INTENTIONAL RELATIONS AND SOCIAL UNDERSTANDING

Abstract

1. Introduction

Humans are adapted to living in social groups with complex patterns of social interaction. In recent years, it has been suggested that the evolution of intelligence in primates that ultimately led to human beings was driven in part by the demands of social information processing (e.g., Humphrey, 1984). To deal with these demands, the reasoning goes, certain primates, most notably humans, have evolved a conceptual system that is used for making sense of, predicting, and manipulating the behavior of conspecifics (Byrne & Whiten, 1988). This conceptual system has been called (adopting a phrase from Premack and Woodruff's, 1978, work with chimpanzees) a "theory of mind" and it has become the focus of increasing attention, especially in primate and human developmental research (e.g., Astington, Harris, & Olson, 1988; Frye & Moore, 1991; Perner, 1991; Wellman, 1990; Whiten, 1991a).

Whether or not one accepts the label 'theory of mind' (for criticisms see Hobson, 1991; Johnson, 1988), there are two essential capacities of the human conceptual system for social understanding. First, the conceptual system represents activities of agents that are directed at objects. Such activities include simple purposive actions or psychological orientations that are directed at real objects, such as seeing, fearing, or poking them, as well as more complex mental activities, such as beliefs and desires, that may be directed at imaginary objects. We refer to all such activities involving an agent, a directed activity and an object as 'intentional relations'. Second, the conceptual system can be used for understanding the activities of both oneself and others. In other words, we have a single conceptual system that can equally well represent our own activities and the activities of others. Such uniform representation occurs despite the fact that the information available about one's own activities is qualitatively different from the information available about others' activities. It is the relationship between these two capacities of the system that is our main concern in this paper. As we will argue below, the uniform representation of the intentional relations of both self and other requires that the information available about one's own activities be combined with the information available about others' activities. In elaborating our view on how this is possible, we present a framework of forms of representation of intentional relations that identifies four different levels of social understanding. This framework is then applied to evidence from animal phylogeny and human development and then to the developmental disorder of autism, which is characterized by a deficit in social understanding.

2. Intentional Relations and their Representation 2.1. Intentional Relations

Organisms engage in a variety of activities. Consider the following examples:

(1) The bird builds a nest.

(2) The monkey fears the snake.

(3) Columbus sees land.

(4) The infant wants food.

(5) The frog mistakes a beebee for a fly.

(6) John thinks that Mary wants a computer for her birthday.

(7) Bob remembers that he vacationed in Jamaica last year.

(8a) Beatrice knows that Amanda likes Colin.

(8b) Beatrice knows that she likes Colin.

In each example an organism directs its activities towards an actual or possible object or state of affairs and these activities depend on the sensorimotor or higher mental capacities of the organism. We refer to such directed activities as 'intentional relations' and all such relations involve three constituents: the agent, the object, and the activity connecting agent to object.1

The examples listed here illustrate a number of features of intentional relations which it is important to distinguish. Examples 1 - 3 present three main types of simple intentional relations. Example 1 involves the action intentional relation of building a nest and example 2 involves the emotional/motivational intentional relation of fear of a snake. Example 3 involving seeing, is an illustration of epistemic intentional relations, which include sensing, perceiving, knowing, and believing. The difference between examples 1-3 and 4-5 shows that intentional relations can take objects or states of affairs that are current or they can take objects or states of affairs that are in the past or future (or, even just possibilities, including counterfactual ones).

Example 6 illustrates that intentional relations can themselves be the object of other intentional relations. In such cases, the latter intentional relation may be referred to as 'second order'. In this example, John is the agent of a second order intentional relation for which the object is the first order relation, 'Mary wants a computer for her birthday'. Example 7 shows that, agents can have second order intentional relations to their own first order intentional relations. Finally, examples 8a and 8b show that a single agent can be in second order intentional relations to comparable first order intentional relations of self and other.

It is second order intentional relations such as 6, 7 and 8 with which we are ultimately concerned in this paper for it is in these examples that an agent exhibits social understanding. In these particular examples an agent is in a second order epistemic relation which takes as its object a first order intentional relation involving either the self as agent or another agent . It is under such circumstances that it is usually understood that the agent "represents" or encodes the embedded first order intentional relation of the object-directed activity of an agent (cf. Dennett, 1987 for an example of a similar hierarchical system involving e.g., beliefs about beliefs). Furthermore, and most important, it is typically assumed (and explicitly by those who take a "theory of mind" perspective, e.g., Gopnik, 1993) that the same representational form is used in encoding the embedded intentional relation, whether the latter involves the self as agent or another agent. For example, in 8a and 8b the same agent (Beatrice) is in second order epistemic relations to two different first order intentional relations. In one of the first order relations, the agent is a different person (Amanda); whereas in the other the agent is the same (Beatrice herself). These examples assume that Beatrice is able to represent the intentional relations of both herself and other agents in the same manner. However, this seemingly innocuous assumption of a common representional system that is applied to self and others actually has a significant interpretive problem, which is that the information we have about the first order intentional relations of self and other is of qualitatively different kinds. For instance, in examples 8a and 8b, the information that Beatrice has about her own liking of Colin is not of the same kind as the information that she has about Amanda's comparable liking of Colin. So the problem is: How is it possible that the same representational form can be applied equally to self and other?

2.2 First and Third Person Information about Intentional Relations

An individual organism's information about intentional relations is available from two different sources. An organism has direct 'first person' information about its own intentional relations and direct 'third person' information about the intentional relations of other agents. In what follows we will argue that such first person information is qualitatively distinct from third person information.2 Before we begin, it is crucial that the reader not presuppose an adult human understanding of intentional relations. There are two illusions associated with adult human knowledge of intentional relations. In the "illusion of first person knowledge" (Gopnik, 1993; Moore & Barresi, 1993), adults humans believe that they have direct access to their own mental states, such as their beliefs and desires, but that they must infer comparable mental states in others. In addition, there is the equally pervasive "illusion of third person knowledge" whereby adult humans believe that they can perceive directly certain intentional activities of others such as the latters' seeing an object. In both cases, however, the illusion of direct access may be the product of acquired knowledge of intentional relations. To appreciate the information that is available to simpler organisms, such as animals and young children, it is necessary to dispose of a priori assumptions about the direct accessibility of intentional relations. Instead, it is necessary to consider the information that is directly available to an organism when it attends to the intentional relations of itself or another agent.

When an organism attends to the intentional relations of another agent the third person information that is immediately available consists most importantly in the behavioral expressions of the other's intentional relations. This information about the other's intentional action is available as gross physical movement; information about the other's emotions is available, in part, as facial transformations; and information about the other's perception may be available as orientation of head and eyes. These behavioral expressions may occur in spatiotemporal relations with other objects. For example, an observer attending to another agent who is looking at an object has direct visual access to the latter's head and eye orientation as well as to the spatial array of objects. In cases where the agent is looking at a static object, third person information of the agent's looking may be entirely about the agent's head and eye behavior and there is no direct information available to the observer that the agent's head and eye orientation are directed at an object. In other cases, third person information may also include the spatiotemporal relations between the agent's behavior and the object, as when the agent visually tracks a moving object, thereby moving its eyes or head in tandem with the object. Even though information about the object may be included in such cases, there is still no information about the object-directedness of the agent's activity.

Because third person information depends so heavily on spatiotemporal relations involving the agent's movements and their relations to objects, the extent to which such relations are available to the observer varies. In particular, for objects that are physically handled or near the agent it is relatively simple for the observer to form an associative connection between the agent's movements and the object around which its movements are organized. However, for objects at some distance in space, and especially in time, such associative connections are much more complex. So the observation of an agent's head turn may lead the observer to turn in the same direction, thereby allowing the observer to see the object of the other agent's gaze but the information about the object may not be incorporated as part of a relation between the agent and the object. Although the object of the other agent's gaze is now available to the observer, the latter may not link the object to the agent. In sum, third person information about intentional relations consists entirely in the movements of agents and certain spatiotemporal relations with other objects. Although third person information may be in part information about the organization of the agent's movements in relation to objects, it is not information about the intentional directedness of these relations.

When we consider the information that is available to an agent about its own intentional relations, what is most striking is that this information is qualitatively very different from that which is available about the relations of others. Consider again an agent tracking a moving object. The information available to the agent about its own looking is not visual information about its head and eye movements. Rather the first person information is predominantly about the object. Nevertheless, there is some first person information about the agent's own activity in relation to the object (Russell, 1994). This point is most evident from a comparison of world-caused and self-caused perceptual effects. In terms of the visual experience, the apparent movement of an object caused when that object travels in the world from left to right is no different from the apparent movement of an object when the agent moves its eyes from right to left. Yet, organisms are able to distinguish real movements in the world from apparent movements caused by self- motion, so some form of monitoring of the difference must occur. Agents must be able to distinguish perceptual events that are dependent on their own activity from perceptual events that are independent of their own activity. It is not the case that such discrimination requires an understanding of self as agent. Indeed, the above example demonstrates that any motile organism must have this capacity, and it has been shown to exist in relatively primitive organisms such as Drosophila as the phenomenon of afferent feedback (see Russell, 1994). In higher organisms, a variety of forms of monitoring exist, including kinesthetic and proprioceptive feedback. Through these forms of feedback the organism is provided with information about its own actions or possible actions in relation to perceived or imagined properties of the object. The point here is that an agent's experience of its own intentional relations must always involve objects and events in the world in relation to its own activities directed at those objects or events but need not include itself as an agent.

In brief, then, first and third person kinds of information about intentional relations, not already infected by the development of adult human forms of knowledge, are qualitatively different. Nevertheless, adult humans do represent their own intentional relations to objects and those of others in the same manner. It is our view that essential to the development of this capacity is the existence of an "intentional schema". This schema takes as its inputs both first person information about intentional relations and third person information about intentional relations. Because third person information tends to be about agents and their movements in spatiotemporal relations to objects, while first person information tends to be about objects in relation to the agent's actions or potential actions, it becomes possible, using the schema, to integrate these two sources of information into a single representation that can be applied equally to the activities of self and other.3 2.3 The Intentional Schema

Schemas are structures that organize information according to rules to yield new perceptual or conceptual units. Whereas some schemas organize information within a single perceptual modality, such as vision or audition, others provide rules for integrating information across modalities. By the intentional schema we mean an intermodal perceptual and conceptual structure with the capacity to coordinate and integrate first and third person sources of information about object-directed activities into representations that link agents to objects through intentional relations. Because first person information about these activities consists primarily of kinesthetic and proprioceptive information whereas third person information about these activities consists primarily of visual information, the intentional schema is a special case of an intermodal schema. Intermodal integration yields representations that preserve the quality of both forms of informational contribution. For example, integration of the visual information about a speaker's mouth movements and auditory information consisting of speech sounds yields a cross-modal representation of speech for which neither form of information is redundant (McGurk & McDonald, 1976). Kinesthetic feedback information may be similarly combined with information from other modalities, for example, when the visual image of oneself in a mirror becomes integrated with one's somatosensory awareness of one's movement so that one can use a mirror to shave or put on makeup. In these two examples, the information that is being integrated originates from a single individual - another person in the speech perception case; the self in the mirror case. What is different about the intentional schema is that informational inputs from both self and other are integrated, and the resultant representation can be applied to either self or other. Thus, consider our previous example of an agent who turns in a certain direction to look at some object, followed by an observer of the first agent who, seeing the latter, also turns in the same direction and sees the same object. Although these two events may occur without the observer being aware of their cooccurrence, the observer might also attend to the joint activity of self and other, and integrate the first and third person information that is available. Through the use of the intentional schema, the observer could generate a representation of the intentional relation that could be applied equally to the looking activity of either self or other. Without the intentional schema, the representation of the looking activity of others would necessarily remain distinct from the representation of one's own looking and neither would resemble the adult human concept of an agent looking at and seeing an object.

We assume that two critical conditions must be met for the intentional schema to generate an integrated representation of a particular intentional relation. First, there must be available to the schema both first and third person information about the intentional relation. If one agent is in a situation with another agent where both agents are engaged in the same intentional relation with an object then these two sources of information would be presented in current experience. The agent would have first person information about its own intentional relation with the object concurrently with third person information about the other agent's comparable intentional relation with the object. Although there are various means through which this situation can come about, for example, emotional empathy and joint attention, the imitation by one agent of another's actions on objects provides the most general one. Thus, when two individuals exhibit matched behavior towards a shared object while engaged in interaction with each other, as in imitating actions on objects, there is available to each individual at least both first person kinesthetic and proprioceptive information about its own current intentional relation to the object, and third person visual information about the comparable intentional relation of the other. We suppose that it does not matter which individual first generates the matching action as far as the schema is concerned. As we shall see later, an adult might match an infant's action, and thereby provide the infant with its initial opportunity to use the schema by presenting third person information of a comparable action coincident with first person information of the infant's own action.

It is important to emphasize that mere imitation of movements will not suffice to provide an opportunity to generate a representation of an intentional relation. Rather, the imitation must be of the object-directedness of activities. Although imitation per se may not be a necessary condition for the use of the schema, we suggest that it was only with the emergence in certain species of a capacity to imitate the object-directed activity of others that the appropriate general conditions occurred for the use of the schema. This mimetic capacity provides a way to generalize the matching of actions between self and other and allows the organism to generate first person information of one's own actions to match with a wide range of third person information obtained from observing the actions of others. Indeed, as we shall soon show, the evolutionary emergence of such an object-directed imitative responsivity - at least among primates - seems to have occurred in just those species for which we have independent evidence of a type of social understanding which suggests the use of the intentional schema.

We also suggest that in cognitively advanced organisms, not all uses of the schema require that both observer and actor currently engage in matched actions. Rather one or both of the sources of information may be derived not from perception but from the imagination. By 'imagination', we mean the capacity to mentally represent or think about objects in their absence, by using memory- based information. Through imagination noncurrent first or third person information which can act as input for the intentional schema may be provided. For example, one individual may observe another's behavior directly (the third person information) and imagine the first person information associated with the object or situation at which the action is directed. Conversely, one individual may engage in some action directed at an object or situation and thereby directly obtain the first person information of that action and at the same time that individual may imagine how such an action may appear to an observer, thereby imagining the third person information appropriate to that action. Such imaginative acts will also provide matched first and third person information which can then be integrated through the intentional schema to generate a representation of the intentional relation. The resultant representation unifies these two sources of information in a form that would apply equally to self or other regardless of the original source of the current non-imagined information.

The second condition for the generation of an integrated representation of an intentional relation by the intentional schema is that when the first condition of matched first and third person informational input is met either in actuality or in imagination, the observer is able to attend to both informational elements. For example, in the case of imitation, regardless of who first generates the matching response, one individual must attend to the other's behavior, while at the same time attending to the first person information about its own comparable intentional relation. Attention to both informational sources provides both inputs so that a representation of the intentional relation may be generated through the intentional schema.

It is only when both of these conditions are satisfied (i.e.: 1) matched first and third person inputs, 2) attention to these inputs) that the intentional schema can be used to integrate first and third person information into a single representation of an intentional relation. Whether the intentional schema should be considered to be a special module evolved for the particular purpose of social understanding or whether general purpose information processing mechanisms are involved is not our particular concern at this point. That is an empirical issue. What is essential is that some such mechanism is in place for some species (most obviously humans) and not for others, even though the latter may already satisfy the two conditions that have been considered. Moreover, this mechanism must join first and third person information of agents' relations to objects in a particular way so that the resultant represention can be applied in a comparable manner to self and other thus facilitating the kinds of social understanding found in human beings and some other primates. 2.4 A Four-Level Framework of Representations of Intentional Relations

In this section we present a framework of the forms that representations of intentional relations take. This framework identifies four levels of representation of intentional relations (see Table 1). It is derived in part from an empirical base which will be discussed in detail in later sections. At the lowest level are representations that treat first person information about the organism's own intentional relations separately from third person information about the intentional relations of others. Because, as we have seen, first person information has little in common with third person information, knowledge about the intentional relations of self gained through first person information is quite distinct from knowledge about the intentional relations of others which is based on third person information. At this level, all representation of intentional relations treats first and third person information about intentional relations differently. As a result, there are two distinct types of knowledge about intentional relations, one that applies only to other individuals and one that applies only to self. ------------------------------ Insert Table 1 about here ------------------------------

Given the capacity to integrate first and third person information through the intentional schema, we see three additional levels in the form of such representations. These two kinds of information may be derived from direct or current experience or from noncurrent, imaginative sources. The three levels at which the schema is used differ in the degree to which imagination is involved in the generation of the first and third person information that is integrated.

At level 2, both inputs are provided through current immediately available sensory and perceptual information as opposed to noncurrent, memory based, imaginative information. An organism can be said to be at level 2 when it is capable of attending to and integrating the first and third person sources of information available when a particular intentional relation to an object is shared with another individual. At this level, the similarity between self and other need not be recognized, but the organism must show evidence of knowledge of the shared relation in the sense that its actions will differ depending upon whether the relation is shared or not. In some circumstances the organism may attempt to establish a shared relation with a conspecific. For example, 12-month-old human infants, while not showing a clear awareness of themselves as agents, will make communicative attempts to engage others in shared attention to an object. At level 2, then, intentional relations are not understood to be associated with a single individual, either self or other. Nevertheless the representations of activities do necessarily involve both first and third person information and thereby captures the intentional nature of the relation between an agent and an object.

At level 3, current experience of an intentional relation provides one input to the schema and imagination provides the second. The current input, whether from self or other, instead of leading to an attempt to generate matched input through an action, leads to the generation of this input through an act of the imagination based on memorial representations of similar intentional relations. The intentional schema is then used to yield a new representation of an intentional relation based on only one perceptual input. The presence of imaginal input to the intentional schema means that an integrated representation of a particular intentional relation can be generated without a matching of action. As a result, at this level it is possible to represent the current intentional relation of an agent, whether it be self or other, and distinguish it from the intentional relations of other agents. Thus the self-other distinction can become explicit and the individual can be said to have a concept of an intentional agent. The understanding of the intentional relations of both self and other, however, is still tied to current experience. In the case of the other, the third person information is typically given by current perceptual input, and the first person information is imagined. For example, an individual is observed to display a particular facial expression in the direction of an object and the observer imagines the first person component of the emotional attitude corresponding to that display. In the case of the self, typically, it is the third person information that is imagined, while the first person information is provided by the intentional relation in which the actor is immediately engaged. Thus feeling an emotional orientation to an object becomes integrated with an image of how that orientation would appear to an observer. The understanding of intentional relations that results at this level is one involving diversity in the intentional orientations of different agents to a common directly accessible situation (though it may be a "pretend" situation). Because direct access to a current situation constrains this form of intentional representation, it cannot represent diversity in intentional relations of agents to different mental objects - some of which may misrepresent the current situation.

Finally, at level 4, both inputs to the schema come from the imagination rather than through current experience. Situations necessitating level 4 are those which require the representation of an intentional relation of either self or other where neither current first nor third person information enters into the intentional relation. In order to resolve this problem, the intentional relation, either of self or other, must be abstractly represented with an agent's orientation to a mental object. In these cases, as with level 3, current experience typically activates stored representations of intentional relations and it is these representations that provide the inputs for the schema. In contrast to level 3, however, the novel schematic representation is now entirely composed in the imagination, in the sense that both first and third person information are imagined and integrated through the schema. Thus, the observer can imagine the third person information of an agent in some situation and, at the same time, can imagine that agent's first person orientation to that situation. In this way, the agent is placed in a different situation from that which the observer faces and the novel representation of the imagined intentional relation can be compared to current experience of the observer. As a result, the mental object in the imagined situation is cleanly distinguished from the actual object in the situation as perceived by the observer. The individual at this level, then, may be said to have a concept of a mental agent. Although there may be further levels, such as that of a reflective agent (see Barresi & Moore, 1993; Tomasello, Kruger, & Ratner, 1993), our account of the understanding of intentional relations - particularly in human development - terminates at this level of representation since it is at this level that misrepresentation can be understood by an agent.

Now that we have a framework which indicates various levels of the representation of intentional relations, we can use it to describe how organisms come to understand themselves and others as intentional beings. In the next section, we suggest that most animals operate at the lowest level of the framework. Among nonhuman animals, only great apes appear to exhibit a general social understanding beyond level 1, though they do not attain level 4. In the following section, we present evidence that during the first four years of life, children progress through a series of three stages, equivalent to the later three levels of the framework. Finally, we argue that, perhaps as a result of a failure in information integration, autistic individuals fail to develop representations of intentional relations that can be applied uniformly to self and other.

3. The Phylogenesis of Social Understanding

Nonhuman species can be divided into two classes based on whether or not they can categorize their own and others' intentional relations using a single conceptual system. Because most organisms lack this special capacity, they tend to interpret their own and others' actions using different conceptual systems. For these organisms, most representations based on first person information are distinct from those based on third person information and there is no general understanding of the similarity in intentional relations between self and other. We suggest that these organisms operate for the most part at level 1 of our framework, though sometimes possibly at level 2, typically in communicative domains, where they may recognize the comparability of their own activities and those of others (see Table 1). Members of a smaller class of organisms are able to interpret the activities of themselves and others using the same conceptual system. The evidence that this capacity is present is the ability of some species - in particular great apes - to imitate novel goal-directed actions. These species can operate at levels 2 and 3 though not yet at level 4 of our framework (see Table 1). We do not think that this mimetic capacity alone is sufficient to allow them to appreciate the representative character of mental activity. Only humans have the capacity to think about the representational nature of mental relations and thus to achieve level 4 of the present framework. 3.1 Level 1 or nonimitative organisms.

Insofar as a level 1 organism is unable to apply the same conceptual system to self and other, such an organism will be severely restricted in what it can appreciate of the meaning of its conspecifics' activity and what it can learn from that activity. Although we do not wish to restrict ourselves to particular models of learning, it is worth considering the nature of level 1 processing with respect to Pavlovian and instrumental conditioning from the perspective of the expectancy theory of Dickinson (1989; Heyes and Dickinson, 1990). For level 1 organisms, learning about the intentional relations of others takes the form of Pavlovian, stimulus- stimulus generalizations or expectancies about one stimulus given another. Learning about their own intentional relations takes the form of instrumental expectancies or expectancies about outcomes given particular actions. In observing others, one organism can form stimulus-stimulus generalizations and expectancies between different observable behaviors of the others, for example, between a grimace and a likely response of another organism in a certain context. It can also notice stimulus-stimulus relations between observable behavior of the organism and objects or events in the world, for example, the orientation of the other organism's head or eyes in a direction where an interesting object might be found. In this way, much can be learned about the intentional relations of others, both general and specific. It may be possible to learn to detect deceptive acts by others from the absence of the usual contingency between an observed act and consequence. An organism will not be able to interpret these stimulus-stimulus complexes in terms of the motives of the other organism, however, nor to see the other's movements as actions that are analogous to its own. Though it may act in response to the deceptive behaviour, it will not be able to recognize that it is responding to the behaviour as deceptive.

In contrast to the stimulus-stimulus learning that can occur through the observation of others, a level 1 organism will be able to acquire knowledge of its own instrumental actions and their expected consequences. Many organisms are capable of forming expectancies which represent the relation between their own instrumental behaviour and a consequence (see Dickinson, 1989). Unlike Pavlovian conditioning, which does not require the representation of one's own potential behaviour, instrumental conditioning has been interpreted as involving a representation of the action in its relation to the expected consequent or goal. Thus, for instance, if the reward that is "expected" to occur following a particular response is decreased, rats but not goldfish, will begin to exhibit an instrumental response depressed below the control level for the new reward. This result has been interpreted to suggest that rats have representational knowledge of the behaviour-reward contingencies but goldfish do not (Bitterman, 1987; Dickinson, 1989; Mackintosh, 1987; see Macphail, 1987, for a contrary view). Assuming that this is the case for instrumental learning for some organisms, an organism with this capacity, but still at level 1 of our framework, would have a representation of its own goal-directed action in a manner that it could not generalize to the comparable goal-directed actions of others. Nor could it translate the goal- directed actions of others into actions it could adopt for itself. A result of this lack of comparability between the actions of self and others is that such an organism could learn to engage in fairly elaborate deceptive behaviours without recognizing that its behavior was, in fact, deception; at the same time it could be easily deceived by a comparable deceptive act of the other.

In short, the activities of self and other are processed in different ways by level 1 organisms. The visual representation of the activity of others which may act as a conditioned stimulus is not directly translatable into the representation of their own instrumental activities. We have suggested that the advent of the ability to interpret the activities of both self and other within a single conceptual scheme comes with a generalized capacity to imitate goal-directed activity. Such a capacity demonstrates crossmodal matching not only of body movements based on visual and kinesthetic information (Mitchell, 1994) but also of those directed at objects or goals. It is the latter forms of movement that must be matched for uniform processing of the intentional relations of self and other.

In evaluating our framework with respect to lower organisms it is important to distinguish similarity in the behavioural manifestations of intentional relations because of a general capacity to imitate goal-directed actions and similarity because of alternative mechanisms. Among many - if not all - social species a wide variety of mechanisms produce similarity in behaviour in the absence of a general capacity to imitate novel behaviors, whether expressive or goal-directed. In a recent review of the literature, B. Moore (1992) has described ten mechanisms by which two organisms might exhibit similar behaviour in a particular context without having the capacity to imitate novel actions (also see, Galef, 1988; Whiten and Ham, 1992). In what follows we will give a brief overview of some of the matched activities we believe can be accounted for at level 1 (or at most level 2) of our framework. We will then go on to consider more complex phenomena by nonimitative organisms that might be interpreted as higher levels of our framework, but we will suggest how they might be accounted for at level 1. Finally, we will consider recent experimental studies which are consistent with this view, suggesting that monkeys do not exhibit the generalized capacity to understand the viewpoint of others indicative of level 3 processing.

A simple form of such sharing of intentional relations occurs when an organism is a member of a group and responds to the perceived third person aspects of the intentional relations of others by adopting the same intentional relation itself. Examples of such sharing of action relations can be seen in the herding behavior of mammals and schooling behavior in fish. Examples of sharing of expressive behaviour that serve to communicate emotional/motivational intentions are the croaking of frogs, the barking of dogs, and the howling of wolves. There are also examples of the sharing of epistemic relations in animal systems. Upon seeing a conspecific stand to look in a particular direction, baboons will themselves adopt a similar orientation; thus this mechanism allows each baboon to benefit from the earliest exposure of any member of the troop to a significant environmental event. Such shared intentional relations come to play a quite precise role in more complex forms of communication of emotional as well as epistemic relations. Cheney and Seyfarth (1990) have described how vervet monkeys use particular calls that not only express an emotion - such as fear - but indicate the presence of particular predators. Such signals elicit appropriate evasive action in a hearer, but they also cause it to generate the same call, thus producing a unified group activity of communication and action.

We believe that most of these phenomena can be accounted for at level 1 of our framework, without proposing the existence of an intentional schema that applies uniform representations to the actions of self and other. In these cases, we would suggest that each group member only has a third person representation of any other's activity and a first person representation of its own activity. However, the other's acts, such as movements or emotional expressions, come to be associated with the first person aspects of those expressions as experienced concurrently or in close synchrony. Because of the close synchrony of third and first person representations of common intentional relations, the individual organism does not seem to act as an individual but as a member of a group. Such associated, but not integrated, first and third person representations of intentional relations act nearly as well for group activity as an integrated version at the next level of our framework. For such complex communicative systems as the vervet calls, it seems possible that the organisms recognize the similarity of calls being used by self and other (e.g. the 'Moving into the Open' grunt studied by Cheney & Seyfarth, 1990, and described by Dennett, 1987; Cheney & Seyfarth, 1992). If so, then these organisms are operating at level 2 of our framework; however such cases seem to be the exception rather than the rule.

With the possible exception of complex communication systems, we believe that the majority of the behaviours of self and other are not represented in the same representational system by nonimitative organisms. To further support our claim that even monkeys' social skills can be accounted for at level 1, it is worth evaluating some recent research on the ability of monkeys to attribute intentional relations to conspecifics. Cheney and Seyfarth (1990) conveniently provide a review of this research, particularly in their chapter 8 on attribution. Their summary indicates that monkeys are capable both of interpreting current stimulation from conspecifics that might have consequences for their immediate goal attainment and of anticipating the responses of conspecifics to their own actions and expressive behavior. Insofar as intentional relations of others can be understood based on perceived properties of objects and the contingent relations of other monkeys to those objects, or, in other words, from a third person perspective, these monkeys seem highly skilled in anticipating the actions of others based on these relations. However, monkeys seem incapable of interpreting the behavioural tendencies of others as a reflection of the observed monkey's motives, emotions, and epistemic relations in situations where the observers are - in a sense - disinterested. They exhibit virtually no signs of empathy, defined as the capacity to recognize the emotional relations of others which they are not themselves experiencing, in that they do not exhibit compassion such as comforting a monkey in grief over a loss of a close relative or providing food to the old and disabled.

Self-knowledge of monkeys is likewise limited and of a different form from their knowledge of others. For instance, though they can learn to use mirrors to perceive the activities of others, they do not recognize themselves in mirrors. That their knowledge of their own intentional relations is represented differently from their knowledge of the intentional relations of others is indicated by a number of the findings. For instance, they are sufficiently aware of their own goals that they act to avoid being deterred from achieving them when they anticipate that another will interfere. Consistent with expectancy theory (Dickinson, 1989), the agent's own goals and knowledge of how to achieve them are not immediately visible to the other, yet do determine its own actions.

The degree of sophistication that monkeys can achieve while remaining at level 1 of our framework is probably best exhibited in the various forms of tactical deception that can be found in baboons. Whiten (1991b) cites an example provided by Strum (1989). A male baboon, Dr. Bob, was tricked by Peggy, who groomed him until he released his grip on an antelope, which he refused to share, so that she could grab it and run. Later a similar situation arose and so Dr. Bob held on tightly, despite the grooming. Peggy, not to be thwarted, then attacked a female associate of Dr. Bob, until he - apparently with trepidation - abandoned the meat to come to her defense, and Peggy again grabbed the meat and ran off. Now, it is easy enough to imagine that Dr. Bob and especially Peggy, understood the intentional relations of each other and themselves in the same conceptual terms. But this is by no means necessary in the present case. Both Dr. Bob and Peggy could perceive the other's actions and potential actions entirely from a third person perspective and their own goals and possible actions from a first person perspective, without in the least being able to imagine how their own actions appeared to the other, or how the other's actions appeared to that other. In both directions, the appropriate actions to take and the likely actions of the other could be based on the immediate perception of the situation and an acquired expectancy of the relation between one's own potential action and the behaviour likely to be exhibited by the other organism in response to it. Hence, without further evidence that the organism has knowledge of the perspective of the other animal that could not have been acquired through instrumental learning or stimulus-stimulus generalizations, even such sophisticated acts of deception are easily handled by representations acquired at level 1 of the present framework.

There are two kinds of phenomena that would indicate that monkeys really understood their own and the other's behaviour from the perspective of the other animal. First, such understanding would be shown by the imitation of novel purposive actions of conspecifics based purely on the visual observation of the performance of those actions. Second, it would be shown by the ability to switch roles with another organism with whom they have jointly learned a cooperative task. Yet, the evidence to date indicates that monkeys show neither of these abilities.

Evidence of the monkey's ability to imitate novel purposive action is quite limited (Cheney and Seyfarth 1990; Galef, 1990; Whiten and Ham, 1992). For instance, consider the famous case of the potato washing technique that was acquired through a kind of cultural transmission among Japanese macaques. What is notable, besides the extremely slow diffusion of this skill, is that the monkeys seemed to acquire only knowledge of third person contingent relations involving stimulus complexes of water, edible potatoes and monkey movements with these objects, not first person knowledge of how to wash the potatoes as well as the direct awareness of the function or goal of the washing. As a result each monkey independently had to make sense of the stimulus complex to which it was exposed and to re-invent how to wash the potatoes and thus to develop its own peculiar method of washing. If its perception of the actions of others represented those actions in the same way as its own actions, as we suggest occurs at level 2 or higher levels of our framework, then we believe that imitation would have presented the 'how' and 'why' as well as the 'what' of potato washing.

Further evidence that monkeys are probably not able to represent their own and the other's actions in a common representational form is provided by a recent study by Povinelli, Parks, & Novak (1992). They constructed a paradigm where a rhesus monkey had to cooperate with a human, each performing different actions in order to generate a reward for the monkey. Different monkeys were trained on each of the two possible roles. After having acquired the ability to play one of the two roles, the monkeys were assigned to the opposite role. If the monkeys had represented the actions of both agents within a common representational system then the role reversal should have produced strong positive transfer, as it did in a previous study with chimpanzees (Povinelli, Nelson, & Boysen, 1992). However, for the monkeys, none showed immediate comprehension of their new role. This suggests that the monkeys' representation of their own role was different from their representation of the role of the other agent and that the difference is because monkeys - unlike chimpanzees - are restricted to level 1 representations. Hence the representation of their own actions uses only first person information while the representation of the actions of the other agent uses only third person information and there is no direct translation between these representational forms that can be used in the role transfer task of the present experiment.

3.2 Imitative organisms.

Cheney and Seyfarth (1990) suggest that only higher primates have a general ability to imitate a functionally useful technique. B. Moore (1992), in his review of the imitation literature, also suggests that the evidence of imitation in monkeys is too limited to attribute this capacity to them, and that, of nonhuman species, it is only the great apes, cetaceans (e.g. dolphins), and certain avian species, in particular parrots, that have this capacity. For our purposes, it is interesting to note that only the most intelligent and social mammals are capable of a generalized imitative ability. We would go further and suggest that the imitative ability of avian species, which functions predominantly for display, is intrinsically different from that of the social mammals. While avian imitation involves only the form of the movement or sound, mammalian imitation is centred around the goal directedness of activities and appears to function predominantly for social learning. Hence we believe that the general capacity to imitate actions directed at objects for a purpose may be an important skill showing an ability to interpret intentional relations of self and other within a common conceptual system.

In what follows, we will evaluate the capacities for social understanding in great apes with a particular focus on chimpanzees (for Gorillas see Patterson & Linden, 1981; for Orangutans see Miles, 1986, 1990; Russon & Galdikas, 1993; for a more general survey which includes dolphins see Mitchell, 1994), and argue that the capacities that they exhibit, indicate that they can operate at levels 2 and 3 of our framework as well as at level 1. We would suggest that these capacities imply that they have acquired a concept of an intentional agent (See Barresi & Moore, 1993; Tomasello, et. al., 1993) which they can apply to self and other.

Several interrelated phenomena indicate that adult chimpanzees have the concept of an intentional agent and make use of the intentional schema that joins first and third person information of an agent engaged in a current intentional relation. Consistent with this hypothesis, chimpanzees are able to determine the goals of others' actions when they observe their movements directed at particular objects, for example, reaching for a distant object (Premack & Woodruff 1978). They can also imitate the actions of others based on these same observations because the representation of the action is in a format that can be applied to the self as well as to the other. Hence, they are able to imitate both the means and the ends of the actions (Tomasello, Savage-Rumbaugh, & Kruger, 1993). Since they appreciate the other's point of view as well as their own relative to a cooperative task, they can reverse roles in the Povinelli task and spontaneously adopt the behaviour of the other agent (Povinelli, Nelson, & Boysen, 1992). Furthermore, chimpanzees have also been known to show emotional sensitivity and compassion (Cheney & Seyfarth 1990; Goodall 1990), indicating that their observation of a conspecific's expressive behaviour is sufficient to provide an understanding of the other's emotional intentional relations directed at goals. Chimpanzees have been reported to show quite sophisticated acts of deception and counterdeception (Whiten & Byrne 1988; Whiten, 1993). The latter acts seem to require the capacity to distinguish the "real" goal of the action of self or other from the "apparent" goal of the behaviour, and to recognize that the apparent goal oriented act can function to achieve the real goal. Although not yet requiring a notion of misrepresentation, such deceptive behaviour depends on the concept of an intentional agent (cf. Whiten, 1993). Finally, although the interpretation of mirror self-recognition remains uncertain (Heyes, 1994; Mitchell, 1993), the fact that chimpanzees but not monkeys do show such self-recognition suggests that the former are able to interpret their own image in the same manner as they interpret the images of others, which would follow if they acquired a concept of intentional agent that they could apply to self as well as other (Gallup 1977, 1982).

Premack and Woodruff (Premack and Woodruff 1978; Woodruff and Premack 1979) originally suggested that chimpanzees have a "theory of mind". Gallup (1982) went further, hypothesizing that any organisms capable of self-recognition would also have an introspective awareness of their own mental states and the ability to ascribe mental states to others. In a recent article, Povinelli (1993) appears sympathetic to Gallup's interpretation. These suggestions imply that chimpanzees are able to understand their own and other individuals' behavior not only at level 2 and 3 of our framework, but at level 4 as well. If Gallup's model were correct then chimpanzees would not only be aware of themselves as intentional agents of the same kind as their peers, but they would also have introspective knowledge of their mental states and be able to attribute such states to others. We believe that Gallup's model of chimpanzee social understanding is too rich given the current state of the evidence (see also Heyes, 1993). Some of the evidence from Povinelli's studies and those of others (see Cheney & Seyfarth 1990) suggests that chimpanzees have the ability to represent in some fashion the intentional relations of others, in particular, to represent the motives, emotions and epistemic relations as perceived in the others' actions. They may also be capable of representing diversity in the knowledge relations of others, for example, differentiating an actor who has knowledge of information from one who does not (Povinelli, Nelson & Boysen 1990).

There is little evidence, however, to suggest that chimpanzees can appreciate the representational nature of mental activities, and therefore misrepresentation, which require level 4 representations. Premack's original work has been contested (see e.g. Bennett 1978; Dennett 1978) and, indeed, led to the interesting developmental studies on human "theory of mind" which we discuss later. Premack's own subsequent research (e.g. Premack & Dasser 1991) suggests that the chimpanzee, at its outer limit, behaves like a three year old child in its capacities, a level congruent with level 3 of our framework as we will show in the next section. An attempt to assess the chimp's ability to attribute false belief failed to provide positive evidence (Premack, 1988). Given the limited evidence of knowledge of intentional relations beyond those expressed in immediate action, and the absence of positive evidence in favor of chimpanzees' and other great apes' ability to handle abstract epistemic relations, such as false belief, we are not convinced that the intentional ascriptions of these species are represented beyond level 3 of our framework. Hence neither the assumption of a theory of mind nor that of introspective awareness seem warranted.

When we turn to human social understanding, we find evidence for all four levels of our framework. In particular, humans are capable of abstractly representing intentional relations of self and other, characteristic of level 4. However, these level 4 representations of intentional relations should not be considered to be replacements for lower level representations. It is likely that humans exhibit social understanding at all levels of the framework depending on the demands of the circumstances. Furthermore, the human capacity for level 4 representations of intentional relations does not appear full-blown at birth. We turn now to the ontogenesis of human social understanding and apply the framework to a variety of the phenomena seen in the study of human development.

4. The Development of Social Understanding

In what follows, we give a schematic account of the development of social understanding during infancy and the preschool years, using a three stage ontogenetic model based on the latter three levels of the framework introduced earlier (see Table 1). Let us summarize the main points of our account here. First, we will briefly review the social abilities of the infant during the first year of life, which, from our point of view, indicate the use of the intentional schema. These behaviors, such as joint attention and social referencing, are ones in which a particular intentional relation is shared with another; at this stage, the infant can, therefore, be said to be operating at level 2 of our framework. Once the first representation of intentional relations is in place, we suggest that subsequent developments in the representation of intentional relations depend on domain-general changes in imaginative ability. These changes may well be the result of maturational changes in information-processing capacity (cf. Olson, 1989, 1993). During the second half of the second year, the child becomes capable of imagining one informational component of an intentional relation while attending to the other. This ability marks the emergence of level 3 of the framework, where intentional relations are ascribed to individual agents, including self and other. The child can therefore be said to have a concept of an intentional agent. Later again, at about 4 years, a further increase in imaginative capacity allows both informational components of an intentional relation to be imagined and integrated. At this stage, corresponding to level 4 of our framework, the child is able to represent intentional relations to mental objects and thus can be said to have a concept of a mental agent and a 'representational theory of mind' (Perner 1991). 4.1 Stage 1 (level 2): Shared Intentional Relations

Infants become interested in social stimulation and can engage in mutually arousing social interactions with adults from at least two months of age (e.g. Brazelton, Koslowski, & Main, 1974; Schaffer, 1984; Trevarthen, 1979; Watson, 1972). From this time, adults will interpret infant behavior within an interaction as being social and communicative (Frye, Rawling, Moore, & Myers, 1983). From about 4 to 5 months of age, as their perceptual and motor skills develop, infants start to pay more attention to nonsocial objects in their environments (Schaffer, 1984). During the middle part of the first year, however, infants do not coordinate person-directed behavior with object-directed behavior. Coordinated attention to both a person and an object generally appears towards the end of the first year of life (Bakeman & Adamson 1984; Trevarthen & Hubley, 1978). It is at about this time that infants will turn to look in the same direction as other people. This activity is commonly termed 'joint attention' (Butterworth & Jarrett 1991; Scaife & Bruner 1975). At about the same time social referencing develops: the infant uses emotional information from another to regulate his or her own behavior in an uncertain situation (e.g. Feinman 1982; Sorce, Emde, Campos & Klinnert 1985). In addition to using others' behavioral cues to guide their own behavior, infants will seek to direct others' behavior, including their direction of gaze. For example, communicative gestures, such as protodeclarative pointing, also start to appear towards the end of the first year (e.g. Bates, 1979). The typical recent interpretations of joint attention, social referencing, and communicative gestures share the notion that the infant understands that others can exist in intentional relations with objects (e.g. Baron-Cohen, 1991; Bates, 1979; Bretherton, 1991; Bruner, 1983; Harris, 1989; Reddy, 1991; Tomasello et al., 1993). For example, most recently, Tomasello et al. (1993) have claimed that all these phenomena are based on the infant having a concept of an 'intentional agent'.

Although we agree with all of these authors that these phenomena mark a very important development in the infant's social understanding, our interpretation differs. For us, these phenomena indicate the first use of the intentional schema at level 2 (see Table 1) of our framework and in what follows, we make this case. At this level, the intentional schema is used to integrate first person and third person information derived from the matched intentional activities of self and other. The resulting representation of the intentional relation is one which spans the self and other without the intentional relation being identified as existing in either individual alone.

Because the operation of the intentional schema involves intermodal integration of information, it is necessary that infants during the first year be capable of such information processing. Research directly examining infants' intermodal integration has shown that from early in the first year, infants are able to integrate information from more than one source in order to detect amodal invariances. Such information integration is seen clearly in cases of intermodal perception of temporal properties such as rhythm (e.g. Spelke 1979). With respect to social information, Walker (1982) has shown that by 5 to 7 months infants are able to detect the correspondence between visual and auditory information specifying the expression of emotions. However, in this case both pieces of information were of other people and were of the third person variety. Can infants integrate information of the third person sort with information of the first person sort, when those pieces of information are presented via different modalities? Bahrick and Watson (1985) have shown that infants of five months are able to detect the relationship between their own leg movement and a real- time video display of that movement, even when they cannot simultaneously see their leg. They suggest that the basis for this detection is contingency. These results show that intermodal integration of information from vision and proprioception is occuring at this stage in infancy, and it seems quite likely that similar integration could occur for information provided by different individuals if there was a high degree of contingency in the presentation of this information. It could be, for example, that the strong tendency of mothers and others to imitate infants' behavior in the first year (e.g. Moran, Krupka, Tutton & Symons, 1987) provides the necessary contingent experience for infants to recognize the match between their own and others' movements.

Recently, Meltzoff (e.g. 1990; Meltzoff & Gopnik, 1993) has gone further and proposed that innate knowledge of the equivalence of first and third person information has an important role in the development of social understanding. He suggests that "infants, even newborns, are capable of apprehending the equivalence between body transformations they see and ones they feel themselves perform" (Meltzoff, 1990, p.160) through the existence of an innate intermodal representational system that codes the body movements of both self and other. The evidential basis for Meltzoff's claim for such an innate intermodal representational system is the apparent neonatal imitation of facial gestures such as tongue protrusion and mouth opening (Meltzoff & Moore, 1977, 1983). However, a recent review and meta-analysis of the research on neonatal imitation by Anisfeld (1991) has shown that reliable evidence has been obtained for only one facial gesture - tongue protrusion. Without strong evidence for the imitation of a range of facial expressions, there is no need to postulate the existence of an innate intermodal representational system which would allow the recognition of the equivalence between others' movements and one's own.

In any case, the issue of whether the understanding of the equivalence of movements of self and other is innate or acquired from experience is not critical to the present discussion. Whether or not Meltzoff is right in proposing the existence of an innate representational system for coding invariance in the movements of self and other, and whether or not infants learn to imitate movements, neither could generate an understanding of the object- directedness of such movements, because no object is included in the representation. The important consideration is that during the second half of the first year, when objects first start to be brought into interactions between infants and others, infants are able to match the behavior of others. In this way, object-directed activities can be imitated. Once the infant can match others' behavior towards objects, they will have available the kind of experience necessary for the first representation of intentional relations (see also Moore & Barresi, 1993). It seems plausible that the first intentional relations to be represented are those involving body parts in contact with objects. For example, the infant who is participating in an imitative game involving a particular action on an object will have concurrently available both first and third person information about that action intentional relation. These two types of information may then be combined through intermodal integration into a representation of that intentional relation.

The representation of intentional relations will not be limited to cases where the agent makes contact with the object, however. Although the infant's first tendency to turn to look in the same direction as another, or to use another's emotional expression in social referencing, may come about as a result of learning or an innate orienting response (Campos, 1983; Gewirtz & Pelaez-Nogueras, 1992; Moore & Corkum, 1994), we suggest that the infant's representation of these events is not limited to that seen in lower primates. Once acquired, these behaviors will provide the infant with both first and third person information about the shared intentional relation. Thus, in joint attention, infants observe the head/eye orientation of the other and experiences the kinesthetic feedback from their own head turning and the sight of the object. Here, intermodal integration will combine these two types of information into a representation of the intentional relation between viewer and object. In the case of social referencing, the infant observes another's emotional facial expression and adopts a corresponding emotional orientation to the object of interest. In this case, intermodal integration will combine the first and third person information into a representation of the emotional relation between agent and object. In all these cases, then, the infant's participation in shared intentional relations with others allows the intentional relation to be represented by integrating the third person information available from the observation of others' behavior and the first person information available from actually being in an intentional relation with an object. We suggest that the evidence for this integration is seen most clearly in the fact that, unlike the majority of cases of apparent shared intentional activity in other species, infants will check back between observation of the object and observation of the adult (cf. Gomez, 1991), thereby indicating concurrent attention to both informational elements.

In sum, we are proposing here that during the first year of life, infants become able to integrate information of the first and third person sort. Towards the end of the first year, the inclusion of objects into social interaction provides the first understanding of intentional relations. Initially, infants understand these intentional relations to the extent that they can participate in episodes of shared relations, with the first person component corresponding to the information available from their own intentional relation to some object or state of affairs, and the third person component constituted by the information available to them from the other's relation to that same object or state of affairs. At this stage there is a lack of differentiation of the sources of the first and third person information that enters into the experience of an intentional relation, so the intentional relation is not attributable to individual agents. This state of affairs occurs because only immediately experienced first person and third person information enter into the representation of the intentional relation at this time. This situation gradually gives way during the second year of life to one in which self and other are differentiated, with each understood to have both first and third person aspects. 4.2 Stage 2 (level 3): Imagination and the self-other distinction

The second year of life sees the development of a variety of phenomena, persuading many authors that major changes in cognitive abilities are occurring. The development of flexible manual search for hidden objects (e.g. Piaget 1953), deferred imitation (e.g. Piaget 1962), language use (e.g. Bates 1979; Greenfield & Smith 1976), and pretend play (Leslie 1987; Piaget 1962) have all been interpreted in terms of a general change in representational capacity. During the first 18 months of life, infants show an increasingly sophisticated ability to interpret and respond to perceptual events in terms of stored representations and can show recognition and discrimination. However, what is new about the phenomena appearing during the second half of the second year is that not only can infants respond to some perceptual object or event but they can also simultaneously 'hold in mind' a representation of a nonpresent or noncurrent object or event (e.g. Olson 1989, 1993). The immense advantage of this form of representation is that infants are now able to hold in mind objects and events from other places that occurred in the past (as in memory), as well as those that have not yet occurred, but could (Perner, 1991). Furthermore, the mental representation can control behavior, so that activity can now be conducted in relation to imaginary objects or states of affairs rather than real ones, as seen, for example, in pretend play. Rather than 'representation', we prefer to use the term 'imagination' to refer to this capacity, because the latter term more adequately captures the idea that this cognitive advance allows the mind to range over both past and future as well as other possible objects and events.

The development of the imagination makes a new stage of social understanding possible. In stage 1 of social understanding, infants understand intentional relations to the extent that that they can participate in episodes of shared intentional relations with others, such as joint attention and imitation. Because the understanding of the intentional relation involves both first person information about the intentional relation of the self and third person information about the intentional relation of the other, it cannot be assigned independently to either the self or the other. In contrast, in stage 2, the development of the imagination allows the intentional relation that is not given in current experience - third person for self, first person for other. In this way, on seeing another's behavior, the child will be able to imagine the first person information of the other. Similarly, while experiencing first person information about their own intentional relation, the child will be able to imagine the corresponding third person information. With both informational components available at the same time for either self or other, the child can integrate them so that a full intentional relation is represented for either self or other. In short, during stage 2 of social understanding, children can be said to have a concept of intentional agency (Barresi & Moore, 1993) in the sense that they understand that self and other are separate but similar in that they are both objective entities which exhibit behaviors directed at objects with accompanying first person experience of those objects. In this context, it is noteworthy that children start to use personal pronouns and gesures such as pointing to pick out individual others or the self at about 18 months of age (Bates, 1990).

As mentioned, there are two complementary sides to the understanding of intentional relations in stage 2. On the one hand, the child becomes aware of third person information about the self, which is integrated with first person information about the self. On the other hand, the child becomes aware that others have first person information about their activities in relation to objects, which gets integrated with the third person information about them. Perhaps the clearest sign that the child is aware of third person information about the self is self-recognition. Self-recognition is usually tested using the marked face paradigm developed by Gallup (1970) with chimpanzees and Amsterdam (1972) with infants, in which the subject is surreptitiously marked with rouge on the nose and then observed in front of a mirror. Behavior directed towards the subject's own nose as opposed to the mirror in such a situation appears during the second half of the second year (e.g. Amsterdam 1972; Lewis & Brooks-Gunn, 1979). Self-recognition indicates that the child is able to imagine self as an object like other objects and in particular like other persons whose third person information the child directly perceives. With the development of the imagination such imagined third person information of the self becomes a possibility. Strictly speaking, such self-recognition can be achieved without the use of intermodal integration through the intentional schema because it does not require the imagined third person information of self to be integrated with the first person information of the self (cf. Mitchell, 1993). However, at about the same time, children also show evidence of self-conscious emotions, such as coyness and embarrassment (Lewis, Sullivan, Stanger & Weiss 1989). The latter forms of behavior suggest that children's own intentional relation is affected by their imagination of how they appear from the intentional perspective of another. If this interpretation is appropriate then these emotions would require not only imagined third person information of self but an integration of first with third person information of the self through the intentional schema.

To determine unambiguously that children understand that others have first person information, it is necessary to show that the children can behave appropriately in relation to the other's first person information rather than their own. Again, the first clear signs of this understanding appear from the second half of the second year on. Perhaps the best example is in the development of empathy (e.g. Hoffman 1977). Younger infants, on perceiving another's exhibition of distress, will typically get distressed themselves and seek comfort for themselves. As the second year proceeds, infants in the same situation, while showing some distress themselves, will try to do something to bring comfort to the other. It is at this point, therefore, that they appear to understand that the feeling belongs to the other. In understanding the other, unlike in understanding the self, it is the first person information that has to be imagined and integrated with the immediately available third person information of the other. We assume that in the most primitive or novel circumstances in which the intentional relation of another is to be understood, the first person information of the other is imagined using the child's first person information of self as a model (Harris 1989, 1991). In other words, the imagined first person information of the other is essentially the same as the subject's own first person information, but marked as belonging to the other. For example, when viewing another's distress early in this period, the child empathically feels distress, but attributes it to the other. However, with experience, it is likely that the understanding of intentional relations of others will no longer require an appropriate empathic response from the child. Rather, the child will recognize the third person information of the other and the first person component will be imagined. In this way, young preschoolers can recognize the diversity of intentional orientation of self and other to a particular object in some circumstances. For example, Flavell, Flavell, Green & Moses (1990) have shown that 3-year-olds, presented with a favorite cookie and then shown another person express disgust after taking a bite of the cookie can understand that, even though they like the cookie, the other does not.

We are proposing here that the understanding of third person information of the self and the understanding of first person information of others both depend on the capacity for imagination. This proposal suggests that the understanding of self and other develop in parallel (an idea that goes back at least as far as Baldwin, 1894) and there is some empirical support for this (see also Perner, 1991). A number of studies have shown a relationship between self- recognition and empathy (e.g. Bischof-Kohler, 1988; Johnson, 1982). Most recently, for example, Bischof-Kohler (1988) found a correlation, independent of age, between self-recognition behavior using the marked face paradigm and empathic behavior directed at a distressed adult in an experimental setting.

Although the imaginative abilities of stage 2 allow differentiation of the intentional relations of self and other, intentional understanding is by no means complete. The limitation of this stage is that, because of restrictions on the child's imaginative power (e.g., Olson, 1993), the understanding of the intentional relations of both self and others involves combining one current perceptual component and one imagined component. Thus, while children at this stage can recognize diversity of intentional relations, their understanding is still tied to current experience. 4.3 Stage 3 (level 4): Children's theories of mind and false belief

Much recent research on social understanding during the preschool period has been conducted under the heading of 'the child's theory of mind' (see Astington et al. 1988; Frye & Moore 1991; Wellman 1990; Whiten 1991a). Rather than giving a comprehensive review of the research on children's theories of mind, we shall concentrate here on one central issue in this literature - the understanding of false belief. False belief has been considered a kind of litmus test (e.g., Chandler 1988; Wellman 1988) for the presence of a theory of mind since Premack and Woodruff's (1978) seminal work with chimpanzees. Premack and Woodruff's target article generated intense interest among commentators with respect to what would be required as firm evidence for a theory of mind (e.g., Bennett 1978; Dennett 1978). That interest spread into developmental psychology with the work of Wimmer and Perner (1983) who, responding to the suggestions of the commentators, developed the first false belief task.

In their task, Wimmer and Perner (1983) presented children with a story in which a character, 'Maxi', places some chocolate in one cupboard in the kitchen and leaves. While Maxi is away, another character removes the chocolate from the first cupboard, replaces it in a second cupboard, and then also leaves. Maxi then returns, and the child subject is asked to predict where he will look for the chocolate. The correct answer, of course, is to predict that he will look in the empty cupboard and not in the cupboard where the chocolate really is. Wimmer & Perner (1983) found that it was not until about four to five years of age that children performed correctly on this task. Younger children typically judged according to the real location of the object. On the basis of this result and others showing a developmental shift in performance on various measures of false belief understanding at about 4 years of age (e.g. Gopnik & Astington 1988; Moore, Pure & Furrow 1990; Perner, Leekam & Wimmer 1987), some have argued that what develops at this age, as indexed by success on false belief tasks, is the understanding that beliefs are attitudes towards representations of reality rather than towards reality itself, or what has been termed 'a representational theory of mind' (Astington & Gopnik 1991; Perner 1988; 1991).

While the finding that it is not until about four years that children understand false belief is a fairly robust one (but see Freeman, Lewis & Doherty 1991; Mitchell & Lacohee 1991; Siegal & Beattie 1991), younger children do have some knowledge of belief- desire reasoning. Most important, perhaps, they appear to know a considerable amount about desire and true belief or knowledge. Wellman (1990) has provided the most extensive coverage of three year old children's belief-desire reasoning. Using brief story tasks, in which the subject is told a story about a character who wants something or thinks something, Wellman and his colleagues (Wellman & Bartsch 1988; Wellman & Woolley 1990) have shown, among other things, that three-year-olds know that people will act on the basis of their desires and knowledge. As previously noted, three-year-olds can also recognize the distinctness of intentional relations between self and other to a particular object in some circumstances (Flavell et al. 1990).

The questions of theoretical importance to us here are: what makes the false belief task particularly difficult, and what accounts for the apparent change in social understanding at about four years of age? Let us consider what it means to fail the false belief task. This task requires children to predict what the story character will think or do and children younger than about four years old, as noted, typically fail the standard version because they answer on the basis of the real location of the hidden object. But what is meant by 'the real location' in this context? 'The real location' clearly does not refer to the state of reality independent of any agent or observer, because one could imagine that if, unbeknownst to the 3-year-old child subject, the object in the false belief task were moved again, that subject would still predict that the agent will search in the location where the subject believed the object to be. For the purposes of the task, 'reality' essentially means 'the content of the child subject's own current belief' or the situation as it appears to the child through current first person information. There are, then, two critical aspects to note about the task. First, the task requires the child to generate a representation of the agent's epistemic relation to the situation, for which that agent's third person information is not perceptually given. Second, the task also requires the child to imagine the first person information of the agent's intentional relation, while the first person information about the child's own intentional relation is different. Both these aspects are critical because together they ensure that neither current third person information nor current first person information are available to generate the requisite representation of the intentional relation. As a result, both components of the agent's to-be-predicted intentional relation must be imagined.

Our argument, then, is that the Wimmer and Perner (1983) false belief task and its variants (e.g. Perner et al. 1987) are particularly difficult because they require the subject to generate a representation of the intentional relation of an agent when neither the first nor third person informational component is provided by current experience. It is under this particular set of conditions that the younger child has difficulty because, as we have already suggested, younger children are able to imagine only one component of an intentional relation. When presented with perceptual third person information about others, they are able to imagine the coordinate first person information. However, in the absence of perceptual third person information about the other, the representation of the other's intentional relation can only be generated by using the current first person information of self, and in doing so the children make the wrong prediction.

The tasks on which younger children show reasonably good abilities tend not to have both of the critical aspects just outlined. On the one hand, Flavell et al.'s (1990) task, which shows understanding of emotional diversity, presents the child with perceptual third person information about the other; hence the child need not generate a representation of the other's intentional relation, which involves both imagined third person information and imagined first person information. The child can deal with the diversity of emotional orientation by comparing two attitudes, both of which involve one perceived and one imagined component. On the other hand, tasks assessing true belief or knowledge, while they may require imagining the third person information about the other, are relatively straightforward because the subjects' use of their own current first person information to generate the representation of the intentional relation of the other will lead to an accurate prediction. Similarly, if one examines the research on desire, it can be argued that there have been few, if any, tasks used to assess desire understanding that are clearly analogous to the false belief task. In other words, in virtually no case have subjects been presented with a situation in which they must predict the desire of another person while themselves holding a conflicting current desire. The only case that comes close to being an assessment of desire understanding in the presence of a current conflicting desire is a study by Gopnik and Slaughter (1991), in which children had to recall their own previous desire while holding a different current desire. In this task, there was no available perceptual information, of either first or third person kind, on which to base an answer to the critical test question. Half of the 3-year-olds failed this task, answering that they had previously held the desire they now held.

If our argument so far is correct, then what develops at about four years is the capacity to perform what amounts to a double imaginative act, whereby the child can generate, through the use of the intentional schema, a representation of an intentional relation for which both first and third person information are imagined. As with the previous stage, we believe that the change in social understanding at about four years is made possible by a domain- general cognitive advance (see also Frye, Zelazo, & Palfai, 1993; Olson, 1993). The double imaginative act needed to understand false belief would require the subject to hold in mind or imagine the two informational components - first and third - of an agent's epistemic relation. The use of the intentional schema would then generate a novel representation for the agent's intentional relation to a mental object. Comparing this imagined epistemic relation with the subject's own would yield the notion of false versus true belief for cases in which the imagined relation mismatched or matched perceptual reality. This double imaginative ability would also allow older children not only to fully imagine the intentional relations of others, as in the false belief task, but also to represent their own intentional relation as an intentional relation existing 'here, now, and for me', something that is potentially false. Hence the diversity of intentional relations across space, time, and person can be represented. These are the hallmarks of what has been termed 'a representational theory of mind' (Perner 1988; 1991).

5. Pathology of Social Understanding: Autism

In recent years, there has grown a large literature assessing the possibility that autism is a kind of tragic 'null phenotype' with respect to social understanding. Autism is a severe childhood psychopathology with onset prior to three years. It is typically characterized as involving three main behavioral abnormalities - impaired social behavior, symbolic disability, including language impairment and lack of symbolic play, and a strong need for stereotyped routines in daily life. On the basis of recent research from the theory-of-mind perspective demonstrating that autistic children perform considerably worse than mentally retarded controls on a variety of tasks assessing an understanding of other minds, a number of authors have proposed that autism is a relatively specific deficit with respect to the understanding of other minds (e.g., Baron- Cohen 1991; Baron-Cohen, Leslie & Frith 1985; 1986; Frith 1989; Harris 1989; 1991; Hobson 1989; 1990; 1993; Leslie 1987; 1988; Leslie & Frith 1990).

It is clear, however, that the problems of autism do not commence at 3 to 4 years, when normal children have a concept of mental agent and start to pass false belief tasks. Retrospective studies suggest that, on average, autism becomes noticeable to parents as an abnormality in social behavior during the first two years of life (e.g., Short & Schopler 1988). Autistic children do not show social pretend play, a type of behavior that emerges in normal children during the second year of life (e.g. Sigman & Ungerer 1984; Wing, Gould, Yeates & Brierley 1977). Other research has shown that autistic children show abnormalities in behaviors such as joint attention, which may well be precursors to the difficulties seen in standard theory of mind tasks (Loveland & Landry 1986; Sigman, Mundy, Ungerer & Sherman 1986). In a similar vein, autistic children produce protoimperative but not protodeclarative communicative acts (e.g. Baron-Cohen 1989; Curcio 1978; Landry & Loveland, 1989; Mundy, Sigman, & Kasari, 1990), showing that, although they express their own desires, they do not act as though they recognize that there is a potential for joint attention or that others have intentions.

Imitation has also been shown to be impaired in autism and this impairment may point to an even earlier origin of the autistic problem (for a review see Smith & S.E. Bryson, 1994). Dawson & Adams (1984) found that about half of their autistic subjects were operating imitatively at stages 2 and 3 of the sensorimotor period, stages that are reached in normal children at about 2 to 4 months (Piaget, 1962). When attempts to imitate are seen in autistic children, they may show some curious abnormalities. For example, Ohta (1987) found a pattern of "partial imitation" of manual gestures in a substantial proportion of autistic subjects. In these cases, the subjects, who were seated face-to-face with the model, produced a gesture which reversed the orientation of the hands so that the visual appearance of their own hands most closely resembled the model's hands. Such attempts at imitation, therefore, appeared to be controlled only by visual information.

Other authors have supposed that autism is primarily a deficit existing in the affective domain (Kanner 1943; Hobson 1989; 1990; 1993; Mundy & Sigman 1989). For example, Hobson (1993) marshalls a number of kinds of evidence in support of its origins in affective interpersonal relatedness. First, clinicians since Kanner's (1943) original work have described how interactions with autistic children seem to lack affective contact. Such clinical impressions are supported by empirical work which shows that autistic children's emotional expressiveness, both vocal and facial, when in the company of others is abnormal (e.g., Dawson, Hill, Spencer, Galpert & Watson 1990; Kasari, Sigman, Mundy, & Yirimiya, 1990; Yirmiya, Kasari, Sigman & Mundy 1989). In addition, autistic children are significantly worse than control children in categorizing and determining the meaning of emotional expressions (e.g. Hobson 1986; Hobson, Ouston, & Lee 1989).

On the basis of these varied results, we suggest that if there is a single simple deficit responsible for the difficulties in social understanding seen in autism, it is probably that first and third person information are not integrated in the representation of own and other agents' intentional relations. This inability exists at the most basic level; in other words, autistic individuals never attain an understanding of intentional relations at level 2 of our model. The failure of social understanding to proceed beyond level 1 of our framework could occur for a number of reasons. As we argued in previous sections, the representation of intentional relations between agents and objects at level 2 requires that the individual be able to integrate matched first and third person information with respect to objects. Given that the occurrences of matched states in interactions in early infancy are largely maintained by the adult rather than the infant (Schaffer 1984), it seems likely that matching of first and third person information does occur in the early developmental contexts of autistic children. Thus, it is more likely that the problem is with the extent to which the autistic infant attends to and processes such matched information. In other words, the intermodal integration of first and third person information is either disordered or absent. The extent of this information-integration problem is so far unknown. It may be specific to the processing of certain types of matched intentional information, for example, the epistemic information available in joint attention (Baron-Cohen 1991) or emotion information (Hobson, 1989, 1990). However, given that the literature has implicated action, emotional and epistemic deficits in autism, we suspect that the problem in autism is a general one with respect to the integrated representation of matched first and third person information. Further, given that autistic children appear disordered in very basic affective and imitative abilities (Dawson & Adams 1984; Hobson, 1993; Ohta 1987), we suggest that the problem is not limited to the integration of first and third person information about intentional relations, but affects all intermodal social information processing, even that prior to the stage when objects are normally brought into the interaction (see also Rogers & Pennington, 1991). It may even be that the problem is a very general one in information processing, not limited to social information at all. Given that there is some evidence about the problems that autistic individuals suffer in the integration of information from more than one modality, such as vision and audition (e.g. C. Q. Bryson 1970; 1972), it is entirely possible that autism is the result of a pervasive problem in integrating information across any modalities. Such a general difficulty with intermodal integration may itself be the product of an attentional impairment that prevents the relatively rapid shifts of attention across sensory modalities that are required for bimodal integration of information (Courchesne et al. 1993).

Whether the problem with information integration is a general or limited to social and perhaps just intentional information, it is relatively straightforward to see, from our account, how such an information-processing deficit would generate the severe problems in social understanding that occur in autism. As a result of this failure to integrate first and third person information, the autistic person is unable to enter into the shared intentional experience of level 2, normally evident in infants towards the end of their first year in behavior such as joint attention, and protodeclarative communicative acts. This would not interfere with the representation of their own desires from a first person perspective as well as the actions of others from a third person perspective. It would not, therefore, affect their acquisition of proto-imperative communicative skills, although the latter would tend to be quite primitive, taking advantage only of the mechanical causal properties of the other, e.g. pulling or dragging a person. As an inevitable result of failure to integrate first and third person information, autistic individuals are unable to imagine intentional relations as properties of separate agents, including both self and other, at level 3 of our framework. Furthermore, although they may be able to recognize differences in knowledge among different others (Kazak, 1992) on the basis of third person information alone, the representational theory of mind characteristic of level 4, which requires imagining divergent epistemic relations across self and other would remain beyond the ability of the person with autism. Despite the sophistication of the knowledge that autistic individuals acquire about people including themselves, the form of their representations of people remains entirely of either a first or third person sort.

Although according to our framework the autistic person use only level 1 representations, we are not suggesting that their representation of the intentional relations of self or other is limited to the same range as that found in lower organisms. We assume, for example, that their ability to use the imagination is intact or at least in line with their general level of mental retardation. The use of the imagination would support abilities such as mirror self-recognition (see Mitchell, 1993), which develops in autistic individuals at approximately the same time as in mentally retarded controls (Neuman & Hill 1978; Spiker & Ricks 1984). Furthermore, they acquire some use of language, which they apply to self and other (Frith 1989). Despite these indications of a lack of asymmetry in the understanding of intentional relations between self and other, however, we believe that such an asymmetry maintains itself. The autistic person acquires an ability to imagine first person information and generalize it to others, and to imagine third person information and generalize it to self without ever integrating these informational components into a representation of intentional relations that can distinguish diversity in perspectives. A clear indication of this failure to integrate first and third person information is seen in their behavior in front of a mirror. Autistic individuals can imagine third person information about the self in order to recognize themselves in the mirror. However, they do not show any signs of self-conscious emotions (Neuman & Hill 1978; Spiker & Ricks 1984), implying a failure to integrate first and third person information for both the self and others. Consistent with this interpretation is the fact that autistic individuals refer to themselves as well as to others by name, but they rarely use personal pronouns, which, being deictic, require an appreciation of the intentional perspective of the speaker (Hobson, Lee & Chiat in press).

In sum, we believe that social understanding deficits in autism are the result of an inability to form representations of intentional relations with integrated first and third person components. This inability, we believe, is a general one, so that the understanding of both own and others' epistemic, affective, and action intentional relations is disordered. The origin of the disorder appears to be in the extent to which autistic individuals are able to attend to or integrate information across sensory modalities so as to extract information about perceptual invariants and build representations using such invariants.

6. Conclusion

In this target article, we have presented a theory of social understanding and the forms of representation of intentional relations. The theory describes a system of such forms that can be applied uniformly to self and other, and that requires the use of an intentional schema that can integrate the first person information about an intentional relation with the third person information about that relation. We have suggested a four level framework which describes the various forms of representation that do and do not require the intentional schema. The framework was applied to empirical evidence on the phylogeny and ontogeny of social understanding, and on human autism.

Although some aspects of our theory will no doubt be disconfirmed, we believe that it addresses certain issues in social understanding, which are not, in themselves, open to disconfirmation. For example, it is a necessity of organismic existence that one have a qualitatively different kind of information about one's own intentional relations and about the intentional relations of others. Nevertheless, human beings have developed forms of representation of intentional relations that can be applied almost uniformly to self and other. How is this possible?

The current literature offers three apparent answers to this question, none of which seems to us entirely satisfactory. Those who postulate a "theory of mind" (e.g., Gopnik, 1993; Perner, 1991) succeed in accounting for the uniformity in applying the theory to self and other, but at the expense of providing a coherent account of how the theory takes hold of the first and third person information about intentional relations (Moore & Barresi, 1993). Those who postulate a process of imaginative simulation to account for social understanding typically focus on using one's own mind in simulating the mind of the other (e.g., Goldman, 1989, 1992; Harris, 1989, 1991; Humphrey, 1984), but they do not make it clear how one can know one's own mind as such. Other simulation theorists suggest that we acquire knowledge about both self and other through simulation (e.g. Gordon, 1986; 1987; 1992), but they are not explicit on how this understanding comes about, especially in its original nonimaginative form. A third view, which postulates an original empathic relation between human persons (e.g. Hobson, 1990; 1991), although dealing with origins of social understanding, fails to be explicit about its central mechanism of empathic relatedness.

Our proposal contains aspects of all three of these approaches without reducing to any one of them. We suggest that in human beings there has evolved a capacity to enter into shared intentional relations with another and an associated intentional schema to integrate first and third person information about these intentional relations. The resulting representations could be applied uniformly to self and other in comparable situations where only part of the informational source is available perceptually. This is one possible solution to the problem; there may be others. In any event, once knowledge of particular intentional relations has been acquired, the door is open to use the imagination and folk theories to further expand the understanding of intentional relations where only one, and eventually where no, direct input is available. At root, however, such understanding will always involve aspects of experience that have their source in the first person information about our selves and the third person information about others, and our ability to integrate these two forms of information into a single representation. This capacity to integrate knowledge about self and other is the basis for human social understanding and our appreciation for all aspects of the human condition.

Footnotes

1. The term "intentional relation" derives in part from the concept of intentionality. Intentionality is a property of mental activities that has been described in various ways since Brentano (1874/1973) brought it into modern usage in an attempt to distinguish mental phenomena from physical phenomena (See Barresi, 1989; Chisholm, 1967; Dreyfus & Hall, 1984) . Brentano's definition of intentionality focused on the intentional "inexistence" of the objects of mental acts, by which he meant that mental acts are "directed toward", make "reference to" or are "about" objects that need not exist in reality but exist only as objects contained within these acts. These "inexistent" objects occur in mental phenomena in various modes, such as being thought about or being desired. In the present paper we focus not on intentionality per se, but on intentional relations. In our terminology, intentional relations exist between an organism and actual or possible objects or states of affairs. They depend on sensory-motor and higher mental capacities of the organism and have their roots in certain activities that can be defined in nonintentional causal terms (see, e.g., Agar, 1993; Dretske, 1981, 1988; Fodor, 1987, 1990; Millikan, 1989; Shapiro, 1992). Our presupposition is that any empirical account of social understanding will have to begin with how organisms appreciate intentional relations involving agent's activities directed at real rather than with inexistent objects and that the latter understanding must depend on the former. Hence we begin with intentional relations rather than with intentionality. 2. This argument is rooted in the philosophical tradition of Merleau- Ponty (1964), Nagel (1974), Scheler (1954), Searle (1992), Strawson (1959), Wittgenstein (1958). Like ourselves, these authors were concerned with the problem of how our knowledge of mental activities of self and others can be based on two qualitatively different kinds of information yet apply equally to self and other. Their answers differ from each other and ourselves, yet all presuppose that there are two distinct kinds of information about mental activities. For instance, Nagel (1974) seems to have the view that there are two distinct forms of representation of mental activities, subjective and objective, but that both of these can be applied to self and other. In contrast, Wittgenstein (1958) seems to believe that there is single linguistic form of representation for mental activities, but that our criteria for the application of this form is different when we apply it to ourselves and others. 3. Although we are claiming that there is a uniform conceptual system for understanding the intentional relations of self and other, despite the differences in informational sources, this is not to say that there are no differences between understanding self and understanding others. Research in social cognition has revealed a number of instances of such differences and some of these instances provide clear support for the idea that first and third person information tend to focus on different aspects of the intentional relation. One such example is the actor-observer effect (Jones and Nisbett, 1972). In this effect, attributions of the causes of one's own behavior tend to focus on the situation that one is in, in other words, on the objects of one's intentional relations, whereas attributions of the causes of others' behavior tends to focus on the characteristics of the person acting. Thus, attributions for self tend to inherit the quality of first person information for self and those for others tend to inherit the quality of third person information.

REFERENCES

Agar, N. (1993). What do frogs really believe? Australasian Journal of Philosophy 71:1-12.

Amsterdam, B. K. (1972). Mirror self-image reactions before age two. Developmental Psychology 5:297-305.

Anisfeld, M. (1991). Neonatal imitation: A review. Developmental Review 11:60-97.

Astington, J. W. & Gopnik, A. (1991). Developing understanding of desire and intention. In: Natural theories of mind. Evolution, development, and simulation of everyday mindreading, ed. A. Whiten. Blackwell.

Astington, J. W., Harris, P. L. & Olson, D. R., eds. (1988). Developing theories of mind. Cambridge University Press.

Bahrick, L. E., & Watson, J. S. (1985). Detection of intermodal proprioceptive-visual contingency as a potential basis of self- perception in infancy. Developmental Psychology 21:963-973.

Bakeman, R. & Adamson, L. (1984). Coordinating attention to people and objects in mother-infant and peer-infant interactions. Child Development 55:1278-1289.

Baldwin, J. M. (1894). Mental development in the child and the race. Macmillan.

Baron-Cohen, S. (1989). Perceptual role taking and protodeclarative pointing in autism. British Journal of Developmental Psychology 7:113-128.

Baron-Cohen, S. (1991). Precursors to a theory of mind: Understanding attention in others. In: Natural theories of mind. Evolution, development, and simulation of everyday mindreading, ed. A. Whiten. Blackwell.

Baron-Cohen, S., Leslie, A. M. & Frith, U. (1985). Does the autistic child have a "theory of mind"? Cognition 21:37-46. Baron-Cohen, S., Leslie, A. M. & Frith, U. (1986). Mechanical, behavioral, and Intentional understanding of picture stories in autistic children. British Journal of Developmental Psychology 4:113-125.

Barresi, J. (1989). Prolegomena toward a causal theory of mind and meaning (Review of: J. Fodor, Psychosemantics: The Problem of Meaning in the Philosophy of Mind.). American Journal of Psychology 102:122-130.

Barresi, J. & Moore, C. (1993). Sharing a perspective precedes the understanding of that perspective. Behavioral and Brain Sciences 16:513-514.

Bates, E. (1979). Intentions, conventions, and symbols. In: The emergence of symbols, ed. E. Bates, L. Benigni, I. Bretherton, L. Camaioni & V. Volterra. Academic Press.

Bates, E. (1990). Language about me and you: Pronominal reference and the emerging concept of self. In: The self in transition. Infancy to childhood, ed. D. Cicchetti & M. Beeghly. University of Chicago Press.

Bennett, J. (1978). Some remarks about concepts. Behavioral and Brain Sciences 1:557-560.

Bischof-Kohler, D. (1988). Uber der Zusammenhang von Empathie und der Fahigkeit, sich im Spiegel zu erkennen. Schweizerische Zeitschrift fur Psychologie 47:147-159.

Bitterman, M. E. (1987). Evidence of divergence in vertebrate learning. Behavioral Brain Science 10:659-660

Brazelton, T. B., Koslowski, B. & Main, M. (1974). The origins of reciprocity. In: The effect of the infant on its caregiver, ed. M. Lewis & L. Rosenblum. Wiley.

Brentano, F. (1874/1973). Psychology from an Empirical Standpoint. London: Routledge & Kegan Paul.

Bretherton, I. (1991). Intentional communication and the development of mind. In: Children's theories of mind. Mental states and social understanding, ed. D. Frye & C. Moore. Hillsdale, N.J.: Erlbaum.

Bruner, J. (1983). Child's talk. Learning to use language. New York: Norton.

Bryson, C. Q. (1970). Systematic identification of perceptual disabilities in autistic children. Perceptual and Motor Skills 31:239-246.

Bryson, C. Q. (1972). Short-term memory and cross-modal information processing in autistic children. Journal of Learning Disabilities 5:25-35.

Butterworth, G. & Jarrett, N. (1991). What minds have in common is space: Spatial mechanisms serving joint visual attention in infancy. British Journal of Developmental Psychology 9:55-72.

Byrne, R. W. & Whiten, A. eds. (1988). Machiavellian intelligence. Social expertise and the evolution of intellect in monkeys, apes, and humans. Oxford University Press.

Campos, J. J. (1983). The importance of affective communication in social referencing: A commentary on Feinman. Merrill-Palmer Quarterly 29:83-87.

Chandler, M. J. (1988). Doubt and developing theories of mind. In: Developing theories of mind, ed. J. Astington, P. Harris & D. Olson. Cambridge University Press.

Cheney, D. L. & Seyfarth, R. M. (1990). How monkeys see the world: Inside the mind of another species. University of Chicago Press.

Cheney, D. L. & Seyfarth, R. M. (1992). Precis of How monkeys see the world. Behavioral and Brain Sciences 15:135-182.

Chisholm, R. (1967). Intentionality. In: The Encyclopedia of Philosophy, Vol. IV, ed. P. Edwards. Macmillan Publishing Co.

Courchesne, E., Akshoomoff, N. A., Townsend, J., Yeung-Courchesne, R., Lincoln, A. J., James, H. E., Haas, R. H., Schreibman, L. & Lau, L. (1993). A new finding: Impairment in shifting attention in autistic and cerebellar patients. In S.H. Broman & J. Grafman, J. (Eds.), Atypical cognitive deficits in developmental disorders: Implications for brain functions. Hillsdale, NJ: L. Erlbaum.

Curcio, F. (1978). Sensorimotor functioning and communication in mute autistic children. Journal of Autism and Childhood Schizophrenia 8:281-292.

Dawson, G. & Adams, A. (1984). Imitation and social responsiveness in autistic children. Journal of Abnormal Child Psychology 12:209-225.

Dawson, G., Hill, D., Spencer, A., Galpert, L. & Watson, L. (1990). Affective exchanges between young autistic children and their mothers. Journal of Abnormal Child Psychology 18:335-345.

Dennett, D. C. (1978). Beliefs about beliefs. Behavioral and Brain Sciences 1:568-570.

Dennett, D. C. (1987). The intentional stance. M.I.T. Press.

Dickinson, A. (1989). Expectancy theory in animal conditioning. In: Contemporary learning theories: Pavlovian conditioning and the status of traditional learning theory, ed. S. Klein & R. Mowrer. Lawrence Erlbaum Associates.

Dretske, F. (1981). Knowledge and the flow of information. Blackwell.

Dretske, F. (1988). Explaining behavior: Reasons in a world of causes. M.I.T. Press.

Dreyfus, H. L. & Hall, H. eds. (1984). Husserl, intentionality, and cognitive science. M.I.T. Press.

Feinman, S. (1982). Social referencing in infancy. Merrill-Palmer Quarterly 28:445-470.

Flavell, J. H., Flavell, E. R., Green, F. L. & Moses, L. J. (1990). Young children's understanding of fact beliefs versus value beliefs. Child Development 61:915-928.

Fodor, J. A. (1987). Psychosemantics: The problem of meaning in the philosophy of mind. M.I.T. Press

Fodor, J. A. (1990). A theory of content and other essays. M.I.T. Press.

Freeman, N. H., Lewis, C. & Doherty, M. J. (1991). Preschoolers' grasp of a desire for knowledge in false belief prediction: Practical intelligence and verbal report. British Journal of Developmental Psychology 9:139-157.

Frith, U. (1989). Autism: Explaining the enigma. Blackwell.

Frye, D. & Moore, C. eds. (1991). Children's theories of mind. Erlbaum.

Frye, D., Rawling, P., Moore, C., & Myers, I. (1983). Object-person discrimination at 3 and 10 months. Developmental Psychology 19:303-309.

Frye, D., Zelazo, P., Palfai, T. (1993). The cognitive basis of theory of mind. Unpublished manuscript, New York University.

Galef, B. G. (1988). Imitation in animals: History, definition, and interpretation of data from the psychological laboratory. In: Social learning: Psychological and biological perspectives, ed. T. Zentall & B. Galef. Hillsdale, NJ: Erlbaum.

Galef, B. G. (1990). Tradition in animals: Field observations and laboratory analyses. In: Comparative perspectives, ed. M. Beckoff & D. Jamieson. Westview Press.

Gallup, G. G. (1970). Chimpanzees: Self-recognition. Science 167:86- 87.

Gallup, G. G. (1977). Self-recognition in primates: A comparative approach to the bidirectional properties of consciousness. American Psychologist 32:329-338.

Gallup, G. G. (1982). Self-awareness and the emergence of mind in primates. American Journal of Primatology 2:237-248.

Gewirtz, J. L. & Pelaez-Nogueras, M. (1992). Social referencing as a learned process. In: Social referencing and the social construction of reality in infancy, ed. S. Feinman. Plenum.

Goldman, A. I. (1989). Interpretation psychologized. Mind & Language 4:161-185.

Goldman, A. I. (1992). In defense of the simulation theory. Mind & Language 7:104-119.

Gomez, J. C. (1991). Visual behavior as a window for reading the mind of others in primates. In: Natural theories of mind. Evolution, development, and simulation of everyday mindreading, ed. A. Whiten. Blackwell.

Goodall, J. (1990). Through a window. My thirties years with the chimpanzees of Gombe. Houghton Mifflin Company.

Gopnik, A. (1993). How we know our minds: The illusion of first- person knowledge of intentionality. Behavioral and Brain Sciences 16:1-14.

Gopnik, A. & Astington, J. W. (1988). Children's understanding of representational change and its relation to the understanding of false belief and the appearance-reality distinction. Child Development 59:26-37.

Gopnik, A. & Slaughter, V. (1991). Young children's understanding of changes in their mental states. Child Development 62:98-110.

Gordon, R. M. (1986). Folk psychology as simulation. Mind and Language 1:156-171.

Gordon, R. M. (1987). The structure of emotions. Cambridge University Press.

Gordon, R. M. (1992). The simulation theory: Objections and misconceptions. Mind & Language 7:5-27.

Greenfield, P. M. & Smith, J. H. (1976). The structure of communication in early language development. Academic Press.

Harris, P. L. (1989). Children and Emotion. Blackwell.

Harris, P. L. (1991). The work of the imagination. In: Natural theories of mind: The evolution, development, and simulation of everyday mindreading, ed. A. Whiten. Blackwell.

Heyes, C. (1993). Anecdotes, training, trapping, and triangulating: Do animals attribute mental states? Animal Behavior 46:177- 188.

Heyes, C. (1994). Reflections on self-recognition in primates. Animal Behavior 47:909-191.

Heyes, C. & Dickinson, A. (1990). The intentionality of animal action. Mind & Language 5:87-104.

Hobson, R. P. (1986). The autistic child's appraisal of expressions of emotion. Journal of Child Psychology and Psychiatry 27:321- 342.

Hobson, R. P. (1989). Beyond cognition. A theory of autism. In: Autism: Nature, diagnosis, and treatment, ed. G. Dawson. Guilford.

Hobson, R. P. (1990). On acquiring knowledge about people and the capacity to pretend: Response to Leslie. Psychological Review 97:114-121.

Hobson, R. P. (1991). Against the theory of 'Theory of Mind'. British Journal of Developmental Psychology 9:33-51.

Hobson, R. P. (1993). Understanding persons: The role of affect. In: Understanding other minds: Perspectives from autism, ed. S. Baron-Cohen, H. Tager-Flusberg, & D. Cohen. Cambridge University Press.

Hobson, R. P., Lee, A. & Chiat, S. (1991). Autism and "the self": An experimental study of personal pronoun comprehension and use. Paper presented at Society for Research in Child Development, Seattle.

Hobson, R. P., Ouston, J. & Lee, A. (1989). Naming emotion in faces and voices: Abilities and disabilities in autism and mental retardation. British Journal of Developmental Psychology 7:237-250.

Hoffman, M. L. (1977). Empathy, its development and prosocial implications. In: Nebraska symposium on motivation. Vol. 25. Social cognitive development, ed. C. B. Keasey. University of Nebraska Press.

Humphrey, N. (1984). Consciousness Regained. Oxford University Press.

Johnson, C. N. (1988). Theory of mind and the structure of conscious experience. In: Developing theories of mind, ed. J. Astington, P. Harris, & D. Olson. Cambridge University Press.

Johnson, D. B. (1982). Altruistic behavior and the development of the self in infants. Merrill-Palmer Quarterly 28: 379-388.

Jones, E. E. & Nisbett, R. E. (1972). The actor and the observer: Divergent perceptions of the causes of behavior. In E. E. Jones, D. E. Kanouse, H. H. Kelly, R. E. Nisbett, S. Valins, & B. Weiner (Eds.), Attribution: Perceiving the causes of behavior. Morristown, NJ: General

Learning Press.

Kanner, L. (1943). Autistic disturbances of affective contact. Nervous Child 2:217-250.

Kasari, C., Sigman, M., Mundy, P., & Yirimiya, N. (1990). Affective sharing in the context of joint attention interactions of normal, autistic, and mentally retarded children. Journal of Autism and Developmental Disorders 20:87-100.

Kazak, S. (1992). Understanding knowledge as a mental state in normal and autistic children. Unpublished Ph.D. thesis, University of Warwick.

Landry, S. H. & Loveland, K. A. (1989). The effect of social context on the functional communication skills of autistic children. Journal of Autism and Developmental Disorders 19:283-299.

Leslie, A. M. (1987). Pretense and representation: The origins of "theory of mind." Psychological Review 94:412-426.

Leslie, A. M. (1988). Some implications of pretense for mechanisms underlying the child's theory of mind. In: Developing theories of mind, ed. J. Astington, P. Harris, & D. Olson. Cambridge University Press.

Leslie, A. M. & Frith, U. (1990). Prospects for a cognitive neuropsychology of autism: Hobson's choice. Psychological Review 97:122-131.

Lewis, M. & Brooks-Gunn, J. (1979). Social cognition and the acquisition of self. New York: Plenum.

Lewis, M., Sullivan, M. W., Stanger, C. & Weiss, M. (1989). Self development and self-conscious emotions. Child Development 60:146-156.

Loveland, K. & Landry, S. (1986). Joint attention and language in autism and developmental language delay. Journal of Autism and Developmental Disorders 16:335-349.

Mackintosh, N. J. (1987). From null hypothesis to null dogma. Behavioral and Brain Sciences 10:689-690.

Macphail, E. M. (1987). The comparative psychology of intelligence. Behavioral and Brain Sciences 10:649-656.

McGurk, H. & McDonald, J. (1976). Hearing lips and seeing voices: A new illusion. Nature 264:746-748.

Meltzoff, A. N. (1990). Foundations for developing a concept of self: The role of imitation in relating self to other and the value of social mirroring, social modeling, and self practice in infancy. In: The self in transition. Infancy to childhood, ed. D. Cicchetti & M. Beeghly.University of Chicago Press.

Meltzoff, A. N. & Gopnik, A. (1993). The role of imitation in understanding persons and developing a theory of mind. In: Understanding other minds: Perspectives from autism, ed. S. Baron-Cohen, H. Tager-Flusberg, & D. Cohen. Cambridge University Press.

Meltzoff, A. N. & Moore, M. K. (1977). Imitation of facial and manual gestures by human neonates. Science 198:75-78.

Meltzoff, A. N. & Moore, M. K. (1983). Newborn infants imitate adult facial gestures. Child Development, 54, 702-709.

Merleau-Ponty, M. (1964). The primacy of perception. Northwestern University Press.

Miles, H. L. (1986). How can I tell a lie?: Apes, language and the problem of deception. In: Deception: Perspectives on human and nonhuman deceit, ed. R Mitchell & N. Thompson. State University of New York Press

Miles, H. L. (1990). The cognitive foundations for reference in a signing oranutan. In: "Language" and intelligence in monkeys and apes: Comparative developmental perspectives, ed. S. Parker & K. Gibson. Cambridge University Press.

Millikan, R. (1989). Biosemantics. The Journal of Philosophy 86:281-297.

Mitchell, P. & Lacohee, H. (1991). Children's early understanding of false belief. Cognition 39: 107-128.

Mitchell, R. W. (1993). Mental models of mirror self-recognition: Two theories. New Ideas in Psychology 11: 295-325.

Mitchell, R. W. (1994). The evolution of primate cognition: Simulation, self-knowledge, and the knowledge of other minds.