Notes towards a deep history of narrative
Francis F. Steen
University of California Los Angeles
Under submission; please do not quote
January 2004

1. Introduction

Narrative is a ubiquitous cultural artifact, generated in copious quantities by all known cultures past and present. Alongside acknowledged fictional narratives, causal explanations are frequently presented in narrative format: the creation of the world, why the bear has a stumped tail, the origin of fire. Literary narratives still have a pride of place in most cultures, whether in the form of orally delivered stories or silver screen fantasies. As a method of developing, encapsulating, and communicating causal models of nature and society, narrative has recently lost ground to ways of thinking that dispense with agents and their beliefs and desires. Yet while science shifts the burden of causality onto impersonal forces, narrative remains the favorite everyday mode of thinking for handling the complex causal networks of social reality.

In cognitive terms, narrative is a methodology for compressing information about a large number of cases into a single prototypical case. Listeners can uncompress the patterned significance of this prototype by projecting it back onto their own real and imagined pasts and futures. In this manner, narratives function as abstract models that structure, simplify, and lay out causal connections among otherwise unintelligible events. By trying out a succession of stories on a particular situation and seeing which ones most credibly generate the observed facts, narratives function as testably hypotheses in thought experiments. In these and other ways, narrative helps us orient ourselves as agents in a complex natural and social world.

These common uses of narrative and its pervasive presence in the times and places of recorded history indicate that narrative is profoundly integrated into the operation of the human mind. This integration expresses itself in part as a spontaneous preference for narratives over other forms of symbolic representation, in part in the wonderful ease and skill with which people make sense of their world by narrating it. Through narrative, we know how to make impressively effective use of the information we glean from a situation.

This claim will strike most people as so self-evidently true that it conveys no new information. However, it remains a puzzle that human beings should be so designed. In what sense are narratives computationally effective? Stories are in no way a favored form of representation for our artifically constructed thinking machines, even though computers are built by people and might by default be expected to share the unconscious cognitive biases of their designers. Unlike people, computers do not think it terms of stories. To bring this point home, Gregory Bateson used to tell the story of a man who asked his latest technological wonder, "Do you compute that you will ever think like a human being?" After a long pause the machine responded, "That reminds me of a story..." (Bateson 1980). The fact that computers don't employ the narrative method for organizing, remembering, and communicating information, or for generating inferences, highlights the oddity of the human reliance on stories.

In the following, I will suggest that the spectacular narrative performance we see in any human culture is in part made possible by well-established and tested adaptations with a deep biological history. In a nutshell, my proposal is that narrative bears the hallmark of natural selection operating on the ability to run simulations. More precisely, I will argue that the basic structure of narrative is a result of inherent contraints in the design of cognitive systems that evolved to accomplish the biological function of training by means of simulations. The initial changes required to allow the mind to run simulations in turn opened up new possibility spaces that were selectively exploited by a succession of cognitive adaptations. Cultural uses of narrative are thus able to piggyback on and recruit a set of neurobiological circuits that were subject to natural selection over various periods, some relatively recent and others stretching all the way back to the early mammals.

The deep history of narrative is complex and hard to reconstruct, and I will focus on highlighting one or two of the central threads of this story. I begin by giving an illustrative analysis of how stories function cognitively, spell out my proposals for the basic structure of narrative, and then trace the evolutionary history of some of the key cognitive adaptations that make narrative possible.

2. Little Red Riding Hood

Prompted by the wolf, Rotkäppchen "opened her eyes and saw the sunlight breaking through the trees and how the ground was covered with beautiful flowers" (Grimm, 1812). As she does, the listener conjures up in his mind's eye images of what it is she sees. These images must be constructed out of the reader's memories, and will vary from individual to individual. Perhaps in one person's mind, the scene is reconstructed from above, looking down on the little blonde girl standing amid the green grass and looking out and up on the beams of sunlight illuminating the flowers ahead of her, to the left of the path, which leads off the frame at around 110 degrees. In another's, the scene is seen from a position ahead of the little girl on the path, observing her brown ringlets as she looks for flowers on either side. In yet another's, the perspective is that of the little girl herself, the light coming down in shafts through dense fir trees, sheltering clusters of deep blue wood anemones. The forest itself may be the listener's prototypical forest -- light birchwoods, or dense redwoods, perhaps a specific location he remembers, and into which he places the little girl. In brief, if we could plug a television into the visual cortex of each listener and display their mental movies in a bank of screens, each monitor would show a different movie.

In spite of the fact that not a single frame will be identical, however, all the listeners to the story of Little Red Riding Hood will be very confident that they have all heard and understood the same story and participated in a common and shared experience. For the purpose of comprehending the story, it appears that the specifics of the mental imagery can be overlooked. This is the central puzzle of narrative: different movies, and yet with great conviction the same story.

On all the television screens, there will be a little girl with red headgear walking alone in a forest, picking flowers. It won't be the same girl or the same forest in any normal sense of the word "same" -- that is to say, she will look different for each one, it will be a different girl, and not a single tree or flower will stay constant across two monitors. It is not that these differences in visual instantiation are somehow negotiated and a consensus reached, or averaged with some slack given; these differences are not attended to. They are not subject to communication, and this startling fact is not even noticed. It suggests that differences in mental imagery are not a matter of lacking precision, some type of misunderstanding, or missing information -- rather, the details are in some radical sense not part of the story. In spite of the rich internal phenomenology the story generates, they carry little or no part of the significance of the story. At a simple propositional level, they don't mean anything.

This is an extravagant state of affairs. When the brain is generating moving, three-dimensional images on the fly, it is doing something that the most sophisticated computers still struggle to accomplish. While 3D movies are being produced, they are painstakingly put together frame by frame and then rendered; the computational power required to generate 3D on the fly is enormous. Now, the brain is doing this at the drop of a hat -- and for the purpose of a shared understanding, the specifics don't seem to matter one bit.

Narratives are individually instantiated in sophisticated and detail-rich private worlds. We don't talk about them, and the attempt to do so may lead to discomfort, as if we were competing over whose imagery is the most valid. One might infer, of course, that two people in fact don't ever reach a shared understanding of a story, and instinctively paper over their differences to minimize unpleasant disagreement, or worse, cover up the terror of an unbridgeable chasm between solipsistic minds. The interesting point is that this does not appear to be what's happening. Countless generations of children and their grandparents have no problems reaching a shared understanding of the story of Little Red Riding Hood. The reason they don't talk about their private imagery is simply that this would only introduce a distraction. What allows us to see the story as the same is its propositional structure, a structure that in itself is uninstantiated, a set of abstract relations.

3. The paradox of narrative thinking

A large literature from Propp (1928) onwards examines the distinction between an underlying narrative structure, the syuzhet or plot, and its surface manifestation, the fabula or story. The distinction I'm proposing between structure and instantiation is related but different. Where Propp looked for patterns between folktales, I'm asking for the basis on which listeners are able to reach a common understanding of a single tale, given that they each visualize a different mental movie.

Let's say that the structure of the narrative is whatever remains invariant across the multiple television screens that we've left plugged into our listeners' brains all this time. Now the point is that it would be extremely difficult for a computer program to pull out this structure. Remember, not a single pixel remains invariant across two screens. It's not a matter of seeing the same object from different angles, at different distances or speeds, under different types of lighting. These are also demonstrably different objects, at best vaguely similar. To pull the story out of these displays, you would need highly sophisticated dedicated equipment, far exceeding anything currently in existence in visual analysis software. You couldn't just approach the task with a blank slate; to find the pattern, you would need to know in advance exactly what to look for. The basic structure of narrative is simple: an agent in some circumstance, a goal, an obstacle to that goal, and some finite resources to deal with the obstacle.

Of course any child would do fine. She would know, without realizing that she knew, that stories are about people facing some difficulty, and needing to come up with a way of overcoming this difficulty. Seeing a movie of Little Red Riding Hood, her mind would effortlessly abstract these elements and make a series of complex inferences regarding appropriate strategies of action in some class of similar circumstances. Listening to the story read aloud, she would generate her own imagery and utilize it in a similar manner to generate inferences -- to construct a series of implicit morals, similar to the one that is explicitly provided in Perrault's 1697 version of the tale. Human minds are wired precisely in the requisite manner to solve this particular task -- in fact show a perverse preference for processing information in this roundabout fashion.

The point could be put even more strongly. The child appears to rely on a mental representation of the agent (Little Red Riding Hood), the setting (the forest), the goal (survival), the obstacle (the wolf) and the little girl's resources (her imperfect understanding of the danger) in order to generate the appropriate inferences. Even though the details of the mental imagery don't matter, the imagery itself is mandatory. That is to say, the plot must undergo a peculiar form of processing to be understood.

But why would natural selection produce a computational device that uses moving 3D images to make inferences about something that has nothing to do with the specifics of the images produced? It would be one thing if you generated complex imagery that would visualize the information for you, but this is not at all what's happening in narratives -- on the contrary, almost every feature of the visualized information has nothing to do with the target inferences. Only on a very abstract level can we say that the mental imagery instantiates the structural components in such a way as to make them intelligible. The general puzzle is, why do we want information presented in a narrative format? In what sense could this possibly be an efficient method of information processing? It's as if a computer, to make the right inference about 2 + 2, had to represent the numbers as dancing penguins. Inherent in its processing loop is the requirement that it generate moving three-dimensional dancing penguins, and when this display is sensed by some second component, it generates the correct inference: 4. It's an absurd situation that the mind should go to such extraordinary lengths to make a few generic inferences about wolves. How hard a task can this be to solve?

In the case of the fairly tale, for instance, let's say the target inference is "Don't stray from the path" and "Don't stop to talk to wolves". Or you might argue the inferences are really more complex; Perrault (1697), for instance, suggests that the "wolf" could also be understood as a friendly but wicked man out to take advantage of young girls. Yet however complex these morals are, it is not hard to state them briefly. Why not skip the story and just provide the moral? Why this fantastically circuitous route?

4. A functional take on consciousness

The brain we have today is an accretive structure, a result of a series of successive innovations added onto a continuously operating plant. New capacity is typically built as extensions of existing operations, constraining the scope of available solutions. When narrative, in spite of its extravagant resource consumption, remains a favored and effective use of the human cognitive machinery, the causes must be sought in our biological history.

The central role of the brain was and remains to enable and dispose the organism to respond to its environment in a manner that promotes survival and reproduction. The simplest way to accomplish the task of connecting sensory information to appropriate action is a response system that is triggered by a particular range of stimulus values. Under adaptive pressure, genetic mutations may arise that build cognitive mechanisms to broaden the scope of the data acquired, improve its quality, and produce a better targeted response. In mammals, primary sensory data acquisition systems are complemented by secondary sensory processing systems that refine the incoming data stream, extracting meaningful patterns, mapping the data onto a spatial grid, adding color, and multiplexing sound, vision, smell, and touch. Tertiary cognitive adaptations in turn rely on this processed information: inference systems that assess danger, make decisions about what to do next, and select mates on the basis of information that is already highly processed, synthesized, and multi-modal.

The situation is roughly analogous to the operation of a radio receiver. Everywhere we go, we are bathed and penetrated by a sea of broadcast signals that we have no direct sensory access to, and could make no sense of if we did. A specialized device, such as a radio or a television set, is required to process the signal into a type of information we're able to understand and utilize. Higher level cognitive mechanisms may find the raw feed from the senses as much gibberish as we find a stream of zeros and ones in an HDTV signal.

Consciousness apprears to play a role in mediating between such secondary sensory postprocessing systems and the tertiary inference engines that rely on their output. Perceptual consciousness presents these tertiary systems with a highly refined and processed version of sensory data, Sophisticated algorithms map the presentation back onto the incoming raw sensory signal in real time, compensating as best it can for the body's own movements. This produces the entirely mistaken but normally extremely useful impression that the consciously experienced "present", generated and communicated within the brain itself, is a direct and unmediated -- if inexplicable -- apprehension of the external world. As Fodor (1993) pointed out, the delusion is robust and encapsulated; the realization that conscious experience, full of light and color, must logically be taking place within the darkness of the brain has no effect on our subjective phenomenology. The advantage of mapping the interpretation back onto the raw signal is that you get rapid interactivity, with just a tiny lag. You can get a sense of this lag by turning your head rapidly: the visual construction blurs and objects may double, losing resolution and color. In a fraction of a second after movement stops, the high-resolution image is restored. Although the image on the retina moves violently, the external world is perceived to stand still and movement correctly attributed to the body.

A controversial aspect of consciousness is the question of whether it constitutes a type of display. Dennett (1991) has argued against the notion that consciousness is a "Cartesian theatre", as this would appear to presuppose a homuncular viewer. In the present model, however, the "audience" of the theater of consciousness is a large number of tertiary inference systems. The "presentation" provides them with the richly and appropriately structured information they use as input and require to operate. Now, why would these high-level inference systems need a display-type solution? The force of the metaphor of a display is that the coupling is loose: there is a many-to-many relation between the information that is shown on the display and the behavior generated in response. A loose coupling may be favored in situations where there are multiple inference systems working in parallel -- if you don't know beforehand which one will give you the response you need, you need to respond to several types of information at the same time, information gathering resources are finite and under pressure, and you must constantly reassign them new relative priorities. While this may sound complicated, it is also familiar, and recognizably the normal, everyday life for any mammal.

The upshot of this is that natural selection has favored a multi-tiered cognitive system dedicated to producing an analysis of sensory data that is oriented towards extracting information that is potentially relevant to the organism as an agent. The primary sensory data is refined by secondary processes into multi-modal objects and events, and at some stage -- as Marr (1982) argued, not necessarily the final -- this analysis is presented in consciousness. The display is not in itself committed to any particular action; rather, it maintains a loose coupling to the tertiary inference systems that drive behavior. In J.J. Gibson's terms, we perceive the world in terms of affordances, or possible actions (Gibson 1979). The key point here is that the high-level cognitive analysis that detects, defines, and acts in response to such affordances require as its input information that is already highly processed. Consciousness appears to play a crucial role in making this processed information available.

5. The origins of narrative

How does narrative become a part of this cognitive architecture? In brief, my suggestion is that it starts out as a training device. Mammals are born motorically inept and cognitively half unformed; the task of self-construction is incomplete at birth and must be sustained through infancy and youth. A range of skills can reliably be acquired through trial and error, but in a class of cases the cost of failure is unsustainably high. In high-stakes adversarial encounters, such as predation, you may not be given a second chance; you must succeed the first and every time. Nature has come up with an ingenious solution to this seemingly intractible problem: simulation. In a simulation, skills can be acquired in a safe environment where the targeted threat is absent and the cost of failure negligible. What I would like to propose is that narrative owes its basic design to the adaptive constraints present in constructing a device for simulation training.

Such simulation training, however, builds on prior art. The basic structure of narrative is commonly described as consisting of an agent in some initial condition, who has at their disposal some set of resources and capabilities; a goal, an obstacle to that goal, and one or more strategies to circumvent the obstacles, using the resources available, to reach the goal. Srini Narayanan, with the Neural Theory of Language group at Berkeley, has proposed a general control system structure shared by all higher-level motor control schemas. Before we perform an action, we need to get into a state of readiness; this initial state is followed by a starting process and a main process. The control system provides us with an option to stop, an option to resume, and an option to iterate or continue the main process. Finally, we check to see if a goal has been met, and complete the finishing process to reach the final state (personal communication; cf. Feldman and Lakoff, in preparation). Such a control system structure embodies key features of narrative structure. Universally experienced from infancy by any embodied organisms, motor control structures may provide us with neural circuitry that is subsequently recruited to reason conceptually about narrative.

A key qualification of Narayanan's model is that it applies to higher-level motor control schemas. It is clearly not the case that all actions need to have goals. In their simplest form, actions are reflex responses to stimuli, built into organisms by natural selection. If my hand is burned, I will withdraw it from the fire long before I have time to formulate a goal, thanks to an evolved reflex. It may be nevertheless be useful to describe this action as having the biological function of removing my hand from the fire; in this description, I rely on a metaphor of an imagined agent who projects a future goal. Yet such a projection is not an inherent part of action; it must be constructed and instantiated in optional cognitive machinery. Elementary actions don't have the structure of narrative, simply because they lack a key feature of narrative, that of virtualization. It is not until actions are orchestrated by an imagined goal that they fully acquire a narrative structure. What characterizes such actions is that they are driven by simulations.

In evolutionary terms, the ability to simulate represents a revolution in the scope and use of consciousness. Perceptual consciousness, which is imbued with the robust phenomenology of presenting information located outside of the body, is here harnessed to an entirely new purpose: the simulation of action. It is this revolution, I suggest, that marks the birth of narrative. The structure of narrative derives, in this view, from the adaptive constraints inherent in the design of a device for running effective simulations. A key domain for such simulations is training.

The principal requirement for a simulation system is that it must access and activate the same cognitive structures that handle the target situation. This constraint may be considered definitional; if it isn't met, what we have is not a simulation. In the case of predation, the simulation must activate the suite of systems that would respond to an actual attack by a predator. Call it the principle of virtual duplication.

A second, related constraint on the design of a simulation system is that it must produce relevant learning. It is not sufficient, for instance, merely to reproduce through simulation a particular past incident. It is the structure of the event that must be reproduced; the concrete details are incidental. Call it the principle of generativity.

By implication, the biological function of training will necessarily limit the machinery of simulation to fit into pre-existing cognitive structures and purposes. As argued in the previous section, evolution began by producing agent-centered cognitive adaptations. At the level of sensory organs, perceptual analysis, and high-level inference engines, cognitive processes are sculpted by natural selection into contributing to effective strategies of action. The principles of virtual duplication and generativity imply that a simulation must incorporate into itself the very structure of action, but translated into a virtual form. In the simulation, the agent must to some degree be conceptualized or constructed. In the simplest form of simulation, the dream, the agent is constructed as a virtual agent within the imagined scenario of the dream. In predation play, which enacts some component of an encounter with a predator, the players must take on respective roles of predator and prey. In self-handicapping behavior -- a feature seen in both human and animal play -- a virtual agent is constructed that is less capable along some dimension than the animal playing.

The third constraint relates to the mechanisms regulating the transfer of lessons learned in a simulation into the repertoire of real-life behavior. On the one hand, the simulated world must be kept clearly distinct from the real world, to protect us from psychosis. In the case of dreams, this is accomplished by decoupling the skeletal muscles from the motor cortex, and by a mechanism that erases the content of dreams on waking. In play, we enter a distinct ludic, self-constructive mode, where we construct a pretend space embedded within an executive space that tracks the real world (Steen & Owens, 2001). Yet maintaining perfect shutters between simulation and reality would defeat the biological function of training. On the other hand, then, the design of a simulation system must incorporate a mechansism for incorporating elements of play back into the individual's behavioral repertoire.

Evidence from the study of dreams (Revonsuo 2000), animal play (Fagen 1981), and children's play (Steen and Owens 2001), indicate that the adaptive pressure of predation played a central role in the evolution of cognitive mechanisms to enable simulation training. What is enacted in dreams, in animal's play, and in children's games is prototypically a story of an agent in dire circumstances, facing the threat of being killed and eaten by a predator, and looking for ways to get away. This is the mother of all narratives.

6. Mammalian dreams

To my proposal that narrative originates in the virtualization of action in dreams and in play, it may be objected that we do not know the content of animal dreams, and have no evidence that non-human animals tell stories. Let me begin by clarifying that I'm proposing only that the basic structure of narrative can be traced back to early mammalian adaptations for simulation training, not that there are no distinctive features of the human capacity for narrative. In the second half of this article, I will sketch out some of the structural elements that belong to our common mammalian narrative heritage, and then go on to examine some of the unique circumstances in primate evolution that eventually led to the capacity for full-fledged storytelling in humans.

Although we have yet to discover the video-out port of the visual cortex, and cannot display on a screen an animal's subjective experience of its self-produced visual simulations in the dream state, we have good reasons to think that other mammals dream like we do. Some of the earliest research, starting with Michel Jouvet's lab in Lyons, was conducted with cats. The first logs of EEG recordings of sleeping cats entering the dream state show that the researchers initially assumed that the cats had woken up, although they were lying perfectly still. They were astonished to realize they had discovered a third state of mind that was neither wakefulness nor quiescent rest. The brain waves in this state are very similar to the waking state and indicate a high level of brain activation (Jouvet and Delorme 1965). Following on the work of Dement, the state became known as the REM state, as it was typically accompanied by rapid eye movements (Dement 1958).

In humans, the REM state is associated with vivid, hallucinatory dreams -- that is to say, simulations that are mistaken for the real thing. The process is initiated in the brain stem, and waves of activation sweep through the limbic system, and flood the neocortex, starting at the back of the head in the visual cortex (Hobson 1999). These PGO (pontine-genicular-occipital) waves demonstrate that dreaming is built into the structure of the brain at several levels and suggest an ancient history. It is extremely implausible that dreams should serve a fundamentally different biological function in humans than in other mammals. Although other mammals are unable to describe the content of their dreams to us, the neurological evidence indicates that, just like us, they run mental simulations in their dreams, and, like us, are unaware in this state that they are merely dreaming.

Running mental simulations while the body is resting is an ingenious use of resources, not just because it taps into spare capacity. One of the problems with simulations is that they must by design access and activate cognitive response systems that are vital for survival. In order to avoid a confusion between what is perceived and what is generated by memory -- a risk Leslie (1987) called "representational abuse" -- the brain must allocate working memory resources to monitoring the sources of conscious representations. Non-human mammals appear to lack the cognitive equipment required to accomplish this task, a circumstance that puts a severe crimp on their ability to make use of simulations for training and other purposes. In the dream state, this obstacle finds an elegant solution: instead of devoting scarce or simply unavailable working memory resources to source monitoring, the skeletal muscles are decoupled from the brain. This is accomplished by depolarizing the cells in the brain stem that transmit the signals from the motor cortex. The hallucinatory dreams characteristic of the REM state are, in Hobson's term, psychotic, in that the dreamer fails to make the vital distinction between what she produces internally and what she perceives (Hobson 1999). This psychosis, however, is controlled and harmless. On waking, the content of the dream, however dramatic, is typically erased.

That animal dreams have a narrative structure was suggested by an ingenious if cruel surgical interventions. By interrupting the process of depolarization, Michel Jouvet was able as it were to switch the cat's body back on in the dream state. As the cats acted out their dream content, they engaged in species-typical activities of hunting, including hiding and pouncing (Jouvet and Delorme 1965). What Bekoff (Animal Planet 2002) notes in the case of play -- that prey animals typically play games of predator evasion and predators games of hunting -- is likely to also hold for dreams.

In the dream state, we are conscious, yet this consciousness is fundamentally psychotic: it fails to make a distinction between what is perceived and what is imagined. The confusion is harmless, since the communication between the motor cortex and the skeletal muscles has been suspended. Decoupled from the body, temporarily freed from its task of guiding the body in its interactions with the environment, the brain can devote all its resources to the task of simulation. The ability to run simulations dramatically lowers the cost of training, as rare but important events can be repeated and response systems honed. At the same time, the absence of any feedback from the real world places a limit on the utility of dream training.

7. Mammalian play

While the content of mammalian dreams remains somewhat elusive, we can observe play directly, which makes it easier to pin down its narrative structure. I suggested above that the prototypical mammalian narrative is that of an agent in dire circumstances, pursued by a predator, and looking for ways to escape. By calling mammalian play a narrative, I mean three things: that the behavior has a plot structure with a beginning, middle, and an end; that it imitates, in a virtualized form, an action that is serious, in the sense that the stakes are high; and that the solutions to the problem posed in the plot are loosely constrained by what would be possible in actuality. These pseudo-Aristotelian dimensions of mammalian play are a result of its biological function of simulation training.

In play, the act of simulation must be carefully managed within the limited cognitive resources available to the player. I argued briefly above and at length elsewhere (Steen and Owens, 2001) that play is characterized by a distinct self-constructive mode that differs systematically from the executive mode in which the individual accomplishes survival-related tasks such as feeding, fighting, reproducing, and resting. In the self-constructive mode, the biological goal of behavior is simulation training; from a subjective point of view, the activity is experienced as intrinsically enjoyable and an end in itself. In other words, while the distal cause is self-constructive learning, the proximate cause is simply enjoyment.

At the beginning of a play sequence, the self-constructive mode must first be established, and during the course of the game it must be sustained. The plot structure of play is thus embedded within a series of ongoing negotiations aimed at virtualization, or what we call "pretense," the key cognitive innovation in play. In dogs, the eliciting cues include [cite Bekoff on tape]. Preschoolers use broad play faces, sideways glances, and laughter to initiate play, along with "You can't catch me!" taunts, "Nana nana boo boo", and outright pleas: "Please chase me!" (Steen and Owens 2001). In the course of the game, chasers sustain the play mode by exaggerated stalking gaits, self-handicapped running, arrested lunges, and a running commentary: "Mmmm, snacktime!" (Steen and Owens 2001)

These cues also sustain the plot. Anecdotal evidence suggests that a game of chase with a child or a dog can take place without any conscious imagery (Steen and Owens, 2001). Simple cues of pretend danger suffice to trigger a response in play, suggesting that our brains may contain what Barrett (1999) has called a "predator-prey schema." Children have a deep-seated and intuitive understanding of what a "monster" is -- in brief, a generic, scary predator that wants to harm and maybe eat you. In pretense, the input threshold for the predator inference engine is is lowered: the grasping hand of a friend is allowed to trigger terror.

The plot itself is structurally simple. In pretense, your best friend turns into a monster, setting off a fear and flight response, but admixed with that peculiar emotion of play, thrill. Thrill is contingent on the coexistence of real safety and pretend danger: it is only when the child feels fully cared for and out of danger that she is able to immerse herself in the pretended crisis. This high level of required comfort is extended to control of the game itself. It is the chased who determines the pacing of the plot: the chaser will be told off if he catches his victim too early, and the child may interrupt the game at any time. At the end, if the child is finally exhausted and feels sufficiently comfortable with the chaser, the game may end with a climactic scene of ritual eating, where the pretend monster digs into the soft and vulnerable belly of the child, to mutual hilarity.

Play, even as it is experienced as frivolous, inconsequential, and fun, gravitates towards serious life and death plots. In the paradoxical world of play, the fatally dangerous becomes a seemingly perverse source of the greatest enjoyment. The threat of being caught by grasping hands and eaten alive is tirelessly sought out by generations of young children, in faithful memory of the sustained threat of predation in our ancestral past.

The most fascinating aspect of play, however, emerges not out of the cognitive constraints imposed by evolutionary path dependencies, but as a revolutionary opportunity. This is the use of play for exploring the possible. In our daily lives, in the executive mode, we are focused on actually accomplishing goals, and must inevitably be drawn towards strategies and modes of being that in the past have successfully brought us to these goals. There is an inherent bias towards the known and familiar, simply because the cost of trying out new strategies is unknown and could be unacceptably high. In play, the cost of failure is sharply reduced, as nothing is really at stake and the actions are undertaken merely for pretend. This opens up for play not merely to practice a set of known skills, but to explore possible strategies of action that are genuinely novel.

In describing a simple physical device, such as a pendulum, you can adopt one of two stances. The first is to measure the movements that the pendulum actually makes, as it moves back and forth. While this approach tells you the facts, the facts are in some broader sense arbitrary, contingent on the particular act that set the pendulum moving. The second approach is to interact with the pendulum yourself, and explore the full range of motions and positions it can occupy. What results from such an analysis is a description of what is called the pendulum's "phase space" -- the totality of the pendulum's possible behavior. Such a description necessarily goes beyond what the pendulum at any time is doing; indeed, there may be areas of the phase space that are never realized.

Play, I propose, opens the door to the organism's own phase space. The chief significance of this is that play can be utilized to generate new behavior. Especially in the race between predators and prey -- where, as Mark Ridley pointed out, you may have to run to stay in the same place -- new behavior may be rewarded with survival. The vast majority of what an animal could do is junk: the phase space is largely made up of territory to avoid. Yet hidden in places where nobody had gone before may be gems, strategies that handily defeat your opponent -- at least for a few rounds. On the boundaries of the ridiculous, as Napoleon is said to have remarked, is the sublime.

A human being's species-typical phase space -- the sum total of actions each one of us can undertake, by virtue of being human -- is staggeringly large, cannot be realized by a single individual, and has not been exhausted by history. Nevertheless, the phase space is in a broader sense real and knowledge of it true -- indeed, knowledge of what you or someone else could do is at times precisely the kind of truth that matters. In Aristotle's Poetics, the possible and the probably are criteria of dramatic quality that reflect the role of narrative in exploring the human phase space. Its limits are unknown, and narratives cannot in a simple manner be censored for exceeding them; indeed, as Aristotle puts it, "it is probable enough that things should happen contrary to probability" (Aristotle. Poetics. Chapter 25).

In play, the organisms utilizes safe environments and surplus resources to practise skills required for high-stakes adversarial encounters. By simulating dangerous events, the animal's response system is primed and trained. Because the cost of failure is low -- nothing is done for real; play is just for pretend -- the young animal is able to explore new strategies of action that may be too risky to try out in real hostile encounters. This exploration of the individual's phase space is by design experienced as intrinsically enjoyable; the reward system favors novel strategies that work. In contrast, boredom signals that the activity lacks the edge of innovation and new areas of the phase space should be opened up. Although animals don't tell stories, they enact the plot of the chase in a thousand stories.

8. Vervetspeak

Jouvet's studies of dreaming cats open for the possibility that, just like us, cats experience a simulated world in dreams. The close neurobiological homologies between human and feline dream states certainly do not contradict this inference. Yet if cats possess the cognitive machinery for generating mental simulations from memory, why don't they have more to show for it in their waking lives? Human beings use their waking imagination to plan the future, to model other minds, to communicate, and to transform their natural environments, Cats, in their waking lives, appear singularly focused on the perceived present.

The absence of evidence for widespread use of waking simulations in non-human mammals highlights the evolutionary puzzle of why humans have ended up with such distinctive and elaborate capabilities in this area. In general terms, the reason for the stark differential in the use of the imagination in cats and men is that cats lack the requisite cognitive adaptations for source monitoring. As we have seen, source monitoring is not an issue in the dream state, as the skeletal muscles are functionally decoupled from the brain. Elementary forms of play, such as those seen in cats, also do not appear to require mental imagery. In order to plan the future, however, or to model other minds and communicate what you are thinking, you need to be able to distinguish reliably between what you think and what you perceive -- and indeed, between what you remember and what you merely imagine.

A key component of such source monitoring is working memory. In order to sustain a mental simulation in consciousness in parallel with perceptual input, you need at the very least the working memory to hold the representation. From our own experience, we know that closing our eyes, or looking up at the sky or at a featureless surface, helps us create more detailed and vivid mental imagery. Could cats similarly just close their eyes for a moment and imagine their next clever move? There is little evidence they do, and for good reason. In addition to working memory constraints, a waking imagination requires cognitive systems for tracking the representational status of mental simulations. An imaginative cat that failed to distinguish between a remembered or imagined and a real mouse would quickly be weeded out by natural selection.

The broader argument is that under most circumstances, there are no advantages to a capacity for running waking mental simulations. It is a skill that is off the horizon of the adaptive landscape of most mammals at most times. The use of narrative, consequently, is restricted to dreams and play in all species but ours. To reconstruct the next stage in this history of narrative, we need to pinpoint some of the factors that played a role in setting our ancestors up for a cognitive revolution. The key dimensions to watch on the environmental front are feeding and predation, and how these affect working memory and group size.

Consider the case of our primate cousins, the monkeys. Subsisting on a diet of seeds, insects, and roots, they have successfully colonized a vast range of ecological niches, from the dense rainforest to treeless savannah. Because their food is fairly evenly distributed in the environment, there is no premium on devoting cognitive resources to tracking it. Simple behavioral rules like "move on if you spend more than a few minutes without finding food" suffice. Since no animal has better information than another, it's safe to follow whoever makes a move. Such flocking behavior permits large group sizes, which provides protection against predators.

Yet if feeding and maintaining group size are cognitively undemanding tasks, predation can drive new skills. Field experiments by Seyfarth and Cheney (1992) showed that vervets learn a basic repertoire of predator-specific warning cries. In the wild, they will respond appropriately to each warning, for instance by standing up and peering into the grass in response to the snake cry. When subjected to a repeated "false alarm" recording, they quickly figured out it should be ignored. Seyfarth and Cheney tentatively concluded that the vervets responded to the cries by calling up a mental image of the target predator. It was this mental image, they argued, that allowed the vervets to respond with the observed flexibility and contextual appropriateness.

Their reasoning betrays a tacit theory of consciousness similar to the one outlined above. A purely cue-based system, it implies, would lack the flexibility of a system that relies on a conscious display. By simulating a perception of the martial eagle, say, the vervets were able to activate the full range of high-level inference systems that would have been activated and available if they had seen the eagle themselves. Their behavior demonstrated a functionally significant loose coupling between the sensory signal and the behavioral response, a looseness that may be explained by appealing to the reliance of high-level inference engines on input from consciousness. This implied argument highlights why simulations are an extremely effective way of getting sophisticated computational and thus behavioral results: they simply tap into the sophisticated existing cognitive architecture of the mammalian brain.

This efficiency, however, is highly domain specific. The same argument would also explain why, unlike computers, we are constrained in a large number of cognitive tasks to engage in exorbitantly expensive simulations, rather than to work out the problem more directly. When a computer adds 2 + 2, it doesn't really even add 2 + 2 -- that's just a display it generates for our benefit. It actually computes directly in binary format. Because the human brain is designed to solve problems of survival in a rich sensory environment, it is wired in a manner that makes it extremely inefficient at solving basic numerical problems a computer finds trivial. While the dancing penguin example I used above may sound like a wild fantasy with no basis in reality, the fact is that if you want to teach a child to add 2 + 2, using dancing penguins may well be an effective pedagogy. Such is the absurdity of human cognitive architecture when taken out of its environment of evolutionary adaptedness.

Seyfarth and Cheney's work provide a powerful paradigm for understanding the power of mental imagery in building a communication system. A primary requirement, however, for increased sophistication in communication is that there is something to talk about. In the case of warning cries, it is easy to make the case that natural selection on a consistent basis favors individuals who respond to cries, and inclusive fitness or group selection would suffice to build the capacity and proclivity to signal. Yet in other areas of vervet life the benefit might just not be there for increased communication. As for the act of comprehension itself, it may be accomplished by no more than a brief flash of the memory of the predator in consciousness, only to be rapidly swamped by perceptual input. There are good reasons for consciousness to remain solidly anchored in the perceptual present. Any sustained deviation from this runs the imminent risk of degenerating into disfunctional, psychotic states. Although the machinery of waking simulations in some sense is in place, it doesn't scale, and it doesn't go anywhere. There's no persistent pressure to change. For tens of millions of years, monkeys have done well with the smarts they've already got.

9. A geographical imagination

While monkeys typically feed on things that are scattered loosely across the landscape, apes prefer foods that are harder to find and that go in and out of season. Foraging resources unevenly distributed in space and time places a premium on improved cognitive abilities for simulating features of your natural environment. If you are able to recall where a certain tree was located, and even better, when its fruits may be ripe, you have an edge. As in the case of vervets, chimpanzees appear to be capable of communicating about predators, but their minds provide a much richer simulation that includes invisible locations. Sue Savage-Rumbaugh (Dreifus 1998) recounts an exchange between the bonobo Panbanisha and a researcher at the Georgia State University Language Research Center:

"A few weeks ago, one of our researchers, Mary Chiepelo, was out in the yard with Panbanisha. Mary thought she heard a squirrel and so she took the keyboard and said, 'There's a squirrel.' And Panbanisha said 'DOG.' Not very much later, three dogs appeared and headed in the direction of the building where Kanzi was.

Mary asked Panbanisha, 'Does Kanzi see the dogs?' And Panbanish looked at Mary and said, 'A-frame.' A-frame is a specific sector of the forest here that has an A-frame hut on it. Mary later went up to 'A-frame' and found the fresh footprints of dogs everywhere at the site. Panbanisha knew where they were without seeing them.

And that seems to be the kind of information that apes transmit to each other: 'There's a dangerous animal around. It's a dog and it's coming towards you.'"

The exchange indicates that chimpanzees are capable of tracking out-of-view predators in a simulated landscape, a capacity we might call a geographical imagination. It appears to be focused on simulating absent but real events in their environment.

Thanks to the work of Savage-Rumbaugh and others, we are given a relatively direct insight into the content of the chimpanzee mind. The language itself, however -- the symbol boards, the gestures -- has been designed for the apes; humans have assigned the symbols' communicative significance and imparted them through patient training (Savage-Rumbaugh 2000). For this reason, the types of communicative relationships the language researchers have been able to establish with the apes don't generalize to the content of ape communication in the wild.

Ape play shows some dramatic new features. At the Language Research Center, the chimpanzees favorite remains the aboriginal game of chase, which may not require mental imagery. But they also enjoy being chased by a lab staffer dressed up in a gorilla suit (Dreifus 1998). The prop is of course created for them and the role enacted by a human, as a form of Vygotskian scaffolding for imaginative role play (cf. Berk & Winsler, chapter 3). Nevertheless, the ability to understand this level of pretense indicates that with some experience (they tend to be scared for real at first), they are able to imagine the person behind the mask. This provides further evidence that the chimpanzees can simulate a visually unavailable reality in their minds. Interestingly, the apes at the Language Research Center enjoy watching television, suggesting they have no difficulties distinguishing between a representation and reality. Such a basic source-monitoring capacity is a requirement for a geographical imagination. They are said to particularly enjoy home movies with themselves and their caretakers, as well as movies that involve interactions between apes and humans (Dreifus 1998).

On the social front, however, these impressive capacities are not without drawbacks. In the course of daily life, individual apes utilize information displayed in a private conscious simulation to guide their behavior. In narrative terms, the simulation is an action schema, one that may involve other individuals as well as oneself. The narrative schema, however, is not visible to others, rendering the behavior of each individual in the group comparatively unpredictable. Social coordination requires that everyone is responding to the same information: monkeys achieve this by responding largely to publicly accessible environmental cues; humans have developed sophisticated systems for developing common narrative schemas and for communicating the contents of their simulations. Neither of these strategies are viable options for apes. While language-trained chimpanzees can convey aspects of their mental simulations to their human companions by means of a relatively sophisticated symbol set, extensive field studies have uncovered no native ape language of comparable complexity. It may be that such communication requires the scaffolding not only of human language creators, but of human mental-state modeling interloqutors.

The difficulty is the following. In order to predict the behavior of others and coordinate with them, an ape would need to be able to imagine that another is imagining a certain course of action. Such recursive modeling may also be required to develop and sustain a communicative system focused on conveying the content of mental simulations. The evidence from behavioral studies in captivity and in the wild suggest that recursive modeling is not a core feature of ape cognition. It may simply be that apes lack the working memory capacity to simulate what another ape is simulating. Yet the cognitive equipment required is unlikely to be generic; mental-state attribution in humans involves a suite of specialized adaptations (Baron-Cohen 1995, Vogeley et al. 2001). Selection pressures don't appear to have constructed such a capacity in the apes to any elaborate degree.

In fact, the incentives are mixed at best. Foraging for fruits is inherently competitive, and on average you may be better off solitary -- a solution adopted by orangutans. In Africa, the pressures of predation may have precluded this solution, forcing the formation of larger groups. Yet social group size among the apes has declined sharply relative to their monkey ancestors, to a mean below a dozen individuals. An increase in individual size represents an alternative strategic solution: instead of bearing the costs of developing the necessary mental equipment for being able to sustain large groups, the apes have compensated by growing larger, and by sheltering in dense forests.

Once you have a creature whose behavior is directed by their own personal narratives, you get a dramatic increase in social complexity. Individuation makes it harder to achieve collective action; it no longer pays to follow whoever makes a move, since they may have inferior knowledge to another, or to yourself. Apes quickly run up against cognitive constraints in their abilities to construct recursive mental models of each other. Yet the ecological niche the apes entered into could be stably exploited once certain clearly definable problems had found a satisfactory solution. For millions of years, apes enjoyed an unbroken rainforest stretching from the Cape to northern China. Within the dense jungle, small groups could handle the pressures of predation, and their foraging niche favored the private utilization of individual knowledge. A balanced compromise -- smart enough to be individuated, but not smart enough to harness this individuation for a common good -- constrained the apes to a rainforest environment.

10. Hominid role play

Starting some ten million years ago, a cooling and drier climate began to shrink the edenic, fruit-bearing forests of the miocene apes. The eviction from paradise took place over millions of years, repeatedly and unrelentingly, starting in northern China and proceeding towards east Africa. Deprived of cover and of their staple foods, the apes were forced to adapt or perish. From an ape perspective, the new world was not only poorer in resources, it was also far more dangerous. On the spreading grassy woodlands, vast herds of ungulates emerged, hunted by swift and sharp-toothed predators. The first intact skulls of Australopithecines were found in breccia-filled cave deposits in South Africa, mixed in with smashed bones -- remnants of leopard's lairs, evidence not of early hominid aggression but of our ancestors' vulnerability to predation.

What were the options open to the emerging hominids for solving the problems of foraging and defense from predators in this new environment? In the fragmented habitats of the gradually shrinking rainforest, foraging became cognitively more demanding, and the challenge more open-ended: new types of food had to be exploited and tracked over expanded home ranges. In the gracile Australopithecus Africanus, a likely human ancestor, an overall increase in brain size was accompanied by expanded frontal lobes (Falk et al. 2000), the center for working memory in primates (Cabeza and Nyberg 2000; Haxby et al. 2000) and as such a critical component for running simulations (Mellet et al. 2000). Yet a greater ability to imagine and to formulate alternative courses of action increased the unpredictability of the individual and exacerbated the difficulties of social coordination found already among the apes. On the open woodlands of the pliocene, however, neither increased body size nor hiding in the trees presented a viable alternative: predation placed a palpable and persistent premium on larger and better-coordinated groups.

Two kinds of solutions for enabling collective action were available to these budding hominids: on the one hand, adapatations that make each individual more predictable; on the other, adaptations that improve each individual's ability to model others. While these solutions are radically different, they can play complementary parts in securing the adaptive outcome of group coordination. Natural selection is outcome-based; broad features of human behavior indicates both solutions were chosen.

Social coordination techniques makes consistent use of ways to render the individual's behavior more predictable to the others in the group. Consider the case of command structures. The reason the military is effective in collective action is not that each soldier builds a complex and accurate model of his companions that allows him to anticipate what they will do next and to coordinate his own behavior with theirs. Rather, each individual sharply constrains his own behavior by adopting a certain role, a central element of which is following orders from those who occupy other roles. The imperative towards such simplified solutions of the challenges of social coordination appears to have deep evolutionary roots. Young children spontaneously pick up the mannerisms, postures, and emotional of the people around them. Once their social circle expands beyond their family, they typically select the common dialect of the group in preference to that of their parents. Being different is experienced subjectively as undesirable and problematic.

Key to conformity is the development of social roles, templates of expected and predictable behavior. In children, social roles start out as "imitations of actions" -- that is to say, as acts of pretense, where a virtual agent is constructed in a simulated scenario. Such a virtual agent is present in an elementary form already in classic chase play: the mother of all narratives requires that one person adopt the role of the predator, the monster, and the other that of the prey. The construction of a virtual agent lies at the root of narrative; it is here that a narrative figure with certain resources can be constructed, and it is with reference to this virtual agent that the narrative challenges can be formulated and achieve appropriate calibration.

Because play permits a cheap exploration of the organism's phase space, it can function as the leading edge of cognitive and behavioral innovation. The development of social roles extends the achitecture of playful narratives into the reality of executive social interactions. Play itself embodies a radical simplification of interaction: the chaser is conveniently predictable to the chased. A particular narrative script is followed, and self-handicapping is employed to provide the other with the most appropriate degrees of freedom. Social roles in a similar manner prescribe a particular narrative script and similarly involves types of self-handicapping. Once the repertoire of social roles expands beyond the predator-prey schema utilized in elementary forms of play, social interactions become play-like without actually being playful. The architecture of the game has been recruited to simplify the task of social coordination.

By expanding the architecture of play into the social sphere, hominids became squatters in their own possibility space. In addition to social coordination, social role play can facilitate a modest pace of cultural innovation. Through a selective imitation of successful innovators, new strategies and modes of being can become part of an imperfectly shared cultural repertoire. This can take place without any explicit pedagogy; there is no requirement that the adults recognize or realize that the children have an appetite for social roles and that they are learning through their play.

A second way in which hominids were able to achieve the elusive goal of collective action is by evolving specialized machinery for modeling other minds. Elementary forms of mental-state modeling takes place in simple forms of play: to understand the pretend attack of the chaser, it is necessary to model his intention to attack. Barrett (1999) has proposed that mental-state attribution skills at first appear in predator-prey interactions. Gaze-detection, for instance, yields the inference that the attacker's epistemic state has changed: he now knows where you are and hiding is futile. As working-memory capacities expand, a richer subset of another's mental simulations can be recursively simulated, and the other can be emulated as a virtual agent in one's own consciousness. By simulating the resources, goals, and obstacles facing the other, your own inference systems kick in and provide you with emotional feedback and factual predictions about what the other is likely to want to do next. In collaborative situations, this facilitates group coordination.

Social role play facilitates mental-state attribution, in that a person playing a role can as an approximation be assumed to have the mental states conventionally associated with that role. In its simplest forms, role play becomes a proxy for mental-state modeling: you are able to model another simply because you know the narrative yourself.

Yet a richer and more individuated modeling was likely a prerequisite for the development of language. In order to understand another, you adopt what Sperber and Wilson (1995) call the "principle of relevance": the assumption that your interlocutor is conveying information she believes you need. This simple principle presupposes a recursive and mutual mental-state modeling. Although language is to some degree a code, much of the actual work of comprehension is inferential, and it is such inferences that rely on mental-state modeling. In the exchange between Mary Chiepelo and the bonobo Panbanisha, we see how the human interpreter supplies the mental modeling that permits the exchange to remain meaningful. Mary, modeling minds, asks, "Does Kanzi see the dogs?" The topic of her question is the content of Kanzi's perceptual consciousness, whose perspective she simulates as part of her concern for him. Panbanisha, in contrast, is simulating the geographical location of the dogs, and replies "A-frame." By modeling Panbanisha's mind, her human caregivers can conclude that "Panbanisha knew where they were without seeing them" (Dreifus 1998). It is only by filling in the disjunction in the dialogue with a rich appreciation of the nature of Panbanisha's mind that the conversation remains intelligible.

The first, unfortunate hominids found themselves in a harsh environment where there was a sustained natural selection for improved cognition. As the rainforests dried out, they were constrained to track seasonally variable resources scattered over large geographical areas. Predators, sustained by vast herds of ungulates occupying the emerging grasslands, had to be confronted on the ground, leaving no other option than a collective and coordinated defense. While feeding and defense had unequivocally pulled in opposite directions among the apes, keeping group sizes small, they began to converge among the hominids: new opportunities for scavenging and hunting favor and help sustain larger and collaborative groups. In addressing the challenge of maintaining group coordination among highly individuated members, two complementary strategies were open to our ancestors: to make the individual more predicable through the development of social roles, and to improve the ability to understand each other through the development of mental state modeling. Both required a domestication of the power of play for the purpose of collective action.

11. Little Red Riding Hood revisited

In the story of Little Red Riding Hood, an aboriginal mammalian drama is given a hominid twist. The narrative retains the elementary structure of a chase: a vulnerable victim is spotted, pursued, and -- depending on the version of the story you read -- is either caught and eaten or gets away.

Yet the story is not a chase. If a play chase is a simulation in action of a real chase, the story of a chase substitutes an imagined predator for the pretended predator. The mental image of the wolf is different for every child, yet every image emerges out of and makes explicit and comprehensible the concept of a wolf. The function of narrative consciousness is to communicate a conceptual, narrative structure to higher-order inference engines through a simulation of sensory perception. This act of communication within the mind itself is required because our higher-level (tertiary) inference systems evolved to take highly processed (secondary) perceptual information as their input. The thrill of the chase is thus conveyed over to the physically passive act of listening to a story.

In the geographical imagination, the story of Little Red Riding Hood is constructed as an event taking place in time and space: she walks from home through the forest towards the village where her grandmother lives, to bring her some food. She takes her time to enjoy the forest, "gathering nuts, running after butterflies, and making nosegays of such little flowers as she met with" (Lang 1889). The geographical level of the story must be understood before the more complex modeling can begin to make sense.

The predation theme is put to novel and specifically hominid uses. The evocation of this ancient narrative serves first of all the dramatic purpose of activating the excitement, fear, and thrill of predation play, thus ensuring the child's rapt attention. This narrative usage is very different from the original biological function of predation play, which is to train predator-evasion skills (Steen and Owens, 2001). In the story, the predator -- the wolf -- has become a metaphor for a deceitful and ill-intentioned man. His is a blend of a wolf and a human being. selectively drawing features from each (cf. Fauconnier and Turner 1998). While the predator features activate a primordial set of cognitive and physiological responses, the human features serve to explore aspects of the uniquely complex hominid possibility space. This possibility space is made possible by our expanded capacities for running recursive simulations for modeling other minds.

The wolf in the story, unlike real wolves, is a skilful mindreader, and he very subtly uses his skills to achieve his goals. When he first encounters Little Red Riding Hood in the forest, he doesn't attack her "because of some faggot-makers hard by in the forest" (Lang 1889). Now, why would the sound of woodmen discourage you? To make sense of the wolf's behavior, we must model the wolf's mind. As he encounters the girl, he quickly generates a conscious simulation of a possible future in which he at once attacks the girl. As he attacks, she screams in terror; he cannot stop her. Her screams are heard by the faggot-makers -- the wolf now adopts the perspective of the out-of-sight woodmen, and infers that since he can hear them, they will hear the child's desperate cries. What will the woodmen do when they hear the cries? The wolf knows that in this particular species of primate, the default behavior of adults is to come to the aid of children threatened by predators, even when they are not the adults' own offspring; he knows that such a defense is going to be well coordinated, that these are strong and large males, and that they are armed with deadly axes. The wolf's higher-level inference systems compute quite accurately that this scenario is altogether unappealing; his emotions notify his whole body that such an attack is extremely risky and should be avoided; and he instantly abandons this initially promising course of action.

Because the wolf, being a human blend and having the mental capacities of a human being, has already covered the overhead costs of building a powerful simulation machine, the marginal costs of running a particular simulation is tiny. It is cheap for him to run through this complex, counterfactual scenario; in his mind, he can explore possibility spaces that in real life would have been fatally expensive. The simulation of the woodmen that come to the child's aid is an embedded tale within the tale itself. Within the world of the tale, it never actually happens; in fact, the story is not even explicitly told. Yet for the child to understand the wolf's behavior, she must model the wolf's mind modeling the woodmen's mind, and reach the same conclusion.

The wolf runs through this possibility and rejects it fast, as they are still approaching each other. Before they begin to speak, he initiates a second counterfactual scenario. In this alternative simulation, he projects a future further ahead, a situation where he will be able to attack her and eat her under conditions of his own chosing, in a location where the woodmen won't hear her. This course of action produces far more desirable emotions in the wolf, as success seems far more probable. But how can the wolf, who intends to eat the little girl, reliably obtain information from her about where she is headed? He quickly realizes that if he communicates his intentions to her, or even allow them to shine through, she will become suspicious and afraid, and withhold this information from him, thwarting his new scenario. The wolf must not only delay his gratification; he must as best he can conceal his intentions from her, and make her think he has a different set of intentions than he actually does. How can he accomplish this? He must adopt the role of a trustworthy adult, someone the child can implicitly rely on. By manipulating her mind in this subtle manner, he increases his chances of successfully killing her eventually.

Innocent little Rotkäppchen is no match for this superpredator. His complex and rapid simulations are invisible and inaccessible to her; all she sees is the outward behavior of a friendly man. She takes this appearance at face value and fatally provides him with the information he requires, thereby not only endangering her own life, but placing her grandmother in imminent and mortal danger. Ignorant of her disasterous failure, she fritters away her time gathering beautiful flowers for a woman whom she has already indirectly killed.

The listening child, however, is shielded from the price that Little Redcap has to pay. He is given the tools to understand the wolf's intentions, as the story provides him with information about the wolf's mind that is unavailable to her. A key and distinctive function of hominid stories is to reveal the hidden connections between thought and action, so that the child can improve his skills at mental modeling. Narrative, for this reason, presents its readers with transparent minds (Cohn 1978), rendering the sequence of events hyperintelligible. The listening child is therefore led to infer that communication is not always a good thing. He needs to understand in a visceral manner that the wonderful gift of communication is also a peril. If you freely provide information to those who want to hurt you, you help them to destroy you and those you love. Little Redcap should have communicated to the woodmen that she needed protection, or she should have refused to speak to the wolf. If she really had her wits about her, she could have outwitted him by pretending in turn to believe and accept his pose of friendship and then provided him with information that was incorrect, saving herself and her grandmother and sending him off to some other location, perhaps a place where he would encounter a stronger adversary and be killed instead. These are alternative stories open to the listening child, once he has understood the challenge posed by the tale.

The story itself is an enactment of pretense. In chase play, the chaser pretends to be a monster by emitting cues such as stalking, grasping, and growling. In the story of Little Red Riding Hood, the roles are reversed: it is the wolf, the predator, that pretends to be the safe and beloved grandmother. Yet in the telling of the story, the storyteller -- who may herself be the listener's grandmother -- must pretend to be and enact the wolf pretending to be her. The situation is of course entirely absurd: no child would ever make the fatal categorical mistake of confusing a wolf for her grandmother. What makes this absurdity tolerable, indeed perfectly natural and thrilling, is that the wolf is in fact the grandmother -- it is the grandmother pretending to be the wolf. As the storyteller simultaneously recounts and enacts the story, the wolf is conjured into the present by invoking his salient features as cues. His big eyes, his big ears, and -- climactically -- his big teeth bring him alive and allows the child the experience the safe and terrifying thrill of being eaten while your grandma hugs you.

In listening to another, we construct a simulation out of the raw material of our personal memories. On the one hand, this construction provides our higher-level inference systems with the material they need to respond to the story in some way as if it actually happened. On the other hand, the conceptual grasp of the story that allows us to affirm a shared understanding is prior to and not dependent on the details of the simulation. Concepts have an interesting relation to consciousness: they must necessarily be instantiated in a particular form, drawing on personal memories, in order to be present in consciousness. Yet this instantiation is not in itself the concept. The image you utilize to represent the concept in consciousness does not exhaust the concept, which can be instantiated in an infinite number of ways. Most interestingly, human beings have what appears to be a very robust if entirely implicit understanding of the distinction between a concept and its simulated instantiation in consciousness. The ability to distinguish between a concept and its particular instantiation would appear to be a requirement for symbolic communication beyond some elementary level of complexity, since the instantiation cannot be communicated. This line of reasoning produces the somewhat surprising conclusion that a narrative, if what we mean by this term is the shared understanding a group of people have of a story, is not in itself conscious. What is conscious is only the individual instantiation of narratives, an instantiation that in itself is uncommunicable.

The proposal of this paper is that the reason conscious simulations play such a prominent role in our subjective experience of narrative, even though they play almost no role in our shared understanding, relate to the evolutionary origins of narrative as a simulation of perceptual consciousness. Through millions of years of evolution, our brains evolved a complex network of inference systems responding to highly processed information presented in perceptual consciousness. Mental simulations are able to tap directly into this machinery, activating the full range of cognitive responses as if (with reservations) the imagined event had been experienced and perceived in reality. From a pure information-processing perspective, this solution is hideously wasteful. It would be much better if the mind could operate directly on the conceptual structure of narrative and derive the appropriate inferences. The integrated architecture of consciousness, in which our emotions, bodies, and thoughts are intimately tied to conscious sensations, makes this unappealing and hard in practice, perhaps even impossible. We might call this perspective "stupid design theory": natural selection, constrained by prior choices, often favors very local optima. In the case of consciousness and narrative, perhaps we shouldn't mind: their wonderfully idiosyncratic design is hard to reproduce in machinery, making us irreplaceable to each other.

Works cited

Baron-Cohen, Simon (1995). Mindblindness: An essay on autism and theory of mind. Cambridge, MA: MIT Press.

Barrett, H.C. (1999). Human cognitive adaptations to predators and prey. Doctoral dissertation, Department of Anthropology, University of California at Santa Barbara.

Bateson, Gregory (1980). Mind and Nature. London: Fontana.

Berk, Laura E. and Adam Winsler (1995). Scaffolding children's learning: Vygotsky and early childhood education. Washington, DC: National Association for the Education of Young Children.

Boinski, Sue, and Paul A. Garber (eds.) (2000). On the move: how and why animals travel in groups. Chicago, IL: University of Chicago Press.

Cabeza, R. and L. Nyberg (2000). Imaging cognition II: An empirical review of 275 PET and fMRI studies. Journal of Cognitive Neuroscience 12: 1-47.

Cohn, Dorrit (1978). Transparent minds: narrative modes for presenting consciousness in fiction. Princeton, NJ: Princeton University Press.

Dement, William (1958). The Occurrence of Low Voltage, Fast, Electroencephalogram Patterns During Behavioral Sleep In the Cat. Electroencephalography and Clinical Neurophysiology 10: 291-96.

Dennett, Daniel C. (1991). Consciousness Explained. Boston: Little, Brown and Co.

Dreifus, Claudia (1998). She talks to apes and, according to her, they talk back. New York Times, National/Metro section, 14 April 1998.

Fagen, R. (1981). Animal play Behavior. New York, NY: Oxford University Press.

Falk D, Redmond JC Jr, Guyer J, Conroy C, Recheis W, Weber GW, Seidler H. (2000). Early hominid brain evolution: a new look at old endocasts. Journal of Human Evolution 38. 5: 695-717.

Feldman, Jerome, and George Lakoff (in preparation). From Molecule to Metaphor: The Neural Theory of Language.

Fodor, Jeffrey (1983). The Modularity of Mind: An Essay On Faculty Psychology. Cambridge, MA: MIT Press.

Gibson, James J. (1979). The ecological approach to visual perception. Boston, MA: Houghton Mifflin.

Haxby, J.V., L. Petit, L.G. Ungerleider, S.M. Courtney (2000). Distinguishing the functional roles of multiple regions in distributed neural systems for visual working memory. Neuroimage 11: 145-56, 380-91.

Hobson, J. Allan (1999). Dreaming as delirium: How the brain goes out of its mind. Cambridge, MA: MIT Press.

Jouvet, Michel, and Delorme, F. (1965). Locus coeruleus et sommeil paradoxal. Comptes Rendus des Séances et Mémoires de la Société de Biologie 159: 895-99.

Lakoff, George (1972). Structural Complexity in Fairy Tales. The Study of Man 1: 128-50.

Lang, Andrew (ed.) (1889). The blue fairy book. New York, NY: Longmans, Green & Co.

Leslie, Alan M. (1987). Pretense and representation: The origins of "Theory of Mind". Psychological Review 94. 4: 412-26.

Marr, D. (1982). Vision. San Francisco: W.W. Freeman.

Mellet E., N. Tzourio-Mazoyer, S. Bricogne, B. Mazoyer, S.M. Kosslyn, M. Denis (2000). Functional anatomy of high-resolution visual mental imagery. Journal of Cognitive Neuroscience 12. 1: 98-109.

Animal Planet (2002). The Power of Play. Broadcast 12 May.

Propp, Vladimir L. (1928). Morfologiia skazki. English tr. Morphology of the folktale. International Journal of American linguistics 24 (1958), no.4, pt.3.

Revonsuo, Antti (2000). The reinterpretation of dreams: An evolutionary hypothesis of the function of dreaming. Behavioral & Brain Sciences 23. 6: 877-901, 904-1018, 1083-1121.

Savage-Rumbaugh, Sue (2000). Linguistic, Cultural and Cognitive Capacities of Bonobos (Pan paniscus). Culture & Psychology 6. 2: 131-153.

Seyfarth, Robert M., and Dorothy L. Cheney (1992). Meaning and mind in monkeys. Scientific American 267. 6: 122-28.

Sperber, Dan and Deirdre Wilson (1995). Relevance: communication and cognition. 2nd ed. Cambridge, MA: Blackwell.

Vogeley, K., P. Bussfeld, A. Newen, S. Herrmann, F. Happé, P. Falkai, W. Maier, N. J. Shah, G. R. Fink, and K. Zilles (2001). Mind Reading: Neural Mechanisms of Theory of Mind and Self-Perspective. NeuroImage 14. 1: 170-181.