sign in sign up
<<
Home , Embodied cognition, Situated cognition

Psychonomic Bulletin & Review 2002, 9 (4), 625-636

THEORETICAL AND REVIEW ARTICLES

Six views of embodied

MARGARET WILSON University of California, Santa Cruz, California

The emerging viewpoint of holds that cognitive processes are deeply rooted in the body’s interactions with the world. This position actually houses a number of distinct claims, some of which are more controversial than others. This paper distinguishes and evaluates the following six claims: (1) cognition is situated; (2) cognition is -pressured; (3) we off-load cognitive work onto the environment; (4) the environment is part of the cognitive system;(5) cognition is for action; (6) off- line cognition is body based. Of these, the first three and the fifth appear to be at least partially true, and their usefulness is best evaluatedin terms of the range of their applicability.The fourth claim,I argue, is deeply problematic. The sixth claim has received the least attention in the literature on embodied cognition, but it may in fact be the best documented and most powerful of the six claims.

There is a movement afoot in to grant has emphasizedsensory and motor functions,as well as their the body a central role in shaping the . Proponents of importance for successful interaction with the environment. embodied cognitiontake as their theoretical starting point Early sources includethe viewof 19thcentury psychologists not a mind working on abstract problems, but a body that that there was no such thing as “imageless ” (Good- requires a mind to make it function. These opening lines win, 1999); motor theories of such as those sug- by Clark (1998) are typical:“Biologicalbrains are first and gested by William James and others (see Prinz, 1987, for a foremost the control systems for biologicalbodies. Biolog- review); the developmental of Jean Piaget, ical bodies move and act in rich real-world surroundings” which emphasized the emergence of cognitive abilities out (p. 506). of a groundwork of sensorimotor abilities; and the ecologi- Traditionally,the various branches of cognitivescience cal psychology of J. J. Gibson, which viewed perception in have viewed the mind as an abstract proces- terms of — potential interactions with the envi- sor, whose connections to the outside world were of little ronment. In the 1980s, linguists began exploring how ab- theoretical importance. Perceptual and motor systems, stract conceptsmay be based on metaphorsfor bodily,phys- though reasonable objects of inquiry in their own right, ical (e.g., Lakoff & Johnson, 1980). At the same were not considered relevant to understanding “central” time, within the field of artificial , - cognitive processes. Instead, they were thought to serve based began to emphasize routines for interacting merely as peripheral input and outputdevices.This stance with the environment rather than internal representations was evident in the early decades of , used for abstract thought (see, e.g., Brooks, 1986). when most theories of human thinking dealt in proposi- This kind of approach has recently attained high visi- tional forms of knowledge. During the same time period, bility,under the banner of embodied cognition.There is a artificial intelligencewas dominated by computer models growing commitment to the that the mind must be un- of abstract symbol processing. of mind, too, derstood in the context of its relationship to a physical made its contribution to this zeitgeist, most notably in body that interactswith the world. It is argued that we have Fodor’s(1983) modularityhypothesis.According to Fodor, evolved from creatures whose neural resources were de- central cognitionis not modular, but its connectionsto the voted primarily to perceptual and motoric processing, and world are. Perceptual and motor processing are done by whose cognitive activity consisted largely of immediate, informationally encapsulated plug-ins providing sharply on-lineinteractionwith the environment.Hence human cog- limited forms of input and output. nition, rather than centralized, abstract, and sharply However, there is a radically different stance that also has distinct from peripheral input and output modules, may in- roots in diverse branches of cognitive science. This stance stead have deep roots in sensorimotor processing. Although this general approach is enjoying increasingly Correspondence should be addressed to M. Wilson, Department of broad support, there is in fact a great deal of diversity in Psychology, University of California, Santa Cruz, CA 95064 (e-mail: the claims involved and the degree of controversy they at- [email protected]). tract. If the term embodied cognitionis to retain -

625 Copyright 2002 Psychonomic Society, Inc. 626 WILSON ful use, we need to disentangle and evaluate these diverse ceptual information continuesto come in that affects pro- claims. Among the most prominent are the following: cessing, and motor activity is executed that affects the 1. Cognition is situated. Cognitiveactivitytakes place environment in task-relevant ways. Driving, holding a in the context of a real-world environment, and it inher- conversation, and moving around a room while trying to ently involves perception and action. imagine where the furniture should go are all cognitiveac- 2. Cognition is time pressured. We are “mind on the tivities that are situated in this sense. hoof” (Clark, 1997),and cognitionmust be understoodin Even with this basic definition of what it means for cog- terms of how it functions under the pressures of real-time nition to be situated, we can note that large portions of interaction with the environment. human cognitive processing are excluded. Any cognitive 3.We off-load cognitivework onto the environment. activity that takes place “off-line,” in the absence of task- Because of limits on our information-processing abilities relevantinputand output,is by definitionnot situated.Ex- (e.g., limits on attention and working ), we exploit amples include planning,remembering, and day-dreaming, the environmentto reduce the cognitiveworkload.Wemake in contexts not directly relevant to the content of plans, the environment hold or even manipulate information for , or day-dreams. us, and we harvest that information only on a need-to- This observation is not new (see, e.g., Clark & Grush, know basis. 1999; Grush, 1997), but given the rhetoric currently to be 4. The environment is part of the cognitive system. found in the literature, the point is The information flow between mind and world is so dense worth emphasizing. By definition, situated cognition in- and continuous that, for scientists studying the of volves interaction with the things that the cognitive activ- cognitive activity, the mind alone is not a meaningful unit ity is about. Yetone of the hallmarks of human cognition of analysis. is that it can take place decoupledfrom any immediate in- 5. Cognition is for action. The function of the mind is teraction with the environment. We can lay plans for the to guide action, and cognitive mechanisms such as per- future, and think over what has happened in the past. We ception and memory must be understood in terms of their can entertain counterfactuals to considerwhat might have ultimate contribution to situation-appropriate behavior. happenedif circumstanceshad been different. We can con- 6. Off-line cognition is body based. Even when de- struct mental representations of situations we have never coupled from the environment, the activity of the mind is experienced,based purely on linguistic input from others. groundedin mechanisms that evolvedfor interactionwith In short, our ability to form mental representations about the environment—that is, mechanisms of sensory pro- thingsthat are remote in time and space, which is arguably cessing and motor control. the sine qua non of human thought, in cannot Frequentlyin the literatureon embodiedcognition,sev- yield to a situated cognition analysis. eral or all of these claims are presented together as if they An argument might be made, though,that situated cog- represented a single point of view.This strategy may have nition is nevertheless the bedrock of human cognition,due its uses, as for example in helping to draw a compelling to in our evolutionary history. Indeed, it is popular to try picture of what embodied cognition might be and why it to drive intuitions about situated cognition by invoking a might be important. This may have been particularly ap- picture of our ancestors relying almost entirely on situated propriate at the time that attention first was drawn to this skills.Before we got civilized,the argument goes, the sur- set of , when audiences were as yet unfamiliar with vival of our mental abilities depended on whether this way of conceptualizingcognition.The time has come, they helped us to act in direct response to immediate situ- though,to take a more careful look at each of these claims ations such as obtaining food from the environment or on its own merits. avoiding predators. Thus, situated cognition may repre- sent our fundamentalcognitive architecture, even if this is Claim 1: Cognition Is Situated not always reflected in the artificial activities of our mod- A cornerstone of the embodied cognition literature is ern world. the claim that cognitionis a situated activity(e.g., Chiel & This view of early humans, though, most likely exag- Beer, 1997; Clark, 1997; Pfeifer & Scheier, 1999; Steels gerates the role of these survival-related on-line activities & Brooks, 1995; a commitment to situated cognition can in the daily lives of early humans. With respect to obtain- also be found in the literature on dynamical systems— ing food, meat eating was a late addition to the human e.g., Beer, 2000; Port & van Gelder, 1995; Thelen & Smith, repertoire, and even after the onset of hunting, the large 1994; Wiles & Dartnall, 1999). Some authors go so far as majority of calories were probably still obtained from to complain that the phrase “situated cognition” implies, gathering. Evidence for this claim comes from both the falsely, that there also exists cognition that is not situated fossil record and the dietary of hunter/gatherers (Greeno & Moore, 1993, p. 50). It is important, then, that today (Leaky, 1994), as well as from the dietary patterns we be clear on what exactly it means for cognition to be of our nearest relatives, the chimpanzees and bonobos situated. (de Waal, 2001). It might be more appropriate, then, to Simply put, situated cognition is cognition that takes consider gathering when trying to construct a picture of place in the context of task-relevant inputs and outputs. our cognitive past. But gathering lends itself much less That is, while a cognitiveprocess is being carried out, per- well to a picture of human cognitionas situated cognition. SIX VIEWS OF EMBODIED COGNITION 627

Successful gathering might be expected to benefit a great tions from early on. Indeed, once the representational ca- deal from human skills of reflective thought—remember- pacity of emerged, it is unclear why its full ca- ing the terrain, coordinating with one’s fellow gatherers, pacity in this respect would not be used. consideringthe probableimpact of last week’srain, and so Along different lines, Brooks (1999, p. 81) argues that on. During the actual act of gathering, though, it is not because nonsituatedcognitiveabilitiesemerged late in the clear what situated cognitiveskills humans would bring to history of animal life on this planet, after extremely long bear beyond those possessed by any foraging animal. (Put periods in which no such innovationsappeared, these were in this light, we can see that even hunting, early human therefore the easy problems for evolution to solve (and style, probably involved considerable nonsituated mental hence, by implication,not of much theoreticalinterest). In activity as well.) fact, exactly the oppositecan be inferred. Easy evolution- In addition to chasing food, though, being chased by ary solutionstend to arise again and again, a process known predators is also supposed to have been a major shaping as convergent evolution. In contrast, the late emergence force, according to this pictureof the early human as a sit- and solitary status of an animal with abilitiessuch as man- uated cognizer. Yet while avoiding predators obviouslyhas ufacturing to a mental template, language,and artistic de- a great deal of survival value, the situated skills of fight- pictionattests to a radical and complex innovationin evo- or-flight are surely ancient,shared with many other species. lutionary engineering. Again, it is not clear how much mileage can be gotten out In short, an argument for the centrality of situated cog- of trying to explain human intelligencein these terms. In- nition based on the demandsof human survivalin the wild stead, the cognitive abilities that contributed to uniquely is not strongly persuasive. Furthermore, overstating the human strategies for avoiding predation were probably of case for situated cognitionmay ultimatelyimpede our un- quite a different sort. As early humans became increas- derstanding of the aspects of cognitionthat in fact are sit- ingly sophisticated in their social abilities, avoiding pre- uated. As be discussed in the next two sections, there dation almost certainly involvedincreasing use of off-line is much to be learned about the ways in which we engage preventative and communicative measures. in cognitive activity that is tightly connected with our on- Finally,we should consider the mental activitiesthat are going interactionwith the environment.Spatial cognition, known to have characterized the emerging human popu- in particular, tends to be situated.Trying to fit a piece into lation and that set them apart from earlier hominid species. a jigsawpuzzle,for example,may owe more to continuous These included increasingly sophisticated tool-making, reevaluating of spatial relationships that are being contin- particularly the shaping of tools to match a mental tem- uously manipulated than it does to any kind of disembod- plate; language, allowing communicationabout hypothet- ied matching (cf. Kirsh & Maglio, 1994). For cer- icals, past events, and other nonimmediate situations; and tain kindsof tasks, in fact, humans may activelychooseto depictiveart, showing the abilityto mentally represent what situate themselves (see Section 3). is not present, and to engage in representation for repre- sentation’ssake rather than for any situated functionality Claim 2: Cognition is Time Pressured (see Leakey, 1994,for further details).All of these abilities The previous section considered situated cognitionsim- reflect the increasingly off-line nature of early human ply to mean cognition that is situation bound. There ap- thought. To focus on situated cognition as the fundamen- pears to be more, though, that is often meant by “situated tal principleof our cognitivearchitecture is thus to neglect cognition.”It is frequentlystated that situatedagents must these species-defining features of human cognition. deal with the constraints of “real time” or “runtime” (see, A few counterarguments to this can be found in the lit- e.g., Brooks, 1991b;Pfeifer & Scheier, 1999, chap. 3; van erature. Barsalou (1999a), for example, suggests that lan- Gelder & Port, 1995). These phrases are used to highlight guage was used by early humans primarily for immediate, a weakness of traditional models, situated, indexical purposes. These situated uses of lan- which are generally allowed to build up and manipulate guage were intended to influence the behavior of others internal representations of a situation at their leisure. A during activities such as hunting, gathering, and simple real creature in a real environment,it is pointedout, has no manufacturing.However, some of the examples that Barsa- such leisure. It must cope with predators, prey, stationary lou gives of situated uses of language appear to be in fact objects,and terrain as fast as the situationdishes them out. off-line uses, where the referent is distantin time or space— The observation that situated cognitiontakes place “in real as, for example, in describingdistant terrain to people who time” is, at bottom, an observation that situated cognition have never seen it. One can easily think of further nonsitu- must cope with time pressure. ated uses of language that would serve adaptive functions A belief in the importance of time pressure as a shap- for early humans: absorbing parental edicts about avoiding ing force in cognitive architecture underlies much of the dangerousbehaviors;holdingin mind instructionsfor what situated cognition literature. For example, in the field of materials to go fetch when helping with tool manufactur- behavior-based robotics, “autonomous agents” have been ing; deciding whether to join in a planned activity such as built to perform tasks such as walking on an uneven sur- going to the river to cool off; and comprehending gossip face with six legs (Quinn & Espenschied, 1993), brachi- about members of the social hierarchy who are not present. ating or swinging “branch to branch” like an ape (Saito & It seems plausible,then,that languageserved off-line func- Fukuda, 1994), and navigating around a cluttered envi- 628 WILSON ronment lookingfor soda cans without bumpinginto any- are necessary, but they are at the leisure of the cognizer. thing (Mataric, 1991). Each of these activities requires (Of course, any task can be performed in a hurry, and real-time responsiveness to feedback from the environ- many often are. But the state of “being in a hurry” is one ment. And althoughthese activitiesare not especially“in- that is cognitively self-imposed, and such tasks are gener- telligent” in and of themselves, it is claimed that greater ally performed only as fast as they can be, even if this cognitive complexity can be built up from successive lay- means being late.) Situations in which time pressure is in- ers of procedures for real-time interaction with the envi- herently part of the task, such as playing video games ronment (for reviews, see Brooks, 1999; Clark, 1997; or changing lanes in heavy traffic, may actually be the Pfeifer & Scheier, 1999). exception. A similar emphasis on time pressure as a principle that This is not to say,though,that an understandingof real- shapes cognitioncan be seen as well in human behavioral time interaction with the environmenthas nothing to con- research on situated cognition. For example, Kirsh and tribute to our understanding of human cognition. A num- Maglio (1994) have studied the procedures that people use ber of important domains may indeed be illuminated by in making time-pressured spatial decisions while playing considering them from this standpoint.The most obvious the video game Tetris (discussedin more detailin Section3). of these is perceptuomotorcoordinationof any kind.Even This research is conducted with the assumption that situ- such basic activities as walking require continuousrecip- ations such as Tetris playingare a microcosm that can elu- rocal influence between perceptual flow and motor com- cidate general of human cognition. mands. Skilled hand movement, particularly the manipu- One that time pressure is thoughtto is that lation of objects in the environment, is another persuasive it creates what has been called a “representational bottle- example of a time-locked perceptuomotor activity. More neck.” When situations demand fast and continuously sophisticatedforms of real-time situated cognitioncan be evolving responses, there may simply not be time to build seen in any activity that involves continuous updating of up a full-blown mental model of the environment, from plans in response to rapidly changing conditions. Such which to derive a plan of action. Instead, it is argued, being changing conditions often involve the activity of another a situated cognizer requires the use of cheap and efficient human or animal that must be reckoned with. Examples tricks for generating situation-appropriate action on the include playing a sport, driving in traffic, and roughhous- fly. (In fact, a debate has raged over whether a situated ing with a dog. As interesting as the principles governing cognizer would make use of internal representationsat all; these cases may be in their own right, though, the argu- see Agre, 1993; Beer, 2000; Brooks, 1991a; Markman & ment that they can be scaled up to provide the governing Dietrich, 2000;Vera & Simon,1993.)Thus,takingreal-time principles of human cognitionin general appears to be un- situated action as the starting point for cognitive activity persuasive. is argued to have far-reaching consequencesfor cognitive architecture. Claim 3: We Off-Load Cognitive Work Onto the The force of this argument, though, depends upon the Environment assumption that actual cognizers (humans, for example) Despite the fact that we frequently choose to run our are indeed engineered so as to circumvent this represen- cognitive processes off line, it is still true that in some sit- tationalbottleneckand are capable of functioningwell and uations we are forced to function on line. In those situa- “normally” in time-pressured situations.But although one tions, what do we do about our cognitive limitations? One might wish an ideal cognitive system to have solved the response, as we have seen, is to fall apart. However, hu- problem, the assumption that we have solved it is dis- mans are not entirely helpless when confronting the rep- putable. Confronted with novel cognitive or perceptuo- resentational bottleneck, and two types of strategies ap- motor problems, humans predictablyfall apart under time pear to be available when one is confronting on-line task pressure. That is, we very often do not successfully cope demands. The first is to rely on preloaded representations with the representational bottleneck. Lift the demands of acquired through prior (discussed further in Sec- time pressure, though, and some of the true power of tion 6). What about novel stimuli and tasks, though? In human cognitionbecomes evident.Given the opportunity, these cases there is a secondoption,which is to reduce the we often behave in a decidedlyoff-line way: steppingback, cognitive workload by making use of the environment it- observing,assessing,planning,and onlythen taking action. self in strategic ways—leaving information out there in It is far from clear, then, that the human cognitive system the world to be accessed as needed, rather than taking time has evolved an effective engineering solution for - to fully encode it; and using epistemic actions (Kirsh & time constraints of the representational bottleneck. Maglio, 1994) to alter the environment in order to reduce Furthermore, many of the activitiesin which we engage the cognitive work remaining to be done. in daily life, even many that are clearly situated,do not in- (The environmentcan also be used as a long-termarchive, herently involve time pressure. Cases include mundane as in the use of reference books, appointment calendars, activities,such as making sandwiches and paying bills, as and computer files. This can be thought of as off-loading well as more demanding cognitive tasks, such as doing to avoid memorizing, which is subtly but importantly dif- crossword puzzles and reading scientific papers. In each ferent from off-loading to avoid encoding or holding ac- of these cases, input from and output to the environment tive in short-term memory what is present in the immedi- SIX VIEWS OF EMBODIED COGNITION 629 ate environment.It is the lattercase that is usually discussed In fact, though,potentialuses of off-loading may be far in the literatureon off-loading.Althoughthe archival case broader than this. Consider, for example, such activities certainly constitutes off-loading, it appears to be of less as countingon one’sfingers, drawing Venn diagrams, and theoretical interest. The observation that we use such a doingmath with penciland paper. Many of these activities strategy does not seem to challenge or shed light on exist- are both situated and spatial,in the sense that they involve ing theoriesof cognition.The present discussionwill there- the manipulation of spatial relationships among elements fore be restricted to what we may call the situated exam- in the environment. The advantage is that by doing actual, ples of off-loading, which are the focus of the literature.) physical manipulation,rather than computing a solution in Some investigators have begun to examine how off- our heads, we save cognitive work. However, unlike the loading work onto the environmentmay be used as a cog- previousexamples,there is also a sense in which these ac- nitive strategy. Kirsh and Maglio (1994), as noted earlier, tivities are not situated. They are performed in the service have reported a study involvingthe game Tetris, in which of cognitive activity about something else, something not falling block shapes must be rotated and horizontally present in the immediate environment. translated to fit as compactly as possible with the shapes Typically, the literature on off-loading has focused on that have already fallen. The decision of how to orient and cases in which the world is being used as “its own best place each block must be made before the block falls too model” (Brooks, 1991a, p. 139). Rather than attempt to far to allow the necessary movements. The data suggest mentally store and manipulateall the relevant detailsabout that players use actual rotation and translationmovements a situation, we physically store and manipulate those de- to simplify the problem to be solved, rather than mentally tails out in the world, in the very situation itself. In the computing a solution and then executing it. A second ex- Tetris case, for example, the elements being manipulated ample comes from Ballard, Hayhoe, Pook, and Rao (1997), do not serve as tokens for anything but themselves, and who asked subjects to reproduce patternsof colored blocks their manipulation does not so much yield information under time pressure by draggingrandomly scattered blocks abouta solutionas produce the goal state itselfthroughtrial on a computerscreen into a work area and arranging them and error. In contrast, actionssuch as diagramming repre- there. Recorded eye movements showed repeated refer- sent a quite different use of the environment. Here, the encing of the blocks in the model pattern, and these eye cognitivesystem is exploitingexternal resources to achieve movements occurred at strategic moments—for example, a solution or a piece of knowledge whose actual applica- to gather information first about a block’scolor and then tion will occur at some later time and place, if at all. later about its precise location within the pattern. The au- Notice what this buys us. This form of off-loading— thors argue that this is a “minimal memory strategy,” and what we might call symbolic off-loading—may in fact be they show that it is the strategy most commonly used by applied to spatial tasks, as in the case of arranging tokens subjects. for armies on a map; but it may also be applied to non- A few moments’ thought can yield similar examples spatial tasks, as in the case of using Venndiagrams to de- from daily life. Not all of them involve time pressure, but termine logical among categories. When the pur- other cognitive limitations, such as those of attention and pose of the activity is no longer directly linked to the working memory, can drive us to a similar kind of off- situation,it also need not be directly linked to spatial prob- loading strategy. One example, used earlier, is that of lems; physical tokens, and even their spatial relationships, physically moving around a room in order to generate so- can be used to represent abstract, nonspatial domains of lutions for where to put furniture. Other examplesinclude thought. The history of mathematics attests to the power laying out the pieces of something that requires assembly behind this decoupling strategy. It should be noted, too, in roughly the order and spatialrelationshipsthat they will that symbolic off-loading need not be deliberate and for- have in the finished product, or giving directions for how malized, but can be seen in such and automatic to get somewhere by first turning one’sself and one’slis- behaviorsas gesturing while speaking.It has been found that tener in the appropriate direction.Glenberg and Robertson gesturing is not epiphenomenal,nor even strictly commu- (1999) have experimentally studied one such example, nicative, but seems to serve a cognitive function for the showing that in a compass-and-map task, subjects who speaker, helpingto grease the wheels of the thoughtprocess were allowed to indexicallylink written instructionsto ob- that the speaker is trying to express (see, e.g., Iverson & jects in the environment during a learning phase per- Goldin-Meadow,1998; Krauss, 1998). As we shall see in formed better during a test phase than subjects who were Section 6, the use of bodily resources for cognitive pur- not, both on comprehension of new written instructions poses not directly linked to the situation has potentiallyfar and on performance of the actual task. reaching consequencesfor our understandingof cognition As noted earlier, this kindof strategyseems to applymost in general. usefully to spatial tasks in particular. But is off-loading strictly limited to cases in which we manipulatespatial in- Claim 4: The Environment Is Part formation? Spatial tasks are only one arena of human of the Cognitive System thought. If off-loading is useful only for tasks that are The insightthat the body and the environmentplay a role themselves spatial in nature, its range of applicabilityas a in assisting cognitiveactivityhas led some authors to assert cognitive strategy is limited. a strongerclaim: that cognitionis not an activityof the mind 630 WILSON alone, but is instead distributed across the entire interact- ments must in addition have properties that are affected ing situation,includingmind, body,and environment(see, by their participationin the system. Thus, the various parts e.g., Beer, 1995, pp. 182–183; Greeno & Moore, 1993, of an automobile can be considered as a system because p. 49; Thelen& Smith, 1994,p. 17; Wertsch, 1998,p. 518; the action of the spark plugs affects the behavior of the see also Clark, 1998, pp. 513–516, for discussion). In fact, pistons, the pistons the drive shaft, and so on. relatively few theorists appear to hold consistently to this But must all things that have an impact on the elements position in its strong form. Nevertheless, an attraction to of a system themselves be considered part of the system? somethinglike this claim permeates the literatures on em- No. Many systems are open systems, existing within the bodiedand situatedcognition.It is therefore worth it to bring context of an environment that can affect and be affected the core idea into focus and consider it in some detail. by the system. (No system short of the entire universe is The claim is this: The forces that drive cognitive activ- truly closed, although some can be considered closed for ity do not reside solely inside the head of the individual, practicalpurposes.) Thus, for example,an ecologicalregion but instead are distributed across the individual and the on earth can be considered a system in that the organisms situation as they interact. Therefore, to understand cogni- in that region are integrallydependenton one another;but tion we must study the situation and the situated cognizer the sun need not be considered part of the system, nor the together as a single, unified system. rivers that flow in from elsewhere, even thoughtheir input The first part of this claim is trivially true. Causes of is vital to the ecological system. Instead, the ecological behavior (and also causes of covert cognitive events such system can be considered an open system, receiving input as )are surely distributedacross the mind plus en- from something outside itself. The fact that open systems vironment. More problematic is the reasoning that con- are open is not generally considered a problem for their nects the first part of the claim with the second part. The analysis, even when mutual influence with external forces fact that causal control is distributed across the situation is continuous. is not sufficient justification for the claim that we must From this description,though,it should be clear that how studya distributedsystem. Science is not ultimatelyabout one defines the boundaries of a system is partly a matter explainingthe of any particularevent.Instead, it of judgment and depends on the particular purposes of is about understanding fundamental principles of organi- one’sanalysis.Thus, the sun may not be part of the system zation and function. when one considers the earth in biologicalterms, but it is Consider, for example, the goal of understandinghydro- most definitely part of the system when one considers the gen. Before 1900,hydrogenhad been observed by scientists earth in terms of planetary movement. The issue, for any in a large number of contexts,and much was known about given scientific enterprise, is how best to carve nature at its behavior when it interacted with other chemicals. But its joints. none of this behavior was really understood until the dis- Where does this leave us with respect to defining a cog- covery in the 20th century of the structure of the atom, in- nitive system? Is it most natural, most scientifically pro- cluding the protons, neutrons, and electrons that are its ductive,to considerthe system to be the mind; or the mind, components and the discrete orbits that electrons inhabit. the body, and certain relevant elements in the immediate Once this was known, not only did all the previous obser- physical environment, all taken together? To help us an- vations of hydrogen make sense, but the behavior of hy- swer this question, it will be useful to introduce a few addi- drogen could be predicted in interactions with elements tional concepts regarding systems and how they function. never yet observed.The causes of the behaviorof hydrogen First, a system is defined by its organization—that is, are always a combination of the nature of hydrogen plus the functional relations among its elements. These rela- the specifics of its surrounding context; yet explanatory tions cannot be changed without changing the of satisfaction came from understandingthe workings of the the system. Next, systems can be described as either fac- narrowly defined system that is the hydrogenatom. To have ultativeor obligate.Facultativesystems are temporary, or- insisted that we focus on the study of contextualized be- ganized for a particular occasion and disbanded readily. havior would probably not have led to a theoretical under- Obligate systems, on the other hand, are more or less per- standingwith anythinglike this kind of explanatoryforce. manent, at least relative to the lifetime of their parts. Distributed causality, then, is not sufficient to drive an We are now in a position to make a few observations argument for . Instead, we must ask about a “cognitive system” that is distributed across the what kind of system we are interested in studying. To an- situation.The organizationof such a system—the functional swer this, we must consider the meaning of the word sys- relations among its elements, and indeed the constitutive tem as it is being used here. For this purpose, the contri- elements themselves—would change every time the per- butions of systems theorists will be of help. (For a lucid son moves to a new location or begins interacting with a summary of the issues discussed below,see Juarrero, 1999, different set of objects. That is, the system would retain its chap. 7.) identity only so long as the situation and the person’stask For a set of things to be considered a system in the for- orientation toward that situation did not change. Such a mal sense, these things must be not merely an aggregate, system would clearly be a facultativesystem, and faculta- a collection of elements that stand in some relation to one tive systems like this would arise and disband rapidly and another (spatial, temporal, or any other relation). The ele- continuouslyduringthe daily life of the individualperson. SIX VIEWS OF EMBODIED COGNITION 631

The distributed view of cognition thus trades off the ob- ory,” as Glenberg (1997) similarly argues, “evolved in ser- ligate nature of the system in order to buy a system that is vice of perception and action in a three-dimensionalenvi- more or less closed. ronment” (p. 1). If, on the other hand, we restrict the system to include First, let us consider the case of visual perception. The only the cognitive architecture of the individual mind or traditionalassumptionhas been that the purpose of the vi- brain, we are dealing with a single, persisting, obligatesys- sual system is to build up an internal representation of the tem. The variouscomponentsof the system’sorganization— perceived world. What is to be done with this representa- perceptual mechanisms, attentionalfilters, working mem- tion is then the job of “higher” cognitiveareas. In keeping ory stores, and so on—retain their functionalroles within with this approach, the ventral and dorsal visual pathways that system across time. The system is undeniably open in the brain have been thoughtof as the “what” and “where” with respect to its environment, continuously receiving pathways, generating representations of structure input that affects the system’sfunctioning and producing and spatial relationships, respectively. In the past decade, output that has consequences for the environment’sfur- though, it has been argued that the dorsal stream is more ther impact on the system itself. But, as in the case of hy- properly thought of as a “how” pathway. The proposed drogen, or an ecosystem, this characteristic of openness function of this pathway is to serve visuallyguided actions does not compromise the system’s status as a system. such as reaching and grasping (for reviews, see Goodale & Given this analysis, it seems clear that a strong view of dis- Milner, 1992; Jeannerod, 1997). tributed cognition—that a cognitivesystem cannot in prin- In support of this, it has been found that certain kinds ciple be taken to comprise only an individualmind—will of visual input can actually prime motor activity. For ex- not hold up. ample, seeing a rectangle of a particular orientation facil- Of course we can reject this strong version of distrib- itates performance on a subsequent grasping task, pro- uted cognitionand still accept a weaker version, in which vided that the object to be grasped shares that orientation studying the mind-plus-situation is considered to be a (Craighero,Fadiga,Umiltà, & Rizzolatti,1996).This prim- promising supplementary avenue of investigation, in ad- ing occurs even when the orientationof the rectangledoes dition to studying the mind per se. Two points should be not reliably predict the orientation of the object to be noted, though. First, taken in this spirit, the idea of dis- grasped. A striking corollary is that visual input can acti- tributed cognition loses much of its radical cachet. This vate covert motor representations in the absence of any view does not seek to revolutionize the field of cognitive task demands. Certain motor neurons in monkeys that are science, but simply adds to the list of phenomena that the involved in controllingtool use also respond to seen tools field studies.Likewise, chaos theory did not revolutionize without any motor response on the part of the or overturn our understandingof physics, but simply pro- (Grafton, Fadiga, Arbib, & Rizzolatti,1997;Murata et al., vided an additional tool that helped to broaden the range 1997).Behavioraldata reported by Tucker and Ellis (1998) of phenomena that physics could characterize success- tell a similar story. When subjectsindicatewhether common fully. (Indeed, some examples of research on distributed objects (e.g., a teapot, a frying pan) are upright or inverted, topics appear to stretch the bounds of what we would rec- response are fastest when the response hand is the ognize as cognition at all. The study of the organized be- same as the hand that would be used to grasp the depicted havior of groups is one such example; see, e.g., Hutchins, object (e.g., the left hand if the teapot’shandleis on the left). 1995.) A similar proposal has been advanced for the nature of Second, it remains to be seen whether, in the long run, memory storage.Glenberg (1997) argues that the traditional a distributed approach can provide deep and satisfying in- approach to memory as “for memorizing” needs to be re- sights into the nature of cognition.If we recall that the goal placed by a view of memory as “the encoding of patterns of science is to find underlyingprinciples and regularities, of possible physical interaction with a three-dimensional rather than to explain specific events, then the facultative world” (p. 1). Glenberg seeks to explain a variety of mem- nature of distributedcognitionbecomes a problem.Whether ory phenomenain terms of such perceptuomotorpatterns. this problem can be overcome to arrive at theoretical in- Short-term memory, for example, is seen not as a distinct sights with explanatorypower is an issue that awaits proof. memory “system,” but as the deploymentof particularac- tion skills such as those involved in verbal rehearsal. Se- Claim 5: Cognition Is for Action mantic memory and the formation of concepts are simi- More broadly than the stringent criteria for situated larly explained in terms of embodied memory patterns, cognition, the embodied cognition approach leads us to differing from only in frequency of the consider cognitive mechanisms in terms of their function pattern’suse across many situations. in serving adaptive activity (see, e.g., Franklin, 1995, This approach to memory helps make sense of a vari- chap. 16). The claim that cognitionis for action has gained ety of observations,formal and informal, that we concep- momentum from work in perception and memory in par- tualize objects and situations in terms of their functional ticular. “Vision,”according to Churchland,Ramachandran, relevance to us, rather than neutrally or “as they really and Sejnowski (1994),“has itsevolutionaryrationalerooted are.” These observations range from laboratory experi- in improved motor control” (p. 25; see also Ballard, 1996; ments on encoding specificityand functional fixedness, to O’Regan, 1992; Pessoa, Thompson, & Noë, 1998). “Mem- the quip attributed to Maslow that when all you have is a 632 WILSON hammer everything looks like a nail, to the fanciful The answer being critiqued here is that the connectionsto Umwelt drawings of Uexküll (1934; reprints can be found action are quite direct: Individual percepts, concepts, and in Clark, 1997)showing what the environmentmight look memories are “for” (or are based on) particular action pat- like to creatures with different cognitive agendas. Our un- terns. The evidence discussed above, though,suggests that derstanding of the “how” system of vision suggests how this is unlikely to hold true across the board. An alterna- this type of embodied memory might work. As we have tive view is that cognition often subserves action via a seen from the work on priming of motor activity, the vi- more indirect,flexible,and sophisticatedstrategy, in which sual system can engage motor functionswithout resulting information about the nature of the external world is in immediate overt action. This is precisely the kind of stored for future use without strong commitments on what mechanism that would be needed to create the perceptuo- that future use might be. motor patterningthat Glenberg argues comprises the con- In support of this, we can note that our mental concepts tents of memory. often contain rich information about the properties of ob- The question we must ask, though, is how far this view jects, information that can be drawn on for a variety of of perception, memory, and cognition in general can take uses that almost certainly were not originallyencoded for. us. Can we dispense entirely with representation for rep- We are in fact capable of breaking out of functionalfixed- resentation’s sake, neutral with respect to a specific pur- ness, and do so regularly. Thus, I can notice a piano in an pose or action? We need not look far for evidence suggest- unfamiliar room, and being a nonmusician, I might think ingthat we cannot.Tobeginwith,althoughthe “how”system of it only as having a bench I can sit on and flat surfaces I of perceptual processing appears to be for action, the very can set my drink on. But I can also later call up my knowl- of the “what” system suggests that not all infor- edge of the piano in a variety of unforeseen circumstances: mation encodingworks this way.The ventralstream of vi- if I need to make a loud noise to get everyone’sattention; sual processing does not appear to have the same kinds of if the door needs to be barricaded against intruders; or if direct links to the motor system that the dorsal stream we are caught in a blizzard without power and need to does. Instead, the ventral stream goes about identifying smash up some furniture for fuel. Notice that these novel patterns and objects, apparently engaging in perception uses can be derived from a stored representation of the for perception’ssake. This pointis driven home if we con- piano.They need not be triggered by direct observationof sider some of the things that this system is asked to en- the piano and its affordances while one is entertaining a code. First, there are visual events, such as sunsets, that new action-based goal. are always perceived at a distance and do not offer any op- It is true that our mental representations are often portunity for physical interaction (cf. Slater, 1997). Sec- sketchy and incomplete, particularly for things that we ond, there are objects whose recognitiondepends on holis- have encountered only once and briefly. The literature on tic visual appearance, rather than on aspects of physical change blindness, which shows that people can entirely structure that offer opportunitiesfor perceptuomotorinter- miss major changes to a scene across very brief time lags, action. Human faces are the showcase example here, al- makes this point forcefully (see Simons & Levin, 1997, though the same point can be make for recognizing indi- for a review). But the fact that we are limited in how much viduals of other categories,such as dogs or houses. Third, we can attend to and absorb in a single brief encounter there is the case of reading, where sheer visual pattern does not alter the fact that we can and do build up robust recognition is paramount and opportunities for physical detailed representations with repeated exposure. Further- interaction with those patternsare virtuallynil. Thus, per- more, it is unclear that the sketchiness of a representation ceptualencodingcannot be accountedfor entirely in terms would prevent it from being a “representation for repre- of direct perception-for-actionprocessing channels. sentation’s sake.” Our mental representations, whether The problems get worse when we look beyond percep- novel and sketchy or familiar and detailed,appear to be to tual processing to some of the broader functions of mem- a large extent purpose-neutral, or at least to contain infor- ory. Mental concepts, for example, do not always or even mation beyond that needed for the originally conceived usually follow physical concrete properties that lend purpose. And this is arguably an adaptive cognitive strat- themselves to action, but instead often involve intangible egy. A creature that encodes the world using more or less properties based on folk-scientific theories or knowledge veridical mental models has an enormous advantage in of causal history(see, e.g., Keil, 1989; Putnam, 1970;Rips, problem-solving flexibility over a creature that encodes 1989). A classic example is that a mutilated dollar bill is purely in terms of presently foreseeable activities. still a dollar bill, but a counterfeit dollar bill is not. Simi- larly, cheddar cheese is understood to be a dairy product, Claim 6: Off-Line Cognition Is Body Based but soy milk, which more closely resembles milk in its Let us return now to the kinds of externalized cognitive perceptual qualities and action affordances, is not. activities described in Section 3, in which we manipulate In an ultimate sense, it must be true that cognitionis for the environment to help us think about a problem. Con- action. Adaptive behavior that promotes survival clearly sider the example of countingon one’sfingers. In its fullest must have driven the evolution of our cognitive architec- form, this can be a set of crisp and large movements, un- ture. The question, though, is the following: In what way ambiguously setting forth the different fingers as coun- or ways does our cognitive architecture subserve action? ters. But it can also be done more subtly, differentiating SIX VIEWS OF EMBODIED COGNITION 633 the positionsof the fingers only enough to allow the owner appears to be the only viable way to account for the large of the fingers to keep track. To the observer, this might body of data on working memory (Wilson, 2001a). Early look like mere twitching. Imagine, then, that we push the evidence for the sensorimotor nature of working memory activity inward still further, allowing only the priming of included effects of phonologicalsimilarity (worse memory motor programs but no overt movement. If this kind of for words that sound alike), word length (worse memory for mental activity can be employed successfully to assist a long words), and articulatorysuppression (worse memory task such as counting, a new vista of cognitive strategies when the relevant articulatorymuscles are kept busy with opens up. anotheractivitysuch as repeatinga nonsenseword). More Many centralized,allegedlyabstract cognitiveactivities recently, a similar set of effects, but in a different sensori- may in fact make use of sensorimotor functionsin exactly motor modality, has been found for working memory for this kind of covert way. Mental structures that originally sign language in deaf subjects: Performance drops when evolvedfor perceptionor action appear to be co-opted and to-be-remembered signs have similar hand shapes or are run “off-line,” decoupled from the physical inputs and temporally long, or when subjects are required to perform a outputsthat were their original purpose, to assist in think- repetitivemovement with their hands (Wilson & Emmorey, ing and knowing. (Several authors have proposed mecha- 1997, 1998). Furthermore, research on patient popula- nisms by which this decouplingmight take place: Dennett, tions and brain imaging of normals indicates the involve- 1995, chap. 13; Glenberg, 1997; Grush, 1996, 1998; Stein, ment of speech perceptionand speech production areas of 1994.) In general, the function of these sensorimotor re- the brain in working memory rehearsal (see Wilson, sources is to run a simulation of some aspect of the physi- 2001a, for a review). Thus, working memory appears to be cal world, as a means of representing information or draw- an example of a kind of symbolic off-loading, similar in ing inferences. spirit to that discussed in Section 3. However, instead of Although this off-line aspect of embodied cognitionhas off-loading all the way out into the environment, working generated less attention than situated cognition, evidence memory off-loads information onto perceptual and motor in its favor has been mounting quietly for many years. control systems in the brain. Sensorimotorsimulationsof external situationsare in fact Episodic memory. Long-term memory, too, is tied in widely implicated in human cognition. certain ways to our bodies’ with the world. Mental imagery. Imagery, includingnot only the well- The point is most obviousin the case of episodicmemory. studied case of visual imagery but also those of auditory Whether or not one posits a separate episodic memory sys- imagery (Reisberg, 1992) and kinesthetic imagery (Par- tem, episodic memories are a of memories defined by sons et al., 1995), is an obvious example of mentally sim- their content—they consistof records of spatiotemporally ulating external events. It is a commentary on the histori- localized events, as experienced by the rememberer. Phe- cal strength of the nonembodied viewpoint, then, that nomenologically, recalling an episodic memory has a during the 1980s the study of imagery was dominated by of “reliving,” with all the attendant visual, kines- a debate over whether images were in fact image-like in thetic, and spatial impressions. This is especially true any meaningful sense. An elaborate defense had to be when memories are fresh, before they have become crys- mounted to show that imagery involves analogue repre- tallized by retelling into something more resembling se- sentations that functionally preserve spatial and other mantic memories. properties of the external world, rather than consisting of . Implicit memory also appears to be bundlesof propositions(see Kosslyn, 1994, for a review). an embodied form of knowledge, consisting of a kind of Today, this issue has been firmly resolved in favor of the perceptual and/or procedural fluency (see, e.g., Cohen, analogue nature of images, and evidence continues to Eichenbaum, Deacedo, & Corkin, 1985; Johnston, Dark, mount for a close connection between imagery, which & Jacoby, 1985). Implicit memory is the means by which takes place in the absence of relevant external stimulation, we learn skills, automatizing what was formerly effortful. and the machinery of ordinary perception (see, e.g., Farah, Viewed in this light,implicitmemory can be seen as a way 1995; Kosslyn, Pascual-Leone, Felician, & Camposano, of taking off line some of the problems that confront the 1999). situated cognizer.I notedearlier that when humans are con- Working memory. A second example of simulating fronted with novel complex tasks under time pressure, the physical events through the off-line use of sensorimotor representational bottleneck comes into play and perfor- resources is short-term memory. Early models referred mance suffers. With practice, though, new skills become abstractly to “items” maintained temporarily in memory. automatized, reducing and circumventing Baddeley and Hitch (1974; Baddeley, 1986), however, the representationalbottleneck.(See Epelboim, 1997, for built a persuasive case for a multicomponent working evidence that automatizinga task reduces the need for off- memory system that had separate storage componentsfor loadingwork onto the environment.)In effect, prior expe- verbal and for visuospatial information, each of which rience allows whatever representations are necessary for was coded and maintained in something resembling its task performance to be built up before the fact. This strat- surface form. The particulars of the Baddeley model have egy involves exploitingpredictabilityin the task situation been challenged on a variety of grounds, but, as I have ar- being automatized—hence the fact that tasks with consis- gued elsewhere, some version of a sensorimotor model tent mapping between stimulus and response can be au- 634 WILSON tomatized,but tasks with varied mapping cannot (Schnei- Talmy, 2000; see Tomasello, 1998, for a review). Of par- der & Shiffrin, 1977). ticular interest for the present purpose, this linkage be- Viewing automaticityas a way of tackling the represen- tween syntax and rests in part on image schemas tational bottleneck ahead of time can help explain one of representing embodied knowledge of the physical world. the apparent paradoxes of automaticity. Traditionally, These image schemas make use of perceptual principles automaticprocessinghas been consideredthe polaropposite such as attentionalfocus and figure/ground segregationin of controlled processing (Schneider & Shiffrin, 1977; order to encode grammatical relations between items Shiffrin & Schneider,1977); yet highlyautomatizedtasks within the image . appear to allow greater opportunityfor fine-tuned control A second example is an embodied approach to explain- of action, as well as more robust and stable internal repre- ing mental concepts. We saw earlier that there are prob- sentations of the situation (cf. Uleman & Bargh, 1989). lems with trying to explain concepts as direct sensori- Compare, for example, a novice driver and an expert driver motor patterns. Nevertheless, it is possible that mental making a left turn, or a novice juggler and an expert jug- concepts may be built up out of cognitive primitives that gler trying to keep three balls in the air. In each case, the are themselves sensorimotor in nature. Along these lines, degree of control over the details of the behavior is quite Barsalou (1999b) has proposed that perceptual symbol poorfor the novice,and the phenomenologicalexperience systems are used to build up concepts out of simpler com- of the situation may be close to chaos. For the expert, in ponents that are symbolic and yet at the same time modal. contrast, there is a sense of leisure and clarity, as well as a For example, the chair, rather than comprising high degree of behavioral control. These aspects of auto- abstract, arbitrary, representationsof the componentsof a matic behavior become less mysterious if we consider the chair (back, legs, seat), may instead comprise modal rep- process of automatizingas one of buildingup internal rep- resentationsof each of these componentsand their mutual resentations of a situation that contains certain regulari- relations, preserving analoguepropertiesof the thing being ties, thus circumventing the representational bottleneck. represented. Whereas this example is quite concrete, the Reasoningand problem-solving. There is considerable inclusion of introspection as one of the modalities helps evidence that reasoning and problem-solvingmake heavy support the modal representation of concepts that we use of sensorimotor simulation. Mental models, partic- might think of as more abstract, such as (e.g., ularly spatial ones, generally improve problem-solving hungry) and mental activities (e.g., compare). relative to abstract approaches. A classic example is the A slightly different approach to abstract concepts is Buddhist monk problem: prove that a monk climbing a taken by Lakoff and Johnson and others, who argue that mountain from sunrise to sunset one day and descending mental conceptsare deeply metaphorical,based on a kind the next day must be at some particular point on the path of second-order modeling of the physical world and rely- at exactly the same time on both days. The problem be- ing on analogiesbetween abstract domainsand more con- comes trivial if one imagines the two days superimposed crete ones (e.g., Gibbs, Bogdanovich, Sykes, & Barr, on one another. One instantly “sees” that the ascending 1997; Lakoff & Johnson, 1980, 1999). As one example, monk and the descending monk must pass one another consider the concept communication. The internal struc- somewhere. Other examples of spatial models assisting ture of this concept is deeply parallel to our physical un- reasoning and problem-solving abound in undergraduate derstanding of how material can be transferred from one cognitive psychology textbooks. Furthermore, recent container to another. The parallels include metaphorical work by Glenberg and colleagues explores how the con- movement of thoughts across space from one person’s struction of mental models may occur routinely, outside head to another, metaphorical barriers preventing suc- the context of formal problem-solving, in tasks such as cessful transfer (as when someoneis being “thick-headed”), text comprehension (Glenberg & Robertson, 1999, 2000; and so on. According to this view, our mental representa- Kaschak & Glenberg, 2000; see also commentaries on tion of communication is grounded in our knowledge of Glenberg & Robertson, 1999: Barsalou, 1999a; Ohlsson, how the transfer of physical stuff works. Thus, even highly 1999; Zwaan, 1999). abstract mental concepts may be rooted, albeit in an indi- The domains of cognitionlisted above are all well estab- rect way, in sensory and motoric knowledge. lished and noncontroversial examples of off-line embodi- A third exampleis the role that motoric simulationmay ment. Collectively,they suggest that there are a wide vari- play in representing and understanding the behavior of ety of ways in which sensory and motoric resources may be conspecifics. Consider the special case of mentally simu- used for off-line cognitiveactivity.In accord with this, there lating something that is imitatible—that can be mapped are also a number of current areas of research exploringfur- isomorphically onto one’sown body. Such stimuli in fact ther ways in which off-line cognition may be embodied. primarily consistof our fellow humans.There are goodrea- For example, the field of cognitivelinguisticsis reexam- sons to believe that this isomorphism provides a special ining linguistic processing in terms of broader principles foothold for robust and noneffortful modeling of the be- of cognitive and sensorimotor processing. This approach, havior of other people (see Wilson, 2001b, for review). in radical contrast to the formal and abstract syntactic Given that we are a highly social species, the importance of structures of traditional theories, posits that syntax is such modeling for purposes of imitating, predicting, or un- deeply tied to semantics (e.g., Langacker, 1987, 1991; derstanding others’ behavior is potentially quite profound. SIX VIEWS OF EMBODIED COGNITION 635

We need not commit ourselves to all of these proposals Barsalou, L. W. (1999b). Perceptual symbol systems. Behavioral & in their present form in order to note that there is a general Brain Sciences, 22, 577-660. Beer, R. D. trend in progress. Areas of human cognition previously (1995). A dynamical systems perspective on agent– environment interaction. Artificial Intelligence, 72, 173-215. thoughtto be highly abstract now appear to be yielding to Beer, R.D. (2000). Dynamical approaches to cognitive science. Trends an embodied cognition approach. With such a range of in Cognitive Sciences, 4, 91-99. arenas where mental simulation of external events may Brooks, R. (1986). A robust layered control system for a mobile . play a role, it appears that off-line embodied cognition is Journal of Robotics & Automation, 2, 14-23. Brooks, R. (1991a). Intelligence without representation. Artificial In- a widespread phenomenon in the human mind. The time telligence Journal, 47, 139-160. may have come when we must consider these not as iso- Brooks, R. (1991b). New approaches to robotics. Science, 253, 1227- lated pieces of theoretical advancement, but as reflecting 1232. a very general underlying principle of cognition. Brooks, R. (1999). Cambrian intelligence: The early history of the new AI. Cambridge, MA: MIT Press. Chiel, H., & Beer, R. (1997). The brain has a body: Adaptive behavior Conclusions emerges from interactions of nervous system, body, and environment. Rather than continue to treat embodied cognition as a Trends in , 20, 553-557. single viewpoint,we need to treat the specific claims that Churchland,P.S., Ramachandran,V.S., & Sjenowski, T.J. (1994). have been advanced, each according to its own merits. A critique of pure vision. In C. Koch & J. L. Davis (Eds.), Large-scale One benefit of greater specificity is the ability to distin- neuronaltheories of the brain (pp. 23-60).Cambridge, MA: MIT Press. Clark,A. (1997). Being there: Putting brain, body, and world together guish on-line aspects of embodied cognitionfrom off-line again. Cambridge, MA: MIT Press. aspects. The former include the arenas of cognitive activ- Clark, A. (1998). Embodied, situated, and distributed cognition. In ity that are embedded in a task-relevant external situation, W. Bechtel & G. Graham (Eds.), A companion to cognitive science including cases that may involve time pressure and may (pp. 506-517). Malden, MA: Blackwell. Clark,A., & Grush,R. (1999). Towards a cognitiverobotics. Adaptive involve off-loading information or cognitive work onto Behavior, 7, 5-16. the environment. In these cases, the mind can be seen as Cohen, N. J., Eichenbaum, H., Deacedo, B. S., & Corkin, S. (1985). operating to serve the needs of a body interacting with a Different memory systems underlying acquisition of procedural and real-world situation. There is much to be learned about declarative knowledge. In D. S. Olton, E. Gamzu, & S. Corkin (Eds.), these traditionally neglected domains, but we should be Memory dysfunctions: An integration of animal and human research from preclinical and clinical perspectives (Annals of the New York cautious about claims that these principles can be scaled Academy of Sciences, Vol.444,pp. 54-71).New York: New YorkAcad- up to explain all of cognition. emy of Sciences. Off-line aspects of embodied cognition,in contrast, in- Craighero,L., Fadiga,L., Umiltà, C. A.,& Rizzolatti,G. (1996).Ev- clude any cognitiveactivitiesin which sensory and motor idence for visuomotor priming effect. NeuroReport, 8, 347-349. Dennett, D. (1995). Darwin’s dangerous idea. New York: Simon & resources are brought to bear on mental tasks whose ref- Schuster. erents are distant in time and space or are altogetherimag- de Waal, F. B. M. (2001). The ape and the sushi master: Cultural reflec- inary.These include symbolicoff-loading,where external tions by a primatologist. New York: Basic Books. resources are used to assist in the Epelboim, J. (1997). Deictic codes, embodiment of cognition, and the and manipulationof things that are not present, as well as real world. Behavioral & Brain Sciences, 20, 746. Farah, M. J. (1995). The neural bases of mental imagery. In M. S. Gaz- purely internal uses of sensorimotor representations, in zaniga (Ed.), The cognitive neurosciences (pp. 963-975).Cambridge, the form of mental simulations.In these cases, rather than MA: MIT Press. the mind operatingto serve the body,we find the body (or Fodor, J. A. (1983). The modularity of mind. Cambridge, MA: MIT its controlsystems) serving the mind. This takeoverby the Press. Franklin, S. (1995). Artificial . Cambridge, MA: MIT Press. mind, and the concomitant ability to mentally represent Gibbs, R. W., Bogdanovich, J. M., Sykes, J. R., & Barr, D. J. (1997). what is distant in time or space, may have been one of the Metaphor in idiom comprehension. Journal of Memory & Language, driving forces behind the runaway train of human intelli- 37, 141-154. gence that separated us from other hominids. Glenberg,A. M. (1997).What memory is for. Behavioral & Brain Sci- ences, 20, 1-55. REFERENCES Glenberg,A.M., & Robertson,D.A. (1999). Indexical understanding of instructions. Processes, 28, 1-26. Agre, P.E. (1993). The symbolic worldview: Reply to Vera and Simon. Glenberg,A.M., & Robertson,D.A. (2000). Symbol grounding and Cognitive Science, 17, 61-69. meaning: A comparison of high-dimensional and embodied theories Baddeley, A. (1986). Working memory. Oxford: Oxford University of meaning. Journal of Memory & Language, 43, 379-401. Press. Goodale,M. A., & Milner,A.D. (1992). Separate visual pathways for Baddeley, A., & Hitch, G. (1974). Working memory. In G. H. Bower perception and action. Trends in Neurosciences, 15, 20-25. (Ed.), The psychology of learning and motivation (Vol. 18, pp. 647- Goodwin, C. J. (1999). A history of modern psychology. New York: 667). Hillsdale, NJ: Erlbaum. Wiley. Ballard, D. H. (1996). On the function of visual representation. In Grafton, S. T., Fadiga, L., Arbib, M. A., & Rizzolatti, G. (1997). K. A. Akins (Ed.), Perception (pp. 111-131). Oxford: Oxford Uni- Premotor cortex activation during observation and naming of famil- versity Press. iar tools. NeuroImage, 6, 231-236. Ballard,D. H., Hayhoe, M. M., Pook, P.K., & Rao, R. P.N. (1997). Greeno, J. G., & Moore, J. L. (1993). Situativity and symbols: Re- Deictic codes for the embodiment of cognition. Behavioral & Brain sponse to Vera and Simon. Cognitive Science, 17, 49-59. Sciences, 20, 723-767. Grush, R. (1996). Emulation and cognition. Unpublished doctoral dis- Barsalou, L. W. (1999a). Language comprehension: Archival memory sertation, University of California, San Diego. or preparation for situated action? Discourse Processes, 28, 61-80. Grush, R. (1997). Yet another design for a brain? Review of Port and 636 WILSON

van Gelder (Eds.), Mind as . Philosophical Psychology, 10, Quinn, R., & Espenschied, K. (1993). Controlof a hexapod robot using 233-242. a biologically inspired neural network. In R. Beer, R. Ritzman, & Grush, R. (1998). Perception, imagery, and the sensorimotor loop. T. McKenna (Eds.), Biological neural networks in invertebrate neuro- www.pitt.edu/~grush/papers/%21papers.html. English translation of: and robotics (pp. 365-381). San Diego: Academic Press. Wahrnehmung, Vorstellung und die sensomotorische Schleife. In Reisberg, D. (Ed.) (1992). Auditory imagery. Hillsdale, NJ: Erlbaum. F. Esken & H.-D. Heckmann (Eds.), Bewußtsein und Repräsentation. Rips, L. (1989).Similarity, typicality,and . In S. Vosniadou Paderborn, Germany: Ferdinand Schöningh. & A. Ortony (Eds.), Similarity and analogicalreasoning (pp. 21-59). Hutchins, E. (1995).Cognitionin the wild. Cambridge, MA: MIT Press. Cambridge: Cambridge University Press. Iverson, J. M., & Goldin-Meadow, S. (1998). Why people gesture Saito, F., & Fukuda, T. (1994). Two link robot brachiation with connec- when they speak. Nature, 396, 228. tionist Q-learning. In D. Cliff (Ed.), From animalsto animats3 (pp.309- Jeannerod,M. (1997).The cognitiveneuroscience of action. Cambridge, 314). Cambridge, MA: MIT Press. MA: Blackwell. Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic Johnston, W. A., Dark, V. J., & Jacoby, L. L. (1985). Perceptual flu- human : I. Detection, search, and attention. ency and recognition judgments. Journal of Experimental Psychol- Psychological Review, 84, 1-66. ogy: Learning, Memory, & Cognition, 11, 3-11. Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic Juarrero, A. (1999). Dynamics in action: Intentional behavior as a human information processing: II. Perceptual learning, automatic at- complex system. Cambridge, MA: MIT Press. tending and a general theory. Psychological Review, 84, 127-190. Kaschak, M. P., & Glenberg, A. M. (2000). Constructing meaning: Simons,D.J., & Levin, D.T. (1997). Change blindness. Trends in Cog- The role of affordances and grammatical constructions in sentence nitive Sciences, 1, 261-267. comprehension. Journal of Memory & Language, 43, 508-529. Slater, C. (1997). Conceptualizing a sunset ¹ using a sunset as a dis- Keil, F. C. (1989). Concepts, kinds, and cognitive development. Cam- criminative stimulus. Behavioral & Brain Sciences, 20, 37-38. bridge, MA: MIT Press. Steels,L., & Brooks,R. (1995). The artificial life route to artificial in- Kirsh, D., & Maglio,P. (1994). On distinguishingepistemic from prag- telligence: Building embodied, situated agents. Hillsdale, NJ: Erl- matic action. Cognitive Science, 18, 513-549. baum. Kossly n, S. M. (1994). Image and brain: The resolution of the imagery Stein, L. (1994). Imagination and situated cognition. Journal of Exper- debate. Cambridge, MA: MIT Press. imental Theoretical Artificial Intelligence, 6, 393-407. Kossly n, S. M., Pascual-Leone, A., Felician, O., & Camposano, S. Talmy, L. (2000). Toward a cognitive semantics: Vol. I. Conceptual (1999). The role of area 17 in visual imagery: Convergent evidence structuring systems. Cambridge, MA: MIT Press. from PET and rTMS. Science, 284, 167-170. Thelen,E., & Smith, L. B. (1994). A dynamic systems approach to the Krauss, R. M. (1998). Why do we gesture when we speak? Current Di- development of cognition and action. Cambridge, MA: MIT Press. rections in Psychological Science, 7, 54-60. Tomasello,M. (1998).Cognitivelinguistics.In W.Bechtel & G. Graham Lakoff, G.,& Johnson, M. (1980).Metaphorswe live by. Chicago: Uni- (Eds.), A companion to cognitive science (pp. 477-487).Malden, MA: versity of Chicago Press. Blackwell. Lakoff, G., & Johnson, M. (1999). Philosophy in the flesh: The em- Tucker, M., & Ellis, R. (1998). On the relations between seen objects bodied mind and its challenge to western thought. New York: Basic and components of potential actions. Journal of Experimental Psy- Books. chology: Human Perception & Performance, 24, 830-846. Langacker,R. (1987,1991).Foundationsof cognitive grammar (2 vols.). Uexküll, J. von (1934). A stroll through the worlds of animals and men. Stanford: Stanford University Press. In C. H. Schiller (Ed.), Instinctive behavior: The development of mod- Leakey, R. (1994). The origin of humankind. New York: Basic Books. ern concept (pp. 5-80). New York: International Universities Press. Markman, A.B., & Dietrich, E. (2000). In defense of representation. Uleman, J., & Bargh, J. (Eds.) (1989). Unintended thought. New York: Cognitive Psychology, 40, 138-171. Guilford. Mataric, M. (1991). Navigating with a rat brain: A neurobiologically van Gelder, T., & Port, R. (1995). It’s about time: An overview of the inspired model for robot spatial representation. In J.-A. Meyer & dynamical approach to cognition. In R. Port & T. van Gelder (Eds.), S. Wilson (Eds.), From animals to animats (pp.169-175).Cambridge, Mind as motion: Explorationsin the dynamics of cognition(pp. 1-43). MA: MIT Press. Cambridge, MA: MIT Press. Murata, A., Fadiga, L., Fogassi, L., Gallese, V.,Raos, V.,& Rizzo- Vera, A. H., & Simon, H. A. (1993). Situated action: A symbolic inter- latti, G. (1997). Object representation in the ventral premotor cortex pretation. Cognitive Science, 17, 7-48. (area F5) of the monkey. Journal of Neurophysiology, 78, 2226-2230. Wertsch, J. V. (1998). Mediated action. In W. Bechtel & G. Graham Ohlsson, S. (1999). Anchoringlanguage in : Observations on ref- (Eds.), A companion to cognitive science (pp. 518-525).Malden, MA: erence and representation. Discourse Processes, 28, 93-105. Blackwell. O’Regan, J. K. (1992). Solving the “real” mysteries of visual percep- Wiles, J., & Dartnall, T. (1999). Perspectives on cognitive science: tion: The world as an outside memory. Canadian Journal of Psychol- Theories, experiments, and foundations. Stamford, CT: Ablex. ogy, 46, 461-488. Wilson,M. (2001a).The case for sensorimotor coding in workingmem- Parsons, L. M., Fox, P. T., Downs, J. H., Glass, T., Hirsch, T. B., ory. Psychonomic Bulletin & Review, 8, 44-57. Martin, C. C., Jerabek, P. A., & Lancaster, J. L. (1995). Use of Wilson, M. (2001b). Perceiving imitatible stimuli: Consequences of implicit motor imagery for visual shape discrimination as revealed by isomorphism between input and output. Psychological Bulletin, 127, PET. Nature, 375, 54-58. 543-553. Pessoa, L., Thompson, E., & Noë, A. (1998). Finding out about filling- Wilson, M., & Emmorey, K. (1997). A visuospatial “phonological in: A guide to perceptual completion for visual science and the phi- loop” in working memory: Evidence from American Sign Language. losophy of perception. Behavioral & Brain Sciences, 21, 723-802. Memory & Cognition, 25, 313-320. Pfeifer, R., & Scheier, C. (1999). Understanding intelligence. Cam- Wilson, M., & Emmorey, K. (1998). A “word length effect” for sign bridge, MA: MIT Press. language: Further evidence for the role of language in structuring Port, R.F., & van Gelder, T. (1995). Mind as motion: Explorations in working memory. Memory & Cognition, 26, 584-590. the dynamics of cognition. Cambridge, MA: MIT Press. Zwaa n, R. A. (1999).Embodied cognition,perceptual symbols, and sit- Prinz, W. (1987). Ideo-motor action. In H. Heuer & A. F. Sanders uation models. Discourse Processes, 28, 81-88. (Eds.), Perspectives on perception and action (pp. 47-76). Hillsdale, NJ: Erlbaum. Putnam, H. (1970).Is semantics possible? In H. E. Kiefer & M. K. Munitz (Eds.), Language,belief and (pp.50-63).Albany,NY: State (Manuscript received September 12, 2000; University of New York Press. revision accepted for publication October 5, 2001.)