Cognition, Vol. 3, No. 1

Editorial board

Azuma of Education. University, Bunkyo-ku, Japan

Jerome S. Bruner Center for Cognitive Studies, Harvard University, Cambridge, Mass. 02139, U.S.A.

Merrill Garrett Department of Psychology, M.I.T. EIO-034, Cambridge, Mass. 02139, U.S.A.

Paul Bertelson Laboratoire de Psychologie Experimentale, UniversitP Libre de Bruxelles II 7, Av. Adolphe Buyl, B-1050 Bruxelles, Belgique

Noam Chomsky Dept. Modern Languages and Linguistics, MI-T., Cambridge, Mass. 02139, .!%S.A.

Pierre Greco Laboratoire de Psychologie, 54, bvd. Raspail, F- 75006 Paris, France

Hiroshi Fact&v TokyoHongo, Tokyo,

T. G. R. Bower Dept. of Psychology, University of Edinburgh, 60, Pleasance, Edinburgh EH8 9TJ, Great Britain

Peter D. Eimas Walter S. Hunter Laboratory of Psychology, Brown University, fiovidence, R. I. 02912, U.S.A.

Ned Block Dept. of Philosophy, M.I.T., Cambridge, Mass. 02139, US.A.

Gunnar Fant Lab. of Speech Transmission, Royal Institute of Technology, S-1 0044 Stockholm 70, Sweden

Frarqois Bresson Laboratoire de Psychologie, 54, bvd. Raspail, F- 75006 Paris, France

Jerry Fodor Dept. of Psychology, MI. T. El O-34 Cam bridge, Mass. 02139, U.S.A.

Roger Brown Dept. of Psychology, Harvard University, Cambridge, Mass. 02138, U.S.A.

Kenneth Forster Dept. of Psychology, Monash University, CTayton, Victoria 3168, Australia

Jean-Blaise Grize, 1, Ghan temerle, Neuchdtel, Suisse

David T. Hakes Department of Psychology, University of Texas, Austin, Tex. 78712, U.S.A.

Henry Hecaen Directeur d’Etudes, Ecole Pratique des Hau tes Etudes, Unite’ de Recherches Neuropsychologiques, I.N.S.E.R.M., 2, rue dillesia, F- 75014 Paris, France

Michel Imbert Laboratoire de Neurophysiologic, College de France, Il. place Marcelin Berthelot, F- 75005 Paris, France

A. R. Luria Faculty of Psychology, University of Mosco w, 13, Frunze Street, Moscow G.19, U.S.S.R.

Robert Shaw Center for Research and Human Learning, University of Minneso taI Minneapolis, Minn. 554.555, U.S.A.

John Lyons Dept. of L~~ist~es, Adam Ferguson B&ding, Edinburg~l EH8 9LL, Great Britain

Dan I. Slobin Department of Psychology, Un~versit_vof ~liforn~, Berkeley, #hf. 94720, U.S.A.

James Jenkins Center for Research and Human Learninn, University of M&neso ut, Minneapolis, Minn. 55455, U.S.A.

Humberto Maturana Eseuela de Medicina, Universidad de Chile, A. Sanartu 1042, Santiago, Chile

Jan Smedslund Jnstitute of Psychology, Universitet i Oslo, Box 1094, Oslo 3, Norway

Daniel Kahneman Dept. of Psychology, The Hebrew University of Je~salem, Jerusalem, Israel

John Morton Applied Psychology Unit, 15, Chaucer Road, Grmbridge CB2 2EF. Great Britain

B&be1 Inhelder Institut des Sciences de PEducation, Universite de Geneve, CHV1211 GenPve 14, Suisse Marc Jeannerod Laboratoire de Neuropsychologie Expertmentale, Doyen Lepme, F-69500 Bran, Frame

Jerrold J. Katz Dept. of Philosophy, MLT., Gzmbridge, Mass. 02139, U.S.A. Edward Klima Dept. of Linguistics, La Jolla, University of California, San Diego, Calif 92031, USA. Eric H. Lenneberg Dept. of Psychology, Cornell University, Ithaca, N. Y. 14850, U.S.A. Atexei Leontiev Faculty of Psychofogy, University of ~us~ow, 13, Fnrnze Street, Moscow G.19, U.S.S.R. Wilhelm Levelt Psychological Laboratory, Nqmegen University, Erasmuslaan 16, Ngmegen, Netherlands

George Noizet Laboratoire de Psychologie Experimentale, F-13 Aiw en Provence, France Domenico Parisi fstituto di Ps~c~log~a, Consigho Nazionale delle Ricer&e, Piazzale delle scienze 7, Rome, Italy Michael Posner

Dept. of Psychology, University of Oregon, Eugene, Ore. 9 7403, K S.A. Nicotas Ruwet Dept. de L~~~isti~u~, Centre Univ. de Vincennes, 12, Rue de Tourelle F - 75012Paris, France Harris B. Savin Dept. of Psychology, University ofPennsylvania, Philadelphia, Pa. 19104 U.S.A.

Sidney Strauss Department of ~d~i&ationa~ Sciences, Tel Aviv Unive~ity, Ramat Aviv, Israel Alina Szeminska Olesiska S/3, Warsaw. Poland Yoshihisa Tanaka Dept. ofPsychology, University of Tokyo, ~nkyo~u, Tokyo 123, Japan Hans-Lukas Teuber Dept. of Psychology, M.I.T. EIO-034, Cambridge, Mass. 02139, (I.S.A. Peter Wason ~y~holin~~tics IJniversity College London, Research Unit, 4, Stephenson Way, London NW1 2HE. Great Britain Hermina Sinclair de Zwart Centre d’Epistemologie GPnPtique, Universitd de Geneve, CH-1211 Geneve, St&se

1

Interactional

aspects of cognitive organization*

JONAS LANGER University

of California,

Berkeley

Abstract Further propositions are formulated towards a comprehensive structural developmental theory of cognitive change, begun in Langer (1969a, 1969b). Here, the analytic focus is upon: (a) 7he organization of the subject’s assimilatory operations and accomodatory figurations; (b) the intrinsic coordinations between the theoretical and empirical cognitions constructed, respectively, by these two kinds of functional structures; and (c) the cognitive developmental changes produced by instrinsic coordinations. Derivative empin’cal hypotheses are considered in light of findings summarized here and elsewhere. The development of any sequence of psychological stages is an interactive process of equilibrating the intrinsic functional structures of the organism with the extrinsic structures of the environment &anger, 1969a, Ch. 5). A variety of approaches has been used to study the interactive process. Typically these approaches have established external disparity - usually produced by an experimenter or a teacher - between the child’s existing level of conceptions and the actual physical or social character of the environment. To illustrate, the experimental situation may be designed to establish a disparity between the child’s predictive judgment of the outcome of a physical deformation of an object and his observation of the actual outcome. Often, this type of approach presumes that such external disparity is reflected by intrinsic ‘cognitive conflict’ when in fact there is little if any theoretical basis, and usually no independent empirical evidence, for the presumption. In most of these situations the more parsimonious assumption is that the external disparity remains just that - external to the child’s conceptual operations - and is therefore not cognized as anomolous. Our approach has been to focus upon intrinsic disequilibrium as a source of the * Expanded version of national conference on of the Child, Urbana, while the author was Rockefeller Universitv.

a paper presented at the The Natural Curriculum March 1969, prepared a Visiting Scholar at 1970, and supported in

part by USPHS Grant HD 03617 on The Intergenerational Studies in Development and Aging at the Institute of Human Development, University of California, Berkeley.

Cognition

3/I),

pp. 9-28

10

Jonas Langer

development and interiorization of mental operations (Langer, 1969b). This has included inquiring into how interactive disparity between intrinsic (organismic) and extrinsic (environmental) functional structures may perturb the existing cognitive organization of the child. This required studying how interactive disparity may provide the occasion for perturbative feedback and the intrinsic disequilibrium necessary for conceptual transformation to a more advanced stage. Consequently, these efforts have been partially directed towards distinguishing, theoretically and empirically, between external disparity and intrinsic disequilibrium. The equilibration model implies the hypothesis of progressive or regressive cognitive transformation when intrinsic disequilibrium or disparity has been demonstrated by independent means. Only then does the theoretical expectation follow that the energetic conditions are present for the formation and interiorization of more or less advanced mental operations and concepts. In order to examine intrinsic disequilibrium more closely, two major theoretical parameters of equilibrating intrinsic functional structures were proposed, an organizational and an energetic parameter &anger, 1969b). An additional distinction was then made between two complementary aspects of the organizational parameter. The first is the interaction between the organism’s systems of action, and the second is the interaction of the media in which acts and the environment are embodied. The present essay is directed towards some further analyses of these two interactive aspects of the organizaThese analyses will be guided by developmental tional parameter of equilibration. considerations that are relevant to learning and education.

1. Figurative and operative functional

structures

Previously we were led to the conclusion that intrinsic integration within functional structures may lead to progressive cognitive alterations, while intrinsic segregation may result in temporary regressive-like cognitive effects (Langer, 1969b). Here we shall explore the transformational consequences of intrinsic interaction between two classes of functional structures. These we shall call accommodatory figurations and assimilatory operations. Piaget (Piaget and Inhelder, 1969a, 1969b; Piaget, 1969) has most clearly articulated the differences between two kinds of structures, figurations and operations. Figurations are the basis for the range of action schemata that cognitively apprehend, extract and reproduce aspects of the physical and social environment. The action schemata involved include components of perception, speech as an expressive medium, imagery and memory. Basically, figurative acts are directed towards reproducing empirical information. Operations are the basis for the range of action schemes that construct logical transformations. These logical schemes of transformation operate upon environmental events or upon one’s own logical operations. Logical operations are directed toward producing conceptual knowledge.

Interactional aspects of cognitive organization

11

Piaget (Piaget, 1951; Piaget and Inhelder, 1966; Piaget, 1967) has proposed that figurative and operative structures are like two parallel streams that have their genetic origins in the same source, the sensorimotor structure of schemata that develop during the first two years of life. The proposal of parallel streams suggests that operative structures of conception do not derive from figurative schemata of perception, speech, imagery or memory. Reciprocally, figurative structures do not derive from operative schemes. Furthermore, figurative structures do not derive from each other but have their individual sources in sensorimotor schemata. For example, imagery is not a derivative of perception but rather of deferred sensorimotor imitation (Piaget, 195 1; Piaget and Inhelder, 1966). Perception, itself, is a derivative of immediate sensorimotor fixations and not deferred sensorimotor imitation (Piaget, 1969). Piaget (1969) draws a strong division between the operative and figurative structures that develop after the sensorimotor stage. Operative structures are constructive while figurative are not. Operative structures produce logical concepts. Figurative structures reproduce environmental configurations. This hypothesis led Piaget and his colleagues to examine the influence of operative conception upon perception (Piaget, 1969), imagery (Piaget and Inhelder, 1966) and memory (Piaget and Inhelder, 1968). They have also somewhat examined the influence of figurative speech upon operative conceptions (Sinclair, 1967). Piaget (1950) also makes a strong distinction between structure and function. Operative, but not figurative, structures are variant. The functions of assimilation and accommodation are invariant throughout life. The variant operative structures (of intuitive, concrete and formal operations) form the discontinuous sequence of stages that composes cognitive development. The invariant functions of assimilation and accommodation provide cognitive development with its continuity. The theoretical formulation @anger, 1964, 1969a, 1969b, 1970a 1970b, 1970~) adopted here makes two fundamental assumptions. First, all developmental processes are continuous and discontinuous. Second, all cognitive activity is constructive. Consequently it follows that both operative and figurative structures are constructive systems and that the development of both structures and functions is marked by continuity and discontinuity. The special sense in which figurative activity is constructive is that it consists of extractive judgment and/or expressive representation about environmental configurations selected for interaction by the subject. And that interaction is always reconstructive, never merely impressionistic and/or mimetic. Psychological structures and functions are but complementary aspects of a constructive organism whose development is a dialectical process of maintaining its continuous integrity while elaborating discontinuous forms. The central thesis is that the organization of a person at any given stage of his development is that of a multi-leveled network of finctional structures. Each stage is the progressive transform of the functional structures that composed the stage out of which it materialized. Thenetwork of each stage is composed of two major sets of functional structures that

12

Jorzas Langer

construct knowledge. One set is composed of accommodatory figurations (to mix Piaget’s metaphors) while the other set is composed of assimilatory operations. Accommodatory figurations are the intrinsic bases for action systems directed towards constructively extracting and representing empirical ~formation. Extracting schematic knowledge, such as information about the appearance, reality and pre~ctab~ity of the physical and social enviro~ent, is accomplished by systems of action such as attending, examining, selecting, verifying and discriminating. Expressing the knowledge achieved by these extractive means is accomplished by symbolic embodiments in a variety of media that have representational and communicative value. Assimilatory operations are the intrinsic bases for the action systems directed towards constructing theoretical knowledge or understanding. First approx~ations to theoretical understanding are obtained by performing mental transformations upon the information and representations achieved via aceommodatory figurations. Higher-order logica conceptions are first constructed in adolescence by perfor~ng mental operations upon the first-order mental tr~sformations. There are three logically possible relations between accommodatory figurations and assimilatory operations. The first is that they are totally different and unrelated. The psychological implication of such a dualistic theory would seem to be that they are mentally segregated functional structures which do not interact with each other. Therefore, they set no limits to each other’s operations and development. It is difficult to find serious representatives of this kind of dualistic theory of normal mentation, at least in Western thought. Rather, evidence of such a split is more likely to be taken as symptomatic of abnormal mentation (e.g., Werner, 1948). A second possibility is that of theoretically reducing one set to the other. There has been little attempt to reduce all psychological phenomena to assimilatory operations, except possibly by radical subjective idealists. There have been comprehensive attempts to reduce all mental phenomena to accommodatory ~gurations. Gestaltists and phenomenologists have typically reduced mentation to figurative perception. Michotte (1963) attempted to reduce the cognition of causality to its perceptual components. More recently, Bower (1967) has been using a similar reductionistic approach in studying object permanence. In these theories concepts are usually considered to be epiphenomenal or distorted forms of percepts. AlternativeIy, sensationists and social learning theorists have attempted to reduce psycholo~cal phenomena to sensory impressions and/or to motoric imitations (e.g., Bandura and Walters, 1963; Bandura, 1969). The third theoretical possibility, and the one taken most seriously here, is partial communication between accommodatory figurations and assimilatory operations. The major implication for a theory of mental organization is that each type of functional structure may interact with the operations and development of the other. It is hypothesized, and later we shall present evidence in support of the hypothesis, that the nature of the interaction will differ depending upon the developmental stage of the person. In principle, the person’s stage consists of both his operational and figurational levels of development; but, in research practice, the assessment is usually limited to diagnosing one

Interactional aspects of cognitive organization

13

or the other. It is assumed that the interaction takes an informational form. This means that the development of each type of functional structure has logical implications for, but not direct causal effects upon, the operations and development of the other. From the present perspective of genetic epistemology, the evolution of these two sets of functional structures is the cognitive basis for the history of ideas, whether social, ethical, aesthetic, physical or logical. Accommodatory figurations construct our empirical facts and assimilatory operations construct our theoretical knowledge. This epistemological perspective seeks to supersede the academic and traditional categorization of mental faculties and phenomena such as perceptual, conceptual, linguistic, mnemonic and learning. Useful as they may have been when viewed in historical perspective, these categories have seemed artificial and psychologically unreal ever since Kant wrote his Ch’tique of pure reason. In their place we therefore propose accommodatory figurations and assimilatory operations that construct facts and theory, whether physical or social. The theoretical task before us is to formulate the general rules of constructing information and concepts, whether perceptual, conceptual, linguistic, mathematical, logical, etc. The study of cognitive development, then, becomes in great measure an inquiry into the development of these two sets of functional structures, their genetic relations and their organizational (systems) interactions. From the perspective of genetic psychology, the developmental relation between accommodatory figurations and assimilatory operations is an orthogenetic process (Werner, 1948; Langer, 1970a). The basic organization of the sensorimotor stage is a global fusion of assimilatory operations (such as playful transformations) and accommodatory figurations (such as imitative reproductions). The observable result is syncresis of perception, action and affect. The sequence of stage development, starting with the sensorimotor stage, is marked by progressive differentiation and integration of these functional structures. The orthogenesis of accommodatory figurations and assimilatory operations constitutes a central feature of the intrinsic equilibration that is the source of stage development, i.e., progressive alteration of functional structures. The general hypothesis is that the operation of one system of action may lead to the feedback of information that modifies its underlying functional structures. In turn, this may result in intrinsic disequilibrium between it and other functional structures. This is a necessary condition for progressive reorganization and feed-forward to the child’s schemes of action. The manifest result should then be some observable modifications and advances in intelligent acts and their products, constructed facts and theories.

2. Actions in interaction The issues are twofold, organizational and developmental. The general organizational issue is to determine the results of interaction between functional structures for the

14

Jonas Langer

formation of any given functional structure. The general developmental issue is to determine whether the results of interaction lead to some change, whether progressive or regressive, in any given functional structure. Here the organizational and developmental considerations will be limited to the unidirectional implications of accommodatory figurations for assimilatory operations. The next four sections will outline some of the ways in which we have been investigating the problem experimentally. Specifically, each section will be devoted to the discussion of one figurative means of mentally extracting or representing empirical information about physical and social objects and the consequences of such empirical activity for the construction of operational concepts. The first three figurative means are primarily extractive. They are imitating an observed event, comparing one’s predictions about the result of a physical deformation with one’s observation of the actual outcome and comparing one’s observations of the way things appear with one’s observations of the way things really are. The fourth figurative means is primarily representational and involves symbolization of empirical presentations and reasonings about them. In the concluding section we will consider some of the educational implications of our findings.

3. Imitation We begin with and give most attention to the implications of imitation for cognitive development. The reason is that imitation is the most radical form of figuratively accommodating physical and social information. The psychological literature on imitation may be classified into two groups: The first group considers imitation as a type of behavioral phenomenon. Like all other behavioral phenomena, imitative phenomena are the product of other mechanisms such as conditioning and reinforcement. As such, one should be able to plot an ontogenetic growth curve for imitative behaviors. The second group considers imitation as a mechanism that produces behavior. For example, Miller and Dollard (1941) hypothesized that imitation is a secondary drive to behave. As such, imitation is a mechanism for the acquisition and growth of behavior. But imitation itself should not show an ontogenetic growth curve: It is a universal mechanism of learning. From the genetic epistemological point of view (Cassirer, 1953; Kohlberg, 1969; Langer, 1969a; Piaget, 1967; Werner, 1948) imitation represents perhaps the purest case of figurative accommodation, that is, directly extracting and reproducing aspects of the physical and social environment. Functionally, imitation represents the subject’s radical form of accommodating to objects. Structurally, it represents his radical form of Iiguratively schematizing parts of environmental configurations. Entailed in this epistemological analysis is a broad-gauged definition of imitation. Imitation may involve all or some of the following components: Following, tracing, selecting, getting an impression of and reproducing parts of physical or social objects or events. Thus defined, imitation plays a

Interactional

aspects of cognitive organization

15

significant part in the construction of object permanence (Cassirer, 1944; Piaget, 195 I), in the intersensory identification of form (Zaporotchets, 1965), etc.’ Imitation has its developmental roots in sensorimotor activity. Its original precursory form may be that of contagious activity. Piaget (195 1) reports observations in which his one- to six-day&d infant cried in response to hearing other infants cry or Piaget simulating crying sounds. The infant did not cry in response to other sounds of equal intensity. Clearer manifestations of precursory imitative activity is found by one month. Many observers report co-action by this age. For example, Zazzo (1953) found the following circular reaction. An infant sticks his tongue out, then the observer sticks his tongue out, the infant follows by sticking his tongue out, and so the cycle may continue for three or four co-active units. It should be observed that co-action seems to be found at this age only when (1) the neonate initially performs an act already present in his sensorimotor repertoire and (2) the social environment imitates the infant’s act. This leads to the hypothesis that the neonate’s co-action is already guided by a constructive schema in which the imitator is cognized (albeit in a nonconscious form) by the neonate as performing acts which are analogous to results that the neonate has already obtained himself. Deferred imitation, the most advanced form of sensorimotor imitation, develops around a year-andone-half when children begin to reproduce aspects of the performance of, or characteristics of, previously observed but presently absent physical or social models. Of special interest from the constructivist point of view are two features of deferred imitation: (1) The delay between the observation and the reproduction and (2) the often-times partial and transformed nature of the reproduction. These features are sufficient to suggest that the child’s schema already involves imaginal and memory components that guide his selective reproduction. Between the ages of two to seven years the child is relatively intuitive and egocentric. Piaget (195 1, p. 73) therefore hypothesized that, although somewhat influenced by his operative level of understanding, ‘the child often imitates without being aware of it, merely through confusion of his activity or his point of view with those of others’. During this stage the primary determinant of the child’s imitative activity is his figurative structures. The intuitive child’s imitation is marked by three features: (1) It is relatively automatic and nonconscious; (2) it is relatively direct and nonanalytic; and (3) it is relatively indiscriminate and nonselective. Between the ages of seven and eleven, with the development of concrete operations, imitation comes under greater sway of the child’s operative structures. Consequently, imitation becomes increasingly conscious, selective, analytic and discriminate. 1. We have already noted that accommodatory figurations are complemented by assimilatory operational functional structures. We should briefly add that perhaps the most pure form of assimilatory operations is play (Piaget, 1951: Werner, 1948). Functionally, play represents

the subject’s radical form of assimilating objects, that is, transforming the environment. Structurally, play represents his radical form of operatively constructing meaning and attributing significance to the environment.

16

Jonas Langer

From the perspective of social learning theory (e.g., Bandura and Walters, 1963; Bandura, 1969) the Piagetian hypothesis about the two- to seven-year-old (1) attributes too much rationality to the process of imitation and (2) underestimates the importance of imitation for development. Social learning theory makes two basic claims about imitation. The first is a behavioral claim. The claim is that as a behavior imitation is composed of two phases, an acquisition phase and a performance phase. The acquisition phase consists of the child observing a model (and probably storing a memory trace of the sensory impression). The performance phase consists of the child overtly reproducing his observation (and depends upon the schedule of reinforcement to the child or the model). The second claim is about the mechanism of development. The claim is that imitation is the mechanism of much, if not most, of the child’s behavioral development, particularly his social development. Furthermore, imitation is the mechanism by which all novel social behaviors are acquired by the child. The empirical basis for the behavioral claim is observational and experimental. Many observers have reported that young children may imitate in what appears on the surface to be an automatic fashion. Recently, these observations have been poured into an experimental mold. Numerous variations on the theme have been performed in the last ten years, but the following procedure used by Bandura and his colleagues seems paradigmatic. A child is brought into a room in which some toys are scattered and where an adult model is beating up a hobo doll. After performing this act the model leaves the room saying, ‘I’ll be back in a few minutes. You can do anything you want while I’m gone. I’ll have a prize for you when I get back’. The basic finding is that four- to six-year-old children imitate the model’s performance while he is gone. Our theoretical perspective led to a reinterpretation of the significance of such findings and the experimental design on which they are based. Essentially the reinterpretation was that four- to six-year-old children understand what they have been presented (in the bobo doll situation) as an implicit message, but an instruction nevertheless, to reproduce the presentation. Based upon this reinterpretation we performed an experiment (Kuhn and Langer, 1968) in which we introduced two additional variables to the standard procedure, an ontogenetic variable and an instructional variable: A. Age: Groups of three-year-olds (XC 3; 10) and four-year-olds (x= 4; 10) were tested. The expectation was that the threes would not understand the standard social-learning instructions as an implicit message to reproduce as compared to the fours. B. Instructional: In addition to replicating the social-learning condition, positive and negative explicit conditions were tested. The most explicit positive and negative conditions were the same as the implicit condition except that before leaving the room the model added that the child should or should not do ‘exactly what I did’, depending upon the condition. These two variables yielded three basic findings. First, very few of the three- and four-year-olds reproduced the model’s performance in the implicit condition. Second, the

fours imitated somewhat more than the threes. There was no difference in the reproductive behavior of the threes and fours in the explicit conditions, whether positive or negative. Third, almost all threes and fours reproduced in the positive explicit conditions. Almost none of the threes and fours reproduced in the negative explicit conditions. Taken together, the force of these findings is to seriously question the behavioral claim of social learning theory that imitation is an arational, automatic process of acquisition and performance. These findings also suggest that even the Piagetian hypothesis underestimates the constructive rationality that guides the young, intuitive child’s reproductive activity. Now let us consider the mechanism claim of social learning theory, namely, that imitation is the mechanism for the acquisition of novel behaviors and growth. With respect to progress from one developmental stage to the next, the claim rests empirically upon the findings of a study by Bandura and McDonald (1963). Taking off from an early work by Piaget (1968) on the ‘Moral judgment of the child’, Bandura and McDonald categorized children’s moral judgments into predominantly low judgments based upon consequences or predominantly high judgments based upon intentions. They tested fiveto eleven-year-odds on pairs of stories. Using modeiing as a training technique, ~andu~ and McDonald report about a 30% change in the children? judgments from their predominant category of response to their minor category of response. In a replication and extension of this study (Cowan, Langer, Haevenrich and Nathanson, 1969) two major findings emerged. First, it was extremely difficult to categorize many of the children’s judgments on the posttest. It was only possible to classify clearly the judgments based upon consequences as low. The rest of the judgments ranged vastly in their characteristics. These judgments were unclassifiable using the Bandura and McDonald two-category system. This makes much sense in light of Kohlberg’s (1963) extensive reanalysis of moral developments as involving many more stages and aspects than phenomenal consequences and intentional motivation. Second, even when we used the invalid procedure of arbitrarily assigning high scores to the children’s uncategor~zable judgments, the finding is a change of only about 30%. Almost 50% of the children’s judgments remained at their initial pretest level. These findings suggest that the children were confused by the apparent discrepancy between their own judgments and those of the model. This does not mean that their cognitive organization was transformed by imitation. Rather, it poses the possibility that such discrepancy may feed back different conceptual knowledge. That is, the presented discrepancy plus the (inferred) childrens’ confusion may provide the appropriate interactive conditions for figurative accommodation to anomolous social information. In turn this may perturb the existing cognitive organization if the developmental jump is a small elaboration. From the present epistemolo~~al perspective, then, modeling becomes a method of presenting conceptual objects at a less advanced, equal or more advanced level than the child’s own cognitive level. The purpose is to determine whether these conceptual objects

18

Jonas Langer

have any implications for the subject’s cognitive activity. This purpose is twofold. The first is to determine the effects of disparity and nondisparity upon the subject’s cognitive progress and regress. The second IS to determine the optimal range of disparity for generating intrinsic disequilibrium that may lead to progress. We already know that modeling conceptual disparity results in small progress in some subjects’ production of moral concepts (Turiel, 1966, 1969). We also know (Kuhn, Langer, Kohlberg and Haan, in press; Rest, Turiel and Kohlberg, 1969) that when presented with reasoning by a model, subjects are able to recapitulate in their own words the model’s reasoning as long as it is below the level or the same level as the subject’s stage. They are able to recapitulate some but not all reasoning one stage above their own stage. Subjects are not able to recapitulate reasoning two stages above their own stage. One general conclusion that may be drawn from these findings on the effect of modeling on production and recapitualation is that the subject can imitate, with some fidelity, reasoning that is not new to him. However, he tends to assimilate reasoning that is new to him (that he has not himself previously produced) to his own structural stage. Thus, these findings seriously question (even more so than the findings of the Cowan et al., 1969 study) the mechanism claim of social learning theory that imitation leads to the development of novel moral behavior. We still did not know, however, whether modeling conceptual disparity would also result in some small progress when the concepts are more clearly the direct products of operative assimilation. This required investigating the implications of figuratively accommodating conceptual disparity for the production of purely logical concepts, rather than the more social concepts involved in moral judgments. The first investigation focused upon classification (Kuhn, 1969, 1972). FOUL- to eight-year-olds were divided into four groups and were presented with models dealing with classification problems (taken from Inhelder and Piaget, 1964) at different levels (--1 , 0, +l , +2) in relation to the children’s predominant level. Very little change was found. To the extent that change was observed, the children who were presented with small conceptual disparity in the progressive direction (tl) were most affected. Substantial progressive disparity (+2) resulted in even less change. And most of that change was not to the models’ t2 level but to the tl level. This particular finding is especially significant for the understanding of imitation because these subjects had not observed +l performances. Small regressive disparity (-1) resulted in the least change. Thus, the findings so far on both moral and logical concepts suggest that a model may present reasoning to a child whose form and content is novel to the child. If the form of the modeled concept is at the same level as the child’s conceptual stage (0), then novel content transfer is possible. If (a) the form is at a more advanced level than the child’s stage and (b) the structural disparity is within the limits of figurative accommodation (+I but not +2), then (c) interactive conditions for organizational progress are present. The central developmental and educational problem then becomes that of working out in a more refined fashion the limits of interactive disparity that can be transformed into

Interactional aspects

intrinsic disequilibrium (where modeling presenting conceptual disparity). 4. Appearance

becomes

ofcognitive

organization

one of the figurative

techniques

19

for

and reality

Imitation comes closest to the definition of extractive activity as direct representation of empirical knowledge, whether social or physical. A less direct but still extractive means for obtaining information about the empirical character of the environment is to compare the way things look with the way they really are. Consider the horizontal-vertical illusion. The vertical strip looks ionger. But if you take the strips apart and place one on top of the other you see that they are really equal in length. This is, then, an empirical way of evaluating the validity of judgments and may therefore play an important role in the logical concepts formed. As an initial approximation it seems reasonable to make three structural comparisons between the characteristics of figuratively accommodating configurations, such as perceptual illusions, and operatively assimilating events, such as physical deformations. These three refer to presentational, judgmental and ontogenetic features of cognitive structures (cf., Piaget, 1969): A. As environmental or experimental presentations, illusory configurations do not involve any actual deformation of a property of a static, single-state configuration. As presentations, deformation events, such as those presented in the well-known conservation tests, always involve two states in which the initial state is deformed into a resultant state. The presentation is a dynamic, multiple-state event. B. As judgments, illusions are the product of the accommodatory distribution of visual fixations upon the configurations. The mental products are percepts, such as the percept that one line is longer than another. Such percepts are probabilitic: The magnitude of the illusory effect fluctuates from moment to moment, and the subject is never completely certain of the accuracy of his judgment. As judgments, conservations are the product of assimilatory operations. Such concepts are necessary in the sense that the judgments are all or none - either the lines must logically be the same despite the observed deformation or they do not have to be the same. The judgmental difference between illusions and conservations is consistent with the comparative findings (Piaget, 1969) that: (1) Five-year-old children perceive one of two lines as a bit longer in a single-state illusion task when conceiving their length as unequal in a parallel multiple-state conservation task; while (2) eight-year-olds perceive one of the lines as much longer in an illusion task but conceive their length as equal in a parallel conservation task. Thus, the perceptual illusion increases during the same ontogenetic period in which the conceptual distortion disappears. C. As ontogenetic phenomena, illusory judgments persist throughout life. The change is merely quantitative, that is, the magnitude of some increase with age while that of others decrease. Illusory judgments persist notwithstanding the distinction that is

20

Jonas Lunger

acquired at about seven between subjective appearance and empirical reality. The illusory experience persists in the horizontal-vertical illusion, for example, even when the observer is fully aware that he is viewing an illusion. As ontogenetic phenomena, judgments of nonconservation are no longer produced after the equilibration of concrete logical operations at about eleven years old. The change is qualitative, that is, conceptions of nonconservation are completely rejected as illogical, and conservation concepts are substituted as logically necessary and universal. Nor is there any persistence of the subjective experience of nonconservation. This analysis of the functional structures of illusory and deformation phenomena was the basis for testing a series of experimental conditions @anger and Strauss, 1972).* They were designed to examine the systematic relations of (1) the acquisition of the empirical distinction between appearance and reality, with (2) the development of the logical operations of identity. The results are most pertinent. First, children who were diagnosed on the pretest as at the stage of intuitive operations did not change their nonconservation concepts. Successful training in discriminating between the appearance and reality of (a) perceptual illusions or (b) conceptual distortions did not induce progress in conservation judgments in the training group as compared with a control group of intuitive children. Second, some children who were diagnosed, on a pretest, as at a transitional stage (between the intuitive and the concrete operational stage) seemed to profit from the training procedure. Training may lead to advances in understanding the relations between the subjective appearance and the empirical reality of physical phenomena. However, such an advance in accomodatory figurations does not induce parallel progress in the assimilatory operations of intuitive children. It seems rather to feed back to the nascent concrete operations of some transitional children who have already developed partial concrete operational structures. Of course, the claim can be made for many types of training strategies that intuitive children are less susceptible and transitional children are more susceptible to cognitive change (cc, Strauss, 1973). Nevertheless, it is of no small interest to our considerations that one of the training strategies which may interact with the assimilatory operations of the transitional child, if not the intuitive child, involves accomodatory figurations.

5. Prediction

and outcome

A partially related and a most promising figurative means of accommodating empirical information is to evaluate hypotheses in light of data. Such verification activity often leads to reformulation in our empirical knowledge. Furthermore, it seems plausible that such empirical knowledge feeds back to the organization of assimilatory operations, leading in certain crucial instances to conceptual reorganization, Thus, viewing certain 2. The reader concerned with the literature on appearance/reality will recognize the difference

between the present position and Shanks (1965a, 1965b).

and that of Braine

Interactional

aspects of cognitive organization

situations may mislead the child into focusing upon empirical variables that he would be better off excluding from his logical calculations in certain conceptual situations. The reason would seem to be that the character of the presentation leads to the figurative accommodation of empirical misinformation that feeds back to the child’s operations. Some possible methodological consequences are that screening the pouring of liquid in a conservation task may facilitate conservation conception (Bruner, 1966) and presenting class-inclusion problems in a verbal rather than pictorial form may facilitate part-whole reasoning (Wohlwill, 1968). In order to look at the independent and interactive effects that these two factors prediction/outcome and screening - might have upon the operative conception of conservation, four training conditions were investigated (Strauss and Langer, 1970): Prediction and screening; prediction and no screening; prediction, outcome and screening; prediction, outcome and no screening. The relevant findings were: (1) Whatever progressive change occurred in the intuitive children as a result of the various experimental conditions, the control children have caught up by the second posttest. (2) The transitional children seem to profit from the various conditions and maintain some advantage over the control children through the second posttest. (3) Prediction and outcome experience without screening seems to be most effective with both intuitive and transitional children. (4) Screening without prediction and outcome experience is not effective with intuitive children; it may be effective with some transitional children. Thus, the prediction and outcome activity seems to be more influential in changing assimilatory operations than screening. The implications of prediction and outcome experience for the alteration of operative assimilation seems to be limited to the transitional child who has already partially developed concrete operations. It does not seem to change the assimilatory operations of the child who is at the stage of intuitive operations. Experience in comparing predictions with outcomes and in relating appearance with reality are, at least in part, ways of teaching children empirical verification procedures. The evidence so far suggests that if they have any implications for the formation of assimilatory operations they only influence the disequilibrated, that is, partial functional structures of transitional children. There are at least three plausible reasons why the purely intuitive children in these experiments were not affected. First, the verification procedures do not involve active or spontaneous experimenting on the part of the subject but rather only permit following what the trainer is doing. Second, the verification procedures do not lead to any strikingly observable anomalies that would perturb an intuitive child’s existing accommodatory figurations. Third, the intuitive child’s functional structures are not competent to appreciate the cognitive significance of the empirical verification procedures he is experiencing. A study by Coie (1969) provides some preliminary hints on these issues. Older (elevenyear-old) but not younger (seven-year-old) children seem able to profit from active verification procedures. The older children’s level of verification activity is more advanced

22

Jonas Lunger

on a somewhat anomalous event, regardless of whether they predicted correctly or incorrectly. The more advanced their level of verification activity, the greater the progressive change in their subsequent level of explaining the event.

6. Symbolic medium So far we have focused upon three means of sccommodatory figurations that may be considered as primarily systems for extracring empirical information. We have paid special attention to some of the implications that these means may have for assimilatory operations. Now we will turn to some parallel considerations, but we will focus upon a means of accommodatory figurations that is primarily representational. Previously (Langer, 1969b) we reported some work on the interaction between the medium in which a conceptual problem is presented and the medium in which the child represents the problem and thinks about it. Here we will also be concerned with the unidirectional implications of the medium in which a conceptual problem is presented for the child’s assimilatory operations. To begin with let us consider some research on the development of classificatory activity. Jennings (1969) examined class-inclusion reasoning when the presentation of the objects to be considered were embodied in either the pictorial or the verbal medium. The pictorial condition took the form, for example, of presenting subjects with pictures of six roses and two violets and asking ‘Are there more flowers or more roses?’ In the comparable verbal condition the subjects were zimply asked ‘If I had six roses and two violets, would I have more flowers or more roses?’ Kindergarten (x= 5; 11 years) and first-grade (x = 7; 1 years) boys were tested. The childrens’ classificatory reasoning was examined for both the operational level of their solutions and their explanations. Consider their solutions first. Kindergarten children produced more correct concrete-operational type solutions to the verbal than to the pictorial forms. Although the difference is not statistically significant, it is in the direction of replicating Wohlwill’s (1968) findings. However, the frequency of correct concrete-operational type solutions produced by first graders is about the same for the pictorial and verbal forms. Now consider the childrens’ operational level of explanation. Only the first graders produced concrete-operational explanations for their solutions. Concrete-operational explanations occurred more frequently on the pictorial than the verbal presentations. A child’s explanation was judged to be concrete operational in accordance with the Inhelder and Piaget (1964) criterion of a solution that is justified in terms of subordinate/superordinate or part/whole relations, such as ‘cuz dogs and horses are both animals’. Finally, consider these two results together. It appears that the medium of presentation influences the operational level of the explanations of seven-year-olds but not that of their solutions. Conversely, the medium seems to influence the level of solutions of

Itlteractional aspects

ofcognitive

organization

23

kindergarten children but not that of their explanations. There are numerous methodological problems with the studies mentioned so far. A central methodological problem for the theory of cognitive development is to distinguish between (a) the presentational medium in which conceptual problems are posed and (b) the symbolic medium in which the cognizer represents the presentation. Consider a brief illustration. The screening procedure used in some investigations (e.g., Frank in Bruner, 1966) were designed, at least in part, to vitiate ikonic representation by the children and to induce them to symbolize verbally. It is true that they cannot perceive the water level behind the screen. But there is no way of knowing that the children are not imagining the water level; and imaging is a mode of ikonic representation. As a matter of fact screening is the type of technique used by Piaget and Inhelder (1963, 1966) in order to study the child’s mental imagery. We know from their extensive investigation of this problem that the visual imagery which the child produces is usually no more advanced than his operational stage of competence. Consequently, if the screening technique leads the intuitive child to figurative imagery then it should not have a facilitative effect upon his assimilatory operations. Recall that this is precisely what we found for intuitive children on the screening without prediction and outcome condition: It had no progressive effect on operational conservation reasoning (Strauss and Langer, 1970). The issue, then, is a delicate one that has important theoretical ramifications, particularly with respect to the organizational relations between assimilatory operations and accommodatory figurations. It poses delicate problems because it is immensely difficult to control the medium of representation in which the child thinks. The ideal empirical approach would be to systematically investigate the consequences for cognitive competence and performance (i.e , comprehension, production and appreciation; cf, Langer, 1969~) of the intersection of the: (A) Medium of presentation - pictorial, practical, gestural (signs), verbal, notational, with the (B) medium of representation - graphic, gestural, imaginal, verbal, notational. The intersection of the presentation and symbolic media would then be considered in relation to the structure of the problem: (C) Figurative - empirical (D) operative - logical. It should, of course, be remembered that few problems are purely figurative or operative but can be posed so as to be predominantly one or the other. An earlier paper @anger, 1969b) reported some of our work along these lines. To give a brief additional illustration, consider part of one follow-up study conducted by Schwartz (1970). The study investigated the effect of the child’s drawing the verbal presentation of a set of objects prior to questioning versus after questioning. The questioning was on his logical conceptualization of a class-inclusion problem about the objects. What is most interesting to note for our purposes is that sixth-grade subjects seemed to correctly solve the problem at a concrete-operational level almost twice as

24

Jonas Langer

frequently if they drew the verbal presentation before being questioned about it (17 out of 20 Ss) than if the drawing followed questioning (10 out of 20 5’s). Such results suggest that the medium of presentation and representation does not have the primary significance for intellectual development attributed to it by Werner and Kaplan (1963) and Bruner et al. (1966). On the other hand, the forms of symbolic activity seem to have greater significance for the development of assimilatory operations than that attributed to them by Piaget (e.g., Piaget and Inhelder, 1966). In general, operative assimilation is the primary system of intellegence, and figurative accommodation is the secondary system. As such, operative assimilation should have greater implications for figurative accommodation than the reverse. As we have seen, however, the medium of presentation has some significant differential influence upon the solution of and reasoning about crass-inclusion problems depending upon the age of the child. Furthermore, the medium of representation effects the formation of logical class-inclusion concepts depending upon the child’s stage of producing such concepts.

7. ConcI~ions The findings summarized here and elsewhere (Langer, 1969b, 1970a, 1970b) permit the formation of a number of hypotheses on the development and organization of functional structures. In turn, they lead to certain hypotheses about education. The d~e~~~~e~~a~ hypothesis is that the organization of the child’s stage sets Limits to his interaction with his environment. This means that the or~a~zation of functional structures at each stage is resistant to extrinsic inducement to alterations in its intrinsic stage sequence and rate of development. Extrinsic experience and training should have minimal effects upon progressive or regressive development. To the extent that the child assimilates extrinsic inducements to change, he is most susceptible to movement in the same direction as his natural progress to one stage ahead of his predominant stage level. Reciprocally, (a) the child is most resistant to regressive movement in the opposite direction of his natural intrinsic development, that is, to one stage behind his predominant stage level, and (b) the child cannot skip any of the stages in the sequence of development. This developmental hypothesis also implies that the functional structures of a given stage must be at least partially present for interaction to lead to learning, in the triple sense of consolidating, elaborating and generaliz~g these functional structures. This is the reason why empirical training such as relating predictions to outcomes and comparing appearance to reality remains extrinsic inducement for the intuitive child. Such empirical training can have little implication for his development of concrete operations. It is also part of the reason why such empirical training may lead to some progress in the transitional child since he is in ~sequ~ibrium, and he has already partially developed the functional structures of the more advanced stage.

Interactional aspects

of cognitive

organization

25

The organizational hypothesis is that cognition is a multi-leveled network of functional structures. The first corollary hypothesis, that this network is composed of two major sets of functional structures, assimilatory operations and accommodatory figurations, has already been considered at the outset of this essay. The second corollary hypothesis, that the network of functional structures is multileveled, has not yet been described directly. This is the hypothesis that ‘stage mixture’ (Turiel, 1969) or mixture of functional structures is the intrinsic source of development &anger 1969b, 1970b). Not all the functional structures are at the same developmental level. Some accommodatory figurations may be more developed than other accommodatory figurations and assimilatory operations. More important for cognitive development, some assimilatory operations may outstrip and regulate other functional structures. This constant state of intrinsic disequilibrium or dynamic equilibrium is the source of self-generated development. A third corollary hypothesis is that the network of functional structures is not only multi-leveled, but it is also an open gridwork (Langer, 1970b). The gridwork radiates outward to interact with the physical and social environment. For structural and adaptive reasons this interaction leads to discrepancies between parts of the gridwork and parts of the environment. This constitutes the interactive basis for organizational disequilibrium. The optimal conditions for progress would then be an intersection of intrinsic and interactive disequilibrium. These conditions of relating structural form to empirical content seem to be met when (1) the person is in a transitional phase, that is, a multi-leveled state of functional structures, and (2) the person is engaged in a variety of figurative accommodatory activities that provide him with empirical information. It is the feedback and feed-forward between form and content that leads to developmental and organizational change. An educational hypothesis flows from this developmental and organizational analysis. Children and adolescents diagnosed to be in a transitional phase, where they have already partially developed more advanced funtional structures than those of their dominant stage, are somewhat susceptible to progressive elaboration of their more advanced functional structures via a variety of empirical training procedures. But the findings point in the direction that the progress can only be a small step and is probably not very stable, at least when the procedures are short-term as in the studies described here. An important educational consideration is the apparent discrepancy between the genotypical potential for cognitive development and its phenotypical actualization. The theory of cognitive development asserts genotypic development to a stage of formal, logical operations and principled moral judgments @anger, 1969a, 1970~). The crosscultural data on moral development indicate that individuals in all cultures actually develop to the level of conventional morality that precedes the stages of principled morality (Kohlberg, 1969). The rate of attainment varies somewhat between cultures. Fewer than a third of all people progress further to the level of principled morality. The cross-cultural data on logical development indicate that individuals in all cultures develop

26

Jonas Langer

to the level of concrete operations that precedes the stage of formal operations (Dasen, 1972). Again, the rate of progress varies somewhat between cultures. It has not yet been ascertained precisely what proportion of the population continues progressing to the stage of formal operations, but it is likely that not more than half do so (Kuhn, Langer, Kohlberg and Haan, in press). Apparently, then, all people actualize their potential for concrete logic and conventional morality regardless of whether their rate of progress is a bit faster or a bit slower. More significant, and much more disturbing, is the fact that most people do not actualize their potential for principled morality and, possibly, formal logic. This fact points to an obvious educational conclusion. The primary goal is not to focus upon accelerating the rate of developmental progress. The goal is to focus upon actualizing adolescents’ genotypic potential for formal logical and principled moral thinking and action. A promising approach is to build upon the adolescents’ competence to benefit from active verification procedures and his intrinsic state of structural disequilibrium. It just might turn out that children will actualize to the concrete operational and conventional moral stages regardless of the education they obtain. On the other hand, the kind of education adolescents are given might have a much greater effect upon their development - particularly whether they actualize their potential for formal operations and principled morality.

REFERENCES

Bandura, A., and McDonald, F. J. (1963) Influence of social reinforcement and the behavior of models in shanina children’s moral judgments. J. abn. sol. Psychoi., 67, 274-281. -and Walters, R. (1963) Social learnirgand personality development. New York, Holt, Rinehart & Winston. (1969) Social-learning theory of identiticatory processes. In D. A. Goslin (Ed.), Handbook of socialization. Chicago., Rand McNally. Bower, T. G. R. (1967) The development of object permanence: Some studies of existence constancy. Pert. Psychophy., 2, 411--418. Braine, M. D. S., and Shanks, B. L. (1965a) The development of conservation of size. J. verb. Learn. verb. Beh., 4, 227-242. and Shanks, B. L. (1965b) The conservation of a slope property and a proposal about the origin of the conservations. Can. .I. Psychol., 19, 197-207.

Brainerd, C. J., and Allen, T. W. (1971) Experimental inductions of the conservation of ‘first-order’ quantitative invariants. Psychol. Bull., 75, 128-144. Bruner, J. S. et al. (1966) Studies in cognitive growth. New York, Wiley. Cassirer, E. (1944) The concept of group and the theory of perception. Phil. phenom. Res.. 5, l-35. _ _ -- (1953) Philosophy of svmboli: forms. New Haven, YaleVUniversity Press.- Originally published in 1923. Coie, J. D. (1969) An ontogenic study of the use of explanation and related verification activities. Unpublished doctoral dissertation, University of California, Berkeley. Cowan, P. A., Langer, J., Heavenrich, J., and Nathanson, J. (1969) Social learning and Piaget’s cognitive theory of moral development. J. Pers. sot. Psychol., 11, 261-214. Dasen, P. A. (1972) Cross-cultural Piagetian research: A summary. J. cross-cult. Psychol. 3,23-39.

Interactional

Inhelder, B., and Piaget, J. (1964) Early growth of logic in the child. New York, Harper & Row. Jennings, J. R. (1969) The effect of verbat and pictorial presentation on class inclusion competence and performance. Unpublished manuscript, University of California, Berkeley. Kohlberg, L. (1963) The development of children’s orientation toward a moraI order. Vita Hum., 6, 11-33. (1969) Stage and sequence: The developmental approach to socialization. in D. Go&n (Ed.), Handbook of socialization. New York, Rand McNafiy. Kuhn, D. (1969) Patterns of imitative behavior in children from 3 to 8: A study of imitation from a cognitive-developmental perspective. Unpublished doctorai dissertation, University of California, Berkeley. -(1972) Mechanisms of change in the development of cognitive structures. Child Devel., 43,833-844. -and Langer, J. (1968) Cognitive devefopmental determinants of imitation. Unpublished manuscript, University of California, Berkeley. -Langer, J., Kohlberg, H., and Haan, N. (in press) The development of formai operations in logical and moral judgments. J. gen. Psychol. Mono. Langer, J. (1964) Implications of Piaget’s talks for curriculum. J. Res. Science Teach., 2, 208. 213. (1969a) Theories of development. New York, Halt. Rinehart & Winston. (1969b) Disequilibrium as a source of development. In P. H. Mussen, J. Langer and M. Covington (Eds.1, Trends and issues in developm>rttal. psychology. New York, Holt, Rinehart &Winston. (1969c) (1969c) Some structural aspects of the development of moral conduct. Paper presented at the meetings of SRCD. (1970a) Werner’s comparative organismic theory. In P. H. Mussen (Ed.), Gzrmichael’s manual of child psychology. New York, Wiley. (19’70b) Mental regeneration. in M. Wohns and M. Gottesman (Eds.), Group care. New York, Gordon. (1970~) The development of the individuaf. Chapter 10 in Psychology today: An introduction. Del Mar, California, CRM.

aspects

of’ wgnitive organization

27

-- - and Strauss, S. (1972) Appearance, reality, and identity. Cog., 1, 105--128. Michotte, A. (1963) The perception of causalify. New York, Basic Books. Miller. N. E.. and Dollard. J. (19411 Social learning and imitation. ’ New’ Haven, Yale University Press. Plagct, J. (1950) Psychology of intelligence. New York., Harcourt, Brace & World. (1951) Play, dreams and imitation in childhood. New York. Norton. (1967) Biologie et ‘connaissance. Paris, Galhmard. -- (1968) The moral judgment of the child. New York, Free Press. Originally published in 1932. (1969) The mechanisms of perception. London, Routledge & Kegan Paul. Originally published in 196 1. and Inhelder, B. (1969a) Mental images. In P. Fraisse and J. Piaget (Eds.), h’xperimental psychology: VII. Intelligence. New York, Basic Books. and Inhelder, B. (1969b) Intellect~Iral operations and their development. In P. Fraisse and J. Piaget (Eds.), Experimental psychology: VII. Intelligence. New York, Basic Books. and Inhelder, B. (I 966) L ‘image mentate chez i’enfant. Paris, P.U.F. and Inhelder, B. (1968) Memotie et intelligence. Paris, P.U.F. Rest, J., Turiel, E., and Kohlberg, L. (1969) Level of moral development as a determinant of preference and comprehension of moral judgments made by others. J. Pers., 37.225-252. Schwartz, C. (1970) Developmental aspects of class inclusion. Unpub~sh~d doctoral dissertation, University of California, Berkeley. Sinclair, H. (1967) Acquisition du langage et development de la pensee. Paris, Dunod. Strauss, S. (1973) Inducing cognitive development and learning. Cog., 1, 329-357. and Langer, J. (1970) Operational thought inducement. Child Devel., 41, 163-175. Turiel, E. (1966) An experimental test of the sequentiality of developments stages in the child’s moral judgments. J. Pers. sot. PsychoI.. 3,611.-618.Turiel, E. (1969) Developmental processes in the child’s moral thinking. In P. H. Mussen, J. Langer and M. Covington (Eds.), Trends and issues in developmental psy-

28

Jonas

Langer

cho[ogy. New York, Holt, Rinehart & Winston. Werner, H. (1948) Comparative psychology of mental development. New York, International Universities Press. Werner, H., and Kaplan, B. (1963) Symbol formation. New York, Wiley. Wohlwill, I. F. (1968) Children’s responses to class inclusion questions with verbally and pictorially presented items. Child Devel., 39,449-467.

Zaporotchets, A. V. (1965) The development of perception in the preschool child. Mono. sot. Rex child Devel., 30 (2), 822 101. Zazzo, R. (1953) The behavior of new-born anencephalics with various degrees of anencephaly. In J. M. Tanner and B. Inhelder (Eds.), Discussions on child develop men& Vol. 1. New York, IUP.

RbumP Dans cet article I’auteur avance des propositions pour completer la theorie du developpement structural du changement cognitif que Langer a prksente en 1969. L’analyse se centre ici sur a) l’organisation des operations d’animation et des representations d’accomodation; b) les coordinations intrinseques entre les approches cogni-

tives theoriques et empiriques respectivement construites avec les deux formes de structures fonctionnelles; c) les changements dans le developpement cognitif dus aux coordinations intrinseques. Des hypotheses theoriques d&iv&es sont etudies dans le cadre des ddcouvertes resumees ici et ailleurs.

2

Conservation

of weight in infants*

PIERRE MOUNOUD University

of Geneva

T. G. R. BOWER University

of Edinburgh

Abstract Conservation of weight can be defined as the ability to affirm that the weight of an object remains invariant during the transformations of the form of the object. It is known to be achieved at a conceptual level at about 9 years of age. The behavior of infants seems to indicate that between 6 and 18 months of age they develop a sensorimotor form of conservation.

1. Introduction Piaget (1937, 1941a, 1967) has described the development of the concept of conservation in children between 4 and 14 years of age. At the beginning of this period children are incapable of affirming that the volume, weight or substance of an object is independent of its arrangement in space. If such a child is shown two identical balls of plasticine, he will agree that there is the same amount in each and that they weigh the same. If one ball is then rolled out into a sausage shape, with the child watching the transformation, the child will typically say that there is more plasticine in the sausage than in the remaining ball and that the sausage will weigh more than the ball because it is longer or that there is less plasticine and the sausage will weigh less because it is thinner. The child does not realize that weight, volume and substance are invariant under transformation of shape. At the end of this stage of development, children are aware of this invariance during transformations. These acquisitions originate in the sensorimotor behavior of the baby. The baby elaborates through his actions what the child between 4 and 10 years elaborates by means of his thought processes (Piaget, 1937, 1941a, 1967). Consequently, it is important to study the way in which this construction is effected at these more fundamental levels (Mounoud, 1971,1973). * This research was supported by M. R. C. Grants Nos. G969/559/C and G912/982/C. Cognition 3(l), pp. 29-40

30

Pierre Mounoud and T. G. R. Bower

Conservation or the lack of it must have effects on the success or failure of simple, everyday behavior. Consider the act of picking up an object and transporting it to the mouth, an act that children engage in from the age of six months or so. For that act to be performed successfully, the child must adjust the force of his grip to the weight of the object - if the grip is too light, the object will slip out of the hand; if too strong, the object may be crushed. In order for the baby to transport various different objects to his mouth with accuracy, he must appropriately adjust the tension of his arm muscles to the weight of the objects. He must relate the variations in size and form with variations in weight and also must recognize that transformations of observable aspects of the object or of the distribution of its elements do not entail a modification of its weight. Most children in most cultures play with plastic substances ~ playdoh, plasticine, flour and water or plain clay - whose shape changes in the course of the play, leaving weight, volume and substance unchanged; every time this happens the child is faced by a conservation problem at a behavioral level. While conservation of weight at the concrete operational stage is evidenced at 8 or 9 years, casual observation would indicate an awareness of conservation in behavior well before that age. The experiments to be reported here were designed to make such observations in a systematic way. The results indicate that conservation, realized through actions, is achieved during infancy by the age of 18 months. Conservation of weight was selected for study. As was mentioned above, accurate transport of an object requires at least two adjustments to the weight of the object, the degree of muscular contraction of the arm and the force of the grasp. Before testing for conservation of weight it is necessary to ensure that the relevant behaviors have certain other characteristics. First of all the subject must show a differentiated response to weight; if a subject simply applies the same pressure or arm tension every time any object is presented, giving the same response on every occasion regardless of weight, there is no point in testing for conservation since conservation behavior would necessarily appear. Suitable response differentiation can be demonstrated by adaptive changes in response to a single object. The first time anyone, even an adult, is presented with an unfamiliar object which is not part of a set, there is no basis on which to judge the weight of the object. The weight might be overestimated or underestimated, but only by purest fluke could even an adult correctly gauge the weight of such an object. On repeated presentations, however, force of grasp and arm tension should adjust to the weight of the object. To be relevant to conservation, such adjustments must be anticipatory and based on visual information. Thus the improvement in performance must be specific to objects which can be identified as the same on the basis of visual information, so that anticipation of weight is made prior to actual manipulation of the object. However, adapted responses to a single familiar object do not demonstrate sufficient capacity to make conservation testing worthwhile. Before beginning conservation testing it is necessary to establish that the adaptation is based on visual size. An infant could adapt to a single object, demonstrating perfect behavior, by recognizing the object on the

basis of pattern or markings on it. If the relevant pattern or mark were invariant under the shape transfarmation, then we would necessarily obtain conservation behavior with no need for attainment of the concept that weight is invariant under shape transformations. Only if the visual basis for adapted behavior is seen size is conseffation testing meaningful. One can only demonstrate that seen size is the basis of adaptation to the weight of a single object if there is some transfer of adaptation to a new object of the same material but different size. If the baby adapts to an object of size x and weighty, one could then give him an object of size 2x and weight 2y. Ideally there would be no error at alf on first presentation of 2x, if seen size were the basis of the adaptation, An acceptable criterion of performance would be that the error on first presentation of 2x be no greater than the error on the last presentation ofx. A minimal criterion would be that the error with 2x for a group given experience with x would be that they showed an error indicating greater expectation of weight than a group given no previous experience ofx. In the context of behavior this could mean a lessened drop of the arm on taking 2X. If these criteria are met one can proceed to test for conservation behavior. A paradigm for test would be to adapt the subject to a particular object and then to transform the object. If there is conservation of weight, the first response after transformation should ideally show no errors, or no greater error than the last response prior to transformation. To summarize then, it is necessary before testing for conservation to ensure that the subject is capable of differentiated adaptive responses to weight, which are anticipatory and based on seen size. If these criteria are met one can proceed to conservation testing. In the context of the two adjustments necessary for accurate transport of an object, error as used above means a drop or elevation of the arm on tak’lng an object or else application of too little grasp force to hold an object or more force than is required to hold the object.

2.1 Subjects Six groups of five infants experiment.

aged between

6 and

Ifi months

served as subjects

in this

2.2 Procedure The objects used in the arm tension experiment were a series of brass rods, all 2.5 cm in diameter, of length 2.5 cm, 5.0 cm, 7.5 cm and 10.0 cm, whose weights were 110 grams, 220 grams, 330 grams and 440 grams, respectively. They constituted the seriation set. The third cylinder in this series (330 gr.) was paired with a visually identical hollow cylinder weighing 100 gr. There was also an object which could be transformed, a

32

Pierre Mounoud and T. G. R. Bower

15 .O cm high 2.5 cm diameter brass rod, weighing 550 grams, hinged in the center so that it could be doubled over and locked to make a double rod 7.5 cm high. An additional transformable object that was occasionally used consisted of a lump of playdoh that had lead bearings concealed in it. This object could be rolled into a ball or a sausage without revealing the lead. Its weight was 250 grams, its volume 50 ccs. Subjects were initially given the seriation set in ascending and then descending order (item 1). Each object was presented several times in a row, followed by the next in the series, again presented several times in a row. Objects were presented by hand in such a way that the infants were forced to reach out to take them. The arm could thus drop, raise or rest stable. The item involving the illusory identity between the two cylinders (item 2), consisted of presenting three times in succession the heavy cylinder (item 2a) and immediately after presenting the hollow cylinder (item 2b). (Response to this sequence tells us whether or not the infant expects visually identical objects to weigh the same.) After this substitution item has been presented, the transformable object was introduced, either fully extended or doubled (item 3). After three presentations it was transformed with the infants watching and then presented again. This terminated the experiment, save for a few infants who were given a conservation test with the playdoh object. Behavior was recorded on videotape. The measure adopted was the amplitude of hand drop or hand elevation on presentation of an object, measured by comparing the position of the hand on taking the object with its position 250 msecs later. The time interval was chosen to ensure that we were obtaining a measure of anticipation, our assumption being that 250 msecs was too short a time to allow for recovery from an initial error. Response to item 1 was intended to tell us whether the baby was capable of adapted, differentiated responses to weight. Items 2a and 2b were intended to tell us whether or not these responses were cued by visual size, while item 3 was the conservation test.

2.3 Results The arm drop measure worked well, except for infants of 11-13 months (see Figs. 1,2 and 3). We would expect that if the infants were able to adjust their reaching and grasping behavior for the same object when it is given several times in succession, there would be a diminution in the amount of arm drop or arm elevation between the first and last presentations of the same object. Table 1 gives the results for item 2a. They indicate that at all the ages there was such an adjustment.’ 1. Rather than taking the object and then transporting it, as the younger and older babies do, infants in the age range 11-13 months integrated taking and transport into a single movement. This made it impossible to use the simple arm-drop measure. A measure based on path

and speed of movement would be required, and it would be difficult to give a simple quantitative measure of the aberrant trajectories produced. It is not therefore possible to include the data from this age group in Table 1.

33

Conservation of weight in infants

Table 1.

Age

Mean drop (-) or elevation (+) on first presentation of object (item 2a)

Mean drop (-) or elevation (+) on third presentation of object (item 2a)

6 months 7 months 8 months 9.5 months 15 months

-50 -35 -10 -20 -30

-20 -10

_

0 +15 -15

t* 5.0 6.25 4.0 9.84 4.75

P <.005 <.005 <.Ol <.OOl <.005

*df=4

The next question asked concerned the basis for adjustment. If the babies were using the visual appearance of the object as a basis for inferring its weight and thereby utilizing a rule that the same object weighs the same each time it is presented, we would expect that presentation of item 2b, visually identical to but lighter than 2a, would result in an effective elevation of the arm when compared to the elevation or drop occurring on the third presentation of item 2a. Table 2 gives the results of the comparison of response to these two objects. As can be seen, it was only by the age of 9.5 months that the appearance of the object became critical in determining response to it. Only at that age did presentation of item 2b result in significant elevation of the arm. The seemingly successful adjustments of the- younger infants were therefore not based on the use of visual indices. It is possible that they were based on a very subtle use of proprioceptive ones. However since these could hardly produce anticipation and the time interval between measures was too short for recovery from an initial error, it seems that there must have been some other basis for the adjustment. Inspection of their behavior indicated that these babies tended over trials to lock their arm against their body, holding the forearm tightly against the body, so that the weight of their body was acting as a counterweight to the weight of the object placed in their hand thereby cutting the possible range of elevation or drop. Thus on the criteria we have adopted, it is not possible to say whether or not these infants were showing visually based prediction of weight. However we can say that from 9.5 months on there is visually based prediction of weight for single objects, that infants above 9.5 months know that the visually same object weighs the same each time it is presented. We must turn to the seriation data to discover whether visually based prediction of weight goes beyond prediction of the weight of a single familiar object and can be based on familiarity only with the material from which an object is made, with size used as an indicator of weight, regardless of familiarity with the specific size presented. If size can be so used, we would expect the first presentation of the later objects in the seriation set to elicit lesser arm elevations or drops than first presentation of the initial object in the series. The possible criteria for size determination were set out in the introduction. Sample data, Table 3, indicates that the 9.5month-old group did not meet the two

34

Pierre Mounoud and T. G. R. Bower

stronger criteria. Unfortunately the procedure used made it impossible to use the weakest criterion. The 15-month-old group by contrast met the second strongest criterion proposed, zero decline in accuracy. This group also showed zero decline in precision after transformation in the conservation test.’

Table 2. ~Mean drop (-) or elevation (f) on third presentation of item 2a ___ -20 -10 0 +15 -15

Age 6 months 7 months 8 months 9.5 months 15 months

Mean drop (0 or elevation (+) on first presentation of item 2b -2o* -1o* 0* +40** +20***

* = NS **t=3.01,@=4,p<.05 ***t=7.0,df=4,p<.005

Table 3.

Difference between movement on last presentation presentation of third object in seriation task

Age

Difference

6 months 7 months 8 months 9.5 months 15 months

6 -16 8 20 0

2. There are interpretation testing their

difficulties in a straightforward of such zero differences and in representativeness of true mean

of second object and first

values. We will return section of Experiment

to this point in the results 2.

Figu Ire 1.

Each frame shows two arm positions. I) The position on taking the c>bject and 2) the position of the arm at the end of its first excursion. At top is the reaction to the first presentation of an object; center - reaction to third presentation of that object. Note diminished arm excursion. Bottc ,rn ~ reaction to the visually identical but lighter object (item 2b). Not e the extreme elevation shown in the bottom frame

Figure 2.

The top two frames show the response ofan injkt of 11 months to a normal object, the lower two frames the reaction of the same infant to the visually identical lighter object (item 2b). The rapid grab typical of injkts of this age means that the illusory object produces aberrant trajectory rather than a simple drop or elevation

Figure 3.

Each jkame shows two arm positions superimposed. presentation

of conservation

Top - response on first

object; center -~ response on third presentation

oj’conservation object, note diminished arm excursion; bottom - response on jirst presentation of conservation object after transformation. Note size of arm drop

38

Pierre Mounoud and T. G. R. Bower

3. Experiment

2

3.1 Subjects A total of 30 infants (6 each at 9 months, 12 months, 15 months, 18 months and 21 months) were run through the entire series. A further 24 (12 each at 15 months and at 18 months) were run with the conservation set alone, half with the long-to-short, half with the short-to-long transformation. This was done to ensure that the seriation set was not inducing a conservation error that would otherwise not have occurred. 3.2 Procedure The basic procedure and assumptions of the force of grip experiment were the same as those of the experiment just described. The seriation set consisted of 4 rods each 2 cm X 2 cm, with lengths 2.5, 5.0, 10.0 and 15.0, weighing 45,90, 180 and 275 grams. The transformable object was a 2 cm X 2 cm X 17 cm rod, weighing 280 grams, hinged in the center. The objects were placed on a table top within reach of the infant. All the rods were cloth covered. Each contained a pressure transducer made of resilistor foam to pick up force of grip. The output of the pressure transducers was recorded on a Beckman polygraph. The polygraph was calibrated so that when no pressure was applied the pens were at their maximum downward deflection. The reading for the minimum pressure necessary to hold each object was determined empirically. The basic measure taken was the initial applied pressure, defined as the maximum pressure applied within 300 msec of contact with the object. This was converted to a percentage error by subtracting the minimum pressure necessary to hold the object from the obtained pressure, dividing it by the minimum pressure and multiplying by one hundred. Our assumptions were the same as before: That anticipation that the same object would weigh the same on repeated presentations would show up as increasingly precise initial applied pressure over the three presentations of an object, that serial anticipation would show up as increasingly precise initial applied pressure with objects presented late in the series, that lack of conservation would show up as a change in the pressure applied to a transformed object which could increase or decrease depending on whether the baby centers on the variation in length or width. Conservation would appear as no effect of transformation in either direction. 3.3 Results The results for the two basic forms of anticipation are shown in Table 4. As can be seen there, sequential presentation of the same object results in improvement at all ages. However, serial prediction does not improve in the same way until 15 months. At this age the improvement for the lightest object is significant beyond the .OOl level. For the

Conservation of weight in infants

39

heaviest object the improvement just escapes significance (t = 2.42 u’f = 2 p < .l > .05). In part this negative result is an artefact since the initial response to the heaviest object was better than we would have expected by chance. If we take response to the lightest object as our criteria1 test, then we can conclude that serial prediction is within the infant’s repertoire by 15 months. Table 4. 7%err01

9

12

15

18

21

1st presentation 3rd presentation

of lightest object of lightest object

Are

400 0

300 5

200 0

200 5

200 0

1st presentation 3rd presentation

of heaviest object of heaviest object

-20 12

-30 5

-18 0

-18 0

-18 -5

1st presentation after experience

of lightest object with other 3

400

300

10

10

10

1st presentation after experience

of heaviest with other

-30

-30

0

5

object 3

-15

The basic results for the conservation test are shown in Table 5. There are problems in evaluating these responses. The expectation is that non-conservers will expect the transformed object to be a different weight from itself when untransformed. However we have no a priori reason to believe that they will think a particular transformation results in either an increase or a decrease in weight. If half of a group of non-conservers thought that the transformation resulted in an increase and half in a decrease, the mean change could be zero, a result that on the face of it would indicate conservation. Table 5. % error

months

Last presentation prior to transformation

First presentation after transformation (long-toshort)

First presentation after transformation (short-tolong)

9 12 15 18 21 15* 18*

-25 -10 -3 -4 -5 -10 -5

25 30 -43 -4 4 45 -5

44 66 97 -3.5 4 95 -5

Age

* Group

with no seriation

pretraining

As can be seen in Table 5 the problem does not arise until the infants are 18 months old. The younger infants did not show conservation behavior. Further, at 15 months infants seemed to equate weight with length, in that the applied pressure was increased when the

40

Pierre Mounoud

and T. G. R. Bower

object was elongated and decreased when the object was shortened. This was true even in the group with no seriation pre-training, in that all 6 infants given the short-to-long transformation increased their applied pressure while all 6 given the long-to-short transformation decreased their pressure. The probability of either result occurring by chance is less than .016. At 18 months, 4/6 infants increased pressure on the short-to-long transformation and 4/6 decreased pressure on the long-to-short transformation. The probability of such a result is ,454, virtually chance. Under these circumstances the mean difference is less relevant than a comparison of the two variances. If the infants saw the transformed object as different, we would expect the variance to increase as outlined above. If they saw it as the same object, weighing the same, we would expect the variance to decrease, as the judgments grew more precise. The variance for the last judgment for the 18-month-old inexperienced group prior to transformation was 91.6 whereas the variance for the judgment after transformation was 10.4 df’= 8.8, p < .Ol). We can thus conclude that the 18-monthold group showed conservation behavior.

4. Discussion These experiments indicate that infants develop a behavioral form of weight conservation, the ability to detect that weight is invariant under transformations of the shape of the object whose weight is in question, by 18 months. The sequence of development is the same as that observed at a verbal level in children between 4 and 8 years. It would seem that we are dealing with the first phase of a vertical dtcalage (Piaget, 1941 b).

REFERENCES

Mounoud, P. (197 1) Developpement des systkmes de representation et de traitement chez I’enfant. Bull. Psychol., 296, 25, 5

I. Mounoud, P. (1973) Les conservations physiques chez le b&be. Bull. Psychol., 27, 13-14. Piaget, J. (1955) The construction of reality in the child. London, Routledge & Kegan

La conservation des poids peut etre definie comme la capacitd d’aftirmer que le poids d’un objet reste inchange malgrd ses transformations morphologiques. On sait, qu’i un niveau conceptuel, cette capacitt est acquise approxima-

Paul. Original French edition in 1937. Piaget, J. (1969) The child’s conception of number. London, Routledge & Kegan Paul. Original French edition in 1941a. Piaget, J. (1967) Biologic et connaissance. Paris, Editions GaJlimard. Ch.V, St 6.11. Piaget, J. (1941b) Le micanisme du developpement mental et les lois du groupement des operations. Arch. Psychol., 28.

tivement a l’age de neuf ans. Le comportement des enfants semblerait indiquer qu’une forme sensori-motrice de la conservation est acquise entre 6 et 18 mois.

3

Acquisition

of a non-vocal

‘language’

JENNIFER

by aphasic children*

HUGHES

Medical Research Council Developmental Psychology Unit, London * *

Abstract Aphasic children with serious language deficits were taught to communicate

via a system

of visual symbols originally devised by David Premack (I 969) for use with chimpanzees. The subjects,

lacking normal language, readily learned to express several language func-

tions in this way (word, sentence, classconcept, question, negation). The linguistic status of ‘Premackese’ is questioned, and it is suggested that it is better viewed as a communication system,

It may,

therefore,

be that the aphasic children

lack some

specifically

linguistic ability.

The essential feature of developmental aphasia is the failure of the individual to acquire speech although his hearing and non-verbal IQ are within the normal range (Myklebust, 1963b). It is not yet clear whether this failure represents an inability to handle symbolic material, a disturbance of communicative functioning in general or a purely linguistic deficit. By the application of techniques devised to investigate the communicative and linguistic capacities of other species, it might be possible both to gain some insight into the nature of developmental or early acquired aphasia and to provide a relevant empirical contribution to the following problem: ‘What essentially is language and how can it be learned?’ (Brown, 1973). Although it has long been known that animals are capable of communication, it is only within the last fifty years that concerted attempts have been made to teach human language to a non-human animal. Early experiments with chimpanzees met with little success since they were concerned with spoken language and demanded vocal responses (Kellogg and Kellogg, 1933; Hayes, 1951). It is now apparent that these animals are not * I should like to thank Dr. B. Hermehn for suggesting this experiment and Dr. R. F. Cromer and Dr. N. O’Connor for their helpful comments and criticisms. I should also like to thank the staff of Moor House School. Hurst

Green, Surrey, for their assistance with subjects and facilities. ** Present address: Department of Psychology, University College London.

Cognition 3(I), pp. 41-55

42

Jennifer

Hughes

biologically fitted for such a task. Lieberman, Klatt and Wilson (1969) have stressed the differences between the vocal apparatus of the chimpanzee and that of man. In addition it appears that chimpanzee vocalizations are not under voluntary control (Yerkes and Learned, 1925; Hayes and Hayes, 1952). The natural calls are only emitted in the presence of the appropriate emotion (Goodall, 1963). A more recent experiment by-passed the difficulties outlined above by adopting a system of communication which did not involve any articulation at all. Instead, Gardner and Gardner (1969) took advantage of the high manual dexterity of the chimpanzee and the natural predisposition of this species to use gesture in their social communication and began teaching a chimpanzee the sign language of the deaf. By the time their subject was 4 years old, she was reliably using more than 80 signs, singly and in combination. While her sign strings carry semantic meaning (Brown, 1970) it has been argued that they do not constitute true grammatical combinations (Bronowski and Bellugi, 1970). The latest attempt to teach a chimpanzee a formal system of communication rejected natural language altogether. David Premack (1969, 1970, 1971) created an artificial ‘language’ for the purposes of the experiment. Premack’s system is visual, consisting of plastic shapes (‘plastic words’) metal-backed to cling to a magnetic board. Each word is unique, the shapes being varied along the dimensions of color, shape and texture. Messages were written vertically, since this was the mode the chimpanzee seemed to prefer. Premack asks the question ‘What must animals do to give evidence of language?’ and answers it by drawing up a list of ‘language functions’. This list is not exhaustive but includes the functions which Premack felt to be important, e.g., word, sentence, question and metalanguage. To the extent to which an animal could be trained to produce and comprehend these functions, it would have language, as defined by Premack. Using a method of operant conditioning, Premack was largely successful in that his chimpanzee subject acquired most of the language functions in which she was trained. Premack attributed the few failures to inadequacies in the training technique. Among the language functions established were word, sentence, question, metalanguage, the use of class concepts, the copula, pluralization and logical connectives. In the present experiment the aim was to find out whether similar language functions could be acquired by human children who have failed to develop normal language or who have lost it at an early age. Congenitally aphasic children do not learn to talk at the usual age, and their babbling may sometimes appear abnormal. They seem chronically inattentive to the spoken word (Lenneberg, 1964). Aphasia is not characterised as a speech disorder as such but rather as an inability to relate a language symbol to experience (Myklebust, 1963a). It is thought to result from an organic condition. In adults, various types of aphasias occur, according to the site of brain damage. Usually only certain aspects of speech are affected. In children, speech is affected in a general way, even when the lesion is relatively localized (Geschwind, 1970). The main feature of aphasia is the failure to develop speech although other symptoms are usually present.

Acquisition

of a non-vocal ‘language ’ by aphasic children

43

Aphasic children differ trom normal children chiefly in that they do not possess language despite adequate intelligence and hearing and in the absence of any other peripheral defects (such as inarticulation). Can they, like Premack’s chimpanzee, acquire ‘language functions’ even though they cannot acquire language itself? If they are able to do this, it suggests that human language and Premackese differ in some important way, at least as far as this group is concerned. It may be that the difference is basic, lying perhaps in the scope and structure of the two systems (Brown, 1973). Alternatively, they may differ primarily in modality, human language being essentially vocal-auditory and Premackese visual. Two possible methods of presentation existed, Manual Sign Language and Premackese. The latter was selected as the medium for the experiment for the following reasons. The use of plastic ‘words’ by-passes the problem of limited short-term memory, since the words are present for reference. In addition the children have to learn only the simple action of placing the words on the board as opposed to the much more complex task of learning manual signs. Problems of judging correct responses are also minimized. Premack’s system is better adapted to an experimental context since the materials are available only during testing sessions and practice cannot therefore occur outside them. Manual Signing had recently been introduced into the school’s curriculum. Such a system of communication is clearly more appropriate in a social situation. However, problems of experimental control would have arisen had this, or too similar a system, been used in the present study.

1. Method For the present experiment five of Premack’s language functions were selected: Word, sentence, class concept, negation and question. These were chosen since they appear to embody some of the major characteristics of language. ‘Word’ introduces the principle of arbitrary reference, with a single sign denoting a class of actions, objects or ideas. ‘Sentence’ involves combining words to communicate a more complex meaning. Premack would hold that it also involves syntax. ‘Class concept’ is included since the principles of hierarchical organization and classification are important in both thought and language. ‘Negation’ is a relatively early feature of child speech, possibly the first grammatical transformation. ‘Question’ is another early-acquired transformation, and the asking and answering of questions is an activity which may contribute to the growth of language as well as the growth of knowledge (Greenfield, Smith and Laufer, 1972). The aim of this experiment was to discover whether aphasic children with minimal speech could acquire the above five language functions via a process of mapping a simplified system of visual symbols onto a simplified world. The method, following Premack, was operant conditioning, applied rather less strictly in order to accommodate the higher capacities of the child subjects for insight and the communication of their

44

Jennifer Hughes

knowledge. The use of operant techniques in the experiment in no way implies a belief that it is the method by which unimpaired children normally acquire language. Rather, the technique is used merely as a tool for attempting to train children who have not acquired language in the normal way.

1 .l Subjects

Subjects attended a school for children classified as aphasic. The basis for subject selection was language deficit. The children selected were the ones who, in the opinion of the school staff (borne out by test results), had the least linguistic ability. Their ages at the beginning of the study were as follows: Girls: 91 years; 12 years Boys: 84 years; 13 years All subjects had been diagnosed as aphasics, the aphasia being either congenital or of early onset. All had non-verbal IQs well within the normal range. Their linguistic abilities had been assessed by a variety of tests, including the Peabody Picture Vocabulary Test (PPVT) and the Reynell Developmental Language Scales (RDLS). The children’s scores are shown in Table 1. Table 1.

Summary

of subjects’ linguistic abilities prior to experiment

Subject sex Age (in years, months) at testing RLDS Comprehension RLDS Expressive Language PPVT Note: Scores all expressed

Sl Female

s2 Male

s3 Female

s4 Male

Mean

9:o

7:lO

11:7

12:2

10:2

1:lO 2:4 < 1:9

2:6 2:9 2:l

1:8 1:3 2:3

1:ll 1:9 2:o

in terms of age levels in years:

1:lO 0:lO 2:l

_-

months.

It will be seen that functioning was generally at or below the two-year-old level. In terms of the language functions with which the present study is concerned, these scores represent the following state of affairs. All children could use at least three words; for all but one child, these were restricted to nouns and verbs. Only one child could construct a sentence of four syllables. None could use prepositions, and only one showed comprehension of them. No child could ask questions. Simple questions about familiar objects or pictures could be answered by pointing by some of the children (e.g., ‘Where is the doll?‘). None of these questions approached the complexity of those involved in the present study, since it was possible for the child to respond on the strength of comprehending only the noun. There was no evidence of use or comprehension of any construction similar to the Class Concept and Negation functions described below. All the subjects had been given instruction in deaf sign language starting about

Acquisition

of a non-vocal ‘language’ by aphasic children

45

6 months

before the experiment began. At the time of the experiment, their use of sign language consisted in the main of single signs: Nouns, verbs and adjectives. Combinatious rarely exceeded two signs, and no apparent syntactical order was observed. Comprehension is reported to have been at a similar level. Longer and more complex sign sequences seem only to have been understood by the children if they consisted of familiar questions of the rote-learned variety. It should be noted that the school authorities report that continued instruction in Sign Language has resulted in its use by aphasic children as an effective means of communication. An analysis of the form of this sign system and of the functions it is capable of expressing remains to be made. In any event, the relationship between the structure of a Manual Sign Language and natural languages is as yet unclear (see review by Bonvillian, Charrow and Nelson, 1973). There is an implicit assumption that the subjects were familiar with the concepts underlying the language functions of the present experiment, although they were unable to give expression to them in verbal language or, at that point in time, in Manual Sign Language. Their comprehension of communications in both media was only slightly less limited. 1.2 Apparatus The ‘words’ which made up the ‘language’ were all shapes cut from 5 X 5 cm. squares of black, magnetic-backed plastic. Texture and color were held constant, shape and size being varied to give a set of stimuli which were easily discriminable. The shapes were assigned to words on a random basis to avoid the introduction of any iconic relationship between plastic ‘word’ and referent. The shapes were intended not to resemble any particular objects, letters or numbers. During training the words were placed on language boards. These were metal sheets (approximately 30 X 35 ems. mounted on slightly larger pieces of hardboard). Because the plastic words were magnetic and clung to the boards, they could be used either placed flat on a table top or vertically. Two boards were always in use. One was the vehicle of communication between experimenter and subject, and the other acted as a lexicon (the ‘storage board’). On this were placed the words which the experimenter wanted to make available to the child at any one time. 1.3 Procedure The method employed to teach the five language functions was basically similar to that outlined above in the account of Premack’s work. Rewards usually took the form of social reinforcement, sweets being used with some of the children. Speech was not used during the course of the experiment; all communication between experimenter and subject was either improvised gesture or effected via the language functions which were being taught. Subjects were seen in 20 to 30 minute sessions twice weekly.

46

Jennifer

Hugh.&

The procedure described here is taken function by function, in the order that these were introduced to the subjects. For ease of reference, Premack’s nomenclature has been adopted, but this usage of linguistic terms should not be taken to imply that these language functions are equivalent to ‘language’. This point will be expanded in the discussion below. 1.3.1 Word Training in the language function ‘word’ was begun in the following way. The experimenter handed a toy to the child who took it, looked at it, perhaps played with it and finally handed it back. Then, the child was offered a piece of plastic, the word which was to stand for that toy. He was required to take the word and place it on the language board. He was then given the toy to look at or play with and also praised. This procedure was repeated for each of five objects (car, doll, book, horse, engine). When all five objects had been named in this way, they were again offered to the child, one by one, the child’s task being to select the correct word from the storage board. Errors were corrected, either spontaneously by the subject or, failing this, by the experimenter. This procedure was continued until the subject was consistently choosing the appropriate words. At this point the situation was reversed. Instead of production, comprehension was demanded. The child’s task was now to select a toy named by the word which the experimenter had placed on the language board. Generalization was tested by introducing, from time to time, different class exemplars from those which had originally been used to teach the word, e.g., a toy soldier instead of the original rag doll. This was done for both comprehension and production trials. 1.3 2 Sentence Three types of sentence were used in this phase of the experiment. The first was a simple subject-verb-object construction, for example: (a) Jenny give doll (b) Diane point-to car As in Premack’s experiment, the sentence was built up, word by word, each word mapping a certain aspect of the situation. First, the experimenter ascertained that the child was able to remember all the vocabulary introduced at the single-word stage. Next, all the known words were removed from the storage board and replaced by two new shapes which were to stand for ‘give’ and ‘point-to’. One of the nouns, say ‘doll’, was then placed on the language board. The experimenter then produced the real doll and gave it to the child, exaggerating the action. The experimenter pointed to the words on the storage board and indicated to the child that one of them was to be chosen and placed on the language board, to the left of ‘doll’. If the child chose the correct word (in this example ‘give’), the experimenter indicated that he was right. If he was incorrect he was encouraged to try again. A similar procedure was followed using the other objects for

Acquisition

of a non-vocal ‘language’ by aphasic children

47

which names had already been learned. ‘Give’ and ‘point-to’ were each used on roughly half the trials, the order in which they occurred being random. When the child was consistently supplying the correct verbs, all the nouns (excluding the proper nouns) were placed on the storage board, along with the two verbs. The child was now required to select both verb and noun and to place them in the correct order on the language board, in correspondence to actions performed by the experimenter. The target sentences (a) and (b) were thus approximated by: (c) Give doll (d) Point-to car When two-word strings were being produced and comprehended consistently, the subject of the sentence was introduced, in an analogous manner. Training was continued until the child was able to form entire three-word sentences from the selection of object and action words and proper names available on the storage board. The second type of sentence to be established involved four-word strings: Subject verb - direct object - indirect object. Two examples of the kind of sentence taught are: (e) Diane insert doll dish (f) Jenny takeout car bucket The child was required both to perform the action called for by the sentence (comprehension) and to construct complete four-word strings describing actions performed by the experimenter (production). It should be noted that during testing for comprehension of sentences of this kind, more than one object of the kinds represented on the board was present. For example, for (e) at least one doll was already inside a dish. If only one doll and one dish had been present, only one action would have been possible on each trial, according to the original relative positioning of the objects. The third type of string was rather different in that it contained no verbs at all, although its English ‘translation’ does have an implied copula. This third type involved a preposition ‘on’. Colored cards were used as the referents for these prepositional strings. Since no arrangement of colors is inherently more likely than any other, this necessitated understanding and making appropriate use of word order. This contrasts with the first two types of sentences which were taught. In those, little information was carried by the word order. In the first type of sentence, for example, certain words referred to agents, i.e., people capable of performing actions; these were always the subject of the sentence. Certain words referred to the actions themselves, the verbs, and a third set of words referred to objects which could be acted upon. Because no single word appeared in more than one class, the order of the words held no particular significance. Similarly, in the second type of sentence there was a class of objects which could be inserted and a second class of objects into which insertions could be made. In introducing the third type of sentence, the subject was first taught the plastic symbols for a number of color words. Color stimuli were rectangles of glossy colored paper, pasted to cards. When all the color words were being used reliably, the preposition ‘on’ was introduced. The experimenter put up on the language board ‘red on green’; the

48

Jennifer

Hughes

green card was laid on the table in front of the child who was then given the red card and encouraged to place it on top of the green one. A similar procedure was then followed for ‘green on red’. Several repetitions followed. Once the child realized what was required of him, both cards were given to him at once. The next stage involved production. The three words ‘red’, ‘green’ and ‘on’ were placed on the storage board, and the child was required to place them in the correct order, i.e., corresponding to the experimenter’s arrangement of the cards. Comprehension and production established, the remaining two colors were introduced. All possible combinations were subsequently used on both production and comprehension trials. 1.3.3 Class concepts: Color, shape The names of three geometrical shapes, were taught to the children. The exemplars were simple figures, draw-n with black ink on white cards, and consisted of a circle, a square and a triangle. When the child had shown that he knew both the color words he had learned earlier and the new shape words, the two new class concept words were introduced: ‘color-of and ‘shape-of. Twelve stimulus cards were constructed, each bearing a figure (a circle, triangle or square) in one of the four colors. One of the cards was placed on the language board. The color word corresponding to the color of the stimulus figure was placed on the left of the card, a gap being left between the two. The class concept words were placed on the storage board. The child’s task was to select ‘color-of (through trial and error, if necessary) and place it in the space left for it on the board to give, for example: (g) RED color-of RED CIRCLE A battery of such trials followed, using the standard reinforcement/correction procedure. Stimulus cards were shuffled following each successive series of 12 trials. Half the trials involved shape and half involved color, the sequence in which they appeared being random. This procedure was continued until the subject demonstrated that he was using both words appropriately. As a test of transfer, an alternative form of the task was given. Instead of the child being required to supply the class concept word, he was required to supply the appropriate color or shape word, the class concept word and the stimulus figure having already been laid down. For example: __ shape-of YELLOW SQUARE (h) When competence in the use of class concept words had been established, some informal tests were carried out to find out whether their usage could be extended to describe objects other than the stimulus cards which had been used during the training. Using the plastic words, the children were asked to describe such things as the shape of the wheel and the color of the toy car. 1.3.4 Negation The class concept

construction

was used to introduce

the negative particle ‘not’. Training

Acquisition ofa non-vocal ‘language’ by aphasic children

49

began with a demonstration series of four examples placed successively on the language board by the experimenter. These are shown below: (i) (1) RED c&r-of RED CIRCLE (2) RED not color-o~GREEN CIRCLE (3) TRIANGLE shape_ofGREEN TRIANGLE (4) TRIANGLE Motshape-ofGREEN CIRCLE Twelve training trials followed, in which each stimulus card was used once. As before, the class concept word was omitted. The case could either be positive or negative, i.e., the color or shape word either did or did not correctly describe the stimulus figure. Examples of each case are as follows: YELLOW CIRCLE (j) YELLOW RED TRIANGLE (k) SQUARE Half the trials involved color and half shape. Within these, half the trials were positive cases and half negative. The child’s task was to insert the relevant class-concept word, with or without the negative particle, as appropriate. When the negative was being used consistently, examples resembling (h) above were given. Again, half the examples involved negative instances and half positive, half color and half shape. 1.3.5 Question The three types of questions used are summarized below. On the left is given the schematic form, ‘A’ representing some object and ‘?’ the plastic interrogative mark. On the right is shown the English paraphrase. (1) A’A ..I....... ‘What is the relation between A and A?’ (2) A = ? . . . . . . . . . . ‘A is the same as what?’ (3) A=A? . . . . . . . . . ..‘IsAthesameasA?’ The first question has two possible answers, ‘same’ and ‘different’. The first step was to establish these two words and, in doing so, a preliminary form of the question. This was done by using, to begin with, concrete objects: Cups, spoons, dishes. The two new words were placed on the storage board, and two of the above objects on the language board with a space between them. Subjects were, by now, quite familiar with the idea of filling the gap. The usual reinforcement/correction procedure followed the selection and placing of a word by the child. A block of six such training trials was given, using all possible combinations of pairs of objects. Testing was carried out in a similar way, this time using line drawings instead of actual objects. This procedure was repeated, if necessary, until the child was using the two words consistently, even with novel stimuli. At this point the space between the two comparison stimuli was replaced by ‘?‘. Instead of placing the answer in the space, the child was now required to place it in a square drawn on the language board, below the stimuli and word sequence. Transfer to this version of the question was accompanied by a change in the stimulus set. The second form of question was not answered using a plastic word at all. Instead the child was given a choice of several stimuli similar to the one used to frame the question,

50

Jennifer Hughes

one of them being identical. In the case of the ‘same’ form (A = ?) it was the child’s task to pick out the card identical to A. In the complementary form (A # ?), the child had to point to those selection stimuli which were not identical to A. Type (3) questions, framed as indicated above, required either ‘yes’ or ‘no’ as an answer. A complementary frame using ‘different’ (A # A?) also demanded these answers. ‘Yes’ was a new piece of vocabulary for the subjects. ‘No’ had already been introduced in the shape of the negation particle, the same plastic word serving both functions. Two identical stimuli (objects for the first few trials, cards bearing line drawings later) were placed on the language board. The word ‘same’ was placed between them, and ‘?’ was placed on the right of the second stimulus. The child had available two words on the storage boards, ‘yes’ and ‘no’. The experimenter indicated to the child that one of these words was to be placed in the box drawn on the board below the ‘sentence’. The usual reinforcement/correction procedure was followed. The next trial involved a queston framed around the word ‘different’, the stimuli being identical with those used on the previous trial. A series of question sets then followed. For each identical pair of stimuli, the question could now be asked in two ways. Similarly, for each different non-identical pair. With six sets of pictures, twelve questions could therefore be asked of any set of stimuli. These were always taken in random order. This procedure was continued until the subject showed himself to be competent in answering questions.

2. Results A function is rated as having been acquired if the subject met the following criterion: Correct use of the function on at least 70% of the trials following training, this level of competence being demonstrated during at least two sessions. In most cases, tests of generalization provided further evidence of acquisition (see section 1). It should be borne in mind that a level of 70% correct usage is fairly impressive, since most functions demanded the selection and correct ordering of several words from a relatively much larger number on the storage board. The probability that such results could be obtained by chance is extremely small. In all vocabulary acquisition a 100% correct criterion was enforced. The acquisition of the functions is shown for all four subjects in Table 2.

2.1 Word, sentence, and class concept It can be seen that the functions ‘word’, ‘sentence’ and ‘class concept’ were acquired by all four subjects with little difficulty. All were able to learn the initial elements of vocabulary during their first session and acquired further vocabulary in a similar way throughout the experiment. Some forgetting occurred from session to session, especially when vocabulary had only been taught and had not been used in the course of

Acquisition

Table 2.

of a non-vocal ‘language’ by aphasic children

Summary of language functions

acquired by each subject

Sl Female 9:

s2 Male 8;

s3 Female 12

s4 Male 13

Class Concept Negation Question (1) (2) (3)

X X X X X X X X X

X X X X X 1 X X ?

X X X ? X 1 X X ?

X X X X X X X X _

Total No. Sessions

12

15

11

12

Subject Sex Age (in years) Word Sentence

(1) (2) (3)

51

X: Function acquired; = 70% correct usage during at least 2 sessions. ?: Some evidence of acquisition but criterion not reached ~-: Function not acquired.

establishing or testing other functions. Vocabulary so used was rarely forgotten, even after a lapse of several sessions. All subjects were able to generalize their use of a word from the referent originally paired with it to other class members. Sentence forms (1) and (2) were acquired more rapidly than type (3) for three out of the four subjects. This can probably be attributed to the increased complexity introduced into this type of sentence by making word order a crucial source of information. A second source of difficulty with sentences involving ‘on’ seemed to stem not from the inability to understand the relation but to lie in the strategy the children adopted in discovering how it was expressed. Typically, on early trials, a child would pick up the colored card named on the left of ‘on’ and would lay this on the table. He would subseqently select the second color and lay this down on the first, thereby producing an incorrect sequence. When the relation was expressed correctly the children still tended to operate on the left-most color first but now picking up the card and retaining it before selecting and laying down the second one. It seemed as if they initially perceived the relation as a global one which encompassed both ‘on’ and ‘under’, since the cards were always stacked together, one above the other, on comprehension trials. 2.2 Negation All subjects gained some competence in negation, but only two reached the 70% criterion. There were, however, indications that the other two children understood the concept of negation, their failure to reach criterion being partially a function of the method employed. The most frequent error with all subjects was the use of ‘not’ to indicate a mismatch between the color/shape word placed on the board and the actual

52

Jennifer

Hughes

stimulus presented, omitting the ‘concept word’ (i.e., color-of, shape-of). Although responses of this kind were scored as errors, since they did not correspond with the intended structure, they were consistent and did express a kind of negation, namely denial. The children were effectively communicating comments such as (in English translation) ‘A blue circle is not red’. The same kind of argument applies to later ‘errors’ where the concept word was present but incorrect, since statements such as ‘Ked is not the shape of a blue circle’ are undeniably true. If these types of ‘errors’ are allowed as correct, the remaining two children also met criterion. Further, the children’s expression of negation might well be considered to be neater and more logical than the experimenter’s. As a test of ingenuity, some subjects were given a single stimulus card on the language board, with all the color, shape, concept words and ‘not’ available on the storage board. The children were able spontaneously to ‘describe’ several stimulus cards with the plastic symbols without any assistance. It is interesting that this task was performed quickly and accurately. Color, shape and negation were all used. 2.3 Questions All the subjects performed well on question types (1) and (2). All found (3) difficult, only one subject reaching criterion. Few errors were made by any subject on the initial stage of learning ‘same’ and ‘different’ via the classification of identical and non-identical pairs of objects. Transfer from concrete objects to figures was made without difficulty by all subjects. Similarly, the introduction of ‘?’ had no effect on performance. Questions of the second type were handled without difficulty, some subjects making no errors at all. Type (3) questions were much more difficult for all subjects. Results were well below the near 100% rate for the previous types. One subject was highly inconsistent, but the other three showed fairly consistent error patterns. Questions which were phrased using ‘same’ were generally answered correctly, whether the appropriate response was ‘yes’ or ‘no’, but questions involving ‘different’ frequently produced confusion. They were generally answered with ‘no’ whether or not this answer was correct. It seems that ‘same’ is in some way an easier concept to understand.

3. Discussion We have seen that aphasic children are able to learn, via a program of training procedures, a communication system which parallels several functions of human language. The finding reflects the original work of David Premack, who evolved the system and succeeded in teaching it to a chimpanzee. It should be emphasized that the children who participated

Acquisition of a non-vocal ‘language’ by aphasic children

53

in the experiment possessed only rudimentary spoken language. There was no evidence that the children’s progress on the various language functions was the result of rote learning; this would have been virtually impossible, especially in the case of functions involving several ‘words’. The fact that the functions underwent generalization, both within and outside the strict experimental context, also militates against this kind of interpretation. Through the system of symbols the children learned to communicate effectively in a number of ways: Naming, describing, carrying out requested actions and answering questions. Is this tantamount to saying that aphasic children can, after all, learn language? The answer to this question depends upon the status which we assign to ‘Premackese’ and, of course, what we mean by language. It has been suggested that the structure of language and the process of first language acquisition in children are closely linked in the following way. The child brings to the problem of language learning some prior knowledge of linguistic structure (McNeil, 1970). Since any normal child is able to acquire any natural language, these prior principles must be common to them all, that is, they are innate and therefore universally found. If this is the case, the task of every child is to discover how the linguistic universals are expressed in the particular language to which he is exposed. The special structure of language is significant for a second reason. It ‘provides the means for expressing indefinitely many thoughts and for reacting appropriately in an indefinite range of new situations’ (Chomsky, 1965). Here, Chomsky is reiterating the long-recognized creative quality of language. Bronowski and Bellugi (1970) stress the fact that the acquisition and use of language accurately reflect the more general human capacity for analyzing the environment into parts and recombining them in a different way. Language is certainly well able to map such operations. This is not to say that communication of such ideas cannot be achieved in the absence of language (as in the case of the deaf) but rather that spoken language is the most efficient means of communicating them. The suggestion which I am making is that human beings are specially attuned to language in a way which enables its acquisition and use to take place efficiently and reliably. This potential for language exists in addition to the cognitive factors which must also be present for language to develop, for example, sensorimotor intelligence and the semiotic function as described by Piaget (Piaget and Inhelder, 1969). In the absence of such a ‘language tuning device’ communication would still be possible. Such a system would be capable of expressing the relations apprehended by the organism’s intelligence but might lack the elegance, efficiency and precision of a fully developed language. The communication systems acquired by the ‘educated’ chimpanzees mentioned earlier, deaf sign language and the symbol system of the aphasic children all utilize the various cognitive and communicative capacities of their users, but none of these systems has the essential structural characteristics of language. Language itself is a species-specific skill, backed up by certain biological adaptations, the human answer to the communication problem.

54

Jennifer

Hughes

The validity of the distinction drawn between a functional system of communication, and language as one particular way in which communication may be achieved, is borne out by the results of the present experiment where aphasic children, having failed to acquire speech, learned to use a system of symbolic communication. The characteristic linguistic tuning seems to be absent in these children. It is not possible to be much more precise. It may be that some basic linguistic ability is missing, i.e., that the strategies normally applied to the language corpus have failed to operate. It is also possible, however, that the deficiency is confined to the processing of auditory material. Further experimentation will be necessary to test these possibilities.

REFERENCES

Bonvillian, J. D., Charron, V. R., and Nelson, K. E. (1973) Psycholinguistic and educational implications of deafness. Hum. Devel, 16,321-345. Bronowski, J., and Bellugi, U. (1970) Language, name and concept. Science, 168, 669673. Brown, R. (1970) Psycholinguistics: Selected papers by Roger Brown. New York, The Free Press. .-.- (1973) A first language. Cambridge, Mass., Harvard University Press. Chomsky, N. (1965) Aspects of the theory of syntax. Cambridge, Mass., M.I.T. Press. Gardner, A. R., and Gardner, B. T. (1969) Teaching sign language to a chimpanzee. Science, 165,664-672. Geschwind, N. (1970) Disturbances of language, perception, and memory. In C. S. Keefer and R. W. Wilkins (Eds.), Medicine: Essentials of clinical practice. Boston, Little, Brown & Co. Pp. 973-980. Goodall, J. (1963) My life among wild chimpanzees. Nat. Geo., 124,272-308. Greentield, P. M., Smith, J. H. and Laufer, B. (1972) Communication and the beginnings of language. Unpublished draft. Hayes, C. (1951) The ape in our house. New York, Harper & Row. Hayes, K. J., and Hayes, C. (1952) Imitation in a home-raised chimpanzee. J. Comp. Physiol. PsychoI., 45,450-459. Kellogg, W. N., and Kellogg, L. A. (1933) The ape and the child. New York. McGraw Hill.

Lenneberg, E. H. (1964) Language disorders in childhood. Harvard educ. Rev., 34, 1522 177. Lieberman, P. H., Klatt, D. H., and Wilson, W. H. (1969) Vocal tract limitations on the vowel repertoires of rhesus and other non-human primates. Science, 164, 118551187. McNeill, D. (1970) Theacquisition of language. New York, McGraw Hill. Myklebust, H. R. (1963a) Aphasis in children: Language development and language pathology. In L. E. Travis (Ed.), Handbook of spee& pathology. New York, Appleton-CenturvCrofts. Pp. 503.-513. (1963b) Aphasis in children: Diagnosis and training. In L. E. Travis (Ed.), Handbook of speech pathology. New York, Appleton-Century-Crofts. Pp. 5144530. Piaget, J., and Inhelder, B. (1969) Thepsychology of the child. London, Routledge & Kegan Paul. Premack, D. (1969) A functional analysis of language. Unpublished paper based upon an Invited Address (Division I), A.P.A., Washington, D.C. (1970) The education of Sarah. New Sot., 768-770. (1971) Language in chimpanzee? Science, 112,808-822. Yerkes, R. M., and Learned, B. W. (1925) Chimpanzee intelligence and its vocal expression. Baltimore, Wilkins & Wilkins.

Acquisition

of’ a non-vocal ‘language’ by aphasic children

On a enseigne’ i des enfants aphasiques ayant des deficits de langage serieux, ri communiquer par l’intermediaire du systeme de symboles visuels mis au point par D. Premack pour les chimpanz&. Les sujets, qui ne possedaient pas de langage normal, ont rapidement pu de cette facon exprimer plusieurs fonctions du langage (mot, phrase, concept de classe, interrogation,

55

negation). On peut s’interroger sur le statut linguistique du ‘systeme Premack’dont on se rend sans doute mieux compte comme d’un systbme de communication. En consequence, il est possible alors de penser que les enfants aphasiques manquent d’une capacite sp&ifiquement linguistique.

Discussion

Minds,

machines

and phenomenology: on Dreyfus’

‘What

Some

computers

reflections can’t

do’*

ZENON W. PY LYSHYN University

of Western Ontario

years cognitive psychology has been criticized from below by empiricists (e.g., Skinner), from above by rationalists (e.g., Chomsky) and now is being subjected to an attack from behind by phenomenologists ~ especially in a recent book by Dreyfus (1972). Empiricists have found the information-processing approach too mentalistic, too removed from the data of observed behavior and insufficiently interested in the problems of learning and the modification of behavior. The more rationalist critics have found it, on the contrary, too data-bound, too dedicated to concrete cases of observable performance and too unwilling to make certain strategic epistemological distinctions - as between competence and performance (see, for example, Pylyshyn, 1972, 1973a). Cognitive psychology has rather self-consciously steered a middle ground in which the formal aspects of the computer metaphor furnishes the conceptual structures around which empirical observations are organized. In opposing all three of these approaches the phenomenological world-view asserts that the structure of immediate experience, as apprehended in the naive everyday world and uncolored by the philosophical traditions of the last two millenia, should be not only the starting point but the ultimate referee of the success of any system for understanding the mind. It is not an easy path to travel for anyone schooled in Western science. As a philosophical movement begun by Edmund Husserl at the turn of the century it achieved considerable importance in Germany and France and left its imprint on that psychological school known as Gestalt psychology - which is probably the only form in which it is familiar to most North American psychologists. The influence of Gestalt psychology and of the recent ‘subjectivist’ epistemologies (e.g., Polanyi, 1958) is apparent in Dreyfus’ book. Rather than provide a general review of Dreyfus’ critique this article will concentrate on certain fundamental criticisms which its author directs at the information-processing approach to cognitive psychology and will try to point out the unique conception of what it means to ‘understand’ cognition which separates a phenomenologist from the

In recent

* Reprints Pylyshyn,

may be obtained from Zenon W. Department of Psychology, Univer-

sity of Western Ontario, Canada, N68 3K7.

London,

Ontario,

Cognition 3(l),

pp. 57-77

58

Zenon W. Pylyshyn

typical cognitive psychologist. Before proceeding with this task, however, I might point out that the book itself is an exasperating mixture of nonsense, well-deserved criticism, ~sunderstanding and insight. Unfo~unately, however, I suspect that it may be largely ignored except by latter-day Luddites, phenomenologists and misguided humanists who see in it an expert’s vindication of their prejudices concerning the cold, hard, technocratic sciences of the mind. If this turns out to be the case it will be due in no small part to Dreyfus’ harsh polemics [some of which is warranted but much of which is misdirected or mistaken, such as his claim that ‘artificial intelligence is the least self-critical field on the scientific scene’ (p. 63) which follows over 60 pages of direct critical quotes from artificial intelligence (AI) workers; or his repeated reference to the deveIopments of the last five years when the last project cited dates back to 19641 or to some simple misunderstanding - e.g., concerning the nature of analogue computation. Mostly, however, it will be due to the completely alien epistemological position which the author implicitly accepts and without which most of his assertions would be read as simply nonsensical. It would be a pity if the book were to be ignored by its victims since it is one of the most explicit and studied attacks from the phenomenological position, and its author is at pains to try to meet the opposition on mutual ground. In fact the latter attempt leads to what I believe is generally a serious weakness of the book. By arguing on performative grounds (i.e., what computers will never be able to do) Dreyfus lays himself vulnerable to rapid obsolescence. If Dreyfus had stuck to epistemological questions he might have been able to mount an argument (along the lines suggested by Von Neumann, 1966, or argued by Shaw, 1971) that even if we could build powerful intelligent computers we might still be unable to satisfactorily understand the human mind because in a sense we might not understand the machines we had built. Although I believe such an argument to be false, there is no question that it could be made persuasive. But this does not appear to be Dreyfus’ point at all: He makes substantive claims about the limits of artificial reason. However, even if all the current AI work were on the wrong track, it would still not follow from a phenomenolo~cal analysis that a computer could never be a chess master. This non sequitur should be obvious from Dreyfus’ own belief concerning the relation between the theoretical and the practical or his approving reference to Leibniz’s claim that one could not have a ‘theory of practice’. Before examining Dreyfus’ attack on cognitive psychology we might note a certain similarity between his protestations and those which psychologists frequently made against the linguistic claim concerning the psychological reality of generative grammar. Psychologists maintained the irrelevance of such formalisms on the grounds that ‘people do not do things that way’. In discussing cognitive simulation Dreyfus argues over and over that people do not do things the way a program does: They do not search through a tree of alternatives ~ they ‘zero-in’; they do not learn merely to perform ~ they learn through understanding. Computer simulations differ from the processes they purport to describe because they fail to account for understanding: ‘The only successful case of

cognitive simulation simulates a process which does not involve comprehension, and so is not genuinely cognitive’ (p. 22). Regarding Bobrow’s (1968) STUDENT he claims, ‘Again it is easy to show that what has been acquired by the machine can in no way be called “understanding”‘. Evans’ geometric analogy system (Evans, 1968) also operates differently from the way people do: Its failure to surpass people shows that ‘As in the case of GPS, there is no evidence that human beings proceed in this way, and descriptive, psychological evidence suggests that they do not’ (p. 52). When psychologists criticized the assumption (which linguists never actually held) that grammars describe the way in which people construct sentences, they did so on a variety of grounds some of which were in fact of the same phenomenological type as Dreyfus’ criticism of computer simulation theories (which we shall consider below). But there are also very straightforward empirical reasons for believing that psychological mechanisms for sentence production differ from generative linguistic mechanisms - reasons that do not simply rest on the belief that an adequate psychological account must not contain such formalisms as rewrite rules or that there cannot be such things as internal ‘mental symbols’. These reasons (discussed in Pylyshyn, 1972, 1973a) rest on the fact that generative grammars do not address themselves to certain important and legitimate bodies of empirical evidence. They do not, for example, account for the order in which linguistic structures are acquired, the relative difficulty in understanding or producing various types of structures or the conditions under which different utterances will occur. Most cognitive psychologists nevertheless believe that when more general theories are developed which do account for a broader base of evidence they will very likely contain such things as uninterpreted formal symbols and rewite rules (or some equivalent formalism). Such a theory, however, would be dismissed by Dreyfus as still lacking the alluring property of ‘psychological reality’. After reading through the long list of present and even possible formal theories that Dreyfus rejects as not accurately reflecting how people do things, one might well wonder what the essential but illusive aspects are which never appear in formal theories. Much of the mystery surrounding the claims made by Dreyfus would be eliminated if we could make explicit exactly why he believes that computer simulation (CS) theories lack psychological reality, in a manner which parallels the arguments made against the psychological reality of generative grammars. To some extent Dreyfus has tried to do this, He has argued that CS theories do not represent how people do things on the grounds that: (a) If they did faithfully reflect the way people do the simulated tasks, then the CS theories should be able to reach human performance levels - which they do not; (b) CS theories are largely ad hoc and lack generality; and (c) CS theories do not account for phenomenal evidence - i.e., they do not represent the steps which we are consciously aware of going through in cognitive activity. There is little to say about the first point. There is no simple and obvious relation between level of performance and fidelity of simulation. It is possible to construct an important partial theory or even an excellent complete theory in the form of a computer

60

Zenon W. &lyshyn

simulation which nevertheless leaves out such things as a large tedious data base or some special heuristics which people have learned that enable them to achieve a higher level of performance. Conversely high levels of performance are routinely reached on computers in a manner which reveals nothing about human cognition. In any case this kind of argument from present limits is a special form of the ‘first step fallacy’ which Dreyfus deplores. Regarding the second point (that AI and CS systems lack generality), this is a relative matter. They may well be far from achieving the kind. of generality which some spokesmen would claim for them. On the other hand, there is an important sense in which some current systems are highly general - because of the uniform principles underlying their design (e.g., Winograd, 1972). Nevertheless, it is in this area that Dreyfus presents some of his strongest arguments and we shall touch on several of these later. In the next section, however, we shall consider the third argument regarding the status of phenomenal evidence and examine the phenomenological point of view in some detail.

1. Phenomenology

and scientific understanding

It is on the question of the status of experiential evidence that phenomenologists and cognitive psychologists clash most strongly. In fact, because of the particular position which phenomenologists take on this issue, they are led to question the possibility of a science of cognitive psychology conceived along the lines commonly referred to as the ‘information processing approach’. This skepticism is raised because, according to Dreyfus, there is no intelligible level of description between the purely physical and the purely phenomenal. He asserts: ‘All that is given empirically are continuous physical inputs to the organism, on the one hand and the world of ordinary objects given to the perceiving subject, on the other. No cognitive psychologist has succeeded in defining another sort of input between these two which would provide the ultimate bits of information to which rules are to be applied. All accounts offered thus far turn out to be an incoherent mixture of physical description in terms of energy, and phenomenalist description in terms of crudely defined sense data’ (p. 199). Because this is an argument on which Dreyfus places considerable importance, especially in respect to his criticism of cognitive psychology, it deserves some attention. According to his account the problem arises because the psychologist is caught in a ‘conceptual squeeze’. On the one hand there is the brain ~~ ‘an energy-transforming organ’. This represents one acceptable form of description, but one not very useful to the psychologist since ‘On this level one would not be justified in speaking of human agents, the mind, intentions, perceptions, memories, or even of colors or sounds, as the psychologists want to do. Energy is being received and transformed and that is the whole story’ (p. 89). On the other hand there is the phenomenological level on which one can speak of human agents, of acting or of perceiving objects, colors, sounds and so on. But, as

Minds, machines and phenomenology

6I

Dreyfus points out, ‘. . . this level of description is no more satisfactory to a psychologist than the physiological level, since here there is no awareness of following instructions or rules; there is no place for a psycholo~cai ex~lana~u~z of the sort the cognitive simulationist demands. Faced with this conceptual squeeze, psychologists have always tried to find a third level on which they can do their work, a level which is psychological and yet offers an explanation of behavior’ (p. 90). Having thus given what appears to be a reasonable description of the psychologist’s dilemma, Dreyfus proceeds to show how cognitive theorists have tended to blur the distinction between phenomena1 and physical descriptions by using mental terms in speaking about physical processes and vice versa. There is some merit in this criticism: The casual glossing from phenomenoIogica1 to physical senses of the same terms can create a great deal of confusion (see, for example, a similar point in Pylyshyn, 1972a, 1973b). However, the main argument that there is no coherent level of description apart from the physical and the phenomenological rests on a number of assumptions which need to be exposed. In the first place the distinction which is really at issue in this discussion is not, as Dreyfus puts it, between the physical and the phenomenal. This is clear from the way Dreyfus partitions his concepts, placing objects like chairs and tables among the phenomenal terms and energy, electrons, and neurons among the physical terms. Surely there is no important difference to be made, strictly along the physical-phenomena1 dimension, between a neuron and a chair or between an electron and a person. The difference which Dreyfus is most concerned with has to do with the fact that at the phenomenological level we deal with ‘meaningful actions in a context already charged with meaning’ (p. 90) and ‘with objects in an already organized field of experience’ (p. 100). In other words the meaning of terms in the ‘phenomenological’ class comes from their role in the field of purposive human actions and, even more importantly, from their role in the world of interpersonal relations and informal communication. On the other hand, the meaning of the other terms (the ones he calls physical - i.e., energy, electrons, neurons, etc.) comes from their role in rationally constructed systems called theories. In other words the distinction is between ordinary language terms and theoretical terms. This distinction is further highlighted by the following claim, which is typical of many similar assertions in Dreyfus’ book: ‘No amount of complication can bridge the gap between shifting energy inputs and the perception of an enduring sound’ (p. 96). Whether or not we believe this basically dualistic position depends on what we mean by ‘energy’ and ‘perception’ on the one hand and what we mean by a ‘bridge’ on the other. In science the sort of bridge we wish to build is an intellectual bridge. Furthermore the bridge is not between shifting energy and phenomenal experience (which may be unbridgable simply because one might wish to stipulate the context in which certain words are to be appropriately used) but between some measurable aspect of the incoming causal physical event and a functional theoretical event. The former need not be one that is easily measurable using one of our current stock of physical instruments (which were, after ah,

62

Zenon W. Pylyshyn

designed to provide indexes which are useful in certain physical and technical contexts) and may not be easily describable in the physicists’ language (i.e., in terms of length, mass and time), but it is one which should be measurable by some constructable device or machine. The second term of the bridging operation, on the other hand, need not indeed in a natural science will not - be an experience but a theoretical entity like any other theoretical entity in science. It will have its ‘cash value’ in its ability to logically account for certain observed phenomena and in its function within a system of theoretical ideas. Thus if we see the critical distinction as one between ordinary language terms which derive their meaning from personal experiences and social context, on the one hand, and theoretical terms which derive their meaning from explanatory scientific theories, on the other, we see that Dreyfus’ dualism puzzle is no different from similar puzzles arising in all areas of science (cJ: Russell, 1960). Science has always been concerned with showing that there is a fundamental uniformity in nature even though appearances are often to the contrary. For example, in Galileo’s time it took an immense conceptual step to accept that there might be real physical objects which could not be seen directly with the naked eye. For Galileo to consider pointing his telescope toward the heavens required a great leap of faith in the uniformity of nature. It entailed the belief that there could exist objects which cannot be seen but which could be rendered visible by a process which is continuous with seeing (i.e., it was not a magical image-creation but rather a rendering of a potential image). In the subsequent scientific tradition people have come to believe that theoretical entities such as molecules are like things which can be seen even though in fact they can never be seen directly. In other words if we could become aware of them in our perceptual or phenomenal field they would be fundamentally no different from the objects which we see around us all the time. A very similar belief is held by cognitive theorists when they speak of such things as plans. The analogous claim here would be that there are certain strategies and thinking processes of which we are phenomenally aware and others of which we are not aware (some are so abstract that we can never be aware of them) but that the two are not fundamentally different entities. In other words if we could become aware of them we would class them with the class of ‘thoughts’ and not with such things as biological processes or anatomical structures. A similar kind of continuity also appears in the relation between theoretical terms (such as those of information-processing theories of cognition) and the terms used to describe certain experiential phenomena. One of the reasons the phenomenologist draws a sharp distinction between experiential and theoretical concepts is that a different descriptive vocabulary appears to be appropriate in these two domains. Phenomenal qualities may be described by such words as ‘immediate’, ‘meaningful’, ‘familiar’, ‘purposeful’ or ‘threatening’, and phenomenal events may be spoken of as ‘being recognized’, as ‘appearing’, as ‘being understood’ or as ‘being desired’. Such words cannot be used directly to speak of formal procedures or data structures. Thus in a purely formal context if one says

Minds, machines and phenomenology

63

that a data structure is ‘meaningful’ to a program or that an input has a certain ‘appearance’, one is then not using these words in the precise sense in which they are used in the phenomenal context. This is simply because in the phenomenal account the ‘appearance’ of something is an experience and is quite distinct from its biological, physical or behavioral correlates. But it is nevertheless possible to give these phenomenal terms a theoretical interpretation by identifying them with certain formal processes. Thus we can say that for a system to ‘recognize’ some object as ‘familiar’ is for it to establish a certain relationship between a formal representation of that object and stored representations of objects previously encountered. Similarly what we call an ‘object’ may also be given a theoretical definition in terms of the way the system systematically treats certain physical patterns of events occurring at its transducers (i.e., ‘eyes’). Now there is nothing unusual about this kind of reductive analysis; in fact it is routine at all levels of science. Pressure and temperature are macroscopic properties defined in terms of global measurements on large quantities of matter. Nevertheless they are theoretically interpreted in terms of aggregate properties of motions of molecules. Such properties are not equivalent to pressure and temperature in any direct sense; they are theoretical interpretations - which means that the identification is established through a series of deductive steps. Exactly the same may be said of the relations between phenomenal properties and their formal theoretical interpretive counterparts. While there is surely nothing incoherent about this position, it is nevertheless one which phenomenologists find extremely repugnant. They point out that in such an approach we are no longer using the theoretical terms in their original authentic sense: We have abandoned the original problem and find ourselves talking about a counterfeit world in place of the one we initially intended to study - a formal world of logic, mathematics and ‘bits’ of information in place of a human world of experience, knowledge and purpose. But this is a somewhat misleading way to describe the human world which is in the first instance the source of the psychologist’s curiosity. The ‘human world’ is not made up entirely of knowledge which is immediately and phenomenally given. In fact the human world which Dreyfus speaks of is continuous with the scientific world - it is full of inferences and rational reconstructions even though we are not always fully conscious of such processes. When we are confronted with the Muller-Lyer illusion one ‘immediate given’ is the different lengths of the two lines. But an equally immediate ‘given’ replaces it as soon as we are given a ruler and allowed to observe by direct comparison that the two Lines are of equal length. One does not have to be a sophisticated and trained scientist to make the switch in beliefs. Similarly while the phenomenologist may insist that in the human world which is his primary concern such things as ‘perceiving people’ is an immediate phenomenal event, the intelligent layman - who may be said to inhabit a more unbiased ‘human world’ than the phenomenologist - may have doubts. He may recognize even in his phenomenal experience the fleeting remnants of deductive in-

64

Zenon W. Pylyshyn

ferences and may be surprised later that his inferences had been mistaken. The point I am making is that even granting the primacy of phenomena, in the sense of how the world presents itself to the naive observer, not all of the terms which Dreyfus calls phenomenological are accurately described in phenomenological terms. They also contain aspects of logical and rational analysis. Thus the phenomenologist has no more right to claim his description as characterizing the authentic human world than does the scientist. But let us put aside this question of the authenticity of the world being described and consider how we are to interpret such phenomenal descriptions as those Dreyfus (pp. 5254) quotes from Arnheim on the question of how people solve geometric analogy problems. Arnheim states that when a person is confronted by the initial pair of geometrical figures he ‘may have a rather rich and dazzling experience’ during which certain unstable patterns combine, interfere, but fail to add up to an understandable whole. ‘Suddenly, however, the observer may be struck by the simple . . arrangement . . .’ of certain parts of the figures and begin to apprehend a relational pattern. Then as he turns to the missing part of the analogy problem he finds ‘the family resemblance is great, the relation comes easily’. But Arnheim is ‘. . . shocked to learn’ that the computer program which Evans (1968) wrote to simulate this function is extremely complex since ‘For us the problem is not hard; it is accessible to the brain of a young student’. How does one develop a scientific theory which will account for such observerations as those quoted above (as well as for the observed performance)? No doubt it would be possible to construct a simulation (such as those reported in Loehlin, 1968) which demonstrates certain symptoms of emotion or in which some function may be identified with a ‘rich and dazzling experience’ or with the event of ‘being struck by’ a particular pattern. But Dreyfus makes it clear that this would not be good enough. It is not enough that the system simulate the function which people carry out, it must also do it in the way that people do: It must, in Dreyfus’ terms, not simulate but represent the process. It must, in particular, deal with the same information as people deal with ~ which, according to Dreyfus, is analogue and not digital. A digital simulation of such processes must therefore be operating differently: ‘A digital computer solving equations describing an analogue information-processing device and thus simulating its processing”. It is not processing the function is not thereby simulating its “information information which is processed by the simulated analogue, but entirely dijferent information concerning the physical or chemical properties of the analogue’ (p. 107). (We shall examine the notion of representing and of analogue computation in the next section.) But what would it mean for a device to process the ‘same information’ as people do? What is the nature of such information? Apart from scattered references to such qualities as that it does not consist of independent ‘bits’, there is no explicit answer to this question. However I think we can get a good idea of what Dreyfus has in mind from his many discussions of the inadequacies of present conceptions. For example he asserts: ‘Just as in chess the acceptance of the digital model led to the assumption that the chess player must be using unconscious heuristics, even when the player reported that he was

Minds, machines and phenomenology

65

zeroing in on patterns of strength and weaknesses, the assumption of the pre-existence of rules for disambiguation introduces a process of which we have no experiential evidence, and fails to take seriously our sense of oddness of certain uses.’ Clearly then the ‘information’ which Dreyfus is concerned to have represented involves that of which we have ‘experiential evidence’ including such subjective phenomena as the feeling of ‘zeroing-in’ and our ‘sense of oddness’. In order to deal with the same kind of information as people do, artificial systems would have to deal directly with the sorts of appearances which Arnheim describes. It would not be enough to describe the function, but one would have to simulate the appearances. But this amounts to a request that we reproduce the phenomena rather than simulate them. This can only reveal a basic misunderstanding as to the function of scientific understanding. As Einstein is said to have remarked, it is not the function of science to produce the taste of the soup! me scientist’s task is not to duplicate phenomena but to make them accessible to the intellect. In contemporary Western science this can mean only one thing: The scientist must substitute for the ‘real thing’ a system built on principles which he can understand. The ‘ultimate reality’ is approachable in its manifest entirety by neither science nor revelation, neither by poetry nor mystic illumination. There is no limit to the questions which man can ask and no limit therefore to what in principle can be revealed. The scientist’s task is a never-ending one of unfolding a description which relates both to the phenomena (i.e., the evidence of his senses) and to his capacity to intellectually grasp the description (i.e., to his rational capacities). The phenomenologist sees the task of science somewhat differently. He is not satisfied with an intellectual substitute but wishes to get to the essence of the phenomenon directly or not at all. Since the risks of misunderstanding phenomenologists are great (most have a penchant for obscurity) let us see how a phenomenologist expresses this division. In his essay on phenomenology and linguistics Verhaar (1970) addresses himself to the question of whether language can be formalized in the way Chomsky has proposed, but his argument applies equally to the more empirical type of theories. ‘A theory . . . is a descriptive COUNTERPART of the material (facts, data) (to be) explained; it is a COUNTERPART in the sense that it is, as it were, the epistemological (or ‘scientific’) mirror-image of the material under discussion . . . In contrast, it is typical of the phenomenological frame of reference that a description amounts, in some sense which is still obscure, to a theory; a description is supposed to reflect so immediately the matter described that it cannot be called a COUNTERPART in any methodologically relevant sense; it therefore needs no justification, in its turn, on theoretical ground’ (p. 45). But this is the epistemological side of the problem. While it underlies most of the otherwise inexplicable claims which Dreyfus makes, we must still face a number of substantive questions raised in the book. Again we have to be selective: The marginalia in my copy of Dreyfus’ book cannot be dealt with in one article. We shall pass over such claims as that psychological processes can never be formalized, since these claims are largely a by-product of his epistemological position.

66

Zenon W. Pyljshyn

2. On the question of analogue computation In rejecting the possibility of a digital computer being programed to be intelligent, Dreyfus frequently refers to the analogue nature of mental activity. He rejects the obvious counterargument that a digital computer can be made to do anything which an analogue device can do on the grounds that while a digital computer can simulate an analogue processor it cannot ‘represent’ it because it proceeds in a different manner. For example, he claims that the only way a digital machine could ‘represent’ the division function of a slide rule would be by ‘. . . assigning (colinear) spatial coordinates to the mantissas of two log tables, and effecting a “translation” by subtracting’ (p. 233). (It is not clear if this means that we cannot tell whether the computer ‘represents’ the slide rule without ascertaining the precise location of data in the machine’s hardware!) But this is simply to define the notion of ‘representing’ in such a manner that its only usefulness is to ensure that analogue devices can only be ‘represented’ in an analogue manner. Such a notion is much too strong for any interesting sense in which a system can be thought of as a model of another system. The relation of model to system-being-modelled must be partial and incomplete in important ways. As Sellars (1963, p. 187) has put it, the interpretation of the model as some kind of analogue of a system is possible only if the model is accompanied by a commentary which tells the user which aspects of the system are mapped onto the model and which aspects of the model are relevant to its analogy with the system. Dreyfus wants the relation of ‘representing’ to be so complete and transparent that no such commentary would be necessary. But his is impossible unless the model and the system are so close to being identical that the model can no longer serve as an instrument for understanding the system ~ which, in the case of CS, is its sole purpose. But there are other problems with the proposal that some kind of analogue device may be a better model of mind than is a digital computer. Apart from being guilty of the kind of switching of levels of description which in another .context (e.g., p. 91) Dreyfus finds intolerable, this argument represents a fundamental misunderstanding of the nature and limitations of analogue computation. In this respect, however, Dreyfus is in good company: Virtually the whole of an otherwise important movement in psychology ~ the Gestalt school - was built on a very similar error. In introducing the Gestalt concept of dynamic organization, Kiihler (1929) makes the following observation: ‘In a physical system events are determined by two sorts of factors. In the first class belong the forces and other factors inherent in the processes of the system. These we will call the dynamic determinants of its fate. In the second class we have characteristics of the system which subject its processes to restricting conditions. Such determinants we will call topogruphicd factors. In a conducting network, for instance the electrostatic forces of the current represent its dynamic phase. On the other hand, the geometrical pattern and the chemical constitution of the network are topographical conditions which restrict the play of those forces. It will at once be seen that, while in all systems of nature dynamic forces

Minds, machines and phenomenobgy

67

are at work, the influence of special topographical conditions may be at a minimum in one case and predominant in another’ (p. 64). This leads Kohler to a classification of systems which he feels helps to show why biological systems are different from typical machines: It is because ‘. . . we do not construct machines in which dynamic factors are the main determinants of the form of operation’ (p. 69, whereas this is the dominant form of organization in biological systems. This is in many ways a useful distinction. It enabled Kbhler to introduce an interesting clarification of the nativist-empiricist controversy. Taken as an argument against computer simulation, of intelligence, however (and it is not clear that Kohler would have used it in the manner that Dreyfus does), it has a number of serious weaknesses. In the first place the distinction between the two modes of organization, while useful in those cases where the relation between architecture and mode of operation is simple and transparent, has little force in the case of digital computers. In the second place dynamically organized ‘analogue’ systems are simply incapable of the function required of intelligent activity so long as they retain their pure analogue form. We shall consider these two points below. Viewed as a physical object, the digital computer is indeed highly topographical in its organization. The wiring in it is fixed and completely constrains the flow of electricity among parts. From this perspective its operation, to use Kohler’s terminology, depends on ‘anatomically prescribed conditions’ rather than on ‘dynamic self-distribution of processes’. Described thus it might perhaps come as a surprise that what we actually have is not some fixed-function machine but in fact a universal machine capable of reproducing the input-output behavior of any describable machine - including one which is predominantly dynamic in operation (although it might have to be fitted with appropriate analogue/digital transducers). This occurs because, while the machine is anatomically fixed, it is so constructed that transitions among possible machine states are specified by non-anatomical factors - by setting the machine’s starting state and providing it with external inputs (i.e., programs). Note that the discrete nature of the discrimination of inputs and production of outputs in such a machine is no limitation to its potential for simulating human behavior since human sensory systems are similarly limited and since the size of the discrete steps in the machine can be made arbitrarily small. The more serious objection is one which reflects on the complexity of digital simulations of analogue systems: A function which can be simply computed by an analogue machine may be enormously complex and time consuming when carried out by a digital system. The point is an important one, for such considerations may be more than practical inconveniences ~ especially when we are constrained by real-space and real-time requirements. Furthermore there is the important fact that the form in which we express our theories often makes the decisive difference between our finding the solution to a problem and getting sidetracked into a conceptual dead-end. There is a sort of linguistic relativity principle operating in science: The likelihood of our finding a good solution depends strongly on how easy and natural it is

68

Zenon W. Blyshyn

to express this solution in the theoretical language in which we happen to be working even though it may in principle be expressable in any language. For such reasons 1 would not be surprised if future practical artificial intelligence systems made use of some hybrid analogue-digital machine configurations. This admission that some analogue computation may possibly be expedient in AI systems brings us to our second point, for it might be tempting to argue from this admission that the whole system may have to be analogue. But this does not follow. In the first place the complexity argument works in both directions: There are functions which are much more efficiently computed by digital machines. In the second place an analogue computer is not a universal machine and so does not have the flexibility of function which a digital machine enjoys. This is not a mere technicality but a fundamental limitation of any purely analogue device. It rests on the fact that such a device cannot make a contingent branch in its functioning, doing one thing under one set of circumstances and a completely different thing under a discretely different set of circumstances. Similarly it cannot make decisions because it is incapable of executing the discontinuous function implied by the logical statement ‘If X occurs then do Y’. Of course one could easily add such a capability by building a discrete threshold element into the device. While this gives the device at least two discrete states and therefore may make it possible to convert it to a universal Turing machine, it does not thereby make it a practical device for AI or a perspicuous way of representing cognitive processes. A fundamental limitation to the analogue approach - and the reason that the Gestalt program and its neurophysiological isomorphism postulate ran into difficulty - is that it is at best a way of representing a class of sensory patterns. While various field dynamics may be developed to account for certain sensory transformations and perhaps even constancies, the representation remains always a representation of a particular. But cognition requires the recognition of universals and the processing of concepts. Even if the field effects postulated by Gestalt psychologists were able (in some manner that has never been spelled out) to account for the transformation of the retinal pattern of a chair it would still require a homunculus to into some canonical ‘typical’ chair-pattern, recognize this as an instance of the concept ‘chair’ and to associate it with its discrete verbal label. The arguments I have used against the notion of an ‘image’ as a cognitive representation apply here (Pylyshyn, 1973b).

3. The role of the body The assertion that the body is intimately involved in intelligence is not a claim which is foreign to psychology. It is the cornerstone of Piaget’s psychology and of many contemporary models such as the motor theory of speech perception (Liberman, Cooper, Shankweiler and Studdert-Kennedy, 1967). But the point which needs to be understood (as it is well understood by Piaget) is that the importance of the body is in the genesis of

Minds, machines and phenomenology

69

intelligence and not in its eventual practice. The physico-biological properties of the body (including, of course, the brain) together with the structure of the environment in which it finds itself (e.g., one that contains graspable, moveable, more-or-less permanent objects) determines the way in which conceptual structures are built up. It leads to grouping of sensory patterns into perceptual integrals and even (contrary to Kant) to the intuitions of space, time and causality. But this is not to claim that no machine can be intelligent unless it has a human body. Having built up brain structures, formerly motoric operations become internalized (in Piagetian terms the child moves from concrete sensorimotor intelligence through operational to symbolic and formal intelligence). By the time he is an adult a person’s intelligence depends on his possessing a body only in the obvious sense that his body contains the mechanisms in which intelligence is realized and provides the means for perception, locomotion, etc. To claim otherwise is to suggest that a person who is paralyzed has lost his intelligence! Now if we accept this view that the body is necessary for the development of intelligence (as we know it) then Dreyfus’ claim reduces to the claim that in order to have acquired intelligence in the same manner as people do a machine would need to have had a body sufficiently like that of a human. Except for the question of what we mean by ‘sufficiently like’, this is a perfectly reasonable statement which, however, in no way argues against the possibility of a machine being able to display the type of intelligence which an adult could display if he were restricted to the type of sensors and effecters which we provide the machine. There is no reason why one cannot simulate the current state of an organism (including glandular activity) without having simulated the actual genesis of this particular state. Such a possibility does not depend on any of the assumptions mentioned by Dreyfus although it does depend on a belief in some sort of deterministic materialism - viz., in the principle that all that is effective in the history of the organism is embodied in the present material state of the organism. I do not know whether Dreyfus would wish to argue this principle but I do not see any alternative to it except to reintroduce vital forces.

4. Heuristics and ‘zeroing-in’ Dreyfus is adamant in his rejection of any form of rules as a legitimate way of describing cognitive processes. Most of the attack, however, is aimed at ‘heuristics’ on the grounds that, however sensible the heuristic criteria may appear, heuristics involve searching and searching involves ‘counting-out’ and trial-and-error filtering of alternatives. And, of course, this is not how people proceed. It should be noted, however, that while heuristic strategies typically involve some blind searching, this is by no means an essential aspect of the notion of heuristics in general. In fact it has generally been recognized by AI workers that the best heuristics are ones which result in little or no explicit trial-and-error search. Such search can often be

70

Zenon W. Pylyshyn

reduced or even eliminated by structuring the representation of the problem domain so that the heuristic criteria move the process ever closer to the solution without the need to back-track. The trick is to bring as much available knowledge as possible to bear at each decision node in the problem-solving process. This is the essential difference between, for example, the way in which symbolic integration problems are solved by Slagle’s (1963) SAINT system in contrast to Moses’ (1967) more ‘knowledgeable’ system. More modern programing systems (e.g., Sussman and McDermott, 1972) are explicitly designed to facilitate the use of all available knowledge in working towards their goal - including knowledge gained from the analysis of interim failures. Such systems are approaching Dreyfus’ criterion of ‘zeroing-in’ on the goal. The important point is that the ‘blindness’ of some searches is not a result of the mere fact that heuristics are being used but has to do with the difficult general problem of going from an evolving internal problem structure to the generation of subsequent steps with minimum redundancy. A more serious problem arises if we confuse the sort of heuristic processes which people are aware of using (i.e., ones which appear in thinking-out-loud protocols) with the knowledge that enables people to solve problems. The problem here is that the knowledge people have is not adequately characterized by describing such heuristics. Such knowledge is more abstract and not available to conscious inspection: It is what enables people to generate these heuristic strategies when they are needed (some of them are probably thought up after the fact) and to know that they are the appropriate ones for that occasion. It is also what enables people to restructure a problem representation as part of the process of solving the problem. As I have argued elsewhere (Pylyshyn, 1973a) this epistemological aspect of cognition is much more poorly understood than is the performative aspect. To the extent that we have no theory of how appropriate heuristics are generated, Dreyfus is correct in pointing out that the heuristics incorporated in most problemsolving systems are of limited generality. He correctly .points out that in problem-solving systems it is usually not the system which decides which heuristics are appropriate or which features of the problem are to be attended to, but the programmer who has foreseen the possible circumstances which could be encountered. To the extent that the heuristics are tailored to the forseeable circumstances and are of the type which people are typically aware of using, the system will be limited in its generality. It should be emphasized, however, that what is at issue here is the abstractness and generality of the knowledge which drives the heuristics and not the fact that it was the programmer who introduced the initial structure. As we have argued earlier, unless we wish to simulate the genesis of intelligence, the system can be made ahistoric by providing it with a starting state. That it is the programmer who builds in this starting state by providing the system with the benefits of his experience becomes irrelevant in this case. But he must do so in the most general manner possible, otherwise the accusation that the system is ad hoc may be rightly made. The designer must, in other words, provide a deep structure of knowledge rather than a surface structure of particular methods in his

Minds, machines and p~en#~e~al#~

71

initialization of the system. Such an epistemological deep structure, by the way, is precisely what is referred to by the terms ‘competence theory’ (see Pylyshyn 1972, 1973a).

5. Fringe consciousness and focusing of attention Phenomenologists have always been fascinated by the way in which whole patterns appear suddenly into consciousness from an uneasy and initially confusing perceptual field and by the fact that the meaning and the integrity of the pattern requires the presence of a background which, while it remains outside of the focus of attention, plays an important role in structuring the pattern. To phenomeno~ogists, to subjective epistemologists like Polanyi (1958) and to Gestalt psychologists, such phenomena are at the heart of perception, and no perceptual theory can be adequate unless it captures such qualities. Current cognitive simulation theories do not appear to give such an account of the processes involved in perception and understanding. There are several aspects to this criticism of the CS system. Some of the disagreements have to do with different conceptions of the role of theories as we have already mentioned. But there is more to it than this. There are at least two other difficulties involved in a comparison between CS systems and such phenomenological accounts as that described above. One is that there are certain substantive functional shortcomings in current theories: There are some general problems which phenomenologists can point to which current theories cannot adequatefy deal with. Such problems, to which we shah return below, go beyond our mere inability to simulate states of awareness. The second point is that we do not understand formal programing systems well enough from a philosophical point of view to give a description of what they are doing in terms which might show how such concepts as ‘fringe consciousness’ could be explicated. It is possible that the function of sufficiently complex existing systems (e.g., Winograd, 1972) could be described in such a way that we could see in them aspects corresponding to fovea1 or fringe consciousness, to outer and inner horizons, to self-consciousness, to expectation, to meaningfulness and even to Husserl’s ‘transcendental consciousness’. We cannot rule out such a possibility simply because the machine’s operation, as it is known to us in electronic terms, appears so alien to our human nature. We must remind ourselves that the artificial method by which a machine comes into being cannot be taken as a priori grounds that it must forever remain indescribable in human terms. On the question of functional shortcomings, it has long been known to psychologists that the phenomenon of attention-focusing exhibits some remarkable properties which are still inadequately understood. From the experiments on the ‘cocktail party problem’ through the ‘tip of the tongue phenomenon’ to the ‘new look in perception’, it has been clear that there are levels of attention involved in perception, apprehension and recall whose interrelationship is far from being understood. Attempts to describe the role of

72

Zenon W. olyshyn

attention in perception which utilize such metaphors as sweeping searchlights or tunable filters (e.g., Broadbent, 1958) have failed consistently because the operation of the attention-focusing mechanism itself must depend on the entire field - including parts outside of the focus of attention. Even more sophisticated systems of hierarchical decision functions and continuous attenuation of incoming signals fail because all such systems partition the field at some level into two independent parts - an attended and a rejected set. Such a partition has always proven to be unworkable since it precludes the contents of the rejected set from influencing the processing of the accepted set at a level above the selection mechanism. Similar difficulties are encountered in attempts at building more comprehensive models of memory and language comprehension (for a summary of some of this work see Norman, 1969). Attempts at intermediate levels of attention have been made in several CS systems by the use of activation tags (e.g., Reitman, 1965; Baylor, 1971) or by carrying along a local context (as in Winograd, 1972). Such attempts are only beginnings. Other promising alternatives are also under current development in AI (e.g., Sussman and McDermott, 1972) as well as in logic (e.g., Montague, 1970). As yet, however, there is no serious answer to this particular problem which is perennially being raised by phenomenologists.

6. Parts and wholes Another closely related problem raised by Dreyfus concerns the relation between parts of a perceptual field and the organization of the field as a whole. This issue has been enshrined in the well-known Gestalt slogan that ‘the whole is different from the sum of its parts’. How can the parts of a perceptual event (e.g., the notes of a melody) constitute the whole percept (e.g., the melody itself) when this same percept can occur without these particular parts ~ or in fact without any particular collection of parts. The perceptual event appears somehow to depend only on the relationship among parts. Even here, however, the matter is not so straightforward because in practice what constitute atomic parts of a pattern often cannot be determined without first knowing something about the whole pattern. Thus phonemes cannot be identified without first recognizing words - or at least distinct phonologically well-formed sounds. Consider for example the problem of identifying objects in a scene - say from a high-quality photograph. The qualities and features which are ‘physically distinguishable’ (i.e., distinguishable by a first-level analysis of the output of some readily available measuring device) include such things as reflectivity (grey-level), contrast gradients, light or dark regions, etc. We could go on from such qualities to recognize ‘lines’ (defined as loci of maximum intensity gradients), ‘vertices’ (defined as points of intersection of lines) and eventually ‘bodies’ (using methods such as those developed by Guzman, 1968). The trouble is that these ‘bodies’, ‘lines’, ‘vertices’ and ‘regions’ are not the ones which people would see in the picture! For example lines defined in this manner do not correspond to

Minds, machines and phenomenologv

73

figure boundaries, and figure boundaries more often than not do not produce lines on photographs. Why then do we see lines, regions and objects? The phenomenologist answer to this puzzle is disarmingly simple: Perception is holistic - we do not perceive by constructing a percept like a chair out of elements like lines or intensity gradients, but rather we perceive the chair first as a whole Gestalt and then, if we are so inclined by a misguided atomistic philosophy, reconstruct or reperceive the atomic elements. We see an optical edge at the side of the table even though there may be no contrast change there because we perceive that it is a table and thereby ‘see’ its outline. Unfortunately, however, we are given no account of how this pervasive ability to grasp wholes operates. Gestalt psychologists and Dreyfus feel that it must be a property of ‘dynamic organizing principles’ which can be exhibited only by analogue devices. But, as we have argued, such an explanation gets us nowhere. The problem is that even if perception occurred in some sense holistically and from the top down (i.e., starting with the whole percept) we cannot give a formal description of the percept in such holistic terms. So long as the set of percepts has some structure, so that one percept resembles another in certain respects and yet another in other respects, we know of no way to capture this quality without describing the individual percepts in terms of relations among more primitive elements or qualities. In fact the very notion of structure presupposes the existence of such a reductive relational description. Structures occur in pairs: Relations among distinct objects (paradigmatic structure) implies a set of relations among parts of individual objects (syntagmatic structure). It does not help to speak of a ‘complicated network of overlapping similarities’ or of ‘recognition of the typical’ or of ‘family resemblances’ since these are precisely the symptoms which one wants to explicate. There have been two distinct approaches to the problem of describing perceptual structure. One starts from phenomenal structures and uses intuitions as primary evidence. From the relational structure among percepts it infers a set of rationally derived primitives and defines the structure of individual percepts in terms of relations among these primitives. This is the rationalist approach exemplified by modern generative linguistics. It produces what is essentially an intensional description of percepts. The second approach starts from empirical structures and uses physical measurements as its primary evidence. From the structure among measurements made on local parts of individual fields of stimulation it infers the empirical laws by which local features combine to form global patterns. This is the empirical approach characteristic of most early work in AI as well as some work in psychology. It produces what is essentially an extensional description of percepts. The first approach is more faithful to the phenomenal facts of perception while the second is more useful from a pragmatic point of view since it takes readily available physical measurements as its starting point. The problem which Dreyfus poses (and suggests is unsolvable) is that of bridging the two approaches. We have already alluded to some of the peculiarities of this request as Dreyfus sees the

74

Zenon W. fyvlyshyn

problem. Here we will point out that the problem is not unique to perception, or even to psychology, and that much has already been learned about how such a bridge might be built because of new formal techniques which have been recently developed to describe the problem in more exact terms. For example if we know the set of combinations of atoms or primitives (be they empirically or rationally derived) which qualify as instances of a particular global concept or pattern, we can say something about the formal properties of any mechanism capable of representing (recognizing or generating instances of) that pattern. For example, if we take the global concept to be that of a well-formed formula in the sentential calculus and the primitives to be individual letters and logical symbols, we can assert that the relation between primitive terms and global concept can only be expressed by a recursive mechanism and therefore that it cannot be represented by a finite automaton. The same holds true of the relation between the concept ‘grammatical sentence’ and the set of primitives known as formatives (see Pylyshyn 1972b, 1973a). In the case of more empirical primitives there are similar examples. For example Minsky and Papert (1969) have shown that from a retina consisting of an array of local detectors which define the system primitives, certain classes of geometrical patterns or concepts (e.g., all figures of a given size) can be recognized while others (e.g., all continuous figures) cannot be recognized by certain kinds of devices. In other words the relation between primitives and certain classes of patterns demands that certain ‘computational requisites’ be met. Some devices do not meet these requirements while other devices do. Thus as Minsky and Papert (1972) point out, while the slogan that the whole is different from the sum of its parts represents an important insight, the force of this principle depends on precisely how we interpret the term ‘sum’. Computer scientists and mathematicians are just beginning to make some progress in making this notion precise. In the process they are discovering that there is indeed a sense in which certain patterns are not specifiable in terms of their parts - but that this is a relative statement whose truth depends on the formal properties of the specifying mechanism. Even if this formal study were to make significant progress in the next few years, however, Dreyfus would not see this as a resolution to the part-whole dilemma which he poses. This is because Dreyfus refuses to accept that independently defined primitives are even relevant to the manner in which people perceive whole patterns. His position is that there is never a fixed set of primitives: Rather, what serves as a set of primitives on a particular occasion depends on the whole ‘situation’. A particular physical event presents the human perceiver with an infinity of potential primitives (or what Dreyfus calls ‘traits’) from among which he selects only the essential ones. These, however, cannot be identified in advance since (a) they depend on the situation, and (b) they get their meaning, not from some independent definition, but only in terms of the whole pattern in which they are imbedded. This then brings us back to the same apparent paradox: How can we recognize the pattern as a whole when the identification of the elements which constitute the pattern itself depends on already knowing what the pattern is?

Minds, machines and phenomenologv

75

But this is not a problem unique to perception. It has been an age-old problem in epistemology and the philosophy of science. Here the problem may be stated as follows: What we attend to in the world and how we perceive things depends on our conceptual categories and our theoretical framework, yet at the same time our conceptual categories and theories seem to be influenced by the empirical observations. The two extreme positions on this issue would correspond to the two different approaches to perception mentioned earlier, namely the rationalist extreme and the empiricist extreme. But in spite of the philosophical puzzle, science does progress and the influence does seem to go in both directions - from observation to conceptualization and vice-versa. The recent resurgence in emphasis on the rationalist direction has been largely a response to an overemphasis on the empiricist direction by the contemporary orthodoxy. Our problem has been that we have been misled into thinking of these as mutually exclusive alternatives - as an either-or proposition. Furthermore the most common view has been that neutral signals enter our relatively passive nervous systems and go through a series of progressive abstractions and interpretations. Dreyfus is not the only person who has been active in pointing out the fallacy of this view. Cyberneticians, AI workers and many psychologists (e.g., Piaget) as well as philosophers have been decrying this fallacy for years. For some reason, however, Dreyfus seems to feel not only that this is the predominant view in AI (which is false - see Minsky and Papert, 1972) but that this is the only way in which machines can function. This is certainly not the case. It has long been acknowledged that such an approach is unreasonable, although the actual development of systems which are characterized by a ‘heterarchical’ as opposed to hierarchical organization is relatively new. Modern perceptual systems (see Minsky and Paper-t, 1972) reflect this kind of approach in which low-level systems propose elements which are examined by higher-level systems, which are also in communication with other higherlevel systems and which, in turn, influence both lower, higher, and related systems. Thus a concept such as ‘line’ may be low level in some cases but in others may derive from the most abstract and complete analysis (e.g., it may ‘appear’ only after an object has been tentatively recognized and the higher-level system has passed on its decision to a ‘line proposer’). Such a heterarchy, which is doubtlessly too discrete and logical for the phenomenologist’s tastes, is nevertheless capable of meeting his objections since it shows how general knowledge and contextual factors in the ‘fringe consciousness’ can be brought to bear at all levels of perception - including the pre-attentive field where the ‘fine and numerous traits’ are all potentially effective.

7. Conclusion We have examined Dreyfus’ claims that the information-processing approach to cognitive psychology is fundamentally inadequate and that digital computers are fundamentally

76

Zenon W. ~lyS~.yn

incapable of demonstrating intelligent behavior except in a very narrow and artificial sense. The first claim appears to rest on a notion of ‘understanding’ which is foreign to science - demanding too strict a criterion of equivalence between the process we seek to understand and the theoretical system through which we seek to gain this understanding. We have argued that the type of phenomenological description which fills the pages of Dreyfus’ book can be logically explicated (in the same sense as global terms are reductively explained in physics) in such a way that all the functional aspects are retained and the only thing lost is the poetic style. While it is true that the Loss incurred in this reduction may appear to a non-scientist as a violation of the Dionysian side of our human nature, it nevertheless remains irrelevant to the goal of scientific understanding or to the question of the limits of machine performance. The second claim made by Dreyfus is at best an indictment of the present achievements of AI. It is difficuh to see how one can build an in-principle argument against the possibility of machine intelligence on Dreyfus’ premises. We know that we cannot produce an artifact with all the qualities of a human being without in fact artificially creating a human being. But at the same time we also know that we can produce devices which display a wide spectrum of functions previously considered to be the sole prerogative of human beings. Between these two extremes we shall no doubt find that there are human capacities which we will not be able to reproduce in a device of any one particular type so long as that type is sufficiently unlike the human machine. But where the line is between functions which are reproducible and those which are not is at present a matter of sheer speculation. Perhaps a better theoretical understanding of the nature of computers and programs will give us some idea. Certainly a little theoretical understanding can already rule out such alternative models of mental functions as those based on purely analogue devices. On the other hand examining the implicit assumptions that form the foundation of a discipline is always a valuable exercise, especially in an area as young, exciting and full of wishful optimism as cognitive simulation. We must be thankful to Dreyfus for acting as an antidote to complacency.

REFERENCES Baylor, G. W. (1972) A treatise on the mind’s eye: An empirical investigation of visual mental imagery. Doctoral dissertation, Carnegie-Mellon University. Ann Arbor, Michigan, University Microflims, No. 72% 12,699. Bobrow, D. G. (1968) Natural language input for a computer problem-solving system. In M. Minsky (Ed.), Semantic information processing. Cambridge, Mass., M.I.T. Press.

Broadbent, D. E. (1958) Perception and communication. London, Pergamon Press. Dreyfus, H. L. (1972) What computers can’t do: A critique of artificial reason. New York, Harper & Row. Evans, T. G. (1968) A program for the solution of geometric-analogy intelligence test questions. In M. Minsky (Ed.), Semantic information processing. Cambridge, Mass., M.I.T. Press.

Minds, machines and phenomenology

Guzman, A. (1968) Decomposition of a visual scene into three-dimensional bodies. In Proceedings AFIPS I968 Fall Joint Computer Conference. New York, Spartan Books. Kohler, W. (1929) Gestalt psychology. New York, Liveright. Liberman, A. M., Cooper, F. S., Shankweiler, D. P., and Studdert-Kennedy, M. (1967) Perception of the speech code. Psychol. Rev., 74,431-461. Loehlin, J. C. (1968) Computer models of personality. New York, Random House. Minsky, M., and Papert, S. (1960) Perceptions.’ An introduction to computational geometry. Cambridge, Mass., M.I.T. Press. (1972) Research at the laboratory in vision, language and other problems of intelligence. (A. I. Memo 252). Cambridge, Mass., M.I.T. A.I. Lab. Montague, R. (1970) Pragmatics and intensional logic. Synthese, 22,68-94. Norman, D. (1969) Memory and attention. New York, Wiley. Polanyi, M. (1958) Personal knowledge. Chicago, Ill., Univ. of Chicago Press. Pylyshyn, 2. W. (1972a) Can subjective experience cause brain activity? Amer. Psycho/. , 27,509-510. (1972b) Competence and psychological reality. Amer. PsychoZ., 27,546-552. (1973a) The role of competence theories in cognitive psychology. J. psycholing. Res., 2, 21-50.

-

77

(1973b) What the mind’s eye tells the mind’s brain: A critique of mental imagery.Psychol. Bull., 80, l-23. Reitman, W. R. (1965) Cognition and thought: An information-processing approach. New York, Wiley. Russell, B. (1960) Our knowledge of the external world. New York, New American Library (Mentor Book). Sellars, W. F. (1963) Science, perception and reality. New York, The Humanities Press. Shaw, R. E. (1971) Cognition, simulation and the problem of complexity. J. struct. Learn., 2,31-44. Slagle, J. R. (1963) A heuristic program that solves symbolic integration problems in freshman calculus. In E. A. Feigenbaum and J. Feldman (Eds.), Computers and thought. New York, McGraw-Hill. Pp. 191-203. Sussman, G. J., and McDermott, D. V. (1972) Why conniving is better than planning. (A.I. Memo No. 255A), Cambridge, Mass., M.I.T. A.I. Lab. Verhaar, J. W. M. (1970) Method, theory and phenomenology. In P. L. Garvin (Ed.), Method and theory in linguistics. The Hague, Mouton. Pp. 42-82. Von Neumann, J. (1966) Rigorous theories of control and information. In Theory of self-reproducing automata. Urbana, Ill., Univ. of Illinois Press. Pp. 42-56. Winograd, T. (1972) Understanding natural language. Cog. Psychol., 3, 1- 19 1.

The decline of reason*

ROGER SCHNAITTER Illinois

Wesleyan

University

The editorial initiating Volume 2 of Cognition presents a rationale for a type of argumentation which has become increasingly visible in our discipline over recent years. In short, it suggests that logical and methodological criteria for the evaluation of scientific research be disregarded in those cases where we disagree with the political or social implications of that research and advocates instead a political response. The reasons given for the proposed strategy are utterly without merit. Mehler and Bever state correctly that ‘science does not function in a political vacuum’ and that it is an important matter to understand ‘what motivates scientists to follow certain lines of research’. However, it does not follow that the motivation for a particular line of research is itself relevant to the evaluation of that research, as Mehler and Bever suggest it does when they remark in regard to the current IQ controversy that ‘a line of “research” such as this can only be politically motivated, and it is with political arguments that it must be countered’. Science has a well known history of badly motivated research, and it is clear that such motivation has not always led to bad research or to faulty conclusions. Kepler’s concern with the ‘music of the spheres’ and Fechner’s desire to justify a panpsychist metaphysics may both be considered ill-founded motivations for scientific research, but in neither case can the results of the investigations be faulted for that reason. Scientific conclusions are faulty if they do not follow logically from the evidence or if for methodological reasons the evidence put forward is itself inadequate. The motivation of the scientist can contribute to either problem but does not necessarily do so. While commission of errors may or may not be a function of motivation, the defectiveness of the conclusion is just a function of logical and methodological error, for whatever reason error occurs. The adequacy of scientific and political argumentation is assessed with different criteria. While the scientist evaluates the correctness of a conclusion by looking at its logical and methodological foundations, the adequacy of political argumentation depends on its persuasiveness. Ignoring the importance of a scientific assessment to pursue a * I express appreciation to L. W. Colter helpful comments on the manuscript.

for

Cognition 3(l), pp. B-81

80

Roger Schnaitter

political response removes the possibility of demonstrating logical and methodological flaws in a line of research and its conclusions. While opposition to racism may be unarguable (both Jensen and Herrnstein assert that they are not racists), a number of other politically and socially motivated disagreements may not be as clearcut. For instance, not everyone believes that socialism is a better economic order than capitalism. If a given body of research may be construed as supporting capitalism (let’s say, a cross-cultural study showing consistent patterns of material possessiveness in young children), then an argument to the effect that the research is unsound because it was done by a capitalist and thus presupposes capitalist ideology will not be a convincing refutation to an ideological capitalist. If the research is flawed for logical or methodological reasons, the political line of attack fails, necessarily, to make this clear. On the other hand, one may be persuaded for political reasons that research conclusions are in error when in fact they have no demonstrable logical or methodological flaws. The confusions described above are combined by Mehler and Bever in a curious way to impute question begging in the current IQ controversy. It is asserted that much current IQ research ‘presupposes racist and elitist ideas’. I presume that the editors only mean to suggest that Jensen’s research (1969) as an example, presupposes racist ideology, not that Starr-Salapetek’s research (1971), as an example, does so. Just what is meant by ‘presupposes’ is not clear. If it can be shown that the assumptions, methods and computations of IQ heritability research entail conclusions compatible with racism, then it would not be necessary to do any research at all for these conclusions to be apparent. All one would have to do is inspect the assumptions, methods and computations and deduce the inevitable conclusion. Empirical evidence would be as irrelevant to the conclusion as in any case of analytical truth. If conclusions compatible with racism are so entailed, then anyone doing IQ research, aware of the predetermined outcome of his investigation, would be doing it for ideological reasons presumably; but if conclusions compatible with racism are not so entailed, it cannot be assumed that anyone doing the research is a racist. I do not know of any evidence that this entailment exists for the assumptions, methods and computations used by any of the researchers in question. Thus it cannot be inferred that just because an investigator has engaged in IQ research he is a racist and an elitist and that his research ‘presupposes’ these ideas. Since evidence has not been presented that Jensen, as an example, held racist and elitist ideas many years ago and decided to engage in IQ research to support them, the only reason that may be adduced for the conclusion that Jensen is doing research with racist presuppositions is that his research has come out in a certain way, that is, in a way compatible with racist and elitist ideology. Such reasoning is patently circular. Even if Jensen were clearly and avowedly motivated by racist ideology, however, it does not follow that his research must result in racist conclusions (see paragraph two), nor does it follow that conclusions drawn from his research are in error just because they are politically repugnant (see paragraph three). It is true that due to the confused status of many of the assumptions, methods and computations involved in IQ heritability research, the research is undoubtedly

The decline of reason

81

susceptible to outcomes biased to accord with the investigator’s prejudices. We should all be terribly aware of that danger today, in virtually every research area (Rosenthal and Rosnow, 1969). The proper way to demonstrate these difficulties is not to engage in the sort of political argumentation subject to fallacious reasoning but to demonstrate through clear and rational argument, and perhaps through additional experimentation, just how the research and its conclusions are unsound, logically and methodologically. Science exists in a political and social context, and we should be sensitive to the influence of that context on scientific research. While political considerations can play a valuable role by signalling to us areas where research outcomes may be notably biased, our response should be a careful analysis, according to logical and methodological criteria, of the research in question and its conclusions and implications. The discovery that science is not as objective as some have thought should not turn us to the equally erroneous belief that it is nothing but a miasma of subjectivity. Reason must prevail.

REFERENCES

Jensen, A. R. (1969) How much can we boost IQ

and

scholastic achievement? educ. Rev., 39 (I), l-23. Mehler, J., and Bever, T. (1973) Editorial. 2 (l), 7-11.

Harv. Cog.,

Rosenthal, R., and Rosnow,

R. L. (1969)

Arti-

fact in behavioral research. New York,

Academic Press. Starr-Salapetek, S. (1971) Race, social and IQ. Science, 174, 1285-1295.

class,

Reason

and un-reason

THOMAS

G. BEVER

Columbia

University

JACQUES MEHLER C.N. R.S.

In a recent editorial’ we suggested that every piece of behavioral research must be understood as presupposing a socio-political stance. Often the issues addressed in such stances are diffuse or benign. However, in certain cases, e.g., ‘pure’ research on ‘racial’ differences, the presupposition is definite and may be socially obnoxious. We suggested that such research presupposes racism and must be assessed in that light. Dr. Schnaitter’ has replied to this with a number of points: 1) A scientist’s personal motivation is not relevant to judging his work. For example, Kepler’s romantic concern with the ‘music of the spheres’ did not invalidate his laws of planetary acceleration. 2) Political questions are not matters of logic but of persuasion and should be kept out of scientific interpretation. 3) IQ research does not logically ‘entail’ racism or elitism. 4) Science is not as objective as some have thought, but it is not a ‘miasma of subjectivity’ either. The reply to the substantive points (l-3) is stated or entailed in our editorial. Dr. Schnaitter’s failure to understand our arguments is particularly remarkable in light of his obvious mastery of logic and English. We return to this matter below in conjunction with the fourth point. First we reply to his first argument. We are not concerned with the personal motivations of researchers like Jensen and Herrnstein but with the logical distinctions presupposed by their research program. Similarly the presuppositions inherent to Kepler’s laws of acceleration are scientifically acceptable regardless of his personal motivation, e.g., that there are planets and a sun, that one can view a planetary orbit in time as subtending an angle in a two-dimensional plane, that in conjunction with the orbit as the third side the angle defines a trianguloid area. These presuppositions were necessary prerequisites for the formulation of the constant-area-law governing acceleration changes in orbits. If we turn to Jensen’s writings and ask the same sort of question, we find that his work 1. See Editorial. 2. See preceding

Cog., 2 (I), 7-11 article in this issue. Cognition 3(l), pp. 83-84

84

Thomas G. Bever and Jacques Mehler

must presuppose that there are ‘races’, that there is a society-free capacity called ‘intelligence’, that ‘IQ’ measures it, etc. One may or may not find such presuppositions justified or repugnant; however, they are necessary prerequisites for Jensen’s conclusion that one race is more intelligent than another. In this sense this research ‘presupposes’ racism. Herrnstein presents the parallel presupposition with respect to ‘elitism’. To conclude that the ‘upper classes’ are where they are because of biologically transmitted capacity makes the same assumptions as Jensen (excepting the racial one). In addition this line’of argument presupposes a social world in which the biologically fittest become members of the ‘upper classes’, i.e., that the elite biologically belong where they are. One may agree or disagree that Western society is elitist in this way. However, it is a necessary prerequisite for Herrnstein’s conclusion that social standing is becoming increasingly determined by biological capacity. We agree that a researcher who argues against these authors may choose to accept such presuppositions (e.g., Starr-Salapetek, as suggested by Schnaitter); they may then argue that the conclusions do not follow, either for reasons of logic or of fact. Many authors have done this, and we do not judge their intentions any more than we do Jensen’s or Herrnstein’s. However, it remains the case that many such counter-arguments also presuppose racism or elitism, albeit pro tern argumentis. Often they show that such a presupposition leads to a logical or empirical fallacy: Such a demonstration is a classic method in logic to demonstrate that a proposed presupposition is in fact invalid. In this sense all research on race or IQ does presuppose racism or elitism, at least as a starting point; however, sometimes it is useful to tight real fire with temporarily controlled fire. We do not see our approach as indicating that science is a ‘miasma of subjectivity’. It is instructive that Dr. Schnaitter does see our view as suggesting this. We suspect that he is representative of most of our colleagues. His concern reflects a point that we have emphasized in our editorials and elsewhere: So long as there is no theory of the relation between behavioral science and its social context, scientists will be at the mercy of the latter and society of the former. It is the absence of such a theory that makes our colleagues shrink back from viewing behavioral science as a social phenomenon because such a view indeed makes science look like a subjective miasma. This need not be perhaps the one positive result from the current tedious discussions of ‘race’ will be initial steps towards such a theory. We urge our colleagues to address this problem above all others - it would be nice to know what we are about.