Philosophy Department
University of Waterloo
Waterloo, Ontario, N2L 3G1
pthagard@watarts.uwaterloo.ca
© Paul Thagard and Allison Barnes, 1997
To go directly a particular section of this paper, click on a section title below.
| 1. The Concept of Empathy |
| 2. Analogy as a Cognitive Process |
| 3. Empathy as Analogy |
| 4. The Elusiveness of Empathy |
| 5. Cognitive Architecture |
| 6. Simulation Theory versus Theory-Theory |
To return to the Coherence Articles Table of Contents page, click here.
Abstract
We contend that empathy is best viewed as a kind of analogical thinking of the sort described in the multiconstraint theory of analogy proposed by Keith Holyoak and Paul Thagard (1995). Our account of empathy reveals the Theory-theory/Simulation theory debate to be based on a false assumption and formulated in terms too simple to capture the nature of mental state ascription. Empathy is always simulation, but may simultaneously include theory-application. By properly specifying the analogical processes of empathy and their constraints, we are able to show how the amount of theory needed to empathize is determined.
The distinguished psychologist Robin Dawes (1994) recently surveyed the empirical evidence concerning the effectiveness of psychotherapy. He determined that therapy does help people, but that the training and approach of the therapist have no statistically significant influence on the success of the therapy. Length of therapy is also unrelated to success. Dawes concluded, however, that "empathic" therapists are more effective, although he did not provide an account of what constitutes empathy.
The difficulties that psychotherapists have met in explaining empathy are shared by philosophers. (1) Philosophical discussion of empathy largely takes place in the phenomenological tradition (e.g., Stein 1964, Woodruff-Smith 1989). Recently, however, Alvin Goldman has argued convincingly that appreciation of empathy is important for ethical questions such as altruism, and also for the epistemological problem of understanding other minds. Put to use in the other minds problem, Goldman's work on empathy is part of a lively debate on the nature of mental state ascription. (2) Goldman develops an account of empathy which is intended to deflate the well-established position (known as "theory-theory" or TT) that understanding other people is a matter of applying some "theory of mind" to them. According to TT, our ability to ascribe states to others reflects the fact that we possess a primitive theory of mind, often called a "folk psychology". The "simulation theory" (ST) which Goldman espouses as an alternative says that we habitually understand others' actions in the absence of any theory of mind, by using the resources of our own minds to simulate the beliefs and intentions of others. Goldman's simulation alternative claims that we understand others by simulating them in a way that produces empathy, but like the psychotherapists, he is not able to describe in sufficient detail the nature of empathy or of simulation.
Like Goldman, we believe that an accurate account of empathy can help solve philosophical questions about the nature of mental state ascription. We contend that empathy is best viewed as a kind of analogical thinking of the sort described in the multiconstraint theory of analogy proposed by Keith Holyoak and Paul Thagard (1995). Our account of empathy as analogy shows that the assumption that TT and ST are mutually exclusive alternatives is a false assumption. Empathy always involves simulation, but may simultaneously include theory application. By properly specifying the analogical processes of empathy and their constraints, we are able to show how the amount of theory needed to generate an accurate simulation is determined. As we shall see, empathy is independent of theory application when an analog of the other's mental state is easily retrieved from memory. Processes of rule-based reasoning are required when empathy is achieved by constructing an analog. The constraints of analogical mapping determine when an analog is likely to be retrieved and when it is likely to be constructed. We conclude that the debate between proponents of ST and TT is effectively transformed, since we can ascertain the balance of simulation and theory for any single act of empathy. Moreover, our account indicates how this balance can be determined in other sorts of mental state ascription.
After presenting a conceptual study of empathy, we will show how empathy is a cognitive process that is fundamentally analogical. Our account explains why empathy can be difficult to achieve, but also suggests how it can be possible between people of very different backgrounds and even with animals.
Since its inception, the sense of the term "empathy" has wandered. As a result, defining empathy is difficult and somewhat arbitrary. Still, a core formation can be reached by looking closely at etymology, and by focusing on conceptual differences between empathy and closely related terms, primarily sympathy.
The concept of sympathy has a long history (Aristotle used it), but the term empathy is quite recent. In the late nineteenth and early twentieth century, the German term "Einfühlung" was used by Rudolph Lotz and Wilhelm Wundt in an aesthetic doctrine, which was then elaborated by Theodore Lipps. (3) For Lipps, Einfühlung is a mode of inner imitation. When empathizing with a work of art, the beholder physically imitates the object and imaginatively projects himself into the object. Lipps also extended Einfühlung to the domain of interpersonal understanding. (4) E.G. Tichener, a student of Wundt, coined the English translation "empathy" in 1910.
The German "Einfühlung" has a distinctive meaning of "feeling into" something. Unfortunately, Einfühlung was often translated as "feeling with", which is the usual meaning of Mitfühlung (sympathy). M.F. Basch (1983) notes that in the Greek derivation of empathy, the prefix em- means "in" or "within", while the prefix sym- means "with", "along with" or "together". Etymologically, therefore, the term sympathy means to share an experience with someone else. When one sympathizes with others, one "feels with" or shares their suffering. In contrast, Einfühlung is a more extensive concept than sympathy. It signifies the ability to comprehend another's state without actually experiencing that state. Basch tells us "its German synonyms, 'sich hineinversetzen' (to put oneself in another's place) and 'Fremdwarhrhnehmung' (to come to know the other or the stranger), imply an understanding of another person that includes, but is not limited to an affective experience, and there is nothing of the irrational or primitive implied by these terms." (1983, p. 110). The term sympathy, then, refers to our awareness and participation in the suffering of another person, while empathy refers to the attempt to comprehend either positive or negative states of another. Lauren Wispe (1991, p.80) describes the difference this way:
In empathy the self is the vehicle for understanding, and it never loses its identity. Sympathy, on the other hand, is concerned with communion rather than accuracy, and self-awareness is reduced rather than enhanced....In empathy one substitutes oneself for the other person; in sympathy one substitutes others for oneself. To know what something would be like for the other person is empathy. To know what it would be like to be that person is sympathy. In empathy one acts "as if" one were the other person....The object of empathy is understanding. The object of sympathy is the other person's well-being. In sum, empathy is a way of knowing; sympathy is a way of relating. (5)
Sympathy is performed with altruistic ends, but empathy may or may not be motivated by good intentions. Indeed, one may empathize solely with narcissistic ends. It is said, for example, that Hitler empathized with the Jews in order to annihilate them.
The centrality of feeling distinguishes empathy and sympathy from other kinds of interpersonal acts. Other concepts in this feeling category are emotional contagion and emotional identification. Both terms overlap with sympathy, but have the added sense of being involuntary. One undergoes emotional contagion, for example, when one passively becomes jubilant in a crowd of people. Other modes of mental state ascription involve no feeling whatsoever. What we shall call "ordinary mental state ascription" is a pure intellectual process consisting of a dispassionate inference or simulation.
Beyond this point, there is little consensus as to what constitutes empathy. We will now argue that achieving empathic understanding involves making a comparison of emotions. We will show how the process of "feeling into" is essentially analogical.
In most informal logic textbooks, analogy is viewed as a comparison of objects that have similar features in common. Research in cognitive science, however, has produced a much richer account of analogy as a computational process of finding correspondences between complex structures that involve relations between objects and causal relations between relations. (6) For example, Darwin's analogy between natural selection and artificial selection, which played a major role in the development of his theory of evolution, involves a mapping between structures such as the following:
| Source | Target |
|---|---|
| select (breeders, organisms) name: select-1 | select (nature, organisms) name: select-2 |
| develop (breeds) name: develop-1 | develop (species) name: develop-2 |
| cause (select-1, develop-1) | cause (select-2, develop-2) |
Just as artificial selection by breeders using the natural variability of organisms explains how new breeds of plants and animals can arise, so variability and natural selection explain how new species arise.
The analogical comparison in this example involves more than seeing the correspondences between attributes such as develop and relations such as selects. The explanatory power of the analogy derives from the correspondence between the high-level causal relations: just as human selection of traits causes new breeds to develop, so natural selection of traits causes new species to develop. In this example, natural selection is the target analog which needs to be understood and developed, while artificial selection is the source analog that is intended to further explanation and problem solving. Adapting distinctions made by Dedre Gentner and Graeme Halford, Holyoak and Thagard (1995) use the term system mapping to describe analogical comparisons that involve correspondences between relations among propositions such as "cause" and "implies". In the next section we will describe the role of system mappings in empathy.
In addition to mapping between structures, analogical thinking involves stages of selection, evaluation, and learning. Faced with a target problem, a thinker must somehow select a source analog to contribute to a solution. Often, selection comes by retrieval from memory, when the thinker recalls a similar problem that can be applied analogically to the target. This application requires mapping the source to the target to make the appropriate comparisons. Ideally, mappings are isomorphisms, involving a perfect structural correspondence between source and target; but analogy generally has to be more flexible and has to settle for imperfect correspondences. The evaluation stage requires the analogist to step back and critically ask how effective the analogy really is in accomplishing its purpose. Finally, if the analogy is effective, the analogist can learn from the identified correspondence and produce a schema abstracted from the source and target analogs for general future use.
Holyoak and Thagard argue that all four stages of analogical thinking can be understood in terms of three cognitive constraints: similarity, structure, and purpose. Such constraints on analogy are essential because in principle anything can be viewed as similar to anything else in some trivial respects. The problems of retrieval of potential source analogs from memory and of mapping between complex structures are both computationally intractable unless strong constraints are applied. To summarize briefly, the constraint of similarity encourages the finding of correspondences between objects that have properties which are perceptually or semantically similar. For example, a cat is more likely to map to a dog than to a fish, since cats and dogs share perceptual features such as legs and fur and semantic features such as being kinds of mammals. There is much more to analogy, however, than this basic kind of similarity. As in the Darwin example, we want source and target analogs to have corresponding structure, especially that involving higher-level relations such as cause. The third constraint on analogical thinking is purpose: the value of an analogy depends ultimately on whether it accomplishes its cognitive end, which may include problem solving, explanation, or communication. An analogy may be useful even if it only weakly satisfies the constraints of similarity and structure. Holyoak and Thagard have described how these constraints operate at all stages in many different kinds of analogical thinking, including children's problem solving, scientific explanation, political decision making, and cultural uses of metaphor. They have also developed computational models that show precisely how the different constraints can be simultaneously satisfied. Here, we have provided only enough detail to set the stage for showing how analogy is essential to empathy.
We can have empathy for literary characters as well as real individuals. To take a familiar example, consider Shakespeare's Hamlet. One of the reasons that this character is so compelling is that many people can understand Hamlet's distress and indecision not only abstractly but also empathically: we can imagine feeling an approximation to Hamlet's emotions. How this comes about requires an appreciation of the multifaceted nature of emotion.
Keith Oatley (1992) has presented a theory of emotion that incorporates the phenomenological and physiological aspects of emotions, but that nevertheless emphasizes their cognitive aspects. Building on joint work with Philip Johnson-Laird, he contends that the basic human emotions are all intimately connected with goals. For example, happiness occurs when individuals are accomplishing their goals, sadness occurs as the result of failure to accomplish goals, and anger is directed at whatever blocks the accomplishment of goals. We can put his insights about the relations between goals and emotions to work in seeing how empathy operates analogically.
When we watch or read Hamlet, our ability to "feel-into" the lead character requires our being able to produce a system mapping between his situation and some aspect of our own life. Like analogical processing in general, production of this mapping may be unconscious. Our understanding begins with an appreciation of his situation: his uncle has killed his father and married his mother:
To be, or not to be- that is the question;
Whether 'tis nobler in the mind to suffer
The slings and arrows of outrageous fortune
Or to take arms against a sea of troubles
And by opposing end them. To die, to sleep-
No more- and by a sleep to say we end
The heartache and the thousand natural shocks
That flesh is heir to.
Although few of us have experienced Hamlet's exact situation, most people can nevertheless imagine, at least roughly, what it would be like to be in Hamlet's shoes. Most people have experienced situations where loss made us sad, betrayal made us angry, and difficult situations made us indecisive. For Hamlet as the target analog, we can easily retrieve from memory potential source analogs of varying degrees of similarity. Candidates might include having been abandoned by a lover or having been fired from a valued job. The important thing is that these source analogs should have involved situations where you experienced an emotion that you can project onto Hamlet. In line with Oatley's theory, projection involves mapping over the system of causal relations that ties together the situations (including beliefs and behaviour) and goals with the emotions produced. Here is a rough approximation to what might be involved, presenting propositions and their names to express the causal relations between propositions:
| Source: you | Target: Hamlet |
|---|---|
| fire (boss, you): s1-fire | kill (uncle, father): t1-kill |
| lose (you, job): s2-lose | lose (Hamlet, father): t2-lose |
| marry (uncle, mother): t3-marry | |
| cause (s1-fire, s2-lose): s3 | cause (t1-kill, t2-lose): t3a |
| angry (you): s4-angry | angry (Hamlet): t4-angry |
| depressed (you): s5-depressed | depressed (Hamlet): t5-depressed |
| cause (s2-lose, s4-angry): s6 | cause (t2-lose, t4-angry): t6 |
| cause (s2-lose, s5-depressed): s7 | cause (t2-lose, t5-depressed): t7 |
| indecisive (you): s8-indecisive | |
| cause (s5-depressed, s8-indecisive): s9 |
Here we have used extended predicate calculus to make clear the relational structure of the source and target, including the crucial causal relations. On the multiconstraint theory of analogy, mapping between source and target does not require semantic identity of relations such as fire and kill, or exact structural correspondence between source and target. (Note the mismatches: the boss fired you, but the uncle killed the father, not Hamlet.) Holyoak and Thagard's computer program ACME (Analogical Constraint Mapping Engine), which implements the multiconstraint theory of analogical mapping, would have no problem taking the above structures as input and concluding, not only that you correspond to Hamlet and your boss corresponds to Hamlet's uncle, but also inferring that Hamlet will be indecisive. This last inference involves a process that Holyoak and Thagard call copying-with-substitution : to expand the Hamlet target, we copy over from the source the predicate indecisive, and substitute Hamlet for you on the basis of the mapping already established.
The crucial difference between ACME (or someone who has only a theoretical interest in Hamlet) and someone who understands Hamlet empathically is that the analogical transfer involves not only a verbal correspondence but also a projection of emotions. Your memory of being fired involves not merely the fact that you were angry, but also your feeling of being angry, and you attribute to Hamlet an image of this feeling. Thus empathy involves an unusual sort of analogical thinking, in that the correspondences involve emotions as well as verbal representations. Extending analogical thinking beyond verbal thinking should not be surprising, since analogies can be visual and even auditory.
Analogical transfer need not be based on unique cases. Holyoak and Thagard describe how it is often effective to use multiple analogs to understand a complex situation. A full understanding of Hamlet may accordingly require you to map his situation to a number of situations that you have faced in order to capture the causal complexity of his emotional situation. In addition, transfer can take place, not from a particular case, but from a schema that is constructed from two or more previous cases. As we mentioned in our discussion above of learning, the fourth stage of analogical thinking, abstract schemas can be formed from two or more analogous cases. One may, for example, have a loss schema formed from what is common to losing a job and losing a lover, including the crucial common information that loss causes anger and depression. For the purposes of empathy, the representation of these emotions must not be purely verbal, but should involve some approximation to the feeling that goes with the original experiences from which the schema was formed. Then mapping the loss schema to Hamlet's situation will have the desired empathic result: you actually simulate what Hamlet feels.
We have seen how empathy can be possible when there is a system mapping that draws correspondences between two persons' situations, goals and emotions. But our account also explains why empathy is often not achieved. The basic pattern of empathic explanation of a person P's situation is:
The person P is in situation S, which is like your situation S'.
P has goals G which are like your goals G'.
When you faced situation S' which affected your goals G', you felt emotion E'.
(S' and G' caused E'.)
So P is probably feeling emotion E, which is like your E', caused by S and G.
Putting it this schematically is misleading. What is important for empathy is not that you know explicitly that your situation S' and goal G' caused my emotion E' (you do not need to have Oatley's theory of emotion), but only that you remember that when faced with S' and G', you felt E'. That is usually enough for you to know what P is feeling.
Empathy can involve not only inferring that someone has an emotion, but also using the attribution of the emotion to explain or predict another's behavior. Having understood that P is feeling E, you can then predict P's behavior as the result of possession of that emotion. For example, empathy with Hamlet produces appreciation of his emotional distress, which then can explain his subsequent odd behavior, such as spurning Ophelia. Again, the explanation pattern can be analogical, based on a system mapping:
The person P has emotion E in situation S.
When you were in a similar situation S' with a similar emotion E', you did action A'.
(S' and E' caused you to do A'.)
Hence P's emotion E may lead to a similar action A.
This kind of empathic understanding of actions is important in legal trials when a jury refuses to convict someone of a crime such as the killing of a husband by a battered wife; the members of the jury may feel that they would have done the same thing under the same circumstances. A complex act of empathy, such as that performed by a sensitive therapist in extended treatment, will produce understanding of a whole complex of interacting emotions and actions.
But empathy can easily fail, either because you do not find a source analog that corresponds to the goals, situations, and emotions of the other person, or because a retrieved source does not in fact correspond very well, and should be constructed instead. During the 1992 Canadian election, Prime Minister Kim Campbell gave a speech at a shelter in Vancouver's Skid Row. She told the residents of the shelter that she too had known loss and disappointment, for she had once wanted to be a concert cellist. Perhaps from her perspective this was the best source analog available, but it maps poorly to the long-term desperate situation and disparate goals of the inhabitants of Skid Row. According to the multiconstraint theory of analogy, the success of an analogy depends on how well it satisfies the constraints of similarity, structure, and purpose. Accordingly, empathy will be weak to the extent that the source analog of the empathizer:
1. has disparate goals, situations, and emotions to those in the target analog;
2. has causal relations with different structure than those in the target analog;
3. does not contribute to the cognitive purposes of the empathizer.
We shall see how these constraints determine the balance of simulation and theory-application in empathic understanding.
In order to address further questions about the nature of mental state ascription, we need to locate the processes that produce empathic understanding in a general account of mental operations. In current cognitive science, there are two main accounts of mental structures and processes. The initially most influential kind of cognitive architecture involved rule-based systems, that is systems in which the major structures are IF-THEN rules and the major processes are ones that involve selecting rules and firing them to make inferences (Newell and Simon, 1972; Newell, 1990; Anderson, 1983; Holland et al., 1986). An alternative connectionist view of cognitive architecture conceives of thinking not as rule firing, but as parallel constraint satisfaction in networks of units connected by excitatory and inhibitory links (Rumelhart and McClelland, 1986; Kosslyn, 1994 ; Holyoak and Thagard, 1995). These two views of cognitive architecture are not necessarily incompatible, for it can be argued that human thinking involves both the highly parallel associative processes modeled by connectionist networks, and the more serially, more constructive kinds of inference modeled in rule-based systems (Thagard, forthcoming).
In particular, empathic understanding can involve both relatively automatic parallel constraint satisfaction and more deliberate rule-based reasoning. In cases where you want to gain an emotional understanding of someone whose nature and experiences are like yours, you merely need to activate some elements of your experience to produce an emotional representation of the person you are trying to understand. If you have a good friend struggling to make a difficult decision that requires integration of various actions and goals, you may be able to appreciate the friend's anxiety by retrieving a situation in which you were also struggling with a difficult decision and felt considerable anxiety. Retrieving and mapping analogs can be modeled in terms of parallel constraint satisfaction (Holyoak and Thagard, 1995), and so can decision making (Thagard and Millgram, in press; Millgram and Thagard, in press). Emotions are very important to this sort of process, since we have no conscious access to the mental operation of parallel constraint satisfaction, and feelings such as happiness, relief, fear, and anxiety provide consciousness with a reading of the overall state of constraint satisfaction. (This view is compatible with recent accounts of the cognitive function of emotion offered by Oatley, 1992, and Damasio, 1994.) Your emotional reaction is what tells you how the process of parallel constraint satisfaction went for you, and by analogy, how the process of parallel constraint satisfaction went for the person you are trying to understand. Obviously, you do not need any theory of how your mind works to be able to do a simulation of your friend's decision. Nor do you already need to know the person's emotional state in order to retrieve a relevant analog, since other aspects of the situation serve as retrieval cues that bring to mind an analog enabling you to infer the other's emotional state.
But empathic understanding, like thinking in general, is not always so automatic. If parallel constraint satisfaction fails to come up with a coherent analog of another's situation, a more deliberative, constructive process produces new hypotheses and new concepts that establish new elements and constraints to contribute to parallel constraint satisfaction. If you fail to retrieve from your own experience a situation relevant to that of the person you are trying to understand, you may be able to use your general knowledge about people and yourself to construct a new situation in which you can imagine placing yourself. This process of construction requires the kind of rule-based reasoning that is often used to account for human problem solving, planning, and explanation. Therapists trying to understand patients may need to make many inferences based on rule-like knowledge of human psychology before being able to place themselves in the situation of the patients and simulate what they feel.
Both parallel constraint satisfaction and rule-based reasoning often result in mental state ascriptions with no corresponding feeling. Empathy is more than ordinary mental state ascription, since it involves having an emotion that is analogically ascribed to someone else. The feeling provides a deeper level of interpersonal knowledge than ordinary mental state ascription provides.
Achieving empathic understanding by parallel constraint satisfaction and by rule-based reasoning both involve having feelings, both involve an analogical mapping from one person's situation to another, and both involve a kind of simulation of the other person. But the simulations are different, in that parallel constraint satisfaction is relatively automatic and unconscious and does not apply explicit knowledge and inference. In contrast, simulation of someone's emotions that requires rule-based reasoning will typically be more deliberate and theoretical in character. We will now show that appreciating these two ways of achieving empathy reconciles two views of mental state ascription that are taken to be competing.
To conclude, we want to use our analogical account of empathy to contribute to a recent controversy concerning how people attribute mental states to other people. (7) Goldman has argued that the simulation heuristic that he sees at the basis of analogy provides an alternative to the "theory-theory"(TT), according to which people use an implicit theory of mind to understand what other people are thinking or feeling. The simulation-theory (ST) holds that interpreting other people requires no folk-psychology, and no possession of any conscious or unconscious theory. Rather, says Goldman, we ascribe states to others by feigning their beliefs and desires and feeding these inputs into our own practical reasoning mechanisms. The output states which we generate are subsequently taken "off-line" and attributed to the other person. Goldman sees empathy as a special case of simulation, where the output states are emotions.
Our analogical account of empathy could be construed as a more detailed version of ST, where the practical reasoning mechanisms are specified as analogical processes. What we shall call "local" analogies, or cases when empathy is easy, appear to involve little or no theoretical work. (I do not need much theory to understand that my best friend was angry when she missed her flight, especially since I was angry when I missed my flight last year.) However, although we believe ST loosely describes what is happening when we claim to know what another person is feeling or thinking, we also believe that the hypotheses of TT are also compatible with our account. "Long-distance" analogies inevitably involve the application of theory to compensate for disparate situations and goals, as well as missing target information. In these cases, rule-based reasoning guides the construction of an appropriate analog, thereby making simulation possible. The proportion of theory involved in intermediate cases will again be determined by the constraints of analogical mapping.
Suppose you are trying to understand the feelings of someone of a different culture, or someone who is physically or mentally impaired. Because the other person is very different from you, it may be very hard for you to find a source analog from your own experience that has features and causal structure that are similar to those of the other person. Empathy is particularly likely to fail if you are not motivated to go to the effort of constructing an appropriate source analog when simple retrieval produces inaccuracy.
Consider empathy with an autistic person. In her recent autobiography, Donna Williams describes how her autistic traits (averted gaze, blank expression, and anti-social behaviour) were almost always misunderstood. She writes: "I learned eventually to lose myself in anything I desired- the patterns on the wallpaper or the carpet, the sound of something over and over again.... Even people became no problem. Their words became a mumbling jumble, their voices a pattern of sounds."(1992, p.4) For non-autistics, empathic understanding of Donna's affective state will require a constructed source analog or affective schema that includes, at minimum, feelings of intense isolation and frustration. Empathy with autistics is an example of a "long-distance" analogy. Only those of us who are intrinsically shy or withdrawn will be able to simply retrieve situations when the goal of unfettered interaction with others caused an emotion which is something like Donna's. For many of us, empathy with the affective life of autistics will remain elusive. An attempt to empathize with Donna's experience will require background theory in the form of rule-based reasoning, specifically about the nature of autism. Rule-based reasoning is needed to bridge the disparity between goals and situations and direct the creation of an appropriate, more sophisticated analog.
Surprisingly, empathy with another species may be easier to achieve than empathy with autistic people. Suppose we wish to interpret the affective states of East African Vervet monkeys. Empathy with another species involves immediate dangers of over-anthropormorphizing our analogies. However, in their recent work How Monkeys See the World, Cheney and Seyfarth (1990) insist that in the case of Vervet monkeys, anthropormorphizing works. Cheney and Seyfarth argue that the features and structure of human family relations are analogous in many respects to those of Vervets. To take a simple example, a human mother's emotional need to nurture and protect her infant is easily mapped to the Vervet mother's defense of infants who show signs of distress. The Vervet grooming habits are understood roughly in terms of human situations and goals: just as the female members of a human family preserve their friendships through various rituals, the Vervets throughout the maternal line maintain close social bonds which are expressed by grooming. Empathy with Vervets is an intermediate case of analogy, involving less rule based reasoning than in the autistic case to generate an accurate simulation.
Empathy with Vervet monkeys and with autistics are examples of analogies that require greater conceptual leaps. These analogies are just as possible in empathic understanding as they are in scientific understanding. Empathy is of course much easier when it is based on "local" analogies, ones where there is a high degree of similarity between the concepts in the source and target analogs. Empathizing with a friend who has similar experiences and aspirations will often be effortless. It will be a simple task, for example, to empathize anger in a best friend who has missed an important appointment due to traffic. Local analogies are accomplished by pure simulation and no rule-based reasoning is needed to retrieve a source analog. Rather, retrieval is spontaneously guided by the close similarity of features and causal structure.
It might be objected that analogy is not really necessary for empathic simulation: you just place yourself in someone's situation and see how you feel. But the placing in fact requires that there be a systematic correspondence between the situations of the empathizer and the other, so that processes of analogical mapping, which may be conscious or unconscious, are required. In order to imagine ourselves in another's shoes, it is necessary to use the materials of our own experience. Selecting which bits of experience to use may or may not involve the use of a theory. Sometimes we interpret or understand people by using quick simulations, sometimes we use more sophisticated models, and sometimes we resort to dispassionate theory-laden explanations. As we have shown, we can make generalizations about the conditions under which these various mental processes obtain based on the constraints of analogical mapping. The ST/TT debate is transposed once mental state ascription is viewed as an analogical activity.
We can see now that the assumption that ST and TT are mutually exclusive alternatives is untenable. (8) Adults undoubtedly have theories and are likely to apply them when making long-distance analogies. Our investigation has shed light on when empathy is easy, when it is hard, and how understanding the connection between empathy and analogy changes the debate between proponents of ST and TT.
* We thank two anonymous reviewers for comments on an earlier draft of this paper. This research is supported by a grant from the Natural Sciences and Engineering Research Council of Canada.
1. Empathy is studied extensively in psychoanalytic theory and clinical psychology. See for example: Lichtenberg, Bornstein and Silver (1984); Kohut (1959); Truax (1967); Hoffman (1987).
2. The debate is documented in Mind and Language (1992) and is also presented in Davies and Stone (1994).
3 . J. Engell, (1963), p.157. His etymology is confirmed by Wispe (1987), pp. 17-37.
4. For a more complete account of Lipps' use of empathy, consider Katz (1963).
5. For more on the distinction between empathy and sympathy see Wispe (1986) and Chismar (1988).
6. The psychological and computational literature on analogy is vast. For reviews, see: Gentner (1983, 1989). Hall (1987); Holyoak and Thagard (1989, 1995); Thagard, Holyoak, Nelson, and Gochfeld (1989).
7. There are slightly different versions of ST and TT. Here, we will use these terms as they are defined by Alvin Goldman.
8. Gordon(1992), Stich and Nichols(1992) consistently maintain that ST and TT are exclusive. Goldman seems to waver on this issue in (1992), allthough he thinks simulation can provide a "uniform account" of all interpretations. Paul Harris(1992), however, claims that children improve their grasp of folk psychology by means of simulation, although he does not state that both theory application and simulation could be present in the same act of understanding.
Anderson, John. R.
1983
The Architecture of Cognition. Cambridge, MA: Harvard University Press.
Basch, Micheal.F.
1983
"Empathic Understanding: a Review of the Concept and some Theoretical Considerations". Journal of the American Psychoanalytic Association, 31, pp. 101-26.
Cheney, Dorothy.L. and Robert M. Seyfarth
1990
How Monkeys See the World. Chicago: University of Chicago Press.
Chismar, Douglas
1988
"Empathy and Sympathy: the Important Difference". The Journal of Value Inquiry, 22, pp.257-266.
Damasio, Antonio. R.
1994
Descartes' Error. New York: G. P. Putnam's Sons.
Dawes, Robyn. M.
1994
House of Cards: Psychology and Psychotherapy Built on Myth. New York: Free Press.
Engell, James
1981
The Creative Imagination: Enlightenment to Romanticism. Cambridge: Harvard University Press.
Gentner, Dedre
1983
"Structure-mapping: A Theoretical Framework for Analogy."Cognitive Science,7, pp. 155-170.
_____ . 1989
"The Mechanisms of Analogical Learning." In S. Vosniadou and A. Ortony eds., Similarity and Analogical Reasoning (pp. 199-241). Cambridge: Cambridge University Press.
Goldman, Alvin
1993
Philosophical Applications of Cognitive Science. Boulder: Westview Press.
_____ . 1992
"In Defense of the Simulation Theory". Mind and Language, 7, pp. 104-19.
Gordon, Robert. M.
1992
"The Simulation Theory: Objections and Misconceptions". Mind and Language, 7, pp. 11-34.
Hall, Rogers
1987
"Computational Approaches to Analogical Reasoning: A Comparative Analysis." Artificial Intelligence, 39, pp. 39-120.
Harris, Paul. L.
1992
"From Simulation to Folk-Psychology: the Case for Development". Mind and Language, 7, pp. 120-44.
Holland, John. H., Keith J. Holyoak, R. E. Nisbett and Paul Thagard
1986
Induction: Processes of Inference, Learning, and Discovery. Cambridge, MA: MIT Press/Bradford Books.
Holyoak, Keith, J., and Paul Thagard
1989
"Analogical Mapping by Constraint Satisfaction." Cognitive Science,13, pp. 295- 355.
Holyoak, Keith, J., and Paul Thagard
1995
Mental Leaps. Cambridge, M.A.: MIT Press.
Katz, Robert
1963
Empathy: Its Nature and Uses. London: The Free Press of Glencoe.
Kohut, Heinz
1959
"Introspection, Empathy and Psychoanalysis". Journal of the American Psychoanalytic Association, 7, pp. 459-83.
Oatley, Keith
1992
Best Laid Schemes: the Psychology of Emotions. Cambridge: Cambridge University Press.
Lichtenberg, Joseph., Melvin Bornstein and Donald Silver, eds.
1984
Empathy I &II. Hillsdale: The Analytic Press.
Millgram, Elijah. and Paul Thagard
(forthcoming)
Deliberative Coherence. Synthese.
Newell, Allen
1990
Unified Theories of Cognition. Cambridge, MA: Harvard University Press.
Newell, A.llen and Herbert. A. Simon
1972
Human Problem Solving. Englewood Cliffs, NJ: Prentice-Hall.
Rumelhart, David. E. and James L. McClelland, eds.
1986
Parallel Distributed Processing: Explorations in the Microstructure of Cognition. Cambridge MA: MIT Press/Bradford Books.
Smith, David Woodruff
1989
The Circle of Acquaintance: Perception, Consciousness and Empathy. Dordrecht: Kluwer Academic Publishers.
Stein, Edith
1964
On the Problem of Empathy , W. Stein, (trans.). The Hague: Martinus Nijhoff.
Stich, Stephen.P. and Shaun Nichols
1992
"Folk Psychology: Simulation or Tacit Theory?". Mind and Language, 7, pp. 35- 71.
Thagard, Paul
(in press)
Mind: Introduction to Cognitive Science. Cambridge, MA: Bradford/MIT Press.
Thagard, Paul and Elijah Millgram
(forthcoming)
"Inference to the Best Plan: A Coherence Theory of Decision." In A. Ram and D. B. Leake (Eds.), Goal-directed Learning. Cambridge, MA: MIT Press.
Thagard, P., Keith J. Holyoak, Nelson and Gochfeld
1989
"Analog Retrieval by Constraint Satisfaction." Artificial Intelligence,46,pp.259- 310.
Truax, Charles and Robert.R. Carkhuff
1967
Toward Effective Counseling and Psychotherapy. Chicago: Aldine Publishing Company.
Williams, Donna
1992
Nobody Nowhere: the Extraordinary Autobiography of an Autistic. Toronto: Doubleday Canada Limited.
Wispe, Lauren
1987
"History of the Concept of Empathy", in N.Eisenberg & J.Strayer (eds.), Empathy and Its Development. Cambridge: Cambridge University Press, pp. 17- 37.
_____. 1991
The Psychology of Sympathy. New York, Plenum Press.