Language, Anthropology and Cognitive Science Author(s): Maurice Bloch Source: Man, New Series, Vol. 26, No.

2 (Jun., 1991), pp. 183-198 Published by: Royal Anthropological Institute of Great Britain and Ireland Stable URL: http://www.jstor.org/stable/2803828 Accessed: 04/08/2009 13:52
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=rai. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

Royal Anthropological Institute of Great Britain and Ireland is collaborating with JSTOR to digitize, preserve and extend access to Man.

http://www.jstor.org

LANGUAGE, ANTHROPOLOGY AND COGNITIVE SCIENCE

MAURICE BLOCH

& PoliticalScience LondonSchoolof Economics

It is argued in this article that anthropology has relied upon a psychologically mnisleading and overly linguistic model of culture. Psychological studies show that concepts are principallyformned independently of language. Moreover, studies of expertise show that in order to become an expert at a familiartask or a set of tasks a person needs to organise his or her knowledge in a way which is not language-like. The article drawson the theory of connectionism to suggest that most cultural knowledge cannot therefore be organised in the sentential logical form characteristicof language. It is suggested that the traditional anthropological method of participantobservation enables the cognitive scientist to understand cultural knowledge without the dangerous intermediary of language.

Cognitive science is usuallydescribedas the attemptto bring cognitive psychology, philosophy, neurophysiology, artificialintelligence, linguistics and anthropology together in order to understandcognition. In this alliance anthropology is, in fact, a rathershadowypartner.Only cognitive anthropologyis usuallytakeninto account by cognitive scientists and, even then, only the fiirther reaches of cognitive anthropology which many anthropologists, especially European anthropologists, would fail to recognise as their business at all. This state of affairsis unfortunate, because some of the theories emerging in cognitive science are central to the concerns of anthropology, whether social or cultural,and should lead anthropologiststo re-examine many ofthe premissesoftheir work. Other cognitive scientists, however, have much to learn from the central concerns of anthropology and particularly,if ratherunexpectedly, from the traditionalmethod of social anthropology: participantobservation. Culturalanthropologistsstudy culture. This can be defined as that which needs to be known in order to operate reasonably effectively in a specific human environment.' Social anthropologists traditionallystudy social organisation and the behaviour by means of which people relate to each other. Both cultural and social anthropologists, however, are well aware that the distinction between the two branchesof the discipline is not absolute. Culturalanthropologistsknow that they cannot get at culture directly, but only through the observation of communicative activity, verbal or otherwise, natural or artificiallysimulated. Social anthropologistsare aware that they cannot understandaction, verbal or otherwise, if they do not construct, probably in imagination, a representationof the culture
*

FrazerLecture, 1990 Man (N.S.) 26, 183-198

184

MAURICE BLOCH

of the people they study, since this is the only way to make sense of their activities (Winch 1958). Some concept of culture is therefore essential to all social and culturalanthropologists. However, a further assumption of anthropology, sometimes stated and sometimes unstated, is that this culture is inseparablylinked to language, on the grounds either that culture is thought and transmittedas a text through language, or that cultureis ultimately 'languagelike', consisting of linked linearpropositions. It is these two assumptionsabout culture that I want to challenge here. If culture is the whole or a part of what people must know in a particularsocial environment in order to operate efficiently, it follows first, that people must have acquired this knowledge, either through the development of innate potentials, or from external sources, or from a combination of both, and secondly that this acquiredknowledge is being continually stored in a mannerthat makesit relatively easily accessiblewhen necessary.2These obvious inferences have in turn a further implication which is that anthropologists'concerns place them right in the middle of the cognitive sciences, whether they like it or not, since it is cognitive scientists who have something to say about learning, memory and retrieval.Anthropologists cannot, therefore, avoid the attempt to make their theories about social life with what other cognitive scientists have to say about the processes of compatible learning and storage. Perhapsall this may seem commonplace but it is strikinghow often anthropologists' theories of learning, memory and retrievalhave not been compatible with those of other cognitive scientists. The exception is the small group of cognitive anthropologists, largely confined to the United States,3 who have paid serious attention to recent developments in cognitive science. Things are nevertheless beginning to change, and I hope that in this article I am swimming with the tide, but a tide which has only just startedto flow and which has not reached, especially in Europe, the heartland of anthropological theory where it is perhaps most needed.4 Of course I do not claim that other cognitive scientists have figured out how the mind works, and that anthropologists have only to slot culture into this well-advanced model. Cognitive scientists' understanding of the mind-brain is dramaticallyincomplete and tentative. Nonetheless, some findings are fairly clear and we should take these into account. Moreover, the hypotheses of cognitive scientists, however speculative, fundamentally challenge many unexamined anthropological assumptionsin a way that should not be ignored. II As noted above, my main concern is with only one aspect of the broad topic of the relation between cognitive science and anthropology, namely the importance or otherwise of language, or language-like phenomena, for cultural knowledge. A good place to start is with a consideration of concept formation, particularly classificatory concept formation, a topic about which much has been written, and which is often recognised as of relevance to anthropology. In discussingthis topic I am not concerned with the traditionalanthropological issue, recently reviewed by Atran, of the extent to which classificatoryconcepts have an innate basis (Atran 1990), though it is quite clear that in the past,

MAURICE BLOCH

185

anthropologistshave grosslyexaggeratedculturalvariation,and that the traditional questions of cultural anthropologists concerning very broad areas of knowledge should be rephrasedfrom 'How are these things learned?' to 'How is culturally specific knowledge produced out of universal predispositions?'.Such rephrasing is too easily obscured in anthropological references to 'culturalmodels' (Holland & Quinn 1987). Leaving this fundamentalquestion aside, it is nonetheless clear that all classificatory concepts are at least partially learned, and recent work in this area has brought about certain fundamental changes in the way we envisage the process. The old idea that the child learns classificatory concepts as minimal and necessary definitions, an idea taken for grantedin most of anthropology and which was more particularly implied in structuralism and ethnoscience, was shown to be untenable some time ago (Fillmore1975; Rosch 1977; 1978; Smith 1988). The more generally accepted position now is that such concepts are formed through reference back to rathervague andprovisional'prototypes'which anchorloosely-formed 'families' of specific instances. For example, the concept of a house is not a list of essential features(roof, door, walls, and so on) which have to be checked offbefore deciding whether or not the thing is a house. If that were so we would have no idea that a house which haslost its roof is still a house. It is ratherthatwe considersomething as 'a house' by comparing it to a loosely associatedgroup of 'houselike' features, no one of which is essential, but which are linked by a general idea of what a typical house is. Thus, as suggested by Markham and Seibert (1976), classificatoryconcepts are in fact based on an appraisalof their referentsin the world, on how we think of the construction and make-up of these referents, or on our understandingof the way they are constituted. It therefore seems that the mental form of classificatory concepts, essential building blocks of culture, involves loose and implicit practical-cum-theoretical pattern networks of knowledge, based on the experience of physical instances sometimes called 'best exemplars' (Smith et al. 1988: 372). A significant aspect of looking at classificatoryconcepts in this way is that it makes them isomorphic with what are known as 'scripts'and 'schemata',although these latter may be on a much larger and more elaborate scale. These 'scripts' and 'schemata'are,in effect, chunked networks ofloose proceduresand understandings which enable us to deal with standard and recurringsituations,for example 'getting the breakfastready', that are clearly culturallycreated (Abelson 1981; Holland & Quinn 1987; D'Andrade 1990).5 If classificatoryconcepts such as 'scripts'and 'schemata'are not like dictionary entries, but are instead small networks of typical understandingsand practices concerning the world, then the question of their relation to words becomes more problematicthanit was with the old 'checklistofnecessaryandsufficientconditions' view. That there is no inevitable connexion between concepts and words is shown by the now well-established fact that concepts can and do exist independently of language. This is made clear in the many examples of conceptual thinking in pre-linguisticchildren,firstpresentedby Brown (1973). Children have the concept 'house' before they can say the word. We also have studies which show that the acquisition of lexical semantics by children is very largely a matter of trying to match words to alreadyformed concepts. This is the so called 'concept first'theory.

186

MAURICE BLOCH

Contrary to earlier views in cognitive anthropology (Tyler 1969), therefore, language is not essentialfor conceptual thought.6 It is possible to go beyond this initial distancing between the lexicon and mental concepts, however, thanks to work on semanticacquisitionby Bowerman (1977). This demonstratesa continual back and forth movement between aspects of classificationwhich are introduced through languageand mental concepts, as the child learnsto expressthese concepts through words. This dialecticalmovement is not only interesting in itself but also suggestsa much more generalprocess, to which I shallreturn, by which originally non-linguistic knowledge is partly transformedas it becomes linguistic, thereby among taking on a form which much more closely resembles what structuralists, other anthropologists,had assumed to characterisethe organisation of all human knowledge (Keil & Batterman 1984). This brief review of concept formation enables us to reach the following provisional conclusions: (1) that much of knowledge is fundamentally non-linguistic; (2) that concepts involve implicit networks of meanings which are formed through the experience of, and practice in, the externalworld; and (3) that, under certaincircumstances,this non-linguistic knowledge can be renderedinto language and thus take the form of explicit discourse, but changing its characterin the process. III Another area of joint concern to anthropology and cognitive psychology also reveals the importance of non-linguistic knowledge. This is the study of the way we learn practical,everyday tasks. It is clear that we do not usually go through a point-by-point explanation of the process when we teach our children how to negotiate their way aroundthe house or to close doors. Much culturallytransmitted knowledge seems to be passed on in ways unknown to us. Perhaps in highly schooled societies this fact is misleadinglyobscured by the prominence of explicit instruction, but in non-industrialisedsocieties most of what takes people's time and energy - including such practicesas how to wash both the body and clothes, how to cook, how to cultivate, etc. - are learned very graduallythrough imitation and tentative participation. The culturalspecificity, complexity and embeddedness of such tasks,and their character as not linguistically explicit, have often been commented upon by anthropologists,for example by Mauss (1936), Leroi-Gourhan (1943) and Haudricourt (1968), but have rarelybeen studied satisfactorily. The few studies we do have tend to deal with unusual tasksnecessaryfor specialisedcraftswhich require formalapprenticeships. In these cases,also,anthropologistshave noted thatlanguage seems to play a surprisinglysmall role in the transmission of knowledge. For example, in her study of weavers in Ghana, Goody (1978) was amazedat the small part played by questioning and speaking in teaching apprentices.Similarly,Lave, in her study of Liberiantailors,notes that what she calls 'apprenticeshiplearning', which relies on the 'assumptionsthat knowing, thinking, and understandingare generatedinpractice' (Lave1988; 1990:310), is more effective thanformalteaching based on linguistic, socratic forms. If this is so for these relatively specialisedtasks there is no doubt that the same conclusions would be reached even more

MAURICE BLOCH

187

emphatically in studies of learning more common, though not necessarily less skilled, everyday tasks. The significance of such findings is much more important than we might at first suppose. This is because the fact that the transmissionof knowledge in West Africanweaving or tailoringis largely non-linguistic may have less to do with the culture of education in these places than with a general feature of the kind of knowledge thatunderliesthe performanceof complex practicaltasks,which requires that it be non-linguistic. That this is so is suggested by various studies of learning in which, by contrast to the examplesjustmentioned, the originalteaching is received through language, or at least in propositional form, but in which the process of becoming an expert seems to involve the transformationof the propositions of the teacher into fundamentally non-linguistic knowledge (Dreyfus & Dreyfius 1986: ch. 1).7 Thus Anderson (1983) points out how people who are taught driving through a series of propositions have to transformthis knowledge into non-linguistic, integrated proceduresbefore the task can be effected rapidly, efficiently and automaticallyone might say properly. Only when they do not think about what they are doing in words are drivers truly experts.8 Probably some teaching needs to be done verbally,but there arealso advantages in the non-linguistic transmission ofpractical skillstypicalof non-industrialsocieties since such transmission by-passesthe double transformationfrom implicit to linguistically explicit knowledge made by the teacherand from linguisticallyexplicit to implicit knowledge made by the learner. It could be objected thatmy stresson the non-linguistic side ofpracticalactivities is somewhat exaggerated. After all, language also plays a role in the performance of many familiarpracticalactions, though not necessarilydriving. Even this fact, however, may bearless on the extent to which knowledge is linguistic than might appearat first sight. Let me take an example which is in part derived from the semantic work of Johnson-Laird (1988: ch. 18; see alsoJohnson-Laird1983: 396-447), modified by Sperber and Wilson's (1986) theory of relevance and which is also indirectly inspired by Malinowski's (1935) study of the role of language in Trobriand agriculture. Imagine a Malagasyshifting cultivator with a fairly clear, yet supple mental model, perhapswe could say a script,stored in long-term memory, ofwhat a 'good swidden' is like; and that this model, like that of a roundaboutin the mind of the expert driver, is partly visual, partly analytical(though not necessarilyin a sentential logical way), partly welded to a series of procedures about what you should do to make and maintain a swidden. This Malagasyis going through the forest with a friend who says to him 'Look over there at that bit of forest, that would make a good swidden'. What happens then is that, after a rapid conceptualisationof the bit of forest, the model of 'the good swidden' is mentally matched with the conceptualised area of forest, and then a new but related model, 'this particularplace as a potential swidden' is established and stored in long-term memory. This stored mental model, however, although partly created by the linguistic event and understood in terms of the relevance of the statement to both the mental model of the 'good swidden' and to that of the area of forest, is not likely to include the linguistic statement tagged onto it. The intrusion of language has therefore not made the mental model any more linguistic.

188

MAURICE BLOCH

To return to the example of car driving: we have seen that driving expertly seems to require that the information be stored non-linguistically if it is to be accessible in an efficient way. Why should this be so? In order to begin to answer this question we need to turn again to the process involved in becoming an expert. It is not surprisingthat practice in performing a complex task makes the practitionermore efficient, but studies of expertise show that the increase in efficiency is more puzzling than might at first appear. For example, when people are repeatedly asked to read a page of text upside down they graduallydo this fasterand faster,but the increasein speed is not continuous, nor does it go on for ever. At first there is a rapid increase in efficiency which continues for a while, then it begins to slow down, until eventually there is no furtherincrease. The shape of the curve of increasedefficiency suggests JohnsonLaird 1988: 170, citing Newell & Rosenbloom 1981) that the process of learning involves the construction of a cognitive apparatus dedicated to cope with this sort of task. The establishmentof that apparatus is slow, and while it is in construction there is significant improvement; however once it has been set up no further improvement becomes possible. A chunk or apparatus concerned with a familiar activity has thus come into existence in the brain as a result of repeated practice (Simon 1979: 386 sqq.). A more complex and much discussedexample of what happenswhen someone becomes an expert comes from studies of master chess players. It has been convincingly argued that expert chess playersdo not differfrom novices (who are not complete beginners) in knowing the rules of chess or in performing such motor tasks as moving one piece without knocking the others down. What seems to distinguishthe expert from the novice is not so much an ability to handle complex strategic logico-mathematical rules, but rather the possession, in memory, of an amazingly comprehensive and organised store of total or partialchessboardconfigurations, which allows the expert to recognise the situation in an instantso as to know what should be done next (Dreyfus & Dreyfus 1986: 32-5). However, bearingin mind the example of driving, what is surely happeningis that the expert is not just remembering many games but that she has developed through long practice a specific apparatuswhich enablesher to remember many games and configurations much more easily and quickly than the non-expert. She has learned how to learn this kind of information. This would explain how the expert can cope, not only with situationswhich she recognises, but also with situationswhich are new, so long as they fall within the domain which she has learned to cope with efficiently. Learning to become an expert would therefore be a matter not simply of rememberingmany instances,but of constructinga dedicatedcognitive mechanism for dealing with instances of a particularkind. Such a mechanism, because it is concerned only with a specific domain of activity, can cope with information relatingto that domain of activity remarkablyquickly and efficiently, whether this be information about chess-piece configurations, motorway scenarios, or potentially useful areasof forest, and even though the specific cases of chess, motorway or forest have not been previously encountered in exactly that way (Dreyfus & Dreyfus 1986: ch. 1).

MAURICE BLOCH

189

If becoming an expert does involve the creation of apparatusesdedicated to handle families of related tasks,then this is surely something which an anthropologist must bear in mind. For what she studies is precisely people coping with familiaryet ever novel situations (Hutchins 1980). It seems reasonableto assume for dealing expertly with certain that the construction of dedicated apparatuses areas of activity is going on in the process of cultural learning of all common practicaltasks.Indeed some recent work suggeststhatlearningto become an expert in familiarareasis a necessarypreliminaryto other types of learning and to being able to cope with the less familiar and less predictable. The reason seems to be partly neurological. In the case of car driving, it seems that as a person becomes an expert, not only does she dnrve better, not only does she transform what was once linguistic propositionalinformationinto something else, she also seems to employ much less neurological potential in doing the necessary tasks (Schneider & Shiffrin 1977), therebyfreeing her for other mental tasks,such as talkingon a carphone. Similarly, the extraordinaryfeats of memory of the chess master seem to be made possible by the efficient packing of information through the use of the expert apparatus for coping with novel situations of play. Such observationssuggest the general conclusion that the ability to learn more is largely a matter of organisingwhat one has alreadylearned in packed chunks so that one has room for the new. Some cognitive scientists have therefore argued that the problemsyoung children have in doing the taskswhich, as Piaget showed, only adults can do, stem not from the immaturity of the children's brains, but from the fact that this 'packing' has not yet taken place. Once the essential preliminaryprocedureshave been sufficientlyorganised,their implementation will only requirelimited neurological capacity,leaving enough 'room' for the child to performfurthersupplementarytasks.These arethe taskswhich the child had earlier seemed unable to perform;but reallythe problem was that they could not normally be performedsimultaneouslywith their necessarypreliminaries(Smith et al. 1988). There is therefore considerable evidence that learning is not just a matter of storing received knowledge, as most anthropologistsimplicitly assumewhen they but that it is a matterof constructing equate culturaland individualrepresentations, for the efficient handlingand packing ofspecific domains of knowledge apparatuses and practice. Furthermore,as suggested by the case of learning to drive, evidence shows that once these apparatuses are constructed, the operations connected with these specific domains not only arenon-linguistic but also mustbe non-linguistic if they are to be efficient.9 It follows that much of the knowledge which anthropologists study necessarilyexists in people's heads in a non-linguistic form. Before proceeding further, an ambiguity in what has been argued so far must be removed. To say that knowledge concerned with the familiarmust be nonlinguistic could mean one of two things. It could mean simply that this knowledge is not formulatedin naturallanguage. On the other hand it could mean something much stronger. It could mean that this knowledge is in no way 'language like', sententiallogic of naturaland computer thatit is not governed by the characteristic languages.10Here I adopt the stronger of the two alternativesbecause I believe that the studies on expertise discussedabove suggest that the knowledge organised for efficiency in day-to-day practice is not only non-linguistic, but also not

190

MAURICE BLOCH

language-like in that it does not take a sentential logical form. To argue this and to make my argument less negative I now turn to the admittedly controversial assumptionsof what has been called 'connectionism'. IV What is particularly interestingfor anthropology in connectionism is not so much connectionism itself, but the reasonswhy a theory like it is necessary.Simply put, a theory such as connectionism is necessarybecause a sentential linear model of the mind-brain, sometimes called the sentence-logic model (Churchland & Sejnowski 1989), which is broadlysimilarin form to the semanticsof naturallanguage, cannot account for the speed and efficiency with which we perform daily tasks and cope with familiarsituations. The sentence-logic or sentential linear model is intuitively attractive for a number of reasons, and these explain why it is implicit in anthropological theory and was accepted unchallenged for a long time in cognitive psychology. First of all, and probably most significantly,it is themodel of folk psychology, as anthropologists have nicely demonstrated(D'Andrade 1987). Sententiallogical forms are how we think we think. Secondly, sentential logical models work well for the semantics of naturallanguage; and thirdly, they worked well for those metaphors for the mind - the von Neumann or digital computers. However, these three argumentsin favour of the sentence-logic model arevery weak. First, folk psychology, whatever the majority of anthropologistsmay say, need not, as Churchland and Churchland (1983) point out, be any more valid thanfolk physicsasa basisfor scientific accounts. Secondly, what appliesto language need not apply to other forms of mental activity. Thirdly, digital computers come nowhere near to doing the jobs we humans can do, and so they must be, in some way, inappropriateas metaphorsfor the mind-brain. The case for connectionism is best made by considering an example, and by using this type of example we can see the relevance of the theory for anthropology. Remember the Malagasypeasant. When the man said to his friend, 'look over there ... that piece of forest would make a good swidden', an unbelievable mental feat seems to have then been achieved by the man addressed.He recalled from long-term memory the complex yet highly flexible mental model or schema 'the good swidden', then he conceptualised the piece of forest indicated, taking in information about the vegetation, the slope, the surrounding countryside, the hydrology, the soil, etc., then he matched the two intricate conceptualisationsin what could not be just a simple comparison but a highly complex set of transformations. When put in this way, the total taskseems Herculean, but in realityeven the moderately talented Malagasyfarmerwould come to some assessmentin just a few seconds. Furthermore,this computationalfeat is no more difficultthan many other similartaskswhich human beings perform all the time. Why then does such a taskappearso impossibly complicated when we think about what it entails?The reason is that we are explaining the behaviour of the Malagasyfarmerin terms of our own folk psychology, including a model of language-like semantics. This makes an easy task, which we know the farmercan perform in an instant or two, seem absolutely awesome. There must, indeed, be something wrong about how we think we think.

MAURICE BLOCH

191

Connectionism is an alternativetheory of thought which makes such commonplace feats as that of our Malagasyfarmerpossible to envisage. It suggests that we go about the whole process of thought in a quite differentway from what we had previously and loosely assumed. The problem with the folk way of describing thought procedures, for example of how a decision is reached, is that we tend to see the activity as a serial process of analysiscarried on along a single line by a singleprocessor.Forcomplex yet familiartaskssuch processingwould be impossibly clumsy and lengthy. Instead, connectionism suggests that we access knowledge, either from memory or as it is conceptualised from perception of the external world, through a number of processing units which work in paralleland feed in information simultaneously. It suggests, too, that the information received from these multiple parallel processorsis analysedsimultaneouslythrough alreadyexisting networks connecting the processors. Only with this multiple parallelprocessing could complex understandingsand operations, like those about the swidden, be achieved asfastasthey are. Otherwise, given the conduction velocities and synaptic delays in neurons, it is a physical-cum-biological impossibility for the number of steps required by a logical-sentential model of the mind-brain to be carried out in the time within which even the simplestmental tasksare ordinarilyperformed. A connectionist brain, on the other hand, could (at least hypothetically) work sufficiently fast (Feldman & Ballard 1982). It is much too early to say whether connectionism will prove to be an accurate analysisof the working of the brain and, in any case, I am not in a position to be able to evaluate its neurological validity. What mattershere, above all, is that the theory offersthe kind of challenge to sententiallogical models which anthropology requires, and it offers the kind of answerswhich would cope with the situations we seek to understand. Support for connectionism does not, however, all come from first principles. Some psychological experimentalwork seems to confirm the theory (Rumelhart et al. 1986). The fact that computer programswhich enable digital computers to work on something like parallelprocessing seem to be able to get them to do tasks which ordinarydigital computers working with classicalprogramsare unable to do, such as reading and reproducing in three dimensions grey shaded images, is also encouraging. There are, however, two other aspects of the theory which make it particularlyattractiveto the anthropologist. The first is that connectionism can cope well with whatJohnson-Laird (1983: 438-47) called the 'provisional'characterof mental models, while sententiallogical models imply a much greaterrigiditywhich is quite unlike what we find in natural situations. The mental model of 'the good swidden' cannot be a checklist of characteristics to be found in a particularconfiguration or even an example of the kind of 'fuzzy' digital models recently proposed. With such fundamentallyfixed models the Malagasyfarmerwould never find the right plot. The model cannot require absolutely any particular characteristicor configuration;just a general flexible theoretical-practicalhypothesis. Connectionism can handle this type of 'fairly loose' practical-theoreticalthinking, which as we saw is also implied in prototype theory, while other theories cannot. Secondly, a connectionist model can account for the length of the process of becoming an expert at a particulartask, a fact which itself is quite a puzzle. With

192

MAURICE BLOCH

such a model we can understand what would be happening when a person is learning to handle a family of related tasks, such as learning how to learn chess configurations. The person would be creating connected networks dedicated to specific domainsofcognition, andprocedureswhich, once set up, could be accessed quickly and efficiently by multiple parallelprocessing. Such a process for complex tasks,such as becoming a car driver or a chess expert, would require a good deal of packing and quite a bit of connecting, but once the job was done it would be highly efficient. Since much of culture consists of the performanceof these familiarprocedures and understandings,connectionism may explain what a great deal of culture in the mind-brain is like. It also explains why this type of culture cannot be either linguistic or 'languagelike'. Making the culture efficient requiresthe construction of connected domain-relevant networks, which by their very nature cannot be stored or accessedthrough sententiallogical forms such as govern naturallanguage. Furthermore, as the discussion of apprenticeshiplearning shows, it is not even necessaryfor thistype ofknowledge ever to be put into words for it to be transmitted from one member of the community to another. The highly efficient non-linear packing in purpose-dedicated domains, formed through the practice of closely related activities, also explains why the transferof one type of knowledge from one specific domain to another is so difficult. Lave's (1988) observation that there is no carry-over between school maths and coping with mathematicalproblems in a supermarketmay well be due to the fact that the latter are dealt with in such a well chunked and connected domain that it cannot easily admit knowledge of such a different kind as sentential-logically organised school maths.11 V To claim that much of culture is neither linguistic nor 'language like' does not imply thatlanguageis unimportant.Nevertheless, contraryto what anthropologists tend to assume, we should see linguistic phenomena as a partof culture, most of which is non-linguistic. Instead of taking language for granted, we should see its presence as requinrng explanation. I cannot, in the space of this article, review all the circumstances and reasons for the occurrence of language in cultural life. Nonetheless, even here, cognitive science can offer provocative suggestions. As we have seen, everyday practicalactions and knowledge are probablypackagedfairlyhermeticallyinto networks thattakethe form presumedby connectionist models. This packaging works very well for quick and efficient operations in familiardomains, but occasionallythese networks can also be unpacked into linear sentential sequences which can then be put into words. If such transformations are commonplace, as I believe they are, we should then see culture as partly organisedby connectionist networks andpartlyconsistingofinformation organised boundary between the two.12 by sentential logic, with a fluid transformative The processofputting knowledge into words must requiresuch a transformation in the natureofknowledge that the words will then have only a distantrelationship to the knowledge referredto. But the process may also involve gains in different areas, as suggested by work, cited earlier in this article, on the transformationof prototype concepts into classicalconcepts (concepts which can be defined by a checklist of necessaryand sufficient features).

MAURICE BLOCH

193

For example, the extension of aspects of knowledge, normally chunked in a particulardomain, into another domain may be one of the processes that require andsuch extension may well be linked with the processofinnovation, verbalisation, assome of the work on the significanceofanalogical thinking for creativitysuggests (Steinberg1985). Indeed, we should perhapssee cultureasalwaysbalancedbetween the need for chunking for efficiency, on the one hand, and, on the other, linguistic explicitness which allows, interalia, for innovation and for ideology. With such a perspectivewe might be able to envisagea kind ofgeneral 'economy' ofknowledge, which social anthropologists could then place in specific social and historical situations. VI In mentioning one ofthe possiblereasonsfor explicitnessI am suggestinga direction which would takeme farfrom the limited concerns ofthis article.But even without going into these issuesour discussionof the ways in which knowledge is organized has fundamentalimplications for anthropology. The first point is that culture is probablya differentkind ofphenomenon from what it haspreviouslybeen thought to be, with the result that our understandingof culture has remained partialand superficial. Up to now, anthropology has tried to analyse culture through folk models of thought applicableonly to sententiallogical knowledge which, as noted, is but a small part of all knowledge. Dreyfus and Dreyfus (1986: 28) point out how social scientists use a 'Hamlet model' of decision making where the actor is assumed to weigh up and analyse alternativesin a self-conscious, logical fashion. In this respect it is interesting to note that unlike most of our informants doing familiartasks,when Hamlet was trying to decide what to do next he was putting his thoughts into words. It is striking how many of the theories which have been popular in anthropology, such as transactionalisttheories and other forms of methodological individualism, fall foul of Dreyfus and Dreyfus's strictures. Then there are methodological implications. If the anthropologist is often attempting to give an account of chunked and non-sentential knowledge in a linguistic medium (writing), and she has no alternative,she must be aware that in so doing she is not reproducing the organisationof the knowledge of the people she studies but is transmutingit into an entirely different logical form. To effect such a transmutationis not impossible - afterall we can describe things which are not linguistic. But in the attempt to evoke such knowledge we should avoid stylistic devices which turn attempts at description into quasi-theory, as was the case with structuralism and transactionalism. Perhapswe should make much more use of description of the way things look, sound, feel, smell, taste and so on drawing on the realm of bodily experience - simply for heuristic purposes, to remind readersthat most of our materialis taken from the world of non-explicit expert practice and does not only come from linear, linguistic thought. Above all we should beware the temptations of folk psychology. Folk psychology is indeed a form of competence employed in certain limited circumstancesby the people we study, and it is therefore an object of researchfor anthropologists,but it is not something we want, or need, to carryover into anthropological explanation. Thus, when our informants honestly say 'this is why we do such things', or 'this is what this means', or 'this is how we do such things', insteadof being pleased

194

MAURICE BLOCH

knowledge is this which can we should be suspiciousand ask what kind of peculiar take such an explicit, linguistic form?13Indeed, we should treatall explicit knowledge as problematic, as a type of knowledge probablyremote from that employed in practicalactivities under normal circumstances. Of course such conclusions raise the question of how we are ever going to get at this connected, chunked knowledge. But here I believe anthropologistshave an advantageover other cognitive scientistsin that they alreadydo have a technique advocated by Malinowski but perhapsnever followed by him: participantobservation. Because of its long-term character,involving continuous and intimate contact with those whom we study, participantobservationmakes us learn the procedures which these people have themselveslearnedand enablesus to check up on whether we are learning properly by observing our improving ability to cope in the field with daily tasks,including social tasks,as fast as our informants. For example, as a result of fieldwork I too can judge quickly whether a bit of forest in Madagascar would make a good swidden. Indeed, I find that as I walk through the forest I am continually and involuntarily carryingout this sort of evaluation. Once this level ofparticipationhasbeen reachedwe can attemptto understandchunked knowledge through introspection. Introspection is, of course, a notoriously dangerous procedure, and fear of introspection is what led anthropologists,especially cognitive anthropologists,to adopt researchprocedures which imitate the laboratorystudies of others in order to 'harden'their findings. To me this approachseems misguided for two reasons. First, even in psychology, now that it has emerged from its behaviourist phase, the value of'ecologically' basedstudies(Neisser 1975) andthe merit ofintrospection as a theoreticalstartingpoint are once againbecoming recognised. Secondly, much 'hard' evidence has proved on critical examination to be much 'softer' than the information obtained through participantobservation. For example, one lesson which seems to follow from my argument is that ethnography backed up by our informants' 'actualwords' may often be quite misleading. In any case, I believe that anthropologistswho have done prolonged fieldwork have always obtained the basis of their knowledge about the people they study from informal and implicit co-operation with them, whatever they might have pretended. I am fairly sure that the way I proceed in giving an account of the MalagasyculturesI study is by looking for facts, and especiallyfor statements,that what I already know to be right because I know how to live efficiently confirm with these people, or, if you will, because I have establishedin my brain non-linguistic chunked mental models which enable me to cope with most things in daily life at great speed. Like other anthropologists, I then pretend that the linguistic confirmationsof these understandings, which I subsequentlyobtain from what my informants say, are the basisof what I understand, but this is not really so. My knowledge was establishedprior to these linguistic confirmations.14 To recognise this is not shamefuland should help us to avoid what is our greatest betrayalas anthropologists.One fact we always and rightly stresswhen explaining how our way of going about things contrastswith that of other social or cognitive scientists is the importance we attach to the everyday and how we believe that the most important aspectsof culture are embedded in the basic mental premisses

MAURICE BLOCH

195

of action. Anthropologists are particularlyaware of this because many work in foreign cultures and so the exotic-in-the-everyday cannot but be prominent. We are also reminded of the importance of everyday practicalculture because we do long-term fieldwork and participatefor very long periods indeed. This learning about the practicalis the best thing about anthropology, yet it often hardlyfeatures in our ethnographies;ratherwe rush to studies of ritualswhich could have been done in a week, or to analyses of economic organisation which could be done better by geographersand for which participantobservation is, in any case, unsuitable. Anthropologists worried about the difficulty of applying the kind of rigorous confirmation procedures of other disciplines should not worry too much. They certainlyshould not give up altogetherandpretend, ashasrecentlybeen fashionable, that they are merely involved in some literary exercise. They should learn to respecttheir own work rathermore, andin cognitive mattersthey should remember that the basisof cognitive science is a combination of differentdisciplines, each with a contribution to make, but with a single aim. The greatest contribution of anthropology to this totality can perhaps be to specify how cognition is employed in a naturalenvironment, and in this way to help to decide what kinds of hypotheses and findings are requiredfrom those other cognitive sciences which work in more controlled conditions. It is because I am familiarwith daily life in a Malagasyvillage, and also because I look for general explanations of what is going on, that I see the central problems to be solved as those of accounting for such intellectual feats as that of the peasantlooking at the forest in searchof a swidden site. Dreyfus and Dreyfus are quite right to point out that, by failing to pay attention to the way real human experts operate, cognitive scientistshave attempted to create artificialintelligences along lines that ignore the very characteristics of human practice, but this is not a reason for arguing, as they do, for giving up the attempt altogether. It is a reason why, in the same way as anthropologists need other cognitive scientists, these other cognitive scientists would also benefit from co-operation with anthropologistshaving experience of participantobservation.
NOTES The research on which this article was based was financed by The Spencer Foundation. I also benefited greatly from the hospitality of the anthropology department of Bergen University where I was able to develop some of the ideas presented here. I would also like to thank F. Cannell, C. Fuller, D. Holland, N. Quinn and G. Straussfor helpful comments on an earlier draft. Above all I would like to thank D. Sperber for introducing me to the subject and making many useful suggestions for the improvement on the text of the lecture as it was originally delivered. 1 I do not want to imply that all members of a community need possess all cultural knowledge. Discussions of 'distributedcognition' by Cicourel and others suggest that this may not be so. 2 See Sperber (1985) to find this point fuhlly argued. 3 Among American anthropologists who have discussed connectionism are Quinn and Strauss (1989), Hutchins (1988) and D'Andrade (1990). 4 It is true that structuralism,at least in its Levi-Straussianform, paid attention to memory in that it assumed that the source of structurationwas the need to encode information so that it could be kept and retrieved. In retrospect, however, as was shown by Sperber, the structuralist's view of memory was much too simple. In particular it assumed, with other types of cognitive anthropology, that what an individual received from others was stored in the same form as it had been communicated and that all information was equally memorable (Sperber 1985).

196

MAURICE BLOCH

The theories of learning implicit in structuralismare even more problematic since they cannot account for the inevitably gradualconstruction of structuredknowledge, a criticism which has been made in a variety of ways by Piaget (1968), Turner (1973), Sperber (1985) and myself (1985). Admittedly, such writers as Bourdieu (1972) have tried to remedy this situation, but in his case with a theory of learning habituswhich is psychologically vague and which, because of its reintroduction of the notion of individual 'rational' choice, runs into some of the difficulties which I go on to discuss. 5 Holland and Quinn (1987: 19) draw attention to the significance of these for our understanding of culture. 6 Work on deaf and dumb children also seems to show that advanced conceptual thought does not require language (Petitto 1987; 1988). I am grateful to L. Hirschfeld for pointing out the significance of this work for my argument. 7 Although I am relying extensively on Dreyfus and Dreyfus's characterisation of expertise (1986), I do this to reach quite different conclusions. 8 I am told that in Norway, although the normal driving test involves both a practical section and a question and answer section, some people seem totally unable to do the language part of the test, and so a more difficult practical test has been devised for them which eliminates the need to do the language part. There is apparentlyno evidence that people who have obtained their test in this way are worse drivers than the others. 9 The point has alreadybeen made by Schank and Abelson (1977). 10 After all, it is possible to argue, as Fodor does, that although thought is not a matter of speaking to oneself silently, it still is ultimately 'language like' and involves series of 'grammatically' (though not the grammar of the surface structure of natural languages) linked representations and propositions. This suggestion enables Fodor to talk of a 'language of thought', though it might be better to say a 'quasi-languageof thought' (Fodor 1987). 11The fact that people who are unable to perform Wason's task (a logical test performed under laboratory conditions with two-sided cards) when faced by his cards can do it when it is reproduced in a context familiar to them points to similar conclusions Johnson-Laird & Wason 1977). See the discussion of this point in Dreyfus and Dreyfus (1986: 19). 12 Such a reconciliation of rule and representation models with connectionist ones has been proposed by Bechtel (1990). 13 These explicitly linguistic forms may have different relations to knowledge (see Barth 1990). 14 An attempt to write an ethnography along these lines is to be found in Bloch (forthcoming.)

REFERENCES

Abelson, R. 1981. Psychological status of the script concept. Am. Psychol.36, 715-29. Anderson, J.R. 1976. The architecture of memory. Cambridge, MA: HarvardUniv. Press. Atran, S. 1990. Cognitivefoundations of natural history.Cambridge: Univ. Press. Barth, F. 1990. The guru and the conjurer: transactionsin knowledge and the shaping of culture in southeast Asia and Melanesia. Man (N.S.) 25, 640-53. Bechtel, W. 1990. Connectionism and the philosophy of mind: an overview. In Mind and Cognition (ed.) W. Lycan. Oxford: Blackwell. Bloch, M. 1985. From cognition to ideology. In Powerand knowledge: anthropological and sociological approaches (ed.) R. Fardon. Edinburgh: Scottish Univ. Press. in press. What goes without saying: the conceptualisation of Zafimaniry society. In Conceptualising society(ed.) A. Kuper. London: Routledge. Bourdieu, P. 1972. Esquissed'une theorie de la pratique.Paris:Droz. Bowerman, M. 1977. The acquisition of word meaning: an investigation in some current concepts. In Thinking:readings in cognitive science (eds) P. Jonnson-Laird & P. Wason. Cambridge: Univ. Press. Brown, R. 1973. Afirst language: the earlystages.Harmondsworth: Penguin. Churchland, P.M. & P.S. Churchland 1983. Stalking the wild episteme. Nous 17, 5-18 (reprintedin Mind and cognition (ed.) W.L. Lycan. Oxford: Blackwell, 1990). Churchland, P.S. & T. Sejnowski 1989. Neural representation and neural computation. In Neural mentalcomputations connections, (eds) L. Nadel et al. Cambridge, Mass.: M.I.T. Press.

MAURICE BLOCH

197

D'Andrade, R. 1987. A folk model of the mind. In Culturalmodelsin language and thought (eds) D. Holland & N. Quinn. Cambridge: Univ. Press. 1990. Some propositions about the relations between culture and cognition. In Cultural psychology (eds)J. Stigler et al. Cambridge: Univ. Press. in the eraof the thepowerof intuitionand expertise Dreyfus, H. & S. Dreyfus 1986. Mind overmachine: computer. Oxford: Blackwell. Feldman,J. & D. Ballard 1982. Connectionist models and their properties. Cogn. Sci. 6, 205-54. Fillmore, C. 1975. An alternative to checklist theories of meaning. In Proceedings of thefirstAnnual Meetingof the Berkeley Linguistic Society,123-31. Fodor, J.A. 1987. Psychosemantics. Cambridge, MA: M.I.T Press. 1983. Modularity of mind:an essayonfacultypsychology. Cambridge, MA.: M.I.T. Press. in socialinteraction andpoliteness: Goody, E. 1978. Towards a theory of questions. In Questions strategies (ed.) E. Goody. Cambridge: Univ. Press. (ed.) Haudricourt, A.-G. 1968. La technologie culturelle:essaide methodologie. In Ethnologiege'ne'rale J. Poirier. Paris:Pleiade. Holland, D. & N. Quinn (eds) 1987. Cultural modelsin language and thought. Cambridge: Univ. Press. Hutchins, E. 1980. Culture and inference: a Trobriand casestudy.Cambridge, MA.: HarvardUniv. Press. 1988. Where's the expertise in expert navigation? Paper delivered at the annual meeting of the American Anthropological Association at Phoenix, Arizona. science and consciousness. Johnson-Laird, P. 1983. Mentalmodels:towards a cognitive of language,inference Cambridge: Univ. Press. to cognitive 1988. The computer and the mind:an introduction science. London: Fontana. & P. Wason 1977. A theoretical analysisof insight into a reasoning task. In Thinking: readings in cognitive science (eds) P. Johnson-Laird & P. Wason. Cambridge: Univ. Press. Keil F.C. & N. Batterman 1984. A characteristic-to-defining shift in the development of word meaning. Verb.Learning and verbal Behav. 23, 221-36. Lave, J. 1988. Cognitionin practice. Cambridge: Univ. Press. psychology 1990. The culture of acquisition and the practice of understanding. In Cultural (eds)J. Stigler et al. Cambridge: Univ. Press. Leroi-Gourhan, A. 1943. Evolutionet techniques. Paris:P.U.F. andtheir Malinowski, B. 1935. Thelanguage ofmagic andgardening (CoralGardens MagicVol. 2). London: Allen & Unwin. Markham,E.M. &J. Seibert 1976. Classesand collections: internal organisationand resulting holistic properties. Cognit.Psychol.8, 561-77. Mauss, M. 1936. Les techniques du corps.J. Psychol.32, n. 3-4. San Francisco: and implications Neisser, U. 1975. Cognitionand reality: principles of cognitive psychology. Freeman. Newell, A. & P.S. Rosenbloom 1981. Mechanisms of skill acquisition and the law of practice. In skillsand theiracquisition Cognitive (ed.) J.R. Anderson. Hillsdale, NJ: Erlbaum. Petitto, L.A. 1987. On the autonomy of language and gesture: evidence from the acquisition of personal pronouns in American sign language. Cognition 27, 1-52. 1988. Language and the prelinguistic child. In The development of languageand language researchers: essaysin honorof RogerBrown(ed.) F.S. Kessel. Hillsdale, NJ: Erlbaum. Paris:P.U.F. Piaget, J. 1968. Le structuralisme. Quinn, N. & C. Strauss1989. A cognitive cultural anthropology. Paper preparedfor the session on "AssessingDevelopments in Anthropology" ofthe AmericanAnthropological Association Annual Meetings, 1989. Rosch, E. 1977. Classification of real world objects: origins and representations in cognition. In in cognitive science Thinking: readings (eds) P. Johnson-Laird & P. Wason. Cambridge: Univ. Press. 1978. Cognitive representationsof semantic categories.J. exp. Psychol.104, 192-223. Rumelhart, D., G. Hinton & J. McClelland 1986. A general framework for parallel distributed processing. In Explorationsin the microstructure of cognition,vol. 1. (eds) D. Rumelhart & J. McClelland. Cambridge, MA.: M.I.T. Press. Schank, R. & R. Abelson 1977. Scripts, plans,goals and understanding. Hillsdale, NJ: Erlbaum. Schneider, W. & R. Shiffrin 1977. Controlled and automatic human information processing, 1: detection, search and attention. Psychol.Rev. 84, 1-66. New Haven: Yale Univ. Press. Simon, H. 1979. Modelsof thought.

198

MAURICE BLOCH

Smith, E.E. 1988. Concepts and thought. In Thepsychology of humanthought (eds) RJ. Stemnberg & E.E. Smith. Cambridge: Univ. Press. Smith, L., M. Sera & B. Gattuso 1988. The development of thinking. In The psychology of human thought (eds) RJ. Stemnberg & E.E. Smith. Cambridge: Univ. Press. Sperber, D. 1985. Anthropology and psychology: towards an epidemiology of representations.Man (N.S.) 20, 73-89. & W. Wilson 1986. Relevance: communication and cognition. Oxford: Blackwell. Stemnberg, RJ. 1985. BeyondI.Q.: a triadic theory of humanintelligence. Cambridge: Univ. Press. Turner, T. 1973. Piaget's structuralism.Am. Anthrop.75, 351-73. Tyler, S. (ed.) 1969. Cognitive anthropology. New York: Holt, Rinehart & Winston. Winch, P. 1958. The ideaof a socialscience. London: Routledge & Kegan Paul.

Langage, anthropologie et science cognitive

Resume
L'anthropologie s'est fi&e a un modele de la culture psychologiquement trompeur et par trop linguistique. Des etudes de psychologie cognitive montrent que les concepts sont principalement formes independamment du langage. En outre, d'autres etudes montrent qu'afin de devenir un expert a une tache familiere ou a un ensemble de taches, une personne organise sa connaissance d'une fa,on qui ne correspond nullement a l'organisationdu langage. Appel est fait au 'connectionisme' pour proposer que la plupart du savoir culturel ne peut pas etre organise de fa,on analogue a la logique sous-ja,ente au langage. Parcontre il est sugg&re qu'une contribution de l'anthropologie a la science cognitive pourrait-etre la technique de l'observation participante qui donne acces au savoir culturel sans passerpar l'intermediaire dangereux du langage.

Department ofAnthropology, LondonSchool of Economics, Houghton Street,LondonWC2A 2AE