From desire coherence and belief coherence to emotions: A constraint satisfaction model Josef Nerb University of Freiburg Department of Psychology 79085 Freiburg, Germany http://www.psychologie.uni-freiburg.de/signatures/nerb/ [email protected]
Abstract The model is able to account for the elicitation of discrete emotions using these five criteria. In the DEBECO archi- Appraisal theory explains the elicitation of emotional reactions as a consequence of subjective evaluations tecture, the five criteria evolve out of three basic principles: based on personal needs, goals, desires, abilities, and belief coherence, desire coherence and belief-desire coher- beliefs. According to the appraisal approach, each emo- ence. Based on these principles, which are all well sup- tion is caused by a characteristic pattern of appraisals, ported by empirical evidence from psychological research, and different emotions are associated with different ap- DEBECO also allows the representation of two elementary praisals. This paper presents the DEBECO architec- types of goals: approach goals and avoidance goals. The ture, a computational framework for modeling the ap- DEBECO architecture is implemented as a parallel con- praisal process. DEBECO is based upon three core straint satisfaction network and is an extension of Thagard’s coherence principles: belief coherence, desire coher- HOTCO model (Thagard, 2000, 2003; Thagard & Nerb, ence and belief-desire coherence. These three coher- 2002). ence mechanisms allow two elementary types of goals to be represented: approach and avoidance goals. By Within DEBECO, emotions are seen as valenced reac- way of concrete simulation examples, it is shown how tions to events, persons or objects, with their particular na- discrete emotions such as anger, pride, sadness, shame, ture being determined by the way in which the eliciting situ- surprise, relief and disappointment emerge out of these ation is construed (cf. Ortony et al., 1988). DEBECO is not principles. The framework allows the integration of a concerned with how desires arise or how beliefs are acquired variety of empirical findings about cognitive emotional in the first place, but rather with how desires and beliefs are congruence effects and shows how phenomena as di- integrated while a situation is being construed. Furthermore, verse as reasoning, belief updating, or eliciting of emo- DEBECO does not deal with physiological or neurological tional reactions can be subsumed under the same basic influences on emotions or with the expression of emotions. underlying processes.
Introduction Basic principles of the appraisal process Emotions are not simple reactions to situational stimuli. Like others (e.g. Smith & Kirby, 2000), we assume that the Highly similar situations can elicit different emotions in a very core process in appraisal is a continuously operating, person over time and different situations can give rise to preattentive mechanism that checks whether newly acquired similar emotional reactions. Instead of considering emo- beliefs are congruent or incongruent with pre-existing be- tion as a fixed response to one’s objective circumstances, liefs and whether newly acquired desires are congruent or appraisal theory, a major theoretical perspective in emotion incongruent with pre-existing desires research, explains the elicitation of emotional reactions as Translating these mechanisms into a parallel constraint a consequence of subjective evaluations based on personal satisfaction network where nodes of the network represent needs, goals, desires, abilities and beliefs. These evalua- propositions or concepts is straightforward. Each node has tions or appraisals of a situation are seen as crucial for both two activation values, representing (a) evaluation, desire or the elicitation and the differentiation of emotions. Summa- valence and (b) the expectancy or belief. Nodes are con- rizing and integrating this line of research, Ellsworth and nected by excitatory or inhibitory links representing positive Scherer (2003) conclude that most theorists agree upon five and negative relationships between propositions. Activation dimensions of emotion antecedent appraisal criteria: goal- spreads between connected nodes in the network until the conduciveness, intrinsic pleasantness, belief, novelty and activations of all nodes asymptote or “relax” into a state that agency. This paper will introduce a computational model satisfies the constraints among the nodes. of the appraisal process called DEBECO (abbreviation of Desire-Belief-Coherence). Psychological roots of coherence principles Copyright c 2004, American Association for Artificial Intelli- The three principles, belief coherence, desire coherence and gence (www.aaai.org). All rights reserved. belief-desire coherence, can be traced back to early theories regarding cognitive emotional consistency in social psychol- tion of the activations of the units to which it is linked and ogy, in particular McGuire’s dynamic model of thought sys- the strength (positive or negative) of these links. Updating tems (1999 for an overview) and the psycho-logic model of happens recurrently in cycles until such systems enter stable Abelson and Rosenberg (e.g., Rosenberg & Abelson, 1960). states in which activations of all units cease to change from McGuire’s theory conceptualizes thought systems as con- one cycle to the other; this is called settling. In this state the sisting of a list of propositions with two attributes, one being overall consistency or coherence of the network is achieved. desirability (evaluation dimension) and the other being like- In DEBECO, three different pathways of activation ex- lihood of occurrence (expectancy dimension). These two change can model the three coherence principles mentioned dimensions express how much the content of a proposition above: (1) activation spreads between two nodes that both is liked and how much it is believed in. McGuire’s model represent belief values (belief coherence); (2) activation assumes that changes in any one of these dimensions of a spreads between two nodes that both represent valences (de- proposition can generate changes in other related proposi- sire coherence); and (3) activation spreads within one node tions according to a “probabilogical” consistency tendency between the belief part and the desire part (belief-desire co- that approximates the principles of logic, and a hedonic con- herence). sistency process that allows motivational inclinations to in- The literature provides numerous examples for the use of fluence reasoning. The thought system is seen as dynamic, constraint networks in belief updating, that is, the spread- meaning that a change that is directly induced in one part ing of activation along the first pathway (Holyoak & Si- of the system results in compensatory adjustments in remote mon, 1999; Thagard, 1989, 2000). Likewise, there is am- parts of the system. Together, these assumptions imply that ple psychological evidence for the spreading of desire or a person’s evaluative and expectancy judgments on a topic “sentiment” (Heider, 1958; Rosenberg & Abelson, 1960; will tend to be assimilated towards one another (McGuire, Thagard, 2003) and the interaction of belief and desirabil- 1999, p. 190): ity (McGuire, 1999). I postulate that causality flows in both directions, re- Recent support for a desire-belief interaction comes flecting both a “wishful thinking” tendency such that from findings within Behavioral Decision Theory, where people bring their expectations in line with their de- probability-outcome interaction effects are found when the sires, and a “rationalization” tendency such that they outcomes of decisions (Rottenstreich & Hsee, 2001) or the bring desires in line with expectations. targets of judgments are affectively rich (Slovic, Finucane, The psycho-logic model of Abelson and Rosenberg as- Peters, & MacGregor, 2002). These findings are particularly sumes mental representations of cognitive elements and pos- crucial because both expected utility and prospect theory itive, negative and indifferent relations between these enti- assume strict independence of probability (belief) and out- ties (Rosenberg & Abelson, 1960). Pairs of elements con- come (desire). The mechanisms proposed in DEBECO aim nected by relations are called cognitive units or bands. In ad- at finding a state of activations within the network that max- dition, elements have affect-arousing significance and bands imizes belief coherence, desire coherence and belief-desire can be balanced, consistent and stable, or imbalanced, in- coherence. consistent and unstable. A band is stable when (a) two con- cepts of identical affects are believed to be positively related Updating of desires and beliefs: Details or (b) two concepts having opposite affects are believed to be negatively related. Conversely, instable bands are those in which (a) two concepts of identical affects are negatively Let beli and desi denote the belief and desire values of related or (b) two concepts of opposite affects are positively a node i. The updating function of the belief parts of a related. Achieving affective or “hedonic” balance is clearly a node is the standard activation rule for constraint satisfac- constraint satisfaction problem. It should be noted here how- tion network (see McClelland and Rumelhart, 1988; Tha- ever, that detecting and fixing such inconsistencies within gard, 2000): large structures was not possible before appropriate compu- tational tools became available (Vallacher, Read & Nowak, netbeli(max − beli(t)) 2002). if netbeli > 0 DEBECO translates the principles and pre-computational notions of McGuire’s dynamic thought system as well as beli(t + 1) = beli(t)(1 − δ) + Abelson’s and Rosenberg’s psycho-logic model into com- netbeli(beli(t) − min) putational terms using a parallel constraint system. Propo- sitions and concepts are represented by nodes within a net- otherwise work. Each node has two activations, representing (a) the (1) evaluation, desire or valence value and (b) the expectancy or Here, d is a decay parameter (say .05) that decrements each belief value of a proposition or a concept. Nodes are con- unit at every cycle, min is the minimum activation of the nected by excitatory or inhibitory links representing positive belief part (−1), max is the maximum activation of the be- and negative relationships between entities. Parallel con- lief part (1). All links between nodes i and j have a weight straint satisfaction is performed by means of equations that wij reflecting the strength of the relationship between these specify how the activation of each unit is updated as a func- nodes. The input to the belief part of a node i, netbeli, can be calculated by: Pre-existing beliefs and desires
n When people encounter a new situation they do not develop X netbeli = wijbelj(t). (2) their cognitive and evaluative assessments all from scratch. j=1 The assessment is rather embedded within and influenced by pre-existing desires and beliefs. People hold beliefs that where netbeli is the net input to the belief part of a node i. something is present or absent, likely or unlikely, and have The updating rule for the desire part desi of a node i is acquired positive or negative affects towards a person, ob- like the updating of beliefs: ject or situation. Likewise, DEBECO networks usually do not come as tabulae rasae when situations are appraised. In- netdesi(max − desi(t)) stead, such networks use special nodes, OBSERVED and LIKED, to express pre-existing beliefs and desire values. if netdesi > 0 As in the IMP model of Kunda and Thagard (1995), ex- desi(t+1) = desi(t)(1−δ)+ pressing that something is given is accomplished by setting a positive link between the belief part of a node and a special netdesi(desi(t) − min) node OBSERVED; expressing that something is not given is otherwise done by setting a negative link between the belief part of a (3) node and OBSERVED. The OBSERVED node has constant where netdesi is defined as: maximum positive activation that remains unchanged by the belief updating equation. Setting desire values is similar to n X setting beliefs: a positive link to a special node LIKED al- netdesi = wijdesj(t). (4) lows a positive desire value to be represented; conversely, j=1 a negative link to the LIKED node stands for a low desire value, expressing that something is undesirable. In addition to these two pathways of activation exchange, Put together, four different states are possible for each the desire part of a node has a weighted unidirectional ex- node: (1) a positive link to LIKED and to OBSERVED ex- citatory link to the belief part of the node. The weight of presses evidence for something positive, while (2) two neg- this connection, desire-belief weight, may have values from ative links express that something negative was avoided; (3) 0 to 1, reflecting the strength of this influence. A value of a positive link to LIKED and a negative link to OBSERVED 1 would amount to an extreme case of wishful thinking; a represents that something positive did not happen, and fi- value of 0 would totally preclude emotional input from the nally, (4) a negative link to LIKED and a positive link to calculation of belief values. Likewise, the belief part of a OBSERVED shows that something negative is given. Note, node has a weighted unidirectional excitatory link to the de- however, that both belief and desire values of a node can sire part of the node. The weight of this connection, belief- change during the updating of the whole network. Although desire weight, may also have values from 0 to 1 reflecting the links to the special units OBSERVED and LIKED will have strength of this influence. Allowing beliefs to influence de- a strong influence on the final activations of a node, this in- sires can lead to rationalization, a tendency to bring desires fluence can be overridden by different influences of other in line with expectations. connected nodes. The flow of activation along these links is already cov- In addition to positive links to OBSERVED, there is a ered in Equations 1–4. As can be seen from these equations, way in which to express that one assumes the truth of a the activation of the belief or desire part of a node increase proposition—namely, to set the activation of the belief part as a function of positively connected nodes with positive to +1; assuming a proposition to be wrong is done by set- activations and negatively connected nodes with negative ting the belief part to −1. All values in-between indicate activations, and decrease as a function of negatively con- respective degrees of assumed credibility of a proposition. nected nodes with positive activations and positively con- In particular, a value of 0, which is the default value, indi- nected nodes with negative activations. cates that nothing is known a priori about the credibility of Depending on the setting of the parameters for desire- a proposition. belief weight and belief-desire weight, isolated nodes would be prone to showing massive wishful thinking effects, where the desirability of a proposition brings about congruent per- Goal achievement ceptions of its likelihood—or rationalization effects, where Within the DEBECO architecture, desire values of nodes the likelihood of a proposition leads to congruent percep- represent whether and to what degree a person wants to tions of its desirability. However, in addition to this desire- avoid or approach something. In this sense, desirability val- belief interaction, beliefs and desires are usually also con- ues determine avoidance goals and approach goals. Acti- strained by other belief links and other desire links. Since vations of belief values determine how plausible, possible, inputs from those two sources have a massive influence on likely, or believable something is—and considered together the activations of nodes, wishful thinking and rationaliza- with desirability values, how successful a person is in ap- tion can be easily overruled by other evidence (beliefs) and proaching a desired state or avoiding an undesired state. preferences (desires). Thus, both wishful thinking and ratio- Thus, an integration of the desire and belief values of a node nalization are not at all unconstrained. indicates the degree of goal achievement of a node, that is, to what degree a state of affairs represented by a node is Desire: consistent with, or conducive to, achieving one’s goals. Goal achievement is realized within the system by a node positive that only has one activation value resulting from the mul- tiplicative integration of the belief and desire part of the referring node. In addition to a goal-achievement value for each goal, there is a global node that reflects the over- + all goal achievement of the system. The activation of this global node results from the integration of all specific goal- Belief: Belief: achievement values and mirrors how successfully all goals false true of the systems are achieved. Figure 1 depicts the relation- ship between belief and desire values, goals, goal achieve- ment, and global goal achievement. +
Belief node-1 Belief Desire node-4 ga-4 Desire: Desire negative
GA
Belief node-2 Figure 2: Positive emotions arise when something is true Desire Belief node-5 ga-5 that is desired (successful approach) or wrong that is not de- Desire sired (successful avoidance); negative emotions arise when something is wrong that is desired (unsuccessful approach) Belief negative link or true that is not desired (unsuccessful avoidance). node-3 Desire positive link Surprise
Figure 1: Belief and desire values, goal achievement and Surprise is elicited by the ascribed novelty or unexpected- global goal achievement. Hypothetical network with five ness of an event (cf. Reisenzein, 2000). An event is ap- nodes. Nodes 4 and 5 are goal nodes with respective goal- praised as unexpected when the integration of new input achievement nodes (ga-4 and ga-5); GA is the global goal leads to a substantial change in the interpretation of a sit- achievement. All nodes are interrelated by inhibitory links uation. In terms of our framework, novelty and the sub- (dashed lines) or excitatory links (solid lines). sequent emotion of surprise arise when new input appears to be incompatible with the currently activated belief-desire network. The degree of novelty determines the intensity of Appraisals and emotions the feeling of surprise. In the following, examples are given of how new situations Novelty is high when new input leads to a coherent in- are appraised within the DEBECO framework. We will fo- terpretation of a situation that substantially differs from the cus first on which emotions arise in a particular situation prior coherent interpretation of that situation. Hence, both and how they arise. Then various effects stemming from the surprise and novelty are a function of how many belief val- interaction among emotions and among emotions and ap- ues and desire values change their activation value when new praisals will be discussed. As we will show, emotions such input is integrated. as surprise, disappointment, relief, anger, sadness, shame The computation of novelty and surprise is therefore and pride emerge without further assumptions out of the ba- straightforward and follows the implementation suggested sic framework of DEBECO. by Thagard (2000). After the network settles, each node records its final activation values for its belief part and its Positive and negative emotions desire part. When new information is added to the network The desire value of a node, that is the pleasantness of a stim- and the network settles again, DEBECO computes the dif- ulus per se, does not allow negative and positive emotions ference between the new activation and the previous activa- to be differentiated. A positive emotional reaction might in- tion. This difference represents the extent to which the new volve pleasant and unpleasant stimuli. Certainly, approach- information has produced a “surprising” change for a node. ing the pleasant and avoiding the unpleasant are both asso- Each change in activation conveys to the surprise node the ciated with positive emotions, whereas the opposite leads extent to which it is locally surprised, so that many nodes af- to negative emotions (see Figure 2). Thus, the appraisal of fect the general surprise node. In a similar vein, changes in goal achieve- ment is different from the appraisal of intrinsic the overall goal-achievement node reflect whether an event pleasantness in an important way. was a pleasant or an unpleasant surprise. Disappointment and relief new information is integrated. The emotions disappointment and relief are closely related According to DEBECO, disappointment and relief are ac- to expectations. Consider someone who has just finished an companied by negative and positive surprise, respectively. oral exam and is waitint in front of the professor’s office for Indeed disappointment and relief are special types of sur- to get his grade. The student desires to get a good grade, prise. This follows from how surprise is modeled within and is very much expecting this to happen, that is, the stu- DEBECO. However, not every surprise is disappointing dent has a positive expectation that something he likes will or relieving; only belief-shifts within a goal node have this come true. Translating such a situation into DEBECO terms quality. In line with Miceli and Castelfranchi (2000), disap- is simple: there is an approach goal for getting good grades pointment is a falsified belief that the desired will come true expressed by a node for “good grade” that has a high desire (falsification of a positive expectation) and relief is a falsi- value and a high value for belief. The desire value is pos- fied belief that the undesired will come true (falsification of itively linked to the special node LIKED. Since the belief a negative expectation). is bolstered by related beliefs, this value remains positive. Thus, the node has a positive goal-achievement value. Now Anger, sadness, shame and pride back to the student. The door opens and the professor says The attribution of agency and the assignment of responsi- that the student has failed the exam. No doubt the student bility are central for differentiating emotions such as anger, will be disappointed when his positive expectations turn out sadness, shame or pride (Nerb & Spada, 2001; Ortony et al., to be falsified by reality. His disappointment will be greater 1988; Roseman et al., 1996; Smith & Ellsworth, 1985). Ac- the more he desired to pass the exam and the more he had cording to these theorists, the following appraisal-emotion positive expectations to do so (see Miceli & Castelfranchi, associations can be derived: a negative event will lead to 2000, for a related and more detailed treatment of expec- sadness if it is caused by situational forces; if someone tations and disappointment). Translating this episode into else is appraised as being responsible for the negative event, DEBECO is straightforward. Reality forces the belief value anger will arise; if the self is seen as responsible for a nega- of “good grade” to become negative. Disappointment is this tive event, shame will arise; and pride will arise if the event shift in the belief part of a node that has a high desire value. is appraised as positive. The intensity of disappointment results from multiplying the In our framework, an event is appraised as negative when value of the desire part of the node by the decrease in abso- its related goal-achievement value is negative and appraised lute terms of the belief part of this node after the new infor- as positive when its related goal-achievement value is pos- mation is integrated. itive. The appraisal of agency, that is the assignment of causality and responsibility, can easily be modeled using Relief is reciprocal to disappointment. Instead of having parallel constraint satisfaction models. We are therefore a positive expectation, one has a negative expectation that taking the result of the agency/responsibility appraisal as a something that is undesirable will happen. In our example given and are exploring more deeply how emotions such as the student has the goal of not failing the exam. When he sadness, anger, pride and shame are related to the appraisals is waiting outside the office, he has a negative expectation of agency/responsibility and goal achievement. At the cen- that he might have failed. The door opens and the student ter of this analysis we will use an abstract description of learns instead that he has passed. Certainly then, the student an event consisting of information about the agent and the will feel relieved when his negative expectation gets falsi- control, knowledge and the positive motive of the agent for fied. The relief will be greater the more the student desired causing an outcome. Such an abstract description of an event not to fail the exam and the more he had negative expecta- has the simple form: tions that he would fail.
anger-2 ga-1 sadness-2
Bel = .9 Outcome ANGER Des = -.8 SADNESS
Pers. Cause anger-1 sadness-1
Con- trol
negative link Neg. positive link Motive
Know- OBSERVED ledge
Figure 3: Relationship between an abstract description of an event and the emotions anger and sadness. Situation-specific emotion nodes (anger-1, sadness-1) are linked to global emotion nodes (ANGER, SADNESS). The illustration shows a situation in which an agent (personal cause) caused a negative outcome, that is, something that is negative (e.g. desire value = -.8) becomes true (e.g., belief value = .9). The agent had control over this action and had a bad motive for doing so (expressed by links to OBSERVED). Nodes are interrelated by inhibitory links (dashed lines) or excitatory links (solid lines).
Figure 3 depicts a situation in which an agent is another Emotional mood and emotional carry-over effects person who has caused something negative; the agent had Sometimes emotional experiences lead to a more diffuse control over this action and a bad motive for doing so, and it overall emotional mood that has little or no reference to is unknown whether the agent could foresee the consequence the antecedent situation originally causing it. By emotional of his action. The illustration also shows the theoretical pos- mood we mean an integration of several emotional situations tulated relations of these entities to the emotions anger and that happen at the same time or within a short period of time. sadness. Both anger and sadness are negatively related to To account for these more general emotions or emotional the goal-achievement node, indicating that both are negative moods, each event specific emotion is connected by a bidi- emotions. However, for the goal incongruent event that pro- rectional link to a global emotion node (Figure 3). Thus, duces sadness, no agent is held accountable for the negative emotions are represented similarly to goal-achievement val- outcome (see Nerb & Spada, 2001). ues (see Figure 1). This global emotion node may be inter- Only if the person locates an accountable agent, will the preted as a specific mood, such as an angry or sad mood. emotion be anger; if a person locates him or herself as the This conceptualization allows for an emotional carry-over accountable agent for the negative event, the emotion will be effect between situations. For instance, if the input given to shame (see again Roseman et al, 1996; Smith & Ellsworth, the network consists of two events which, taken in isolation, 1985). Shame results as an emotion when the event is goal are causing only mild anger, the situation-specific anger for incongruent, and pride results when the event is appraised as each event will be higher when the events are presented to- goal congruent. Simulations within DEBECO confirm these gether than when each of the events is presented alone. This predictions. emotional summation effect results from bidirectional rela- tions between specific emotions and the global emotion. SADNESS (global) Figure 4 presents simulation runs of DEBECO showing 0.8 ANGER (global) how global emotional moods, SADNESS and ANGER, lead 0.6 to “carry-over effects” to specific situations that are consis- 0.4 tent with the underlying appraisal patterns of the emotions. The simulation results are consistent with the empirical data 0.2 reported by Keltner et al. (1993). 0 Effects within one episode. Coherence effects between -0.2 emotions and associated cognitions do not only happen be- Activation -0.4 tween different episodes, but also occur within the evalua- tion of a single event. Nerb and Spada (2001) found such -0.6 affective coherence effects in empirical studies investigat- -0.8 ing the evaluation of environmental accidents. In a series of three studies, the authors found that some non-manipulated Person Situation Control Knowledge Motive Sadness Anger attributes of the reported event were evaluated so as to maxi- Appraisals Emotions mize the overall consistency of cogntive and emotional judg- ments. In particular, the authors report that participants have construed judgments that “go beyond the information given” Figure 4: The interpretation of the same ambiguous event in a way that reflects the appraisal pattern underlying the in different emotional moods. Sample final activations of specific emotion in a situation. In a similar vein, Pfeiffer, DEBECO. The example represents simulation runs of ap- Manz and Nerb (in press) report two experiments that con- praising an negative event in which it is ambiguous who firm the coherence of cognitive and emotional reactions im- is responsible for the negative event. The global emotional mediately after reading a newspaper report and in a recog- moods, SADNESS and ANGER, influence the specific emo- nition paradigm one hour after the reading. DEBECO gives tions for the situation, sadness and anger, and the appraisal a rigorous account of such “construal” effects that are often criteria. In sad mood, situational forces are seen as respon- documented in social psychology. The DEBECO architec- sible for the event, whereas in angry mood, personal forces ture explains such dynamic affective coherence phenomena are considered as the cause. in the evaluation of one event by using bidirectional links for the cognition-emotion relationship. The empirical validity of this coherence bias strongly corroborates the assumption Coherence effects between emotions and appraisals of a bidirectional appraisal-emotion connection. Effects across situations. A prior or concurrent emotional episode may not only influence the emotional reaction but Concluding remarks also the appraisal of a current event. In DEBECO, this Architectures for modeling emotions should not only be able “transfer” effect is again mediated by the global emotion to recognize and express emotions in an appropriate and psy- nodes. Using appraisal theory as a framework for deriv- chologically plausible way, they should also account for how ing hypotheses, the empirical validity of such effects from emotions get influenced and influence reasoning processes. emotion to appraisals was first shown by Keltner, Ellsworth The abundance of psychological evidence that cognitions in- and Edwards (1993). The authors induced anger or sadness fluence and get influenced by emotions (e.g., Bower, 1981; in their participants and assessed causal attributions for an 2000; Isen, 2000; Keltner et al., 1993; Nerb & Spada, 2001) ambiguous event in a subsequent, unrelated task. As pre- calls for viable accounts for integrating models of cognitions dicted, they found emotional “carry-over effects” consistent with theories of emotions. DEBECO attempts to do so by with the underlying appraisal patterns of anger and sadness. incorporating appraisal theories of emotions within the ar- In a series of experiments, sad participants perceived situa- chitecture of a reasoning system. Appraisal theories present tional forces to be more responsible for an ambiguous event a rich and often unexploited source of hypotheses about the than did angry participants. relationship between cognitive appraisals and emotional re- Lerner and Keltner (2000, 2001) extended this line of re- actions (Brave & Nass, 2000; Forgas, 1995). search and developed a more general framework for predict- DEBECO is based upon three coherence principles: belief ing the influences of specific emotional states and disposi- coherence, desire coherence and belief-desire coherence. As tions on judgment and choice. The authors showed that a shown, these coherence mechanisms allow two elementary lingering emotion activates a predisposition to appraise fu- types of goals, approach and avoidance goals, to be repre- ture events in line with the central appraisal dimensions that sented. Furthermore, we demonstrated how discrete emo- triggered the emotion originally. Lerner and Keltner call this tions such as anger, pride, sadness, shame, surprise, relief predisposition appraisal-tendency. The bidirectional rela- and disappointment emerge out of the very same principles. tion between global emotions and the bidirectional relation DEBECO differentiates between three major sources for between emotion and appraisal make predictions that are the interaction of emotion and cognition. First, it assumes a fully in accordance with the appraisal tendency framework. continuously operating appraisal process for monitoring the In DEBECO, activation from a global emotion node such conduciveness of newly acquired beliefs of a person with as anger or sadness spreads to a situation-specific emotion pre-existing beliefs and desires. This basic appraisal process node and from there to situation-specific appraisals. has a tendency to bring beliefs in line with desires (wishful thinking), and a tendency to bring desires in line with beliefs Miceli M., Castelfranchi C. 2000. Nature and mechanisms (rationalization). Second, we showed how important dimen- of loss of motivation. Review of General Psychology, 4, sions of emotion-antecedent appraisal criteria evolve out of 238-263. DEBECO—namely goal-conduciveness, intrinsic pleasant- Nerb, J., & Spada, H. 2001. Evaluation of environmental ness (desire), belief, novelty and agency. The relationship problems: A coherence model of cognition and emotion. between these appraisal criteria and emotions is bidirec- Cognition and Emotion, 15, 521-551. tional and explained interaction effects among these criteria Ortony, A., Clore, G. L., & Collins, A. 1988. The cognitive and discrete emotions. Third, we demonstrated how posi- structure of emotions. Cambridge: Cambridge Univer- tive and negative moods as well as emotionally toned moods sity Press. evolve from this framework. Lingering emotions or moods Pfeiffer,T., Manz, S. & Nerb, J. in press. Wer den allow for carry-over effects from one emotional episode to Schaden macht, hat auch das Wissen: Koharenzeffekte¨ another. der kognitiven und emotionalen Bewertung von Umweltschadensfallen.¨ [The one who causes the dam- Acknowledgments age, knows what he is doing: Coherence effects within This research was supported by a fellowship from the cognitive and emotional evaluation of environmental Alexander von Humboldt-Stiftung. I am grateful to Paul damages] Zeitschrift fur¨ Psychologie. Thagard for providing the source code of HOTCO upon Reisenzein, R. 2000. Exploring the strength of association which the implementation of DEBECO is based. between the components of emotion syndromes: The case of surprise. Cognition and Emotion, 14, 1-38. References Roseman, I. J., Antoniou, A. A., & Jose, P. E. 1996. Ap- Bower, G.H. 1981. Mood and memory. American Psychol- praisal determinants of emotions: Constructing a more ogist, 36, 129-148. accurate and comprehensive theory. Cognition and Emo- Brave, S., & Nass, C. 2002. Emotion in human-computer in- tion, 10, 241-277. teraction. In J. A. Jacko & A. Sears (Eds.), The human- Rosenberg, M.J., & Abelson, R. P. 1960. An analysis of computer interaction handbook: Fundamentals, evolv- cognitive balancing. In M. J. Rosenberg, C. I. Hovland, ing technologies and emerging applications (pp. 251- W. J. McGuire, R. P. Abelson, & J. W. Brehm (Eds.), At- 271). Mahwah: Erlbaum. titude organization and change: An analysis of consis- Ellsworth, P. C., & Scherer, K R. 2003. Appraisal processes tency among attitude components (pp. 112-163). New in emotion. In R. J. Davidson, H. Goldsmith, K. R. Haven, CT: Yale University Press. Scherer (Eds.). Handbook of the affective sciences (pp. Rottenstreich, Y., & Hsee, C. K. 2001. Money, kisses and 572-595). New York: Oxford University Press. electric shocks: On the affective psychology of proba- Forgas, J. P. 1995. Mood and judgement: The affect infusion bility weighting. Psychological Science, 12, 185-190 model (AIM). Psychological Bulletin, 117, 39-66. Slovic, P., Finucane, M., Peters, E. & MacGregor, D.G. Heider, F. 1958. The psychology of interpersonal relations. 2002. The affect heuristic. In T. D. Gilovich, D. W. Grif- New York: John Wiley. fin, & D. Kahneman (Eds.) Heuristics and biases: The Holyoak, K. J., & Simon, D. 1999. Bidirectional reasoning psychology of intuitive judgment (pp. 397-420). Cam- in decision making by constraint satisfaction. Journal of bridge: Cambridge University Press. Experimental Psychology: General, 128, 3-31. Smith, C. A., & Ellsworth, P. C. 1985. Patterns of cognitive Isen, A. M. 2000. Positive affect and decision making. In M. appraisal in emotion. Journal of Personality and Social Lewis & J. Haviland-Jones (Eds.), Handbook of emotion Psychology, 48, 813-838. (2nd ed., pp. 417-435). New York: Guilford. Smith, C. A., & Kirby, L. D. 2000. Consequences require Keltner, D., Ellsworth, P. C., & Edwards, K. 1993. Beyond antecedents: Toward a process model of emotion elic- simple pessimism: Effects of sadness and anger on so- itation. In J. Forgas (Ed.). Feeling and thinking: The cial perception. Journal of Personality & Social Psy- role of affect in social cognition. New York: Cambridge chology, 64, 740-752. University Press. Kunda, Z., & Thagard, P. 1996. Forming impressions from Thagard, P. 1989. Explanatory coherence. Behavioral and stereotypes, traits, and behaviors: A parallel constraint Brain Sciences, 12, 435-467. satisfaction theory. Psychological Review, 103, 284-308. Thagard, P. 2000. Coherence in thought and action. Cam- Lerner, J.S., & Keltner, D. 2000. Beyond valence: Toward a bridge, MA: MIT Press. model of emotion-specific influences on judgement and Thagard, P. 2003. Why wasn’t O. J. convicted? Emotional choice. Cognition and Emotion, 14, 473–493. coherence in legal inference. Cognition and Emotion, Lerner, J.S., & Keltner, D. 2001. Fear, anger, and risk. Jour- 17, 361-383. nal of Personality and Social Psychology, 81, 146-159. Thagard, P., & Nerb, J. 2002. Emotional gestalts: Appraisal, McClelland, J. L., & Rumelhart, D. E. 1988. Explorations in change, and emotional coherence. Personality and So- parallel distributed processing: A handbook of models, cial Psychology Review, 6, 274-282. programs, and exercises. Cambridge, MA: MIT Press. Vallacher, R. R., Read, S. J., & Nowak, A. 2002. The dy- McGuire, W. J. 1999. Constructing social psychology: Cre- namical perspective in personality and social psychol- ative and critical processes. New York, NY, USA: Cam- ogy. Personality & Social Psychology Review, 6, 264- bridge University Press. 273.