The Act of Creation

by Arthur Koestler

  Is that all? Or may we at least add, with Hilgard, that 'without language it would be well-nigh impossible to think of such abstract notions as justice or reciprocal tariffs' (Hilgard, 1957, p. 315).

  To p. 601. In Head's gesture-imitation experiments, on the other hand, some kind of implicit verbal symbolization seems to be indispensable, because reliance on visual or kinesthetic clues leads only to the familiar mirror-confusion -- as any normal person will discover by repeating the experiment.

  To p. 602. One might further speculate that the patient's inability to name the object shown on the printed card, while he is perfectly capable of using words to explain that he cannot name it, may be due to the fact that single object-names like tree, moth, etc., have a more arbitrary character qua phonetic labels learnt by rote, and have more 'brittle' traces, than verbs and propositions which form syntactic sub-wholes. A 'tree', after all, is originally a nonsense syllable. One of Penfield's patients, shown a card picturing a foot, said, 'Oh, I know what it is. That is what you put in your shoes'. After withdrawal of the electrode, he said correctly 'foot'. Now why should 'foot' be more difficult to name than 'shoe'? Perhaps because the word 'shoe' occurred in a meaningful context -- Burt's Factor II of Verbal Ability -- whereas 'foot' occurred in a naming context (Factor I) which was blocked (see next note). In other words, the naming-code was out of action, while the semantic code continued to function.

  To p. 603. Burt divided Verbal Ability into a word factor dealing with words in isolation and a language factor dealing with words in their context. The former he sub-divided into a receptive factor of recognizing and understanding words and into an executive factor for finding and selecting the right word. The language factor is similarly sub-divided into a receptive factor for understanding statements and an executive factor for literary expression and verbal fluency. Burt's divisions followed in broad outline Head's classifications of aphasia and substantiated the validity of the latter, (The British Journal of Educational Psychology, Vol. XIX, June and November, 1949).

  XV

  LEARNING TO THINK

  Let me return to the early development of verbal matrices.

  From about the eighteenth month onward, the child acquires new words at a faster rate; about the same time it begins to correlate single words into word-sequences, and later on into sentences. The earliest word-sequences are again produced by spontaneous, undirected vocal activity, in which meaningful words alternate with strings of nonsense syllables and with words imitated but not yet understood. Some children have the uncanny gift of imitating the phonetic patterns of adult speech so well that from the next room their babbling sounds like a meaningful monologue. If we are to believe Bertrand Russell, his daughter at the age of eighteen months, 'when supposed to be sleeping, was overheard saying to herself: "Last year I used to dive off the diving-board, I did." Of course, "last year" was merely a phrase repeated without understanding. . . .' [1] This philosophical comment seems to imply, oddly enough, that the remainder of the sentence was pronounced with understanding.

  Frequently the first meaningful string of words refers to a sequence of events. Fenton's unusually precocious son, also at the age of eighteen months, 'uttered soon after seeing his father climb into an automobile with another man and drive away, the words, "Daddy, school, man, auto".' [2] This sounds exactly like the 'picture-strip language' of the primitive (as Kretschmer called it [3]) -- the unrolling of a visual sequence, where each single word symbolizes a complete event. 'The speech units of the child belong to no single class of words because they are (i.e. stand for) not single words but sentences.' [4] Even the first verbal labels have a not merely denotative, but an operative character; they do not refer to objects in vacuo, but to 'action-objects' [5], that is to say, to the functional relevance of the object to the child. 'Chamberlain, Tracy, Dewey, Binet and othen have shown that the child's symbols are action-words, i.e., their content is action. There is also practically universal agreement on the fact that the first symbols of the child are in reality word-sentences designating action and object or subject, or all three at once.' [6]

  This, of course, is again a far cry from any S.-R. theory of language learning. The command TURN RIGHT is a simple and definite verbal stimulus, which, one would think, should be easy to associate with a definite motor response; but many children and some normal adults have considerable difficulty in distinguishing between right and left. This fact is used by Hebb as an elegant proof that the concept (of right- left-sidedness) precedes the word for it: 'The child can very readily learn at the age of three that "right" and "left" each refers to a side of the body -- but ah me, which one? . . . What is set up first is a conceptual organization. By the age of six the word "right" clearly and immediately means sidedness to the child. A considerable conceptual elaboration has already occurred, and the stimulus effectively arouses that structure; but it arouses no prompt, specific response. . . . With such facts, it becomes nonsense to explain man's conceptual development as exclusively consisting of verbal associations.' [7]

  The point becomes clearer when we realize that between the ages of two and four many children are equally confused about 'up-down', 'back-front', etc.; but they only confuse opposites with each other, never 'left' with 'up' or 'front' with 'down'. [8] This shows that the child forms at an early stage sensory-motor matrices of 'up-downness', 'left-rightness', etc.; and that the verbal labels acquired later on may become attached to the matrix as a whole, before directions within the matrix are verbally discriminated. This does not mean, of course, that abstraction preceded discrimination, since both are aspects of the same process; it only means that there is often delay and confusion in the verbal labelling of pre-verbal concepts.

  This brings us to the central problems of the evolution of symbolic thought: abstraction and concept-formation.

  Abstraction, Discrimination, and Transfer

  In common usage 'abstract' thought is regarded as a specifically human faculty, and more particularly as a prerogative of the scientific mind. The Concise Oxford defines abstraction as 'the process of stripping an idea of its concrete accompaniments'. There are, however, obvious analogies between the abstraction of ideas -- the formation of concepts -- and perceptual generalization -- the extraction of invariant features, stripped of their accidental accompaniments, from varied situations. It is therefore frequently asserted or implied that abstract thought is merely the extension, along a continuous scale, of selective processes which operate already in the rat and even lower down. I shall first briefly recapitulate those aspects of the abstractive process which are indeed continuous with animal learning -- until we arrive at the point where the line again breaks. I shall again use the term 'generalization' in the sense indicated above -- and not in the sense of 'spreading of responses' (see p. 537 ff.).

  The continuum of abstractive processes can be extended down into the inorganic domain. The green-grocer's balance abstracts from a pound of peaches and a pound of potatoes the one feature which it has been programmed to recognize as relevant, their weight. A barometer, programmed to compensate for variations of temperature, can be regarded as a model for perceptual size-constancy -- the apparatus programmed to compensate for distance. The same barometer used as an altimeter may serve to illustrate the reversible figure-ground relation. Each of these programmes is represented by a code which determines which type of stimulus should be regarded as relevant and abstracted from experience, and which not.

  Built-in perceptual analysers for the recognition of species-specific sign-releasers may be described as the result of the phylogenetic abstraction of biologically relevant stimulus-patterns. We may regard this abstractive power, with Lashley, as 'one of the primitive basic functions of organized nervous tissue'. [9]

  Turning to acquired behaviour, we must distinguish between learning under natural and artificial conditions. In both cases the animal must learn to abstract relevant patterns from its environment as a conditio sine qua non of properly reacting to them.
With animals in their natural habitat this happens either during maturation, or, in adult animals, after a few repetitions of a certain kind of experience. Since the animal attends only to stimuli which are biologically relevant to it -- to which its perceptual organization is 'attuned' -- the learning process consists essentially in the sharpening and modification of its built-in perceptual analysers. In the laboratory situation, however, the animal must in the first place readjust to an artificial universe, in which stimuli which it would normally treat as irrelevant, become all-important. When this revaluation is well under way, the cat in the box will begin to abstract from its bewildering enviroament the significant patterns of loops and bolts in general, regardless of variations in their position and shape. But this will be a slow, repetitive process because abstraction is a function of relevance, and relevance in this case must be inferred by induction; the cat is caught in a vicious circle. Even so, its powers of abstraction are considerable -- see the perfunctory lick reduced to a symbolic performance (p. 570). The rat in the maze is in a more favourable position: it likes exploring highways and byways; and the cognitive map which it abstracts from a number of concrete tries provides it with a matrix of remarkable plasticity -- see Lashley's extirpation experiments, pp. 458 f.

  The child's pre-verbal concepts are derived from abstractive processes which form a continuous series with animal learning -- from the early development of its perceptual constancies and its 'sensory-motor intelligence' to the crystallization of 'object concepts'. The child learns to recognize its mother before it learns the word 'ma'. Unlike the gosling whose nervous system is ready, a few hours after its birth, to receive the imprint of the mother-goose all in one piece, the helpless infant must slowly, and perhaps painfully, abstract the concept of its mother as a stable, unvarying entity from a series of her very varied appearances, all different in shape, and disconnected in time. The breast first, more tactile than visual, monopolizing the whole perceptual field; a fully-clad figure with conceded breasts, standing or bending over; a series of faces appearing at intervals, never looking quite the same -- such must have been the fragmentary, concrete experiences out of which the unitary mother concept was extracted in the first three weeks of our existence -- if only we could remember.

  However, owing to the nursing mother's exceptional importance to, and sustained bodily contact with the child, the formation of the mother-concept must also be regarded as an exceptional process in which several successive stages are telescoped into one. Other object and action concepts emerge more slowly and hesitantly, even if the object is a person. The adult's awareness of other people's personal identity is based on empathy -- which is a projective phenomenon, in some respects comparable to the stroboscopic effect. But in the babe the experience of its own identity is still hazy, and so is its awareness of the personal identity of others. When a feature F, which is important to the child, is common to several individuals whose other attributes are less important to it, then the child will abstract, conceptualize, and name that shared feature in preference to the collection of features which constitutes each individual's personal identity. This can happen even if one of the persons concerned is the father -- at the stage before the father becomes functionally important to the baby. A baby is often taught the word 'da-da' before it has learned to recognize its father -- except for some vague features such as largeness or bulkiness, which are equally found in other appearances. As a result, it applies the label 'da-da' 'widely and often embarrassingly to large individuals of all shapes, sizes, ages, and colours'. [10] Abstraction and discrimination are guided by relevance; and the relevant experience in this case is some feature of dada-ness shared by lots of visitors who -- their colour, age, etc., being as yet irrelevant -- are not discriminated as individuals. It is the same with the gosling, which in the first stage of imprinting follows all human shapes, and only later on confines its attention to the keeper. During the first stage individual differences in appearance are seen, but not 'taken in' and retained, because they are not yet relevant to the gosling; during the second stage the distinguishing marks become relevant became the gosling has learned that keepers produce food, while other humans do not. Thus the sharpenings and modifications of discriminatory codes are the result of changes in the scales of relevance. In the first stage of development, all 'dada-ish' visitors to the baby -- and all human beings to the gosling -- are regarded as 'for all intents and purposes the same thing'; in the second stage they no longer are, because the intents and purposes have changed.

  I have laboured this point at the risk of repetitiveness because once accepted, it allows us to dispense with terms like transfer, spreading, 'generalization' (in sense b, cf. p. 537), and 'association by similarity' -- and to close the lid of the semantic Pandora box. Take, for instance, Watson's famous experiment intended to establish a 'conditioned fear reflex' in an infant eleven months of age, by striking an iron bar with a hammer each time the child touched its pet animal, a white rat. After this was repeated several times within the span of a week, little Albert responded with fear, crying, etc., whenever the rat was shown to him. But he also showed fright-reactions in varying degrees to rabbits, fur coats, cotton wool, and human hair -- none of which had frightened him before. Watson concluded, and a number of textbooks with him, that the conditioned reflex had 'spread' or been 'transferred' to all furry things -- words which have the connotation of motion in space, conveying the image that the child had somehow lifted its fear-reaction from the rat and put it down on the piece of cotton wool. In fact the idea of the 'spreading' of the conditioned response was originally derived from Pavlov's notion of the irradiation of excitatory processes spreading across the cortical tissues. This physiological theory of Pavlov's has long been abandoned [11], but its unconscious echoes still haunt the laboratories. On a metaphorical level there is of course no objection to saying that the child's dread had spread from rats to cotton wool, but on the technical level the concept of 'spreading' has no explanatory value, and has been the source of endless confusion. (The same applies to 'transfer', and to 'generalization' used as a synonym for spreading.) Instead of pretending that the child has shifted its response, along a so-called generalization gradient, from rat to cotton wool, we should say that it has abstracted the tactile quality of furriness, and recognized it in the cotton wool. And its reaction to it was perfectly logical, because Watson had taught it that furriness always signalled the dreaded bang. It would of course be nonsensical to pretend that this child of eleven months was incapable of seeing the difference between rabbit, cotton wool, and human hair. But at that time the tactile quality alone was relevant, and with regard to that, rabbit, rat, and fur coat were all 'the same thing' (cf. pp. 537 ff.).

  We can observe the operation of the same principles on the verbal level at a more advanced age. Stern's daughter, Hilda [12], at the age of nineteen months, had been in the habit of pulling at her parents' noses; when she discovered that the tips of their shoes (before having leaned the name 'shoe') offered the same satisfactory opportunities, she promptly named them 'noses'. Again, instead of speaking of 'transfer', etc., we shall say that the child recognized an abstracted quality that could be found in various objects. Koffka mentions a boy of twenty-six months' to whom "la-la" first meant song or music; later when he heard a military band, it meant soldier, and finally all kinds of noises, including sounds like claps and thuds. . . . Another little boy uttered the word "atta" at the end of his eleventh month whenever anything disappeared -- when a person left the room or when a light was turned out.' [13]

  The first object- and relation-concepts to be abstracted and named are those which have the greatest functional relevance to the child. At that stage no verbal distinctions are made between objects and attributes, between qualities and things in which these qualities are vested, between nouns, adjectives, and verbs. Lala refers to music; it functions as a collective name for bells, soldiers, instruments; and if we feel that Lala is a silly name for a soldier, but regard 'r
edcoat' or 'poilu' as reasonable, then only because wearing a red coat or being unshaven appear to us more relevant martial qualities than making music. When the child learns to use words, the functionally most important aspect of an as yet unnamed object or event will provide its name; the less important aspects of secondary qualities are then relegated to the role of 'going together with it' -- its 'attributes' or 'parts' or 'functions'. But the criteria of relevance change with age; when Stern's little girl learned to walk, shoes acquired a more important function than that which they shared with noses; and with it a new name.

  Of course the visual experience of Watson's baby is different when it touches a rat and when it touches a fur coat; but it is the touch that matters, the visual aspect is irrelevant, and accordingly, in its overt reactions, the baby does not discriminate between the two. Similarly, the baby mentioned by Koffka saw the difference between the shapes of its Teddy bear and of its stuffed rabbit; but this difference is irrelevant at the early age when all that matters is manipulating a soft toy; hence the name 'dolly' was sufficient to symbolize the whole class, and no motivational need arose for explicit verbal discrimination. Adults behave much the same way. The Eskimoes have several words for various kinds of snow where we have only one; but Malinovski's savages had only one word, manna wala, for all insects and birds -- except those that could be eaten. Building workers, who shift from job to job, cannot be bothered to learn one another's names; all electricians are 'Sparks' and all carpenters 'Chippies'. Elderly ladies addicted to romantic novels from the lending library feel that the names of authors are irrelevant; all that matters is that it should be a 'nice book'.