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