HUMAINE Plenary Newcastle, May 24-27, 2004 Computational models of emotion Klaus R. Scherer University of Posing the problem: Three types of computational models Appraisal cirteria Integration rules Sequential process Decoding model Decoding agent Physiol. Reactions Motor expression Goal state Event Elicitation & Differentiation model EMOTION What are the best indicators or cues? Encoding model Physiol. Reactions Motor expression How can the emotion be predicted and how should it be displayed? Encoding Agent 1
Direct links to dimension mapping Appraisal criteria checking Criteria-specific outcomes Componential patterning Outcome-specific responses Integration to unique feeling Integration and synchronisation Qualia emotions Specific to individuals Semantic feature rules Labelled emotions Specific to language/culture Semantic field rules Modal (basic) emotions Universal Semantic dimension rules Affective dimensions Universal Component process model - Internal representation and feeling Physiological symptoms Cognitive appraisal Motor expression Action tendencies Unconscious reflection and regulation Zone of valid self-report measurement Conscious representation and regulation Verbalization and communication of emotional experience 2
Major emotion processes Significance detection Implication evaluation Response Preparation Stimulus Event Resource Allocation Behavioral meaning Multimodal Integration Episodic Storage Rule Extraction Multimodal Response Subjective Feeling Episodic Memory Procedural Knowledge Component process model - Appraisal Attention Memory Motivation Reasoning Self Concept Event Relevance Implication Coping Normative Significance Neuroendocrine/Autonomic/Somatic Nervous System 3
Levels of processing in the appraisal process Component process model - Response patterning Event Components Novelty CNS Pleasantness Information Goal Significance Processing Coping Potential (Appraisal) Norm Compatibility Support (ANS physiology) Motivation (Action tendencies) Execution (Motor expression) Monitoring (Feeling state) Time System interaction and regulation 4
CPM response patterning predictions for efferent effects of the conduciveness check NES ANS SNS FACE VOICE Conducive EEG synchronisation decrease in respiration rate, slight HR decrease, vasodilatation in sexual organs, increase in glandular secretion, bronchial constriction, increase in gastro-intestinal motility, relaxation of sphincters, decrease in general muscle tone relaxation of facial muscle tone overall relaxation of vocal apparatus (F0 at lower end of range, low-to-moderate amplitude, balanced resonance with slight decrease in highfrequency energy - "relaxed voice") Obstructive corticosteroid and catecholamine, particularly adrenaline secretion deeper and faster respiration, increase in HR and heart stroke volume, vasoconstriction in skin, gastro- intestinal tract, and sexual organs, vasodilatation in heart and striated musculature, increase of glucose and free fatty acids in blood, decreased gastro-intestinal motility, sphincter contraction, bronchial dilatation, con-traction of m. arrectores pilorum, decrease of glandular secretion, increase in SC level, pupillary dilatation SNS: increased muscular tonus AUs 4 (brow lowerer, frown), 7 (lids tighten), 23 (lips tighten), 17 (chin raising); gaze directed overall tensing of vocal apparatus (F0 and amplitude increase, jitter and shimmer, increase in high frequency energy, narrow F1 bandwidth, pronounced formant frequency differences - "tense voice") Component process model - Monitoring and regulation via multimodal integration Appraisal SNS ANS patterning and amplitude patterning and amplitude Regulation patterning and weight of criteria Feeling Quality Intensity Duration Motivation urgency 5
What is the organizing principle for integration? What is basic feeling? Discrete emotion theories: Feeling is organized around the basic emotions which constitute the fundamental building blocks of the emotion system (specific circuitry and neuromotor programs organize physiology, expression, and associated feeling) Dimensional theories: Wundt - feeling constituted by three dimensions. Russell & Feldman Barret - core affect constituted by valence and arousal, enriched by additional cognitions Component process theory: Multimodal integration of appraisal processes and their motivational, physiological, and expressive consequences in the form of qualia. From the latter dimension judgments can be extracted or discrete states can be categorized according to semantic conventions. An actuarial study of Swiss emotions Representative sample of the Swiss population, N = 1003. Question: Describe the strongest emotion you had yesterday. How would you call it? Happiness 14,8 Anger 14,1 Anxiety 9,6 Joy 8,8 Sadness 8,7 Disappointment 6,5 Tension / Stress 6,1 Desperation 6,0 Contentment 5,5 Uncodable 4,8 Irritation 4,1 Negative 3,7 Positive 2,7 Compassion 2,4 Pleasure / Enjoyment 2,1 Pride 2,0 Fear 2,0 Stupefaction 2,0 Surprise 1,9 Guilt 1,7 Relaxation / Serenity 1,7 Relief 1,4 Love 1,1 Amusement 1,1 Gratitude 1,0 Hate 0,8 Interest 0,7 Disgust 0,6 Longing 0,6 Being touched 0,6 Admiration / Awe 0,5 6
Norman Anderson: Information integration Norman Anderson claims that the regression model is inadequate to describe information integration in emotion-antecedent appraisal and pleads for the detailed study of integration rules N.H. Anderson (1989). Information integration approach to emotions and their measurement. In R. Plutchik & H. Kellerman (Eds.) : Theory, Reserach, and Experience. NY:Academic. Modeling and measuring the integration of appraisal results -- based on Anderson (1989) intentional accidental Intensity of Anger not responsible Agency Extent of Goal Obstruction Parallel lines = additive function 7
Modeling and measuring the integration of appraisal results - based on Anderson (1989) intentional accidental Intensity of Anger not responsible Agency Extent of Goal Obstruction Fan-shaped lines = multiplicative function Catastrophy Theory An Example of Hysteresis Anger Hysteresis Frustration 8
Modeling and measuring the integration of appraisal results -- a nonlinear approach intentional accidental Intensity of Anger not responsible Extent of Goal Obstruction Catastrophy Model of Fear/Anger Conflict (Zeeman, 1976) 9
Catastrophy Model of Appraisal (based on Zeeman, 1976) BEHAVIOR OR RESPONSE SURFACE Anxiety Sadness Fear Satisfaction Joy Elation Triumph Hope Irritation Anger Rage ABRUPT CHANGE INACCESSIBLE REGION HYSTERESIS RESPONSE VARIABLES Conducive Low power Obstructive CONTROL SPACE High power Scherer, K. R. (2000). s as episodes of subsystem synchronization driven by nonlinear appraisal processes. In M. D. Lewis & I. Granic (Eds.), development, and self-organization: Dynamic systems approaches to emotional development (pp. 70-99). New York: Cambridge University Press. 10