Functional Origins of Religious Concepts:

Ontological and Strategic Selection in Evolved Minds

Pascal Boyer

What is the origin of religious concepts? How come we can find concepts of supernatural agency more or less the world over, with important recurrent features? This lecture is a ‘progress report’, an account of how these previously intractable questions are now a matter of empirical, indeed experimental inquiry. What brought about this remarkable change is substantial progress in our understanding of how human minds work. This allows a naturalistic account of cultural representations that describes how evolved conceptual dispositions make humans likely to acquire certain concepts more easily than others. The aggregated result of these individual acquisition processes channels cultures along particular paths, with the result that some concepts are both relatively stable within a group and recurrent among different groups.

Cultural transmission, like other forms of human communication, does not consist in a ‘downloading’ of concepts from one mind to another. It requires inferential processes, whereby people attend to cues in other people’s behaviour, infer their communicative intentions, build concepts on this basis of what they inferred (Tomasello, Kruger, & Ratner, 1993; Sperber, 1996). As a result, cultural transmission is a process whereby people constantly create variants of other people’s representations. Concepts that are stable in one group and recurrent between groups are concepts that were selected in the transmission process, against a whole variety of variants that were forgotten, discarded and modified.

Obviously, many different factors contribute to stability and recurrence, even if we factor out local conditions. Here I describe two selection factors that result from the way human minds acquire and store information. Both reduce the space of possible cultural concepts to a narrow range of ‘culturally fit’ ones.

A first selection stems from the fact that human minds are equipped with a particular ontology, a set of intuitive expectations about the kinds of things to be found in the world. Among the indefinitely many concepts individuals can imagine and combine, some connect with this ontology in a particular way I describe below. As a result, they stand better chances than other concepts of being successfully transmitted. The outcome of this conceptual selection is a short ‘catalogue’ of supernatural notions, among which religious concepts but also culturally viable folklore, superstition, fiction and fantasy.

A second selection pressure stems from the fact that humans are greatly dependent upon cooperation and information about potential cooperators. This creates specific ‘strategic’ problems. Various cognitive and emotional adaptations help solve these problems. These specialised mechanisms make a small sub-set of our ‘catalogue of the supernatural’ more likely to be associated with representations of group-identity, ritual, morality and social interaction.

These selection pressures are very general. In selective models of culture, one assumes that the aggregation of many individual acquisition processes ‘washes out’ all sorts of local biases. In any group at any time, we will find particular causes that boost the diffusion of one particular version of religious concepts against others. But in the long run, and in the comparison of many different human groups, these local biases cancel each other out. What we find as recurrent features, over time and between groups, are concepts that, all else being equal, tend to resist distortion better than others. Explaining why this is so is what I mean by describing the ‘functional origins’ of religious concepts.

Religious concepts and intuitive ontology

The present account is about mentally represented concepts. These concepts do not always match official descriptions of what people are supposed to think, in this case commonly sanctioned theologies. Perhaps more important, people’s concepts are not entirely accessible to conscious inspection. This is why we need and use experimental protocols to uncover their contents and organisation.

Three features are generally present in the mental representation of such concepts.

First, religious concepts activate a set of ontological categories that we know are present in normal minds from an early stage of cognitive development and remain active as the main categories of things in the world: PERSON, ARTEFACT, ANIMAL, etc. We now have much better evidence concerning intuitive ontological categories and associated theories, coming from developmental psychology, from experimental studies with adults, from neuroscience and from the study of cognitive pathologies. In the mental representation of conceptual knowledge we can distinguish between two levels of organisation, that of ‘kind-concepts’ (or ‘entry-level’ concepts) on the one hand, and that of ‘ontological categories’ on the other. The former are concepts like ‘giraffe’ or ‘telephone’ or ‘uncle’ or ‘tree’ or ‘river’. Their representation activates higher-level categories of the fundamental kinds of things in the world, like animal, artefact, person, plant, natural object. Any object in the word that is identified as belonging to a kind-concept thereby activates a particular ontological category. In the same way, concepts of imaginary objects and beings are intuitively associated with particular ontological categories. The concept of ‘spirit’ we find in so many cultures activates the category PERSON. If you pray to a particular statue of the Virgin, you are standing or kneeling in front of an ARTEFACT. If you think that some antelopes can disappear at will, you must activate your ANIMAL category to represent these special beings.

Second, religious concepts invariably specify information that violates intuitive expectations associated with the relevant ontological category. Ontological categories are not just large classes of things. They constitute what psychologists like to call ‘naive’ or ‘intuitive theories’. For instance the motions and interaction of solid physical objects is one such domain, that we generally call ‘intuitive physics’ and develops from the first months after birth. Principles of intuitive physics stipulate that objects have continuous trajectories in space and time, that they collide but do not merge on contact, that unsupported objects fall, etc. Also, the fact that some objects are identified as members of the categories animal or person triggers ‘intuitive biology’ principles. These stipulate that live beings have ‘essential’ qualities inside them that make them members of one particular species, that they are propelled by internal force, that they belong to mutually exclusive classes in a taxonomic hierarchy. This ‘intuitive biology’ also develops in early childhood. The most important domain of such expectations is what we usually call ‘theory of mind’ or ‘intuitive psychology’. This is the domain of expectations and principles that allow us to represent the behaviour of animals and people as guided by internal representations: beliefs, intentions, emotions. All these specialised skills are called ‘theories’ because they inform complex, non-trivial inferences from the available information. The skills are ‘intuitive’ in the sense that subjects are generally not aware of the principles involved. All they are aware of are the output of these devices, e.g. representations of particular events as caused by other events. Finally, intuitive theories are ‘domain-specific’ because they are selectively triggered by particular aspects of the environment. Theory of mind inferences are spontaneously applied to the behaviour of animals and persons, but not of artefacts. Biological inferences are triggered when some object is identified as an animal, not if it is identified as an artefact. Now religious concepts describe members of these categories that have special characteristics. For instance, spirits are persons that are mostly invisible and go through physical obstacles. Some statues and other artefacts are special because you can talk to them and they will listen and understand. Some mountain are special because they eat food and digest it (see more illustrations in (Boyer, 1994)). These violations are generally explicit and is transmitted by cultural input. One is told that spirits are invisible or that ghosts can go through walls.

Third, a religious concept also activates the intuitive expectations that are not violated, among those associated with the relevant ontological category. By contrast with the features above, this remains generally tacit and need not be acquired via social transmission. For instance, people tacitly represent the spirits as having minds. That is, spirits are assumed to perceive events. They supposedly remember what they perceived. They have beliefs and form intentions on the basis of their beliefs, and so on. There is an intuitive ‘theory of mind’ that is spontaneously extended to spirits because they are identified as a special kind of PERSON. Note that our theory of mind works very well without us ever representing what its principles are, how it computes intentions from behaviours, and so on.

Templates for religious concepts

We can now distinguish between supernatural concepts and a small series of templates . Mentally represented supernatural concepts are extremely diverse. They vary not just from group to group but also from a member of a group to another and even within the same individual, depending on the cognitive task at hand. But diversity and similarity, obviously, are a matter of explanatory viewpoint. To understand why we find both diversity and some recurrent features in religious concepts, we must focus, not just on the concepts themselves but also on the cognitive processes that allow people to acquire, represent and communicate them. Templates are just principled ways of handling information from intuitive ontology, with the following structure:

[0] lexical label;

[1] pointer to an ontological category;

[2] violation of expectations, either:

[2a] breach of expectations for that particular category, or

[2b] transfer of expectations from another category;

[3] activation of non-violated expectations for the category;

[4] additional encyclopaedic details, that vary from place to place.

There are several such templates. Many different religious concepts may correspond to one template. For instance, this is the template for the many concepts of ‘spirit’ that we find in so many cultures:

[1] an ontological category: PERSON

[2] a violation of intuitive physics, e.g. spirits are invisible

[3] activation of non-violated expectations: being persons, spirits have a mind, they can perceive events, form beliefs, have intentions, etc.

[4] place-holder for additional (local) detail

We can apply the same operation to another kind of supernatural concept, that of objects with intentionality, like for instance a statue that listens to prayers:

[1] an ontological category: ARTEFACT

[2] a transfer of expectations associated with the PERSON category: this special artefact has thought and perception.

[3] activation of non-violated expectations: being an artefact, this object was made by someone, it is made of a particular material, it is a solid object with one position in space, etc.

[4] place-holder for additional (local) detail

Templates are not ‘archetypes’ or ‘ideal types’ or a ‘family resemblance’ for religious concepts. In fact templates are not concepts at all. They are just procedures for the use of information provided by intuitive ontology. As I will show below, the cultural success of concepts seems to depend, not on their specific features, but on what intuitive ontological information is used and how, in other words which template is used.

This account predicts that there are not that many different templates, simply because there are not that many ontological categories and sets of associated intuitive principles. In other words, there are not that many ways of ‘tweaking’ intuitive ontology so as to produce supernatural concepts, so that a general ‘catalogue of the supernatural’ should be rather short. If one sticks with the description of intuitive ontology given above (which may of course change depending on better psychological evidence for categories and intuitive principles), five categories are involved: person, animal, plant, artefact, natural object (i.e. non-man-made, non-living parts of the environment like rivers, rocks and mountains) and three main domains of inference: intuitive physical expectations, intuitive biology and intuitive ‘theory of mind’. Violations are produced either by breach or by transfer. A breach contradicts intuitive expectations associated with the ontological category (e.g. a table that suddenly disappears, thereby violating intuitive physics activated by the artefact category). A transfer extends to a category information that is intuitively associated with another category (e.g. a table that breathes, using biological information associated with the animal category). Categories, types of inferences and possible violations produce only a small list of templates:

[1] Person with breach of physical expectations.

[2] Person with breach of biological expectations.

[3] Person with breach of psychological expectations.

[4] Animal with breach of physical expectations.

[5] Animal with breach of biological expectations.

[6] Animal with breach of psychological expectations.

[7] Plant with breach of physical expectations.

[8] Plant with breach of biological expectations.

[9] Plant with transfer of psychological expectations.

[10] Natural object with breach of physical expectations.

[11] Natural object with transfer of biological expectations.

[12] Natural object with transfer of psychological expectations.

[13] Artefact with breach of physical expectations.

[14] Artefact with transfer of biological expectations.

[15] Artefact with transfer of psychological expectations.

The template account is, on the whole, compatible with the anthropological record. Note that it is by no means a simple task to evaluate the relative frequency of templates. One serious difficulty is that the kind of evidence provided by anthropological reports, precious though it is, covers only one aspects of religious concepts, the overt, socially transmitted violations. This makes sense as this is what makes such concepts salient. A cognitive description of the templates also requires the tacit side, those background expectations that are used in producing inferences about religious objects and agents. Moreover, we have little or no reliable statistical information about the distribution of different types. However, we can be fairly confident about both ends of the distribution: concepts that we find virtually everywhere and concepts that are not reported so far in the anthropological record. This evidence confirms that most religious concepts are indeed based on one of the templates in this catalogue, given the ontological categories and principles normally developed by human minds. The evidence also suggests that the features that correspond to these templates are invariably essential to the representation of the religious concepts. In psychological terms, they are the core features that make such imagined objects and agents what they are.

Moreover, the template account also predicts what we will not find in the core features of culturally successful religious concepts. This is an advantage of cognitive explanations over accounts phrased in terms of ‘general types’ or ‘common traits’ or a ‘family resemblance’, which provide no description of the contrast set, that is, the set of possible representations that you should not expect to find as central features of religious concepts. If the present account is valid, we should not expect to find:

1. Concepts that include no violation of expectations for an ontological category (no slot [2] in the template).

2. Concepts that include a violation but do not activate background assumptions (no slot [3] in the template).

Because there are two levels of conceptual information, that of kind-concepts and that of ontological categories, one can imagine strange situations that violate information associated with either one of these levels. For instance, a tiger with five legs and green fur constitutes a violation of the basic concept ‘tiger’. By contrast, a tiger that was born of a cow violates, not specific information about tigers but a principle of intuitive biology (‘all living things reproduce within their species’) that is associated with the ontological category animal. In the same way, compare (i) a table made of chocolate (basic-level violation) and (ii) a table that understands what people say (an ontological violation). Compare for instance the following concepts:

[1a] there are people who never die

[1b] there are people who turn blue when they are thirsty

[2a] there are animals that are invisible

[2b] there are animals that feed on sand

[3a] there are statues that listen to you

[3b] there are statues so tall they reach the clouds

In each pair, the first concept [a] corresponds to one of our templates by including a violation of ontological information and the second example [b] specifies information that is unusual and contradicts information about our basic kind concepts. Now it is striking that, although we find such ‘oddities’ in the representation of religious entities and agencies, they are not the central features, those that are prominent in the representation of the concept.

The second prediction is that violations must be accompanied by activation of non-violated expectations for a category. Here too we can contrast template-generated examples with concepts that do not allow such activation:

[1a] there are special people who never die

[1c] there are special people who exist only on Wednesdays

[3a] there are special statues that listen to you

[3c] there are special statues that are nowhere at all

The [c] examples block the background expectations and make inferences about such situations particularly difficult. A person who never dies remains a person so it makes sense to wonder what he or she knows or wants. That a person exists only sporadically makes such inferences difficult. The same goes for the statue. That it listens to people does not block physical inferences; the artefact is somewhere, has a mass, etc. Such inferences are difficult with [3c]. Although such ‘inference-blocking’ concepts are strange, intriguing, etc., we do not find them among culturally successful religious notions.

The central features of religious concepts violate information from ontological categories and from that source only. Religious and other supernatural concepts may or may not include basic-level violations, but they invariably include a violation of ontological information. Also, they are such that they support non-violated inferential principles of intuitive ontology. Concepts that do not abide by these two conditions are not found among successful religious concepts. To sum up, religious concepts strike at the core, as it were, of people’s ontological commitments but generally in the form of limited violations of particular intuitive expectations.

As I said at the beginning, what we find in religion is by necessity a matter of selection in transmission. We cannot just say that the concepts we find in many cultures are ‘better’ than other possible variants. We have to demonstrate that they enjoy some advantage in acquisition, keeping in mind that ‘non-starters’, concepts with little cultural viability, are in constant production. What makes people either ignore or forget those variants? or, What makes people ‘normalise’ them to better versions? Given utterances, stories, images, etc. about various imaginary situations, this account would require that people tend to represent (or mis-represent) them in terms of limited violations of expectations associated with one ontological category. Multiple iterations of this process result in the small number of combinations found in culturally recurrent supernatural concepts. So we must turn to experimental evidence that highlights such processes in individual acquisition.

Experimental evidence: Recall and inference

We now have some evidence that individual processes indeed tend to favour supernatural concepts of the format described above. This is not the place to describe protocols, materials or statistics in any detail, so I will concentrate on the main findings.

[1] Violations are recalled better than standard associations

Ideally, all sorts of individual factors involved in transmission should be tested. But recall is most important because it is a necessary condition for cultural transmission. All else being equal, concepts that are recalled better than others have a higher potential for transmission. An advantage for violations as described above was found both in straightforward free recall tasks (Boyer & Ramble, 9999) and in serial transmission studies (Barrett, 1996). The materials for these studies consisted in stories that included descriptions of both violations and standard associations. In recall tasks, participants are asked to list what they can recall of the story after a distraction task. In serial transmission, the participants’ recall of a story is used as material for a second ‘generation’ of participants, and so on. Barrett and I were careful not to use material that could trigger direct associations with familiar religious concepts. We found a significant advantage for violations over standard associations for both person and artefact categories.

[2] Recall depends on violations, not on ‘oddity’

The previous results could be interpreted as the fact that violations are simply ‘strange’ because they are unexpected. So in additional studies we included items that are simply ‘odd’, that is unexpected but without ontological violations. This was tested in serial transmission (Barrett, 1996) and free recall (Boyer & Ramble, 9999). Results show that violations are better recalled and better transmitted than ‘odd’ material. This difference suggests that strangeness cannot be the only explanation for recall of violations. The recall difference may also explain why we find violations rather than mere oddities in recurrent religious assumptions.

[3] Violations do not contaminate people’s intuitions

In this account, religious templates are procedures for handling information from intuitive ontology that produce salient concepts. This requires that intuitive ontology itself is not affected by the presence of such salient violations (otherwise they would not be salient). This contrasts with culturalist accounts for instance, which would suggest that people’s intuitions are modified by local cultural concepts. A good test of this would be whether we find situations in which people (i) represent a violation of intuitive expectations in the limited context of some religious notion and (ii) refuse similar violations in all other contexts. An illustration is Walker’s study of transformations of natural kinds (an animal changes species) in both ritual and non-ritual contexts among the Yoruba of Nigeria (Walker, 1992). She found that people who were willing to accept such transformations as ritual occurrences were equally confident that they could not happen in other contexts.

[4] Inferences are governed by background expectations

Those intuitive expectations that are not violated constitute the source of people’s inferences, for instance about the ancestors’ likely reactions to a particular situation. There is good anecdotal evidence for that, confirmed by ingenious experiments designed by Barrett & Keil on concepts of God (Barrett & Keil, 1996). They elicited from the participants features that make God special. Subjects generally mention violations of theory-of-mind expectations, e.g. that God attends to everything at once. They then tested recall for stories that used these violations. They found that in the subjects’ recall such features were generally replaced with more intuitive descriptions of cognitive functioning, taken from intuitive ‘theory of mind’. So subjects se their default expectations about a category even when the explicit concept (what they think they think) contradicts these expectations.

[5] Violations with no inferential potential are not recalled

Another prediction of this model is that religious concepts should maintain background inferential principles that are not violated (see above illustrations). This has only been tested once, using extremely weird concepts that violate intuitive expectations but do not allow default inferences for the category (e.g. a god that exists sporadically, a statue that is nowhere). If recall favours combinations of violations + default inferences, we should expect such concepts to be recalled less than simple violations, although they are far stranger. This is indeed what we found (Boyer & Ramble, 9999).

[6] Sensitivity to violations is cross-culturally stable

Barrett replicated the God-concepts studies in India and found the same effects (Barrett, 1998). In a series of recall studies, I used stories that included similar items to the French studies. This was tested in two contrasted settings in Gabon and in Nepal. The results of these studies were the following. First, in the same way as in France and the US., we could observe that violations were recalled much better than common associations. Second, we found that ‘odd’ material did not enjoy the same advantage in transmission as violations. Note that we had compared very different groups. Our Gabon subjects had massive exposure to a variety of religious concepts, mainly transmitted through informal channels. Our Nepal group consisted in Tibetan monks who are exposed to religious concepts though massive training in one particular tradition, and whose sources are mostly literate. These massive differences in cultural settings did not result in any significant differences in recall performance.

To conclude, the major result from these studies is that there is a cross-cultural sensitivity to violations of intuitive expectations for ontological categories, that is not affected by the range of religious concepts used in the culture, by their variety or mode of transmission, or by people’s commitment to them. This would suggest that the range of religious concepts people are likely to entertain, acquire and store seems is indeed taken from the same ‘catalogue’ the world over. What specific ‘choices’ were made by the previous generations affects one’s sensitivity to these concepts only marginally. To put this in crude terms, everyone is available for any religious ontology, provided it is in the catalogue. The fact that in each human group some choices are made rather than others is a consequence of history, not of cultural differences in sensitivity or cognitive functioning.

So far, our experimental evidence supports the template account of religious concepts. It predicts that we should find certain recurrent features in religious concepts and we find them in the anthropological record. It excludes certain concepts and we do not find them. It comes with an account in terms of memory processes, a crucial factor of cultural spread. A cognitive account of religious concepts, based on experimental evidence, gives us a first approximation of a reasonable account of religion. But that is only a first step.

Further selection in the supernatural imagination

Our ‘catalogue of the supernatural’ lists a small number of conceptual ‘gadgets’ that are used in different cultural domains: in religion, certainly, but also in fiction, fantasy, folklore, dreams and ‘superstition’. What we usually call ‘religion’ is not just any cultural institution where supernatural imagination is active. We generally use the term to denote an association between some supernatural concepts and particular social and psychological effects. Religion seems to differ from fiction and other forms of folklore in that it triggers stronger commitment. It differs from other kinds of supernatural beliefs in that socially important rituals are associated with its ontological commitments. Religion ties in with concepts of group-identity. It may create powerful social bonds and strong emotional reactions. Also, what we call ‘religion’ establishes a close association between supernatural imagination and moral prescriptions. A frequent association between at least some of these features constitutes the family-resemblance of ‘religion’. Considerable energy has been spent (or wasted) trying to turn this ‘polythetic’ characterisation of religion into a more proper definition in terms of necessary and sufficient features. It is perhaps more profitable to consider such probabilistic associations for what they are. In many human groups, some supernatural concepts may become associated with moral understandings, strong commitment, group-identity, etc. In trying to determine whether this association obtains in all places and at all times in the same way, when in fact it does not, we may have lost track of a more significant point. The association seems to involve only a specific subset of possible supernatural concepts. That is, only some of our templates seem to lend themselves to that kind of ‘serious’ social use. Moreover, I think that some very general cognitive mechanisms make this association very likely and stable in human cultures.

These privileged templates have one feature in common. They all require activation of ‘intuitive psychology’ expectations. This is true for instance of concepts of gods, ancestors and spirits, which correspond to:

[1] Person with breach of physical expectations, and/or

[2] Person with breach of biological expectations.

In these cases ‘intuitive psychology’ is activated as the intuitive background associated with the category (slot [3] in the template). These religious agents are presumed to have a mind because they are special persons and intuitive psychology is activated by the person category. In other cases, such as concepts of listening statues, activation of ‘intuitive psychology’ constitutes the violation part of the template:

[15] Artefact with transfer of psychological expectations.

It seems clear from the anthropological record that ‘socially significant’ supernatural concepts are largely about agents spontaneously (and in large part tacitly) represented as having psychological processes (perception, belief, intention) that agree with our intuitive psychology.

A good illustration of this preference for templates with activation of intuitive psychology is the difference in frequency between concepts of ‘spirits’ and of ‘zombies’. These constitute symmetrical violations. A spirit is a person with standard psychological properties and special physical ones. A zombie is a person with standard physical properties but special psychological ones. Now the spirit type of religious concepts is much more frequent than the zombie-type. Further, wherever people have a concept of zombie, they invariably complement that with the notion of some spirit or witch of god who ‘remote-controls’ the zombie (Boyer, 1996).

The idea that socially significant supernatural imagination is principally about imagined intentional agents is not new, to say the least. But why is that so? That is, why should these concepts rather than others become associated with moral understandings, high commitment, group-identity or emotional arousal in the polythetic cluster we intuitively identify as ‘religion’? The phenomenon seems so obvious that we can easily lose track of the fact that it requires an explanation, preferably based on plausible independent evidence. As Guthrie pointed out, the strong ‘anthropomorphic’ tenor of most religious thinking requires an explanation that must be general – the phenomenon is found the world over – and psychological – since this is a question of some concepts being simply easier to acquire and represent than others – (Guthrie, 1993). However, classical answers are of little help here. In anthropology we used to think, following Piaget, that projecting intentional agency onto non-intentional domains was a pervasive form of cognitive activity, especially prevalent in early childhood (Piaget, 1954). This latter point is not entirely plausible in the light of recent evidence concerning cognitive development. It seems that, on the contrary, children start with a strict demarcation between the domain of intentional agents and the rest (Bullock, 1982; Rochat, Morgan, & Carpenter, 1997). They may sometimes be confused about some objects but this stems from a lack of relevant information, not from principled animism. Explanations in terms of spontaneous projection of human features to imaginary beings are not satisfactory either. The concept of anthropomorphism implies that we take human characteristics and project them onto non-human objects. But this is not exactly what happens here. First, what is projected is not specifically human: Intuitive psychology is spontaneously activated by the presence of, not just humans, but also most animals. Second, not all human characteristics are projected, only intuitive psychology. Many religious concepts that project intentional agency fail to extend to the imagined agents other human characteristics, such as having legs and arms, living in kin-based groups, eating cooked food, etc.. So we are left with this simple question, what makes projection of that special characteristic (standard cognitive processes) so good?

At this point we must not lose track of the fact that we are talking about selection in cultural transmission. That is, the problem is not to explain how people can easily conceive of such agents. That much is explained by the connection with intuitive ontology described above. The problem is to explain why a connection between socially important domains and supernatural concepts seems better transmitted when the latter include activation of intuitive psychology. Nothing in what we said so far, in terms of conceptual structure, would make it especially difficult to establish such associations with other templates for supernatural concepts. Stories that connect ritual, moral prescriptions or group-identity to zombie-like animals or plants with strange physics would be equally salient, counter-intuitive, intriguing, etc.. So there must be some other special feature that makes psychology-based concepts special. To understand what that is, we must turn to another domain of evolved mental machinery, that of cooperation and information.

Mental mechanisms for social interaction

Let me start with a few commonplace remarks about likely consequences of our evolutionary history:

[1] Humans depend more than any other species upon information about their environment. Most human behaviour is based on a rich and flexible data-base that gives parameters for action. Very little human behaviour can be explained or even described without taking into account the massive acquisition of information about surrounding situations. The proper ecological niche of humans is a ‘cognitive niche’ (Tooby & Cosmides, 1990).

[2] Humans depend upon finely-tuned cooperation with other human beings. Humans have for a long time – long enough to make a difference in evolutionary terms – lived in small groups and in intense social interaction. This interaction is not accomplished through stereotyped action-sequences and in fact could not be. They depend on co-ordination of different people’s actions.

[3] Because of these two facts, humans depend to an enormous degree on information about other people’s mental states, that is, what information they have and what their intentions are. No joint hunting expedition, war raid or marriage negotiation can be organised without precise monitoring of what other people want and believe.

That humans depend on co-operation creates all sorts of specific problems. Co-operation problems are strategic problems, where the value (the expected benefit) of a particular move depends on whether someone else makes a particular move (not necessarily the same one) (Schelling, 1960). This among other things creates ‘commitment problems’, that is, it becomes crucial to be able to estimate one’s potential partner’s willingness to cooperate or defect. A whole range of typically human characteristics is explained by these evolutionary factors, including:

[1] A hypertrophy of social intelligence. The human mind-design is characterised by an obvious ‘hypertrophy’ of intuitive psychology. The set of intuitive principles that infer mental states – perceptions, beliefs and desires – from observed behaviour is incomparably more complex than in any other species. Also, it appears very early and seems to engage dedicated brain structures.

[2] Use of ‘un-fakeable’ (or difficult to fake) signals. For organisms that depend on co-operation, signals of trust-worthiness in others are crucially important, as well as detection mechanisms for faked signals (Bacharach & Gambetta, 1999). In humans and other species we find that evolution favours the development of signals that are either impossible or at least difficult and costly to fake. In the case of humans, there is now good evidence that some clues concerning people’s sincerity (tone of voice, facial expressions, gestures) are difficult to fake, indeed to a large degree outside voluntary control (Ekman, 1985). People are often sensitive to these clues, even though they generally have little conscious access to the processes whereby a person’s sincerity is automatically gauged.

[3] Capacity to detect non-cooperators. As evolutionary biologists have pointed out, the evolution of co-operation requires mental structures that allow (i) a precise identification of individuals, (ii) memories of past interaction and reliability of these individuals, (iii) some way of detecting cheating or likelihood of cheating, and (iv) some motivation to punish or exclude cheaters (Trivers, 1985). Many experimental results have confirmed that we find such capacities in humans. A spectacular demonstration of the importance of cheater-detection is Cosmides’ series of experiments on the logical task of finding out what situation could validate or refute an ‘IF p THEN q’ statement (Cosmides, 1989). This simple ‘selection task’ proves very difficult for most subjects, however familiar the format of the problem, with one exception. Cosmides showed that subjects readily solve the problem when the formulation of the rule evokes a social contract (‘IF someone gets a benefit THEN she must pay a cost’).

[4] A repertoire of moral feelings. The existence of moral feelings, like a disposition to honesty in exchange, an emotional preference for fair dealings, a disposition to feel deep anger at the mere suggestion of cheating, is not directly opportunistic. That is, it creates many situations where people forgo immediate benefits of relatively safe cheating or profitable dealings with cheaters. However, as some economists have argued, there is reliable evidence that such moral feelings are beneficial in long-term co-operation (Frank, 1988). In particular, they create emotional states that are not voluntarily controlled and whose signals are difficult to fake. Such signals advertise reliability and therefore to a certain extent make it possible to by-pass commitment problems.

[5] Emotional rewards for gossip. Humans spend a great deal of time acquiring information on other people. The ubiquity and importance of gossip suggest that there is an evolutionary disposition to pay attention to information concerning others, in particular in domains of great adaptive value such as sex, resources and status. Because we live in groups and depend crucially on information, and because most of the crucial information is information about other people’s actions and intentions, we are predisposed to enjoy communicating about third-party actions and intentions, although the information conveyed may be of no immediate benefit. This also explains why gossip is as universally despised as it is enjoyed. It undermines our great efforts at ‘impression management’ and threatens our control over the information we want to transmit about our own actions and intentions.

[6] Easily triggered self-deception. The difficulty of efficient trickery is somewhat offset by another mechanism that is pervasive and extensively documented in experimental social psychology, and perhaps misleadingly labelled ‘self-deception’ (Trivers, 1985). When it would be advantageous for an agent A if another agent B perceived a situation in a particular way, we find that A often tends to feel committed to this way of perceiving the situation. Controlled experiments show that people tend to adapt their way of seeing situations to the construal that it would be in their interest that others adopt. This is not ‘self-deception’ in a hypocritical or manipulative sense, since such perceptions of situations are spontaneous and genuine. Yet they constitute a strategic mechanism that allows sincere communication of a desirable perception.

This description of specialised mental machinery explains how people assess particular situations and use evolved dispositions to help them make (roughly) optimal choices given the circumstances. But people do not consciously use such computations when evaluating situations of interaction. We may have an emotional distaste for cheating but we do not represent that this kind of moral feeling is in fact a highly adaptive trait. That there are mechanisms that help us maintain cooperation is beyond doubt; but we have very poor explicit concepts to justify the choices we actually make.

Hirschfeld coined the term ‘naive sociology’ to describe the set of intuitive understandings brought to bear on the social domain. His experimental studies demonstrate early-developed, intuitive concepts of social groups such as kin-based units and ethnic categories (Hirschfeld, 1993; Hirschfeld, 1994; Hirschfeld, 1996). Intuitive expectations guide our inferences about such groupings and make certain kinds of cultural representations particularly easy to acquire and transmit. In this domain as in the ontology of material objects, we find both intuitive principles and some consciously accessible concepts. The kinds of strategic computations described above are handled by automatic processes that focus on one particular kind of information, yield a particular assessment of a situation and particular emotional preferences. They can be investigated and demonstrated experimentally, yet they remain largely inaccessible to conscious access. Social psychologists have observed this in many different domains of social interaction. People for instance evaluate potential mates in terms of very specific criteria, for which evolutionary biology has some good explanations (Buss, 1989). But they do not always represent these computations. What is accessible to consciousness is typically a feeling that a certain person is attractive, or that another one has something missing that excludes them from the list of possible partners. The same applies to matters of trust and co-operation. People feel that some people are reliable or not. They do not need consciously to represent the complex cues that lead to such intuitions, although experimental situations show that computations of reliability are complex and principled (Bacharach & Gambetta, 1999).

Strikingly, efficient intuitive judgements in this domain are accompanied by very vague and poor explicit concepts. In all human groups we find that people have consciously accessible concepts of social relations, folk-theories about how they are built and maintained, culturally specific ways of construing them. But these accessible concepts lag far behind the intuitions they are supposed to explain. As I said, we readily assume that specific qualities of people we encounter are sufficient to explain our attitudes towards them, when decades of experimental social psychology show that implicit computations and judgements were involved all along. The crude approximations of naive sociology are even more conspicuous in the description of group dynamics. Whenever complex interaction is involved, we ‘mythologise’ and consider social groups or parts of these groups as big agents (‘the committee thought that…’, ‘society will make sure that…’). This allows us to use our intuitive psychology to describe group behaviour, although practice shows that this gives us very little predictive and explanatory power.

To sum up, we have (roughly) adequate social intuitions and a very weak set of explicit social concepts. This discrepancy between intuitions and concepts is not specific to sociological problems. We find a similar situation in intuitive physics. We have intuitive principles that help us predict the motion of solid objects, the consequences of collisions, the right amount of push need to throw a stone at a target. Yet our explicit concepts of ‘force’ and ‘speed’ are hopelessly poor in their content and often downright false. Cognitive models of co-operation, however much we back them with relevant empirical evidence, always seem rather alien. They are just different from the way we represent social interaction. This ‘we’ includes social scientists, who are human beings too.

Strategic information and imagined strategic agents

Co-operation creates situations where people have to make strategic decisions, that is, choices the value of which depends on the information, intentions and actual moves of other participants. It is a moot point, whether humans are equipped with very general heuristics that evaluate all such strategic decisions, or with a host of specialised devices that keep track of particular cues (facial cues, reputation, consistency, etc.), or with strategic principles specialised by domain (kinship relations, social exchange, mate-choice, etc.). What is certain, on the other hand, is that some specialised machinery keeps track of that domain of information about the world and others that is relevant to strategic decisions. Let us call this the domain of strategic information.

A major difference between humans and most other species is that there is no easy way to sort out strategic information from all other available information. Among chimpanzees for instance, specific signals indicate a willingness to engage in peaceful co-operation (grooming) or to challenge some other individual’s status (shaking branches and emitting special cries). In monkeys, alarm cries are not just segregated from other voicings but specialised in different kinds of hazards. But in humans, any information or behaviour could turn out to include strategic information. It all depends on inferences derived from that specific information by the specialised machinery described above. These inferences are in turn based on a representation of ongoing social interaction. Depending on this representation, that I have meat in my refrigerator may be non-strategic to you (in most cases) or strategic (if meat was stolen from your pantry, if you are hungry, if I always declared I was a vegetarian). That you went to the next village may be non-strategic (if all I infer is that you were away) or strategic (if I think you went there to meet a potential sexual partner). That you talked with So-and-so may become strategic if I suspect that the two of you are involved in some plot against me or a possible coalition with me. There is no a priori way to determine whether a given piece of information has a strategic value and what that value is. Both decisions are made by the specialised cognitive dispositions I described above. So the definition of strategic information can be expressed in purely functional terms:

• Strategic information is that subset of all the information currently available (about a particular situation) that happens to activate intuitive mental processes that regulate cooperation.

In other words, strategic information is not a natural kind of information. The distinction between strategic and non-strategic is all in the eye of the beholder (who may be wrong).

We can now return to the question, Why are associations with morality, group-identity, emotion and ritual generally restricted to a sub-set of possible supernatural concepts? Taking the strategic dimension of communication into account affords a better understanding of how psychology-based concepts (that is, corresponding to templates that activate intuitive psychology) are actually transmitted and why. This requires that we go a bit further in our description of these concepts. What are the contents of the thoughts and intentions attributed to imagined agents? In other words, what do people think spirits and gods think about? Remember that we are talking here about people’s actual representations rather than an officially sanctioned theology. The question is not, What do local ideologies tell you about the spirits’ thoughts? but What are people’s actual assumptions about these thoughts? Also, we must not assume that all these assumptions are accessible to conscious inspection. We are generally not aware of the complex computations of our intuitive theory-of-mind, we are only aware of their output. The same applies to how we intuitively represent what imagined agents know or want.

Imagined religious agents are in general implicitly credited with good access to information. That they appear at several places at the same time or become invisible, often gives them the possibility to acquire information that real agents have more difficulties acquiring. I do not want to suggest that such agents are always considered to be wiser than mere mortals. Actually, we know of many cases where they are represented as intrinsically stupid. So the point is not that they know better. The point, more simply, is that they often seem to know more. Indeed, in the many narratives (anecdotes, memories, myths, etc.) that include such agents as well as human ones, the scenarios in which a religious agent has information that a real agent does not possess greatly outnumber descriptions of the converse situation. So far, this is more or less what local ideologies tell us and what people would readily mention as special properties of religious agents. God knows more than we do, the ancestors are watching us.

Now I want to introduce a further specification that is not explicitly represented but is, in my view, crucial to the acquisition and cultural survival of religious concepts:

• In people’s representation of imagined agents, it is tacitly assumed that these agents have access to strategic information (as defined above).

Imagined agents may be represented as having access to other information as well but, I would claim, it is difficult to find concepts of religious agents without access to strategic information. If any aspect of information about a given situation is strategic, then an imagined religious agent is assumed to represent it. To coin a phrase, religious agents are ‘strategic agents’.

Consider this: In most local descriptions of spirits and other such agents, we find the assumption that they have access to information that is not available to ordinary folk. But this, in general, is in fact information about people’s motivations, their intentions to harm or benefit other people, their actions in that respect, and so on. Ancestors are said to know of people’s behaviour, and so are gods and spirits. In some cases they also know of people’s intentions, in so far as these are good predictors of behaviour. In general, what strategic agents know is what matters to social interaction in a human group.

An interesting limiting-case is provided by those concepts of gods who know everything. In such cases, the theological, literate version of the concept is that the god has access to all information about the world from all possible angles. But that is not the way such agents are actually represented by people, as Barrett and Keil’s experiments demonstrated (Barrett & Keil, 1996). For instance, when Christians say that ‘God knows everything’ they rarely if ever assume that God knows the contents of every refrigerator in the world, or that God perceives the state of every machine in operation, or that he knows what every single insect in the world is up to. If asked, Christians will of course admit that God knows all this. But the fact that such questions are strange shows that the proposition ‘Gods knows everything’ does not produce all the inferences it should produce. In people’s spontaneous imagination, it is generally assumed that God knows primarily about morally relevant aspects of human situations. That is why God may in fact be thought to represent the contents of your refrigerator (if that includes items you stole from your neighbours), the state of some machines (if you use them to harm people) and of insects (if they are a plague you sent your enemies). In such situations the information is strategic. Intuitively, people who represent such situations and represent God’s reactions assume that he represents the strategic information.

More generally, my claim is that, in a given situation, if a subset of information about that situation is relevant to strategic mechanisms, and if people imagine a supernatural agent, then they tacitly assume that the imagined agent has access to the strategic information. Note that this is a spontaneous inference. This quality of agents does not need to be explicitly transmitted. Indeed, I would suspect it is so ‘natural’ that it is rarely if ever transmitted, and rarely if ever explicitly represented.

Selecting strategic agents: causes and consequences

With this plausible conjecture, it is easier to understand the advantage of supernatural concepts that activate intuitive psychology. Let us accept, for the moment, that there is a real selective advantage in cultural transmission for concepts of ‘something with access to strategic information’. This would seem to require the assumption that this something has standard psychological processes. The least costly (and inferentially richest) way of representing ‘something’ in the world that has access to any information is to apply to it our intuitive psychology. This is why people routinely produce semi-serious psychology-based explanations of what computers or photocopiers do. The same goes for strategic information, which is not a different kind of information but just a subset of the available information about a particular situation. If there is ‘something’ that has strategic information, the story that this something has a mind is the simplest one available.

But this does not explain why there should be a selective advantage to such concepts. Why would people want to represent that ‘something’ in the world, over and above the real agents they deal with, has strategic information? The question is, in my view, misguided. The fact that some kinds of notions are selected in does not depend on people’s inclinations but rather on the aggregation of acquisition and memory processes over which they have no control. So a better formulation would be, Why are accounts of such strategically-informed imaginary ‘somethings’ more likely to be acquired and transmitted than other possible accounts?

This, as in the first ‘selection’ described above, is a question of differential cognitive effects. Some pieces of information are more relevant than others. This requires three conditions. First, these different pieces of information must be couched in a format that can be handled by some mental mechanism (relevance is relevance to some particular mechanism and memory store). Second, more relevant pieces of information are the ones that generate more inferences when combined with stored information in that mechanism and/or require less processing to generate inferences. So let us see how concepts of strategic agents fare in terms of these requirements.

Obviously, the first requirement is met since there are special mechanisms that regulate co-operation and track information that could have consequences in that domain. Also, it is quite plausible that concepts of strategic agents generate more inferences than concepts of non-strategic agents. In most situations of social interaction, we need access to other people’s actions and intentions, but we also need to protect ourselves by broadcasting only a certain description of our own intentions and actions. A strategic agent typically sees through all this and has access to real actions and intentions, rather than the public version. Imagining such agents creates a distinct representation of any situation. Note that this description from an imagined agent’s viewpoint may be completely wrong. What matters here is not that it is true but that it is richer in inferences than the description yielded by a concept of non-strategic agent. Finally, the cost of such inferences is minimal since it is achieved by ordinary intuitive psychology processes applied to the imagined agent. So the hypothesis could be summed up as follows:

• Cultural material that includes concepts of imagined agents is made more relevant by the tacit assumption that these agents have access to strategic information.

This could make sense of the variety of features that make up the family resemblance of ‘religious’ activity, in particular the association with moral understandings, with potential emotional arousal, and with group-identity. In this article I cannot do more than outline these connections.

First, the assumption provides a straightforward connection with moral concepts. We find link between some supernatural concepts and moral prescriptions in many human groups, although the way it is explicitly described is highly variable. In some places it is a theological connection (the gods laid down the rules we live by). In most human groups it just goes without saying that ancestors and spirits and other such agents are concerned with the way people behave and use their powers against those who violate moral prescriptions. It may be tempting to think that this connection results from cultural ‘axioms’ or that morally relevant supernatural agents are useful for social cohesion. But there is a simpler account. Supernatural ontology occurs in minds that have particular strategic interests; moral rules are represented by minds with special dispositions for the handling of strategic situations. These two domains of representations stem from adaptations to social interaction. They are intuitively connected because they largely overlap in the cognitive processes that they activate. But this can be true only inasmuch as religious agents are represented as strategic agents.

Second, the connection between religious agency and group-identity, though it is far too complex to examine here, may be better understood in the light of the present description of strategic agents. Mental mechanisms that regulate co-operation are involved in evaluating a realistic level of trust, given kin-based and ethnic distance between partners. If concepts of strategic agents are (in part) signals of reliability, we should predict, not necessarily that people will trust potential partners who share their religious commitments, but that they should distrust people who do not.

Third, at least some of the emotional arousal that is sometimes connected to religious agents makes more sense in the context of strategic agency connected to moral understandings and cooperation. For instance, we observe that people often pay considerable attention to (and often resent) the fact that others are not committed to particular beliefs. This may be less surprising if we recall that adaptations for cooperation are themselves emotional mechanisms, and that clues (however indirect) of non-cooperation have notable emotional effects.

More generally, we often think that people have concepts of ‘powerful’ supernatural agents and connect such agents to socially important phenomena like morality, ritual and group-identity because of that alleged power. This is also how many people explicitly describe their religious notions. From a cognitive viewpoint, we know that such explicit connections are often based on post-hoc rationalisation. Indeed, in this case things seem to work the other way around. We have evidence that specialised mental mechanisms track cues for cooperation, and highlight information that is relevant to cooperation (strategic information). We also have evidence that these mental mechanisms are tightly connected to moral feelings and to group-identity. The present proposal is just that some products of supernatural imagination are particularly relevant to these mental mechanisms, and this is consistent with anthropological evidence concerning what people think gods and spirits know. So imagined agents may not be connected to social interaction because they are powerful; they may be represented as powerful because they were tacitly represented in a way that connects them to adaptations for social interaction.

Functional origins of religion

What is the origin of religion? We find a plethora of functional answers in anthropology and the philosophy of religion. Religious concepts, it is argued, came from a need to understand various natural phenomena, from an urge to comprehend puzzling mental events like dreams, from the need to keep society cohesive, from a desire to make the world more human-like, from anxiety triggered by our mortality, from childhood concepts or from people’s superstitious proclivities. As we teach our students, none of these answers is really satisfactory, for several reasons. First, they are all untestable, since they generally refer to historical scenarios for which there is no conceivable evidence. Second, they are often ethnocentric, describing cognitive and emotional aspects of religion that are not really general of human societies. Third, they project onto some original context a variety of questions, explanatory needs and feelings observed in cultural settings where people already have religion. Indeed, one may suspect that such questions (e.g. What is the ultimate meaning of the Universe? How can we save our souls?) make sense only in cultural contexts where people are given the answers to start with. Fourth, most classical accounts suffer from explanatory over-kill. They predict a whole variety of possible concepts that could fulfil the functions described. But we only find a small subset of those in the anthropological record.

Progress in our knowledge of evolved human capacities affords us with a better account that is testable, based on what we find in the anthropological record rather than in familiar settings, and that predicts only the narrow range of concepts observed rather than a variety of other possible concepts. The arguments presented here are based on experimental evidence for the special relevance of particular concepts. Cultural concepts are mostly triggered by communication, so that which ones (or which variants) survive many cycles of transmission is a consequence of the inferential processes whereby fragmentary cues are completed and re-described using resources in memory (Sperber & Wilson, 1995). But these resources are themselves informed by stable evolutionary and environmental pressures. It is not too surprising that we find roughly similar intuitive ontological categories and principles. These make some types of imaginary situations more likely to be selected in iterated transmission. It is not suprising either that we find specific resources for strategic interaction, as should be expected in a complex social species. These resources make some versions of supernatural concepts more relevant than others. In both of these selection processes, all we need to produce relative stability of some concepts is a slight advantage for concepts that afford rich inferences at a minimal cost. The processing cost is low when the inferential resources are already there, which is the case for both supernatural selection and selection of imagined strategic agents.

This is a functional account in the sense that we explain some recurrent features of cultural representations in terms of what effects entertaining these representations can have on human minds. Functional accounts are not very popular with anthropologists anymore. Functionalism is studied as a period piece more than a viable research programme. This is partly for bad reasons. Functionalism has been found guilty of ignoring many social institutions with no clear function at all; peddling ad hoc stories (anyone with enough ingenuity could find a social function for any cultural institution); making historical change incomprehensible; depicting societies as harmonious organic wholes which is obviously false. These were all criticisms of specific versions of functionalist explanations rather than the programme itself. A more fundamental criticism was directed against the teleological assumption that cultural institutions were around because of their effects.

However, the real problem with this, in my view, was not so much that functionalists exaggerated the functional aspects of cultural institutions as the opposite, that they failed to go far enough in exploring these aspects. This is clear in a comparison between anthropological functionalism and scientific programmes founded on serious functionalist arguments, like evolutionary biology (McCauley & Lawson, 1984). Evolutionary thinking uses ‘reverse engineering’ as an initial rule of thumb: To understand the evolution of a particular organ, try and work out what it does, what it is for. But biologists do not stop there. First, they specify a process whereby a particular functional design could originate in other designs. In biology, most of this work is done by mutations and genetic inheritance. Second, they specify a selection mechanism, generally in terms of differential effects of variants on reproductive potential.

Now functionalist anthropology had neither the mutation nor the selection mechanism. It could not say how particular institutions would emerge out of other ones, and it had no clear way of explaining why supposedly ‘functional’ institutions would survive better than non-functional ones. All it could say was that ‘society’ would eventually make the right choices. This rather implausible claim was grounded in the ‘naive sociology’ described above. We anthropologists often use this weak ontology uncritically, assuming that anthropomorphising groups is a good way of thinking about social relations.

But things have changed in the social sciences. We do not make mistakes anymore. So we can explain the recurrence of particular types of concepts in religion in a more precise way. Some types of concepts are more frequent than others in religion. They are the ones that pass through the filters of conceptual and strategic selection. Religious concepts are not around because they are good for people or for society or because of an inherent need or desire to have them. They are around because they are more likely to be acquired than other variants. A benefit of this functional account is that it explains why we find these religious concepts and not others. There is a cost, too, since we have to provide independent evidence for the cognitive processes involved in cultural selection. We have some such evidence already. Much remains to be done, but progress in our understanding of the mind is gradually transforming many mysteries of culture into mere problems, including that of the origins of religion.





Atran, S. (1990). Cognitive Foundations of Natural History. Towards an Anthropology of Science. Cambridge: Cambridge University Press.

Atran, S., & Medin, D. (Eds.). (1999). Folkbiology. Cambridge, MA: The M.I.T. Press.

Bacharach, M., & Gambetta, D. (1999). Trust in signs. In K. Cook (Ed.), Trust and Social Structure . New York: Russell Sage Foundation.

Barrett, J. L. (1996). Anthropomorphism, Intentional Agents, and

Conceptualizing God. Unpublished PhD dissertation, Cornell University.

Barrett, J. L. (1998). Cognitive constraints on Hindu concepts of the divine. Journal for the Scientific Study of Religion, 37, 608-619.

Barrett, J. L., & Keil, F. C. (1996). Conceptualizing a non-natural entity: Anthropomorphism in God concepts. Cognitive Psychology, 31, 219-247.

Bartlett, F. C. (1932). Remembering. A Study in Experimental and Social Psychology. Cambridge: Cambridge University Press.

Bloch, M. (1985). From Cognition to Ideology. In R. Fardon (Ed.), Power and Kowledge. Anthropological and Sociological Approaches . Edinburgh: Scottish Academic Press.

Bloch, M. (1998). How we think they think. Anthropological Approaches to Cognition, Memory and Literacy. Boulder, CO: Westview Press.

Boyer, P. (1994). The Naturalness of Religious Ideas: A Cognitive Theory of Religion. Berkeley/Los Angeles: University of California Press.

Boyer, P. (1996). What Makes Anthropomorphism Natural: Intuitive Ontology and Cultural Representations. Journal of the Royal Anthropological Institute (n.s.), 2, 1-15.

Boyer, P. (1998). Cognitive Tracks of Cultural Inheritance: How Evolved Intuitive Ontology Governs Cultural Transmission. American Anthropologist, 100, 876-889.

Boyer, P., & Ramble, C. (9999). Cognitive Templates for Religious Concepts: Cross-cultural Evidence for Recall of Counter-Intuitive Representations. .

Bullock, M., Gelman, R. & Baillargeon, R. (1982). The development of causal reasoning. In W. J. Friedman (Ed.), The developmental psychology of time . New York: Academic Press.

Buss, D. (1989). Sex Differences in Human Mate Preferences: Evolutionary Hypotheses Tested in 37 Cultures. Behavioral and Brain Sciences, 12, 1-49.

Cosmides, L. (1989). The logic of social exchange: Has natural selection shaped how humans reason? Studies with the Wason selection task. Cognition, 31, 187-276.

Ekman, P. (1985). Telling lies. New York: Norton.

Frank, R. (1988). Passions within Reason. The Strategic Role of the Emotions. New York: Norton.

Gambetta, D. (1994). Godfather’s gossip. Archives Européennes de Sociologie, 35, 199-223.

Guthrie, S. E. (1993). Faces in the Clouds. A New Theory of Religion. New York: Oxford University Press.

Hirschfeld, L. A. (1993). Discovering Social Difference: The Role of Appearance in the Development of Racial Awareness. Cognitive Psychology, 25, 317-350.

Hirschfeld, L. A. (1994). The acquisition of social categories. In L. A. Hirschfeld, & Gelman, S.A. (Ed.), Mapping The MInd: Domain-Specificity in Culture and Cognition . New York: Cambridge University Press.

Hirschfeld, L. A. (1996). Race in the making: cognition, culture and the child’s construction of human kinds. Cambridge, MA: The M.I.T. Press.

Hirschfeld, L. A., & Gelman, S. A. (Eds.). (1994). Mapping The Mind: Domain-Specificity in Culture and Cognition. New York: Cambridge University Press.

McCauley, R. N., & Lawson, E. T. (1984). Functionalism reconsidered. History of Religions, 23, 372-381.

Piaget, J. (1954). The Child’s Construction of Reality. New York: Basic Books.

Rochat, P., Morgan, R., & Carpenter, M. (1997). Young infants’ sensitivity to movement information specifying social causality. Cognitive Development, 12, 441-465.

Schelling, T. (1960). The Strategy of Conflict. Cambridge, MA: Harvard University Press.

Sperber, D. (1985). Anthropology and Psychology. Towards an epidemiology of representations. Man, 20, 73-89.

Sperber, D. (1996). Explaining Culture: A Naturalistic Approach. Oxford: Blackwell.

Sperber, D., & Wilson, D. (1995). Relevance. Communication and Cognition [2nd edition]. Oxford: Blackwell.

Sperber, D. S., Premack, D., & James-Premack, A. (Eds.). (1995). Causal cognition. A multidisciplinary debate. Oxford: Clarendon Press.

Tomasello, M., Kruger, A. C., & Ratner, H. H. (1993). Cultural Learning. Behavioral and Brain Sciences, 16, 495-510.

Tooby, J., & Cosmides, L. (1990). The past explains the present: Emotional adaptations and the structure of ancestral environments. Ethnology and Sociobiology, 11, 375-424.

Tooby, J., & Cosmides, L. (1992). The Psychological Foundations of Culture. In J. Barkow, L. Cosmides, & J. Tooby (Eds.), The Adapted Mind: Evolutionary Psychology and the Generation of Culture . New York: Oxford University Press.

Trivers, R. (1985). Social evolution. Menlo Park, CA: Cummings.

Walker, S. J. (1992). Supernatural beliefs, natural kinds, and conceptual structure. Memory and Cognition, 20, 655-662.