Cognition, Vol. 3, No. 3

1

“If you want to get ahead, get a theory”*

ANNETTE KARMILOFF-SMITH BARBEL INHELDER University of Geneva

Abstract Although

Genevan research has provided a detailed analysis of cognitive structures, our

knowledge of cognitive processes remains fragmentary. The focus macro-development but also on changes occurring in children? sequences

in micro-formation.

behavior is in progress.

A series of experiments

designed

is now not only on spontaneous action to study goal-oriented

This paper describes the action sequences of 67 subjects between

4;6 and 93 years in a block balancing task. It is not a study of children’s understanding of a specific notion in physics, but an attempt to pave the way towards understanding the more general processes of cognitive behavior. The analysis focuses on the interplay between the child’s action sequences and his implicit theories which the observer infers from the sequences

rather than from his verbal comments.

Emphasis is placed on the role

of counterexamples and on shifts in attention from goal to means. The construction and overgeneralization of ‘theories-in-action’ appear to be dynamic and general processes which

are not stage-linked.

structural analogies between

The results also suggest certain functional

rather

than

the acquisition of phqrsical knowledge and the acquisition of

language.

Introduction How can we go about understanding children’s processes of discovery in action? Do we simply postulate that dynamic processes directly reflect underlying cognitive structures or should we seek the productive aspects of discovery in the interplay between the two? This is not an entirely new concern in the Genevan context since in the preface of 27re Growth of Logical Thinking (Inhelder & Piaget, 1958) it was announced rather prematurely that “... the specific problem of experimental induction analyzed from a func*This research was supported by Grant No. 16610.72 of the Fonds National Suisse de la Recherche Scientifique. The authors wish to express thanks to G. Cellerier, R. Garcia and H. Sinclair for useful comments on the research manuscript covering this study. They are particularly grateful to Morris Sinclair for editing the final English version of this paper. Cognition 3(3). pp. 195 - 212

196

Annette Karmiloff-Smith and Birbel Inhelder

tional standpoint (as distinguished from the present structural analysis) will be the subject of a special work by the first author”. Two decades have elapsed. With hindsight it is realized how much more experimentation and reflection were required to undertake the structural analysis. Operational structures are clearly an important part of the picture. They provide the necessary interpretative framework to infer the lower and upper limits of the concepts a child can bring to bear on a task. But they obviously do not suffice to explain all facets of cognitive behavior. Our first experiments that focused directly on processes were undertaken as part of some recent work on learning (Inhelder, Sinclair & Bovet, 1974). The results illustrated not only the dynamics of interstage transitions but also the interaction between the child’s various subsystems belonging to different developmental levels. Though the learning experiments were process-oriented, they failed to answer all our questions (CellCrier, 1972; Inhelder, 1972). What still seemed to be lacking were experiments on children’s spontaneous organizing activity in goal-oriented tasks with relatively little intervention from the experimenter. The focus is not on success or failure per se but on the interplay between action sequences and children’s ‘theories-in-action’, i.e., the implicit ideas or changing modes of representation underlying the sequences. Although what happens in half an hour cannot be considered simply as a miniature version of what takes place developmentally, it is hoped that an analysis of the processes of microformation will later enable us to take a new look at macrodevelopment. While not very typical of our current studies on goal-oriented behavior* where the child needs to devise lengthy and complex action plans, the block-balancing experiment reported here has been selected as a first sample of the current work. Because of the fairly simple nature of this experimental task, a clearer initial picture could be drawn of the interplay between action sequences and theories-in-action.

Experimental

procedure

Subjects were requested to ‘balance so that they do not fall’ a variety of blocks across a narrow bar, i.e., a 1 X 25 cm metal rod fixed to a piece of wood. There were seven types of blocks, with several variants under each type. Some were made of wood, others of metal, some were 15 cm long, others 30 cm. One example of each type is illustrated in Fig. 1. Type A blocks had their weight evenly distributed; B blocks consisted of two identical overlapping blocks glued together, weight being evenly distributed in each block. In A and B blocks, the center of gravity thus coincided with the geometric center of the length of the solid as a whole. We shall refer to A and B types as ‘length blocks’ since dividing the length in half gives the point of balance. The child can succeed without being *See list of Internal’Research Manuscripts, experiments are to be published_later.

following

the list of references.

The results

of these

“Ij’you

want to get ahead, get a theory ”

197

Figure I .*

Motal

-1

TYPQ

-

bar

A

dlfforont

*Arrow underneath each block indicates the point of contact with the bar when the block is in equilibrium. For Type F blocks, this point is a function of the weight of the blocks to be inserted into the cavity.

aware that weight is involved. C types consisted of a block glued to a thin piece of plywood; D types were similar, except that the plywood was much thicker and thus the weight of the glued block had less effect. We shall call C and D types ‘conspicuous weight blocks’ since the asymmetrical distribution of the weight could readily be inferred. E blocks were invisibly weighted with metal inside one end, and F types had a cavity at one end into which small blocks of various weights could be inserted. E and F types

198

Annette Karmiloff-Smith

and Btirbel Inhelder

will be referred to as ‘inconspicuous weight blocks’. An ‘impossible’ block (C type) was also used which could not be balanced without counterweights. The experiment took place in two phases. In phase I, subjects were left free to choose the order in which they wished to balance each block separately on the bar. It was hoped in this way to gain insight into the ways in which children spontaneously endeavor to apprehend the various properties of the blocks: whether they group analogous blocks together, how they transfer successful action sequences from one block to another, and how they regulate their actions after success or failure. Once each block had been placed in equilibrium, children were requested to repeat certain items in a new order proposed by the experimenter. As an experimental precaution to make sure that no psychomotor difficulties would affect the results, subjects were first asked to balance two identical cylinders (2 cm diameter) one on top of the other. When analyzing the results of phase I, it was hypothesized that children interpret the results of their actions on the blocks in two very different ways: either in terms of success or failure to balance the blocks which will be referred to as positive or negative actionresponse, or in terms of confirmation or refutation of a theory-in-action, which will be called positive or negative theory-response. A negative theory-response, for instance, implies contradiction of a theory either through failure to balance when the theory would predict success, or through successful balancing when the theory would predict failure. In other words, the same result was interpreted by children either as positive actionresponse or as negative theory-response and vice versa. Phase II focused on this problem. About half of the phase I subjects in each age group were interviewed again some twelve months later. The purpose was twofold. First, to verify the cross-sectional analysis, the interpretative hypotheses we had made, as well as to determine the progress achieved by each subject. Secondly, since we now had a detailed description of the phase I developmental trends, we wanted to intervene rather more systematically in phase II by providing increased opportunity for positive and negative action- and theory-response, in order to study their interplay during the course of a session. Apart from balancing each of the blocks separately on the bar, subjects were also asked to leave one block in balance and try to balance in front of it on the same bar another block that looked similar but which had a strikingly different center of gravity (e.g., A types with E types), to add to blocks already in balance several small cubes of various size and weight, etc. In both phases, a written protocol was taken by one observer and a continuous commentary was tape-recorded by a second observer on all the child’s actions, corrections, hesitations, long pauses, distractions, gross eye movements and verbal comments.* Unlike many other problem-solving studies, tnis new series of experiments did not strive to keep tasks untainted by conceptual aspects. Indeed, we purposely chose situa*Having completed this level of experimental analysis, our team is presently making a very detailed analysis from videotapes of two subjects undergoing all the different experimental tasks.

“If you want to get ahead, get a theory ”

199

tions in which physical, spatial or logical reasoning was involved but which we had already analyzed from a structural point of view, thus providing additional means for interpreting data. Both for constructing the material and interpreting results of the blockbalancing task information was used from previous research (inter alia, Inhelder & Piaget, 1958; Vinh Bang, 1968; Piaget & Garcia, 1971; Piaget et al., 1973) about the underlying intellectual operations and children’s modes of interpreting weight and length problems. The exact order of presentation of items and the types of problems set were not standardized in advance. Indeed, just as the child was constructing a theory-in-action in his endeavor to balance the blocks, so we, too, were making on-the-spot hypotheses about the child’s theories and providing opportunities for negative and positive responses in order to verify our own theories!

Population Sixty-seven children between the ages of 4;6 and 9;s years from a Geneva middle-class state school were interviewed individually .* Phase I covered 44 subjects; 23 of these children were interviewed again in phase II. Five young subjects between 18 and 39 months were observed in provoked play sessions with the blocks. In the results, some rough indications are given of the ages at which the various action sequences and theories-in-action are encountered, but this should not imply that the processes described are considered to be stage-bound. 22 subjects from this experiment were also asked to perform quite a different task, that of constructing toy railway circuits of varying shapes, and it was quite clear that children interpreting success or failure as theory-responses in one task might be interpreting success or failure as action-responses in the other task. Furthermore, similar action-sequences for block-balancing were encountered not only in many children of the same age but also during the course of a session with children of very different ages. On the basis of children’s verbal explanations of the relevant physical laws, the developmental trend falls into neat stages. If the analysis is based on children’s goal-oriented actions, this is not so obviously the case. However, both the nature and the order of action sequences as described in our results were overwhelmingly confirmed by changes during sessions as well as the longitudinal results of phase II.

*Appreciation is extended to the staff and pupils of Pinchat Infants’ and Primary School, Geneva. Faculty students M. H. Lavanchy, C. Mainardi, M. Mergulies, C. Mulliez, P. A. Schneider and J. Ziircher are thanked for their assistance in data collection.

200

Annette

Observational

KarmilofjSmith

and Barbel Irlhelder

data

We felt it would be instructive to have an indication of how very small children go about balancing blocks. Accordingly, five subjects between 18 and 39 months were observed in provoked play sessions with our experimental material. This led us to interpret the older subjects’ seemingly anomalous behavior in conflict situations as rather clearcut regressions to earlier patterns. None of the five subjects failed in balancing the two cylinders used for checking psychomotor problems. As far as the experimental blocks were concerned, it was possible to coax the children into trying to balance a few blocks, but only for very short periods of time. Nevertheless, what they did was often organized. The following was the basic pattern: Place the block in physical contact with the bar at arzy point (e.g., extremities, center, pointed edge, side, etc.), let go, repeat. Their attention was frequently diverted to the noise made by the falling block; indeed, the two youngest subjects rapidly made their goal that of causing a loud noise. Gradually, with the first chance successes on the 15 cm “length blocks” (easier than the 30 cm ones), the three oldest subjects (32 - 39 months) lengthened their action sequences in a systematic way, as follows: Place the block at any point of physical contact with the bar, push hard with finger above that point of contact, let go, repeat immediately. However, these subjects did not move the block to another point of contact before letting go, although they had consistently done so when trying to balance the two cylinders or when building towers or houses with wooden cubes. It would appear that in such cases they were simply forming the parallel plane surfaces; in other words the problem of finding the appropriate point of contact between two objects of different shape did not arise. Nonetheless, even though the subjects placed the blocks at random points of contact, further development of action sequences by pushing on the block above the point of contact (i) seemed to denote that the need for spatio-physical contact had become clearer for the child, his finger acting. somethat like a nail and thus simplifying the balancing problem and, (ii), provided the child with an indication through diffuse proprioceptive information of the fact that blocks have properties that counteract his actions. The experimental material was in fact designed to allow for proprioceptive information. The experimental material was in fact designed to allow for proprioceptive information. In previous research (e.g., Inhelder & Piaget, 1958; Vinh Bang, 1968) on equilibrium with a fixed fulcrum, subjects could only obtain visual information which then had to be expressed via another mode of representation. Experimental

data

Unlike previous Genevan research articles in which extensive quotations were given from what children said, this study’s protocols consist mainly of detailed descriptions of children’s actions. We shall, however, occasionally refer to children’s spontaneous comment when it is particularly illustrative. Here we will describe those action sequences

‘Tf you want to get ahead, get a theory”

201

that were encountered among most children of a given age and repeated several times by each child on the various blocks. Many of the subjects in the experiment proper started the session in a similar way to the young subjects just described in the observational data. However, what seemed to take place developmentally between 18 and 39 months was observed during part of a single session among 4 - 6 year olds. Thus the initial approach was as follows: Place block at any point of contact, let go; this was immediately followed by a second attempt with the same block: Place at any point of contact, push hard above that point, let go. As the block kept on falling, the children gradually discovered through their act of pushing that the object had properties independent of their actions on it. Negative response sparked off a change from an action plan purely directed at the goal of balancing, to a subgoal of discovering the properties of the object in order to balance it. These children then undertook a very detailed exploration of each block trying one dimension after another, as follows: Place the block lengthwise, widthwise, upended one way, then the other and so forth. Such sequences were repeated several times with each block. Although the order of the dimensions tried out differed from one child to the next, each child’s exploration of the various dimensions became more and more systematic. Yet only one point of contact was tried for each dimension - for quite some time the children would never change the point of contact along any one dimension. Frequently, even when children were successful in balancing an item on one dimension (e.g., at the geometric center of length-blocks or along the length of the bar), they went on exploring the other dimensions of each block. It was as if their attention were momentarily diverted from their goal of balancing to what had started out as a subgoal, i.e., the search for means. One could see the children oscillating between seeking the goal and seeking to ‘question’ the block. Successful balancing was certainly registered since, in the course of their investigation of the blocks, the children returned more and more frequently to the dimension that had yielded the solution. Yet they continued to explore, as if seeking an alternative solution. Although such behavior seems to indicate that the child is beginning to differentiate between his own actions and the properties of the block, this seeking for alternative solutions may be interpreted to mean that the child does not yet understand that a physical law, unalterable by his action, governs the object’s behavior. We hypothesize that during his exploration the child is endeavoring to make a sort of ‘catalogue’ of the different actions he may make on the blocks; once he has set the limits of these actions, he is then in a position to narrow them down to those that appear to be more relevant to his goal. More advanced subjects did not lose sight of their goal; once they found a solution, they registered the positive response and no longer explored the block in any observable way. As balancing of length-blocks became progressively easier, children attempted to balance all items on their longest, flat dimension, only reverting to ‘exploration’ and even to ‘pushing hard above the contact point”when grappling with the difficult item G. After opting for the longest dimension and retaining some form of representation of a previously balanced block on that dimension, children began their first real search for the

202

Annette Karmiloff-Smith

and Birbel Inhelder

Figure 2.

Succa~sfulbalancing

Unsuccessful from

provnus

attempt

- posltlon

transferrod

success

effective point of balance. Here is where spatial symmetry, already so prevalent in young children’s behavior in solving other problems, is used by the child as possibly relevant for success. Action sequences were as follows: first place more or less symmetrically on the support bar, i.e., close to the geometric center, correct in the right direction guided by the sensation of falling (adjustments were rarely made in the wrong direction), readjust in the other direction (corrections were frequently excessive), continue correcting back and forth but gradually more carefully until equilibrium is achieved. It was already suggested that for some time children treat the problem as one not governed by a physical law. By contrast, what was particularly apparent at the next level, either developmentally or later in a session, was that all blocks were systematically first tried at their geometric center. Almost all children aged 6 - 7;5 and also some younger subjects, did this. Here we witness the beginnings of an important theory-in-action (i.e., spatial symmetry or as the children put it “I’ll try the middle first”, “...half-way along”), which was generalized to all blocks and was to pervade behavior at the next level. Action sequences ran as follows: Place at geometric center, release hold very slightly to observe result, correct very slightly, correct a little more, return carefully to geometric center, repeat until balance is achieved. Depending on the block, the further the child had to move away from the geometric center, the more often he returned to it before further adjusting. These frequent returns to the geometric center seem to denote that the child is using a form nf spatial notation as a known location from which to orient his corrections and are an obvious prelude to the ‘geometric center theory’ that develops next. Another important change took place at this level. The children no longer seemed to be merely registering the fact of balance but beyond that something about the balancing position of the block. The new action sequence, which alternated with the one described previously, was as follows: First place at geometric center, next place at the point of contact corresponding to the previous success (irrespective as yet of the differences between the two blocks and often far removed from possible empirical success - see Fig. 2), return to geometric center, continue as in previous sequence. At this point, some corrections were made away from the center of gravity, a feature to be discussed later.

“If you want to get ahead, get a theory”

203

Notable in the various action sequences of these children was the interplay between the endeavor to use information acquired from previous actions and the gradual introduction of a coherent, analogous approach to all items. Interestingly enough, most of those subjects who had been successful in balancing conspicuous or inconspicuous weight blocks because they had been concentrating on the goal, began later in the session to experience serious difficulties in repeating the successful action as their attention shifted to the means. They now placed the very same blocks more and more systematically at the geometric center, with only very slight corrections around this point. They showed considerable surprise at not being able to balance the blocks a second time (“Heh, what’s gone wrong with this one, it worked before”). This was in fact to be the dominating behavioral pattern for a long period to follow, either during the session or developmentally. Action sequences then became reduced to: Place carefully at geometric center, correct very slightly around this center, abandon all attempts, declaring the object as ‘impossible’ to balance. It is interesting to note that the child remained unflinchingly at the geometric center and completely ignored the proprioceptive messages that had proved useful earlier. What is striking is that when these very same subjects were asked to close their eyes and attempt to balance inconspicuous weight blocks, success was very rapid. But once they reopened their eyes, they again resolutely tried the geometric center and expressed surprise at their blind success. In fact we observed more failures to balance blocks among 5;6 to 7;s year olds than among 4;s to 5;.5 year olds. How can this be explained? It would appear that a strong theory-in-action, i.e., that the center of gravity necessarily coincides with the geometric center of an object (or as one child put it “things always balance in the middle” though others did not express verbally what was obvious in their actions), pervaded the actions of the older children. Not only did they place all blocks at their geometric center with substantially no corrections of the position, but when asked to add small blocks of varying shapes and sizes to blocks already in balance, they added up to ten blocks precariously one on top of the other at the geometric center rather than distributing them at the extremities, as did both the younger subjects (“it’s like a see-saw, you put a block at each end”) and those 8/9 year olds who placed blocks equidistant from the point of contact. If this ‘geometric center theory’ is so pervasive and persistent despite negative theoryresponse (blocks fall when placed at the center or balance at points other than the center), how does the child ever come to change it? We postulate three interacting causes: (1) The ever-increasing regularity of counterexamples, (2) changes taking place in the child’s general conceptual competence and (3) the integration of the earlier proprioceptive information into a theory-in-action. Let us take up each of these points separately. Frequent counterexamples do not alone induce a change in the child’s behavior. If they did, then progress could be achieved by simply providing a large number of counterexamples. The child must first form a unifying rule based on regular patterns he has observed: In this study, the geometric center theory, which in fact accounts adequately for some blocks and for many situations in his daily life. Only when this theory is

204

Annette

Karmiloff-Sttzith

atd Barbel lthclder

really consolidated and generalized, is he ready to recognize some form of unifying principle for the counterexamples which he earlier rejected as mere exceptions (‘impossible to balance’). One obvious reason for giving up the geometric center theory would be that the child considers weight and no longer exclusively length as being relevant. From previous studies, it is known that around 7 years children consider weight to be a significant property in problems involving equilibrium, and it is not until even later that the child differentiates between weight as an absolute property and weight as a force. Is this to imply that younger children are unaware that objects can be heavy or light or that imbalance is caused by heaviness? No, but what younger children do not understand is the role of weight in situations of balance. As the child attains conservation of weight and recognizes weight as being relevant in other situations, so in recognizing the regularity of counterexamples he looks by way of explanation for some aspect other than length, i.e., weight: (“Oh, it’s always just the opposite of what 1 expect . . . maybe it’s this block glued to the end here”). And the first corrections away from the geometric center took place with conspicuous weight blocks (type D) while inconspicuous weight blocks were still resolutely centered. Though children may not have a conceptual understanding of the role of weight in equilibrium, it should be recalled that the younger children did react to weight communicated proprioceptively. It thus seems plausible that once weight is considered to be relevant, the proprioceptive information is integrated gradually into the child’s theory-inaction. Interestingly enough, many corrections at this level were made away from the center of gravity, denoting that the correction did not stem from proprioceptive information but from a conceptualized need for a change of position. Younger subjects relying on proprioceptive information alone rarely if ever made corrections in the wrong direction. We have seen that three interrelating factors seem to bring about corrections to the weight items as distinct from the length items. Once this happens, do children then easily change their geometric center theory and opt for a new and broader one? It is hypothesized that for some time they tend to hold on to the earlier theory. Among the conspicuous weight blocks, for instance, type D (thick plywood base) were clearly easier for these subjects than type C (thin’plywood base on which the weight of the glued block has more effect), because the center of gravity of D blocks is much closer to the geometric center than in C blocks. Correcting a D block from its geometric center to its point of equilibrium is thus less of a challenge to the child’s geometric center theory than a C block which balances much nearer its extremity. Furthermore, even when adjusting all conspicuous weight blocks children continued resolutely to center the inconspicuous weight blocks. It would thus appear that the geometric center theory was not abandoned when the child started taking negative responses and weight into account; it was retained for most situations where the theory could still hold true and a new theory, quite independent of it, was developed to deal with the most obvious exceptions. Length and weight were thus considered indepen-

“If you want to get ahead, get a theory”

205

dently, which was apparent not only from the child’s actions but also from the explanations he gave at the end of a session. For length blocks, weight was considered to play no role and symmetry of length was the sole property invoked; for conspicuous weight items, weight was used in the explanation. (For A blocks: “it’s the same length each side, there’s no weight”; for C blocks: “it’s heavy on one side and long on the other . .. no in this one (D) there’s weight all the way along the bar, in that one (C) there’s only weight where the block is”). Gradually, and often almost reluctantly, the 7 to 8 year olds began to make corrections also on the inconspicuous weight blocks. It is to be recalled that 4 to 5 year olds did this immediately, but for different reasons. Whereas the young subjects were relying solely on the proprioceptive information, the older subjects (73 - 8;6 years) provided explicit references not only to equal length but simultaneously to equal weight for all items balanced at their geometric center (types A and B). Previously length alone had sufficed to explain equilibrium. At this point, we observed many pauses during action sequences on the inconspicuous weight items: Place at geometric center, correct slightly, pause, lift object, rotate object, pause, place at geometric center, correct position slightly, release hold slightly, readjust carefully, pause longer, glance at a conspicuous weight item, pause, place again slowly at geometric center, shake head, glance again at conspicuous weight item, then suddenly correct continuously and rapidly in the right direction until balance is achieved. Repetition of a success was thereafter immediate, even if the object was rotated. As the children were now really beginning to question the generality of their geometric center theory, a negative response at the geometric center sufficed to have the child rapidly make corrections towards the point of balance. We obviously do not mean to imply that he had a full, explicit understanding of the inverse relationship of weight and distance, but simply that he now implicitly understood the importance of both length and weight. To give an indication of action sequences at the upper level, by 8;7 years of age children paused before each item, roughly assessed the weight distribution of the block by lifting it (“you have to be careful, sometimes it’s just as heavy on each side, sometimes it’s heavier on one side”), inferred the probable point of balance and then placed the object immediately very close to it, without making any attempts at first balancing at the geometric center. These subjects demonstrated, when adding extra blocks to already balanced items, that the point of balance was for them not a fixed, but a relative point, for they could alter it several times by correcting simultaneously the fulcrum and the position of the added blocks. Both negative and positive responses had immediate effect. Adding two equal blocks to an inconspicuous weight block already in balance was not made at the two extremities, as the otherwise successful 7 to 8 year olds would have done, but at an equal distance from the center of gravity. The principles of the law of moments were at this point apparent in the child’s actions. A number of other apparently minor aspects of behavior seem important. There is for example the manner in which children held the blocks. The very youngest children

206

Annette Karrniiofjr-Smith andkirbel

Inhelder

Figure 3.

and many others at the outset of a session tended to hold blocks from above in one hand when trying to balance them and then to push with the same hand on the block above the point of contact. Next children tended to hold the blocks at each extremity with both hands, which is a much more informative method proprioceptively. Thus, the rather diffuse proprioceptive information by pushing with one hand above the point of contact, became a definite feeling of heaviness in one of the child’s hands, indicating the side on which the block would fall (see Fig. 3). However, the child gradually seemed to realize that one of his hands was in fact superfluous. As his attention moved from his own actions on the blocks to the properties of the blocks, the child attempted (not without difficulty) to replace his hand by placing a small cube on top of the block as a counter-weight, or replaced his other hand by a cube underneath the block; thus providing a second support. The latter was not satisfactory for the child since the small cubes were purposely all lower or higher than the vertical distance between the table and the point of support: thus the block remained slanted. This again reinforced the need for either the use of counter-weights (rarely used spontaneously, except by the older subjects) or for corrections of the point of contact along the dimension chosen. A second aspect was the order in which subjects chose to treat items in phase I where choice of order was left to the child. Many children (including all the youngest but also a certain proportion in the other age groups up to 7 years) chose blocks closest to them on the table in a haphazard order, not bothering to group identical or similar ones together. It was as if each block constituted a separate problem for them and as if there was no anticipation of the differences in complexity nor any effort to transfer acquired information intentionally during these initial attempts. Furthermore, when, after completing

“If you want to get ahead, get a theory”

207

the free choice session, we then imposed a more ‘informative’ order (e.g., following C by D) children did finally begin to make transfers from one block to another. Only very gradually during a session, or more systematically for older subjects (as of roughly 7 years and including all subjects above 8;7 years) did children from the outset stretch out for blocks so as to group them in an organized fashion. When the children dealt with two seemingly analogous blocks, either spontaneously or because of the experimenter’s imposed order, there were many erroneous yet informative transfers from one item to the next.

Discussion What are the processes of discovery in history, in ontogenesis and during an experimenta: session? In his interesting book on Darwin’s creativity, Gruber (1974) suggests that “there is nothing necessarily creative in being immediately trapped by every original thought one has”. indeed, in our particular field of enquiry, the processes of cognitive discovery involve far more than simply isolating the properties of weight and length. We have seen that action sequences are not merely a reflection of the child’s implicit theories. The very organization and reorganization of the actions themselves, the lengthening of their sequences, their repetition and generalized application to new situations give rise to discoveries that will regulate the theories, just as the theories have a regulating effect on the action sequences. What is the role of experience or of the object’s ‘responses’ in generating change? It may seem surprising that an event as obvious as the falling of a block is not always clearly evaluated as negative theory-response but merely as negative action-response. While the child is solely success-oriented, all balancing - irrespective of which block will be read as ‘positive’ information and all falling of blocks as ‘negative’. However, if considered from the standpoint of theories-in-action, then the child’s interpretation of the object’s ‘response’ will depend on the block. For instance, for a child with the geometric center theory, conspicuous or inconspicuous weight blocks in balance will, when not rejected as mere exceptions, be read as negative theory-response since such balancing indeed counteracts the child’s theory-in-action. On the contrary, the balancing of a length block and the falling of any block placed at a point outside the geometric center will be read by the child as positive (positive to his theory, though not immediately to his goal). And the reading of identical events can change from positive to negative or vice versa as the child modifies his theories. Positive and negative action- and theory-responses seem to have varying roles at different times. While the child is exclusively success-oriented, i.e., concentrating on balancing, positive action-response is all-important. It encourages the child’s natural tendency to repeat successful actions. Then gradually negative action-response shifts the child’s attention to the means, i.e., ‘how to balance’. At this point we witness experi-

208

Annette Karmiloff-Smith

and Birbel Inhelder

mentation for experimentation’s sake; for attending to the means implies seeking knowledge of the approximate range of possible actions on an object. Koslowski and Bruner (1972) found a similar phenomenon when experimenting on young babies’ use of a rotating level to obtain a distant object. The authors report a stage when babies continued experimenting with the various movements of the lever, thus becoming ‘enmeshed in the means’, ignoring the goal which they had meanwhile moved within their reach, whereas previously they had been solely goal-oriented. As the child gradually begins to construct a theory for interpreting the regularity of positive action-responses, these become positive theory-responses. Negative responses remain action-responses until the child’s theory is generalized and consolidated, after which they progressively become negative theory-responses, once the child becomes aware of their regularity. A further important fact is that younger subjects make use of proprioceptive information in an uncontaminated fashion since they have not yet developed a unifying theory. For the more advanced subject the object’s ‘behavior’ is evaluated conceptually, and they are only able to use the proprioceptive information if they close their eyes. As long as the child is predominantly success-oriented, there are rarely any pauses in his action sequences. As his attention shifts to means, however, pauses become more and more frequent in the course of the sequence. Only when goal and means are considered simultaneously do pauses precede action. Such differences in the occurrence of pauses are potentially meaningful for the observer. Although negative responses are a necessary condition for progress, they are clearly not a sufficient one; in order to be effective, such responses must counteract a powerful theory-in-action, such as the geometric center theory. It should be noted in passing that we purposely chose to speak of ‘theories’ and not of ‘hypotheses’ since the latter tend to have the connotation of intentionally seeking to verify. Previous research (Inhelder & Piaget, 1958) showed that the formal operational child was frequently capable of attempting to test hypotheses and verify theories by deliberately seeking counterexamples. In those experiments, however, tasks were often well beyond the capacities of the concrete operational child; his experimental method was therefore defined in terms of what was lacking as compared to the formal operational child. Our present research was aimed at elucidating the positive aspects of the younger child’s behavior by using simpler situations and looking more closely at how he goes about his task. Our observations indicate that the younger child does not intentionally seek counterexamples; rather, children in this research, whatever their theory-in-action (from the elementary need for physical contact to the more sophisticated geometric center theory) constructed and generalized theories, and gradually recognized counterexamples. However, the earlier claim (op. cit.) that “at the concrete level, the child does not formulate any hypotheses” needs to be reconsidered in the light of the strong tendency of our children to act under the guidance of a powerful theory-in-action which involves far more than mere observation of immediate empirical reality. Nonetheless, their theories remain implicit, since the

‘lf J,ou want to get ahead, get a theory”

209

younger child clearly cannot reflect on hypothetical situations which might confirm or refute his theory. Indeed, although the child’s action sequences bear eloquent witness to a theory-in-action implicit in his behavior, this should not be taken as a capacity to conceptualize explicitly on what he is doing and why. Recent experimental work (Piaget et al., 1974a and 1974b) has confirmed that a developmental gap exists between succeeding in action and being capable of explaining it. Our present analysis is focused less on particular, explicit notions than on the gradual unification of ideas as observed in action sequences. There is no doubt that the generalized application of a theory will ultimately lead to discoveries which in turn serve to create new or broader theories. However, it seems possible for the child to experience surprise and to question his theory only if the prediction he makes emanates from an already powerful theory expressed in action. Our observations indicate that children hold on to their initial theory for as long as they can. Even when they finally do take counterexamples into consideration, they first prefer to create a new theory, quite independent of the first one, before finally attempting to unify all events under a single, broader theory. The tendency to explain phenomena by a unified theory, the most general or simplest one possible, appears to be a natural aspect of the creative process, both for the child and the scientist. The construction of false theories or the over-generalization of limited ones are in effect productive processes. Overgeneralization, a sometimes derogatory term, can be looked upon as the creative simplification of a problem by ignoring some of the complicating factors (such as weight in our study). This is implicit in the young child’s behavior but could be intentional in the scientist’s. Overgeneralization is not only a means to simplify but also to unify; it is thus not suprising that the child and the scientist often refuse counter-examples since they complicate the unification process. However, to be capable of unifying positive examples implies that one is equally capable of attempting to find a unifying principle to cover counterexamples. The history of physics abounds in examples analogous to those we have observed among children. Dugas (1950) recalls that the really creative scientists were those who did not merely study the positive examples of well-known principles but who endeavored to extend the principles to other phenomena. It is in this way that scientists, and children, are able to discover new properties which in turn make it possible for new theories to be constructed. And psychology experiments on adults (Claparede, 1934; Miller, Galanter & Pribam, 1960; Wason and Johnson-Laird, 1972) illustrate the general tendency to construct powerful, yet often inappropriate, hypotheses which adults try to verify rather than to falsify. This temporarily blinds the adult to counterexamples which should actually suffice to have them reject their hypothesis immediately. Our present results on young children seem to point to the fact that constructing and extending a powerful theory-in-action is a very general aspect of discovery which has a deeprooted function. The observations stemming from this study suggest some thought-provoking analogies between the processes of acquiring physical knowledge and language. Ruth Weir’s

2 10

Annette Karmiloff-Smith

and Btirbel Inhelder

fascinating recordings of language in the crib (1962) show how small children use language not only for symbolic representation and direct communication, but also how they ‘explore’ language by trying out in presleep monologues series of word combinations, almost as if they were looking for the most appropriate-sounding ones. Children’s exploration of each block’s dimensions in this study is not an unsimilar process. Furthermore, many authors (e.g., Ervin, 1964; Klima & Bellugi, 1966) have observed that certain morphological markers such as the plural, the past-tense, negation, etc., first enter the child’s corpus in an unanalyzed, isolated fashion, rather than as part of a unified system. Young children are able to assimilate irregular forms they hear such as ‘feet’ and ‘went’, and first produce them in their correct form, alone or in combination with other words. Such utterances seem as yet ‘uncontaminated’ by rules. or by what might be called implicit ‘theories-in-language’, and thus very comparable to the way in which our young subjects were able to make use of proprioceptive input, uncontaminated by a theory-inaction. Just as our subjects gradually recognize regular patterns in the objects’ responses and construct a theory-in-action, so children progressively recognize patterns in their linguistic environment and construct ‘theories-in-language’. The tendency to generalize these implicit theories results in the temporary disappearance of the correct, irregular forms in favor of theory-revealing errors (e.g., ‘foots’, ‘goed’, etc.), just as successful balancing of weight blocks ceases temporarily in favor of the geoinetric center theory. It is clearly not sufficient to provide the child with frequent counterexamples to induce him immediately to drop this theory. A striking example of this phenomenon in language is seen when a child repeatedly produces errors which resist adult corrections until the child is ready spontaneously to recognize that there are exceptions to his implicit theory-inlanguage. When the correct linguistic forms do reappear, they are no longer unanalyzed entities as in the first phase, but exceptions to a unified system, just as subsequent successful balancing of weight blocks is the result of a broader, implicit theory. Attempts to compare language and cognition were initially made by authors comparing syntactical structures and concrete operational structures. The more recent trend has been to analyze, both theoretically and experimentally, the sensory-motor foundations of language and thought (e.g., Sinclair, 1971; Inhelder et al., 1972; Brown, 1973; Sinclair, 1973). It may be that a comparison between the experimental results of the Genevan psycholinguistics team and our new data on goal-oriented behavior, will prove to have a heuristic value in leading us closer to the general processes underlying cognitive and linguistic behavior.

REFERENCES Brown, R. (1973) A first language: The early stages. Harvard University Press, Cambridge, Mass. Celldrier, Guy (1972) Information processing tendencies in recent experiments in cognitive learning Theoretical implications. In S. Farnham-Diggory (ed.), Infkwmafion Processing in Children. Academic Press, New York, pp. 115-l 23.

“If you want to get ahead, get a theory ”

2 11

Claparcdc. 11. (1934) I,u get& de I’/zJ’poth&e. Kundig, Geneve. Switzerland. Dugas, Rent? (1950) Hirtoirede la M&unique. I
Intern ‘1Research Manuscripts on children s goal-oriented strategies (Provisional

titles, translated

into English,

indicate

sphere

of research)

Blanchet, A. (1973/74) Constructing multi-stage mobiles. Coil, C. (1974) Discovering maximum number of routes for loading lorries. Karmiloff-Smith, A. (1973) Building closed railway circuits of varying shapes and sizes. _ (1973/74) Balancing of blocks of varying physical properties. Kilcher, H. (1973) Building bridges with materials of varying physical properties. _ (1974) Constructing instruments to move objects. de Marcellus, 0. (1973) SkittIes. _ (1974) Modifying order of locomotives and carriages in a closed circuit. Montangero, J. (1973/74) Controlling water levels by immersion of combinations of objects with varying properties. Robert. M. (1973) Spontaneous exploratory activities in observed play with Russian Matriona dolls. _ (1974) Building staircases with various sized blocks.

2 12

Annette Karmiloff-Smith

and B&be1 Inhelder

Valladao-Ackcrmann, 1,:. (1973) Simplified chess situations. ._ (1973/74) Combining serial orders of boxes and ladder rungs (1974) Path puzzle leading to combined destinations. Wagner, S. (1974) Constructing resisting chains for hauling weights.

Cette rccherche s’inscrit darts le courant actucl des etudes psychoginctiques gencvoiscs qui, sans negliger le cadre structural des fonctions cognitives, s’oricntent resolument vcrs I’analysc dcr processus. Au-deli de l’etude de la macro-gem&e dcs ctapes du dcveloppcment, nous nous intcrcssons plus particulicrement a la micro-formation des processus de dccouvertc au tours dcs seances expcrimcntales. Nous prdscntons ici un premier excmplc de l’cnsemble de travaux destines i rcndre compte de l’organisation des scqucnccs d’action orientees vers un but. Notre analyse Porte sur 67 enfants dc 4;6 a 93 ans qui ont i rdsoudre un probliimc d’iquilibre de plots SUI un support ctroit. Moms cent& que darts le passe sur l’explication de notions physiques particulieres (en l’occurrcncc Ic principc dc gravitc) notrc etude portc esscntiellcment sur I’aspect dynamique de l’interaction entre l’organication des sequcnccs d’action du sujet ct ses theories implicitcs. Plusicurs moments importants du processus de dccouvcrtc ont retcnu notre attention: les changcmcnts d’orientation du but aux moycns et vice versa, lc role des contrc-exemplcs ct celui surtout de la sur-gencralisation dcs theories implicitcs cn tant quc factcurs d’organisation de la connaissancc. Enfin, les resultats suggcrcnt dcs analogies de nature fonctionnelle plutbt que structurale entre la formation dcs connaissances physiques et cclledu langagc.

2

Language and the ability to evaluate contradictions and tautologies*

DANIEL

N. OSHERSON

University of Pennsylvania ELLEN

MARKMAN

University of Illinois “Logic

does not by any means treat

does not treat about

objects;

and

universal

tion

of logic

objects

of objects

logic is first generated validity,

derives

or better, just

of the totality

at all but only

from

the

the fact

of things,

it

of our way of speaking

by language. irrefutability that

The certainty of a proposi-

it says nothing

about

Hans Hahn

(1959)

of any kind.”

Abstract Children were found to experience difficulty evaluating contradictions of the form p 6 -p, and tautologies of the form p v -p. It was hypothesized that (a) the difficulty of these statements was not due solely to the logical words occurring in them, (6) part of the difficulty is due to the fhct that their truth value derives from their linguistic f&-m rather than from empirical considerations, and (c) the ability to examine language in an objective manner, apart from events and objects to which it refers, is necessary but not sufficient for correct evaluation of contradictions and tautologies. The results of two experiments support the hypotheses. During our informal conversations with elementary school children, we found that they had difficulty judging the truth-value of sentences corresponding to simple contradictions and tautologies. Roughly speaking, a tautological sentence is true by virtue of its logical form. A contradictory sentence is false by virtue of its logical form. Children were asked

*The authors thank Joanne Duncan of the Llanerch Elementary School, and Mary Ostrum of the Portola Valley Elementary School for their cooperation during this study. We also express our appreciation to Rachel Gelman and Burton Rosner for their careful reading of earlier versions of this manuscript. We thank Barbara Abrahams for administering Experiment 1. The first author was supported by an NSF predoctoral traineeship. The second author was supported by PHS Training Grant No. HD 00337-03 from NICHHD. The research was partially supported by Grant No. NIHHD-MH 04598 to Rachel Gelman. The study was conducted while both authors were at the University of Pennsylvania.

Co@tion S(3), pp. 2I3 - 226

2 14 Daniel N. Osherson and Ellen Markman

to say “True”, “False”, or “Can’t tell” in response to statements exemplifying such formulas as p & -p, and p v-p.* To use the tautology p v -p as an example, the children failed to recognize the truth of the statement “Either this chip is red or it is not red” (when the chip in question is entirely of one color). For the most part, the children’s poor performance on tasks such as these did not seem attributable to a misunderstanding of the logical words occurring in them (‘or’, ‘and’, ‘not’). Thus, nontautologous and noncontradictory statements such as “Either this chip is green or it is blue” and “This chip is yellow and it is not red” were readily understood. In addition to the presence of logical words, then, an important variable in these simple problems seemed to be whether the task demanded evaluating a contradiction or a tautology. As a first step towards explaining this phenomenon, notice that no empirical evidence is required to falsify or verify contradictions and tautologies. Instead, sentences corresponding to these kinds of formulas are true or false by virtue of their linguistic form, rather than deriving their truth-value from any extra-linguistic states of affairs. For the remainder of this paper, contradictory and tautological sentences will be called nonempirical, in that their truth value does not depend upon states of the world.** To illustrate the difference between empirical and nonempirical statements, compare the sentence “This chip is green and it is not green” with the sentence “This chip is green and it is not red”. Determining the truth of the latter requires examining the chip in question; determining the truth of the former requires no such examination. This distinction has been the focus of philosophical controversy (Hahn, 1959 ; Quine, 1953), but our interest in it is mainly psychological. In particular, it appeared to us that the children’s difficulty with nonempirical statements stemmed from their failure to appreciate just this distinction. The children indiscriminately sought empirical evidence for both types of sentences. Resisting the tendency to seek empirical support for nonempirical sentences requires an ability to examine the form of the statement itself. This ability in turn presupposes that the child can use language not only as a code for reality but also as an object of thought in its own right. The child must be able to reflect upon his language much as he reflects upon more tangible events; he must be able to regard language independently of the reality it refers to. This amounts to the ability to lookat language rather than through it. That children do have trouble considering language in this objective manner has been documented by Piaget (1929). Although the nature of the child’s trouble is not clear, two findings bear comment. First, it has been shown that young *Throughout this paper, we shall follow standard conventions concerning the interpretation of ‘p’, ‘SC’, ‘v’ and ‘-I. Thus, p and q stand for propositions; & stands for logical conjunction (‘and’); v stands for logical disjunction (‘or’) ; and - stands for negation (‘not’). **We have coined the term ‘nonempirical’ rather than relying on the terms ‘analytic’ or ‘logical for three reasons. First, given our theoretical position, stated below, the term ‘nonempirical’ has mnemonic value. Second, we wish to avoid questions concerning the relationship between analytic and logical sentences. Third, psychologists commonly characterize contingent statements that include logical connectives as ‘logical’.

Language and the ability to evaluate contradictions

and tautologies

21.5

children do not appreciate the arbitrariness of the relation between words and their referents (Piaget, 1929; Vygotsky, 1962). Second, it is well known that children find perceptually immediate events quite compelling (Piaget & Szeminska, 1965; Bruner et al., 1966). It is plausible that they find aspects of linguistic structure less salient. We summarize the above discussion by stating two hypotheses. Hypothesis statements.

la: Children

tend to treat contradictions

and tautologies

as empirical

Hypothesis lb: Much of the difficulty children experience evaluating nonempirical sentences is due to their nonempirical character, and not merely to the logical words occurring in them. Hypothesis

1 will be called the nonempiricality

Hypothesis 2: The ability to examine ability to correctly evaluate nonempirical

hypothesis. language objectively statements.

is necessary

for the

Hypothesis 2 will be called the objectivity hypothesis. The converse of Hypothesis 2 - that language objectivity is sufficient for evaluating nonempirical statements - is not implied in our discussion. Among other requirements, it is obvious that encoding the logical structure of the statement is also necessary for distinguishing nonempirical from empirical statements. For complex contradictions and tautologies the encoding requirement might be formidable.

Experiment

1 : The nonempiricality

hypothesis

Experiment 1 was designed to test the nonempiricality hypotheses la and lb. In order to test Hypothesis la, it is necessary to distinguish subjects who correctly evaluate nonempirical sentences on the basis of linguistic form from subjects who correctly evaluate these sentences by relying on empirical considerations. The latter subjects, right for the wrong reason, do not genuinely comprehend tautologies and contradictions. If a subject recognizes the nonempirical nature of sentences of the form p & -p and p v -p, then withholding the empirical means to judge the truth of p should not interfere with this judgment. Conversely, if we are correct in believing that children tend to treat contradictions and tautologies empirically, then the children should be reluctant to respond either ‘True’ or ‘False’ to statements for which the truth of p cannot be determined (responding ‘Can’t tell’ instead). Accordingly, items based on the forms p v -p and p & -p were constructed so that objects referred to in the proposition p were hidden. Questions in which objects referred to are hidden will be called, barbarously, nonvisible items. Hypothesis 1b states that much of the difficulty children experience evaluating simple nonempirical sentences is due to the nonempirical character of these sentences, and not simply to the logical words occurring in them. To test this hypothesis we asked children

2 16 Daniel N. Osherson and Ellen Markman

to evaluate sentences containing various combinations of the same logical words (i.e., ‘and’, ‘or’, and ‘not’) included in the nonempirical sentences. These control sentences, however, were empirical. If Hypothesis 1b is true then correct eva!uation of the control sentences should not guarantee success with the nonempirical sentences.

Method Subjects: Two classes of second graders, N = 26 (13 males, 13 females) and N = 25 (10 males, 15 females), from a middle class suburban school served as subjects. The mean age for the classes was 7 years, 10 months, and 7 years, 8 months, respectively. No subjects were dropped. Materials: The materials were small plastic poker chips in assorted colors (each chip exemplified only one color). Procedure: Each subject was seen individually for one twenty minute session. The chips were placed between the experimenter and the subject on a table. It was explained that the experimenter would say some things about the chips, and that in each case the subject should indicate whether the experimenter’s statement was true or false. The subject was also given the explicit option of responding ‘Can’t tell’. Every item consisted of a statement about a chip held in the experimenter’s hand. In each case the question pertained to the chip’s color, and was preceded by the interrogative “True, false, or can’t you tell?” For the nonvisible items, the subject was asked to close his eyes while the experimenter selected a chip. This chip was then concealed in the experimenter’s closed hand until after the subject responded. For the remaining items, the chip was continuously in view. In order to minimize the time required for each subject, two distinct but overlapping sets of items were administered, one to each class. Order of presentation of the items was randomized for each subject. Questions were repeated if the subject hesitated or appeared puzzled. Items: The left hand columns of Table 1 provide a description of the items. The description includes (a) the question posed by the experimenter, (b) the color of the chip, if visible, and (c) the correct answer. Items 1 - 4 are nonempirical; all other items are empirical. These remaining items were designed to assess subjects’ ability to evaluate questions sharing properties of the nonempirical items. These properties include negation, conjunction, disjunction, nonvisibility (items 7, 10, 14, 26) and repetition of a constituent proposition (items 21, 22,33,34). Columns 5 and 6 of Table 1 show which of the two groups of subjects (or both) received a given item.

Language and the ability to evaluate contradictions and tautologies

217

Results Hypothesis

la: The righthand columns of Table 1 give the percentage of subjects in each group passing the items. Few subjects answered the nonempirical items (1 - 4) correctly. More importantly, 73% of the mistaken answers to these questions were responses of ‘Can’t tell’. This supports Hypothesis la that children tend to treat contradictions and tautologies as empirical statements; without empirical evidence, the children refused to evaluate them.* The percentage of ‘Can’t tell’ responses to the nonempirical items may even be spuriously low. Items 7 and 10, the simplest, nonvisible, empirical questions, show that the children are in general reluctant to respond ‘Can’t tell’. Only 63% of the subjects did so, even though the ‘Can’t tell’ response seems plainly appropriate in these cases. Moreover, the percent of correct responses (i.e., ‘Can’t tell’) for the more complex nonvisible, empirical items 14 and 26 was slightly lower. ‘Can’t tell’ was almost never given to the visible empirical items.

Hypothesis

Zb: In general, the nonempirical items were substantially more difficult than their empirical counterparts. This supports Hypothesis 1b: The children had more difficulty with the nonempirical questions than with empirical questions of comparable logical structure. Items 28 and 30, involving disjunction, are exceptions to this pattern, being at least as difficult as the tautologous items 3 and 4. However, these items have a property not found in the nonempirical items. In each case, a response of ‘True’ was required despite the visibility of a chip whose color matched neither of those mentioned in the question. In marked contrast, items of identical logical structure, but not possessing this feature (items 26, 27, 29, 31, and 32) posed little difficulty. Therefore, we do not consider these exceptional items to infirm Hypothesis lb. Consistent with this interpretation is the fact that there were no such exceptions with respect to contradictions. That is, all of the empirical items involving conjunction were substantially easier than the contradictions.**

Experiment

2: The objectivity

hypothesis***

Experiment 1 provided evidence supporting the nonempiricality hypothesis. Experiment 2 was designed to test Hypothesis 2 that the ability to examine language objectively is necessary for the ability to correctly evaluate nonempirical statements. *On the other hand, in other experiments we included nonempirical items in which the chip was clearly visible, and thus provided such ‘evidence’. In this situation, children are still unable to appreciate the nonempirical character of these items, and revert to a simple ‘matching’ strategy. Both kinds of nonempirical statements (p v -p and p & -p) were judged as true only when p was true, and judged as false when p was false. **We have conducted three additional studies to test Hypothesis 1. The subjects in these additional experiments ranged in age from 6 years to adulthood. The results are fully comparable to those reported for Experiment 1. **+This experiment is part of a larger study in which the same subjects participated. The data from the larger study are fully consistent with the results reported below.

2 18

Daniel N. Osherson and Ellen Markman

Table 1.

Item Number

Items for Experiment I

Question

Status of chip

Correct answer

Percent correct, Group 1 (N = 26)

_

Percent correct, Group 2 (N = 25) ___.

Nonempirical. 1.

The chip in my hand is white and it is not white.

hidden

2.

The chip in my hand is not blue and it is blue.

hidden

false

3.

Either the chip in my hand is yellow or it is not yellow.

hidden

true

4.

Either the chip in my hand is not red or it is red.

hidden

true

_

3 1% _

24% 56%

23%

28% 24%

Simplr Assertion: 5.

The chip in my hand is blue.

blue

true

100%

6.

The chip in my hand is green.

blue

false

96%

100%

7.

The chip in my hand is yellow.

hidden

can’t tell

73%

56%

Simple Negation. 8.

The chip in my hand is not white.

red

true

8 1 ‘5,

9.

The chip in my hand is not green.

green

false

96%’

68% _

The chip in my hand is not red.

hidden

can’t tell

65%

56%

10.

Conjurlction: 11.

The chip in my hand is red and it is yellow.

red

false

-

80%

12.

The chip in my hand is blue and it is green.

green

false

-

16%

The chip in my hand is white and it is red.

green

false

_

100%

The chip in my hand is yellow and it is not blue.

hidden

can’t tell

58%

__

15.

The chip in my hand is white and it is not red.

white

true

88%

100%

16.

The chip in my hand is not blue and it is yellow.

yellow

true

17.

The chip in my hand is red and it is not white.

white

false

92%

96%

18.

The chip in my hand is blue and it is not yellow.

red

false

73%

96%

19.

The chip in my hand is not red and it is white.

red

false

20.

The chip in my hand is not blue and it is green.

red

false

13.

Conjunction with Negation. 14.

_

100%

96% _

100%

Language and the ability to evaluate contradictions and tautologies

2 19

Table 1. (continued)

Item Number

Question

21.

Conjunction with Repeated Propositions. The chip in my hand is blue and

22.

it is blue. The chip in my hand is green and it is green.

Percent correct, Group 2 (IV= 25)

Status of chip

Correct answer

Percent correct, Group 1 (A’= 26)

blue

true

100%

-

white

false

96%

-

red

true

-

16%

white

true

-

92%

red

false

-

100%

hidden

can’t tell

13%

white

true

81%

96%

green

true

15%

16%

white

true

-

96%

yellow

true

-

0%

green

false

blue

false

white

true

96%

_

false

81%

-

Disjunction:

23. 24. 25.

Either the chip in my hand is red or it is blue. Either the chip in my hand is green or it is white. Either the chip in my hand is yellow or it is blue. Disjunction with Negation:

26.

Either the chip in my hand is red or it is not green.

27.

Either the or it is not Either the or it is not Either the green or it

28. 29. 30. 31. 32.

chip in my hand is white blue. chip in my hand is red yellow. chip in my hand is not is white.

Either the chip in my hand is not red or it is blue. Either the chip in my hand is yellow or it is not green. Either the chip in my hand is not blue or it is yellow.

Si%

_

100% 88%

-

Disjunction with Repeated Propositions:

33. 34.

Either the chip in my hand is white or it is white. Either the chip in my hand is yellow or it is yellow.

Method Subjects: Eight children from each of grades 1,2, and 3 (mean ages: 6 years, 11 months; 7 years, 10 months; and 8 years, 9 months, respectively) were haphazardly selected to serve as subjects from a middle class suburban public school, while eight sixth graders

220

Daniel N. Osherson and Ellen Markman

(mean age, 11 years, 7 months) were selected from a nearby private school. In addition, seven tenth-graders and eleven adults volunteered as subjects. In all, there were 50 subjects, 22 males and 28 females. No subject from Experiment 1 participated in Experiment 2. No subjects were dropped. Materials: The materials were the poker chips described in Experiment

made of blue construction a dog.

paper, a color photograph

1, 15 cm. figurines of a cat, and a color photograph of

Items Nonempirical

items

There were four nonvisible nonempirical items, two based on each of the formulas p & --p and p v -p. These items are described in Table 1, numbers 1 and 3. Language objectivity items

The objectivity items were designed to investigate various components of an objective attitude towards language. One component investigated was the awareness that words bear an arbitrary relation to their referents, for example, the awareness that interchanging the names of common objects does not affect their other properties (c.f., Piaget, 1929; Vygotsky, 1962). This first item is called Arbitrariness of Language. Allied to the awareness assessed by the first item is the realization that words do not share the physical properties of the things they signify. Thus, the word ‘needle’ is not itself sharp. This item is called Nonphysical Nature of Words. In a similar vein, other items assessed the subject’s belief in the stability of the meaning of words in the face of destruction of the word’s empirical referent (consider the word ‘dinosaur’). This is the Meaning and Reference item. In more detail, the objectivity items were administered as follows: 1. Meaning and Reference. The experimenter began by saying: “Do you know what the word glump means? The word glump means a toy creature made out of blue paper that smiles and is round. What does the word glump mean? [The subject responds with the experimenter coaching if necessary.] There are only two glumps in the whole world, and here they are. [The experimenter produces two glumps and expands instructions as necessary.] Now what would you tell a kindergartener who asked you ‘What does the word glump mean?’ ” [After the subject responds, the experimenter proceeds to completely destroy both glumps in front of the subject, and continues questioning as follows] : “Now, there are noglumps left in the world. What would you say now to a kindergartener who asked ‘What does the word glump mean?’ [The subject responds.] Suppose someone said ‘glumps are blue’. Would he be right? [The subject responds.] Has the meaning of the word glump changed? [The subject responds.] Does the word glump now have any meaning?”

Language and the ability to evaluate contradictions and tautologies

221

A subject was scored as correct only if he (1) gave essentially the same definition for the word glump as before, or gave it with an added proviso as to the glumps’ nonexistence; (2) affirmed that glumps are still blue; (3) claimed that the meaning of the wordglump had not changed (or had only changed through the addition of the extinction proviso); and (4) claimed that the word still had meaning after the destruction of the two glumps. After the questions concerning glumps, essentially the same questions were posed concerning the word giraffe. This time, however, the subject was asked to imagine the annihilation of the word’s referent; that is, he was asked to imagine that every giraffe in the world had suddenly disappeared. Scoring was the same as for the glump-subitem. To be counted as passing the Meaning and Reference item, a subject had to pass both the glump and giraffe subitems. 2. Arbitrariness of Language. The subject was instructed as follows: “Suppose everyone in the world got together and decided that from now on we will call the sun ‘the moon’ and we will call the moon ‘the sun’. All we are going to do is change the names. Could we do that if we wanted to?” All subjects acquiesced wiih minimal coaching. The subject was then asked “Now when you go to bed at night, what will you call the thing in the sky?” Virtually all subjects responded correctly, and otherwise they were coached. The subject was then asked to describe what the sky would look like. The correct answer was to describe a night sky. After the sun-moon question, the subject was similarly asked to imagine that the names for cats and dogs were interchanged. He was then shown a picture of a cat and asked to give its new name. All subjects responded correctly (i.e., “dog”). The experimenter then asked: “What sound will this animal make?” The correct answer is “meow”. A picture of a dog was produced and the same questions asked, the correct answers being “cat” and “ruff’, respectively. To be counted as passing the Arbitrariness of Language item, a subject had to pass both the sun-moon and cat-dog subitems. 3. Nonphysical Nature of Words. Subjects were asked the following questions, answer being correct for all of them.

a negative

1. Is the word ‘nickel’ worth five pennies? 2. Is the word ‘book’ made of paper? 3. Does the word ‘bird’ have feathers? 4. Can you buy bubble gum with the word ‘penny’?* *Additional questions were originally included (e.g., “Is the word fire hot?” “Is the word ice a cold word?“), but they were not scored since adult subjects reported confusion as to whether the questions should be interpreted literally or metaphorically. No adults reported confusion for the items scored, above.

222

Daniel N. Osherson and Ellen Markman

All four subitems had to be answered correctly for a subject to be counted as passing the Nonphysical Nature of Words item. Procedure All subjects were seen individually for one 30-minute session. The order of administration was objectivity items 1 - 3 followed by the four nonempirical items, or the reverse. The two orders were counterbalanced within age groups. All subjects received every task. Results The number of subjects passing each task is arranged by grade in Table 2. Since there were no striking effects of task order, the table makes no reference to this variable. Modal errors As in Experiment 1, the modal error for the nonempirical items was a response of “Can’t tell”. In the first meaning and reference problem, after the destruetion of the glumps, the majority of the erroneous responses were that the meaning of the word ‘glump’ had changed (beyond a proviso of nonexistence) and/or that the word no longer had any meaning. Nonetheless, virtually all subjects still agreed that “glumps are blue”. The results of the meaning and reference item concerning giraffes were analogous to the problem concerning glumps. In the first problem concerning the arbitrary connection between a word and what it signifies, after agreeing that the moon is now called ‘the sun’, the universal erroneous response was to describe a day sky in place of a night sky (e.g., to describe the sky as bright blue). Similarly, in the problem wherein the names of cats and dogs were interchanged, the universal erroneous response, with the relevant picture in plain view, was to interchange the sounds that cats and dogs produce (i.e., the cat, now .correctly called ‘a dog’, was incorrectly said to bark; similarly, the dog, now correctly called ‘a cat’, was incorrectly said to meow). In the item concerning the nonphysical nature of words, the nearly universal erroneous response was to reply affirmatively to such questions as “does the word bird have feathers?” The errors to these three language items reflect, we believe, an inability to regard language objectively, to disentangle it from the reality it encodes. Of the total number of responses given to the objectivity items 1 - 3, 56%, 50% and 44%, respectively, were correct. This finding suggests that common psychological processes underlie the solution of all the language items. To provide a further test of this possibility, we calculated phi-correlations between each of the three language items. The three resulting correlations (0.56,0.62, and 0.40) are all significant @ < 0.05). These moderate correlations support the idea that common psychological processes mediate the diverse language items. The correlations thus render less plausible the idea that the difficulty of the objectivity tasks is due to idiosyncrasies of the questions, with resulting misinterpretations by the subjects.

Language and the ability to evaluate contradictions and tautologies

Number of subjects passing each item

Table 2.

ofExperiment 2 by grade

Objectivity items Reference and meaning Grade 1 2 3

N

8

10 adult

I 11

Total

50

6

Table 3.

8 8 8

Nonempirical items

Arbitrariness of language

1 2 3

223

Nonphysical nature of words

4

1 2 2

6

0 2 0

5 10

6 11

25

28

P & -P 1

P ” -P 1

2

2

6 8

0 0 1 1 4 8

0 0 1 2 3 8

1 0 1 2 3 10

0 0 1 4 3 9

22

14

14

17

17

6

Conditional Probabilities of a subject passing each of the three language objectivity items given he has passed every item from the two kinds of nonempirical items (Test of the Objectivity Hypothesis)

Nonempirical items

Language

objectivity

items

Nonphysical nature words

Arbitrariness of language

Meaning and reference

Total number of subjects passing nonempirical category

10/12

12112

11/12

12

lO/lO

9/10

10

Nonvisible P ” -P Nonvisible P&-P Total number of subjects passing language objectivity item:

8110

22

28

2s

The objectivity hypothesis In the introductory section we argued that being able to distance oneself from language, to consider it as independent of the events it refers to, is a necessary condition for comprehending nonempirical statements. If this hypothesis is correct, then those subjects passing the nonempirical items should have also passed the language tasks: In turn, this implies that the conditional probability of a given subject passing language items given that he passed a nonempirical item ideally equals one. Table 3 provides such conditional probabilities in matrix form, the columns representing the three language objectivity items and the rows representing the two kinds of nonempirical items. To be counted as

224

Daniel N. Osherson and Ellen Markman

passing a given kind of nonempirical item, a subject had to pass both items of the given kind. Each cell of the table gives the conditional probability of a subject passing each language objectivity item given that he is successful with the designated nonempirical items.* These conditional probabilities range from S/IO to 12/12, with a mean of 0.91. Compare these results with the results of an analysis designed to test the hypothesis that language objectivity is a sufficient condition for the ability to correctly evaluate nonempirical statements. We suggest that this is not the case, that processes other than language objectivity are likely to be required as well (e.g., logical computation). The hypothesis that language objectivity is sufficient implies that the conditional probability of a given subject passing nonempirical items, given that he passed objectivity tasks, ideally equals one. Using categories identical to those in Table 3, we computed the conditional probabilities associated with this hypothesis. These range from 9125 to 10/22, with a mean of 0.40. Additional analyses were performed in which the criteria for language objectivity and comprehension of nonempirical statements were relaxed. For purposes of this analysis, to be considered as passing language objectivity, a subject need only pass one of the three objectivity items. For the two nonempirical categories in this analysis, a subject was considered to have passed that category so long as he passed at leasLone of the two items in it (rather than both items). The obtained conditional probabilities for the objectivity hypothesis are comparable to those in the previous analysis, with a mean of 0.88. The conditional probabilities for the converse hypothesis, that language objectivity is sufficient for nonempiricality, are again low, with a mean of 0.52.

Discussion Hypothesis la was supported by the large number of “Can’t tell” responses to the nonvisible, nonempirical items in both experiments. A subject recognizing the nonempirical nature of simple contradictions and tautologies should be willing to evaluate these sentences in the absence of empirical evidence. Hypothesis lb is supported by the subjects’ generally good performance on the empirical items in Experiment 1. These control items differed from the contradictions and tautologies mainly in being empirical. postulates” We are aware that the nonempirical items violate certain “conversational described by H. P. Grice, and this may contribute to their difficulty. However, this is not sufficient to account for the relative difficulty of the nonempirical items. For one thing, the empirical items involving disjunction (items 23 - 34 of Table 1) similarly violate

*This conditional probability is represented by the fraction as follows: The numerator in each cell cc gives the number of subjects passing every task indicated by the ith row andjth column, while the denominator gives the number of subjects passing every task in the ith row.

Language and the ability to evaluate contradictions

and tautologies

225

conversational postulates, but in general are not nearly so difficult. For another, one would expect older subjects to be at least as sensitive as younger subjects to conversational postulates. Yet, as seen in Experiment 2, older subjects have more success than younger subjects with the nonempirical items. The results of Experiment 2 support Hypothesis 2. Comprehension of nonempirical sentences predicts well an objective attitude toward language. In contrast, passing the language objectivity items is not a good predictor of success with the nonempiricality items. Comparable results were obtained using weak and strong scoring criteria. The experiments help to clarify an issue raised by Piaget, concerning the relation between advanced logical thinking and language. According to Genevans, advanced logical thinking, including facility with propositional logic, is attained in the stage of “formal operations”, an achievement of adolescence (Inhelder & Piaget, 1958). Piaget is explicit that the acquisition of a natural language is not sufficient for attaining formal operations (Piaget, 1967, pp. 94 - 98; Sinclair-de-Zwart, 1969, p. 320). To the extent that our nonempirical items partake of formal operations, our results are consistent with this claim. Piaget also states that language is at least a trivial prerequisite for formal operations if only because “these operations no longer bear upon the objects themselves .. . but on propositions, on verbally announced hypotheses, etc.“. To what extent and in what way linguistic competence plays a more significant role in the acquisition of advanced logical thinking is left open by Piaget (Piaget, 1969, p. 127; Furth, 1969).* Our study emphasizes the role of “meta-linguistic” skills rather than linguistic competence per se in evaluating contradictions and tautologies. The present study is exploratory, and both empirical and conceptual elaboration is needed. Obviously, our conditional probability analyses are only a weak test of Hypothesis 2. Training studies would be useful to support our conclusions. Also, the source of children’s difficulty with the language objectivity items remains to be determined. Aside from additional experiments, however, the concepts of language objectivity and nonempiricality require clarification. Our treatment of language objectivity is intuitive, rather than based on explicit criteria .** As for nonempiricality, its clarification amounts to a deeper understanding of logical truth.

*On the other hand Piaget is quite explicit that the acquisition of a natural language is neither necessary nor sufficient for attaining the stage of “concrete operations” (Piaget, 1967; Furth, 1969; Sinclair-de-Zwart, 1969). **See Cazden (1972) for a review of research using tasks akin to our language objectivity items.

226

Daniel A? Osherson and Ellen Markman

REFERENCES Bruner, J., Olver, R., Greenfield, P., et al. (1966) Studies in cognitive growth. New York, Wiley. Cazden, C. (1972) Child language and education. New York, Holt, Rhinehart & Winston. Furth, H. G. (1969) Piaget and knowledge. Englewood Cliffs, N.J., Prentice-Hall, Chapter III. Hahn, H. (1959) Logic, mathematics, and knowledge of nature. In A. J. Ayer (Ed.) Logical positivism. New York, The Free Press, pp. 147-161. Inhelder, B. & Piaget, J. (1958) The growth of logical thinking from childhood to adolescence. New York, Basic Books. Piaget, J. (1929) The child’s conception of the world. London, Routledge & Kegan Paul. Piaget, J. (1967) Language and thought from the genetic point of view. In J. Piaget, Six psychological studies. New York: Random House, pp. 88-99. Piaget, J. (1969) Language and intellectual operations. In H. G. Furth, Piaget and knowledge. Englewood Cliffs, N.J., Prentice-Hall, pp. 121-130. Piaget, J. & Szeminska, A. (1965) The child’s conception of number.. New York, W. W. Norton. Quine, W. V. 0. (1953) Two,dogmas of empiricism. In W. V. 0. Quine, From a logical point of view. Cambridge, Mass.: Harvard University Press. Sinclair-de-Zwart, H. (1969) Developmental psycholinguistics. In D. Elkind & J. Flavell (Eds.), Studies in cognitive development. New Yqrk, Oxford University Press. Vygotsky, L. S. (1962) Tholfght and Language. Cambridge, Mass., MIT Press.

On a remarqub que les enfants 6prouvent des difficult& i bvaluer des contradictions de la forme p & -p et des tautologies de la forme p v -p. On a fait les hypothsses suivantes: (a) Les difficult& de ces &non&s ne sont pas uniquement dues aux mots logiques qui les composent. (b) Une partie des difficult& provient du fait que leur valeur de v&it6 repose davantage sur leur forme linguistique que sur des consid&ations empiriques. (c) La capacite i examiner le langage de man&e objective, en dehors des &dnements et des objets auxquels il se r&f&e est une condition n&essaire mais non suffisante pour une &valuation correcte des contradictions et des tautologies. Les rdsultats des deux expkriences confirment ces hypothhes.

3

Factors influencing

assignment

of pronoun antecedents*

CATHERINE ALFONSO

GARVEY CARAMAZZA

JACK YATES

The Johns Hopkins University Abstract In two experiments a group of adult subjects and a group of 10 - 13 year old subjects completed sentences of the format NPl V NP2 because Pro... The fragment, John telephoned Bill because he..., elicited completions such as had good news, which locate the impetus for telephoning in NPl (John). The fragment, John admired Bill because he..., elicited comp~eh’ons such as was a fine athlete, Which locate the impetus for admiring in NP2 (Bill). The assignment of the subject pronoun of the second clause as coreferential with NPI or NP2 shows that this choice depends on a factor, direction of causality, which may be assigned to vet-4roots. This factor was isolated from interacting semantic and syntactic features of sentences in which the verbs occurred. Manipulation of these features resulted in greater or lesser salience of the direction of causality factor. The potential of this technique for investigation of verb semantics and for interaction of semantic, pragmatic and syntactic features in the psychological study of comprehension is discussed. Anaphoric prowords are generally thought to be assigned appropriate referents by some process that matches a symbol of the deep structure to an available antecedent. The matching is in many cases facilitated by explicit cues to the features shared by proword and antecedent. For example, in 1) below: (1) Henry trusted the woman because she was efficient. the pronoun she and the antecedent woman share the critical feature, [Female], as well as other features not critical in this context, e.g., [+Human], [-t-Singular]. In sentence 2) (2) Henry trusted the secretary because she was efficient. we must assume that secretary also bears the feature [Female], but the basis for this assumption is somewhat different. It must be either that we know that secretaries are *This research was supported by the National Science Foundation under Grant #X-31636. Authors’ address: The Department of Psychology, The Johns Hopkins University, Baltimore, Maryland 21218.

CognitionS(3), pp. 22 7 - 243

22% Catherine Garvey, Alfonso Caramazza, and Jack Yates

marked [Female] for linguistic purposes or that we know that she must select a seman_ ticallY feminine antecedent, and secretary is the only available candidate, If we exam_ ine 3), however, (3) Rosemary trusted the secretary because she was efficient. it again appears that the basis for selecting the antecedent

is different. No obvious seman-

tic feature of the antecedents can provide a unique basis for assignment, since both antecedents share all obvious critical features. Here it would appear that the listener may rely on some expected compatibility in the meanings of secretary and efficient to accomplish the matching. A sentence such as 4) reveals that listeners may rely on yet other kinds of information to make an assignment: (4) Rosemary trusted the secretary because she was a good administrator. 4) may be read as meaning that Rosemary, being a good administrator, habitually supported her staff by entrusting them with her confidence. Or, it may be read as meaning that Rosemary, noting the administrative capabilities of the secretary, believed that these capabilities indicated a trustworthy character. The problem is how to account for the way in which a listener understands the pronoun she as coreferential with the first noun phrase (NP 1 = Rosemary) or with the second noun phrase (NP, = the secretary).* Finally, a sentence such as 5) can be produced and comprehended, i.e., assigned a reading. (5) Rosemary trusted the secretary because she was gullible. This example brings us directly to the question of the role of semantic factors in the assignment of pronoun antecedents. Although sentences 5) and 3) are both ambiguous, they appear to differ in the availability or salience of alternative assignments. If this is the case then there are some as yet poorly understood semantic features associated with the matching of noun phrases. Sentences with contrasting assignments, such as 6) and 7) indicate that some property of verb roots may be implicated in the process of pronoun assignment. (6) George telephoned

Walter because he wanted sympathy.

(7) George criticized Walter because he wanted sympathy. Such a property (or properties) would be normally obscured in the presence of disambiguating cues, such as are present in 1) and 2), and, perhaps, in 3). But sentences such as 6) and 7) raise the question of the role of verb semantics in pronoun assignment as well as the question of possible interaction of semantic properties with syntactic or pragmatic *The listener may entertain both readings, simultaneously or sequentially, and subsequently choose one of them, or he may perceive only one reading at the time of responding. In the absence of strong evidence for either alternative process, we will assume that at the time of response only one interpretation is salient, i.e., consciously available to the listener.

Factors influencing assignment of pronoun antecedents

229

features of sentence structure. Specifically, the question being raised concerns the kinds of knowledge a listener might use in selecting the antecedent of a potentially ambiguous pronoun. If the listener employs some sort of pragmatic strategy, it would still be subject to qualifications. For example, on hearing 4) the hearer might predict: “The topic of the first clause is likely to be the topic of the second clause as well. Lacking any counterindications, 1 will assume that the antecedent of the pronoun topic she is the topic Rosemary”. Some such strategy would provide one acceptable reading for 4) and would work smoothly for 5) and 6). But it would produce an unexpected reading for 7).

Study I Before examining the question of listener strategies, however, it is necessary to confirm the informal observation that speakers do indeed produce sentences with supposedly “ambiguous” pronouns and to obtain some information on interspeaker agreement on selection of antecedents. A pilot study was undertaken to provide this basic information. In addition, this first study was designed 1) to identify a number of verbs which appear to influence pronoun assignment and 2) to explore certain pragmatic and syntactic factors which might influence the phenomenon.

Method Subjects:

participated

Twenty-eight volunteer in the experiment.

subjects

from an introductory

psychology

class

Procedures: Subjects were tested in a group in the classroom. They listened to recorded instructions which stated that the objective of the experiment was to find out how people ascribed reasons or motives for actions. Subjects were told that they would hear 24 incomplete sentences. Their task was to complete each sentence, writing a response which supplied a credible motive for the action or attitude presented in the first part of the sentence. A separate set of instructions introduced the second part of the task in which subjects were asked to write answers to 11 recorded questions concerning reasons or motives. In this part they were asked to begin their response with the word because. They were informed that the recorded items allowed only 10 seconds for completion and that a signal tone would be heard before the beginning of each item. Materials. Part 1 consisted of 16 test items and eight distractor items. All were fragments of the format: NPI Past-V NP2 because Pro... All subject and at least one of the non-subject nouns referred to human beings. In the test items no grammatical cues were available to assist pronoun assignment to either NPI or NP2, e.g., The father scolded his

230

Catherine Garvey, Alfonso Caramazza, and Jack Yates

son because he.,. The distractor

items, which were interspersed among the test items, did provide grammatical cues, e.g., Mary returned the records to Sum because she... Items were recorded with approximately equal stress on NPI, NP2 and on the pronoun. Of the 16 test items, eight were expected on the basis of preliminary work to elicit NPl assign_ ment, and eight to elicit NP2 assignment of the potentially ambiguous pronoun. Stan_ tences were h~~~W.W~~~~ as to grammatical form.* The different structures were chosen to allow comparison of positions of assertional focus, e.g., Bill sold the exam questions to John because he... versus Henry sold Bill the bike because he... and to reflect some diver-

sity in the status relations between the two candidate antecedents, e.g., prisonet-guard, mother-daughter. Status was conceived as a semantic dimension reflecting relative ranking of two [+Human] nouns within a single field or system of reference, e.g., within the family, father-son; within job or institutional system, director-actor. Part 2 consisted of eight test items and three distractor items, (11 questions) which requested a ‘because’ answer, e.g., Why did the mother punish her daughter? These items were designed to test 1) the effect of interrogative structure on the selection of antecedent and 2) the effect of implicit negativeness in the verbs, e.g., director praised versus criticized the actor. Response sheets were scored independently by two judges with 99% interjudge agreement. Scoring consisted of judging the completed sentence as indicating either NPl or NP2 assignment of the pronoun of the second clause. Judges also scored a response as ambiguous (A) if it was not clearly interpretable as NPl or NP2; as unintelligible (U) if illegible or if it indicated lack of understanding the test item; or as no response (NR). The last three scoring categories did not appear to be systematically distributed in relation to test items. Total number of possible responses was 652; 1.5% of these were comprised of A+U+NR responses. The results report the percentages of the dominant response obtained (either NPl or NP2) for each item after A+U+NR responses were subtracted from the possible total for each item.

Results and Discussion Subjects were indeed able to complete the potentially ambiguous sentence fragments. Informal questioning after the experiment revealed that the ostensible objective of the experiment had been accepted. None reported noticing that the second clause pronoun was potentially ambiguous. Thus, Study I suggests that the processing of such sentences lies within the competence of speakers and that the technique may be useful for investigating the processes involved in assignment of pronouns to antecedents.

*A list of the test from the authors.

items as well as a list of the items used in Study II may be obtained,

on request,

Factors influencing assignment of pronoun antecedents

23 1

Total number of Nl+N2 responses; number of Nl and N2 responses; percentage of dominant response/subordinate response; and level of probability.

Table 1.

1. Sentence Completion Sentence Number

11 4 18 13 19 10 6 16 24 3 22 1 14 9 20 7

(verb) tie up help distrust confide in pick up rush to criticize selI scold fear confess to call join blame sell kill

Totala

Nl

N2

%

p (2-tailed)

18 27 21 :z 26 28 23 26 24 25 13 23 23 24 27

10 18 6 14 6 6 4 20 3 2 23 13 23 0 24 0

8 9 15 5 20 20 24 3 23 22 2 0 0 23 0 27

55 66 71 74 77 77 85 87 88 92 92 100 100 100 100 100

0.814 0.230 0.078 0.064 0.014 0.014 0.004 0.002 0.001 0.001 0.001 _ _ _ -

Totala

Nl

N2

%

p (2-tailed)

II. Questions Sentence Number

7 2 8 2 4 11 1

(verb) help kick fear

23 21 21

9 7 7

14 14 14

61 67 67

0.404 0.190 0.190

give confess to praise punish selI

25 27 26 27 21

251 1 1 21

242 25 26 0

96 93 96 96 100

0.001 0.001 0.001 -

a N = 28.

Table 1 presents the total responses obtained (maximum number of responses for an item = 28), the number of NPl and NP2 responses obtained for each item, the percentage of the dominant response, and the probability level for the dominant response. Part I presents the data for the sentence completion task and Part II for the answer-to-question task. The results for Part I indicate that most of the test items elicited a biased response. One third of the items received only NPI completions (three items) or only NP2 completions (two items). The majority of items showed an intermediate response bias to NPl or NP2 completions, and only a few revealed no clear bias. NPl items ranged from 55% to

232

Catherine Garvey, Alfonso Caramazza, and Jack Yates

100% and NP2 items from 71% to 100%. It would appear that any psychological rule designed to account for antecedent assignment in this type of sentence must reflect the continuous (rather than simply dichotomous) nature of this distribution. The answer-to-question task revealed a similar patterning of results. As shown in Table 1, Part II, two questions received a majority of NPl responses (93% and lOO%), three were moderately biased toward NP2 (61%-67%), and three were strongly biased toward NP2 (96%). The existence of moderate as well as extreme values in the dominant responses confirms the continuous nature of the bases for antecedent assignment under conditions of potential ambiguity and extends the phenomenon to the interrogative structure. Part II results suggest that the rule or rules which determine antecedent assignment apply to intersentential discourse as well as to the sentence types of Part I. Results obtained for specific items generally supported the notion that verbs play a role in pronoun assignment and do so in some consistent way. Verbs from test items are indicated in Table 1. The variations in syntactic structure and lexical composition of the test items prevent us from drawing any conclusions at this point about the influence of the particular verb on the responses. However, comparison of Part I items 3 Cfear), 16 and 20 (seU), and 22 (confess to) with Part II items 8 (jear), 1 (se//), and 5 (confess to), does indicate that a dominant response is not unique to a particular sentence, but is associated with the verb in some way. Only one verb elicited different responses in the completion task as compared with the answer task. Help elicited NPl dominant responses in Part I (item 4) and NP2 dominant responses in Part II (item 8). In both tasks the dominant response was weak and was not statistically significant. Several other tentative observations emerged from the item analysis. Briefly, 1) implicit negative or positive value of the verb was not an important factor in determining the response bias, e.g., items with criticize, praise, and blame all elicited strong NP2 responses; 2) the position of assertional focus may have some influence on strength of response, but it did not appear to influence the type of response. Subjects responded similarly to Bill sold the exam questions to John and Henry sold Bill the bike. 3) The status relationships of the antecedent noun phrases, though not systematically compared in Study I, appeared to be a potential factor in the strength of the response.

Study II A second study was designed to replicate the results obtained for specific verbs (see 1 below) and to pursue several factors (see (2)-(.5) below) which Study I and a previous linguistic analysis (Garvey & Caramazza, 1974) suggested might interact with the influence of the verb in the process of pronoun assignment. The notions to be tested were the following: (1) Certain verbs (i.e., those identified in Study I and perhaps others) bear a semantic feature which imputes the cause of an event or situation to either the subject or non-

Factors influencing assignment of pronoun antecedents

233

subject of the clause in which it serves as a main verb. The feature, however, is best described as a bias or valence rather than as a simple binary feature.* (2) A linguistic analysis (Garvey & Caramazza, 1974) had suggested that sentences completed wi?h NP2 assignments were in general amenable to passive transformation (P-T) whereas those with NPl assignments were not compatible with P-T. Since this observation relates the semantic phenomenon to an important syntactic property of sentences (i.e., applicability of P-T), comparisons between active (A) and passive (P) sentences were included. (3) Although the implicit positive or negative valences of verbs such as praise or blame did not strongly affect the pronoun assignment responses in Study I, explicit negation is an important factor in semantic judgments. Thus, a comparison of explicit negative (-) versus positive (+) verb phrases was included in Study II as a factor likely to influence the bias of implicit causality for a given verb. (4) The postulated causality feature operated in the selection of antecedents of anaphoric pronouns in both intra- and intersentential sequences. In the latter, the feature was unaffected by the presence of [+ interrogative] in the first sentence. To further substantiate this result comparisons between interrogative (?) and declarative (!) sentences were included. (5) Finally, the relative status of the candidate noun phrases of the first clause might have been a factor in the strength of the responses obtained in Study I. Certain actions or attitudes referenced by the verb may be viewed as congruent with those relations, e.g., father praised son, or noncongruent with those relations, e.g., son praised father. Contradiction of such congruence would be expected to attenuate or modify the attributed direction of causality to some extent. Therefore, comparisons of congruent and noncongruent collocations of NPl-V-NP2 were included. In summary, then, the effects of the following variables on the pronoun assignment response were tested: Variable I - Active-Passive, e.g., (A) John recognized Michael because he... (p) Michael was recognized by John because he... Variable II - Positive-Negative, e.g., (+) The soldiers feared the natives because they... (-) The soldiers did not fear the natives because they... Variable III - Interrogative-Declarative, e.g., (?) Why did the mother punish her daughter? (!) The mother punished her daughter because she... Variable IV - Congruent and Noncongruent, e.g., (Cong) The prisoner confessed to the guard because he... (NonC) The guard confessed to the prisoner because he...

*In using the term ‘semantic feature’ we do not espouse any particular theoretical position as to the dividing line between semantic knowledge and ‘knowledge of the world’. We leave open the question of whether the phenomenon investigated in these studies should be classed as a grammatical or extragrammatical one.

234

Catherine Garvey, Alfonso Caramazza, and Jack Yates

Method Subjects: 110 junior high school students, ages 10-13, who participated ment as part of a battery of tests administered in an independent study.

in the experi-

Materials: Twenty-six verbs were selected so as to compose a heterogeneous set. No a priori classification was possible, since pilot work failed to reveal categories which consistently partitioned verbs into those that exhibit the proposed causality feature and those that do not. Nor could any systematic categorization of NPl and NP2 verbs be made. Four sets of five ‘kernel’ sentences were each transformed to provide contrasting sentences for each sentence in each variable set. The resulting 40 sentences were divided into two sets of response sheets, each containing 20 sentences. Form A, for example, contained five A-P items, three passive and two active sentences; and Form B contained the converse of this set. Order of sentences within and between variable sets was randomized as was the order of verbs. For example, the order of sentences in Form A proceeded as follows: 1. chase (P); 2. sail (+); 3. sell (?); 4. confess to (NonC); 5. hit (A); whereas Form B proceeded as follows: 1. recognize (A); 2. kill (+); 3. give (?); 4. sail (+); 5. sell (!). Of the 20 verbs used in the 20 ‘kernel’ sentences, 11 had been employed in Study I. A second set of items was composed of 10 sentence frames of the format: “ verb because....” In this set one verb (blame) was taken from Study I, one verb (telephone) was a synonym of a verb in Study I (call), and the remainder were included for the first time. These items were presented in the same order to all subjects. Materials were presented in two parts. Part I included the 20 sentences described above. Part II included the 10 frames of the format: “ verb because...” in which subjects were required to supply NPl, NP2 of the first clause as well as the entire ‘reason clause’. All items were test items and no distractor items were used. However, three items of Part II which did not elicit [fHuman] NP2’s were omitted from the analysis. Procedure: Subjects were tested together in a large room. Fifty-five subjects received a copy of Form A of the written test materials and 55 received Form B. They were asked to provide written responses. The instructions were also written as follows: Part I. Instructions: Think of a good way to complete each of the sentences below, then write it down in the blank space provided. There are no right or wrong answers, but the sentence should make good sense after you complete it. Work rapidly. The sentences are easy, but, if you have trouble completing one, continue on to the other sentences and return later. Part II. Instructions: For each ‘sentence frame’ below, fill in the blanks to make a sensible sentence. Fill in the first two blanks with names of people or things and then

Factors injluencing assignment of pronoun antecedents

235

complete the sentence by giving a reason for the event, just as you did before. If you have trouble filling in a sentence frame, don’t spend too much time on it. Go on to the reate the apple because -he maining sentences, and then return. Example: Tom was hungry. Scoring: The same scoring procedure as in Study I was used in Study II. Again, two scorers categorized the sentence completions into surface NPl and NP2 type responses (99% agreement). The total possible number of responses for any single sentence was 110 (5.5 in Form A plus 55 in Form B). Thus there were 2200 possible responses. A third category comprising responses which were either ambiguous (A), unintelligible (U), or absent (NR) was also used. A+U+NR responses equalled 122 (6%) of total possible responses). A+U+NR responses did not appear to be distributed systematically in relation to test items and were omitted from further analysis. The NPl and NP2 type completion data for Part I are presented separately for each of the four variables investigated. A nonparametric statistic, chi-square, was used since the type of data collected did not meet requirements for parametric analyses: First, the data are not interval scale data, and second, we do not know anything about the distributions of these data. A chi-square analysis is sufficient given that it reveals changes in the distribution of NPl to NP2 relative to the various experimental factors manipulated. The analyses focus on the differences in distributions as a function of experimental factors rather than absolute shifts in directionality of antecedent choice.

Results I. Active-Passive. Table 2 presents the number of surface NPl and NP2 responses for each of the five sentences in the active and passive voice. The distribution of NPl and NP2 responses suggests that voice influenced the assignment of pronoun antecedents (x: = 43.4, p < 0.001). Comparison of column 1 with 4 and column 2 with 3 in Table 2 indicates that: (a) sentences which were strong NPl’s in the active voice appeared as weak NP2’s in the passive voice (sentences 2,4 and 5); (b) the sentence which in the active voice was a weak NPl appeared as a stronger NPl in the passive voice (sentence 1); and (c) the sentence with weak NP2 in the active voice appeared as a weak NPl in the passive voice. Thus, with passivization, which reverses the surface order of logical subject and nonsubject of the sentence, there is a general drift in the assignment of pronoun antecedents toward the grammatical subject or ‘topic’ of the sentence (in this case, the surface structure NPl). Separate x2 tests for each sentence to further evaluate the statistical reliability of this effect reveal highly significant effects for sentences 1,2,4 and 5 0, < 0.001 in each case) and marginal significance for sentence 3 @ < 0.10).

236

Catherine Garvey, Alfonso Caramazza, and Jack Yates

Table 2.

Completions for active and passive sentences”

Sentence

NPl

Active

1 2 3 4 5 Total

Passive NP2

NPl

NP2

30 52 23 48 53

18 2 29 2 1

43 24 37 14 13

9 25 13 36 39

206

52

131 -___

122

a NPl refers to the first noun corresponding Passive sentence

phrase of the surface structure: is the same lexical item.

(Verb) chase hit recognize strike approach

thus NPl in the Active and NP2 in the

These results clearly implicate the surface structure properties of sentences as a determinant of antecedent assignment. Thus, in addition to whatever other factors may contribute to the directionality of pronoun assignment, the overt marking of sentence topic (at least as achieved by P-T) introduces a further independent dimension along which antecedent selection is made. II. Positive-Negative

The effects of verb negation on the assignment of pronoun antecedents on each of five sentences tested are presented in Table 3. (The appropriate comparison in Table 3 is that of column 1 with 3 and column 2 with 4.) With the exception of sentence 1, the positive verb sentences show NP2 preference, which is attenuated in the verb negated versions (XT = 31.5, p < 0.001). x2 tests for each individual sentence revealed highly significant effects for sentences 2 and 3 (p < 0.001 for both) and marginal significance for sentence 5 (p < 0.10). Sentence 4 was also affected in this direction but only slightly 0; < 0.20). Sentence 1 stands in contrast to the other four sentences in several respects. First of all, it was the only sentence that showed preference values for NPl in its positive version; Table 3.

Completions for positive verbs and negative verbs _ Positive

Sentence 1 2 3 4 5 Total

Negative

NPl

NP2

NPl

52 3 11 13 22

1 50 42 41 33

101

161

NP2

(Verb)

56 29 32 19 29

0 26 21 31 20

sail (past) kill fear blame pick (UP)

163

98

Factors injluencing assignment of pronoun antecedents

237

secondly, it differed from sentences 2 to 5 in that they all involved animate subjects and nonsubjects while sentence 1 had an inanimate subject and nonsubject; thirdly, and most importantly, sentence 1 involved a movable NPl (rocket) and, for our purposes, a fixed NP2 (moon). This latter juxtaposition apparently restricted drift in attribution of cause independently of other factors that may be relevant to the antecedent choice. In general, then, we may conclude that verb negation affects pronoun antecedent assignment. This obviously reinforces our earlier claim that the verb implicitly specifies a causality relationship between the nouns of a sentence, which to a major extent determines the direction of pronoun antecedent assignment. III. Interrogative and Declarative In Table 4 are presented the pronoun antecedent assignments for five sentences in the interrogative and declarative forms. As is immediately obvious from these results, there is no difference in the direction of pronoun assignment as a function of the two forms investigated. These results stand in contrast to those obtained for the active/passive variable discussed earlier. There a surface structure factor, presumably purely syntactic, significantly influenced the directionality of the pronoun assignment. In the present case, however, the surface structure variable we manipulated did not have any noticeable effects.

Table 4.

Responses to interrogative and declarative sentences

Sentence

Interrogative NPl NP2

Declarative NPl

NP2

(Verb) sell punish

2 3

40 0 25

12 55 26

39 3 19

14 52 3.5

4 5

32 23

18 26

34 36

13 16

120

137

131

130

1

Total

help miss give

IV. Congruent-Noncongruent The effects of congruent and noncongruent collocation of noun status with the lexical features of the verb are presented separately for verbs identified (in pilot work) as NPl’s and NP2’s. Table 5 presents the results of sentences with NPl’s and Table 6 for those with NP2’s. In both types of sentences the Congruent-Noncongruent variable has a large effect on choice of antecedent (xt = 15.3, p < 0.001 for data in Table 5 and x:= 24.7, p < 0.001 for data in Table 6).

238

Catherine Garvey, Alfonso Caramazza, and Jack Yates

Table 5.

NPl verbs: Responses to congruent and honcongruent noun phrase-verb phrase collocations Congruent

Sentence 1 2 3

NPl

Noncongruent NP2

NPl

NP2

(Verb) argue (with) confess (to) join

41 55 50

9 0 1

20 45 50

30 3 1

Total

146

10

115

34

Table 6.

NP2 verbs: Responses to congruent and noncongruent noun phrase-verb phrase collocations Congruent

Sentence 1 2 Total

Noncongruent NP2

NPl

NP2

(Verb)

8 I

44 46

22 28

30 21

praise criticize

15

90

50

51

NPl

_

x2 tests for each sentence in Table 5 showed a highly significant effect for sentence 1 (p < 0.001) and sentence 2 reveals only a slight drift in assignment. Sentence 3 reveals no difference between the congruent and non-congruent collocations. x2 tests for the two sentences in Table 6 were both highly significant (p < 0.001). Thus, Tables 5 and 6 together show that the status values of the NPl and NP2, as these interact with the lexical meaning of the verb, appear to be a factor in determining pronoun assignment. These results provide independent confirmation of the interaction of components in sentential structures (Heise, 1970). The patterning of results for this complex variable suggests that the evaluative loading of the verb plays a role in producing a congruent collocation. The results of Part II, the task which employed a framework containing verb and conjunction only, provided further evidence for the contention that sentences such as we have discussed can indeed be produced by subjects. The response scores (numbered in order of item presentation with items read, won, scratch omitted) are presented in Table 7. Response bias for verbs previously studied confirms earlier findings on directionality, i.e., blame, NP2, (p < 0.02) and telephone NP1, (p < 0.001). Further, the range of response bias obtained ranged from 61% for blame to 96% for congratulate, adding weight to the hypothesis that verb roots differ in the degree to which a bias toward causality is present.

Factors inj7uencingassignment of pronoun antecedents

Responses to unaccompanied verbs: Total number of Nl + N2 responses; number of NI and N2 responses; percentage’ of dominant responselsubordinate response; and level of probability

Table 7.

Sentence

239

Number

Total”

Nl

N2

70

p(2-0

96 87 91 93 90 86 100

31 55 20 20 16 81 4

59 32 71 73 74 5 96

61 63 78 78 82 94 96

0.0200 0.0182 0.001

(verb) blame anger

envy love admire telephone congratulate aN=

1 -

110.

Discussion Aside from the specific concern with factors affecting pronoun antecedent assignment, the results reported in the present studies bear on two general issues central to the explication of a comprehension model of language. First, the identification of structural components of a sentence that have psychological import; and, second, a specification of how these components relate to each other such as to permit the assignment of a semantic reading (comprehension) to a sentence. To the extent that we have identified specific factors that affect pronoun antecedent assignment, we have also identified linguistic elements that must play a role in a psychological theory of comprehension. The significance of this rests upon the distinction that can be drawn between a linguistic and a psychological explanation of the relationship between sentence structure and meaning. It is considered the task of a linguistic grammar to assign a unique structural representation to any and all interpretations of a sentence, but the linguistic grammar is not required to account for a listener’s selection of one or another possible meaning of an ambiguous surface structure. Rather, it is the task of a theory of comprehension to determine the means by which a listener assigns one or another possible readings to a given sentence token. To do so, it must explain how these operate in the selection of a particular interpretation. By identifying linguistic factors which contribute to obtained respon’se bias in the comprehension of sentences with potentially ambiguous pronouns, we have attempted to uncover clues to the nature of the interpretive process. Consider the example (6) cited in the introduction: (6) George telephoned

Walter because he wanted sympathy.

As we pointed out, this sentence has at least two underlying representations: The pronoun he can be coreferential with either of the two noun phrases in the antecedent clause. With potentially ambiguous sentences of this type, it is often the case, however,

240

Catherine Garvey, Alfonso Caramazza, and Jack Yates

, that subjects seem to prefer one reading over another. At the level of linguistic description this bias for one reading would be inconsequential ~ it would be ascribed to performance variables. At the psychological level, on the other hand, the bias supplies an important source of information on how a subject may actually determine the specific meaning of a particular sentence. Thus, to pursue this example further, if we find that for sentence (6) subjects generally tend to interpret the pronoun as coreferential with the first NP (George), then we must assume that there are some aspects of this sentence that “restrict” the interpretation of the sentence. * The identification of those variables that restrict the interpretation of the sentence and a description of the interaction of these variables is a positive step towards the development of a comprehension model of language. In examining the factors that affect the assignment of pronoun antecedents, several interesting results emerge. First, individual sentence fragments appeared to evoke consistent biases in pronoun assignment in the sentence completion task. Even when sentences were presented in the form of questions to be answered rather than incomplete statements, there was little change in the proportion of NPl to NP2 responses. In addition, these sentences that were common to Study I and II maintained remarkably consistent biases even though the nature of experimental materials and subject populations were clearly different. In contrast to the stability of the effect over variables of age and context, several linguistic variables produced marked changes in the direction and extent of bias and thus are indicated as governing factors in the assignment of pronoun’ antecedent relations. The operation of these linguistic variables has been heretofore neglected, in part perhaps because purely grammatical cues, such as gender or number marking, often operate to assist the process of antecedent assignment. For example in (8) (8) George telephoned

Anne because she wanted sympathy.

the grammatical use of gender directly blocks alternative interpretations. In particular, the semantics of verbs seems to be important, perhaps dominant, in determining antecedent choice. When verbs were given alone, they produced distributions comparable to those produced by questions or statements, i.e., they produced biases from extreme NPl to extreme NP2. Also, a verb occurring in isolation, in clauses or in questions, was always biased in the same direction. (The single exception to this statement was the verb help. See page 232 above.) These results support the notion that a semantic property of the verb influences the selection of antecedents for ambiguous pronouns and further suggest that the influence operates both within and across sentential structures. The property itself was identified as ‘implicit causality’, a feature that selects one or the other of the available candidate nouns as primarily responsible for instigating the action

*In normal language use, of course, both linguistic and important role in the interpretation process. We are concerned to (though perhaps not limited to) intrasentential structure.

extralinguistic context here with those aspects

would play an which are basic

Factors injluencing assignment of pronoun antecedents

24 1

or state denoted by the antecedent clause (Garvey and Caramazza, 1974). This feature establishes a primary direction of cause in the antecedent clause; and, unless some modulating influence is exerted by other linguistic elements, the feature will proceed to ass&Y the Pronoun according to the direction established by the verb. The actual values taken by the causality direction feature are not binary. Rather, as indicated by the data, the strength of imputed cause should be conceptualized as varying continuously. Thus, as shown in Table 7, each verb varies in terms of the degree to which it restricts or determines a particular reading, e.g., the verb telephone strongly determines the choice of Nl whereas anger does so only weakly. Further evidence that supports our analysis of the relative importance of the verb as a major determinant of antecedent choice accrues from the results obtained with explicit negation of the verb. The fact that negation strongly influenced bias in pronoun-antecedent assignment clearly indicates that at least one important locus of antecedent determination is indeed the verb. A possible objection to the existence of an implicit causality feature is that some other linguistic property of verbs might account more simply for the results reported here. For example, underlying relations of case (Fillmore, 1968) might explain the tendency of certain verbs to attract pronoun assignment to the underlying Agent or Experiencer of the antecedent clause. We propose that the sensitivity of bias to semantic properties of noun status in N-V-N collocations, e.g., praise, Table 6, and argue with, Table 5, rules out any direct relation between case structure and the implicit causality feature. In the congruent collocation, The assistant argued with his boss, a strong NPl response was obtained. In the noncongruent collocation, The boss argued with his assistant, a weak NP2 response was obtained. The underlying case relations, however, would presumably be the same in both variants. Similarly, it does not seem possible to attribute the results of the active-passive sentence variants to an underlying case structure, since the structure would be the same for both variants. Thus, the postulated feature of implicit causality cannot be reduced in any direct way to deep case relations. As for other possible semantic interpretations of these results, we should note that many ‘NP2’ type verbs reference emotional states on the part of the subject of the antecedent clause, e.g., admire. Some, however, do not, e.g., punish. Also, verbs which have performative potential in other constructions may be either NPl types, e.g., confess to, or NP2 types, congratulate. These observations suggest but, of course, do not confirm the independence of the implicit causality feature in the semantic structure of verbs investigated here. A final point on the analysis of the verb concerns the generality of the phenomenon described. The results obtained for the feature ‘direction of causality’ is not an artifact of the connective because. The direction of bias for specific verbs can be predicted to hold under other conditions. For example, congratulate led to NP2 responses and confide to NPI responses when marked for past tense and in sentences connected by because. In (9) and (10) these verbs appear to maintain the same direction in future verb phrases and in sentences connected by if and when.

242

Catherine Garvey, Alfonso Caramazza, and Jack Yates

(9) NP2: Arthur will congratulate

( if ) he gets the raise. Henry (when)

(10) NPl : Alice will confide in Mary ( if ) she gets in trouble. (when) These examples clearly attest to the generality of the phenomenon. Two other factors that significantly affected the choice of antecedents were (1) the relative status of the person referenced in the two candidate noun phrases and (2) pas_ sivization. These factors are best thought of as modulating the causal valence of the verb rather than as operating independently in determining antecedent choice. That is, the ef’h.2 of these last two factors is to attenuate or augment the causal valence of the verb and not to establish causality independently of the verb. Thus, in the case of the effect of the re1ative status of the two nouns in the antecedent clause, it seems that the status contrast between the two nouns interacts with the causal valence of the verb to produce a sentence-specific causal valence. The process through which the interaction actually emerges seems to involve the perceived plausibility of verb-established direction of causality in the context of permissible social relations obtaining between the referents of the two nouns. Passivization is the variable examined which, at least superficially, seems to suggest a direct effect of a syntactic factor on pronoun antecedent selection. Whether this effect should be attributed directly to the syntactic properties of the passive versus the active forms or to the concomitant pragmatic effect of ‘topicalization’ which the passive form marks (Anisfeld and Klenbort, 1973) depends on whether a linguistic or a psycholinguistic level of explanation is stressed. A linguistic analysis would suggest that the passive element is either present or not present in the deep structure. If the surface structure reflex of this element interacts with the direction of causality of the verb in sentence comprehension, however, it does not apparently do so in an all or none fashion. We may consider that the marked topic (ordered by the syntactic transformation) exerts - for the listener - a bias value for NPl. This value then interacts with the directionality value of the verb to produce a sentence-specific value. This explanation is parallel to that advanced for the operation of the Congruence variable. To conclude, we have demonstrated a technique whereby an implicit semantic feature can be related to a grammatical alternative (pronoun-antecedent assignment) and thereby made explicit. We have also demonstrated that pragmatic, syntactic and other semantic features interact in an orderly way with this implicit feature of causality in verbs. The next steps are to determine whether the interacting influences can be more precisely isolated in sentence comprehension processes and to study directly how listeners understand or comprehend what we believe to be not wholly ambiguous sentences.

Factors injluencing assignment of pronoun antecedents

243

REFERENCES Anisfeld, M., & Klenbort, I. (1973) On the functions of structural paraphrase: The view from the passive voice. Psychol. Bull., 79, 117-l 26. Fillmore, C. J. (1968) In E. Bach and R. Harms (Eds.), Universals of’linguisric theory. New York, I-90. Garvey, C., & Caramazza, A. (1974) Implicit causality in verbs, Ling. Inq., 5, (in press). Heise, D. R. (1970) Potency dynamics in simple sentences. J. Person. SOC. Psychol., 16. 48-54.

RCsumC Les deux experiences present&es ici, utilisant d’une part un groupe de sujets adultes, d’autre part un groupe de sujets ages de 10 i 13 ans, consistent a faire complkter des phrases de la forme: NP,-V-NP,

P&tom...

Le fragment de phrase: “John a telephone a Bill parce qu’il... ” induit des completions du type, “...avait de bonnes nouvelles”. L’impetus du verbe telephoner est localis en NP, (John). Le fragment de phrase, “John admirait BilJ parce qu’il” induit des completions du type, “...etait un bon athlete”. L’impetus est alors en NP, (Bill). Le choix de la reference du prinom de la subordonnee, en NP,, ou en NP,, montre le r61e du facteur de directionalite de la causal&& Celui-ci peut dtre determine par la racine du verbe. On a isold ce facteur des traits semantiques et syntaxiques avec lesquels il peut interagir. L’incidence plus ou moins importante du facteur de directionalite de la causalitk, depend de la manipulation de ces traits. Dans la discussion, on examine la portee de la technique utilisee dans cette recherche, pour l’etude de la semantique des verbes et pour la mise en Evidence, dans la comprChension syntaxiques.

psychologique

du langage,

de I’interaction

de traits

sgmantiques

pragmatiques

et

“in ‘, “on”, and “under” revisited

STEPHEN WILCOX DAVID

S. PALERMO”

The Pennsylvania State University

Abstract The present study presents evidence that young children S comprehension of’the locatives ‘in’, ‘on’, and ‘under’ is, at least in part, contextually determined. Children aged 1.63;0 were given tasks with verbal instructions which were either contextually congruent or incongruent, The results were interpreted in terms of’ the non-linguistic as well as linguistic strategies apparently used to interpret speech. The results and interpretation are in contrast to those oj’earlier research.

In recent years a large amount of work has been directed toward shedding light upon the processes through which a child acquires the semantic system of his first language. One of the most consistent findings reported in the literature is that young children have a tendency to overgeneralize the use of words (e.g., Clark, 1973). For instance, a two year old child might use the word ‘dog’ to designate all animals or he might use ‘dad’ to refer to all men. This phenomenon has been reported for relational terms as well as for words referring to concrete objects. For example, Wales and Cambell (1970) and Clark (1973) found that when children are asked to give the antonyms for such words as tall, short, long, or narrow, they respond most often with one of the pair ‘big-wee/small’. Clark (1973) has attempted to account for these overgeneralizations (or overextensions) in terms of semantic features into which a given meaning may be analyzed. The features are assumed to be hierarchically ordered and the hypothesis is that initially the child interprets words only in terms of the most general features. These latter features are thought to be derived from the perceptual system of the child. As the child progresses, he gradually adds more specific features to his meanings for words. For example, at first the child would only know that a word such as ‘big’ refers to size. At this stage he would use ‘big’ to refer to all relations of size. A corollary of the semantic feature theory is that the child learns the positive or unmarked end of an antonym pair before he learns the negative or marked end, and at one stage in his linguistic development, he will take both members of an antonym pair to *We would like to thank C. N. Cofer for suggestions which led to this research. Reprint requests should be sent to the second author, 441 Moore Building, Pennsylvania State University, University Park, Pennsylvania 16802. Cognition 3(3/, pp. 245 - 254

246

Stephen

Wilcox and David S. Palermo

have the same meaning. Studies by Donaldson and Balfour (1968) and Palermo (1973) have given support to this last prediction. These investigators found that a sizable portion of children between the ages of 3 and 8 treat ‘less’ as if it were synonymous with ‘more’. Similar results were obtained with ‘same-different’ (Donaldson and Wales, 1970) and ‘before-after’ (Clark, 1971). Other studies have shown that children respond more accurately in comprehension tasks when more general terms are used (e.g., ‘big-small’) than when more specific terms are used (e.g., ‘tall-short’) (Wales and Cambell, 1970). One problem with the semantic feature hypothesis stems from the fact that it clearly predicts overgeneralizations in comprehension of concrete object words. For instance, if the child takes ‘dog’ to mean ‘four-legged animal’, he should sometimes respond to questions regarding the location of the dog by pointing to the cat, particularly if the dog is not present. Few such examples have been reported in the literature. Huttenlocher (in press), in fact, found no evidence of overgeneralizations in comprehension among four children 10 - 19 months old tested over a period of six months in their homes. The testing procedure consisted of asking the children for objects from an array of items. According to Huttenlocher, if the children didn’t know a word, they simply didn’t respond. It is particularly interesting that the same children were reported by their mothers to display overextension in production. Another apparent difficulty with the feature theory is reflected in the relative absence of confusion between polar adjectives except for a select few such as ‘more’ and ‘less’. It seems intuitively unlikely that children take ‘small’ to mean ‘big’ at any stage. A recent experiment by Brewer and Stone (in press) bears on this question. They found that when subjects were given a choice, their errors were much more likely to be different dimensionsame pole than same dimension-different pole. For example, long and tall were confused more often than long and short. When asked to point to the ‘long one’ when both horizontal and vertical rods of varying sizes were presented, subjects were much more likely to point to the tall rod than to the short (horizontal) rod. These results directly contradict the prediction of the semantic feature hypothesis that the ‘polar’ feature will be the last distinction to be acquired. In light of the general paucity in instances of overextensions in comprehension reported in the literature, one study done by Clark (1972) seems particularly striking. It concerns the locative terms ‘in’, ‘on’, and ‘under’. Clark reported evidence that ‘in’ is acquired before ‘on’ which in turn is acquired before ‘under’. This doesn’t seem strange considering that ‘in’ is three times more frequent than ‘on’ which is nine times as frequent as ‘under’, according to the Thorndike-Lorge (1944) compilation of word frequencies. But, according to Clark, a large portion of her subjects (children 1;6-3;O) treated ‘on’ as if it meant ‘in’ and treated ‘under’ as if it meant either ‘in’ or ‘on’. Her explanation of the results was as follows: ‘In’ and ‘on’ are less complex than ‘under’ because ‘in’ refers to a container and ‘on’ refers to a surface; whereas if something is ‘under’ an object, it is on a surface which supports that object. ‘In’ is more general than ‘on’ since a container refers to three dimensions but a surface is only two-dimensional.

‘in’, ‘on’, and ‘under’revisited

247

Thus Clark makes the analogy between ‘in’ and ‘on’, on the one hand, and adjectives such as ‘big’ and ‘tall’ on the other. There is an obvious weakness in such an analogy. Though ‘big’ to some extent subsumes the meaning of ‘tall’, ‘in’ does not subsume the meaning of ‘on’. Likewise, ‘big’ can be substituted for ‘tall’ but ‘in’ cannot be substituted for ‘on’ without a major alteration in meaning. Clark suggests a different interpretation of her findings with ‘in’, ‘on’, and ‘under’ in a more recent paper (Clark, 1974). She offers a partial semantic hypothesis which makes the claim that a combination of limited semantic knowledge and non-linguistic strategy can better account for many of the facts of language acquisition. With regard to the locative terms in question, Clark posits the existence of two hierarchically ordered rules: (1) If the stationary object is a container, the movable object goes inside it; and (2) if the stationary object has a horizontal surface, the movable object goes on it. Though we agree with Clark that her subjects’ responses were determined in part by non-linguistic strategies, we are not convinced of the generality of the particular strategies she proposed. It appears to us that Clark’s results may have been a contextually determined artifact of the particular objects she used. The children were given small toy animals and asked to put them ‘in’ or ‘on’ a tunnel and a box; ‘in’ or ‘under’ a truck and a crib; and ‘on’ or ‘under’ a bridge and a table. It seems reasonable to assume that children would be predisposed to put the objects in a tunnel or box rather than on it, in a truck or a crib rather than under it, and on a bridge or a table rather than under it, thus confirming Clark’s predictions. Furthermore, most of the responses which would confirm the predictions seem to be easier for the subjects in terms of the motor responses involved. Thus, it appears that the contextual support for the linguistic statements presented to the children by Clark was such that the children had no alternative to the specific nonlinguistic strategies which the children in that study selected when they did not know how to respond to the statements including ‘on’ and ‘under’. The present study was designed to determine whether alteration of the contextual support of the linguistic statements used in the Clark study might lead to different results. Specifically, it was anticipated that the child’s knowledge of the usual relation of the movable object to the stationary object might vary with age and that if that knowledge conflicted with the linguistic statement, the contextual relations might be more important in determining the appropriate response than the linguistic relation.

Method Sub]ects. The subjects were 36 children from State College, Pennsylvania. At least one parent of each child was either a faculty member or graduate student at The Pennsylvania State University. The subjects were divided into three age groups: (i) 1;6-1 ;I 1 (mean age 19); (ii) 2;0-2;5 (mean age 23); (iii) 2;6-2;ll (mean age 2;9). Half of the subjects in each group were males, half females.

248

Stephen

Figure 1. Table IA

Wilcox and David S. Palermo

Stimulus objects.

Bridge IIA

Truck IUA

Teapot

Boat

Road

Block

IB

IIB

IIIB

IPB

Neutral IYA

fiaure -

M Materials.

The objects used are shown in Figure 1. The four objects labelled IA-WA are the stationary objects: those labeled IB-IVB are the corresponding movable objects. The relative size of the objects is shown in the figure.

Procedure.

Each subject was tested with each pair of objects. Object Pairs I-III were used twice and object Pair IV was used 3 times with each subject. In the three congruent tasks, the subjects were instructed to put the teapot on the table, the boat under the bridge, and the road under the truck. In the three contextually incongruent tasks, the subjects were asked to put the teapot in the table, the boat on the bridge, and the road in the truck. Half of the 6 subjects in each cell formed by age group and sex were presented with the 3 congruent tasks first and half with the 3 incongruent tasks first. All of the subjects received the presentations with stimulus Pair IV, following all the presentations with each of stimulus Pairs I-III. Pair IV was used with ‘in’, ‘on’ and ‘under’. It was designed to be contextually neutral. Each of the six possible orders of this task was used once within each age-sex cell. The instructions for the first six tasks were as follows: “See the teapot/boat/road (E giving the appropriate movable object to the subject) and the table/bridge/truck (E grasping the appropriate stationary object). Can you put the X (movable object) in/on/ under the Y (stationary object)?” The instructions for the last three tasks were as follows: “Can you put this (E giving S the block) in/on/under that (E grasping object WA)?” If the subject did not respond within a few seconds on any particular task, the instructions were repeated. No more than two repetitions were required on any of the tasks.

‘in ‘, ‘on ‘, and ‘under’ revisited

249

The subjects were given the objects before the testing procedure began. The testing was never begun until the subject in question had been playing spontaneously for a few minutes with the objects. This warmup period was intended to minimize the subject’s fear of both the objects and the experimenter. All of the subjects were tested in their homes by the same male experimenter with their mothers within sight. One prospective subject was eliminated due to failure to respond favorably to the experimenter.

Results

? \q s-

A 2(congruency versus incongruency) X 3(stimulus pairs) X 3 (age groups) analysis of variance was performed, congruency and stimulus pairs being within subject variables and age groups a between-subject variable. None of the main effects were significant but there were significant interactions between age and congruency, F(2,33) = 10.31, p < Figure 2.

Percent correct responses as a function of age group.

100

El c” 758 ii L

I

‘1

\

\

\

\ v \ \ \

-‘-A

&-____*

\

E E 25.

-A y----~

d

k n

-

-

n

0 A

0

congruent

tasks incongruent tasks stimulus pair I stimulus pair II stimulus pair III

1;6-131 2;0-2;5 2;6-231

Age

groups

0.001 and between stimulus pairs and congruency F(2,66) = 10.1, p < 0.001. Figure 2 gives the percent correct responses for the tasks performed with object Pairs I, II, and III. It can be seen from the figure that, though the overall numbers of correct responses are similar for the 3 age groups, the pattern of correct responses changes appreciably. For Pair I, the table and the teapot, there is little difference between the age groups, but for

250

Stephen

Wilcox and David S. Palermo

the stimulus Pairs II and III, the bridge and the boat and the truck and the road respectively, there is an obvious age trend. The youngest subjects performed better on the incongruent tasks, but the older children performed better on the congruent tasks. All of the incorrect responses were those which would have been correct with the other locative term for that pair, except for 5 responses of the youngest age group which involved putting the movable object beside the stationary object. Table I gives the results of the tasks performed with pair IV. The horizontal dimension gives the locative term used in the instructions; the vertical dimension gives the response

Table 1.

The number and types of responses on stimulus pair IV Instructions

Responses

in on under

in

on

under

in

on

under

in

on

under

1 5 0

5 I 0

I 4 1

8 3 1

5 7 0

6 2 4

8 1 3

3 8 1

6 1 5

Age 1

Age I1

Age I11

of the subjects. Correct responses are on the diagonals. The only obvious difference between ages is that the two older age groups were more successful than the youngest age group with the ‘under’ task. The most frequent response, regardless of the instructions, was for the subject to put the block ‘in’ the stationary object. Figure 3 gives the numbers of subjects who gave the same response to all three of the locatives used with object Pair IV (perseverators) and those who gave all 3 possible responses without getting them correct labeled alternators but, perhaps, more appropriately called permuters. As can be seen from the figure, there was a tendency for the younger children to perseverate and for the older children to change responses. Figure 4 shows the per cent correct responses on the three locative terms for the three age groups collapsed across all the tasks. Performance tends to improve with age with ‘under’ but falls off slightly with ‘in’ and ‘on’ as age increases. For all three age groups performance was better with ‘on’ than with ‘in’ and, except for the youngest age group, performance with ‘under’ was either equal to or better than performance with ‘in’, the oldest age group actually performing best with ‘under’.

‘in’, ‘on ‘, and ‘under’revisited

Percent of subjects who perseverated or alternated with stimulus pair IV.

Figure 3.

25 1

Performance over all tasks

Figure 4.

I loo-

. .

alternators porsoverators

E c: 25

0 Age

groups

I;&l;Tl2;0-2;5 2;6-2;ll Age groups

The results of the present study replicate those of Clark (1972, 1974) when context and language are congruent but demonstrate the important influence of context on the interpretation of language when context and language are not congruent. Except for the youngest age group, any apparent differences in performance for the first 3 stimulus pairs favor ‘on’ and ‘under’ rather than ‘in’. Both Clark’s earlier interpretation in terms of semantic complexity and her more recent “partial semantics hypothesis” predict that the majority of the errors with ‘in’, ‘on’, and ‘under’ wiIl be in taking ‘on’ or ‘under’ to mean ‘in’ or in taking ‘under’ to mean ‘on’. In the present study, however, at least for the two oldest groups, the children responded as if ‘in’ meant ‘under’ or ‘on’ or as if ‘on’ meant ‘under’. There are two types of response tendencies in the present task which may account for the kind of errors observed here: (1) The tendency to make the simplest (and thus the easiest) motor response; and (2) the tendency to put the objects in their most normal (congruent) contextual relationship. The former tendency would result in ‘motor’ errors,

252

Stephen Wilcox and David S. Palermo

the latter in ‘contextual’ errors. It is assumed that the younger child is not as likely as the older child to have as clear a notion of the normal relationship between the objects represented. Thus, when presented with this situation the younger child who does not know the normal contextual relations of the objects, nevertheless does recognize that he is expected to place one object relative to the other and makes the motor response which is the easiest for him (e.g., placing the road ‘in’ rather than ‘under’ the truck) resulting in correct responding in the incongruent context. As the child becomes older and thus more familiar with the objects and their normal relationships to each other, he tends to make more semantic or contextual errors and fewer motor errors either because he does not comprehend the locative terms or because he ignores them in favor of the contextual relations which normally hold between the objects represented. It is not clear from the present results which of these alternatives may be the case. In the case of Clark’s objects, both types of response tendencies would coincide with her hypothesis, except for the box and the tunnel in which the motor bias would favor the ‘on’ response. Clark (1972) points out that the children perform better with the truck and the crib than with the box and the tunnel when the ‘in’ response is required. Her explanation is in terms of the canonical orientation of containers being such that the opening is in the top. This makes containers without the opening in the top less salient. We would like to argue that it is less the saliency of the container orientation and more likely the ease of response for the child when the opening is in the top which determines the response. There is, however, little to choose between the two without further evidence. The stimulus objects used in the present study, in contrast to those used by Clark, load the contextual bias in favor of ‘under’ and ‘on’ rather than ‘in’. Teapots normally go on tables (Pair I); boats normally go under bridges (Pair II); and roads normally go under trucks (Pair III). The motor bias is the same as the contextual bias for Pair I but for Pairs II and Ill the motor bias is in the opposite direction from the contextual bias. The direction of errors should therefore be the same for all age groups with Pair I but there should be a crossover with Pairs II and Ill. As shown in Figure 2, this seems to be the case. All of the subjects performed better when asked to put the teapot on the table than when asked to put the teapot in the table. The younger children tend to perform better when asked to put the boat on the bridge or the road in the truck but the older children perform better when asked to put the boat under the bridge or the road under the truck. Stimulus Pair IV was intended to be as neutral as possible and also to allow all 3 responses. The fact that the majority of all three age groups put the block ‘in’ the stationary object tends to give some support to Clark’s notion of saliency. The large size of the object probably tended to equate motor difficulty for the three locatives. The contained space did seem to be more salient. Some of the parents of the subjects pointed out that many of the children’s toys involved putting one object within another one. This could be one of the bases for differences in saliency.

‘in’, ‘on ‘, and ‘under’revisited

253

The existence of perseveration and permutation strategies with stimulus Pair iV is also of interest. There is an apparent developmental trend from perseveration to permutation of responses. These tendencies seem to represent another form of response bias, or strategy, that the child brings with him to the experiment. The fact that some of the youngest children put the movable object beside the stationary object suggests that some of the children did not know either the meanings of the words or the contextual’ relationships which usually hold between the objects. The realization that locatives imply at least a contiguous relationship between objects was not clear from these responses and suggests that the youngest age groups may not have any linguistic grasp of locatives. Such a stage would precede the stage of limited semantic knowledge of locatives as Clark views it. On the other hand, it is possible to conceive of the present results as indicating that the concept is either present in its entirety or absent. Support for such an interpretation,’ however, requires considerably more empirical evidence. As Clark (1974) points out, even the data on polar adjectives can be explained easily in terms of nonlinguistic strategies. This leaves us with a fair amount of negative evidence and little positive evidence for overextension in comprehension. There still remains the problem of overgeneralization in production. Huttenlocher’s (1974) observation that overextension of the use of words reflects a limited vocabulary makes good sense. It seems clear, however, that the overextension of a word does not imply that the meanings are overgeneral. The order in which words are acquired is surely a complex conceptual matter involving both the abstraction of concepts from instances and the abstraction of attributes from instances. The contextual factors and their interaction with hypotheses fostered in the child appear to be the direction toward which further research must be aimed. Nelson (1973), for example, has suggested some of the complexities involved. Donaldson (in press) in a recent paper has put forth a notion similar to Clark’s partial semantic hypothesis and the contextual view advanced here. She posits the existence of 3 types of rules: (1) Lexical rules which specify the meanings of individual words; (2) syntactic rules; and (3) local rules which will be “selected as criteria for assigning of truth values when the linguistic rules leave the matter vague”. The local rules depend upon the circumstances locally obtaining (i.e., context) and thus they account for the child’s apparent differing interpretation of words. It seems logical to assume that the way in which the young child interprets a word in a given situation is a complex function of the context and a number of linguistic and nonlinguistic strategies. Any theory that attempts to handle the facts of language acquisition must address itself to these strategies. It would appear that from a number of different empirical bases researchers in this area are beginning to focus on all of these factors.

254 Stephen

Wilcox and David S. Palermo

REFERENCES Brewer W., and Stone, B. (1974) Non-linguistic strategies and the acquisition of word meanings. J. exp. Child Psychol., in press. Clark, E. (1971) On the acquisition of the meaning of before and after. J. verb. Learn. verb. Behav., 10, 266-275. Clark, E. (1972) Some perceptual factors in the acquisition of locative terms by young children. Papers from the Eighth Regional Meeting, Chicago Linguistic Society. Clark, E. (1973) What’s in a word? On the child’s acquisition of semantics in his first language. In T. E. Moore (Ed.), Cognitive Development and the Acquisition of Language. New York, Academic Press, pp. 65-110. Clark, E. (1974) Non-linguistic strategies and the acquisition of word meanings. Cog., 2, 161-182. Donaldson, M. (1972) Some clues to the nature of semantic development. J. child Lung., in press. Donaldson, M. and Balfour, G. (1968) Less is more: A study of language comprehension in children. Brit. J. Psychol., 59, 461472. Donaldson, M. and Wales, R. J. (1970) On the acquisition of some relational terms. In J. R. Hayes (Ed.), Cognition and the Development of Language, N.Y., Wiley, pp. 235-268. Huttenlocher, J. (1974) The origins of language comprehension. In R. L. Solso (Ed.), Theories in Cognitive Psychology. Lawrence Earlbaum Associates, in press. Nelson, K. (1973) Structure and strategy in learning to talk. Monographs of the Society for Research in Child Development,

38, l-2,

l-36.

Palermo, D. S. (1973) More about less: A study of language comprehension.

J. verb. Leurn. verb.

Behav., 12, 211-221.

Thorndike, E. and Lorge, I. (1944) The Teacher’s Word Book of 30,000 Words. New York, Teachers College. Wales, R. J. and Cambell, R. (1970) On the development of comparison and the comparison of development. In G. B. Flores d’Arais and W. J. Levelt (Eds.), Advances in Psy’cholinguistics. Amsterdam: North-Holland, pp. 373-396.

R&urn6

La presente etude montre que la comprehension par les jeunes enfants, des locatifs ‘dans’ ‘sur’ et ‘sous’ est, pour le moins partiellement, determinee par le contexte. On a soumis des enfants de 1;6 i 3;0 i des thches en leur donnant des consignes verbales conformes et non conformes au contexte. Les rdsultats sont interpret& en termes de strategies non-linguistiques et linguistiques, untilisees pour I’interprktation du discours. Les rdsultats et l’interpretation sont en contradiction avec ceux des recherches precedentes sur le sujet.

5

From communication to language A psychological’/ perspective* Dedicated

to Roman

Jacobson

on his eightieth

birthday

J. S. BRUNER

University of Oxford

Abstract Any realistic account of language acquisition must take into account the manner in which the child passes from pre-speech communication to the use of language proper. For it can be shown that many of the major organizing features of syntax, semantics, pragmatics, and even phonology have important precursors and prerequisites in the prespeech communicative acts of infants. Illustrations of such precursors are examined infour different domains: The mother) mode of interpreting the infant’s communicative intent; the development of joint referential devices en route to deixis; the child’s developing strategy for enlisting aid in joint activity; the transformation of topic-comment organization in prespeech to predication proper. Finally, the conjecture is explored whether the child’s knowledge of the requirements of action and interaction might provide the basis for the initial development of grammar. Whatever view one takes of research on language acquisition proper - however nativist or empiricist one’s bias - one must still come to terms with the role or significance of the child’s pre-speech communication system. What is the nature of that system and does its very nature aid the passage from pre-speech communication to language? What makes for linguistic difficulty in answering such questions is that since the chil.d’s communication before Ianguage is not amenable to conventional grammatical analysis, efforts to trace continuities often seem little more than a hunt for analogies of ‘grammar’ in early action or gesture. On. the psychological side, the difficulty inheres in the tradition of such research - usually to explain away language as “nothing but” the concatenation of simple conditioning, imitation, or other mysterious simplifications. I shall try in the following pages to show that one can establish continuities between pre-speech communication and language, and do so without recourse to inappropriate reductionism.

*This research was supported by a grant from the Social Science Research Council of Great Britain, and parts of this paper have been presented at the Universities of Stirling and London and at the University College of Swansea. 1 am particularly grateful for criticism to Mr. Churcher, Mr. Leslie, Dr. Scaife, Ms. Caudill, and Ms. Garton of Oxford, as well as to Dr. Richard Cromer, Dr. Elisabeth Bates and Susan Sugarman. 1 gratefully dedicate this paper to Professor Roman Jakobson in honor of his eightieth birthday. Cognition 3(3), pp. 255 - 287

256

J. S. Bruner

The resurgent nativism that followed upon the work of Chomsky (1965) nurtured quite falsely the hope that this first step could be by-passed - although there is nothing in Chomsky’s own writing that would lead necessarily to such a conclusion. The reasoning seems to have been that, if language grows from its own roots, it suffices to study the beginnings of language proper if one wishes to understand the nature of its early acquisition. The programme, in effect, was the linguists’ programme: Gather a corpus of speech, with due regard for context (unspecified), and subject it to grammatical analysis. Or, to add an experimental dimension, contrive experimental situations to tap the child’s capacities for producing and comprehending speech in particular contexts, and draw inferences from the child’s responses concerning his underlying linguistic competence. There can be little doubt that this programme has deeply enriched our understanding of early language and of the course of its early development. The work of Brown (1970, 1973) and his group, of Bloom (1970), of McNeil1 (1970a, 1970b), of Slobin (1973) of the Edinburgh group (e.g. Donaldson and Wales, 1970) - all attest to the enormous progress of the last decade. But the early language for which a grammar is written is the end result of psychological processes leading to its acquisition, and to write a grammar of that language at any point in its development is in no sense to explicate the nature of its acquisition.‘Even if it were literally true (as claimed by Chomsky), that the child, mastering a particular language, initially possesses a tacit knowledge of an alleged universal deep structure of language, we would still have to know how he managed to recognize these universal deep rules as they manifest themselves in the surface structure of a particular language. Even an innate “language acquisition device” would require a programme to guide such recognition and it would fall to the psychologist to discover the nature of the programme by investigating the alleged recognition process (Chomsky, 1965, p. 27). For Chomsky, the child’s problem is “to determine which of the (humanly) possible languages is that of the community in which he is placed”. If there were no such recognition problems, the child would obviously learn language immediately and perfectly - at least those portions of it to which he were exposed. This is so far from what happens that we generally agree that it is eminently worth studying what might be called the prerequisites necessary for learning a language or for progressing in the mastery of that language. At the most general level, we may say that to master a language a child must acquire a complex set of broadly transferable or generative skills - perceptual, motor, conceptual, social, and linguistic which when appropriately coordinated yield linguistic performances that can be described (though only in a limited sense) by the linguists’ rules of grammar. Such rules of grammar may bear no closer resemblance to the psychological laws of language production, comprehension, and use than do the principles of optics bear to the laws of visual perception - in neither case can the one violate the other. If we are to concentrate upon the prerequisite sensory, motor, conceptual, and social skills whose coordination makes language possible, we must alas abandon in large part the powerful grammar-writing procedures of the developmental linguist. For it no longer

From communication to language - A psychological perspective

2.51

suffices to collect a corpus of spoken language for which successive grammars may be written, though these grammars may yield valuable hypotheses about the antecedent psychological processes that brought them into being. Instead one must devise ways of investigating the constituent skills involved in language. And typically one begins well before language begins, following the communicative behaviour of particular children until a particular level of linguistic mastery is achieved, testing as well for other, concomitant indices of growth, Not surprisingly, then, there are few such studies available, most still in progress: Trevarthen (1974a, 1974b), Sugamran (1974) Bates, Camaioni and Volterra (1973) Lock (1972) McNeil1 (1974) Dore (1974, 1975) Urwin (1973) and Bruner (1975) though more are starting up. What is peculiarly difficult in conducting such studies is that their design depends upon more or less explicit decisions concerning what beside language should be studied, decisions derived from hypotheses about the precursors and prerequisites of early language. Typically, in such work, an investigator starts by selecting a ‘target’ process in later speech and explores its precursors in prelinguistic communication, concentrating upon forms of communication later realized by linguistic means but earlier fulfilled (partially or fully) by gestural or other expression. Studies of this kind explore the continuity between functionally equivalent forms of communication before and after the onset of speech proper. The investigator, for example, may study the prelinguistic devices a child uses for making a request or for establishing a joint referent before these can be handled through such grammatically appropriate means as interrogatives or demonstratives. Inevitably, such an approach shifts emphasis to functions of language use, to pragmatics and communicative competence (Campbell and Wales, 1970) and away from syntactic competence in the sense employed by Chomsky (1965) and McNeil1 (1970a, 1970b). If one pursues this course, one is tempted to look for the ‘grammar’ inherent in certain forms of social interaction, the emergence of ‘proper’ grammar then being conceived of as the child gaining insight about how to express in language an idea previously held but expressed by other than linguistic means. Both Sugar-man (1974) and Bates et al. (1973) for example use non-linguistic behavioral indicators to infer the presence in pre-speech behavior of such concepts as Agent and Instrument (when the child signals an adult to help him do something that he wishes to accomplish). They see these prelinguistic accomplishments as precursors of case-grammatical categories like Agentive and Instrumental in Fillmore’s (1968) sense of case grammar (of which more will be said later). But this procedure inevitably brings the investigator to the psychological question: What makes it possible for the child to progress, say, from a prelinguistic form of expressing the demonstrative or agentive to a more advanced linguistic fomr of expression? It is at this point that the second side of this type of research emerges: The search for the constituent skills relevant to linguistic mastery. The commonest practice is to turn to Piaget (e.g., Sinclair, 1969). According to his well-know view, language is facilitated by the development of sensori-motor schemas, that represent the joint outcomes of percep-

258

J. S. Bruner

tion and action. These undergo orderly changes that are nourished (though not shaped) by continued experience in acting on the world. In time, for example, the child comes to separate thought from action in his schemas, and his concepts of objects and events in the world become independent of the actions to be performed on them. Sensori-motor schemas also come with experience to transcend space and time, so that the concept of an object is no longer tied to particular contexts, but becomes somewhat more context-free. The acquisition of language is seen as somehow emerging from these developments. Thus a concept of objects that is independent of action on the object should aid the child in mastering such linguistic distinctions as Action and Object in case grammar or, even Noun Phrase and Verb Phrase in the more usual.generative grammar. Bates et al., (1973), for example, attribute prelinguistic progress in signalling imperatives and declaratives to the child’s maturing Piagetian sensori-motor schemas, though their basis for doing so is partly by appeal to coincidence between the times of appearance of different forms of signalling and the dates usually cited in Piagetian norms and partly to their presence in the sample studied. Sugarman does somewhat better in this regard. She notes Piaget’s observation (1952) that the child, in organizing a sensori-motor schema, will first go through a phase of dealing separately with particular objects before he is able to subordinate the use of one object to the other, as in using one as a tool for getting the other. She likens this to the process of early skill development which progresses by the combining of skilled routines that have first developed separately (Bruner, 1973). With this as background, she postulates that the child will first go through a stage in which he treats persons and objects independently, developing schemas for each, will then elaborate these, until finally he will combine them into a unified schema: Person-as-agent-to-help-obtain-object. As this schema develops, the child will acquire signalling techniques appropriate to his level of growth. And indeed, her data indicate that the child first addresses himself separately to objects or to the mother, and finally learns to signal the mother to get her help in obtaining an object, with an intermediate stage in which there is elaboration of signalling toward mother and object separately. My principal concern with the Piagetian approach of these authors and of Sinclair (1969) is that it concentrates almost exclusively on the formal aspects of language at the expense of the functional, the emerging structure of the child’s language without due regard for the uses to which language is put in different contexts. It will be clearer in the following pages why I think this is a serious and distorting difficulty. But in general, one can only applaud the aim of such efforts, for they do indeed represent the kind of ‘middle way’ between extreme nativism that sees no problem because it is all there in advance, and extreme empiricist reductionism that sees no problem because it dismisses what is in fact already there by way of readiness to use language in a particularly structured way. Whoever studies prelinguistic precursors of language must, I believe, commit himself to what Cromer (1974) has recently called the “cognition hypothesis”. The cognition

From communication to language - A psychological perspective

259

hypothesis has two parts to it. The first holds that “we are able to understand and productively to use particular linguistic structures only when our cognitive abilities enable us to do so” (Cromer, 1974, p. 246). The second holds that when our cognitive abilities allow us to grasp a particular idea, we may still not have grasped the complex rule for expressing it freely but may nonetheless express it in a less complex, if indirect form. He provides as an example the perfect tense and the conceptual idea of completed action: A child who has not yet grasped the perfective device embodied in such sentences as Have you peeked? can nonetheless express the same meaning by using the simpler rule of combining known form with the word yet, yielding utterances like,

a

Did you peek yet? Presumably, this capacity to express a cognitive insight by means short of the fully realized linguistic rule can be pushed down in age to the point where one asks whether the child is able to use pre-linguistic devices for expressing a cognitive insight even before sentential grammar is present in his language. Both parts of the cognition hypothesis presuppose the doctrine of functional substitutability or at least of continuity. Neither is a doctrine to be facilely accepted. In semantics, substitutability is represented by Bloomfield’s (1933) ‘fundamental postulate’ that in any given speech community one can find formal and semantic equivalents of certain sentences that serve as ‘glosses’ of each other. The transformational grammarian usually handles these matters by invoking “a common ‘underlying’ structure for semantically equivalent ‘surface’ syntactic arrangements” (Silverstein, 1975). But while in Cromer’s example of the more and less compact versions of the grammatical perfective, one can arguably make the case for a gloss, it becomes progressively more difficult to do so as communicative devices become more separated in ontogenetic time. What is the relationship, for example, between a gestural sign of pointing to an apple and the uttering of the word apple? They are plainly not glosses of each other in any formal sense. Yet, one is prepared, if only intuitively, to grant that they may be continuous with each’other. We say that the two serve the same function of indicating, or at least some aspect of this function. If we make a further separation in time and compare a gestural indication with a simple sentence, That apple (whether or not we assume that the existential copula is absent because of a deletion rule), then the gap becomes so great as to seem discontinuous. For example, the predicational form of the more advanced utterance makes it amenable to truth testing, the word that already presupposes diectic marking, etc., etc., none of which are remotely attributable to ostensive pointing. Yet, again we assume there is some continuity. In what does it consist? I would suggest that continuity can be attributed for two reasons: The first is by a principle of incorporation, that in achieving competence to utter a simple sentential indicative the child necessarily incorporates prior knowledge implied in his mastery of the

760

J. S. Bruner

ostensive indicative. But this is surely a weak form of continuity by incorporation, no stronger than the indubitable claim that an infant must stand before he can take his first step. It is strengthened by two additional considerations, one treated below in examining some bases for attributing continuity, the second being the following. If we can show that the child’s prior grasp of ostensive indicating by pointing provides knowledge that permits him to ‘crack the code’, say, of lexical indicating, that it is in fact a stepping stone in a line of prerequisites leading to the use of a simple sentential form of indicating, then incorporation ceases to be merely logical and becomes psychological. The stronger form would be, then, that lexical indicating occurs if and only if the child has previously grasped a more primitive device of indicating and can be shown to use that device instrumentally in the acquisition of the more advanced form. In a word, for a precursor utterance to become psychologically and linguistically interesting, it must be shown to be an instrumental prerequisite to a more evolved utterance. A second basis for attributing continuity is provided in a more comprehensive view of the nature of language use within any given culture, a subject too readily overlooked in our headlong pursuit of structural regularities in grammar. Stated at its most banal, it is that speech is meaningful social behaviour. But at the same time, it is crucial to bear in mind that articulate phonetic speech is only one of the devices by which meaning is transmitted in such social behavior. Whatever the device employed, “this functional sign mode always involves some aspect of the context in which the sign occurs” (Silverstein, 1975). This pragmatic aspect of sign use is dependent upon a mastery of cultural conventions and it is the linkage of signs to conventions that assures the ‘meaningful’ use of any signalling system, language included. It is not surprising that Cohen (1974) has recently lamented that, in applying speech-act theory, it is very difficult to decide where linguistics ends and where the study of ‘manners’ begins. For, as we shall try to show, many of the conventions that underlie the use of language are learned prior to the onset of articulate phonetic speech. Silverstein (1975), proposing to extend the tradition of pragmatics in the line from Peirce to Jakobson and beyond, puts the matter as follows: To say of social behavior that it is meaningful implies necessarily that it is communicative, that is, that the behavior is a complex of signs (sign vehicles) that signal or stand for something in some respect. Such behavioral signs are significant to someone, participants in a communicative event, and such behavior is purposive, that is goal-oriented in the sense of accomplishing (or failing to accomplish) certain ends of communication... In general, then, we can say that people are constituted as a society with a certain culture to the extent that they share the same means of social communication. He then goes on to point to various of the properties of communicative events - notably the nature of the relationships that prevail between communicator and recipient. These relationships, interchangeability of roles being one of the most obvious, are highly structured by some subtle mix of human endowment and cultural convention. Silverstein is not concerned directly with the ontogenesis of these role relations, but they lie at the

From communication to language - A psychological perspective

261

base of the concern of much of what follows in this paper. These are the functioning communicative acts that give shape to the infant’s discourse with adults in his immediate environment: Referencer and recipient, demander and complier, seeker and finder, task-initiator and accomplice, actor and prohibitor, etc. A close analysis of the first year of an infant’s life provides not only a catalogue of the joint ‘formats’ (see below) in which communicator and recipient habitually find each other, but also provides a vivid record of how roles developed in such formats become conventionalized. The infant is not only learning, as we shall see, what constitutes indicating something to another, or having something indicated to him, but he is also learning how to substitute new means for doing so in order to achieve less uncertain outcomes by the use of more,ritualized techniques. When, finally, he reaches a stage at which lexical indicating is psychologically within his reach, he already knows a great deal about the nature of indicative contexts and conventions for dealing with them. It is in this second sense that continuity becomes of special importance. For if the child, say, already knows (as we shall see) many of the conventions for give-and-take exchanges and how to conduct them by appropriate non-linguistic signalling, he is equipped better to interpret or “crack the code” of linguistic utterances used as regulators of such exchanges. We too readily overlook the fact, perhaps in celebration of the undoubted generativeness of language, that speech makes its ontogenetic progress in highly familiar contexts that have already been well conventionalized by the infant and his mother (or other caretaker). In this sense, it is not extravagant to say that initial language at least has a pragmatic base structure. Let one matter, finally, be made abundantly clear. The point of view that has been set forth in this introductory section is in no sense to be interpreted as a rejection of the role of innate predispositions in the acquisition of language. In the opening paragraphs of this section I commented in passing that there is nothing in Chomsky’s writings that would in any sense deny the role of prelinguistic precursors of prerequisites in aiding the acquisition of language. Indeed, it would be absurd to imagine that the Chomskean Language Acquisition Device could operate without considerable pre-tuning achieved during the period that precedes the use of articulate phonetic grammatical speech. Chomsky comments (1965, p. 58): “The real problem is that of developing a hypothesis about initial structure that is sufficiently rich to account for acquisition of language...“. Surely, part of that richness is the representation built up of communicative requirements established over the long period of interaction between infant and caretaker. I have argued in this section, and will develop the argument further in what follows, that these representations help the child crack the linguistic code. As I have stated elsewhere (Bruner, 1972), there is a long evolutionary history that has shaped human immaturity and many of the elaborated forms of mother--infant interdependence are sufficiently invariant in our species to make inescapable the conclusion that they are in some crucial measure based on innate predispositions, however much these predispositions require priming by experience. What other forms of innateness must be present for the child to

262

J-. S. Bruner

acquire language proper - its grammar and phonology particularly - I cannot debate, though it is worthwhile pointing out that until we discern more clearly the contribution of prelinguistic concepts it is premature to conclude that innate or even acquired ideas about grammar are all that operative. Grammar may itself be a product of the evolution of joint action in the species and one does well therefore to examine how the human ontogenesis of joint action contributes to the mastery of that grammar. The present paper is an attempt to throw some light on some of the persistent problems, that are encountered in the study of the transition from prespeech communication to early language. There are, as this introduction hopefully makes plain, many such problems. I have chosen four of them as illustrative. In a concluding section I shall try to formulate a general conclusion about the role of the three branches of semiotics in such work: Syntax, semantics, and pragmatics (Morris, 1938). The four topics are: (1) The inference of communicative intent, (2) the nature of early reference, (3) the use of language in the regulation of joint action, and (4) the precursors of predication.

Communicative

intentions

Communication, as Silverstein (q.v.) has already noted, presupposes intent or purpose in communicating in the sense that a communication succeeds or fails in its objective. Grammarians usually take intent for granted but one does so at one’s peril. To underline the intentionality of language, linguistic philosophers like Austin (1962), Grice (1968), and Searle (1969) have been particularly insistent on drawing the distinction between the performative or illocutionary functions of utterances, judged both by their conventional felicity and their efficacy in achieving desired results, and the locutionary function, to be judged against such criteria as well-formedness or truth value. But intent in communication is difficult to deal with for a variety of reasons, not the least demanding of which is the morass into which it leads when one tries to establish whether something was really, or consciously intended. Does a prelinguistic infant consciously intend to signal his displeasure or express his delight? To obviate such difficulties, it has become customary to speak of the functions that communication or language serve and to aetermine how they do so. This has the virtue, at least, of postponing ultimate questions about ‘reality’ and ‘consciousness’ in the hope that they may become more manageable. Jakobson (1960) proposes an analysis of language functions based upon the familiar ‘information’ diagram, functions being noted by numbers. Function 1 is expressive and is made up usually of accompaniments to the addresser’s feelings. In a primitive sense, its ‘success’ or ‘failure’ depends upon innate or early learned recognition routines. In time, the form and the recognition of expressions of state become increasingly conventionalized. Function 2 is poetic and involves modes of structuring messages to achieve the illuminative or exhibitive effects of an art form. Again, it comes increasingly to use

From communication to language - A psychological perspective

263

Referent

Addressee

Cdde

the conventions and devices of a language community (Combrich, 1975). Function 3 is conative and is concerned with forming messages in such a way as to produce desired behavior in the addressee. It encompasses the philosopher’s illocutionary force. Function 4 is phatic and relates to the maintenance of a channel of communication between addresser and addressee. Its conduct too may be governed by standard procedures as represented by permissible pause lengths, etc. Function 5 is metalinguistic and it serves to explicate, usually by reference to a code, e.g., Why do you call it metalinguistic? Oh, because it is talk about talk itself. Function 6 is referential and its use is to make clear the referent of a message by clearing up the context for interpreting an utterance. In the stringent terms of the philosophers of language, we may say that “If a speaker refers to an object, then he identifies or is able on demand to identify that object for the hearer apart from all other objects” (Searle, 1969, p. 79). But it is usually much sloppier than that in practice, viz., What did you mean, in front of the house? It’s right in front of the house, by the wall. Any linguistic community has, as noted, conventions for dealing with the functions of language. So do sub-communities. Scientists in communication follow conventions of appearing to ‘avoid’ conative, poetic and expressive functions by the use of meticulous declaratives, passive voice, and words of compact rather than diffuse associative value, etc. The sociologist Garfinkel (1963) notes that in virtually all communities, excessive request for metalinguistic clarification in ordinary discourse is often taken as a sign of hostility or disbelief in one’s interlocutor. To be felicitous requires learning a great many such conventions and rituals. To characterize these conventions Grice (in press) invokes conversational postulates that govern discourse, from which rather loose-fitting maxims are derived - maxims of relevance, of quantity, of quality. Speakers in conversation are expected to stick to the point, to give neither too little nor too much information about context, to speak the

264

J. S. Bruner

truth as they see it. When they depart from these maxims, it is expected that they will do so in a patterned way, with specific intent -- irony, humor, or some effort at manipulation. The pre-linguistic child is probably not much under the sway of such maxims. The postulates governing their communication cannot be taken for granted. But we as their tutors in communication very soon learn their speech proclivities and very early try to shape them to those of the adult community. Unfortunately, there are no studies that have investigated the ways in which this is done, although work on social class differences seems to be making a start (Bernstein, 1960; Hess & Shipman, 1965; Schoggen & Schoggen, 1971; Howe, 1975). Generally (and often unconsciously) adults impute communicative intent to the utterances of infants and children - intent with respect to all the Jakobsonian functions. Indeed, Macfarlane (pers. comm.), in studies of birth ‘greeting behavior’, finds mothers irresistably imputing intent to the cries, gestures, expressions, and postures of newborns. And there is often a strikingly moralistic approach to the imputations. Infants are seen to be showing off, to be asking more than their share, to be ‘buttering up’ mother, to be ‘going on too much about it’. Let us postpone for a bit-the question whether these inferences about intent are ‘correct’ or even ‘consistent’. Joanna Ryan (1974) puts the issue of a child’s communicative intent and its interpretation by an adult in a useful light. She notes “that much of what a child utters in the early stages is difficult to understand, if not unintelligible,” though the “child’s speech and other vocalizations take place within a context of interaction with adults who are motivated to understand the child’s utterances” (p. 199). She continues: “Many young children experience extensive verbal interchanges with their mothers. During these the mother actively picks up, interprets, comments upon, extends, repeats, and sometimes misinterprets what the child has said,” a point which our own observations would certainly confirm as characteristic of even the three-month-old and his mother. Ryan properly complains that the grammarian’s emphasis on well-formedness and semantic sense obscures the role of these interpreted exchanges in preparing the child for language use. Not only do mothers interpret the child’s gestures and vocalizations in conative terms - what he wants - but also in terms of Grice-like maxims like ‘sincerity’ (“He’s really faking when he makes that sound”) and ‘consistency’ (“Won’t you please make up your mind what you want”). Our own observations during the first year of life point to the importance of the creation of what Garvey (1974) has called ‘formats’, habitual exchanges that provide a basis for interpreting concretely the intent of the communications of the child and of the mother. We shall have more to say in a later section about the nature of such formats and their transformations. It suffices to note here that they serve not only to concretize but to socialize and give pattern to the child’s communicative intentions as well as providing the adult with a basis for interpreting them. There is, of course, a great deal of variation in the attitudes of mothers toward their children’s communicative intent, variation that produces considerable disparity in the manner in which mothers interact and talk with their young children. Howe (1975)

From communication

to language - A psychological perspective

265

has shown the extent to which middle-class mothers conceive their role toward their infants as being more “instructive”, not only responding more to their infants efforts to ,vocalize by speaking in return, but also attempting more often to initiate exchanges. The working-class mothers in her study were more often laissez-faire in their approach. By the time, then, that he is two-and-a-half or three years old, the middle-class child is on average not more competent to handle more advanced forms of utterances - propositions of state and of action - but in fact is using such advanced forms more often. For their mothers continued to interpret their child’s utterances as having to do with state and action (in contrast to propositions of naming only) until the child explicitly replied with such propositional forms. The mother’s interpretation of the child’s communicative intent is what seemed to keep the verbal interaction going and it keeps going until the child conforms or the mother gives up. While Howe’s data begin at 18 months, when holophrases and early two-word utterances were appearing, the same principle can be shown to govern the mother’s persistence even when the criterion the mother is applying relates to pre-linguistic communicative exchanges. In interpreting the infant’s communicative intent - correctly or incorrectly - the mother has a rich variety of cues to use. So too the child, for if the mother is at all consistent, he gives forth cues that come increasingly to have a predictable consequence as far as her behavior is concerned. In this sense, they are in a transactional situation; their joint behavior determining its own future course. Ryan (1974) adapts a classification of the cues used by the mother prepared originally by John Austin (1962) for the analysis of performative aspects of speech. (1) Aspecrs of the utterance itself including intonation patterns that suggest insistence, pleasure, protest, request, etc. As Ryan puts it, “what is important is that adults interpret children’s use of intonation in a systematic way, thus allowing children to learn what is conventional usage”. Wolff (1969) was one of the first to indicate that the early cries of infants were interpretable by parents. Ricks (1971) has shown even more convincingly that cries of normal babies obtained under controlled conditions (expressing greeting, pleased surprise, request, and frustration) were correctly categorized ~~ although the cries of the parent’s own children included in the sample were not reliably identifiable. And Dore (1975) has suggested that intonation contours may be the first carriers of primitive illocutionary force in the child’s utterances. (2) Accompaniments of the utterance provide a second set of cues for interpretation ‘pointing, searching, playing with specific objects, refusing’. These are evident enough and need no comment here. We shall meet them again in a later section. (3) And finally, circumstance of the utterance constitutes the third source of cues, the context of the communicative event. Families of the children in our present study at Oxford typically classify their infants’ vocalisations by context; babbling contentedly in his cot on first waking up, calling for attention on waking from the afternoon nap, hunger-fretting before feed time, annoyance at not being able to reach an object, etc. For what it is worth, we have also found distinctive voicing patterns in these calls as early as four months, suggesting that it may not be context alone that is being used as a cue.

266

J. S. Bruner

When, then, does infant come to ‘intend consciously’ to communicate? Early students of prelinguistic communication were given to classifying pre-speech utterances of children into expressive (early cries of discomfort and pleasure), stimulative (producing reactions in others) and representational (Btihler, 1934). The process of going from expressive to stimulative was conceived much as Piaget’s (1952) secondary circular reaction for producing or prolonging a desirable state of affairs previously produced inadvertently. Was intent involved in going from the expressive to the stimulative, and could one tell that the trip had been made? That debate does not seem worth a repeat performance, for we surely have no better basis for deciding than did our forbears. Rather, I think we would do better to concentrate instead on the description of particular intention-imputing situations and their outcomes to determine the child’s and the mother’s course in learning to deal with Jakobson’s communicative functions. How indeed do the child-and-mother cope with the joint reference requirements in communicating? How is the phatic link maintained? How do the child and mother handle misunderstandings and their disambiguation? How do the child and mother express and recognize states of feeling? Is there an early poetic function and what well-turned babbles are rewarded by smiles? Can one discern a systematic trend in the conative devices a child uses to produce desired behaviour in his listeners? If only for methodological reasons, I would propose that we avoid a priori arguments about ‘conscious intent’ and ‘when’ it is born. For questions whose answers are not in principle recognizable are rarely useful, and it is likely that ‘consciousness’ and ‘intention’ are opaque in this way. The issue, rather, is how communicative functions are shaped and .how they are fulfilled. In fact, when one examines the development of specific communicative functions, the issue of conscious intent and its dating seems to wither away. An example is provided by one of our own subjects, Jon A., and the development of a signal pattern involving reaching outward bimanually while in a sitting position, hands prone. It had usually been interpreted by the mother as a signal that Jon wanted some familiar, hand-sized object beyond his reach’s terminus, and she generally provided him with it, often heightening his anticipation by advancing the object slowly or ‘dramatically’ toward his hand with an accompanying rising voice pitch. At eight months, one week, Jon used the signal; M interpreted it as calling for her hand, since there was no object close by, and performed her ‘walking hand’ body-game format, with the fingers walking up Jon’s front to his chin. He tolerated it, though not entering as exuberantly as usual. That over, Jon then reached out again. M interpreted it as request for repetition. He participated even more reluctantly. M, on completion, then repeated the game though Jon had not signalled. He averted his gaze and whimpered a little. She repeated again and he was totally turned off. Pause. Then, 27 seconds after Jon first reached out, he reached again, this time pulling M’s hands to a position where he could take hold of the ulnar edges and raise himself to a standing position. There was a following sequence of 14 episodes extending for slightly over 9 minutes in which M and Jon played a game of alternating irregularly between the two ‘formats’ - M’s hand either walking on fingers to tickle position, or

From communication to language - A psychological perspective

267

M’s hands in stand-supporting position. Under /V’s control, it was made into a ‘surprise’ evoking, alternating format, with her alternative interpretations of his reach gestures being rendered explicit. In the course of such exchanges, as Ryan (1974) has already noted, “The child is developing skills that are at least as essential to speaking and understanding language as the mastery of grammar is supposed to be”. Much of that learning is based upon the mother interpreting the child’s intent, the child sometimes conforming with the interpretation, sometimes not, but learning, en route, what interpretations his efforts evoke and how these may be modified. We shall return to these issues in a more general way in the concluding section. Here it suffices to say, with Dore (1974), that any theoretical framework for understanding language development requires a consideration of pragmatics, and a theory of pragmatics must have some way of coping with the communicative intentions of speakers. Dore proposes that communicative intent be defined as the inducement in a listener of the speaker’s expectation. In this section, we have looked at intent as being realized in a transactional situation, with mother providing an interpretation to which the infant ‘speaker’ can conform, dissent, or which he can attempt to modify by correction or persistence. In the following sections we shall deal with more specific communicative intents - referring, regulating joint action and predicating.

Reference The issues raised in traditional philosophical discussions of reference have often been introduced into the debate by invoking the example of a hypothetical infant learning that a given sound, word, or gesture “stands for” something in the extralinguistic environment. Though such exercises are logically stimulating - else they would not have continued over the centuries - they are, alas, principally empty or irrelevant in.explicating psychologically the infant’s real problem in mastering reference. I find myself strongly in agreement with Harrison’s (1972) contention that the psychological (and even arguably the philosophical) problem of reference is how the child develops a set of procedures for constntcting a very limited taxonomy to deal with a limited set of extralinguistic objects with which he traffics jointly with adult members of the linguistic community. What adults do for the child is to teach him or help him to realize how these taxonomic procedures operate in assuring joint reference in relatively well established situations until, finally, the child can go on quite on his own in coping referentially with larger arrays of objects in novel situations. The procedures of reference, I believe, are generative. The issue is how to differentiate among a set of objects, and how to refer precisely to any single one. I am quite prepared to accept Wittgenstein’s (1953) demolition of empiricistassociationist theories of naming based on pointing or other forms of ostension on his grounds: That ostension, even with negative feedback, can never specify what it is that

268

J. S. Bruner

a sign refers to in the complex welter of properties that any object necessarily displays. The negative feedback, moreover, is rarely in evidence in the data on language acquisition and when it does occur (e.g., Nelson, 1973), it is usually followed by the child abandoning his effort to use a name to indicate an object. Moreover, associative theories of naming or reference are beleaguered by the presupposition that uttering a sound or making a gesture in the presence of a referent somehow evokes a nascent or innate recognition in the child that the name is associated with some feature of something that is at the focus of the child’s attention, so that any concatenation of sign and referent is as likely as any other to be learned, and that is plainly not so. Whatever the reference

triangle is (Ogden 8~ Richards, 1923), it is plainly not an isolated bit of mental furniture produced by the linking of a sign, a thought, and a referent. The objective of early reference, rather, is to indicate to another by some reliable means, which among an alternative set of things or states or actions is relevant to the child’s line of endeavor. Exactitude is initially a minor issue. ‘Efficacy of singling out’ is the crucial objective, and the procedures employed are initially quite independent of the particular nature of objects and their defining or essential properties. If what I have boldly asserted here is even arguably so (and for a more carefully reasoned presentation of the same argument, the reader is referred to Harrison’s 1972 discussion), then we would do well to avoid falling into the classical empiricist trap of the theory of naming or referring (even Quine’s (1960) seductively common sense version of it in Word and Object) and look instead at the procedures earliest used by the infant and adult in indicating and differentiating the very limited set of objects with which they traffic. I shall want to deal with three separate aspects of early reference, and for convenience I shall give them labels. The first we may call indicating (if only to avoid the term ostension!), and it refers to gestural, postural and idiosyncratic vocal procedures for bringing a partner’s attention to an object or action or state. The second is deixis and refers, of course, to the use of spatial, temporal, and interpersonal contextual features of situations as aids in the management of joint reference. The third involves the development of standard lexical items that ‘stand for’ extralinguistic events in the shared world of infant and caretaker and I shall call the process naming. Our task, as already indicated, is to explore the procedures employed in all three of these considerable linguistic accomplishments. Take indicating first. Studies by Collis and Schaffer (in press), by Kaye (1976) and by Scaife and Bruner (1975) all point to a highly primitive form of indicating early in the child’s first year. Collis and Schaffer have shown the extent to which the mother’s line of regard follows the infant’s, she constantly monitoring and following where the child looks as an important feature of inferring what is at the focus of his attention - better to interpret his demands, to elaborate upon what he is attending to, etc. Kaye has shown the extent to which mothers, asked to teach their child a simple task of taking an object from behind a transparent barrier, actively enlist the child’s attention by ‘marking’ the target object in various ways - touching it, shaking it, etc. Strikingly, such manoeuvres occur

From communication to language - A psychological perspective

269

far in excess of chance expectancy when the infant looks away from the task. In sum, she follows his line of regard, and when it diverges from where the. task requires it to be, she uses effective procedures of indicating to reestablish joint attention. The findings of Scaife and firuner provide the final piece in this picture. Not only, as indicated by Collis and Schaffer (in press), does the mother follow the child’s line of regard as an indicator, but in this experiment the infant seems able as early as four months to follow and increasingly does follow an adult’s line of regard when it is turned toward a locus removed from the child. We do not yet know what the processes are that bring this accomplishment about, but there are some tantalizing indications of the kinds of factors that are involved and that will have to be unravelled by experiment and close observation. What Scaife and Bruner (1975) have found is that as early as four months in some children and with high frequency by nine months, the infant turns his regard in the same direction as a facing experimenter turns his. To what extent imitation is initially involved is difficult to say, but it can be said that there is no confusion among their young subjects as to which way to turn, though imitation might lead to head turning in either direction. Work is continuing at Oxford on these tangled issues by Scaife and by Churcher, and hopefully a clearer indication of the origins of thus behavior will emerge. Qualitative analysis of the responses of Scaife and Bruner’s babies seem to suggest that the head turns of the infants are not of a magnitude to match imitatively the degree of head turn by the adult. Yet, it is quite possible that initial imitative turning might lead the child to ‘pick up’ an interesting object and thereby provide a perceptual reinforcement to the child’s head turning. Again, this would very likely depend upon the density and discriminability of targets available to the child who orients in the direction of an adult’s gaze. In any case, what we can say at this early juncture is that there is present from a surprisingly early age a mutual system by which joint selective attention between the infant and his caretaker is assured -- under the control of the caretaker and/or of the child, eventually managed by joint pickup of relevant directional cues that each provides the other. Plainly, such devices for assuring a joint focus are insufficient for indicating what feature of a focus of attention is being abstracted - by the mother or by the child. That, of course, is the shortcoming of all ostensive indicating. But it is far more to the point to note not this shortcoming, which in a practical sense seems trivial in terms of what the child is actually doing, but the nature of the accomplishment (whether.it be innately primed or somehow learned). What has been mastered is a procedure for homing in on the attentional locus of another: Learning where to look in order to be tuned to another’s attention. It is a discovery routine and not a naming procedure. It is totally generative within the limited world inhabited by the infant in the sense that it is not limited to looking at oranges or dolls or rattles. It has also equipped the child with a basis for dealing with space that transcends egocentrism, or, in any case, the child’s egocentrism does not prevent him from following another’s attention. For the child is able to use both a second origin of reference, another’s line of regard, as well as various forms of

210

J. S, Bruner

marking or highlighting of objects (as in the Kaye experiment). These accomplishments would surely qualify as precursors of Piagetian decentration, their earlier occurrence being attributable perhaps to the more personal, less object-orientated testing situations used in Oxford. In this sense, these accomplishments guarantee the first bases for spatial, interpersonal deixis. There is a further procedural accomplishment implied by Kaye’s (1976) study of the ‘implicit pedagogy’ of mothers, their use of ‘marking’ in indicating an object or event to be attended to. Without going into the details, it is plain that mothers of six-month-olds succeed in getting their infants to attend to and capture the object intended, in spite of the barrier. They not only mark the object, but evoke the action either by a process of tempting - putting the object nearer and at the edge of the barrier - or by modelling the behavior themselves. The marking involves a combination of ‘highlighting’ features and exaggerating the structure of acts to be performed. They do both in a manner that is highly contingent on thechild’sstate,his attentional deployment, and his line of activity.* We may now profitably introduce the concept of “natural categories” so interestingly expounded in a recent series of papers by Rosch (1974). She argues that in development, categories of objects are built on the basis of a common sharing of ‘motor programs’ and of those perceptual features required for their execution. In this sense, they are ‘practical’ objects that are marked by features of use and their structuring into equivalence classes is based upon that use. In this sense, jointly managed activities provide an essential contextual basis for parent indicating to child and child to parent. As Nelson (1973) notes, categories of use are the first to develop and the childish definition ‘a hole to dig’ is to be taken as something more than quaint. But what is apparent is that indicating in either direction - child to mother or mother to child - occurs in situations where both are involved jointly in the act of digging or of reaching or of knocking down. The indicating that is used by each is based on the joint knowledge of the course of these actions and initially takes the form of exaggerating or ‘marking’ some phase of the action as a signal (often accompanied by ancillary vocalization and gesture, as in the child indicating a target of reaching by exaggerating the reach toward an object and making an ‘effort’ or ‘fretting’ vocalization, or by the mother shaking or vocally marking by ‘C’mon’ a proffered object). Note again that the procedure is independent of the conventional defining properties of the objects involved and relates instead to the programme of use. Generative procedures for indicating undergo three striking changes over time: Decontextualization, conventionalization, and increased economy. Decontextualization involves the development of indicating strategies that are not so closely linked to the specific action patterns in which they are embedded. Rather than indicating by exaggerating a feature of the reach (like extension) and fretting, the child now uses a more peremptory reach toward the object. This manoeuvre appears to signal more the child’s *For a fuller account and Bruner (in press).

of earlier

implicit

pedagogies

concerned

with such ‘marking’,

see Wood, Ross,

From communication to language - A psychological perspective

211

line of regard and less the next step in his line of activity. %e extended hand becomes an external pointer for noting line of regard rather than direction of activity. It is probably this crucial fact that leads to an increase in economy in indicating. For by eight months, and often earlier in ‘comfortably familiar’ joint action formats, the child holds his hand out toward the object in a non-grasping directional gesture. By a year, when he is presented pictures on the page of a book, he rarely ‘grasps’ at the picture, but touches it, and eventually touches it only with the index finger. With respect to conventionalization, its basis is somewhat problematic. Its origin may be in the phenomenon of visual ‘cross checking’ between mother and infant: Each looking at the other en face while in the process of indicating (present from the start for the mother, but increasingly evident for the child after six months in our own observations as well as in those by Sugarman, 1974, and by Bates et al., 1973). They appear to be seeking agreement on a referent. The term conventionalization may, perhaps, be inappropriately grandiose for such a minimal sign. I use it nonetheless to indicate that mother and child seem increasingly in the second half of the first year to be checking whether their gesturing or marking is ‘getting through’ to the other, as if there were mutual recognition of a correct way to signal. It is after all of this prior learning that holophrastic ‘naming’ comes into the picture. Again, it seems highly unlikely that naming is what in fact is at issue. For as I have argued in a previous paper (1975) and as Bloom (1973) Greenfield and Smith (in press) and others, I think, abundantly illustrated, the child’s holophrases are grammatically contextualized in a Fillmore-like (1968) case form that highlights some aspect of who is doing what with what object toward whom in whose possession and in what location and often by what instrumentality: Agent, Object, Recipient of Action, Location, Possession, Instrument. And it is not surprising, as Eve Clark (1973) has pointed out, that from a sheerly referential point of view, the child’s usage is highly overgeneralized, since he is still grouping objects and actions in terms of function rather than properties - a point well made in Greenfield’s (1973) paper on “Who is Dada?” In this sense, the emphasis is upon a rough taxonomic procedure rather than exactitude of response. Equipped with such useful and generative procedural rules, the child may then get on with the Augustinian business of learning to refer - but in no sense can it be taken as claimed by St. Augustine (cf. Wittgenstein, 1953) as learning language through naming ab initio. For if the child now fails to be able to discern the properties to which orange or rattle or Dada refer, he has an extensive repertory of procedures available for disambiguation - though, to echo Wittgenstein’s critique again, there is no set of ostension procedures that can ever uniquely determine reference or meaning. We may now turn to the issue of deixis and its development. Recall that from the fourth month there is already some basis for spatial deixis in the line-of-regard following of mother and infant and, probably, this implies some appreciation of deixis of person at least the recognition that it is another’s line of regard being followed. Much of the ‘reality of discourse’ depends upon the establishment of what Benveniste (197 1) calls a

272

J. S. Bruner

locution-dependent I/You concept. Adult speech would be impossible without it and Lyons (1974) has argued that reference is dependent for its growth upon it to deal not only with the shifters Z and YOU but with spatial and temporal indicators like here and there, now and later, etc. Benveniste’s point is worth quoting (1971, pp. 217 - 218): Between Z and a noun referring to a lexical notion, there are not only the greatly varying formal differences that the morphological and syntactic structure of particular languages imposes; there are also others that result from the very process of linguistic utterance and which are of a more general and more basic nature. The utterance containing I belongs to that level or type of language which Charles Morris calls pragmatic, which includes, with the signs, those who make use of them. A linguistic text of great length - a scientific treatise, for example - can be imagined in which Z and you would not appear a single time; conversely, it would be difficult to conceive of a short spoken text in which they were not employed. But the other signs of a language are distributed indifferently between these two types of texts. Besides this condition of use, which is itself distinctive, we shall call attention to a fundamental and moreover obvious property of I and you in the referential organization of linguistic signs. Each instance of use of a noun is referred to a fixed and “objective” notion, capable of remaining potential or of being actualized in a particular object and always identical with the mental image it awakens. But the instances of the use of Z do not constitute a class of reference since there is no “object” definable as I to which these instances can refer in identical fashion. Each Z has its own reference and corresponds each time to a unique being who is set up as such. What then is the reality to which Z or you refers? It is solely a “reality of discourse,” and this is a very strange thing. I cannot be defined except in terms of “locution,” not in terms of objects as a nominal sign is. Z signifies “the person who is uttering the present instance of the discourse containing I.” This instance is unique by definition and has validity only in its uniqueness. If I perceive two successive instances of discourse containing I, uttered in the same voice, nothing guarantees to me that one of them is not a reported discourse, a quotation in which I could be imputed to another. It is thus necessary to stress this point: Z can only be identified by the instance of discourse that contains it and by that alone. It has no value except in the instance in which it is produced. But in the same way it is also as an instance of form that Z must be taken; the form of Z has no linguistic existence except in the act of speaking in which it is uttered. There is thus a combined double instance in this process: the instance of Z as referent and the instance of discourse containing Z as the referee. The definition can now be stated precisely as: Zis “the individual who utters the present instance of discourse containing the linguistic instance I.” Consequently, by introducing the situation of “address,” we obtain a symmetrical definition for you as the “individual spoken to in the present instance

From communication to language - A psychological perspective

273

of discourse containing the linguistic instance you. ” These definitions refer to I and you as a category of language and are related to their position in language. We are not considering the specific forms of this category within given languages, and it matters little whether these forms must figure explicitly in the discourse or may remain implicit in it. This constant and necessary reference to the instance of discourse constitutes the feature that unites to Z/you a series of “indicators” which, from their form and their systematic capacity, belong to different classes, some being pronouns, others adverbs, and still others, adverbial locutions. If we can interpret Benveniste as implying psychologically that a grasp of reciprocal roles in discourse is the essential prerequisite for deixis of person, place, and time, then some very interesting questions arise about the development of reference. For one thing, the Benveniste hypothesis places pragmatics - the relation of language to those who are speaking it - at the heart of the problem of reference. Again, the beginnings of a locution-dependent reciprocal concept emerges in action well before it is ever used in formal language. Established and reversible role relationships obviously provide a primitive base for later linguistic deixis. The universal prelinguistic game of ‘Peekaboo’ is a striking example (Bruner and Sherwood, in press; Greenfield, 1972) of such reversible role structures, bound as it is by rule constraints with respect to who is the recipient and who the agent of coverings and uncoverings and how these may be reversed. Give-and-take routines, early established between mother and infant, again with reversibility of roles, and often marked by distinctive vocalizations for marking the giving and the receipt of an object (Bruner, 1975) provide another example. In such games, once developed, the child looks mother directly in the eye for a signal at crucial pauses in the play, as if to calibrate his intended actions with hers and to check which one is playing which role. In the first year of life, then, the child is mastering a convention - checking procedure not unlike that of adults - indeed, even using eye-to-eye contact for determining intent, readiness, and whose ‘turn’ it is (Argyle & Ingham, 1972). But there is a big step from ‘behavioral’ to ‘linguistic’ deixis. In the latter, the context is contained in the message, in the former in the behavioral field of the speakers. Are there any small steps that help the child scale the heights from extralinguistic to intralinguistic deixis? Perhaps one such step is through early phonological marking. One of the children we are studying showed at six months a difference in range of pitch for vocalizations accompanying the manipulation of objects in hand and those accompanying his interactions with his mother. Vocal ‘comments’ or babblings while manipulating objects were higher pitched than those accompanying exchanges with the mother. The mother was observed to look back to the child when the pitch of his vocalization dropped, to check whether he had redirected his attention toward her, though pitch difference did not develop as a systematic calling device to which the child and mother had recourse. The

274

J. S. Bruner

same child used a sharper onset of voicing when reaching toward out-of-reach objects than when taking one in hand that was within easy reach, a distinction akin to Lyons’ (1974) second-order deictic marker, proximal/non-proximal. When the child did use sharper onset voicing the mother responded by moving an out-of-reach object toward him. It may well be, to be sure, that the sharper onset was an accompaniment of the effort of reaching for the more distant object - and in this sense be ‘expressive’ - but the fact remains that the distinction provided a vocal cue to the mother as to topic of the child’s attention. We cannot know whether the child ‘deliberately’ used the distinction for signalling purposes, but the mother responded as if it were deliberate and he continued to use such signalling in appropriate situations. In one other instance with the same child, again at six months, he was observed to use a distinctive vocalization that soon was able to produce a particular act by the mother. The situation was the familiar one in which mother ‘looms’ an object toward the child, the termination of the looming being to touch the object to the child’s chest or hand or forehead. It is a standard body game play ‘format’ for this child-mother pair. The child responded to looming with a pharyageal fricative, shifting to velar stop, and terminating with the achromatic vowel, aah. When the mother delayed looming, the child used this call. If mother delayed too long, the vocalization shifted to a fretting call. This signalling was observed over two observational sessions separated by three weeks. Kaplan and Kaplan (197 1) have set out a plausible case for such phonological marking as a beginning of a semantic referential system that may precede or operate independently of syntax. Cromer’s (1974) admirably concise summary of the Kaplan’s position will serve: (They) propose that the child’s semantic position develops out of the early distinctions present in his communication system. They feel that with adequate data one will be able to identify a set of semantic features and chart the developmental order of their emergence. For example, when the infant makes the early distinction between human and non-human sounds, the Kaplans suggest the feature ‘f human’ has become operative. When the infant differentiates himself from others, as observable in the effect of delayed auditory feedback on crying (i.e., indicating that the infant can distinguish between his own voice and other sounds), he is credited with the feature ‘f ego’. As the child develops his knowledge of object properties he adds such features as ‘* existence’ and ‘5 presence’. Other later acquisitions would include ‘f agent’, ‘ + past’ and the like. These semantic features would place constraints on the child’s language acquisition. May not such devices provide the beginnings of the idea of vocally marking different positions of play in the relation between mother and child? I am all too aware that such instances do not provide a proper ‘tracing’ of the course from behavioral to linguistic deixis, yet I would urge that an effort be made to examine the small steps that might, in combination, provide the big insight that must be involved in learning to handle the classical deictic ‘shifters’ - those expressions whose interpretation varies as a function of

From communication to language - A psychological perspective

2’75

which member of a pair uttered them, ranging from you and me to in front of and behind. We shall pursue this matter further in the following section. With respect to naming proper, finally, there is ample evidence that well before language, the idea of the word or label as an instrument of reference becomes firmly fixed. Indeed, Nelson’s (1973) study of language acquisition indicates that one of the two ‘styles’ of language acquisition is referential - exercises in labelling being at the center of certain mother-infant verbal interactions. (The other style, expressive, will be considered in the following section.) It takes the form, prior to word production, of playing: ‘Where’s your nose’, or ‘Show me your eyes’. Thus the concept of a label must be a very early feature of language competence. Indeed, it too may have a deictic element as evidenced by the very common phenomenon of labelling both the infant’s mouth, eyes, nose, etc., and the infant then indicating the mother’s corresponding parts. How early a start the lexical concept or label may get is indicated in some recent work by Ricks (1971). He distinguishes at eleven to eighteen months three classes of vocalization: Babble sounds with no evident referent, ‘dada’ words with a loose referent, and ‘label’ words. He lists seven properties of the last of these (like ‘bow-wow’): They are not found in babbling, they are used only in the presence of a particular object or event, they are not modified toward conventionality but rather are adopted by the parents, they are frequently generalized and over-generalized (Clark, 1973), the expression of the ‘word’ is often accompanied by excitement, mention of the label word alerts the child to searching and also leads the child to repetition of the word. Ricks’ data start at eleven months. By eighteen or twenty months, some children have even introduced a word that stands for ‘label-lacking’ objects as with Matthew in Greenfield, et al., (1972) who uses ‘Umh’ for unknowns and Bloom’s (1970) Allison whose ‘wida’ is even more ambiguous. We know very little about the onset of labelling as an instrument of reference. It seems highly likely that, at least later and possibly earlier as well, it is related to IQ, for after three or four, the single best indicator of a child’s measurable intelligence is the size of his vocabulary (Raven, 1948). Surely, if we are to understand the origins of and the later elaboration of reference, we shall have to explore more fully the kinds of phenomena reported by Ricks as they begin to manifest themselves in the first year of life. They may be a natural outgrowth of the phonological labels mentioned earlier or may indeed be an elaborated and later accompaniment to mutual pointing and joint gaze direction. All of these phenomena point to the early existence of means for managing joint reference. Yet none of them move very far along the line toward discourse-sensitive, deictically dependent reference of the kind so carefully described by Benveniste. We turn next to a form of development that may explicate the early phases of such reference.

Language and joint action As we have already noted, emphasis upon linguistic competence can easily distort the study of acquisition toward or preoccupation with syntax. Joanna Ryan’s critique (1974,

276

J. S. Bruner

p. 185) is doubtless correct: “Recent psycholinguistic work has neglected the earliest, presyntactic stages of language development, concentrating exclusively on the details of the child’s later mastery of grammar. This approach can be characterized as exclusively cognitive, in the sense that it regards language as something to be studied as the object of the child’s knowledge, and ignores all the other skills that determine actual language use. This neglect of what has come to be known as ‘communicative competence’ (Campbell & Wales, 1970) is not only serious in itself, but has also led to a distorted view of the child’s grammatical abilities.” Perhaps the best antidote to syntactic preoccupation is to examine closely how the infant masters the task of communicating to others his needs, wishes, and objectives in order to assure either assistance or joint action. It is this that constitutes the beginnings of the more elaborated speech acts that are developed to ‘get things done with words’. Rejecting as incomplete Chomsky’s definition of the task of linguistics as the specification of rules that relate sound and meanings, Searle (1975) comments: “I don’t think that his picture is false, so much as it is extremely misleading and misleading in ways which have unfortunate consequences for research. A more accurate picture seems to me this. The purpose of language is communication. The unit of human communication in language is the speech act, of the type called illocutionary act. The problem (or at least an important problem) of the theory of language is to describe how we get from the sounds to the illocutionary acts. What, so to speak, has to be added to the noises that come out of my mouth in order that their production should be a performance of the act of asking a question, or making a statement, or giving an order, etc.?” To the beginning of this process we turn now. From the start, the child is well equipped with communicative routines in what we shall call the demand mode, many derived from innate patterns of expressing discomfort. By the third or fourth month of their baby’s life, most mothers claim to be able to distinguish forms of satisfaction expressed by vocalization. The demand cries almost always include pain or physical discomfort, hunger, demand for social interaction, and fatiguefrustration. Whether these cries have a universal phonological pattern or are idiosyncratic is not clear. Alan Leslie and Christopher Pratt at Oxford are currently carrying out analyses of changes that occur in such cries recorded in the child’s familiar home setting. ‘Pleasure’ vocalizations usually include ‘chatting’ upon awakening and then playing by oneself and also the gurgling accompanying ‘happy’ interactions with a familiar caretaker. We can say little about these as yet save that they are recognizable to the mother and do not sound ‘troubled’ to the naive listener. Characteristically, ‘trouble’ demand cries, on the other hand, are insistent, with no pauses in anticipation of response, are ‘wide spectrum’ in their distribution of energy across a range of audible frequencies, and if unattended are followed by uncontrolled scream-crying. In practice, they are usually responded to, with the effect of establishing an expectancy of response. When such expectancy is established, at least three changes occur, marking the beginning of what I call a request mode. One major change is moderation in the wide-band intensity and

From communication to language - A psychological perspective

2’77

‘insistence’ of initial calling. Its wide band spectrum is reduced. A second change is condensation of the call to a more limited time span with a pause in anticipation of response. If the response is not forthcoming, the infant reverts to the demand mode. There also ensues an increasing ‘stylization’ of initial calling with each infant developing a more recognizable ‘signature’ call. Studies of crying and fretting (e.g., Sander et al., 1970; Ainsworth, 1975) point to the important role of a consistent caretaker in effecting this transition from demand to request. There next appears a distinctive exchange mode. It begins with indicating a demand for an object gesturally and often with vocal accompaniment. By 8-l 0 months, the child not only calls for and receives an object, but hands it back, calling again, receiving it, and handing it back again. As noted in the discussion of deixis, he reverses roles with himself first as recipient of action, then as agent. Indeed, exchange may make an even earlier appearance at the gestural level for as early as 2 weeks of age, an infant will imitate facial and manual gestures (Moore and Meltzoff, 1975). If the child’s imitation of the mother is then imitated in turn by the mother, the rate of the child’s responding with matching gesture can be raised (Rheingold, Gewirtz, and Ross, 1959). Meltzoff (pers. comm.) has preliminary results indicating, moreover, that if the mother responds to the child’s imitative gesture with a non-matching one, the child will either start imitating that gesture or will stop and may show distress or gaze aversion. It is difficult to say whether this early gestural exchange has any role in supporting the later exchange patterns first described, for the later pattern involves tasks with objects rather than directly interpersonal ones. The exchange mode is gradually transformed into what we may call a reciprocal mode. Interactions are now organized around a task that possesses exteriotity, constraint, and division of labour. The two participants enter upon a task with reciprocal, though nonidentical, roles. It may involve nothing more complex than the mother holding steady the ever-present toy pillar box into which different shapes can be inserted. Elsewhere (Wood, Ross, and Bruner, in press) we have referred to this as ‘scaffolding’ activity by the mother on the child’s behalf. In time, and with the development of anticipatory schemas, the child’s conception of the task is elaborated. He may now hold up a form to his mother’s view before placing it in the pillar box. The mother may hand him one to put in. He may hand her one to insert. Much intermittent eye-to-eye checking and vocalizing accompanies these variations. The task is gradually being structured into reciprocating roles, the roles defined into rounds, each composed of turns, often with turns interchangeable (Garvey, 1974). The task and its constituents have become the objects of joint attention. Betimes, more complex tasks emerge and task formats are combined, with a strong quality of play and pleasure. But however playful, the striking thing about such task formats is that they are rule bound and constraining. The progress from demand to request to exchange to reciprocity during the first year is, I believe, of central importance to the development of speech acts (or, more properly, communicative acts) and, as well, to the establishment of a ground work for the later

218

J. S. Bruner

grasping of case in language. An unexpected source of information about the elaboration of communicative acts comes from Ainsworth and Bell’s study of mother-infant pairs (1974). They report that as ‘crying and fretting’ - what we could call the demand mode - recedes, more subtle fomrs of communicating increase, and those children who persist in the demand mode are slow in developing communicatively. More directly relevant are contextualized observations of my own (Bruner, 1975) and of Edwards (1975). The first example is from the Oxford corpus: Ann had learned between 8 and 10 months to play a well modulated exchange game involving the handing back and forth of objects. When, at 13 months, the game was well organized, Ann picked up her mother’s receiving Thnnk you. She used it both when giving and when receiving an object. After two weeks the expression dropped out of the giving position, nothing at first taking its place, but remained in the receiving position. Meanwhile, the demand demonstrative Look was appearing in Ann’s lexicon, used in referential situations, as when looking at pictures in a book. At the end of the thirteenth month, Look was transposed as well into the position at which Ann handed her mother an object. Look was later replaced by There in the giving-taking format. 1 would interpret Ann’s performance in terms of Searle’s (1975) earlier quoted comment on how we go “from sound to illocutionary force”. Initially, she accompanies both roles in the exchange format by a single expression, in recognition of the compact of exchange. In time, each role is appropriately and conventionally accompanied by differentiated vocalizations, one of them borrowed from a demonstrative speech act (Look), but shortly replaced by one more appropriate to the act of bestowal. One is reminded of Cohen’s (1974) discussion of ‘markers’ used for characterizing such semantic force properties as imperative, optative, interrogative, exclamatory, or performative. He goes on to raise the issue of whether there might not also be markers used for differentiating the uses to which performatives are put: Indicating explicirness/implicitness, noting intention/inadvertence, whether a verdict, commitment, or promise is being made, etc. One has the impression in examining protocols that something of this order is occurring, and in the present case, the final shift on the ‘giving’ end from Look to There is a subtle recognition by Ann that exchange and demonstration involve different accompanying performatives. The second example illustrates a somewhat different point: That a joint-action format provides, as already noted, an opportunity for the child to master major elements of case grammar as it relates to specific and familiar action formats. Edwards (1975) shows how a child’s knowledge of a ‘prohibitive’ format - having to do with objects she was not to touch - provided an opportunity for her to develop grammatical concepts. Initially, she used the simple negative No to characterize situations in which she was not to touch an object. This was followed, in comparable formats, by the introduction of possession, Yours or Mine for objects she was forbidden to touch or play with. Still later, a verb form was inserted into the format - Leave it for what was prohibited. And finally, again in the same format type, the adjectival form appeared: Hot or sharp for the object

From communication

to language ~ A psychological perspective

279

that was prohibited. Edwards’ Alice was learning not only grammar, but learning it as an adjunct to social situations whose structure she was having to learn and to manage. The case variants were all embodiments of a self-directed imperative - to keep clear of the object in question. We turn next to predication and its prelinguistic precursors and prerequisites. Predication Looked at in its linguistic sense, “predication involves affirming or asserting something of or about the subject of a proposition’. (Wall, 1974, p. 9). It might seem then somewhat jejune to inquire about the precursors or even the prerequisites for predication in the prelinguistic child, for surely he can in no sense be thought to be dealing in propositions. What has made the issue of pre-propositional predication a persistent and interesting one, however, was the early insistence of DeLaguna (1927) that the single words of holophrastic speech could profitably be treated as compacted sevtence forms, and that single words could be conceived within that framework as comments upon extra-linguistic topics inferent in the contexts in which the child found himself. The primitive topic, then, was implicit rather than explicit. This interpretation of early one-word utterances persisted in the literature - cited principally in general reviews of language development _ until picked up again in McNeill’s (1970a, 1970b) work, and then further developed by Bloom (1973) and more recently in a rather widely circulated manuscript by Greenfield and Smith (in press) who were specifically concerned with interpreting the run-up in early speech development from single word utterances to the patterns that are found when M.L.U.‘s approach two morphs. Like DeLaguna, these investigators were interested in the manner in which the ‘unmentioned’ topic finally found its way into explicitness to be represented by a nominal or other grammatically interpretable form that could carry language development beyond dependence on unspecified context DeLaguna’s famous claim that language development could be conceived as a process of decontextualization.* McNeil1 (1970a, 1970b) carried the argument one step further: Arguing from the existence of the predicational postpositions in Japanese, ga and wa, indicating respectively extrinsic and intrinsic predicates, the latter being habitual or essential and the former temporary or transitory, he proposed that initial predication with unmentioned topic could be conceived of as the intrinsic form, while extrinsic predication was more of the ga type. In Japanese, the postpositions could be noted by the contrast: ‘The dog-wa *With respect to the issue of the sentential or predicational status of the holophrase - still a very lively theoretical issue - the reader is referred to Bloom (1973), whose conclusions will be taken up in the final section, and Dore (1975) who reviews the arguments for and against sentential status for the holophrase and ends by opting for the view that holophrases represent a first step along the hard path from primitive force to grammaticalized illocutionary force, a path one traverses by successive mastery of grammaticalizing devices such as using word order, intonation, etc. in the spirit of Harrison (1972).

280

J. S. Bruner

has hair’, versus ‘The dog-ga is on the chair’. He found that for both Japanese and English speaking children, early sentences contained about twice as many intrinsic as extrinsic predications, and that subject noun phrase topics that referred to the speaker were particularly rare. This led him to conclude that “holophrastic utterances consist largely, if not exclusively, of intrinsic predicates.... Children would add subjects to predicates... when the predicates become extrinsic. Such an event appears to happen first when the children are 18 to 24 months old” (1970b, p. 1093). To this interesting finding (or contention) should be added two others. Quite counter-intuitively, Wall (1968) found that the mean length of dialogue between children and parents was shorter than dialogues between the same children and strangers. Chafe (1970) had, meanwhile, urged that one must make the contrast between ‘new’ information and ‘old’ or shared information, that the two are handled grammatically in different ways. Wall (1974, pp. 232-3) makes the point “It seemed possible that the difference in utterance length might well be explained on the basis of presence or absence of shared information between participants in the conversations. That is, it is necessary for relative strangers to state explicitly whatever it may be that they are trying to communicate verbally for efficient information transfer, whereas among friends and close associates remarks are often greatly abbreviated with little or no resulting loss of information transferred.” Vygotsky (1962, p. 139) has made the same point (en route to presenting his argument that the nature of inner speech is condensed predication, with topical subject left implicit): “Now let us imagine that several people are waiting for a bus. No one will say, on seeing the bus approach, ‘The bus for which we are waiting is coming’. The sentence is likely to be an abbreviated ‘Coming’, or some such expression, because the subject is plain from the situation”. And, indeed, in Wall’s (1974) study too, her 18 to 30 month-olds conformed to the rule in an interesting way. She compares the number of constituents in sentences given in response to a question (where the topic, of course, is shared in advance) and sentences spoken spontaneously. Just half of the spontaneous sentences contained two or more constituents, but only 18 percent of those in reply to a topic-setting question. We may now, in the light of the foregoing, consider afresh the significance of established, mutual-action formats discussed in the preceding section. They constitute the implicit or shared topics on which comments can be made by the child without having to be mentioned. These are the implicit topics about which comments can be made. And as these formats become differentiated into reversible or complementary roles during the growth of exchange and reciprocal modes, implicit topics become that much more contextualized in the action that adult and child share. I used the three terms, division of labor, exteriority, and constraint to characterize the nature of the shared action formats that developed during the onset of the reciprocal mode - terms borrowed, of course, from Durkheim’s (1933) characterization of the requisite properties of social norms - to specify the manner in which formats seemed to take on a shared existence binding the two partners in discourse. And it is this development that is, in my view, crucial to the course of prelinguistic predicational activity to which we now

From communication

to language - A psychological perspective

28 1

turn - particularly to the development of intrinsic predication in McNeill’s (1970b) sense. What, we may first ask, are the forms of ‘comment’ that can be made prelinguistically (or pre-propositionally) on such shared topics as the joint action patterns described in the preceding section? Before we can answer this question, we must first consider the function of predication in a communicative act. Its functions are three in number: (a) To specify something about a topic that is explicit or implicit, (b) to do so in such a way that topic and comment can be rendered separable (e.g., John is a boy and John has a hat), and, (c) to specify something in a way that is subject to truth testing or, more simply, negation. I do not know whether prelinguistic ‘comments’ (in forms we shall consider) upon implicit topics fulfill all three of these functions, and I should prefer to leave out of consideration the last of these, (since it is now just in the process of being studied by Roy Pea in our laboratory), to treat the second rather lightly, and to concentrate principally upon the first function. The first and perhaps simplest form of comment is, I think, giving indication that a topic is being shared in joint action, and it is principally revealed in the child’s management of gaze direction. Typically in our own protocols, the child when involved in a transaction over some object or activity, looks up at some juncture and makes eye-to-eye contact with the mother, often smiling as well. The topic is the joint activity, the comment is the establishment of ‘intersubjective’ sharing in connection with that activity, after which the activity goes on. A good example is provided in the account of glance management in an exchange game reported by Bruner (1975). The ‘comment’ consists of noting whether both partners are ‘with it’, engaged in the game. Similarly, when one of our mothers uses a toy such as a clown that disappears inside a cone, when the clown has disappeared and then reappeared, the child will usually then turn from the clown to the mother for gaze contact. I would interpret this ‘joining’ as an act of reasserting the joint action, a primitive version of the ‘interpersonal concept’ in respect of which Benveniste (197 1) was cited in a previous section. This form of confirming comment is supplemented and extended.at around the ninth month by the emergence of a form of vocalization we have dubbed ‘proclamative’. It occurs at two points during joint action sequences: First, at a point where the infant is about to undertake his part of a jointly attended action, seemingly as an accompaniment to intention; second, when the act is complete. The vocalized babbling may be coincident with the child looking back at the mother or may precede it. The vocalization, in short, appears to be initiating or completive with respect to an act embedded in a jointly attended task. In this sense, it may be considered as a ‘candidate-comment’ on an implicit topic. In time, the pattern becomes further elaborated, and the child may not only vocalize in these positions and make gaze contact, but also hold up an implicated object to show the mother, as when picking up a brick and placing it on a pile. Elsewhere I have commented on the fact that attentional deployment as revealed in eye movement records (Bruner, 1975; Mackworth and Bruner, 1970) may itself predis-

282

J. S. Bruner

pose to topic-comment structures in the very organization of information processing. For typically, large saccades that move attention to a sharply defined focus in the visual field are followed by smaller inspection saccades that play around features of this focus. Heywood and Coles in our laboratory are now exploring this feature of early attention, and while it is too early to say anything definitive about the onset of these focusinspecting eye-movement patterns, their stabilization might surely be thought of as a further predisposing factor to topic-comment communication - linguistically or prelinguistically. Finally a word about the separability of topic and comment achieved in predication. It is by now a common observation that the child’s play with objects takes one of two forms (a point also noted for chimpanzees by Kohler, 1926, and commented upon by Bruner, 1972, 1973). An object is successively placed into as many different actionpatterns as the child can manage: A ball is successively mouthed, squeezed, banged on the table, thrown down, called for, etc. Or an a&on is fitted to as many different objects as it will accommodate: Successively a cup is banged, then a spoon, then a doll, then any other loose object to hand. These play patterns, while in no sense direct precursors of propositional predicating, are nonetheless striking examples of separation and variation of comments on topics, with either the object serving as topic and actions-upon-it as comments, or the action serving as organizing topic and a variety of fitting objects as comments. Typical of the play of both higher apes and children (Loizos, 1967), this focus-variation pattern should not be overlooked as a factor that predisposes action, attention, and eventually language to the pattern that at the propositional level we call predicational. In conclusion, I find myself in strong agreement with Lyons (1966, p. 131) when he comments: By the time the child arrives at the age of eighteen months or so, he is already in possession of the ability to distinguish ‘things’ and ‘properties’ in the ‘situations’ in which he is learning and uses language. And this ability seems to me quite adequate as a basis for the learning of the principal deep-structure relationship between lexical items (the subject-predicate relationship), provided that the child is presented with a sufficient amount of ‘primary linguistic data’ in real ‘situations’ of language use. Before he reaches eighteen months, indeed during the second half of his first year, he is well on the way toward conceptual mastery of these concepts in the extralinguistic sphere.

Conclusion The developmental psychology of language is currently initial optimism that grew out of Chomsky’s formulation

in a rather confused state. The of a generative-transformational

From communication to language - A psychological perspective

283

grammar has not been sustained by the torrent of work that it provoked. His was a powerful idea, one that will want a revisit after other aspects of language acquisition become clearer. The central notion - that the child in some sense ‘has a knowledge’ of the rules of language and that he is attempting to generate from this knowledge hypotheses about a local language - while boldly suggestive, is plainly insufficient. Principally as a result of the studies of Brown (1973) and his students, it has become increasingly apparent that language acquisition is enormously aided by the child’s prelinguistic grasp of concepts and meanings that make it easier for him to penetrate grammatical rules. In a closely reasoned and provocative paper published in 1972, Macnamara formulated the case well, arguing that syntactic rules are discovered by the child with the aid of meaning. His view was that the child roughly determined the referent of principal lexical items in a sentence and then used previously acquired knowledge of these referents to decode the grammar of the sentence. Sinclair’s (e.g. 1969) work too has alerted the psycholinguist to the link between development and the child’s emerging, extralinguistic knowledge of the world. And Bloom’s most recent work has also dealt a strong blow in favor of the early semantic origins of single-word utterances. She concludes that “children develop certain conceptual representations of regularly recurring experiences, and then learn whatever words conveniently code such conceptual notions” (1973, p. 113). The effect of this recent work has been to put the semantic element back into the developmental picture and make more attractive such ideas as Fillmore’s (1968) semantically relevant case categories. But neither the syntactic nor the semantic approach to language acquisition takes sufficiently into account what the child is trying to do by communicating. As linguistic philosophers remind us, utterances are used for different ends and use is a powerful determinant of rule structures. The brunt of my argument has been that one cannot understand the transition from prelinguistic to linguistic communication without taking into account the uses of communication as speech acts. I have, accordingly, placed greater emphasis on the importance of pragmatics in this transition - the directive function of speech through which speakers affect the behaviour of others in trying to carry out their intentions. I find myself in sympathy with Dore’s effort (1975) to understand the process whereby ‘primitive forces’ or ‘orectic intentions’ are gradually conventionalized and ‘grammaticalized’ so that they can be reformed into communications with illocutionary force. I am not dismayed at all by Jonathan Cohen’s (1974) warning that the conventionalizations by which illocutionary force is achieved are often, strictly, extralinguistic ‘manners’, for perhaps, as Silverstein (1975) suggests, there is not so sharp a boundary between social convention and grammatical devices. Dore’s account of how illocutionary skill is augmented by the acquisition of such ‘grammaticalizing devices’ is interesting. He defines a primitive speech act “as a rudimentary referring expression plus a primitive illocutionary force” (such as requesting, answering, etc.) so that the child “communicates what it is he means or wants” through referential tricks and, through prosodic pattern initially and then by other means, “that he intends or wants something”.

284

J. S. Bruner

How the child gets from the primitive to the grammaticalized is left to rather mysterious processes like ‘emergence’ and ‘grammaticalization’ that may do no more than paper over the discontinuous course of language acquisition with some new words. Yet, my sympathies are with Dore’s effort to examine how the requirement of ‘getting different things done with words’ constantly alerts the child to appropriate devices and conventions and, in an evolutionary sense, may even have equipped him with special sensitivities for picking these up. Yet, for all that, I hope I have not seemed to deny that syntactic and semantic precursors can also be explored fruitfully: Grammar-like principles underlying reference, predication, privileges of occurrence, etc. But if there is one point that deserves emphasis, whether one is searching for syntactic, semantic, or pragmatic precursors of early language, it is that language acquisition occurs in the context of an ‘action dialogue’ in which joint action is being undertaken by infant and adult. The joint enterprise sets the deictic limits that govern joint reference, determines the need for a referential taxonomy, establishes the need for signalling intent, and provides a context for the development of explicit predication. The evolution of language itself, notably its universal structures, probably reflects the requirements of joint action and it is probably because of that evolutionary history that its use is mastered with such relative ease, though its theoretical explication still eludes us.

REFERENCES Ainsworth, Mary D. Salter (1975) Social development in the first year of life: maternal influences on infant-mother attachment. Paper presented in Geoffrey Vickers Lecture, London. (Unpublished) Ainsworth, Mary D. Salter and BeIl, Sylvia M. (1974) Mother-infant interaction and the development of competence, In K. Connolly and J. S. Bruner (Eds) The Growth of Competence. London and New York, Academic Press. Argyle, M., and Ingham, R. (1972) Gaze, mutual gaze and proximity. Semiotica, Vol. IV, (1) 32-49. Austin, J. L. (1962) How To Do Things with Words, Oxford, Oxford University Press. Bates, Elizabeth, Camaioni, L., and Volterra, V. (1973) The acquisition of performatives prior to speech. Technical Report No. 129, Consiglio Nazionale delle Ricerche, Rome. Benveniste, E. (1971) Problems in General Linguistics. (Translated by M. E. Meek) Coral Gables, Florida, University of Miami Press. Bernstein, B. (1960) Language and social class. Brit. J. Social., 11, 271-276. Bloom, Lois (1970) Language Development: Form and Function in Emerging Grammars, Cambridge, Mass., M.I.T. Press. Bloom, Lois (1973) One Word at a Time: The Use of Single Word Utterances Before Syntax. The Hague, Mouton. Bloomfield, L. (1933) Language. New York, Holt. Brown, R. (1970) Psycholinguistics. New York, The Free Press. Brown, R. (1973) A First Language: The Early Stuges. Cambridge, Mass., Harvard University Press. Bruner, J. S. (1972) The nature and uses of immaturity. Amer. Psychol., 27, (8), l-22. Bruner, J. S. (1973) Organisation of early skilled action. Child Devel., 44, l-11. Bruner, J. S. (1975) The ontogenesis of speech acts. J. child Lang., 2 (1) 1-19. Bruner, J. S., and Sherwood, Virginia (in press) Early rule structure: the case of peekaboo. In J. S. Bruner, A. Jolly and K. Sylva (Eds) Play: Its Rote in Evolution and Development. London, Penguin. Btihler, K. (1934) Sprachtheorie: die Darstellungsfunktion der Sprache. Jena, Fischer.

From Communication to language - A psychological perspective 285

Campbell, R., and Wales, R. (1970) The study of language acquisition. In J. Lyons (Ed) New Horizons in Linguistics. London, Penguin. Chafe, W. L. (1970) Meaning and the Structure of Language. Chicago, University of Chicago Press. Chomsky, N. (1965) Aspects ofthe Theory ofSyntax. Cambridge, Mass., M.I.T. Press. Clark, Eve (1973) What’s in a word: on the child’s acquisition of semantics in his first language. In T. E. Moore (Ed) Cognitive Development and the Acquisition of Lunguuge. New York, Academic Press. Cohen, J. (1974) Speech acts. In T. A. Sebeok (Ed) Current Trends in Linguistics, vol. 12: Linguistics and the Adjacent Arts and Sciences. The Hague, Mouton. Collis. G. M.. and Schaffer. H. R. (in press) Synchronisation of visual attention in mother-infant pairs. i child PsychoI. Psych., ‘16.Cromer, R. F. (1974) The development of language and cognition: the cognition hypothesis. In B. Foss (Ed) New Perspectives in Child Development. London, Penguin Education Series. DeLaguna, Grace (1927) Speech: Its Function and Development. New Haven, Connecticut, Yale University Press. Donaldson, Margaret, and Wales, R. (1970) On the acquisition of some relational terms. In J. R. Hayes (Ed) Cognition and the Development of Language. New York, Wiley and Sons. Dore, J. (1974) Communicative intentions and the pragmatics of language develop’ment. Unpublished paper. (1975) Holophrases, speech acts and language universals. J. child. Lung., 2(l), 2140. Durkheim. E. (1933) The Division oflubor. Glencoe, Illinois, The Free Press. Edwards, D. (1975) Constraints on action: a source of early meanings in child language. Based on delivered at the Symposium on Language and “The three sources of a child’s first meanings”, Social Context, University of Stirling, 10 - 11 January. Fillmore, C. J. (1968) The case for case. In E. Bach and E. T. Harmes (Eds) Universals in Linguistic Theory. New York, Holt, Rinehart and Winston. Garfinkel, H. (1963) Trust and stable actions. In 0. J. Harvey (Ed) Motivation and Social Interaction. New York, Ronald. Garvey, Catherine (1974) Some properties of social play. Merrill-Palmer Q., 20(3), 164-180. Gombrich, E. (1975) Mirror and map: theories of pictorial representation. Philosophical Transactions of the Royal Society. (Biol. Sci.), Vol. 270, No. 903, 119-149. Greenfield, Patricia M. (1972) Playing peekaboo with a four-month old: a study of the role of speech and nonspeech sounds in the formulation of a visual schema. J. Aychol., 82, 287-298. Greenfield, Patricia M. (1973) Who is “Dada”? . ..some aspects of the semantic and phonological development of a child’s first words. Lung. and Speech, 16, (l), 34-43. Greenfield, Patricia M., Bruner, J. S., and May, M. (1972) Early Words (A Film). New York, Wiley. Greenfield, Patricia M., and Smith, J. H. (in press) Lunguuge Beyond Syntax: The Development of Semantic Structure. New York, Academic, Press. Grice, H. P. (1968) Utterer’s meaning, sentence-meaning and word-meaning. Found. Lung., 4, 1-18. Grice, H. P. (in press) Logic and conversation. The William James Lectures, Harvard University, 1967 68. In P. Cole and J. Morgan (Eds) Syntax and Semantics, Vol. 3, Speech Acts. London and New York, Academic Press. Harrison, B. (1972) Meaningand Structure. New York and London, Harper and Row. Hess, R. D. and Shipman, Virginia (1965) Early experience and the socialisation of cognitive modes in children. Child Devel., 36, 869-886. Howe, Christine (1975) The nature and origin of social class - differences in the propositions expressed by young children. unpublished Ph.D. thesis, University of Cambridge. Jakobson, R. (1960) Linguistics and poetics. In T. A. Sebeok (Ed) Style in Language. Cambridge, Mass., M.I.T. Press. Kaplan, E. and Kaplan, G. (1971) The pre-linguistic child. In J. Eliot (Ed) Human Development and Cognitive Processes. New York: Halt, Rinehart and Winston. Kaye, K. (1976) Infants’ effects upon their mothers’ teaching strategies. In J. C. Glidewell (Ed), The Social Context of Learning and Development. New York, Gardiner Press. Kohler, W. (1926) The Mentality of Apes. New York, Harcourt, Brace.

286

J. S. Bruner

Lock, A. (1972) From out of nowhere? Proceedings of the International Symposium on First Language Acquisition, University of Ottawa Press. Loizos, E. (1967) Play behaviour in higher primates: a review. In D. Morris (Ed) Primate Ethology. London, Weidenfeld & Nicolson. Lyons, J. (1966) General discussion to D. McNeill’s paper, The creation of language. In J. Lyons and R. Wales (Eds) Psycholinguistic Papers. Edinburgh, Edinburgh University Press. Lyons, J. (1974) Deixis as the source of reference. Unpublished paper. Macfarlane, A. (1975) Personal communication. Mackworth, N. H., and Bruner, J. S. (1970) How adults and children search and recognise pictures. Hum. Devel., 13. (3), 149-177. Macnamara, J. (1972) Cognitive basis of language learning in infants. Psychol. Rev., 79, (l), 1-13. McNeill, D. (1970a) The Acquisition of Language: The Study of Developmental Psycholinguistics. New York, Harper and Row. (1970b) The development of language. In P. H. Mussen (Ed) Carmichael’s Manual of Child Psychology, 3rd ed. Vol. 1. New York, Wiley. _ (1974) Semiotic extension. Paper presented at the Loyola Symposium on Cognition, 30 April, Chicago, Illinois. Meltzoff, A. N. (1975) Personal communication. Moore, M. K., and Meltzoff, A. N. (1975) Neonate imitation: a test of existence and mechanism. Paper delivered at the Society for Research in Child Development meeting, Denver, Co., April. Morris, C. W. (1938) Foundations of the Theory of Signs. Chicago. Nelson, Katherine (1973) Structure and strategy in learning to talk. Sot. Res. Child Devel. Mono., 38, Nos. l-2, Serial No. 149. Ogden, J. C., and Richards, I. A. (1923) The Meaning ofMeaning. New York, Harcourt Brace Jovanovich, Inc. Piaget, J. (1952) The Origins of Intelligence in Children. (1st ed., 1936). New York, International Universities Press. Quine, W. V. 0. (1960) Word and Object. Cambridge, Mass., M.I.T. Press. Raven, J. C. (1948) The comparative assessment of intellectual ability. The British J. Psychol., 39-40, 12-19. Rheingold, H. L., Gewirtz, J., & Ross, H. (1959) Social conditioning of vocalisations in the infant. J. camp. Physiol. Psychol., 52, 68-73.

Ricks, D. M. (1971) The beginnings of vocal communication in infants and autistic children. Unpublished Doctorate of Medicine thesis, University of London. Rosch, Eleanor (1974) Basic level objects in natural categories. Paper presented at the Psychonomic Society, Boston, November. Ryan, Joanna (1974) Early language development. In M. P. M. Richards (Ed) The Integration of the Child into a Social World. Cambridge, Cambridge University Press. Sander, L. W., Stechler, G., Burns, P. and Julia, H. (1970) Early mother-infant interaction and 24hour patterns of activity and sleep. J. amer. Acad. Child Psych., 9, 103-123. Scaife, M. and Bruner, J. S. (1975) The capacity for joint visual attention in the infant. Nature, 253, No. 5489,265-266.

Schoggen, M. and Schoggen, P. (1971) Environmental forces in the home lives of three-year-old children in three population subgroups. D.A.R.C.E.E. Papers and Reports, Vol. 5, No. 2, (John Kennedy Center for Research on Education and Human Development, George Peabody College, Nashville, Tenn.). Searle, J. R. (1969) Speech Acts: An Essay in the Philosophy of Language. Cambridge: Cambridge University Press. Searle, J. R. (1975) Speech acts and recent linguistics. Paper read at the Conference on Developmental Psycholinguistics and Communication Disorders, New York Academy of Sciences, January 24-25. Silverstein, M. (1975) Shifters, linguistic categories and cultural description. Unpublished manuscript. Sinclair-de-Zwart. Hermina (1969) Developmental psycholinguistics. In D. Elkind and J. H. Flavell (Eds) Studies in Cognitive Growth: Essays in Honour of Jean Piaget. New York, Oxford University Press.

From Communication to language - A psychological perspective

287

Slobin, D. (1973) Prerequisites for the development of grammar. In C. A. Ferguson and D. Slobin (Eds) Studies of Child LanguageDevelopment. New York, Holt, Rinehart and Winston. Sugarman, Susan (1974) A sequence for communicative development in the pre-language child. Unpublished paper. Trevarthen, C. (1974a) Conversations with a two-month old. New Scientist, 62, (896), 230-235. _ (1974b) Infant responses to objects and persons. Paper presented at the Spring 1974 meeting of the British Psychological Society, Bangor. Urwin, Catherine (1973) The development of a blind baby. Unpublished manuscript presented at Edinburgh University. Vygotsky, L. (1962) Thought and Language. Cambridge, Mass., M.I.T. Press. Wall, Carol (1968) Linguistic interaction of children with different alters. Unpublished paper, University of California, Davis. Wall, Carol (1974) Predication: A Study of its Development. The Hague, Mouton. Wittgenstein, L. (1953) Philosophical Investigations. New York: Macmillan. Wolff, P. H. (1969) The natural history of crying and other vocalisations in early infancy. In B. M. Foss (Ed) Determinants of Infant Behavior, Vol. 4, London, Methuen. Wood, D., Bruner, J. S. & Ross, Gail (in press) The role of tutoring in problem solving. J. Child Psychol. Psych.

Toute approche realiste du langage, se doit de rendre compte du passage de la communication par prelangage de l’enfant h l’utilisation de la langue h proprement parler. Pour cela, on peut montrer qu’il existe de nombreuses bases, pr&lables ou ndcessaires aux traits organisationnels de la syntaxe, de la semantique, de la pragmatique et mdme dc la phonologie, dans les activitds prelangagkres des enfants. Des illustrations de ces bases prClables sont etudikes ici dans 4 domaines differents: le mode d’interprdtation, par la mere des intentions de communication de l’enfant, le developpement de la combinaison des rkferentiels, rendant le langage confonne i l’environnement, l’evolution de strategies permettant l’utilisation de I’activite conjointe au langage, la transformation dune organisation de type topic-comment i la predication. En dernier lieu on propose la conjecture suivante: la connaissance, par l’enfant des besoins de l’action et de l’interaction peut elle fournir la base i l’elaboration initiale de la grammaire.

Discussion

Putnam on reductionism

ANDREW

LUGG

University of Ottawa

It is Putnam’s view that reductionism (i.e., “the doctrine that the laws of such ‘higher level’ sciences as psychology and sociology are reducible to the laws of lower-level sciences - biology, chemistry, ultimately to the laws of elementary particle physics”*) is wrong. In this note, I take issue with some of Putnam’s arguments. Putnam begins with the ‘logical point’ that “from the fact that the behavior of a system can be deduced from its description as a system of elementary particles it does not follow that it can be explained from that description” (131). That a square peg, a fraction less than 1” across, goes through a 1” square hole but not through a 1” round hole can be deduced and explained by appealing to elementary geometry, the rigidity of the peg, and so on. But, even if deducible, it can’t be explained by appeal to elementary particle physics, there being a pragmatic constraint on explanation: The relevant features of a situation must be brought out by an explanation and not buried in a mass of irrelevant information (132). So explanation is intransitive and reduction fails (provided, of course, reduction without explanation is an impossibility). This argument is not as clear as Putnam seems to think. One might cling to the idea that deducibility guarantees explanation and deny that explanation must be revealing. Then, only revealing explanation would fail of transitivity, not explanation itself. (In this respect, note that Putnam himself has occasion to talk of revealing explanation (133).) A more telling complaint is that reduction need not make for explanation (in Putnam’s sense) since derivability, not explainability, guarantees reducibility. (But see below.) Putnam has a second important line of argument, however, which if correct, undercuts the criticisms just mentioned. As he puts it, the laws of higher-level science (e.g., economics, even biology) are autonomous vis&vis the laws of physics, in the sense that there is no deduction of the former from the latter to be had. The former can only be deduced from the latter in conjunction with “ ‘auxiliary hypotheses’ which are occidental from the point of view of the lower-level discipline” (134, Putnam’s italics). Whereas, says Putnam, “given the structure of the peg and board, one can deduce the rigidity... given the microstructure of the brain and the nervous system, one cannot deduce that capitalist *Putnam, H. (1973) Reduction and the Nature of Psychology”, in the text are to this article.

Cog., 2, p. 131. All page references Cognition

3(3), pp. 289 - 293

290

Andrew Lugg

production relations exist... [Th e 1aws of capitalist society] depend on ‘boundary conditions’ which are accidental from the point of view of physics but essential to the description of a situation as ‘capitalism’ ” (ibid.). To see that something has gone wrong here consider the relationship between Kepler’s laws and Newton’s laws. In this case, the ‘higher-level’ discipline (Kepler’s laws) is no less autonomous vis-a-vis the ‘lower-level’ discipline (Newton’s laws) than is biology vis-a-vis physics. For the former cannot be deduced from the latter; they can only be deduced (or rather an approximation to Kepler’s laws can only be deduced - see below) from Newton’s laws plus auxiliary hypotheses, accidental from the point of view of Newton’s theory. (The central mass must be much larger than any of the other masses and all masses must be far apart.) So, provided we agree that Kepler’s laws are reducible to Newton’s, reduction does not entail lack of autonomy. A similar point can be made about the peg and the board. Its behavior cannof be deduced from the laws of elementary particle physics alone. We also need its description as a certain (accidental) configuration of elementary particles. But one might insist (cf. Putnam’s remarks on p. 134) that, whereas rigidity can be deduced from microstructure, capitalistic production relations can’t be deduced from the microstructure of the brain and the nervous system. In reply, note that even if the reported difference between pegs and boards and captialism is genuine, failure of reduction doesn’t follow. To see why consider a container of gas. The pressure of the gas cannot be deduced from its microstructure alone. In addition, we need facts about the container - boundary conditions in the literal sense - which are accidental from the point of view of statistical mechanics. But if we allow that thermodynamics is reducible to statistical mechanics why should we make reduction of the social sciences to the physical sciences more difficult? In the case of economic systems, why should we bar appeal to boundary conditions (e.g. facts about profits and the like) which are accidental from the point of view of the physical sciences? Next note that according to Putnam (134) reductionism gives rise to and reinforces the bad idea that human nature is unchanging. For, he contends, if reductionism is true, the laws of psychology are deducible (via the laws of biology) from the laws of physics, which are unchanging.* So the laws of psychology are unchanging, which is to say that human nature doesn’t change. But it is not at all obvious - not to some scientists and philosophers, at least - that the laws of physics are unchanging. More important: Even if the laws of psychology do not change, it does not follow that human nature is unchanging. The boundary conditions may change. Certainly, the laws of physics have not changed noticeably in the last half century but the nature of, e.g., radios has: They don’t distort so much, they are less noisy, frequency response has improved, and so on.

*Putnam also allows (unchanging) affect the point to be made.

reductive

definitions

in these derivations

(134).

But this does not

Putnam on reductionism

291

Methodological individualism is another bad idea which Putnam believes is nurtured by reductionism. But this is far from evident. Even within.physics, reduction of one theory to another frequently fails to bring about theory replacement. (A vast amount of contemporary work within physics makes use of Newtonian mechanics, a theory long ago reduced to something better.) Furthermore, reduction need not precipitate neglect of the subject matter or the methods of higher-level disciplines. Indeed, even the most thorough-going reductionists may consistently advocate that disproportionately large efforts be expended on higher-level disciplines. Reduction is not the only nor even a major aim of science; to effect a reduction one needs something reasonably well worked out to reduce. Next, some issues of a more general nature. Whatever else it is, reduction surely involves a connectability requirement - the terms of the higher-level theory must be linked in some way with terms of the lower-level theory - and a consequence requirement ~ the laws of the higher-level theory must be linked in some way with laws of the lower-level theory.* But to say this is not to say very much. Can we say more? First, connectability. This would seem to hold only if (i) there is a one-many (not necessarily a one-one) correspondence between higher-level and lower-level structures and (ii) for every open sentence, @x, of the higher-level discipline there is an open sentence, 3/x, of the lower-level discipline such that $x is satisfied by a higher-level structure whenever $x is satisfied by one of the lower-level structures that clause (i) associates with the higher-level structure. Now, given this account of connectability, reduction can fail in a number of ways. For instance one might argue that there are no open sentences of physics which correspond appropriately to open sentences of psychology. Indeed, it has been held that the only appropriate correspondences are between psychology and even higher-level disciplines. (Consider the claim that psychological fact depends on institutional or sociological fact.) But such arguments are difficult to sustain, it always being open to the reductionist to claim that the brain can internafly represent psychological and even sociological facts. (As far as I am aware, what evidence there is rules more in favor of internal representation of this sort than against it.) More interesting is the possibility of there being a many-one correspondence between higher-level and lower-level structures. Putnam seems to hold that such correspondences exist. (Capitalistic and socialistic economic orders, he seems to hold, are both compatible with the same psychological order.) And Hull has suggested on the basis of a detailed study of reduction in genetics that “the same molecular mechanism can produce different phenotypic effects..., (that) the Same molecular situation can result in phenomena which would have to be characterized by different Mendelian predicate terms.“** But, if things *In Mechanistic Explanation and Organismic Biology, Phil. Phenomenal. Res., 1. 1951, E. Nagel mentions two similar conditions: definability and derivability. My conditions are designed to allow for non-definability and non-derivability. (But note that Nagel’s definability condition includes connection by empirical biconditionals.) **Hull, D. L. (1972) Reduction in Genetics - Biology or Philosophy, Phil. Science, 39, 497-498.

292

Andrew Lugg

like this can happen in biology it is not unreasonable to think that they can happen in psychology too, However, once again, matters are not so straightforward as they appear at first sight. The one-many relation may be eliminable by introducing additional factors. For instance, in the biological case this can be done by ‘expanding’ the notion of a molecular mechanism.* But, if this is so, the reductionist can retort: Why shouldn’t we expect something similar to happen in the case of psychology? Surely, what ‘sort of structures are appropriate for correlation is an empirical matter, not something to be decided a priori. In short, it is by no means obvious that connectability, construed reasonably, fails to hold between Jhe social sciences and physics. Earlier I spoke of the consequence requirement as one of deducibiZity. But, as is wellknown, certain incompatible pairs of theories seem (intuitively) reducible one to the other. For example, some of the consequences of Newton’s theory conflict with consequences of Einstein’s theory. Whence, it follows that we must forgo deducibility or relinquish our intuitions. But lack of deducibility is hardly a strike against reductionism. For there seem to be a number of plausible candidates for the consequence constraint other than deducibility. One is that the consequences of the higher-level theory agree approximately with those of the lower-level theory.** Another is that the lower-level theory correct the higher-level theory and explain why it works to the extent that it does.*** What needs to be shown (and as far as I am aware what has not been shown) is that the consequence relation, construed reasonably, - e.g., in the one of the ways just mentioned - fails to hold between the social sciences and physics. One might object that there is an important disanalogy between intra-physics reductions and reductions between physics and higher-level disciplines. Putnam expresses it this way (138): Whereas it is the business of physics to account for inaccuracies in physics, accounting for inaccuracies in the social sciences may or may not be the business of the social sciences. But surely there are ‘levels’ within physics just as much as there are levels within science? It is (in part) the business of microphysics to account for inaccuracies in macrophysics. (Consider, e.g., the relation between statistical mechanics and thermodynamics.) It is (in part) the business of relativistic physics to account for inaccuracies in non-relativistic physics. And so on.

*Cf. ibid. p. 498. Another example is this. The pressure on a region depends not merely on the molecular motion of particles in or near it but on the entire molecular milieu. So, reduction would fail were we to connect pressure with molecular motion in the vicinity. However, all this need show is that our chosen connections are at fault, not that reduction is not to be had. **Cf. Hi Putnam (19651 How Not to Talk about Meaning. In Cohen and Wartofsky (eds.) Boston Studies in the Philosophy of Science, II, Reidel. In this-paper, Putnam makes some suggestions concerning what counts as an approximation. ***W. Sellars (1961) in The language of theories, Feigl and Maxwell (eds.) Currenf Issues in the Philosophy of Science. Holt, Rinehart and Winston, discusses a view similar to this. Note that I don’t intend to suggest that these two approaches are or need be distinct.

Putnam on reductionism

293

Putting all this together we get the following moral: One should not expect to convince a reductionist that he is wrong by showing that a reduction paradigm fails to apply vis-a-vis the social sciences and physics when it also fails to apply to clear cases of reduction within physics itself. In this note, I have not commented on Putnam’s claims concerning intelligence testing. These seem to be both cogent and important. However, I believe they can be and should be launched and developed independently of any considerations concerning reductionism. They can be because reductionism is compatible with human natures which change; they should be because an appeal to anti-reductionism results in a loss of generality.

Reply to Lugg

HI LARY PUTNAM Harvard University

Mr Lugg’s first point is that one might “cling to the idea that deducibility guarantees explanation” and then it would follow that I was wrong. My paper was obviously based on a rejection of the deductive-nomological model of explanation. The point that ifthe deductive-nomological model of explanation is correct after all, then I am wrong is one that I should have thought would be evident to any reader of the paper. My position is that explanation is essentially a pragmatic notion. In my view, whether or not a putative explanation is really an explanation depends on many factors besides the logical relations between the putative explanation and whatever is to be explained; for instance, the explanation must relate in an appropriate way to the interests of the questioner, there are questions of scientific fruitfulness to be considered, etc. To illustrate the way in which explanation is interest-relative, let me use an example due to Alan Gartinkel.* Suppose a priest asks Willy Sutton “Why do you rob banks?“. And Willy Sutton replies “because that’s where the money is”. Clearly something has gone wrong. In my view, what has gone wrong is this: The priest was really interested in knowing why Sutton robs banks us opposed to not robbing at all (this is what Garfinkel calls the explanation space of the question). If Sutton had been asked the same question by another robber, then his answer would have been explanatory. But the other robber would have had a quite different ‘explanation space’. Another example may bring out the point even more sharply. Suppose the question is asked, “Why was Professor McJones, who was previously believed to be an exemplary member of the academic community, found stark naked in the women’s dormitory at 12:OO midnight?” To answer: “He was stark naked in the women’s dormitory at 12:00 midnight because he was stark naked in the women’s dormitory at 12:OO midnight minus epsilon and it was not possible for him either to put on his clothes or to leave the women’s dormitory in epsilon seconds without violating the law that nothing can move faster than the speed of light”, would not be to explain why Professor McJones was found naked in the women’s dormitory at 12:00 midnight, even though all the requirements of the deductive-nomological model of explanation are met. (Again, Garfinkel’s notion of an explanation-space presupposed by a *Cf. Gariinkel, A. (1975) Explanation and Individuals. Ph.D. thesis presented Philosophy, Harvard University, Cambridge, Mass.

to the Department

of

Cognition 3(3), pp. 295 - 298

296 Hilary Putnam

why-question can be easily used to explain why this reply is not a reply to the question that is really intended.) The example in my paper of the hypothetical Laplacean Super Mind who deduces that the square peg will go through the square hole and not through the round hole by surveying all possible trajectories of this particular system of atoms and molecules, illustrates the way in which considerations of scientific fruitfulness and relevance enter into our decision whether to call something an explanation. If we put the Laplacean Super Mind’s ‘explanation’ in canonical deductive-nomological form, then the covering law is something like this: “If one ever has a system consisting of a peg and a board, and the system consists of exactly these sorts of atoms and molecules in exactly these arrangements, then the square peg will go through the square hole and not through the round hole”. It is clear that this ‘covering law’ is not one that we are interested in. But interest here is not just a subjective matter. We are interested in it for good methodological reasons. It is not a fruitful law and it is not even a practically useful one, for the very good reason that there is never going to be another system consisting of a peg and a board which consists of exactly these sorts of molecules and atoms in exactly this arrangement. The difference between Mr. Lugg and myself may at bottom be a difference over our conceptions of philosophy of science. I see philosophy of science as normative description of science. The task of the philosopher of science is to find out what regularities in the actual behavior of scientists contribute to the success of science. To dismiss the interest-relativity and fruitfulness-relativity of explanation on the grounds that these are ‘merely pragmatic’ features of explanation is to assimilate these features to features which are only of ‘psychological’ interest. But that explanations are required to fit the explanation-space of the questioner, and that the covering laws in an explanation must be relevant, are regularities in the way in which we use the notion of explanation which contribute to the usefulness of the notion. It is of methodological, not merely psychological, interest that scientists require explanations to fit the explanation-spaces of the questioner and they require the covering laws in an explanation to be relevant. To summarize, I take my stand on this: That whatever the right theory of explanation may be, the Laplacean Super Mind’s deduction is a paradigm case of a non-explanation. The ‘autonomy’ of higher level disciplines, in my view, resides precisely in this: That the behavior of the higher level systems is sometimes not explainable by their microstructure even though it is deducible from their micro-structure. The case of the peg and the board is an example of this. I take it that on anyone’s economic theory, economics is another example. To deduce from their physical and chemical composition as systems of elementary particles that a bunch of organisms will arrive at an equilibrium price is not to explain why an equilibrium price was arrived at. It might be to explain something else, however. It might, for example, be to explain how it is possible for physical systems to engage in such behavior as economic bargaining. Similarly, when I deny that the laws of psychology, or, for that matter, the way in which your favorite digital computer computes the decimal expansion of pi, is explained by physics and chemistry, I do not mean

Reply to Lugg

297

to deny that physics and chemistry can answer the question “how is it possible for a physical system to obey such and such psychological laws?“, or “how is it possible for a physical system to embody such and such a program. 7” That these are important questions and that they can be answered by physical and chemical investigation is, it seems to me, the sound core of truth in reductionism. But it is a mistake to confuse the question “Why is an equilibrium price almost always arrived at?“, which cannot be answered by a physical and chemical investigation, with the question “How is it possible for physical systems to be models of economic theory?” which can be. Mr. Lugg attributes to me the view that in a successful reduction, the laws of the reduced discipline must be deduced from the laws of the reducing discipline (and, of course, the theoretical definitions). This is not my view. What I pointed out in my paper was that it is commonly asserted in the literature that this is what takes place in a reduction. But I perfectly agree with Mr. Lugg that even in the standard examples of reduction (e.g., the reduction of thermodynamics to statistical mechanics) this is not complied with. What we actually do is deduce the laws of the reduced discipline from the laws of the reducing discipline together with appropriate auxiliary statements (e.g., statements of boundary conditions). I am not arguing that the need for boundary conditions in a reduction by itself shows that the laws of the reduced discipline are ‘autonomous’ or that the ‘reduction’ is not really a successful reduction. The problem here is connected with another suggestion that Mr. Lugg makes. Mr. Lugg suggests that one might agree with me that ‘reductions’ are sometimes not explanatory, and simply give up the idea that explanation is any part of the purpose of reduction. But this seems to me bad philosophy of science. The reduction of thermodynamics to statistical mechanics is an explanation and that is why physicists are interested in it, and, indeed, it is why they say that thermodynamics has been reduced to statistical mechanics. To simply make “reduction” a term of art, so that by definition the silly deduction of the Laplacean Super Mind counts as a ‘reduction’ is precisely to give up the task of figuring out what reductions scientists ought to be interested in and what reductions they ought not to be interested in; it is to give up the task of finding out the significant differences between acceptable reductions and unacceptable reductions. My view is that the reduction of thermodynamics to statistical mechanics is a genuine reduction notwithstanding the need for boundary conditions, whereas the ‘reduction’ performed by the Laplacean Super Mind in the case of the peg and the board, or the analogous ‘reduction’ that the Laplacean Super Mind might perform in the case of the equilibrium price problem, is not a genuine reduction because it fails to explain anything and because explanation is the most important purpose of reduction. Finally, Mr. Lugg takes issue with a number of suggestions in my paper that were meant more as amateur sociology than as logical considerations. I perfectly agree with Mr. Lugg that it is possible to consistently be a strict reductionist and to encourage investigation in higher level disciplines; in fact I said so in my paper. What I pointed out was that as a matter of historical fact, many biology departments fired their naturalists because

298

Hilary Putnam

the departments were carried away with a certain enthusiasm for reduction when the DNA model appeared. Their mistake, moreover, had a logical component: They argued that in principle all of the higher level phenomena could be explained by molecular chemistry. In my view, this logical component of their blunder was wrong. But even if they had been right they would have been making a crude error, as I said in the paper. And, of course, Mr. Lugg is right in saying that one can consistently be a reductionist and think that it may be possible to change human nature in at least some respects by changing boundary conditions. Finally, let me say that I do not see at all why Mr. Lugg feels so sure that there must be open sentences in the language of physics and chemistry corresponding to such psychological predicates as x is angry. Given the certainty that anger can be realized by an infinite number of different physical-chemical structures, it would seem very unlikely that there should be such an open sentence, at least in the language of first order physics. Of course, there may well be a functional analysis of anger; but this is something quite different. And the argument that references to environment can be eliminated because the environment is represented inside the brain, seems to me just a mistake. For the fact is that even if Desdemona and Iago are represented by structures in Othello’s brain, nevertheless, that those structures are representations of a real woman and a real man who really behaved in certain ways is not a physical-chemical property of those structures but depends on the relation of those structures to the surrounding environment. Thus, to predicate the open sentence x is jealous of Othello, is not to make any statement whose truth conditions can be given just in terms of the physics and chemistry of Othello’s brain (unless we redefine jealousy so that one can be jealous of figments of one’s imagination); and we will have even worse problems with the predicate x arrived at an equilibn’um price. *

*John McDowell has put this point well (I quote from personal correspondence), “One might say psychological states aren’t necessarily in the head. Of course that’s a paradoxical way of putting it: if psychological states are neurophysiological states, presumably the head or at least the central nervous system is where they are. The point would be: you can’t arrive at their psychological descriptions by looking nowhere but there. (I think this is Wittgenstein’s point when he says “If God had looked into our minds, He would not have been able to see there what we were speaking of’.)”