Social play: Evolution, pretense, and the cognitive turn

To return to our immediate subject: the lower animals, like man, manifestly feel pleasure and pain, happiness and misery. Happiness is never better exhibited than by young animals, such as puppies, kittens, lambs, &c., when playing together, like our own children. Even insects play together, as has been described by that excellent observer, P. Huber, who saw ants chasing and pretending to bite each other, like so many puppies. (Charles Darwin 1871/1936, p.448)

Pierre Huber (1810, p. 148), in his book about the behavior of ants, claims that if one were not accustomed to treating insects as machines one would have trouble explaining the social behavior of ants and bees without attributing emotions to them. Although we shall skirt the issue of emotion, many observers would agree that animals play because it is fun for them to do so. But even if the issue of emotions is set aside, readers conditioned by the scruples of modern psychology are likely to be skeptical of Darwin's ready acceptance that Huber observed ants playing. Play, as the quotation above indicate, seems to involve pretense, and pretense is commonly thought to require more sophisticated intentions than are usually attributed to ants. How could Huber have seen or inferred pretense from the behavior of the ants? And how could he be sure that the observed behavior was not, in fact, directed toward some very specific and immediate function? These questions raise the difficult issue of what play is, or, as biologists are wont to put it, how to define "play". This issue has proven a great challenge to those who study this interesting behavioral phenotype.

We and others believe that social play is a tractable, evolved behavioral phenotype that lends itself to detailed empirical study. And, the flexibility and versatility of social play makes it a good candidate for comparative and evolutionary cognitive studies including those that center on ways in which animals might negotiate agreements to engage in a cooperative social interaction. As such, cognitive ethological approaches are useful for gaining an understanding of the social play behavior of diverse animals for a number reasons (Jamieson and Bekoff 1993; Bekoff 1995a,b; Allen and Bekoff 1997; Bekoff 1997). First, empirical research on social play has benefited and will further benefit from a cognitive approach because play involves issues of communication, intention, role playing and cooperation. Second, many believe that detailed analysis of social play may provide more promising evidence of animal minds than research in many other areas for it may yield clues about the ability of animals to understand each others' intentions. Third, play is a phenomenon that occurs in a wide range of mammalian species and a number of avian species. Thus it affords the opportunity for a comparative investigation of cognitive abilities extending beyond the all-too-common narrow focus on primates that dominates discussions of nonhuman cognition. Thus, the topic of social play exemplifies many of the theoretical issues faced by cognitive ethologists and may help those who are interested in broadening the evolutionary study of animal cognition.

The study of social play provides an opportunity to pursue the suggestion by Niko Tinbergen (1972) and others (Schaller and Lowther 1969) that we may learn as much or more about human social behavior by studying social carnivores as by studying nonhuman primates. Byrne (1995), who otherwise takes a strongly primatocentric view of animal cognition, observed that we might learn more about the phylogenetic distribution of what he calls intelligence by doing comparative research. Furthermore, Povinelli and Cant (1995) suggest that the performance by arboreal ancestors of the great ape/human clade of "unusual locomotor solutions . . . drove the evolution of self-conception" (p. 400). Many nonprimate mammals also perform complex, flexible, and unusual acrobatic motor patterns (locomotor-rotational movements) during social play, and it would be premature to rule out the possibility that the performance of these behaviors is also important to the evolution of self-conception in nonprimates. In some instances it is difficult to know whether aboreal clambering or the performance of various acrobatic movements during play may more be related to the evolution of (mere) body awareness (e.g. knowing one's place in space) and not a concept of self.

What is play?

As other papers in this volume show, the term "play" covers a wide range of behavior patterns. In this respect it is not different from terms such as "feeding" or "mating", both of which may encompass a variety of quite different behaviors both when comparing members of the same species and when comparing members of different species. However, unlike play, feeding and mating correspond to easily identified biological functions.

Play is not easily defined (Bekoff and Byers 1981; Fagen 1981; Martin and Caro 1985; Burghardt 1996). Attempts to define it functionally face the problem that it is not obvious that play serves any particular function either at the time at which it is performed or later in life. Indeed several authors have been tempted into defining play as functionless behavior. Alternatively it has sometimes been suggested that play serves some general functions such as improving the motor and cognitive skills of young animals, yielding possible payoffs, for example, in the hunting, foraging, or social abilities of these animals from the time of the play through their entire lifespans. Even if this is correct, the reproductive fitness consequences of play may typically be so far removed in the life time of the organisms involved that it would be very difficult to collect data to support the assertion that play increases fitness. Furthermore, there may be different possible evolved functions of play depending on the species being studied. It is difficult to design experiments to test hypotheses about functions of play that are both practicable and ethical. Thus play seems to be either functionless or it can be considered as serving different functions for individuals of different species, ages, and sex (Bekoff and Byers 1981; Fagen 1981; Byers and Walker 1995; Burghardt 1996; Watson and Croft 1996).

These considerations led Bekoff and Byers (1981, pp. 300-301; see also Martin and Caro 1985) to eschew a functional characterization of play by offering the following definition: "Play is all motor activity performed postnatally that appears [our emphasis] to be purposeless, in which motor patterns from other contexts may often be used in modified forms and altered temporal sequencing. If the activity is directed toward another living being it is called social play." This definition centers on the structure of play sequences -- what animals do when they play -- and not on possible functions of play. Nonetheless the definition is not without problems, for it would seem to apply, for example, to stereotypical behaviors such as the repetitive pacing or excessive self-grooming sometimes evinced by caged animals. It is difficult to see how to state a non-arbitrary restriction on the range of behaviors that may constitute play.

Because it is not easily defined, play, both social and nonsocial, has been a very difficult behavioral phenotype with which to deal rigorously. A few people would claim that only humans engage in play, but most agree that nonhumans play despite finding it difficult to offer an exceptionless definition. But this lack of a comprehensive definition need not be an impediment to conducting solid research. Our view is that the study of play ought to be approached like the study of any other (putative) natural kind of behavior (Allen and Bekoff 1994). To study play, one ought to start with examples of behaviors which superficially appear to form a single category -- those that would be initially agreed upon as play -- and look for similarities among these examples. If similarities are found, then we can ask whether they provide a basis for useful generalizations. We therefore propose to precede on the basis of an intuitive understanding of play, guided to some extent by Bekoff and Byers' attempt to define it, but without the view that this or any other currently available definition strictly includes or excludes any specific behaviors from the category of play.

 Can there be an evolutionary biology of play? The possible problem of intentionality

Alexander Rosenberg (1990) presents some challenges to an evolutionary approach to social play. One of his concerns hinges on his claim that play is an intentional activity. For reasons similar to those of Dennett (1969) and Stich (1983), and rejected by Allen and Bekoff (1994, 1997), Rosenberg believes that intentional explanations are not suitable for scientific explanations of behavior. Rosenberg, for instance, suggests that it might be inappropriate to attribute the concept of mouse-catching to a cat by asking "Does it have the concept of mouse, Mus musculus in Linnaean terms?" (p.184). Our view is that possession of the Linnaean concept of a mouse is not a reasonable requirement to be placed on the attribution of beliefs about mouse catching (see Allen and Bekoff 1994).

Rosenberg also argues that there can be no unified evolutionary account of play because actual cases of play have heterogeneous causes and effects, and different underlying mechanisms. He draws an analogy between play and clocks, pointing out that because there are so many different mechanisms that constitute clocks there is no "single general explanatory theory that really explains what clocks do, how and why they do it." (p. 180) The problem with this argument is that the kind of "single general explanatory theory" referred to is not (and should not be) the kind of thing evolutionary biology is necessarily concerned with. While it is the concern of some branches of biology (particularly molecular and cellular) to explain how certain organs do what they do, other branches of evolutionary theory are concerned with what they do and why they do it. So while it would be foolish to expect a singular molecular or cellular account of light-sensing capabilities across species, it is not foolish to expect unity in some aspects of the evolutionary explanations of the development of such organs (although, of course, there will be differences in the evolutionary histories across different species). If Rosenberg was right there could be no general evolutionary theory of predation or sexual selection by mate choice, for these phenomena too depend on a very heterogeneous set of mechanisms. Play, we submit, is in no worse shape than these well entrenched targets of biological explanation.

 Play, pretense, and intentionality

After all, from an evolutionary point of view, there ought to be a high premium on the veridicality of cognitive processes. The perceiving, thinking organism ought, as far as possible, to get things right. Yet pretense flies in the face of this fundamental principle. In pretense we deliberately distort reality. How odd then that this ability is not the sober culmination of intellectual development but instead makes its appearance playfully and precociously at the very beginning of childhood. (Leslie 1987, p. 412)

As we noted above, discussions of play commonly refer to the concept of pretense. Because pretense seems to be a fairly sophisticated cognitive ability it has led some authors to deny that nonhuman animals can be said to engage in play. Rosenberg (1990), for example, associates pretense with "third-order" intentionality (Grice 1957; Dennett 1983, 1987). According to Rosenberg, for animal a truly to be playing with b, it must be that "a does d [the playful act] with the intention of b's recognizing that a is doing d not seriously but playfully. So, a wants b to believe that a wants to do d not seriously but with other goals or aims." (Rosenberg 1990, p. 184) This is third-order because there are three levels of mental state attribution involved, i.e. a believes that b believes something about a 's desires. This requirement might be thought to rule out play not just in nonhuman animals, but also in human infants.

In contrast to this approach, the Bekoff and Byers characterization of play is neutral about the intentionality of play behavior. Ultimately it might be found that play is an intentional activity but it would be premature, in our view, to include this in the definition of play. The relevance of intentionality to play is a matter for empirical investigation, and any empirical investigation of the connection between play and intentionality will be shaped by the account of intentionality that is provided (for discussion see Allen and Bekoff 1997, chapter 6).

From Dennett's intentional stance, organisms are modeled as representing various aspects of their environments and their actions are guided by those representations. For some organisms, these representations may themselves contain information about how other organisms represent their environments. Such a representation of a representation is a case of second-order intentionality in Dennett's scheme. Dennett treats higher-order intentionality as cognitively more sophisticated (and therefore more recently evolved) than first-order intentionality (which in turn is more sophisticated than zero-order or non-intentionality). Thus, to place cognitive capacities into an evolutionary framework, he thinks it is important to identify the distribution of higher-order intentionality among animals.

Millikan (1984) provides a contrasting approach to intentionality. According to Millikan's account, intentionality is a functional property -- attributions of intentionality provide information about the historical role of a particular trait but do not directly explain or predict the operations of that trait. To understand this it is useful to consider a non-intentional example of a functional property: the function of a sperm to penetrate an egg. Even knowing this function, one cannot predict that any particular sperm will penetrate an egg -- it is far more likely that it will not. Likewise, in intentional cases, one cannot predict that any particular organism will act in a way that is rationally predicted by attributing a state with intentional content. While it may be a function of that intentional state to produce the behavior, there is no more guarantee that a state such as a belief or a desire will fulfill its function than there is that a sperm will penetrate an egg. (See Bekoff and Allen 1992 for a discussion of why Millikan's theory is useful for informing and motivating studies in cognitive ethology.)

Different theories of intentionality have different consequences for specifying the contents of intentional states. Consider Dennett's intentional stance first. To attribute a belief in the conjunction of P and Q entails the attribution of the belief that Q for it would be irrational to fail to infer Q from the conjunction. Attributing this (rather minimal) rationality to subjects thus seems to entail that any subject capable of believing a conjunction must also be capable of believing each conjunct separately. But in Millikan's framework it is quite possible to have an intentional icon whose function it is to map onto the conjunction of P and Q without the system having either the ability or the tendency to represent the singular Q. Imagine, for example, a system whose Q-detector only becomes operative once its P-detector registers an occurrence of P. Such a system would be capable of representing the conjunction of P and Q without being able to represent Q alone. Perhaps, because Q rarely occurs in isolation, or when it does its occurrence is normally irrelevant to the organism, it was never important for the members of the species to have evolved isolated Q-detectors or the capacity for representing Q alone.

Our point at present is not to adjudicate between these different conceptions of intentionality. Rather, each provides a framework within which one may ask different kinds of questions about the behavior of animals. As such, each provides opportunities for research. Dennett's framework emphasizes orders of intentionality as a significant evolutionary variable, and Dennett (1983) suggests experiments that one might perform with vervet monkeys to test his ideas. Dennett is also concerned to explain how animals may sometimes show evidence of higher-order capabilities while at other times or in other contexts showing a lack of ability to reason at a similarly high level -- a phenomenon that would be puzzling if the animals were ideally rational. But from within Millikan's perspective this puzzle does not arise. This is because intentional states which are supposed (evolutionarily) to correspond to the intentional states of other organisms (second-order content) need not be related by inference to any general ability to form states with second-order intentional content. An animal may have very specific cognitive abilities with respect to particular intentional states of other organisms, without having the general ability to attribute intentional states to those organisms.

Returning to Rosenberg's third-order account of pretense we see that whether or not one regards it as plausibly attributed to nonhuman animals depends on the general account of intentionality that is adopted. From the intentional stance, if a believes that b believes that a desires to play (third-order) it would seem that ideal rationality would also require that a believes that b has a belief (second-order). But from a Millikanian perspective this more general second-order belief, if it requires a to have a general belief detector, may actually be more sophisticated than the third-order belief which supposedly entails it. A general belief detector may be much more difficult to evolve than a specific belief detector, for the detection of specific beliefs may be accomplished by the detection of correspondingly specific cues.

If this is correct, then on Millikan's account Jethro (Marc's dog) may be capable of the third-order belief that (or, at least, a state with the intentional content that) Sukie (Jethro's favorite canid play pal) wants Jethro to believe that her bite was playful not aggressive, even though Jethro is perhaps limited in his ability to represent and hence think about Sukie's second-order desires in general. Further below we shall argue for such an understanding of the content of play signals using Millikan's approach to intentionality.

If one takes a Dennettian approach to third-order intentionality, then Rosenberg's third-order analysis of pretense seems over-inflated. It is doubtful that many animals could make the general inferences that the rationality assumption seems to require them to be capable of making from any specific third-order belief. A particular behavioral sequence in social play may involve pretense even though neither participant has a general conception of pretense. In social play, an animal, a, may, for example, bare her teeth in a gesture that might also occur during or as a prelude to a fight. The playmate, b, may respond by growling -- another behavior that could occur during a fight. The first animal, a, may then pounce on b and grasp some portion of b's body between her teeth. This sequence involves motor patterns found in fighting, yet the animals are not fighting. What cognitive abilities must a and b possess for this to be possible? They must be capable of discriminating those occasions when a behavior is genuinely aggressive from those when it is playful. This could be achieved by detecting subtle differences between, for example, aggressive teeth baring and playful teeth baring -- if such differences exist. In the only study of its type of which we are aware, Hill and Bekoff (1977) found that bites directed towards the tail, flank, legs, abdomen, or back lasted a significantly shorter time and were more stereotyped during social play than during aggression in Eastern coyotes. Or it can be achieved by providing contextual cues that inform players about the difference between aggression and play. As we shall discuss below, in many species signals have evolved to support the second approach, and such signals may be understood as intentional icons that convey the messages about the intentions of the play participants.

Play signals

When animals play they typically use action patterns that are also used in other contexts, such as predatory behavior, antipredatory behavior, and mating. These action patterns may not be intrinsically different across different contexts, or they may be hard to discriminate even for the participants. To solve the problems that might be caused by, for example, confusing play for mating or fighting, many species have evolved signals that function to establish and maintain a play "mood" or context. In most species in which play has been described, play-soliciting signals appear to foster some sort of cooperation between players so that each responds to the other in a way consistent with play and different from the responses the same actions would elicit in other contexts (Bekoff 1975, 1978, 1995b; Bekoff and Byers 1981; Fagen 1981). Play-soliciting signals also provide aid to the interpretation of other signals by the receiver (Hailman 1977, p. 266). Coyotes, for example, respond differently to threat gestures in the absence of any preceding play signal than they do to threat gestures that are immediately preceded by a play signal or in the middle of sequence that was preceded by a play signals (Bekoff 1975). Given the possible risks that are attendant on mistaking play for another form of activity, it is hardly surprising that animals should have evolved clear and unambiguous signals to solicit and maintain play.

The canid "play bow", a highly ritualized and stereotyped movement that seems to function to stimulate recipients to engage (or to continue to engage) in social play (Bekoff 1977), provides an excellent example of what we are calling a play signal and it has been extensively studied in this context. That play bows are important for initiating play is illustrated by the example of a dominant female coyote pup who was successful in initiating chase play with her subordinate brother on only 1 of 40 (2.5%) occasions. Her lone success occurred on the only occasion in which she had signaled previously with a bow, although on the other occasions she engaged in a variety of behaviors that are sometimes successful in initiating play such as rapid approach/withdrawals, exaggerated pawing toward the sibling's face, and head waving and low grunting (Bekoff, 1975).

To say that the bow is stereotyped is to say that the form that play bows take is highly uniform without implying anything about the contextual versatility with which bows are used. When performing a bow, an individual crouches on its forelimbs, remains standing on its hindlegs, and may wag its tail and bark. The bow is a stable posture from which the animal can move easily in many directions, allows the individual to stretch its muscles before and while engaging in play, and places the head of the bower below another animal in a non-threatening position. Play-soliciting signals show little (but some) variability in form or temporal characteristics (Bekoff 1977). The first play bows that very young canids have been observed to perform are highly stereotyped, and learning seems to be relatively unimportant in their development. The stereotyped nature of the play bow is probably important for avoiding ambiguity.

Play bows occur throughout play sequences, but most commonly at the beginning or towards the middle of playful encounters. In a detailed analysis of the form and duration of play bows (Bekoff 1977) it was shown that duration was more variable than form, and that play bows were always less variable when performed at the beginning, rather than in the middle of, ongoing play sequences. Three possible explanations for this change in variability include (1) fatigue, (2) the fact that animals are performing them from a wide variety of preceding postures, and (3) there is less of a need to communicate that this is still play than there is when trying to initiate a new interaction. These explanations are not exclusive alternatives.

The meaning of play bows

Play bows occur almost exclusively in the context of play, and it is common to gloss play-soliciting signals with the message "what follows is play" or "this is still play". What is the significance of these glosses for the players themselves? Are they in any way aware of the meaning of the play bows, or are they simply conditioned to respond differently, e.g. less aggressively or less sexually, when a specific action such as a bite or a mount is preceded by a play bow?

One way to approach this question is to ask whether play signals such as bows are used to maintain social play in situations where the performance of a specific behavior during a play bout could be misinterpreted. A recent study of the structure of play sequences (Bekoff 1995b) showed that bows in some canids, infant and adult domestic dogs, infant coyotes, and infant wolves, often are used immediately before and immediately after an action that can be misinterpreted and disrupt ongoing social play. Recall that the social play of canids (and of other mammals) contains actions, primarily bites, that are used in other contexts that do not contain bows (e.g. agonistic, predatory, or antipredatory). Actions such as biting accompanied by rapid side-to-side shaking of the head are used in aggressive interactions and also during predation and could be misinterpreted when used in play.

Bekoff asked the following questions: (1) What proportion of bites directed to the head, neck, or body of a play partner and accompanied by rapid side-to-side shaking of the head are immediately preceded or followed by a bow? (2) What proportion of behavior patterns other than bites accompanied by rapid side-to-side shaking of the head are immediately preceded or followed by a bow? Actions considered here were mouthing or gentle biting during which the mouth is not closed tightly and rapid side-to-side shaking of the head is not performed, biting without rapid side-to-side shaking of the head, chin-resting, mounting from behind (as in sexual encounters), hip-slamming, standing-over assertively, incomplete standing-over, and vocalizing aggressively (for descriptions see Bekoff 1974; Hill and Bekoff 1977). Not considered was the situation in which the recipient of bites accompanied by rapid side-to-side shaking of the head performed a bow immediately before or immediately after its partner performed bite accompanied by rapid side-to-side shaking of the head or other action, because these rarely occurred. It was hypothesized that if bites accompanied by rapid side-to-side shaking of the head or other behavior patterns could be or were misread by the recipient and could result in a fight, for example, then the animal who performed the actions that could be misinterpreted might have to communicate to its partner that this action was performed in the context of play and was not meant to be taken as an aggressive or predatory move. On this view, bows would not occur randomly in play sequences; the play atmosphere would be reinforced and maintained by performing bows immediately before or after actions that could be misinterpreted.

The results of Bekoff's study support the inference that bows might serve to provide information about other actions that follow or precede them. In addition to sending the message "I want to play" when they are performed at the beginning of play, bows performed in a different context, namely during social play, might also carry the message "I want to play despite what I am going to do or just did -- I still want to play" when there might be a problem in the sharing of this information between the interacting animals. Species differences were also found that can be interpreted by what is known about variations in the early social development of these canids (Bekoff 1974; see also Feddersen-Petersen 1991). The interspecific differences also are related to the question at hand. For example, infant coyotes are much more aggressive and engage in significantly more rank-related dominance fights than either the infant (or adult) dogs or the infant wolves who were studied. During the course of this study, no consistent dominance relations were established in either the dogs or the wolves, and there were no large individual differences among the play patterns that were analyzed in this study. Social play in coyotes typically is observed only after dominance relationships have been established in paired interactions. Coyotes appear to need to make a greater attempt to maintain a play atmosphere, and indeed, they seem also to need to communicate their intentions to play before play begins more clearly than do either dogs or wolves who have been studied (Bekoff 1975, 1977). Subordinate coyote infants are more solicitous and perform more play signals later in play bouts. These data suggest that bows are not non-randomly repeated merely when individuals want to increase their range of movement or stretch their muscles. However, because, among other things, the head of the bowing individual is usually below that of the recipient, bowing may place the individual in a non-threatening, self-handicapping, posture. Self-handicapping might occur when the bowing animal is dominant or subordinate to her partner: when the bower is dominant she may be sending the message "I do not want to dominate you more" and when the bower is subordinate she may be sending the message "I am not trying to dominate you."

Standing-over, which usually is an assertion of dominance in infant coyotes (Bekoff, 1974) but not in infant beagles or wolves of the same age was followed by a significantly higher proportion of bows in coyotes when compared to dogs or to infant wolves. Because bows embedded within play sequences were followed significantly more by playing than by fighting after actions that could be misinterpreted were performed (unpublished data), it does not seem likely that bows allow coyotes (or other canids) more readily to engage in combat, rather than play, by increasing their range of movement, although this possibility can not presently be ruled out in specific instances.

In addition to the use of signals such as bows, it is also possible that the greater variability of play sequences when compared to sequences of agonistic behavior (Bekoff and Byers 1981) allows animals to use the more varied sequences of play as a composite play signal that helps to maintain the play mood; not only do bows have signal value but so also do play sequences (Bekoff 1976; 1977). Self-handicapping (e.g., Altmann 1962), occurring, for example, when a dominant individual allows itself to be dominated by a subordinate animal, also might be important in maintaining on-going social play. Watson and Croft (1996) found that red-neck wallabies (Macropus rufogriseus banksianus) adjusted their play to the age of their partner. When a partner was younger, the older animal adopted a defensive, flat-footed posture, and pawing rather than sparring occurred. In addition, the older player was more tolerant of its partners tactics and took the initiative in prolonging interactions. While more data are needed, this study also suggests that the benefits of play may vary according to the age of the player.

Putting play in a broader cognitive context

The data presented above suggest that at least some canids (and most likely other mammals) cooperate when they engage in social play, and may negotiate these cooperative ventures by sharing their intentions. Fagen (1993, p. 192) also has noted that "Levels of cooperation in play of juvenile primates may exceed those predicted by simple evolutionary arguments . . . " In general, animals engaged in social play use specific signals to modulate the effects of behavior patterns that are typically performed in other contexts, but whose meaning is changed in the context of play. These signals are often flexibly related to the occurrence of events in a play sequence that might violate expectations within that sequence. Furthermore, the relationship of play to a cognitive appreciation of the distinction between reality and pretense provides an important link to other cognitive abilities, such as the ability to detect deception or to detect sensory error. Given these connections, a detailed consideration of some selected aspects of social play might help promote the development of more sophisticated theories of consciousness, intentionality, representation, and communication.

The ability to engage in pretend play (e.g., to manipulate an object as if it is something else) normally first appears in human children around 12 months of age (Flavell et al. 1987). This is well before children appear to be capable of attributing mental states to others. Human children also seem capable of engaging in social play before they have a developed theory of mind. Leslie, in the quotation at the head of this chapter, expresses surprise about the distortion of reality implied by pretense. We, however, are inclined to suggest that play is one way that an animal may learn to discriminate between its perceptions of a given situation and reality, learning, for example to differentiate a true threat from a pretend threat. From this perspective it would be perhaps more surprising if cognitively sophisticated creatures could get to this point without the experiences afforded by play (for related discussion see Parker and Milbraith 1994).

It is also possible that experiences with play promote learning about the intentions of others. Even if the general capacity for understanding the mental states of others is a specifically human trait, many other species may be able to share information about particular intentions, desires, and beliefs. How might a play bow serve to provide information to its recipient about the sender's intentional state? It is possible that the recipient shares the intentions (beliefs, desires) of the sender based on the recipient's own prior experiences of situations in which she performed bows. Given our earlier discussion of specialized mechanisms, it may be reasonable to attribute a very specific second-order inference of the form "when I bow I want to play so when you bow you want also to play" without being committed to a general capacity for the possession of second-order mental states in these animals.

Recently, Gopnik (1993, p. 275) has argued that " . . . certain kinds of information that comes, literally, from inside ourselves is coded in the same way as information that comes observing the behavior of others. There is a fundamental cross-modal representational system that connects self and other." Gopnik (see also Meltzoff and Gopnik 1993) claims that others' body movements are mapped onto one's own kinesthetic sensations, based on prior experience of the observer, and she supports her claims with discussions of imitation in human newborns.

For example, Gopnik wants to know if there is an equivalence between the acts that infants see others do and the acts they perform themselves, and imagines "that there is a very primitive and foundational 'body scheme' that allows the infant to unify the seen acts of others and their own felt acts into one framework" (Gopnik, 1993, p. 276). If by "primitive and foundational" Gopnik means phylogenetically old, then there should be some examples, or at least precursors, of this ability in other animals. Gopnik and her colleague Andrew Meltzoff also consider the possibility that there is "an innate mapping from certain kinds of perceptions of our own internal states . . . In particular, we innately map the body movements of others onto our own kinesthetic sensations. This initial bridge between the inside and the outside, the self and other, underlies our later conviction that all mental states are things both we and others share" (Gopnik, 1993, p. 275; see also Flanagan 1992, pp. 102ff).

How these ideas might apply to nonhuman animals awaits further study. There are preliminary suggestions that Gopnik's ideas might enjoy some support from comparative research on animal cognition. For example, Savage-Rumbaugh (1990, p. 59) noted that "Likewise, if Sherman screams when he is upset or hurt, Sherman may deduce that Austin is experiencing similar feelings when he hears Austin screams. This view is supported by the observation that Sherman, upon hearing Austin scream, does not just react, but searches for the cause of Austin's distress." This cause-effect relationship is generated after sufficient experience -- if an animal screams when he is upset or hurt he may deduce that another is experiencing similar feelings when he hears a scream. Tomasello, Gust, and Frost (1989) also note that some gestures in chimpanzees may be learned by "second-person imitation" -- "an individual copying a behavior directed to it by another individual" (p. 35). They conclude (p. 45) that chimpanzees " . . . rely on the sophisticated powers of social cognition they employ in determining what is perceived by a conspecific and how that conspecific is likely to react to various types of information . . . "

There is also the possibility that in social play one dog might be able to know that another dog wants to play by knowing what she feels like when she performs a play bow. Among the questions that need to be studied in detail is "Does a dog have to have performed a bow (or other action) before knowing what a bow means and subsequently being able to make attributions of mental states to other individuals?" The following two hypotheses would have to be distinguished: (i) viewing a play bow induces a play mood in the recipient because of kinesthetic mapping and (ii) viewing a play bow induces knowledge in the recipient of how the actor feels. With respect to bows, at least, there are data that suggest that there is a genetic component to them; the first bows that are observed to be performed by young canids are highly stereotyped and occur in the correct social context (Bekoff 1977). Could these data support Gopnik's idea about the "primitive and foundational 'body scheme'"? And, if so, how is learning incorporated into the development of social communication skills? Regardless of how nature and nurture mix, sparse evidence at hand supports the view that studies of animal cognition can inform the study of human cognition, and that much more comparative research is needed.

Concluding remarks: Social play and comparative studies of animal cognition

Because social play is a widespread phenomenon, especially among mammals, it offers the opportunity for much more truly comparative and evolutionary work on intentionality, communication, and information sharing (see also Parker and Milbraith 1994). The collection of new data will provide for a much broader perspective on the origins of intentionality in diverse species. Nonetheless, some primatologists write as if theirs are the only subjects who are capable of recognizing the intentions of others. For example, Byrne (1995, p. 146) writes: "...great apes are certainly 'special' in some way to do with mentally representing the minds of others. It seems that the great apes, especially the common chimpanzee, can attribute mental states to other individuals; but no other group of animals can do so -- apart from ourselves, and perhaps cetaceans." To dismiss the possibility that nonprimates are capable of having a theory of mind, not only do more data need to be collected, but existing data about intentionality in nonprimates need to be reconsidered (see also Beck 1982 on "chimpocentrism). Furthermore, claims about the uniqueness of nonhuman primates are often based on very few comparative data derived from tests on small numbers of nonhuman primates who might not be entirely representative of their species. The range of tests that have been used to obtain evidence of intentional attributions is also extremely small, and such tests are often biased towards activities that may favor apes over monkeys or the members of other species. There is evidence (Whiten and Ham 1992) that mice can outperform apes on some imitation tasks. These data do not make mice "special"; rather they suggest that it is important to investigate the abilities of various organisms in respect to their normal living conditions. The study of social play affords this opportunity.

Acknowledgments

Some of the material in this chapter has been excerpted from Allen and Bekoff (1997) with permission of The MIT Press. Colin Allen was supported by NSF grant SBR-9320214 during the writing of this chapter. Maxeen Biben provided helpful comments.

References

Allen, C. and Bekoff, M. 1994. Intentionality, social play, and definition. Bio. Phil., 9, 63-74.

Allen, C. and Bekoff, M. 1997. Species of Mind: The Philosophy and Biology of Cognitive Ethology. Cambridge, Massachusetts: MIT Press.

Altmann, S. A. 1962. Social behavior of anthropoid primates: Analysis of recent concepts. In: Roots of Behavior (Ed. by E. L. Bliss), pp. 277-285. New York: Harper.

Beck, B. B. 1982. Chimpocentrism: bias in cognitive ethology. J. Human Evol., 11, 3-17.

Bekoff, M. 1974. Social play and play-soliciting by infant canids. Amer. Zool., 14, 323-340.

Bekoff, M. 1975. The communication of play intention: Are play signals functional? Semiotica ,15, 231-239.

Bekoff, M. 1976. Animal play, problems and perspectives. Persp. Ethol., 2, 165- 188.

Bekoff, M. 1977. Social communication in canids, Evidence for the evolution of a stereotyped mammalian display. Science, 197, 1097-1099.

Bekoff, M. 1978. Social play, Structure, function, and the evolution of a cooperative social behavior. In: The Development of Behavior: Comparative and Evolutionary Aspects (Ed. by G Burghardt and M. Bekoff), pp. 367-383. New York: Garland.

Bekoff, M. 1995a. Cognitive ethology and the explanation of nonhuman animal behavior. In: Comparative Approaches to Cognitive Science (Ed. by H. L. Roitblat and J.-A. Meyer), pp. 119-150. Cambridge: Massachusetts: MIT Press.

Bekoff, M. 1995b. Play signals as punctuation, The structure of social play in canids. Behaviour, 132, 419-429.

Bekoff, M. 1997. Playing with play, What can we learn about evolution and cognition? In: The Evolution of Mind (Ed. by D. Cummins and C. Allen) New York: Oxford University Press.

Bekoff, M. and Allen, C. 1992. Intentional icons: towards an evolutionary cognitive ethology. Ethology, 91, 1-16.

Bekoff, M. and Byers, J. A. 1981. A critical reanalysis of the ontogeny of mammalian social and locomotor play, An ethological hornet's nest. In: Behavioral Development, The Bielefeld Interdisciplinary Project (Ed. by K. Immelmann, G. W. Barlow, L. Petrinovich, and M. Main), pp. 296-337. New York: Cambridge University. Press.

Bekoff, M. and Jamieson, D. (Ed.) (1996). Readings in Animal Cognition. Cambridge, Massachusetts: MIT Press.

Burghardt, G. M. 1996. Play. In: Encyclopedia of Comparative psychology (Ed. by G. Greenberg and M. Haraway). New York: Garland.

Byers, J. A. and Walker, C. 1995. Refining the motor training hypothesis for the evolution of play. Amer. Nat., 146, 25-40.

Byrne, R. 1995. The Thinking ape: Evolutionary Origins of Intelligence. New York: Oxford University Press, .

Darwin, C. 1871/1936. The Descent of Man and Selection in Relation to Sex. New York: Random House; Modern Library edition.

Dennett, D. C. 1983 Intentional systems in cognitive ethology, The "Panglossian paradigm" defended. Behav. Brain Sci., 6, 343-390

Dennett, D. C. 1987. The Intentional Stance. Cambridge, Massachusetts: MIT Press.

Fagen, R. M. 1981. Animal Play Behavior. New York: Oxford University Press.

Fagen, R. 1993. Primate juveniles and primate play. In: Juvenile Primates: Life History, Development, and Behavior (Ed. by M. E. Pereira and L. A. Fairbanks), pp. 183-196. New York: Oxford University Press.

Feddersen-Petersen, D. 1991. The ontogeny of social play and agonistic behaviour in selected canid species. Bonn. Zool. Beitr., 42, 97-114.

Flanagan, O. J. 1992. Consciousness Reconsidered. Cambridge, Massachusetts: MIT Press.

Flavell, J., Flavell, E, and Green, F. 1987. Young Children's Knowledge about the Apparent-Real and Pretend-Real Distinctions. Dev. Psych., 23, 816-822.

Gopnik, A. 1993. Psychopsychology. Consciousness and Cognition, 2, 264-280.

Grice, H. P. 1957. Meaning. Phil. Rev., 66, 377-388.

Hailman, J. P. 1977. Optical Signals: Animal Communication and Light. Bloomington, Indiana: Indiana University Press. .

Hill, H. L. and Bekoff, M. 1977. The variability of some motor components of social play and agonistic behaviour in infant eastern coyotes Canis latrans var. Anim. Behav., 25, 907-909.

Huber, P. 1810. Recherché sur les moeurs des fourmis indigenes. Paris, Geneve: J. J. Paschoud.

Jamieson, D. and Bekoff, M. 1993. On aims and methods of cognitive ethology. Phil. Sci. Assoc., 2, 110-124.

Leslie, A. M. 1987. Pretense and representation. The origins of "theory of mind." Psych. Rev., 94, 412-426.

Meltzoff, A., and Gopnik, A. 1993. The role of imitation in understanding persons and developing a theory of mind. In: Understanding Other Minds (Ed. by S. Baron-Cohen, H. Tager-Flusberg, and D. Cohen), pp. 335- 366. New York: Oxford University Press.

Millikan, R. G. 1984. Language, Thought, and Other Biological Categories. Cambridge, Massachusetts: MIT Press.

Parker, S. T. and Milbraith, C. 1994. Contributions of imitation and role-playing games to the construction of self in primates. In Self-Awareness in Animals and Humans: Developmental Perspectives (Ed. by S. T. Parker, R. W. Mitchell, and M.L. Boccia), pp. 108-128. New York: Cambridge University Press.

Povinelli, D. J. and Cant, J. G. H. 1995. Arboreal clambering and the evolution of self-conception. Q. Rev. Bio., 70, 393-421.

Rosenberg, A. 1990. Is there an evolutionary biology of play? In: Interpretation and Explanation in the Study of Animal Behavior, Vol. 1, Interpretation, Intentionality, and Communication (Ed by M. Bekoff & D. Jamieson), pp. 180-196. Boulder, Colorado: Westview Press. (Reprinted in Bekoff and Jamieson 1996.)

Savage-Rumbaugh, E. S. 1990. Language as a cause-effect communication system. Phil. Psych., 3, 55-76.

Schaller, G. B. and Lowther, G. R. 1969. The relevance of social carnivore behavior to the study of early hominids. Southwest J. Anthro., 25, 307- 341.

Stich, S. 1983. From Folk Psychology to Cognitive Science. Cambridge, Massachusetts: MIT Press.

Tomasello, M., Gust, D., and Frost, G. T. 1989. A longitudinal investigation of gestural communication in young chimpanzees. Primates, 30, 35-50.

Watson, D. M. and Croft, D. B. 1996. Age-related differences in playfighting strategies of captive male red-necked wallabies (Macropus rufogriseus banksianus). Ethology, 102, 336-346.

Whiten, A. & R. Ham 1992 On the nature and evolution of imitation in the animal kingdom: Reappraisal of a century of research. Adv. Study Behav., 21, 239-283.