Religious Ontologies and the Bounds of Sense:

A Cognitive Catalogue of the Supernatural

Pascal Boyer

Center for Advanced Studies in the Behavioral Sciences
202 Junipero Serra Blvd
Stanford, CA 94305


Religious ontologies consist of mentally represented assumptions about the identity and powers of stipulated supernatural entities and agencies. Such representations seem culturally specific and it is often assumed that their variability is unbounded. However, it is possible to identify recurrent types and to explain the recurrence by using experimental evidence from developmental psychology. Ontological categories acquired at the earliest stages of conceptual development constitute an intuitive ontology. This is a limited set of expectations concerning particular domains of experience. Most religious assumptions gain psychological salience by violating expectations derived from that intuitive ontology. One may therefore speculate that there is a limited "catalogue" of potentially successful religious ontologies. This psychological prediction can be tested against anthropological evidence for the recurrence of particular assumptions in diverse cultural environments.

Religious Ontologies and the Bounds of Sense:
A Cognitive Catalogue of the Supernatural

Religious ontologies seem to display striking cross-cultural differences. As a result, cultural anthropologists are on the whole rather wary of general propositions on religious representations. Here I will contend that it is in fact possible to put forward a general explanatory framework that makes sense of the apparent variability but also of the recurrent features of religious representations. This requires that we focus, not just on the categories as described in anthropological monographs, but also on cognitive processes underlying their acquisition and representation by individual subjects. Here I will consider the limited domain of mentally represented ontological assumptions, that is, partly tacit assumptions concerning the existence and specific causal properties of supernatural entities. Other types of assumptions are involved in the representation of religious categories, to do for instance with the specific powers of particular human agents (priests, shamans, etc.), with general ideas about ritual action, with particular memories of such actions, etc. I will leave aside these domains, and show that the ontological assumptions which constitute the conceptual core of religious representations belong to a few recurrent types. Their variability is bounded by universal properties of the human mind which make certain conceptual representations much easier to acquire than others.

The argument here starts from the fact that human minds are predisposed to build particular types of conceptual structures, from the earliest stages of conceptual development. These conceptual structures in turn constrain the inferences produced on the basis of experience and cultural input. In this way, cognitive processes impose constraints on the range of cultural representations likely to be acquired, stored and transmitted, and therefore likely to be found in diverse cultural environments (see Boyer 1994 for a general presentation). This explains that religious ontologies tend to display a limited number of recurrent types, why there is a limited "catalogue" of religious ontological assumptions. My aim here is not to offer yet another classificatory scheme for religious representations, but to show in what ways universal properties of human minds limit the range of "culturally viable" representations.

1. Varieties of ontological assumptions
An ontology postulates a certain number of kinds of objects in the world as well as the fundamental properties that differentiate them from other kinds. Ordinary ontological categories denote such broad domains as events, physical objects, animate beings, abstract objects, persons, plants, animals, etc. Obviously, such categories are not just labels for classes of objects thrown together for no good reason. All ontological categories are based on (generally tacit) assumptions about the specific characteristics of the different classes that are distinguished. Similarly, religious categories are based on a series of assumptions about the existence and the specific causal powers of various classes of entities. It may be of help at this point to give a few examples of the type of representations I am talking about, before dealing with their cognitive underpinnings and their proper theoretical explanation.
Let me start with the most common type of religious assumption, which postulates the existence of non-human agents, clearly marked off from ordinary human beings. I will take here an example from my own fieldwork, although the features relevant to the argument are not specific to that particular case. Among the Fang of Cameroon, the "ghosts" or "spirits" (bekong) are commonly characterised as the immaterial presence of dead people (see Boyer 1992; 1994: 91-124 for more detailed descriptions). Some features are central to the characterisation of the bekong. First, the ghosts have extraordinary physical properties. They are invisible, probably immaterial, can easily go through physical obstacles and usually move extremely fast. Also, they have powers that set them off from most other kinds of beings. They can throw tiny "darts" at people, which pierce the skin and "thicken" (i.e. poison) the victim's blood. These powers, and notably the physical properties of the ghosts, are clearly and explicitly construed as out of the ordinary, as what makes the category bekong of particular interest.

Such intentional agents with extraordinary physical properties, are given various names in anthropological descriptions (gods, spirits, souls, etc.); the specific differences should not make us neglect the common features of the genus, in particular the combination of intentional action with non-ordinary physical properties. This type of assumption is so common that some anthropologists, after Tylor, would take it as the most central form of religious representation; I will return to this point at the end of this paper. The range of possible religious ontologies is of course not limited to such agents, as the following examples will show.

Another familiar feature of religious ontologies is the assumption that some inert physical objects have some of the typical properties of animate beings, among which intentionality is central. Consider for instance W. James's account of "ebony divination" among the Uduk-speaking peoples of Sudan (James 1988). The main ritual in this recent and successful cult centres on the interrogation of wands made from branches of ebony. The wands are first kindled on a fire, and then held over a gourd of water. Divination messages are "read" in the way the wand burns and in the smudges on the surface of the water. The crucial assumption here is that the wand gives messages simply by "telling" what the tree from which it was made, has overheard:
"It is in the nature of ebony, people say, as it grows wild in the forest or bush, to 'hear' signals about what is going on in the human world, and it is these secrets which are revealed in the consultation" (James 1988: 10).

People assume that this is a particular feature of ebony trees, which makes them different from other plants:

"Ebony growing far out in the woodland between the hamlets will know of grumbling over debts it will know of the actions of the arum [souls, spirits, including people who were not given a proper burial] and of dhatu/ (witches) and other sources of psychic activity" (James ibid.: 303).

The trees are not just dumb recording devices, however. During the consultation, the ebony stick produces particular smudges in the water; these not only indicate the nature of the problem at hand, but also a solution, for instance by directing the diviner to the place where a particular soul (a "Genius" in James's terminology) is held, separated from the person (James ibid.: 303-305). The concept of "listening ebony" seems to bridge the gap between the ordinary, non-religious categories of plant and person: something that is identified as a member of a particular natural kind is given standard features of intentional beings: perceptions, thoughts, memories and communicative intentions.

In many cultural environments, this projection is directed at artefacts. Many religious traditions include some assumptions concerning the intentional features of particular religious artefacts. As an illustration, consider for instance this description of shamanistic ritual among the Cuna of Panama:
[The shaman's] song is chanted in front of two rows of statuettes facing each other, beside the hammock where the ill person is lying. These auxiliary spirits drinks up the smoke, whose intoxicating effect opens their minds to the invisible aspect of reality, and gives them the power to heal. In this way [the statues] are believed to become themselves "diviners". From the Cuna point of view, the shaman ritually regenerates the "power of vision" of his statuettes. He also speaks both like them and for them their secret and incomprehensible tongue. (Severi 1993: 231)

To turn to another type of assumption, consider the connections established in some religious systems between biological properties of live things and the structure of non-live natural objects. In his account of the religious representations of an Andean community, Bastien for example (Bastien 1978) describes how a particular mountain is construed as a live body, with a trunk, a head, legs and arms. Different villages, but also valleys, fields and rivers are described as different parts of the mountain's body. This spatial metaphor is also relevant to exchanges between the different villages, and to the organisation of ritual activities. The mountain is represented as having physiological properties; it "bleeds" for instance, and also "feeds" on the meat of sacrificed animals that are left in particular places. Sacrifices of llama's hearts or foetuses for instance are made to the mountain, and left in special shrines to "feed" its body, supposedly in exchange for the fertility of the fields. Diviners "pump [sacrificial] blood and fat, principles of life and energy, to the rest of the mountain's metaphorical body" (Bastien 1978: 37). A whole domain of ritual acts and explanatory assumptions are here based on assumptions that transfer properties of live organisms to an inert natural object.

Finally, I must mention a less common, but more spectacular form of religious speculation, with an excerpt from Metraux's classic analysis of Voodoo in Haiti:

Zombi are people whose decease has been duly recorded, and whose burial has been witnessed, but who are found a few years later living with a boko [sorcerer] in a state verging on idiocy. The zombi are the living dead ­ corpses which a sorcerer has extracted from their tombs and raised by a process which no-one really knows [Ša zombi] moves, eats, hears what is said to him, even speaks, but he has no memory and no knowledge of his condition. (Metraux 1959: 280-1).

This notion in a sense constitutes the converse of the notion of spirits described above. Spirits are ontologically peculiar in that they seem to be persons without being physical objects. Zombies have material bodies but they seem to have lost some crucial properties of persons, notably intentions and self-awareness.

These examples, however cursory, convey some of the conceptual diversity to be expected in the domain of religious categories. They also make clear that how ontological assumptions are at the core of such categories. What makes the Cuna statuettes special, for the Cuna, can be described as a special combination of features of two distinct ontological categories, that of artefact (the statuettes are artefacts) and that of person (the statuettes are described as intentional agents); the same remark applies to the ebony sticks used in Uduk divination. The Aymara mountain, too, combines features of two different ontological domains, that of live beings and that of inanimate objects. Zombies are remarkable in the apparent violation of a central feature of the category person, possession of self-awareness and volition. The same applies very generally to concepts of gods and spirits, except that in this case it is another typical property of person, the possession of a body that is a solid physical object, that seems to be violated. In each of these examples, and I would assume in most types of religious categories, it would be difficult to describe the categories without making use of some broad ontological categories and their particular combinations. A straightforward consequence is that we cannot really describe "religious ontologies" unless we have a good description of non-religious ones. Notions like artefact or animal seem hardly mysterious or problematic; the cognitive processes whereby they are acquired and represented may not seem likely to generate difficult problems or indeed great intellectual wonders. As it happens, familiarity is rather misleading here. As we will see, the representation of ordinary ontological categories is a rather complex affair; moreover, it has direct consequences for a plausible explanation of religious categories and their cultural transmission.

2. Intuitive ontology in conceptual development
Ontological assumptions are part and parcel of ordinary, everyday cognitive processes. Even the most trivial activities require a set of fast, automatic decisions about the identification objects in the environment, as well as the inferences made about them. Subjects may not, and in fact need not be aware of these ontological categories and principles. This is why their content and organisation can best be revealed by experimental studies, and by developmental studies in particular. These show how intuitive ontologies develop early, before subjects can formulate explicit, integrated conceptions of the types of objects that can be found in the environment. Moreover, such studies demonstrate the constraining function of intuitive ontologies. They make conceptual acquisition possible by restricting the range of inferences that can be made about given objects, by orienting subjects towards certain preferred inferences.

The development of intuitive ontologies requires the operation of two sets of cognitive mechanisms, which are to a certain extent functionally distinct. First, there are a number of ontological categories, automatically and tacitly activated in the identification of objects. These categories may be perceptually grounded. but this does not mean that their identification procedure is simple or passive. In order to identify something as animal or artefact or person, one must pay selective attention to particular features, but also produce complex inferences on the basis of those features, of their combination, etc. Second, the output of these categorisation processes activates a series of modular inference engines, based on domain-specific principles. These modules specialise in the treatment of particular aspects of objects found in the environment. For instance, an inference engine delivers expectations about the future movement of a solid object, given the force applied to it. Such inference engines are automatically activated and provide the organism with a range of possible inferences about the objects represented, and therefore restrict the range of possible beliefs about those objects. What follows is a very concise summary of the evidence concerning the early development of ontological categories and inference engines.
Broad ontological categories correspond to such domains as persons, artefacts, animate beings, events, abstract objects, etc. There is strong experimental evidence about the presence of such categories from the earliest stages of cognitive development, although early categories are probably more inclusive than adult ones. From infancy, it is quite clear that children can, like most animals, distinguish con-specifics from other types of beings. This is made manifest in face-recognition as well as in a series of typical behaviours and interaction only elicited by the presence of other human beings. Less obvious ontological categories also appear very early, as for instance a distinction between animate and inanimate types of objects (Gelman 1983; Bullock 1985; Richards & Siegler 1986). This seems to be grounded in an early sensitivity to the perceptual difference between self- and non-self-generated movement in physical objects (Massey 1988). With cognitive development, more refined distinctions appear. For example, Mandler and her colleagues have demonstrated the existence of a variety of ontological categories such as plants and artefacts in 18 months children (Mandler 1989; Mandler, Bauer & McDonough 1991)1.

At a later stage of development, pre-schoolers have a complex hierarchy of (mostly unlabelled) ontological categories. This is made manifest for instance by "predicate-spanning" (Keil 1979); if the application of a given predicate is relevant for a certain object, this allows us to predict the possibility of applying other predicates to that object. If for instance an object X, of which we know nothing, is described as "hungry" or "stupid", we may assume that it makes sense to wonder whether it is "omnivorous". but we may also assume that it would not make sense to wonder "who made it" or "how long it lasts". As Keil's studies have shown (1979), pre-schoolers spontaneously make use of such predicate restrictions to make inferences even about imaginary objects they know nothing about2. This shows that pre-schoolers have complex hierarchies of nested ontological categories.
The identification of objects as members of particular ontological categories is an automatic, largely unconscious process that is partly based on complex inferences from perceptual cues. As I said above, such features as typical movement provide the starting-point of an animate-inanimate distinction; it is clear that other distinctions (person-animal, plant-artefact) may also be directed by the child's specific sensitivity to perceptual cues. Ontological categories, however, are not just a means to sort out objects; they have conceptual consequences, directing the type of expectations entertained about different objects. This is why we must describe the way ontological categories are connected to early "theories" about different aspects of reality.

Recent experimental studies of children's early intuitive ontologies have shown that children's inferences about particular aspects of the experienced world have two salient features. First, they are to a certain degree under-determined by experience, and appear much earlier than the child's actual interaction with the world would lead us to imagine. Second, they are domain-specific, involving structurally different principles and expectations according to the aspect of reality they focus on. Generating physical expectations from the experienced physical behaviour of objects is not done in the same way as generating biological expectations on the basis of observed biological features. Experimental studies allow us to describe a variety of domain-specific conceptual structures (for a general survey, see Hirschfeld and Gelman 1994). Here I will only mention three structures of particular importance, namely:

(1) an intuitive physical module, which produces explanations and predictions concerned with the physical properties of solid objects;

(2) an intuitive biological inference module, which specifies principles to do with the essential nature of certain biological properties, and their connection to membership of particular, mutually exclusive and jointly exhaustive, classes of beings;

(3) an intuitive intentional explanation module, which infers speculative explanations of behaviour in terms of the existence and causal properties of hidden entities such as beliefs, intentions, perceptions.

Let me first consider physical processes, the domain of the statics and dynamics of solid objects. There is growing evidence that physical principles appear much earlier than classical Piagetian models would assume, and certainly earlier than sensory-motor capacities could warrant. Four months olds for instance expect solid objects to behave according to a principle of continuity (objects move in continuous paths) and solidity (objects do not coincide in space) (Baillargeon 1987; Baillargeon & Hanko-Summers 1990). Also, a principle of support (unsupported objects fall downwards) seems to appear around 6 months (Spelke 1990). Moreover, infants are sensitive to differences in contiguous events, particularly between causal and non-causal versions of contiguous physical events; they also hold a principle of "no action at a distance" (Leslie 1979; 1988). Such early principles form a skeleton on which more complex notions of physical causation can be then developed. As Bullock et al. point out, "the development of causal understanding is more a process of learning where, when and how to apply the rules of [causal] reasoning rather than figuring out what these rules might be" (Bullock 1982: 216).

Another domain of inference concerns specifically biological aspects of live beings. From the earliest age, there is a categorical distinction between "animate" and "inanimate" objects. This distinction is then enriched with a variety of specific intuitive principles which describe biological aspects of live beings. There are several different kinds of principles at work in the child's "intuitive biology". First and foremost, there is a taxonomic principle, following which the space of all species is divided in a number of mutually exclusive, jointly exhaustive natural classes, arranged in a taxonomy (Atran 1991). Second, the identification of living kinds also activates an "essentialist principle" following which an undefined internal principle, which is exclusive to the species, causes the external features of living objects as well as certain aspects of their behaviour. For instance, even pre-schoolers assume that membership of a kind is more important than observable features, in predicting the typical behaviour of an unknown animal (Gelman 1986), (Massey & Gelman 1988; Becker & Ward 1991)3. These theoretical assumptions are also manifest in children's reactions to putative scenarios of artificial transformation from one kind to another. Such changes are judged more plausible between types of artefacts than from one living kind to another, even in cases where an animal is described as having gradually acquired the other's outside appearance or behaviour (Keil 1986). Third, this essentialist understanding of living things leads children to evaluate a number of properties as directly caused by "inner" mechanisms4.

Finally, one of the most salient sets of intuitive principles is found in the intentional explanation of behaviour. From the earliest stages of cognitive development, children readily interpret the behaviour of animate beings, particularly persons, in terms of the causal role of unobservable entities, such as beliefs and intentions. This requires a series of tacit principles about the nature of mental entities and their connections with observable action (see for instance Astington, Harris & Olson 1988; Wellmann 1990; Perner 1991n; Whiten 1991)n. Beliefs and desires are represented as immaterial entities from the age of three (Wellmann & Estes 1986). At the same age, children have a rudimentary understanding of the causal connections between states of affairs and mental representations. They understand that perceptions routinely cause beliefs which (may) cause intentions which (may) result in actions, whilst the reverse causal links are non-standard. They also understand, around the age of four, the possibility for mental representations to misrepresent actual states of affairs, as well as the behavioural implications of false belief (see Wimmer & Perner 1983 for the original findings). All these principles form the transparent, self-evident basis of intentional explanations. Of course children subsequently acquire vastly more complex notions about mental processes and behaviour. Yet these notions never contradict the fundamental principles observed in five year olds.
Particular activation rues connect the identification of an object as a member of a particular ontological category and the activation of some inference engines. For instance, identifying something as an animal results in the activation of the three modules described above, producing definite expectations about physical constraints, the possession of species-specific essential features and the interpretation of the animal's action as caused by intentions and beliefs. On the other hand, identifying the object as an artefact does not activate the intentional action or biological properties modules, whilst the physical one is activated5. The selective activation of particular inference engines is an automatic process, triggered by ontological identification.
The combination of ontological categories and domain-specific inferential modules provides the child and the adult with what I will call an intuitive ontology, which constitutes a basis and a constraining system for further knowledge acquisition. Obviously, knowledge structures could not be developed without the trigger provided by external stimuli. No child will ever develop an understanding of intentional action unless she is in constant interaction with intentional beings; no understanding of biological properties could develop in a environment that did not include live beings. However, the child's early or intuitive "theories" go far beyond the information given, and are structured, from an early age, by complex, quasi-theoretical hypotheses. Obviously, under-determination by experiential input entails under-determination by cultural input, which is only a subset of the child's experience. This is why the principles underlying intuitive ontologies can be found in a similar form in the most diverse cultural environment. For instance, Walker finds a similar developmental schedule for aspects of ontological development in Yoruba and American children (Walker 1985; 1992). Or again, the principles of taxonomic ordering and essentialist understanding of living things are represented in similar ways the world over (Atran 1990). The understanding of mental processes and intentional action, which after all would seem to depend most on particular social interactions, is seen to undergo the same developmental phases in Cameroon Pygmy and British children (Avis & Harris 1991). The content and developmental schedule of ontological categories and domain-specific principles are the outcome of maturational programmes, triggered but not shaped by experiential input.

3. Religious assumptions as counter-intuitive
Returning to the examples of religious ontological assumptions mentioned at the beginning of this paper, we can now see that they are all based on assumptions which go against the expectations delivered by intuitive ontology. They can be called counter-intuitive in that precise sense. The notion of spirits violates an intuitive expectations concerning the category person, namely that persons are physical objects, which obey the principles of solidity or continuity. The notion of a zombie also violates intuitive expectations for person, in that intentional behaviour inferences, to do with perception, thought and intention, cannot apply in the expected way. Conversely, the notion that a statuette or a twig can "hear" what is said in front of them violates intuitive expectations, in that the ontological category artefact does not activate inference engines concerned with intentional behaviour; artefacts are not intuitively construed as having perceptions and thoughts.

A first hypothesis, then, is that religious ontologies generally include counter intuitive assumptions, relative to the expectations derived from intuitive ontological categories and principles. This quality is not really unfamiliar to anthropologists, and probably constitutes the intuitive basis on which we judge that a given set of representations belong to the "religious" domain. However, the counter-intuitive quality of religious representations is generally not discussed at much length in cultural anthropology. One important factor is that we generally tend to consider that apparently counter-intuitive assumptions are in fact perfectly intuitive, given a set of established cultural conventions. In other words, we assume that the criteria of the counter-intuitive might be culturally variable. What is counter-intuitive to you might be rather intuitive to me. However, this widespread anthropological conception seems to be based on several types of confusion, among which the following are particularly important:

(i) A confusion between intuitions and reflexive conceptions. Religious assumptions violate certain intuitive expectations all humans have, about such things as the physical embodiment of intentional agents, the inevitability of death and decay, the impossibility of action at a distance, the fact that living things grow, that dropped objects fall downwards, etc. These expectations are the outcome of principles which need not be, and in fact are generally not penetrable to conscious introspection. Only their outcome, in the form of expectations, is present to consciousness. Now in some (not all) cultural environments, some people have developed explicit, articulated conceptions of what is "natural" and what is not, that is, of what entities and processes constitute the normal working order of the universe. These explicit conceptions vary from one cultural environment to another. There is no experimental evidence that such explicit conceptions can dislodge or radically alter intuitive expectations; there is in fact a lot of evidence to the contrary. What is claimed here is that religious assumptions include counter intuitive representations relative to intuitive expectations, and nothing else.

(ii) A confusion between the counter-intuitive and the unfamiliar. In many cultural environments, people are made familiar, from an early age, with the counter-intuitive characteristics of religious entities. In this sense, they are not especially "surprised" to hear spirits described as invisible or gods as immortal. This, however, is orthogonal to the question whether the assumptions violate intuitive expectations. Intuitive principles, as I will show below, constitute an early developed, automatically activated and relatively inflexible system of expectations. Developing familiarity with counter-intuitive assumptions does not necessarily make them part of intuitive expectations6.

(iii) A confusion between the counter-intuitive and the unreal. For the people concerned, some features of spirits, ghouls, gods, magical objects or imaginary animals are counter-intuitive. This, however, does not imply that they are construed as unreal. If I drop a piece of chalk in front of my lecture-room audience, and the chalk flies up in the air instead of falling downwards, this event will be in all likelihood counter-intuitive, given people's expectations about the physics of solid objects. This will probably trigger a search for possible explanations; for most of my audience, this physical event, however counter-intuitive, has really occurred.

That religious assumptions are counter-intuitive and construed as actually true is probably what makes them of particular interest, and triggers significant cognitive investment on the part of the people concerned. We often tend to consider that ontological categories, and the relevant inferential mechanisms, may well be culturally specific. This conception is not only rather implausible, in the view of the actual experimental data in this domain, but also paradoxical; it makes it very difficult to account for the obvious attention-grabbing quality of religious representations. To return to our examples, consider for instance the Uduk notion of ebony trees as intelligent voice-recorders. One might want to interpret this as the consequence of a culturally specific ontology, in which trees like people can remember conversations. Yet people who hold such beliefs take great interest in the idea that some trees have such memories, and are totally unimpressed by the fact that people do. This asymmetry should suggest that the two assumptions ("people remember what was said in front of them" and "trees remember what was said in front of them") do not really have the same status. Or, to take a less extreme example, consider the Aymara of the Andes, who claim that the mountain on which they live is a live organism, that breathes and eats and digests (Bastien 1978). Again, if the "Aymara ontology" simply assumed that such physiological features are present in animals, people and mountains, one would not expect the extraordinary interest in, and ritual constructions about the mountain's physiology, as opposed to that of people or animals. Not to put too fine a point on it, these representations attract considerable attention, often command a high degree of commitment and may trigger powerful emotional responses, and all this deserves some explanation. Now, whatever that explanation is, it cannot really be founded on the idea that religious assumptions are "intuitive" in the cultural environments considered7.

4. Background principles and inferences
Counter-intuitive assumptions seem to constitute the core elements of religious categories. It is "constitutive" of spirits to be physically counter-intuitive, and of gods to have counter-intuitive biological properties. This is why such aspects are usually central in people's description of their religious categories, and as a consequence central in anthropological descriptions as well. But do such descriptions exhaust the cognitive processes actually involved in representing those categories?

People could not acquire religious ontologies, and make inferences on the basis of the information given by cultural input, without activating a background of additional representations. These form part of the mental representation of the religious categories, yet are not described because they are "transparent" or self evident to both the people concerned and the anthropologists who describe such representations. This background is built by tacitly applying to the religious entities or agencies, the intuitive principles which are not explicitly violated by the information available from cultural input. For example, in most cultural environments where the notion of spirits is relevant, these are construed as intentional agents, as entities with psychological states and processes. Spirits or ancestors are said to perceive actual states of affairs, to form beliefs on the basis of those perceptions, to form intentions on the basis of those beliefs and so on. It is assumed for instance that causal connections obtain from states of affairs to spirits' beliefs, and from spirits' beliefs to spirits' intentions, and not in the opposite direction. Also, spirits are implicitly assumed to think along the lines of certain pragmatic schemes; if the spirits are said to want p and assumed to know that p cannot be achieved unless q, it will be inferred that they want q. Principled psychological assumptions are projected onto the spirits' putative mental states and behaviours. The point applies to other types of religious ontologies. The artefacts described by W. James as "listening ebony" are certainly counter-intuitive in that they are described as having certain aspects of intentional beings. But one must notice that these aspects are themselves represented in a way that is not counter intuitive. As in the case of spirits, all the assumptions concerning the alleged perceptual cognitive activities of the religious agencies are a straightforward projection of ordinary, self-evident understandings of such processes in persons. The ebony trees have perceptions which are caused by external states of the world, like human beings. Also, the way the ebony twigs "communicate" with human beings complies with ordinary, intuitive principles of intentional communication. Finally, for all its bizarre aspects, even the notion of a zombie includes a host of non-counter-intuitive assumptions. Zombies are characterised as persons and this carries the intuitive assumption that they are solid physical objects, with one definite location in space and with the possibility of interacting with other objects in accordance with intuitive physical expectations.

Background assumptions may seem self-evident, and this is probably why they are neither explicitly mentioned by the people concerned nor described by the anthropologist. That they appear self-evident, however, does not mean that they are causally irrelevant. Without such assumptions, people could certainly not make sense of any of the statements, anecdotes, displays and material objects which constitute the cultural input in the religious domain. One cannot acquire much anything of use, in terms of information about spirits for instance, unless one assumes a host of complex assumptions about their psychological, notably cognitive processes. In other cases, these principles consist in notions of physical properties or biological processes. Background assumptions allow subjects to make inferences about religious entities, inferences without which they could not build a coherent conceptual representation of these entities. Whilst counter-intuitive cues account for the salience of religious ontologies, intuitive principles make them learnable.

5. A succinct catalogue of violations and transfers
This implies that we should be able to make definite predictions as to the range and variability of religious ontologies. The claim here is not that religious ontological representations consist in any odd combination of "something counter intuitive" and "something intuitive". My claim is more specific and concerns the connection between the assumptions underlying religious categories and early developed ontological categories and principles. Counter-intuitive assumptions, which provide the attention-grabbing quality of religious representations, are counter-intuitive only in the sense that they violate expectations from intuitive ontology. The various inferences that complete the representations result from the activation of intuitive ontological principles. But intuitive ontology comprises a small number of domain-specific inferential engines applied to a small number of categories, and religious categories constitute a limited departure from intuitive principles. It follows that there can only be a limited number of "possible" types of culturally viable religious ontologies. It should be possible to describe these combinations systematically, and establish the "catalogue" promised at the beginning of this paper. Obviously, the point of the enterprise is not to establish distinctions between types, but on the contrary to show how the apparent variety of religious categories is in fact constrained, if considered in its proper causal context, that of intuitive ontology. This is a speculative hypothesis that derives its plausibility from the psychological evidence but remains to be tested as an explanation of cultural recurrence.

Let us start with the most salient categories and inference modules of intuitive ontology, that is to say, the categories and modules that seem to appear from the earliest stages of conceptual development. Here I will consider five ontological categories: (1) person, (2) animal (exclusive of human beings), (3) plant, (4) natural object (rocks, rivers, mountains, etc. which are non-living, non artefactual discrete objects) and (5) artefact. As for the aspect-specific inference engines, I will consider the three devices described above, which correspond to different aspects of objects and different modes of explanation: the physical behaviour, biological properties and intentional behaviour inference engines.

As I said in the above section, inferences are produced by activating a particular module, in the representation of an object identified as a member of a given ontological category. For instance, once something is represented as a person, all three modules listed above are activated by default. The default activation links for each ontological category are as follows:

­ For person: physical behaviour, biological properties, intentional behaviour;
­ For animal: physical behaviour, biological properties, intentional behaviour;
­ For plant: physical behaviour, biological properties;
­ For natural object: physical behaviour;
­ For artefact: physical behaviour.

Given these categories and inference engines, the point now is to produce counter-intuitive assumptions, beyond the default activation links. From the examples mentioned at the beginning, we can see that there are two possible ways of producing an assumption that violates intuitive expectations. First, there may be a violation of intuitive expectations: a particular representation is entertained, to the effect that the inferences produced by normally relevant inference modules do not apply to the particular object considered. This is what makes spirits for instance counter-intuitive. They are represented (i) as similar to persons in terms of their psychology and (ii) as different from persons in their physical properties. Another way of producing counter-intuitive assumptions is what could be called a transfer of expectations. An object is represented as belonging to a particular ontological category, and an inference module is activated, although it is not part of the modules activated by default for that category. If we imagine that a statuette can hear what we say or tell us what it sees, we construe the statuette as an artefact, and apply to the statuette all the intuitive principles usually applied to the artefact category. But we also apply to the statuette all the expectations derived from the intentional psychology module. So, in order to generate a catalogue of counter-intuitive assumptions in this particular space, we must systematically apply the following principles:

(1) for each inference module that is activated by default for an ontological category, produce a counter-intuitive assumption by introducing a violation of the expectations generated by that module;

(2) for each inference module that is not activated by default, produce a counter-intuitive assumption by transfer of that inference module.

This produces the list of combinations in table [1], which can be glossed as follows:
[1] Person with violation of physical expectations. This is activated in the notions of ghosts, souls and spirits that can be found in most cultural environments, and were discussed at the beginning of this paper.
[2] Person with violation of biological expectations. This, again, is widespread in categories of "gods" or "spirits", represented as persons whose biological properties are exceptional, being immortal, or feeding on the smell of sacrificed meat, etc. This is also found in the notion that mythical heroes have had organs replaced with artefacts. In the Fang epics, the ancestors make a hero immortal by replacing his liver and heart with iron organs (Boyer 1988). The Borasana of Colombia say that powerful shamans can transform into jaguars (Hugh-Jones 1979: 124).
[3] Person with violation of psychological expectations. This is implied in the concept of a zombie, represented as having the ordinary physical characteristics of persons, but with a specific violation of the psychological expectations; zombies are described as deprived of volition and conscious experience. Construed as a temporary state, however, this assumption is very commonly found in the representation of possession states, in which a person is represented as having no intentional control of his or her actions, including communicative action.
[4] Animal with violation of physical expectations. Mythologies the world over mention invisible animals, combining the ordinary "psychological" characters of intentional action that are applied by default to animals, with counter-intuitive physical properties. Among the Fang of Cameroon for instance, some people claim that the ancestors breed animals, which can suddenly disappear when hunted by humans (Boyer 1988).
[5] Animal with violation of biological expectations. The notion of metamorphoses is of course the most salient illustration of this assumption, which violates one of the central principles of intuitive biology, postulating the stability of kind identity. To take but one example, consider the Malay notion that many animals were once human, or were transformed from other animal species (Endicott 1970: 29-30). As Kelly & Keil point out, the range of metamorphoses is constrained by intuitive taxonomic principles (Kelly & Keil 1985). The cultural recurrence of such transformations seems to be a function of ontological proximity. Transformations occur within families more often than between them, rarely go beyond the animal kingdom (e.g. plants) and almost never from animals or humans to artefacts. Chimeras are another example of this particular combination.
[6] Animal with violation of psychological expectations. This does not seem to be very common in any religious system, although cases of possessed animals may be an illustration of this counter-intuitive assumption.
[7] Plant with violation of physical expectations. Some religious systems assume the existence of invisible plants, e.g. an invisible tree a shaman must climb to reach the heavens where souls have been taken. [8] Plant with violation of biological expectations. This applies to the same cases as above, when the plants in question are for instance described as eternal, i.e. not submitted to intuitive expectations of growth and decay.
[9] Plant with transfer of psychological expectations. The example of the "listening ebony" described by W. James (see above) is based on this type of assumption (although some of the people concerned may in fact represent the ebony twigs used in the divination ritual as artefacts). There are of course many beliefs concerning the "souls" of particular plants, e.g. the soul of taro plants among the Orokaiva (Thompson 1949), and these beliefs in many cases include a transfer of psychological expectations.
[10] Natural object with violation of physical expectations. This would be the case for instance of the notions of "floating islands" found in some Polynesian mythologies.
[11] Natural object with transfer of biological expectations. Many religious representations ascribe biological properties to objects such as mountains or rivers. Mountains for instance may be described as "breathing", containing "blood" and feeding on sacrifices (see above).
[12] Natural object with transfer of psychological expectations. Again, notions of "live" objects such as live mountains are sometimes complemented with descriptions of the intentional states of these objects, described as "desiring" certain things or "resenting" certain human actions.
[13] Artefact with violation of physical expectations. This is involved when some artefacts are described as invisible, or capable of effects which violate intuitive physics, such as some forms of action at a distance.
[14] Artefact with transfer of biological expectations. This is particularly important and widespread, as breathing or bleeding statues and other such miracles are found in many religious systems.
[15] Artefact with transfer of psychological expectations. This, too, is both important and widespread. Statues for instance listen to humans (and sometimes answer back). This assumption is also activated in many divination techniques, in which an artefact is described as the originator of statements, as giving a particular diagnosis or prescribing a particular course of action. Together with [1] and [2], this kind of combination can be found all over the world, and generally at the core of religious representations.

6. Cognitive equilibrium and the explanation of recurrence
This way of listing possible religious ontologies may appear unduly abstract. The formulae in this "catalogue" certainly differ from traditional descriptions of religious ontologies. For one thing, they do not correspond to traditional categories in the anthropology or history of religion. The traditional notion of a "god" for instance is split between two assumptions ([1] and [2]), each of which applies equally well to notions of "spirits" or "ancestors". This is probably not a serious problem, since there is no agreement in either anthropology or the history of religion as to the precise extension of such terms. If anything, there is general agreement in cultural anthropology that such traditional terms are no more than vague glosses of local categories, convenient as a kind of conceptual short-hand and misleading as analytical categories. More importantly perhaps, my description is removed from actual representations in at least three different ways:

First, the formulae leave out part of the representations actually associated with religious ontological assumptions. The "catalogue" lists underlying ontological assumptions, not the variety of additional details that accompany them in their actual representation. Fang ghosts for instance, who induce illness in people by throwing magical darts at them, may seem rather different from their Victorian counterparts, who seemed mainly interested in chatting with the living through ouija boards. People who entertain assumptions about religious agencies do not usually construe them in terms of the "raw" combinations listed above, but as detailed descriptions that entail these combinations of assumptions.

Second, the catalogue makes a distinction between assumptions that are often combined in actual representations. In many cultural environments, religious categories activate more than one of the combinations listed above. For instance, many concepts of gods or spirits are based on both [1] and [2], that is, these agencies are made counter-intuitive both by having non-default physical properties and non default biological properties. Greek gods for instance have both a non-default physics, being generally invisible, and a non-default physiology, as they feed on the smell or fumes of sacrificial meats.

Third, some people in some cultural environments may have their own reflexive notions about their religious ontology, and its connections with other intellectual domains. This is the case, obviously, for all human groups where specialised (and usually literate) scholars explicate and codify various aspects of religious representations. The Greeks did not just have beliefs about gods, but also a variety of people producing (and arguing about) descriptions of these beliefs. My list of combinations leaves aside such aspects and focuses on the connection between intuitive ontology and religious representations.

Far from being problematic, these departures from ordinary descriptions are in fact a necessary condition for a realistic explanation of cultural recurrence. Without dwelling too much on general issues of theoretical construction, we must keep in mind that explanations of cultural representations differ, not by the extent to which they leave out rich contextual details, but by the choice of which details are to be left out. For instance, insisting on the cultural differences between Victorian ghosts and Fang bekong entails that we ignore other aspects, like the fact that in both cases the religious ontology assumes the existence of a class of persons with counter-intuitive physical properties and non-counter-intuitive mental capacities. In the same way, we could assume that there is an integrated conception of "spirits" in which their physical and biological properties are integrated; but then we would ignore the fact that these counter-intuitive properties activate different aspects of intuitive ontology. So the question is not whether it is legitimate to be abstract and reductive, since this is what all theoretically meaningful descriptions require, but whether this particular type of abstract description is theoretically productive. The particular list proposed above is not an absolute classification; it is relevant only in the context of a particular explanatory account of cultural acquisition.

The recurrence of particular features of cultural representations can be explained in two different ways, which can be called generative and selective respectively. Generative accounts postulate a common cause for the occurrence of similar representations. Applied to religious ontologies, we could assume for instance that certain human motivations lead inevitably to the development of a notion of spirit or soul; or one could trace such notions to particular kinds of experience; or, in a more general way, one might see the "origin" of religion in the need to provide explanations for phenomena beyond everyday circumstances, as is assumed in intellectualist models. By contrast, the point of a selective account is that we need not consider the "origins" of cultural representations; indeed, a similar representation that is found in different groups may have originated in entirely different circumstances and for entirely different reasons. What matters in the study of recurrent features is that not all representations are equally likely to survive iterated cycles of acquisition and transmission. Some representations are not easily transmitted or acquired, because they have either a low attention-grabbing potential, or do not support rich inferences. Other representations, being particularly easy to acquire or transmit, will be selected by the repeated cycles of transmission. Religious representations that do not combine intuitive and counter-intuitive assumptions may be found in some cultural environments, particularly as speculations produced by imaginative individuals or theologies transmitted with considerable institutional effort. They will rarely survive in the form of those stable, widespread sets of mental representations we call "cultural systems".

The argument presented here, as concerns religious ontologies, is an example of such a selective account. It specifies cognitive conditions such that, all else being equal, representations which activate the particular combinations of intuitive and counter-intuitive assumptions listed above will be more "viable" than others. Religious categories would probably not survive if they did not include some counter-intuitive assumptions that made them salient. In the course of social communication, they would be gradually discarded. Conversely, religious ontologies that make it impossible to activate the background of intuitive principles would also be discarded, because they could not trigger inferential processes that would make them memorable or usable. Specific combinations of counter-intuitive assumptions and intuitive principles can be described as a cognitive equilibrium. All other things being equal, cognitive systems such as human minds will tend to make such states more permanent than others, once they are reached. Or to put it in converse terms, once such a state is reached, a cognitive system will tend to deviate from it less than from other states. In the idealised situation described here, one should expect representations which reach the cognitive equilibrium to be the ones most easily acquired and transmitted, all else being equal. This acquisition process also accounts for the recurrence of similar religious assumptions in very different cultural environments.

7. Predictions on the properties of religious systems
Whether this explanation is satisfactory, at least as a first step, can be judged by considering the predictions it implies. Let me briefly consider four predictions, concerning (1) the relative rarity of religious ontologies outside the cognitive equilibrium, (2) the trend towards the equilibrium even in cases where the input is far from it, (3) the relative frequency of types of ontologies within the list and (4) general features of the acquisition process.

1. The "catalogue" presented here seems to exhaust the variety of ontologies, in the sense that there do not seem to be any religious ontologies that are not based on one or several of the combinations listed here. More specifically, the claim is that the list exhausts the variety of culturally viable religious ontologies, at least in the standard conditions of an effortless acquisition. That is to say, although there may be examples of people or groups that hold ontological beliefs outside this list, the prediction is that such constructions, barring exceptional effects of outside factors such as systematic training or political constraints, will not spread. Note that the claim is not that it is impossible for human minds to conceive other types of ontologies: philosophers, science-fiction writers or physicists constantly imagine worlds far removed from intuitive ontology or even from the simple combinations listed here. These rarely become widely spread "world-views". On the whole, they remain socially marginal, and often require considerable effort and institutional support to be acquired.

2. Buddhism offers the interesting example of a religious ontology systematically organised by a scholarly literate group, and obviously very far from what I described as a cognitive equilibrium. Buddhism is special in that it puts forth a description of the world that does not include supernatural agents, indeed excludes them, and more generally does not include any of the combinations of assumptions listed here. The prediction of the model is that such a set of beliefs either would not spread at all outside the specialised scholars, or else would be "corrected" by non-scholars towards one of the combinations described here. There would be two parallel transmission tracks: painstaking learning of non-equilibrium assumptions in a small group, effortless acquisition of equilibrium assumptions in the larger population. This is indeed a consistent finding of all anthropological studies of Buddhism "on the ground", in which people readily postulate the existence of various superhuman agents, with all the standard default intentional properties. In the same way, compare the popular spread of different aspects of Christianity: most believers hold those beliefs that correspond to equilibrium points, like the invisible presence of God (combination [1]) or immaculate conception ([2]). By contrast, the unity of Trinity does not seem to be part of the core representations of most believers, although scholarly theologians view it as a fundamental tenet of the system.

3. The cognitive model presented here states that religious assumptions are likely to be acquired and transmitted inasmuch as (i) they have some attention grabbing potential (given by violations of intuitive expectations), and (ii) they have some inferential potential (provided by the default background assumptions). One may want to push this argument further; since both attention-grabbing potential and inferential potential are variables, their variation should result in variable salience of religious ontologies. All else being equal, some combinations of assumptions would be more salient than others. They would be more likely to attract an individual's attention, and probably better remembered than other combinations. One would therefore expect those combinations to be more easily transmitted from one generation to the next; for the same reason, one would then expect them to be more widespread in different cultural environments. Now, if we return to the "catalogue" presented above, it is rather clear that the combinations listed are not equally widespread across cultural environments. For instance, one can expect to find the following combinations:

[1] Person with violation of physical expectations. or
[2] Person with violation of biological expectations. and
[15] Artefact with transfer of psychological expectations
in practically every religious ontology, whilst
[3] Person withviolation of psychological expectations
[6] Animal with violation of psychological expectations

are rather uncommon, and rather marginal in the groups where they are found. This is a familiar aspect of religious ontologies. The most frequent type of religious assumption by far is the postulation of intentional agents which are counter intuitive either because they have counter-intuitive physics, or because they are actual artefacts or material objects. This is so widespread and central that such assumptions are often considered the defining feature of "religion". There seem to be many common-sense reasons why postulating intentional agents is more "productive" than imagining, for instance, disappearing rocks or eternal trees. We could say that intentional agents seem more similar to us, are more relevant to human interests, could be more productive for the explanation of mysterious occurrences, and so on. But all these "explanations" are question-begging, for they presuppose the very familiarity they purport to explain.

The salience of intentional agents may be easier to explain in a cognitive framework. As I said above, the model presented here predicts that religious ontologies should be transmitted and therefore recurrent as a function of (i) their attention-grabbing potential and (ii) their inferential potential. Now we do not really have as yet precise data about the relative attention-grabbing potential of various ontological violations, although this could be easily evaluated in terms of effects on attention and memory. On the other hand, we have ample evidence, particularly from developmental studies, of the differences of inferential potential between domain-specific inference engines. The domains of domain-specific expectations (e.g. biological aspects of live beings, intentional psychology, physical constraints) are not equally complex in the range of expectations delivered, nor are they equally prepared to cope with variable situations. In particular, one must notice that the "theory of mind" inference engine delivers more complex expectations than for instance the principles underlying people's intuitive physics. Also, the explanations produced are invariably of an abductive format, which makes each cause compatible with indefinitely many effects. And a theory of mind module deals with a domain in which the causally relevant states and events (perceptions, beliefs and intentions) are all invisible. All these features make it likely that combinations of assumptions which either maintain the activation of a theory of mind module or transfer its expectations to a new domain should be found more often. This is only a sketch of a speculative explanation; but the main point here is that the explanation is based on independent evidence and does not involve circular claim that intentional agents are intrinsically more "interesting" than other types of counter intuitive entities. 4. Inference-engines are activated automatically, and produce stable inferences, on the basis of cues about the ontological categories objects belong to. If the cultural input gives some cue that a particular religious agency is a person, then all the inference engines that are not explicitly excluded will be activated. So no-one needs to be told that spirits for instance form beliefs on the basis of their perceptions, or intentions on the basis of their beliefs, but not conversely. Such inferences are automatically produced by a "theory of mind" module if that module is activated. All that is needed, in order to generate these particular inferences, is some cue that indicates, either that intentional inferences are relevant, or even more simply the absence of any cue that specific ally excludes its activation. More generally, if religious ontologies are represented in the way described here, it follows that acquiring them does not require the transmission of complex, fully-specified "theories", "world-views" or "conceptual schemes". All that is required is (i) some cues as to the ontological identity of the objects postulated, and (ii) some cues as to which parts of intuitive ontology are violated or transferred. Beyond this, the spontaneous activation of intuitive inference modules should be sufficient to build a representation of the religious entities postulated. This predicted mode of acquisition is consistent with what cultural anthropologists report. In most cultural environments, there is no explicit or observable "learning" of the properties of religious agencies; in cases where there is some explicit transmission, the particular assumptions people share vastly exceeds what is being explicitly transmitted. The present model of religious ontologies explains this without implying that culture is "in the air" or mysteriously transmitted in indirect ways. Here the assumption is that most shared representations are not transmitted at all, but spontaneously inferred by members of the group exposed to similar cues. The process would result in roughly similar results in most members of a social group because the inferential modules are similar in their principles, because everyone shares the same intuitive ontology.

8. Conclusion
Recurrent religious representations are often described in the anthropological literature in the form of inductive generalisations on the basis of ethnographic descriptions. This method is, unfortunately, both indispensable to further research, and fraught with difficulties. One central and obvious problem is that such generalisations are not, as we would like to believe, a-theoretical and data-driven, but theory-laden in a way that is either not perceived or found too "natural" to deserve a justification. The point of a cognitive account is to integrate observed recurrence in an explicit causal framework. The hypothesis is that biases that favour certain types of representations are amenable to experimental studies, and that they probably have an important effect on the distribution of representations in a population, across groups and across generations. In other papers, I have shown how this type of description allows us to understand the particular conceptual structures built upon religious ontologies (Boyer 1993), their link with special social categories (Boyer 1994: 155-184), or the way they are used in causal understandings of particular occurrences (Boyer 1992; 1995). This does not in any way entail that the cognitive processes described are exclusively found in what we usually call "religious" systems. Indeed, the notion of "religion" is not really used here, except as a rough delineation of a set of problems. It is an empirical question, whether the kind of "cognitive equilibrium" described here is relevant to the explanation of other domains of cultural representations.

The relevance of this particular account depends on (i) the experimental evidence for the specific cognitive processes described, and (ii) its capacity to predict distributions of cultural phenomena that approximate the actual distributions. As far as the first point is concerned, the model is based on cognitive claims which are both explicit and empirically supported. As for cultural recurrence, the model is founded on a ceteris paribus clause. This model postulates that the representation of particular combinations of counter-intuitive and background intuitive assumptions is an important factor, in predicting the "cultural survival" of religious ontologies. This does not imply that other factors are non-existent, but only that in the long run, all these additional conditions being varied, the trend will favour assumptions based on the combinations described here, a claim that is supported by their recurrence in religious systems the world over.

One central point of Kant's Critique of Pure Reason was to demarcate the domain of concepts and intuitions that can be represented by human minds, by virtue of their constitution. The inquiry would therefore describe what Strawson (1966) called the "bounds of sense", beyond which there could be no contentful concept or intuition. There is in anthropology, as in other disciplines, a widespread tendency to assume that religious concepts are located, as it were, beyond the bounds of sense, in a domain of unconstrained imagination, limited only by social and historical constraints. This anti-Kantian conception would imply that cognitive constraints are, on the whole, irrelevant to the explanation of religious representations. The aim of the present cognitive account is to show exactly the opposite. Given a particular cultural input, people develop complex conceptual structures by an accumulation of spontaneous non-demonstrative inferences, which are strongly constrained by the content of intuitive principles. These principles result from evolved properties of their brains which delimit the "bounds of sense" and in consequence the bounds of non-sense.

Table 1
List of combinations in religious ontological assumptions

[1] person Dpsy, bio, phys + Vphys
[2] person Dpsy, bio, phys + Vbio
[3] person Dpsy, bio, phys + Vpsy

[4] animal Dpsy, bio, phys + Vphys
[5] animal Dpsy, bio, phys + Vbio
[6] animal Dpsy, bio, phys + Vpsy

[7] plant Dbio, phys + Vphys
[8] plant Dbio, phys + Vbio[9] plant Dbio, phys + Tpsy

[10] nat. obj. Dphys + Vphys
[11] nat. obj. Dphys + Tbio
[12] nat. obj. Dphys + Tpsy

[13] artefact Dphys + Vphys
[14] artefact Dphys + Tbio
[15] artefact Dphys + Tpsy

D = default activation, V = violation of expectations, T = transfer of expectations, psy = intentional behaviour, bio = biological properties, phys = physical properties.


Astington, J.W., Harris, P., & Olson, D.R. (eds.)
1988. Developing theories of mind, Cambridge: Cambridge University Press.
Atran, S.
1990. Cognitive Foundations of Natural History. Towards an Anthropology of Science. Cambridge: Cambridge University Press.
Avis, M., & Harris, P.
1991. Belief-desire reasoning among Baka children: Evidence for a universal conception of mind. Child Development, 62, 460-467.
Baillargeon, R.
1987. Young infants' reasoning about the physical and spatial characteristics of a hidden object. Cognitive Development, 2, 179-200.
Baillargeon, R., & Hanko-Summers, S.
1990. Is the top object adequately supported by the bottom object? Young infants' understanding of support relations. Cognitive Development, 5, 29-53.
Bastien, J. W.
1978. Mountain of the Condor. Metaphor and Ritual in an Andean Ayllu. St Paul: West Publishing Co.
Becker, A. H. &. W., T.B.
1991. Children's use of shape in extending novel labels to animate objects: Identity versus postural change. Cognitive Development, 6, 3-16.
Boyer, P.
1992. Causal thinking and its anthropological misrepresentation. Philosophy of the Social Sciences, 22, 187-213.
Boyer, P.
1993. Pseudo-natural kinds, in Boyer, P., Ed., Cognitive Aspects of Religious Syumbolism, Cambridge: Cambridge University Press.
Boyer, P.
1994. The Naturalness of Religious Ideas: A Cognitive Theory of Religion. Berkeley/Los Angeles: University of California Press.
Boyer, P.
1995. Causal Understandings in cultural representations. Constraints on inference from cultural input, in Premack, D., & Sperber, D., Eds., Causal Understandings in Cognition and Culture, Oxford: Oxford University Press.
Bullock, M., Gelman, R. & Baillargeon, R.
1982. The development of causal reasoning. In W. J. Friedman (Eds.), The developmental psychology of time. New York: Academic Press.
Bullock, M.
1985. Animism in childhood thinking: A new look at an old question. Developmental Psychology, 21, 217-225.
Crocker, J.C.
1985. Vital Souls, Tucson, University of Arizona Press.
Endicott, K. M.
1970. An Analysis of Malay Magic. Oxford: Clarendon Press.
Gelman, R., Spelke, E. & Meck, E.
1983. What preschoolers know about animate and inanimate objects. In D. &. S. Rogers J.A. (Eds.), The acquisition of Symbolic Skills, London: Plenum.
Gelman, S. &. M., E.
1986. Categories and Induction in Young Children. Cognition, 23, 183-209.
Hirschfeld, L.A., & Gelman, S. Eds.
1994. Mapping the Mind: Domain-Specificity in Culture and Cognition, New York, Cambridge University Press.
Hugh-Jones, S.
1979. The Palm and the Pleiades. Initiation and Cosmology in Western Amazonia, Cambridge: Cambridge University Press.
James, W.
1988. The listening ebony. Moral knowledge, Religion and Power among the Uduk of Sudan, Oxford: Clarendon Press.
Keil, F. C.
1979. Semantic and Conceptual Development. Cambridge, MA: Harvard University Press.
Keil, F. C.
1986. The acquisition of natural kind and artefact terms. In A. &. W. D. Marrar (Eds.), Conceptual Change, Norwaood, NJ: Ablex.
Kelly, M. & Keil, F.C.
1985. The more things changeŠ metamorphoses and onceptual structure, Cognitive Science 9, 403-406.
Lambek, M. 1981. Human Spirits: A cultural account of trance in Mayotte, Cambridge: Cambridge University Press.
Lawson, E. T. &. McCauley, R.
1990. Rethinking Religion: Connecting Culture and Cognition. Cambridge: Cambridge University Press.
Leslie, A.
1979. The representation of perceived causal connection. DPhil thesis, Oxford University.
Leslie, A.
1988. The necessity of illusion: Perception and thought in infancy. In L. Weizkrantz (Eds.), Thought without language, Oxford: Clarendon Press.
Mandler, J., Bauer, P. & McDonough, L.
1991. Separating the sheep from the goats: Differentiating global categories. Cognitive Psychology, 23, 263-298.
Mandler, J. &. B., P.
1989. The cradle of categorization: Is the basic level basic? Cognitive Development, 4, 247-264.
Massey, C. &. Gelman, R.
1988. Preschoolers' ability to decide whether pictured unfamiliar objects can move themselves. Developmental Psychology, 24, 307-317.
Metraux, A.
1959. Voodoo in Haiti, New York, Oxford University Press.
Perner, J.
1991. Understanding the Representational Mind. Cambridge, MA: The M.I.T. Press.
Richards, D. D., & Siegler, R.S.
1986. Childen's understanding of the attributes of life. Journal of Experimental Child Psychology, 42, 1-22.
Rosengren, K., Kalish, C.W., Hickling, A.K. & Gelman, S.
1994. Exploring the relation between preschool children's magical beliefs and causal thinking, British Journal of Developmental Psychology, 12: 69-82.
Severi, C.
1993. Talking about souls: The pragmatic construction of meaning in Cuna ritual language, in Boyer, P. ed., Cognitive Aspects of Religious Symbolism, Cambridge, Cambridge University Press.
Spelke, E. S.
1990. Principles of object perception. Cognitive Science, 14, 29-56.
Springer, K., & Keil, F.C.
1989. On the development of specifically biological beliefs: The case of inheritance, Child Development 60: 637-648.
Springer, K., & Keil, F.C.
1991. Early differentiation of causal mechanisms appropriate to biological and non-biological kinds, Child Development, 62: 767-781.
Strawson, P.F.
1949. The Bounds of Sense, Oxford, Clarendon Press.Thompson, L.
1949. The relations of men, animals and plants in an island community (Fiji), American Anthropologist, 51, 253-267.
Walker, S.
1985. Atimodemo. Semantic Conceptual Development among the Yoruba. PhD Dissertation, Cornell University.
Walker, S.
1992. Developental changes in the representation of word-meaning: Cross cultural findings, British Journal of Devcelopmental Psychology , 10: 285-299.
Wellmann, H., & Estes, D.
1986. Early understandings of mental entities: A re-examination of childhood realism. Child Development, 57, 910-923.
Wellmann, H.
1990. The child's theory of mind. Cambridge, MA: The M.I.T. Press.
Whiten, A. (Ed.)
1991. Natural Theories of Mind: The Evolution, Development and Simulation of Everyday Mind-reading. Oxford: Blackwell.
Wimmer, H., & Perner, J.
1983. Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children's understanding of deception. Cognition, 13, 103-128.n


*This paper was prepared while I was a Fellow at the Center for Advanced Study in the Behavioral Sciences, Stanford. The research was supported by a grant from the John D. and Catherine T. McArthur Foundation, grant #8900078. I wish to thank Tom Lawson, Carlo Severi, Michael Houseman, Dan Sperber and Sheila Walker for their comments on previous versions.

1 These results are important in that they go against a widespread assumption, that "superordinate" ontological categories are gradually constructed by abstracting from lower-level "basic" categories. That is to say, we tend to assume that children first acquire such concepts as "dog", "car" and "orange", and then abstract some of their features to build a representation of animal, artefact and plant. This "bottom up" account of conceptual acquisition seems supported by the fact that the earliest linguistic categories acquired and used by children do indeed correspond to such "basic" categories. Experimental evidence, however, seems to suggest that unlabelled, higher order ontological categories are available to the child before she acquires natural language, and probably make the acquisition of basic terms easier.

2 Some experiments for instance show that young children are confident that an imaginary item, described as "sleepy", may well be "furious", but that it is certainly cannot be "made of metal" (Keil 1986).

3 Also, they assume that such kind-based inductive generalisations are more plausible if the properties projected are "inherent" properties of the exemplars, e.g. ways of breathing and feeding, rather than weight or speed (Gelman & Markman 1987 passim, see also S. Gelman 1988).

4 The inheritance of particular external features, and that of colour for instance are not attributed to external physical causes; growth, too, is understood by the pre schooler as a lawful, internally generated process that is specific to living things (Springer & Keil 1989, 1991).

5 By default, identifying something as an artefact implies that the psychological and the biological module are not activated. Obviously, other modular structures may be activated, which would be irrelevant to live things. For instance, artefacts are commonly represented in a way that provides non-trivial abductive explanations of structural features in terms of the intended normal function of each part of the artefact. This is, by and large, irrelevant to the present argument.

6 Conversely, the fact that a situation is extremely unfamiliar does not ncecessarily make it counter-intuitive in th precise sense used here. If I see a giraffe being thrown out of a window in the House of Commons, swimming in the Thames to the Embankment and running away from people, this occurrence may be rather unfamiliar, but does not violate my expectations about solid objects (falling downwards), live beings (that generally try to avoid un unfamiliar milieu, like water for giraffes) and intentional beings (the giraffe intends to avoid people because it fears them).

7 To add one final, a contrario argument, let us suppose for the sake of argument that the criteria of "intuitiveness" are indeed culturally specific (against all the experimental evidence). Let us also imagine that some Westerners, armed with their supposedly "Western" criteria of intuitiveness, create a religious system based on assumptions like the following: that objects dropped will fall downwards, that people have thoughts and feelings, that mountains are solid inanimate objects, that a good deal of misfortune happens by pure chance, that dead men do not talk, and so onŠ I do not suppose anyone will want to bet too much on the cultural success of this truly innovative religious system. This crude thought-experiment does not constitute a sufficient argument, but it highlights a simple point: the notion that religious ontologies are not counter-intuitive for the people who hold them, generates a difficult problem when we want to account for their attention grabbing nature.