To go directly a particular section of this paper, click on a section title below.
|1. Three Ways of Making Sense|
|2. Coherence as Constraint Satisfaction|
|3. Impression Formation as Coherence among Concepts|
|4. Attribution as Explanatory Coherence|
|5. Using One Person to Understand Another via Analogical Coherence|
|6. The Generation of Elements and Constraints|
|7. Automatic and Controlled Processes in Coherence|
All three of these ways of understanding people can be applied to oneself as well as to others. I may gain insight into myself by applying new concepts to myself (e.g., realizing I am impatient), forming new hypotheses about myself (e.g., conjecturing that I may be more upset by a setback than I realized), and by seeing myself as similar to others (e.g., noticing that I am acting just like my father did).
We propose that making sense of people through information integration, explanation, and analogy can all be understood in terms of cognitive mechanisms for maximizing coherence. When integrating information about a person, we attempt to achieve coherence among concepts by reconciling conflicts among the different pieces of information that we have about an individual (Kunda & Thagard, 1996). Knowing that someone who is a lawyer responded meekly to an insult requires us to balance conflicting expectations generated by the stereotype that lawyers are aggressive and the unaggressive behavior (Kunda, Sinclair, & Griffin, in press). Similarly, in causal attribution, we need to reconcile different explanations for an individual's behavior, choosing, for example, between explanations in terms of personality traits and explanations in terms of situational factors. Such choices require us to maximize explanatory coherence, accepting those explanations that fit best with the rest of our beliefs (Read & Miller, 1993; Thagard, 1989). Finally, making sense of people in terms of other similar people requires us to assess analogical coherence, finding a good fit between the complex of attributes of one person and the complex of attributes of another (Holyoak & Thagard, 1995).
In the next section, we first outline a general characterization of coherence that provides a uniform vocabulary for understanding a wide variety of cognitive processes in terms of parallel constraint satisfaction. We then show in more detail how information integration, causal attribution, and analogical understanding (including empathy) can be understood as different kinds of coherence. We argue that connectionist models of conceptual, explanatory, and analogical coherence provide computationally powerful and psychologically plausible explanations of diverse ways in which people think about other people and themselves.
Thagard and Verbeurgt (1996) have provided a general definition of coherence problems. A coherence problem arises when one encounters a set of elements that mutually constrain each other, and wishes to accept some of these elements and reject the remaining ones. For example, one needs to decide which of a set of inter-related traits are characteristic of John (the accepted elements) and which are not (the rejected elements). The constraints among the elements may be positive or negative. A positive constraint among two elements means that the two should go together--they should both be accepted or both rejected. For example, if John is loving he should be kind as well, and if he is not loving he should not be kind either. A negative constraint means that the two elements should not go together--if one is accepted the other should be rejected. For example, if John is loving he should not be hateful; if he is hateful he should not be loving. Each of the constraints carries a weight that reflects its importance.
When partitioning the elements into the accepted set and the rejected set, it is often not possible to satisfy all of the constraints because they may conflict with each other. For example, a person who is manipulative should also be interpersonally skilled. But a person who is interpersonally skilled should also be loving whereas a person who is manipulative should not. It will be impossible to satisfy all of these constraints simultaneously. The coherence problem is to satisfy as many of the constraints as possible, while giving preference to the more important ones. More technically, the aim is to partition the elements into an accepted and rejected set in a way that maximizes the weight of the satisfied constraints. For a more precise definition, see the appendix.
In later sections, we will show how conceptual, explanatory, and analogical coherence can all be understood as special cases of this general characterization of coherence. Each kind of coherence involves different sorts of elements and constraints.
Maximizing coherence is a difficult computational problem: Verbeurgt (1996) has proved that it belongs to a class of problems generally considered to be computationally intractable, so that no algorithms are available that are both efficient and guaranteed to be correct. Nevertheless, good approximation algorithms are available, in particular connectionist algorithms from which the above characterization of coherence was originally abstracted.
Here is how to translate a coherence problem into a problem that can be solved in a connectionist network:
1. Each element is represented as a unit (node) in a network of units. These units are very roughly analogous to neurons or groups of neurons in the brain.This process results in a partition of elements into accepted and rejected sets by virtue of the network settling in such a way that some units end up with activation levels that are above the critical threshold for acceptance, and others do not. The final levels of activation can also be taken to represent degrees of acceptance and rejection.
2. A positive constraint between two elements is represented as an excitatory link between the corresponding units. Each link has a weight representing the strength of the constraint, as determined, for example, by the strength of association between two concepts.
3. A negative constraint between two elements is represented as an inhibitory link between the corresponding units.
4. Each unit is assigned an equal initial activation (say .01). The activation of all the units is then updated in parallel. The updated activation of a unit is calculated on the basis of its current activation, the activation of the units to which it is linked, and the weights of these links. The activation of a given unit is increased with the activation of units to which it has excitatory links, and decreased with the activation of units to which it has inhibitory links. A number of equations are available for specifying how this updating is done (McClelland and Rumelhart 1989). Typically, activation is constrained to remain between a minimum (e.g. -1) and a maximum (e.g. +1).
5. The network goes through many cycles in which the activation of all units is updated. Updating is repeated until all units have settled, that is, have achieved stable activation values that change only minimally from one cycle to another.
6. If a unit's final activation exceeds a specified threshold (e.g. 0), then the element represented by that unit is deemed to be accepted. Otherwise, that element is rejected.
Intuitively, this solution is a natural one for coherence problems. Just as we want two coherent elements to be accepted or rejected together, so two units connected by an excitatory link will be activated or deactivated together. Just as we want the outcome for two incoherent elements to be such that one is accepted and the other is rejected, so two units connected by an inhibitory link will tend to suppress each other's activation, with one activated and the other deactivated. A solution that enforces positive and negative constraints on maximizing coherence is provided by the parallel update algorithm that adjusts the activation of all units at once based on their links and previous activation values. Table 1 summarizes the correspondences between coherence problems and connectionist networks.
|positive constraint||excitatory link|
|negative constraint||inhibitory link|
|constraint satisfaction||parallel updating of activation|
|element accepted||unit activated above threshold|
|element rejected||unit activated below threshold|
Table 1. Comparison of coherence problems and connectionist networks.
Kunda and Thagard (1996) have developed a parallel constraint-satisfaction theory of impression formation. This theory assumes that stereotypes, traits, and behaviors can be represented as interconnected nodes in a spreading activation network. The nodes can have positive, excitatory associations or negative, inhibitory ones. To illustrate the model, consider the well-documented finding that stereotypes can affect the meaning of behavior. For example, when a Black person pushes someone, this is interpreted as violent push. But when a White person performs the identical behavior, this is interpreted as a jovial shove (Sagar & Schofield, 1980) .
Figure 1. Stereotypes affect the meaning of behavior (P1). The network on the left activates "violent push" and deactivates "jovial shove." The network on the right does the opposite. Reprinted from Kunda and Thagard (1996).Figure 1 shows part of the network of concepts that would be used to make sense of the observation that a Black person or a White person pushed someone. The boxes depict the nodes representing the concepts. The lines connecting these nodes depict the associations among them. Bold lines indicate excitatory associations, and thin lines indicate inhibitory ones. Each of the concepts depicted also has many additional associates that are not portrayed in the figure. The observed information, in this case the behavior (pushed someone) and the stereotyped category (Black or White), is connected to a node termed observed to indicate its special status, and to distinguish it from inferred knowledge--in this case, the traits associated with the stereotype and the possible interpretations of the behavior. Observed concepts receive strong activation from the observed node. Inferred knowledge becomes activated or deactivated through its positive or negative associations with the observed information. These associations are based on perceivers' prior beliefs about the interrelationships among characteristics.
When one observes that a person pushed someone, pushed someone activates both violent push and jovial shove. If one also observes that the pusher is Black then, at the same time, Black activates aggressive which further activates violent push while deactivating jovial shove. If, on the other hand, one observes that the pusher is White, White does not activate aggressive. Therefore, both aggressive and violent push end up with less activation when the pusher is White than when the pusher is Black. In this manner, stereotypes color one's understanding of a person's behavior and one's impression of that person.
As this example illustrates, the identical behavior may be interpreted differently in different contexts (Kunda & Sherman-Williams, 1993; Sagar & Schofield, 1980; Wojciszke, 1994 ). Similar shifts in meaning from one context to another have been demonstrated for all the major ingredients of impression formation. These include traits (e.g., Asch, 1946; Asch & Zukier, 1984; Hamilton & Zanna, 1974; Kunda, Sinclair, & Griffin, in press; Zanna & Hamilton, 1977) , stereotypes (e.g., Deaux & Lewis, 1984; Kunda, Miller, & Claire, 1990; ) , facial expressions (Trope, 1986), and self-conceptions (e.g., Sanitioso, Kunda, & Fong, 1990). Thus there is broad support for the notion that the meaning of social constructs varies from one occasion to another.
Traditional models of representation cannot readily account for such shifts in meaning. Leading models of social cognition conceptualize representations as schemas, which are typically understood in terms of a filing cabinet or a storage bin metaphor (e.g., Wyer & Srull, 1986). In such models, each social construct has a fixed and discrete meaning which may be accessed independently, much like one might pull out a single file from a filing cabinet without affecting any of its neighbors (cf. Kunda et al., in press; Smith, 1996; in press). In contrast, the parallel-constraint-satisfaction model of impression formation assumes that there are no discrete, independent representations of constructs, and the meaning of each construct is not defined in the net but, rather, arises from its pattern of associations with other constructs (c.f. Kintsch, 1988). At any time only a subset of a construct's associates are activated, and these constitute its meaning at that time. Thus the notion that social constructs vary in meaning from one occasion to another, which conflicts with the assumptions underlying traditional, schema-based models of representation, constitutes a core assumption of the parallel-constraint-satisfaction model.
The parallel constraint satisfaction model of impression formation assumes that a coherent impression of the person is achieved through parallel satisfaction of the constraints imposed by the many concepts applied to the person. This view of impression formation is quite different from the one advocated by earlier, serial models of impression formation (Brewer 1988; Fiske & Neuberg, 1990). These serial models assume that people first try to make sense of other people by applying stereotypes. They may then use individuating information such as traits and behaviors if they are strongly motivated to understand the person or if they cannot successfully categorize the person as belonging to any particular stereotype. Thus the serial models give special, dominating status to stereotypes. In contrast, the parallel constraint satisfaction model does not. It treats all kinds of information as equal in status, and assumes that their impact depends entirely on their patterns of associations with other pieces of information.
Using this parallel constraint satisfaction model, Kunda and Thagard (1996) were able to account for most of the phenomena emerging from the literature on how people form impressions of others based on stereotypes and individuating information. Their connectionist program, IMP, successfully simulated the results of experiments that demonstrated these phenomena.
The parallel constraint satisfaction model can readily account for several phenomena that are not easily accommodated within previous serial models (Brewer, 1988; Fiske & Neuberg, 1990). For example, Kunda, Sinclair, and Griffin (in press) found that the impact of stereotypes on impressions can depend on the perceiver's judgment task. In line with earlier findings (e.g., Locksley, Borgida, Brekke, & Hepburn, 1980), Kunda et al. found that the effects of stereotypes on trait ratings of an individual were undermined by the individual's behavior. Although construction workers are stereotyped as more aggressive than accountants, a construction worker and an accountant were viewed as equally unaggressive after having failed to react to an insult, an unaggressive behavior. But even though the stereotypes no longer affected trait ratings, they continued to influence predictions about the individual's behavior: The construction worker was still viewed as more likely than the accountant to engage in coarse aggressive behaviors such as punching and cursing.
The parallel constraint satisfaction model predicts such a pattern when the stereotypes are associated with additional traits that are not undermined by the target's behavior and so can continue to influence behavioral predictions. In this case, even though both targets came to be viewed as equally unaggressive, the construction worker continued to be viewed as a member of the working class, and the accountant as a member of the upper-middle class. Punching and cursing are positively associated with working class status but negatively associated with upper-middle-class status. Therefore, the working-class construction worker was viewed as more likely than the upper-middle-class accountant to punch and curse even though the two were viewed as equally unaggressive.
In this manner the parallel constraint satisfaction model can readily account for the differential effects of stereotypes on traits and on behavioral predictions as due to the pattern of associations (constraints) among traits, behaviors, and the different aspects of stereotypes. In contrast, these findings are problematic for the serial models (Brewer, 1988; Fiske & Neuberg, 1990) because these models provide no grounds for distinguishing among different judgment tasks.
The parallel constraint satisfaction model is also better able than the serial models to account for findings showing that newly encountered combinations of stereotypes (e.g., feminist and bank-teller) can jointly influence impressions (Kunda , Miller, & Claire, 1990). Such findings are problematic for the serial models because these assume that only a single stereotype dominates one's impressions at a time. In contrast, the parallel constraint satisfaction model assumes that stereotypes can be integrated with each other just like any other concepts are integrated, through satisfying the constraints imposed by the knowledge associated with them (cf. Miller & Read, 1991; Read & Miller, 1993).
The fact that the connectionist program, IMP, could successfully simulate the diverse empirical findings on how stereotypes and individuating information influence impressions suggests that the parallel constraint satisfaction model of impression formation is computationally feasible and psychologically plausible.
It is important to note that this model accounts only for relatively automatic processes of impression formation, that is, processes carried out with little awareness, intention, or effort. It does not model more elaborate, controlled processes that require more effort, intention, and awareness. Thus this model can account for the relatively automatic processes that take place in the early stages of the attribution process--the identification of behavior (e.g., as a friendly act) and the characterization of the person (e.g., as a friendly person) (Gilbert, 1989; Trope, 1986). But it does not address the more effortful processes that sometimes take place later in the attribution sequence, in which early impressions are corrected by taking situational constraints into account (e.g., the person is on a job interview and so is trying to appear friendly). Similarly, the model explains how racial stereotypes can color initial impressions of members of minority groups, but does not address the processes through which people who are not prejudiced can subsequently attempt to eliminate these automatic influences of stereotypes from their judgments (Devine, 1989). Such higher order reasoning is captured by models of explanatory coherence, which we discuss next.
Note also that this theory of impression formation does not explain how new concepts are formed, only how existing concepts are applied. However, the need for new concepts may be signaled by lack of coherence, as we will discuss in section 6 below.
In sum, we view the automatic aspects of impression formation as resulting from a process of parallel constraint satisfaction in which one understands people by applying to them a set of concepts in a way that maximizes coherence.
Coherence-based explanatory inferences require specification of the positive and negative constraints among the propositional elements. The main source of positive constraints is explanation: if one proposition explains another, then there is a positive constraint between them. Such constraints can operate at many levels, since we can generate hypotheses that explain other hypotheses as well as hypotheses that explain observed behavior. You may hypothesize that your friend screamed at you because of a stressful day at work, and further hypothesize that the stressful day was caused by impending layoffs. The result can be a network of propositions of the sort shown in figure 2. This shows different hypotheses competing to explain the evidence. Positive constraints can be affected by considerations of simplicity: in Thagard's theory, if a number of hypotheses are required to make an explanation, then the positive constraints between hypotheses and evidence are weakened. For example, if you explain Mary's behavior by supposing that she was abducted by aliens who mistreated her, you are making a number of hypotheses whose coherence may suffer as a result of lack of simplicity as well as incompatibility with other things that you believe.
In explanatory coherence, the sources of negative constraints are contradiction and competition. If two propositions logically contradict each other (Mary is in Florida vs. Mary is in Toronto), then there is a strong negative constraint between them. Moreover, in explanatory situations, people tend to treat hypotheses as negatively constraining each other even if they are not strictly contradictory. It is possible that Mary's behavior should be explained because she had a stressful day and she stopped taking her medication and she found out something about you, but normally we treat these as independent competing explanations. Explanatory coherence can also be used to assess hypotheses about oneself, as when Mary herself figures out that she screamed because of some previously suppressed hostility.
Figure 2. Explanatory coherence network. Positive associations are shown as thick lines and negative associations are shown by thin lines. The evidence that Mary screamed can be explained by three competing hypotheses.
Thagard (1989, 1992a, 1992b) presented a set of principles for explanatory coherence. Explanatory coherence is a symmetric relation. A hypothesis coheres with what it explains, and the more hypotheses it takes to explain something, the less the degree of coherence. Similar hypotheses that explain similar pieces of evidence cohere. Contradictory and competitive propositions are incoherent with each other. Propositions that describe the results of observations have a degree of acceptability of their own, and the acceptability of a proposition in a system of propositions depends on its coherence with them.
Explanatory coherence has been modeled using a connectionist program called ECHO which has been applied to many cases of scientific theory evaluation (Thagard, 1989, 1991, 1992b). 1992a; Nowak and Thagard, 1992a, 1992b). It has also been used to model belief revision in the context of science education (Ranney and Thagard 1988, Schank and Ranney, 1991, 1992). More social applications include adversarial problem solving in which people have to make inferences about the hostile intentions and possible deceptions of their competitors in games, business, and international relations (Thagard, 1992a). ECHO has also been used to model a variety of attributions concerning interpersonal relations. Read and Marcus-Newhall (1993) demonstrated empirically that when people evaluate explanations of social behavior such as "Joanne agrees to marry Bill if they elope immediately", they follow the principles of coherence outlined by Thagard (1989). Further, ECHO successfully simulated these empirical results. It appears then that people make explanatory inferences about others in a manner that maximize coherence.
Processes of maximizing explanatory coherence are particularly well suited for accounting for jury decision making, where the task is to evaluate the coherence of accounts presented by the prosecution and the defense (cf. Read & Miller, 1993; Pennington & Hastie, 1992; Byrne, 1995). Such processes may also capture the controlled inferences often required for choosing among dispositional and situational attributions for behavior (Gilbert, 1989).
Viewing causal attribution as an attempt to maximize explanatory coherence highlights questions that have received little attention by attribution researchers. Much of the large literature on attribution has focused on how people choose between dispositional and situational explanations (Jones, 1990). But there has been little discussion of how people may choose among competing dispositional attributions (is this person friendly or ingratiating?) or among competing situational attributions (is this person driven by pressure at work or by tension at home?) (Kunda, in press). And this tradition has left no room for attributions that incorporate dispositional as well as situational explanations (is this the kind of person who collapses under pressure?) (Mischel & Shoda, 1995; Shoda & Mischel, 1993). Yet many of our attempts to make sense of others are of precisely this nature. The work on explanatory coherence provides a language and tools for exploring these important kinds of judgment (cf. Miller & Read, 1991; Read & Miller, 1993).
Just as conceptual coherence does not explain concept formation, explanatory coherence does not explain hypothesis formation. See section 6 below for discussion of how hypothesis formation is related to coherence.
Such analogical mapping can be viewed as a coherence process that maximizes the satisfaction of multiple constraints (Holyoak and Thagard, 1995). The elements are hypotheses about what corresponds to what, for example that Diana corresponds to Anna and that loves in Diana's case corresponds to loves in Anna's case. One constraint is perceptual and semantic similarity: two elements will tend to correspond to each other if they look the same or have similar meaning. Other constraints are structural: in order to map Anna loves Vronsky to Diana loves James we must consistently map Ann to Diana, loves to loves, and Vronsky to James. Mappings should tend to be one-to-one; without strong reason, we should not map Diana to both Anna and Vronsky. Finally, purpose provides a practical constraint on the mapping, since we should try to come up with mappings that will contributed to the cognitive goals that the analogy is supposed to serve, such as providing an explanation or contributing to a decision.
Finding an appropriate mapping between complex analogs is a computationally difficult problem that can be solved using connectionist models (Holyoak and Thagard, 1989) as well as non-connectionist methods (Falkenhainer, Forbus, and Gentner, 1989). Here we omit the computational details, but emphasize that analogical understanding, like impression formation and attribution, can be understood as a process of maximizing coherence. To map between two analogous situations involving sets of interrelated people, we must come up with a coherent set of correspondences between the various people and aspects of the situation. Similarly, when our task is to retrieve from memory a person or situation similar to one we want to understand, we must search for one that has a highly coherent set of correspondences.
Making sense of people analogically always involves comparing two individuals, a target to be understood and a source that provides understanding. In the Princess Diana example, the source and target are both other people, but sometimes the source is oneself and the target is another (e.g. empathy), sometimes the target is oneself and the source is another (e.g. some kinds of social comparison), and sometimes both the source and target are oneself, as when a past situation of one's life is used to make sense of a current situation.
Several diverse lines of social psychological research assume, explicitly or implicitly, that people use analogy to make sense of others, that is, they understand individuals by mapping them onto other individuals. Newly encountered individuals are often understood in terms of more familiar others (Smith & Zarate, 1992). For example, people who resembled a significant other (parent, close friend) on some dimensions were falsely recalled as also resembling that person on other dimensions as well (Andersen, Glassmann, Chen, & Cole, 1995). Significant others may be used spontaneously as sources for understanding others because their representations are highly accessible (Andersen et al., 1995). Other individuals may be used as sources of analogical mapping when one is reminded of them. For example, Gilovich (1981) showed that fictitious college football players were understood in terms of famous players when participants were reminded of the famous player, and when that player could be readily mapped onto the fictitious one (as when both played the same position). When mapping was facilitated in this manner, the fictitious players were rated more highly. It appears then that we often understand strangers by mapping them onto people we know well.
We may also understand others by mapping them onto ourselves. When trying to assess others' attitudes and behaviors, we may do so by assuming that they resemble our own. This may be one source of the well-documented false consensus effect, wherein peoples' estimates of the prevalence of various responses in the population are correlated with their own responses. For example, optimists assume that optimism is more common than do pessimists (Ross, Green, & House, 1977). False consensus effects are exacerbated when the target population is more similar to the self on various demographic dimensions (Marks & Miller, 1987). This may occur because such similarity facilitates mapping the other onto the self.
A particularly important kind of mapping from another to oneself is empathy, in which I establish a correspondence not only between someone else's situation and my own, but also a correspondence between the other's emotional state and an emotional experience of my own. Deep understanding of people's work stress requires not just seeing how their situation corresponds to one that I have been in (unpleasant boss, risk of layoff, etc.) but also appreciating their emotional state (anger, fear). In a purely verbal analogy, I may infer that just as I was angry in my own situation, so the other is likely to be angry in a similar situation. But empathy goes beyond verbal elements by providing a correspondence between some emotional experience of my own and what I can infer analogically to be the emotional experience of the other. By setting up an analogy between another person and myself, I can feel an approximation to what the other feels. Such an analogy should be facilitated if I myself have been in a similar situation. Indeed, Batson et al. (1996) found that women felt greater empathy for someone undergoing a difficult experience if they themselves had a similar experience (though the same was not true for men). Barnes and Thagard (in press) have an extended discussion of empathy as analogy. Like other kinds of analogy, empathy can be understood as a coherence mechanism that evaluates a set of correspondences between two people and their situations; empathy differs from other analogies in that the correspondences link representations that are not verbal or visual, but emotional.
Just as the self can serve as a source for understanding others, other people can be used to enrich one's understanding of oneself. An extensive literature on social comparison suggests that people often attempt to evaluate their own abilities and performance by comparing them to those of others. People are particularly likely to seek comparisons with others who are similar to them in various ways (Wood, 1989). Similarly, one's self-views are particularly likely to be affected by exposure to superior others when these superior others are similar to the self. Tesser and his colleagues (Tesser, 1986; Tesser & Campbell, 1983) reasoned that self-evaluations can be threatened when one is outperformed by others. One then engages in thoughts and actions designed to reduce the threat. Such protective action is especially likely to occur when the other person is similar to the self on dimensions such as age, race, gender, or personality. In other circumstances, one can be inspired by the performance of outstanding others, and view oneself more favorably after exposure to them. This too is more likely to occur when the outstanding other resembles the self. For example, future accountants were inspired by reading about an outstanding accountant, but were unaffected by reading about an outstanding teacher. Future teachers, in contrast, were inspired by the outstanding teacher but were unaffected by the accountant (Lockwood & Kunda, 1996). It appears then that we often make sense of ourselves through comparison to others, and we find others most informative about ourselves when we can readily map the other onto the self. Such mapping is facilitated by similarity between the self and the other.
Thus there is ample evidence that people often make sense of the self and others through analogy to others. And they are more likely to do so when they can construct more coherent analogies among individuals. Similarity between two individuals increases the likelihood that one will be used to make sense of the other. But what determines similarity?
It is important to note that relevant social psychological research has, for the most part, examined the impact of only one kind of constraint on analogical coherence--surface similarity, that is, relatively superficial similarity that is based only on the number of shared features. This research has shown that an individual is more likely to be compared to another when the two are similar in their performance on a given dimension, or when they share one or more attributes. Such surface similarity is indeed an important contributor to the coherence of analogies. But it is not the only one, and it can even be superseded by deeper structural and relational similarity, that is, similarity that is based on the underlying patterns of relations among elements. Structural similarity can lend coherence to analogies among sets of elements even if they are superficially very different from each other (Holyoak & Thagard, 1995). For example, people presented with two pictures, one of a woman receiving a delivery from a food bank employee, and another of a physically similar woman feeding a squirrel will, upon brief reflection, map the woman in the first picture onto the squirrel in the second picture rather than onto the superficially similar woman. The mapping of the first woman onto the squirrel is more coherent because it involves greater correspondence in underlying relations among the objects and characters in each picture, that is, both the woman in the first picture and the squirrel in the second are recipients of food (Markman & Gentner, 1993).
Existing social psychological research cannot speak to the importance of such structural coherence in facilitating analogies among people because it has focused on examining similarity on singular dimensions. This focus has left no room for detecting the operation of deeper structural similarity. It is possible, therefore, that the conclusion that attribute similarity among individuals will determine whether and how they are used to make sense of one another is overly strong. Intuitively, it seems that we often compare ourselves and others to superficially dissimilar individuals who differ from us in background, nationality, even gender. Surely, one need not be Indian and male to be inspired by Mahatma Ghandi. Moreover, we can gain important insight into ourselves and others through analogy to the animals in Aesop's fables, and can be inspired by The Little Engine That Could. People or things that are highly dissimilar to us in their attributes may nevertheless be similar in their purposes and in the structures of their lives, so that they can be highly informative about ourselves and others. Applying this more complex view of analogical coherence to analogies among individuals may enrich the understanding of the circumstances that lead us to compare one individual to another.
To summarize our discussion so far, coherence theory, which views inference as maximizing satisfaction of positive and negative constraints, provides a general way of understanding how people make sense of each other. (1) Table 2 shows how conceptual, explanatory, and analogical coherence can all be viewed as instances of a more broadly defined coherence problem.
|conceptual coherence||explanatory coherence||analogical coherence|
|positive constraints||positive associations||explanation||similarity, purpose, structure|
|negative constraints||negative associations||contradiction, competition||one-to-one mappings|
Table 2. Three kinds of coherence.Two major problems remain. First, where do the elements and constraints come from? Second, how are the different kinds of constraints related to each other. We will address these questions in the next two sections.
Generation of new elements is sometimes driven by incoherence. If I am trying to understand someone but fail to form a coherent impression or attribution, I may be spurred to form new elements that can add coherence to the old set of elements. To take an example from Kunda, Miller, and Claire (1990), if I am told that someone is a Harvard-educated carpenter, it may be difficult to reconcile the conflicting expectations associated with the two concepts. Surprise is an emotional reaction that signals that a satisfactory degree of coherence has not been achieved. This reaction triggers hypothesis formation as I ask myself how someone with a Harvard degree could end up working as a carpenter. People show ingenuity in generating explanations, for example that the Harvard graduate was a counterculture type who preferred a non-professional career path. Hence new hypotheses and possibly also new concepts (the Ivy-league-laborer type) can be added to the set of elements so as to lend greater coherence to the attempt to make sense of this person. In this case, generation of elements is incoherence-driven, since it is prompted by a failure to achieve an interpretation that satisfies an adequate number of the positive and negative constraints. In addition to surprise, other emotions such as anxiety may signal incoherence,
Not all element generation is incoherence-driven, however. Some representations arise serendipitously, based on things we just happen to encounter. I may form the concept of Albanian as the result of meeting various immigrants from Albania, without having experienced any incoherence in my previous attempts to understand them. In other cases, new representations may arise from curiosity-driven thinking that is motivated not by any incoherence but by the desire to find out more about something that interests me. If I am interested in the Balkans, I will learn more about Serbs and Croats and may form stereotypes about them without having tried and failed to fit them with my other social concepts. Motivation may also lead one to generate new concepts. For example, our desire to protect our stereotypes from change in the face of disconfirmation may lead us assign individuals who threaten our stereotypes into novel subtypes that serve to isolate these individuals from their group (Kunda & Oleson, 1995; Weber & Crocker, 1983). And our desire to view ourselves positively may lead us to construct hypothetical individuals to whom we are superior (Taylor, Wood, & Lichtman, 1983). Thus serendipity, curiosity, and motivation, in addition to incoherence, can spur the generation of new representations .
Where do constraints come from? Some may be innate, capturing basic conceptual relations such as that an object cannot be both red and black all over. Most constraints, however, capture empirically discovered relations between elements. For impression formation, I learn that some concepts (e.g. nurse and benevolent) are positively associated, whereas other concepts (e.g. Nazi and benevolent) are negatively associated. Such associations may be learned through direct observation of nurses or Nazis as well as through cultural transmission. For attribution, the positive constraints come from understanding causal relations. The link between the hypothesis that Mary is in love and the fact to be explained that Mary is very happy depends on the causal judgment gleaned from experience that being in love can cause people to be happy. Negative constraints in explanatory coherence arise from logical contradictions (you cannot be both in love and not in love) and from competing hypotheses (maybe instead she's happy because she got a promotion at work).
Since any full account of human cognition will have to include an account of how new concepts, hypotheses, and other representations are formed, a complete cognitive architecture will have to include generation mechanisms as well as coherence mechanisms (see Thagard, in press, for a review of different kinds of learning). Our goal in this paper has not been to propose a cognitive architecture, but merely to show how coherence mechanisms contribute to making sense of people.
We believe that all three ways of making sense may operate in parallel. However, some aspects of impression formation take place automatically, that is, with little awareness, intention, or effort, whereas others may require more controlled, effortful processes. The process of integrating information modeled by IMP is assumed to take place automatically (Kunda & Thagard, 1996). Traits, behaviors, and stereotypes, as well as some aspects of the situation influence each other's meaning and jointly influence impressions in a relatively automatic manner. This view is supported by evidence that a person's behavior can provoke automatic trait inference (Gilbert, Pelham, & Krull, 1988; Winter & Uleman, 1984), that the situation can automatically color the meaning of emotional expressions (Trope, 1986; in press), and that stereotypes can automatically influence trait ratings (Devine, 1989) and affective reactions (Fazio, Jackson, Dunton, & Williams, 1995). It appears, then, that a great deal of impression formation arises from automatic integration of available information, without awareness or intention.
Some aspects of causal reasoning may also be relatively automatic. Some behaviors and traits may be automatically viewed as driven by particular causes (e.g., Susan is crying because Tom hit her; John is marrying Ellen because he loves her). Indeed, it has been suggested that underlying goals are often central to the meaning of traits and behaviors (Read, Jones, & Miller, 1990). However, many behaviors are not associated with any obvious causes (e.g., why did David quit his job?), and others are associated with many conflicting ones (Did Melissa lose interest in the lesson because she was too slow to follow it, or because she has long since mastered it?). Moreover, often the context offers alternative causes that compete with a normally strongly associated one (e.g., John is an illegal immigrant; did he marry Ellen because he loves her or because he wants to avoid deportation?) In such cases, when no plausible cause comes to mind, or when one is entertaining several plausible, competing causes and has trouble choosing among them, one is likely to call upon elaborate causal analysis, in a conscious and effortful attempt to determine which is the most likely cause. Thus controlled causal reasoning can be triggered by the emotional experience of surprise or confusion (Kunda et al., 1990). Note that the availability of multiple competing causes will not necessarily baffle us; often, the context will lead us to favor one cause and suppress the others through relatively automatic constraint satisfaction (see Read & Miller, 1993). But when we are puzzled or stumped in our attempts to make sense of others, we turn to controlled processes involving elaborate causal analysis.
Recent models of attribution point to such an interplay between automatic and controlled processes (Gilbert 1989; Trope, 1986). These models suggest that perceivers can automatically view behavior as caused by underlying personality traits (e.g., she acted nervously while talking, so she must be an anxious person) (Gilbert et al., 1988; Winter & Uleman, 1984). But when the context offers a competing situational explanation for the behavior (she was discussing an embarrassing topic, that's why she acted nervously), controlled processes are required for this information to be taken into account (Gilbert et al., 1988). Similarly, when the most compelling cause for a behavior is the situation, its causal role is inferred automatically, but controlled processes are required for the actor's personality to be taken into account (Krull, 1993). Thus, a single, strongly associated cause may be inferred automatically. But sorting among multiple competing causes can be more cognitively demanding and so may require more controlled processing.
The third way of making sense of people--analogy--seems likely to involve both automatic and controlled processes. Simple similarity mappings may take place automatically, whereas more complex relational mappings may require elaborate reasoning. There is some evidence that both types of processes may be implicated when making comparisons among individuals. When faced with another person in circumstances similar to their own, people tend to compare themselves automatically to this person; they use information suggesting that the comparison is logically inappropriate only when they have sufficient cognitive resources (Gilbert, Giesler, & Morris, 1995). This suggests that people can map one person onto another automatically, but the mapping can also be influenced, even undone, by more controlled processes. In a similar vein, it has been shown that people spontaneously use surface similarity to map one group of individuals onto another, that is, they base their mappings on the number of shared features among individuals. But, following brief reflection on the similarity between the two groupings, they use deep structural similarities instead, that is, they base their mappings on the underlying relations among sets of individuals (Markman & Gentner, 1993).
Our view of how these automatic and controlled processes interact with each other is similar to that proposed recently by Sloman (1996). Sloman argued that people use relatively automatic, associative processes as well as more controlled, rule-based processes. The two systems may operate simultaneously, with the rule-based system sometimes suppressing the outcomes of the associative one. We would add that the products of one way of making sense may feed into another. Based on the way we integrate information, we may conclude that a person behaved in an unfriendly manner. Our causal analysis may then lead us to conclude that the unfriendly behavior was due to stress, and that the person may well be friendly. This characterizing may then be integrated with our other knowledge of the person. Our resulting impression may lead us to map this person onto another person, and this mapping will trigger further inferences about the person's likely traits and attributes, to be integrated with previous knowledge. As this example illustrates, all three kinds of making sense are dynamically interrelated, and all can contribute to the understanding of a person.
We should note, though, that our attempts to outline how the different ways of making sense of people are interrelated remain speculative. Most empirical and theoretical work to date has focused on understanding each of these processes when examined alone. Thus there are well-developed models of information integration (Kunda & Thagard, 1996), of causal explanation (Read & Marcus-Newhall, 1993; Thagard, 1989), and of analogy (Holyoak & Thagard, 1989; 1995), and each is supported by empirical evidence that addresses its particular assumptions. But spelling out the interrelationships among these models remains a major challenge for future theorizing and research.
1. if (ei, ej) is in C+, then ei is in A if and only if ej is in A.
2. if (ei, ej) is in C-, then ei is in A if and only if ej is in R.
Let W be the weight of the partition, that is, the sum of the weights of the satisfied constraints. The coherence problem is then to partition E into A and R in a way that maximizes W. Because a coheres with b is a symmetric relation, the order of the elements in the constraints does not matter. The coherence problem is computationally intractable in that there is no efficient and exact way of solving it, but there are connectionist and other algorithms that provided excellent approximate solutions.
Asch, S. E. (1946). Forming impressions of personality. Journal of Abnormal and Social Psychology, 41, 303-314.
Asch, S. E., & Zukier, H. (1984). Thinking about persons. Journal of Personality and Social Psychology, 46, 1230-1240.
Barnes, A., & Thagard, P. (in press). Empathy and analogy. Dialogue.
Batson, C. D., Sympson, S. C., Hindman, J. L., Decruz, P., Todd, R. M., Weeks, J. L., Jennings, G., & Burris, C. T. (1996). "I've been there, too": Effect on empathy of prior experience with a need. Personality and Social Psychology Bulletin, 22, 474-482.
Byrne, M. D. (1995). The convergence of explanatory coherence and the story model: A case study in juror decision. In J. D. Moore & J. F. Lehman (Eds.), Proceedings of the seventeenth annual conference of the Cognitive Science Society (pp. 539-543). Mahwah, NJ: Erlbaum.
Deaux, K., & Lewis, L. L. (1984). Structure of gender stereotypes: Interrelationships among components and gender label. Journal of Personality and Social Psychology, 46, 991-1004.
Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and controlled components. Journal of Personality and Social Psychology, 56, 5-18.
Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995). Variability in automatic activation as an unobtrusive measure of racial attitudes: A bona fide pipeline? Journal of Personality and Social Psychology, 69, 1013-1027.
Gilbert, D. T. (1989). Thinking lightly about others: Automatic components of the social inference process. In J. A. Bargh & J. S. Uleman (Eds.), Unintended thought (pp. 189-211). New York: Guilford.
Gilbert, D. T., Giesler, R. B., & Morris, K. A. (1995). When comparisons arise. Journal of Personality and Social Psychology, 69, 227-236.
Gilbert, D. T., Pelham, B. W., & Krull, D. S. (1988). On cognitive busyness: When person perceivers meet persons perceived. Journal of Personality and Social Psychology, 54, 733-740.
Gilovich, T. (1981). Seeing the past in the present: The effect of associations to familiar events on judgments and decisions. Journal of Personality and Social Psychology, 40, 797-808.
Hamilton, D. L., & Zanna, M. P. (1974). Context effects in impression formation: Changes in connotative meaning. Journal of Personality and Social Psychology, 29, 649-654.
Holyoak, K. J., & Thagard, P. (1989). Analogical mapping by constraint satisfaction. Cognitive Science, 13, 295-355.
Holyoak, K. J., & Thagard, P. (1995). Mental leaps: Analogy in creative thought. Cambridge, MA: MIT Press/Bradford Books.
Jones, E. E. (1990). Interpersonal perception. New York: W. H. Freeman and Company.
Krull, D. S. (1993). Does the grist change the mill? The effect of the perceiver's inferential goal on the process of social inference. Personality and Social Psychology Bulletin, 19, 340-348.
Kunda, Z. (in press). Parallel processing in person perception: Implications for two-stage models of attribution. In J. M. Darley & J. Cooper (Eds.), Attribution processes, person perception, and social interaction: The legacy of Ned Jones Erlbaum.
Kunda, Z., Miller, D., & Claire, T. (1990a). Combining social concepts: The role of causal reasoning. Cognitive Science, 14, 551-577.
Kunda, Z., & Oleson, K. C. (1995). Maintaining stereotypes in the face of disconfirmation: Constructing grounds for subtyping deviants. Journal of Personality and Social Psychology, 68, 565-579.
Kunda, Z., & Sherman-Williams, B. (1993). Stereotypes and the construal of individuating information. Personality and Social Psychology Bulletin, 19, 90-99.
Kunda, Z., & Thagard, P. (1996). Forming impressions from stereotypes, traits, and behaviors: A parallel constraint satisfaction theory. Psychological Review, 103, 284-308.
Locksley, A., Borgida, E., Brekke, N., & Hepburn, C. (1980). Sex stereotypes and social judgment. Journal of Personality and Social Psychology, 39, 821-831.
Lockwood, P., & Kunda, Z. (1996). Superstars and me: Predicting the impact of role models on the self. Unpublished manuscript, University of Waterloo.
Markman, A. B., & Gentner, D. (1993). Structural alignment during similarity comparisons. Cognitive psychology, 25, 431-467.
Marks, G., & Miller, N. (1987). Ten years of research on the false-consensus effect: An empirical and theoretical review. Psychological Bulletin, 102, 72-90.
McClelland, J. L., & Rumelhart, D. E. (1989). Explorations in parallel distributed processing. Cambridge, MA: MIT Press.
Miller, L., & Read, S. (1991). On the coherence of mental models of persons and relationships. In F. Fincham & G. Fletcher (Eds.), Cognition in close relationships. (pp. 69-99). Hillsdale, NJ: Erlbaum.
Millgram, E., & Thagard, P. (forthcoming). Deliberative coherence. Synthese.
Mischel, W., & Shoda, Y. (1995). A cognitive-affective system theory of personality: Reconceptualizing situations, dispositions, dynamics, and the invariance in personality structure. Psychological Review, 102, 246-268.
Nowak, G., & Thagard, P. (1992a). Copernicus, Ptolemy, and explanatory coherence. In R. Giere (Eds.), Cognitive models of science, (pp. 274-309). Minneapolis: University of Minnesota Press.
Nowak, G., & Thagard, P. (1992b). Newton, Descartes, and explanatory coherence. In R. Duschl & H. R. (Eds.), Philosophy of Science, Cognitive Psychology and Educational Theory and Practice. (pp. 69-115). Albany: SUNY Press.
Pennington, N., & Hastie, R. (1992). Explaining the evidence: Tests of the story model for juror decision making. Journal of Personality and Social Psychology, 51, 189-206.
Ranney, M., & Thagard, P. (1988). Explanatory coherence and belief revision in naive physics. In Proceedings of the Tenth Annual Conference of the Cognitive Science Society (pp. 426-432). Hillsdale, NJ: Erlbaum.
Read, S. J., Jones, D. K., & Miller, L. C. (1990). Traits as goal-based categories: The importance of goals in the coherence of dispositions. Journal of Personality and Social Psychology, 58, 1048-1061.
Read, S. J., & Marcus-Newhall, A. (1993). Explanatory coherence in social explanations: A parallel distributed processing account. Journal of Personality and Social Psychology, 65, 429-447.
Read, S. J., & Miller, L. C. (1993a). Explanatory coherence in the construction of mental models of others. In Proceedings of the Fifteenth Annual Conference of the Cognitive Science Society (pp. 836-841). Hillsdale, NJ: Erlbaum.
Read, S. J., & Miller, L. C. (1993b). Rapist or "regular guy": Explanatory coherence in the construction of mental models of others. Personality and Social Psychology Bulletin, 19, 526-541.
Ross, L., Greene, D., & House, P. (1977). The "false consensus effect": An egocentric bias in social perception and attribution processes. Journal of Experimental Social Psychology, 13, 279-301.
Sagar, H. A., & Schofield, J. W. (1980). Racial and behavioral cues in black and white children's perceptions of ambiguously aggressive acts. Journal of Personality and Social Psychology, 39, 590-598.
Sanitioso, R., Kunda, Z., & Fong, G. T. (1990). Motivated recruitment of autobiographical memories. Journal of Personality and Social Psychology, 59, 229-241.
Schank, P., & Ranney, M. (1991). Modeling an experimental study of explanatory coherence. In Proceedings of the Thirteenth Annual Conference of the Cognitive Science Society (pp. 892-897). Hillsdale, NJ: Erlbaum.
Schank, P., & Ranney, M. (1992). Assessing explanatory coherence: A new method for integrating verbal data with models of on-line belief revision. In Proceedings of the Fourteenth Annual Conference of the Cognitive Science Society (pp. 599-604). Hillsdale, NJ: Erlbaum.
Shoda, Y., & Mischel, W. (1993). Cognitive social approach to dispositional inference: What if the perceiver is a cognitive social theorist? Personality and Social Psychology Bulletin, 19, 574-587.
Sloman, S. A. (1996). The empirical case for two systems of reasoning. Psychological Bulletin, 119, 3-22.
Smith, E. R. (1996). What do connectionism and social psychology offer each other? Journal of Personality and Social Psychology, 70.
Smith, E. R. (in press). Mental representations and memory. In D. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), Handbook of Social Psychology, 4th Edition New York: McGraw-Hill.
Smith, E. R., & Miller, F. D. (1983). Mediation among attributional inferences and comprehension processes: Initial findings and a general method. Journal of Personality and Social Psychology, 44, 492-505.
Smith, E. R., & Zarate, M. A. (1992). Exemplar-based model of social judgment. Psychological Review, 99, 3-21.
Tesser, A. (1986). Some effects of self-evaluation maintenance on cognition and action. In R. M. Sorrentino & E. T. Higgins (Eds.), The handbook of motivation and cognition: Foundations of Social Behavior (pp. 435-464). New York: Guilford Press.
Tesser, A., & Campbell, J. (1983). Self-definition and self-evaluation maintenance. In J. Suls & A. Greenwald (Eds.), Social psychological perspectives on the self (pp. 1-31). Hillsdale, N. J.: Erlbaum.
Thagard, P. (1989). Explanatory coherence. Behavioral and Brain Sciences , 12 , 435-467.
Thagard, P. (1991). The dinosaur debate: Explanatory coherence and the problem of competing hypotheses. In J. Pollock & R. Cummins (Eds.), Philosophy and AI: Essays at the Interface. (pp. 279-300). Cambridge, Mass.: MIT Press/Bradford Books.
Thagard, P. (1992a). Adversarial problem solving: Modelling an opponent using explanatory coherence. Cognitive Science, 16, 123-149.
Thagard, P. (1992b). Conceptual revolutions. Princeton: Princeton University Press.
Thagard, P. (in press). Mind: Introduction to cognitive science. Cambridge, MA: MIT Press.
Thagard, P., & Millgram, E. (1995). Inference to the best plan: A coherence theory of decision. In A. Ram & D. B. Leake (Eds.), Goal-driven learning: (pp. 439-454). Cambridge, MA: MIT Press.
Thagard, P., & Verbeurgt, K. (forthcoming). Coherence. Under review.
Trope, Y. (1986). Identification and inferential processes in dispositional attribution. Psychological Review, 93, 239-257.
Trope, Y. (In press). Dispositional bias in person perception: A hypothesis-testing perspective. In J. Cooper & J. M. Darley (Eds.), Attribution processes, person perception, and social interaction: The legacy of Ned Jones
Weber, R., & Crocker, J. (1983). Cognitive processes in the revision of stereotypic beliefs. Journal of Personality and Social Psychology, 45, 961-977.
Winter, L., & Uleman, J. S. (1984). When are social judgments made? Evidence for the spontaneousness of trait inferences. Journal of Personality and Social Psychology, 47, 237-252.
Wojciszke, B. (1994). Multiple meanings of behavior: Constructing actions in terms of competence or morality. Journal of Personality and Social Psychology, 67, 222-232.
Wyer, R. S., & Srull, T. K. (1986). Human cognition in its social context. Psychological Review, 93, 322-359.
Zanna, M. P., & Hamilton, D. L. (1977). Further evidence for meaning change in impression formation. Journal of Experimental Social Psychology, 13, 224-238.
(1) There are several other kinds of coherence than the three discussed in this paper, but they seem to have less relevance to making sense of other people. In deliberative coherence, actions and goals are evaluated to select plans (Thagard and Millgram, 1995; Millgram and Thagard, 1996). Understanding of other people's decisions may involve appreciation of their assessment of deliberative coherence. In deductive coherence, general principles such as mathematical axioms and ethical rules are evaluated in connection with their deductive implications such as theorems and particular ethical judgments. In visual coherence, interpretations of visual inputs are combined to produce coherent perceptions.