Figure 1. The M-representation (Leslie and Roth, 1993)
Return to
articles on mental architecture
Pretend
Play: Is it Metarepresentational?
Chris
Jarrold1*, Peter Carruthers2, Peter K Smith1
& Jill Boucher1
1Department of Psychology,
and 2Department of Philosophy,
University
of Sheffield
Sheffield,
U.K.
*The first author, to whom requests for reprints should be sent, is now at the following address: Department of Experimental Psychology, University of Cambridge, Downing Street, Cambridge, CB2 3EB, U.K.
Childrens ability to pretend, and the apparent lack of pretence in children with autism, have become important issues in current research on theory of mind, on the assumption that pretend play may be an early indicator of metarepresentational abilities. We review the issue of whether pretend play can be considered metarepresentational, by focusing on both theoretical arguments and relevant empirical evidence. We stress that it is crucial to examine different types and levels of pretend play, and argue that some kinds of pretence are not metarepresentational. We discuss whether certain contextualised abilities may provide the scaffolding by which fully fledged metarepresentational pretend play abilities develop.
Pretend
Play: Is It Metarepresentational?
The view that pretend play is metarepresentational, in such a way as to require the pretender to represent their own or anothers representation of a counter-factual state of affairs, was first explicitly proposed by Leslie (1987), and has since become widespread. There are important reasons for questioning this assumption, however, over and above the fact that it has become generally accepted. Firstly, it is clearly relevant to our understanding of the phenomena of pretence. Does pretend play, when it emerges at around two years of age[1] really require the child to represent anothers counterfactual representations? And even on the assumption that it does, would the child who pretends alone have to be reflexively representing their own non-literal representation of the world, as the metarepresentational account suggests, or could there be a fundamental difference between shared and solitary pretence? The question is thus whether there is something unique about pretence which separates it from other abilities emerging at the same time, such as the ability to seek the best means to satisfy a desire, which are thought to be non-metarepresentational.
Perhaps more importantly in the context of this discussion, pretend play is seen by many as a developmental landmark in the childs acquisition of a Theory of Mind. This term refers to the ability to reason and make inferences about anothers mental states, and presupposes the ability to hold beliefs about anothers beliefs, or to mentally represent anothers mental representation. As will be seen, this constitutes metarepresentational understanding. Consequently, proponents of the view that pretend play is metarepresentational have drawn a functional link between the two phenomena (Baron-Cohen, 1987, 1991; Leslie, 1987, 1988). A potential problem for this account arises from the fact that, while pretend play emerges at around two years, the ability to infer the beliefs of others is not seen in children until later in development. Mindreading abilities of this kind are typically tested by psychologists by asking children to predict the actions of a protagonist who holds a false belief. For example, children are shown a character X who hides an object at location A. In Xs absence the object is then moved to location B, and on their return the child is asked to predict where X will search[2]. It is not until around four years of age that children are able to override their own knowledge and answer correctly in terms of the protagonists intentional stance to the situation (e.g. Wimmer & Perner, 1983). The developmental lag between the emergence of pretence and a workable folk psychology implies one of two things: either that it must be something other than the metarepresentational aspects of false-belief tasks that makes them difficult, or that pretend play (at two) cannot be metarepresentational. The question then arises, whether children first metarepresent at two, or at four?
Those who have defended the metarepresentational nature of pretence have also been 'Theory-Theorists' about our ability to understand the mental states of ourselves and others, arguing that this ability is subserved by an implicit theory of the nature and functioning of the mind. The connection between pretence and theory of mind is that the same metarepresentational mechanisms which are held to underpin the latter are also believed to be involved in the former. However, not everyone believes that our everyday folk psychology is mediated by the use of a theory. In particular, it has been suggested that instead of metarepresentational theorising, common-sense psychology rests on the ability to simulate the position of the other person in question (Goldman; 1989, 1992; Gordon; 1986, 1992; Johnson, 1988), an ability which does not itself require metarepresentational thought (although of course it will characteristically issue in such thoughts). Those who adopt this approach also believe that there is a functional link between theory of mind and pretence, but maintain that the latter is not metarepresentational. Rather, the very same capacity to suppose, or to entertain a counterfactual hypothesis, which (they hold) is involved in pretence will also be employed whenever one simulates the mental states of another. So the question of the metarepresentational nature of pretence is a crucial locus of conflict between these two opposed accounts of common-sense psychology, at least as developed by their current adherents.
A final reason for investigating the nature of pretence comes from research into the psychology of the developmental disorder of autism. Many studies have convincingly shown that such individuals are severely delayed in their ability to pass false belief tasks of the form described above (e.g. Baron-Cohen, Leslie & Frith, 1985). Many other studies have shown a characteristic absence of pretence amongst children with autism (e.g. Baron-Cohen, 1987). This has led some to claim that delayed acquisition of the ability to manipulate mental metarepresentations may lie at the heart of this disorder (Leslie and Frith, 1990; Leslie & Roth, 1993). It has led others to claim that the crucial deficit in autism may be lack of imagination, or an impaired ability to construct and manipulate counterfactual representations (Currie, forthcoming, Harris, 1991). However, there is evidence (reviewed below) to suggest that children with autism can engage in pretence under certain circumstances. This result is problematic for both the theory-theorist's and the simulationist's accounts of the connection between pretence and mindreading ability. We will not attempt, in this paper, to show whether, or how, these difficulties may be resolved.
The status of pretence is therefore a question relevant to our understanding of the processes involved in folk psychology, in both normal and atypical development. Further, there are empirical reasons for suspecting that the metarepresentational assumption is flawed. These will be reviewed fully later in this paper, when relevant evidence from the following three domains will be considered in detail: normal pretend play development, normal theory of mind development, and the ability of children with autism to engage in pretence. Prior to a consideration of empirical issues, theoretical arguments for and against a metarepresentational position will be outlined. However, it is first necessary to define what we mean by metarepresentation.
In order to define metarepresentation, it is important to establish first what is meant by term representation. In his recent book Perner (1991) devotes a chapter to the nature of representations. He draws a distinction between the representational medium and the representational content, and notes that one and the same content can be represented by a number of different media. Thus a particular scene can be represented either pictorially or by description, and a particular situation can be represented by two distinct sentences. When we talk of representations in psychology we usually refer to the representational medium - to the mental entity that does the referring - rather than to its content.
Perner, following Goodman (1976) and Frege (1892/1960), also notes that a distinction must be drawn between representing (referring) and representing-as (sense). These correspond to two distinct constraints on the accuracy of a description of a representation. In some circumstances, and for some purposes, a representation may be described adequately by specifying its worldly referent, whereas in others, an adequate description will also have to convey the way in which it represents what it does. In passing on to someone a report of the weather, it will generally make no difference whether I say that the forecaster has said that it will be sunny tomorrow, or sunny on Tuesday, provided that tomorrow is Tuesday. Whereas it makes all the difference, in reporting Oedipus belief, whether I say that he takes himself to be married to Jocasta, or married to his mother. Perner argues that we need to concentrate on sense rather than reference, and therefore offers the following definition: A representation represents something as something.
There is another distinction that Perner does not draw, but which is highlighted by Currie (in press), and which is worth making at this point since it will be of importance later. This is between the content of a representation on the one hand, and the type of mental attitude involved in that act of representation, on the other. I can have a number of distinct mental attitudes that share the same content. For example I can believe that ice cream is in the fridge, I can wish that ice cream were in the fridge, or I can suppose that ice cream is in the fridge.
If metalanguage is language used to describe language and metacognition is knowledge about what I know, are metarepresentations therefore representations of representations (second- or higher-order representations)? In a sense they are. However, if representation properly defined is representing something as something, metarepresentation is therefore representing a representation as a representation (or more long-windedly, representing a representation as a representation of something as something). This is how Pylyshyn (1978) originally used the term - ability to represent the representational relation itself.
Having arrived at a definition of what constitutes a metarepresentation, we now consider the theoretical arguments for and against applying the term to the psychological processes involved in pretence. Three positions will be discussed in turn, those of Leslie, of Perner and of Harris.
Children who pretend are generally ascribed a double knowledge about the situation (McCune-Nicolich, 1981). That is to say, they are pretending that a banana is a telephone, for example, but at the same time they know that it is a banana really. Leslie (1987) argues that this poses a potentially disastrous problem for the child, namely that of representational abuse. How can a child who holds a primary representation (a literal, factual, representation defined by a direct semantic relation with the world) of a real object or situation, this is a banana, at the same time juggle a second representation, this is a telephone, when engaging in object substitution? How is it that the childs representational system is not totally undermined by this - is this a banana or is it a telephone? Both representations cannot be primary as they contradict each other semantically. To account for the childs ability to substitute readily a wide variety of objects without losing a grip on their literal meaning Leslie argues that the pretend representations must be quarantined off from primary representations in some way. This is done, he proposes, by the use of meta-, or second-order, representations.
Leslie argues that during an act of pretence the primary representation, this is a banana, is copied into another context, this is a banana. This secondary representation is decoupled from reality, and its reference, truth and existence relations are suspended; so representational abuse is avoided. The opacity afforded by the decoupling of the secondary representations input-output relations is supposed to allow the decoupled expression to be transformed without abusing the primary representation, as in this banana is a telephone. Leslie also suggests that the decoupled expression will be a second-order, metarepresentational, one, maintaining that it will be a representation of the primary representation.
In his original paper, Leslie sometimes seems to suggest that the mere fact that the mechanisms underlying pretence are supposed to involve copying a primary representation is sufficient to make pretence metarepresentational. This is clearly a mistake. It may be true that sometimes, a copy of a representation is at the same time a representation of that representation, but when, for example, an artist paints a portrait from a photograph of their subject, the result is a representation of the person in question, not of the photograph. The resulting portrait is not a metarepresentation, despite the fact that it was produced by copying a representation. The status of the copied representation depends on the way in which it is used. So when we consider Leslies postulated mechanism of decoupling, the question to ask is whether the copy of the primary representation, this is a banana, is used as a representation of the earlier representation. And there is, surely, not the slightest reason to believe that it is. On the contrary, it continues to be used as a representation of the banana, only now in connection with some unusual predicates, such as is a telephone, or is a source of sound.
Leslie in fact probably never intended to suggest that simply copying a primary representation constitutes metarepresenting. This is clear from his argument that the onset of pretend play in childhood occurs at the same time as the ability to understand pretence in others (Leslie, 1987). In terms of his theory, to understand pretence a child must be able to compute the relation PRETEND (a, ei, ej), where a is the agent and ei and ej are the secondary and primary representations respectively. Rather than suggest that the decoupled secondary representation is itself the metarepresentation, he proposes that metarepresentations take the general form: Agent - Informational Relationship - Expression. Here the decoupled secondary representation is the expression and the informational relationship is included to indicate the nature of the (non-automatic) relation between the decoupled expression and its corresponding primary representation. For example: Mother - PRETENDS - this is a banana.
Though he makes minor reformulations to his approach in subsequent papers (Leslie, 1988; Leslie & Frith, 1990), this is generally how Leslie defines metarepresentations, and as he is referring to a representation of a representational relation he is correct to do so. Put another way, Leslies metarepresentation is a representation of the secondary representation as a representation of the primary representation. Leslie and Roth (1993) have suggested that metarepresentations have a number of components: the agent, and the informational relationship between an aspect of reality (primary representation) and an imaginary situation (decoupled representation), see Figure 1. The secondary representation is therefore now seen as only one component of the metarepresentation as a whole. Leslie and Roth introduce the term M-representation to refer to this relational structure.
Figure 1 about here
There can be no doubt that Leslies M-representation is indeed metarepresentational. Were a child to have this sort of representation of another person representing something as something else they would be employing a metarepresentation. Does this prove the claim that pretence is metarepresentational? By no means. There may be ways of engaging in simple forms of shared pretence which do not presuppose that one actually represents the other as pretending (this will be discussed later). Moreover, solitary pretending must be considered separately from shared pretence, because even if it were true that a child begins to pretend at around the same time that they understand pretence in others, this need not imply that the same cognitive processes are operating in each case. The only reason for supposing that individual pretence is metarepresentational would be if it were assumed that children must have some self-awareness of their pretending, in other words if they themselves are the agent in the M-representation. Yet it seems highly unlikely that a child necessarily needs to represent the fact that I PRETEND (of) the banana (that) it is a telephone. Clearly they may do this at times, for example when asked what they are doing by an adult, but there is still no reason to suppose that in general they need do anything other than employ a suppositional secondary representation (see below).
It may be said on Leslie's behalf that even young children know the difference between pretending and real acting - they do not, for example, really eat the mud pies that they are pretending to have made - and if they know that they are pretending then they must, surely, represent themselves as engaging in pretence, which is a metarepresentation. But, in fact, the sense in which a young child knows that it is pretending is just that it is able to keep this attitude distinct from others, and act appropriately. The child's cognition may be so structured that it does not perform the actions that would satisfy a pretend desire if it also has a real desire that would be frustrated by that act, and knows this. But this no more shows that the child is metarepresenting its own pretence, than the fact that it knows that mud pies don't really taste very nice shows that it is metarepresenting its own desire - and neither claim is warranted, surely.
A key point in Leslies argument is that a child cannot simultaneously hold two primary representations that contradict each other semantically. However, in defining representation we noted that the attitude associated with a mental act of representation may vary while the content of the representation remains the same. Thinking in these terms we can see Leslie as defining primary representations as having a know or believe attitude to a certain state of affairs. We can then similarly suggest the existence of secondary representations, having a non-literal suppose, what if, or counterfactual attitude. Under this analysis there is no logical reason why non-literal, non-primary representations cannot be first-order (i.e. representational, as opposed to metarepresentational). This distinction is echoed by Currie (in press) who notes two potential types of misrepresentation: weak misrepresentation caused by a conflict between the content of a representation and the true state of the world, and strong misrepresentation which occurs when a representations content is supposed to reflect the real state of the world but conflicts with it. Weak misrepresentation does not result in representational abuse because the holder holds no factive attitude to the non-literal representations content. Such a representation can misrepresent the world, and yet not undermine the childs semantic knowledge of the world precisely because the child is not committed to its literality. This line of argument is central to Perners account.
Perner argues that young children proceed through three levels of semantic awareness. At the initial level of semantic awareness infants have a mental model of the world. This model is said to be determined veridically by perception, and consists of primary representations. It represents the world as it is, and makes up a non-manipulable knowledge base. Around the beginning of the second year children develop the ability to use mental models. They can copy elements from the knowledge base to create new models representing a variety of not necessarily literal situations. These models can be used to represent the world as it could be, to represent hypothetical situations. Because they do not share a direct causal relationship with the world, and are therefore not primary representations, they are secondary representations. The child can compare secondary representations with one another and with their primary knowledge base. What they cannot do, however, is to create models of models. This metarepresentational ability reflects acquisition of the third level of semantic awareness and occurs at around the age of four. This ability is necessary if a child is to compare a model of anothers mental model with their own knowledge of the world (as is required in a false belief task, for example).
One way in which Perner goes wrong in his account is that he, like Leslie, treats primary representations as entirely reality-oriented, forgetting, for the most part, the role of desire. Even quite simple organisms (and certainly, we would say, children during their first year) have desires, which will contain representations of non-actual states of affairs. The point of a desire is to represent, and by representing help to make actual, some state of affairs that does not yet exist. If this is correct, then it must be wrong to claim, as Perner does, that children only become capable of entertaining multiple models during their second year. What may be true, is that children only become capable of constructing such models at will during that year. He is surely correct that supposing, pretending, and wondering whether are all cognitively more sophisticated than believing and desiring, and make their appearance in children rather later. (As Currie points out, in press, the function of supposing is to enable us to test the consequences of alternative courses of action in imagination, before having to perform one of them in reality. This clearly presupposes that one is already a believer/desirer - that is to say, an agent. But the converse presupposition surely does not hold - there can be agents, with beliefs and desires, who lack imagination.) But this is not the same as saying that the capacity to entertain more than one model only makes its appearance later. We therefore suggest that the term primary representation should be understood to cover those attitudes involving models that are cognitively given, or basic (particularly beliefs and desires), reserving the term secondary representation for those mental models that are freely constructed, and manipulable, by the subject (particularly supposings and wonderings whether). This correction need not, however, affect the adequacy of Perners proposed treatment of the phenomenon of pretence, which is as follows.
Perner suggests that in pretence children create a counter-factual model of the pretend situation. Within the scope of his theory this ability is available to children operating at the second level of semantic awareness. In other words, this can be done using hypothetical, counter-factual secondary representations and does not require metarepresentations. It might be argued that these hypothetical models, originating from the knowledge base, are representations of the primary representations in the base, and are therefore metarepresentational after all. Perner counters this possible objection by pointing out that, though drawn from and placed in comparison to the knowledge base, the counter-factual pretend models are still models of the external world (as it could be) and not of the base. This is analogous to our earlier example of a photograph used as the basis of a portrait. Or, to return to the mental attitude/content distinction, Perner is arguing that all that is needed in pretence is the ability to hold a suppose attitude rather than a belief attitude to a certain content.
There are clear parallels between Perners theory and Leslies. Because Perners counter-factual mental models are hypothetical they are detached from reality and are therefore decoupled. Because they are separate from the knowledge base they are quarantined from it. Perner therefore circumvents the problem of representational abuse in much the same way as Leslie; he agrees that a child cannot concurrently hold two semantically conflicting (reality oriented) primary representations. Where he differs from Leslie is in his use of secondary representations as opposed to metarepresentations.
Simulation theorists would agree with Perner concerning the non-metarepresentational nature of pretence; but their line of argument is different to that adopted by Perner, and requires further analysis of the relationship between pretend play and theory of mind.
As noted, a theory of mind can be thought of as the ability to hold beliefs about someone elses beliefs. I have a theory of mind when I am able to understand that other people are minded in their own right; that they have their own intentional states, beliefs and desires, which may or may not be different from my own. Representing anothers belief (whether it be false or true) is undoubtedly metarepresentational. For example, if I believe that my friend Pete believes that his box of cornflakes is in the cupboard, then I represent Pete as having a representation of the cornflakes as being in the cupboard. I therefore have a metarepresentation (properly defined).
Simulators do not argue that theory of mind is non-metarepresentational - it clearly is metarepresentational - but they do contend that even older children (and adults) do not routinely employ a theory of mind when predicting anothers behaviour (Goldman; 1989, 1992; Gordon; 1986, 1992; Johnson, 1988). A theory of mind, or folk-psychology of behaviour, is one way in which I can make explanatory predictions about others. I can predict that my friend Pete will search for my cornflakes in the wrong cupboard because I believe that (he believes that (the cornflakes are there)). An alternative strategy would be for me to put myself in Petes shoes as he comes down hungry for breakfast. What would I do if I were Pete? I would look in the cupboard where the cornflakes are normally kept. In essence this is the simulation view of predicting behaviour. Note that, in contrast to the theory of mind approach, the simulation view does not require me to theorise about Petes mental state; I need only consider what I would do in his place. (Unanswered here is whether in his place implies only generalised knowledge about what people do when they look for cornflakes, or particular knowledge about Petes knowledge or expectations; we will return to this point later). Simulation is therefore first-person centred, and eliminates the need for third-person rationalisation.
If, as suggested by simulation theory, I do not need to use metarepresentations to predict what someone will do on the basis of a false belief, then I would certainly not need to use them in the potentially related, but certainly more simple task of following through my own counter-factual reasoning in pretence. Harris (1991) has used a simulation approach to argue directly for the non-metarepresentational nature of pretence. He highlights the early acquisition of the ability to imaginatively entertain mental states in pretence (by two or three) and notes that children are able to reason with pretend premises at this stage (Dias & Harris, 1988). Simulation, involving imagining a hypothetical situation and then reasoning from this pretend premise, should therefore be available to children at an early age.
Where Harris goes beyond previous accounts is in his appreciation that simulation will be more difficult the more the child has to take into account the idiosyncratic status of the other person (the other person might want something the child would not want, for example)[3]. Harris terms this setting aside default settings and identifies two such types of setting; these are the childs intentional stance towards reality and their specification of reality itself. By this analysis understanding another persons desire calls for setting aside intentional default settings (what one wants), while engaging in a pretence requires setting aside reality defaults (what really is). Predicting behaviour requires both sets to be ignored (one must appreciate what another wants, and what they believe to be the case) and is consequently a developmentally harder task.
What Harris is saying about pretence, therefore, is that all a child need do in order to produce pretend play themselves is to imagine what the world could be like, and to reason from there. This does not require metarepresentations but the ability to simulate counter-factual, hypothetical reality by setting aside what one knows about the world.
We have considered three theoretical approaches to the question of whether pretend play is metarepresentational in nature. How might one go about separating these accounts, and determining their relative merits? There are theoretical reasons to be invoked in favour of one or other standpoint, a number of which have already been highlighted, but the real test of these accounts must come from empirical research. We review evidence from normal pretend play development, normal theory of mind development, and pretend play in autism. Focusing on these domains will also re-emphasise the importance of resolving the question of the metarepresentational status of pretence.
In the discussion of Leslies theory above, it was argued that it is important to consider individual pretence separately from shared pretence. There seems no reason to suppose that a child who pretends alone represents themselves as representing one object as representing another. In contrast, in joint pretence a child may well need to represent the other partner as pretending if they are to understand what is occurring. Crucial to Leslies account is his claim that shared pretence co-occurs with the onset of individual pretend play. Leslie therefore argues that 2-year-old children must be able to compute the M-representation. However Leslies review of the literature on the development of pretence is somewhat cursory, and a number of more recent studies are relevant to this debate.
Many experiments have shown that children of around three years of age are perfectly able to understand or make sense of pretend play acts carried out by another person. For example, Flavell, Flavell and Green (1987) showed three-year-olds a candle that looked like an apple. An experimenter then mimed eating actions with the candle, and children were asked what the experimenter was pretending to eat. 95 % of children answered that it was an apple, rather than a candle, and 75% stated that the candle was a pretend apple rather than a real apple. Similar findings were reported by Woolley and Wellman (1990), who also found that children first used real or really to emphasise a contrast between real and non-real situations from around three years of age. Stronger evidence for an ability to understand pretence, as opposed to the ability to separate the literal from the non-real, comes from a series of studies reported by Harris and Kavanaugh (1993). In these experiments (their experiments 6 and 7) three-year-olds correctly identified a pretend substance that had been poured from a container onto a toy animal, or onto an animals pretend food, in over 70% of cases. When asked about the results of the action three-year-olds answered in terms of a pretend outcome (e.g. the animal was wet), rather than in terms of the literal state of affairs (the animal being dry) in over 85% of cases.
These results indicate that certainly by three years of age children are capable of distinguishing between reality and non-literal situations, and can make pretend sense of imaginary scenarios which are acted out before them. However just because two people are involved in the pretend situation, one acting out the pretence, the other understanding it, there is no need to assume that the understander must necessarily represent the actor as pretending. In other words it is possible, and indeed likely, that young children understand pretend acts of this kind by treating it as individual pretence, albeit pretence not generated by themselves. In terms of Perners notion of suppositional thinking (using secondary representations), a child may pick up cues in the actors intonation and facial expression that indicate that as if behaviour is taking place, and prompt them into suppositional rather than literally-receptive mode. Similarly, Harris (1991; Harris & Kavanaugh, 1993) argues that faced with a potentially puzzling non-literal behaviour young children may simply search for a personal pretend scheme that could explain this. They could then understand that the banana was being used as if it were a telephone while by-passing the metarepresentational attribution of the actors pretence. In doing so the child might rely heavily on scripts, or quite generalised knowledge about behaviour sequences and scenarios, in order to join in a mutual pretend play episode.
By contrast, evidence of metarepresentational understanding in shared pretend play would be provided by children taking account of the others pretend viewpoint, communicating about the pretend nature of the scenario, and responding appropriately to shifts in the pretend game initiated by the partner. There is some debate as to when these abilities develop. Lillard (1993) has argued that children rely heavily on scripts, and do not engage in this form of cooperative pretence until 5 years of age; but she does not cite the recent empirical work on this topic.
Through a series of studies Howes and colleagues (Howes & Matheson, 1992; Howes, Unger & Seidner, 1989; Howes, Unger & Matheson, 1992) have developed a sequential scale of social pretend play development. The two final stages of this scale which are of interest here are cooperative social pretend play - children act out complementary roles within social pretence, and complex social pretend play - complementary roles are acted out with metacommunication.
Howes and Matheson (1992) report evidence of these levels of social pretend play in a three-year longitudinal study of 72 children, initially aged between 13 and 24 months (the sample size fell to 48 children over this time). They found that cooperative social pretend play first emerged between 19 to 23 months, and was present in half the sample by 30 to 35 months. Complex social pretend play was first seen in children at 30 to 35 months, but was not present in half the sample until 42 to 47 months. In an earlier cross-sectional study, which looked at stages up to and including cooperative social pretend play, but excluded complex social pretend play, Howes et al. (1989) observed cooperative social pretend play in children aged between 29 and 38 months.
Howes et al. (1992) summarise and elaborate these findings with reference to mother-child and peer-child play. They claim that before 30 months children will offer pretend scripts to the mother and to peers, but integration of scripts does not occur. Such behaviour would not seem to necessitate metarepresentational understanding. At 31 to 36 months social pretence takes on the joint enactment of complementary roles (Cooperative Social Pretend Play), for example Children discriminate between speech used for enactment and speech about enactment. Children assign roles and negotiate pretend themes and plans" (our italics). Recalling Leslies analysis of the M-representation, this sounds more likely to be metarepresentational. By 37 to 48 months children are established in their ability to communicate and instruct the cooperative integration of roles (Complex Social Pretend Play), children adopt relational roles, are willing to accept identity transformations and generate or accept instruction for appropriate role performance. Children negotiate scripts and dominant roles and use metacommunication to establish the play script and clarify role enactment (our italics). There seems little doubt that metarepresentation is involved here.
In summary, fairly unambiguous evidence for metarepresentation in pretence, i.e. verbal metacommunication of roles and pretence structure, is normal at 42-47 months; some children show signs of it from possibly 30 months and certainly 36 months. The importance of this data is that it undermines one of Leslies key arguments, namely that the co-occurrence of understanding of pretend play in others and individual pretend play indicates that children of two years of age are able to process metarepresentations. While it remains possible that they are able to do so at this age, there is no convincing evidence of this ability in pretend play until the much later age of 3 to 4 years. Lillard (1993) also rejects the view that pretend play is a zone of proximal development in the childs acquisition of metarepresentational abilities, and instead proposes that it is a fools gold which appears to be metarepresentational, but isnt.
This re-evaluation of the cognitive nature of pretend play also raises intriguing questions about the nature of pretend play development. Do children progress from non-metarepresentational individual pretend play to metarepresentational complex social pretend play, and if so, how does this process occur? These questions will be returned to in the final section of the paper.
A second reason for questioning whether early pretend play is metarepresentational is the fact that pretend play emerges at around two years of age, while a workable theory of mind does not appear to develop until around four years. This statement does, however, require considerable qualification. As noted, children may show signs of pretend play well before two years, but it is not until around 20-24 months that pretend play proper, or symbolic play (as defined by Leslie, 1987) is seen. Similarly aspects of childrens understanding of others minds develop before four years. Mental verbs such as know and think appear in childrens speech before three years (Shatz, Wellman & Silber, 1983), and three-year-olds are able to infer anothers actions on the basis of that persons desires ( Wellman & Woolley, 1990). It further appears that an appreciation of how true beliefs may influence behaviour emerges during the third year (Wellman & Bartsch, 1988??).
However, three-year-olds perform poorly on the false belief test outlined earlier. Wimmer and Perner (1983) found that only X% of three-year-olds succeeded on this task, and indeed the performance of four-year-olds was relatively poor on their procedure. Subsequent studies, which do not differ in their basic methodology, but which are perhaps linguistically simpler and contextually more appropriate, indicate that false-belief tests may be passed as early as 36 months (Freeman, Lewis & Doherty, 1991; Lewis & Osborne, 1990; Sullivan & Winner, 1991). Three-year-olds also perform very poorly, in contrast to four-year-olds, on other tasks which require the ability to infer how anothers false belief will influence their behaviour. For example three-year-olds who know that a container (e.g. a Smartie tube) hides surprising contents (e.g. pencils) will claim that an ignorant peer is also aware of the true contents of the container (Perner, Leekam & Wimmer, 1987). Further, while four-year-olds readily deceive a robber about the fact that a box contains a reward, which they wish to keep for themselves, three-year-olds tell the robber the true contents of the box and lose the reward (Sodian, 1993), apparently unaware of the value of implanting an incorrect belief in anothers mind.
The results of these and many other studies indicate that it is somewhere between 36 and 48 months that children come to appreciate how false beliefs influence the actions of others. Whether one sees this as indicative of a qualitative shift in childrens understanding of the mind, or rather reflecting a gradual and quantitative development of mindreading abilities, the important point is that only at this stage can one be certain that metarepresentational abilities are being employed in this domain (Dennett, 1978). This returns us to the question of why it takes children so long to pass these tasks if they are able to metarepresent as early as two years of age as Leslie suggests.
Leslie (Leslie & Thaiss, 1992; Leslie & Roth, 1993) argues that this lag is due to the modular nature of theory of mind and pretend play mechanisms, claiming that while understanding pretend play and false belief both require the ability to manipulate M-representations, predicting actions on the basis of false belief also requires a further mechanism, the Selection Processor, which develops between the ages of two and four. In other words children are able to metarepresent at two, but non-mentalistic aspects of the false-belief task render it more complex than pretence, and it requires the maturation of the Selection Processor to overcome these problems. Leslies justification for proposing this further mechanism is not entirely post-hoc; he argues that differences in the performances of children with autism and three-year-olds on certain theory of mind tasks is due to the fact that the Selection Processor is intact in autism. Nevertheless, there is no direct evidence for the Selection Processor in young children.
Of course the alternative approach is to suggest that pretend play is not metarepresentational. The advantage of this proposal is that the emergence of success on false-belief tasks fits well with the above analysis of social pretence. In fact the claim that metarepresentational pretence does not definitely occur until 3 to 4 years fits particularly well with those studies that indicate that false-belief tests can first be passed at 36 months. While the co-emergence of two abilities is certainly not conclusive evidence of a direct underlying link between them, this approach is more parsimonious than Leslies as there is no need to have recourse to a hypothetical Selection Processor to solve a dilemma which no longer exists.
Autism is a disorder which results in a number of characteristic impairments. These include severe difficulties in dealing with, and withdrawal from, social behaviour, and abnormal language development. A preference for stereotyped and routine behaviours is also seen. Despite the relatively disparate pattern of these impairments, a number of theories have been advanced as to the fundamental cause of the disorder. At present the most influential of these argues that autism results from a failure to manipulate metarepresentations (Leslie and Frith, 1990; Leslie & Roth, 1993).
Support for this hypothesis comes from conclusive evidence of children with autisms severe problems on theory of mind tasks. They fail traditional false-belief tests (Baron-Cohen, Leslie & Frith, 1985), other tests of false-belief attribution (Perner, Frith, Leslie & Leekam, 1989), and deception tasks (Sodian & Frith, 1992), even when they have a verbal mental age of above four years, and should otherwise be expected to pass these tasks. This apparent lack of a theory of mind, which in fact appears to reflect severely delayed development of mindreading abilities, is clearly able to explain the characteristic social deficits seen in autism. It is also claimed that a failure to appreciate the mental states of others underlies the linguistic and communicative impairments observed (Frith, 1989); this argument is based on the assumption that communication is constrained by the need to be relevant to the listener, and therefore needs to take into account their beliefs, ignorance and attitudes.
Further support for this account has, in the past, been drawn from a number of studies which have shown that children with autism show abnormally low levels of spontaneous pretend play (Baron-Cohen, 1987; Riguet, Taylor, Benaroya & Klein, 1981; Sigman & Ungerer, 1984), again based on the assumption that pretend play is metarepresentational. However, in the context of this discussion, and to avoid a circular argument[4], the absence of pretence seen amongst children with autism is support for the metarepresentational nature of pretence only in the light of their clear difficulties with theory of mind tasks. In other words it is the co-occurence of problems in pretence and with theory of mind tasks that suggest a common link. This pattern of impairments is clearly consistent with Leslies accounts, and may be taken to provide corroborating, if not direct, evidence for the metarepresentational nature of pretence.
However the strength of this evidence is clearly undermined if it is shown that a metarepresentational deficit does not lie at the heart of the disorder. There can be no doubt that children with autism are severely impaired in their ability to pass theory of mind tests, but Harris (1991), following the simulationists argument, claims that this is not evidence for a metarepresentational deficit, but rather reflects an inability to simulate or to set aside both reality and intentional default settings. And Currie (forthcoming) argues that the common explanation for autistic children's failures to engage in pretend play and to mindread lies in a failure of imagination, or in impairments in their ability to frame and manipulate counterfactual representations. Similarly Russell (Hughes & Russell, 1993; Russell, Mauthner, Sharpe & Tidswell, 1991) argues that failure on theory of mind tasks is not primarily due metarepresentational deficits. In common with Harris, he claims that it is the need to disengage from the reality of a theory of mind task that renders the tests particularly difficult for autism. This view is supported by evidence of executive dysfunction, or problems in actively monitoring, directing and inhibiting behaviour in autism (see Bishop, 1993 for a review). Where Russell differs from Harris is that he sees theory of mind tasks as requiring metarepresentational competence, but claims that this competence rests on more fundamental executive abilities. Indeed there is some evidence to suggest that executive problems are more widespread and fundamental to autism than theory of mind deficits (Ozonoff et al., 1991), though the precise relationship between the two abilities remains unclear.
The implications of children with autisms joint difficulties in pretence and on theory of mind tasks would be further undermined if it were shown that they can engage in pretence. Indeed, though children with autism normally show abnormally low levels of pretend play, the fact that they show any pretence at all could be seen as problematic for Leslies account (and also, to an extent, for Harris's and Currie's). More importantly there is growing evidence to suggest that a pretend play deficit in autism is confined to spontaneous or free play situations, and that children with autism can pretend when required to do so by an external agent (Lewis & Boucher, 1988; see Jarrold, Boucher & Smith, 1993). It is therefore possible that a failure to pretend in autism reflects a performance (failure to produce) rather than a competence (inability to produce) deficit. Though this proposal remains to be confirmed, studies of children with autisms comprehension of pretence indicate that this is unimpaired, relative to language matched controls (Jarrold, Smith, Boucher & Harris, in press; Kavanaugh & Harris, in press).
These results suggest that children with autism can produce and understand individual pretend play under certain circumstances, though it should be noted that the quality of pretence seen may be poor and may lack the creativity and flexibility seen in normal children. Nevertheless the very fact that children with autism can engage in the mechanics of pretence, given their clear problems on theory of mind tasks, poses a problem for Leslie. As yet no attempts have been made to investigate such childrens ability to produce or comprehend complex social pretend play. This would certainly be a worthwhile area of study. If it were found that children with autism were unable to understand complex social pretend play, in contrast to their apparent ability to understand individual pretence, this would lend support to the view that individual pretence is non-metarepresentational, and complex social pretence is metarepresentational in nature. In this case failure on theory of mind tasks could still be attributed to metarepresentational deficits. Alternatively, if children with autism were found to be able to participate in complex social pretence, this would favour Harris proposal that metarepresentational competence does not play a part in pretence, or by implication in theory of mind tasks; children with autisms problems in these domains would reflect a failure to set aside default settings, rather than an inability to metarepresent. But it would remain a problem for both approaches to explain why autistic children do not engage in spontaneous pretend play, given that they have the capacity to do so.
Why should we suppose that pretence necessitates the use of metarepresentations? As noted above, Leslies account raises two possible theoretical reasons for doing so. The first is the need to circumvent representational abuse. Metarepresentational pretence does manage this quite easily, by decoupling the metarepresentation from the primary representation. But Perners account reflects the mental attitude/content distinction and makes use of secondary representations which are effectively decoupled from the knowledge base. These counter-factual representations are not metarepresentational, yet representational abuse is still avoided.
The second possible reason for adopting Leslies approach, concentrating now on his notion of the M-representation, is that it could be argued that pretence does indeed need to capture the informational relationship between the primary and secondary representations. The crucial question is therefore whether I need to represent the secondary representation ("the banana as a telephone") as being a representation of the primary representation ("the banana"), or whether it is enough to hold that secondary representation alongside the primary representation without having the relationship between the two made explicit.
It would certainly seem that having a decoupled secondary representation would be sufficient for individual pretence. But it could be argued that in order to understand pretence in others the relationship between the two representations must be appreciated. The key question here is what we mean by understanding. Obviously if I represent mother as pretending that something is something else, then I employ a metarepresentation. However, I may be able to make sense of the pretence without having to go this far, by realising that something non-literal is occurring, whether from maternal cues or because of the inherent strangeness of the behaviour, and by interpreting the pretence myself and joining in relying on scripted or stereotyped knowledge.
Crucial here is the issue of the extent to which special knowledge of another person - the play partner in the case of mutual pretend - is needed. In so far as the child relies on generalised knowledge - that bananas are to be eaten, that food is kept in food cupboards, that people like eating bananas, etc. - then a great deal of prediction of behaviour, and a great deal of pretence (whether alone with dolls, or with play partners) can be explained without invoking metarepresentation. It is enough to posit a suppositional and/or simulation process. This may involve the child projecting themselves into another's shoes, for example, using generalised knowledge, or default settings. The analysis of the development of social pretend play indicates that it is not until around three and a half years that children show evidence of going beyond these first-person based strategies to a third-person centered metarepresentational understanding.
These arguments nullify Leslies view that all pretence must be metarepresentational. Further, there are specific reasons for rejecting this view. Firstly, complex social pretence is not seen until around three and a half, an age which ties in much more closely with the onset of metarepresentational competence in other domains. Secondly, an advantage of assuming that individual pretend play is not metarepresentational is that it allows one to explain why children with autism can produce (individual) pretend play yet appear to lack a theory of mind.
Both Perners and Harris theories are consistent with the view that individual pretence is not metarepresentational. However Perner would agree that complex social pretend play does rely on metarepresentational understanding, and would therefore predict that children with autism would be unable to engage in this level of pretence. In contrast, Harris would claim that complex social pretend play simply requires setting aside more default settings, in the same way that understanding false belief demands more than does engaging in individual pretence. While he would no doubt argue that children with autism would have severe difficulties in producing this type of play, his theory might be interpreted to suggest that they would be able to comprehend it in a structured situation. These two accounts do then appear to be theoretically separable.
We conclude by returning to the question of how metarepresentational complex social pretend play[5] might develop from non-metarepresentational individual pretence. Rather than set up a dichotomy, a step-function from not being able, to being able, to metarepresent, we prefer a gradual, transitional model. Interestingly, a similar controversy was debated 20 years ago in the area of children's empathy. Borke (1971) claimed to find evidence for empathy in children as young as 3 years. Chandler and Greenspan (1972) objected to her conclusions. They asserted that mechanisms such as projection, and use of stereotyped knowledge of scenarios (nowadays, scripts) could explain the 'success' of the children. Borke (1972) retorted that projection and stereotyped knowledge were precisely the mechanisms supporting early empathic abilities, from which a truer and fuller version would arise developmentally. Nascent abilities are scaffolded in familiar situations, and can later be used more independently.
Such mechanisms may support early developing theory of mind and metarepresentational abilities beyond just empathy. This form of gradual transition is becoming evident in the literature on false belief tasks, and on ability to deceive others (e.g. Avis & Harris, 1991; Sodian, Taylor, Harris & Perner, 1991). We believe it is also evident in our knowledge of pretend play. In all these cases, children at 24-30 months have not been found to have metarepresentational abilities. Children of 30 months through to 42 months seem to be in transition, able to show metarepresentational abilities only when in easier situations, or when helped or prompted to do so (e.g. Sodian et al., 1991). From 42 months onward, evidence for metarepresentational ability in many children is reasonably clear. This now seems to be true across three main domains of investigation - false belief, deceit, and (mutual) pretence.
Clearly, further examination of the exact nature of shared pretence is still needed. Howes analysis of mutual pretence did not have the question of metarepresentational ability specifically in mind; this could be targeted in future work. Furthermore, even if a range of metarepresentational abilities is gradually emerging in the 30 to 42 month period, the details of this need much more elaboration, in pretend play as in other domains. Also, whether pretend play has any kind of leading edge in this developmental transition is still in question.
What we do conclude is that Leslie's proposal concerning the metarepresentational nature of pretence is unjustified. Non-metarepresentational accounts explain individual pretence and early mutual pretence adequately and parsimoniously; unless new arguments can be advanced in support of Leslies metarepresentational view, there appears to be no reason to continue to adhere to it. Rather, we suggest a refocus of attention on the broad and diverse nature of pretend play, both individual and mutual, and the need to provide a framework for a more cognitive analysis of the developmental trends in this fascinating form of behaviour.
Avis, J. & Harris, P. L. (1991). Belief-desire reasoning among Baka children: evidence for a universal conception of mind. Child Development, 62, 460-467.
Baron-Cohen, S. (1987). Autism and symbolic play. British Journal of Developmental Psychology, 5, 139-148.
Baron-Cohen, S. (1991). The development of a theory of mind in autism: Deviance and delay? The Psychiatric Clinics of North America, 14, 33-51.
Baron-Cohen, S., Leslie, A. M. & Frith, U. (1985). Does the autistic child have a theory of mind? Cognition, 21, 37-46.
Borke, H. (1971). Interpersonal perception of young children: egocentrism or empathy? Developmental Psychology, 5, 263-269.
Borke, H. (1972). Chandler and Greenspan's "Ersatz egocentrism". Developmental Psychology, 7, 107-109.
Chandler, M. & Greenspan, (1972). Ersatz egocentrism: a reply to H. Borke. Developmental Psychology, 7, 104-106.
Currie, G. (in press). Imagination and simulation: Aesthetics meets cognitive science. Mind and Language.
Currie, G. (forthcoming). Simulation-Theory, Theory-Theory and the Evidence from Autism.
Dennett (1978)
Dias, M. & Harris, P. L. (1988). The effect of make-believe play on deductive reasoning. British Journal of Developmental Psychology, 6, 207-221.
Flavell, J. H., Flavell, E. R. & Green, F. L. (1987). Young children's knowledge about the apparent-real and pretend-real distinctions. Developmental Psychology, 23, 816-822.
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.
Frege, G. (1892/1960). On sense and reference. In P. Geach and M. Black (Eds.), Philosophical writings of Gottlob Frege. Oxford: Blackwell.
Goodman, N. (1976). Languages of art. Indianapolis: Hackett Publishing Co.
Goldman, A. I. (1989). Interpretation psychologized. Mind and Language, 4, 161-185.
Goldman, A. I. (1992). In defense of the simulation theory. Mind and Language, 7, 104-119.
Gordon, R. M. (1986). Folk psychology as simulation. Mind and Language, 1, 158-171.
Gordon, R. M. (1992). The simulation theory: Objections and misconceptions. Mind and Language, 7, 11-34.
Harris, P. L. (1991). The work of the imagination. In A. Whiten (Ed.), Natural theories of mind. Oxford: Blackwell.
Harris, P. L. & Kavanaugh, R. D. (1993). Young children's understanding of pretense. Monographs for the Society of Reaserach in Child Development, 58(1).
Howes, C. & Matheson, C.C. (1992). Sequences in the development of competent play with peers: social and pretend play. Developmental Psychology, 28, 961-974.
Howes, C., Unger, O. & Matheson, C. C. (1992). The collaborative construction of pretend. Albany: SUNY Press.
Howes, C., Unger, O. & Seidner, L. B. (1989). Social pretend play in toddlers: parallels with social play and with solitary pretend. Child Development, 60, 77-84.
Hughes, C. & Russell, J. (1993). Autistic children's difficulty with mental disengagement from an object: Its implications for theories of autism. Developmental Psychology, 16, 543-549.
Jarrold, C. (submitted). An investigation into the pretend play shown by children with autism. Unpublished Ph.D. thesis.
Jarrold, C., Boucher, J. & Smith, P. K. (1993). Symbolic play in autism: A review. Journal of Autism and Developmental Disorders, 23, 281-307.
Jarrold, C., Smith, P. K., Boucher, J. & Harris, P. L. (in press) Children with autisms comprehension of pretence. Journal of Autism and Developmental Disorders`
Johnson, C. N. (1988). Theory of mind and the structure of conscious experience. In J. Astington, P. L. Harris & D. P. Olson (Ed.), Developing theories of mind Cambridge: CUP.
Leslie, A. M. (1987). Pretence and representation: The origins of theory of mind. Psychological Review, 94, 412-426.
Leslie, A. M. (1988). Some implications of pretence for mechanisms underlying the childs theory of mind. In J. W. Astington, P. L. Harris & D. R. Olson (Ed.), Developing theories of mind Cambridge University Press.
Leslie, A. M. & Frith, U. (1990). Prospects for a cognitive neuropsychology of autism: Hobsons choice. Psychological Review, 97, 122-131.
Leslie, A. M. & Roth, D. (1993). What autism teaches us about metarepresentation. In S. Baron-Cohen, H. Tager-Flusberg & D. Cohen (Ed.), Understanding other minds: Perspectives from autism Oxford: Oxford University Press.
Leslie, A. & Thaiss, L. (1992) Domain specificity in conceptual development: neurological evidence from autism. Cognition, 43, 225-251.
Lewis , C. & Osborne, A. (1990). Three-year-olds' problems with false belief: Conceptual deficit or linguistic artifact? Child Development, 61, 1514-1519.
Lewis, V. & Boucher, J. (1988). Spontaneous, instructed and elicited play in relatively able autistic children. British Journal of Developmental Psychology, 6, 325-339.
Lillard, A. (1993). Pretend play skills and the child's theory of mind. Child Development.
McCune-Nicolich, L. (1981). Toward symbolic functioning: Structure of early pretend games and potential parallels with language. Child Development, 52, 785-797.
Perner, J. (1988). Developing semantics for theories of mind: From propositional attitudes to mental representation. In J. Astington, P. L. Harris & D. P. Olson (Ed.), Developing theories of mind Cambridge: Cambridge University Press.
Perner, J. (1991). Understanding the representational mind. Cambridge, MA: MIT Press/Bradford Books.
Perner, J., Frith, U., Leslie, A. M. & Leekam, S. R. (1989). Exploration of the autistic childs theory of mind: Knowledge, belief and communication. Child Development, 60, 689-700.
Piaget, J. (1962). Play, dreams and imitation in childhood. London: Routledge and Kegan Paul.
Pylyshyn, Z. W. (1978). When is attribution of beliefs justified? Behavioral and Brain Sciences, 1, 592-593.
Riguet, C. B., Taylor, N. D., Benaroya, S. & Klein, L. S. (1981). Symbolic play in autistic, Downs and normal children of equivalent mental age. Journal of Autism and Developmental Disorders, 11, 439-448.
Russell, J., Mauthner, N., Sharpe, S. & Tidswell, T. (1991). The 'windows task' as a measure of strategic deception in prescholers and autistic subjects. British Journal of Developmental Psychology, 9, 331-349.
Sigman, M. & Ungerer, J. A. (1984). Cognitive and language skills in autistic, mentally retarded, and normal children. Developmental Psychology, 20, 293-302.
Sodian, B., Taylor, C., Harris, P. L. & Perner, J. (1991). Early deception and the childs theory of mind: false trails and genuine markers. Child Development, 62, 468-483.
Sullivan, K. & Winner, E. (1991). When 3-year-olds understand ignorance, false belief and representational change. British Journal of Developmental Psychology, 9, 159-171.
Wimmer, H. & Perner, J. (1983). Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children's understanding of deception. Cognition, 13, 103-128.
Woolley, J. D. & Wellman, H. M. (1990). Young children's understanding of realities, nonrealities and appearances. Child Development, 6, 946-961.
Figure 1. The M-representation (Leslie and Roth, 1993)
[5]Leaving aside the simulationists view that even this level of pretence might not be metarepresentational.