, N. F. (1980). Preconscious processing. New York: Wiley. Dorfinan, D. D. (1967). Recognition of taboo words as a function of a priori probability. Journal of Personality and Social Psychology, 7, 1-10. Dorfinan, D. D., Grossberg, J. M., & Kroeker, L. (1965). Recognition of taboo stimuli as a function of exposure time. Journal of Personality and Social Psychology, 2, 552-562. Easterbrook, J. A. (1959). The effect of emotion on cue utilization and the organization of behavior. Psychological Review, 66, 183-201. Egeth, H. (1977). Attention and preattention. In G. H. Bower (Ed.), The psychology of learning and motivation. New York: Academic Press. Erdelyi, M. H. (1974). A new look at the new look: Perceptual defense and vigilance. Psychological Review, 81, 1-25.

S. Kitayama

247

Erdelyi, M. H. (1985). Psychoanalysis: Freud's cognitive psychology. New York: W. H. Freeman and company. Erdelyi, M. H., & Appelbaum, G. A. (1973). Cognitive masking: The disruption effect of an emotional stimulus upon the perception of contiguous neutral items. Bulletin of the Psychonomic Society, 1, 59-61. Eriksen, C. W. (1963). Perception and personality. In J. M. Wepman & R. W. Heine (Eds.), Concepts of personality. Chicago: Aldine. Eysenck, H. (1967). The biological basis of personality. Springfield: Thomas. Eysenck, M. W. (1976). Arousal, learning, and memory. Psychological Bulletin, 83, 389-404. Fazio, R. H., Sanbonmatsu, D. M., Powell, M. C., & Kardes, F. R. (1986). On automatic activation of attitudes. Journal of Personality and Social Psychology, 50, 229-238. Field, T. M. (1985). Neonatal perception of people: Maturation and individual differences. In T. M. Field & N. Fox (Eds.), Social perception in infants. Norwood, NJ: Ablex. Fiske, S. T. (1980). Attention and weight in person perception: The impact of negative and extreme behavior. Journal of Personality and Social Psychology, 38, 889-906. Fiske, S. T., & Taylor, S. E. (1984). Social cognition. Reading, MA: Addison-Wesley. Fodor, J. A. (1983). The modularity of mind Cambridge, MA: MIT Press. Fowler, C. A., Wolford, G., Slade, R., & Tassinary, L. (1981). Lexical access with and without awareness. Journal of Experimental Psychology: General 110, 341-362. Freeman, J. T. (1954). Set or perceptual defense. Journal of Experimental Psychology, 48, 282-288. Freud, S. (1895/1966). Project for scientific psychology. In J. Strachey (Ed. and Trans.), The standard edition of the complete psychological works of Sigmund Freud (Vol. 1). London: Hogarth Press. (Unpublished manuscript, originally entitled "Psychology for neurologists," completed in 1895). Frijda, N. (1986). The emotions. Cambridge: Cambridge University Press. Goldiamond, I. (1958). Indicators of perception: I. Subliminal perception, subception, unconscious perception: An analysis in terms of psychophysical indicator methodology. Psychological Bulletin, 55, 373411.

248

Chapter 5

Green, D. M., & Swets, J. (1966). Signal Detection Theory and psychophysics. New York: Wiley. Greenwald, A. G., Klinger, M. R., & Lui, T. J. (1989). Unconscious processing of diehoptieally masked words. Memory and Cognition, 17, 35-47. Grier, J. (1971). Nonparametric indexes for sensitivity and bias: Computing formulas. Psychological Bullean, 75, 424-429. Hansen, C. H., & Hansen, R. D. (1988). Finding the face in the crowd: An anger superiority effect. Journal of Personality and Social Psychology, 54, 917-924. Hastie, R., Ostrom, T. M., Ebbesen, E. B., Wyer, R. S., Hamilton, D. L., & Carlston, D. E. (Eds.), Person memory: The cognitive basis of social perception. Hillsdale, NJ: Erlbaum. Helmholtz, H. v. (1884). Ober das Sehen des Menschen. Vortrage und Reden, 1,365-396. Higgins, E. T. (1987). Self-discrepancy: A theory of relating self and affect. Psychological Review, 94, 319-340. Higgins, E. T., & Bargh, J. A. (1987). Social cognition and social perception. Annual Review of Psychology, 38, 369- 425. Holender, D. (1986). Semantic activation without conscious identification in dichotic listening, parafoveal vision, and visual masking: A survey and appraisal. Behavioral and Brain Sciences, 9, 1-66. Humphreys, M. S., & Revelle, W. (1984). Personality, motivation, and performance: A theory of the relationship between individual differences and information processing. Psychological Review, 91, 153-184. Isen, A. M. (1984). Toward understanding the role of affect in cognition. In R. S. Wyer & T. K. SruU (Eds.), Handbook of social cognition (Vol. 3). Hillsdale, NJ: Lawrence Erlbaum. Isen, A. M. (1987). Positive affect, cognitive processes, and social behavior. In L. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 20). New York: Academic Press. Isen, A. M., & Daubman, K. A. (1984). The influence of affect on categorization. Journal of Personality and Social Psychology, 47, 1206-1217. Isen, A. M., Shalker, T. E., Clark, M., & Karp, L. (1978). Affect, accessibility of material in memory, and behavior: A cognitive loop. Journal of Personality and Social Psychology, 36, 1-12.

S. Kitayama

249

Jacoby, L. (1983). Perceptual enhancement: Persistent effects of an experience. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 21-38. Johnston, W. A., & Dark, V. J. (1985). Selective attention. Annual Review of Psychology, 3 7, 43-75. Kahneman, D. (1973). Attention and effort. Englewood Cliffs, NJ: PrenticeHall. Kakar, S. (1978). The inner world: A psychoanalytic study of childhood and society in India. Delhi: Oxford University Press. Kihlstrom (1987). The cognitive unconscious. Science, 238, 1445-1452. Kihlstrom, J. F. (1990). The psychological unconscious. In L. A. Pervin (Ed.), Handbook of personality: Theory and research. New York~ondon: The Guilford Press. Kitayama, S. (1989). Affect and cognition: The role of frequency of occurrence in the perception of affectively toned words. Unpublished manuscript, University of Oregon, Eugene, OR. Kitayama, S. (1990). Interaction between affect and cognition in word perception. Journal of Personality and Social Psychology, 58, 209-217. Kitayama, S. (1991). Impairment of perception by positive and negative affect. Cognition and Emotion, 5, 255-274. Kitayama, S. (1996). Remembrance of emotional speech: Improvement and impairment of incidental verbal memory by emotional voice. Journal of Experimental Social Psychology, 32, 289-308. Kitayama, S., & Bumstein, E. (1989). The relation between opinion and memory: Distinguishing between associative density and structural centrality. In J. Bassili (Ed.), On-line cognition in person perception. Hillsdale, NJ: Erlbaum. Kitayama, S., & Howard, S. (1994). Affective regulation of perception and comprehension. In P. M. Niedenthal & S. Kitayama (Eds.), The heart's eye: Emotional influences in perception and attention. New York: Academic Press. Kleinsmith, L. J., & Kaplan, S. (1963). Paired-associate learning as a function of arousal and interpolated interval. Journal of Experimental Psychology, 65, 190-193. Kucera, H., & Francis, W. N. (1967). Computational analysis of presentday American English. Brown University Press. Kunst-Wilson, W. R., & Zajonc, R. B. (1980). Affective discrimination of stimuli that cannot be recognized. Science, 207, 557-558.

250

Chapter 5

LaBerge, D. (1983). Spatial extent of attention to letters and words. Journal

of Experimental Psychology: Human Perception and Performance, 9, 371-379. LaBerge, D., & Brown, V. (1989). Theory of attentional operations in shape identification. Psychological Review, 96, 101-124. LaBerge, D., & Samuels, J. (1973). Toward a theory of automatic information processing in reading. Cognitive Psychology, 6, 292-323. Lacey, J. I., & Laeey, B. C. (1968). Verification and extension of the principle of autonomic response sterotypy. American Journal of Psychology, 71, 50-73. Lacey, O. W., Lewinger, N., & Adamson, J. E. (1953). Foreknowledge as a factor affecting perceptual defense and alertness. Journal of Experimental Psychology, 45, 169-174. Lang, P. (1984). Cognition in emotion: Concept and action. In C. Izard, J. Kagan, & R. B. Zajone (Eds.), Emotion, cognition, and behavior. New York: Cambridge. Lang, P., Bradley, M. M., & Cuthbert, B. N. (1990). Emotion, attention, and the startle reflex. Psychological Review, 97, 377-395. Lawrence, D. H., & Cole, G. R. (1954). Accuracy of recognition with alternatives before and after the stimulus. Journal of Experimental Psychology, 47, 208-214. Lazarus, R. S. (1982). Thoughts on the relations between emotion and cognition. American Psychologist, 3 7, 1019-1024. Lazarus, R. S., & MeCleary, R. A. (1951). Autonomic discrimination without awareness: A study of subception. Psychological Review, 58, 113-122. LeDoux, J. E. (1987). Emotion. In F. Plum (Ed.), Handbook of Physiology,

Section I. The nervous system. Vol. V: Higher functions of the brain. Bethesda: American Physiological Society. LeDoux, J. E. (1989). Cognitive-emotional interactions in the brain. Cognition and Emotion, 3, 267-289. Lewicki, P. (1986). Nonconscious social information processing. New York: Academic Press. Lindsley, D. B. (1951). Emotion. In S. S. Stevens (Ed.), Handbook of experimental psychology. New York: Wiley. Loffus, E. F. (1979). Eyewitness testimony. Cambridge, MA: Harvard University Press. McGinnies, E. M. (1949). Emotionality and perceptual defense. Psychological Review, 56, 471-482.

S. Ki tayama

251

MacLeod, C., Mathews, A., & Tata, P. (1986). Attentional bias in emotional disorder. Journal of Abnormal Psychology, 95, 15-20. Marcel, A. J. (1983a). Conscious and unconscious perception: Experiments on visual masking and word recognition. Cogmtive Psychology, 15, 197-237. Marcel, A. J. (1983b). Conscious and unconscious perception: An approach to the relation between phenomenal experience and perceptual processes. Cognitive Psychology, 15, 238-300. Markus, H. R., & Kitayama, S. (1991). Culture and the self: Implications for cognition, emotion, and motivation. Psychological Review, 98, 224-253. Markus, H., & Zajonc, R. B. (1985). The cognitive perspective in social psychology. In G. Lindzey, & E. Aronson (Eds.), Handbook of social psychology (3rd r Vol. 2). New York: Random House. Marr, D. (1976). Early processing of visual information. Philosophical Transactions of the Royal Society of London G, 275, 483-524. Mathews, A., & Wertheimer, M. (1958). A "pure" measure of perceptual defense uncontaminated by response suppression. Journal of Abnormal and Social Psychology, 57, 373-376. Matthews, G., Pitcaithly, D., & Mann, R. E. (1995). Mood, neuroticism, and the encoding of affective words. Cognitive Therapy and Research, 19, 563-587. Merikle, P. M. (1982). Unconscious processing revisited. Perception and

Psychophysics, 31,298-301. Minard, J. G. (1965). Response-bias interpretation of "perceptual defense": A selective review and an evaluation of recent research. Psychological Review, 72, 74-88. McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: Part 1. An account of basic findings. Psychological Review, 88, 375-407. Moray, N. (1959). Attention in dichotic listening: Affective cues and the influence of instructions. Quarterly Journal of Experimental Psychology, 11, 56-60. Mozer, M. C. (1988). A connectionist model of selective attention in visual perception. Proceedings of the Annual Meeting of the Cognitive Science Society, 195-201. Murphy, S. T., & Zajonc, R. B. (1990). Affecave primacy and affective priming of preferences. Unpublished manuscript, The University of Michigan.

252

Chapter 5

Natsoulas, T. (1965). Converging operations for perceptual defense. Psychological Bulletin, 64, 393-401. Navon, D. (1984). Resources - a th~retical soupstonr Psychological Review, 91, 216-234. N~ly, J. (1977). Somantie priming and retrieval from loxical memory: Roles of inhibitionloss spreading activation and limited-capacity attention. Journal of Experimental Psychology: General, 106, 226-254. Newr A., Rosenbaum, P. S., & Laird, J. E. (1989). Symbolic architectures for cognition. In M. I. Posner (Ed.), Foundaaons for cognitive science. Cambridge, MA: MIT Press. Niedenthal, P. M. (1990). Implicit perception of affoctive information. Journal of Experimental Social Psychology, 26, 505-527. Norman, D. A. (1968). Toward a theory of memory and attention. Psychological Review, 75, 522-536. Ohman, A. (1985). Face the beast and fear the face: Animal and social fears as prototypes for evolutionary analyses of emotion. Psychophysiology, 23, 123-145. Osgood, C. E. (1962). Studies on the generality of affectivr meaning systems. American Psychologist, 17, 10-28. Paulhus, D., & Lovit, K. (1987). Desirable responding triggered by affect: Automatic egotism? Journal of Personality and Social Psychology, 52, 245-259. Pollak, I., & Norman, D. A. (1964). A nonparametrie analysis of recognition experiments. Psychonomic Science, 1, 125-126. Posner, M. I. (1978). Chronometric explorations of mind. New York: Oxford University Press. Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Experimental Psychology, 32, 3-25. Posner, M. I. (Ed.) (1989). Foundaaons of cognitive science. Cambridge, MA: MIT Press. Posner, M. I., & Petorson, S. E. (1990). The attention system of the human brain. Annual Review of Neuroscience, 13, 25-42. Posner, M. I., Potorson, S. E., Fox, P. T., & Raichle, M. E. (1988). Localization of cognitive operations in the human brain. Science, 240, 1627-1731. Posnor, M. I., & Snyder, C. R. R. (1975). Attention and cognitive control. In R. Solso (Ed.), Informaaon processing and cognition: The Loyola Symposium. HiUsdalr NJ: Erlbaum.

S. Ki tayama

2 53

Postman, L., Bronson, W. C., & Gropper, G. L. (1953). Is there a mechanism of perceptual defense? Journal of Abnormal and Social Psychology, 48, 215-224. Pratto, F., & John, O.P. (1991) Automatic vigilance: The attention-grabbing power of negative social information. Journal of Personality and Social Psychology, 61, 380-391. Rapaport, D. (1967). The theory of attention cathcxis: An economic and structural attempt at the explanation of cognitive processes. In M. M. Gill (Ed.), The collected papers of David Rapaport. New York: Basic Books. Rcvcllc, W., & LoRus, D. A. (1990). Individual differences and arousal: Implications for the study of mood and memory. Cognition and Emotion, 4, 209-237. Roland, A. (1988). In search of self in India and Japan: Toward a crosscultural psychology. Princeton: Princeton University Press. Roscman, I. (1984). Cognitive determinants of emotion: A structural theory. In P. Shaver (Ed.), Review of personality and social psychology: emotions, relationships, and health. Beverly Hills, CA: Sage. Rumclhart, D. E. (1989). The architecture of mind: A conncctiomst approach. In M. I. Posncr (Ed.), Foundations of cogmave science. Cambridge, MA: MIT Press. Russell, J. A. (1980). A circumplcx model of affect. Journal of Personality and Social Psychology, 39, 1161-1178. Sagi, D., & Julcsz, B. (1985). "Where" and "what" in vision. Science, 228, 1217-1219. Sales, B. D., & Habcr, R. (1968). A different look at perceptual defense for taboo words. Perception & Psychophysics, 3, 156- 160. Salovcy, P., & Bimbaum (1990). Influence of mood on health-relevant cognitions. Journal of Personality and Social Psychology, 57, 539-551. Schachtcr, S., & Singer, J. E. (1962). Cognitive, social and physiological determinants of emotional state. Psychological Review, 69, 379-399. Schcrcr, K. R. (1984). On the nature and function of emotions: A component process approach. In K. R. Schcrcr & P. Ekman (Eds.), Approaches to emotion. Hillsdalc, NJ: Erlbaum. Schcrcr, K. R. (1986). Vocal affect expression: A review and a model for future research. Psychological Bulletin, 99, 143-165. Schneider, D. J., Hastorf, A. H., & EUsworth, P. C. (1979). Person perception (2nd exi.). Reading, MA: Addison-Wesley.

254

Chapter 5

Schneider, W. (1988). Micro experimental laboratory: An integrated system for IBM PC compatibles. Behavior Research Methods, Instruments, & Computers, 20, 206-207. Schwartz, G. E. (1990). Psychobiology of repression and health: A systems approach. In J. L Singer (Ed.), Repression and dissociation:

Implications for personality theory, psychopathology, and health. Chicago: The University of Chicago Press. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge, England: Cambridge University Press. Shevrin, H., Smith, W. H., & Fritzler, D. E. (1972). Average evoked response and verbal correlates of unconscious mental processes. Psychophysiology, 8, 149-162. Shevrin, H. (1990). Subliminal perception and repression. In J. L Singer (Ed.), Repression and dissociation: ImpBcatiom for personality theory, psychopathology, and health. Chicago: The University of Chicago Press. Smith, C., & Ellsworth, P. C. (1987). Patterns of appraisal and emotion related to taking an exam. Journal of Personality and Social Psychology, 52, 475-488. Smith, C., & Lazarus, R. S. (1990). Emotion and adaptation. In L. A. Pervin (Ed.), Handbook of personality: Theory and research. New York: The Guilford Press. Spence, K. W. (1956). Behavior theory and conditioning. New Haven, CT: Yale University Press. Stembach, R. A. (1966). Principles of psychophysiology. New York: Academic Press. Tassinary, L. G., Orr, S. P., Wolford, G., Napps, S. E., & Lanzetta, J. T. (1984). The role of awareness in affective information processing: An exploration of the Zajonc hypothesis. Bulletin of the Psychonomic Society, 22, 489-491. Taylor, S. E. (1991). Asynunetrieal effects of positive and negative events: The mobilization-minimization hypothesis. Psychological Bulletin, 110, 67-85. Tomldns, (1962). Affect, imagery and consciousness. Vol. 1: The posiave affects. New York: Springer. Tomkins, (1980). Affect as amplification: Some modification in theory. In R. Plutchik & H. Kellerman (Eds.), Emotion: Theory, research, and experience. New York: Academic Press.

S. Kitayama

255

Tranel, D., & Damasio, A. R. (1985). Knowledge without awareness: An autonomic index of facial recognition by prosopagnosics. Science, 228, 1453-1454. Treisman, A. M. (1964). Verbal cues, language, and meaning in selective attention. American Journal of Psychology, 77, 206-219. Treisman, A. M. (1988). Features and objects: The fourteenth Bartlett Memorial Lecture. Quarterly Journal of Experimental Psychology, 40,4, 201-237. Tucker, D. M., & Williamson, P. A. (1984). Asymmetric neural control systems in human self-regulation. Psychological Review, 91, 185-215. Uleman, J. S., & Bargh, J. A. (1989). Unintended thought. New York: Guilford Press. Van Egeren, L. (1968). Repression and sensitization: Sensitivity and recognition criteria. Journal of Experimental Research in Personality, 3, 1-8. Walker, E. L. (1958). Action decrement and its relation to learning. Psychological Review, 65, 129-142. Walker, E. L., & Tarte, R. D. (1963). Memory storage as a function of arousal and time with homogeneous and heterogeneous lists. Journal of Verbal Learning and Verbal Behavior, 71, 190-193. Weinberger, D. A. (1990). The construct validity of the repressive coping style. In J. L Singer (Ed.), Repression and dissociation: Implications for personality theory, psychopathology, and health. Chicago: The University of Chicago Press. Weiner, B. (1982). The emotional consequences of causal ascriptions. In M. S. Clark. & S. T. Fiske (Eds.), Affect and cognitmn. Hillsdale, NJ: Erlbaum. Wike, E. L., & Church, J. D. (1976). Comments on Clark's "The languageas-fixed-effect faUaey." Journal of Verbal Behavior & Verbal Learning, 15, 249-255. Worthington, A. G. (1969). Paired comparison sealing of brightness judgments" A method for the measurement of perceptual defense. British Journal of Psychology, 60, 363-368. Yerkes, R. M., & Dodson, J. D. (1908). The relation of strength of stimulus to rapidity of habit formation. Journal of Comparative Neurology and Psychology, 18, 459-482. Zajonc, R. B. (1962). Response suppression in perceptual defense. Journal of Experimental Psychology, 64, 206-214. Zajonc, R. B. (1965). Social facilitation. Science, 149, 269-274.

256

Chapter 5

Zajonc, R. B. (1980). Feeling and thinking: Preferences need no inferences. American Psychologist, 35, 151-175. Zajonc, R. B., & Dorfinan, D. D. (1964). Perception, drive, and behavior theory. Psychological Review, 71,273-290. Zajonc, R. B., Murphy, S. T., & Ingelhart, M. (1989). Feeling and facial efference: Implications of the vascular theory of emotion. Psychological Review, 96, 395-416. Author Notes

This research was supported by National Institute of Mental Health Grant 1R01 MH50117-01. This chapter was completed while the author was a fellow at the Center for Advanced Study in the Behavioral Sciences.

S. Kitayama

257

A p p e n d i x A . W o r d p a i r s u s e d in E x p e r i m e n t 1.

Long words Positive

Negative

FASCINATING - OUTSTANDING

COMFORTABLE

ATTRACTIVE - SATISFYING

D E D I C A T I O N - ENTHUSIASM

-

MAGNIFICENT

EXCELLENCE - DELIGHTFUL

F A V O R A B L E - PROMINENT

FANTASTIC - WONDERFUL

B R I L L I A N T - DESIRABLE

HAPPINESS

MARVELOUS - INTEGRITY

-

HONEYMOON

D E S T R U C T I O N - THREATENING

A L I E N A T I O N - DISTURBING

INADEQUATE - CONSPIRACY

RESENTMENT - CORRUPTION

ISOLATION

-

CONDEMNED

CRITICISM

-

PRISONERS EMERGENCY

S U F F E R I N G - DESTROYED

DANGEROUS

T R E M B L I N G - EXECUTION

R E V O L U T I O N - SLAUGHTER

-

V I O L E N C E - WEAKNESS Neutral

C O N C E P T I O N - IMPRESSION

INTENTIONS - OBJECTIVES

MECHANISM - APPARATUS

D I M E N S I O N - MAGNITUDE

A L T E R N A T I V E - MEASUREMENT

P R O B A B I L I T Y - TRANSLATION ARRANGEMENT - MANUFACTURE

APPLICABLE - IN ITATANT RESIDENTS - FORMATION

VARIABLES - PERTINENT

Short words Positive

Negative

Neutral

C O M E D Y - WISDOM

T A L E N T - MATURE

L U C K Y - HUMOR

SMILE - CHARM

FUNNY - AWARD

T R U S T - ENJOY

G L O R Y - PRIZE

P R O U D - EAGER

JOKE

-

CASH

C A N C E R - INJURY

DAMAGE - HATRED

V I C T I M - TERROR

D E V I L - PANIC

ERROR

F A L S E - BLAME

FAULT - GUILT SNAKE - CRIME

ANGRY WASTE - STORM H U N G - UGLY

B O R D E R - MARGIN

LOCATE - DETECT

S T A M P - LABEL

T R A C K - ROUTE

S P A R E - EXTRA

SWITCH - CUSTOM

HABIT - TREND SHEET

-

FENCE

-

C H A I R - TRACE PAUSE

-

TREND

F O O L - PALE SHAME

-

WORSE

S T O N E - PANEL W I R E - BONE

Chapter 5

258

A p p e n d i x B. W o r d p a i r s u s e d in E x p e r i m e n t

1.

Long words Positive Neutral

FASCINAIqNG - ALTERNATIVE COMFORTABLE - RESIDENTIAL A T T R A C T I V E - MECHANICAL DEDICATION FAVORABLE

-

M A G N I H C E N T - SURROUNDING SATISFYING - OCCASIONAL

OBJECTIVES

-

EXCELLENCE

OUTSTANDING - REPRESENTED

-

E N T H U S I A S M - INTENTIONS

APPLICABLE

DELIGHTFUL - INHABITANT

SCHEDULED

HAPPINESS

FANTASTIC - DIMENSION B R I L L I A N T - RECORDING

N e g a t i v e

-

N e u t r a l

-

ASSEMBLED

W O N D E R F U L - MAGNITUDE

H A P P I N E S S - VARIABLES

D E S I R A B L E - FORMATION HONEYMOON- COMPONENT

M A R V E L O U S - PERTINENT

I N T E G R I T Y - RESIDENTS

DESTRUCTION - ALTERNATIVE ALIENATION - OBJECTIVES I N A D E Q U A T E - MECHANICAL RESENTMENT - APPLICABLE

T H R E A T E N I N G - REPRESENTED DISTURBING - OCCASIONAL C O N S P I R A C Y - INTENTIONS C O R R U P T I O N - SURROUNDING

D E S T R U C T I O N - RESIDENTIAL

S L A U G H T E R - MAGNHZIDE

ISOLATION - A S S E M B L E D C O N D E M N E D - SCHEDULED

C R I T I C I S M - DIMENSION PRISONERS

S U F F E R I N G - FORMATION

D E S T R O Y E D - VARIABLES

DANGEROUS - PERTINENT T R E M B L I N G - RECORDING

E M E R G E N C Y - COMPONENT E X E C U T I O N - INHABITANT

-

RESIDENTS

Short words

TRUST - HABIT

T A L E N T - LOCATE H U M O R - LABEL F U N N Y - SPARE E N J O Y - TREND

GLORY - CHAIR

P R I Z E - TRACE

P R O U D - EXTRA

EAGER

J O K E - WIRE

C A S H - BONE

C A N C E R - BORDER

I N J U R Y - MARGIN

DEVIL - STAMP

D A M A G E - LOCATE

H A T R E D - DETECT

V I C T I M - SWITCH

TERROR

PANIC - LABEL

E R R O R - TRACK BLAME - EXTRA

C O M E D Y - BORDER MATURE - DETECT

Positive Neutral

S M I L E - TRACK AWARD - STONE

N e g a t i v e

N e u t r a l

-

-

PANEL

-

CUSTOM

A N G R Y - ROUTE FAULT - HABIT S T O R M - TRACE WORSE - STONE FOOL - SEED

W I S D O M - MARGIN L U C K Y - STAMP CHARM- ROWI~

FALSE

-

SPARE

G U I L T - TREND S N A K E - FENCE H U N G - WIRE PALE - GEAR

WASTE - SHEET C R I M E - PANEL U G L Y - BONE

Cognitive Science Perspectives on Personality and Emotion- G. Matthews(Editor) 9 1997 Elsevier Science B.V. All rights reserved. CHAPTER 6

Levels of Processing in Emotion-Antecedent Appraisal Carien M. van Reelcum and Klaus R. Scherer

Given the widely different scope of phenomena that are currently described by the word "emotion" - ranging from fleeting preferences to existential life crises - it seems advisable to propose a brief working definition of the psychological construct to be dealt with in this chapter. Following a suggestion by Soberer (1993a), we define an emotion episode as a sequence of interrelated, synchronized changes in the state of all organismic subsystems in response to the evaluation of an external or internal stimulus event as relevant to central concerns of the organism. As highlighted in the definition, the evaluation or appraisal of an event with respect to its relevance to the individual is seen as responsible for the elicitation and differentiation of emotion. This chapter is concerned with the nature of such "emotionconstitutive" appraisals and the role of individual differences. Lazarus (1968) and Arnold (1960) were the first to elaborate on the notion of "appraisal" in an effort to explain and differentiate between the positive and negative emotional consequences of an event. Following their pioneering work, a large number of "appraisal theories of emotion" have been developed in an attempt to predict the elicitation and differentiation of emotion on the basis of a detailed set of appraisal criteria (De Rivera, 1977; Frijda, 1986; Oatley & Johnson-Laird, 1987; Lazarus, 1991; Roseman, 1984, 1991; Soberer, 1982, 1984, 1986, 1988; Smith & Ellsworth, 1985; Solomon, 1976; Weiner, 1982, 1986). All of these authors consider the appraisal process generally as a subjective, cognitive evaluation of potentially emotion-eliciting events. In spite of divergent disciplinary and historical traditions of the authors involved, one finds a high degree of convergence with respect to the nature of the appraisal dimensions or criteria postulated by different theories (see Lazarus & Smith, 1988; Manstead & Tetlock, 1989; Reisenzein & Hofmann, 1990, 1993; Roseman, Spindel, & Jose, 1990; Scherer, 1988). These include the perception of a change in the environment that captures the subject's attention (novelty and expectancy), the perceived pleasantness or unpleasantness of the stimulus or event (valence), the importance of the stimulus or event to one's goals or concerns (relevance and goal

260

Chapter 6

conduciveness or motive consistency), the notion of who or what caused the event (agency or responsibility), the estimated ability to deal with the event and its consequences (perceived control, power or coping potential), and the evaluation of one's own actions in relation to moral standards or social norms (legitimacy), and one's self-ideal. At the very least, this convergence on a standard set of appraisal criteria suggests high face validity of the underlying assumptions. In addition, a number of empirical studies have provided support for the idea that a small number of relatively general appraisal dimensions or criteria can explain the differentiation of major emotions. A variety of different paradigms has been used to study the relationship between particular configurations of appraisal results and the nature of the ensuing emotional reaction. The results generally confirm the fundamental assumptions of appraisal theories as well as specific predictions about emotion-specific appraisal profiles. In general, appraisal profiles allow the discrimination of 40 to 50% of the emotion episodes studied (see Scherer, 1997 for an overview of this research).

Critique of Appraisal Notions Despite substantial empirical evidence demonstrating the explanatory value of appraisal theory in emotion differentiation, the appraisal approach has been heavily criticized with respect to both its presumed emphasis on cognitive, i.e. conscious or voluntary, processes (e.g. Zajonc, 1980, 1984; Berkowitz, 1994), and the research methods generally used to study appraisal in emotion (particularly the emphasis on verbal reports of presumed appraisal processes in emotion experiences recalled from memory). The critique of exaggerated cognitivism is in large part based on the kinds of experimental paradigms used in this research tradition: 1) asking subjects to recall specific emotional experiences and questioning them about the outcome of antecexlent evaluation processes (Ellsworth & Smith, 1988; Folkman & Lazarus, 1988; Frijda, Kuipers, & ter Schure, 1989; Gehm & Scherer, 1988; Mauro, Sato, & Tucker, 1992; Reisenzein & Hofinann, 1993; Reisenzein & Spielhofer, in press; Roseman, Spindel, & Jose, 1990; Smith & Ellsworth, 1985; Tesser, 1990); 2) obtaining judgments on appraisal processes after naturally occurring or emotion-producing events such as examinations or emotions induced experimentally (Folhnan & Lazarus, 1985; Smith, 1989; Smith & Ellsworth, 1987);

C.M. van Reekum and K.R. Scherer

261

3) having emotion words judged as to the appraisal implications of the underlying concepts (Conway & Bekerian, 1987; Frijda, 1987; Parkinson & Lea, 1991; Smolenaars & Schutzela.ars, 1986/87), and; 4) using vignettes or scenarios that have been systematically manipulated with respect to appraisal relevant dimensions and asking subjects to indicate the emotional reactions that they - or a fictitious other might experience in this situation (McGraw, 1987; Roseman, 1984; Russel & McAuley, 1986; Smith & Lazarus, 1993; Stipek, Weiner, & Li, 1989; Weiner, Amirkhan, Folkes, & Verette, 1987; Weiner, Graham, & Chandler, 1982; Weiner, Russel, & Lerman, 1979). In all of these cases, participants are required to engage in conscious, complex inference or imagination processes, followed by verbalization. Clearly, all of these processes require a fairly high level of conceptual processing. It has been argued (Parkinson & Manstead, 1992, 1993; Parkinson, 1997) that participants are unlikely to be able to report upon antecedent appraisal processes which mostly occur outside awareness. Furthermore, these critics point out that the very process of imagining an emotional event depends to a large extent upon cognitive interpretation and memory. In such circumstances, participants are likely to construct a rationale for their emotional response, a situation which is possibly made worse by providing the participants with ready made dimensions of appraisal. These problems parallel those encountered by Cognitive psychologists and knowledge engineers when attempting to extract information from experts about problem-solving and decision-making strategies which are often overtrained and therefore implicit. It has been shown that experts consistently report having used a logical strategy which bears only limited resemblance to the one they actually used (see Berry, 1987, for an excellent summary of this literature). Furthermore, Frijda (1993) has argued that the empirical evidence supporting appraisal, rather than describing emotion antecedent processing, has to date been limited to describing the content of an emotion or the meaning of the respective labels. Specifically, post-hoc analyses of an emotional event are susceptible to logical inferences of why an emotion arose, but may not include an accurate description of the actual antecedent evaluation which occurs in part outside of awareness. While these criticisms are well taken, they reflect the difficulties faced by any attempt to observe and measure cognitive processes. Since thoughts are neither directly observable nor measurable, the only available access so far is to request individuals to verbalize whatever parts of their mental processes are available to consciousness and can be encoded in language.

262

Chapter 6

Cognitive psychologists have used ingenious experimental paradigms using systematic variations of exposure conditions and reaction times to study cognitive processes without recourse to conscious verbal report. However, they have mostly studied formal aspects of cognitive operations often using highly formalizexl rule systems such as syntactic and semantic structures. Even in these highly constrained cases, the use of reaction time as a "royal road" to understanding cognitive processing has been called into question since its interpretation is often highly conjectural (see Scherer, 1992). Appraisal theorists face the additional complication that the evaluation and inference processes studied are largely independent of language or other formally structured systems and often occur outside of awareness. Thus, the issue here is not to study formal aspects of the respective operations, but the assessment of the content of the appraisal processes. Clearly, there are, so far, few alternatives to studying these processes other than through the use of verbal report. In addition, it is probably fair to state that the critics of appraisal theory have so far failed to suggest convincing alternatives to conceptualizing and measuring the elieitation and differentiation of the entire gamut of emotional states. The theoretical statements concerning the structural components of the appraisal profiles underlying specific emotional states and the empirical evidence obtained through verbal report so far have laid solid foundations for further work in this area. In summary, the reliance upon questionnaire data, which by its very nature reflects high-level, conscious processes, combined with the large number of relatively sophisticated appraisal dimensions typically posited, has led many critics to erroneously deduce that appraisal processes are necessarily deliberate, conscious and thus "eognitivistie". In addition, many of the appraisal theories are worded in such a way as to reinforce this misconception. However, some appraisal theorists have pointed out quite explicitly that appraisal processes are not necessarily conscious or voluntary (e.g. Frijda, 1993; Lazarus, 1984; Seherer, 1984b). In fact, many of the criticisms discussed above seem to stem from the inherent ambiguity associated with the word "cognition", and the lack of a consensual definition thereof. What is evidently needed in this situation is not a debate concerning the semantics of the word "cognition" but rather a recognition that different types or levels of processing are implicated in emotion-antecedent appraisal. In this spirit, Leventhal and Scherer (1987) have suggested the adoption of a "levels of processing" approach to more precisely specify the mechanisms underlying appraisal, trying to correct the misconception that appraisal is predominantly a high-level, conscious, or controlled cognitive activity.

263

C.M. van Reekum and K.R. Scherer

Levels of Processing in Appraisal Leventhal and Soberer (1987), in trying to demonstrate the futility of "cognition-vs.-emotion" debates, integrated their respective approaches; on the one hand Leventhal's (1979, 1984) perceptual-motor theory, suggesting that emotions are produced by the constructive activity of a multi-component, hierarchical processing system; and, on the other hand, Scherer's (1982, 1984a) component process theory, arguing for emotion elicitation and differentiation through a sequence of stimulus evaluation or appraisal "checks" (SECs). Specifically, they suggested that appraisal can occur at three different levels - sensory-motor, schematic, and conceptual. Table 1 reproduces the central figure from Leventhal and Scherer (1987, p. 17) suggesting a preliminary model of appraisal occurring at different levels of processing.

Table 1. Levels of processing for Stimulus Evaluation Checks. ,=

Novelty Conceptual Level

Pleasantness Goal/need Conducive

Expectations: Recalled, cause/effect, anticipated, probability or derived estimates positive-

Conscious goals, plans

Coping Potential L ,

Norm/self

Problem solving ability

Self ideal, moral evaluation

Cgmpatib!lity

negative evaluations Schematic Level

Familiarity: Learned Acquired schemata preferences/ needs, matching aversions motives

Body schemata

Self/social schemata

Sensorymotor Level

Sudden, Innate Basic needs intense preferences/ stimulation aversion

Available energy

(Empathic adaptation7)

Note. ReproducedfromLeventhal and Scherer, 1987, p. 17.

Evaluation of events at the sensory-motor level mostly involves processing by innate, unconditioned, hard-wired feature detectors, giving rise to reflex-like, often uncontrolled reactions. The underlying assumption is that,

264

Chapter 6

through evolution, the sensory-motor mechanisms were tuned in such a manner as to quickly recognize and respond to intrinsically harmful or beneficial stimuli. The rudimentary adaptive responses, in particular orienting and defense reactions, are generated without the interference of more controlled processes to ensure rapid adaptive responses. Affective responses exclusively elicited through sensory-motor processing are likely to be short-lived and are often of a reflex-like nature as in the ease of startle (Lang, Bradley, & Cuthbert, 1990). One can ask to what extent such response patterns constitute "true" emotions in the sense of the definition suggested above, i.e. involving changes in all organismic subsystems in a synchronized fashion (see also Leventhal & Soberer, 1987). Even though this might be expected in some eases, most adult emotions observed in daily life are unlikely to be exclusively accounted for by sensorymotor processing. The schematic level of processing is based on structures acquired during the past learning experiences of an organism, which have ot~en been called "schemata". Bartlett (1932) described the schema as the active, organized setting within which new experiences are influenced by those previous reactions and experiences that are connected by some common aspect. These schemata are expected to be of central importance for the accommodation to or assimilation of objects and events (see Piaget, 1937/1955). With respect to schemata created in emotional encounters, Leventhal and Scherer (1987, p. 10) suggested that they "are concrete representations in memory of specific perceptual, motor (expressive, approach-avoidance tendencies, and autonomic reactions), and subjective feelings each of which were components of the reactions during specific emotional episodes". Newly experienced events or stimulus patterns can lead to the activation of schemata which an organism has acquired throughout its life, and responses are likely to be generated largely in accordance with the reaction components of the existing schema. The conceptual level of processing is characterized by more abstract, active, reflective processes, as opposed to the template matching-like processes which were specific to the first two levels. This level comprises capacities to infer generalities from and reason about events on the basis of propositionally organized memory structures, which are developed by comparisons between past experiences. Processes such as anticipating or problem-solving are typical for the conceptual level. While inherently more effortful and less "automatic" than the preeexting two levels, Leventhal and Scherer (I 987) expect this type of processing to become, with maturation of

C.M. van Reekum and K.R. Scherer

265

the individual and performance experience, "increasingly rapid, automatic, and 'lazy and mindless' (Langer, Blank, & Chanowitz, 1978)". In other words, conceptual processing of frequently encountered situations is thought to become schematic and thus conceptual "content" can become "schematized". Leventhal and Scherer (1987, p. 22) argue that: Indeed, we expect most if not all emotional processing to be initiated at the middle, or schematic level. Experience with specific persons, objects, and events develops schemata which encapsulate their history in perceptual memories or "identities" (Hebb, 1949) that then organize current emotional experience (Bruner, 1957; Yates, 1985). One of the criticisms that is frequently raised with respect to Leventhal and Scherer's proposal concerns the assumption that all checks can occur at all levels. It may seem odd, indeed, to claim that a painful stimulus is evaluated as obstructive to an individual's "central concern" (Berkowitz, personal communication). As in the classic Lazarus-Zajonc debate, the problem seems to be one of semantics. Most of the terms used in psychology unfortunately have a strong cognitivistic or mentalistic bent, obscuring the fact that the underlying processes can be very primitive indeed. The fact that a reaction to pain may be very automatic or reflexive does not contradict the idea that these defensive reflexes are at the service of very fundamental needs of the organism such as being unharmed, dry, warm, and fed. The "central concerns" of the organism thus range all the way from very basic motivational constructs (such as basic needs) to the most elaborate goals and plans (of. Maslow's, 1970, hierarchy of motives). Unfortunately, the neglect of motivation in psychology and the diffmulties inherent in functional argumentation is responsible for the relative lack of sophistication in our treatment of these motivational phenomena. Leventhal and Scherer made a preliminary attempt to label the appraisal criteria in a manner that is appropriate to each level. For example, with respect to the goal conduciveness check, they mention "basic needs" for the sensorimotor level, "acquired needs, motives" for the schematic level, and "conscious goals, plans" for the conceptual level. This exercise proved to be rather difficult, particularly for the coping potential and norm/self compatibility checks (see Table 1), and further efforts are required to elaborate on these concepts. Leventhal and Scherer's proposition of a levels of processing model, even though preliminary and highly conjectural, as emphasized by the

266

Chapter 6

authors, recognized the need to address different levels of emotion-anteceAent appraisal and argued for a new approach to studying appraisal. Although this early suggestion provided a potential mechanism to explain the more or less automatic elicitation and differentiation of emotion, the adoption of multilevel frameworks by appraisal theorists remains rather limited to date. One of the possible reasons might be that most theorists found it difficult to go beyond the rather general suggestion by Leventhal and Scherer, given the uncharted nature of the territory and the difficulty of operationalizing the concepts for empirical study. Appraisal theory, although claiming to be a cognitive approach to emotion, has generally not contributed much to the discussion of exactly which cognitive processes underlie appraisal. For instance, attentional processes (i.e. the allocation of resources to a potentially significant stimulus and the inhibition of irrelevant information sources ot~en controlled by the general concerns of the individual; see Williams, Mathews, & Macl_exxt, 1996) are not addressed by appraisal theorists even though they emphasize the evaluation of a situation with respect to its relevance to one's goals. The adoption of more elaborate cognitive processing models would seem to be the next logical step in the development of appraisal theories in order to better explain the nature of the processes underlying the elicitation and differentiation of emotion and to thus address some of the criticism of "classical" appraisal approaches mentioned above. Hierarchical Process Notions in Related Traditions

Contrary to the relative neglect of hierarchical process models by appraisal theorists, these notions have received strong attention from theorists and researchers concerned with affective phenomena in the areas of social cognition, clinical psychology, and neuropsychology. In envisaging a hierarchical, process-oriented model for appraisal, it is useful to review the approaches which might shed light on aspects of emotion-antecedent cognitive processing. There are a number of different research traditions, developed quite independently of one another and characterized by emphasizing specific aspects of the emotion process, that qualify for this review. In what follows, we will outline some of the approaches that seem directly pertinent to the issue of levels in emotion-antecextent processing.

C.M. van Reekum and K.R. Scherer

267

Implicit perception and affective preferences Zajonc, one of the major critics of the "exaggerated cognitivism" of appraisal theories, bases his argument (Zajonc, 1980, 1984) on research findings showing that affect is generated for stimuli which cannot be consciously perceived (e.g. Kunst-Wilson & Zajonc, 1980; Murphy & Zajonc, 1993). His "affective primacy hypothesis", claiming that affective reactions towards a stimulus can be elicited with minimal stimulus input and virtually no cognitive processing, has since triggered much research in the domain of implicit perception and emotion (e.g. Niedenthal, 1990; Niedenthal & Kitayama, 1994). For instance, Murphy and Zajonc (1993) carried out an experiment in which they established positive and negative valence for a previously neutral stimulus using an affective priming procedure. They compared the effects of what they called "affective" and "cognitive" priming on the judgment of novel stimuli under extremely brief and longer exposure duration of the primes. Pictures of faces expressing happiness and anger were used as primes, while Chinese ideographs, selected as being affectively neutral, were used as targets. Among other effects, they found that only the affective primes presented at a subliminal level produced significant shifts in subjects' evaluations of "good" versus "bad" towards the targets. These results were explained by claiming that "when affect is elicited at levels outside of conscious awareness, it is diffuse and its origin and address are unspecified. Because of its diffuse quality, nonconscious affect can "spill over" onto unrelated things" (p. 736). One can argue that Zajonc's affective primacy hypothesis is concerned with the role of consciousness or awareness rather than that of cognition, emphasizing again the importance of definitional issues (see Leventhal & Scherer, 1987). Furthermore, most of the empirical studies by Zajonc and followers concern affective preferences and are thus not directly pertinent to the study of emotion as defined above. Yet, the research in this tradition addresses two important issues, notably a) the fact that stimuli are often evaluated in an automatic manner, that is, without much cognitive effort, and b) that there may be an affective response disposition to certain stimuli, such as facial expressions of certain emotions. Clinical approaches to emotional disorder The study of pathology has always been of great importance to understanding the fundamental mechanisms of behavior, providing - through highlighting malfunctioning and its effects - a privileged window on the

268

Chapter 6

underlying mechanisms. Not surprisingly, then, the study of emotional disturbance can greatly contribute to the understanding of normal emotion. Emotion and attention. Recently, clinical research on emotion and attention has applied information-processing modds to the question of the role of emotional states and traits in attontional processes (sdection and intensive processing of stimuli, see Wells & Matthews, 1994) and sustaining emotional thoughts (Teasdale & Barnard, 1993; Wells & Matthews, 1994). Experimental research in the fidd of attentional bias as a result of an emotional disturbance has been inspired by paradigms devdopexi within the tradition of cognitive psychology (e.g. the Stroop color naming task; Stroop, 1935), resulting in interesting adaptations of these paradigms to problems of emotional malfunctioning (e.g. using an emotion version of the Stroop task). This extensive body of work cannot be reviewed here; for more complete reviews of theoretical models and research results, the reader is referred to Teasdale and Barnard (1993), Wells and Matthews (1994) and Williams, Mathews, and M a c ~ (1996). In brief, evidence so far shows that emotionally disturbed patients often show an attentional bias towards information relevant to their concerns or worries. This effect can also be found in non-clinical groups with high trait emotion (e.g. trait-anxiety), or for normal participants following emotion induction, but tends to be less pronounced than in clinical patients (Wells & Matthews, 1994) I. Furthermore, relevant information earl be presented subliminally, eliciting a comparable attentional bias (although the effect is small in magnitude). This emphasizes the automatic nature of the detection of relevance (see MaeLexxt & Hagan, 1992; Mogg, Bradley et al., 1993; Mogg, Kentish et al., 1993; all in Williams et al., 1996). Automatic, low-level processes, triggered by specific stimulus inputs and often interacting with higher-order controlled processes, are held responsible for this shill of attention toward potentially relevant information (e.g. Wells & Matthews, 1994). The fact that attentional bias is found for persons high on certain emotional traits (especially trait anxiety) but not for normal individuals experiencing a particular emotional state (such as state anxiety), is explained by Wells and Matthews in terms of unusual and extensive S-R learning experiences, which are supposedly anchored in low-level memory traces. 1 This phenomenon,however,cannotbe replicated in single stimulus tasks, such as lexical decision tasks with emotionallyrelevant words (see Mathews, 1988)or homophone/ homographspelling (e.g. Mathews, Richards, & Eysenck, 1989for homophonespelling; French & Richards, 1992for homographassociations; in Wells & Matthews, 1994).

C.M. van Reekum and K.R. Scherer

269

Whether or not most phobics have actually encountered the object of their phobia in a more striking manner than nonphobics is an open question (see Rachman, 1974, cited in Frijda, 1986). What seems clear, however, is that through a process of repeated exposure, reflection, and rehearsal, associations between objects related to the phobia and the resulting emotional response tend to become highly automatized or schematized. Even though strong attentional bias effects have been shown mostly in clinical patient groups rather than nonpatient groups, these studies imply that information which is relevant to the organism is given processing priority. The fact that personal worries or concerns play a significant part in the attentional bias effect fits with one of the basic tenets of appraisal theory, namely that information is assessed on the basis of its need, goal, or concern relevance. Appraisal theory would thus predict that attentional bias effects will be more pronounced in individuals with affective disorders than controls, to the extent that they appraise certain stimuli as being more concernrelevant. What current appraisal theory fails to specify are the precise mechanisms whereby the appraisal of a stimulus as highly relevant regulates the organism's attention. Ohman's "'preparedness" model of emotion-generation. Ohman (1986, 1988) is centrally concerned with models of fear, panic, and anxiety disorders. His research, based on Pavlovian conditioning of fear-rdevant stimuli, reveals that physiological responses can be activated pre-attentivdy, and that emotional learning can take place without the participant being aware of its source. Based upon this research evidence, 0hman argues that evolution has tuned the perceptual mechanisms in such a manner as to promptly generate a response as soon as a stimulus relevant to the organism is perceived, even at a very low level of analysis of the stimulus. According to this view, the perceptual systems detect relevant stimuli automatically and independently of the current attentional focus. As soon as a relevant stimulus is perceived, ongoing actions are interrupted and more conscious, i.e. controlled, processing mechanisms are activated to further analyze the situation's significance before directed action is taken. This "request" for controlled processing is accompanied by an unspecific physiological response, preparing the organism for subsequent emotional responses. The controlled processes in 0hman's (1986) model comprise a primary appraisal mode, a secondary appraisal mode and a response selection mode. Primary appraisal is defined as a further evaluation of the emotional significance of an event using any related information held in memory at a conscious or explicit level. Secondary appraisal evaluates the events in

270

Chapter 6

relation to available action alternatives before a response is selected (response selection mode). This sequence of modes results in physiological activation, motor behavior, and, potentially, verbal responses. Research findings reported by Ohman and eoUcagues (e.g. 0hman, 1988) underscore the prc-attcntivc activation aspect in 0hman's model, in contrast with the controlled (appraisal) processes which have not been thoroughly addressed. The research paradigm generally used in this work consisted of conditioning of emotionally relevant and irrelevant pictures to a light electric shock, followed by subliminal presentation of the conditionod pictures (using backward masking procedures) in order to maximally exclude controlled or conscious processing of the pictures. Skin conductance response measures revealed increases in autonomic nervous system (ANS) activity only for fear-relevant pictures, such as spiders and snakes (0hman, 1986) or angry facial expressions (0hman, Dimberg, & Estcves, 1989). No such effects were found with the fear-irrelevant pictures, such as mushrooms or happy facial expressions. Ohman and Soarcs (1994) have been able to extend these findings to spider and snake phobias where the phobics show similar physiological responses when presented either optimally (i.e., clearly visible) or subliminally with the object of their phobia. Interacting cognitive subsystems (ICS). In an attempt to better model cognitive-affective relationships, Teasdale and Bamard (1993; Bamard & Teasdale, 1991) have presented their framework of interacting cognitive subsystems (ICS). Briefly, ICS consists of 9 subsystems, including sensory/proprioceptive, effector, propositional, and implicational subsystems, each of which handles a different type of information. Contained in every subsystem is a memory for that system's previously processed information, processes to update and access the memory, as well as processes to transform information into a representation (or coding) suitable for processing by the other subsystems. For a full description of the ICS framework and the application of ICS to emotion, the reader is referred to Teasdale and Bamard (1993). Of particular relevance to emotion is their idea that emotions are generated primarily within the implicational subsystem. The implicational subsystem directly receives and integrates information from many of the other subsystems, such as the sensory and the propositional subsystems, about the occurring event. The implicational subsystem then makes inferences based upon these diverse sources of information by matching the patterns of information with previously stored schemata, which include emotion-related, "affective themes" (e.g. the theme of "potential severe self-related threat";

C.M. van Reekum and K.R. Scherer

271

see Bamard & Teasdale, 1991, p. 19), and generates emotional reactions including overt manifestations such as facial and bodily responses by activating patterns in the effector subsystems. In ICS, it seems that the type of emotion elicited is highly dependent upon "matching" of incoming or inferred information to stored information about past emotional experiences. And despite the fact that any final inferences occur in the implicational subsystem, ICS makes explicit the possibility that different types of information, being processed at different "levels", contribute to this final inference. Also, Teasdale and Bamard's (1993, Bamard & Teasdale, 1991) reference to the affective themes which are extracted from previous experiences or innately prepared responses is reminiscent of Lazarus' (1991) notion of Core Relational Themes (CRT). Adopting an ICS approach, one could infer that the implieational system is the seat of appraisal. The ICS is unique in emphasizing the importance of the implieational system for emotion in that this subsystems "adds" emotional connotations to otherwise "cold" cognitions. "Classic" cognitive psychology

In general, mainstream cognitive processing models have tended to emphasize either the way in which information is represented in memory (e.g. Tulving, 1972; Johnson, 1983), or the way in which information is processed (e.g. Leventhal, 1984; Shiffrin & Schneider, 1977). Both approaches are pertinent for our current topic. Emotion and memory. Recent advances in the study of memory have greatly influenced work on emotion-related cognitive processing (see Christianson, 1992). For instance, the dissociation between implicit and explicit memory as suggested by Graf and Schacter (1985) and Schacter (1987) has been used by emotion psychologists in order to address and further explain the seemingly unconscious elicitation of affect, particularly in the case of emotional disorders (Teasdale & Bamard, 1993). Similarly, Bower's (1981) influential work on the role of emotion on thinking and remembering has been directly influenced by memory models in cognitive psychology. Of particular importance to the notion of hierarchical processing is Johnson's (Johnson, 1983: Johnson & Multhaup, 1992) model of a MultipleEntry, Modular memory system (MEM). Johnson proposes that memory consists of a perceptual system (dealing with perceptual activities such as seeing, heating etc.), and a reflective system (which is actively involved in the

272

Chapter 6

creation and recording of self-generated activities such as planning, comparing, imagining etc.). Based on this conceptualization of a multi-level memory system, Johnson and Multhaup (1992) also made a first attempt to describe the elieitation of emotion within this kind of cognitive structure. This approach is briefly illustrated below. The goals and plans of an organism, which have a central place in appraisal theories of emotion, are considered as "agendas" by Johnson and Multhaup. These agendas control processing in different (sub)systems, and they can vary in complexity and in the degree to which they are perceptually (situationally) controlled or reflectively (self-) controlled. For example, an activated perceptual agenda guides one as to where to look, whereas activated reflective agendas, providing the individual with a "sense of self' (Johnson & Multhaup, 1992), might give rise to one blaming others for an event. Emotions, according to Johnson and Multhaup, arise from processing withinsubsystems (whether or not controlled by agendas), accompanied by autonomic and motor responses. All subsystems give rise to emotions, but there are differences as to which emotions arise from which systems. Some emotions, like fear, anger or joy, can arise from all subsystems. But the quality of the emotion that arises from reflective processing will be different from that arising from perceptual processing. As Johnson and Multhaup note, the reflective subsystems are responsible for the occurrence and experience of more complex emotions. Emotions arising from activity of the reflective subsystems unfold more slowly than emotions arising from the perceptual subsystems. To demonstrate that a multi-level process is involved in the generation of an emotional state, and that the complexity of an event can influence the automaticity of some but not other aspects of its evaluation, Johnson and Multhaup investigated the acquisition and retention of preferences in amnesic patients in different situations. They found that whilst amnesics developed smaller preferences for imaginary men than controls (provided by biographical information depicting the men as "good" or "bad"), they showed no deficit in the development of preferences for melodies. Johnson and Multhaup concluded that whereas preference acquisition of melodies depends largely on perceptual processes (which are intact in amnesics), the acquisition and especially the retention of evaluative impressions for people are mainly influenced by reflective processes (which are impaired in amnesics). While certainly pertinent to the topic under discussion, the experiments reported by Johnson and Multhaup, as many others in this area, are concerned with preferences, not with emotion in the sense of the definition

C.M. van Reekum and K.R. Scherer

273

given above. Similar dissociation effects for bona fide emotions remain to be demonstrated, even though it is tempting to conclude at this stage that at least two levels of information processing can work in a dissociated manner. Effort and control in cognitive processing. Cognitive models that specifically describe processing mechanisms are of particular importance to the modeling of emotion-antecedent appraisal and might help to explain the dynamics underlying emotion elicitation. The relative amount of control or effort required for certain types of processing has acquired great importance in the study of (social) cognition. Distinctions such as "automatic" or "schematic" versus "controlled" or "reflective" processing provide explicit conceptualizations of different processing levels. For example, Shiffrin and Schneider's (1977) notion of "automatic processing" focuses on the aspect of effort required, and the constraints imposed by the allocation of limited resources to a specific event or stimulus. Such concepts lend themselves to addressing the question of how bottom-up processing of sensory information and top-down processing of memories of past experiences can generate or modify an emotional response. Logan (1988) extends Shiffrin and Schneider's notion by proposing an instance theory o f automatization. This model provides a potentially useful theoretical construct for the formation of emotion schemata. In summary, Logan relates automaticity to the retrieval of instances from memory rather than to resource limitations, as often is proposed in the literature on attention (e.g. Shiffrin & Schneider, 1977). Each encounter with a stimulus is encoded and stored as a specific trace in memory. The use of a general algorithm or rule to perform a task is, after some practice, replaced by retrieval from memory of specific stimulus-response mappings. Thus, the more often similar stimuli are encountered, the more easily one can retrieve an appropriate response to anyone of those stimuli. This can be illustrated by the example of a person finding him/herself in an entirely novel, highly obstructive situation. On the basis of the previously discussed theories of automatic versus controlled processing, one could infer that elaborate cognitive processing is required in order to appraise the situation as goal obstructive and thus to generate an appropriate emotional response, for example anger. Instance theory, in contrast, predicts a somewhat different process. If one assumes that emotional situations are better remembered than comparable non-emotional events (see Bower, 1992; also presumably because the attention is more centrally focused, see Christianson, 1992), the apparently novel situation might activate a trace of the previous appraisal-response pairs from generally comparable situations.

274

Chapter 6

As a consequence, on the basis of partial similarity of stimulus cues, an anger response might be generated without elaborate controlled processing. Such an appraisal process is likely to occur in an implicit fashion, that is without the person being able to report exactly why he or she became irritated or angry. Although unsubstantiated, this example demonstrates how Logan and others' (e.g. Schneider, 1985; for further references, see Logan, 1988) description of the automatization of processes by practice might be useful in explaining how appraisal becomes automatised. It should be noted that this account is a somewhat more elaborate model of what many authors have subsumed under schematic processing - the assimilation of novel instances to genetic, prototypical patterns on the basis of partial similarity.

Brain mechanisms in emotion Multiple pathways in the neural processing of conditioned fear responses. One of the most important bodies of neuropsyehological literature pertinent to multi-level notions of appraisal is the work of LcDoux (e.g. 1986, 1989, 1993, 1994), who traced two separate but interdependent routes in the rat brain that are held responsible for emotional learning and emotional memory. He conditioned fear in rats by the application of a standard conditioning paradigm (presentation of an auditory stimulus followed by an electric shock), and studied the r fear responses by producing lesions at systematically chosen locations and measuring the electrical activity of neurons at different sites in the rats' brains. This research shows that fear in the rat can be elicited through neural activity involving two pathways; subeortieal and cortical (or "low" and "high" roads respectively; see LeDoux, 1996). In both eases, the amygdala plays a central role in the evocation of fear responses to a conditioned auditory stimulus. The subcortieal pathway consists of projections from the thalamus which pass rudimentary information directly to the amygdala. The amygdala then activates the peripheral nervous system. The cortical pathway consists of trajectories from the thalamus to the sensory (either visual, or in the case of the research mentioned above, auditory) cortex, and then via the association cortices to the amygdala. Research from LeDoux and colleagues reveals that lesions in the auditory cortex do not interfere with the conditioned response to the sound, implying that the more direct, subeortieal route is sufficient to evoke behavioral fear responses. LeDoux (e.g. 1989, 1994) argues that the subcortical pathway processes information very quickly due to the fact that it only involves very few neural links, but only provides

C.M. van Reekum and K.R. Scherer

275

coarse-grained, global information about the stimulus. In contrast, signals processed via the "high road" provide much more detailed information but require much more processing time given that these pathways involve several links b ~ o c n the thalamus and the amygdala. With respect to emotion in general, LeDoux (1986) argues that subjective experience results from a conscious, higher-level of cognitive processing, but that every emotion is precodexl by unconscious, low-level emotional processing of the stimuli. The unconscious processes, i.e. the "low road", "prepare" the organism for fight/flight behavior; the "high road" modifies the responses, i.e. gives the reaction "direction", or inhibits the processes triggered via the subcortical pathway in the case that they are maladaptive. While the research on fear conditioning in rats by LeDoux and his collaborators makes a strong case for the existence of different levels of emotion-antecedent processing, it remains to be established whether the contribution of the amygdala to emotional processes in general is as important in humans as in rats (see Christianson, 1992, for some counter arguments). The subcortical pathways identified by LeDoux might be less important in human emotional experiences than in rats. For instance, Halgren (1992) argues that sensory input to the amygdala via a direct thalamoamygdaloid pathway is very weak or even non-existent in humans. Rather, studies measuring amygdala unit activity as well as stimulation of the amygdala reveal that the human amygdala functions within the cortical system, emphasizing cortico-amygdala communication in emotional evaluation. Measurement of electrical activity during word and face recognition show high activation in the amygdala. Also, the amygdala seems to receive visceroscnsory input. The stimulation studies show that visceromotor and hormonal responses are produced by the amygdala as well as images, thoughts and feelings. Amygdala processes might, however, still precede the conscious evaluation (Halgren, 1992). Interactions between the amygdala and other brain structures. It is possible that the direct pathway from the sensory cortex to the amygdala which LeDoux traced in the rat also plays a role in the emotion-antecedent sensory processing of information in humans. However, here it is likely to play only a relatively minor role, since the pathway seems to transmit mostly information about simple stimulus characteristics. In contrast, the hippocampal formation has been shown to be strongly involved in the processing of spatial and contextual information, that is, information concerning the relations between sensory stimuli and their embedding in

276

Chapter 6

specific spaces and places (Kolb & Whishaw, 1990). Consequently, the complex, contextual information that is at the basis of most human emotions is likely to be mediated via connections from the hippocampus to the amygdala (LcDoux, 1996).These connections might well reflect something like a schematic or automatic mode of processing. There is strong evidence that the frontal lobes are also centrally involved in fear conditioning. Lesions in the rat's medial prefrontal cortex prolong the extinction process of a fear conditioned stimulus (LeDoux, 1996; see also Kosslyn & Koenig, 1995). LcDoux (1996) describes a study showing that once the visual stimulus (a light) had been conditioned to shock, the lesioned rat kept responding fearfully to the light, whereas after a couple of days the normal rat ceased showing any fear-related behavior. Apparently, the frontal lobes in the rat monitor and regulate the outputs of the amygdala. In humans, the frontal lobes seem to be involved in the inhibition of inappropriate (emotional) behavior. LeDoux (1996) reports that frontal lobe damaged patients performing rule-directed tasks often have trouble changing from one rule to the other. For example, once they have discovered a certain rule, these patients continuously fall back on the use of that rule (e.g. once used to sorting on shape in a card-sorting solution, it is difficult for them to change to color-based sorting). Damasio (1994) illustrates the crucial role of the frontal lobes in human emotional functioning with a number of case studies, such as the classic case of Phineas Gage (see also Stuss & Benson, 1983), who produced a range of anti-social and "careless" behaviors atter having suffered severe injury to the frontal lobes. In recent work done with Beehara (e.g. Bechara, Damasio, Damasio, & Anderson, 1994; see also Damasio, 1994), this group found that humans with damage to the ventromedial prefrontal cortex show a more oblivious attitude towards a simulated gambling situation than controls. Based on these experiments, Damasio (1994) concludes that frontal lobe damaged patients are unable to develop affective responses which are suitable to a new situation, even though they have stable representations or factual knowledge of future outcomes. He argues that the marking of a positive or negative value is lacking, resulting in the inability to reject or accept a future outcome. Damasio takes this evidence as support for his "somatic marker" hypothesis, which states that basic body-regulatory systems "prepare" cognitive processes determining what is considered to be good or bad for the organism. If these impressions can be empirically confirmed, the frontal lobe can be considered as a crucial relay station in emotion-related processing in the sense of affectively priming conceptual processes.

C.M. van Reekum and K.R. Scherer

277

Issues in Rewriting Appraisal Theory The recent work in the neurosciences, as well as in social and clinical cognition research, provides additional support for the idea that multi-level cognitive processing is implicated in the generation of emotion. Among other issues, a multi-level framework could account for phenomena such as attention-deployment involved in the selection of relevant information, the allocation of higher levels of appraisal to particular events, the interaction of lower, automatic and higher, conceptual or controlled processes, and the resulting activation of bodily responses. Together with the critique of classical appraisal notions, as outlined in the beginning of the chapter, it becomes imperative to integrate these advances in our knowledge of cognitive processing into appraisal theory. In what follows, two central issues for the development of multi-level appraisal theories will be discussed. Towards general process models of appraisal.

Recently, Smith and collaborators (Smith, Griner, Kirby, & Scott, 1996) have reported an ongoing effort to reformulate basic concepts suggested by Leventhal and Scherer (1987) by constructing a preliminary version of an appraisal process model. They emphasize the function of emotion to motivate and regulate attention and argue that since emotion alerts the organism to relevant information, the scanning or monitoring of the environment must occur outside attention; if this evaluative scanning process were attentional, then the demand upon attentional resources would be too high, restricting the resources available for ongoing processes. This implies that appraisal models should consider non-attentive mechanisms for the elicitation of emotion as well as the encoding of event-specific information for the purposes of appropriately motivating the individual for action. In the model, the authors emphasize two modes of processing, namely schematic and conceptual. Schematic processing is described to be fast, automatic, and functioning in a parallel fashion, yet inflexible and relatively concrete. The conceptual mode on the other hand, is expected to function serially, to be slower, under voluntary control and therefore flexible, and relies upon semantically accessible information. As in the case of the automatic/controlled distinction made by Shiffrin and Schneider (e.g. 1977) and in general agreement with Leventhal and Scherer (1987), the two modes proposed by Smith et al. are not independent of one another, but are assumed to interact in several ways. Most important

278

Chapter 6

in this context is the creation of schemata within the conceptual mode through the process of learning from experience. Conversely, when existing schemata are sufficiently activated by incoming sensory data, the information they represent might become available to conceptual processing. The appraisal information output from the conceptual and schematic modes, in addition to perceptual information, is described to be monitored by a so-called appraisal register. This register combines appraisal information from these three input modes, on the basis of which a specific emotional output is generated. The register is not considered as an active processor which computes information, but rather as an information detector and response selector which combines the appraisal outcomes and initiates the response accordingly. Attention is seen as being regulated by the registering of a subjective feeling state, when sufficiently intense, into conscious awareness. In this manner, conceptual processes are called into play, in order to further analyze the significance of the stimulus and to refine the emotional response accordingly. While the model suggested by Smith et al. is still preliminary and so far lacks both detail and precision (in particular with respect to the notion of a passive "appraisal register"), it constitutes an important effort to move appraisal theory from largely structural to dynamic process modeling. In addition, this effort is likely to result in greater convergence of appraisal research with other branches of emotion and cognition research.

Bottom-up vs. top-downprocessing Apart from developing general process models that describe the dOails of information processing on different levels, the obvious interaction between the levels needs to be theor~ieally conceptualized in a more stringent fashion. As has become apparent in the discussion above, the issue of bottom-up vs. top-down processing is one of the central concerns in this respect (see Leventhal & Soberer, 1987, pp. 21-23). The work of LcDoux, 0hman, and Damasio on the role of pre-attemive or amygdaloid processing together with general advances in the field of attention and emotion indicates that a bottom-up process is involved in the evaluation of the significance of sensory stimuli. This process is responsible for signaling to the organism that attention needs to be focused upon an external event in order to further specify the results of the rather coarse, unspecific analysis performed at the lower level. At the same time, the lower level initiates a rudimentary preparation for action, by activating the AN S

C.M. van Reekum and K.R. Scherer

279

and SNS in a rather general, non-specific manner. However, the propositions of the authors mentioned above, although firmly based on empirical evidence, have thus far only been tested for one emotion, namely fear. Research on implicit perception of affective information, even though it is currently limited to addressing positive-negative valence distinctions, also suggests that affective information is evaluated in a highly automatised, largely preattentive manner. It thus seems that relevance checking might precede and prepare, in a bottom-up fashion, other appraisal processes. While there is good reason to assume bottom-up processing of the significance of an event, there is evidence that top-down processes are involved in the priming of the relevance "detectors" as well as subsequent modification of the emotional response. LeDoux (1996) notes that the amygdala is not only activated by bottom-up processes evaluating the significance of an incoming simple sensory stimulus, but that is also receives information from various other parts of the cortex. In consequence, the appraisal performed at low levels might well be influenced by higher level processes such as the monitoring of the current concerns and goals of the individual. In fact, evidence in the field of attentional bias (see above) suggests that the individual actively inhibits any information not relevant to its current concerns or goals. Also, according to Kosslyn and Koenig (1995), working memory is thought of as being the interface between bottom-up and top-down processing. LeDoux (1996) has suggested that the lateral prefrontal cortex, presumably involved in working memory, plays a role in the selection of relevant stimuli. Thus, whatever is active in working memory determines which stimuli will be selected for further processing. There can be little doubt that in emotion-antecedent appraisal, both bottom-up and top-down processes are relevant. Low level appraisal can prime higher-level appraisal by bottom-up processing of potentially relevant stimuli (e.g. by attention focusing or activation spreading to related information in long-term memory). Also, low-level appraisal might be primed by higher level processes. Leventhal and Scherer (1987, p. 21) suggest that top-down effects of conceptual processing may sensitize classes of schemata relevant to situations that are related at a more abstract or semantic level. They provide the example of hearing a noise that fails to match the schemata of noises normally encountered upon entering one's home, with the result of priming sensory-motor reactions and strengthening the fear response. Frijda (1993) provides a similar example: "Emotions indeed rarely have the oneshot, immediate, and fast character of the paradigmatic case of being startled by a crackle in the solitary woods. And even there one has been walking

280

Chapter 6

around with all sort of expectations against which the crackle is perceived" (p. 382). Thus, the implications of being alone in the woods facilitate or enhance low level appraisal of the crackle by allotting more than the usual share of attention to the stimulus. Also, one's self concept (which presumably relies upon episodic or explicit memory; see Schaeter, 1996, and Johnson & Multhaup, 1992) or personality (which is based more upon implicit processes; Schacter, 1996) have an influence upon what information is selected and further processed and, more importantly, how this information is appraised. These personality dispositions can thus influence, in a top-down manner, what information will be attended to and how this information will be further processed and appraised. This important, and hitherto largely neglected, field of appraisal research is discussed in the next section. Individual Differences in Appraisal Processes

One of the paradoxes in the area of appraisal research is the lack of concern with individual differences in perceiving the same type of event. Appraisal theory partly started with Lazarus' (1968) insistence on the "transactional" nature of appraisal, linking the objective event and the subjective appraisal, strongly affected by the perceived coping ability of the individual. In consequence, one might have thought that appraisal theorists were particularly interested in individual difference factors that can explain divergences in appraisal outcomes under similar objective conditions. Indeed, most appraisal theorists will stress that appraisal is highly subjective and depends on the individual's perception and evaluation of events rather than their objective characteristics, predicting that the resulting emotion will be determined by the subjective interpretation. Yet, there has been little effort to more systematically identify stable individual traits that might predispose persons to show systematic appraisal tendencies or even biases in the appraisal process. Among the exceptions has been Scherer's (1987) suggestion of conceptualizing different types of emotional disorders on the basis of appraisal malfunctioning (see also Kaiser & Scherer, in press). The underlying assumption is that although appraisal is subjective and may vary from individual to individual, it must remain - within certain limits appropriate to the objective situation (e.g. through reality testing) and to the coping potential that is commonly perceived to be within the individual's means. Violation of these appraisal reality constraints, as one might call

C.M. van Reekum and K.R. Scherer

281

them, will lead to the resulting emotion being considered as abnormal or disordered, at least by an individual's social environment, if not by him/herself. For example, Seherer (1987) has suggested that one particular form of depression, helplessness, might be partly due to a consistent underestimation of one's coping potential. It is important to stress that an individual who truly lacks the means to deal with a particularly difficult situation would be described as "dejected" whereas someone described as "depressed" is implicitly assumed to appraise the situation, particularly one's coping potential, in an inappropriate, disturbed fashion. Scherer (in press) reviews some of the individual difference factors that are likely to systematically affect emotion-antecedent appraisal. The following overview summarizes some of the major suggestions. First, theoretically postulated individual differences in both formal, process-related appraisal characteristics and content-related differences will be reviewed. Second, the personality characteristics that might underlie the tendency to show systematic appraisal biases are discussed, including pertinent findings in the literature.

Appraisal biases with respect to the form or process of appraisal One of the most basic formal variables in appraisal is the speed of the emotion-antecedent appraisal processes. It is possible that there are potential differences with respect to general processing speed in the central nervous system (CNS) as well as other factors that produce consistent differences in the speed of appraisal, resulting, for example, in differential onset of reaction patterns. A related factor is the relative predominance of automatic vs. controlled processing which might be consistently different across individuals (see below). Individuals may also differ with respect to the thoroughness or completeness of the appraisal following a particular event. For example, it is conceivable that one individual fairly rapidly arrives at a specific appraisal result, whereas another one engages in repeated re-appraisals before settling on an emotion-specific action tendency. These formal or process aspects of the appraisal may be related to the degree of cognitive effort that is expended, which in turn is linked to possible differences with respect to potential tendencies to engage in relatively effortful controlled processing on a conceptual level (a tendency often described as "intellectualizing"). Another related issue concerns the relative complexity of the analysis within the appraisal, i.e. gross vs. more fine-grained appraisal, a dimension

282

Chapter 6

directly related to the width of the categories used in inference and classification. For example, causal attribution, one of the major criteria in appraisal, might be evaluated quite differently depending on whether broad or narrow categories are used for assigning responsibility and inferring intentionality. A factor related to the breadth of category might be the absolute or temporary (interim) character of attributions (also linked to the readiness to engage in re-appraisals). Major individual differences could exist with respect to vigilance, i.e. the detection of events that are marginally pertinent to an individual and the nature of the attention deployment strategies used. Whereas a highly vigilant individual would be likely to have a very low threshold for the activation of more elaborate appraisal processes following detection of a potentially relevant stimulus, coupled with immediate direction of attention to this new stimulus, another, a less vigilant individual might not even notice a comparable stimulus. This suggests that there may be differential levels of reactivity to implicit or subliminal stimuli. This dimension could also be linked to the degree of feexlforward guidance, or bottom-up processing of higher-level conceptual processing. Similarly, there may be systematic differences in the degree of top-down control of lower level processing, such as sensitization for schema formation or automatization. Thus, the issue of levels of processing might be of fundamental importance for individual differences in appraisal, not only with respect to systematic preferences for certain processing levels but also relative to the kind and frequency of interactions between the levels. Appraisal biases with respect to content

Individuals may also differ with respect to tendencies to lean more towards one rather than another direction in appraising events with respect to certain dimensions, i.e. to attribute responsibility to oneself rather than others, or underestimate or overestimate one's power. Such biases concern the content of appraisal more than its form. Soberer (1987, Kaiser & Soberer, in press) has proposed a classification of such content-related appraisal biases and their potential contributions to clinical syndromes. The main points of this classification will be summarized bdow, using Scherer's stimulus evaluation chock (SEC) appraisal criteria (Table 1, for further details consult Schcrcr, 1984a, 1988, 1993b). With respect to the Novelty cheek one might expect differences, possibly linked to fundamental characteristics of the ANS and CNS, on speed of

C.M. van Reekum and K.R. Scherer

283

habituation and/or extent of inhibition. Slow habituation and lack of inhibition could result in ovcrscnsitizationand ovcrvigilance with respect to incoming stimulation. If extreme, this might correspond to pathological nervousness or jumpiness, provoking frequent and massive startleresponses. Conversely, very rapid habituation or strong inhibition potential might correspond to a more sluggish appraisal system, requiring fairly strong and unusual stimulus characteristics to diagnose novelty or change. Again, extreme forms might produce clinicalsyndromes of stupor or lethargy. On the Intrinsic Pleasantness check one can imagine systematic differences with respect to hcdonic evaluation, due to differences in the tuning of valence detectors, unusual learning experiences, or specific states. The anhcdonia often found in depressive patients provides a clinically relevant example. Several subchecks of the of Goal Conduciveness appraisal can be affected by biases. A central issue concerns the appraisal of relevance or pertinence of stimuli or events with respect to one's needs and goals. This may depend partly on individual differences in the intensity of motivational striving - individuals with weak drive structure and lack of strong concerns may systematically underestimate the pertinence of events, whereas strongwilled individuals with intense motivational impetus might show a tendency to see relevance everywhere accompanied by excessive personalization. Biases in relevance appraisal might also be due to lack of ability to evaluate consequences of and establish links between events, resulting either in overassimilation and overgeneralization, or in lack of concern or caring. The frontal lobe lesion syndromes described above are good examples for clinically relevant cases, yielding blandness and lack of ego-involvement. Paranoia, obviously, is an example of the other extreme where any event is personalized. Another aspect of the goal conduciveness check concerns the appraisal of outcome probabilities, the estimation of the likelihood that certain consequences will follow certain events. The optimism-pessimism personality dimension provides an example for systematic biases in this context (Scheier & Carver, 1985). Systematic biases may also occur with respect to the conduciveness judgment itself, i.e. the estimation to which extent a certain state of affairs will help or hinder attainment of a goal or satisfaction of a need. For example, perfectionists or individuals lacking realism may well systematically underestimate the conduciveness of events or states of affairs. The opposite might be true for individuals considered to be naive or overly credulous. Major differences could be expected for the Urgency appraisal, depending on time perspective and assessment of

284

Chapter 6

consequences, and giving rise to either overreaction and panic or apathy or passivity. Some of the best documented individual differences are found for the Coping potential appraisal, particularly the Causal attribution subcheck. The external/internal control personality dimension is a prime example for powerful biases with respect to causality or responsibility assessment. For example, overestimation of self agency has been at the core of cognitive theories of depression etiology (Abramson, Seligman, & Teasdale, 1978; Teasdale, 1988). Other potential biases include faulty ability or effort attribution. Similar biases may occur with the appraisal of intention. Again, paranoia is a ease in point where overattribution of intention is frequent. Important differences between individuals can be expected for perceived Control, i.e. the degree to which certain events or consequences are under human control, independent of the amount of power that is available to the person. Some individuals may have illusions of control or may overestimate control potential whereas others may systematically denigrate the possibility of control, the extreme ease being the clinical syndrome of hopelessness. The Power check is one of the most important subchecks of the Coping Potential appraisal since its result ot~en powerfully determines the nature of the emotion that is elicited. In many eases, the result of the power appraisal may make the difference between fear and anger responses. Given the highly subjective nature of power appraisals, and the many dimensions that are involved (e.g. types of power such as physical force, money, knowledge, ability to recruit help), this appraisal dimension is a prime candidate for individual differences and biases. Such systematic misjudgments of one's own or other persons' power earl be due to a large number of different personality characteristics and cognitive style variables (see below). Most important, is the person's self-image, including persistent tendencies to overor underestimate one's power or ability to influence a given situation. The helplessness syndrome as described by Seligman and his collaborators (Abramson et al., 1978; Peterson & Seligman, 1984) is a classic example in the clinical literature. A final check related to Coping Potential is the appraisal of Adjustment, the estimated capacity to accept and live with the consequences of a given situation. Clearly, individuals tending toward fatalism, espousing more passive life philosophies, will appraise the respective situations very differently from people with more deterministic philosophies. The Compatibility with Standards check concerns the relationship of one's own or someone else's behavior to both external (norms, morals) and

C.M. van Reekum and K.R. Scherer

285

internal standards. Potential sources for systematic biases might be undersocialization, normlessness, amorality, or asociality, all tending toward minimizing the appraisal of behavior-norm discrepancies, thus reducing guilt potential and producing, in extreme cases, shamelessness and antisocial behavior. Conversely, high moral and ethical standards would tend to augment perceived discrepancies and might lead to exaggerated feelings of shame or guilt which could become part of a neurotic syndrome. With respect to internal standards, the major individual difference variable would be the definition and the strength of the ideal self. Potential sources for individual differences in appraisal tendencies

The preceding two sections have presented hypotheses as to possible systematic differences between individuals with respect to the form and content of the appraisal process. In this section, we will look at this issue from another vantage point - the likelihood that dimensions of individual difference and personality that have been shown in the literature to have powerful effects on behavior and cognition may systematically influence appraisal processes. This review is organized by the nature of the underlying individual difference dimension, here broadly classified into organismic predispositions, cognitive styles, and personality traits. Organismic predisposition. At the very lowest level of individual differences factors are innate characteristics of the CNS and/or AN S. As mentioned above, rapidity of habituation and efficiency of inhibition are among the candidates in this category. There is reason to assume that differences in efficiency of inhibition and interference sensitivity might be related to working memory capacity, cognitive strategy use, and knowledge (Hamishfeger & Bjorklund, 1994). Lord and Levy (1994) have attempted to show that connectionist-level mechanisms related to activation and inhibition can model the information selection needed to provide continuity in thoughts and may explain individual differences in selective attention capacities. Another important variable might be general speed of processing. Hale and Jansen (1994) have shown that individual performance on diverse tasks can be predicted on the basis of a single processing-time coefficient. Another CNS variable of interest for differential questions is habitual cortical arousal (Matthews & Amelang, 1993; Matthews & Harley, 1993). All of these cognitive mechanisms are likely to be directly involved in appraisal processes, particularly at the lower, more automatic levels.

286

Chapter 6

Cognitive styles. Another classic area of individual differences on a fairly low level of processing are persistent cognitive styles. Among the styles that are likely to have a bearing on appraisal processes are holistic vs. analytic processing (Durra, Dunn, Andrews, & Languis, 1992), parallel vs. serial processing (Tous, Fuste, & Vidal, 1995), field dependence (Messick, 1994), overgeneralization, and personalization (Dritschel & Teasdale, 1991). The latter authors suggest that the persistent differences in response to emotional events measured by the traits of affect intensity and ncurotir may be mex~ated by particular styles of cognitive processing, a relationship that is conceivably influenced by differences in the emotion-antecedent appraisal processes. Directly pertinent for appraisal is the tendency to use certain ways of categorizing events, linked to the classic issue of cognitive complexity (Sommers, 1981; Schroder, Driver, & Streufert, 1967; Tetlock, 1983; Harris, 1981; Press, Crockett, & Delia, 1975). Among the possible variables are wide vs. narrow category use, rigidity, or need for structure. For example, Neuberg and Newson (1993) demonstrated how individual differences in the desire for simple structure may influence how people understand, experience, and interact with their worlds. Participants high in Personal Need for Structure were especially likely to organize social and nonsocial information in less complex ways, stereotype others, and complete their research requirements on time. Chronic information-processing motives may obviously strongly influence formal aspects of the appraisal process. A variable that received much attention in social cognition research is nee~ for cognition (Cacioppo, Petty, Feinstein, & Jarvis, 1996). Waller (1994) showed that low need for cognition-participants reported a significantly greater reliance on normative information (indicative of peripheral processing) while participants high in need for cognition reported a significantly greater expenditure of cognitive effort (indicative of central route processing). Similarly, data reported by Thompson, Chaiken, and Hazlewood (1993) suggest that (1) nee~ for cognition involves intrinsic motivation for effortful cognitive processing, (2) need for cognition may predict such processing mainly in contexts with minimal extrinsic incentives for processing, and (3) control motivation may be related causally both to extrinsic undermining effects and to in&vidual differences in neexl for cognition. Clearly, the disposition to engage in effortful cognitive processing may greatly determine the likelihood of switching to higher levels of appraisal. Differences on this dimension may appear rather early. Berzonsky (1993) showed that adolescents with an information-oriented, self-exploratory

CM. van Reekum and K.R. Scherer

287

identity style were significantly more motivated to engage in active information processing (need for cognition) and were more receptive to new ideas, personal feelings, and actions (experiential openness), even though this might threaten hard-core areas of the self, than were their normative or diffuse/avoidant counterparts. Personality traits. Many classic personality traits such as extroversion, repression-sensitization, neuroticism, rigidity, dysphoria, worrying, sensationseeking, or openness, can be seen as pertinent to appraisal. The effects of trait disposition can be expected to directly affect cognitive operations. Sensation-seeking (Schroth & Lurid, 1994) and openness (McCrae, 1993) have also been shown to correlate with cognitive processing preferences that may play a major role in appraisal, such as creativity and intellectual orientation. Pruzinsky and Borkovec (1990) found that worriers reported more negative daydreaming, greater difficulty with attentional control and greater obsessional symptoms, public self-consciousness, and social anxiety. Worriers evidenceA significantly more negatively affect-laden cognitive intrusions during relaxed wakefulness and a focused attention task. Findings reported by Lorig et al. (1994) suggest that repressors exhibit an absence of cognitive activity when faced with the recall of negative memories. Extroversion has been frequently studied with respect to cognitive processing. Matthews and his collaborators have been examining relationships to cognitive arousal (Matthews & Amelang, 1993; Matthews & Harley, 1993; Matthews, Davies & Lees, 1990). Neurological differences between extroverts and introverts, while hotly debated, continue to be reported. Stenberg, Wendt and Risberg (1993) confirmed findings from earlier studies of cerebral blood flow (CBF) at rest showing higher blood flow in the temporal lobes for introverts than for extroverts, and a negative correlation between extroversion and global CBF among women. Tobacyk, Driggers and Hourcade (1991) suggest that the perceptual/cognitive information processing preferences associated with extroversion, intuition, and perceiving complement high self-monitoring, while the processing preferences associated with introversion, sensing, and judging complement low self-monitoring. Of direct relevance to the issue of differences in the form of the appraisal process is the finding by Stelmack, Houlihan and McGarry-Roberts (1993) showing that higher neuroticism scores were associated with faster P300 latency, a measure that is regarded as an index of stimulus evaluation time that is independent of response production. Contrary to what one might expect on the basis of this finding, higher neuroticism scores were associated

288

Chapter 6

with slower reaction time (RT), a measure that is also regarded as an index of speed of information processing. The authors speculatively explain this asynchronieity with an overly hasty and worried evaluation of the stimulus requiring further processing before response initiation. This explanation fits the notion of oversensitization to changing stimulation mentioned above. Furthermore, affective processes linked to the neexl for response initiation may intervene in such a way as to slow down or inhibit the reaction. Apart from personality dispositions in the narrow sense, individual differences in appraisal could also be due to specific attitudinal structures. For example, Zucker and Weiner (1993) showed that conservatism correlated positively with a belief in the importance of individualistic causes, controllability, blame, and anger, and negatively with perceptions of the importance of societal causes, pity, and intentions to help. Personal help was emotionally determined, whereas welfare judgments were directly related to attributions of responsibility and political ideology. Obviously, these and other types of generalized attitudes are rather likely to color the way in which events are appraised. Another set of individual difference variables likely to influence emotion-antecedent appraisal are linked to the structure of the self, in terms of self concept or self image. For example, individuals high on self-efficacy (Bandura, 1982; Berry, 1989) are likely to be systematically biased, across a large range of situations, towards auto-attributing higher internal control and power than those low on this dimension. In addition, they are likely to consistently check outcomes against their internal standards. Conversely, individuals with more fragile selves might have the opposite tendencies, particularly in social contexts. Thus, Downey and Feldman (1996) show that people who are sensitive to social rejection tend to anxiously expect, readily perceive, and overreact to it. Apart from structural variables concerning the self, social psychologists have studied a variety of process variables such as self-monitoring or selfhandicapping. Again, it takes little fantasy to imagine ways in which individual differences on these dimensions might affect appraisal, in particular with respect to assessing coping potential and the compatibility with internal standards. An interesting example is provided by Rhodewalt, Morf, Hazlett and Fairfield (1991) who examined the effects of selfhandicapping on ability attributions and self-esteem. High self-handicapping participants generally discounted ability attributions in response to failure feedback. The study illustrates how individual differences in motives to

C.M. van Reekum and K.R. Scherer

289

engage in self-protective or self-enhancing behavior affects attribution of achievement and affect. In addition to these more or less stable personality dispositions, momentary mood differences are likely to systematically affect cognitive processes that are directly pertinent to emotion-antecedent appraisal (Basso, Schefft, & Hoffmann, 1994; Isen, Niedenthal, & Cantor, 1992; Murray, Sujan, Hirt, & Sujan, 1990; Woodfield, Jones, & Martin, 1995). This compilation of potential individual difference dimensions which can be reasonably expected to affect appraisal processes shows the promise of further theoretical efforts in this direction, accompanied by more focused empirical investigations. Such theoretical efforts will need to first systematize the nature of the individual differences. One pertinent dimension is form versus content. While some of the stable dispositions described seem to affect the nature of the appraisal process as a whole, others seem to be specific to content such as the predominance of certain schemata, certain values, or certain goals. A second pertinent dimension concerns differences with the "horizontal" organization of differences by processing level - some of the findings reported above seem to concern the way in appraisal functions on a particular level, for example the role of cognitive style in category formation on the conceptual level. Finally, although there has been little direct evidence for this in the findings reported above, one may expect a "vertical" organization with respect to the levels, for example differences in the speed and extent of "schematization" or "automatization" of the appraisal processes. While work specifically dedicated to these issues has hardly begun, advances in this area may well provide important insights to our understanding of the appraisal process as a whole. Conclusions

This chapter could do little more than survey a construction site. Deep holes are gaping everywhere but some of the construction elements can already be found strewn all over the site. We attempted to show how appraisal theory, which we consider as a solid foundation for a future theoretical edifice, needs to add stories or levels (mostly below the conceptual floor that has been reasonably consolidated so far). In drawing up the blueprints for such future development we had, by necessity, to be quite speculative in showing the promise of moving towards hierarchical, multilevel models of appraisal, buttressing our arguments with some of the existing approaches in related areas. Similarly, we have been trying to outline

290

Chapter 6

ways in which individual differences can be made fruitful for an understanding of both appraisal biases and appraisal pathology, potentially leading to emotional disorder. Again, individual building blocks found in the literature come in handy to illustrate the way in which one might piece together a more systematic approach to predicting how neurophysiologieal and personality characteristics might predispose an individual towards certain modes of appraisal with foreseeable effects on emotional experience. This review of the blueprints and some of the pertinent building blocks has hopefully demonstrated that future advances on the tower of appraisal will not materialize by a compartmentalized activity of the appraisal specialists other crafts need urgently to be involved: neuropsychologists, cognitive scientists, social psychologists, and clinical researchers, to name but a few. As on any construction site, the speeA of building and the quality of the resulting workmanship will depend in large part on the cooperation of the crafts that are immersed in this effort. References

Abramson, L. Y., Seligman, M. E. P., & Tcasdale, J. (1978). LcameA helplessness in humans: Critique and reformulation. Journal of Abnormal Psychology, 87, 49-74. Arnold, M. B. (1960). Emotion and personality. Vol.l : Psychological aspects. New York: Columbia University Press. Arnold, M. B. (1960). Emotion and personality. Vol.2: Neurological and physiological aspects. New York: Columbia University Press. Bandura, A. (1982). Self-efficacy mechanisms in human agency. American Psychologist, 37, 122-147. Barnard, P. J., & Teasdalr J. D. (1991). Interacting Cognitive Subsystems: A systemic approach to cognitivc-affcctivc interaction and change. Cognition and Emotion, 5, 1-39. Bartlett, F. C. (1932). Remembering: A study in experimental and social psychology. London and New York: Cambridge University Press. Basso, M. R., Scheflt, B. K., & Hoffmann, R. G. (1994). Mood-moderating effects of affect intensity on cognition: Sometimes euphoria is not beneficial and dysphoria is not detrimental. Journal of Personality and Social Psychology, 66, 363-368. Bcchara, A., Damasio, A. R., Damasio, H., & Anderson, S. W. (1994). Insensitivity to future consequences following damage to human prefrontal cortex. Cogmaon, 50, 7-15.

C.M. van Reekum and K.R Scherer

291

Berkowitz, L. (1994). Is something missing: Some observations prompted by the cognitive-n~associationist view of anger and emotional aggression. In L. R. Huesmann (Ed.), Aggressive behavior: Current perspectives (pp. 35-57). New York: Plenum. Berry, D. C. (1987). The problem of implicit knowledge. Expert Systems, 4, 144-151. Berry, J. M. (1989). Cognitive efficacy across the life span: Introduction to the special series. Developmental Psychology, 25, 683-686. Berzonsky, M. D. (1993). Identity style, gender, and social-cognitive reasoning. Journal of Adolescent Research, 8, 289-296. Bower, G. H. (1981). Mood and memory. American Psychologist, 36, 129148. Bower, G. H. (1992). How might emotions affect learning? In S.-A. Christianson (Ed.), The handbook of emotion and memory: Research and theory (pp. 3-31). Hillsdale, NJ: Lawrence Erlbaum. Cacioppo, J. T., Petty, R. E., Feinstein, J. A., & Jarvis, W. B. G. (1996). Dispositional differences in cognitive motivation: The life and times of individuals varying in need for cognition. Psychological Bulletin, 119, 197-253 Christianson, S.-A. (1992). Remembering emotional events: Potential mechanisms. In S.-A. Christianson (Ed.), The handbook of emotmn and memory: Research and theory (pp. 307-340). Hillsdale, NJ: Lawrence Erlbaum. Conway, M. A., & Bekerian, D. A. (1987). Situational knowledge and emotions. Cognition and Emotion, I, 145-191. Damasio, A. R. (1994). Descartes' error: Emotion, research and the human brain. New York: Avon. De Rivera, J. (1977). A structural theory of the emotions. Psychological Issues, 1O, No 4, Monograph 40. Downey, G., & Feldman, S. I. (1996). Implications of rejection sensitivity for intimate relationships. Journal of Personality and Social Psychology, 70, 1327-1343. Dritschel, B. H., & Teasdale, J. D. (1991). Individual differences in affectrelated cognitive operations elicited by experimental stimuli. British Journal of Clinical Psychology, 30, 151-160. Dunn, B. R., Dunn, D. A., Andrews, D., & Languis, M. L. (1992). Metacontrol: A cognitive model of brain functioning for psychophysiological study of complex learning. Educational Psychologist, 27, 455-471.

292

Chapter 6

Ellsworth, P. C., & Smith, C. A. (1988). Shades of joy: Patterns of appraisal difforentiating pleasant emotions. Cognition and Emotion, 2, 301-331. Folkman, S., & Lazarus, R. S. (1985). If it changes it must be a process: Study of emotion and coping during three stages of a college examination. Journal of Personality and Social Psychology, 48, 150170. Folkman, S., & Lazarus, R. S. (1988). Coping as a mediator of emotion. Journal of Personality and Social Psychology, 54, 466-475. Frijda, N. H. (1986). The emotions. Cambridge and New York: Cambridge University Press. Frijda, N. H. (1987). Emotion, cognitive structure, and action tendency. Cognition and Emotion, 1, 115-143. Frijda, N. H. (1993). The place of appraisal in emotion. Cognition and Emotion, 7, 357-387. Frijda, N. H., Kuipers, P., & ter Schure, E. (1989). Relations among emotion, appraisal, and emotional action readiness. Journal of Personality and Social Psychology, 57, 212-228. Gehm, T. L., & Scherer, K. R. (1988). Relating situation evaluation to emotion differentiation: Nonmetrie analysis of cross-cultural questionnaire data. In K. R. Scherer (Ed.), Facets of emotion: Recent research (pp. 61-78). Hillsdale, NJ: Erlbaum. Graf, P., & Schacter, D. L. (1985). Implicit and explicit memory for new associations in normal subjects and amnesic patients. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 501518. Hale, S., & Jansen, J. (1994). Global processing-time coefficients characterize individual and group differences in cognitive speed. Psychological Science, 5, 384-389. Halgren, E. (1992). Emotional neurophysiology of the amygdala within the context of human emotion. In J. P. Aggleton (Ed.), The amygdala: Neurobiological aspects of emotion, memory, and mental dysfunction (pp. 191-228). New York: Wiley-Liss. Hamishfeger, K. K., & Bjorklund, D. F. (1994). A developmental perspective on individual differences in inhibition. Special Issue: Developmental perspectives on individual differences in learning and memory. Learning and Individual Differences, 6, 331-355. Hams, R. M. (1981). Conceptual complexity and preferred coping strategies in anticipation of temporally predictable and unpredictable threat. Journal of Personality and Social Psychology, 41, 380-390

C.M. van Reelcum and K.R. Scherer

293

Isen, A. M., Niedenthal, P. M., & Cantor, N. (1992). An influence of positive affect on social categorization. Motivation and Emotion, 16, 65-78. Johnson, M. K. (1983). A multiple-entry, modular memory system. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research theory (Vol. 17, pp. 1-26). New York: Academic Press. Johnson, M. K., & Multhaup, K. S. (1992). Emotion and MEM. In S.-A. Christianson (Ed.), The handbook of emoaon and memory: Research and Theory (pp. 33-66). Hillsdale, NJ: Erlbaum. Kaiser, S., & Scherer, K. R. (in press). Models of 'normal' emotions applied to facial and vocal expressions in clinical disorders. In W. F. Flack, Jr. & J. D. Laird (Eds.), Emotions in psychopathology. New York: Oxford University Press. Kolb, B., & Whishaw, I. Q. (1990). Fundamentals of human neuropsychology (3rd ed.). New York: Freeman. Kosslyn, S. M., & Koenig, O. (1995). Wet mind: The new cognitive neuroscience. New York: Free Press. Kunst-Wilson, W. R., & Zajonc, R. B. (1980). Affective discrimination of stimuli that cannot be recognized. Science, 207, 557-558. Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1990). Emotion, attention, and the startle reflex. Psychological Review, 97, 377-395. Langer, E. J., Blank, A., & Chanowitz, B. (1978). The mindlessness of ostensibly thoughtful action: The role of "placebic" information in interpersonal interaction. Journal of Personality and Social Psychology, 36, 635-642. Lazarus, R. S. (1968). Emotions and adaptation: Conceptual and empirical relations. In W. J. Arnold (Ed.), Nebraska Symposium on Motivation (Vol. 16, pp. 175-270). Lincoln, NE: University of Nebraska Press. Lazarus, R. S. (1984). Thoughts on the relation between emotion and cognition. In K. R. Soberer & P. Ekman (Eds.), Approaches to emotion (pp. 247-258). Hillsdale, NJ: Erlbaum. Lazarus, R. S. (1991). Emotion and adaptation. New York: Oxford University Press. Lazarus, R. S., & Smith, C. A. (1988). Knowledge and appraisal in the cognition-emotion relationship. Cognition and Emotion, 2, 281-300. LeDoux, J. E. (1986). Sensory systems and emotion: A model of affective processing. Integrative Psychiatry, 4, 237-248. LeDoux, J. E. (1989). Cognitive-emotional interactions in the brain. Cognition and Emotion, 3, 267-289.

294

Chapter 6

LcDoux, J. E. (1993). Emotional networks in the brain. In M. Lewis & J. Haviland (Eds.), Handbook of emotions (pp. 109-118). New York: Guilford Press. LcDoux, J. E. (1994). Emotion, memory, and the brain. Scientific American, June, 32-39. LeDoux, J. E. (1996). The emotional brain: The mysterious underpinnings of emotional life. New York: Simon and Schuster. Leventhal, H. (1979). A perceptual-motor processing model of emotion. In P. Pliner, K. Blankstein, & I. M. Spigel (Eds.), Advances in the study of communication and affect: Perception of emotion in self and others (Vol. 5). New York: Plenum. Leventhal, H. (1984). A perceptual motor th~ry of emotion. In K. R. Scherer & P. Ekman (Eds.), Approaches to emotion (pp. 271-292). Hillsdalr NJ: Erlbaum. Leventhal, H., & Scherer, K. R. (1987). The relationship of emotion to cognition: A functional approach to a semantic controversy. Cognition and Emotion, 1, 3-28. Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review, 95, 492-527. Lord, R. G., & Levy, P. E. (1994). Moving from cognition to action: A control theory perspective. Applied Psychology: An International Review, 43, 335-367. Lorig, T. S., Singer, J. L., Bonanno, G. A., Davis, P., & r al. (1994). Repressor personality styles and EEG patterns associated with affr162 memory and thought suppression. Imagination, Cognition and Personality, 14, 203-210. Manstead, A. S. R., & Tctlock, P. E. (1989). Cognitive appraisals and emotional experience: Further evidence. Cognition and Emotion, 3, 225240. Matthews, G., & Amelang, M. (1993). Extraversion, arousal theory and performance: A study of individual differences in the EEG. Personality and Individual Differences, 14, 347-363. Matthews, G.A., & Harley, T. A. (1993). Effects of r and selfreport arousal on semantic priming: A connectionist approach. Journal of Personality and Social Psychology, 6.5, 735-756. Matthews, G., Davies, D. R., & Lees, J. L. (1990). Arousal, r and individual differences in resource availability. Journal of Personality and Social Psychology, .59, 150-168.

C.M. van Reekum and K.R Scherer

295

Mauro, R., Sato, K., & Tucker, J. (1992). The role of appraisal in human emotions: A crosscultural study. Journal of Personality and Social Psychology, 62, 301-317. McCrae, R.-R. (1993). Openness to experience as a basic dimension of personality. Imagination, Cognitmn and Personality, 13, 39-55. McGraw, K. M. (1987). Guilt following transgression: An attribution of responsibility approach. Journal of Personality and Social Psychology, 53, 247-256. Messick, S. (1994). The matter of style: Manifestations of personality in cognition, learning, and teaching. Educational PsyChologist, 29, 121136. Murphy, D., & Zajonc, R. B. (1993). Affect, cognition, and awareness: Affective priming with optimal and suboptimal stimulus exposures. Journal of Personality and Social Psychology, 64, 723-739. Murray, N., Sujan, H., Hirt, E. R., & Sujan, M. (1990). The influence of mood on categorization: A cognitive flexibility interpretation. Journal of Personality and Social Psychology, 59, 411-425. Neuberg, S. L., & Newson, J. T. (1993). Personal need for structure: Individual differences in the desire for simpler structure. Journal of Personality and Social Psychology, 65, I 13-131. Niedenthal, P. M. (1990). Implicit perception of affective information. Journal of Experimental Social Psychology, 26, 505-527. Niedenthal, P. M., & Kitayama, S. (1994). The heart's eye: Emotional influences in perception and attention. San Diego, CA: Academic Press. Oatley, K., & Johnson-Laird, P. N. (1987). Towards a cognitive theory of emotions. Cogniaon and Emotion, 1, 29-50. Ohman, A. (1986). Face the beast and fear the face: Animal and social fears as prototypes for evolutionary analyses of emotion. Psychophysiology, 23, 123-145. Ohman, A. (1988). Preattentive processes in the generation of emotions. In V. Hamilton, G. H. Bower, & N. H. Frijda (Eds.), Cognitive perspectives on emotion and motivation (pp. 127-144). Dordrecht: Kluwer Academic Publishers. Ohman, A., & Soares, J. J. F. (1994). "Unconscious anxiety": Phobic responses to masked stimuli. Journal of Abnormal Psychology, 103, 231-240. Ohman, A., Dimberg, U., & Esteves, F. (1989). Preattentive activation of aversive emotions. In L. G. Nilsson & T. Archer (Eds.), Aversion,

296

Chapter 6

avoidance, and anxiety: Perspectives on aversively motivated behavior (pp. 169-193). Hillsdale, NJ: Erlbaum. Parkinson, B. (1997). Untangling the appraisal-emotion connection. Personality and Social Psychology Review, 1, 62-79. Parkinson, B., & Lea, M. F. (1991). Investigating personal constructs of emotion. Briash Journal of Psychology, 82, 73-86. Parkinson, B., & Manstead, A. S. R. (1992). Appraisal as a cause of emotion. Review of Personality and Social Psychology, 13, 122-149. Parkinson, B., & Manstead, A. S. R. (1993). Making sense of emotions in stories and social life. Cognition and Emotion, 7, 295-323. Peterson, C., & Seligman, M. E. (1984). Causal explanations as a risk factor for depression: Theory and evidence. Psychological Review, 91, 347374. Piaget, J. (1937/1955). The child's construction of reality. London: Routledge and Kegan Paul. Press, A. N., Crockett, W. H., & Delia, J. G. (1975). Effects of cognitive complexity and of perceiver's set upon the organization of impressions. Journal of Personality and Social Psychology, 32, 865-872. Pruzinsky, T., & Borkovec, T.-D. (1990). Cognitive and personality characteristics of worriers. Behaviour Research and Therapy, 28, 507512. Reisenzein, R., & Spielhofer, C. (1994). Subjectively salient dimensions of emotional appraisal. Motivation and Emotion, 18, 31-77. Reisenzein, R., & Hofmann, T. (1990). An investigation of dimensions of cognitive appraisal in emotion using the repertory grid technique. Motivation and Emotion, 14, 1-26. Reisenzein, R., & Hofmann, T. (1993). Discriminating emotions from appraisal-relevant situational information: Baseline data for structural models of cognitive appraisals. Cognition and Emotion, 7, 271-293. Rhodewalt, F., Morf, C., Hazlett, S., & Fairfield, M. (1991). Selfhandicapping: The role of discounting and augmentation in the preservation of self-esteem. Journal of Personality and Social Psychology, 61, 122-131. Roseman, I. J. (1984). Cognitive determinants of emotion: A structural theory. In P. Shaver (Ed.), Review of personality and social psychology (Vol. 5, pp. 11-36). Beverly Hills, CA: Sage. Roseman, I. J. (1991). Appraisal determinants of discrete emotions. Cognition and Emotion, 5, 161-200.

C.M. van Reekum and K.R. Scherer

297

Roseman, I. J., Spindel, M. S., & Jose, P. E. (1990). Appraisal of emotioneliciting events: Testing a theory of discrete emotions. Journal of Personality and Social Psychology, 59, 899-915. Russel, D., & McAuley, E. (1986). Causal attributions, causal dimensions, and affective reactions to success and failure. Journal of Personality and Social Psychology, 50, I 174-1185. Schacter, D. L. (1987). Implicit memory: History and current status. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 501-518. Schacter, D. L. (1996). Searching for memory: The brain, the mind, and the past. New York: Basic Books. Scheier, M. F., & Carver, C. S. (1985) Optimism, coping, and health: Assessment and implications of generalized outcome expectancies. Health Psychology, 4, 219-247 Scherer, K. R. (1982). Emotion as a process: Function, origin, and regulation. Social Science Information, 21,555-570. Scherer, K. R. (1984a). Emotion as a multicomponent process: A model and some cross-cultural data. In P. Shaver (Ed.), Review of personality and social psychology (Vol. 5, pp. 37-63). Beverly Hills, CA: Sage. Scherer, K. R. (1984b). On the nature and function of emotion: A component process approach. In K. R. Scherer & P. Ekman (Eds.), Approaches to emotion (pp. 293-318). Hillsdale, NJ: Erlbaum. Scherer, K. R. (1986). Vocal affect expression: A review and a model for future research. Psychological Bulletin, 99, 143-165. Scherer, K.R. (1987). Vocal assessment of affective disorders. In J.D. Maser (Ed.), Depression and expressive behavior. (pp. 57-82). Hillsdale, NJ: Erlbaum. Scherer, K. R. (1988). Criteria for emotion-antecedent appraisal: A review. In V. Hamilton, G. H. Bower, & N. H. Frijda (Eds.), Cognitive perspectives on emotion and motivation (pp. 89-126). Dordrecht: Nijhoff. Scherer, K. R. (1992). What does facial expression express? In K. Strongman (Ed.), International Review of Studies on Emotion (Vol. 2, pp. 139-165). Chichester: Wiley. Scherer, K. R. (1993a). Neuroscience projections to current debates in emotion psychology. Cognition and Emotion, 7, 1-41. Scherer, K. R. (1993b). Studying the emotion-antecedent appraisal process An expert system approach. Cognition and Emotion, 7, 325-355.

298

Chapter 6

Schcrcr, K. R. (1997). Profiles of cmotion-antccexicnt appraisal: testing theoretical predictions across cultures. Cognition and Emotion, I 1, 113150. Scherer, K. R. (in press). Appraisal theories. In T. Dalgleish, & M. Power (Eds.), Handbook of cogmtion and emotion. Chichester: Wiley. Schneider, W. (1985). Toward a model of attention and the development of automatic processes. In M. I. Posner & O. S. Matin (Eds.), Attention and performance )21 (pp. 475-492). Hillsdale, NJ: Eflbaum. Schroder, H. M., Driver, M. J., & Streufert, S. (1967). Human information processing: Individuals and groups functioning in complex social situations. Now York: Holt, Rinehart and Winston Sehroth, M. L., & Lund, E. (1994). Relationships between need achievement, sensation seeking and cognitive performance. Personality and Individual Differences, 16, 861-867. Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attention and a general theory. Psychological Review, 84, 127-190. Smith, C. A. (1989). Dimensions of appraisal and physiological response in emotion. Journal of Personality and Social Psychology, 56, 339-353. Smith, C. A., & Ellsworth, P. C. (1985). Patterns of cognitive appraisal in emotion. Journal of Personality and Social Psychology, 48, 813-838. Smith, C. A., & Ellsworth, P. C. (1987). Patterns of appraisal and emotion related to taking an exam. Journal of Personality and Social Psychology, 52, 475-488. Smith, C. A., & Lazarus, R. S. (1993). Appraisal components, core relational themes, and the emotions. Cognition and Emotion, 7, 233269. Smith, C. A., Griner, L. A., Kirby, L. D., & Scott, H. S. (1996). Toward a process model of appraisal in emotion. In N. H. Frijda (Ed.), Proceedings of the IXth Conference of the International Society for Research on Emotions, ISRE'96. Storrs: ISRE Publications. Smole aars, A. J., & Schutzelaars, A. J. H. (1986/87). On 'cognitive' semantics of emotion words: Solomon quasi ecologically tested. Journal of Semantics, 5, 207-231. Solomon, R. C. (1976). The passions: The myth and nature of human emotion. Garden City, NY: Doubleday. Sommers, S. (1981). Emotionality reconsidered: The role of cognition in emotional responsiveness. Journal of Personality and Social Psychology, 41,553-561.

C.M. van Reekum and K.R. Scherer

299

Stelmaek, R. M., Houlihan, M., & McGarry-Roberts, P. A. (1993). Personality, reaction time, and event-related potentials. Journal of Personality and Social Psychology, 65, 399-409. Stenberg, G., Wendt, P. E., & Risberg, J. (1993). Regional cerebral blood flow and extraversion. Personality and Individual Differences, 15, 547554. Stipek, D., Weiner, B., & Li, K. (1989). Testing some attribution-emotion relations in the People's Republic of China. Journal of Personality and Social Psychology, 56, 109-116. Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18, 643-662. Stuss, D. T., & Benson, D. F. (1983). Emotional concomitants of psychosurgery. In K. M. Heilman & P. Satz (Eds.), Neuropsychology of human emotion (pp. 111-140). New York: Guilford Press. Teasdale, J. D. (1988). Cognitive vulnerability to persistent depression. Cognition and Emotion, 2, 247-274. Teasdale, J. D., & Bamard, P. J. (1993). Affect, cognition, and change: Remodelling depressive thought. Hove (UK): Erlbaum. Tesser, A. (1990). Smith and Ellsworth's appraisal model of emotion: A replication, extension, and test. Personality and Social Psychology Bulletin, 16, 210-223. Tetlock, P. E. (1983). Accountability and complexity of thought. Journal of Personality and Social Psychology, 45, 74-83 Thompson, E. P., Chaiken, S., & Hazlewood, J. D. (1993). Need for cognition and desire for control as moderators of extrinsic reward effects: A person • situation approach to the study of intrinsic motivation. Journal of Personality and Social Psychology, 64, 987-999. Tobacyk, J. J., Driggers, E. C., & Hourcade, J. (1991). Self-monitoring and psychological type: A social cognitive information processing model. Journal of Psychological Type, 22, 33-38. Tous, J. M., Fuste, A., & Vidal, J. (1995). Hemispheric specialization and individual differences in cognitive processing. Personality and Individual Differences, 19, 463-470. Tulving, E. (1972). Episodic and semantic memory. In E. Tulving & W. Donaldson (Eds.), Organisation of memory. New York: Academic Press. pp. 381-403. Waller, J. E. (1994). Dual-processing in value self-confrontation: The role of need for cognition. Journal of Social Issues, 50, 175-188.

300

Chapter 6

Weiner, B. (1982). The emotional consequences of causal attributions. In M. S. Clark & S. T. Fiske (Eds.), Affect and r (pp. 185-209). Hillsdale, NJ: Erlbaum. Weiner, B. (1986). An attributional theory of motivation and emotion. New York: Springer. Weiner, B., Amirkhan, J., Folkes, V. S., & Verctte, J. A. (1987). An attributional analysis of excuse giving: Studies of a naive theory of emotion. Journal of Personality and Social Psychology, 52, 316-324. Weiner, B., Graham, S., & Chandler, C. (1982). Pity, anger, and guilt: An attributional analysis. Personality and Social Psychology Bulletin, 8, 226-232. Weiner, B., Russel, D., & Lcrman, D. (1979). The cognition-emotion process in achievcment-rdated contexts. Journal of Personality and Social Psychology, 37, 1211-1220. Wells, A., & Matthews, G. (1994). Attention and emotion: A clinical perspective. Hove (UK): Erlbaum. Williams, J. M. G., Mathews, A., & M a c L ~ , C. (1996). The emotional Stroop task and psychopathology. Psychological Bulletin, 120, 3-24. Woodfield, E., Jones, G. V., & Martin, M. (1995). Selective processing of stress-related words in colour naming. Personality and Individual Differences, 19, 385-387. Zajonc, R. B. (1980). Feeling and thinking: Preferences need no inferences. American Psychologist, 2, 15 l- 176. Zajonc, R. B. (1984). On the primacy of affect. In K. R. Scherer & P. Ekman (Eds.), Approaches to emotion (pp. 259-270). Hillsdale, NJ: Erlbaum. Zucker, G. S., & Weiner, B. (1993). Conservatism and perceptions of poverty: An attributional analysis. Journal of Applied Social Psychology, 23, 925-943. Author Notes

The authors would like to thank Leonard Bcrkowitz, Craig Smith, Leslie Kirby, and in particular Tom Johnstone, for their extremely helpful comments and suggestions on earlier versions of this chapter. Preparation of this manuscript was supported by a grant (11-37504.93) of the Swiss National Scientific Research Foundation to Klaus R. Schcrcr.

Cognitive Science Perspectives on Personality and Emotion - G. Matthews (Editor) 9 1997 Elsevier Science B.V. All rights reserved. CHAPTER 7

Modeling Individual Differences in Negative Information Processing Biases Greg J Siegle and Rick E. Ingram Research on depression and research on personality have long been very different fields. Research on personality and individual differences has concentrated on figuring out why individuals respond differently to similar stimuli, process similar information differently, and act differently in similar social situations. It has often focused on understanding differences between individuals who are considered "normal". In contrast, research on depression has focused on commonalties among people who are considered depressed. Questions such as why depressed people respond similarly to negative stimuli and why depressed people appear similarly biased in their processing of negative information have pervaded this literature. The disciplines of studying differences among otherwise similar people and similarities between groups of very different people have recently begun to converge as theory begins to suggest that depressed people differ qualitatively on a number of trait-like dimensions, and that mood may affect the expression of individual differences other than depression itself. For example, Klein, Wonderlich, and Shea (1992) suggest that enduring personality characteristics such as dependency and perfectionism appear to be associated with qualitative differences in the expression of depression. Similarly, researchers such as Persons and Miranda (1992) have found that biased memory for negative things, which was once thought to be trait-like, appear to vary with affective state. These paths have begun to suggest that depression can affect the assessment of personality, and that personality is intimately associated with the onset, maintenance, and recovery from depression. An area in which this synthesis has been extremely productive regards the mediating role of personality variables upon biased information processing in dysphoric individuals. Some dysphoric individuals appear to systematically attend to negative over positive or neutral information (MacLeod & Mathews, 1991), remember disproportionate amounts of negative information (Blaney, 1986), and engage in excessive processing of negative information or "rumination". Such information processing biases

302

Chapter 7

have been linked by numerous researchers to a vulnerability for depression (e.g., Ingrain, 1984). Other dysphoric and even clinically depressed individuals do not appear to show these biases (e.g., Siegle, Ingram, & Matt, 1996a). To the extent that personality and individual difference variables may mediate these biases, it will be important to understand their role in depression. Such variables may be particularly valuable in predicting who will and will not be vulnerable to depression after a period of dysphoria. The goal of this chapter is to illustrate some of the ways in which personality factors might help to explain variation in information processing in dysphoric and depressed individuals. Towards this end, we will introduce a number of proposed roles for personality and individual differences in explaining vulnerability to depression. We then describe a framework for evaluating these roles using computational neural networks. To illustrate the use of this framework, a computational neural network described by Siegle, Ingram and Matt (1995; 1996b) will be employed to simulate the roles of three personality variables in mediating observed information processing biases on two simulated information processing tasks. Each simulated personality variable will be related to the goal of integrating depression and personality research through the following steps. First, ways a personality variable may be simulated by an analogous aspect of the network model will be explained. The effect of the parameter on the network's performance on simulated information processing tasks will be assessed before and atter the network has undergone a simulated analog of prolonged dysphoria. Ways in which the parameter affects the network's simulated behavior will be generalizeA to ways in which the analogous personality characteristic might affect performance. Evidence will be provided for this type of a relationship in dysphoric and nondysphoric people. After discussing each variable, we will provide a few integrative conclusions regarding the applicability of this type of framework for investigating the role of personality in vulnerability to depression.

Personality Research and Vulnerability to Depression: A History Many individual difference variables appear to influence how people process information, including what they attend to, what they remember, and how they interpret their life-experiences. Two such variables are cognitive structures and coping processes. A third individual-difference variable which may bear distinct correlates on the onset, maintenance, and recovery from

G.J. Siegle and R.E. lngram

303

depression is intellect. While the latter variable has not received a great deal of attention from depression theorists, the former variables have. A common theme underlying several contemporary models of depression is that individuals who experience this disorder, or who are at risk for its experience, are characterized by cognitive structures that affect information processing in a dysfunctional manner. For example, models by Beck (1967, 1976), Ingram (1984), and Teasdale (1988), although differing on some dimensions, converge on the notion that the onset and maintenance of depression is influenced, to a significant degree by dysfunctional information processing that is determined by negative or depressive cognitive structures or schemas. A wealth of empirical data has, in fact, shown that depressed individuals process information in a variety of domains in a negative and selfdefeating manner (e.g., Lester & Schaeffler, 1993; Schill & Kramer, 1991). From a functional standpoint, such schema driven information processing is thought to lead to life experiences being processed in a fashion which induces and maintains depression. Specifically, positive and potentially selfenhancing experiences are processed poorly or not at all while neutral experiences take on a negative tone, and genuinely negative life experiences are viewed as catastrophic. The negative cognitive triad proposed several decades ago by Beck (1967) still adequately sums up the interpretation of life experiences by depression-prone individuals; one's self, future, and world are viewed in quite negative terms. It is hard to overestimate the implications of this way of structuring life experiences for understanding depression. Although cognitive structures have received a great deal of attention from depression theorists, other cognitive processes are also considered important. Of particular relevance for this chapter, is coping. The literature on coping is extremely large and has been reviewed in depth elsewhere (Zeidner & Endler, 1996). Instead, our focus is on a particular style of cognitive coping that several depression researchers have suggested may have important implications for either the alleviation of a depressive state or for its prolongation. Ingram (1984) suggested that one of the ways in which the activity of dysfunctional cognitive structures could be either exacerbated or diminished was through active efforts at cognitive coping. Ingram proposed that once activated, negative cognitive structures are consistently reactivated by recycling energy through a network of other interconnected cognitive structures. Phenomenologically, this process may be experienced as a ruminative state by the individual. Cognitive efforts and behaviors intended to interrupt this activity may be considered to constitute coping. Alternately, individuals who make little effort to interrupt this cognitive recycling process

304

Chapter 7

are apt to experience longer, more severe periods of depression. Coping is thus seen in this model as a major determinate of several important parameters of the depressive state. More recently, Nolen-Hoeksema, Morrow, and Fredrickson (1993) have expanded the notion of cognitive coping to give it a central role in what she terms the "response styles theory" of depression. In brief the authors propose that an important individual difference variable in depression is the extent to which people are inclined to ruminate on the symptoms of depression. Those who ruminate longer and deeper are proposed to experience more protracted and severe periods of depression while those who ruminate less will have correspondingly shorter and less intense depressive experiences. Other models of depression have given rumination a similarly central role (Pyszczynski & Greenberg, 1987). While we would view rumination in a broader fashion than do Nolen-Hoekserna et al. (1993; e.g., rumination on a variety of depression related phenomena in addition to symptoms may be important), we believe that the central construct of rumination is an extremely important variable in the maintenance, if not the onset, of depression.

Simulating Aspects of Depression and Personality on a Computer The research described above has provided many useful hypotheses about potential relationships between vulnerability to depression and personality. Most of these hypotheses reflect causal models (e.g., rumination prolongs depression). A common criticism of clinical causal models is that they are often fundamentally unsatisfying to people wishing to understand how aspects of a model interact. For example, diagrams composed of circles and arrows are valuable tools for helping readers understand an author's conception of qualitative relationships between measured or hypothetical constructs. However, such diagrams do not convey the precise nature of mediational interactions between causal agents. Moreover, the entirety of the processes which are assumed to interact, and potentially even feed back to reinfluence each other are rarely directly observed in the generation of causal explanations; rather specific steps in the causal path are observed and large causal diagrams are constructed from these paths. Thus, the technologies used to specify and evaluate theoretical causal models have rarely caught up with our complex intuitions of causal models themselves. A model constructed within a framework which allows direct inspection of the processes involved in causal mediation of clinical phenomena would be much more desirable. Computational models provide this technology, allowing their

G.J. Siegle and R.E. Ingrain

305

creators to interactively observe feeAback between modeled constructs and to experiment with how changes m these constructs might change observed behaviors. They also allow users to investigate how the addition of variables, which were not empirically measured, might influence the observed behaviors. The use of computer simulations to understand behavior has thus gained a growing following in recent years for addressing causal mediation processes in this more satisfying and rigorous manner. Computational modeling has many advantages for psychologists interested in understaadmg causal mediation processes in abnormal behavior. Some include its ability to aid in the translation of theory to a rigorously specified, empirically testable causal model, to aid in understanding behaviors, to generate new hypotheses about abnormal processes, to promote the creation of novel clinical interventions, and to integrate various "granularities" of research (Siegle, 1996). One particular type of model which has sparked a great deal of interest among the computational clinical community is called a neural network model. Neural network models represent information throughout a connected network of units. "Activation" travels between units representing a flow of information. Units in a conneetionist network have thus been likened to biological neurons which receive information on dendrites and send out an aggregate of that information over axons. Neural network models have become popular for many reasons including their predictive power (Sarle, 1994), biological congruity (Cohen & Servan-Schreiber, 1992), ability to handle noisy data (Cohen & Servan-Sehreiber, 1992), and the natural way in which they can be used to model information processing tasks (McClelland, Rurnelhart & Hinton, 1986). Additionally, some neural networks can be used to model phenomena which are difficult to represent using more traditional symbolic modeling techniques, such as behaviors for which explicit governing rules are not known (Heeht-Neilson, 1990). Moreover, Tryon (1993) has shown that neural network models help to provide resolutions for traditional schisms in psychology such as the mind/body problem and the nature vs. nurture debate.

Why use computer simulations to investigate dysphoria? While computational models, and specifically neural networks, are interesting vehicles for psychologists in general, they are particularly useful for researchers interested in understanding the role of cognition in depression because neural networks are natural biologically congruent extensions of

306

Chapter 7

common models of depressive information processing. Since the "cognitive revolution" of the late seventies and early eighties, semantic network models have been proposed to explain information processing biases in depression. Adopting Collins and Loflus's (1975) notion of mental events as propositions spreading between loci representing bits of knowledge in a "semantic network", Bower (1981) modeled the normal experience of emotion by suggesting that emotions might also be nodes in such a semantic network. Ingram (1984) suggested that when such emotion nodes become strongly activated, cognitive aspects of depression might occur. This approach appears to account for many types of biased information processing observed in depressed people as well as the onset and maintenance of depression (Ingram, 1984) and its eventual treatment (Ingram & Hollon, 1986; Morrow & Nolen-Hockscma, 1990) and potential recurrence (Tcasdale, 1988). The distinction between semantic networks composed of information processing nodes, each representing beliefs or logical propositions, and distributed conncctionist networks, in which each node is individually meaningless, is important. Yet, work relating the two representations has served to provide important bridges between subdisciplines of experimental psychology (Anderson, 1990; Blank ct al., 1992; Yatcs & Nasby 1993). For example, Yates and Nasby (1993) identify six fundamental assumptions common to semantic network and neural network models of associative memory including: l) Both types of models use nodes representing aspects of propositional knowledge; 2) Propositions are associated at the time of encoding; 3) Information in the network becomes conscious when it is activated above some threshold; 4) Environmental stimuli activate some nodes in the network; 5) Activation can spread in varying degrees; 6) Consciousness may be understood as involving the activation of nodes. The semantic network approach thus provides an important historical context for understanding the subsequent development of neural network models of depression. The cognitive orientation, and the focus on the roles of memory and attention in depression used throughout semantic network models may provide a basis for their use in corresponding neural network models. Neural network reformulations of semantic network representations of depression preserve the benefits of semantic network models while

G.J. Siegle and R.E. Ingram

307

providing a biologically congruous approach to examining mechanisms responsible for the onset and maintenance of depression. More practically, many types of experiments can be performed on a computational neural network model which it would be unethical or difficult to perform on a depressed human. For example, it would often be desirable to test theories of depression by progressively depressing a person, bit by bit, until he or she were on the brink of complete despair. At each step of the induction, the person could be assessed on a number of variables to ascertain whether the person's behavior follows that predicted by a theory. Such an experiment would, of course, be unethical. As a substitute, if a computational model of analogs of processes thought to operate in depression could be created, the computer's behavior on some task could be analyzed at various levels of operation of these processes with relatively few ethical complications.

Why simulate personality factors and individual differences? Our goal in simulating aspects of personality within a computational model is to generate a number of possibilities for the role of personality in mediating depressive information processing biases. It is our hope that simulations can help to show how many avenues for differential information processing might exist based on individual differences in just a few constructs. R is not the attempt of this paper to provide any concrete empirical demonstrations of the role these factors may play in depression. Rather, the following simulations serve as hopeful "might be's" and avenues for empirical confirmation. By simulating personality variables on a computer, theories about them can be rigorously specified. Hidden ambiguities may correspondingly become apparent, and thus theories of both normal and pathological behavior will stand a better chance of validation. Moreover, computational models which are biologically motivated often serve to suggest physiological correlates of simulated phenomena, in this case, personality variables. Another reason to consider simulating personality factors on a computer involves the determination of vulnerability markers for depression. If some simulated personality factors are associated with greater expression of characteristics associated with depression after the model is subjected to a computational analog of a depression induction, we can examine the effects these factors have on a model before it assumes a state analogous to

308

Chapter 7

depression. By examinm"g the pre-pathologized model, we may determine markers for people at risk for depression. Two final reasons for simulating the role of personality and individual difference variables in depression will become more apparent later in the chapter. First, neural networks are particularly suited for investigating the role of personality variables because parameters inherent to network simulations correspond neatly to aspects of personality. Without adding variables to an existing simulation, aspects of personality implicit in the models can be investigated. Second, by simulating characteristic individual differences in depressive information processing biases, variation in the behavior of a computational model of depression earl be attributed to factors other than random noise. In this way, the specific variation in human information processing experiments can be approximated more closely than if personality factors were not simulated. A short introduction to neural networks

To appreciate the types of models which are discussed in the following sections, a basic familiarity with computational neural network models is assumexi. The following section discusses some of the basic concepts necessary for understanding these models. A number of excellent review articles are available which summarize specific types of conncctionist networks (Rumelhart, Hinton, & Mr 1986), the mathematics and mechanics of computational simulations of connectionist networks (e.g., Arbib, 1987; Hecht-Neilson, 1990), and the relative strengths of this type of modeling over traditional modeling technologies (Barnden, 1995), for the interested reader. Neural network models represent information throughout a connected network of individually meaningless units nodes. Information is represented in a distributed fashion, as a function of the simultaneous activation of multiple nodes. Each unit receives "activation" from other nodes to which it is connected in response to the activation of these nodes. The unit then sends a transfer function of the activations coming into it to other nodes to which it is connected, much as a biological neuron sends a function of activations from its dendrites to other neurons through its axons. For example, a collection of nodes, when activated together, may be said to represent a concept such as "sun". Activation of these nodes may spread to another collection of nodes which, together, represent the concept "moon". This process might be interpreted as a computational analog of a mental

G.J. Siegle and R.E. Ingram

309

association between sun and moon. Environmental or internal factors (e.g., visual stimuli) which cause some nodes to be activated arc dctcrmmexi and modeled explicitly by the neural network modeler. The pattern of connections between nodes, or the network's "architecture" governs what types of information may enter the network, the manner in which nodes may activate each other, and what information may be said to flow out of the network. Thus, neural network's response to a stimulus generally involves successive activation of a number of nodes in the network. This pattern of activations can represent an association in memory between two stimuli, a reaction to an external stimulus, or some other construct based on the network's architecture and the designer's conception of the network. The processes operating within connectionist networks can be assumed to correspond to neuronal, cognitive, or behavioral events based on the intuitions of the network's designer. By strategically modifying the strengths of connections between nodes, the network can be made to produce a specific set of activations in response to another set of activations. This process has been likened to making the network "learn" an association of a stimulus with a response. Numerous procedures for allowing a network to learn associations in this fashion have been proposed (e.g., Fallman & Lebiere, 1991; Rumelhart, Hinton & Williams, 1986). Once a connectionist network model has been created, its behavior can be evaluated on a number of dimensions. The choice of what dimension is to be evaluated generally reflects the processes which the network is designed to simulate. For example, if the network is designed to simulate performance on some information processing task, associations made by the network could be compared to associations made by humans to a stimulus; the frequency of the network's erroneous associations could be measured as an analog of human error rates. Similarly, a network may take a number of associative steps to settle on a learned association. The number of associative processing cycles or "epochs" the network needs to associate a stimulus with a particular response can be examined as an analog of reaction time. As may be inferred from this introduction, connectionist models are generally associated with a number of parameters which can be varied such as the relative strengths of connections within the network, the rate at which the network can "learn", and the number of nodes devoted to representing types of information within the network. Systematic manipulations of these parameters will be termed "perturbations" throughout this review. Such perturbations can be used to make the network respond in different ways to

310

Chapter 7

similar stimuli. Thus, perturbations of network parameters may be seen as one way of approximating individual differences. Often, perturbations of network parameters are used to make a network function in a manner associated with psychopathology. Individual differences will be simulated using three types of "perturbations" to a network model, representing three factors which are believed to induce individual differences in humans. As humans are expected to differ in coping styles, or the techniques they use to process negative stimuli, network parameters corresponding to the ways in which information feeds through the network in "response to a stimulus will be varied. By examining the network's performance on simulated analogs of information processing tasks given more or less of some method for processing negative information, claims may be made about how different humans might process similar negative information. Another way in which individuals are hypothesize~ to differ reflects different learning styles. Some individuals learn slowly and thus may take longer to assimilate new information. To gauge the effects of such a learning style on information processing in depression, parameters affecting the network's learning style can be manipulated. Finally, just as human associations are based on what they have learned, associations made by the network are based on the stimulus-response pairs on which it has been trained. Thus, to approximate characteristics of different individuals, the stimulus-response pairs on which the network is trained will be modified. For example, to represent an analog of depression, more training will be given to the network on stimuli which are deemed to be negative. Similarly, to represent the types of experiences which might help an individual recover from depression, the network can be trained on stimuli deemed not to be negative, after having been originally overtrained on negative information. Manipulations of the sequence of training examples will be important in simulating a variety of sequences of life events which may lead to information processing biases. Because the number and types of parameters which may occur in a conneetionist model are too numerous to illustrate in detail here, the parameters which are manipulated for each simulation will be explained as they are referenced in the text.

A computational framework for investigating affective information processing Before simulating the effects of personality variables on affcetive information processing, it will be useful to have a computational model of

G.J. Siegle and R.E. Ingrain

311

affective information processing which captures basic notions about how most dysphoric and nondysphoric individuals behave. Towards that end, Siegle, Ingram, and Matt (1995; 1996b) have developed a computational neural network model of affective information processing which simulates the behavior of dysphoric and nondysphoric individuals on two information processing tasks. Their model is depicted m Figure 1.

9

.-'.~. ' ~ . - ~ i . , ~

g~

....... ,:~'.,~,'":,i~~"'~,-~:~ ..~.~:~@.~i~

~}~

'

. . . . . .

'~:~.~r

~,~,~, .............

J ..ll,,~t...,,t~,.

! ...................

I.[.

I.

.1

.

.

.

-'..~..,,..-.~t ~?~;..~ia..,.~ii-..,~ ..... ~,,.~.-.:f~,~::.~.~,.,,.~}l

,

i i;. . . . ~iiiii~i~}~i i!i~ ....

~i

i iiiiii:iili!i!i!iii

~,~i~".~!.".'-"~:i::~.'-i~:~,~i~:~.".-'-~:~ i:~','~" .......... !!!,i,,i~i!~iiiiii!iiii!i!iill

....

.....~.',% ~

~.~.

~'. ~ ~."

~:.

r~NN

~ ~ ~

~.~:.~ :~.i~.,:."~i

~

" ,

~ i l i

i!!i!iiiiiiiiiii!ii!i! iiiili?i/iiiiii'iii;iiiiiii/!ii~i

~N~I.. ~ii}i!i;i;i!iiiiiii . . . . ~ili! iiiii!ii!iiiiiligi'>'~!i~ii:

/ The affectivesemantic loop

.,iiiii!ii!iiiii!, ii!i!ii!!!!i!!ii!ili!li!iii!ili!i!!ii!i! !iiii!!iiiiiiiiii!iiiii!ii!i!ii!iiiiiill

Orthographic Features

Figure 1. A neural network platform for investigating affective information processing.

The model is an attempt to formalize Bower's (1981) notion of feedback between a person's internal representations of affective and semantic information. In Siegle et al.'s (1996b) computational model, 10 nodes representing the most basic perceptual characteristics of stimuli ("orthographic" nodes) are made to activate 10 other nodes which represent

312

Chapter 7

the semantic content of the stimulil . T h e orthographic aspects of a stimulus may be thought of as the patterns of light and dark which make up an observed visual image whereas its semantic content represents the mental representation of a learned concept with which the visual representation is associated. The semantic nodes activate two more nodes which represent the affective valence of stimuli. One valence node represents the positivity of a stimulus and the other valence node represents its negativity. A neutral valence is representeA by neither of the valence nodes being active. To capture Ingram's (1984) notion of feedback between the affective and semantic aspects of a stimulus, the nodes representing the affective valence of the stimulus can reactivate the nodes representing the stimulus's semantic content a variable number of times. When activation feeds back to the semantic or valence nodes the activation is computed as the sum of the previous activation plus one half of the activation from the incoming connections. In this way, the feedback from other nodes does not immediately exceed the original activation from the stimulus. The feedback between the affective and semantic representations of a stimulus is congruous with Tucker and Derryberry's (1992) notion of feedback between the parts of the brain responsible for the affective and semantic representation of a stimulus, namely the hippocampus and the amygdala, though we have not tried to model other aspeOs of either the hippocampus or the amygdala in this framework. To simulate the knowledge base that a random individual might have, the network was trained on three positive, three negative and three neutral stimuli. Stimuli were represented in a localist fashion in which only one orthographic, one semantic and one valence node was expected to be active for a given stimulus. While the restriction to a localist representation is not essential for the following simulations, it is useful for illustrating how network connections changed when various aspects of personality were simulated. Training involved presenting a simulated orthographic representation of a stimulus to the network, observing the network's semantic and valence nodes, and adjusting the weights within the network until the desired semantic and valence representations were achieved using a modified back-propagation learning algorithm (Rumelhart, Hinton, & Williams, 1986) in which weights were adjusted after feedback occurred 10 times between the 1 For most simulations, only the first nine nodes were used. The tenth node was reserved for simulations involving "novel" stimuli to which the network was not exposed during its initial training period.

G.J. Siegle and R.E. lngram

313

affective and semantic nodes. No claim is made here that human learning actually takes place via a back-propagation algorithm. Rather, it is assumed that as in the back-propagation algorithm, humans change their associations with stimuli based on their experiences, and more association to some stimulus means that it is learned better. Training continued until the sum of the network's mean squared error in the semantic and valence nodes was below 0.004 for a block of all nine stimuli.

Simulating the inductmn of depression in the network In order to use of the neural network model to represent information processing biases in depression, it is necessary to discuss how the network is to assume a simulated analog of depression. A popular theory suggests that the induction of depression involves a single, pervasive negative life event or loss experience (e.g., Beck 1974, Brewin, Andrews, & Gotlib, 1993, Paykel 1979) which is continuously thought about. This process can be operationalized by allowing the network to be exposed to one negative stimulus for a prolonged period after it has been trained on equal numbers of positive, negative and neutral stimuli. Thus, to represent the induction of depression, the network was trained on a single negative stimulus for 70 epochs at~er the network's initial training was complete. The network described above was implemented in the PLANET neural network simulation environment (Miyata, 1991) on a Sun SPARC 1 computer. PLANET is an environment in which neural network simulations may be constructed using a language developed specifically for that purpose. Users can interactively examine activations within the network and its contents, interactively "train" the network to associate inputs with outputs, and observe the resulting error rates. The PlaNet code representing the network and for presenting stimuli are available from the first author upon request. To assess the validity of this framework for representing information processing biases in dysphoric individuals it is useful to examine how the network performs on computational analogs of information processing tasks on which dysphoric humans are known to be biased. Specifically, it is useful to examine whether the network exhibits similar biases to humans on tasks involving the recognition of the semantic and affective content of stimuli.

314

Chapter 7

Simulating the recognition of semantic information Assessment of the recognition of the semantic content of stimuli is traditionally done by presenting an individual with a stimulus, and measuring the time it takes the individual to recognize it. To measure semantic recognition independent of the time it takes an individual to recognize other aspects of the stimulus (e.g., its size), pronounce the stimulus, move mouth parts, etc., semantic recognition tasks are often conducted in the form of a two alternative forc~ choice task in which an individual is shown a stimulus, and is asked to push a button corresponding to whether the stimulus does, or does not, spell a word. A task in which an individual determines the lexicality of a stimulus is thus termed a "lexical decision task". To gauge an individual's recognition of affective information of one affective valence or another, stimuli of different affective valences may be included in a lexical decision task. Bower's (1981) network theory suggests that individuals who have strong connections between their internal representations of sadness and some personally meaningful negative event should be facilitated at identifying the semantic content relevant to that event. This is because information relevant to the event is expected to activate the mental representation of sadness, which would, in turn, more strongly activate the representation of the event. Some authors use this argument to predict that depressed individuals should therefore be facilitated in responding to all negative stimuli on an affective lexieal decision task (e.g., Challis & Krane, 1988; Mar Mathews, & Tata, 1986; Matthews & Southall, 1991; Ruiz Caballero & Bermudez Moreno, 1992) with little confirmation. In contrast, Siegle, Ingram, and Matt (1996a) suggest that depressed individuals will only be facilitated in identifying the semantic content of stimuli which represent personally meaningful negative events, on the affective lexical decision task. They suggest that negative stimuli which are not personally relevant will activate the mental representation of negativity which will, in turn, activate the mental representation of personally relevant negative events to which it is strongly connectext. They propose that the concurrent activation of personally relevant negative events may interfere with the identification of negative stimuli which are not personally relevant. They therefore predict that depressed individuals will, in general, be delayed in identifying the semantic content of most negative stimuli when the stimuli are presented for a brief period. Siegle et al. (1996a) demonstrate this type of interference meta-analytieally and empirically for results using an affective

G.J. Siegle and R.E. Ingram

315

lexical decision task. Specifically, Siegle et al. (1996a) show that a sample of 30 dysphoric research participants were, in general, delayed in responding to negative words with respect to positive or neutral words presented for a duration of 150ms on an affective lexical decision task. The same delays were not present for a sample of 46 nondepressed individuals. Their results are illustrated in Figure 2. To test whether Bower's theory predicts that being overexposed to negativity leads to a general facilitation towards negative information or facilitation only for personally relevant negative information and delays for

9

NonDepressed-Lexieal Decision

:

Depressed-Lexieal Decision

- -~- - NonDepressed-Valenee Idenfifr.ation - Depressed-Valence ldentilieation

0.9

0.85 S"

9

r

9

0.8 S , . . J 'r

%%%~

r162 r "J'

"~ 9 % l ~

0.75 S ~ S

0.7

-%

SS

"N,

0.65

0.6

0.55

0.5

0.451 Positive

.....

I

',

Neutral

Negative

Word Valence

Figure 2. Reaction times for lexical decision (LD) and valence identification (VID) tasks, 15 0ms condition.

316

Chapter 7

other negative information, Sicglc ct al. (1996b) created a simulation of the affcctivc lcxical decision task. They represent lcxical decisions in a manner analogous to that used by Cohen, Dunbar, and McClclland (1990) to represent word and color naming in a r model of the Stroop task. Following Ratr (1978) notion that semantic identification is a diffusion process, they suggest that a semantic identification occurs when the activation of the mental representation of a stimulus reaches a threshold. They therefore define counters representing the accumulated evidence for each possible item their network might identify. The counters add evidence for a given item proportional to the difference between the activation of the semantic representation of that item and the maximum activation of any other semantic representation, subject to gaussian noise. When any counter cxcce,ds a threshold (arbitrarily sot to 1), the network is said to have made a semantic identification. The presentation of a stimulus for a limited duration is simulated by turning off input from nodes representing the orthographic content of the stimulus after a brief period (90 epochs). In this way, the network is forced to identify the stimulus using only the residual activation within the nodes representing the semantic and affcctivc content of the stimulus. Sicglc ct al. (1996b) show that when the network is trained for 70 epochs on a negative stimulus after having received its original training, the network assumes relatively similar information processing biases to those observed by Sicglc ctal. (1996a) for dysphoric individuals on the affcctivc lcxical decision task. The results for this case arc dcpicte~ in Figure 3. As may bc seen from the figure, the network which has bccn ovcrtraincd on a negative stimulus is delayed in responding to other negative words on the lcxical decision task, but is facilitated in responding to the particular negative word on which it was ovcrtraincd. As negative stimuli on most lcxical decision tasks arc probably not especially personally relevant for the individuals who arc lacing tested, most depressed individuals would thus bc expected to bc delayed in responding to negative words on the task, though facilitation is predicted for any words which arc particularly meaningful for a given depressed person. While Sicglc ct al. (1996b) report results for groups of simulations using the same parameter configurations, Figure 3, and the rest of the figures depicting simulations in this paper show the simulated performance of a single run of the network. A single simulation is a valid indicator of the nctwork's performance because the parameters governing noise in the network arc not high enough to cause more than 1-2 epochs of variability in overall simulated reaction times, and thus, there is no nccd to average over

G.J.. Siegle and R.E. Ingram

317

NonDepressed-Lexical Decision Depmssed-Lexical Decisbn - - m - - NonDepressed-Valeme Idemifr~tion 136

- ' ~ - Depressed-Valence Identification

,A,

131 S

126 #

epochs

J

S

J d#

J

N\%\ \

J N

\\\

\

J

121

\ \

116

111

106 "'1 Positive

! Neutral

I Negative-Other

NegativePersonally Relevant

Stimulus Valenc.

Figure 3. Simulated lexical decision and valence identification tasks.

multiple simulated subjects. As the primary objective of this paper is to show the effects of parameters governing individual differences, effectively simulating one subject at a time sccmeA appropriate. Additionally, each of the rcportexi simulations was run multiple times to insure that results were not the product of a single anomalous trial.

Simulating the recognition of affective information Assessment of an individual's recognition of the affective valence of a stimulus may be accomplished in much the same way as recognition of semantic information was assessed. Specifically, the individual may be presented with an affective stimulus and asked to respond by pushing a button representing the affeetive valence of the stimulus. The time it takes the individual to respond is assumed to be analogous to the time it takes the

318

Chapter 7

individual to recognize the affective valence of the stimulus. This task will be termed an affective valence-identification task. As with the lexical decision task, the valence identification task appears to be a relatively pure measure of an individual's ability to associate a stimulus with a mental representation of affective valence. Despite predicting interference on the identification of the semantic content of most negative stimuli, Bower's (1981) network theory predicts facilitation for the identification of the affective content of negative stimuli on a valence identification task. That is, when a depressed person is exposed to a negative stimulus, their representation of sadness is expected to become activated. The mental representation of events associated with sadness, in turn, become activated, and via the feedback process described by Ingrain (1984), reactivate the mental representation of sadness. In contrast, when a non-negative stimulus is presented, any incidental activation of the nodes representing a negative affective valence begins the spiraling negative loop which may increase the activation the node representing negativity and cause interference with the identification of the stimulus as positive. Thus, Bower's (1981) theory predicts that depressed individuals may be delayed in identifying the affective valence of positive words. By contrasting results of an affective lexieal decision task with that for a valence identification task, it may be possible to determine the relative magnitude to which the affective valence of stimuli interferes with an individual's ability to identify their semantic content. The idea that the affective content of a stimulus can interfere with its semantic identification can be referred to as affective interference. The possibility that such affective interference is responsible for the attention biases that may occur in people with features of depression can be referred to as the affective interference hypothesis. The affective interference hypothesis, and its implications for understanding depressive information processing are described in detail by Siegle et al. (1996a). Siegle et al. (1996a) demonstrate affective interference empirically using an affective valence identification task, in which participants are asked to identify the affective valence of briefly presented stimulus (150ms) which may be positive negative, or neutral. They show that the same sample of 30 dysphoric research participants as were used for their affective lexical decision task were, in general, delayed in responding to positive words with respect to negative words on the affective valence identification task. The same delays were not present for the 46 nondepressed individuals. Their results are illustrated in Figure 2.

G.J. Siegle and R.E. Ingram

319

Sicglc et al. (1996b) simulate these results similarly to the way they simulated the lcxical decision task. Counters arc kept for the activation of the positive and negative valences. When any counter exceeds a threshold (arbitrarily set to 1), the network is said to have made a valence identification. The network judged a stimulus to be neutral when little evidence was accumulated for either valence (both accumulators less than 0.8) after a temporal threshold of 132 epochs plus gaussian noise. Sicgle ct al. (1996b) show that when the network is trained for 70 epochs on a negative stimulus after having received its original training, the network assumes relatively similar information processing biases to those observed for dysphoric individuals on the affectivc valence identification task. Results for this ease are depicted in Figure 3. Sicgle ct al (1996b) show that simulated reaction times to negative words in the network which received ovcrtraining more closely resembles those for the network which did not receive ovcrtraining, when a distributed representation of stimuli is used 2. A distributed representation was not used for the current simulations so as to preserve as interpretable a representation as possible. The authors also show that the magnitude of information processing biases increases in direct proportion to the ovcrtraining on negativity received by the network. The network parameters used in the simulations of the affcctivc lcxical decision task and affectivc valence identification task are shown in Table 1. Sicglc et al. (1996b) show that the nctwork's performance is qualitatively similar when a number of these parameters arc manipulated.

A weakness in the model: The need for investigation ofsubpopulations The simulations described above provide some validation for the proposed connectionist framework. All variation around mean reaction times in the model is induced by allowing random noise to be present in various aspects of the network. This feature might, at first, suggest that all variation about mean reaction times in the human data is effectively due to entirely random variation. Yet, the human data on which the model is based does not reflect entirely random variation about mean reaction times. Instead, observed reaction times for a given condition are often multimodal and the variance about means does not reflect homoscedasticity (Siegle et al., 1996a). These observations suggest that multiple defined subpopulations contributed to the - -

,

,

,

2 While each node in the network represents a single concept, in the current simulations, each stimulus is represented by a pattern of activations throughout a set of nodes in a distributed representation.

320

Chapter 7

observed data. By modeling subpopulations explicitly, the observed characteristic variation about mean reaction times may be explained.

Table 1. Parameters used in the neural network simulations. Parameter Network construction Number of input nodes Number of semantic nodes Number of valence nodes Activation parameters . (input diffusion rate) 13(affective-semanticloop diffusion rate) maximum network activation minimum network activation network noise Task.parameters accumulation noise temporal threshold for "Nonword" decisions temporal threshold noise positive determination accumulationthreshold nesative determination accumulationthreshold Learning Parameters 11(learning rate) ~x(leaming momentum) error threshold for initial learning additional epochs of training on negative stimuli Activations in one trainin$ elx~h ...... Training set Number of stimuli Number of nesative stimuli representing depresso~enic loss

Value 10 10 2 '0.5 0.2 1.0 0 0.05

1.0 200 elx~hs 10 1.0 1.0 0.2 0.4 0.004 70 10 9

1

Simulating Personality Factors There are a number of factors to consider in choosing what personality variables to simulate to best understand how personality affects information processing in depressed people. The goals of this chapter are to use variables which are central to theories of personality, for which some theoretical role in depression exists, and which have distinct correlates in the proposed model.

G.J Siegle and R.E. Ingrain

321

As discussed in the beginning of this paper, three personality variables satisfy these goals including aspects of coping, cognitive structure, and intellect. Each will be addressed in succession. Factors associated with coping - the case o f rumination and distraction

We have suggested that trait-like coping styles could affect the ways in which information is processed by a person coping with adversity. Potentially, then, coping style mediates the results of affective information processing tasks such as the valence identification and affective lexical decision tasks conducted by Siegle et al. (1996a). The following section describes attempts to simulate aspects of two methods by which theorists suggest that individuals cope with adversity. One method, rumination, in which individuals focus on salient negative events, may be conceived of in a number of ways, each of which has distract cognitive correlates. Another method, distraction, in which individuals focus on things other than salient negative events, is often considered to be the opposite of rumination. Each of these methods of coping can be operationalized in the network model. Ruminative coping. Nolen-Hoeksema (1987) has suggested that rumination is one way people deal with adversity. People who ruminate, she proposes, think extensively about particular salient negative events. NolenHoeksema suggests that rumination can increase vulnerability to depression, though, she shows, it is a common coping style. While empirical research has demonstrated a number of negative aspects of rumination, little consideration has been given to why people do it, i.e., the benefits of rumination. We suggest that rumination can act to either help or hurt an individual. We further suggest that the time at which an individual begins to ruminate governs how beneficial or hurtful rumination will be to the individual. To examine how rumination might mediate information processing biases in people who experience a loss, we can observe how a computational analog of rumination affects the computational neural network's analogs of performance on the simulated affective lexical decision and valence identification tasks. To represent a computational analog of rumination in the neural network platform, we must to specify what is meant by rumination more explicitly than has been done in the past. As a start, recall that negative information is represented in the model in terms of a semantic and affective component. Feedback between the representations of the affective and semantic aspects of a stimulus can be thought of as a cyclic concentration of the network's resources upon a particular stimulus. Thus, allowing a great

322

Chapter 7

deal of feexiback between the affcctivc and semantic layers in the computational network, in response to a stimulus might be thought of as corresponding loosely to an analog of "ruminating" on the stimulus. The question now arises as to when this analog of rumination should occur during the nctwork's simulated life. To simulate a "ruminative" personality (possibly akin to having an "obsessive-c, ompulsivc" personality style in which the affcctive consequences and implications of minute details are meticulously considered) we would allow the increased feedback to occur throughout the nctwork's training. This modification would simulate an individual who ruminated from their early childhood. In contrast, we might simulate rumination as a way of coping with a negative event. To do this, we could allow the feedback to occur when, or slightly after we begin to simulate a dcprcssogcnir loss. Finally, we might assume that when an individuals experience a loss events, they arc so overcome by grief that they do not intellectually process (or ruminate upon) the loss event immediately. Rather, the rumination might occur sometime alter the loss event has occurred, and thus, after cognitive biases associated with the loss have been adopted. To simulate this condition we could allow the computational analog of rumination to occur only after the network is ovcrtraincd on a particular loss event. Each of these possibilities lead to different predictions for the nctwork's behavior, and are investigated separately in the following sections. The ruminative personality - a vulnerability factor. The effects of "ruminative personality" on information processing were simulated by allowing the network to engage in 10 cycles of feedback between the affective and semantic loop alter the presentation of stimuli throughout its training. This behavior is meant to represent the contemplation of a stimulus after the stimulus has been presented. The effects of this modification on the network's information processing were assessed by measuring its simulated reaction times on the computational analogs of the affectivr lexical decision and valence identification tasks. The network's performance after training which included the extra feedback cycles, is displayed in Figure 4. As may be seen from the figure, the computational analog of a ruminative personality experience~ a great deal more interference on the affective lexical decision task than the nonruminating network (Figure 3). Because this characteristic of personality appears to exaggerate information processing biases in the context of a loss experience, it might be considered a vulnerability factor for the cognitive aspects of depression. The network's performance in the condition in which it received no overtraining on a negative stimulus is also somewhat different from the

G.J. Siegle and R.E. Ingrain

323

NonDcpressed-Lexical Decision :

Dcprcssed-LexicalDecision

- -dr- - NonDcprcssed-Valence

Idenfifr.ation

141 - Dcprcssed-Vabrr.c Identification

136 .

.

.

.

- -

- A

131 epochs 126

121

116

111

106 I Positive

v

! Neutral

"

I

Negati~-O~r

'

I

NegativePersonally Relevant

Stimulus Valence

Figure 4. Simulated lexical decision and valence identification tasks" always ruminating.

network which did not engage in the excessive affective-semantic feedback loop. Specifically, the network which engaged in fccdback was delayed at recognizing both positive and negative words on the valence identification task. This delay occurred because the network which engaged in an analog of rumination throughout its training effectively used its ruminative periods to modify its representation of incoming stimuli to be close to representations it had already learned. Thus, it did not learn the valence of new incoming stimuli as strongly as the original nctwork. Inspection of the network's weights revealed that nearly all weights throughout the semantic and affcctive loop were lower in the network which did cngage in feedback than in the nctwork which did not. Potentially, this result could be used to suggest that individuals who are particularly slow at recognizing the valence of stimuli might be ruminating on their perceptions. Hence, they may bc vulnerable to

324

Chapter 7

depression. Were future research to validate this speculation, simple information processing tests such as a valence identification task might be used to assess for vulnerability to depression. Coping by ruminating - a protective factor. To simulate the effects of ruminative coping, brought on by a loss experience, the network was made to engage in 10 cycles of feedback between the affective and semantic representations only during the ovortraining on a negative stimulus. This network's performance on the computational analogs of the lexical decision and valence identification tasks is illustrated in Figure 5. When the network was overtrained on a negative stimulus, with 10 cycles of feexlbaek between the affective and semantic layers, its performance on the valence identification task was less biased than the network which had been overtrained on a negative stimulus but incurred no extra feedback between the affective and semantic representations (Figure 3). The reason that the overtraining did not affect the network's valence identification a great deal is that feeAbaek between the network's affeetive and semantic components effectively minimizeA the effects of noise, and thus, the network did not need to adjust its weights a great deal upon overtraining. In this way, the computational analog of rumination prevented the network from learning the negative stimulus in a way which would distort its information processing a great deal. Thus, feedback analogous to rumination protected the network from information processing biases characteristic of depressed individuals. Practically, this result suggests that rumination might be an effective way of coping with the possibly distorted affect common after a loss experience! Simply put, if you think about how to reasonably interpret your life experiences immediately after a loss with respect to what you have learned in the past, rather than with respect to just the current loss, you may be prevented from acquiring the negative information processing biases characteristic of depression. While there is little empirical support for rumination ever being a helpful coping style, potentially its benefits have not been investigated for individuals who restrict rumination to the moments directly after a traumatic event. Anecdotally, many individuals speak of the benefits of "seeing the larger picture" or "not letting an event bother me" as a way of dealing with negative incidents. Potentially, these coping mechanisms, which appear to involve allowing associations with the affective content to be subsumed by less negative associations from an individual's knowledge base, might correspond to this type of rumination.

G.J. Siegle and R.E. Ingrain :

NonDcprcsscd-Lexical

325

Decision

Deprcsscd-Lcxical Decision - --~ - NonDepressed-Valcnce

136

-'~

- Depressed-Valence

Idenfifr.ation

Identification

131

/.-" 126

-,..X

/-"

epochs

/

121

"

----- - ------ ---I

116

111

106

I

Positive

"

I'Neutral

I' Negative-Other

NegativePersonaly Relevant

Stimulus Valence

Figure 5. Simulated lexical decision and valence identification ruminating during overtraimng.

tasks:

Ruminating during depression - the prevention of healing. The person who has often coped with negative events by ruminating, and.thus warded off cognitive correlates of a possible depression, might believe that rumination is always a beneficial way of dealing with loss. Suppose, that this same person experiences a loss, is so overcome by the loss that he or she does not processes it intellectually immediately, and only begins to ruminate later. Now, because the rumination tends to preserve learned knowledge (including learned biased information processing), the depressed person is expected to have exceptional difficulty learning new positive information! As a consequence, the depressed person may have difficulty recovering from depression. This phenomenon can be shown in the network by increasing feedback between the affective and semantic layers after overtraining the network on a negative stimulus. In this case, the network performs no

Chapter 7

326

differently on the simulated affective lexical decision and valence identification tasks from the original network which was overtrained on a negative stimulus, but in which excessive feedback did not occur. The lack of information processing biases does affect the network though, in a more insidious way. The network in which an analog of rumination occurs displays more difficulty learning a new positive stimulus than the network which has not overlearned the negative stimulus, as shown in Figure 6, and in fact, never learns the new positive information quite as well as the nonruminating network 3. This phenomenon may be likened to the clinically observed difficulty which many ruminating depressed individuals experience in learning positive information. Potentially, this result suggests that while rumination may not directly affect how a person processes stimuli, it may affect his or her ability to recover from depression. In this way, rumination may help to maintain depression. Evidence from clinical research exists for the ideas that rumination may help to prevent individuals from recovering from depression. While there are few reliable empirical data on rumination-indueeA maintenance of depression, intervention efforts designed partially to reduce rumination attest to the clinical relevance of the rumination process. Beck's (1976) cognitive therapy of depression, for instance, has among one of its most important goals the interruption of automatic cognitive processes; processes that are quite conceptually similar to the concept of rumination. As Ingram and Hollon (1986) have noted, the meta-cognitive strategy advocated by Beck seeks to substitute a more controlled information processing for the automatic/ruminative information processing that contributes to the maintenance of the depressive state. By "distancing" oneself from one's thoughts, the ruminative process is thought to be reduced. While Beck suggests that the reduction of rumination in this manner is but one component of cognitive theory, it is indeexl an important one. Likewise, other noted cognitive therapy theorists have suggested the importance of reducing ,

,,

,,

3 For this simulation, the network's momentum parameter was set at 0.9 and learning parameter was decreased proportionally to 0.1 to smooth the learning curve. Initial learning was done to a threshold of 0.01 to exaggerate difficulty in learning a new stimulus. For cautions regarding the modification of these parameters see the section on Openness to Experience. Additionally, Siegle et al. (1995) show even more dramatic increases in error rates when a distributed representation is used. In this case, the network which received overtraining on negativitybut did not ruminate experienced was successfully able to relearn previously learned positive information, but the ruminating network took more than 5 times as long, and had much higher error rates.

G.J Siegle and R.E. Ingram

0,]

---r

Normal

---n--

Negative Overtraining N e gative

327

Ove rtraining+R umination

0.25 -'

0.2 J

elTOr

0.15

0.1

,, 0.05

4

,...,,,.. i

0

epochs

Figure 6. Error rate for learning positive information over time.

depressive rumination. Indeed modifying "what individuals say to themselves" is a common theme running through a number of time-honored cognitive intervention procedures (e.g., cognitive relabeling: Goldfried & Davison, 1976; stress inoculation training: Meichenbaum, 1985). Coping by ruminating - a vulnerability factor. Individuals who cope by ruminating were suggested above to be protected from some of the most detrimental cognitive correlates of depression. The argument for this conclusion assumes that individuals who ruminate change their cognitive structure (i.e., learn) only aider they ruminate on a stimulus. Another possibility is that individuals who ruminate learn during the rumination process. That is, while an individual engages in rumination he or she actually reinforces the ideas which are being ruminated upon. This phenomenon is simulated in the network model, by allowing the network to learn during the feedback between the network's affective and semantic nodes. When this

Chapter 7

328

technique is used, the computational analog of rumination no longer acts as a protective factor, and in fact, increases the network's negative information processing biases a great deal. Figure 7 shows the relative increases in the network's reaction time on the lexical decision task to a negative stimulus on which the network is not overtrained, as it begins to overleam a negative stimulus. The nonruminating network showed no appreciable change in reaction times over the first 8 epochs (delays begin to appear around 10 epochs). The network in which learning occurs during a cycle of feeAback between the affective and semantic layers 10 times for each stimulus presentation is delayed after even one epoch of overtraining. By 8 epochs of overtraining, the ruminating network is virtually unable to distinguish the nondepressotypie negative word from the negative word on which it is being trained. This happens because the ruminating network has effectively received 10 epochs of overtraining for 206 - --*-- Rumimtiag 196 -

[

NonRmnkating !

186 - 176 - 166 - epochs 156 146 136 - 126 116 106

,b

1

2

3

4

5

6

7

8

9

epochs o f overtraining on nondepressotypic negative stimulus

Figure 7. Simulated lexical decision task performance on a nonovertrained negative stimulus, for a network which ruminates while learning.

G.J. Siegle and R.E. Ingrain

329

each single epoch at which the negative stimulus was presented. Clinically, this phenomenon might reflect the idea that individuals who use rumination to teach themselves negative information are extremely vulnerable to the cognitive correlates of depression. After even a very brief exposure to some loss event, learning through rumination could engender large information processing biases. Distractive coping. Nolen-Hoeksema (Nolen-Hoeksema & Morrow, 1993) also discusses distractive coping. She demonstrates that people who cope by distracting themselves from the issues which are troubling them seem to be less vulnerable to features of depression than individuals who ruminate. Nolen-Hoeksema does not provide a theoretical model for why this phenomenon should occur. Simulating aspects of distraction in the current neural network model will tie rumination and distraction into the same theoretical model, and thus the two types of coping can be compared in a theoretically meaningful manner. Nolen-Hoeksema and Morrow (1993) conceive of distraction as a process by which individuals think of potentially random information other than their symptoms of depression. This phenomenon was modeled in the network by varying the amount of random noise which entered into the processing of stimuli during overtraining on negative information. More noise is assumed to correspond to more distraction when information is processed, because noise leads the network to randomly associate incoming information with information other than that which it has learned. Figure 8 shows the network's performance on the affective lexical decision task and valence identification task when noise is increased from 0.05 to 0.12. As shown in Figure 8, when the network was overtrained on one negative stimulus for 70 epochs, noise did not appear to greatly affect performance on the lexical decision task with respect to its original performance (Figure 3). In contrast, affective interference on the valence identification task was substantially decreased in the network in which noise was increased. Potentially, this result suggests that individuals who engage in distractive coping may not be subject to the same sorts of affective interference which plague individuals who do not distract themselves from associations with memories which depress them. A similar question regards how people who distract themselves from thinking about affect throughout their lives, rather than just in response to a particularly negative event, will cope with a negative event. These people might correspond to the classic idea of "repressors" who avoid thinking about his or her emotions, and about the emotional significance of events, via

Chapter 7

330 :

NonDepressed-Lexical Decision Depressed-Lexieal

Decisbn

- -~- - N o n D e p r e s s e d - V a l e n c e ---K - D e p r e s s e d - V a l e n c e

Identification

lderlifi~tion

A s S

131

%x

4./0

126 '

epochs

S

SS

121 ~x

116

111

106

~

Positive

....

I Neutral

I

,,.

Negative-Other Stimulus

I NegativePersonal~ Relevant

Valence

Figure 8. Simulated lcxical decision and valence identification tasks, noise during ovcrtraining.

distraction. Such a trait could be modeled by increasing noise in the network throughout its training. With respect to the network with low noise, the network with increased noise throughout its training performed more slowly on the computational analog of the affcctivc valence identification task as shown by comparing its performance (Figure 9) to that of the original network (Figure 3). This finding reflects the clinical idea that a person who distracts themselves from thinking about emotional information throughout their life may process this information less effectively. Moreover, information processing biases on the valence identification task are greatly exaggerated for the network in which distraction is used throughout its over-training. Biases on the lcxical decision task are decreased for this network. Clinically, this result suggests that a person who generally uses distraction to avoid avoid emotional information may perform perfectly normally after a loss on

331

G.J. Siegle and R.E. Ingram 9

NonDepressed-LexicalDecision

-

Depressed-LexicaiDecision

- - a - - NonDepressed-Valence ldenfifr.ation - Depressed-Valence Identification

131 s

\

126 - [ - . . " k\\N

epochs

\ k \\\

121

k 116

111

106 I Positive

\\\ \

\

\

\ x~.

.A

v

I

I

I

Neutral

Negative-Other

NegativeP~sonally Relevant

Stimulus Valence

Figure 9. Simulated lexical decision and valence identification tasks, high noise always.

tasks which require semantic processing (e.g., traditional tasks which might be found in the work place) but might have exaggerated difficulty assimilating emotional information. This description reflects many theorists' intuitions regarding the nature of repression. Conclusions about ruminative and distractive coping. The preceding simulations have suggested a number of possible roles for trait variables representing aspects of rumination and distraction. The simulations suggest that based on individuals' life experiences, when they begins to use a coping strategy, and whether or not they learn from that coping strategy while they are using it, the same ways of coping may either increase or decrease the chances that they experience pervasive information processing biases characteristic of depression. Similarly these same factors may govern how easily an individual who becomes dysphoric or depressed can recover from

332

Chapter 7

this state. By carefully attending to the interaction of trait variables such as coping style with life experiences, researchers may be able to explain a great deal of variation in vulnerability to, and recovery from depression.

Factors associated with cognitive structure The vast majority of the literature regarding vulnerability to depression concerns how an individual's previous life experiences impact his or her current life experiences. This literature is based on theories of cognitive psychology which suggest that an individual's "cognitive structure" or cognitive schema which the individual has derived from previous experiences, helps to govern how he or she interprets and assimilates new information (e.g., Winfrey & Goldfried, 1986). For example, literature in cognitive psychology shows that individuals who have learned a great deal about music tend to perceive songs very differently from individuals who have not learned a great deal about music (Newell & Simon, 1972). Presumably this difference occurs because people who have learned a great deal about music have a much different internal representation of stimuli such as notes than those who have not. Cognitive structure has frequently been suggested to govern an individual's personality (e.g., Cantor, 1990) and interpretation of everyday life experiences (Sehank & Abelson, 1977). Depression researchers such as Beck (1967; 1976) have extended the idea of cognitive structure to suggest that individuals who have experienced loss events create detailed representations of loss experiences for themselves, which become the foundation of their depression. Their perception of negative events may therefore be very different from people who have not experienced a loss. As for cognitive psychologists who study normal functioning, Beck's notion of distorted cognitive schemas in depression is based on the learning of negative information from one's environment. Potentially, the cognitive .structures that are derived from different types of previous experiences contribute to different reactions to qualitatively similar negative events. Yet, it is not clear exactly how previous experiences will affect an individual's reaction to an event. For example, an individual who has experienced death in the past may be more disturbed by the death of a relative than the individual who has not, precisely because he or she has a more well developed notion death. That is, his or her conception of death might be strongly associated with to his or her mental representation of sadness and other aspects of the cognitive network. Alternately, an individual who has experienced the death of many individuals may be able to assimilate

G.J. Siegle and R.E. Ingram

333

information from the recent death of a relative because he or she has a well developed sense of loss. An individual who has never experienced a loss may have more difficulty dealing with the same event. To better understand the role which cognitive structure, as governed by previous life experiences, plays in the onset and maintenance of depression, we have chosen to simulate one factor associated with previous life experience; the role of previous negative experiences. New negative experiences are more damaging. The first question relevant to cognitive structure which we addressed through simulation was whether an individual's previous experience with a negative stimulus would be expected to affect his or her reaction to a loss involving that experience. The question was operationalized by overtraining the network on a novel negative stimulus (one to which it had not been exposed in the past) rather than a stimulus it had previously learned. This procedure was meant to simulate a loss involving something new to a person, (e.g., the first time a person experiences the death of a loved-one). Figure 10 shows the network's performance on the simulated lexical decision and valence identification tasks after being overtrained on a novel negative stimulus for just 10 epochs. As may be seen from the figure, information processing biases due to the new stimulus are far more apparent than they were for the network which had received 10 epochs of overtraining on a previously learned negative stimulus (Figure 3). Similarly, when the network is trained for 59 epochs on the novel negative stimulus, its information processing biases are greatly exaggerated with respect to the network which was trained for 70 epochs on a previously learned negative stimulus. When the network is trained for any more than 59 epochs on the novel negative stimulus, it attempts to match the valence of any incoming stimulus to that of the novel stimulus, and thus displays extremely exaggerated simulated reaction times. It labels many positive stimuli as negative on the valence identification task. Effectively, the network "forgets" how to identify anything but the novel negative stimulus. Inspection revealed that during overtraining, all of the network's weights except those relevant to the new stimulus decayed, corresponding to an actual loss of previously learned knowledge not relevant to the stimulus representing the loss event. The idea that old information is lost when a neural network is overtrained on new information has been explored extensively by Ratcliff (1990) in a discussion of phenomena related to "forgetting" in neural networks. This sort of forgetting takes on new meaning in relation to the current model. In effect the network forgets the positive information it has learned when it experiences a

Chapter 7

334 .~

NonDepfessed-Lexi~al Derision

-

I ) e p r e s ~ - L e x i ~ Decisbn, 10 epochs overtraking

- - * - - NonDepressed-Valence l d ~ n --*( - Depressed-Valence ldenfilkation, 10 epochs o v e m a H ~ - - x - - Depressed-Lexical Decisbn-59 epochs overuaini~ = 9 =~

-

V

~

ld~n-59

epodm o~'a"akJmg

135 m

n

an

m

!

m

~

130 epochs 125 120 ~ 115

-

110 105 100

'

Positive

I

.

Neutral

.

.

.

.

.

I

Negative-Other

Stimulus Valence

Figure 10. Simulated lexical decision and valence identification tasks, novel negative stimulus.

computational analog of depression. The nctwork's behavior suggests that individuals who experience a loss which is very new to them may be more profoundly affected than individuals who experience a more easily interpretable loss. This finding might suggest that sometimes individuals who arc dealing for the first time with concepts such as death, divorce, or losing a job are more strongly affected by these events than individuals who have lost multiple jobs, been previously divorced, etc. Empirical studies could address this hypothesis in the future. The overtrained nctwork's inability to identify the affcctive valence of positive information resembles the finding that people who are depressed have difficulty recalling positive information (Blancy, 1986). The current model suggests that only individuals whose depression is due to a loss experience with which they have not had previous experience would have such a

G.J. Siegle and R.E. Ingram

335

difficulty. Individuals who have learned about loss similar to the one which has sparked their depression in the past might be able to recall more positive information in the context of their depression than those who have not, an idea that merits further research. The role of previous negative experiences. From the idea that previous experiences with loss might help an individual to remember with positive information when they are depressed, the creative researcher might be tempted to speculate that previous depressions help to prevent future depression. Empirical evidence does not support this idea. Instead, previous depression appears to be associated with vulnerability to future depression (Weissman ~ al., 1991), especially if the depressogemc loss occurs at a developmentally critical time. It is therefore interesting to more closely examine the idea that some familiarity with a negative stimulus prevents the network from incumng extreme information processing biases associated with depression, but a great deal of previous overtraining on a negative stimulus might leave the network more vulnerable to information processing biases than a non-overtrained network. Empirically, this situation might be represented by an "optimal" range of familiarity with loss. Individuals who have experienced some minimal losses and developed adaptive reactions to these losses might be better prepared for new losses than individuals who have never experienced loss, or who have experienced a large number of losses. To explore this situation in the computational model, the model was first made to learn the full set of 9 positive, negative, and neutral stimuli to an error threshold of 0.01, representing a time approximately half way through the network's original training. Then, the model was overtrained for 100 epochs on one negative stimulus. Finally, the network was retrained on all 9 stimuli to the error threshold of 0.004, the value which was used for all other simulations. This retraining might be thought of as having many new experiences after an initial episode of depression, or possibly as having "gotten past" the depressive incident. The retrained network's performance on the tasks is shown in Figure 11. Many researchers find that it is difficult to detect depressive attentional biases in individuals who have recovered from depression, without manipulations such as negative mood inductions (Segal & Ingram, 1995). Some researchers have thus concluded that many cognitive correlates of depression do not persist after the depressive episode (Persons & Miranda, 1992). Consistent with this idea, the retrained network shows almost no biases on the valence identification task or lexical decision task for negative

336

Chapter 7

stimuli on which it was not overtrained. Yet, the retrained network's performance is still facilitated on the particular negative stimulus on which it was originally overtrainedf This finding might suggest that individuals who have recovered from depression may not be biased in the way they process most negative information but might still be biased in the way they process negative stimuli which were related to the onset of their depression. This finding can be explained by observing that the weights within the network leading from the semantic content of the node on which the network was overtrained to negativity are stronger than from the other nodes representing negative semantic content (3.06 as opposed to 2.76 and 2.74) and are stronger than the connections from the semantic nodes representing positive stimuli to positivity (2.83, 2.87, 2.88). Similarly, connections from the negativity node to the node representing the semantic content on which it was overtrained are more positive (-.063) than for all other negative stimuli (-.24 and -.19) and for the positivity node to positive semantic stimuli (-. 17, -. 16, -. 17). The network's biases also suggest that the network is less biased away from this information (-1.952) than from other information (biases ranging from -1.987 to -1.989). Were the network to change the bias term representing its propensity to activate negativity, as happened during its initial overtraining, the weights suggesting facilitation on the stimulus on which it was overtrained would thus still be in place. Potentially, these results suggest that structural correlates of depression may be observable even when information processing is not biased in formerly depressed individuals. Moreover, when the network is retrained on the same negative stimulus on which it was initially overtrained for 70 epochs, its performance, shown in Figure 11, was much more biased on the lexieal decision task than the network which had not received previous overtraining, shown in Figure 3. Moreover, on some confirmatory runs of the same simulation, this network was unable to identify a negative stimulus on which it had not been overtramed. The clinical analog of these simulations would be that individuals who experience a prolonged dysphoria resulting from overleaming some negative experience once may be become even more depressed at a similar stimulus later. Such a behavior could correspond to previously depressed individuals being more vulnerable to future depressions than other individuals, as discussed by researchers such as Weissman et al. (1991).

Having a good understanding of positive information helps. The preceding simulations suggest that previous experience with negative information may help to govern the magnitude of information processing

G.J. Siegle and R.E. Ingrain

NonDcprcssed-Lexieal

Dcpresscd-Lcxical Decision

Decision

- -~- - NonDcpresscd-Valence IderSifr~tion -=~ - Depressed-Valence ldenti~ation

136 T

/

_11

-:

337

J

~

%

epochs

[ SS

X

121 J

~L. ~

-"

116 -

%,~

111 -

I06 Poslive

I

t

Neutral

Negative-Other

NegativePersonally Relevant

Stimulus Valence

Figure 11. Simulated lexical decision and valence identification tasks after 100 epochs negative overtraining and retraining to an error threshold of 0.004 on all stimuli.

biases in a depressed individual. Other empirical research suggests that individuals' previous experiences with positive stimuli may affect how they deal with loss. For example, Lcwinsohn and Hobcrman's (1982) behavioral theory of depression suggests that depression is at least as much a function of a lack of positive stimuli as it is ovcrlcaming negative stimuli. Similarly, Schwartz and Garamoni (1989) show that clinically depressed individuals tend to have fewer positive cognitions as wcU as more negative cognitions than nondcprcsscd individuals. To simulate a cognitive structure less attuned to positive than negative information before a loss the network was initiallytrained one fourth as much on the three positive stimuli as it was on the negative and neutral stimuli. Its reaction times on the simulated Icxical decision and valence identification

Chapter 7

338 :

NonDepressed-Lexirad Decision

:

Depcessed-Lexical Decision

- -*- - NonDelm:ssed-Valew~ Idemifr~tion 145

- ' ~ - Depressed-Valence Identifr~tion

140 135 130

\

epochs 125

%.

120 115 110 105 100 Positive

:

I

Neulral

Negative-Other

. . . .

I

NegativePersomlly Relevant

Stimulus Valence

Figure 12. Simulated lcxical decision and valence identification tasks, few positive examples in training set.

tasks, before and after being ovcrtraincd for 70 epochs on a negative stimulus are shown in Figure 12. As may be seen from the figure, the network which received less initial training on positive stimuli was initially delayed in responding to a stimulus with a positive valence on the valence identification task. Information processing biases on the network performance on the simulated valence identification task after ovcrtraining on negative information arc greatly exaggerated with respect to the network which had previously experienced as much positive training as negative and neutral training. Yet, the network shows little interference on the simulated lcxical decision task. This behavior suggests that individuals who do not have a great deal of initial positive experiences may be especially vulnerable to some cognitive correlates of depression but not others. Potentially, this cognitive profile is indicative of a particular cognitive subtype of depression.

G.J. Siegle and R.E. lngram

339

A similar behavior can be found for a network which does not receive a great deal of initial training on any stimulus. For example, when the network is initially trained to an error threshold of 0.01 rather than 0.004 as in the original simulations (that is, it is given less initial training), its performance on the simulated affective lexieal decision and valence identification tasks after having been overtrained on negative information, shown in Figure 13, is much more biased than that of the original network shown in Figure 3. This behavior might reflect the idea that individuals without much experience to fall back on are greatly affected by loss events. The observation that less experience might be a vulnerability factor for negative information processing biases might suggest that children, who have fewer experiences to guide the structure of their cognitive network than adults, might be especially vulnerable to loss. This hypothesis might be tested by determining whether children are more vulnerable to depression than adults when confronted with similar losses. NonDepressed-Lexical Decision - - 4 - - Depressed- Lvxical Decision - -~- - NonDepressed-Valem~ ldenfifr~atbn

140

-'~

epochs

135

-

130

-

125

- Depressed-Valence Iderfi~ation

..........~ - "dg~i~

.~. N

120 115

\

_

~

~,~

110 105

100

Pos~ive

I

Neutral

"

'l

I

Negative-Other

NegativePersomlly Rekvant

Stimulus Valence

Figure 13. Simulated lexical decision and valence identification tasks, low initial error rate.

340

Chapter 7

Conclusions from simulations of cognitive structure. Our simulations of cognitive structure suggest that individuals who have either learned an extraordinary amount of negative information or learned little positive information in the past might be vulnerable to experiencing negative information processing biases when confronted with a loss. Less intuitively, the results also suggest that when a loss event is new for an individual, because the individual has little experience with loss or little experience with life in general, it may be particularly detrimental for the individual. Each of these factors could contribute to individual differences in expressed information processing biases on measures such as the lexical decision and valence identification tasks. To better account for variation in performance due to previous experience, it may thus be useful for researchers investigating negative information processing biases to account for an individual's previous life history. Life events schedules and previous depression inventories may be useful in this regard. Factors associated with intellect- the case of openness to experience

Historically, one of the best recognized sources of individual differences in information processing is mtelle~. The joint effects of intellect and mood, though, are neither well studied nor well understood. Potentially, computational models can help to integrate these areas by providing theoretically motivated predictions regarding the relationship between aspects of intelle~ and mood. Recent research suggests that people with affective disorders often experience information processing deficits traditionally associated with low intelligence. For example, a hallmark of depression is psychomotor retardation (American Psychiatric Association, 1995); processing speed is one of the primary variables thought to contribute to intelligence (Sattler, 1993, p. 77). Similarly, people who are depressed often display impaired problem solving ( M a c ~ & Mathews, 1991), difficulty in learning rules ( M a c ~ & Mathews, 1991), and diminished attentional capacity (Gotlib & McCann, 1984), all of which are often associated with low intelligence. Yet, many information processing biases present in depressed people appear to go away when they are not depressed (Persons & Miranda, 1992) suggesting the impairments are a function of the disordered mood state rather than the "trait" termed low intelligence. Potentially, observing how factors associated with intellect help to mediate the information processing biases expressed in

G.J. Siegle and R.E. Ingrain

341

the model which is overtrained on negativity can help to shed light on this relationship. Openness to experience as a risk and protective factor. Openness to new experiences is an aspect of personality which appears to be related to intellect (McCrae & Costa, 1985). One of the "Big Five" factors repeatedly identified on personality inventories, openness to experience is traditionally characterised by traits such as curiosity and creativity (Goldberg, 1993) or willingness to explore new ideas. Its origin stems from a series of items on personality inventories originally associated with mteUect, and in fact, many personality theorists equate these items directly with intellect (e.g., Digman & Takemoto-Chock, 1981). Individuals who possess more of this trait are said to learn faster and be more willing to change their beliefs in the face of new information than those who do not. The neural network framework proposed here suggests that openness to learning new experiences may also be especially important in determining depressive information processing biases. Because depression is operationalized as the overleaming of negative information, individuals who are more open to learning new information would be expected to incur the types of information processingbiases characteristic of depression more readily than those who do not. Similarly, these same individuals might be expected to be able to "unlearn" their information processing biases more easily than individuals who do not possess the same openness to new experiences. It is not clear whether being open to experience means that a person readily changes any and all of his or her beliefs based on new experience, or whether a person changes only certain, more mutable beliefs. The former definition may be operationalized in a neural network by modifying a parameter common to many neural network models called the "learning rate" (rl). The latter definition may be operationalized by modifying a related, but different common parameter termed the network's "momentum" (~t). To understand the function of these parameters, some knowledge of the backpropagation algorithm which is used to allow neural networks to learn information is important. During training, the network updates the weights of connections between nodes proportional to the amount of error incurred in the network's output, according to the formula: 6 i = r l , ( l a y e r l)T,errorlayer2 + t~,6 i

where/5 i is the change in the weight of a given connection which connects layerl to layer2. The formula states that the change in weights of layer 1 is a

342

Chapter 7

function of both the learning rate (11) times the difference between the desired and actual outputs of layer2, and the momentum ((z) times the previous change to the connection's weight aiter previous stimuli have been presented. Thus, the learning rate governs how much a connection will change in response to the presentation of a single stimulus. When TI is small, even if the network incurs a great deal of error in processing a stimulus, the network may not change its weights a great deal with respect to the stimulus. Alternately, if vI is large a single stimulus could greatly affect the n~work's responses. The momentum governs how much previous changes in the network affect its current responses. When tz is high, a connection is more likely to change if it has changed in the past. Thus, recently learned information has an aspect of mutability which is not attributed to previously learned information. The effects of changes in the learning rate and momentum in the network were illustrated by training the network on all stimuli, and then on a characteristically negative stimulus, in a set of 324 simulations in which 11 and (z varied continuously between 0.05 (very low) and 1 (very high) by steps of 0.05. For previous simulations, 11 was 0.2 and ct was 0.4. The network's simulated reaction times to the tasks for low values of ot (0.05) and r I (0.05) are shown in Figure 14. To depict the magnitude of the network's information processing biases as a continuous function of ot and 11, Figure 15 shows (z on the X axis and rl on the Y axis. The Z axis on the upper figure represents the difference in the overtrained and nonovertrained network's simulated reaction time to a negative stimulus, on which the network was not overtrained. The Z axis on the lower figure represents the difference in the overtrained and nonovertrained network's simulated reaction time to a positive stimulus. Reaction times are superimposed upon the best fit interaction surface, found through linear regression. On the lexieal decision task, the overtrained network was unable to correctly identify negative stimuli when cz or 11 were increased above about 0.5 and thus these data were not plotted. As in previous simulations, the overtrained network was, in general, delayed in responding to negative stimuli on the lexical decision task and positive stimuli on the valence identification task. As the learning parameters increased, the magnitude of the network's information processing biases increased for both the simulated lexical decision task (R2=.53, F(2,85)=49.3, p<0.001) and valence identification task (R2=.61,/7(2,85)=257.5, p<0.001). The interaction between ct and rl was statistically significant only for the

Low Alpha

lm

G.J. Siegle and R.E. lngram

epochs

Low Eta

'-

m

la, Poriivc

la,

N&

NqG&c-oms

N

I

e

p-

Stimulus Valence 343

Figure 14. Simulated lexical decision and valence identification tasks, low a, variable a and q.

Chapter 7

344

Difference in rt to a negative stimulus on the simulated lexical decision task

20 [7 ,4 1

epochs

1.0

0.0

cut

Difference in rt to a positive stimulus on the simulated valence identification task

epochs

--'----

1.0

0.0

eta

Figure 15. Reaction time surfaces for data shown in Figure 14.

lexical decision task (R2=.11, F~ (1,83)=28.8, p<0.001). As shown in the figure, the interaction was due to the greatest effect of 11 occurring for mextium values of cz. These findings may be understood in the following manner. Increasing either T1 or oc tended to allow the nc~work to learn more quickly. Thus, it became more likely to change its weights such that its information processing

G.J. Siegle and R.E. Ingrain

345

was biased in a manner characteristic of depressed people upon even minimal experience with essentially negative information. Clinically, this phenomenon might suggest that people who are more open to learning from their experiences are more likely to become depressed. Conversely, decreasing either ot or rl was associated with a decrease in the amount of information processing biases adopted by the network. The one exception was that when both cx and rl were decreased considerably, the network never learned its initial training very well, even aRer very large numbers of presentations. Thus, when the network was overtrained on negative information, the overtraining had an extreme impact on the network's largely unstructured weights, and the network adopted extreme biases characteristic of depressed people. Clinically this finding might suggest that individuals who are cautious about changing their cognitive structure based upon their life experiences are somewhat protected from the cognitive correlates of depression. Yet those people who do not have enough cognitive resources to learn well at all are most vulnerable to depression after a loss, since this experience can become central to their cognition very easily. That is, such individuals do not have a great deal of other information to fall back on to support them when they experience a loss. Clinical evidence also supports the idea that individuals with low overall cognitive resources may be vulnerable to cognitive correlates of depression. Specifically, a number of researchers have found that individuals categorized as "retarded" or having decreased cognitive functioning appear to be more vulnerable to depression than individuals of moderate intellectual functioning, though the relationship between cause and effect is often dubious in these studies (e.g., Harper & Wadsworth, 1990; Marx, Williams, & Claridge, 1994; Sackeim et al., 1992). It may seem counterintuitive to suggest that decreasing a network parameter analogous to intellectual functioning could help to protect the network against the effects of depression. A different way of interpreting the role of the momentum parameter may shed more light on the network's behavior. One possible interpretation is as a governing factor for emotional reactivity. If an individual's affective state varies a great deal with the stimuli which are presented to him or her, the individual may be said to be reactive. If the individual's affective state is based more on their state before the presentation of some negative stimulus than after it, the individual is not said to be reactive. Likewise, the momentum parameter governs how much change takes place in the network based on the network's previous changes to its simulated affective state. In this way emotional reactivity could be seen as

346

Chapter 7

related to openness to experience. Thus by decreasing the nctwork's analog of openness to experience we implicitly decrease an analog of emotional reactivity. Using this formalism for momentum it is expectexi that more reactive individuals would be vulnerable to cognitive correlates of depression, a phenomenon which is often observed clinically. This prediction could be tested by investigating whether individuals with low emotional reactivity are also low on traditional measures of openness to experience and vice-versa. Were this true, literature describing aspects of intellect might be interpreteA as relevant to the investigation of depression in new and thought-provoking ways. Yet another way of interpreting the momentum parameter is as a mediator of "kindling" (Cytryn & MeKnew, 1996; Ferrier, 1991) in depression. Kindling theorists suggest that each time an individual is exposed to a depressogenie stimulus, it becomes progressively easier for the individual to become depressed. Thus, the intensity of loss needed to make an individual depressed might be decreased with each exposure to loss. Because the momentum parameter governs how much previous stimulus-indueeA change will promote future stimulus-induced change, a high momentum parameter might be analogous to an individual who is vulnerable to kindling effects. That is, a network in which the momentum is high is likely to attain negative information processing biases after subsequent exposures to a stimulus deemed to be negative. Openness to experience affects recovery too. Interestingly, while decreasing the learning rate and momentum tend to decrease the likelihood that a network will assume information processing biases characteristic of depression, these same factors make information processing biases which are acquired very difficult to get rid of, because decreasing learning parameters tends to decrease the network's ability to learn information. For example, Figure 16 shows the learning curves for two networks which have received a large amount of training on negative information (100 epochs), followed by 75 epochs of overtraining on positive information. One network has the learning rate used in previous simulations (0.2), and the other, a high learning rate (0.5). The network with the low learning rate may take longer to become biased as a result of overtraining on negativity, but also takes longer to learn positive information after having done so. Clinically, this result suggests that learning slowly from one's environment may prote~ one against the effects of depression initially, but may also hinder recovery if such an individual does become depressed. Such a phenomenon may provide evidence for Teasdale's (1988) differential activation hypothesis, which states that different factors

G.J. Siegle and R.E. Ingram

347

0.18 ] Fta=0.2 0.16

--tr- Eta=0.5

0.14 0.12 0.1 elTor

0.08 0.06 0.04 0.02 0

I I I II I1'11 I I I I

IIIIIII

II j l I I i I I I I I I I I

II

epochs

Figure 16. Error rate for learning positive information over time after having received overtraining on a negative stimulus.

contribute to the onset and maintenance of depression, and that clinical research should address these processes separately. Conclusions from simulatwns of openness to experience. The simulations of factors associated with openness to experience suggest that like coping strategies, aspects of intellect could either help or hurt an individuals chances of adopting information processing biases characteristic of depression. Also, similar to the simulations of coping style, these simulations suggest that the same factors which initially protect individuals from depression by causing them not to overleam (or overinterpret) their experiences, can also hinder their recovery from depression. By realizing that the factors such as intellect and coping can act differently for at-risk and currently depressed individuals different types of preventative treatments vs. crisis intervention type treatments may be devised. Still, further investigation of differences in the effects of trait variables on vulnerable and currently

348

Chapter 7

depressed individuals is necessary to better clarify the claims derived from simulation of intellectual factors. A Brief Conclusion

By examining how network parameters relevant to coping, intellect and cognitive structure govern the network's performance on information processing, the simulations have revealed a rich array of possible variation in the network's performance. Some of these factors appear to increase the network's vulnerability to negative information processing biases. Other factors appear to protect the network against these biases. In some cases the same factor earl act as a risk or protective factor based on other aspects of the network. The simulations thus suggest that individual differences can play a key role in understanding both depressive information processing biases, and information processing in individuals who may be at risk for depression. Simulations of variables representing types of individual differences can help explain variation in behaviors, and provide valuable insights for future research. References

American Psychiatric Association (1994). Diagnostic and statistical manual of mental disorders - Fourth edition. Washington, DC: American Psychiatric Association. Anderson, J. A. (1990). Hybrid computation in cognitive science: Neural networks and symbols. Applied Cognitive Psychology, 4, 337-347. Arbib, M. (1987). Brains, machines, and mathematics. New York: SpringerVerlag. Bamden, J. (1995). Artificial Intelligence and Neural Networks. In M. Arbib (Ed.), The handbook of brain theory and neural networks. Cambridge, MA: MIT Press. Beck, A.T. (1967). Depression: Clinical, experimental, and theoretical aspects. New York: Hoeber. Beck, A. T. (1976). Cognitive therapy and the emotional disorders. New York: International Universities Press. Blaney, P. (1986). Affect and memory: A review. Psychological Bulletin, 99, 229-246. Blank, D. S., Meeden, L. A., & Marshall, J. B. (1991). Exploring the symbolie/subsymbolie continuum: A case study of RAAM. In J.

G.J. Siegle and R.E. lngram

349

Dinsmore (Ed.), Closing the gap: SymboBsm vs. connectionism. Hillsdale, NJ: Erlbaum. Bower, G. (1981). Mood and memory. American Psychologist, 36, 129-148. Brewin, C. R., Andrews, B., & Gotlib, I. (1993). Psychopathology and early experience: A reappraisal of retrospective reports. Psychological Bulletin, 113, 82-98. Cantor, N. (1990). From thought to behavior: "Having" and "doing" in the study of personality and cognition. American Psychologist, 45, 735750. Challis, B. H., & Krane, R. V. (1988). Mood induction and the priming of semantic memory in a lexieal decision task: Asymmetric effects of elation and depression. Bulletin of the Psyehonomie Society, 26, 309312. Cytryn, L., & McKnew, D. H., Jr. (1996). Growing up sad: Childhood depression and its treatment. New York: W. W. Norton & Co, Inc. Cohen, J. D., & Servan-Schreiber, D. (1992). Introduction to neural network models in psychiatry. Psychiatric Annals, 22, 113-118. Cohen, J., Dunbar, K., & MeClelland, J. (1990). On the control of automatic processes: A parallel distributed processing account of the Stroop effect. Psychological Review, 97, 332-361. Collins, A., & Loftus, E. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82, 407-428. Fallman, S. E., & Lebiere, C. (1991). The cascade-correlatwn learning architecture. (Teeh. Rep. No. CMU-CS-90-100). Pittsburgh, PA: Carnegie Mellon University, Computer Science Department. Femer, I. N. (1991). Chronic depression: Future prospects from recent research. International Clinical Psychopharmacology, 6, 83-89. Digman, J. M., & Takemoto-Chock, N. K. (1981). Factors in the natural language of personality: Re-analysis, comparison, and interpretation of six major studies. Mulavariate Behavioral Research, 16, 149-170. Goldfried, M. R., & Davison, G. C. (1976). Clinical behavior therapy. New York: Holt. Goldberg, L. (1993). The structure of phenotypic personality traits. American Psychologist , 48, 26-34. Gotlib, I. H., & McCann, C. D., (1984). Construct accessibility and depression: An examination of cognitive and affective factors. Journal of Personality and Social Psychology, 47, 427-439.

350

Chapter 7

Harper, D. C., & Wadsworth, I. S. (1990). Dementia and depression in elders with montal retardation: A pilot s t u d y . Research in Developmental Disabilities, 11, 177-198. Hecht-Neilsen, R. (1990). Neurocomputing. Roading, MA: Addison-Wesley Publishing Company. Ingram, R. E., & HoUon, S. D. (1986). Cognitive therapy of depression from an information processing perspective. In R. Ingram (Ed.), Information processing approaches to clinical psychology (pp. 261-284). New York: Academic Press. Ingram, R. (1984). Toward an information processing analysis of depression. Cognitive Therapy and Research, 8, 443-478. Klein, M. H., Wondorlich, S., & She,a, M. T. (1992). Modols of rolationships between personality and depression: Toward a framework for theory and research. In M. H. Klein, D. I. Kupfer & M. T. Shea (Eds.), Personality and depression. Now York: Guilford. Lester, D., & Schaofflor, I. (1993). Sdf-defeating personality, depression, and suicidal ideation in adolescents. Psychological Reports, 73, 113114. Lewinsohn, P. M., & Hoberman, H. M. (1982). Depression. In A. S. Bellack, M. Herson, & A. E. Kazdin (Eds.), International handbook of behavior modification and therapy. New York: Plenum Press. Mac~, C., & Mathews, A. M. (1991). Cognitive-experimental approaches to the r disorders. In P. R. Martin (Ed.), Handbook of behavior therapy and psychological science: An integrative approach. Now York: Pergamon Press. MacLcod, C., Mathews, A.M., & Tata, P. (1986). Attcntional bias in emotional disorder. Journal of Abnormal Psychology, 95, 15-20. Marx, E. M., Williams, I. M.G., & Claddge, G. C. (1994). Social problemsolving in depression. European Review of Applied Psychology~Revue Europeenne de Psyehologie Appliquee, 44, 271-279. Matthews, G., & Southall, A. (1991). Depression and the processing of emotional stimuli: A study of somantir priming. Cognitive Therapy and Research, 15, 283-302. McClr I. L., Rumr D. E., & Hinton, G.E. (1985). The appeal of parallel distributed processing. In I. L. McClelland & D. E. Rumelhart (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition (Vol. 2, pp. 3-44). Cambridge, MA: MIT Press.

G.J. Siegle and R.E. Ingram

351

McCrae, R. R., & Costa, P. T. (1985). Updating Norman's "adequacy taxonomy"" Intelligence and personality dimensions in natural language and in questionnaires. Journal of Personality and Social Psychology, 49, 710-721. Meichenbaum, D. (1985). Stress inoculation training. Elmsford, NY: Pergamon Press. Miyata, Y. (1991). A user's guide to PlaNet version 5. 6: A tool for constructing, running, and looking into a PDP network. (Available from Yoshiro Miyata, Department of Computer Science, Univ. of Colorado at Boulder, Boulder, CO 80309-0430). Morrow, J., & Nolen-Hoeksema, S. (1990). Effects of responses to depression on the remediation of depressive affect. Journal of Personality and Social Psychology, 58, 519-527. Newell, A., & Simon, H. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall. Nolen-Hoeksema, S. (1987). Sex differences in unipolar depression: Evidence and theory. Psychological Bulletin, 101,259-282. Nolen-Hoeksema, S., & Morrow, J. (1993). Effects of rumination and distraction on naturally occurring depressed mood. Cognition & Emotion, 7, 561-570. Nolen-Hoeksema, S., Morrow, J., & Fredrickson, B. L. (1993). Response styles and the duration of episodes of depressed mood. Journal of Abnormal Psychology, 102, 20-28. Paykel, E. S. (1979). Causal relationships between clinical depression and life events. In Barrett, J. E. (Ed.), Stress and mental disorder (pp. 7186). New York: Raven Press. Persons, J., & Miranda, J. (1992) Cognitive theories of vulnerability to depression: reconciling negative evidence. Cognitive Therapy and Research, 16, 485-502. Pyszczynski, T., & Greenberg, J. (1987). Self-regulatory perseveration and the depressive self-focusing style: A self-awareness theory of reactive depression. Psychological Bulletin, 102, 1-17. Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85, 59-108. Rateliff, R. (1990) Conneetionist models of recognition memory: Constraints imposed by learning and forgetting functions. Psychological Review, 97, 285-308.

352

Chapter 7

Ruiz Caballero, J. A., & Bermudez Moreno, J. (1992). Individual differences in depression, induced mood, and perception of emotionally toned words. European Journal of Personality, 6, 215-224. Rumelhart, D. E., Hinton, G. E., & McClelland, J.L. (1986). A general framework for parallel distributed processing. In D. E. Rumelhart, J. L. MeClelland, & the PDP Research Group (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition (Vol. 1, pp. 45-76). MA: MIT Press. Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986). Learning internal representations by error propagation. In D. E. Rumelhart, J. L. McClelland, & the PDP Research Group (Eds.), Parallel distributed processing: Explorations in the microstructure of cogniaon (Vol. 1, pp. 318-362). MA: MIT Press. Sackeim, H. A., Freeman, J., Mcelhiney, M., & Coleman, E. (1992). Effects of major depression estimates of intelligence. Journal of Clinical & Experimental Neuropsychology, 14, 268-288. Sarle, W. S. (1994). Neural networks and statistical models. Proceedings of the Nineteenth Annual SAS Users Group International Conference. Sattler, J. M. (1993). Assessment of children. San Diego, CA: Jerome M. Sattler. Schank, R. C., & Abelson, R. P. (1977). Scripts, plans, goals, and understanding. Hillsdale, NJ: Erlbaum. Schwartz, R. M., & Garamoni, G. L. (1989). Cognitive balance and psychopathology: Evaluation of an information processing model of positive and negative states of mind. Clinical Psychology Review, 9, 271-294. Schill, T., & Kramer, J. (1991). Self-defeating personality, selfreinforcement, and depression. PsychologicalReports, 69, 137-138. Segal, Z. V., & Ingram, R. E. (1995). Mood priming and construct activation in tests of cognitive vulnerability to depression. Clinical Psychology Review, 14, 663-695. Siegle, G. (1996, in press). Why I make models (or what I learned in graduate school about validating clinical causal theories using computational models). The Behavior Therapist. Siegle, G., Ingrain, R. E., & Matt; G. (1995). A neural network model of informaaon processing biases in depression. Presentation at the workshop" Neural modeling of cognitive and brain disorders. College Park, Maryland.

G.J. Siegle and R.E. Ingram

353

Siegle, G., Ingram, R. E., & Matt, G. E. (1996a, submitted). Affect interference: Cause for negative attention biases in depression? 3 S iegle, G., Ingram, R. E., & Matt, G. E. (1996b, in preparation). Affect interference in depression: A neural network model. 3 Teasdale, J. (1988). Cognitive vulnerability to persistent depression, Cognition and Emotion, 2, 247-274. Tryon, W. W. (1993). Neural networks: I. Theoretical unification through connectionism. Clinical Psychology Review, 13, 341-352. Tucker, D. M., & Derryberry, D. (1992). Motivated attention: Anxiety and the frontal executive functions. Neuropsychiatry, Neuropsychology, & Behavioral Neurology, 5, 233-252. Weissman, M. M., Bruce, M. L., Leaf, P. J., Florio, L. P., & Holzer, C. (1991). Affective disorders. In L. N. Robins & D. A. Regier (Eds.), Psychiatric disorders in America (pp. 53-80). New York: Free Press. Winfrey, P. L., & Goldfried, M. R. (1986). Information processing and the human change process. In R. E. Ingram (Ed.), Information processing approaches to clinical psychology (pp. 241-258). New York: Academic Press. Yates, J., & Nasby, W. (1993). Dissociation, affect, and network models of memory: An integrative proposal. Journal of Traumatic Stress, 6, 305326. Zeidner, M., & Endler, N. S. (Eds.) (1996). Handbook of coping: Theory, research, applications. New York: John Wiley & Sons.

3 Much of this informationis also presented in Siegle, G. (1996). Ruminationon affect: Causefor negativeattention biases in depression? UnpublishedMaster's Thesis, San Diego State University.

This Page Intentionally Left Blank

Cognitive Science Perspectives on Personality and Emotion - G. Matthews (Editor) 9 1997 Elsevier Science B.V. All rights reserved. CHAPTER 8

Emotion and Reason. The Proximate Effects and Ultimate Functions of Emotions Timothy Ketelaar and GeraM L. Clore

There once was a peasant farmer whose prized possession was a white horse. One day the King's procession passed his farm and the King remarked, "I wish to purchase this animal. I will give you 60 pounds." The humble peasant replied, "I am sorry, this horse is not for sale." The king's procession departed. Several days later the horse ran off into the woods and was nowhere to be found. The townsfolk gathered around the peasant and remarked, "l.xmk how foolish you have been, you could have earned 60 pounds, yet now your horse has run off and you have nothing. Truly this is not good!" Several days later, the horse returned and brought with it three more white horses. The townsfolk again gathered around the peasant and remarked, "Look how wise you have been, you would have earned only 60 pounds by selling your horse, yet now you have three additional steeds. Truly this is good!" Several days later the peasant's son broke both legs while attempting to train the new horses. The townsfolk now remarked, "Look how foolish you have been, you could have earned 60 pounds if you had sold this beast, yet now your only son is crippled. Truly this is not good!" The point of this story is that what is good or bad is often relative. In emotion research the effects of emotion on judgment and cognitive processing are oiten interpreted as disruptive and bad relative to a single normative standard of what constitutes good reasoning. In the current chapter we suggest how other standards of reference, such as those suggested by evolutionary psychological concerns about "good design," provide another interpretation of research on emotion and cognition. We suggest that the distinction between proximate effects . (what a mechanism can do) and

356

Chapter 8

ultimate functions (what a mechanism is designed to do) gives rise to the intriguing question of whether some of the proximate information processing effects of emotion are not in fact evolved "design features" of emotion. It is in this sense that one might argue that a focus on what emotions can do (proximate effects) does not necessarily give an accurate picture of what emotions are designed to do (their ultimate functions).

Why Does Emotion Affect Cognition? Does emotion affect reasoning? The last two decades of the 20th century were predicted to be the "decades of emotion" (Tomkins, 1981). From the vantage point of approaching the second millennium we can pause and ask ourselves what some of the conclusions are from this period of heightened focus on emotion. In the domain of judgment and decision-making, investigators have traditionally made no mention of emotion at all (e.g., Arkes & Hammond, 1986; Dawes, 1997). But among the important findings of the last two decades has been the demonstration that consciously accessible feelings otten serve as important input into judgment processes that were previously considered non-emotional and purely cognitive (see Clore, Schwarz, & Conway, 1994; Schwarz & Clore, 1988, 1996; for reviews). A variety of judgments, from ratings of satisfaction with life (Schwarz & Clore, 1983) to ratings of satisfaction with one's electrical appliances (Isen, Shalker, Karp, & Clark, 1978), are sensitive to influence from momentary moods. In addition, mood also influences styles of information processing. Positive moods are generally associated with heuristic processing, including reliance on stereotypes (Bodenhausen, 1994), scripts (Bless et al., 1996), and expectations (Isbell, Clore, & Wyer, 1997), while negative moods are associated with more systematic processing of all sorts of tasks, such as solving syllogisms (Melton, 1994), assessing persuasive arguments (Sinclair, Mark, & Clore, 1994), evaluating movies (Kaplan, Kickull, & Reither, 1996), and so on. Thus, to the extent that reasoning includes judgment, decision-making, and problem solving, there is ample evidence that emotion does influence reasoning. Does emotion bias reasoning? A long-standing belief in Western culture is that emotion is the enemy of reason. Indeed, a hallmark of sound judgment is that it be uninfluenced by emotion. This is true regardless of whether the judgments take place in courtrooms, county fairs, ballparks, or voting booths. Thus, emotion is readily seen as a source of bias and distortion. In the research cited above, mood and emotion may be viewed as impairing

T. Ketelaar and G.L. Clore

357

judgment by causing more positive and negative ratings than would otherwise be given, and distorting reason by making people more gullible and less able to process information systematically. Bodenhausen (1994, p. 59), for example, interpreted the reliance on stereotypes that he observed in his research on judgments of individuals accused of wrong-doing as suggesting "reduced motivation for thoughtful analysis" or even "reduced capacity for such analysis" brought about by emotion. Taken at face value, social psychological research does suggest that emotion leads to bias and distortion, because it often focuses on the role of situationally irrelevant emotion. For example, the positive and negative feelings experienced by patrons as they lett movie theaters were found to influence various judgments, including their views on political figures, future events, crime, and life satisfaction (Forgas & Moylan, 1987). The purpose of studying such irrelevant affect is to separate the influence of emotional feelings from the influence of cognitive beliefs. Ordinarily our feelings about something are hopelessly intertwined with our beliefs about it. To examine the role of affective feeling independently of such cognitive content, experiments are carefully arranged so that participants will misattribute their feelings caused by mood as reactions to the object being judged. In other words, research participants are maneuvered into showing judgmental errors and reasoning biases in order to examine the effects of feelings independently of thematic content. Do such samples of behavior provide an accurate picture of the role of emotion in reason? In some ways yes, and in some ways no. Yes, in that they show that emotional influences are often mediated by emotional feelings themselves rather than by their associated cognitive content. No, in that they imply that everyday emotional influences generally involve errors and biases. In fact, in the real world, there is usually a good deal of redundancy between feelings and beliefs. Rather than being irrelevant, the feelings that influence our judgments about something are usually reactions to the those same beliefs. Indeed, that is why it was necessary for social psychologists to develop misattribution procedures in the first place. Although most psychologists view demonstrations of emotional influence as evidence of bias and distortion, some others are beginning to focus on functional rather than dysfunctional relationships between emotion and cognition (Hirshleiefer, 1987; Frank, 1988; Nesse, 1990; Nesse & Williams, 1994). These contrasting orientations make it difficult to amve at a general psychological answer to the question: "Do emotions actually serve important information processing purposes?" Indeed, as in the opening fable, some

358

Chapter 8

emotion researchers gather around such findings and proclaim: look how important emotion is for judgment, truly this is "good" evidence that emotions are functionally involved in information-processing. But because research has focused mainly on emotional biases in judgm,ent and limitations in processing, others also flock to these same findings' and say, look how disruptive emotions are to information processing, truly emotional reasoning is "bad", dysfunctional, and to be avoided. How do we know? What constitutes good or bad information processing depends, of course, on the standard of reference one adopts. It turns out that there is more than one vantage point from which to judge the rationality of emotional reasoning. Recent advances in the emerging perspective of evolutionary psychology (Barkow, Cosmides, & Tooby, 1992; Buss, 1995; Gigerenzer, 1995) have suggested a new set of standards for differentiating those emotional processes that lead to bad reasoning from those that lead to good reasoning. These evolutionary approaches stress the distinction between proximate effects (what a mechanism can do) and ultimate functions (what a mechanism is designed to do; see Barkow et al., 1992; Buss, 1995; Dawkins, 1982, 1986; Hinde, 1970; Tinbergen, 1963). The proximate effect-ultimate function distinction suggests that even though some biases may be disruptive and problematic, that may not be the end of the story. Certain emotional biases may serve as a useful mechanism for focusing attention and drawing inferences about the meaning of one's current situation. Although many emotion theorists have assumed that emotional information processing is somehow adaptive (see Lazarus, 1991), there have been few attempts to identify cognitive problems that emotions actually help solve. Instead we have demonstrations that emotions simply change cognitive processing (a proximate effect). In this chapter, we seek to demonstrate that broadening the standard of reference for evaluating good vs. bad reasoning to include evaluations of both proximate effects and ultimate functions can turn some examples of apparently irrational emotional reasoning into examples of well-designed reasoning mechanisms. In doing so, we may be able to address more adequately questions of whether emotions are designed to shape information processing.

Specific Aims of this Chapter In this chapter, we address questions about whether emotional effects on reasoning arc functional or dysfunctional on both conceptual and empirical grounds. First, we show that psychologists often assume that emotions are

T. Ketelaar and G.L. Clore

359

obstacles to good reasoning. Second, we provide a brief overview of emotion and cognition research, showing that the proximate link between emotion and cognition (that emotions affect cognitions) is well established. Third, we suggest that emotions provide information that is of value in cognitive tasks. We do this by examining the logic of specific information-processing problems and showing how emotions could, in principle, and actually do, in laboratory experiments, help solve them. We conclude with a summary of this view and suggest directions for future research. Views o f emotion as dysfunctional are common. Traditional views have sometimes stressed automatic behavior or action-preparedness as the main functions of emotion. That perspective stresses that emotions prepare organisms to engage in specific behaviors such as fight or flight to meet certain adaptive challenges and problems (e.g., Lang, 1995; Frijda, 1986; Scherer, 1984). Presumably we do possess emotions, in part, to guide behavior that is critical to survival. For example, fear results in escape from danger and parental love results in care of offspring. But in contrast to behavioral views, the present cognitive perspective stresses the informational functions of emotion. We argue that emotions provide information, focus attention, and guide information processing. In the present chapter we emphasize cognitive rather than behavioral functions of emotion. Compared to studies of emotional behavior (LeDoux, 1996) or emotional expressions (e.g., Ekman, 1982), the role of emotion in judgment and reasoning is less obviously adaptive. That the influence of emotions on judgment and decision-making is evidence of "good design" is at odds with the views of many: Today the image of man is no longer that of an individual enslaved by his passions, but rather that of a philosopher making decisions on the basis of logical deduction and inference. In this tradition, emotion is seen as a regrettable flaw in an otherwise perfect machine (Scherer, 1984, p. 293). An economist (Elster, 1995, p. 1394) adds: The standard view of the relation between rationality and emotions is, of course, that emotions interfere with rationality. They are, as it were, sand in the machinery of action. Nobody would deny that this often true.

Chapter 8

360

In discussing how psychologists have attempted to understand the role of emotions in human reasoning, cognitive scientist Piattelli-Palmarini (1994, p. i 65) suggests that: When powerful emotions combine, the cognitive scientist walks about on tiptoe. Emotions are not her field of expertise, and the picture is hard enough without having emotions interfering. Broader views of emotion commonly portray affective states as symptomatic of irrationality or worse yet, disease (see Nesse & Williams, 1994, for a good review). And some others that do not see emotion as a threat to reason, treat the passions (emotions, moods, and affects) as conceptually separate from reasoning and cognition (Zajonc, 1980, 1984). Rather than viewing emotions as separate from cognition or as symptomatic of irrational and maladaptive thought, we argue that emotion is integral to good reasoning (for a neurological route to the same conclusion, see Damasio, 1994). One might argue that both of these qualities - our ability to feel and our ability to reason - are important aspects of human nature. We have not only a capacity to be rational, but also the capacity to be passionate in our decision-making. Rather than treating passions and emotions as "limitations upon self-interested rationality, might not these seeming disabilities actually be functional?" (Hirshleifer, 1987, p. 321). Although empirical examples of obviously functional relations between emotion and information processing are few in number (see Isen, Daubman, & Nowicki, 1987; Platt & Spivack, 1974), there is little doubt that emotions and reasoning are commonly intertwined.

Consequences of Mood

Mood and judgment A reliable example of affectivc influences on cognition is the effect of mood on judgment (see Clore, 1992; Clorc ct al, 1994). Evaluative judgments are usually found to be more positive when people are in good moods than when they are in foul moods (Forgas & Moylan, 1987). According to the affect-as-information hypothesis (Schwarz & Clorc, 1983) affcctivc reactions arc a useful source of information when making evaluative judgments. Contrary to traditional accounts by judgment and decision theorists, this view holds that people often make everyday judgments by asking themselves,

T. Ketelaar and G.L. Clore

361

"How do I feel about this?" (Schwarz & Clore, 1988). As indicated above, traditional theories assume that judgments reflect people's beliefs about the object of judgment. To separate the roles of beliefs and feelings, affective states are sometimes experimentally induced from irrelevant sources, such as having participants view happy or sad films or write descriptions of recent happy or sad events. These experiments show that evaluative judgments can be influenced by whatever feelings are available at the time unless the irrelevant source of the feelings is made salient. Thus, the impact of our feelings on evaluative judgments depends on their perceived informational value. Tests of this affect-as-information hypothesis often include an attribution condition in which the true source of the feelings is made salient. Generally it is found that mood influences can be eliminated by this manipulation, suggesting that they depend on the apparent information provided by the associated feelings. For example, Schwarz and Clore (1983) found in two experiments that the effects of mood on judgments of life satisfaction disappeared when respondents attributed their feelings to irrelevant, situational causes. In one experiment, subjects reported greater life satisfaction and more positive moods during telephone interviews conducted on spring days that were warm and sunny than on days that were cold and rainy. But this difference disappeared when the interviewer directed their attention to the true source of their feelings by asking about the weather. In another version of the experiment, mood was induced in the lab by asking participants to recall a happy or sad life event. Again, the influence of mood on judgments of life satisfaction disappeared when they were led to attribute their feelings to an irrelevant, situational source. In this instance, they misattributed them as reactions to an unusual soundproofed room in which the experiment was conducted. These experiments and others like them (e.g., Keltner, Locke, & Audrain, 1993; Schwarz, Servay, & Kumpf, 1985; Siemer & Reisenzein, 1994) show that irrelevant feelings can influence judgment. Mood effects are generally described in the literature as examples of "emotional biases," and in truth they hardly provide compelling evidence that emotions are adaptive. However, uncovering emotional biases was not the motivation for conducting those studies. Irrelevant feelings were reduced merely as a tool to enable us to trace the role of feelings in the judgment process. Inducing moods is a useful procedure because the feelings of mood have a long half-life and are easily misattributed. But in ordinary life, most affective influences do not come

362

Chapter 8

from irrelevant sources and do not depend on misattribution. The affective reactions that people usually experience are caused by whatever is currently their focus of attention. Of course, for individuals who are depressed or whose affect is chronically skewed, feelings may indeexl represent sources of irrationality and bias. Moreover, the fact that mood can have the effects we have suggested does indicate potential pitfalls in the judgment process. Nevertheless, the results do not answer dearly questions about whether affective influences on judgment and decision making are adaptive or maladaptive.

Mood and processing In addition to the research on mood and judgment, recent findings also show that moods influence styles of information processing. It would be easier to make a ease that emotions undermine reason from this literature. For example, Ellis and colleagues have shown that individuals in sad moods often do more poorly on recall tasks than those in happy or neutral moods. In one such experiment, individuals in sad moods were shown to be less able to recall target words from complex sentences they had read, although mood had no effect when the sentences were simpler (Ellis, Thomas, & Rodriguez, 1984). As an explanation, the authors hypothesized that sad moods reduce cognitive capacity that could have been allocated to remembering the target words. The resource allocation hypothesis (Ellis & Ashbrook, 1988) has since become a standard explanation for mood effects. Sometimes, the same resource-based explanation is used to account for the effects of positive rather than negative mood (e.g., Isen, 1987; Mackie & Worth, 1989). Isen, for example, hypothesized that positive material in memory "is more extensive and at the same time better integrated, so that positive affect is able to cue a wide range of thoughts" (Isen, 1987, p. 217). As a result, being in a positive mood is believed by some to limit cognitive resources due to intruding positive thoughts. Evidence supporting this idea comes from studies of persuasion in which individuals in positive moods show less systematic processing of counter attitudinal messages than those in negative moods (e.g., Mackie & Worth, 1989; Worth & Mackie, 1987). In such studies, happy recipients are usually found to be equally persuaded by strong and weak arguments, suggesting that happy moods may have either depleted their cognitive resources or reducexl their motivation to engage in systematic processing.

T. Ketelaar and G.L. Clore

363

On the other hand, an explanation based on the informational properties of affect does not require assumptions about reduced capacity or motivation. In this view, affective feedback in problem solving situations informs individuals about the adequacy of their current construction of the situation. Experiencing positive cues may lead one to heuristic processing, including reliance on one's own expectations, intuitions, and inclinations, while negative affective cues may imply problems, leading one to inhibit reliance on such prior knowledge and encourage collecting new information and processing systematically (Clore et al., 1994; Schwarz, 1990). As a test of this application of the affect-as-information hypothesis, Sinclair, Mark, and Clore (1994) replicated Schwarz and Clore's (1983) original study showing mood effects for sunny and rainy weather. However, this time they examined persuasion instead of judgments of life satisfaction. Students were approached on early spring days when the weather was sunny and pleasant or on subsequent days when it had turned cloudy and unpleasant. The persuasive messages they received had previously been established to be either strong or weak arguments. To vary the informational implications of the feelings of some subjects, Sinclair et al. drew their attention to the weather. When subjects' attention was not drawn to the weather, the previously obtained effects were observed. Sad respondents were persuaded by strong but not weak messages, while happy respondents were equally persuaded by both. However, when the weather was made salient as a potential cause of subjects' momentary feelings, mood no longer played a role and only a general effect of message strength remained. In addition to studies of persuasion, mood has also been shown to influence processing in a variety of other contexts. That research has been summarized elsewhere (e.g., Clore et al., 1994; Schwarz & Clore, 1996). In general, happy subjects depend on their prior knowledge, whereas sad subjects rely on new information. For example, in an election study, happy individuals were found to be more likely to depend on their prior knowledge of the party identification of political candidates as opposed to what candidates actually said (Marcus & MacKuen, 1993). In a consumer study, subjects in happy moods were more likely to rely on prior knowledge about brand names as opposed to specific features of consumer goods they considered (Adaval, 1996). In studies of film preferences, happy subjects have been found to rely for their choices on prior knowledge of the genre of the films (horror, comedy, adventure, etc.) as opposed to the quality of the films (Kaplan, 1995). And in a study of stereotypes, happy subjects were more likely to rely on their prior expectations about a person as opposed to

364

Chapter 8

basing their impressions on the person's actual behavior (IsbeU, Clore, & Wyer, 1997). There appears to be a consistent pattern in these studies showing the debilitating effects of good mood on reasoning. But damning as these proximal effects seem, we suggest that they are actually by products of a useful affective function. We assume that it is generally useful to attend to our affective cues as we appraise the appropriateness of our knowledge and expectations for the tasks in which we are engaged. Ordinarily it would be adaptive to rely on what we know when we feel confident, but to pick up new information when we do not feel confident. We suggest that affect provides this information, positive affect telling us to assimilate incoming information to our existing conception, and negative affect to accommodate our conception to the data. From a different starting point, Gray (1971) has made analogous interpretations of the role of positive and negative affect. According to Gray, positive affect leads organisms to behave on the basis of habit and negative affect leads them to engage in learning. These two processes, habit and learning, may be what some evolutionary psychologists refer to as two distinct styles of information processing: selectional and instructional learning, respectively (see Gazzaniga, 1992). Summary. The research on the effects of mood on judgment and processing is ambiguous in its implications for whether emotional influences are functional or not. The mood studies do involve mistaken attributions by subjects about the sources of their feelings, but the use of irrelevant moods is simply intended to separate subjective experience from other factors. In addition, several different kinds of experiments on information processing showed that some states - usually states of happiness or positive affect inhibited systematic processing. But from the standpoint of the affect-asinformation hypothesis (Schwarz & Clore, 1983), engaging in heuristic rather than systematic processing does not necessarily indicate a shortage of cognitive capacity or process'rag motivation (Bless et al., 1996). Instead it may be a manifestation of a normally adaptive process whereby one is led to rely on existing knowledge and expectations when one senses that a task is going well and to acquire new information when one senses that it is not going well. The larger point is that it can be perilous not to consider alternatives to the proximal effects of emotions seen in laboratory experiments when trying to deduce function. To illustrate the problem, imagine that you and a colleague stand before a John D~re riding lawn mower and are asked to explain to a hunter-gatherer native from New Guinea just what this thing

T. Ketelaar and G.L. Clore

365

does. Imagine that your colleague, with the help of several sturdyassistants, proceeds to affix the lawn mower - by its seat - to the ceiling. Your colleague then starts the engine and exclaims that what lawn mowers do is circulate air, that they are elaborate ceiling fans! You would think your colleague crazy, not because he has not successfully demonstrated something that lawnmowers can do (a possible proximate effect), but because he has not accurately described what lawn mowers are designed to do. We argue that the study of emotional feeling states is no different. The difficult task is to distinguish what emotions can do, from what they are designed to do. Although both levels of description are essential to a complete understanding of "What affect does" it would seem that the more ultimate functions of affect may require further analysis. If evolutionary psychologists are correct in assuming that our minds are "designed" to solve the adaptive problems of ancestral, not modem, environments (see Bowlby, 1969), a failure to distinguish what emotions can do and what they are "designed" to do, could be problematic for several reasons. Just as particular perceptual systems can sometimes be fooled when they operate in environments in which they were not evolved to operate (e.g., in an Ames room; see Ames, 1951), it should not be surprising to observe that particular emotions sometimes wreak havoc on cognitive tasks removed from the environments in which they evolved (Bowlby, 1969; Sperber, 1995; Tooby & Cosmides, 1990). What would be surprising would be the conclusion that emotions were designed, by evolution, to render faulty judgments. Perhaps the proximate effect-ultimate function distinction can help researchers to distinguish between what emotions can sometimes do (i.e., they can sometimes "bias" judgments and lead to limited processing) and what emotions were designed by evolution to do (i.e., perhaps emotions were designed, in part, to "bias" judgments and alter processing in adaptive ways). Because the proximate effects-ultimate function distraction is not yet explicit in contemporary emotion-cognition research, it is not clear whether many of the well-known demonstrations of emotional effects on information processing are demonstrations of what emotions are designed to do or merely illustrations of what they can sometimes do in certain circumscribed conditions.

Consequences of Emotions Useful as mood research has been in illuminating these processes, it would be a mistake not to distinguish the effects of mood from those of emotion. What about specific emotions? The primary difference between

366

Chapter 8

moods and emotions, from our point of view, is that moods are general and emotions are specific. The difference in the generality and specificity of moods and emotions is apparent both in their objects and in their inherent structure.

Emotions have objects Moods and emotions differ in specificity partly because emotions have objects and moods do not. That is, emotions are intentional states, mcamng that they arc necessarily about something or have objects (Harr6, 1986). One can say that one is glad about finishing a project or disappointeA that it took longer than expected, but one cannot say that one is in a glad or disappointed mood. Moods have causes, of course, but the causes are generally not a salient part of the experience. As a result of the difference in specificity contributed by having or not having an object, there are relatively few kinds of moods (e.g., happy, sad, irritable), but many distinct emotions. Also, in terms of their cognitive effects, one consequence is that the feelings associated with moods arc easily misattributexl to a particular object or cause, while the contingent nature of the feelings associated with emotions means that they carry their object with them and hence are less likely to be misattributcd to some other object (Kcltncr, Locke, & Audrain, 1993). This distinction between moods and emotions is not just a semantic nicety, but carries with it practical implications for action. The presence of an object means that one can engage in what Lazarus and Folkman (1984) have called "problem-focused" coping. That is, one can maintain or diminate the emotion by engaging in relevant action toward the object of the emotion. Thus, when angry at a co-worker one can complain to him or to the boss and seek redress. But when merely in a bad mood in which things in general seem unsatisfactory, it is unclear what steps might be taken. Moods are therefore more likely to elicit "emotion-focusexl" coping in which one can deal only with the feelings and not with their causes (e.g., trying to distract oncsdf). Thus, cvcn though both moods and emotions involve pleasant and unpleasant feelings, the consequences for action differ because of the relative salience of an object.

Emotions have structure Both emotions and moods are affeetive states, which means states concerned with the goodness or badness of something. But emotions not only

1". Ketelaar and G.L. Clore

367

have objects that moods do not, but emotions also convey in what way something is good or bad. That is, emotions have some inherent structure than moods do not have. Emotions can be thought of as bodily, cognitive, and experiential representations of personally important aspects of situations. The emotionally critical features of such situations have been specified by various cognitive theories of emotion (e.g., Ortony, Clore, & Collins, 1988; Roseman, 1984, Smith & Ellsworth, 1987) that attempt to discriminate the different emotions in these terms. For example, according to Ortony et al. (1988), feelings of frustration concern situations in which one is displeased because events have thwarted one's goals. Feelings of reproach, however, concern situations in which one disapproves of someone's blameworthy actions. And feelings of anger concern the two together; situations involving the perception that one's goals have been thwarted by someone's blameworthy actions. Such accounts give linguistic characterizations of the deep structure of particular emotions. In one version of his transformational theory of grammar, Chomsky (1965) suggested that speech acts with different surface structures might have the same meaning by virtue of various transformations in the same deep structure. Analogously one might imagine a deep structure of an emotional situation that makes it an angry rather than a fearful or a joyful situation. Perhaps this deep structure can be represented in terms of the distinct pattern of appraisals and responses that is elicited by a particular emotional situation (see Ortony et al., 1988, for one such structural model). When a person is angry, for example, the anger may have multiple manifestations, including distinctive thoughts and feelings, facial expressions and posture, speech and tone of voice, physiology and neurochemistry, and perhaps behavior. What gives coherence to these occurrences is that all are representations of a single emotional meaning, in the case of anger a meaning involving themes of loss and blame. Emotions, then, are more specific than moods not only because what they represent is directed at specific objects, but also because they are particular in their meaning, a particularity that is represented in multiple ways that include distinctive and discriminable feelings. If moods are states of feeling, then emotions are states of feeling with specific objects and cognitive structures. The fact that emotions have both cognitive and experiential content means that when one is present, the other is often generated. For example, cognitions that one's goals have been thwarted by someone's blameworthy actions is likely to be accompanied by feelings of anger. Similarly, feelings of anger are likely to be accompanied by cognitions

368

Chapter 8

that there must be undesirable outcomes caused by someone's blameworthy action. If so, then inducing feelings of anger should elicit beliefs about blameworthy action, because feelings of anger provide information that bad things have happened and that someone is to blame. Experimental evidence confirms that feelings of anger do lead to increased attributions of human agency (Kcltncr, Ellsworth, & Edwards, 1993) and blame (Gallagher & Clore, 1985). Emotions-as-moavaaon One of the important things about emotions is that they are felt and that the feelings are pleasant or unpleasant. We have emphasized the information value of such feelings as feedback that guides evaluative judgments and strategic choices. But such feelings not only inform judgment, they also motivate action. Emotional feelings are pleasant and unpleasant and as such they serve as incentives and disincentives for action. In this way they are similar to bodily feelings like hunger. States of hunger and exhaustion feel unpleasant, and states of satiety and rest feel pleasant, and each have motivational properties. As indicated above, progress has been made in figuring out some of the eliciting conditions or inputs to the emotion system. Ortony et al. (1988) and many others have proposed comprehensive accounts of the situational elicitors of specific emotions. What has yet to be illuminated is the output of this system. Anger and fear differ not only in the situations that elicit them but also in the kinds of actions they sometimes cause. Is there order to the responses of emotions that complements the order we have found in their elicitors? Some psychologists have tried to systematize the consequences of emotion (e.g., Frijda, 1986; Plutchik, 1980), but no generally compelling structure has emerged. Perhaps the problem has resisted a solution because the right tools were not available. In that regard, the kind of analyses that evolutionary psychologists often undertake may be useful. As in explanations of other psychological phenomena (see Barkow, Cosmides, & Tooby, 1992; Cosmides & Tooby, 1987), it may be useful to think about the problems that particular emotions might have evolved to solve.

T. Ketelaar and G.L. Clore

369

Functionality and emotional motivation There are many facets to the functionality of emotion. Consider fear, for example. In the face of possible harm or loss, responding with fear is surely adaptive to the extent that it motivates escape or avoidance. This is not only true of threats to life and limb but also of more minor threats. Fear is adaptive to the extent that it makes one study before taking examinations, rehearse before going on stage, and make plans before embarking on complex ventures. Fear also helps keep one from leaning too far over balcony railings, closing one's eyes while driving, and doing innumerable other stupid things. Conversely, hope and optimism are important counterweights to fear. Taylor and Brown (1988), for example, discussed the adaptive role of unrealistic hope and optimism for leading one to get up each day and pursue one's plans and dreams. Fear created by credible threats plays an important role in restraining aggressive behavior that could otherwise seriously threaten solidarity within groups and peace between them. Many emotional expressions of animals apparently signal a willingness to react aggressively so that fear can inhibit action that might otherwise result in mutual injury or destruction. A more complex set of reactions can be seen in the relations between groups of humans. For example, Gould (1997) has studied vendetta violence on the island of Corsica during the 19th century. He suggests that in the absence of a strong civil government, the threat of group retaliation sometimes serves as an effective deterrent to violence between individuals who belong to families with a history of intergroup conflict (see also Hardin, 1995). Emotions as representations of long-term consequences of action An interesting feature of human society that befuddles many rationalists is the fact that, "a person can sometimes best further his self-interest by not intending to pursue it." In defending this conclusion, Hirshleifcr (1987) focuses on anger and gratitude. These are emotions that guarantee the execution of contracts and that serve as guarantors of threats and promises. Threats and promises differ from forecasts in that they involve forgoing immediate self-interest to engage in something the individual would not otherwise be motivated to do. Presumably the emotions of anger (in the case of threats) and gratitude (in the case of promises) are immediate experiences that provide the motivation to follow through.

370

Chapter 8

If emotions are designexl, in part, to operate as information processing devices we would expect that one can demonstrate that emotions not only affect judgments (proximate effects), but that they appear to be designed to do so (ultimate functions). In other words, if emotions serve some ultimate information processing function, we should observe that emotions do not simply disrupt or interrupt cognitive activity, they actively shape mental activity in an adaptive way. To support such a functional view of emotions, one must identify information processing problems that particular emotions can solve (not an easy task) and then describe how these emotions actually solve those problems. Before turning to an example of a specific information processing problem that is better solved when an individual is experiencing an emotion, than when they are not, we provide an analogy for the hypothesize~ process. There are a number of situations where imme~ate consequences differ substantially from future consequences (Gigerenzer, 1996). One view of the "rationality of emotions" emphasizes the fact that emotions make palpable and immediate the long-range effects of decision alternatives. Consider, as an analogy, a comparable effect regarding food. The Garcia effect. An individual may respond positively to a particular food on the basis of its immexliate taste, cvcn if it happens to be contaminated with a toxin that will later lead to illness. But in a process referred to as the "Garcia effect" (see Garcia, 1990), a single experience of illness can produce subsequent aversion to the food. Since this can occur even when the onset of illness occurs hours after ingestion, the effect is not easily accounted for by classical conditioning. The experience of distaste provides the individual with information about the long term costs of ingesting the food and with the motivation not to do so. When the organism encounters the food again, the past nausea, now moved to the beginning of the sequence signals that, in the past, this good tasting food made me sick. By virtue of moving the long term costs of ingesting the food into the present, in the form of an aversive feeling state, the individual now has a new unpleasant experience that serves as a cost against responding to a short range desire to eat the food. The process is analogous to the behavioral control processes that some see in emotion (Frank, 1988). Both provide a conscious output, in the form of a feeling state that conveys information about the long term consequences of a particular action, and that motivates one to engage in (or avoid engaging in) the action in question.

T. Ketelaar and G.L. Clore

371

Emotion-as-motivation and Frank's (1988) Commitment Model The economist Robert Frank (1988) makes a similar analysis of a functional role for emotions in what he refers to as commitment problems. Commitment problems arise when immediate incentives run contrary to one's long-term interests. Consider the following anecdote about a so-called "town fool" (adapted from Gigerenzer, 1996): The story goes that there once was a town fool. He was so foolish that whenever he was offered the choice between a pound and a shilling (1/20th of a pound) he would always choose the shilling. The townsfolk were quite amused. Repeatedly they would offer him the choice of a shilling and a pound, every time he would take the shilling. People traveled from miles around to witness this phenomenon. Day in and day out, the town fool always took the shilling, never the pound. This continued for years. He retired a very rich man. In this example, the immediate incentive of choosing the higher value coin competes with the more distal incentive of being offered this choice repeatedly in the future. The town fool anecdote nicely illustrates the idea that what is good for you in the long-run often conflicts with what is good for you in the short-run. When viewed as a single choice, the town fool's behavior appears irrational. Yet, when this same choice is repeated within a social context, where a particular choice determines the probability of getting to choose again, the so-called fool's behavior looks quite different (Gigerenzer, 1996). What constitutes a "rational" value-maximizing choice depends on whether the individual is committed to maximizing value over the long-run or in the short-run. By their very nature Commitment problems pit immediate rewards against long-term incentives (Frank, 1988). Commitment problems are numerous and are not limited to anecdotal descriptions of foolish behaviors. The dieter faces the immediate attraction of a piece of cake, weighed against the long-term cost of gaining weight. The married individual faces the immediate attraction of an extra-marital affair, balanced by the long-term threat to the stability of one's marriage. The diner in a restaurant faces the immediate benefit of not leaving a tip, weighed against the potential long-term damage to one's reputation of being labeled a stingy individual, or worse yet, a cheater. Each of these problems share a common structure. They all involve a dilemma where the choice that

372

Chapter 8

maximizes one's immediate self-interest is at odds with the choice that maximizes one's long-term self-interest. Frank (1988) refers to such situations as Commitment problems because the individual is often faced with the dilemma of committing to a strategy that conflicts with one's immediate self-interest. Frank's (1988) Commitment model illustrates how emotional reasoning helps solve the problem of overcoming the attraction of immediate rewards.

Immediate rewards are often more attractive than future rewards. There is considerable evidence that organisms in general, not just human decision-makers, are built to favor immediate rewards over more distal rewards. Thus, giving more weight to immediate outcomes, as compared to future outcomes, is "apparently part of the hard-wiring of most animal nervous systems" (Frank, 1988, p. 80). There would appear to be some adaptive logic to such a mechanism, given that surviving into the future is contingent on surviving today. In light of the considerable empirical support for such a reward mechanism 1, instances in which individuals forgo immediate rewards in favor of future rewards can be seen as exceptions to the norm that must be explainexi. In the real world, individuals do sometimes behave in a manner that goes against their immeAiate self-interests. People oitcn forgo a piece of cake in the name of diet, do not cheat on their spouses even when the probability of being caught is low, and invariably leave a tip for one's waiter even when visiting a strange town where there is little chance of being rcmembcrexi. Frank (1988) argues that emotional reasoning can account for these examples. He suggests that social-moral emotions such as guilt become activated in such circumstances and serve as commitments to one's long-term 1 The mechanism through which this focus on immediate rather distal rewards works appears to be a discounting process (see Frank, 1988, Hernstr 1970; Loewenstein, 1987). The typical c x ~ c n t a l demonstration of the attractiveness of immediate rewards involves presenting individuals with two choice situations. In situation A they are asked to choose between two rewards: (1) $100 to be received in 28 days or (2) $120 to be received in 31 days. In this situation, the vast majority of individuals choose the second option. Clearly $120 is worth more than $100 dollars. HoweveLwhen confronted with a slightly different scenario, one which shifts the delay in receiving the rewards toward the immediate present, their choices are strildngly different. In situation B individuals are asked to choose between two rewards: (1) $100 to be received today or (2) $120 to be received 3 days from now. In this new situation, the vast majority of individuals now choose the first reward. For any interest rate below 20%, the second reward will always be worth more than the first, yet the majority of individuals shift their preference in situation B to the first reward. Frank (1988) argues that the psychologicalprocess driving these choices is a reward mechanismwhich presents immediate rewards as "speciously"attractiverelative to more distal rewards.

T. Ketelaar and G.L. Clore

373

interests. When the structure of commitment problems is modeled gametheoretically, one can see the problem with the pursuit of immediate (as opposed to future) self-interest. Within such a game-theoretic framework, Frank illustrates the strategic role of emotions in shifting one's commitments away from immediate rewards toward more distal future rewards. The game theoretic structure of commitment problems. In the classic Prisoner's Dilemma game, there are two players, and each is faced with the same set of choices (cooperate or defect). In this game, the payoff matrix is symmetrical, one's payoff depends not just on what one chooses to do (cooperate or defect), but also on what the other player chooses. One's best possible payoff occurs when one chooses to defect and the other player chooses to cooperate (the cheater's payoff; see Axelrod, 1984). The worst possible payoff occurs when one chooses to cooperate and the other player chooses to defect (the sucker's payoff). When playing only a single iteration of the Prisoner's Dilemma game, one can maximize the immediate expected payoff by choosing the dominant strategy of defection. What makes the Prisoner's Dilemma game interesting is that if both players attempt to maximize their immediate rewards by enacting their dominant strategy (defect), they both end up worse off because mutual defection results in the second worse outcome in the payoff matrix for this game. However, if the game is repeated, the dominant strategy changes from the nasty (defect) strategy to a "nicer" tit-for-tat strategy that rewards cooperation and retaliates (tit-for-tat) against defection. In an extensive series of computer simulations, Axelrod (1984) and others have shown that in repeated Prisoner's Dilemma games, this very simple strategy outdoes most other strategies, including more complex strategies (see also Godfray, 1992). Apparently the tit-for-tat strategy works well precisely because it favors expected rewards in the long-run (gained through mutual cooperation) over short-term rewards. Pursuing short-term rewards through defecting invariably results in endless cycles of mutual defection over the long haul (Axelrod, 1984). Consistent with this logic, recent work in experimental economics (Hoffman, McCabe, Shachat & Smith, 1994; Hoffman, McCabe, & Smith, 1996) shows that individuals often cooperate more in social dilemmas than would be predicted by short-term self-interest models. To the extent that commitment problems provoke individuals to choose between strategies where the immediate payoffs differ from the long-run payoffs, then one might use the logic of game theory to justify how behavior that is contrary to one's immediate self-interest is, in the long-run, beneficial to one's material wellbeing. This leaves open the possibility that not all behavior stems from a

374

Chapter 8

single reward mechanism that emphasizes immediate material self-interest. Although game-theory can explain why it is often good to forgo immediate rewards, it does not deal with the psychological mechanisms that actually accomplish this commitment to long-term interests. This is where Frank's model of emotions as commitment devices enters the picture. Emotwns as commitment devices. According to Frank (1988), emotions provoke individuals to make binding commitments to behave in ways that run contrary to our immediate self-interests. Thus, emotional commitments that appear foolish and irrational in the short-run may be quite rational and adaptive over the long haul. Frank (1988, p. 82, emphasis in original) notes: The idea is that if the psychological reward mechanism is constrained to emphasize rewards in the present moment, the simplest counter to a specious reward from cheating is to have a current feeling that tugs in precisely the opposite direction. Guilt is just such a feeling. And because it coincides with the moment o f choice...it can negate the spurious attraction of the imminent material reward. The key "problem" of commitment problems centers around the fact that the psychological reward mechanism automatically produces a representation of one's circumstances that displays the rewards of cheating right now. The activation of the reward mechanism can be an attractive lure for behavior. According to the Commitment model, emotional feeling states such as guilt serve as competing information representing the long-term consequences of cheating (Frank, 1988). Because this competing information is in the form of a feeling state, it is experienced fight now. By virtue of moving the costs of cheating into the present, in the form of a feeling state, guilt coincides with the activation of the reward mechanism and the individual now has two pieces of information that can be taken into account in making a decision on how to behave. One informs the individual about the immediate consequences and the second about future consequences. We turn now to a test of the idea that conscious feelings can help solve an adaptive problem by providing information about the long term consequences of behavior. Guilt in the prisoner's dilemma game

Does the experience of guilty feelings influence the strategy adopted in the prisoner's dilemma game? Although the affect-as-information model

T. Ketelaar and G.L. Clore

375

(Schwarz & Clore, 1983) has been applied primarily to the influence of mood, the affective cues of specific emotions should also be informative. Presumably individuals who experience guilt when considering defection should be less likely to choose that option. By serving as a stand-in for the long-term consequences of defection, the unpleasant experience of guilt should foster commitment to a more adaptive strategy. There is no reason to assume that this unpleasantness would lower the attractiveness of all choices in the game, but only those that match the guilt schema. Although undifferentiated negative mood might have such effects, the particular experience of guilt should be harder to misattribute as a reaction to any and all objects of judgments. We would expect that the more particular the subjective experience of affect, the more narrow the range of judgment alternatives to which they are likely to be attributed. According to the Ortony et al. (1988) account, shame or guilt involve reactions of disapproval triggered by appraisal of one's own actions as blameworthy, an appraisal that depends on a perception that the actions violate important standards. Similarly, Higgins (1987) discusses guilt in relation to the violation of social norms. Thus, guilty feelings should serve as a source of information and motivation for decisions to obey (cooperate) or to violate (defect) social norms. The experience should inform individuals that defection has been appraised as a violation of their standards, and motivate them to avoid that alternative (because guilt is an unpleasant experience). For the guilt to have such effects requires that it be experienced as a reaction (i.e., attributed) to entertaining the option to defect (according to the affectas-information view). Therefore, on occasion, individuals experiencing guilt for other reasons (or persons experiencing chronic guilt), may misattribute that experience as a reaction to any action seen as self-benefiting. According to an affect-as-information perspective, guilty feelings produced by one source (say, writing a story) could be misattributed to a second source (deciding which strategy to play in the prisoner's dilemma). A misattribution study employing feelings of guilt in a prisoner's dilemma game could examine whether affective information can promote adaptive behavior in situations involving social dilemmas. If guilty feelings do not ordinarily serve such informational and motivational functions, we would expect them to have no effect on strategy choice in a prisoner's dilemma game. Such a model assumes, of course, that the individual is unaware of the source of their feelings, or that the source is somehow not salient (Schwarz & Clore, 1983, 1988). In cases where the source of the emotion is extremely salient, one expects that the individual might correct, or sometimes over

376

Chapter 8

compensate, for the influence of emotion. In such cases, one might expect that opposite effects of those predicted here, namely, that guilt inductions will trigger less cooperation because participants arc trying to compensate for a truly irrelevant bias towards cooperation. However, if guilty feelings arc normally used to inform choices in such social dilemmas and if research participants arc relatively unaware of the irrelevant (in this experiment) source of their feelings, then wc might expect participants induce~ to fccl guilty to show increased cooperation. Such findings would constitute indirect evidence that guilty feelings may naturally serve such functions. Empirical support. In a recent study (Kctelaar & Au, 1997), undergraduates that came to the laboratory in small groups wcrc informed that they would interact in pairs via computer in a decision-making task where they could each receive a small amount of money based upon their individual performance. In actuality, they played thc prisoner's dilemma game against a computer program that always employed the same (tit-for-tat) strategy. Prior to each trial, participants wcrc presented with the standard prisoner's dilemma payoff matrix and asked to select their response, A or B. Aitcr apparently waiting for the other player to respond, the computer displayed the outcome of their choices. This information was always presented in the form of "You received X dollars and the other person received Y dollars". They wcrc also reminded of the response they had selected (A or B) and informed of their partner's selection (A or B). The payoff matrix was then again displayed on the screen and the participant was again asked to select a response, and so on. Guilty mood inductions. After the first session of 40 trials, participants wcrc asked to engage in a second, ostensibly unrelated task involving writing a detailed description of an event. Participants wcrc randomly assigned to one of two conditions - either writing a detailed description of a recent experience when they felt really guilty, ashamed, or self-blaming, or simply writing about a typical day. Alter writing for l0 minutes, they played a second set of prisoner's dilemma trials. Provoking cooperation or defection. During the second set, half of the participants in each mood condition wcrc randomly assigned to a nice or nasty computer partner. The nice computer partner was programmed to begin the second session by selecting the cooperate response on the first five trials regardless of what the other person responded. The nasty partner began by selecting the defect response on the first five trials. In both cases, the computer program returned to the simple tit-for-tat strategy after these first five trials.

T. Ketelaar and G.L. Clore

377

This manipulation of nice vs. nasty partners during the second set of trials was done in part to provoke each participant in a manner that could tease apart the role of guilt in accentuating or diminishing propensities to cooperate or defect when their partner displayed cooperative or uncooperative behaviors. Measuring social motives. Because previous work on cooperation in social dilemma's has stressed the importance of individual differences in response strategies - referred to as social value orientation (see Van Lange & Kuhlman, 1994; Messick & McClintock, 1968), - the effects of guilt were assessed separately for cooperative and uncooperative individuals. These individual differences in social motives were taken into account by comparing those individuals who cooperated more than 50% of the time on the first set of 40 trials (roughly half of the participants fell into this category) and those who cooperated less than 50% of the time on the first set of trials. We expected that social motives might interact with guilt. Emotions are ot~en evoked in situations where one's own strategy, say defection, conflicts with one's partner's strategy, say cooperation (see Nesse, 1990), and emotions can provide valuable information about the meaning of the situation. Thus, one expects that guilt reductions might show their greatest effects in conditions where the subjects' social motives (cooperative or uncooperative) conflict with the computer's strategy (nice or nasty). Guilt increases levels of cooperation. Significant results showed that emotion (guilt vs. neutral mood), social motives (cooperative vs. uncooperative) and computer strategy (nice vs. nasty) all interacted to influence cooperation (see Table 1). Guilt had little or no effect on the choices made in games in which the computer's style matched the player's inclinations; the mean differences between the guilty and neutral subjects on the cooperative-cooperative or noncooperative-noncooperative pairs was 0%. But when there was a mismatch of styles (cooperative vs. uncooperative or uncooperative vs. cooperative), the presence of guilt feelings was telling (mean guilt-neutral difference = 23%). Hence, guilt kept uncooperative players from taking advantage of the cooperative computer and kept cooperative players from retaliating against the uncooperative computer. Implications. These results show that the effects of guilt on cooperative behavior vary with the social motives of the participant and the behavior of their computer partner. They are congruent with both Frank's commitment model of emotional reasoning and with a functional interpretation of Schwarz and Clore's (1983) affect-as-information model. Consistent with Frank (1988), guilty feelings biased choices in a repeated prisoner's dilemma game

378

Chapter 8

Table 1. Percentages of cooperative responses in a prisoner's dilemma game. Strategyof ComputerPartner Cooperative (Nice)

Uncooperative (Nasty)

Guilty Mood

Neutral Mood

Guilty Mood

Neutral Mood

Uncooperative

56

30

44

35

Cooperative

90

1O0

69

49

Social Motives of the Player:

by provoking a cooperative strategy. Consistent with a functional interpretation of the affect-as-information model, they show both proximate effects and perhaps the ultimate function of an emotion. Specifically, guilty feelings biased strategy choices (a proximate effect) leading to an adaptive solution to a version of the commitment problem (the hypothesized ultimate function). In general, if guilty feelings provide conscious access to the long-term consequences of pursuing an otherwise attractive short-term strategy, individuals not experiencing such feeling should have less immediate and vivid access to this information. Individuals high on measures of psychopathy arc apparently less likely to generate such experiences (Williamson, Harpur, & Hare, 1991). Even when generated, they may be less persuasive with individuals who have addictions or cravings that make a particular short-term reward have especially high incentive value. Such incentives may either successfully compete with guilt or so dominate the attention of the individual that they lower the likelihood of experiencing guilt. In this regard, Patterson and Newman (1993) have shown that when entrained by pursuit of an incentive, organisms appear to have a reduced sensitivity to punishment cues. Affect-as-Information and Behavior

The notion that affect provides information is an apt explanation of the effects of mood and emotion on judgment, because judgments arc presumed to bc based on information. Moreover, we know from social psychological

T. Ketelaar and G.L. Clore

379

research that persuasion and attitude change is more successful when the information comes from a highly credible source, and no source is more credible for most people than their own feelings. Indeed, all the logic in the world will not convince one of an argument unless one feels convinced. Otten, feeling is believing, which may explain the robustness of mood effects on judgment. When moods are attributed as reactions to an object, the feelings may be experienced as liking or disliking, and this reaction may influence evaluative judgments of the object. Emotions, in contrast to moods, are often stronger and always convey something more than a general positive or negative reaction. Guilty feelings, experienced as one entertains a given choice alternative, not only provide information about the distastefulness of making that choice. But, in addition, because guilty feelings are also experienced (perhaps viscerally; see Damasio, 1995) as aversive and unpleasant, they may lead one to avoid that choice regardless of one's judgment about it or one's knowledge that it is distasteful. It seems plausible to suppose that the system evolved both cognitive and visceral routes to encourage adaptive choices. Thus emotions may hit us both high and low, so that whether we think about it or simply go with our inclinations, consideration of long-term outcomes is encouraged. Much has been written about two process models of cognitive functioning (for a good review see Smith, 1994), and perhaps this is another example. Feelings provide highly credible information about the appealingness of choice alternatives, but in addition to this high road to decision-making in which affective information is included in judgments and deliberations, affect also takes the low road, serving directly as a pleasant incentive or an aversive punishment making one approach or avoid certain choices. This alternative route leads one to continue on the path toward virtue just as one continues eating tasty food regardless of whether one has cognitively decided to do so, because it feels good. Thus, when considering the role of affect in behavior, it may be useful to expand the informational view to include motivation. Unless contradicted by strong ideology, we tend to believe that things that feel good are good (affect as information), but in addition, we want things that feel good (affect as motivation). Thus, as Frank (1988) suggests, the unpleasantness and distress of being angry is a cost against not doing something about it (reacting only to immediate incentives). The way to turn off this unpleasantness is to seek retribution or restitution and to right the wrong.

380

Chapter 8

Anger and punitiveness A common, but limited, model of personality and of emotion is one that explains external behavior on the basis of some more or less identical internal entity, as when hostile action is said to be a manifestation of a hostile trait, instinct, or propensity. While such assertions are not without value, and can be defended philosophically (Hirschberg, 1975), much human behavior is not well explained by simple inner forces that simply turn up the volume on a particular class of behavior. Such a model does work in the case of emotional expression because happy and sad expressions can generally be taken as outward reflections of some inner happy and sad affect. But some consequences of emotion are more complex and contingent. Hirshleifer (1987), for example, has distinguished action independent affective inclinations that he refers to as "malevolence" and "benevolence" from action dependent emotions such as anger and gratitude. He suggests that the contingent nature of anger and gratitude give them power to bring the behavior of others into line, a power that noncontingent affective inclinations such as malevolence and benevolence do not have. Like moods and temperaments, affective inclinations such as malevolence and benevolence should exert their influence through main effects rather than through interactions with the specific situational factors as we might expect for anger and gratitude. This point is also evident in a recent study by Goldberg, Lerner, and Tetlock (1997), which examined the conditions under which anger would influence judgments of appropriate punishment. They reasoned that punishment would be seen as appropriate only when both anger and the perception of injustice were present. Participants saw a video of a bully beating up a teenager and were given one of three level of justice feedback. They were told either that the bully was caught and appropriately punished, that he was caught but left unpunished due to a technicality, or were given no feedback about punishment. They then read four vignettes depicting acts of negligence, recklessness, or intentional harm. Finally, they completed a series of questions on their perception of responsibility, blame, and punishment. The results showed that knowledge of punishment did not affect the anger felt by participants. But in combination with knowledge of the fate of the bully in the anger-inducing film, anger did predict inclination to punish the wrong-doers described in the later vignettes. That is, anger was correlated with punitiveness only when the bully in the original induction video had gone

T. Ketelaar and G.L. Clore

381

unpunished. When the wrongdoer had been punished, they were equally enraged, but there was no relationship between anger and punitiveness. Some of the vignettes involved intentional action and some simply negligence or recklessness. Knowledge that the original bully had been punished limited subsequent punitiveness in response to the vignettes in which intentional harm was involved. However, when participants believed that justice had failed and the bully in the original anger induction film had gone unpunished, they were more indiscriminately punitive regardless of apparent intentionality of harm in the vignettes about which they made judgments. The authors conclude that emotion plays a central role in justice judgments. They suggest that norm violations resulting in harm elicit anger, which triggers a more general impulse toward punitiveness only when a schema for injustice (nonpunishment of a harm doer) is also aroused. Anger appears to make the cognitive construct of injustice salient and to predispose one to identify and punish wrongdoers. Goldberg et al. suggest that beliefs that justice had been served because the wrong-doer was punished may have created a kind of cognitive closure that deactivated the construct so that it no longer influence~ how subjects judged subsequent situations. If the goal of anger is to reduce the continuing threat of harm from another, then punishment, restitution of the loss, or evidence of sorrow, regret, or contrition on the part of the responsible agent might be expected to turn off anger or reduce its intensity and duration. We assume that the emotion serves as information about the nature and importance of the perceived injustice, that it is not only an injustice, but one that the subject cares about. Presumably the idea of punishment (or the inclination to punish) elicited by the video was deactivated (or satisfied) by knowledge that the bully had been punished, but was intensified by knowledge that he had escaped justice. In the latter case, the idea or inclination was apparently still active and perceived to be relevant when judging the subsequent vignettes. So a tendency to engage in aggressive or punitive behavior may not be automatically triggered in anger. Rather punitiveness is elicited when one both feels angry and explicitly learns that principles of justice are not being upheld. More indiscriminate scapegoating was found to be predicted, not by anger, but by less specific feelings of distress, especially when accompanied by relevant social motives. In our view, distress may operate more like a mood and less like a specific emotion in that it has less cognitive structure than anger. As a result it may be relatively more susceptible to misattribution. Thus, subjects who based their judgments of punishment on their general

382

Chapter 8

distress were more indiscriminant than those basing their judgments on specific anger. Summary. The experiments on guilt and anger and Hirshleifer's (1987) arguments about anger and gratitude, are based on the assumption that emotional experiences represent long-term outcomes. The feelings of anger, sadness, fear and so on are all distinctive. They represent experientially the deep structure of each emotion, a structure of the kind that Ortony et al. (1988) and other appraisal theorists have attempted to capture prepositionally. We suggest that what turns an emotion on and off are perceptions of a situation that match the features of the deep structure of the emotion, features reflecting the affective significance of the situation.

Love and perceptual satisficing in the mate choice problem Optimal vs. satisficing strategies of mate choice. Imagine the ultimate computer dating service. It might involve a complete, up-to-date file of all of your potential long-term (and short-term) mating partners. It would describe in luxurious detail all relevant cues for selecting a successful long-term partner, including assessments of the probability that any given partner would eventually accept your offer of a date (or marriage). The mate choice problem would be reduce~ to an electronic data-base search, a bit like searching for a rare volume in some vast, electronic library. After you had constructed a list of your quality criteria, all that would be required to find your optimal mate would be a lot of time to search through your data base, identifying and integrating cues until you arrived at your best possible choice. For many species, however, the criteria for choosing a mate is much more fast and frugal. Many species, such as termites, simply mate with the next fertile appropriately sexed individual that comes along (see Wilson, 1971). As many entomologists and a few home owners know, this mating strategy has historically been quite successful, despite its simplicity. For humans, mate choice is, of course, a bit different. Because human males and females invest a relatively large amount of resources in raising offspring to reproductive maturity, it is not surprising to observe that humans are relatively choosy in selecting candidates for a long-term mate. Humans rarely choose as their "one and only" the next fertile individual of the opposite sex that comes along. But where do human long-term mating strategies lie on the dimension ranging from termites to our ultimate computer dating service? One view would suggest that emotions, such as romantic love, actively shape mate choice decisions so that they are better characterized as fast and frugal

T. Ketelaar and G.L. Clore

383

satisfieing strategies, rather than absolute optimal information-search strategies 2 . While it may be the case that human decision makers are neither optimal nor capricious (remember termites) in their mating decisions, an evolutionary perspective suggests that mate choice algorithms are likely to fall closer to the fast and frugal end of the satisficing vs. optimizing dimension of possible strategies. One can view the absolute optimal solution to the mate choice (and subsequent mate investment) problem as involving a thorough survey and comparison of all attributes of every possible mate, analogous to our ultimate computer dating service. Using this optimizing strategy, an individual selects their single "best" mate only after considering and weighing all possibilities. The opportunity costs associated with adopting such a strategy may be so large, however, that much simpler strategies, which merely search for a mate that is very good, rather than the absolute best choice, will always outeompete them (Todd, 1997). According to this view, individuals who make a simple satisficing mate choice may be at a selective advantage, evolutionarily speaking, over other individuals who attempt to make a single best choice by evaluating all possible mates. The basic logic of this argument is that the costs of mate search are born out mainly in terms of opportunity costs. The more time an individual devotes to mate choice, the less time that same individual can devote to investment in a particular mate and to any offspring that might result from this partnership. Romantic love as a perceptual satisficing mechanism. A mechanism that limits the search space of possible cues (and mates) and provokes an individual to make a fast and frugal mating decision based upon a few, rather than all, relevant pieces of information might prove quite effective in many mating decisions 3. We refer to a mechanism that could in principle solve such a problem as a "perceptual satisficing" mechanism. Perceptual satisficing is defined as the process by which a perceiver relies upon a few good cues rather than all possible cues to form a preference. One mechanism by which 2 We do not wish to imply that all mate choices are somehow malaadaptive or sub-optimal.

Rather, we wish to explore the possibility that the mate choice problem, for a given individual, may not have a single best solution, or perhaps, if one exists, it would be computationally expensive to determine this solution. Thus, we emphasize that psychological adaptations, like other evolved mechanisms, are not necessarily the absolute optimal design for solving a particular problem. This is the case because of the design constraints (costs) imposed by other design features of the organism (see Dawkins, 1982). 3 The ideas developed here were part of a research project conducted at the Center for Adaptive Behavior and Cognition under the supervision of Gerd Gigerenzer and benefited from numerous critical feedback from Jennifer Davis (see Ketelaar & Davis, 1997).

384

Chapter 8

perceptual satisficing could be achieved is for the perceiver to prioritize cues such that: (a) the best cues get entered into the decision algorithm first and (b) the decision-process stops when a good enough choice has been made. Herbert Simon (1956, 1987) referred to these types of decision algorithms as "satisficing" procedures, because they were a mix of optimizing and a satisfying procedures. Perhaps emotions such as romantic love play a role in perceptual satisfieing by shaping both the selection and ordering of cues that one inputs into mate choice algorithms. The cue relevance problem. Mate choice is a decision-problem that, while distinct from other choice problems (food choice, habitat choice, etc.) shares a problem faeexl by all domain-specific modules: the problem of information selection and prioritization. Any domain specific psychological mechanism has, by definition, a restricted range of inputs, a delimited set of stimuli which activate the module (see Buss, 1991; Sperber, 1994). Which cues are relevant to the mate choice problem? In any given mating decision, some cues are more relevant than others. When searching for a long-term mate, some information, like the person's eye color, may be less relevant than other information, like whether the individual is willing and able to commit.to a long-term partnership (Buss, 1989). Moreover, the validity of person characteristics such as "committed" or "coy" depends on whether the judgment concerns a long-term relationship or a short-term, one night stand (Buss & Sehmitt, 1993). Having identified a particular set of cues as relevant to a particular adaptive problem (i.e., choosing a long-term mate), individuals must then prioritize them according to their relevance for the specific problem at hand. This problem of cue selection and prioritization can be referred to as the cue relevance problem. Why is cue selection and prioritization (cue ordering) so important? In situations where targets (e.g, mates) are evaluated sequentially using a lexicographie decision proe~ure, the ordering of cues inputted into the algorithm is essential; if the cues are inputted in the rank-order of their validities, a simple satisfieing algorithm can often perform just as well as a more complex procedure (such as multiple regression), but use considerably fewer cues (Gigerenzer & Goldstein, 1996). It stands to reason that a simple satisficing algorithm that performs well using fewer cues is more efficient than an algorithm which requires more information (cues) to achieve the same level of performance. If mating decisions are made using a lexieographie algorithm such as those identified by Gigerenzer and others (1996), then cues are inputted into the decision algorithm in order of their perceived validities. There are several

T. Ketelaar and G.L. Clore

385

ways of conceptualizing cuc validities, the ecological validity of a cuc, including the Brunswikian (1955) correlational/regression method and the Gigcrcnzcr and Goldstcin (1996) proportion of correct inferences method. For the case of presentation wc will consider the first method, the Brunswikian ( 1955) method. The cue relevance model o f emotion. The cuc relevance model takes as its starting point, the Brunswik (1955) lens model in which stimulus attributes arc differentially weighted on the basis of their validities in a particular decision context. According to the Brunswikian (1955) method, the ecological validity of a cuc is defined as the correlation between the cuc value and the target variable. The validities of the cues used to make a judgment or decision depend, however, on the problem context. For example, to a hungry individual, the height of a fruit trcc might bc seen as a negative cuc, while to a frightened individual, the height of the trcc might bc positive because it affords safety and concealment from predators. In other words, the validities of particular cues for making particular judgments should depend on the context. Romantic love might function as one such context, shifting an individual's prioritization of cues to make them congruent with the adaptive problem of long-term (rather than short-term) mate choice, when it is perceived to bc pertinent. Because this process of cuc prioritization makes certain cues overwhelmingly salient and others vanishingly insignificant, it has the effect of temporarily reducing the search space of cues that one considers in making a particular decision. In this way, romantic love can act as a perceptual satisficing mechanism by reducing the number of cues rcquirext to make an effective mating decision. Because satisficing strategies rely on relatively few cues (rather than the entire set), emotional affects on the selection and prioritization of cues could have a large impact on subsequent decision-making. This effect might determine whether a given cuc enters earlier or later in the decision process. For example, borrowing from Frank's (1988) model, one might expect that feelings of romantic love put a heightened focus on perception of the longterm (as opposed to short-term) benefits of a partner's attributes. Perhaps individuals who arc "in love" consider cues to long-term benefits before, or instead of, using cues to short-term, immediate benefits when assessing their romantic partner. This view makes some intuitive sense. Have you cvcr wondered why one's perceptions of an individual arc often different early in a relationship when one is madly in love with their partner as comparexl to months or years later? There would appear to bc strong shifts in evaluations of a partner that

386

Chapter 8

covary with changes in emotional responses toward an individual as a function of changes in the assessment of important cues. How can these perceptual changes bc explained? The cuc relevance model suggests that affcctivc states, such as romantic love, produce strong perceptual biases in how wc attend to and prioritizc cues in our environment. As an individual comes to possess a greater (perhaps more accurate7) knowledge of their partner's attributes, assessments of the long term benefits of those attributes can also change. To the dcgre,c that feelings of romantic love covary with attention to and prioritization of cues to long-term (rather than short-term) mate value, one might expect that feelings of romantic love wax and wane as the assessments of these cues are up-date~ and modified over time. In specific emotional states, individuals cannot help but attend to and prioritizc cues in terms of their relevance to the perceived adaptive context. Perhaps the individual who is "in love" is compelled to assess cues in terms of their perceived validities relative to the problem of obtaining a long-term mate rather than some other adaptive problem. In the absence of romantic love, the individual might attend to: (1) pre-existing cues from the most recent problem confronted, (2) use a random sdcction of cues, or (3) use cues associated with an alternative emotional state (e.g., lust). If the cuc relevance model of emotion is correct, varying intensities of romantic love should result in different selections of cues in a mate choice decision. As a romantic relationship progresses, or digresses, wc would expect that selection of cues (but not the actual cuc values) shifts over time. This implies that knowing an individuals' emotional state will allow us to predict which cues are used in subsexlucnt judgments of romantic partners. This knowledge could help in predicting mate choice decisions. Summary. In a particular adaptive context, when a specific emotion is triggered, some features of the context will bc seen as more relevant than others. That is, emotions may function to shift perceptions of cues so that cues arc rank-ordered in terms of their relevance to that spccific adaptive problem. A fast and frugal strategy for decision-making would bc to input a limited set of relevant cues, rather than all cues, into a simple lcxicographic algorithm that takes the most relevant cues first and stops when it has just enough information to make a good decision. From this evolutionary psychological perspective, context-dependent rank-ordering of cues associated with the emotion of romantic love could serve the specific adaptive function of"pcrccpmal satisficing" in the mate choice domain.

T. Ketelaar and G.L. Clore

387

The Future of Affect and Information Processing We have suggested that research has begun to move beyond the investigation of positive and negative mood states to explore the informational properties of specific emotions. A working assumption of this emotion-as-information research is that emotions involve distinctive cognitive and experiential representations of a situation. These outputs of emotional appraisals are available to individuals in the form of particular feeling states for use in subsequent information processing. Implicit in this view, is the assumption that such emotional information is functional, that it aids individuals in solving particular kinds of problems. Thus, in the situation in which one is cheated by another, the appraisal of blameworthiness produces a conscious experience of anger/reproach. The conscious feeling state of anger which accompanies such appraisals serves to alert the individual to, and to bias attention toward, aspects of the situation relevant to this emotion. If this emotion-propelled search produces supporting information, then the unpleasant set of feelings that characterize anger is maintained until new appraisals turn them off. An important direction for research on anger (and other emotions) is to determine what appraisals constitute the off-switch for each emotion type. It is assumed that such emotions evolved to help solve recurrent but specific problems of survival that occur and reoccur across species, individuals, and situations. Against this assumption that affect aids rather than impairs information processing, are data showing that emotions elicited in one situation can sometimes be carried over to other, irrelevant situations. Such data are important because they allow us to see the role played by subjective experience independently of cognitive content, but they also highlight a shortcoming of the emotional feedback system. They show that the subjective experience of anger in response to the wrong-doing of one individual can trigger blameworthy judgments about the behavior of an unrelated individual. This kind of misattribution of anger, when visited on powerless others, amounts to scapegoating, an unhealthy way of resolving conflict at the expense of others. This pattern can either be conscious and intentional, as when a judge punishes an offender to the limit of the law as a lesson to others, or it may be unconscious and automatic, as in the F reudian defense mechanisms of displacement. The consequences of the latter were perhaps best described in the classic work on the Authoritarian Personality (Adomo, Frenkel-Brunswik, Levinson, & Sanford, 1950). Attempting to uncover the psychodynamic roots of fascism, Adorno and his colleagues described how an

388

Chapter 8

over-concern with order and control led authoritarian individuals to react to the frustrations of everyday disorder, and to their own repressed anger toward authority, with anger and hostility toward those with less power. Although they used essentially Frcudian concepts in their analysis, much of the emotional process of such scapcgoating conforms also to the misattribution logic of our experiments. Thus specific emotions are believed to provide information relevant to particular environmental challenges. Though functional in those settings, such evolved mechanisms can also exert a disruptive influence. The difficult task of identifying the proper situational domains of each emotion remains for future research. This chapter has summarized some potentially useful beginnings in that direction.

Conclusion: Deficits, Biases, and Functions A famous neurologist once wrote that "Neurology's favorite word is 'deficit'" (Sacks, 1985). Inde~ much of our current understanding of functional brain anatomy is derived from early work that studied brain functions indirectly by studying the dysfunctions and deficits brought about by disease or injury (Luria, 1966, 1973). This history reveals that studying deficits can be a useful starting place for explorations of function, but it is also clear that the study of brain damage is not isomorphic with the study of normal brain functioning. Fortunately, due to rapid advances in new techniques such as magnetic resonance imaging (MRI), traditional studies of deficit are being supplemented with images of activation in normal brains during cognitive and emotional processing (Davidson, 1992, Davidson & Sutton, 1995, Sutton & Davidson, 1997). If "deficit" has long been ncurology's favorite word, "bias" has been the favorite word in the study of judgment and decision-making, in which there has been a focus on deficits and illusions of reasoning (Kahncman, Slovic, & Tvcrsky, 1982; Piattclli-Palmarini, 1994). As such, it is not surprising that the earliest attempts to integrate affcctivc science and cognitive science have focused on the "biasing" role of emotion in information-processing. In this chapter we argued that one might be suspicious of a view that claims that the affcctivc contribution to mental activity is best described as terms of various biases and deficits in rational thought. We pointed out that this need not be the case. By distinguishing between the proximate and ultimate functions of emotions, we tried to iUustratc how research paradigms focused on biases could also be used to explore function.

T. Ketelaar and G.L. Clore

389

References

Adomo, T. W., Fronkol-Brunswik, E., Levinson, D., & Sanford, N. (1950). The authoritarian personality. New York: Harper. Adaval, R. (1996). Information processing, mood and reactions to the price of consumer goods. Unpublished doctoral dissertation, University of Illinois.

Ames, A. (1951). Visual perception and the rotating trapezoidal window. Psychological Monographs, 65, 234. Arkes, H. R., & Hammond, K. R. (Eds.) (1986). Judgment and decision malang: An interdisciplinary reader. New York: Cambridge University Press, 1986. Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books. Barkow, J. H., Cosmides, L. & Tooby, J. (1992). The adapted mind. New York: Oxford University Press. Bless, H., Clorr G. L., Golisano, V., Rabol, C., & Schwarz, N. (1996). Mood and the use of scripts: Do happy moods really make people mindless? Journal of Personality and Social Psychology, 71, 665-678. Bodenhausen, G., Sheppard, L. A., & Kramer, G. P. (1994). Negative affect and social judgment: The differential impact of anger and sadness. European Journal of Social Psychology, ~24, 45-62. Bowlby, J. (1969). Attachment and Loss. Vol. I." Attachment. New York: Basic Books. Bower, G. H., Monteiro, K. P., & Gilligan, S. G. (1978). Emotional mood as a context of learning and recall. Journal of Verbal Learning and Verbal Behavior, 17, 573-585. Buss, D. M. (1989). Sex differences in human mate preferences: Evolutionary hypotheses tested in 37 cultures. Behavioral and Brain Sciences, 12, 1-49. Buss, D. M. (1991). Evolutionary personality psychology. Annual Review of Psychology, 45, 459-491. Buss, D. M. (1995). Evolutionary psychology: A new paradigm for psychological science. Psychological Inquiry, 6, 1-30 Clore, G. L. (1992). Cognitive Phenomenology: The role of feelings in the construction of social judgment. In A. Tesser & L. L. Martin (Eds.), The construction of social judgments. Hillsdale, NJ: Erlbaum. Clore, G. L., Schwarz, N., & Conway, M. (1994). Affective causes and consequences of social information processing. In R. S. Wyer & T. Srull

390

Chapter 8

(Eds.), The handbook of social cognition (2nd ed.). Hillsdale, NJ: Erlbaum. Cosmides, L., & Tooby, J. (1987). From evolution to behavior: Evolutionary psychology as the missing link. In J. Dupr6 (Ed.), The latest on the best: Essays on evolution and optimality. Boston: Bradford Books-MIT Press. Damasio, A. R. (1994). Descartes' error: Emotion, reason and the human brain. New York: Grosset/Putnam. Davidson, R. (1992) Prolegomenon to the structure of emotions: Gleanings from neuropsychology, Cognition and Emotion, 6, 245-268. Davidson, R. J., & Sutton, S. K. (1995). Affective neuroscience: The emergence of a discipline. Current Opinion in Neurobiology, 5, 217224. Dawes, R. M. (1997) Behavioral decision making and judgment. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), The handbook of social psychology. Boston: McGraw-Hill. Dawkins, R. D. (1982). The extended phenotype. New York: W. W. Norton & Company. Dawkins, R. D. (1986). The blind watchmaker. Oxford: Oxford University Press. Ekman, P. (Ed.) (1982). Emotion in the human face. New York: Cambridge University Press. Ellis, H. C., & Ashbrook, P. W. (1988). Resource allocation model of the effects of depressed mood states on memory. In K. Fiedler & J. Forgas (Eds.), Affect, cognition, and social behavior. Toronto: C. J. Hogrefe. Ellis, H. C., & Thomas, R. L., Rodriguez, I. A. (1984). Emotional mood states and memory: Elaborative encoding, semantic processing, and cognitive effort. Journal of Experimenal Psychology: Learning, Memory, and Cognition, 10, 470-482. Elster, J. (1995). Rationality and the emotions. The Economic Journal 116, 1386-1397. Forgas, J. P., & Moylan, S. (1987). After the movies: The effects of mood on social judgments. Personality and Social Psychology Bulletin, 13, 467477. Frank, R. H. (1988). Passions with reason: The strategic role of the emotions. New York: W. W. Norton. Frijda, N. H. (1986). The emotions. New York: Cambridge University Press.

T. Ketelaar and G.L. Clore

391

Gallagher, D., & Clore, G. L. (1985). Effects of fear and anger on judgments of risk and blame. Paper presented at the meeting of the Midwestem Psychological Association, Chicago, May 1985. Garcia, J. (1990). Learning without memory. Journal of Cognitive Neurocience, 2, 287-305. Gazzaniga, M. S. (1992). Nature's mind: The biological roots of thinking, emotions, sexuality, language, and intelligence. New York: Basic Books. Gigerenzer, G. (1996). Rationality: Why social context matters. In P. Baltes & U. M. Staudinger (Eds.), Interacave minds: Life-span perspectives on the social foundatmn of cognition. Cambridge: Cambridge University Press. Gigerenzer, G,. & Goldstein, D. G. (1996). Reasoning the fast and frugal way: Models of bounded rationality. Psychological Review, 103, 650669. Godfray, H. C. (1992). The evolution of forgiveness. Nature, 355, 206-207. Goldberg, J., Lemer, J., & Tetlock, P. (1997). Anger in the face of injusace: The psychology of the intuitive prosecutor. Unpublished manuscript, University of California at Berkeley. Gould, R. V. (1997). Individual and collective conflict in an honor society. Paper presented at the Center for the Advanced Study of the Behavioral Sciences, Standford, CA, February 1997. Gray, J. A. (1971). The psychology of fear and stress. London: Weidenfeld & Nicholson. Harr6, R. (1986). An outline of the social constructionist viewpoint. In R. Harr6 (Ed.), The social construction of emotions. Oxford: Basil Blackwell. Hernstein, R. (1970). On the law of effect. Journal of Experimental Analysis of Behavior, 13, 242-266. Higgins, (1987). Self-discrepancy theory: A theory relating self and affect. Psychological Review, 94, 319-340. Hinde, R. A. (1970). Animal behaviour: A synthesis of ethology and comparative psychology. Kogakusha: McGraw-Hill. Hirschberg, N. (1978). A correct treatment of traits. In H. London (Ed.), Personality: A new look at metatheories. New York: Wiley. Hirshleifer, J. (1987). On the emotions as guarantors of threats and promises. In J. Dupr6 (Ed.), The latest on the best: Essays on evolution and optimality. Boston: Bradford Books-MIT Press.

392

Chapter 8

Hoffman, E., McCabe, K., Shachat, K., & Smith, V. (1994). Preferences, property rights, and anonymity in bargaining games. Games and Economic Behavior, 7, 346-380. Ho~, E., MeCabe, K., & Smith, V. (1995). Social distance and otherregarding behavior in dictator games. American Economic Review, 86, 653-660. Isbell, L., Clore, G. L., & Wyer, R. S. (1997). Mood-mediated uses of stereotypes in impression formation. Manuscript submitted for publication. Isen, A. M., Daubman, K. A., & Nowicki, G. P. (1987). Positive affect facilitates creative problem solving. Journal of Personality and Social Psychology; 52, 1122-1131. Isen, A. M. (1987). Positive affect, cognitive processes, and social behavior. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 20). New York: Academic Press. Isen, A.M., Shalker, T., Clark, M., & Karp, L. (1978). Affect, accessibility, of material in memory, and behavior: a cognitive loop? Journal of Personality and Social Psychology, 36, 1-12. Kahneman, D., Slovic, P., & Tversky, A. (1982). Judgment under uncertainty: Heuristics and biases. New York: Cambridge. Kaplan, M., Kiekul, J., & Reither, A. (1996). Mood and extent of processing plot and visual information about movies. Paper presented at Midwestem Psychological Association, Chicago, May 1996. Keltner, D., Ellsworth, P., & Edwards, K. (1993). Beyond simple pessimism: Effects of sadness and anger on social perception. Journal of Personality and Social Psychology, 64, 740-752. Keltner, D., Locke, K. D., & Audrain, P. C. (1993). The influence of attributions on the relevance of negative feelings to satisfaction. Personality and Social Psychology Bulletin, 19, 21-30. Ketelaar, T., & Au, W. T. (1997). The role of guilty feelings in cooperaave behavior in the prisoner's dilemma: An emotion-as-information interpretation of Frank's (1988) Commitment model. Unpublished manuscript, Max Planck Institute, Munich, Germany. Ketelaar, T., & Davis, J. N. (1997). A general framework for understanding the role of romanac love in mate choice decisions. Unpublished manuscript, Max Planck Institute, Munich, Germany. Lang, P. J. (1995). The emotion probe: Studies of motivation and attention. American Psychologist, 50, 372-385.

T. Ketelaar and G.L. Clore

393

Lazarus, R. S. (1991). Emotion and adaptation. New York: Oxford University Press. Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal and coping. New York: Springer. LeDoux, J. E. (1996). The emotional brain. New York: Simon and Schuster. Loewenstein, G. (1987). Anticipation and the valuation of delayed consumption. Economic Journal. Luria, A. R. (1966). Human brain and psychological processes. New York: Basic Books. Luria, A. R. (1973). The working brain. New York: Basic Books. Mackie, D. M., & Worth, L. T. (1989). Processing deficits and the mediation of positive affect in persuasion. Journal of Personality and Social Psychology, 57, 27-40. Marcus, G. E., & Mackuen, M. B. (1993). Anxiety, enthusiasm, and the vote: The emotional underpinnings of learning and involvement during presidential campaigns. American Political Science Review, 87, 672685. Melton, R.J. (1995). The role of positive affect in syllogism performance. Personality and Social Psychology Bulletin, 21, 788-794. Messick, D. M., & McClintock, C. G. (1968). Motivational basis of choice in experimental games. Journal of Experimental Social Psychology, 4, 1-25.

Nesse, R. (1990). Evolutionary explanations of emotions. Human Nature, 1, 261-289. Nesse, R., & Williams, G. (1995). Why we get sick: The new science of Darwinian medicine. New York: Times Books. Ortony, A., Clore, G. L., & Collins, A. (1988). The cognitive structure of emoaons. New York: Cambridge University Press. Patterson, C. M., & Newman, J. P. (1993). Reflectivity and learning from aversive events: Toward a psychological mechanism for the syndromes of disinhibition. Psychological Review, 100, 716-736. Piatelli-Palminrini, M. (1994). Inevitable illusions: How mistakes of reason rule our minds. New York: John Wiley & Sons. Platt, J. J., & Spivack, G. (1974). Means of solving real-life problems: I. Psychiatric patients vs. controls and cross-cultural comparisons of normal females. Journal of Community Psychology, 1, 45-48 Plutchik, R. (1980). A general psychoevolutionary theory of emotion. In R. Plutchik & H. Kellerman (Eds.), Emotion, theory, research and experience (Vol. 1). New York: Academic Press.

394

Chapter 8

Roseman, I. J. (1984). Cognitive determinants of emotion: A structural theory. In P. Shaver (Ed.), Review of personality and social psychology. Vol. 5: Emotions, relationships, and health. Beverly Hills, CA: Sage. Sacks, O. (1985). The man who mistook his wifefor a hat. London: Picador. Scherer, K. R. (1984). On the nature and function of emotion: A component process approach. In K. R. Scherer & P. Ekman (Eds.), Approaches to emotion. Hillsdale, NJ." Lawrence Erlbaum. Schwarz, N. (1990). Happy but mindless? Mood effects on problem solving and persuasion. In R. M. Sorrentino & E. T. Higgins (Eds.), Handbook of motivation and cognition (Vol. 2). New York: Guilford. Schwarz, N., & Clore, G. L. (1983). Mood, misattribution, and judgments of well-being: Informative and directive functions of affective states. Journal of Personality and Social Psychology, 4.5, 513-523. Schwarz, N., & Clore, G. L. (1988). How do I feel about it? Informative functions of affective states. In K. Fiedler & J. Forgas (Eds.), Affect, cognition, and social behavior. Toronto: Hogrefe International. Schwarz, N., & Clore, G. L. (1996) Feelings and phenomenal experiences. In E. T. Higgins & A. Kruglanski (Eds.), Social psychology: A handbook of basic principles. New York: Guilford Press. Schwarz, N., Servay, W., & Kumpf, M. (1985). Attribution of arousal as a mediator of the effectiveness of fear-arousing communications. Journal of Applied Social Psychology, 15, 74-78. Schwarz, N., Strack, F., Kommer, D., & Wagner, D. (1987). Soccer, rooms and the quality of your life: Mood effects on judgments of satisfaction with life in general and with specific life-domains. European Journal of Social Psychology, 17, 69-79. Siemer, M., & Reisenzein, R. (1994). Effects of mood on evaluative judgments: Influence of reduced processing capacity and mood salience. Manuscript under review. Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63, 129-138. Simon, H. A. (1987). In J. Eatwell, M. Milgate, & P. Newman (Eds.), The new Palgrave: A dictionary of economics. London: Macmillan. Sinclair, R. C., Mark, M. M., & Clore, G. L. (1994). Mood-related persuasion depends on misattributions. Social Cognition, 12, 309-326. Smith, E.R. (1994). Procedural knowledge and processing strategies in social cognition. In R. S. Wyer, Jr. & T. K. Srull (Eds.), Handbook of social cognition (Vol. 1, 2rid ed.). Hillsdale, NJ: Lawrence Erlbaum.

T. Ketelaar and G.L. Clore

395

Smith, C. A., & Ellsworth, P. C. (1987). Patterns of appraisal and emotion related to taking an exam. Journal of Personality and Social Psychology, 52, 475-488. Sperber, D. (1995). The modularity of thought and the epidemiology of representations. In L. A. Hirshfeld & S. A. Gelman (Eds.), Mapping the mind: Domain specificity in cognition and culture. Cambridge: Cambridge University Press. Sutton, S. K., & Davidson, R. J. (in press). Prefrontal brain asyrmnetry: A biological substrate of the behavioral approach and inhibition systems. Psychological Science. Taylor, S. E. & Brown, J. (1988). Illusion and well-being: A social psychological perspective on mental health. Psychological Bulletin, 103, 193-210. Tinbergen, N. (1963). On aims and methods of ethology. Zeitschrifi far Tierpsychologie, 20, 410-433. Todd, P. M. (1997). Searching for the next best mate. Unpublished manuscript, Max Planck Institute, Munich, Germany. Tomkins, S. (1981). The quest for primary motives: Biography and autobiography of an idea. Journal of Personality and Social Psychology, 41,306-329. Tooby, J., & Cosmides, L. (1990). The past explains the present: Emotional adaptations and the structure of ancestral environments. Ethology and Sociobiology, 11,375-424. Van Lange, P. A. M., & Kuhlman, D. M. (1994). Social value orientations and impressions of partner's honesty and intelligence: A test of the might versus morality effect. Journal of Personality and Social Psychology, 67, 126-141. Williamson, S. Harpur, T. J., & Hare, R. D. (1991). Abnormal processing of affective words by psychopaths. Psychophysiology, 28, 260- 273. Wilson, E. O. (1971). The insect societies. Cambridge, MA: Harvard University Press. Worth, L. T., & Mackie, D. M. (1987). Cognitive mediation of positive mood in persuasion. Social Cognition, 5, 76-94. Zajonc, R. B. (1980). Feeling and thinking. Preferences need no inferences. American Psychologist, 35, 151-175. Zajonc, R. B. (1984). On primacy of affect. In K. R. Scherer & P. Ekman (Eds.), Approaches to emotion. Hillsdale, NJ: Lawrence Erlbaum.

396

Chapter 8 Author Notes

The authors thank Herbert Bless, Jennifer M. Davis, Klaus Fie~er, and Peter Todd for helpful comments. Support is acknowledged for Gerald Clore from NSF Grant SBR 93-11970, NIMH Grant MH 50074, and John D. & Catherine T. Maearthur Grant 32005-0 to the Center for Advanced Study in the Behavioral Sciences, and for Timothy Kctclaar from NIMH Grant T32 MH18931 to the Postdoctoral Training Program in Emotion Research and a Postdoctoral Fellowship at the Center for Adaptive Behavior and Cognition, Max Planck Institute for Psychological Research.

P A R T III PERSPECTIVES FROM PERSONALITY TRAIT RESEARCH

This Page Intentionally Left Blank

Cognitive Science Perspectives on Personality and Emotion - G. Matthews (Editor) 1997 Elsevier Science B.V. CHAPTER 9

Extraversion, Emotion and Performance: A CognitiveAdaptive Model Gerald Matthews

Extraversion-mtroversion is almost universally acknowledged as one of the major personality traits (Matthews & Deary, in press). Like most broad concepts it is somewhat fuzzy. Central features of extraversion are described in terms of social psychology (sociability, assertiveness), emotion (positive affectivity) and response style (impulsivity). Varied though these attributes are, Hans Eysenck's (1957) pioneering studies showed that the psychological basis for extraversion may be investigated through laboratory studies of performance on simple tasks. Originally, these studies were behaviourist in orientation, but, more recently, studies of extraversion and informationprocessing have flourished. There is also renewed interest in the affective correlates of extraversion. It has even been suggested that positive affect is simply "state extraversion" (Meyer & Shack, 1989). In this chapter, I will discuss the cognitive science of extraversionintroversion that is emerging from this work. I will argue that at least two perspectives are required. The essential first step is to develop detailed information-processing theories of extraversion effects. Both the affective and the performance correlates of extraversion may derive from individual differences in processing. However, such theories do not suffice to explain how cognitive correlates of extraversion relate to the central, defining characteristics of extraversion, such as sociability. I will outline also an adaptive explanation which considers the functional significance of individual differences in processing in supporting "extraverted" and "introverted" behaviours. These two perspectives loosely correspond to the architectural and knowledge levels of explanation discussed in the introduction to this book (Matthews, this volume).

Extraversion, emotion and performance A review of the inter-relationships between extraversion, performance and emotion would fill a substantial tome. In discussing the literature in this area, I will emphasise four central points only. First, extraversion and

400

Chapter 9

emotion are inter-related but the magnitude and implications of the relationship are open to debate. Second, the relationships between individual differences in appraisal and coping and affect suggest a transactional perspective on extraversion. Third, extraversion effects on performance are task-dependent, varying with information-processing demands. Fourth, extraversion effects are context-dependent, varying with environmental properties such as level of stimulation and potential for reward or punishment. A cognitive theory of extraversion must integrate the roles of emotion, task demands and context in personality expression. Extraversion and Affect

Extraversion and positive emotion Extraverts are happier than introverts (Argyle & Lu, 1990), but how much happier? Psychometricians and experimentalists tend to disagree. Watson and Clark (1992) provide data from large samples of college students suggesting that extraversion correlates at about .6 with positive affect or mood, and shows much smaller correlations with negative affect or mood 1. Similarly, Meyer and Shack (1989) propose that extraversion and positive affect define a common factor, such that the tendency to experience positive emotions is the basis for extraversion, which might be relabelled "positive affectivity". Such an alignment between fundamental dimensions of personality and affect is tidy, but not necessarily correct. Experimental studies run in performance contexts show considerably smaller correlations between extraversion and mood, falling to below +.2 in large samples (Matthews, Jones & Chamberlain, 1990a). Matthews et al. (1990a) propose a three-factor model for mood which distinguishes (1) energetic arousal vs. tiredness, (2) tense arousal vs. relaxation, and (3) contentment vs. depression (hedonir tone). Energy and tension were originally identified by Thayer (1989), and are similar to Watson and Clark's positive and negative affect dimensions. Hedonie tone correlates at about .4 with high energy and with low tension, but is factorially distinct from these dimensions. Matthews et al. (1990a) developed the UWIST Mood Adjective Checklist (UMACL) to assess the three dimensions. The UMACL comprises a list of mood-related adjectives, to which the subject responds by indicating how well 1 I will not distinguish rigorously between the terms ~affect", "emotion" and "mood" here, although in other contexts the distinctions are important (see Ketelaar & Clore, this volume)

401

G. Matthews

each adjective describes their mood, at that moment. Table 1 shows correlations between extraversion and the three UMACL scales before and after task performance (Matthews et al., in press). Data were aggregated across six studies in which subjects performed typical laboratory tasks (e.g. vigilance and working memory). Extraversion tended to relate to better mood in general, rather than to any single mood dimension: both more positive mood, and less negative mood. Table 1 also shows that extraversion wass generally a weaker predictor of mood than neuroticism.

Table 1. Mood correlates of extraversion (E) and neuroticism (N), measured with the EPQ-R (Eysenck, Eysenck & Barrett, 1985). Pre-task (N=517)

Post-task (N=762)

,,,

E

N

E

N -07

Energy

14"*

-13"*

06

Tension

-16"*

38**

-18"*

28**

12"*

-26**

14"*

-18"*

Hedonic Tone ** p<.O1

Dora and Matthews (1995) identify two reasons for the discrepancies in results. First, the studies showing larger correlations tend to ask about affect over relatively long periods, such as a whole day, which may be influenced by biases in memory. Second, the emotional correlates of extraversion are context-dependent. Brandst~tter (1994) reports a diary study which shows considerable variation in the correlation between extraversion and overall mood, which ranged from -.50 to +.78 across different contexts. Extraverts were particularly happy, with respect to introverts, in social situations with acquaintances and strangers at home, but extraverts actually tended to be less happy than introverts when working outside the home. Experimental studies using positive mood inductions (Larsen & Ketelaar, 1991; Rusting & Larsen, 1997) suggest that extraverts' mood may be especially sensitive to pleasant stimuli. Hence, studies which do no more than have subjects complete questionnaires are hard to interpret because of the lack of control of context. Possibly, Watson and Clark's (1992) large correlations between extraversion and positive emotion simply indicate that extraverts enjoy American campus

402

Chapter 9

life more than introverts do. It is important to distinguish extraversion and emotion effects on behavioural indices, and to investigate how these factors may interact in their effects on information-processing. There are also conflicting explanatory accounts for associations between extraversion and mood. Watson and Clark (1992) suggest that extraversion may be associated with sensitivity of a neural reward system, as discussed in more detail below. However, extraversion also relates to cognitive aspects of subjective state. Matthews et al. (1997) found that, prior to performance, extraversion was associated though greater self-esteem and greater confidence in competence to perform. Extraverts are more likely to appraise situations as challenging rather than threatening (e.g. Gallagher, 1990), and are more likely to use adaptive coping strategies such as task focus (Costa, Somerfield & McCrae, 1996). Social factors such as extraverts' greater participation and competence in social activities may also contribute to relationships between oxtraversion and mood (Argyle & Lu, 1990). That is, extraversion - mood correlations reflect two distinct influences: (1) individual differences in affective response within a given situation, and (2) individual differences in exposure to different types of situation. Relationships between extraversion and cognitive stress process variables (i.e. appraisal and coping) suggest that the transactional stress theory of Lazarus and Folkman (1984) may be applied to personality. In brief, transactional theory (Lazarus, 1991) identifies affective states as abstracted representations of the person-environment transaction, influenced by cognitions. Traits may then relate to biasing of the person's representations of their interactions with demanding environments. In fact, transactional theory tends to emphasise situational over person factors as predictors of stress response. Although this emphasis may be overstated (see Costa et al., 1996), the relationship between traits and stress reactions should be moderated by the nature of the environment. Context-dependence of stress responses in extraverts and introverts

Next, I discuss some recent data which illustrate how relationships between extraversion, cognitive process variables and mood vary with the environment or context. Two studies assessed fundamental dimensions of appraisal and coping relate~ to the Lazarus and Folkman (1984) model of stress, using items from validated scales (Endler & Parker, 1990; Ferguson, Matthews & Cox, submitted). In the first study (Mohamed, 1996: N=170), the UMACL was given to first year undergraduates as a part of set of

G. Matthews

403

questionnaires on homesickness. These subjects are prone to homesickness, and individual differences in mood were substantially correlated with a homesickness measure used in the study. Appraisal and coping items referred to subjects' cognitions of being away from home. In the second, unpublished study (N=180), conducted by Sian Campbell and myself, subjects performed a speeded, demanding information processing task (Battig & Buzzi, 1986), requiring identification of digit targets, which tended to induce emotional distress. (Coping items were modified somewhat in this study to refer to strategies relevant to the performance context). Table 2 shows correlates of the three UMACL mood dimensions. The appraisal and coping measures were correlated with mood in both studies, jointly explaining from 25-44% of mood variance. Cognitive predictors of mood were fairly similar across the two studies. Energy related most strongly to challenge, task-focus and low avoidance, tension to threat and emoti0n-focus, whereas hedonic tone was associated with all three appraisal dimensions, and with emotion-focus.

Table 2. Cognitive and personality predictors of mood in studies of homesickness (Study 1) and of task performance (Study 2). Energy

Tension

Hedonic Tone

Study 1 Study2

Study 1 Study2

Study 1 Study2

Appraisal Threat

-26**

-10

43**

46**

-44**

-51"*

Challenge

29**

42**

-21

13

45**

20**

Control

24**

13

-47**

-26**

44**

47**

-31"*

-11

37**

41"*

-39**

-53**

Coping Emotion -focus Task -focus

18"

41"*

-16"

03

13

19"

Avoidance

-30**

20*

-05

-15

-25**

-43**

Extraversion

29**

06

-14

00

27**

07

Neuroticism

-26**

-13

46**

35**

-44**

-32**

Personality

* p<.05, ** p<.01

Chapter 9

404

In contrast, extraversion was predictive of mood only in the homesickness context. There were significant cross-study differences (p<.05) between the extraversion- mood correlations for energy (z=2.23) and for hedomc tone (z=1.93). Table 3, which shows personality correlates of the cognitive variables, suggests a possible explanation. Extraversion was more strongly related to appraisal and coping in the homesickness study than in the performance study. Regression analyses reported by Mohamed (1996) show that relationships between extraversion and mood in that study were fully mediated by individual differences in the cognitive variables. Tables 2 and 3 also show that neuroticism was a more consistent predictor of both mood and cognition.

Table 3. Correlations between personality and cognitive stress variables, in two studies. ,.

Extraversion Study 1

Study 2

-33**

Challenge Control

.

,

Neuroticism Study 1

Study 2

-07

42**

29**

19"

-03

-42**

-05

-19"

-16"

-28**

-15"

Emotion-focus

-20*

-09

40**

35**

Task-focus

-02

-03

-02

-10

Avoidance

-04

-11

09

-15"

Appraisal Threat

Coping

..,

* p<.05. ** p<.01

These studies of controlled comcxts show that mood is not just a direct expression of personality. The transactional perspective suggests that within a given context, personality influences mood to the extent that personality fce~s into the processing which support affective states. In some circumstances only do extraverts seem to have more positive cognitions of environmental demands, and more positive moods. The task for a cognitive theory of

G. Matthews

405

extraversion is to explain the context-dependence of relationships between this personality trait and mood. Extraversion and Performance

The cognitive patterning of extraversion-introversion I have concluded from previous reviews (Matthews & Deary, in press; Matthews, 1992; Matthews & Dora, 1995) that extraversion-introversion is associated with a cognitive patterning of performance. Extraversion effects are critically dependent on the information-processing demands of the task. Cognitive patterning is a concept derived from stress research (l-loekey, 1984) and refers to the tendency for particular stressors to enhance some processing functions, impair others and leave still others unaffected. The cognitive patterning shown in Table 4 characterises extraverts as superior, with respect to introverts, in paradigms such as dual-task performance and resistance to distraction (Eysenck, 1982), whereas introverts are better at other tasks such as vigilance and insightful or reflective problem-solving (Koelega, 1992;

Table 4. Cognitive-patterning of extraversion-introversion: Performance characteristics of the extravert, compared with the introvert.

CHARACTERISTICS OF EXTRAVERSION: 9

Superiorityin...

Divided attention Resistance to distraction Retrieval from memory Short-term memory

9

Inferiorityin...

Vigilance Reflective problem-solving Long-termmemory

9

Lowerresponse criterion Little systematiceffect on...

Attentional selectivity Reaction time tasks General intelligence

406

Chapter 9

Kumar & Kapila, 1987). Tasks such as reaction time tasks often fail to show cxtraversion effects, although some specific task versions may do so (Amelang & Ullwcr, 1991). Extraverts also tend to adopt a lower response criterion than introverts, and a more risky style of response (e.g. Koelega, 1992). Data of this kind suggest a need for information-processing models to explain the various elements of the cognitive patterning. Another important strand of research addresses variation of extraversion effects across different environmental contexts. It is well-established that extraverts tend to perform better than introverts in environments in which arousal is raised through agents such as noise or stimulant drugs (Eysenck & Eysenck, 1985). However, interaction between extraversion and arousal is further moderated by time of day. In the evening, the typical interaction reverses, and high arousal benefits extraverts but damages the performance of introverts (Revelle, Humphreys, Simon & Gilliland, 1980). This modal interaction of extraversion, arousal and time of day generalises across a variety of qualitatively different tasks, although, as I shall discuss below, it is far from ubiquitous. Reward and punishment signals may also moderate extraversion effects on performance (Derryberry & Read, 1994). The cognitive correlates shown in Table 4 appear to be relatively robust across different contexts (although there may be some context-dependence). For example, extraverts' superiority in dual-task performance is obtained irrespective of whether or not loud noise is delivered (Eysenck & Eysenck, 1979). With other tasks, such as verbal intelligence test performance, extraversion-introversion effects are almost wholly dependent on level of stimulation (e.g. Revelle et al., 1980). Context-dependence may, in part, be a consequence of interaction between extraversion and the internal context of affective state. Matthews (1985) obtained results similar to Revelle et al. (1980) simply by dividing subjects into those high and low in self-report arousal. In this study, the moderating effect of subjective arousal was stronger than that of noise, an external stressor. In summary, the evidence indicates that extraversion effects on performance are highly contingent upon information-processing demands, internal emotional state, and external contextual factors such as level of stimulation and motivational signals. Next, I consider the strategies adopted by theorists in attempting to explain the data.

G. Matthews

407

Theories of extraversion and performance

The best-known theory of extraversion, Eysenck and Eysenck's (1985) arousal theory, exemplifies the psychobiological approach to understanding personality traits. Extraverts tend to be low in cortical arousal, due to insensitivity of a cortico-reticular neural circuit. With the inclusion of the Yerkes-Dodson Law the theory neatly handles both task-dependence and context-sensitivity. Because the optimal level of arousal is lower for more demanding tasks, extraverts tend to out-perform introverts on difficult tasks but not on simple ones. Because arousal is affected by external stressors as well by personality, extraverts are more likely to be close to optimal arousal in stimulating environments, whereas introverts perform better in de-arousing contexts. The theory is highly parsimonious in generating a wide range of predictions from simple assumptions, and some of its major predictions are at least approximately confirmed by empirical data (Eysenck & Eysenck, 1985; Matthews, 1992). It has more difficulty with the tendencies discussed previously for extraverts to feel happier and more vigorous than introverts. Close examination of the performance data shows serious shortcomings. At an empirical level, arousal does not seem to mediate extraversion effects on performance as supposed. Characteristic extraversion effects are found even when (1) independent observations fail to confirm lower arousal in extraverts, and (2) individual differences in arousal are statistically controlled (Matthews, 1985, 1992; Matthews & Amelang, 1993). Arousal appears to moderate rather than mediate extraversion effects, and the arousal performance relationship differs in extraverts and introverts (Matthews, 1992). At a conceptual level, the utility of arousal theory as an explanation for variation in cognitive function has been severely criticised (Neiss, 1988). The essential point is that the Yerkes-Dodson Law is silent on the information-processing mechanisms affected by arousal, which, in any case, vary with the nature of the arousal or stress state (Hockey, 1984). The neural bases for arousal theory are also increasingly questioned, with most neuropsychologists preferring to w o r k with more finely-differentiated constructs (e.g. Robbins, 1986). There have been several responses to these theoretical diffmulties. One strategy is to develop a better psychobiological model. Gray's (1987, 1991) personality theory associates extraversion mainly with an impulsivity dimension (extraversion lies at 30 ~ to impulsivity in factor space). Individual differences in impulsivity reflect the activity of a Behavioural Activation System controlling response to reward signals. Plausibly, such a system

Chapter 9

408

might be supported by dopaminergic pathways which influence positive affect (Depue, 1995). Integrating psychobiological theories with human cognitive neuroscience may provide better prediction of performance data. Derryberry and Read (1994) extend Gray's theory by suggesting that extraversion and introversion relate to difficulties in attentional disengagement from reward and punishment signals respectively. Disengagement is related to a specific brain system described by Posner and Raichle (1994). More generally, the theory has little to say about the many instances of relationships between extraversion and performance in the absence of obvious reward or punishment signals. In addition, the evidence for the Gray theory is decidedly mixed, with several well-designed studies failing to support it (e.g. Bartussek et al., 1996). A recent comparative review of the two theories (Matthews & Gilliland, submitted) concluded that extraversion may relate to multiple neural mechanisms. Tentatively, we may be able to distinguish two clusters of psychophysiological correlates of extraversion which may be related to "reticulo-cortical" and "dopaminergic" bases for extraversion, respectively, as shown in Table 4. However, it is unclear that either neural mechanism directly explains more than a small part of the behavioural correlates of the trait.

Table 4. Two clusters ofpsychophysiological correlates of extraversion.

,

"Cortico-reticular"extraversion Low cortical arousability Low autonomicarousability Insensitivityto transmarginalinhibition Poor eyeblinkconditioning High sensorythreshold ,

,,

,

,

"Dopaminergic"extraversion Decreased motoneuronalexcitability Conditioningto reward Faster movementtime Multiple channel detection Subjective energy ,,,

,

Another avenue is to integrate arousal theory with cognitive constructs. Revelle (1993; Humphreys & Revelle, 1984) retained a mediating role for arousal, but suggested that arousal effects on performance are, in turn, mediated by availability of two multiple resources, for sustained-information transfer (SIT) and short term memory (STM) respectively. Necka (this volume) provides a more detailed account of this theory, and develops it to explain additional performance data. The intention here is to maintain a biological basis for extraversion, but to add explanatory depth and predictive

G. Matthews

409

power by specifying how individual differences in brain function influence components of the cognitive architecture. A final strategy is to account for extraversion effects in terms of cognitive constructs alone. Eysenck (1982), for example, suggests that extraverts typically have more attentional resources available than introverts. Other researchers have developed '!cognitive mini-theories" for specific paradigms. Dickman and Mayer (1988) conducted an elegant processing stage model analysis of impulsivity effects on a visual comparison task, and concluded that impulsivity relates to a relatively early stage. Weinman (1987) showed that extraverts' difficulties in problem-solving arise from use of impulsive exit strategies which terminate processing prematurely. Work of this kind is essential in building up an accurate description of the cognitive patterning of extraversion, but it is not immediately obvious how the various components of the pattern may be synthesised within a more general explanation.

Extraversion, Arousal and Attention: Empirical Studies It is reasonable to see extraversion research as being in a state of transition between the early, cognitively-naive psychobiological theories and emerging theories informed by cognitive science. Two major unresolved issues are the role of arousal, and the level of information-processing construct to which extraversion is to be related. This section reviews a series of studies which addressed these issues. Measurement of arousal

I have argued that extraversion cannot be seen simply as a "trait energy" or "trait happiness" factor, whose effects on performance are mediated entirely by state arousal or mood. At the same time, external stimulation and internal arousal operate as powerful contextual factors which moderate extraversion effects. The traditional method for investigating arousal is to manipulate the person's state with experimental stressors. However, this method may not be adequate for discriminating extraversion and arousal effects, given the limitations of interpreting stressor interactions post hoe (Hockey, 1984), and the probability that extraverts and introverts differ qualitatively in their physiological reactions to arousing agents (Smith, 1983). Matthews (1985) argues that arousal is only a scientifically useful construct if it can be operationalised through direct measures, probably of multiple

410

Chapter 9

arousal dimensions. Studies of extraversion require independent measures of individual differences in arousal also. Attention as a key cognitive construct

The second contentious issue is the type of cognitive construct to which cxtraversion should be related. It is a truism that information-processing models are required, but what kind of model? The cognitive patterning approach suggests that we can compile a set of cognitive mini-theories relating to extraversion effects on different tasks (and over different timescales: Revelle, 1993). I return subsequently to the problem of how we can relate such a dissected view of extraversion to the integrated trait most personality researchers believe it to be. Alternatively, there may be more general mechanisms sensitive to extraversion-introversion which can encompass some, if not all of the mini-theories. Such mechanisms might relate both to individual differences in cognitive architecture and to strategy (see Matthews, this volume). At the strategic level it seems clear that extraverts are "geared to respond" (Brebner, 1983) somewhat impulsively, although cognitive accounts of responsivity in extraversion remain somewhat paradigm-bound (e.g. Weinman, 1987). The range of very different tasks sensitive to extraversion implies that we should investigate general-purpose components of the architecture, such as attentional capacity and working memory. As we have seen, Eysenck (1982), Humphreys and Revelle (1984) and Necka (this volume) focus on processing resources as a key mediating construct. Resources may be defined as a reservoir of a metaphorical "energy source" required for processing, which may be flexibly allocated to a variety of different processing routines. The concept is promising as a means for integrating data findings different task paradigms, because dependence of functionally distinct processes on common a resource pool is intrinsic to it. Eysenck's (1982) hypothesis of greater resource availability in extraverts is supported by data from various dual-task paradigms, including learning while subject to distraction (Eysenck, 1982), text processing (Bermudez, Perez & Padilla, 1988) and dichotic listening (Dunne & Hartley, 1985). These and other studies also show the importance of distinguishing different aspects of attention (see also Derryberry & Read, 1994). Both dual-tasks and vigilance are demanding attentionally, but extraversion is associated with performance enhancement on the former but impairment on the latter (Koelega, 1992). Furthermore, in dual-task paradigms, extraversion tends to affect attentional efficiency rather

G. Matthews

411

than selectivity. Effects on selectivity differ across paradigms, and appear to be a secondary consequence of attentional efficiency effects (Matthews, 1992). For example, extraverts appear to be less distractible during learning (i.e. increased selectivity) because of their greater attentional efficiency in this paradigm. Extraverts also show a performance advantage on certain attentionally demanding real-world tasks, and during job training, which presumably tends to require investment of resources in skill acquisition (Matthews, Jones & Chamberlain, 1992; Matthews, 1997a). Various methodological critiques of resource theory (e.g. Pashler, 1994), demonstrate that caution is necessary in inferring resource allocation change from performance change. It is important that resource mechanisms are distinguished from other processing mechanisms to which extraversion may also relate. Careful experimentation is required to demonstrate that resourcelimitation of the task controls its sensitivity to extraversion-introversion effects. Designing a series of studies

Next, I review results from a series of studies intended to investigatethe inter-relationships of cxtraversion, arousal and attentional resource availability. These studies were designed to incorporate three essential features: (1) Use of an explicit theoretical framework for attentional processes, required for distinguishingmultiple mechanisms for attentionconceptually; (2) Use of explicittests for resource-limitation,required to show that a resource model of attention is to be prefcrrcd to alternative types of explanation; (3) Independent measurement of energetic arousal using a validated adjective checklist,required to distinguishcxtraversion and mood effects. The theoretical framework was provided by a dual-level model of information-processing and action. There are a number of different variants of this class of model (e.g. Posner & Snyder, 1975; Shiffrin & Schneider, 1977; Norman & Shallice, 1985; see also van Reekum & Schercr, this volume). Norman and ShaUice (1985), for example, distinguish an "upper", executive level which supervises and regulates operation of a "lower" level, comprising a network of processing modules, with local controls over conflict resolution. The levels arc distinguished in that operation of the executive is voluntary and strategy-driven, resource-limited and partially accessible to consciousness ("controlled"). Lower level processing is involuntary and

412

Chapter 9

stimulus-~ven, requires few or no resources, and is unconscious ("automatic"). Performance of undemanding, well-learnt tasks is largely under lower level control, but the upper level is called into operation for more demanding and novel tasks. A major aim of the research was to test whether extraversion influence~ the lower or the upper level of control. Tasks whose level of control was understood were chosen for the studies, including sustained attention and search tasks whose resource demands could be varied by task parameter manipulations. Arousal was measured by self-report, using the UMACL (Matthews et al., 1990a). Thayer (1989, 1996) has argued the arousal dimension most predictive of performance is energetic arousal, contrasting feelings of vigour and tiredness. Matthews et al. (1990a, in press) showed that the energy scale was appropriately sensitive to stress manipulations such as sleep deprivation. In the studies reported, energetic arousal was usually measured prior to task performance, to avoid contamination by the subjects' appraisals of their success or failure on the task. Vigilance tasks in particular tend to depress energy, but individual differences are reasonably consistent across the time interval of 30-60 minutes during which tasks are performed. To analyse pretask energy effects on performance, subjects were divided into high and low energy groups on the basis of a median-split, with groups differing in energetic arousal by 1.5-2 SD, typically. This technique allows for individual differences in arousal response to the experimental context, and ensures that effects are genuinely associated with energetic arousal, rather than other elements of stress state. Next, the results of studies of energy and extraversion effects on performance are reviewed. Tasks believed to be under upper level control are discussed first, followed by lower level tasks. Upper level tasks: High-event rate sustained attention and controlled search Sustained attention. See, Howe, Warm and Dember (1995) review the literature on vigilance decrement, within a signal detection theory framework. A meta-analysis showed that, in general, the r of perceptual sensitivity decrement increases with task difficulty. Parasuraman, Warm and Dember (1987) attribute sensitivity decrements to depletion of resources over time, when the task is highly demanding. The resource hypothesis is supported by dual-task studies (Matthews & Davies, in press). Overall demands in vigilance depend on several parameters, notably event rate and stimulus degradation. Another important parameter is whether the task requires a

G. Matthews

413

simultaneous or successive discrimination. In simultaneous tasks, each trial

presents sufficient information for a decision on whether or not the stimulus is target or a non-target (e.g. detection of a single target digit). In successive tasks, information from two or more successive trials must be integrated to detect the target, as in the Bakan (1959) task which requires the subject to detect a sequence of three consecutive odd digits. The essence of the resource hypothesis is that overall demands are a stronger determinant of the sensitivity decrement than individual task parameters. Hence, if extraversion and energy affect resources, these factors should have stronger effects on more demanding tasks, but the effect should not be contingent upon any individual task parameter. Six studies of extraversion and energy effects on demanding sustained attention tasks performed singly were conducted. In each case, the subject required was to press a key when a pre-specified visual target was presented. Event rate was fast - typically 1 stimulus per second. The first three studies used degraded stimuli to which visual noise was added (Matthews, Jones & Chamberlain, 1989; Matthews, Davies & Lees, 1990b). The task was to detect a specified digit, in a sequence of single, briefly presented digits. In the next three studies (Matthews, Davies & HoUey, 1990c) type of discrimination (simultaneous vs. successive), and type of coding (visual or symbolic) were manipulated, giving four distinct tasks. Symbolic tasks required digit detection, whereas visual tasks required subjects to judge the lengths of lines degraded with visual noise. Type of coding was manipulated to test the generalisation of results" it has been suggested (Wickens, 1984) that visual and symbolic tasks may draw on different multiple resources. In general, these studies showed that individual differences in energy were a much more consistent predictor of perceptual sensitivity than was extraversion-introversion. There was a weak tendency for extraverts to perform poorly on visually coded tasks, such as discriminating lengths of flickering lines (Matthews et al., 1990c; Experiment 3), but better on symbolically coded tasks, such as a nine-minute version of the Bakan task with degraded stimuli (Matthews et al., 1990c; Experiment 2). In general, though, the data suggested that the normal finding of poorer vigilance in extraverts (Koelega, 1992) does not generalise to these particularly demanding tasks. Energetic arousal effects depended on task demands. Tasks which show a significant perceptual sensitivity decrement also show a facilitative effect of energy (Matthews et al., 1990c). Tasks which were too easy to evoke a P(A) decrement (Matthews et al., 1989; Matthews et al., 1990c: Experiment 2) a

414

Chapter 9

failed to induce an energy effect, even though stimuli were degraded and performance levels were below ceiling. Provided there is a sensitivity decrement, the energy effect generalises across a variety of task types. Across the various studies, the effect has been demonstrated for externally and selfpaced tasks, symbolic and visual tasks, and for tasks of varying duration and signal probability. None of the task parameters appears to be either a suftieient or a necessary condition for significant energy effects. Take visual degradation as a task demand manipulation: Matthews et al. (1990b; Experiment 1A) found that energy related to perceptual sensitivity with degraded but not undegraded stimuli. However, degradation is not a sufficient condition for the energy effect. When degradation does not induce overall performance decrement, energy does not facilitate performance (e.g. Matthews et al., 1989; Matthews et al., 1990c; Experiment 3). It is not necessary, in that energy effects may be obtained through other task demand manipulations. Matthews et al. (1990b; Experiment l b) found that energetic arousal facilitates detection of undegraded digits when a self-paced task was used and subjects spontaneously made the task difficult for themselves by adopting a high event rate (about three digits per second). Matthews and Davies (in press) report two dual-task studies. In the first, auditory or visual secondary probe stimuli were presented during performance of the degraded digit detection task. Resource theory predicts that probe RT should slow as performance of the primary task deteriorates. This pattern of results was obtained only for visual probes. Correspondingly, energy affected performance in the visual condition only, in which low energy subjects showed poorer signal detection, and slower probe RT. A second dual-task study used undegraded stimuli, and compared simultaneous and successive tasks. All tasks required discriminations between visually coded targets and non-targets (see Gluckman, Dember & Warm, 1988, for a description). As predicted, energy had no effects on the relatively undemanding single tasks, but facilitated dual-task performance of successive tasks. Figure 1 shows that low energy subjects were most disadvantaged when required to perform two concurrent successive tasks. Overall, demands for resources appear to be more important than specific task parameters in controlling whether or not energy improves sustained attention. Search tasks. It might be argued that the results discussed so far suggest that energy affects sustained attention specifically, rather than some more general resource. Stronger evidence for the resource hypothesis requires a demonstration that the effect generalises to qualitatively different tasks. Three studies have tested whether energy affects resource-limited visual and

415

G. Matthews SIMULTANEOUS TASK 0.94

-

0.92

-

SUCCESSIVE TASK

El

m

m

0.90 0.88 P(A) 0.86 0.84

E]

. -a. - Low energy --

0.82 0.80

I Single

High energy

I Dual- 1

I

I.

Dual- 2

4 Single

I Dual- 1

I

Dual-2

Task pairing D u a l - l ' Paired with simultaneous task Dual-2: Paired with successive task

Figure 1. Perceptual sensitivity (PA) levels for groups high and low in energetic arousal while performing a simultaneous or successive task singly, or paired with another simultaneous or successive task (Matthews & Davies, in press).

memory search tasks. The Shiffrin and Schneider (1977) paradigm was used, in which varied mapping (VM) and consistent mapping (CM) conditions are said to promote controlled and automatic search respectively. Task demands are varied either by manipulating the number of display items to be searched (attentional demands) or the memory load. The more demanding VM tasks should bo the most strongly rcsourc~-limite~l. Again, energy rather than extraversion was the most reliable predictor of speed of search. Matthews et al. (1990b; Experiment 2) showed that energy influenced VM but not CM search, as predicted from the resource hypothesis. However, energy only enhanced VM visual search, and not memory search. A third experiment reported in the same paper found that energy facilitated speed of letter transformation, but there was no moderating effect of memory load. Matthcws and Margctts (1991) conducted a second study of controlled search only, which tested whether the effect generalized to a task requiring semantic processing: search for instances of categories. A dual-task methodology was used, with subjects required to search two sets of words

416

Chapter 9 RT (a)

Positive trials

Negative trials

_

2.5High arousal E]

_

--n.. Low arousal

1.5-

i

Single

i

l

Dual Task demands

I

i

Single

Dual

Figure 2. Response time on a controlled category search task, as a function of energetic arousal, single vs. dual task performance and trial type (Matthews & Margetts, 1991).

presented in different colours in the dual-task condition. Figure 2 shows that energy cnhance~ speed of search more strongly in the dual-task than in the singlo-task condition, for both positive and negative trials. Matthcws and Margctts (1991) also constructed Performance Operating Characteristics (POCs: Wickcns, 1984) to show that energy affected resource availability rather than resource allocation, although there was some evidence for energy increasing selectivity of allocation in addition to enhancing total resource availability. Further analysis showed that energy did not affect the specific search strategy used. The last study (Matthcws & Wcstcrman, 1994) showed that energy did enhance controlled memory search for characters provided the memory set size was sufficiently large: 6 characters as opposed to the 4 characters used by Matthcws ct al. (1990b). This study also showed that high tension tended to block facilitative effects of energy. Extraversion, energy and upper level control: Conclusions. The data show a clear dissociation between energy and extraversion effects, reinforcing the conclusion that extraversion is not simply trait energy or positive affect.

G. Matthews

417

Extraversion was not consistently related to performance, although there were occasional task-specific effects (Matthews et al., 1990c). There is good evidence from other sources that the tasks used were resource-limited, so the data conflict with other findings suggesting greater resource availability in extraverts (of. Eysenck, 1982). However there is an important difference between the tasks used in the current research and those suggesting greater resource availability in extraverts. Studies such as those of Eysenck and Eysenck (1979) and Bermudez et al. (1988) used verbal stimuli, implying that extraverts show a particular facility on demanding verbal tasks. Only one of the studies reviewed above used verbal stimuli (Matthews & Margetts, 1991), and its sample size (N=36) may have lacked sufficient power. Undemanding verbal processing tasks, such as lexical decision, show no general advantage for extraverts (Matthews & Harley, 1993). Extraverts appear superior at the management and control of processing multiple streams of verbal stimuli, when processing is demanding, but does not necessarily require reasoning or intellectual abilities. We can loosely attribute this facility to greater availability of resources specifically for verbal processing, although the data are also consistent with an alternative hypothesis, that extraverts have more efficient executive routines for handling multiple verbal inputs. In contrast to extraversion, energy was reliably positively correlated with performance efficiency of a range of attentionally demanding tasks, of varying information processing characteristics. Generalisation of the task across tasks and task version implies that energy affects a general resource, rather than any single task parameter or process (see also Necka, this volume). However, energy did not, in general, seem to affect task strategy. It had no consistent effects on speed-accuracy tradeoff or criterion setting, for example. Hence, energy appears to influence the efficiency with which supervisory executive control is effected, but not the specific computations performed by the executive. The data are inconsistent with the YerkesDodson Law, because energy facilitated demanding rather than easy tasks. They partially fit the Humphreys and Revelle (1984) model, which proposes that arousal enhances availailability of the SIT resources required for attentional resources. However, the data do not support the further hypothesis that arousal reduces availability of STM resources.

Lower level tasks: Routine encoding and semantic priming Simple stimulus encoding tasks. Other studies investigated extraversion effects on various tasks in which stimuli were highly discriminable, error

418

Chapter 9

rates were low, and little "mental effort" was ne~od to perform the task. Matthews and Chappelow (1986) carried out an initial study, requiring subjects to detect the target digit "7". Two versions of the task were run, both of which included distraction conditions in which additional characters were presented. In one version, the digits to be attended were distinguished by spatial location, and the other by alphanumeric category (distractors were letters). Results showed the characteristic interactive effect of r and energetic arousal on frequency of missed targets: low energy r and high energy introverts missed more targets, irrespective of the type of selection, presence or absence of distraction, and spatial separation of locations. In other words, r and energetic arousal affected accuracy of detection of briefly presented, easily porccive~, single characters, but personality had no effect on the subject's ability to resist distraction. Extraversion and energy affect character encoding, but not selective attention. These findings contrast with those from more demanding distraction paradigms using word stimuli, in which r show superior selective attention (Eysenck, 1982). This first study suggested that r and energetic arousal may affect lower level processes of stimulus encoding and detection. To test this hypothesis explicitly, the next study (Matthews, 1989) used a levels of control task which allows discrimination of upper and lower level errors, as defmod by Norman and Shallicr (1985). The subject views a sequence of single letters, searching for a specified target. On timing the target, the subject switches to looking for a second target, using an explicit rule. Failure to respond to a target constitutes a "miss error" and is attributed to failure of lower level stimulus-driven detection. Failure to update the current target, a "rule error", is attributed to failure of an upper level executive mechanism. The modal triple interaction between time of day, extraversion and energetic arousal was obtained for miss errors (see Figure 3). In the morning, extraverts missed more targets when low in energy, but introverts committed more miss errors when high in energy, with the opposite pattern of interaction occurring in the evening. No effects of r and energetic arousal on rule errors were found. In other words, lower levd detection processes appear to be more sensitive than executive control of performance to interactive effects of r and energetic arousal. The third study (Matthows r al., 1989) assessed the generality of the effect across six tasks differing in their demands on attention and STM. Two tasks showed the predicted time of day x r x energetic arousal interaction: five-choice serial reaction and a digit recall task similar to digit

419

G. Matthews PM

Arl

3.0--

~'-

[i'lExtraverts [~] Introverts

2.5-

r

Q

#s L G

2.0

L. .io L I0

v

4O L. L

o L.

1.5-#

1.0--

%

d

a.} (f) .,-

:x:::

0.5--

0.0

I

'I

Low

High

'

I

'

'1

Low

'

I. . . .

I

High

Energetic Arousal

Figure 3. Missed target frequency as a function of time of day, cxtravcrsion and energetic arousal (Matthcws, 1989).

span. However, the two most demanding attcntional tasks, sustained attention to degraded digits, and search for six letters in a display of 30 letters, were insensitive to the interaction. This study confirmed that the typical interaction is found with attcntional tasks requiring a simple response to easily perceived stimuli, but not with more demanding tasks which perhaps require more voluntary effort. The gcncralisation of the effect to digit recall might appear surprising, but possibly reflects the contribution of individual differences in identification or encoding of items to digit span (Dcmpstcr, 1981). Lexical decision and semantic priming. Matthews et al. (1989) proposed a lower level encoding hypothesis for extraversion • energetic arousal interactions. Extraversion, energy and time of day may influence the stimulus-driven processing units which analyse the target status of routine, easily-perceived stimuli, under lower level control, as described by Shallice and Norman (1985). However, as so far described, the mechanism is somewhat vague and loosely specified. A more precise hypothesis is that extraversion and energy interactively affect individual differences in the properties of an interactive activation network, of the kind proposed by parallel distributed processing (PDP) theories of cognition. Such theories are well-suited to modelling performance on simple recognition and

420

Chapter 9

discrimination tasks (Harley, 1993). If so, two kinds of predictions may be made. First, cxtravcrsion and arousal should relate to measures of spreading activation within the network. Priming of Icxical decision provides a technique for assessment of spreading activation. Second, it should be possible, through simulation, to identify parameters controlling network function whose variation generates individual differences in performance similar to those obtained in real data. Matthcws and Harley (1993) investigated whether r and energy intcractivcly affect semantic priming of lcxical decision (Nccly, 1991). The subject decides whether a target letter string is a valid English word or not. On primed trials, the target string is preceded by a semantically related prime, which on positive trials reduces decision latency. For example, presentation of the primo "DOCTOR" would speed the decision that "NURSE" is a valid word. The priming effect can be attributed to the spread of activation from the lcxical unit corresponding to the prime, to the unit for the target word. The magnitude of priming may then be used as an index of the spread of activation. Across two studies, Matthcws and Harley (1993) showed that extravcrsion and energy reliably affected priming magnitude but not speed of unprimcd lcxical decision, as shown in Figure 4. The direction of the effect varied across time of day, such that subject groups who normally perform well (e.g. aroused cxtravcrts in the morning) also show more priming, consistent with the spreading activation hypothesis. In one study, half of the stimuli were pattern-masked (though still above recognition threshold). Although masking had some effects on response time, it did not interact with the cxtravcrsion effect on priming, implying that cxtravcrsion and energetic arousal do not simply affect feature extraction processes operating prior to lcxicalisation. Current conceptions of priming rccognisc that it may reflect multiple processes. Nocly (1991) distinguishes automatic priming taking place at short stimulus onset asynchronies (SOAs) between prime and target strings (<300 ms) priming from expectancy-or strategy-controlled priming operating at longer SOAs. Within PDP models, expectancy effects may be controlled by top-down activation spreading from "task demand units", whose activation depends on strategy and task instructions (Cohen, Scrvan-Schrcibcr & McClclland, 1992). Matthcws and Harley (1993) manipulated SOA, but found no significant interaction between individual differences in priming and SOA. They inferred that personality affects the sensitivity of network units to activation, rather than the connections associated with automatic or strategydriven routes for priming. Nccly (1991) also points out that priming may be

421

G. Matthews PRIMING

Extraverts

....n .... I n t r o v e r t s

(ms) 100

Expt. 1 (PM N = 40)

Expt. 2

90-

(AM" N = 5 6 )

80 706050403020100

!

Low

-

f

High

Low

High

AROUSAL LEVEL

Figure 4. Priming magnitude as a function of time of day, extraversion and energetic arousal (Matthews & Harley, 1993).

affected by checking following lexical access: the verification and comparison of one or more candidate words retrieved from lexical memory. Harley and Matthews (1992) tested whether post-lexical checking processes might account for the interaction, by manipulating the r of words and non-words, which increases the need for checking. The effect of individual difference factors was stronger when confusability was low, implying that it is not the checking mechamsm which is responsible for the effect of extraversion and energetic arousal on priming. An interactive activation simulation o f individual differences in priming. The hypothesised spreading activation mechanism was further investigated in the simulation phase of the research (Matthews & Harley, 1993). A modification of McClelland and Rumelhart's (1981) architecture, comprising semantic, phonological and lexical units, was used to simulate lexical decision and priming effects. Word recognition required an integration across time of lexical unit activation to reach a threshold value. Three key network parameters were then varied: (1) the strength of the excitatory connections linking semantic and lexical units;

Chapter 9

422

(2) the decay constant which governs rates of change of activation at the individual unit; (3) the level of random variation in activation ("noise"). Figure 5 shows the effects of parameter variation on unprimed lexical decision and priming magnitude, roughly calibrated against real data so as to express the findings in ms rather than simulation cycles. The leftmost data point represents the initial baseline condition. The three plots show how network performance degraded as each parameter was independently varied. Priming magnitude was most strongly affected by random noise level. Furthermore, increasing noise initially affected priming more than speed of unprimed lexical decision. It is thus plausible, though somewhat speculative, that the extraversion effect is associated with individual variation in noise along the left-hand part of the noise plot shown in Figure 5, generating reliable effects on priming, but not on unprimed response.

PRIMING (ms) 60

Connection

""

s t r e n gt h 50

""

40

"

30

-

20

"-

10

""

Decay constant

Noise

o 540

I

I

560

580

I"=600

=-= 620

=

~

',

640

660

680

UNPRIMED RT (ms)

Figure 5. Effects of varying three network parameters on performance, in a simulation study (data from Matthews & Harley, 1993).

G. Matthews

423

Extraversion, arousal and attention: A recapitulation

The studies reviewed so far demonstrate that there are two qualitatively different effects of extraversion and energetic arousal on attention, summarised in Table 5. The first effect, the facilitative effect of energy, is associated with efficiency of upper-level control over attention. It represents enhancement of resource availability, as in the Humphreys and Revelle (1984) model, although it remains to be seen whether the resource metaphor is useful in developing more detailed, computational accounts. Effects of extraversion on "resources" may be limited to certain types of verbal task. The second effect, the interactive effect of time of day, extraversion and energetic arousal, is associated with individual differences in the efficiency of lower level detection or encoding processes. This effect might be modelled in terms of individual differences in noise levels in an interactive activation network. This characterisation of extraversion and energetic arousal effects has both strengths and limitations. Its strengths are that it accommodates a substantial part of the existing empirical data. Enhancement of vigilance by stimulants which increase energy is well-established (Davies & Parasuraman, 1982). Under some circumstances, subjective energy may also enhance performance of more complex real-world tasks such as semi-automated mail sorting (Matthews, Jones & Chamberlain, 1992) and multi-source monitoring during simulated flight (Singh, Molloy & Parasuraman, 1993). However, energy effects tend to disappear when performance depends on strategic factors rather than resource availability (Matthews & Westerman, 1994). Various relatively complex processing tasks, such as combined memory and visual search (Matthews et al., 1990b; Experiment 2), and the cross-modal dual task used by Matthews and Davies (in press), are insensitive to energy. Spilsbury (1992) draws an important distinction between complexity and demands as properties of tasks. The prototypical energy-sensitive task is demanding but not complex. Similarly, the characteristic interaction between extraversion and arousal is found using other arousing agents, such as loud noise and caffeine, in place of self-reported energy. Tasks plausibly dependent on lower-level encoding are especially sensitive, such as the more traditional, low event rate vigilance task (e.g. Davies & Hockey, 1966), and letter cancellation (Blake, 1971). The effect also generalises to more complex tasks such as intelligence and creativity test performance (Matthews, 1986; Revelle et al., 1980). Consistent with the lower-level encoding hypothesis, it is the easier "speed" intelligence

Studv

No.

N

1A 1B 1 2

100 36 60 60

Task (condition)

a

424

Table 5. Summary of attentional tasks showing (1) facilitative effects of energy on sustained attention and search and (2) the modal interaction between time of day, extraversion and energy.

!2

Error rate (%)

Energyfacilitation (mstained attention)

Matthew, Davies & Lees (1990b) Matthew, Davies & Holley (1990~) Matthews & Davies (in press)

1 2

Simultaneous degraded digits Self-paced simultaneous digits Simultaneous degraded digits Simultaneous degraded lines Successive degraded lines 102 Degraded digitdvisual probe 108 Successive, dual-task

Chapter 9

Energyfacilitation (search)

Matthews, Davies & Lees (1990b) Matthew & Margetts (1990) Matthew & Westerman (1994)

2

100 Controlled visual search 36 Controlled category search (dual-task) 50 Controlled memory search (dual-task)

-

36 Digit detection (early selection ) Digit detection (late selection ) 60 Levels of control (miss errors) 116 Five-choice serial reaction (RT) 60 Primed lexical decision (with non-conhsable non-words) 36 Primed lexical decision 40 Primed lexical decision

-

The "modal"intemction

Matthew & Chappelow (1986) Matthews (1989) Matthews, Jones & Chamberlain (1989) Harley & Matthews (1992) Matthews & Harley (1993)

1 2

1.6 3.6 0.3 1.7 1.O 0.7 1.5

G. Matthews

425

intelligence test items which are more sensitive to the effect than more difficult "power" items (Matthews & Dora, 1995). Fagerstrom and Lisper (1977) report a study implying generalisation to real-life tasks: playing the car radio enhanced attention to a probe stimulus during driving in extraverts but not in introverts. Both types of effect can also be demonstrated using physiological rather than subjective measures of arousal. Munroe, Dawson, Schell and Sakai (1987) showed that performance on the degraded digits vigilance task was higher in subjects characterised by high electrodermal reactivity. Matthews and Amelang (1993) showed that extraversion interacts with individual differences in EEG alpha much as it does with subjective arousal. The strongest such effect was found with two tasks dependent on routine encoding (letter cancellation and reaction time during memory consolidation), with a weaker effect on verbal intelligence test performance. Thus, further explanation for extraversion and energetic arousal effects may be possible through cognitive neuroscience. Matthews and Davies (in press) suggest that energy may act through dopaminergic afferents to an anterior attentional system responsible for event detection, located in mid-prefrontal cortex (Posner & Raichle, 1994). It appears too that the functional significance of c.n.s, arousal varies with personality (Matthews & Amelang, 1993). Connectionism may facilitate the development of neural explanations for individual differences in information-processing. The two-level account is limited in two respects. First, it explains only a subset of the elements of the cognitive patterning of extraversion. Performance correlates such as poor problem-solving and faster retrieval from semantic memory are outside its terms of reference. Further work would be necessary to develop cognitive mini-theories of these effects. For example, the role of impulsive exit strategy (Weinman, 1987) in extraverts' problemsolving impairment implies an upper-level mechanism, relating perhaps to individual differences in routines for terminating strategic control. Retrieval phenomena might be modelled in terms of either voluntary search strategies or activation properties of semantic networks. However, even if we had a complete set of information-processing models, we would still lack an explanation for the patterning as a whole, and its relationship with the core qualities of extraversion as a personality trait, such as sociability and positive emotionality. In the next section, I develop an adaptive framework which seeks to provide this level of explanation.

426

Chapter 9 An Adaptive Framework for Cognitive Correlates of Extraversion-lntroversion

Psychobiological models of extraversion typically seek a critical parameter of neural functioning which will explain the full range of correlates of the trait, such as retieulo-cortieal arousability or the balance of sensitivities of reward and punishment systems (Gray, 1981). However, the cognitive patterning of extraversion does not seem to be reducible either to any single processing component, or to any gross brain property (Matthews, 1992). What is required is a rather different kind of trait theory, which acknowledges that extraversion is distributed across a variety of largely independent cognitive and physiological functions. Matthews and Dom (1995) propose that the cognitive pattemings of traits are associated with adaptive functions. Traits must be understood with reference to environments within which the behaviours associated with the trait are expressed. To understand extraversion as a quality of the person, we must seek the environments within which extraverted behaviours are adaptive. It is not difficult to think of plausible examples: soeialising with strangers, high-pressure occupations, seeking a sexual partner, and the like. The more difficult issue is to find the common element inter-linking these different types of encounter. Matthews and Dora (1995) propose that extraverts are adapted especially for environments characterised by high information flows. These include many social environments, because other people tend to deliver multiple (and often ambiguous) stimuli in parallel, through both verbal and nonverbal channels. More precisely, the adaptation is not so much to high workload per se, but to information streams which are difficult to manage because they are poorly segregated, and it is unclear which stimulus attributes are personally relevant. In this section, I will refer to "overload" in this restricted sense. A party full of strangers is a protypical high information environment of this kind, and one where we would expect extraverts to prosper (as their activity preferences suggest: Fumham, 1981). To support the adaptive argument it must be shown that cognitive characteristics of extraversion are in fact likely to enhance the person's ability to gain rewards and avoid losses in such situations.

Adaptive significance of performance correlates of extraversion Matthcws (1997b) points out that successful handling of real-life encounters requires the acquisition of context-specific skills. Extravcrts may have a high degree of aptitude for initiating and managing rewarding

G. Matthews Trait .

.

.

.

.

Information Processing .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Skill .

.

.

.

.

.

Divided attention Extraversion

427

Rapid retrieval

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Adaptive Benefit .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

> Conversation skills

Working memory

~-Success

in IIL -~_.~socialand

Low response criterion -

" - Rapid response

Interaction with arousal and time of day

~-~ Efficient functioning J l

~

high information envirinments 'High-pressure' occupations

IPhysiological stress resistancel Introversion ~

Vigilance Strategic caution

Success in solitary and Problem-solving/ .......--'1~ low information Capacity for reflection occupations Sustained monitoring ~

I Artistic/scientific occupations

Figure 6. Adaptive functions of information-processing correlates of extraversion-introversion (adapted from Matthews, 1997a).

conversations with strangers, but success nevertheless depends on acquiring culture-bound conversational skills. The proposal here is that these skills are built on the foundation of the cognitive components associated with the trait. Skills such as speaking effectively in multi-person conversation require verbal STM to keep track of the conversation, fast retrieval of topics to talk about, and rapid initiation of utterances to avoid being excluded by other speakers. The impulsivity of extraverts serves to prevent behavioural paralysis induced by attempting full processing of all messages. Figure 6 shows how the cognitive correlates of extraversion may relate to acquired skills required for successful extraverted behaviour in various contexts, and how the overall adaptation may be supported by a variety of skills. This analysis provides a knowledge-level explanation for the data on extraversion and informationprocessing (see Matthews, this volume). The cognitive correlates of extraversion reflect the way the processing system has been designed to

428

Chapter 9

support extraverted activities. The question of how it is "designed" will be addressed subsequently. The advantage of the adaptive approach is its integration of disparate findings. For example, the interactive effects of time of day, extraversion and energetic arousal on routine stimulus enexxling appear unrelated both to other cognitive correlates of extraversion and to the defining qualities of extraversion: extraverts do not become introverts as the day progresses (Gray, 1981). A key observation is that in everyday life the interaction is superimposed on what is probably a biologically-based circadian rhythm in energetic arousal (Matthews & Harley, 1993). People are low in energy early in the morning, and for a longer period during the evening (Thayer, 1978, 1996). In consequence, extraverts will tend to perform poorly shortly after awaking, whereas introverts will be disadvantaged in the evening, as confirmed by Blake (1971). Matthews and Harley (1993) argue that it is more important to extraverts than to introverts to function effe~ively at parties and social gatherings, which typically take place in the evening. Hence the time of day/arousal interaction helps to support extraverts in one of the activities characteristic of the trait. Late in the evening, most people are likely to suffer depleted resources due to decline in energy, but extraverts may be able to maintain social interaction on the basis of efficient low-level processing. Loosely speaking, the extravert may be able to "make friends and influence people" on autopilot. Conversely, introverts may have an advantage in the first hour or two of the working day.

The role of positive emotion So far, the adaptive perspective provides a rather bloodless view of extraversion, in that I have not addressed the relationship between extraversion and positive emotion. I will use the term "positive emotion" loosely in this section to refer to the general enhancement of mood evident in extraverts previously described. Two kinds of question arise. First, I will consider the knowledge-level question of how a context-dependent positive affeetivity might relate to the adaptive hypothesis for extraversion just described. Second, I will take another look at more fine-grained explanations for extraversion - emotion associations, within the adaptive framework. In particular, I will re-assess the respective roles of cognitive and physiological processes. Adaptive accounts of positive moods are well-known. Thayer (1996) presents an adaptive account of the energy states related to extraversion.

G. Matthews

429

Energy is related to an integrated biopsychological system for regulating motor activity. Similar to Ketelaar and Clore (this volume), Thayer sees mood states as having both informational and motivational aspects. Energetic arousal both signals that the person is ready for activity, and motivates motor activity. The enhancement of attentional function in energy states discussed above is commensurate with this adaptive view of energy. More generally (and more cognitively) we can see positive emotion as an index of the status of important plans, signalling ongoing plan success (Oatley & Johnson-Laird, 1987; Wells & Matthews, 1994). However, the correlation between extraversion and energy is context-dependent, and otten weak (Matthews et al., in press), and extraverts do not seem to be particularly prone to positive moods following demanding task performance (see Tables 1 and 2). Extraversion and energy also have rather different effects on performance. As previously discussed energy especially enhances "simple but demanding" attentional tasks, whereas extraversion seems to benefit somewhat more complex multiple verbal task performance. Hence, extraversion is not just a predisposition to the activity-readiness associated with energy, and may require a rather different adaptive explanation. Tentatively, I suggest that a wider view of adaptation is required, one which takes into account voluntary exposure to environments, in addition to performance once actually in the environment. People seek out (and create) environments which match their interests, capabilities and self-concepts (see Caspi & Bern, 1990). The position so far has been that all people are exposed to potential overload situations, but extraverts are better equipped to learn the skills required to handle them. However, extraverts may also be more motivated to enter those environments in the first place. Extraverts are disproportionately represented in "overload" jobs such as financial dealing and police work (Furnham, 1992; Kahn & Cooper, 1993). Conversely, introverts appear to prefer more reflective occupations such as being a scientist or artist (Eysenck, 1995). Real-life overload environments afford context-specific gains and losses, and perhaps also more general rewards and punishments. On the one hand, successful management of overload is likely to enhance beliefs in selfcompetence and mastery, but failure to cope leads to appraisals of loss of control, which appears to be a particular potent driver of stress outcomes (Lazarus & Folkman, 1984). Potential overload poses the adaptive question: "Is it worthwhile engaging with this environmental demand?" Empirically, the answer seems more likely to be positive in extraverts than in introverts. Indeed, extravert, appear to be especially sensitive to reward stimuli when the

430

Chapter 9

overall context offers both punishments and rewards (Nichols & Newman, 1986). The adaptive model suggests an explanation; that extraverts' more positive affeetive responses to overload and "stress" increase the likelihood of exposure to the type of environment in which they function best. Conversely, tolerance for somewhat boring, low-stimulation environments may work to the advantage of introverts. Extraversion and energy relate to qualitatively different kinds of adaptive problem. Energy, as a state variable, relates to level of committed, goal-directed activity within (a variety of) environments. Extraversion, as a trait, relates to long-term choices between environments differing in the processing characteristics they require for adaptation.

Underpinnings of adaptation: Information-processing or neural systems? Thus far, I have linked positive affectivity to a motivational function resembling the arousal-seeking generally attributed to extraverts (e.g. Eysenck & Eysenck, 1985). I have emphasised informational rather than arousing attributes of the environments concerned: extraverts find handling multiple sources of information more rewarding (or at least less aversive) than introverts. Having arrived at an adaptive explanation, we can look for processing- or physiology-based accounts of the processes underlying the overall adaptation. As previously discussed, extraverts exposed to demanding environments show rather different styles of cognition to introverts, characterised by challenge appraisals (Gallagher, 1990), use of task-focus rather than emotion-focused coping (Endler & Parker, 1990), and perceived control (see Table 3). Such cognitions may independently facilitate both performance under overload, and willingness to engage with high information flows. Challenge and task-focus are likely to elicit more task-directed effort, and beliefs in control and lack of emotion-focus protect against distracting worries (Matthews & Wells, 1996). Several unresolved issues remain. First, data are largely based on selfreport, rather than on objective behavioural measures. Extraversion research would benefit from paradigms akin to those used in research on anxiety and cognitive bias. Second, the environmental attributes which elicit positive cognitions and emotions in extraverts remain unclear. Speculatively, I suggest that it may be a combination of potential gain, elieiting challenge appraisals, and a degree of uncertainty and open-endedness, elieiting negative appraisals and emotion-focus in introverts. Real-life "stressors", such as leaving home, are perhaps more likely to have these qualities than is the performance of a well-defined task in a laboratory setting (of. discussion of Tables 2 and 3). So

G. Matthews

431

too do the social environments favoured by extraverts. Third, appraisals are influenced by pre-existing knowledge structures such as schemas and production systems (Lazarus & Smith, 1988; van Reekum & Scherer, this volume; Wells & Matthews, 1994). Little is known about how extraverts and introverts differ in their representations in memory of emotional or personally significant events and concepts. Fourth, the current account does not do justice to the dynamic interplay between emotion, cognition and exposure to demanding environments. Extraverts seem to be more attracted to environments, especially social environments, which are generally affectively uplitting (of. Watson, Clark, Mclntyre, Hamaker, 1992), but it is unclear how individual differences in cognition and affect drive such choices of exposure. In the spirit of Lazarus (1991), the cognitive-adaptive account of extraversion downplays the biological level of explanation. However, it would be premature to dismiss neural processes as underpinnings of adaptation. Plausibly, the low reticulo-cortical arousability Eysenck (1967) relates to extraversion may provide a buffer against discomfort induced by excess stimulation. However, the present account differs from Eysenck's in seeing low arousability as just one of many mechanisms contributing to adaptation to high levels of stimulation. Its principal function may be to enhance stress tolerance, rather than to increase performance competence, given the failures of arousal theory to accommodate performance data (Matthews, 1992). Similarly, the "dopaminergic" extraversion compatible with Gray's (1981) personality theory may contribute to positive affectivity and attraction to potentially rewarding overload environments. Overall, the general picture of the affective-motivational aspect of extraversion is similar to that for cognitive aspects. The general function of approaching and enjoying informationally challenging environments is likely to be supported by a variety of specific mechanisms including cognitive, stress-process mechanisms and multiple neural systems. A more detailed account of the inter-relationship of the mechanisms concerned is a task for future research. In the physiological context, Zuckerman (1991) aptly cautions that there may be no isomorphism between brain systems and personality traits. Complexity of mappings is the rule, and simplicity the exception. The adaptive perspective shares this orientation.

Causal hypotheses The adaptive explanation outlined above rests on an analysis of the correspondence between the cognitive, emotional and physiological correlates

432

Chapter 9

of extraversion and the personal qualities demanded by multiple-source information-overload situations. Hence, the explanation is transactional (Lazarus, 1991), in that extraversion is related to the person-situation interactions associated with a certain class of situations. However, the explanation is non-causal, in that it does not specify how the correlation between extraversion and the toolkit of overload-adaptive functions arises. At the present, any causal explanation is necessarily tentative, but some plausible suggestions may be advanced (Matthews, 1997b; Saldofske et al., in press). A good starting point is provided by the now compelling evidence for partial heritability of extraversion (Loehlin, 1992). A simple, feedforward hypothesis is that multiple genes code for the various (independent) functions related to extraversion. Coding for neural net properties (i.e. parameters of the cognitive architecture) influences the various processing functions which comprise the cognitive patterning of extraversion. The person is predisposed to extraversion to the extent that the majority of genes code for "extraverted" values of functions. The gross characteristics of extraversion then derive from the situationdependent learning influence~ by elementary processing functions. For example, a child who has good verbal STM, reacts quickly to complex demands, and tolerates stimulation well is likely to develop skills for effective performance in socially demanding and other overload situations. Success is motivating, so the extraverted child is likely to actively seek such situations. Conversely, the genetically-programmed functions of the introverted child fail to support effective skill acquisition, the child finds overload situations threatening or unfulfilling, and is motivated to avoid them. Characteristics of extraversion-introversion are likely to reflect both skills and interests. We are most likely to appraise someone as sociable if they both like interaction with other people and are good at it. Thus, the core extraverted qualities such as impulsivity, sociability and assertiveness are distributed across the skills and interests biased by the genotype. This initial causal account is simplistic in several ways. Naturally, any genetic effect is statistical in nature; both elementary processing and physiological functions and the skills they feed into are also environmentally influenced. More subtly, personality development is likely to be affected dynamically through various types of interaction with the environment (Caspi & Bern, 1990). In particular, success or failure in a particular context is likely to influence future exposure to the context, and opportunities for further skill acquisition. Over time, extraverts have more opportunity for learning social skills through their greater exposure to social interaction.

G. Matthews

433

Hence, skills influence real-world adaptive behaviours, but the outcomes of those behaviours feed back into skills. Furthermore, individuals build up knowledge in long-term memory relating to motivations, preferences and beliefs about of self-efficacy in the environments concerned. This knowledge base is a primary determinant of emotional response (Wells & Matthews, 1994). Extraverts develop beliefs about their competence within social and other demanding situations which encourage them to enter those situations. Typically, there will be a degree of positive feedback, such that people tend to seek out the types of environment congruent with their skills and selfknowledge, subject to the constraints opposed by other factors such as social pressures, abilities etc. (cf. Furnham, 1992). Figure 7 sketches out the tentative causal model developed here. Motivation, self-beliefs and emotion are packaged together under "knowledge" although a more detailed model would separate these constructs (of. Mayer et al., this volume). The stability of extraversion may reflect not only stability of genetically coded functions but the tendency for extraverts to allocate their time to high information load environments which maintain "extraverted" skills. The model predicts that influencing skills (in the broad sense used here) should also influence personality. Thus, consistent with the feed-forward part of the model, extraverts appear to be attracted to sales work, and, in some studies, perform better at it than introverts (Barrick & Mount, 1993). However, sales training elevates extraversion (TumbuU,

"'~. Arousal ~ .-" A dap tationiTrait \ J functions[)~ 'Knowledge' ~ n/ets ~ ~~\? ~ ~" ~__-...... Emotion ~, Information- ,~ processing ~_/~~ /

\

\

Genes

Neural

~

Skills

, -"

"--

- Motivations - Efficacy

~ J

I

beliefs

/

/

Early learning

Real-world adaptive behaviours

Figure 7. A tentative causal model of the neural and cognitive bases for individual differences in adaptation (adapted from Saklofske et al., in press).

434

Chapter 9

1976), presumably because the skills acquired generalise to demanding environments other than interaction with customers. Similarly, social introversion is a risk factor for depression, but introversion rises during a depressive episode (Barnett & Gotlib, 1988). Limited social skills may increase vulnerability to depression, which in turn disrupts the translation of social competence into actual performance (Wells & Matthews, 1994). Conclusions

Extraversion-introversion is a surprisingly elusive construct. Although robust psychometrically, it is difficult to identify its key processing components. Like a mosaic, the finer the level of analysis, the more difficult it is to see the larger picture, i.e. a coherent integration of the cognitive, emotional, motivational and physiological aspects of the trait. The starting point for a cognitive science analysis is the development of informationprocessing models linking cxtraversion to behavioural response. Useful cognitive mini-theories have been developed for many of the independent processing functions which relate to extraversion, which jointly define a "cognitive patterning". Information-processing analyses have also contributed to better understanding of contextual factors moderating extraversion effects, such as level of arousal and motivational signals. It is likely that extraversion relates to individual differences both in cognitive architecture and to strategy. Connectionism provides a powerful tool for modelling architectural differences between extraverts and introverts; differences which may vary with arousal level (Matthews & Harley, 1993). It may become possible to link extraversion to individual differences in neural net functioning, with a greater emphasis on cortical processes than provided by current psychobiological theories of personality and sub-cortical functioning. Explaining the cognitive patterning as a whole requires an adaptive perspective. The central question is how the various cognitive correlates of extraversion help the individual to function successfully as an extravert. I have suggested that the processing characteristics associate~ with extraversion provide the foundation for the acquired skills needed in certain "overload" environments, those associated with multiple information sources and social interaction. Physiological and emotional characteristics associated with extraversion may also serve the same overall adaptation, loosely by facilitating stress tolerance. More precisely, the positive affectivity of extraverts may have the motivational function of raising the likelihood of engagement with environments characterised by overload but potential reward

G. Matthews

435

also. Conversely, correlates of introversion may support adaptation to lowstimulation environments offering opportunities for reflection. Causal bases for the adaptive qualities associated with extraversion and introversion are speculative. However, it is plausible that, first, genetic bases for the various component processes feed forward into phenotypic extraversion, and, second, extraversion is shaped also by person-environment interaction as the person succeeds or fails in acquiring the skills for handling overload. References

Amelang, M., & Ullwer, U. (1991). Correlations between psychometric measures and psyehophysiological as well as experimental variables in studies on extraversion and neuroticism. In J. Strelau & A. Angleitner (Eds.), Explorations in temperament. New York: Plenum. Argyle, M., & Lu, L. (1990). Happiness and social skills. Personality and Individual Differences, 11, 1255-1262. Bakan, P. (1959). Extroversion-introversion and improvement in an auditory vigilance task. British Journal of Psychology, 50, 165-178. Barnett, P. A., & Gotlib, I. H. (1988). Psychosocial functioning and depression: Distinguishing among antecedents, concomitants, and consequences. Psychological Bulletin, 104, 97-126. Barrick, M. R., & Mount, M. K. (1993). Autonomy as a moderator of the relationships between the Big Five personality dimensions and job performance. Journal of Applied Psychology, 78, 111-118. Bartussek, D., Becker, G., Diedrich, O., Naumann, E., & Maier, S. (1996). Extraversion, neuroticism, and event-related potentials in response to emotional stimuli. Personality and Individual Differences, 20, 301-312. Battig, K., & Buzzi, R. (1986). Effect of coffee on the speed of subject-paced information processing. Neuropsychobiology, 16, 126-130. Bermudez, J., Perez, A. M., & Padilla, M. (1988). Extraversion and task properties as determinants of incidental recall. European Journal of Personality, 2, 57-66. Blake, M. J. F. (1971). Temperament and time of day. In W. P. Colquhoun (Ed.), Biological rhythms. London: Academic. Brandst~tter, H. (1994). Well-being and motivated person-environment fit: A time-sampling study of emotions. European Journal of Personality, 8, 75-94. Brebner, J. (1983). A model of extraversion. Australian Journal of Psychology, 35, 349-360.

436

Chapter 9

Caspi, A., & Born, D. J. (1990). Personality continuity and change across the life course. In L. A. Pervin (Ed.), Handbook of personality: Theory and research. New York: Guilford. Cohen, J. D., Servan-Schreiber, D., & MeCldland, J. L. (1992). A parallel distributed approach to automaticity. American Journal of Psychology,, ] 05, 239-269. Costa, P. T., Jr., Somerfidd, M. R., & MeCrae, R. R. (1996). Personality and coping: A reconceptualization. In M. Zeidncr & N. S. Endler (Eds.), Handbook of coping: Theory, research, applicaaons. New York: Wiley. Davies, D. R., & Hockey, G. R. J. (1966). The effects of noise and doubling the signal frequency on individual differences in visual vigilance performance. British Journal of Psychology, 57, 381-389. Davies, D. R., & Parasuraman, R. (1982). The psychology of vigilance. London: Academic Press. Dempster, F. N. (1981). Memory span: Sources of individual and developmental differences. Psychological Bulletin, 89, 63-100. Depue, R. A. (1995). Neurobiological factors in personality and depression. European Journal of Personality, 9, 413-439. Derryberry, D., & Reed, M. A. (1994). Temperament and attention: Orienting toward and away from positive and negative signals. Journal of Personality and Social Psychology, 66, 1128-1139. Dickman, S. J., & Mayer, D. E. (1988). Impulsivity and speed-accuracy tradeoffs in information processing. Journal of Personality and Social Psychology, 54, 274-290. Dorn, L., & Matthews, G. (1995). Prediction of mood and risk appraisals from trait measures: Two studies of simulated driving. European Journal of Personality, 9, 25-42. Dunne, M. P., & Hartley, L. R. (1985). The effects of scopolamine upon verbal memory: Evidence for an attentional hypothesis. Acta Psychologica, 58, 205-217. Endler, N., & Parker, J. (1990). Multi dimensional assessment of coping: A critical review. Journal of Personality and Social Psychology, 58, 844854. Eysenck, H. J. (1957). The dynamics of anxiety and hysteria. London: Routledge & Kegan Paul. Eysenck, H. J. (1995). Creativity as a product of intelligence and personality. In D. H. Saklofske & M. Zeidner (Eds.), International handbook of personality and intelligence. New York: Plenum.

G. Matthews

437

Eysenck, H. J., & Eysenck, M. W. (1985). Personality and individual differences: A natural science approach. New York: Plenum. Eysenck, M. W. (1982). Attention and arousal: Cognition and performance. New York: Springer. Eysenck, M. W., & Eysenck, M. C. (1979). Memory scanning, introversionextraversion, and levels of processing. Journal of Research in Personality, 13, 305-315. Eysenck, S. B. G., Eysenck, H. J., & Barrett, P. (1985). A revised version of the Psychoticism scale. Personality and Individual Differences, 6, 2129. Fagerstrom, K. O., & Lisper, H. O. (1977). Effects of listening to car radio, experience, and personality of the driver on subsidiary reaction time and heart rate in a long-term driving task. In R. R. Mackie (Ed.), Vigilance. New York: Plenum. Ferguson, E., Matthews, G., & Cox, T. (submitted). The Appraisal of Life Events (ALE) Scale: Reliability and validity. Furnham, A. (1981). Personality and activity preference. British Journal of Social Psychology, 20, 57-60. Furnham, A. (1992). Personality at work: The role of individual differences in the workplace. London: Routledge. Gallagher, D. J. (1990). Extraversion, neuroticism and appraisal of stressful academic events. Personality and Individual Differences, 11, 10531058. Gluckman, J. P, Dember, W. N., & Warm, J. S. (1988). Capacity demand in dual-task monitoring of simultaneous and successive vigilance tasks. In Proceedings of the Human Factors and Ergonomics Society 40th Annual Meeting. Santa Monica, CA: Human Factors and Ergonomics Society:. Gray, J. A. (1981). A critique of Eysenck's theory of personality. In H. J. Eysenck (Ed.), A model for personality. Berlin: Springer. Gray, J. A. (1987). The psychology of fear and stress (2nd ed.). Cambridge: Cambridge University Press. Gray, J. A. (1991). Neural systems, emotion and personality. In J. Madden IV (Ed.), Neurobiology of learning, emotion and affect. New York: Raven Press. Harley, T. A. (1993). Connectionist approaches to language disorders. Aphasiology, 7, 221-249.

438

Chapter 9

Harley, T. A., & Matthews, G. (1992). Interactive effects of extraversion, arousal and time of day on semantic priming: Are they pre-lexical or post-lexical? Personality and Individual Differences, 13, 1021- 1029. Hockey, G. R. J. (1984). Varieties of attentional state: The effects of the environment. In R. Parasuraman & D. R. Davies (Eds.), Varieties of attention. New York: Academic. Humphreys, M. S., & Revelle, W. (1984). Personality, motivation and performance: A theory of the relationship between individual differences and information processing. Psychological Review, 91, 153-184. Kahn, H., & Cooper, C. L. (1993). Stress in the dealing room: High performers underpressure. London: Routledge. Koelega, H. S. (1992). Extraversion and vigilance performance: 30 years of inconsistencies. Psychological Bulletin, 112, 239-258. Kumar, D., & Kapila, A. (1987). Problem solving as a function of extraversion and masculinity. Personality and Individual Differences, 8, 129-132. Larsen, R. J., & Ketelaar, T. (1991). Personality and susceptibility to positive and negative emotional states. Journal of Personality and Social Psychology, 55, 132-140. Lazarus, R. S. (1991). Emotion and adaptation. Oxford: Oxford University Press. Lazarus, R. S. & Folkman, S. (1984). Stress, appraisal and coping. New York: Springer. Lazarus, R. S., & Smith, C. A. (1988). Knowledge and appraisal in the cognition-emotion relationship. Cognition and Emotion, 2, 281-300. Loehlin, J. C. (1992). Genes and environment in personality development. Newbury Park, CA: Sage. Matthews, G. (1985). The effects of extraversion and arousal on intelligence test performance. British Journal of Psychology, 76, 479-493. Matthews, G. (1986). The interactive effects of extraversion and arousal on performance: Are creativity tests anomalous? Personality and Individual Differences, 7, 751-761. Matthews, G. (1989). Extraversion and levels of control of sustained attention. Acta Psychologica, 70, 129-146. Matthews, G. (1989). The factor structure of the 16PF: Twelve primary and three secondary factors. Personality and Individual Differences, 10, 931-940.

G. Matthews

439

Matthews, G. (1992). Extraversion. In A. P. Smith & D. M. Jones (Eds.), Handbook of human performance. Vol. 3: State and trait. London: Academic. Matthews, G. (1997a). The Big Five as a framework for personality assessment. In N. Anderson & P. Herriot (Eds.), International handbook of selection and appraisal (2nd ed.). London: Wiley. Matthews, G. (1997b). Intelligence, personality and information-processing: An adaptive perspective. In W. Tomic & J. Kingsma (Eds.), Advances in cognition and educational practice (Vol. 4). Greenwich, CT: JAI Press. Matthews, G., & Amelang, M. (1993). Extraversion, arousal theory and performance: A study of individual differences in the EEG. Personality and Individual Differences, 14, 347-364. Matthews, G., & Chappelow, J. E. (1986). Extraversion, arousal and attentional resources. Paper presented to the Third European Conference on Personality Psychology, Gdansk, Poland, Matthews, G., & Davies, D. R. (in press). Arousal and vigilance: The role of task factors. In R. B. Hoffman, M. F. Sherrick, M. F., & Warm, J. S. (Eds.), Integrating perception across psychology: Perspectives on motivation, choice, and individual differences. Washington, DC: American Psychological Association. Matthews, G., Davies, D. R., & Holley, P. J. (1990b). Extraversion, arousal and visual sustained attention: The role of resource availability. Personality and Individual Differences, I 1, 1159-1173. Matthews, G., Davies, D. R., & Lees, J. L. (1990b). Arousal, extraversion, and individual differences in resource availability. Journal of Personality and Social Psychology, 59, 150-168. Matthews, G., & Deary I. (m press). Personality traits. Cambridge: Cambridge University Press. Matthews, G. & Dorn, L. (1995). Personality and intelligence: Cognitive and attentional processes. In D. H. Saklofske & M. Zeidner (Eds.), International handbook of personality and intelligence. New York: Plenum. Matthews, G., & Gilliland, K. (submitted). The personality theories of H. J. Eysenck and J. A. Gray: A comparative review. Matthews, G., Jones, D. M., & Chamberlain, A. G. (1989). Interactive effects of extraversion and arousal on attentional task performance: Multiple resources or encoding processes? Journal of Personality and Social Psychology, 56, 629-639.

440

Chapter 9

Matthews, G., Jones, D. M., & Chamberlain, A. G. (1990a). Refining the measurement of mood: The UWIST Mood Adjective Checklist. British Journal of Psychology, 81, 17-42. Matthews, G., Jones, D. M., & Chamberlain, A. G. (1992). Predictors of individual differences in mail coding skills and their variation with ability level. Journal of Applied Psychology, 77, 406-418. Matthews, G., Joyner, L., Gilliland, K., Huggins, J., Falconer, S., & Campbell, S. E. (in press). Validation of a comprehensive stress state questionnaire: Towards a state Big Three? In I. Mervielde, I. J. Deary, F. De Fruyt, & F. Ostcndorf (Eds.), Personality psychology in Europe (Vol. 7). Tilburg: Tilburg University Press. Matthews, G., & Harley, T. A. (1993). Effects of r and self-report arousal on semantic priming: A r approach. Journal of Personality and Social Psychology, 65, 735-756. Matthews, G., & Margetts, I. (1991). Self-report arousal and divided attention: A study of performanc~ operating characteristics. Human Performance, 4, 107-125. Matthews, G., & Wells, A. (1996). Attentional processes, coping strategies and clinical intervention. In M. Zeidner & N. S. Endler (Eds.), Handbook of coping: Theory, research, applications. New York: Wiley. Matthews, G., & Westerman, S. J. (1994). Energy and tension as predictors of controlled visual and memory search. Personality and IndivMual Differences, 17, 617-626. McClelland, J. L., & Rumelhart, D. E. (1981). An interactive model of context effects in letter perception: Part 1. An account of basic findings. Psychological Review, 88, 375-407. Meyer, G. J., & Shack, J. R. (1989). Structural convergence of mood and personality: Evidence for old and new directions. Journal of Personality and Social Psychology, 57, 691-706. Mohamed, A. A. R. (1996). Stress processes in British and overseas students. Unpublished doctoral dissertation, University of Dundee. Munroe, L. L., Dawson, M. E., Sehell, A. M., & Sakai, L. M. (1987). Electrodermal lability and rapid vigilance decrement in a degraded stimulus continuous performance test. Journal of Psychophysiology, 1, 249-257. Neely, J. H. (1991). Semantic priming effects in visual word recognition: A selective review of current findings and theories. In D. E. Besner & G. Humphreys (Eds.), Basic processes in reading. Hillsdale, NJ: Erlbaum.

G. Matthews

441

Nciss, R. (1988). Rcconccptualizing arousal: Psychobiological states in motor performance. Psychological Bulletin, 103, 345-366. Nichols, S. L., & Newman, J. P. (1986). Effects of punishment on response latency in extraverts. Journal of Personality and Social Psychology, 50, 624-630. Norman, D. A., & Shallice, T. (1985). Attention to action: Willed and automatic control of behaviour. In R. J. Davidson, G. E. Schwartz & D. Shapiro (Eds.), Consciousness and self-regulation: Advances in research (Vol. 4). New York: Plenum. Oatlcy, K., & Johnson-Laird, P. (1987). Towards a cognitive theory of emotions. Cognition and Emotion, 1, 29-50. Parasuraman, R., Warm, J. S., & Dembcr, W. N. (1987). Vigilance: Taxonomy and utility. In L. Mark, J. S. Warm, & R. L. Huston (Eds.), Ergonomics and human factors: Recent research. New York: Springer. Pashler, H. (1994). Dual-task interference in simple tasks: Data and theory. Psychological Bulletin, 116, 220-244. Posner, M. I., & Snydcr, C. R. R. (1975). Attention and cognitive control. In R. L. Solso (Ed.), Information processing and cognition: The Loyola Symposium. Hillsdale, NJ: Erlbaum. Posncr, M. I., & Raichle, M. E. (1994). Images of mind. New York: Scientific American Library. Rcvellc, W. (1993). Individual differences in personality and motivation: ~Non-cognitive' determinants of cognitive performance. In A. Baddcley & L. Weiskrantz (Eds.), Attention: Selection, awareness and control. Oxford: Oxford University Press. Rcvelle, W., Humphrcys, M. S., Simon, L., & Gilliland, K. (1980). The interactive effect of personality, time of day and caffeine: A test of the arousal model. Journal of Experimental Psychology: General, 109, 131. Robbins, T. W. (1986). Psychopharmacological and neurobiological aspects of the energctics of information processing. In G. R. J. Hockey, A. W. K. Gaillard, & M. G. H. Coles (Eds.), Energetics and human information processing. Dordrecht: Martinus Nijhoff Rusting, C. L., & Larsen, R. J. (1997). Extraversion, neuroticism, and susceptibility to positive and negative affect: A test of two theoretical models. Personality and Individual Differences, 22, 607-612. Saklofske, D. H., Matthews, G., Zeidner, M., Deary, I. J., Austin, E. J., & Steinberg, R. J. (in press). The intelligence-personality interface: Prospects for integration. In I. Mervielde, I. J. Deary, F. De Fruyt, & F.

442

Chapter 9

Ostendorf (Eds.), Personality psychology in Europe (Vol. 7). Tilburg: Tilburg University Press. See, J. E., Howe, S., Warm, J. S., & Dember, W. N. (1995). Meta-analysis of the sensitivity decrement in vigilance. Psychological Bulletin, 117, 230-249. Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory. Psychological Review, 84, 127-190. Singh, I. L., Molloy, R., & Parasuraman, R. (1993). Individual differences in monitoring failures of automation. Journal of General Psychology, 120, 357-373. Smith, B. D. (1983). Extraversion and electrodermal activity: Arousability and the inverted U. Personality and Individual Differences, 4, 411-419. Spilsbury, G. A. (1992). Complexity as a reflection of the dimensionality of a task. Intelligence, 16, 31-45. Thayer, R. E. (1978). Toward a psychological theory of multidimensional activation (arousal). Motivation and Emotion, 2, 1-34. Thayer, R. E. (1989). The biopsychology of mood and arousal. Oxford: Oxford University Press. Thayer, R. E. (1996). The origin of everyday moods. New York: Oxford University Press. Tumbull, A. (1976). Selling and the salesman: Prediction of success and personality change. Psychological Reports, 38, 1175-1180. Watson, D., & Clark, L. A. (1992). On traits and temperament: General and specific factors of emotional experience and their relation to the fivefactor model. Journal of Personality, 60, 441-476. Watson, D., Clark, L. A., Mclntyre, C. W., & Hamaker, S. (1992). Affect, personality, and social activity. Journal of Personality and Social Psychology, 6, 1011-1025. Weinman, J. (1987). Non-cognitive determinants of perceptual problemsolving strategies. Personality and Individual Differences, 8, 53-58. Wells, A., & Matthews, G. (1994). Attention and emotion: A clinical perspective. Hove: Edbaum. Wickens, C. D. (1984). Processing resources in attention. In R. Parasuraman & D. R. Davies (Eds.), Varieties of attention. New York: Academic. Zuckerman, M. (1991). Psychobiology of personality. Cambridge: Cambridge University Press.

Cognitive Science Perspectives on Personality and Emotion - G. Matthews (Editor) 9 1997 Elsevier Science B.V. All fights reserved. CHAPTER 10

Motivational and Attentional Components of Personality Douglas Derryberry and Marjorie A. Reed

Contemporary theorizing can be viewed as approaching personality from two general perspectives. The cognitive perspective emphasizes structural representations of the self and the world that support appraisal, attributional, and other cognitive processes considered central to personality. In contrast, the biological perspective approaches personality in terms of evolved neural systems whose activity underlies individual differences in motivational, emotional, and attentional processes. Although both approaches have made progress in recent years, it is fair to say that they are developing in parallel with relatively little integration between them. One of the challenges facing the study of personality is to build bridges that help integrate these complementary approaches. In this chapter, we discuss some potential links between biological and cognitive approaches. We suggest that the models and methods developed within cognitive science may prove particularly helpful in facilitating such an integration. Our optimism is based on the rather simple idea that cognitive science constructs are framed at an analytic level intermediate to those developed by cognitive and biological approaches. This places such models in a useful intermediate position for linking underlying brain processes to complex cognition. More specifically, cognitive science models aim to specify in detail the information processing functions of separable systems related to attention, perception, memory, and so on. In most cases, theorists attempt to develop models that are compatible with what is currently known about the brain, and that can be decomposed into component processes that might be linked to specific neural systems. This compatibility between cognitive science and biological approaches is perhaps best seen in the rapid developments within the field of "cognitive neuroscience" during the last decade (Gazzaniga, 1995). At the same time, an implicit assumption within cognitive science is that their detailed, componential models of basic processes will prove valuable to understanding the more complex cognitive and emotional processes involved in personality. As evident in the previous chapters, this assumption appears to be shared by many personality and social researchers, who are drawing more and more on constructs and

Chapter 10

444

methods from cognitive science. As these trends continue, we should see an accelerating linkage between cognitive science, biological, and cognitive approaches, and thus increasingly integrated models of personality. This chapter consist of three sections. We begin with an overview of biological approaches to personality, and suggest that in their emphasis on the motivational control of attention, these models provide a foundation for understanding more complex cognitive processes. Next, we examine the construct of attention in light of recent developments within cognitive science, and present several studies relating motivational and attentional processes to the dimension of trait anxiety. In the final section, we explore some ways in which such motivated attentional processes may contribute to more complex personality processes involving appraisals, attributions, and representational development. Although the chapter's organization will emphasize bottom-up influences of motivation on cognition, we should make it clear that the reciprocal effects of cognition on motivation are equally important, and hope that the readers will keep such top down effects in mind.

Biological Approaches to Personality Modem biological approaches to personality are based on the notion that major personality dimensions arise from individual differences in the reactivity of underlying neural systems. These "temperament" approaches focus primarily on subcortical systems related to arousal, motivation, and attention, with relatively little emphasis on the cortical mechanisms related to cognition. In describing temperament models, we emphasize the systems related to motivation and attention, because these provide the most direct links to higher level cognition.

Motivational systems Among the most central of the neural systems emphasize~ by temperament approaches arc those involved in incentive motivation. These systems arc based within the brain's limbic regions, which places them in an intcrmodiatr position relative to the cognitive functions of the cortex and the behavioral functions of the brainstcm. They receive simple perceptual inputs from the thalamus, as well as more complex perceptual and conceptual information from the cortex. Based upon these inputs, the limbic incentive systems regulate brainstem mechanisms that serve motor, autonomic, and attcntional functions, thereby promoting an adaptive response to the situation.

D. Derryberry and M.A. Reid

445

One system is specialized for detecting information that predicts positive events and for mobilizing approach behavior, and has been discussed in terms of a "behavioral activation system" (Gray, 1987a), a "behavioral facilitation system" (Depue & lar 1989), and an "expectancy-foraging system" (Panksepp, 1992a). Although formulations vary, the basic idea is that upon detecting a signal that predicts a positive event, the limbir circuits activate dopaminergir projections that extend from the brainstem's ventral tegmental area to the nucleus aceumbens. This dopaminergic activation serves to facilitate approach responses within the nucleus ar162 and can promote approach behavior directed toward rewarding stimuli or active avoidance behavior toward non-punishing stimuli. The resulting emotional state has been described in terms of "hope" or "anticipato.ry eagerness" (given anticipated reward) and "relief' (given anticipated non-pumshment) (Gray, 1987b). Some theorists suggest that variability in this system's reactivity underlies a personality dimension of "positive emotionality" which increases in strength as one moves from the introverted to the extraverted pole of the extraversion dimension (Larsen & Ketelaar, 1989; Watson & Clark, 1992). Others propose a rotated dimension of "impulsivity", which in terms of the two-dimensional space defined by extraversion and neuroticism, increases diagonally from the stable introvert to the neurotic extravert quadrant (Gray, 1987b; Wallace, Newman & Baehorowski, 1991). Complementing this positive system, many authors have discussed neural mechanisms related to negative incentive motivation. Examples include Gray's (1982) "behavioral inhibition system", Panksepp's (1982, 1986a) "fear system", and Gilbert and Trower's (1990) "defense system". In Gray's model, the negative incentive system is centered upon the hippoeampus and responds to novel signals and to signals that predict punishment or non-reward (Gray, 1982, 1987a, 1994). Upon detecting such an input, the system inhibits ongoing motor behavior to promote passive avoidance, increases arousal, and directs attention toward relevant information in the environment. In emotional terms, these projections set up a state of anxiety (given novelty or anticipated punishment) or frustration (given anticipated non-reward). Other researchers have emphasized circuitry centered upon the central nucleus of the amygdala (Davis, 1992; LeDoux, 1995). This nucleus receives inputs from the hippocampus, thalamus, and cortex, and orchestrates fearful behavior via its widespread projections to motor and autonomic circuits within the brainstem. In addition, the central amygdala has connections to reticular and cortical circuits involved in attentional functions. Individual variability in these systems is oRen thought

446

Chapter 10

to undcrly to undcrly a dimension of "negative cmotionality" that increases as one moves from low to high neuroticism (Larscn & Kctclaar, 1989; Watson & Clark, 1992). Alternatively, others have argued for a rotated dimension of "anxiety" that increases from stable cxtravcrsion to neurotic introversion (Gray, 1987b, Wallace ct al., 1991). While the positive and negative incentive systems have received the most attention, it is generally believed that systems related to other motivational processes arc also central to personality. One of these is related to aggression and has been discussed in terms of a "rage system" (Pankscpp, 1982, 1986a) and a "fight/flight" system (Gray, 1987b, 1994). The crucial circuitry appears to be focused on the central grey region of the brainstcm, which activates aggressive behaviors by means of projections to lower level motor and autonomic mechanisms. According to Gray (1994), this fight-flight system responds to unconditioned non-reward to promote anger and aggression, but can also promote panic and escape behavior given unconditioned punishment. Panksepp's model emphasizes inhibitory projections from the ventromedial hypothalamus to the central grey that allow for the control of aggression and the expression of prosocial behaviors (see below). Other inhibitory controls arise from serotonin projections from the brainstem (Spoont, 1992) and from peptide projections from limbic and cortical regions. Potentiating influences appear to arise from circulating androgens. Individual differences in these circuits appear related to the aggressive facets of the psychoticism dimension (Gray, 1987b) of three factor models and to the hostile pole of the agreeableness dimension of five factor models. A related motivational system is involved in nurturant and affectionate behavior. The underlying circuitry remains unclear, but appears to utilize projections from the cortex and amygdala to various hypothalamic and brainstem regions. Panksepp (1986b) proposed that opiate peptide projections enable the ventromedial hypothalamus to inhibit aggressive tendencies and thereby promote friendly, trusting, and helpful behaviors between members of a species. Such reciprocal relations between the mechanisms underlying prosocial and aggressive behavior fit well with the two poles of the agreeableness-hostility dimension. Within caregiving situations, Panksepp (1992b) suggests that oxytocin pathways may evoke warm feelings of acceptance and nurturance that promote social bonding. A similar approach can be found in MacDonald's (1992) discussion of an "affeetional system". MacDonald suggests that this specialized reward system evolved to facilitate close family relationships by promoting feelings of warmth. In addition,

D. Derryberry and M.A. Reid

447

warmth promotvs feelings of empathy, identification with the parents, and adoption of the parents' values. This system is thought to be related to dimensions such as agreeableness-hostility and Cloningcr's (1987) reward dependence. It can be seen that temperament approaches attempt to identify motivational systems that contribute, either alone or in combination, to the major personality dimensions. An underlying assumption is that the systems have been selected through evolutionary history to serve the adaptive needs of our species. In the case of defensive motivation, for example, the relevant systems are shaped to help the individual cope with dangerous situations that threaten a loss of life, health, status, and so on. Such coping potential is enhanced through a mechanism that quickly recognizes these threats and sets up an appropriate response involving inhibited approach and facilitated avoidance behavior. Although such a general defensive capacity may be shared by all members of a species, an evolutionary perspective would nevertheless emphasize variability across individuals. On the one hand, while individuals high in fearfulness may benefit in dangerous situations, low fearfulness can also prove advantageous in less threatening or appetitive contexts. Since adaptive value arises from relatively low as well as high fearfulness, variability across the population would tend to be preserved. In addition, individual differences may arise from structural variations within the motivational system itself. Marks and Neese (1994) have suggested that while a general anxiety system evolved to deal with a range of nonspecific threats, the system may have differentiated to promote coping given particular kinds of dangers. Evidence of such differentiation can be found in human phobias, which tend to be based upon evolutionarily significant stimuli such as heights, blood injury, social rejection, and contamination. Thus, individuals may differ not only in their tendencies to experience general anxiety, but also in the kinds of events that become the focus of their fear. To better understand the adaptive value of the motivational systems, it is worth emphasizing that the systems function by regulating attention as well as response processes. For example, Gray's (1982) behavioral inhibition system helps the individual cope with threat by directing attention to relevant environmental information. Attention facilitates the processing of information related to the threat, and thus promotes a more efficient situational evaluation and a more appropriate response. Similarly, MacDonald's (1992) affectional system can promote family cohesiveness by generating feelings of warmth, and at the same time, by directing the child's attention to the beliefs and values of the parents. In addition to such effects upon ongoing perceptual

448

Chapter 10

processing, motivated attention appears crucial in the long range storage of information in memory. Research from psychology and neuroscience suggests that attention facilitates the stabilization of cortical synapses involved in a wide range of learning processes. By regulating attention, the motivational systems can in a sense function as learning mechanisms, stabilizing synapses that are important to ongoing motivational functions. As a result, the individual can develop representations of information relevant within threatening and appetitive situations, and these representations should enhance the future functioning of the motivational systems (Derryberry & Reed, 1994b, 1996).

Attentional systems Given the important role of attention in motivated behavior, some theorists have focused on additional attcntional systems as central to personality. Several of those mechanisms arc components of the "reticular activating system" that ascends from the brainstcm to the cortex. Posncr has discussed a "vigilance" system involving norcpincphrinc projections from the locus cocrulcus to the cortex (Posncr & Raichle, 1994; Posncr & Rothbart, 1991). This mechanism is thought to be involved in the tonic maintenance and phasic adjustments in general alertness. The vigilance system also facilitates a "posterior attcntional system", distributed across the parietal cortex, thalamus, and superior collicuhs, that controls the orienting of attention from one spatial location to another. Tucker has described a "tonic activation" system involving dopaminc projections from the ventral tcgmcntal area to object processing pathways in the left hemisphere (Tucker & Dcrrybcrry, 1992; Tucker & Williamson, 1984). This mechanism is thought to facilitate defensive behavior by focusing attention on important stimuli and preventing distraction. It is important to note that the vigilance and tonic activation systems arc recruited by higher level motivational systems in carrying out their adaptive functions. If individuals differ in these attcntional systems, they may also differ in the efficiency of their motivated behavior. For example, anxious individuals with strong tonic activation may be effective in focusing on the details of stressful situations, but may also tend to neglect more peripheral information. In contrast, anxious individuals with weaker tonic activation be better able to attend to peripheral information, but less able to take advantage of central information. Another important attcntional system is Posncr's "anterior attcntional system". Located within the frontal and r regions of the cortex, this

D. Derryberry and M.A. Reid

449

is viewed as an executive system responsible for regulating the posterior spatial attentional system and controlling attention to semantic information (Posner & Raiehle, 1994; Rothbart, Derryberry & Posner, 1994). Like the reticular attentional mechanisms, the executive system receives extensive inputs from underlying limbic circuits and should thus influence the efficiency of motivational functions. Moreover, Rothbart and Posner suggest that the anterior system underlies the conscious, "effortful control" of behavior through which the individual can regulate more reactive motivational functions. Examples include the person who can resist temptation or delay gratification by directing attention away from the rewarding properties of an attractive object, or who can control their anxiety by directing attention toward the beneficial aspects of a stressful situation. Such high level skills may regulate multiple motivational functions, and probably contribute to a number of personality dimensions. In general, however, effortful control seems particularly relevant to Tellegen's "constraint" dimension and to the "conscientiousness" dimension of five factor models (Ahadi & Rothbart, 1994). It should be clear from this brief overview that temperament models are only beginning to understand the contribution of motivational and attentional systems to personality. Fortunately, these models are being investigated on multiple fronts within psychology and the neuroseienees. Within the neuroseienees, relevant biological systems are being studied in terms of their anatomy, physiology, chemistry, genetics, and so on, and progress that will surely accelerate in the coming years (Gazzaniga, 1995). Within developmental psychology, advances have been made in mapping the onset and stability of temperament variables during infancy and early childhood (Rothbart & Bates, in press). Additional research is relating constructs such as negative emotionality and effortful control to complex developmental processes involving attachment, empathy, and conscience (Rothbart, Ahadi & Hershey, 1994). Within clinical psychology, temperament constructs have provided new perspectives for viewing a range of disorders. Problems involving anxiety, depression, irnpulsivity, and schizophrenia are being approached in terms of overreactive (or underreaetive) motivational and attentional systems (Fowles, 1994; Gray, 1994), and a greater appreciation of the continuity between abnormal and normal personality is arising. Given this interdisciplinary enterprise, it is surprising that the there has not been more eommumeation between the biological and cognitive approaches to personality. There are of course many historical and conceptual reasons for this separation. But in part, it appears to be due to the

450

Chapter 10

assumption that these approaches focus on different domains of personality. The biological models arc usually viewed as emphasizing subcortical processes related to motivation, emotion, and behavior, whereas the cognitive models emphasize more advanc~ cortical processes related to appraisals, attributions, and coping. What is not always recognized, however, is that both approaches share common interests related to motivation, memory, and especially attention. In the next section, we discuss attention in more detail, and consider its potential as a common ground between the biological and cognitive approaches.

Assessing Attentional Processes in Anxiety As mentioned earlier, biological approaches suggest that motivational systems function by regulating attention as well as behavior. Since attention is involved in many complex cognitive functions, its control by motivation is likely to influence appraisals, attributions, and other cognitive processes considered central to personality. For a number of years, however, the biological and cognitive approaches were unable to explore these connections. The complexity of neural systems involved in attention led many biological researchers to focus on simpler behavioral functions. At the psychological level, problems persiste~ in conceptualizing and measuring attention, and in relating it to complex cognitive processes. Fortunately, recent information processing models developed within cognitive psychology have provided more precise conceptualizations of attention that provide a basis for relating neural systems to more complex cognition. Researchers are making more and more use of these conceptualizations to link temperament and cognitive variables (for reviews, see Eyscnck, 1992; Wells & Matthews, 1994). These conceptualizations are based on a simple, metaphorical view of attention as an internal "spotlight". This attcntional spotlight can be directed toward different processing pathways, such as those dealing with spatial locations, perceptual objects, and conceptual information. When aligned with one of these pathways, attention facilitates the selected information. Such facilitation promotes the linkage of the selected information to response mechanisms and its integration with other information. In addition, attention promotes conscious awareness of the selected information and its storage in memory (Posner & Raichle, 1994).

D. Derryberry and M.A. Reid

451

Component attentional operations Although clearly an oversimplification, this spotlight view of attention is being elaborated along a number of lines. Some of the most important advances have been made in differentiating the spotlight in terms of its component operations. Posncr and his colleagues have identified three component operations involved in orienting attention to different spatial locations: attention must first "disengage" from its current location, "move" to the next location, and then "engage" or enhance that location (Posner & Raichlc, 1994). Other researchers have investigated adjustments in the size of the spotlight (Bennett, Waterman, Scarpa & Castiello, 1995; Erikscn & Yeh, 1985). The breadth of attention can be "focused" or concentrated to provide more localized detail, or "expanded" to provide a more global view. Still other research has addressed inhibitory operations involved in attention. Examples include the inhibition that leads to delays in returning attention to just attended locations (" inhibition of return"; Rafal & Hcnik, 1994) and to delays in processing previously ignored distractors (" negative priming"; see Beech & Williams, this volume.) In subsequent sections, we suggest that the greater specificity afforded by viewing attention in terms of component operations makes it easier to relate biological to more complex cognitive processes. But to make this case, it is first necessary to consider whether the attcntional operations can be linked to specific neural mechanisms. Although much more research is necessary, preliminary evidence looks promising. Posner and his colleagues have employed techniques from neuropsychology and cognitive neurosciencc to relate attentional orienting to distributed circuits within the posterior attentional system. The disengage operation appears to be centered upon circuits within the parietal lobe, the move operation upon bramstcm circuits involving the superior colliculus, and the engage operation upon the pulvinar nucleus of the thalamus (Posncr, Inhoff, Fricdrich & Cohen, 1987; Posner & Raichle, 1994). In addition, evidence suggests that the attentional focusing operation is promoted by dopaminergic projections from the ventral tcgmental area to object processing pathways within the left hemisphere, while attentional broadening depends upon right hemisphere mechanisms (Fink ct al., 1996; Tucker & Dcrryberry, 1992). Finally, physiological evidence is consistent with the notion that these attentional operations are in part controlled by the neural systems related to motivation. For example, the fearrelated circuitry of the central amygdala projects directly to the dopamincrgic pathways involved in attentional focusing. The amygdala also projects to the

452

Chapter I0

noradrenergie system involved in vigilance and the executive attentional system within the anterior cortex. Since the vigilance and executive systems regulate the posterior attentional system, projections from the amygdala allow fear-related motives to influence the disengage, move, and engage operations (Rothbart, Derryberry, & Posner, 1994). Given these relations between motivational and attentional processes, one of the goals of our research is to investigate individual differences in attentional operations during positive and negative motivational states. Our approach is basically compatible with that of Ketelaar and Clore (this volume) who suggest that emotional states modulate the organization of cognitive processing. In its focus on attention, our approach is closest to that of Posner and Rothbart, although we are not yet concerned with the strategic mechanisms emphasized in their construct of effortful control. At this point, our research aims to assess the more basic constraints that motivational processes exert upon different attentional operations, especially those involving the posterior attentional system. These basic constraints shape the attentional reactions on which the voluntary mechanisms operate. The studies reported below focus on the dimension of trait anxiety. Each used undergraduate subjects who were divided into low and high anxious groups based on a median split on Spielberger's State Trait Anxiety Inventory (Spielberger, 1983). Temperament models propose that trait anxiety reflects variability in neural systems, such as Gray's (1982) behavioral inhibition system, that regulate attention in response to threatening signals. Cognitive approaches suggest that anxiety reflects variability within representations of prior knowledge, such as Beck's "danger schemas" (Clark & Beck, 1989), that facilitate the processing of congruent information. Consistent with both biological and cognitive approaches, many studies across the past ten years have found clinically and trait anxious subjects to show attentional biases in favor of threatening compared to neutral stimuli (for reviews, see Eysenek, 1992; Mathews, 1990; Wells & Matthews, 1994). These studies indicate that the anxiety-relateA biases can be highly specific, with different individuals (e.g., physical versus social phobics) showing enhanced attention to different types of threat (e.g., physical versus social threat words; Hope, Rapee, Heimberg, & Dombeek, 1990). They suggest that the biases can be elicited very early in the course of information processing, at times without conscious awareness of the threatening stimulus (Mogg, Bradley, & Williams, 1995). In addition, they suggest that in some situations the biases related to trait anxiety can be amplified by anxious or stressful states (MacLeod & Mathews, 1988). Although these studies have contributed

D. Derryberry and M.A. Reid

453

much to our understanding of anxiety and attention, they have tended to view attention in terms of a single mechanism, and the roles of its component operations remain unclear. One of our paradigms employs a computerized game designed to activate positive and negative incentive processes (for earlier research, see Derryberry & Reed, 1994a; Reed & Derryberry, 1995b). Subjects move through a sequence of twelve 48-trial games, with their score reset to zero at the start of each game. Within a game, each trial carries a positive or negative incentive value. On the positive trials, subjects gain 10 points if their response is fast and gain no points if their response is slow. On negative trials, 10 points are lost if the response is slow and no points are lost if the response is fast. The computer scores each response as "fast" or "slow" by comparing it to a criterion based on the subject's median reaction time (RT) on the previous game given trials of similar difficulty. Each trial consists of a warning cue (indicating whether the trial carries a positive or negative incentive value), a target requiring a speeded response (described below), and a feedback signal (indicating whether the response was fast or slow). In addition, the incentive proportions are varied to generate positive and negative games. On positive games, positive incentives occur on 75 % of the trials, while on negative games 75 % of the trials involved negative incentives. This manipulation is intended to promote relatively prolonged motivational states, allowing an analysis of trait anxiety effects during negative compared to positive states. On the negative games, for example, subjects experience three times as many negative as positive incentives, end up with a game score well below zero, and describe the state as "stressful".

Anxiety and attentional focusing The first study used a "global/local" perceptual task (Navon, 1977) to examine attentional focusing and expanding (Dcrrybcrry & Reed, in preparation). Each trial began with a square cue in the screen's center signaling the trial's positive (a green square) or negative (a red square) incentive value. After 500 milliseconds (ms), the cue was replaced by a hierarchical figure consisting of seven small letters (the local elements) grouped to form a larger letter (the global form). For example, seven small T's could be arranged to form a large L. Subjects pressed one key if an L was present (at either the global or local levels) and another key if an H was present. These targets appeared randomly at either the global or local levd. Finding targets in the global shape requires a broad focus of attention so that

454

Chapter 10

the entire figure can be processed as a whole. In contrast, finding local targets requires greater focusing in order to process the smaller details. Reaction times to global and local targets can thus be used as an indicator of subjects' tendencies to focus or expand attention. The subject's response was followed aider one second by a positive or negative feedback signal (an arrow pointing up or down), and the next trial began after an interval of 1500 ms. The results showed that on positive games there were no differences between low and high anxious subjects in processing global or local targets (see Figure 1). On negative games, there were no differences given global targets, but the high anxious subjects were faster to identify local targets than were low anxious subjects. It is worth noting that all subjects were substantially faster in responding to global than local targets, which suggests a strategy of searching first at the global level and, if a global target is not found, then focusing on the local elements. Since there were no anxietyrelated differences for global targets, the effect on local targets suggests that high anxious subjects were fast in executing the focusing operation. This effect did not depend on the trial's incentive value or on the feedback received on the previous trial. This facilitated focusing in anxious subjects is consistent with other findings of perceptual "narrowing" in dual task paradigms under stressful conditions (see reviews by Eysenck, 1982; Hockey, 1979). The present study is informative in showing that the focusing effect is related to the motivational state generated by increasing the density of negative incentives to form negative games, rather than by more phasic processes related to the incentive and feedback signals occurring within each trial. This type of staterelated motivational effect is generally compatible with Tucker and Williamson's (1984) concept of "tonic activation", a prolonged state in which focusing is promoted by means of dopaminergic projections to the left hemisphere. To examine Tueker's model more closely, we ran a second study that was identical to the first, except that the targets were presented randomly in the leR or right visual field. High anxious subjects again showed facilitated local processing on the negative games, but only for targets delivered to the left hemisphere (i.e., right visual field). It thus appears that the anxiety facilitates focusing operations within the lef~ hemisphere. This does not imply that anxious individuals have a chronically narrow focus of attention, for they did not differ from low anxious individuals on the nonthreatening positive games, and showed no global deficits on the negative games. Instead, facilitation within the left hemisphere allows them to focus attention more

455

D. Derryberry and M.A. Reid 450

440 Local Targets

430 RT

(ms)

420 410 400

Global ~

390 I

j

Negative

~

i .........

Targets I

Game

Positive Negative Game

LOWANXIOUS

HIGHANXIOUS

Positive

Figure 1. Mean RTs for low and high anxious subjects m identifying targets at the global and local levels during positive and negative games.

quickly in threatening situations, resulting in enhanced processing of detailed information. Attentional focusing during anxious states can viewed as basically adaptive. In threatening situations, it is otten the details that are crucial in evaluating whether an object (e.g., a facial expression, an approaching animal) is dangerous or safe. In addition, rapid or strong focusing may help to prevent distraction by environmental stimuli that are irrelevant to coping with a threat. Although our studies found that anxious focusing does not interfere with processing global information, this may due to strategies in which focusing occurred atter the targets were processed globally. Other research using dual task paradigms suggests that anxious focusing does impair the processing of secondary or peripheral information (e.g., Hockey, 1979). Thus, a focusing bias can also be maladaptive when it limits attention available for processing important peripheral or contextual information. We discuss such problems in the last section of the chapter.

456

Chapter 10

Anxiety and attentional orienting We have also employed the incentive paradigm to investigate the attentional processes involved in spatial orienting, namely the engage, move, and disengage operations. As in the global/local studies, incentive densities were manipulated to create positive and negative games, and each trial consisted of a cue, a target, and a feedback signal. The basic display consisted of a pair of blue outlined boxes, one on either side of a central fixation point. At the beginning of each trial one of the boxes changed color. This color cue indicated the trial's incentive value, with green cues indicating trials where points could be gained and red cues signaling trials where points could be lost. In addition, the cue predicted the probable location of the target, with 75% of the targets appearing in the cued box. Targets consisted of a small circle appearing in one of the two boxes, to which subjects responded with a simple key press. If the target appears in the cued location (a "valid" cue), RTs tend to be fast because attention has already moved to and engaged that location. But if the target appears in the other box (an "invalid" cue) RTs are slow because attention must disengage from the cued (attended) location, move, and engage the targete~ location. Thus, the general strength of attentional orienting can inferred from the RT difference between targets following valid and invalid cues. The results were similar to the global/local studies in that low and high anxious subjects did not differ on the positive games. On the negative games, which are graphed in Figure 2, anxious subjects showed a stronger attentional effect following negative incentive cues than low anxious subjects. This is consistent with many other findings that anxious subjects show an attentional bias favoring threatening information, a bias that is sometimes enhancexl by state anxiety (MacLeod & Mathews, 1988). In addition, our use of valid and invalid cues provides a closer view of the underlying operations. Specifically, low and high anxious subjects did not differ given valid negative cues, which suggests that the two groups were similar in moving to and engaging a negative location. Instead, the effect was limited to invalid cues, which gave rise to slower RTs in the high anxious subjects. This suggests that the anxiety-related attentional effect involves delays in disengaging from negative cues. In earlier studies, a similar disengage deficit was found when neurotic introverts (who are also prone to trait anxiety) were presented with negative cues (Derryberry & Reed, 1994a). The present study replicates the disengage effect and demonstrates that a stressful state is required to enable the bias in trait anxious subjects.

457

D. Derryberry and M:A. Reid 340

Invalid Cues

320 RT (ms)

300

280

//

Valid Cues

260 I

.

Positive

I,.

,

Negative

Cue Incentive LOW ANXIOUS

l. . . . .

Positive

I

Negative

Cue Incentive HIGH ANXIOUS

Figure 2. Mean RTs on negative games for low and high anxious subjects in detecting targets following positive and negative cues that were valid or invalid in predicting the target's location. It might be suggested that the delays in disengaging result from anxious subjects' tendency to focus strongly on negative cues, and that we are actually dealing with a single underlying mechanism. Although the two effects are similar in being enabled during negative motivational states, at this point we feel that they reflect distinct operations. In particular, they are dissociable in that the focusing effect is independent of the trial's incentive value whereas the disengage effect depends on a negative incentive cue. In addition, focusing is enhanced in the right visual field while disengagement appears equally across the two visual fields. This last distinction is consistent with neuropsychological research suggesting two separable neural systems, one involving parietal circuits within both hemispheres (disengagement), and the other ventral object pathways within the left hemisphere (focusing). It is also worth noting that our findings suggest that low and high anxious individuals do not differ in their capacity to move toward and engage threatening cues. This is important in suggesting that high anxious persons are not at an advantage (relative to low anxious persons) in detecting and orienting to threat. It is only after the threat has been engaged that the bias

458

Chapter 10

appears (see Wells & Matthews, 1994, for an additional discussion of "post attentional" biases). In itself, the disengage bias can be viewed as adaptive in providing the anxious individual with a somewhat longer "look" at a potentially threatening stimulus. At the same time, however, it can put the person at a disadvantage in processing information arriving from unattended locations. We will return to these types of problems in the chapter's last section. To summarize, our incentive studies suggest several general conclusions concerning trait anxiety and attention. First, they suggest that low and high trait anxious individuals do not differ in attention to positive incentive information. This is consistent with models proposing that anxiety arises from systems speeializeA for processing negative information. Second, the studies suggest that although anxious individuals may show biases related to threatening information, these biases are not always present, and appear to be enhancexl under negative state conditions. Such trait by state interactions are generally consistent with physiological models (e.g., Gray, 1982) proposing that trait anxiety depends on the same neural mechanisms that generate state anxiety. Third, our findings suggest that the attentional bias is also specific in that it involves only two of the five attentional operations tested so far. Anxious individuals show facilitated focusing on the details of objects along with impaired disengaging from threatening cues. Such a pattern should favor certain types of information at the expense of others, and may thus exert important effects on the higher level processing that follows the initial event. To better characterize these effects, however, we need to move beyond the attentional operations to consider their consequences. A first step in this direction is to consider their immediate effects on subsequent processing. Again, constructs and methods from cognitive science prove useful.

Immediate consequences of attention A common assumption within cognitive science is that attention provides a central mechanism for regulating or guiding the flow of information processing. When attention is aligned with a perceptual input, the resulting facilitation should influence the impact of that perceptual information on subsequent pathways processing conceptual and response information. Unattended inputs may still affect subsequent processes, but their impact should be less than that of an attended input. Such attcntional effects have been modeled in conne~onist approaches by adjusting the activation levels for attended and unattcndexl units (Cohen, Dunbar & McClelland, 1990). In

D. Derryberry and M.A. Reid

459

neurophysiological recording studies, such effects appear in the form of substantially increased firing rates for cells responding to attended information (e.g., Treue & Maunsell, 1996). From a more complex perspective, attention also contributes to the integration of parallel channels of information. This "binding" function has been studied in detail for the integration of perceptual features into unified objects (Treisman, 1988), and has also been discussed in relation to the coordination of conceptual and response information (Keele & Neil, 1978). In neurophysiological studies, attentional binding has been related to the synchronized firing rates that arise within spatially distant neurons responding to the features of a coherent object (e.g., Munk, Roelfsema, Konig, Engel & Singer, 1996). But regardless of whether attention is viewed in terms of facilitatory or integrative processes, it can be seen that lower-level attentional processes are capable of regulating the pattern of activation across the processing system. The resulting pattern is likely to have important effects on more complex cognitive processes that the system generates. Several of our studies have aimed at investigating such activation patterns, particularly in regard to concepts involving the self. The initial studies demonstrated that neurotic and anxious subjects show enhanced attention to negative trait reformation that is relevant to the self (Reed & Derryberry, 1995a). This was found in several studies using a modified version of the "dot probe" task. Two trait descriptive adjectives were sequentially presented on the screen, followed by a detection target in one of the two words' locations. When attention was first drawn to negative word, high anxious subjects were slow to disengage and detect targets in the location of the other word. As is the case with incentive cues, anxiety appears to delay disengagement from negative information related to the self. Anxiety and semantic activation Given these initial findings, several subsequent studies used priming tasks to explore the impact of the negative attentional bias on further processing. In the basic priming paradigm, a response irrelevant prime word is presented prior to a target word requiting a speeded response. Processing of the target is facilitated when it is preceded by a related (e.g., doctor-nurse) compared to unrelated (e.g., tree-nurse) prime (Neely, 1991). While some semantic priming effects are due to the automatic activation of related information with little attentional involvement, additional studies indicate that the amount of facilitation depends upon the extent to which attention is

460

Chapter 10

directed to the primo's meaning. For example, priming of one meaning of visual homonyms is gre~ater given attended than unattended auditory primes (Johnston & Dark, 1982). In addition, priming is suppressed if attention is directed away from a visual primo's meaning by requiring subjects to search it for certain letters (Hcnik, Fricdrich & Kellogg, 1983). These studies suggest that attention to the meaning of a stimulus can facilitate the activation of related information. This in turn suggests that high anxious subjects may show greater activation of related negative information when exposed to negative primes. One study used positive, negative, and neutral adjectives as primes and targets. On each trial the prime preceded the target by 100 ms. Subjects responded to the target by pressing one key if it described a personality trait and another key if it described a neutral object (50% of the trials involved positive or negative primes followed by neutral targets). Because we wanted to examine the extent of semantic activation, we varied the "distance" between primes and targets to form pairs that were closely or distantly related. This was done by selecting the positive and negative adjectives from lists representing five trait domains (brave-fearful, kind-cruel, moralimmoral, happy-sad and energetic-lazy), and then presenting prime-target pairs from the same or different domains. In conditions involving negative primes and targets, for example, a negative prime could be followed by a closely related target from the same domain (e.g., fearful/scared) or a more distantly related target from a different domain (e.g., fearful/lazy). Such a design allowed us to assess whether anxiety facilitates activation of related negative information, and if it does, whether this facilitation impacts only closely related information or also more distantly related information. Consistent with our other studies, the results showed no anxiety effects given positive primes (Dcrrybcrry & Rce~, submitted). When a negative prime was presented, however, greater facilitation of negative targets was found in high than low anxious subjects (see Figure 3). This is consistent with work by Richards and French (1992), who found that when presented with a polysemous prime, anxious subjects showed greater facilitation of targets related to the prime's threatening than neutral meaning. Our study builds on this evidence by showing that the anxiety-related priming was extensive, including distant as well as closely-related targets. In anxious subjects, for example, attention to a negative prime denoting anger facilitated negative targets from all five conceptual domains (denoting anger, fear, laziness, immorality, and sadness). In low anxious subjects, attention to a

D. Derryberry and M.A. Reid

461

negative prime resulted in a facilitation limited to targets from the same domain. It might be argued that rather than reflecting attentional effects, the enhanced priming in anxious subjects is an automatic byproduct of their stronger connections between negative concepts in semantic memory. Although we cannot rule out this structural account, we favor an attentional interpretation for several reasons. First, a surprising aspect of the results was that the low anxious subjects actually showed no overall priming following negative primes (see Figure 3). This is particularly hard to explain in terms of automatic structural connections, for it would lead to the unlikely conclusion that negative trait concepts are not associated in low anxious subjects. Second, we have replicated the pattern in a second study that used letter stimuli representing positive (A and B) and negative (D and F) feedback on the previous trial of an incentive-like task. Long term structural associations between these letters should be quite strong, but they did not interact with the anxiety effect (Derryberry & Reed, submitted).

680

Negative Targets

670 RT

(ms)

660

650

Positive Targets

640 I

....

I. . . .

Negative

I

I

Prime

Positive Negative Prime

LOW ANXIOUS

HIGH ANXIOUS

Positive

Figure 3. Mean RTs for low and high anxious subjects m categorizing positive and negative targets (combined across the five trait domains) following positive and negative primes.

462

Chapter 10

Although more research is required, what we think is happening in this paradigm involves differences in attention to two components of the prime's meaning. The semantic meaning of the primes can be viewed as consisting of a conceptual or denotative component (i.e., referring to a particular category of behavior, such as threat-related compared to morality-related behavior) and also an affective or connotative component (i.e., referring to its positive or negative valence, such as brave compared to fearful). The conceptual and affective meanings can be viewed as separable but interacting pathways within a semantic network. In physiological terms, such pathways appear related to "paralimbic" regions of the cortex, in which information is processed diffusely in close relation to body states, as compared to "neocoRical" regions, in which information is more detached from emotional states (Derryberry & Tucker, 1991). In any event, those high in anxiety may preferentially attend to the affective meaning of negative primes, perhaps because they are slow to disengage. Because affective meaning cuts across the different conceptual domains, the resulting facilitation is widespread. In contrast, low anxious subjects may attend primarily to the prime's conceptual meaning, resulting in less spread across domains. But beyond the underlying mechanisms, the widespread activation of negative information is likely to have important consequences on the subsequent cognition of anxious individuals. As discussed below, the activation may increase the accessibility of a wide range of negative information for incorporation into subsequent appraisal and attributional processes. One can again argue that this type of effect is basically adaptive. The widespread activation of negative information may provide a broad context for subsequent evaluations, and perhaps enhance the anticipation of upcoming negative events. However, it is also easy to see how such activation might promote overgeneralized forms of thought.

Extensions to Complex Cognitive Processing Our findings indicate that under negative incentive conditions, individuals high in trait anxiety show enhanee~ attentional focusing along with delays in disengaging from threatening cues. When their attention is drawn to negative self-relevant information, they are slow to disengage, and a wide range of negative conceptual content becomes activated. These findings complement the biological approaches in relating trait anxiety to distinct attentional operations, processes that appear to arise from specific neural systems. But in some ways the most interesting links arise between the

D. Derryberry and M.A. Reid

463

attentional processes and the complex cognitive processes emphasized by many personality theorists. In this final section, we consider two general effects of motivated attention. We first discuss ways in which attention might influence ongoing cognition as it unfolds during a stressful situation, and then conclude by considering the impact of these biases on the long-term memory storage and representational development. In the case of ongoing processing, we begin by returning to the widespread activation of negative information that accompanies trait anxiety. Such a context of primed negative content may contribute to a variety of subsequent cognitive processes, most likely supporting cognition that overgeneralizes or overextends negative anticipations and interpretations. For example, it has often been noted that the appraisal and evaluative processes of anxious individuals tend to be pessimistic and in some cases even catastrophic (Clark & Beck, 1989). In addition, trait anxiety is related to the shameful feelings that arise when a person evaluates themselves negatively in terms of multiple aspects of the self (Tangney, Burggraf & Wagner, 1995). Similarly, anxious individuals (like depressives) appear to make global rather than specific attributions for negative outcomes, assigning responsibility in terms of their general character rather than a specific behavior (Clark, Watson & Mineka, 1994; Ganellan, 1988). These and other overgeneralized characteristics of anxious thought may be supported by the wide range of activated negative content that is available for ongoing appraisals and attributions. Such cognition may also be promoted within individuals who have formed strong associations between the negative aspects of the self (e.g., Showers, 1995). Nevertheless, negative attentional biases would still be expected to amplify such structural effects. The facilitation of focusing and delayed disengagement are also likely to contribute to negative appraisals and evaluations by amplifying specific negative content. The tendency to rapidly focus on details may contribute to the perfectionism and even obsessive concern with details that sometimes accompany trait anxiety (Lundh & Ost, 1996). Narrowing may also contribute to the common states of" self-focused" attention in which anxious individuals focus inwardly upon their personal concerns and feelings, often to the neglect of the external environment (Schwarzer & Wicklund, 1991). In stressful social interactions, anxious persons can become highly selfconscious and concerned about slight deficiencies in their appearance or behavior. It should be kept in mind, however, that in our studies the enhanced focusing evident under stressful conditions influences positive as well as negative information, and may thus impact many types of information. Given

464

Chapter 10

positive information, for example, some anxious individuals focus narrowly upon a potentially relieving coping strategy, at times pursuing it in an obsessive manner (Wachtr 1967). More generally, the restriction of attention may contribute to anxious individuals' tendencies to employ relatively narrow categories and to perceive less rdatodness between categories (Mikulinccr, Kodem, & Paz, 1990). Also, our studies suggest that trait anxiety is related to a rapid focusing rather than a chronically narrow focus. As mentioned above, the delays in disengaging may promote the widespread activation of negative informational and ovcrgcncralizcd forms of thought. Even though positive information is not directly influenced, delays in disengaging from negative content can limit the person's ability to engage cooccumng positive information. Delayed disengagement may also maintain self-focused attention when the person has difficulty shifting from their personal concerns or negative feelings. Even when the self is not the primary focus, the delays may contribute to the prolonged ruminative and worrisome thought that is prevalent in anxiety (TaUis, Eysenck & Mathews, 1992; Wells & Matthews, 1994). Again, it is worth noting that our studies indicate that anxiety does not directly influence the tendency to move attention toward and engage a negative event. This argues against a view of anxious persons as hypervigilant in the sense that they are constantly searching for potential dangers or things that can go wrong. Although their motivational or cognitive systems may render anxious people more sensitive to such dangers, we have found no evidence that they differ from less anxious people in actively searching them out. While this may seem like subtle difference, it suggests that high and low anxious people may not differ in many situations. What appears most distinctive is the individual's tendency to gain additional information (by focusing on details or by prolonging inspection) after noticing a threat. While these examples emphasize influences on negative content, it can be seen that the focusing and disengage biases also contribute to anxious cognition by limiting the flexibility of attention. Self-report studies have found consistent negative correlations between measures of anxiety and attentional flexibility (Derryberry & Rothbart, 1988; Wells & Matthews, 1994). Most higher level cognition requires attention to move fluidly in order to integrate information from various sources. If the individual becomes overfocused or is slow to disengage, the coordination of relevant information is likely to be impaired. For example, in many threatening situations, effective coping requires that attention shift between sources of danger and potential

D. Derryberry and M.A. Reid

465

sources of relief or safety. If attontion is slow to disengage from a threat, the anxious person may be unable to take advantage of relieving information that becomes available. Similarly, many appraisal processes require attention to different aspects of the threat (primary appraisal) and capabilities for coping (secondary appraisal; Lazarus & Folkman, 1984). If attentional flexibility is constrained, the appraisal may be biased toward anticipated harm because certain aspects of the threat are emphasized at the expense of more relieving aspects or available coping options. Inflexibility may also contribute to anxious individuals' tendencies to attribute negative outcomes to internal rather than external causes (Clark, et al. 1994; Ganellan, 1988). Rather than moving attention between internal and external information, the anxious individual may focus on their feelings of negative affect (Dienstbier, 1984), and may thus emphasize internal at the expense of external causes. Finally, as emphasized by Wells and Matthews (1994), states of self-focused attention may ot'ten impair appraisals and attributions because they limit the flexible distribution of attention to relevant environmental information. These examples illustrate some of the ways in which anxiety-related attentional processes may influence complex cognition. While such influences upon ongoing cognition are important, the long range consequences of attention on learning and memory are likely to be equally significant. Physiological and cognitive evidence strongly suggests that attention is central to certain forms of cortical plasticity and learning. This implies that attention should progressively shape the individual's memory representations in a manner that supports the future functioning of the anxiety-related motivational system (Derryberry & Reed, 1994b, 1996). Thus, anxious individuals would be expected to develop representations that emphasize the details of stressful situations, that emphasize negative at the expense of positive information, and that interrelate a wide range of negative content. Research indicates that anxious individuals form stronger short term memory representations for attended negative words (Reed & Derryberry, 1995a), and a number of studies have demonstrated that trait anxious individuals show enhanced recall of negative information (e.g., Eysenck & Byme, 1994; Eysenck & Mogg, 1992; Kennedy & Craighead, 1988). These findings are consistent with the notion that anxious individuals show facilitated storage of threatening information, but more research is needed to assess the role of attentional and nonattentional processes in memory effects. Especially valuable would be developmental studies of threat-related representations within anxious and non-anxious children. If it turns out that motivated attentional operations are crucial to these representations, then temperamental

466

Chapter 10

differences in these motives may prove fundamental to our developing representations of the world. We have suggested elsewhere that the representations or schemas emphasized by cognitive approaches can be viewed as cortical structures that provide inputs to subcortical motivational systems (Derryberry & Reed, 1994b, 1996). Although the fear system can be triggered by direct featural inputs from the thalamus (LeDoux, 1995), in most instances the motivational systems can rely on existing cortical representations to evaluate the significance of environmental events. By exploiting the more differentiated circuitry of the cortex, the representations provide the motivational systems with greater resolution in anticipating and evaluating incoming information and in guiding behavior accordingly. As a child develops, these cortical representations are clearly shaped by environmental inputs reflecting the beliefs and values of their culture. However, motivational systems come on line very early in life, and by means of attention, function to selectively stabilize the cortical information that is most relevant. Thus, individual differences in motivational system reactivity shape individual differences in cortical representations. These reciprocal influences allow for a selforganizing system within which motivational and cognitive processes guide one another. Conclusions

This chapter has attomptext to approach three general goals. The most specific goal was illustrat~ the value of a componcntial approach to approach to human personality. This approach views personality in terms of distinct motivational processes, component attentional operations, and different types of cognitive .processes. We tried to show that by decomposing attention into component operations, more specific predictions can be made for linking motivational and cognitive processes. Hopefully, our studies of trait anxi~y help to clarify how individual differences in attentional focusing and disengaging contribute to some characteristics of anxious cognition. However, future research mw,ds to focus on additional operations, such as those arising from Posner's vigilance and anterior attcntional systems, that may also be regulated by anxiety. In addition, the effects of other motivational systems and personality dimensions need to be assessed. Systems related to appetitive, aggressive, nurturant, and other motives may regulate different sets of attentional operations with different consequences for cognition.

D. Derryberry and M.A. Reid

467

A second and more general goal was to illustrate the value of cognitive science research in linking biological and cognitive approaches to personality. We attempted to show that in its emphasis on relatively detailed information processing models, cognitive science provides intermediate level frameworks for translating between biological and cognitive constructs. By characterizing different attentional patterns in different situations, information processing models not only improve the chances of drawing connections between cognitive processes and biology, but also allow more specific prediction of cognitive difficulties. While our componential model of attention illustrated one of these frameworks, there are a number of others discussed in the chapters of this volume that are valuable to personality psychologists. We look forward to future contributions from cognitive science, and to the broadening of cognitive science that personality psychology will provide. The third and most general goal was to illustrate the complementary nature of biological and cognitive approaches. The biological perspective is helpful in its emphasis on evolved motivational systems that arise from subcortieal circuits to regulate ongoing information processing and storage within the cortex. However, it is the cognitive perspective that provides the best view of actual processing within in the cortex, including our representations of self, others, and cultural values. Even though we have emphasized how anxiety may shape these processes and representations, it is clear that cortical circuits feed back on subcortieal mechanisms to allow cognition to shape motivation. Represented information will influence both the general evaluation of threatening situations and the identification of threat signals in subsequent situations. Given these kinds of reciprocal relations, there should be no fundamental incompatibility between the constructs of cognitive schemas and motivational systems. These can be viewed as part of the same distributed system, with different theoretical perspectives focusing on different levels. References

Ahadi, S. A., & Rothbart, M. K. (1994). Temperament, development, and the Big Five. In C. F. Halverson, Jr., G. A. Kohnstamm, & R. P. Martin (Eds.), The developing structure of temperament and personality from infancy to adulthood (pp. 189-207). Hillsdale, NJ: Erlbaum. Bennett, K. M. B., Waterman, C., Scarpa, M., & Castiello, U. (1995). Covert visuospatial attentional mechanisms in Parkinson's disease. Brain, 118, 153-166.

468

Chapter 10

Clark, D. A., & Beck, A. T. (1989). Cognitive theory and therapy of anxiety and depression. In P. C. Kendall & D. Watson (Eds.), Anxiety and depression: Distinctive and overlapping features (pp. 379-411). San Diego, CA.: Academic Press. Clark, L. A., Watson, D., & Mineka, S. (1994). Temperament, personality, and the mood and anxiety disorders. Journal of Abnormal Psychology, 103, 103-116. Cloninger, C. R. (1987). A systematic method for clinical description and classification of personality variants. Archives of General Psychiatry, 44, 573-588. Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990). On the control of automatic processes: A parallel distributed processing account of the stroop effect. Psychological Review, 97, 332-361. Davis, M. (1992). The role of the amygdala in fear and anxiety. Annual Review of Neuroscience. 1992, 15, 353-375. Depue, R. A., & lacono, W. G. (1989). Neurobehavioral aspects of affective disorders. Annual Review of Psychology, 40, 457-492. Derryberry, D., & Reed, M. A. (1994a). Temperament and attention: Orienting toward and away from positive and negative signals. Journal of Personality and Social Psychology, 66, 1128-1139. Derryberry, D., & Reed, M. A. (1994b). Temperament and the selforganization of personality. Development and Psychopathology, 6, 653676. Derryberry, D., & Reed, M. A. (1996). Regulatory processes and the development of cognitive representations. Development and Psychopathology, 8, 215-234. Derryberry, D., & Rothbart, M. K. (1988). Affect, arousal, and attention as components of temperament. Journal of Personality and Social Psychology, 55, 958-966. Derryberry, D., & Tucker, D. M. (1991). The adaptive base of the neural hierarchy: Elementary motivational controls on network function. In R. Dienstbier (Ed.), Nebraska symposium on motivation, Vol. 38: Perspectives on motivation (pp. 289-342). Lincoln, Nebraska: University of Nebraska Press. Dienstbier, R. A. (1984). The role of emotion in moral socialization. In C. E. Izard, J. Kagan, & R. B. Zajone (Eds.), Emotions, cogniaon, and behavior (pp. 484-514). Cambridge: Cambridge University Press.

D. Derryberry andM.A. Reid

469

Eriksen, C. W., & Yeh, Y. (1985). Allocation of attention in the visual field. Journal of Experimental Psychology: Human Perception and Performance, 11,583-597. Eysenek, M. W. (1982). Attention and arousal. New York: Springer-Verlag. Eysenck, M. W. (1992). Anxiety: The cognitive perspective. Hillsdale, NJ: Erlbaum. Eysenck, M. W., & Byme, A. (1994). Implicit memory bias, explicit memory bias, and anxiety. Cognition and Emotion, 8, 415-431. Eysenck, M. W., & Mogg, K. (1992). Clinical anxiety, trait anxiety, and memory bias. In S. Christianson (Ed.), The handbook of emotion and memory: Research and theory (pp. 429-450). Hillsdale, NJ: Erlbaum. Fink, G. R., Halligan, P. W., Marshall, J. C., Frith, C. D., Frackowiak, R. S. J., & Dolan, R. J. (1996). Where in the brain does visual attention select the forest and the trees? Nature, 382, 626-628. Fowles, D. C. (1994). A motivational theory of psychopathology. In W. G. Spaulding (Ed.), Nebraska symposium on motivation, Vol. 41: Integrative views of motivation, cognition, and emotion (pp. 181-238). Lincoln, Nebraska: University of Nebraska Press. Ganellan, R. J. (1988). Specificity of attributions and overgeneralization in depression and anxiety. Journal of Abnormal Psychology, 97, 83-86. Gazzaniga, M.S. (Ed.), The cognitive neurosciences. Cambridge, MA: MIT Press. Gilbert, P., & Trower, P. (1990). The evolution and manifestation of social anxiety. In W. R. Crozier (Ed.), Shyness and embarrassment: Perspecaves from social psychology (pp. 144-177). Cambridge: Cambridge University Press. Gray, J. A. (1982). The neuropsychology of anxiety. London: Oxford. Gray, J. A. (1987a). Perspectives on anxiety and impulsivity: A commentary. Journal of Research in Personality, 21,493-509. Gray, J. A. (1987b). The psychology offear and stress (2nd ed.). New York: McGraw-Hill. Gray, J. A. (1994). Framework for a taxonomy of psychiatric disorder. In S. H. M. van Goozen, N. E. Van de Poll, & J. A. Sergeant (Eds.), Emotions: Essays on emotion theory (pp. 29-60). HiUsdale, NJ: Erlbaum. Henik, A., Friedrich, F. J., & Kellogg, W. A. (1983). The dependence of semantic relatex~ess effects on prime processing. Memory and Cognition, 11,366-373.

470

Chapter 10

Hockey, R. (1979). Stress and the cognitive components of skilled performance. In V. Hamilton & D. M. Warburton (Eds.), Human stress and cogniaon: An information processing approach (pp. 141-177). New York: Wiley. Hope, D. A., Rapee, R. M., Heimberg, R. G., & Dombeek, M. J. (1990). Representations of the self in social phobia: Vulnerability to social threat. Cognitive Research and Therapy, 14, 177-190. Johnston, W. A., & Dark, V. J. (1982). In defense of intraperceptual theories of attention. Journal of Experimental Psychology: Human Perception and Performance, 8, 407-421. Keele, S. W., & Neil, W. T. (1978). Mechanisms-of attention. In E. L. Carterette & M. P. Friedman (Eds.), Handbook of perception. (Vol. 9, pp. 3-47). New York: Academic Press. Kennedy, R. E., & Craighead, W. E. (1988). Differential effects of depression and anxiety on recall of feedback in a learning task. Behavior Therapy, 19, 437-454. Larsen, R. J., & Ketelaar, T. (1989). Extraversion, neuroticism and susceptibility to positive and negative mood induction procedures. Personality and Individual Differences, 10, 1221-1228. Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal and coping. New York: Springer. LeDoux, J. E. (1995). In search of an emotional system in the brain: Leaping from fear to emotion and consciousness. In M. S. Gazzaniga (Ed.), The cognitive neurosciences (pp. 1049-1062). Cambridge, MA: MIT Press. Lundh, L., & Ost, L. (1996). Stroop interference, self focus and perfectionism in social phobies. Personality and Individual Differences, 20, 725-731. MacDonald, K. (1992). Warmth as a developmental contruct: an evolutionary analysis. Child Development, 63, 753-773. MacLeod, C., & Mathews, A. (1988). Anxiety and the allocation of attention to threat. Quarterly Journal of Experimental Psychology, 40, 653-670. Marks, I. M., & Neese, R. M. (1994). Fear and fitness: An evolutionary analysis of anxiety disorders. Ethology and Sociobiology, 15, 247-261. Mathews, A. (1990). Why worry? The cognitive function of anxiety. Behavioral Research and Therapy, 28, 455-468. Mikulincer, M., Kedem, P., & Paz, D. (1990). Anxiety and categorization 1. The structure and boundaries of mental categories. Personality and

Individual Differences, 11,805-814.

D. Derryberry and M.A. Reid

471

Mogg, K., Bradley, B. P., & Williams, R. (1995). Attentional bias in anxiety and depression: The role of awareness. British Journal of Clinical Psychology, 34, 17-36. Munk, M. H. J., Roclfscma, P. R., Konig, P., Engel, A. K., & Singer, W. (1996). Role of reticular activation in the modulation of intrar synchronization. Science, 272, 271-274. Navon, D. (1977). Forest before trees: The precedence of global features in visual perception. Cognitive Psychology, 9, 353-383. Nccly, J. H. (1991). Semantic priming effects in visual word recognition: A selective review of current findings and theories. In D. Besner & G. W. Humphrcys (Eds.), Basic processes in reading: Visual word recognition (pp. 264-336). Hillsdale, NJ: Erlbaum. Pankscpp, J. (1982). Toward a general psychobiological theory of emotions. Behavioral and Brain Sciences, .5, 407-467. Pankscpp, J. (1986a). The anatomy of emotions. In R. Plutchik & H. Kellerman (Eds.), Emotion: Theory, research and experience. Vol. 3: Biological foundations of emotions (pp. 91-124). San Diego, CA: Academic Press. Panksepp, J. (1986b). The psychobiology of prosocial behaviors: Separation distress, play, and altruism. In C. Zahn-Waxler, E. M. Cummings, & R. lannotti (Eds.), Altruism and aggression: Biological and social origins (pp. 19-57). Cambridge: Cambridge University Press. Panksepp, J. (1992a). A critical role for "affectivc neuroscicnce" in resolving what is basic about basic emotions. Psychological Review, 99, 554-560. Panksepp, J. (1992b). Oxytocin effects on emotional processes: Separation distress, social bonding, and relatioships to psychiatric disorders. Annals of the New York Academy of Sciences, 652, 243-252. Posner, M. I., Inhoff, A. W., Friedrich, F. J., & Cohen, A. (1987). Isolating attcntional systems: A cognitive-anatomical analysis. Psychobiology, 15, 107-121. Posncr, M. I., & Raichle, M. E. (1994). Images of mind. New York: Scientific American Library. Posner, M. I., & Rothbart, M.. K. (1991). Attentional mechanisms and conscious experience. In D. Milner & M. Rugg (Eds.), The neuropsychology of conscmusness (pp. 91-111). San Diego, CA: Academic Press. Rafal, R., & Hcnik, A. (1994). The neurology of inhibition: Integrating controlled and automatic processes. In D. Dagcnbach & T. H. Carr

472

Chapter 10

(Eds.), Inhibitory processes in attention, memory, and language (pp. 151). San Diego, CA." Academic Press. Reed, M. A., & Derryberry, D. (1995a). Temperament and attention to positive and negative trait information. Personality and Individual Differences, 18, 135-147. Reed, M. A., & Derryberry, D. (1995b). Temperament and response processing: Facilitatory and inhibitory consequences of positive and negative motivational states. Journal of Research in Personality, 29, 59-84. Richards, A., & French, C. C. (1992). An anxiety-related bias in semantic activation when processing threat/neutral homographs. Quarterly Journal of Experimental Psychology, 45, 131-139. Rothbart, M. K., Ahadi, S. A., & Hershey, K. L. (1994). Temperament and social behavior in childhood., 40, 21-39. Rothbart, M. K., & Bates, J. E. (in press). Temperament. In W. Damon & N. Eisenberg (Eds.), Handbook of child psychology: Vol. 3: Social emotional and personality development (Sth ed.). New York: Wiley. Rothbart, M. K., Derryberry, D., & Posner, M. I. (1994). A psychobiological approach to the development of temperament. In J. E. Bates & T. D. Wachs (Eds.), Temperament: Individual differences at the interface of biology and behavior (pp. 83-116). Washington, DC: American Psychological Association. Schwarzer, R., & Wicklund, R. A. (Eds.). (1991). Anxiety and self-focused attention. London: Harwood Academic Publishers. Showers, C. J. (1995). The evaluative organizaiton of self-knowledge: Origins, processes, and implications for self-esteem. In M. H. Komis (Ed.), Efficacy, agency, and self-esteem (pp. 101-120). New York: Plenum. Spielberger, C. D. (1983). Manual of the State-Trait Anxiety Inventory, Form Y. Pale Alto, CA: Consulting Psychologists Press, Inc. Spoont, M. R. (1992). Modulatory role of serotonin in neural information processing: Implications for human psychopathology. Psychological Bulletin, 112, 330-350. Tallis, F., Eysenck, M., & Mathews, A. (1992). A questionnaire for the measurement of nonpathological worry. Personality and Individual Differences, 13, 161-168. Tangney, J. P., Burggraf, S. A., & Wagner, P. E. (1995). Shame-proneness, guilt-proneness, and psychological symptoms. In J. P. Tangney & K. W.

D. Derryberry and M.A. Reid

473

Fischer (Eds.), Self-conscious emotions: The psychology of shame, guilt, embarrassment, and pride (pp. 343-367). New York: Guilford. Treisman, A. M. (1988). Features and objects: The Fourteenth Bartlett Memorial Lecture. Quarterly Journal of Experimental Psychology, 40A, 201-237. Treue, S., & Maunsell, J. H. R. (1996). Attentional modulation of visual motion processing in cortical areas MS and MST. Nature, 382, 39-541. Tucker, D. M., & Derryberry, D. (1992). Motivated attention: Anxiety and the frontal executive mechanisms. Neuropsychiatry, Neuropsychology, and Behavioral Neurology, 5, 233-252. Tucker, D. M., & Williamson, P. A. (1984). Asymmetric neural control systems in human self-regulation. Psychological Review, 91, 185-215. Wachtel, P. L. (1967). Conceptions of broad and narrow attention.

Psychological Bulletin, 68,417-429. Wallace, J. F., Newman, J. P., & Bachorowski, J. (1991). Failures of response modulation: Impulsive behavior in anxious and impulsive individuals. Journal of Research in Personality, 25, 23-44. Watson, D., & Clark, L. A. (1992). On traits and temperament: General and specific factors of emotional experience and their relation to the fivefactor model. Journal of Personality, 60, 441-476. Wells, A., & Matthews, G. (1994). Attention and emotion: A clinical perspective. Hillsdale, NJ: Erlbaum.

This Page Intentionally Left Blank

Cognitive Science Perspectives on Personality and Emotion - G. Matthews (Editor) 9 1997 Elsevier Science B.V. All fights reserved. C H A P T E R 11

Investigating Cognitive Processes in Schizotypai Personality and Schizophrenia Anthony Beech and Leanne Willmms

...my trouble is that I've got too many thoughts. You might think about something, let's say that ashtray and just think, oh! yes, that's for putting my cigarette in, but I would think of it and then I would think of a dozen different things connected with it at the same time. My thoughts get all jumbled up. I start thinking or talking about something but I never get there. Instead I wander off in the wrong direction and get caught up with all sorts of different things that may be connected with the things I wanted to say but in a way I can't explain. I can't control my thoughts. I can't keep thoughts out. It comes automatically. It has to do with what is going on around me - taking in too much of my surroundings - vital not to miss anything. I can't shut things out of my mind and everything closes in on me. It's as if I am too wide awake - very, very alert. I can't relax at all. Everything seems to go through me. I just can't shut things out. The above self-reports of schizophrenics talking about their symptoms (Chapman, 1966; McGhie & Chapman, 1961) describe experiences of thought disorder and perceptual disturbance, which form part of the constellation of positive symptoms of schizophrenia. Other positive symptoms include hallucinations (a more severe form of perceptual disturbance) and delusions. In contrast, negative schizophrenic signs are: flattened levels of emotional responsiveness, poverty of content of speech, apathy, anhedonia and asocial characteristics (Andreasen, 1981). Although there is a large body of evidence to suggest that schizophrenia

476

Chapter 11

is in part associated with problems of attention (see Straube & Oades, 1992, for a review), very few models have been proposed that provide a direct link between deficits in cognitive processes and overt schizophrenic phenomena. One such model was proposed by Frith (1979). He proposed that specifically positive symptoms could arise as the result of a defect in the selective inhibitory mechanisms that "control and limit the contents of consciousness". According to this model, thought disorder may be interpreted as an inability to inhibit irrelevant word associations, ones that usually do not enter consciousness. The thought disorder is described by one sufferer as: you might think about something, let's say that ashtray and just think, oh! yes, that's for putting my cigarette in, but I would think of it and then I would think of a dozen different things connected with it at the same time. This self report dearly indicates that this individual is unable to suppress material that is connected with, but not directly relevant to, what they are thinking about. That irrelevant intrusions stem from associations that would normally remain at an unconscious level is suggested in other reports of thought disorder: (I) get caught up with all sorts of different things that may be connected with the things I wanted to say but in a way I can't explain. I can't control my thoughts. I can't keep thoughts out. It comes automatically.

Hallucinations, according to Frith, arise from a faulty perception of real sensations. Descriptions of perceptual disturbances like, "it has to do with what is going on around me", "taking in too much of my surroundings", "vital not to miss anything", suggest that schizophrenics may experience a sensory overload of unconscious stimuli that would usually be inhibited. It is entirely feasible that in such a state of overload an individual can misperceive (or misinterpret) various sensory inputs and that these incorrect interpretations are experienced consciously as hallucinations. Delusions stem from the individual's neexl to explain and interpret and put into context these strange thoughts and perceptions. On the basis of the examples, Filth's ideas have intuitive appeal. Our

A. Beech and L. Williams

477

interest was therefore in undertaking an empirical investigation of his model. First, we drew on descriptions of inhibitory processes described by Frith. Here we drew on models of selective attention described in the information processing literature. However, given the frequent revisions to models in this area it is not possible here to do justice to the complexity of the topic. However, selective coverage of the major issues involved may be sufficient to highlight those that are most relevant to our own ideas. Here, inhibition is used as an information processing term and is not necessarily equated with a particular aspect of brain function. However, it has been recently suggested that the frontal-limbic area of the brain (specifically the prefrontal lobes and the hippocampus) handles executive and learning systems such as filtering of information, maintenance of cognitive focus and the shifting of cognitive set (Stuss & Benson, 1986; Venables, 1992). The next section examines information processing explanations of filtering in more detail. Mechanisms of Selective Attention

Even a quick scan of the attentional literature will reveal that a substantial research effort has been directed at understanding the mechanisms that allow people to attend to certain features in their environment and, at the same time, screen out other features that are irrelevant. Common to these models is the view that both relevant and irrelevant inputs to the human information processing system are analyzed initially by automatic processes. This analysis occurs without conscious awareness or intentionality on the part of the individual and results from parallel spreading of activation of various processing pathways (Anderson, 1976; Collins & Loflus, 1975; Neely, 1977). By contrast, evidence that simultaneous task requirements produce interference, indicates that later, and higher, levels of processing have relatively limited capacity. Such a limited-capacity system necessitates the application of selective attentional mechanisms in order to maintain an efficient level of conscious processing (Posner, 1978). Recent models of attention reject the notion that automatic processing is an all-or-none phenomenon and suggests instead that is can occur in varying degrees (Cohen, Dunbar & McClelland, 1990). Here it is argued that conscious attention modulates the automatic activation of information in relation to the relevance of this information to the task at hand. Thus automatic processing, under this view, can be subject to some attentional control. In the last decade there has been a particular interest in explaining what

478

Chapter 11

happens to information that is ignored when a person selectively attends to another source of information. In other words, when we select information to pay conscious attention to, what are the concomitant mechanisms involved in effectively ignoring irrelevant information? It is this aspect of selective attention that is most crucial to the investigation of Filth's theory. A common example from the university context would be, how do students manage to concentrate on a lecture, while simultaneously screening out competing (and distracting) dialog between students sitting behind them? Psychologists interested in selective attention mechanisms tend to fall into one of two main camps. Some theorists have suggested that ignored information passively fades away (Neisser, 1967). Supporters of this view include: those that argue irrelevant information is screened out at an early, preattentive stage of processing in which only the physical and sensory features of stimuli have been processed (e.g., Broadbent, 1958, 1971); and those that maintain screening occurs much later when stimuli have reached a conscious level of representation (Allport, 1980; van der Heijden, 1981). Theorists in the other camp argue that selection works by the active suppression of information not specifieaUy attended to (Neill, 1977; Lowe, 1979; Tipper, 1985). In their model of attentional processes, Cohen et al. (1990) provide an account of the fate of unattended information that supports this view. They suggest that the filtering of potentially interfering information requires effortful (conscious) processing, in other words, as well simultaneously facilitating the processing of attended information, these mechanisms inhibit the processing of unattended (and potentially interfering) information. It is this concomitant process of active inhibition that we view as analogous to Frith's mechanism that "controls and limits the contents of consciousness'.

Experimental Investigations of Inhibitory Processes It has been the work of Tipper on the fate of ignored information (Tipper, 1985; Tipper & Cranston, 1985) that has made it possible for us to identify an experimental method for the exploration of Filth's ideas. Tipper and his colleagues found in experiments where information that has previously been "screened-out" is quickly re-presented, the time taken to attend to the re-presented stimulus is longer than if it had not been previously ignored. This finding is strong evidence for the argument that unattended information is actively inhibited. The time course of the inhibitory process is approximately two seconds (Neill and Westberry, 1987). Others report that

A. Beech and L. Williams

479

the effect persists at a lower level for up to 8 seconds (Neill, Valdes, Terry & Gorfein, 1992). The effect takes time to develop, with the effect that responses are delayed to material that has just been ignored. Tipper (1984) has termed this effect "negative priming". This is in contrast to the wellknown effect of "positive priming" where re-presentation of material speeds up the response to previously presented material. Negative priming tasks otten use Stroop (1935) words. These are color words written in a different color (for example, the color word RED written in blue ink). If a participant is asked to report the ink color of the word, the argument is that the selection process involves active inhibition of a response to the color word itself. That is, by attending to the ink color blue there is an active inhibition of response to the concept RED. If red is the ink color of the next Stroop word it takes longer to report this, than if the last color word had been GREEN. In contrast, under conditions of positive priming, responses to naming the word red would be faster if they were preeeAed by responses to the congruent ink color red.

Investigation of inhibitory processes using schizophrenic-proneness personality (schizotypaO participants There are a number of problems when conducting experiments with clinical samples of individuals diagnosed with schizophrenia. By the very nature of their illness they are extremely distractible so it can be very difficult for them to concentrate on the task at hand. It is also possible that antipsychotic medication had an impact upon the inhibition effect that we were interested in measuring. For these reasons, initial investigations of inhibitory processes relied on non-clinical samples of individuals who reported differing levels of personality type that suggest proneness to schizophrenia (i.e., schizotypy). Schizotypy is measured by using scales designed to tap milder versions of schizophrenic symptoms, particularly positive symptoms. In line with the continuum model of schizophrenia (Claridge, 1985, 1987) it is argued that, by endorsing items related to sehizotypal experiences, those with high schizotypy scores are showing less severe analogs of schizophrenic/ schizotypal symptomatology. Claridge and Broks (1984) devised one of the instruments used in the studies described below. This is the STQ (schizotypy questionnaire) which comprises two scales: STA, containing items that describe the diagnostic criteria for Schizotypal Personality Disorder (SPD) - a personality disorder seen as a milder version of schizophrenia (DSM-III, American Psychiatric

480

Chapter 11

Association, 1980); and STB, comprising items describing the diagnostic criteria for Borderline Personality Disorder (DSM-III). In an item analysis of the STA scale, Hcwitt and Claridge (1989) found that the scale contained three distinct clusters of items: magical ideation, unusual perceptual experience, and paranoid ideation and suspiciousness. Jackson and Claridge (1991) found that a sample of 56 schizophrenics in relative remission scored highly on the STA scale lending support for the criterion validity of the scale. Chapman and Chapman have also developed a number of scales used by the current authors. These scales, in contrast to the STA, are specifically based on schizophrenic symptomatology and include: the Perceptual Aberration scale (Chapman, Chapman & Rawlm, 1978), measuring perceptual distortions; the Magical Ideation Scale (Eckblad & Chapman, 1983), measuring unusual perceptual and ideational disturbances; the Physical Anhedonia Scale (Chapman, Chapman & Rawlin, 1976), measuring the inability to experience pleasure from physical stimuli; and the revised Social Anhedonia Scale (Eckblad, Chapman, Chapman & Mishlove, 1982), measuring social withdrawal. The Chapmans found that psychiatrically healthy, high scorers on the Magical Ideation and Perceptual Aberration scales reporte~ a greater number of psychotic-like experiences such as paranormal encounters and mild auditory and visual hallucinations these subjects were also more likely to see professional help for mild psychopathology than low scorers over a two year period from the initial administration of the questionnaires (Chapman & Chapman, 1980). In a ten year follow-up study the incidence of psychosis or having features of SPD was significantly greater in high scorers these scales (Chapman et al., 1994). A pilot study (reported by the first author (Beech & Claridge, 1987) investigated inhibitory processes. This study employed a priming procedure where participants had to report the color of a bar presented in the center of a screen and ignore flanking stimuli. These were either a series of crosses or color words. In the second part of the procedure participants were asked to name the ink color of a Stroop word presented for 100 ms. This procedure was first reported by Tipper (1984). He found that normal participants exhibit a measurable inhibition effect, i.e., increased reaction time to name the ink color of the Stroop word (in the regions of 40 ms) when the to=be= named color in the second part of the procedure was the same as the to-be= ignored color word (distracter) in the priming condition. This was compared to a control condition where the ignored priming was a series of crosses. Participants were divided into high and low schizotypy groups on the basis of a median split on the STA scale. R was found that while low scorers

A. Beech and L. Williams

481

showed an inhibitory effect (again of about 40 ms), high scorers did not. In regard to the high schizotypy group's performance a small facilitation effect (15 ms) was found. A significant negative correlation between schizotypy score on the STA scale and the level of inhibition was also observed, indicating that the inhibition decreases with higher levels of schizotypy. It was, however, decided that this particular experimental design was unsatisfactory because of the different nature of the stimuli in the priming and control trials. In a following study (Beech, Baylis, Smithson & Claridge, 1989a), lists of Stroop words were presented under a number of conditions, the two main ones of interest being, control and negative priming conditions. Participants had to name the ink color of Stroop words presented at one of three presentation times 100, 250 and 500 ms. In the negative priming condition the to-be-ignored Stroop color word was the same as the to-benamed ink color of the next Stroop word. In the control condition the to-beignored Stroop color word was unrelated to the prior to-be-named ink color of the next Stroop word. A measure of inhibition was obtained by comparing the difference in reaction time between control and negative priming trials. Subjects were again divided into low and high schizotypal groups. At the 100 ms presentation time low schizotypal subjects showed a substantial inhibition effect (about 30 ms). A significant facilitation effect was found in very high schizotypal scorers (those scoring higher than one standard deviation above the mean on the S T A scale); mean value 49 ms. This result suggests that distracting information is not inhibited and helps subsequent processing in this group. It should also be noted that at the 250 ms presentation time both groups showed a measurable inhibition effect. At the 500 ms presentation no inhibitory or facilitatoryeffect was measured in either group, even though the time course of the inhibition effect can be reliably measured for up to two seconds. These results may have occurred because at longer presentation times conscious biases begin to influence performance. Although it appears that distracting information in high schizotypcs is not inhibited and so is available for further processing, an alternative explanation for the these results needs to be considered. In that high schizotypcs, in the task, may have shown reduced or reversed negative priming because they may only be able to analyze a fragment of the irrelevant distractcr information. In other words, high schizotypes might not be able to complete their perceptual analysis of distractcr information in this prcattentive time frame. This explanation would be consistent with evidence from other studies that schizotypal participants and schizophrenics use local

482

Chapter I 1

rather than global pre-attentive grouping strategies (Schwartz-Place & Gilmore 1980; Rawlings & Claridge, 1984). A second possibility is that the masking stimulus, routinely used in such experiments to prevent further processing at the end of the stimulus presentation time, had a greater effect upon the processing of information in high schizotypes than low sehizotypes. This proposal would appear to fit with the findings from backward masking studies that suggest that there is evidence for slowed information transfer in both schizotypal and schizophrenic subjects (Merritt, Balogh & Leventhal, 1986; Saccuzzo, Hirt & Spencer, 1974). However, it seems likely that the significant facilitation effect found for those very high schizotypy scorers could have only been produceA if they were able to analyze distracters to a semantic level of representation. But to fully discount the possibilities of incomplete perceptual analysis and the differential impact of the masking stimulus, and to provide a more definitive test of the reduceM inhibition hypothesis, a further negative priming study was carried out (Beech et al., 1991). In this study there were a number of conditions using either repeated words or semantically related words (CAT - DOG). The task was to verbally name the semantic category of each words (e.g., say "animal" when the word DOG was presented). The experimental procedure was as follows, two overlapping words were presented in a prime condition shortly followed by a probe condition. The ink color of one of these words in the prime condition was red, the other green. The participants' task in half the trials was to categorize the red word and ignore the green word, in the other half of the trials the opposite procedure applied. Two negative priming conditions were used. In the repetition negative priming condition words that had been ignored in the prime condition were re-presented to be categorized in the probe condition (e.g., red DOG - green DOG). In the semantic negative priming condition, target words in the probe condition were semantically related to words that had previously been ignored (i.e., red CAT green DOG). The use of this condition meant that if facilitation occurred in high schizotypal participants it could not be attributed to their failure to fully analyze the distracting information. Participants were again split on the basis of high and low schizotypy. Their results revealed a double dissociation in the data. High schizotypes showed significant facilitation in the semantic negative priming condition compared to a control condition, but no effect when the distracter was identical to the target. Low schizotypes, on the other hand, exhibited significant inhibition in the repetition negative priming condition, but no effect in the semantic negative priming condition. The observation, that -

A. Beech and L. Williams

483

irrelevant material that should have been actively inhibited in the prime condition, facilitates the recognition of subsequently presented material suggests a way of understanding what might be happening in the positive symptoms of schizophrenia. In that irrelevant stimuli may set off a chain of associations that are experienced as unwanted intrusions into consciousness, viz., hallucinatory experience. Recently, Ferraro and Okerlund (1996) put forward another explanation as to why reduced negative priming was found in high schizotypes. They suggested that the apparent reduction in negative priming among high schizotypes was due to the high and low schizotypy groups not being not equated on simple motor reaction time. They hypothesized this lack of equivalence could produce either an under or overestimate of negative priming. To examine this notion they used a case discrimination, letter identification task reported by Tipper and Cranston (1995). A measure of simple motor reaction time was also obtained for the participants. Consistent with previous findings, they found a significant negative correlation between negative priming and schizotypy score. Importantly, they also ruled out the possibility that group differences in negative priming could be accounted for by differences in motor response, because no differences were found between high and low groups in motor reaction time.

Inhibitory processes in schizotypal subgroups Recently the second author (Williams, 1995) has extended this work in several ways. Further attention was given to semantic as well as repetition priming conditions. Data obtained from here, could provide additional support for the suggestion that positive schizophrenic symptoms result from an inability to inhibit the chain of associations produced by irrelevant stimuli. Instead of relying on a single measure of schizotypy to identify high and low groups, subgroups were defined by their pattern of scores on scales that loaded highly on the positive and negative dimensions of schizotypal traits revealed in factor analytic studies (e.g., Bentall, Claridge & Slade, 1989; Kendler & Hewitt, 1992; Muntaner, Garcia-Sevilla, Fernandez & Torrubia, 1988). Here three scales were used: the Magical Ideation Scale (Eckblad & Chapman, 1983); the Physical Anhedonia Scale (Chapman, Chapman & Rawlin, 1976); and the revised Social Anhedonia Scale (Eckblad, Chapman, Chapman & Mishlove, 1982). Cluster analysis of participants' scores on these scales produced four schizotypal subgroups. The groups were identified as: Low Schizotypy,

484

Chapter 11

Ideational/Perceptual Disturbance (positive schizotypy), Physical Anhcdonia (negative schizotypy) and Cognitive Disorganization/Social Withdrawal (positive and negative schizotypy). It was hypothesized that the negative priming performance of these subgroups, would provide a more definitive test of the hypothesis that reduced cognitive inhibition is spccificaUy associated with positive symptoms. The priming procedure was the same as used in the Beech ctal. study (199 I) and included both repetitionnegative and semantic negative priming conditions. However, to examine the possible influence of the priming task itself, semantic categorization was replaced by verbal identification (word naming). This task was used to encourage an orthographic focus (requiring only a lexical level of analysis for response); as opposed to the focus on semantic processing produced by semantic categorization. An orthographic focus was further encouraged by the inclusion of both concrete and abstract words and the use of low and medium frequency words (Balota & Chumblcy 1984; Scidcnbcrg & McClclland 1989). Also a larger number of prime and target words were used to decrease the potential response benefit that could arise from familiarity with the experimental words (May, Kant & Hasher, 1995, Scidcnbcrg & McClclland, 1989). The Low and the Physical Anhedonia subgroups showed inhibitory effects in the repetition negative priming and semantic negative priming conditions. The performance in the Low subgroup was consistent with that for the low STA group in Beech et al.'s study, indicating that efficient inhibitory processes are associated with a low level of schizotypy generally, and not just with a low level of positive schizotypal features. The performance of the Physical Anhedonia subgroup suggests that inhibitory mechanisms can also operate efficiently for individuals with a high level of purely negative schizotypy. By contrast, the Ideational/Perceptual Disturbance and Cognitive Disorganization/Social Anhcdonia subgroups performance was associated with either reduced inhibition or facilitation (reversed negative priming) in both the repetition negative priming and semantic negative priming conditions. This pattern of negative priming for these subgroups corresponded to the performance of the high STA group in the Beech ct al. study, particularly under the semantic negative priming condition. Given that the Ideational/Perceptual subgroup was defined by positive schizotypal traits such as magical thinking, the presence of rcduce~ inhibition in this subgroup is consistent with the proposal that positive schizophrenic symptoms results from weakened inhibitory processes. Interpretation of

A. Beech and L. Williams

485

reduced inhibition in the Cognitive Disorganization/Social Anhedonia subgroup is not as straightforward. Because this subgroup was characterized by coexistent positive and negative schizotypal features the most parsimonious interpretation would be that reduced cognitive inhibition may underlie some negative, as well as positive, symptoms. However, this interpretation would be inconsistent with the evidence for efficient inhibitory mechanisms in the Physical Anhedonia group. An alternative explanation is that reduced inhibition in the Cognitive Disorganization/Social Anhedonia subgroup is solely accounted for by the presence of positive schizotypal traits. This explanation would concur with the findings reported by Peters, Picketing and Hemsley (1994) that negative priming was inversely related to level of positive symptomatology. Therefore the results from the investigations of negative priming in schizotypy, indexed by both single and multiple scale investigation, provide converging evidence for the proposal that reduced cognitive inhibition underlies positive symptoms of schizotypy,

Inhibitory Processes in Schizophrenia The first author (Beech, Powell, McWilliam & Claridge, 1989b) has investigated schizophrenic performance using the Stroop word list procedure of Beech et al. (1989a) outlined earlier. This study compared with performance on the task by schizophrenics (with predominantly positive symptoms) and a neurotic control group. The latter group showed a level of cognitive inhibition that was equivalent with results found in non-psychiatric participants in previous studies (roughly 30 ms). The schizophrenic group showed reduced negative priming (approximately 9 ms). However, they did not show the facilitation effects observed in highly schizotypal participants in the Beech, et al. (1989a) study. This result may be due to testing the schizophrenic sample when in comparative remission. In an unpublished study, conducted by the first author, it was found that when testing five sufferers over a period of weeks (on the same task as reported above) it was observed that when a sufferer appeared to be more floridly ill they showed less cognitive inhibition than when they seemed comparatively well. However, these observations did not concord with the sufferers self-report, or psychiatric ratings by ward staff, of their current mental state. A second explanation of the failure to find facilitation effects in the schizophrenic group is that all of the schizophrenic sample were on some form of neuroleptic medication and this may have had the effect of normalizing inhibitory processes.

486

Chapter 11

Inhibitory processes in schizophrenic subgroups The second author's study of schizophrenic subgroups (Williams, 1996) again used cluster analysis, being derived from symptom ratings. The Scale for the Assessment of Positive Symptoms (SAPS; Andr~sen 1984) was used to produce subscalo ratings for hallucinations, delusions, positive formal thought disorder and bizarre behavior. The Scale for the Assessment of Negative Symptoms (SANS, 1981) was used to rate affe~ive flattening, alogia, avolition, anh~onia and attentional impairment. Cluster analysis of the scores on the symptom rating scales produc~ four distinct subgroups, defined by: a. high ratings on hallucinations and delusions; b. high ratings for deficit negative symptoms; c. high ratings for all positive symptoms, as well as non-deficit negative symptoms (alogia, attention); d. low ratings overall. Because the first three subgroups showed a broad correspondence with Liddle's (1987) identification of schizophrenic syndromes, they have been rcfcrrexl to by the same names: Reality Distortion, Disorganization and Psychomotor Poverty. Participants in the fourth subgroup exhibited a general absence of symptoms, reporting only intermittent episodes of primarily positive symptoms, hence this cluster was termed Episodic. The parallelism between these subgroups suggests validity for the identified schizotypal typology discussed earlier. Given this phenomenological correspondence, it was predicted that the schizophrenic subgroups would display a pattern of negative priming that mirrored that for the schizotypal subgroups on the task. Table 1 shows the mean and standard deviation priming scores for the schizophrenic groups. The central traits in each schizotypy subgroup that show correspondence with the defining symptoms of the schizophrenic subgroups are also indicated. Table 1 shows that the Reality Distortion and Disorganization subgroups had reduced levels of inhibition compared to the Psychomotor Poverty group. Varying degrees of facilitation can be seen, in these groups, in the repetition negative priming and semantic negative priming conditions. This pattern of facilitated responses paraUeled the observation of either reduced inhibition or facilitation in the corresponding schizotypal Ideational/ Perceptual Disturbance and Cognitive Disorganization/Social Anhedonia subgroups. The Episodic subgroup also showed facilitation in the repetition negative priming and s~mantie negative priming conditions, this finding

487

A. Beech and L. Williams

Table 1. Mean (and standard deviation) of priming scores for schizophrenic and schizotypal subgroups, in ms. Priming Type Subgroup (Schizotypal in italics) .

.

.

.

Repetition/ negative .

.

.

.

.

.

.

.

.

.

.

Semantic/ negative .

.

.

.

.

.

.

.

Repetition/ positive .

Semantic/ positive

.

Reality Distortion*

-29.61 (55.49)

-11.85 (60.32)

59.87 (81.24)

-48.50 (71.13)

Ideational~Perceptual Disturbance*

6.34 (49.80)

- 10.93 (69.43)

26.70 (90.10)

- 11.82 (39.05)

Disorganization**

-20.54 (32.32)

-15.61 (41.30)

25.86 (74.64)

-27.35

Cognitive Disorganization/ Social Anhedonia* *

-12.92 (68.85)

-7.45 (54.32)

55.75 (39.01)

-25.07 (37.18)

Psychomotor Poverty***

37.67 (33.64)

33.16 (51.87)

-26.53 (36.54)

-37.58 (30.34)

Physical Anhedonia* **

18.52 (66.53)

27.28 (38.34)

-18.26 (51.65)

-33.99 (41.92)

Episodic****

-5.96 (55.28)

-25.20 (35.03)

125.61 (58.25)

-26.76 (32.62)

(40.44)

* Defined by positive symptoms [traits]: viz., hallucinations [perceptual disturbances] and delusions [magical thinking] ** Defined by deficit negative symptoms [traits]: viz., affective flattening, avolition and anhedonia [physical anhedonia] *** Defined by above [*] positive symptoms in combination with additional positive and non-deficit negative symptoms [traits]: viz., thought disorder and bizarre behaviour, alogia and attentional impairment [cognitive disorganisation, social anhedonia] **** Defined by generally low level of all symptoms

perhaps reflecting the latent presence of positive symptoms. By contrast, the schizophrenic Psychomotor Poverty group exhibited the normal negative priming effect for these two conditions, which paralleled the results found in

488

Chapter 11

the Physical Anhexlonia schizotypal group. The existence of a similar pattern of negative priming for schizophrenic subgroups and their schizotypal analogs suggests that the schizotypal typology has construct validity. The results also suggest that the priming data obtained from psychosis-prone individuals can bc intcrprctexi with some dcgrcc of confidence. Specifically, the pattern of negative priming for the positive symptom subgroups (Reality Distortion and Disorganization in particular) is consistent with indications from schizotypy results that positive symptoms result from a faihrc of cognitive inhibition. On the other hand, the observation of negative priming in the Psychomotor Poverty subgroup adds weight to the previous suggestion that purely negative symptoms cannot bc accounted for in terms of reduced inhibition. In order to confirm the observations that the positive symptom schizophrenic subgroups showed reduced (or reversed) negative priming, Williams (1996) compared the performance of schizophrenic subgroups to a nonclinical sample. The latter sample comprised students unsclcctcd for their level of schizotypy. The Reality Distortion, Disorganization and Episodic subgroups each differed significantly from non-schizophrenic students for repetition and semantic priming conditions. The Psychomotor Poverty subgroup did not differ from the control sample, having the normal pattern of negative priming.

Inhibitory processes and neuroleptic medication In the first author's study (Beech et al., 1989a) schizophrenic participants showed r e d u ~ inhibition, but no facilitation. A further study was therefore carried out to test the proposal that ncurolcptic medication may have a "normalizing" influence upon performance (Beech, Powell, MeWilliam & Claridge, 1990). All the schizophrenics in the 1989a study were on some form of such medication. In the 1990 study the effects of neuroleptics on inhibitory processes were examined in non-clinical sample. Participants completed the Stroop task twice, with a one-week interval between sessions. Half the participants were administered 25 mg of chlorpromazine before the first testing session, the other half a syrup placebo. The procedure was reversed in the second testing session. In line with prediction, medication increased the amount of measured inhibition compared with the placebo condition. These results appear to support the proposal that neuroleptic medication has a "normalizing" effect on negative priming in schizophrenia. However, it should be noted that the chlorpromazine dosage of

A. Beech and L. Williams

489

25 mg used in this study is substantially lower than that prescribed for most schizophrenics. It could have been that the effect of medication on psychiatrically healthy subjects may have differed from that for schizophrenics. Anecdotally the first author found that some subjects in the experiment reported a narrowing of attentional focus for sometime after taking the drug, akin to one subject's description "as sitting at the bottom of a well". However, given that 80% of schizophrenic subjects were taking some form of neuroleptic medication (ranging from 100 to 1750 mg, chlorpromazme equivalent) in the Williams (1996) study where facilitation effects were found, it would appear that medication does not have a normalizing effect on priming performance in clinical samples. Medication also did not covary significantly with subgroups and priming conditions. It could be argued that this null finding simply reflects the limited within-group variation produced by a heavily medicated sample. However, given that 20% of the sample were unmedicated, and that the range of dosages for medicated participants was positively skewed, this explanation can be discounted. These findings suggest that a failure of inhibition is a stable phenomenon for sufferers experiencing manifest as well as latent positive symptoms. In this light, it is possible that the findings reported by Beech et al. (1990) reflect the differential impact of neuroleptics in a healthy non-schizophrenic sample, compared to long-term use in a clinical sample. Positive priming in schizotypal personality and schizophrenia The focus of this chapter so far has been on negative priming as a measure of cognitive inhibition. However, unexpected results concerning positive (or attended) priming in schizotypy and schizophrenia by the second author (Williams, 1995, 1996) also warrant brief attention. These positive priming results have implications for a model of schizophrenic symptomatology. In the Beech et al. (1991) study attended repetition and attended semantic priming conditions were included to confirm that the priming paradigm would produce the usual positive priming effects (i.e., facilitation of responses) and to minimize the possibility that participants would notice the negative priming manipulation. Here significant positive priming effects were found in the attended repetition priming and the attcndexl semantic priming condition in both high and the low schizotypal groups. However, in the Williams (1995) study the Ideational/Perceptual Disturbance and

490

Chapter 11

Cognitive Disorganization/Social Anhedonia groups displayed an inhibitory effect in the attended repetition condition. To avoid confusion with inhibition produced under negative priming conditions, this inhibitory effect will be referred to as "reversed positive priming". All four schizotypy subgroups displayed facilitated responses in the attended semantic priming condition. The unexpected findings of reversed positive priming were also found in three of schizophrenic subgroups: Reality Distortion, Disorganization and the Episodic subgroups. These results will be discussed later. Towards a "Reduced Cognitive Inhibition" Model of Schizophrenic Symptomatology Previously we have discussed the results of the study where there was a double dissociation of the priming data: high schizotypes showed significant facilitation in a semantic negative priming condition, but no effect when the distracter was identical to the target; low schizotypes exhibited significant inhibition in the repetition negative priming condition, but no effect in the semantic negative priming condition. While Filth's (1979) ideas account, in a broad sense, for the performance of high schizotypes, it does not explain these precise results. Therefore, a modal was constructed to account for these findings (Beech et al. 1991). From the observation that ignored information had significant effects in both groups, it would seem that initial perceptual analysis is equivalent in both high and low schizotypal groups. Therefore any differences that give rise to inhibitory or facilitatory effects must result from disruptions in later selective inhibitory processes. Based on this idea, it has been possible to develop a model of the processes producing inhibitory and facilitatory processes. This modal relics on two basic assumptions, derived from modds of selection attention described previously: a. Spreading activation takes place prior to sdective inhibitory processes, drawing on evidence that inhibition takes time to appear (Lowr 1985); b. Spreading inhibition to related concepts is not as great as spreading activation: as the former is part of a limited capacity active system responsible for controlling various forms of information processing, including selective attention; while the latter is part of a relatively large capacity, passive system responsible for encoding environmental stimuli (Johnson & Dark, 1986). Figure 1 shows how the inhibition effect found in low schizotypes (1A)

A. Beech and L. Williams

491

and the facilitation effect in high schizotypes (1B) are explained by the model. In Figure 1A low schizotypes' initial perceptual analysis produces an activation level of 10 units in the semantic representation of the ignored stimulus DOG. Selective inhibition then begins to prevent response to, and awareness of this irrelevant stimulus, the amount of inhibition = 12 units. Therefore, the final output = (10 - 12) produces 2 units of inhibition to the DOG representation. It is assumed that 60 per cent of the initial activation and 50 percent of the inhibition has spread closely to the concept of CAT. Therefore the amount of priming associated with CAT is 10 • 0.6 = 6 units of excitation and 12 • 0.5 = 6 units of inhibition, which cancel each other out. This accounts for the failure to find any semantic priming effects for semantically related stimuli in low schizotypes. In Figure 1B high schizotypes' initial perceptual analysis produces the same activation level of 10 units but there is a lower level of inhibition (10 units), which would explain the absence of repetition priming effects in this group. Again, it is assumed that 60 percent of the initial activation and 50 percent of the inhibition has spread to the closely related concept - CAT. Therefore the amount of priming associated with CAT is 10 • 0.6 = 6 units of excitation and 10 x 0.5 = 5 units of inhibition. Thus the final output is one unit of semantic facilitation. However, this model relies on the notion, inherent in earlier models of attention (e.g., Collins & Loftus, 1975), that the automatic activation of information occurs as an all-or-none phenomenon. Here, the automatic spreading of activation is not influenced by conscious attention. Given that the development of this reduced inhibition model relies on priming evidence it is pertinent to apply this view of automatic activation to a more specific explanation of priming effects. Until recently, most explanations of priming effects were based on the premise that the mere presentation of a word will activate the lexical representation of that word. In repetition priming, the ability to respond more quickly to a second presentation of this word occurs because its representation has already been activated. The facilitation of responses under semantic priming conditions is thought to results from the automatic spreading of this activation to semantically related representations in memory (Neely, 1977); as described in the above example where 60 percent of the activation from DOG spreads to the related word CAT. This reduced inhibition model, which draws on such explanations, requires some modification in view of more recent theoretical developments. As touched upon earlier, the attentional models that are currently gaining

Chapter 11

492

(A)

Perceptual Input

l

+10

Spreading Activation

-12

Spreading Inhibition

+6 P" -6

! (+1

12)

Selective Inhibition

(

Output 2 units of

- 6)

Output = 0

Negative Priming

(B) Perceptual Input

1

(+1

1

10)

Output = 0

+10

Spreading Activation

+6

-10

Spreading Inhibition ~

-5

!

Selective Inhibition

......

'1

(+6-)

Output 1 unit of Facilitation

Figure 1. A model of inhibitory and faeilitatory processes in (A) low schizotypes and (B) high schizotypes.

A. Beech and L. Williams

493

ground suggest that automatic activation of information may be a continuous, as opposed to an all-or-none, process that is subject to some degree of attcntional control (e.g., Cohen ct al., 1990). From this view, presentation of a word is thought to result in the parallel activation of several different processing codes (or sources of reformation) rather than the automatic access to an individual lcxical code stored in memory (Seidcnberg & McClelland, 1989). In the simplest form of this model, word recognition, for example, revolves the relative activation of orthographic, phonological and semantic codes. Processing word stimuli (as used in the priming tasks) would involve an interaction of these codes, including the influence of feedback from one code to another. Seidcnbcrg and McClelland argue that the nature of the priming task will in part determine which code is the most relevant and is thus given attcntional focus. For instance, the tasks reported in Williams (1995, 1996) used a word naming task to encourage an orthographic attcntional focus, whereas the semantic categorization task reported in Beech ct al. (1991) promotes a semantic focus. It is in this sense that attention is seen to play a role in its conscious influence over which source of activation is focused upon, and allocated the bulk of processing resources (Cohen ctal., 1990). Evidence from both repetition and semantic masked priming procedures has shown that, in some priming tasks, the activation of multiple types of information may cause interference and subsequent inhibition of responses. For others, if the additional information is relevant to the task, it may be used to provide an indirect route to the appropriate response, resulting in response facilitation (Besncr, Smith & McLcod, 1990; Carr & Dagcnbach, 1990; Dagcnbach, Carr & Wilhclmsen, 1989). For example, lcxical decision tasks would require a direct computation from orthography to semantic analysis, while word naming tasks would promote direct computation from orthography to phonology. For these tasks the less relevant information (phonological or semantic) would potcntiaUy interfere with the relevant information and would therefore need to be consciously suppressed. If the suppression of irrelevant reformation detracts from the ability to facilitate relevant information, responses will be inhibited (Cohen ct al., 1990). Inhibitory priming effects have also been found under conditions of semantic priming when the task requires a lcxical decision but the focus is directed to orthographic information by using a new set of learned words (Carr & Dagcnbach, 1990). Similarly, unfamiliar or low frequency words, as well as abstract rather than concrete words (as used in the Williams, 1995, 1996 studies) will delay word naming responses. It is argued that when words

494

Chapter 11

are familiar one can directly access the phonological output (response) from their initial activation, whereas unfamiliar words require additional time to compute the phonological from the orthographic code. The relevance of this revised view of automatic activation to the reduced cognitive inhibition model outlined above is explored in the final subsection.

Revising the Model Because evidence for the view that automatic processing is under some degree of attcntional control comes largely from masked priming studies its application to the negative priming paradigm (and thus the reduced cognitive inhibition model) is somewhat indirect. In particular, the effects of task requirements on automatic processes involved in negative priming conditions may be minimized because the to-be-ignored information is clearly not given conscious attention. Nonetheless, a dynamic association between automatic processes and inhibition would be consistent with the proposal that the inhibition of irrelevant information occurs at the interface between preattentive (primarily automatic) analysis of this information and the more conscious application of selective attentional mechanisms (Dixon, 1981). In this context it is important to emphasize the point raised previously, that reduced inhibition in high schizotypes and schizophrenics is observed only for stimulus presentations of 100 ms, the same time frame within which this interface is thought to occur (Turvey, 1973). Thus, it is possible that the impact of task requirements on the automatic activation of information may differ according to an individual's level of schizotypy, and that this differential impact may play some role in the subsequent application of inhibitory mechanisms (Neely, 1977). The rationale for the initial reduced cognitive inhibition model outlined above, that high schizotypes (and by extension schizophrenics) are equivalent in their preattentive analysis of information, may therefore apply only when both the task and the experimental stimuli promote a congruent focus on semantic information. The second author's findings concerning positive priming could be explained in terms of the conscious influence of task demands on the automatic activation of information. That is, while the tasks used in all of our studies relied on verbal responses (i.e., phonological codes), other aspects of the tasks employed may have promoted a difference in attentional focus. This could have led to differences in the automatic processing of information, these having an effect on later selective inhibitory mechanisms. It will recalled that in some studies, a verbal naming task was used so that conscious attention

A. Beech and L. Williams

495

would be allocated to orthographic information, in others, using a semantic categorization task, the preference would be for meaning. In the verbal naming task, the characteristics of the stimuli themselves (lower frequency, less familiar, and abstract words) were also selected to promote a focus on orthographic information (Seidenberg & McClelland, 1989). In the attended repetition priming condition, therefore, the simultaneous automatic activation of "less relevant" semantic information might have interfered with the computation from orthography to phonology (verbal naming) for individuals in the Ideational/~erceptual Disturbance and Cognitive Disorganization/ Social Anhedonia schizotypal personality subgroups, thus delaying response times. That is, the attentional resources required for suppression of semantic information in order to focus on orthographic information may have taken away from the resources required to facilitate responses to orthographically identical stimuli. This explanation could also account for the inhibitory responses of the schizophrenic Reality Distortion, Disorganization and Episodic subgroups under conditions of attended repetition priming. Similarly inhibited responses would not have occurred in the attended semantic priming condition because responses here rely on the otherwise irrelevant semantic information, rather than on orthographic similarity. In the semantic categorization task study, the fundamentally different nature of the task requirements allow for the usual positive repetition priming (facilitation) effect instead of the reversed positive priming (inhibition). So that the requirement, in this task that participants semantically categorize the target word may well have focused attention on the activation of semantic rather than orthographic information codes. In addition, predominantly high frequency concrete words were used here and each stimulus word was presented to participants a number of times, factors that may have minimized participants' need to attend closely to the orthographic or perceptual features of the word stimuli. That is, familiarity with the stimulus words may have reduced the demands of orthographic processing and encouraged deeper semantic analysis. To summarize the above speculations, depending upon the task the focus of conscious attention might have a different effect on automatic processes involved in priming for those exhibiting high levels of positive traits, as compared to those with low schizotypy generally or with a strong presence of predominantly negative traits. Similarly, conscious influences may operate differentially for those exhibiting positive symptoms of schizophrenia, as compared to those with deficit negative symptoms. Nevertheless, these explanations must be viewed as tentative because they are based on indirect

496

Chapter 11

evidence from conventional masked priming studies (Carr & Dagenbach, 1990; Dagcnbach ct al., 1989; Fricdrich ct al., 1991). However, they do accord with indications from other priming studies into schizophrenia (e.g., Kwapil, Hcglcy & Chapman, 1990; Manschrcck ct al., 1988) that prcattentivc (primarily automatic) activation may in fact be heightened in high schizotypy and schizophrenia, an alteration that may produce interference under certain task conditions. Kwapil ct al. used a word naming task, which would have encouragext participants to focus attention on the orthographic nature of the stimuli, hero schizophrenics (particularly those with thought disorder) showexi heightened prcattcntivc (primarily automatic) activation of semantic associates. Evidence that alterations in preattcntive processing, produced by task demands, may have an effect on later selective mechanisms. In the Kwapil study there were no differences in inhibition between schizophrenics and controls. This was taken to suggest that the heightened semantic activation may have impacted on inhibitory mechanisms in the schizophrenic group. If one accepts the assumption that prcattentive analysis of stimuli does indeed have an impact on the operation of inhibitory mechanisms, it is clearly be necessary to extend the reduced cognitive inhibition model. Interestingly, a more generalized model would be consistent with the recent conclusion from May ct al.'s (1995) review of negative priming studies, that the inhibitory mechanism associated with negative priming is flexible and adapts to shifts in participants' strategies induced by experimental demands. In suggesting a revised model the assumption is made that semantic activation is heightened in individuals with either a strong presence of positive schizotypy or with positive schizophrenic symptoms. From this assumption it is possible to add two components to the original model as follows: 1) In priming tasks that promote a focus on stimulus orthography, the heightened semantic activation in high schizotypy and schizophrenia will produce interference, such that there is response inhibition when attended priming stimuli are re-presented as target stimuli. When the task focus is on semantic information, however, responses will be facilitated. 2) For unattended priming conditions, the combination of heightened semantic activation and reduced cognitive inhibition in high schizotypes and schizophrenics may produce facilitation, rather than inhibition, of responses. Facilitation will most likely occur under unattended semantic priming conditions. These revisions are clearly very general in nature. Hopefully future studies will investigate more systematically the impact of differential task

A. Beech and L. Williams

497

requirements, so that it will be possible to refine the predictions of such a model. Conclusion

It has been suggested that positive symptoms (i.e., hallucinations, delusions and thought disorder) of schizophrenia may arise due to the suffcrcr's inability to screen out distracting information. Specifically, Frith (1979) has suggested that these more cognitive symptoms arise from a defect in the mechanism that controls and limits the contents of consciousness. We suggest that the basis of this mechanism is a weakened ability to actively suppress irrelevant information. This notion draws on ideas from cognitive psychology where it has been suggested that, in the process of becoming consciously aware of some aspect of our environment, there is active inhibition of information that wc do not become consciously aware of. The experimental paradigm used in our investigations of this proposal relics on the phenomenon of negative priming, the delayed reaction time observed when participants respond to a target stimulus that was previously ignored (and thus inhibited from response processes). Several studies have been conducted with individuals assessed for their level of proneness to schizophrenia (schizotypy). We have consistently found participants with high levels of positive schizotypal personality traits show reduced inhibition or even facilitation under negative priming conditions. Specifically, facilitation occurred under conditions of semantic negative priming. In contrast, those with either a low level of schizotypy or with high levels of negative traits show the expected inhibition effects. The normal performance of the negative schizotypy subgroup suggests that a failure of cognitive inhibition is associated with positive schizotypal personality features and, by extension, with positive schizophrenic symptoms. These results wcrc confirmed in investigations of inhibitory processes in schizophrenic participants. Subgroups defined by positive symptoms (including the episodic experience of these symptoms) showed a similar lack of measured cognitive inhibition, whereas the subgroup characterized by deficit negative symptoms displayed efficient inhibitory processes. Thus, findings from schizophrenic groups provided further support for the proposal that positive symptoms result from a failure of cognitive inhibition. Interestingly, ncurolcptic medication did not normalize performance in the positive symptom subgroups. On the basis of our negative priming results, we developed a model

498

Chapter ] 1

based on excitatory and inhibitory processes where we propose that, in high sehizotypes and in schizophrenics, irrelevant information is associated with a chain of semantic activation which is not suppressed. This information thus intrudes on conscious awareness. Such a process could account for the unusual semantic content of positive symptoms, such as thought disorder. This initial model uses concepts from earlier models of attention: automatic spreading activation and inhibition of associates. In our revised model, prompted by unexpected findings produc~ when the priming task was changed, we suggest that the impact of this automatic activation may differ according to conscious task requirements. Under this revised model, therefore, positive symptoms might reflect a dynamic association between the automatic activation of information and conscious selective processes associated with inhibition. References

Allport, D. A. (1980). Attention and performance. In G. Claxton (Ed.), Cognitive psychology: New directions (pp. 112-153). London: Routledge and Kegan Paul. American Psychiatric Association (1980). Diagnostic and statistical manual of mental disorders, third Edition. Washington, DC: APA. Andreasen, N. C. (1981). Scale for the assessment of negative symptoms (SANS). Iowa City: University of Iowa. Andreasen, N. C. (1984). Scale for the assessment of positive symptoms (SAPS). Iowa City: University of Iowa. Balota, D. A., & Chumbley, J. I. (1984). Are lexical decisions a good measure of lexical access? The role of word frequency in the neglected decision stage. Journal of Experimental Psychology: Human Perception and Performance, 10, 340-357. Beech, A., & Claridge, G. (1987). Individual differences in negative priming: Relations with schizotypal personality traits. British Journal of Psychology, 78, 349-356. Beech, A., Baylis, G. C., Smithson, P., & Claridge, G. (1989a). Individual differences in sehizotypy as reflected in measures of cognitive inhibitiorL British Journal of Clinical Psychology, 28, 117-129. Beech, A., Powell, T., McWilliam, J., & Claridge, G. (1989b). Evidence of reduc~ "cognitive inhibition" in schizophrenia. British Journal of Clinical Psychology, 28, 109-116. Beech, A., PoweU, T. J., McWilliam, J., & Claridge, G. S. (1990). The effect

A. Beech and L. Williams

499

of a small dose of chlorpromazme on a measure of cognitive inhibition. Personality and Individual Differences, 11, 114 l- 1145. Beech, A., McManus, D., Baylis, G., Tipper, S., & Agar, K. (1991). Individual differences in cognitive processes: Towards an explanation of schizophrenic symptomatology. British Journal of Psychology, 82, 417426. Bentall, R. P., Claridge, G. S., & Slade, P. D. (1989). The multidimensional nature of schizotypal traits: A factor analytic study with normal subjects. British Journal of Clinical Psychology, 28, 363-375. Besner, D., Smith, M. C., & MacLeod, C. M. (1990). Visual word recognition: A dissociation of lexical and semantic processing. Journal of Experimental Psychology: Learning, Memory and Cognition, 16, 862-869. Broadbent, D. E. (1958). Perception and communication. London: Pergamon Press. Broadbent, D. E. (1971). Decision and stress. London: Academic Press. Carr, T. H., & Dagenbach, D. (1990). Semantic priming and repetition priming from masked words: Evidence for a centre-surround attentional mechanism in perceptual recognition. Journal of Experimental Psychology: Learning, Memory and Cognitmn, 16, 341-350. Chapman, J. (1966). The early symptoms of schizophrenia. British Journal of Psychiatry, 112, 225-231. Chapman, L. J., & Chapman, J. P. (1980). Scales for rating psychotic pyschotic like experiences. Schizophrenia Bulletin, 6, 476-489. Chapman, L. J., Chapman, J. P., Kwapil, T. R., Eckblad, M., & Zmser, M. C. (1994). Putatively psychosis-prone subjects 10 years later. Journal of Abnormal Psychology, 103, 171-183. Chapman, L. J., Chapman, J. P., & Rawlin, M. L. (1976). Scales for physical and social anhedonia. Journal of Abnormal Psychology, 85, 374-382. Chapman, L. J., Chapman, J. P., & Rawlin, M. L. (1978). Body image aberration in schizophrenia. Journal of Abnormal Psychology, 87, 399407. Claridge, G. S. (1985). Origins of mental illness. Oxford: BlackweU. Claridge, G. S. (1987). "The schizophrenias as nervous types" revisitext British Journal of Psychiatry, 151,735-743. Claridge. G. S., & Broks, P. (1984). Sehizotypy and hemisphere function - I. Theoretical considerations and the measurement of schizotypy. Personality and Individual Differences, 5, 633-648.

500

Chapter 11

Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990). On the control of automatic processes: A parallel distributed account of the Stroop Effect. Psychological Review, 82, 407-528. Dagenbach, D., Carr, T. H., & Wilhelmsen, A. (1989). Task-induced strategies and near-threshold priming: Conscious influences on unconscious perceptiorL Journal of Memory and Language, 28, 412443. Dixon, N. F. (1981). Preconsciousprocessing. New York: Wiley. Eckblad, M., & Chapman, L. (1983). Magical ideation as an indicator of schizotypy. Journal of Consulting and Clinical Psychology, 51, 215225. Eekblad, M. L., Chapman, L. J., Chapman, J. P., & Mishlove, M. (1982). The Revised Social Anhedonia Scale. Unpublished test, University of Wisconsin-Madison. Ferraro, F. R., & Okedund, M. (1996). Failure to inhibit irrelevant information in non-clinical schizotypal individuals. Journal of Clinical Psychology, 52, 389-394. Friedrich, F. J., Henik, A., & Tzelgov, J. (1991). Automatic processes in lexieal access and spreading aetivatiorL Journal of Experimental Psychology: Human Perception and Performance, 17, 792-806. Frith, C. D. (1979). Consciousness, information processing and schizophrenia. British Journal of Psychiatry, 134, 225-235. Hewitt, J. K., & Claridge, G. (1989). The factor structure of schizotypy in the normal population. Personality and Individual Differences, 10, 323329. Johnson, W. A, & Dark, V. J. (1986). Selective attention. Annual Review of Psychology, 37, 43-75. Kendler, K. S., & Hewitt, J. (1992). The structure of self-report schizotypy in twins. Journal of Personality Disorders, 6, 1-17. Kwapil, T. R., Hegley, D. C., & Chapman, L. J. (1990). Facilitation of word recognition by semantic priming in schizophrenia. Journal of Abnormal Psychology, 3, 215-221. Liddle, P. F. (1987). The symptoms of chronic schizophrenia. A reexamination of the positive-negative dichotomy. British Journal of Psychiatry, 151, 145-151. Lowe, D. G. (1979). Strategies, context, and the mechanism of response inhibitiort Memory and Cognition, 7, 382-389. Lowe, D. G. (1985). Further investigations of inhibitory mechanisms in attentiorL Memory and Cognition, 13, 74-80.

A. Beech and L. Williams

501

McGhie, A., & Chapman, J. (1961). Disorders of attention and perception in early schizophrenia. British Journal of Medical Psychology, 34, 103116. Manschreck, T. C., Maher, B. A., Milavetz, J. J., Ames, D., Weisstem, C. C., & Schneyer, M. L. (1988). Semantic priming in thought disordered schizophrenic patients. Schizophrenia Research, 1, 61-66. May, C. P., Kane, M. J., & Hasher, L. (1995). Determinants of negative priming. Journal of Experimental Psychology: Learning, Memory and

Cognition, 21,422-435. Merritt, R. D., Balogh, D. W., & Leventhal, D. B. (1986). Use of a m~tacontrast and para-contrast procedure to assess visual information processing of hypothetically schizotypic college students. Journal of Abnormal Psychology, 95, 74-80. Muntaner, C., Garcia-Sevilla, L., Fernandez, A., & Torrubia, R. (1988). Personality dimensions, schizotypal and borderline personality traits and psychosis proneness. Personality and Individual Differences, 9, 257268. Neely, J. H. (1977). Semantic priming and the retrieval from lexical memory: Roles of inhibitionless spreading activation and limited-capacity attentio~ Journal of Experimental Psychology: General, 106, 226-254. Neill, W. T. (1977). Inhibitory and facilitatory processes in attention.

Journal of Experimental Psychology: Human Perception and Performance, 3, 444-450. Neill, W. T., Valdes, Terry, K. M., & Gorfein, D. S. (1992). The persistence of negative priming: If. Evidence for episodic trace retrieval. Journal Of Experimental Psychology: Learning Memory and Cognition, 18. 9931000. Neill, W.T., & Westberry, R. L. (1987). Selective attention and the suppression of cognitive noise. Learning, Memory and Cogmtion, 13. 327-334. Neisser, U. (1967). Cogmaon and reality. San Francisco, CA: Freeman. Peters, E. R., Pickering, A. D., & Hemsley, D. R. (1994). Cognitive inhibition and positive symptomatology in schizotypy. British Journal of Clinical Psychology, 33, 33-48. Posner, M. I. (1978). Chrometric explanations of mind. Hillsdale, NJ: Erlbaum. Rawlings, D., & Claridge, G. (1984). Schizotypy and hemisphere functionIII. Performance asymmetries on tasks of letter recognition and localglobal processing. Personality and Individual Differences, 5, 657-663.

502

Chapter 11

Saccuzzo, D. P., Hirt, M., & Spencer, T. J. (1974). Backward masking as a measure of attention in schizophreni~ Journal of Abnormal Psychology, 83, 512-522. Schwartz-Plac,, E. J., & Gilmore, G. C. (1980). Perceptual organisation in schizophrenia. Journal of Abnormal Psychology, 89, 409-418. Seidenberg, M. S., & McClelland, J. L. (1989). A distributed, developmental model of word recognition and naming. Psychological Review, 96, 523568. Straube, E. R., & Oades, R. D. (1992). Schizophrenia: Empirical research andfindings. New York, NY: Academic Press. Stroop, j. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18, 643-62. Stuss, D. T., & Benson, D. F. (1986). The frontal lobes. New York: Raven Press. Tipper, S. P. (1984). Negative priming in visual selective attention. Unpublished D. Phil thesis, University of Oxford. Tipper, S. P. (1985). The negative priming effect: Inhibitory priming by ignored objects. Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 37, 571-590. Tipper, S. P., & Cranston, M. (1985). Selective attention and priming: Inhibitory and faeilitatory effects of ignored primes. Quarterly Journal of Experimental Psychology, 37A, 581-611. Turvey, M. T. (1973). On peripheral and central processes in vision: Inferences from an information processing analysis of masking with patterned stimuli. Psychological Review, 80, 1-52. Van der Heijden, A. H. C. (1981). Short-term visual informationforgetting. London: Routledge and Kegan Paul. Venables, P. H. (1992). Hippoeampal function and schziophrenia: Experimental psychological evidence. Annals of the New York Academy of Science, 111 - 126. Williams, L. M. (1994). The multidimensional nature of schizotypal traits: A cluster analytic study. Personality and Individual Differences, 16, 103112. Williams, L. M. (1995). Further evidence for a multidimensional personality disposition to schizophrenia in terms of cognitive inhibition. British Journal of Clinical Psychology, 34, 193-213. Williams, L. M. (1996). Cognitive inhibition and schizophrenic symptom subgroups. Schizophrenia Bulletin, 22, 139-151.

Cognitive Science Perspectives on Personality and Emotion - G. Matthews (Editor) 1997 Elsevier Science B.V.

CHAPTER 12 Attention, Working Memory and Arousal: Concepts Apt to Account for the "Process of Intelligence" Edward Necka

According to some cognitive psychologists, the concept of intelligence has lost its explanatory power, if it ever had any (Neisser, 1979). Instead of trying to "explain" human performance by saying that a person is intelligent, they try to reveal basic cognitive mechanisms that underlie intelligent behavior (Hunt, 1978, 1980; Hunt & Lansman, 1986; Steinberg, 1985). From this point of view, intelligence is an everyday rather than scientific concept (Neisser, 1979); as such, it is fuzzy, unclear and has too many meanings. R is a word we use instead of precise but "technical" descriptions of the way in which an individual deals with cognitive tasks. These "technical" descriptions pertain to the cognitive processes of perception, memory, attention, thinking, and problem solving. In this way, we can reformulate the dispute concerning the trait versus process distinction. If the trait of intelligence exists, it probably determines human performance in cognitive tasks permanently. If it does not exist, one can speak only about continually changing processes. Psychological assessment and everyday observations seem to suggest that the theory of intelligence needs to operate with both stable traits and unstable processes. The aim of this chapter is to provide a model of intelligence that combines the trait stance, represented, for instance, by Eysenck (1988, 1994) and Jensen (1987a, 1992), with the information-processing approach, as originated by Hunt or Sternberg. The model assumes that there exist some structural determinants of intelligent behavior that may be understood as foundations of the stable "trait of intelligence." These determinants refer to the structural limitations of attention and working memory. The model also assumes that - in spite of the structural limitations - an individual's ability to solve problems (including IQ tests) varies continuously as the level of arousal gets higher or lower. The structural limitations define the absolute capability of an individual, whereas the actual level of arousal, and especially its fluctuations, define the actual ability to deal with certain cognitive tasks in certain circumstances. "The process of intelligence" may therefore be defined in terms of the way in which a person deals with cognitive task while being

504

Chapter 12

determined both by stable structural limitations and constantly changing level of arousal. Since the distinction between absolute and actual capabilities is important for our considerations, it seems worth further clarification. An analogical distinction serves to differentiate between capacities and abilities. The former refer to what an individual would be able to achieve at best, i.e., in the most advantageous circumstances and when all necessary conditions are met. The latter refer to what an individual is really able to achieve, taking into account the actual conditions and level of motivation, which is usually intermediate (Ackerman, 1994). The distinction between capacities and abilities relates to Cattelrs (1971) theory of fluid and crystallized intelligence, although these two distinctions are not synonymous. The problem may also be described in terms of competence and performance. Normally, people's actual performance is much lower than their competence level, due to many factors that impede perfect performance, like psychophysiological states of the organism, lack of motivation, and unfavorable external circumstances (stress, noise, temperature, etc.). It seems that cognitive science has focused too much on competence, and has tended to neglect performance. Our distinction is therefore necessary in order to describe "the process of intelligence" more completely. In this way, our stance is similar to Ackerman's (1994), who proposes using the term "intelligence-as-typicalperformance." After presentation of basic notions and assumptions, we will describe a new theoretical model of "the process of intelligence." The empirical verification of the basic assumptions and propositions of the model will be presented subsequently. The chapter ends with conclusions concerning the validity of the proposed model of intelligence, as well as its relevance for the cognitive science approach. Theoretical Notions We will use three basic concepts in this chapter: attentional resources, working memory capacity, and arousal. The concept of attenaon is threefold. First, it refers to the ability to select relevant stimuli and to ignore irrelevant ones (Broadbent, 1958; Cherry, 1953). This aspect of attention is typically investigated with the use of dichotic listening tasks. Second, attention means the ability to maintain mental effort during a considerable period of time (Nuechtcdein, Parasuraman & Jiang, 1983). This aspect is normally studied with continuous performance tasks. Third, it denotes the ability to employ

E. Necka

505

cognitive control over the task at hand, particularly, over competing tasks performed simultaneously (Kahneman, 1973; Norman & Bobrow, 1975). This aspect of the phenomenon is investigated in the dual task paradigm. It is the third aspect of attention that has proved to matter for intelligence: psychometrically assessed intelligence correlates with efficiency of dual task performance (Hunt, 1980; Hunt & Lansman, 1982, 1986; Lansman & Hunt, 1982; Necka, 1996; Stankov, 1983). These findings are interpreted in terms of the resource availability model of attention (Kahneman, 1973; Norman & Bobrow, 1975). According to this explanation, a higher level of intelligence is determined by an increased amount of cognitive resources. Thanks to this advantage, some people are more able to deal with tasks that require the efficient control of many simultaneous actions or processes. Such people are thus called "intelligent," because intelligence tests, and also real-life criterion tasks, typically require that a great amount of attentional resources be allocated simultaneously. Necka (1996), in addition to empirical data supporting this line of reasoning, provides a theoretical model that explains how individual differences in attentional resources manifest themselves in differences concerning reaction times in dual task experiments. The concept of attentional resources has been criticized for its alleged futility in psychological theory (Hirst & Kalmar, 1987; Navon, 1984). Undoubtedly, if "resources" are to explain human performance in the dual task paradigm, they should refer to some kind of non-specific, general "mental energy" distributed among simultaneous actions on the basis of their importance or urgency, according to the "allocation policy" adopted by the individual. Of course, descriptions such as "mental energy" or "allocation policy" are purely metaphorical, to an extent which is unacceptable for many theoreticians used to strict language and distinct notions. Resource descriptions also rely on the tacit assumption of the existence of a "homunculus" who adopts and executes the "resource allocation policy." But the most important criticism refers to the alleged non-specificity of resources. It may well be that different types of action require specific kinds of resources, so the "homunculus" cannot decide freely about feeding a chosen action or process at the expense of another. One also has to take into account the possibility that different action or processes rely on specific mental structures, e.g., specific regions of knowledge or even specific regions of the brain's anatomy. And the more specificity we have to ascribe to mental resources the less useful the notion appears, eventually serving as a "theoretical soup stone" (Navon, 1984).

506

Chapter 12

The "homunculus" problem is sometimes conceptualized in terms of executive processes, i.e., the processes that allocate attentional resources and exert activo control over a number of concurrent cognitive tasks. In the dual task paradigm, attention has to switch from one task to another. It is sometimes claimed that the very act of switching of attention elongates reaction time and causes errors, so worse performanc~ in the dual task condition, as compared to the single task condition, may be ascribed to the switching operation. There is a dispute among theoroticians concerning the existence of a specific "time-sharing" ability, which might be responsible for the efficient control of concurrent tasks (Brookings, 1990; Zeljko & ~verko, 1994). Executive functions and processes, responsible for switching attention, are of course different from resources. They may be conceptualized as active operations performed on passive reservoir of resources. The question arises, whether the empirical effects of dual task performance, i.e., the increase of reaction time and error rate, can be accounted for by executive functions merely. Similarly, individual differences in dual-task performance might relate to specific factors like the alleged time-sharing ability. In my opinion, such explanations are rather unlikely, due to the fact that performance in the dual task paradigm depends not only on the presence or absence of a concurrent task, but also on the difficulty level of the primary task. In other words, reaction time and error rate typically increase in the dual task condition, as compared to the single task condition. However, they also increase when one of these tasks, called the primary task, is set to higher difficulty levels. For instance, performance depends on the set size of letters that have to be detected and compared with the probe letter, or on the presence or absence of distractors, particularly - but not solely - in the dual task condition (Necka, 1996; Szymura & Necka, 1996). So, if the increased difficulty level of the primary task causes lower performance indices although the secondary task is already present, and the "resources" were already "switched" to it earlier, these empirical effects cannot be attributed solely to control and executive functions or to specific time-sharing abilities. Therefore, attentional resources should remain in psychological theory as a useful concept accounting for human performance in the dual task paradigm, although the role of executive processes and time-sharing ability should not be neglected (ef. Gopher, 1986). in this chapter, we wish to relate intelligence to attentional resources rather than to executive functions. The amount of attentional resources is conceptualized as one of structural determinants of human intellectual performance. However, one should not forget the possibility that the

E. Necka

507

efficiency of executive functions may also determine our intellectual functioning. It may well be that, to some extent, the more efficient we are in exerting cognitive control over simultaneous tasks the more intelligent we appear in tests and other criterion tasks. These mechanisms are close to the area of cognitive strategies, though, and cannot be discussed extensively in this chapter (el. Baron, 1978; Hany, 1991; Kossowska & Necka, 1994). The concept of working memory, as introduced by Alan Baddeley (1986), refers to a structure that is responsible for two fimctions. The "central executive" performs basic cognitive operations, due to which the present task (or simultaneously tackled multiple tasks) is performed, whereas its "slave subsystems" (articulatory loop and visual-spatial "scratch-pad") are involved in short-term retention of information that results from past processing and may be needed for further elaboration. Opposing the classic notion of STM, the concept of working memory stresses the dynamic role of the memory structure, which, metaphorically speaking, is like an active processor rather than passive data store. Performance in working memory tasks correlates with psychometric intelligence - an effect that has been shown in many experiments (e.g., Kyllonen & Christal, 1990; Necka, 1992; Vernon, 1983, 1985; Vernon, Nador & Kantor, 1985). Subjects who score high on IQ tests appear to be quicker and more accurate in working memory tasks. For instance, in Necka's (1992) experiments, more intelligent subjects committed fewer errors in tasks that required retention of a set of digits or letters in STM for several seconds. This relationship was particularly salient when the set size, i.e., the number of items to memorize, was bigger, whereas in the easier versions of the tasks no relationships with IQ were observed. Such data are usually interpreted in terms of the limited capacity theory of working memory. If the human capacity to store relevant information for a short time is severely limited, individual differences in such capacity may determine success or failure in complex cognitive tasks. In other words, if an individual can keep more "chunks" of information in the STM store, he or she is able to deal with more complex tasks, these may be the tasks that best indicate the measured level of intelligence, e.g., IQ tests. Notwithstanding its fundamental role in contemporary cognitive theorizing, the conception of working memory also raises fundamental doubts and questions. Is the "central executive" a new version of the homunculus? How do the structures constituting the working memory system communicate with each other? In particular, does the necessity of consulting the slave subsystems increase the burden to which the central executive is subjected

508

Chapter 12

while doing the current task? How long and to what extent can slave subsystems continue to hold information for a short time without any active intervention by the central executive? Is the articulatory loop independent of the scratch-pad in terms of its ability to retain information, and is the overall capacity of working memory just the sum of the two capacities provided by the loop and the scratch-pad? Or maybe the overall capacity of working memory just reflects the ability of the whole structure, including the central executive and its slave subsystems, to retain as many pieces of information as possible? And if the answer to the last question was positive, what would be the difference between the notion of working memory capacity and the traditional notion of the restricted capacity of the short term store (Miller, 1965)? But the most important problem amounts to the fact that it is often very difficult to separate - on the empirical level - the central executive from its slave subsystems. The central executive performs current processing, which always refers to particular content, like words, numbers or other pieces of information. These "chunks" of information are undoubtedly held in the articulatory loop or scratch-pad, where they wait to be processed in due course. How, then, can we study the central executive without studying the articulatory loop or scratch-pad at the same time? And if we want to measure the capacity of the slave subsystems, we have to give a subjects a task to do, e.g., a typical short term memory task, which is impossible to do without engaging the central executive. So, neither part of the system can operate on its own; therefore, its processing capacity cannot be assessed purely, without unintentionally assessing the efficiency with which other parts of the system operate. Vandierendonck, De Vooght and Van tier Goten (1995) recently proposed that the burden of the central executive's current activity be assessed with the use of the random interval generation (RIG) task (see also: Baddeley, 1996). Subjects were asked to tap a key at random during mental activities of different complexity. The more demanding the task was, the less random a subject's tapping behavior became, thus allowing indirect assessment of the burden to which the hypothetical central executive was subjected. This measure is thought to be a "pure" one, since it does not depend on the activity of the slave subsystems. The technique proposed by Vandierendonck and his colleagues is very promising, but a new question arises, whether it refers to the central executive, as hypothesized by Baddeley, or to some kind of general executive function already discussed with reference to the attentional mechanisms. In fact, there is some overlap between attention

E. Necka

509

and working memory in contemporary theory. For this reason, I will suggest in the following sections of the chapter - possible ways of development of the theory of working memory, particularly concerning the problem of cooperation between central executive and its slave subsystems. Anyway, what is meant here under the term "working memory capacity" refers to the overall ability of the cognitive apparatus to hold as many items of information for a short time as possible, regardless of which function or structure is primarily responsible for this task. The third notion of interest to us is the concept of arousal. Physiologically, arousal is a complex syndrome of excitation of certain bodily parts, mainly the central and autonomous nervous systems, but also blood circulation, respiration, and endocrine systems. The critical aspect of physiological arousal refers to the activation of the brain cortex, which results from non-specific afferent transmission performed by the reticular formation (Hebb, 1955). Early conceptions of arousal (e.g., Duffy, 1962) assumed its homogeneity as a unidimensional continuum of states: from coma through deep sleep, light sleep, drowsiness, vigilance, alertness, up to intensive excitation. The level of arousal was believed to depend on the amount of energy dispensed by an organism at the given moment. However, later conceptions of arousal stressed the qualitative differences between various aspects and kinds of physiological excitation. For instance, Lacey (1967) proposed the differentiation of at least three kinds of arousal: cortical, autonomic, and behavioral. He argues that, although all kinds of excitation normally co-exist, there are significant exceptions to this rule. For example, some stimulants and tranquilizers affect behavioral, but not cortical, excitation. Many arguments for the existence of different physiological mechanisms of arousal were found in experiments with animals, which showed that, after specific brain lesions, subjects might get excited cortically but not behaviorally (Klonowiez, 1984). Eysenck (1967), too, stressed the necessity of not treating arousal as a unidimensional phenomenon. He distinguishes activation, which refers to the autonomic nervous system and accompanies emotional states, from arousal, which refers to the brain cortex and results from non-specific afferent transmission. According to Eysenck, heightened autonomic activation usually raises the level of cortical arousal due to non-specific afferent transmission; however, the reverse mechanism does not exist, thus showing the functional dissimilarity of two kinds of excitation. Psychologists doing research on arousal had to take into account the above mentioned facts suggesting the existence of qualitatively different

510

Chapter 12

phenomena. They also had to face another problem, i.e., the limited applicability of physiological indices of arousal to psychological research. Variables typically measured in psychophysiological experiments, like heart rate (HR), r response (EDR), or dectrocncephalography (EEG), are sometimes regarded as more objective, more valid, and more reliable than typical psychological variables, particularly self-report data (Klonowicz, 1984). However, the interpretation of psychophysiological variables may raise serious problems, for two reasons. First, objectively the same stimulus may cause different patterns of physiological response, depending on its meaning and depending on the state of expectancy (set) of the subject. Second, people show individual patterns of psychophysiological responding to external stimuli. These differences appear to be qualitative rather than quantitative, and show significant stability over successive trials and situations. Therefore, a researcher may be tempted to register many variables in order to obtain the integrated index of arousal, but such an approach apart from being rather costly- also shows its limited applicability due to low correlations between various aspects of psychophysiological arousal (Klonowicz, 1984; Lacey, 1967; Thayer, 1970). The most popular contemporary theory of arousal has been formulated by Thayer (1967, 1970, 1978, 1989), who also developed a self-report assessment technique. Thayer assumes that arousal is rooted in the general dimension of energy disbursement; however, he argues that the general energy of the organism is likely to be channeled into qualitatively different modes of responding. The author proposed a distinction between energetic and tense arousal. The first one is responsible for the amount of effort and energy invested in an activity or task at hand, and is the "reverse probability of falling asleep" (Corcoran, 1965). The second one is responsible for immediate actions undertaken in order to escape from a dangerous situation or to prevent an organism from threatening events. As we can see, the division into energetic and tense arousal is psychological rather than physiological. In his Activation-Deactivation Adjective Check List Thayer (1978, 1989) differentiated four dimensions of arousal: 1). Energetic (which he calls "general activation"); 2). Vigilant (antagonistic to the state of drowsiness, which he calls "deactivation"); 3). Tense (called "high activation"); 4). Relaxed (called "general deactivation"). "General activation" reflects one's readiness for work, and is a good predictor of cognitive efficiency. "High activation" measures anxiety and

E. Necka

511

withdrawal tendencies. "Deactivation" reflects fatigue, drowsiness, and lack of motivation. Its opposite state, i.e., vigilant arousal, correlates positively with "general activation," with which it comprises the broader concept of energetic arousal. "General deactivation" reflects one's adaptation to the forthcoming events and one's readiness to take an effort. Along with "high activation," with which it is negatively correlated, this dimension fits within the broader concept of tense arousal. Although the adjective check list consists of four scales, Thayer is still convinced of the validity of a twodimensional model of energetic and tense arousal (cf. Thayer, Newman, & McClain, 1994), the more so that the four scales show mutual mtercorrelations, as indicated. The lack of specificity of physiological symptoms of arousal prompted Thaycr to switch to psychological understanding of the term. His self-report assessment tool, which has been used in our own study (see the subsequent sections of the chapter), is also purely psychological in nature. The author takes advantage of the assumption that the physiologically rooted states of activation arc consciously perceived by people as subjective states of being asleep, tense, aroused, energetic, etc. He believes that our central nervous system integrates many different - and dispersed - symptoms of physiological arousal and uses them as the basis for the construction of a generalized emotional state or mood that can be consciously perceived and described using the self-report assessment techniques (see also: Eysenck, 1975; Klonowicz, 1984). In this way, Thaycr does not deny the physiological foundations and mechanisms of arousal. However, his definition and assessment technique refer to the psychological aspects of the phenomenon. From our own point of view it is important to stress that the current level of arousal is probably determined by many different factors. For instance, some individuals are permanently more cortically aroused than others, and there are good reasons to relate these differences to the psychological trait of introversion/extraversion (Eysenck, 1967; Stenberg, Wendt & Risberg, 1993). Arousal is also determined by the intake of substances like caffeine, alcohol etc. External circumstances, e.g., noise (including experimentally induced "white noise") and other forms of intensive stimulation, also raise our level of arousal. We get more aroused when we try hard to achieve a goal, that is, when we exert effort. There arc also spontaneous rhythms of high and low arousal, related to the time of the day. So, there are constitutional, substance-related, situational, motivational, and time-of-the-day related sources of high or low arousal. For the sake of clarity we will ignore these differences for the moment and concentrate on the question whether the actual

512

Chapter 12

level of arousal accounts for "the process of intelligence." We will also ignore, for the moment, the qualitative differences bctwezn various dimensions of arousal. This decision is not due to negligence of the importance of such differences, but is due to the lack of theoretical reasons to construct different models of intelligence concerning various aspects of arousal. In other words, the theoretical model described in the next section deals with arousal as a general construct. However, the question of multidimcnsionality of arousal will be addressed later on, during the discussion of the experimental results.

Assumptions Our basic assumptions refer to the relationship between arousal and attentional resources, on one hand, and arousal and working memory capacity, on the other hand. First, we differentiate between the absolute and momentary values of attentional resources and working memory capacity. The absolute values are not observable, and although they probably differ from person to person, we are not able to measure these differences. What we can measure, though, are the intra- and interindividual differences in the actual, momentary, or transitory values of these parameters. Second, we assume that - according to the experimental findings cited above - intelligence is related to the increased absolute values of both attentional resources and working memory capacity. In other words, intelligent people have more attentional resources to their disposal, and they are more capable of retaining information for a short time, but these relationships refer to the absolute values of both parameters, which we cannot ascertain. The actual ability of an intelligent person to store information for a short time may fall below the level of a person who is much less intelligent. And the actual amount of attentional resources that an intelligent person is able to invest at a certain moment may drop much below the level of somebody who is normally less intellectually able. This may happen because the momentary values of the parameters we talk about vary from situation to situation, and from task to task. Of course, the empirical fmdings suggest an absolute advantage of high IQ people because the testing situation motivates people to do their best, i.e., to invest the majority of their cognitive resources into test performance. Besides, if high IQ subjects start from the higher level, they would be likely to show their advantage over their less intelligent peers even if the fluctuations concerning momentary values of attention and STM capacity were entirely random. Both more and less

E. Necka

513

intelligent persons are presumed to be susceptible to such fluctuations. Anyway, our assumptions are that the absolute values of working memory capacity and attentional resources are differentiated interindividuaUy, whereas the momentary values of these parameters are differentiated both inter- and intraindividually. Third, we assume that the more aroused one is the more attentional resources one can invest into the current task. In other words, the momentary value of attentional resources increases with arousal. On the other hand, high arousal makes a person less able to utilize the content of his/her short term store, which means that the momentary value of working memory capacity decreases with arousal (Figure 1). As already discussed, the distinction between the absolute and actual levels of processing resources relate to the distinction between capacities and abilities, on the one hand, and between competence and performance, on the other hand. According to these definitions, arousal affects abilities or performance, but does not affect capacities or competence. In short, when arousal increases, the efficiency of the attention system amplifies, whereas the efficiency of working memory system diminishes. Possible theoretical mechanisms for such effects will be outlined in the following sections of the chapter. The assumptions stated above have been borrowed from the article by Humphreys and Revelle (1984). The authors proposed that the general inverted U curve relationship between arousal and performance, characteristic of the Yerkes and Dodson law, might be replaced by two monotonic relationships, whose joint operation should result in the well known inverted U shape. The independent variable of their model was the actual level of arousal, whereas the two dependent variables were working memory capacity and "speed of information transfer," understood as a velocity with which the systems processes information. Humphreys and Revelle (1984) review various studies of the influence of stressors and drugs on performance, which they see as supporting the hypothesized relationships between processing resources and arousal. They are, however, aware of the fact that the theoretically postulated relationships are far from being proven. In order to make our model more relevant to intelligence, the "speed of information transfer" has been replaced by attentional resources, i.e., the momentary, actual level of this parameter. Other basic assumptions of the Humphreys and Revelle model were not modified. The assumptions depicted in Figure 1 are questionable but testable, as it will be shown later. The dependence of attentional resources on arousal is probably less controversial, since it may be understood as a consequence of

Chapter 12

514

0 0 0 0

0

E 0 E

0 c 0 e-

=0

Task A [diff.}

0

0 0

0),

Task B (easy)

(. I !

c

I I I I I I I I I I I I I

E O

E

I ! ! I I I

l I I l I I

I,

!,

! ! ! I I I I I I I I I I I I I

! ! ! I I I I I I I I I I I I I

min

0

t._

e-

l !

E E

I I I I t I I I I I I I I

0

max

level of arousal

Figure 1. Momentary values of attentional resources and working memory capacity as a function of arousal. The inverted U shaped curve (bold) depicts the relationship between arousal and performance on cognitive tasks.

increasing monopolization of the system's resources by the current activity when arousal rises. Normally, our resources are divided into many simultaneous actions; even if there is one dominant action, e.g., an experimental task, the resources are "drained" by concurrent actions, accidental thoughts, worries, etc. However, when arousal increases the system may invest everything it has to its disposal to one dominant action, because the increase of arousal may indicate that the situation is more serious, demanding or dangerous. There are, therefore, good reasons to expect the increase of the amount of attcntional resourcos allocated to the task at hand when the situations get tougher.

E. Necka

515

Matthews, Davies and Lees (1990) attempted to verify the Humphreys and Revelle model. They found that high self-report arousal was associated with efficient performance of attentional tasks, particularly in their difficult versions. The authors wished to contrast the Humphreys and Revelle model with the traditional Yerkes and Dodson model, but they failed to find any support for the Yerkes and Dodson law, since the relationships between arousal and performance appeared clearly linear. The predictions based on the Humphreys and Revelle model, concerning the processes of attention (or "sustained information transfer") were also confirmed in various other studies (e.g., Anderson, ReveUe & Lynch, 1989; Matthews, Davies & Holley, 1990; Matthews & Margetts, 1991; Revelle & Loftus, 1990). These findings support the hypothesis that arousal increases availability of attentional resources (Matthews, this volume). As more resources become available, they can be allocated to a task at hand, thus improving its performance. This line of reasoning allows us to understand the mechanism for the postulated relationship between arousal and the momentary value of attentional resources. This momentary value is just the proportion of the absolute amount of resources that is available for the currently performed task. The arousal/working memory relationship seems more contentious, since its psychological mechanism is less obvious, but this hypothetical relationship has obtained some empirical support. For instance, in the Anderson and Revelle (1983) study, subjects performed a short-term memory scanning task under two conditions: after caffeine intake and when treated with placebo. Caffeine appeared detrimental for the STM task, but only when the set size, i.e., the number of items to memorize, equalled six. No effects of caffeine were observed in the less demanding two-item condition. Studies of the influence of time of the day and sleep deprivation, reviewed by Humphreys, Lynch, Revelle and Hall (1983), also suggest that increased arousal exerts detrimental influence on performance of short term memory tasks. Motivation boosted by large incentives is also reported to hurt short-term memory performance (M. W. Eysenck, 1980; cited in Humphreys & Revelle, 1984). Empirical evidence for the detrimental effects of arousal on STM capacity has been obtained in other studies, too (e.g., Anderson & Revelle, 1994; Anderson et al., 1989; Dwivedi, 1990; Eysenck & Calvo, 1992; Revelle & Loftus, 1990). Evidence concerning the relationships between arousal and short term memory is not consistent, though (of. Domic, 1990). However, the important theoretical question is why arousal acts as it does; i.e., what is the nature of the psychological mechanism that makes short-term memory vulnerable to detrimental effects of arousal. According to

516

Chapter 12

our hypothesis, it is probably availability of information processed by the working memory system that decreases with arousal. The absolute capacity of working memory probably remains unimpaired by arousal, but only a proportion of its absolute value may be available for the currently processed task. It is our hypothesis that this proportion gets smaller and smaller as arousal gets higher and higher. So, heightened levels of arousal make attentional resources more available, whereas the information kept in short term store gets less available in such circumstance. Let us discuss the availability hypothesis more thoroughly. The working memory system may be able to keep a fixed number of units of information for a short time regardless of the level of arousal. As we know, this number is severely limited (Miller, 1956), but there is no reason to hypothesize that it depends on transient factors, like arousal. However, the system cannot make use of all of the information stored in STM, i.e., not every portion of information is equally available for processing. And the more the systems becomes aroused the fewer the number of portions it is able to take into account for processing. Short term memory may thus be imagined as consisting of a fixed number of slots, each holding one item or chunk of information. The number of slots is a characteristic of the individual person and defines his/her absolute capacity of working memory. However, some slots may not be activated at any moment; therefore, the pieces of information held in such slots are not available for processing. The number of slots that are open, i.e., activated enough to be used in current processing, defines the actual capacity of one's short term store. The question now arises, of what may be the psychological mechanism due to which the number of open slots depends on the current level of arousal. To answer this question, we have to elaborate our hypothesis even more. Baddeley (1986; also: Baddeley & Hitch, 1974) suggests the existence of trade-off between two basic functions of working memory: current processing of information, performed by the central executive, and storage of data for a short time, performed by the slave subsystems of articulatory loop or visualspatial scratch pad. If the system concentrates on current processing, it may neglect micro-rehearsals or other procedures for keeping the contents of the articulatory loop alive; in effect, the information kept in the loop or scratch pad inevitably decays with time. If, on the other hand, the system concentrates on storage in order to prevent decay, it may lose efficiency of current processing. The existence of the "storage-processing trade-off" may result from the necessity, on the part of the central executive, to control the activity of its "slave subsystems," or at least to communicate with them, so that

E. Necka

517

information might be placed into the loop or scratch pad, or retrieved from them for further processing. Developing Baddeley's ideas, we propose the hypothesis that - as level of arousal increases - the working memory system becomes more and more concerned about current processing, at the expense of storage mechanisms. Therefore, the units of information kept in the articulatory loop or visual-spatial scratch pad are less and less available for the central executive. Put another way, arousal probably makes the system intolerant to the lack of outcomes of the current cognitive activity (Necka, 1989). The system is urged to produce an outcome, so it invests all its attentional resources into processing the present task. It also decides to give the priority to central executive's activity; therefore, it loses control over the processes of rehearsal that determine the efficacy of storage mechanisms. So, the very same factor that causes the increase of the amount of attentional resources allocated to the task, i.e., increased level of arousal, also makes the articulatory loop or scratch pad less able to contribute to the final result of the cognitive activity. As we can see, due to the division of working memory system into the central executive and its slave subsystems, and due to the recognition of the storageprocessing trade-off, we are able to suggest the theoretical explanation of the detrimental effects of arousal on the momentary level of working memory capacity. One should bear in mind, however, that this explanation is highly speculative because it lacks empirical evidence. Apart from being speculative, the proposed explanation raises the question of whether the prioritization of processing over storage in high arousal states is a matter of voluntary choice or of entirely automatic, lowerorder processes. A strategic hypothesis is supported by evidence that sometimes noise may affect the use of short term store. Domic (1990) reviews research of the effects of arousal, caused by white noise, physical effort etc., on performance in cognitive tasks, mainly the tasks involving short term memory processes. Since performance on such tasks is sometimes better in the states of high arousal (cf. Baker & Holding, 1993), Domic hypothesizes that processing strategies may play an important role in such situations. For instance, people may use a strategy of subvocal rehearsal in the noise condition in order to compensate for the effects of stress and thus to facilitate their memory. According to our hypothesis, if people behave in such a way they do so at the expense of their processing capability, because processing and storage seem to remain in a kind of structural conflict. So, one can compensate for the detrimental effects of stress through concentration on either processing or storage, but not both. If a task is purely mnemonic,

518

Chapter 12

concentration on storage may be a better strategy to choose. However, in the case of a more complex intellectual task, when memory processes just serve the final output and do not constitute the main task to do, concentration on current processing seems to be a much better choice. It seems quite possible that the priority of processing over storage - or the reverse in states of low arousal - is a matter of strategic choice rather than automatic switch. However, strategic choice of this kind is not synonymous with the voluntary decision of a person. The notion of cognitive strategy does not necessarily require conscious decision. On the contrary, cognitive strategies may be chosen and used without any knowledge on the part of the person who does so. Strategy research usually requires elaboration of objective methods of strategy identification and excludes simple methodological solutions, like just asking subjects which strategy they prefer (Baron, 1978; Hany, 1991; Hunt, 1980; Kossowska & Necka, 1994). Maybe we are not able to verbalize our strategic choices, or maybe our knowledge of cognitive strategies that we use is only fragmentary. Anyway, the hypothesized processes taking place in human working memory are rather nonconscious, even though pertaining to some kind of strategic choice. On the other hand, it seems possible, too, that the switch from storage to current processing may be automatic in nature. It may be so because - at least in real-life conditions - increase of arousal means that the situation is getting tougher and the processing demands are becoming more and more serious. In such situations the system gets less tolerant of the lack of the outcome of the current cognitive activity. It is therefore quite "natural" for the system to privilege current processing over storage. In effect, the working memory system (precisely: the central executive) is likely to fail altogether, because it cannot process information efficiently without frequent and communication with its slave subsystems. In other words, when arousal increases, more and more slots in the articulatory loop or scratch pad become closed, and the pieces of information held there become unavailable, so that the central executive can concentrate on current processing. But the current processing may fail due to the fact that some chunks of information are not available. So, high arousal reduces just the general momentary capacity of working memory, but if arousal is too high it also reduces overall capability to perform the task. Another explanation of the relationship between arousal and working memory has been outlined in the original paper by Humphreys and Revelle (1984), and further elaborated by Eysenck and Calve (1992). The authors suggest that arousal, particularly the tense arousal connected with anxiety,

E. Necka

519

causes worry, and worry impairs performance of tasks that are highly short term memory dependent. Worrisome thoughts probably "drain" some amount of processing resources, thus impairing one's cognitive functioning. Effects of this kind have been found with subjects suffering from both clinical and nonclinical types of depression and anxiety (see: Eysenck, 1992; Wells & Matthews, 1994; Williams, WaRs, MacLeod & Mathews, 1988). However, the effect of "resource drain" should apply both to attention and memory. In fact, the mechanism postulated by Eysenck and Calvo refers to "processing efficiency" in general. The authors do not differentiate between attentional and mnemonic aspects of processing capacity, and the empirical evidence provided by Eysenck (1992), Wells and Matthews (1994), and Williams et al. (1988) pertains primarily to anxiety induced attentional bias. Besides, it is not clear to what extent this mechanism could operate in the case of people who do not suffer from emotional disturbances. For these reasons, the postulated mechanism of the priority of processing over storage in the states of high arousal seems the one more applicable to our model of intelligence.

"The Process of Intelligence" Taking into account the assumptions stated above, we can now formulate the hypothesis that "the process of intelligence" is simply a process of oscillation within the levels of arousal that are acceptable for a particular person solving a concrete task in certain circumstances. Let us consider in brief the three factors taking part in "the process of intelligence": the person, the task, and the situation. Each person is characterized by his/her own absolute level of attentional resources and working memory capacity. People with a greater amount of resources lose relatively less than people with a smaller amount of resources if their level of arousal drops or climbs too much, so they are less dependent on transient states of arousal (Figure 2). For instance, the amount of attentional resources they can invest into a task is rather modest in the states of lowered arousal, but because they possess plenty of resources altogether they have still more to invest at the given level of arousal, as compared with people characterized by fewer resources. An analogical argument might be formulated concerning working memory capacity. As we can see, a hypothetical person X has more to invest into a task than a hypothetical person Y regardless of the arousal level, providing that both persons are equally aroused, and despite the fact that both persons are dependent on arousal in the similar manner. For the sake of clarity, persons with a greater

Chapter 12

520

0

g 0 0 >. 0

E 4) E

O cO c-

person X

=*

O)

.c_

Task A [d~ff.)

O

0 O O

person Y 4) :3 0

>o

Task B (easy]

>. c 4)

E 0 E

'i i i i i i i i ! i i i i i i

I I I I I I I I I I I I I I I

Ii II II II II II II II Ii II Ii II II II II

i

i

Ii

min

c O

E O E

max level of arousal

Figure 2. The relationships between momentary values of attcntional resources, working memory capacity, and arousal in the case of two hypothetical pcrsons, X and Y. These persons differ in the absolute measures of attcntional resources and working memory capacity.

amount of attcntional resources but lesser capacity of working memory, or vice versa, arc not taken into account here. According to the model, tasks also differ, namely, in the amount of attcntional resources and working memory capacity they require for correct performance. For instance, a hypothetical task A requires more resources than task B; therefore, the first one may bc regarded more difficult. For the sake of clarity, the intermediate instances wcrc omitted, i.e., the task that is challenging for attention but less demanding for memory, or vice versa. Anyway, the theoretical model suggests two consequences of the differences

E. Necka

521

in task's requirements. First, task A may require more restricted boundaries of arousal that are acceptable for successful performance (Figure 1). Second, task A may require higher absolute levels of attentional and working memory parameters in order to be successfully performed (Figure 2). These are two ways in which a person can tackle task A, and the choice may be determined by the structural prerequisites of this person, i.e., by the absolute levels of his/her attention and memory functioning, or by his/her cognitive styles or strategies. Situation is the third party of the game. In this way we refer to many different factors that influence transitory states of arousal. There are probably constitutional sources of high or low arousal; however, the majority of factors influencing our being "high" or "low" refer to time of the day, intensity of external stimulation or substance intake. Even the motivational arousal, caused by increased effort to do one's best, may be treated as rooted in situational factors, because it is transitory in nature and results mainly from external pressure or requirements, in addition to intrinsic motivation. Thus, we can consider the situation in which a person tackles the tasks - and specifically, the level of arousal caused by situational factors - as the third important determinant of the level of performance. Let us consider two particular problems involved in the model of intelligence sketched above. The first problem refers to the precise meaning of the inverted U-shaped relationships, symbolized in Figure 1 and Figure 2 by the bold lines. It has been already suggested (see the caption of the Figure 1) that these lines represent the level of performance of the cognitive task. However, at least two possibilities seem to exist as to the actual dependence of performance on arousal. For both tasks depicted in Figure 2, the horizontal lines define the quantity of resources required. According to classical resource theory (Norman & Bobrow, 1975), further increases in the availability of resources should have no further effect on performance, since the task becomes data-limited. So, between the two vertical lines for, say, Task A, variation in arousal should have no effect on the performance of person X, so this specific fragment of the bold line should be entirely flat rather than Gaussian-like. But if the measures of task performance are continuous in nature (like reaction time, for instance), performance will still change with the arousaldependent fluctuations of resource availability, defined by the vertical lines. In this case, the horizontal lines define the minimum o f resources required, so additional supply of resources seems likely to improve performance to some extent. There are, of course, natural bamers to performance (e.g.,

522

Chapter 12

anatomical, physiological) but, until they are met, performance should still depend on arousal even above the mim'mal level of supply. This issue cannot be solved properly without modifications of the general theory of resource availability. However, if our line of reasoning makes sense, it applies only to person X (i.e., the more "endowed" one) doing the task A (i.e., the more demanding one). Person Y must be regarded as structurally unable to solve the Task A, because he/she does not reach the minimal level of resource supply even in the most favorable circumstances. The second problem amounts to the question of automaticity of processes involved in intelligent behavior. Some measures of intelligence, including the typical IQ tests, definitely engage "higher-level" aspects of cognition, e.g., thinking, concentration and controlled rather than automatic processing. Other measures of intelligence, on the other hand, may not require higher-order cognitive processes. For instance, vocabulary tests require "only" the retrieval of information from semantic memory; retrieval, however complex, is mostly an automatic process. To what extent, one could ask, does the proposed model apply to the more "automatic" aspects of intelligence? Is it possible to apply its basic notions and assumptions to tasks that do not rely on the controlled aspects of cognition? Searching for tenable answers to such questions, one should bear in mind that certain valid measures of intelligence do not necessarily require complex thinking and problem solving. For instance, reaction time (Jensen, 1987b) or nerve conduction velocity (Reed & Jensen, 1991, 1992) proved to correlate with the general mental ability, while being entirely nonintellectual in themselves. It would be risky to claim that intelligence amounts to reaction time or nerve conduction velocity. Such simple measures may correlate with general mental ability for reasons which are more or less accidental. Vocabulary span is a somewhat special ease, because it is probably a result of past intellectual processes (Steinberg & Powell, 1983; Stemberg, 1985, 1990). According to Stemberg's hypothesis, high IQ people have larger vocabulary span - not because retrieval of the word meaning form semantic memory is a difficult, complex, and "controlled" mental process, but because the acquisition of the word meaning is a very difficult and complex task. We learn vocabulary mainly from context, i.e., we infer about the word meaning on the basis of its repeated usage in various real-life situations. The more "intelligent" we are, the more efficiently and quickly we acquire our vocabulary; in effect, the vocabulary span correlates with IQ, although it is far from being synonymous with "intelligence." So, our model seems applicable to the "nonintellectual measures of intelligence" too, if we look at

E. Necka

523

such measures as the results of past intellectual processes. To be precise, the model applies to these past processes rather than to the current use of its results. Let us now concentrate on one particular implication of the model depicted in Figure 2. It seems that, at moderate levels of arousal, people who are endowed with high absolute values of attentional resources and working memory capacity should be less sensitive to emotional state and to arousal from the environment. We could simply call such people "intelligent," if we agreed to reduce the meaning of the term to the structural limitations of one's processing apparatus. I am trying to argue, though, that intelligence is a process, in which at least three elements take part: the person with his/her limitations, the task with its processing demands, and the situation with many factors influencing the current level of arousal. Anyway, "intelligent" people should be regarded as less arousal-dependent - but only in moderate states of excitation. At the extremes of arousal, "intelligence" becomes less important, because performance primarily reflects arousal level. Thus, "intelligence" understood in terms of the processing limitations appears useless when arousal becomes extremely high or low; however, "intelligence" decides which levels of the state of arousal are still "moderate," i.e., acceptable from the point of view of task requirements. To sum up, "the process of intelligence" can just be defined as a process of oscillation between the limits of arousal that are acceptable for task requirements under the current circumstances. It is also the process of choosing tasks that can be performed successfully, taking into account both a person's structural limitations and the situational factors that determine the current state of arousal. In other words, "the process of intelligence" amounts to manipulation of either arousal level or the task's difficulty - depending on what may be controlled under the circumstances - because one cannot change the structural limitations on attention and working memory capacity. Thus, the model predicts that a person X, endowed with a great amount of resources, may act below the level that is structurally available to him/her. It also predicts that a person Y, with lesser amount of resources, may surpass person X in some circumstances. All depends on one's competence in controlling the arousal level, as well as on one's ability to choose the proper task at a proper moment. Efficient manipulation of the current level of arousal, as well as the choice of a task whose complexity is adequate in the given circumstances, probably engages strategic and metacognitive factors. For instance, people can deliberately regulate their mood, and, consequently, their level of arousal

524

Chapter 12

(Thayer et al., 1994; Thayer, Peters, Takahashi & Birkhead-Flight, 1993). Although there is no evidence of this kind, we can speculate that the mood regulation strategies have to be adjusted to the level of constitutional arousal, because the effectiveness of various modes of regulation is likely to depend on the arousal-related personality traits (e.g., extraversion, neuroticism). In other words, people can regulate their mood and arousal to some extent, providing that they have learned the regulation techniques that work in their case. The process of regulation therefore demands a high level of self-knowledge, as well as the ability to control one's cognitive processes. Misehel (1984) suggests that impulse control and the ability to delay gratification may be critical in child social and emotional development. It seems that the analogical mechanism, when applied in the cognitive domain, might be responsible for intellectual development. Skills and areas of expertise probably play a role in these processes, too. It seems to be so because expertise usually allows us to reduce the complexity of the task, either through its redefinition or through taking a new perspective in looking at it (cf. Hany, 1991; Kossowska, Matth~ius & Necka, 1996; Ohlsson, 1984a, 1984b). When we gain experience in a domain, we start to perceive regularities and similarities that are invisible for novices. Herbert Simon calls this the process of "familiarization" and thinks it may be responsible for the act of insight (after Langley & Jones, 1988). Due to the reduction of complexity, tasks that were formerly too complex to tackle suddenly become less demanding and allow effective solutions. And, according to our model, when the task becomes less demanding it becomes automatically less dependent on arousal, that is, it may be effectively tackled in states of activation that are far from optimal. So, "the process of intelligence" depends on whether a person understands his/her constitutional arousal, whether he/she knows how to regulate transient states of arousal, and whether he/she is able to reduce the complexity of the task due to redefinition, "familiarization," or acquisition of expertise. Mechanisms of this sort are probably rooted in the processes of metacognition and strategic choice, which are not covered by the proposed model. It means that the model does not cover the whole area of intelligence research. In fact, it pertains only to the formal level of analysis. Necka (199 l) suggested that, since the concept of intelligence is extremely heterogeneous, it has to be analyzed at four distinct levels: biopsychologieal, formal, strategic, and value-related. The proposed model refers only to the second level, at which the basic formal characteristics of the cognitive apparatus are taken

E. Necka

525

into account. In consequence, it is a model of the formal aspects of intelligence rather than a complete model of intelligence. Although the model does not specify some important aspects of intelligence, it probably accounts for the human intellectual functioning at least to some extent. Three important facets of this model seem worth pointing out. First, it tries to join the processing aspects of intelligence with the structural ones. In other words, it suggests that intelligence is a matter of both transitory processing factors and stable structural traits. Second, the model tries to seek for the intelligence-personality and intelligence-emotion interface. In this way, it views intelligent behavior as resulting from the joint operation of cognition, personality, and emotions. Such a stance is not quite frequent among the differential psychologists, who usually regard intelligence to be separated from emotions, temperament and personality; at the most, they try to investigate the mutual relationships (Saklofske & Zeidner, 1995). Third, the model is able to account for the fact that apparently intelligent people may behave under their natural level of performance, and vice versa. Now, it is time to present some empirical data gathered as a preliminary attempt to verify the model.

Preliminary Empirical Data A series of experiments was carried out in order to check the model of intelligence and its predictions. We chose one single study for this chapter to illustrate the methodological approach that was applied in these experiments. Method Subjects. Eighty one college candidates, aged 19-22, took part in the experiment as volunteers. Materials. We used two computerized procedures and four paper and pencil instruments. First, we employed the modified Saul Steinberg (1969) STM scanning task. Subjects were presented with series of four, six, or eight digits, which appeared in the center of the computer screen one by one. The presentation of the first digit was preceded by a mask, which also followed the last digit of the series. After two seconds, a probe letter appeared on the screen, and subjects were supposed to say YES if they thought it belonged to the series, or NO, if they though it did not. In order to prevent subjects from utilizing specific strategies of encoding (e.g., "chunking"), the presentation of digits was as short as 250 ms (therefore, the task was called Steinberg

526

Chapter 12

"turbo"). Sternbcrg's task has been already applied in various experiments on STM and intelligence (e.g., Necka, 1992; Vernon, 1983, 1985). However, it appeared that the increased speed of presentation made the task rather difficult. The computer registered accuracy and reaction time of subjects' responses. Second, the DIVA ("DIVided Attention") task was employed in order to assess subjects' attcntional parameters (Nccka, 1996). The task consisted of presentation of an uppercase target letter in the center of the computer screen, together with three, four, or five other letters which appeared and vanished in locations around the central target at the pace of one letter per second. Subjects were supposed to press the left-hand mouse key whenever they could see a letter that was semantically identical to the target letter but differed in case (e.g., lowercase "c" when uppercase "E" servexl as target). All other letters were to be ignored as noise. Letters identical with the target both in meaning and in case (e.g., "E" versus "E") were not utilized in this version of the DIVA task. Usually, their appearance serves to introduce distraction conditions, which normally increases the difficulty level of the task (Nccka, 1996). In this study, the presence or absence of distractors of this kind was not manipulated as an independent variable, since we wished to simplify the experimental design. Instead, every single trial was repeated twice, so as to guarantee the necessary number of trials over the whole task. Apart from the primary detection task, subjects simultaneously had to perform a simple psychomotor task, defined as a secondary one. This task demanded the control of the position of two bars located left and fight of the central panel containing the letters. Unexpectedly, one of these bars would start to drop down, and subjects had to prevent from further descent by pressing the right-hand button of the mouse. If they pressed the button too frequently, though, the bar would ascend above the central point, which was also prohibited. In other words, subjects were investigated in a typical dual task paradigm. The comparison of their performance in single versus dual task conditions allowed an assessment of how much the individual subject "suffered" from the necessity to control two simultaneous tasks. In this way, we were able to assess how much attcntional resources the person possesses. The computer registered reaction time of accurate responses; it also counted the number of hits, misses, and false alarms as accuracy measures of performance. Two measures of intelligence were administered: Raven's Advanced Progressive Matrices (Raven, Court & Raven, 1983) and the verbal Analogy Test constructed in our laboratory. The second tool has not been subjected to

E. Necka

527

standard psychometric investigation yet, but it has proveA its validity as an experimental measure of the general mental capacity (Nccka, Gruszka & Orzechowski, 1996). We decided to include it into this study as a test contrasting with the nonverbal spatial material on which the progressive matrices are based. Arousal was assessed with Thayer's Activation-Deactivation Adjective Check List (ADACL, Thayer, 1989), in the Polish adaptation prepared by Klonowicz (1984). As we have already mentioned, Thayer distinguished four dimensions of arousal: 1. energetic (which he calls "high activation"), 2. tense (called "high activation"), 3. vigilant (antagonistic to the state of drowsiness, which he calls "deactivation"), and 4. relaxed (called "general deactivation"). The Activation-Deactivation Adjective Check List consists of twenty items (adjectives), five for each dimension of arousal. It allows quick assessment of these dimensions, understood as momentary states rather than stable traits. In order to assess arousal which is believed to be rooted in stable, constitutional, physiologically determined traits, we decided to apply the Eysenck Personality Questionnaire-Revisexl (EPQ-R; Eysenck & Eysenck, 1975), in the Polish adaptation made by Brzozowski and Drwal (1995). The EPQ-R consists of three scales, two of which are relevant to arousal. According to Eysenck's theory, introversion is rooted in permanently increased cortical arousal, while neuroticism relates to the ease of instigation of visceral activation. Thus, the scores obtained on the E and N scales of EPQ-R were expected to provide some additional reformation about our subjects' level of arousal. Procedure. The experiment was conducted in the following order: ADACL, attention task (DIVA), ADACL, short term memory task (Steinberg), ADACL, intelligence tests (Raven's matrices and the Analogy Test), EPQ-Q, ADACL. In this way, we obtained four consecutive measures of arousal referred to subsequently as ADACL1, ADACL2, ADACL3 and ADACL4, respectively, so that performance in every cognitive task could be related to the ADACL results obtained just before or aider this task. Results

The results for the attention test (DIVA) showed that the manipulation with independent variables was very effective. Response latcncies were longer in the dual task condition than in the single task condition (p<. 0001), and they also increased with set size, i.e., the number of letters that subjects were supposed to process (p<.05). The number of correct responses (hits), as well

Chapter 12

528

as the number of false alarms, also depended on the single versus dual task condition (p<.0001 and p<.001, respectively). Thus, the manipulation of independent variables appeared very efficient, and DIVA may be regarded a good measure of attention (r Necka, 1994, 1996; Szymura & Necka, 1996).

3.6

Number o{ h i t s

3.5

3.4

3.3

3.2

3.1 ...... Low

IO

..... H i g h ]0

3.0

........

. . . . . .

4 '

;

.....

Set 5 i z e

Figure 3. Number of correct responses (hits) in the DIVA task obtained by subjects differing in IQ test scores.

As to the relationships between attention and intelligence, we found unexpectedly - that high IQ subjects (so designated due to a median-split division) did not differ from the low ability subgroup concerning latcnr Normally, intelligent subjects respond faster, particularly in the dual task condition (Necka, 1996). In this study, the advantage of the intelligent group referred only to the accuracy measures. For instance, in the case of less intelligent subjects the number of hits declined linearly with the set size, whereas in the ease of the more intelligent subgroup this deterioration was observable only when the set size increased from four to five letters (Figure 3). As we can see, the two subgroups did not differ in the relatively easy three-letter condition, but they started to differ in the more demanding fourand five-letter conditions (p<.05). It also appeared that intelligent subjects

529

E Necka

committed fewer number of false alarms as compared to the less intelligent subgroup (means: 0.65 versus 0.91, p<0.02). We could therefore conclude that the assumption about the superiority of intelligent subjects in the dual task paradigm was confirmed. However, we also found that intelligent subjects were less effieiem in controlling the secondary task (p<0.02). It seems as if they had to "pay" for better performance of the primary task with poorer control of the secondary task. Thus, the conclusion about the increased efficiency of attentional processes of the intelligent persons must be treated with caution, even though experimental results obtained in other studies favor such a conclusion (cf. Hunt, 1980; Hunt & Lansman, 1982: Lansman & Hunt, 1982; Nccka, 1996; Stankov, 1983).

I0.0

Number of e r r o r s

7.5

5.0

2.5

" ---

0,0 4

6 Set

IQ h i g h IQ low

8

size

Figure 4. Number of errors committed in the STM task by subjects differing in IQ test scores.

As for the STM task, the Steinberg proczxturc appeared very difficult in its "turbo" modification. Response latencies did not depend on set size but were affected by the "yes or no" factor: positive responses required less time than negative ones (p<.05). Accuracy of response, on the other hand, strongly

530

Chapter 12

d e e d e d on set size (p<.0001) and also on the "yes or no" condition, in that positive responses were much less accurate than negative ones (p<0.0001). We can therefore argue that the experimental manipulation was very effe~ive, so the procedure may be used as a valuable test of STM capacity. We found that the high IQ subgroup of subjects, obtained after the mexiiansplit division, responded more accurately in this task than the less intelligent subgroup (p<.001). This tendency was even stronger when the set size increased to 8 digits (the IQ by set size intera~ion was significant at p<.02). These relationships (see Figure 4) justify the assumption that intelligence is related to the increased capacity of working memory to store information for a short time. Let us procx~ to the assumptions concerning the influence of arousal on attcnfional resources and STM capacity. DIVA was preceded by the first application of Thayer's Activation-Deactivation Adjective Check List and it was followed by the second application (see the "Procedure ~176section), therefore, it seems reasonable to check how the DIVA scores related to the ADACL1 and ADACL2 results first of all. The STM task was preceded by the second application of the Thayer list and followed by the third application; so, the results obtained in ADACL2 and ADACL3 should be of primary importance for this task. However, arousal measured not immediately before or after the cognitive task may also be relevant to our considerations. Although arousal is defined as a transient state rather than a stable trait, its value may change at a relatively slow pace during the experimental session. For instance, the effects of init~l arousal may endure until tho end of the experiment, so the final score should also bring about interesting information, particularly from the differential approach point of view. If the between-subject range of the final scores is greater than preceding scores, individual differences in arousal may be more likely to appear, and more likely to correlate with other variables. For these reasons, we will also consider final arousal scores. We found that subjects who scored high on the energetic arousal ("general activation") subscale of ADACL1 responded with increased accuracy, particularly in the repeated trials (p<.01, Figure 5). Due to the the deletion of the distraction condition, the experimental design was simplified but the overall number of trials remained unchanged, due to the immediate repetition of every trial. The effect shown in Figure 5 suggests that the arousexl subjects responded with increased accuracy just in the repeated trials. In other words, generally activated subjects appeared more susceptible to the learning opportunity provided by the repeate~ measure design.

E. Necka

3.5

531

Number o f h i ts

3,4 3.3

3.2 '~More I aroused 18~ Less aroused

3,1

3,0

First trials

Repeated trials

Condi t ion

Figure 5. Number of correct responses (hits) in the DIVA task, as a function of energetic arousal ("general activation" assessed before the task) on first and repeated trials.

Unexpectedly, reaction time did not depend on arousal, measured either before or after the attention task. However, subjects who scored high on the general activation subscale of ADACL4 (i.e., in the fourth and final measurement) responded faster in the DIVA task than subjects who were less energetically aroused at the end of the experimental session (693 ms versus 717 ms, p<. 03). Why only the final score of energetic arousal mattered seems not an easy question to answer. Maybe the reason pertains to the distribution of scores in the ADACL4. It appeared that the mean score in ADACL4 was significantly lower, but the value of the standard deviation was much higher, as compared to ADACL1, ADACL2 and ADACL3. Such a distribution of scores could facilitate the appearance of individual differences. It is very important to stress that the increased response speed of more aroused subjects (ADACL4) was accompanied by indices of better secondary task performance (;,<.05). In other words, aroused subjects did not have to "pay" for better performance of the primary task with inefficient control of the secondary task. On the contrary, they showed their superiority over the less

532

Chapter 12

aroused subjects on both tasks, which is a result strongly supporting our hypothesis. It also appeared that subjects who scored high on the vigilant arousal (reverse of"deaetivation") subseale of ADACL1 committed less false alarms in the single task condition but made more mistakes of this kind in the dual task condition, as compared to their less vigilant peers (p<.05, Figure 6). The first finding is definitely congruent with our assumptions, whereas the second one suggests a somewhat more complicated relationship: it seems as if wakefulness helped to avoid false alarms when the attention task was easy but it acted in the reverse direction when the task became more challenging. However, "deactivation" measured just afar the task (ADACL2) showed a different pattern of relationship with the number of false alarms. More aroused subjects - compared with less aroused ones - committed greater number of false alarms when the set size of letters to detect equalled 3 or 4, but significantly fewer - when the set size equalled 5 (Figure 7, p<. 01).

Number of false alarms _

1.0

' .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

I i ......

0.8

....... i

"~~~ .................. ..___.~

0.6

0.4

.

f

~i~ More aroused

Less aroused Si ng I e

Ilua I Condi i ion

Figure 6. Number of false alarms depending on vigilant arousal ("deactivation" assessed before the DIVA task) and single versus dual task condition.

E. 1. O0

Number o{

Necka

533

f a l s e alarms

/ /

/

/

/

O. 75

"',,,,,,,,,,/

~ --More aroused

O. 50

......Less aroused "i

'

'

Set size

Figure 7. Number of false alarms depending on vigilant arousal ("deactivation" assessed after the DIVA task) and the set size.

The comparison of the relationships illustrated in Figure 6 and Figure 7 suggests that increased arousal before the task helped subjects to avoid false alarms in the first half of the trials (the single task condition) but it could be detrimental in the second half. However, increased arousal after the task helped to avoid false alarms in the difficult 5-1ettor condition but did not help in the relatively easier 3- or 4-1etmr condition. However, states of high vigilant arousal appeared to correlate with worse control of the secondary task 09<.02). Therefore, the conclusion concerning beneficial influence of vigilant arousal on attentional mechanisms should be treateM rather cautiously. No more relationships between arousal and attention test scores were observed. The data obtained with the EPQ-R appeared encouraging, since introverts responded with increased accuracy, primarily m the dual task condition (p<.05). In this way, we succeeded in replicating the results of the third experiment by Szymura and Neeka (1996). Neurotieism was a significant correlate of reaction time, but its interaction with the single versus dual task variable was non-significant. Neurotic subjects needed more time to discriminate signals from noise, regardless of the experimental condition

534

Chapter 12

(p<.03). We also found that extraverts committed more false alarms (p<.05) and were less efficient in control of the secondary task (p<.05). Ineffective control of the secondary task was also characteristic of emotional stability, as compared to neuroticism. It can therefore be stated that cortical arousal, assessed with the E scale of the EPQ-R, is associated with increased accuracy of performance of the attention task. The relative superiority of introverts over extraverts in the dual task condition clearly suggests that cortical arousal, characteristic of introversion, is associated with an increased amount of attentional resources allocated to the task. Visceral arousal assessed with the N scale of EPQ-R, on the other hand, cannot be interpreted in the similar way, because neuroticism, though connected with better control of the secondary task, correlated with a relatively low speed of detection. Now, we will examine the relationships between Stemberg's STM task and arousal measured just before and after this task. In other words, we are looking for relations between spee~ and accuracy of the task performance and the ADACL2 and ADACL3 arousal measures. As with response lateneies, it appeared that tense arousal (the "high activation" scale) in ADACL2 was associated with longer reaction times, particularly in the NO condition (p<.05, Figure 8). We also found that higher levels of vigilant arousal (the reverse of"deaetivation") made subjects respond more accurately in the YES condition, but less accurately - in the NO condition (p<.05, Figure 9). Nota bene, the YES condition was generally more difficult for all subjects, regardless of other factors (p<.0001); however, subjects also responded with increased speed in the YES condition that made (p<. 0001). The most characteristic finding was obtained with the scale of "general activation" (energetic arousal). It appeared that more aroused subjects demonstrated a greater number of errors in the YES condition but made fewer mistakes in the NO condition. Less aroused subjects showed the reverse pattern of responding. The relationship shown in Figure 10 (p<.03) refers to arousal measured after the task (ADACL3), but it was even stronger (p<.01) for arousal level measured before the task (ADACL2). So, if the task was relatively easy (the NO condition), energetic arousal made subjects less error prone, whereas when the task became more difficult (the YES condition) energetic arousal made subjects commit more mistakes. Similar but weaker relationships were found for ADACL1 and ADACIA measures Regarding constitutional arousal measures, introverts responded with increased velocity, but only in the easiest 4-digit condition (p<.04). This effect is not compatible with the hypothesis but we have to realize that reaction time in the STM task is much less informative than accuracy. No

535

E Necka Reaction

time ( s e c . )

1.3 t

........

1,2

I,I

1.0 ~More aroused --~Less aroused

0,9

No

Yes Cond i t i on

Figure 8. Reaction time in the STM task as a function of the level of tense arousal ("high activation") measured before the task. 9.0,

Number ot e r r o r s

8.0f 7.0 i

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

:::::::::::::::::::::::::::::::::::: i:i:i~i:!:!i:i:!:i:i:i:?:i:::!:i:i: .-...-..,r...-.-................. :::::::::::::::::::::::::::::::::::: ................. .:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:. ..................

6.0

5.0 vz///~More aroused 4.0

I

Yes

Less aroused

No Cond i t i on

Figure 9. Number of errors in the STM task as a function of vigilant arousal (reverse of"doactivation") measured aitor the task.

Chapter 12

536 9,0 Number o~ errors

8.0

.....

?.0

6.0 :.

5.O

.

.

.

@i

iiiiiiiiii!iiiiii .

4.0

3.0

.

m

,

,

I ~ I

~ ~.

m

II

_I~ More a roused

Less aroused

.

No

Yes Condi I i on

Figure 10. Number of errors committed by less aroused and more aroused subjects in the STM task, depending on the YES or NO condition ("general activation" measured after the task).

relationship between accuracy and the E scale was found. The N scale also failed to show any significant effect. Let us now check one more prediction inferred from the theoretical dependencies depicted in Figure 1 and Figure 2. The graphs suggest that, while solving Raven's matrices, verbal analogies or other intelligence tests, there are two ways of achieving high scores. The first way amounts to narrowing the arousal level. Clearly, the difference between high and low scoring subjects is expressed in their ability to tackle the most difficult matrices. There is no difference between the ability subgroups at the beginning of the series of matrices; the differences start to show when the progressively more and more demanding items appear in the series. So, more intelligent subjects are the ones who arc able to deal with matrices symbolized in Figure 1 by Task A (more difficult). Accordingly, less intelligent subjects arc the ones who arc able to deal only with easy matrices, symbolized by Task B. More difficult tasks require, by dcfmition, more attentional resources and more STM capacity. However, these mental resources depend on arousal in antagonistic ways. Therefore, only subjects who arc able to oscillate within

E. Ne c ka

537

a relatively narrow range of arousal arc capable to deal with difficult tasks. In effect, they score more points on progressive matrices. The same line of reasoning may be applied concerning the Analogy Test, because it also consists of items of various difficulty, although they are not arranged progressively. The second possible way of achieving high scores in the intelligence tests amounts to possessing higher absolute levels of attentional resources and working memory capacity, like a hypothetical person X compared to a hypothetical person Y in Figure 2. Both persons' actual level of attentional resources and working memory capacity depend on arousal in the antagonistic way assumed by the model. However, person X has "more to spend" than person Y, so the detrimental results of too low or too high arousal are less salient in his/her case. As we can see (Figure 2), person X is able to tackle more difficult tasks than person Y within the same range of arousal. In other words, thanks to his/her structural characteristics of attention and working memory, person X does not have to oscillate within narrow boundaries of medium-level arousal in order to achieve high scores in intelligence tests. In fact, both hypothetical persons may oscillate within the same range of arousal, and still they can differ in their ability to solve difficult tasks. These are two modes of intelligent functioning postulated by the model, so we can hypothesize that at least some people belonging to the high IQ subgroup obtained their test scores due to the first mechanism. In other words, at least some of them should remain within narrow boundaries of arousal. These boundaries may be operationalized as lessened variance of ADACL results, or lessened range (the difference between maximal and minimal score within the subgroup) of ADACL scores. Therefore, we can predict that high scorers in intelligence test should demonstrate lesser variance and smaller range of their ADACL results, compared with low scorers. Table 1 shows these comparisons concerning Raven's matrices, whereas Table 2 shows parallel differences concerning the Analogy Test. As we can see, the prediction was confirmed concerning tense arousal before the test, and energetic and vigilant arousal - aider the test (Raven). On the Analogy Test, it was confirmed for tense and relaxed arousal before the test, and energetic and tense arousal after the test. There was only one difference inconsistent with the prediction; namely, low scorers in the Analogy Test showed lesser variance in vigilant arousal measured before the test. The above mentioned differences refer to variances, which were compared for homogeneity by Bartlett's test. There is no way to make statistical inferences concerning the group differences in range, which are also

Chapter 12

538

shown in Tables 1 and 2. However, the differences in range were consistent with our predictions in five cases out of eight concerning Raven's matrices, and in seven cases out of eight, concerning the Analogy Test. Moreover, even though the data were not entirely consistent with the model, no single effect appeared contradictory to it. There is another prediction based on the theoretical model. Figure 2 suggests that subjects who oscillate within the same boundaries of arousal may differ in their test scores if they also differ in the absolute amount of attentional resources and working memory capacity. At least some members of the high ability subgroup probably obtained their test scores due to this mechanism. Therefore, the second prediction says that people differing in IQ should also differ in the absolute values of attentional and working memory

Table 1. Variance and range of results obtained for four subscales of Thayer's Activation-Deactivation Adjective Check List (ADACL), by subjects scoring low and high on Raven's Matrices (arousal measured before and aRer the intelligence assessment).

Before test

Low Scorers

High Scorers

Variable

Variance

Range

Variance

Range

Energy Tension Vigilance Relaxation

12.72 13.71" 10.91 7.41

15.00 15.00 13.00 12.00

12.14 10.11 9.62 7.66

13.00 12.00 11.00 12.00

After test

Low Scorers

High Scorers

Variable

Variance

Range

Variance

Range

Energy Tension

14.92" 10.95

15.00 15.00

12.92 9.64

15.00 13.00

9.62' 9.66

13.00 14.00

7.95 8.91

13.00 13.00

Vigilance Relaxation

Note. Energy = "general activation"; Tension = "high activation"; Vigilance = . deactivation w ; Relaxation = "general deactivation. * p<.05

539

E. Ne c ka

Table 2. Variance and range of results obtained for four subseales of Thayer's Activation-Deactivation Adjective Cheek List (ADACL), by subjects scoring low and high on the Analogy Test (arousal measured before and after the intelligence assessment).

Low Scorers

Before test

High Scorers

Variable

Variance

Range

Variance

Range

Energy Tension Vigilance Relaxation

12.16 14.75" 8.56 8.48"

14.00 15.00 12.00 13.00

12.13 9.02 11.40" 6.85

13.00 12.00 11.00 12.00

Low Scorers

After test

High Scorers

Variable

Variance

Range

Variance

Range

Energy Tension Vigilance Relaxation

16.08* 13.56* 9.66 8.35

15.00 15.00 13.00 14.00

12.07 7.13 8.24 10.06

13.00 13.00 13.00 13.00

Note: Energy = "general activation"; Tension = "high activation";

Vigilance = "deactivation";Relaxation = "general deactivation." * p<.05

memory parameters. Unfortunately, this prediction is not testable, since we are not able to assess absolute values of these parameters. Discussion o f the data

The data presented earlier in this chapter support the theoretical model only to some extent. The assumptions concerning the relationships between intelligence and attention, on one hand, and intelligence and working memory, on the other hand, have been confirmed. This conclusion seems particularly convincing in reference to working memory, as Figure 4 deafly shows. The relationship of attention task performance with intelligence is also consistent with theoretical expectations, although the results are less clear. First of all,

540

Chapter 12

we were not able to show that high IQ subjects outperformed their less able peers in terms of response latcncics, particularly in the dual task condition. The lack of the intelligence by single versus dual task condition interaction looks puzzling, both from the theoretical point of view and in the light shed by previous experiments (Nccka, 1996). Anyway, the results shown in Figure 3 clearly suggest that more intelligent subjects performed better than less intcUigent ones in the attention task. The problem arises, though, whether one can interpret this advantage in terms of increased amount of attentional resources possessed by more intelligent persons, especially as the intelligent subgroup appeared less efficient in controlling the secondary task. The assumption about the beneficial influence of arousal on availability of attentional resources must be regarded as confirmed only in part. The relationship shown in Figure 5 is consistent with this expectation, because it represents better performance of energetically aroused people in the repeated trials. We have also found that heightened energetic arousal is associated with shorter reaction time and better control of the secondary task. On the other hand, we have to admit that increased levels of vigilant arousal were associated with a greater number of false alarms in the dual task condition (Figure 6). This is not compatible with the model, despite the fact that vigilant arousal was linked to fewer number of false alarms in the single task condition, because it was performance in the dual task condition that was supposed to reflect subjects' attcntional resources. Although the relationship depicted in Figure 7 is more supportive of the soundness of our hypothesis, we have to admit that vigilant arousal did not "behave" according to the model. Fortunately, the measures of constitutional arousal (introversion and, to some extent, neuroticism) brought about the result which appeared compatible with our hypothesis. Clearly, different kinds of arousal affect attention in some specific ways, which should be taken into account in further attempts to reformulate the model. The results concerning the postulated relationships between arousal and STM task performance arc also indefinite. On one hand, we can state that heightened levels of tense arousal make response latcncics in the Steinberg task longer, particularly in the NO condition, in which latencics were generally longer (Figure 8). This finding is consistent with the model, which assumes detrimental impact of arousal on short term memory capacity. On the other hand, vigilant arousal once again brought about results that are contradictory with the model. Aroused people committed more errors but only in the NO condition, which was generally easier in terms of accuracy measures. In the more demanding YES condition it was less aroused people

E. Necka

541

who responded with greater accuracy (Figure 9). To make the picture even more complicated, we should look again at the graph shown in Figure 10. In this case, the theoretical predictions found some support, because energetic arousal was linked with increased error rate in the difficult YES condition, whereas it was connected with decreased number of errors in the relatively easier NO condition. Once again, we have to admit the influence of arousal on processing capacity depends on the kind of arousal, although we did not take this possibility into account while constructing the theoretical model. The predictions concerning variance and range have been confirmed to some extent, too, as Table 1 and Table 2 illustrate. It is worth emphasizing that the differences in variance and range were more frequent and more convincing when arousal was measured after the test, as compared to the instances of measurements preceding test taking. One can hypothesize that the later parts of taking an intelligence test, particularly progressive matrices, are decisive for the final score. Therefore, the differences between ability subgroups concerning the arousal measured after the intelligence test session are more interesting than the differences concerning arousal measured before the test was presented to subjects. Such differences, although not huge and not always statistically significant, are consistent with the theoretical model. To sum up, the data reported earlier in this chapter appeared supportive of the model in some respects, but at the same time they raised many methodological and theoretical questions. Future replication studies should, firstly, concentrate on the question of the mode in which different kinds of arousal affect attention and memory. The dimension of energetic arousal fits the model quite well, whereas the dimension of vigilant arousal fits less well, which is surprising since both dimensions positively correlate with each other (r=.40 to .70 in our study). The model also accommodates cortical arousal well, as represented by the Eysenekian E-I scale. The dimension of tense arousal (and, to some extent, also neurotieism assessed with the EPQ-R) brought about some encouraging results, particularly as to their influence on working memory, rather than attention. It is intriguing why the dimension of relaxed arousal ("general deactivation") appeared entirely unrelated to both attention and working memory, although it correlates negatively with "high activation" (r=-.50 to -.70 in our study), which showed hamfful effects on memory. Probably the intercorrelations between various aspects of arousal, though relatively strong, do not justify simple substitution of one scale by another, thus manifesting the peculiarity of different kinds of arousal. Moreover, in investigating the transient states of arousal we should not, by definition, expect good indices of the stability of measurement. In

542

Chapter 12

consequence, we should admit the qualitative differences between various kinds of activation, regardless of their being strongly intercorrelate~. It is important to stress the results obtained in our study are consistent with those reported by Matthews and Westerman (1994). They presented their subjects with the attention and working memory tasks. The self-report energetic arousal facilitated performance in the attention task, but only in the dual task condition. Energy also boosted performance in the short term memory task. Tense arousal, on the other hand, tended to block the beneficial influence of energetic arousal. The authors found that the best performance in both tasks resulted from the interactive influence of the energetic and tense arousal. It seems therefore plausible to presume that cognition is affected in two qualitatively different ways by two qualitatively different types of arousal: energy and tension. To conclude, the most important task for further research is the modification of the theoretical model so as to make it compatible with the peculiarities of the various kinds of arousal. Second, the replication experiments should be carried out with samples that are less restricted in range concerning intelligence test scores. College volunteers are not the best subjects from this point of view. Third, direct experimental manipulations with arousal, e.g., adrenaline injections or the exposure to white noise, may bring about clearer results. Maybe after such replications the theoretical model presenteA in this chapter will develop into the crystallized theory of intelligence.

Cognitive Science Perspectives In the final section of this chapter, we will try to discuss the theoretical model of "the process of intelligence" from the cognitive science point of view. Cognitive science is not understood here as a subfield of artificial intelligence but as the multidisciplinary approach to the study of cognition, no matter whether human or artificial (of. Steinberg, 1990). The problems undertaken by cognitive scientists are usually approached through computer modelling of cognitive processes. However, the traditional behavioral methods typical of psychology, are also applicable in this field. Both approaches are apt to complement each other, as Ncwcll and Simon (1972) succ,c~cd in demonstrating. Apart from methodology, the main difference between cognitive science and the psychology of human intelligence seems to refer to the question of individual differences. Traditionally, psychologists viewed intelligence as a

E Necka

543

construct accounting for the observable differences between individuals, whereas cognitive scientists were interested in the question of how intelligence works in general, and how one could imitate human cognition with the use of machines. Apparently, the problem of individual differences was not of great importance for computer scientists because computers do not differ from each other very much, and the differences between them may be efficiently manipulated. If we extend our PC's memory or install new software, for instance, we make it "smarter" - an effect hardly possible with people. Human beings, on the other hand, differ from each other very much. These differences may be decisively important for an individuars educational career or job opportunities, and they cannot be manipulated so easily as the differences referring to machines. It seems that psychologists, who could not control human intelligence, learned to treated it as a latent capacity accounting for the observable individual differences, whereas cognitive scientists, being able to engineer artificial intelligent systems, learned to concentrate on the general architecture of cognition. This contrast became less visible within the cognitive approach to intelligence (Steinberg, 1985, 1990) but it is still alive. It is our intention to argue that the proposed model of intelligence may be relevant to both fields of the study of cognition. The model is processoriented; that is, it views intelligence as a process of tackling cognitive tasks, and in particular the process of oscillation within the acceptable boundaries of arousal. On the other hand, the model is structural as well, since it acknowledges the stable, trait-like determinants of the cognitive processes. In other words, the model assumes the process-based nature of intelligence, but it also admits the existence of the structural foundations of this process. The problem may be reformulated as a question of the psychological locus of individual differences, since the very existence of individual differences is just an observable fact of nature. For traditional psychologists, the sources of intra-individual differences have been located in latent mental capacities. For those cognitive scientists interested in the problem, they have been located in the architecture of the cognitive system or in the "software" (e.g., heuristics, strategies) with which the system deals with its tasks. From the point of view of the formal model of intelligence outlined in this chapter, the loci of individual differences in cognition pertain both to the process and structural factors. If further clarifications of the model are accomplished, a unified structural-processing theory of intelligence is likely to emerge. Another problem of importance from the cognitive science perspective refers to the relationships between arousal and cognition. This problem is

544

Chapter 12

much older than the currently investigated question of "personalityintelligence interface" (e.g., Saklofske & Zeidner, 1995), because the bulk of the studies of cognition have ignored noncognitive factors, like emotions, arousal, motivation, and personality. Cognitive models typically operate with purely cognitive constructs, whereas the traditional arousal theory has never properly explained how activation relates to cognition. Perceived from the perspective of psychophysiology, or the psychology of emotions, cognitive models are sometimes regarded too "dry" (Hockey, Coles & GaiUard, 1986), since they lack energetic factors that normally influence human behavior. The study of the interface between cognitive and energetic aspects of behavior is not an easy task, the more so that the general theory of intelligence needs to work out the model of mechanisms of mutual influence, rather than to accumulate the knowledge of correlations between traits, states or dimensions that belong to different domains of psychological research. Only with the knowledge of mechanisms can psychologists really join cognition with personality, emotions, and motivation (cf. Szymura & Necka, 1996). The proposed model is encouraging from the perspective of the "cognition-noncognition" relationship, too. While constructing the model, we have been trying to identify a basic "noncognitive" factor that, hypothetically, might be able to affect cognitive performance. We have chosen arousal because it looked promising as a construct apt to provide a common platform for different noncognitive factors: emotional, motivational, situational, constitutional, etc. Then, we treated arousal as a factor intrinsically involved in "the process of intelligence," that is, we put it into the heart of this process. In this way, we did not consider arousal just as a factor influencing intelligence, or cognition in general, but as an inherent part of the process itself. Finally, we tried to hypothesize in which way arousal is likely to affect other parts of the process, i.e., attention and working memory. We also tried to speculate in which way arousal may be regulated and affected through metacognition. In consequence, the proposed model of intelligence seems likely to connect arousal with cognition to the extent that is not very frequent in the cognitive science approach. The model seems worth testing through computer simulation, in order to make it even more suitable for the cognitive science approach. Although no such attempts have been undertaken yet, we can speculate about possible ways of computer simulation within the connectionist approach. This approach has its well-known characteristics, e.g., the assumption that cognition takes place in the massive network, whose numerous but primitive nodes are activated to various degrees and act in parallel (hence, parallel

E. Necka

545

distributed processing, or PDP). For our purposes, two additional characteristics of connoctionism should be mentioned. First, in contrast to conventional cognitive psychology, this approach does not distinguish separate cognitive structures, like attention and memory. Rather, the phenomena referred to by psychologists as related to attention or memory are simulated through activation of specific nodes in the network, or specific regions of the massive network (of. Cohen, Dunbar & McClelland, 1990; Cohen & Servan-Schreiber, 1992; Just & Carpenter, 1992; Matthews & Harley, 1993, 1996; Wells & Matthews, 1994). Second, conneetionist models always refer to some specific task, e.g., the Stroop task (Cohen et al., 1990; Matthews & Harley, 1996) or the lexieal decision task (e.g., Matthews & Harley, 1993). The architecture of the network is constructed in such a way that allows the network to learn how to do the specific task. Our own model is closer to conventional cognitive psychology than to eonneetionism. However, if it were to be simulated with the artificial network, the following procedure might be adopted. First, a typical IQ task should be chosen for modelling, for instance, a Raven-like task. The input nodes of the network would represent the task structure and demands, as well as the information from environment concerning arousal-related events, whereas the output nodes would represent possible solutions of the task (e.g., the six or eight answer alternatives, as is the case with Raven's matrices). The core of the network would concern the architecture of the hidden nodes, intervening between the input and output nodes. Let us assume that the network consists of two layers of the hidden nodes (Figure 11). The first one is directly activated from the input and might be broadly related to "attention." The more demanding the task, and the greater number of situational factors that increase the arousal level, the more activation the nodes of the first layer receive. In this way, the level of activation of the nodes within the first layer would conventionally represent "attentional resources" allocated to the task. Of course, the activation of the nodes is not a zero-one phenomenon; it is conventionally assumed to take continuous values between -1 and +1. Therefore, what we referred to as "the level of activation of the nodes within the first layer" should be understood in terms of the sum of activation of all the nodes constituting this layer. The nodes of the second layer, which may broadly relate to "working memory", activate the output nodes. If one of the output nodes gets activated above the threshold value, it executes the final decision, which means that the network has worked out the solution to the task. Let us hypothesize that the nodes of the second layer are positively activated only by the ones located in

546

Chapter 12

§

Figure 11. The general idea of the architecture of neural network simulating "the process of intelligence."

located in the first layer. However, due to the reverberation mechanism they are able to inhibit their own activity, that is, they receive negative activation from themselves (Figure 11). The strength of this negative activation determines whether a node is being active or passive at the moment. The function of such a reverberating self-inhibition is, of course, a matter of further speculations. For instance, the network might be programmed in such a way that excludes hyperactivation of the second layer. If the sum of activation of this layer is defined as constant, the self-inhibitory circuits just counteract excessive activation in this region of the network. In fact, too much activation in the second layer could cause above-threshold activation in too many output nodes, which, in consequence, could make the chaotically behaving network unable to decide on any solution, similarly to hyperactive people. It is plausible that the higher the positive activation from the first layer, the stronger the self-imposed, inhibitory activation observed within the second layer. Such a model may provide a speculative explanation for the antagonism between arousal effects on attention and on working memory. Adequate representation of task demands may require a high degree of activation of

E. Necka

547

nodes in the first layer. However, these nodes may then feed forward excessive activation to the second layer, leading to self-inhibition of activation within this layer, and termination of processing. Thus, there is a degree of contradiction between the overall levels of activation required by the two layers, and the network must arrive at a compromise providing for efficiency in both "attention" and "memory". However, consistent with the philosophy of connectionism, we do not postulate the existence of separate cognitive structures but try to show how the attention- and memory-related phenomena might be simulated by the network. Individual differences might be represented by the number of active nodes in both layers of the network. In other words, some networks might be rather rich in terms of the number of nodes and the number of connections existing between them. Other networks might consist of smaller numbers of nodes and connections. The first would be called more intelligent, because they would show increased efficiency in dealing with the Raven-like task. Situational and constitutional arousal could be represented as a factor which increases the number of active nodes within the first layer. In effect, increased arousal would make the network more focused on the task demands, and therefore more efficient, providing that the increased activation of the first layer would not switch out the extensive number of the nodes of the second layer. As we can see, a network built up according to the rules that we have suggested could behave according to the Yerkes and Dodson rule. In general, such a network could perform "the process of intelligence" to some extent, but its performance cannot be demonstrated without simulation. References

Aekerman, P. L. (1994). Intelligence, attention, and learning: Maximal and typical performance. In D. K. Dettennan (Ed.), Current Topics in Human Intelligence (Vol. 4, pp. 1-27). Norwood: Ablex Publishing Corporation. Anderson, K. J., & Revelle, W. (1983). The interactive effects of caffeine, impulsivity, and task demands on a visual search task. Personality and Individual Differences, 4, 127-134. Anderson, K. J. Revelle, W., & Lynch, M. J. (1989). Caffeine, impulsivity, and memory scanning: A comparison of two explanations for the Yerkes and Dodson Effect. Motivation and Emotion, 13, 1-20. Baddeley, A. (1986). Worlang memory. Oxford: Clarendon Press.

Chapter 12

548

Baddeley, A. (1996). Exploring the central executive. The Quarterly Journal of Experimental Psychology, 49,4, 5-28. Baddeley, A. D., & Hitch, G. T. (1974). Working memory. In G. A. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47-90). New York: Academic Press. Baker, M. A., & Holding, D. H. (1993). The effects of noise and speech on cognitive task performance. Journal of General Psychology, 120, 339355. Baron, J. (1978). Intelligence and general strategies. In G. Underwood (Ed.), Strategies of information processing (pp. 403-450). London: Academic Press. Broadbent, D. E. (1958). Perception and communication. London: Pergamon Press. Brookings, J. B. (1990). A confirmatory factor analytic study of time-sharing performance and cognitive abilities. Intelligence, 14, 43-59. Brzozowski, P., & Drwal, R.L. (1995). Kwestionariusz osobowosci Eysencka: Polska adaptacja EPQ-R. Warszawa: Praeownia Test6w Psyehologicznych Polskiego Towarzystwa Psychologieznego. Cattell, R. B. (1971). Abilities: Their structure, growth, and action. Boston: Houghton Mifflin. Cherry, E. C. (1953). Some experiments on the reamgnition of speech with one and two ears. Journal of the Acoustical Society of America, 25, 957-979. Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990). On the control and automatic processes: A parallel distributed processing account of the Stroop effect. Psychological Review, 97, 332-361. Cohen, J. D., & Servan-Schreiber, D. (1992). Context, cortex, and dopamine: A eormeetionist approach to behavior and biology in schizophrenia. Psychological Review, 99, 45-77. Coreoran, D. W. J. (1965). Personality and the inverted-U relation. British Journal of Psychology, 56, 267-273. Domic, S. (1990). Noise and information processing: Findings, trends, and issues. Reports from the Department of Psychology, University of

Stockholm, No. 715. Duffy, E. (1962). Activation and behavior. New York: J. Wiley. Dwivedi, C. B. (1990). Reward-induced arousal differences in retention. Personality and Individual Differences, 11, 685-691. Eysenek, H. J. (1967). The biological basis of personality. Springfield, IL, Thomas.

E. Necka

549

Eysenck, H. J. (1975). The measurement of emotion: Psychological parameters and methods. In L. Levi (Ed.), Emotions: Their parameters and measurement (pp. 439-467). New York: Raven Press. Eysenek, H. J. (1988). The concept of "inteUigcnr Useful or useless? Intelligence, 12, 1-16. Eysenek, H. J. (1994). A biological theory of mtdligence. In D. K. Dettcrman (Ed.), Current topics in human intelligence (Vol. 4, pp. 117-149). Norwood, NJ: Ablex Publishing Corporation. Eyscnek, H. J., & Eyscnck, S. B. G. (1975). Manual for the Eysenck Personality Questionnaire. San Diego: Educational and Industrial Testing Service. Eysenck, M. W. (1992). Anxiety: The cognitive perspective. Hove: Lawrence Erlbaum Associates, Publishers. Eysenck, M. W., & Calvo, M. G. (1992). Anxiety and performance: The processing efficiency theory. Cognition and Emoaon, 6, 409-434. Gopher, D. (1986). In defence of resources: On structures, energies, pools and the allocation of attention. In G. R. J. Hockey, A. W. K. Gaillard and M. G. H. Coles (Eds.), Energetics and human information processing (pp. 353-371). Dodrecht: Martinus Nijhoff Publishers. Hany, E. A. (1991). Strategien der Komplexitatsverarbeitung bei induktiven Problemen in Abhangigkeit von Intelligenz und Alter. Zeitschrift far Differentielle und Diagnosasche Psychologie, 12, 77-92. Hebb, D. O. (1955). Drives and the C.N.S. (conceptual nervous system). Psychological Review, 62, 243-254. Hirst, W., & Kalmar, D. (1987). Characterizing attentional resources. Journal of Experimental Psychology: General, 116, 68-81. Hockey, G. R. J., Coles, M. G. H., & Gaillard, A. W. K. (1986). Energetieal issues in research on human information processing. In G. R. J. Hockey, A. W. K. Gaillard and M. G. H. Coles (Eds.), Energetics and human information processing (pp. 3-21). Dordrecht: Martinus Nijhoff. Humphreys, M. S., Lynch, M. J., Revelle, W., Hall, J. W. (1983). Individual differences in short-term memory. In R. F. Dillon and R. R. Schmeck (Eds.), Individual differences in cognition (Vol. 1, pp. 35-64). Orlando: Academic Press. Humphreys, M. S., & Revelle, W. (1984). Personality, motivation, and performance: A theory of the relationship between individual differences and information processing. Psychological Review, 91, 153-184. Hunt, E. (1978). Mechanics of verbal ability. Psychological Review, 85, 109130.

550

Chapter 12

Hunt, E. (1980). Intelligence as an information processing concept. The

British Journal of Psychology, 71,449-474. Hunt, E., & Lansman, M. (1982). Individual differences in attention. In R. J. Sternberg (Ed.), Advances in the psychology of human intelligence (Vol. 1). Hillsdalr NJ: Erlbaum. Hunt, E., & Lansman, M. (1986). Unified model of attention and problem solving. Psychological Review, 93, 446-461. Jensen, A. R. (1987a). Intelligence as a fact of nature. Zeitschrifi far Padagogische Psychologie, 1, 157-169. Jcnsen, A. R. (1987b). Individual differences in the Hick paradigm. In P. A. Vernon (Ed.), Speed of information processing and intelligence. Norwood, NJ: Ablex Publ. Corp. Jensen, A. R. (1992). Understanding g in terms of information processing. Educational Psychology Review, 4, 271-308. Just M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122-149. Kahneman, D. (1973). Attention and effort. Englewood Cliffs, NJ: PrenticeHall. Klonowicz, T. (1984). Reaktywnosc a funkcjonowanie czlowieka wr6znych warunkach stymulacyjnych [Reactivity and human performance in the conditions differing in stimulation]. Wroclaw: Ossolineum. Kossowska, M., Matthfius, W., & Neeka, E. (1996). The cost of being competent: Expertise and rigidity in coping with novelty. Polish Psychological Bulletin, 27, 25-38. Kossowska, M., & Neeka, E. (1994). Do it your own way: Cognitive strategies, intelligence, and personality. Personality and Individual Differences, 16, 33-46. Kyllonen, P. C., & Christal, R. E. (1990). Reasoning ability is (little more than) working memory capacity?! Intelligence, 14, 389-433. Lacey, J. I. (1967). Somatic response patterning and stress: Some revisions of activation theory. In M. H. Appley and R. Trumbull (Eds.), Psychological stress (pp. 14-37). New York: Appleton-Century-Crofts. Lansman, M., & Hunt, E. (1982). Individual differences in secondary task performance. Memory and Cognition, 10, 10-24. Langley, P., & Jones, R. (1988). A computational model of scientific insight. In R. J. Steinberg (Ed.), The nature of creativity. Contemporary psychological perspectives (pp. 177-201). Cambridge: Cambridge University Press.

E. Necka

551

Matthcws, G., Davies, D. R., & Lees, J. L. (1990). Arousal, cxtravcrsion, and individual differences in resource availability. Journal of Personality and Social Psychology, 59, 150-168. Matthews, G., Davies, D. R., & Holley, P. J. (1990). Extravcrsion, arousal, and visual sustained attention: The role of resource availability. Personality and Individual Differences, I 1, 1159-1173. Matthcws, G., & Harley, T. A. (1993). Effects of cxtraversion and stir-report arousal on semantic priming: A connectionist approach. Journal of Personality and Social Psychology, 65, 735-756. Matthcws, G., & Harley, T. A. (1996). Connectionist models of emotional distress and attcntional bias. Cognition and Emotion, 1O, 561-600. Matthcws, G., & Margetts, I. (1991). Sdf-report arousal and divided attention: A study of performance operating characteristics. Human Performance, 4, 107-125. Matthcws, G., & Westcrman, S. J. (1994). Energy and tension as predictors of controlled visual and memory search. Personality and Individual Differences, 17, 617-626. Miller, G. A. (1956). The magic number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81-96. Mischcl, W. (1984). Convcrgenccs and challenges in the search for consistency. American Psychologist, 39, 351-364. Navon, D. (1984). Resources - a theoretical soup stone? Psychological Review, 91, 216-234. Nccka, E. (1989). Speed, capacity, tolerance: A formal model of the process of thinking and its implications. Zeszyty Naukowe UJ, Prace Psychologiczne, 5, 53-80. Necka, E. (1991). Levels of mind: A multilevel conceptualization of intellect and its implications for identification of the gifted. European Journal for

High Ability, 2, 12-17. Nccka, E. (1992). Cognitive analysis of intelligence: The significance of working memory processes. Personality and Individual Differences, 13, 1031 - 1046. Neeka, E. (1994). Inteligencja i procesy poznawcze [Intelligence and Cognitive Processes]. Krak6w: Impuls. Necka, E. (1996). The attentive mind: Intelligence in relation to selective attention, sustained attention, and dual task performance. Polish Psychological Bulletin, 27, 3-24.

552

Chapter 12

Nocka, E., Gruszl~, A., & Orzechowski, J. (1996). Selective attention in giRod children. Polish Psychological Bullean, 27, 39-51. Neisser, U. (1979). The concept of intdligcncr Intelligence, 3, 217-227. Newr A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall. Norman, D. A., & Bobrow, D. J. (1975). On data-limited and resourcelimited processes. Cogniave Psychology, 7, 44-64. Nuechterlein, K. H., Parasuraman, R., & Jiang, Q. (1983). Visual sustained attention: Image degradation produces rapid decrement over time. Science, 220, 327-329. Ohlsson, S. (1984a). Restructuring revisited: Summary and critique of the Gestalt th~ry of problem solving. Scandinavian Journal of Psychology, 25, 65-78. Ohlsson, S. (1984b). Restructuring revisited: An information processing theory of restructuring and insight. Scandinavian Journal of Psychology, 25, 117-129. Raven, J. C., Court, J. H., & Raven, J. (1983). Manual for Raven's Progressive Matrices and Vocabulary Scales (Section 4, AdvaneeA Progressive Matrices). London: H. K. Lewis. Reed, T. E., & Jensen, A. R. (1991). Ann nerve conduction velocity (NCV), brain NCV, reaction time, and intelligence. Intelligence, 15, 33-47. Reed, T. E., & Jensen, A. R. (1992). Conduction velocity in a brain nerve pathway of normal adults correlates with intelligence level. Intelligence, 16, 259-272. Revelle, W., & Loftus, D. A. (1990). Individual differences and arousal: Implications for the study of mood and memory. Cognition and Emotion, 4, 209-237. Saklofske, D. H., & Zeidner, M. (Eds.) (1995). International handbook of personality and intelligence. New York: Plenum Press. Stankov, L. (1983). Attention and intelligence. Journal of Educational Psychology, 75, 471-490. Stenberg, G., Wendt, P., & Risberg, J. (1993). Regional cerebral blood flow and extraversion. Personality and Individual Differences, 16, 547-554. Steinberg, S. (1969). The discovery of processing stages: extensions of Donders' method. Acta Psychologica, 30, 276-315. Steinberg, R. J. (1985). Beyond 1Q: A triarchic theory of human intelligence. New York: Cambridge University Press. Steinberg, R. J. (1990). Metaphors of mind: Concepaons of the nature of intelligence. Cambridge: Cambridge University Press.

E Necka

553

Steinberg, R. l., & Powell, J. S. (1983). Comprehending verbal comprehension. American Psychologist, 38, 878-893. Szymura, B., & Necka, E. (1996). Visual selective attention and personality: An experimental verification of three models of extraversion (submitted, manuscript available from the author). Thayer, R. E. (1967). Measurement of activation through self-report. Psychological Reports, 20, 663-678. Thayer, T. (1970). Activation states as assessed by verbal report and four psychophysiological variables. Psychophysiology, 7, 86-94. Thayer, R. E. (1978). Toward a psychophysiological theory of multidimensional activation (arousal). Motivation and Emotion, 2, 1-34. Thayer, R. E. (1989). The biopsychology of mood and arousal. New York: Oxford University Press. Thayer, R. E., Newman, I. R., & McClain, T. M. (1994). Self-regulation of mood: Strategies for changing a bad mood, raising energy, and reducing tension. Journal of Personality and Social Psychology, 67, 910-925. Thayer, R. E., Peters, D. P., Takahashi, P. l., & Birkhead-Flight, A. M. (1993). Mood and behavior (smoking and sugar snacking) following moderate exercise: A partial test of self-regulation theory. Personality and Individual Differences, 14, 97-104. Vandierendonck, A., De Vooght, G., Van der Goten, K. (1995). On the utility of random generation of time intervals as a central executive interference task in the double-task paradigm. Paper presented at the VIIIth Conference of the European Society for Cognitive Psychology (ESCoP), Rome, September 23-27. Vernon, P. A. (1983). Speed of information processing and general intelligence. Intelligence, 7, 53-70. Vernon, P. A. (1985). Individual differences in general cognitive ability. In L. C. Hartlage & C. F. Telzrow (Eds.), The neuropsychology of individual differences. New York: Plenum Publishing Corporation. Vernon, P. A., Nador, S., & Kantor, L. (1985). Group differences in intelligence and speed of information processing. Intelligence, 9, 137-148. Wells, A., & Matthews, G. (1994). Attention and emotion: A clinical perspective. Howe: Lawrence Erlbaum Associates. Williams, J. M. G., Watts, F. N., MacLeod, C., & Mathews, A. (1988). Cogmtive psychology and emotional disorders. Chichester: Wiley.

554

Chapter 12

Zeljko, J., & ~verko, B. (1994). Time-sharing factors and their relation to cognitive abilities and personality traits. Personality and Individual Differences, 16, 297-308. Author Note

This study has been supported by the grant No. PB 0900/P 1/94/07 from the Scientific Research Committee (KBN) to Edward Neeka. The author wishes to thank the members of his research group for the invaluable help and support obtained from them during every stage of the project.

Subject Index Acetylcholine 171-1 Achievement, need for 31-32, 50 Adaptation, 8-9, 11-19, 24-25,264, 359, 362, 384, 386, 399, 426-435,447, 455 Affect-related bias 58, 90, 94, 96, 194, 268-269, 279, 301-302, 306, 308, 310, 313-342, 345-348, 361-365, 368-369, 375-378, 387-388, 452, 456, 459, 519 Affective primacy hypothesis 267 Agencies 36-38, 60 Agreeableness 446-447 Alzheimer's Disease 66 Amplification model of affect-cognition interaction 196-243 Amygdala 87, 145, 178, 273-274, 284, 445-446, 451-452 Anger 31, 56-57, 72, 200, 241,273-274, 284, 288, 366-369, 379-382, 387-388 Anhedonia 283, 475,480 Annealing 98 Anxiety 3, 5-7, 14, 17, 20-24, 81, 83, 8688, 96, 1 3 5 , 1 9 4 - 1 9 5 , 2 0 9 , 268, 445446, 449, 452-466, 518-519 Appraisal 4, 7, 14, 22, 59, 237-238, 243, 259-266, 269, 277-290, 367, 375, 382, 387, 402-404,430-431,463,465 Architecture 8-13, 19-24, 76, 89, 163, 167, 172, 174-175, 399, 410, 432, 434, 543, 545 Arousal 4-6, 13-14, 49, 200-201,236, 285,287, 406-412, 417, 425-426, 431, 434, 445, 509-524, 527, 534, 536-547 see a l s o Energy Associative memory network see Semantic network Artificial Intelligence 4, 542 Attention 1 8 , 20-22, 161, 198-203,230, 234-238, 242, 268-269, 301,408, 423, 429, 443-467, 491-493, 504-505 - anterior system 233,425,448-449, 466 -capacity/resources 6, 21,235, 259, 266, 277-278, 282, 340, 362-363,408417, 423,477, 493,505-506, 512-515, -

517, 519-522, 530, 536-541,545 -divided 17, 405-406, 410, 412-417, 423,454, 505-506, 526-534, 540 - orienting (spatial) 233,451-452, 456458 - posterior system (spatial) 233,448449,451-452 - selective/focused 20-21,232-235, 285,411,418, 453-455,477-478, 490, 494, 496, 498, 504 - sustained (vigilance) 405, 410, 412414,419, 423,448, 452, 466 see a l s o Controlled processing, t reattentive processing, Priming Attribution 88-90, 282, 284, 288-289, 357, 361-364, 368, 375, 381,387-388, -

463- 465 - depressogenic attributional style 8890

Authoritarian personality 387-388 Autoassociation 83-84, 163-167 Automatic processing 20-22, 133, 144, 147-149, 172, 177, 196, 264-269, 273277, 281-282, 285,288, 326, 387, 412, 415,461,477, 491-498 see a l s o Controlled processing -

Backpropagation 173-175, 312-313, 341 Bidirectional Associative Memory (BAM) 66-71, 75-101 Brain, evolution of 45-47 Brain imaging 5, 9, 144-145, 150, 233, 388 Circumplex model 71-72 Classical theory of cognitive science 4, 7 8

Codes 9, 14-16, 72-75, 181,196, 199-202, 206-211,217, 230, 237, 270, 413-414, 493-495 Cognitive behaviour therapy 5, 66, 91-94, 326 Cognitive patterning 405-406, 434 Cognitive style 100, 286-288

556

Subject index

Commitment problems 371-378 Connectionism 7, 8, 15, 19, 21, 23-24, 6595, 109-122, 129, 164-176, 198, 243, 285, 302, 305-348, 419-424, 434, 458, 544-547 Conscientiousness 449 Consciousness 38-40, 49-50, 52, 56, 5960, 66, 76, 79, 123, 161, 199, 232-235, 260-261,306, 387, 411,450, 476, 478, 497-498 s e e a l s o Unconscious processing Controlled processing 133, 142, 174,268270, 273-274, 277, 281,326,412, 415416 s e e a l s o Automatic processing Coping 4, 7, 16, 238, 265,280-281,288, 303-304, 310, 321,324-325, 327, 331332, 347, 447, 463,465 distractive 329-332 - emotion focus 366, 402-404, 430 - problem/task focus 22, 366, 402-404, 430 s e e a l s o Rumination Core relational theme 4, 15,270 Cortex 36, 46, 65-67, 87, 127-132, 152, 162, 180, 274-276, 279, 444-446, 448, 462, 466-467, 509 s e e a l s o Frontal lobes Creativity 123, 126, 179, 287, 341,423 -

-

-

-

-

Depression 5, 17, 65, 81, 88-91, 98, 283284, 301-308, 313-315, 325-329, 332340, 345-348, 434, 449, 463, 519 Design 4-5, 355-356, 359, 365, 427-428 Dissociative disorders 76-80 Dopamine 15, 98, 135, 159, 408, 425, 431,445, 448, 451,454 Dtml-task see Attention - divided Dysphoria s e e Depression, Sadness Electroconvulsive therapy (ECT) 98-99 Electrophysiology 78-80, 132, 142, 144, 150, 198, 242, 424, 510 Emotional Stroop 20-22, 268 Enablers 36-38, 50, 52, 56, 60 Energy 52, 57, 400-405, 411-425,428430, 505, 509-11,527, 530-542 Establishments 36-38, 60

Evoked potentials see Electrophysiology Evolutionary psychology 11-13, 18-19, 25, 46-47, 198, 358, 365, 368, 383, 386, 447 Expectations 204-206, 210-212, 215, 217222, 224-231,259, 356, 363, 420, 510 Extraversion-introversion 7, 8, 10-11, 1718, 19, 31, 38, 134-135, 200, 287, 399435, 445-446, 511,524, 527, 533-534, 536, 540-541 - interaction with time of day 406, 418-423,428 Fear 57, 72, 269-270, 274-276, 279, 284, 359, 367-369, 445, 447, 452 Five Factor Model 5, 341 Frontal Lobes 87, 123-126, 129, 145, 151154, 158-161,174-181,276, 283, 424, 448, 477 Genetics 5, 11-12, 18-19, 25, 432-433, 435 Goals s e e Motivation Guilt 56-57, 88, 285, 372, 374-379, 382 Happiness 31, 50, 56-57, 72, 200, 361364, 366-367, 400-404, 407, 429 Hedonic tone see Happiness Hippocmnpus 87, 123-132, 135-148, 151155, 158-172, 176-181,275-276, 312, 445, 477 - dentate gyrus 129, 162-163, 165-173, 181 -CA3 128, 155, 163-168, 170-173 - s e e a l s o Septo-Hippocampal System Homesickness 403-404 Impulsivity 6, 18, 200, 399, 407, 409, 432, 445, 449 Incentive 368, 371,378-379, 444-446, 453-458, 515 Interrupts 4, 58 Intelligence 151,340, 345-346, 406, 423424, 503-507, 512-513, 519-530, 536547 Interacting Cognitive Subsystems 270-271 Interactive activation network 419-423

Subject index

557

Joy see Happiness Judgement 31, 48, 58, 284, 355-357, 359362, 368, 370, 378-379, 381,385-386, 387-388

430, 432-434, 443-454, 464-467, 504, 511,544 see also Incentive, Punishment, Reward

Lexical decision 238, 240, 314-342, 417, 420-422, 493, 545 Limbic system 46-47, 178, 199, 445,449, 462, 477

Neural networks see Connectionism Neuroscience 5, 9, 13-15, 17, 23-24, 4546, 8%88, 126, 274-276, 388, 407-408, 431,443-446, 448-452, 459, 467 Neuroticism 7, 18, 134-135,285, 287, 404, 445-446, 456, 459, 524, 527, 533, 536, 540 Noise (acoustic) 16-17, 406, 423, 511, 542 Nomdrenaline 135,448

Mate choice 382-384 Memory 36, 50, 65-70, 74-75, 264, 271272, 280, 306, 334-345, 362, 415,425, 431,465 -amnesia 78, 80, 96, 99, 124, 126-127, 129, 136-144, 165, 176 - implicit 94, 126-127, 271,280 - intermediate term 129-130, 133, 136, 140-141,143-149, 153-154, 162-164, 170, 236 - long term 7, 16, 22, 24, 87, 99, 124129, 132-136, 141,144, 149, 152-153, 159-164, 171,174, 179-181,199, 236, 433,463 -short term 10, 17, 19, 127, 130, 134144, 153, 167, 170, 179-180, 236, 408, 418-419, 427, 432, 465, 507-508, 512, 515-516, 525-526, 529-530, 534-536, 540 -spatial 130, 152-153, 158-159 - well see Bidirectional Associative Memory (BAM) -working 123-126, 128-135, 137-142, 144-154, 159-162, 164, 171,177-178, 181,279, 285, 503-504, 507-509, 512521,523,537-547 Metacognition 22, 326, 523-524 Modules 14, 18, 243, 384, 411 Mood 6-7, 58-59, 66, 90, 94, 96, 238, 241,243,289, 335, 340, 360-367, 379, 387, 400-405, 411,428-429, 511,523524 Mood-congruence see Affect-related bias Mood disorder see Depression Motivation 6, 11-13, 16-17, 22, 31-32, 3940, 47, 49-51, 56-59, 66-70, 123, 125126, 152, 158-159, 174-179, 181,194, 265,286-288, 362-363, 371-379, 429-

-

Obsessive-compulsive disorder 83-84, 96, 322 Openness to experience 287, 340-341, 346-347 Panic 84-85, 87, 446 Parallel Distributed Processing (PDP) see Connectionism Paranoia 283 Perception 40, 49, 51, 58, 85, 93-95, 110, 120, 123, 126, 172, 193-204, 216, 232, 269, 271 - disorders of 475-476, 480 - "New Look" 194-195 - perceptual defence 194-196, 202, 209-210, 216, 224, 232, 238-240, 242 Pharmacotherapy 91, 97-98, 100, 479, 485,488-489, 497 Phobia 67, 85-87, 100, 269, 447, 452 Posttraumatic stress disorder see Trauma Preattentive processing 194-202, 230, 234-235,239, 482, 494, 496 Prefrontal cortex see Frontal lobes Preparedness 65,269-270 Priming 126, 134, 145-148, 171-172, 174175, 194, 230, 267, 276 - negative 451, 478-497 - semantic 197, 420-422, 459-462, 482484, 489-491,495-496 Prisoner's Dilemma 12, 373-378 Psychopathy 378 Psychoticism 18, 446

558

Subject index

Punishment 7, 39, 378-381,406, 426, 429.430, 445

Subliminal stimuli see Unconscious processing

Rationality 359-360, 371,388 Rehearsal 133-145, 1 7 0 Relational model of personality 32-38, 4860 Relaxation see Tension Response bias 195, 202, 204, 210, 215, 221, 221, 231,406 Reward 7, 39, 158, 371-374, 378, 402, 406, 426, 429.430, 445-447 Rumination 22, 301-304, 321-329, 331, 464

Temperament 444, 447, 449-450, 452, 465 Tension 400, 403, 416, 510-511,518, 527, 534, 537-542 Themes 36-38, 60 Time-sharing 506 Token economy 97 Transactional model 4, 7, 280, 402, 432 Trauma 70, 79-83, 85, 87, 91, 96

Sadness 17, 50, 57, 72, 314, 318, 332, 361-363, 366 Schema 6, 23, 66, 240, 264-265, 270, 278-279, 288, 303-304, 332, 431,452, 466 Schizophrenia 15, 76, 97, 449, 475-476, 479-482,485-490, 494-498 - positive vs. negative symptoms 475476, 479-480, 484-488, 495-496 - reduced cognitive inhibition model 490.498 Sehizotypal personality 479-498 Search tasks 412, 414.416, 419, 423 Self, representations of 6, 22-23, 36, 50, 60, 66, 280, 288, 433, 463, 524 Self-focus 22, 463-465 Self-Referent Executive Function 22-23 Sernantie network 80, 194, 241,306, 314, 318, 425, 461,491 see a l s o Cormectionism Septo-Hippocampal System 14, 20, 445 Serotonin 135,446 Sociability 18, 399, 432 Social cognition 6, 8, 11, 15, 241,266, 286-287 Social skills 19, 432, 434 Spatio-temporal processing 129, 151-174, 176-178 Stimulus Evaluation Check 263, 282-284 Strategy 10-13, 16-17, 20-25, 83, 96, 123, 133, 137, 261,285, 368, 372-378, 409411,416-418, 423,425,435,452,454455, 464, 482, 496, 507, 518, 523-524 -

Unconscious processing 66, 94-96, 193, 197, 199, 239, 242, 267, 270-271,275, 282, 387, 452, 518 see a l s o Consciousness -

Valence 4, 209-213, 216, 219-221,224228, 237-238, 259, 279, 283, 312, 314, 460-462 - identification 316-342 Vigilant arousal see Energy Worry 22, 287, 430, 464, 519