If in the process of memory-formation the input is stripped down to bare essentials, recall requires dressing it up again. This seems to be done by some summary drapings, patched out with surviving fragments of picture-strip, plus some fitting garments borrowed from elsewhere -- we all know how often 'vivid details' are incorporated into the recall of experiences to which they do not belong. Imaging involves imagining, which is a flexible skill. It is triggered-off by an impulse of central origin -- a kind of 'excitation-clang' which unlocks 'memory releasers' and sets off the feedback-circuits of 'inferential reconstruction'; as with other plastic skills, two performances are never quite the same.
Summary
I have treated perception, recognition, and memory-formation as a continuous series. The sensory input is screened, dismantled, reassembled, analysed, interpreted, and stored along a variety of channels belonging to different hierarchies with different criteria of relevance. A tune can be stored stripped of timbre, and vice versa. The de-particularization of experience in the process of memory-formation is compensated to some extent by the multiplicity of abstractive hierarchies which participate in the process, and by the retention of 'picture-strips' -- vivid fragments of emotive or symbolic significance.
Central controls and motor activities participate at various stages in the processing of the input; from stimulus-selection in the end-organ and visual scanning, through resonances from the vocal tract, to the interpretation of Klangbild by Wortschatz, and of the seen by the known. Nowhere are 'stimuli' and 'responses' neatly separable; they form hierarchies of loops within loops. The mechanisms responsible for the processing are partly inborn, mostly acquired; their codes have a high degree of autonomy and show their 'self-asserting' tendencies in the tenacity of optical illusions and of 'seeing in terms of'.
The generalization and retention of perceptual forms has an upper limit where symbolic coding must take over to make further progress in learning possible. The ability of man to form 'number percepts' is not significantly superior to that of some birds; memory images are aggregates of relatively simple schematized forms, i.e. of true perceptual elements, held together by cognitive linkages, as the 'sound-pictures' of speech are given coherence by their meaning. They are double-faced entities: complex perceptual Gestalt-wholes which enter as units into the symbolic hierarchy. We must assume that there are analysing devices of this kind active in the nervous system, 'resonators' which are 'attuned' to a certain configuration in the perceptual input, and respond to it by signals in symbolic coding addressed to the higher echelons. All inputs which are equipotential with respect to that configuration -- e.g. 'triangularity' -- are regarded by the analyser as 'the same thing', and reported by the same signal. The process is thus the reverse of the 'spelling out' activites of the motor hierarchy -- which is in some respects comparable to the unlocking of memory-releasers in recall.
'Generalization' and 'discrimination' are complementary aspects of the same process, and will be discussed, together with the ambiguities of 'spreading' and 'transfer', in later chapters.
NOTES
To p. 513. Cf. Polànyi, 1958, p. 364: 'Behaviourists teach that in observing an animal we must refrain above all from trying to imagine what we would do if placed in the animal's position. I suggest, on the contrary, that nothing at all could be known about an animal that would be of the slightest interest to physiology, and still less to psychology, except by following the opposite maxim. . . .
To p. 516. This response is mediated either by resonance (Helmholz's theory) or, more likely, by the locus of maximum hydraulic pressure in the 'travelling wave'.
To p. 517. The latter assumes that in visual perception a spatial 'picture' is projected on to the primary optical cortex, which reproduces the retinal image. But the excitation-pattern in the auditory cortex has no 'contours' separating figure and background, and it would be difficult to imagine 'field currents' created by them.
To p. 533. There are strong arguments against the segmentation of language according to the letters of the written alphabet (cf. e.g. Paget 1930; Ladefoged in Mechanization of Thought Processes, 1959).
To p. 537. There exist of course both innate and acquired preferences for choosing one system of 'coloured filters' rather than another as a criterion of equipotentiality. Two notes an octave apart sound more 'similar' to man and rat than two notes close together. Evidently the nervous system finds it for its own 'intents and purposes' more convenient to regard two frequencies of the ratio 2p:p as more similar than two frequencies of the ratio p:(p - r).
To p. 539. The protracted controversy about the existence of progressive, systematic changes in perceptual traces ('levelling' and 'sharpening') was unfortunately restricted to one type of change only -- the reduction of 'dynamic stress' in the physical trace, predicted by Gestalt physiology -- see Wulf, quoted by Koffka (1935); Hebb and Foord (1954). No psychologist would dare to deny that 'memory plays us false'; but its confidence-tricks are evidently not of the grossly mechanical type, divorced from the subject's living experience, which Köhler's theory of cortical field-processes demanded.
To p. 540. 'Negative recognition' could be called the unconscious variety of Woodworth's (1938) 'schema with correction'.
XI
MOTOR SKILLS
In the process of becoming an expert typist, the student must go through the whole range of learning processes variously classified as as instrumental conditioning, sign-learning, trial and error, rote and place learning, insight. He is, of course, quite unaware of these categories -- which, in fact, overlap at almost every stage. The essence of the process is the step-wise integration of relatively simple codes of behaviour into complex and flexible codes with a hierarchic structure. This conclusion was actually reached (although expressed in different words) in the 1890s by Bryan and Harter [1] -- then buried and forgotten for nearly half a century. Woodworth was one of the few experimental psychologists who kept harking back to the subject. The following is taken from his summary of Bryan and Harter's Studies on the Telegraphic Language. The Acquisition of a Hierarchy of Habits. [2]
The beginner first learns the alphabet of dots and dashes. Each letter is a little pattern of finger movements in sending, a little pattern of clicks in receiving. It is something of an achievement to master these motor and auditory letter habits. At this stage the learner spells the words in sending or receiving. With further practice he becomes familiar with word-patterns and does not spell out the common words. The transition from the letter habit to the word-habit stage extends over a long period of practice, and before this stage is fully reached a still more synthetic form of reaction begins to appear. "The fair operator is not held so closely to words. He can take in several words at a mouthful, a phrase or even a short sentence." In sending he anticipates, as in other motor performances; but in receiving, he learns to "copy behind", letting two or three words come from the sounder before he starts to copy. Keeping a few words behind the sounder allows time for getting the sense of the message.
Let us call these three stages of habit-formation the 'letter', 'word', and 'context' levels. The letter habit is acquired by 'serial learning'. But no chain-response theory can account even for this first step in acquiring the skill -- for the simple reason that the homogeneous dots and the homogeneous dashes of the Morse sequence offer no distinguishable characteristics for the forming of specific S.-R. connections. The letter 'u' is transmitted by dot-dot-dash; the letter 'w' by dot-dash-dash. In terms of S.-R. theory, the finger-movement made in sending the first dot is the initial part-response which triggers the chain, its kinaesthetic sensation acting as a stimulus which calls out the next response. But the correct response to the same stimulus will be either dot or dash; nothing in the nature of the stimulus itself indicates what the next response should be; the response is determined at this and each following step not by the preceding stimulus but by the total pattern. The habit cannot be represented by a linear series: . -- -- > -- -- -- > . -- . it can only b
e represented as a two-tired hierarchic structure:
The related skill of touch-typing was studied by Book [3], who wired his machines to time every move made by the experimental subjects. In this case the letter habit is acquired by 'place-learning' -- the keyboard is hidden from the student by a screen, and he is required to form a 'map' of its layout in his head. This map, one supposes, is structured by a simple co-ordinate-system: the fixed resting position of the ten fingers on the third row of the keyboard; the result is a kind of simple 'maze' with variable target positions. But when, after a certain amount of hit and miss, the letter habit had been mastered:
further practice gave results unexpected by the learner. He found himself anticipating the sequence of finger movements in a short, familiar word. Habits were developing for groups of letters such as prefixes, suffixes, and short words. . . . 'A word simply means a group of movements which I attend to as a whole. I seem to get beforehand a sort of feel of the whole group'. . . . The single letters were no longer thought of and each word became an automatic sequence. . . . Familiar phrases were similarly organized, the thought of the phrase calling out the whole series of connected movements. [4]
Yet even phrases ending with a full stop did not prove to be the highest units. The records showed 'no pauses between phrases' but an even flow; and here, too, 'the eyes [on the text to be copied] were well ahead of the hands' -- to enable the typist to take in the meaning.
As a third example let us consider learning to play the piano (though I could find no textbook references to this not altogether unusual human occupation). The 'letter habit' here becomes a 'note habit' -- hitting the intended black or white key; for 'word' read 'bar' or 'musical phrase'; and so on to more complex integrated patterns. In this case, however, even the lowest unit of the skill -- hitting the right key -- displays considerable flexibility. There is no longer, as on the typewriter, a rigid attribution of each key to one finger; on the piano keyboard almost any finger can be used, according to circumstances, to hit any key; several keys may be hit and held at the same time; and a hard or soft touch makes all the difference to musical quality. (Needless to say, even the typist's motion-patterns must be adaptable to small portables and large office machines, and the starting position of the finger varies according to the preceding stroke. Flexibility is a matter of degrees; a completely fixed response is, like the reflex arc, an abstraction.)
The skill of hitting the correct piano-key is not acquired by establishing point-to-point correspondences, but primarily by practising the various scales; these superimpose, as it were, structured motions on to the keyboard, sub-structured into triads, septims, etc. At an advanced stage, when improvization has become possible, the left hand will learn to accompany the right, which acts as a 'pace-maker' -- a glorified form of the magnet effect in the gold-fish (p. 438); but the left can also act in relative independence, according to the commands of the score. At this level we have approximately the following state of affairs: the visual input consists in two groups of parallel rows (staves) of coded signals, of which the upper series must be referred to the right, the lower to the left hand. In the course of this procedure both rows of signals must be de-coded and re-coded. The symbols on the two rows are usually in different parallel codes ('violin clef' for the right, 'bass clef' for the left). Moreover, there are 'key signatures' -- sharp and flat signs -- at the beginning of a section, which modify the 'face value' of the notes; there are symbols which indicate the timing and duration of notes; and overall instructions regarding loudness, tempo and mood. All these part-dependent, part-independent de- and re-coding operations for both hands must proceed simultaneously, in the psychological present, and more or less automatically.
On an ever higher level, the concert pianist develops a repertory of oeuvres that he can 'trigger off' as units and play by heart -- though some of these units may be an hour long. Once again we must assume that this is done by a combination of several interlocking hierarchies, each articulated into sub-wholes and the sub-wholes thereof.
Then there is improvisation. It need not be creative; the bar-pianist who, half asleep, syncopates Chopin and trails off into some variation of his own, is not a composer; but he has gained additional flexibility -- more degrees of freedom -- in the practice of his skill. And finally there is the creative act: the composer who weaves his threads into new patterns, and the interpreter who sheds new light on existing patterns.
The learning process is, somewhat paradoxically, easiest to visualize as a reversal of the hierarchic sequence of operations which will characterize performance when learning is completed. When the typist copies a document, the sequence of operations is initiated on the semantic level, then branches down into successive lower levels with increasingly specific 'fixed action-patterns' -- 'word-habits' and 'letter-habits '; terminating in the 'consummatory act' of the finger muscles. The impulses arborize downwards and outwards, whereas learning proceeds in the reverse direction: the tips of the twigs of the future tree are the first to come into existence; the twigs then grow together centripetally into branches, the branches merge into the trunk. It strikes one as a very artificial procedure; but the type of mechanical learning we have discussed, where the discrete base-units must be stamped in bit by bit, is indeed an artificial procedure. The difference between this method of learning through trial and error and learning 'by insight' becomes glaringly obvious if you compare what happens during an elementary violin lesson and an equally elementary singing lesson. The choir boy can rely on his innate, multiple auditory-vocal feedbacks -- operating through the air, through his bones, and through proprioceptive sensations from his vocal tract -- to control his voice. But there exist no innate feedbacks between the violin student's cochlea and finger-muscles, to control their motions. No amount of theoretical insight into the working of the instrument can replace this handicap; it can only be overcome by supplementing insight with trial and error. In other words, human beings are biologically less 'ripe' for learning the violin than for learning to sing. If evolution were to produce a super-cricket or cicada sapiens, the opposite may be true.
To put it in a different way: the built-in feedbacks of the auditory-vocal apparatus provide a direct insight into the rightness or wrongness (singing out of tune) of the response; they permit an immediate 'perception of relations' -- which is Thorpe's definition of insight. But once more, this insight is far from absolute: when it comes to professional singing, a heart-breaking amount of drill is required. The pupil is often taught the proper techniques of breathing with his hand on the teacher's stomach -- because his insight into, and control of, his own physiological functions is limited. Verbal instructions are of little help, and are sometimes a hindrance, in the acquisition of muscle skills; to become clever with one's hands, or one's feet in dancing, requires a kind of muscle training which defies classification as either insightful or trial-and-error learning.
I have repeatedly mentioned the mysteries of riding a bicycle: nobody quite knows how it is done, and any competent physicist would be inclinded to deny a priori that it can be done. However, as a two-legged primate, man has an innate 'ripeness' for the acquisition of all kinds of postural and balancing skills such as skating, rock-climbing, or walking the tight-rope; accordingly, the hierarchy of learning processes in the case of the cyclist starts on a higher level of already integrated sub-skills, than in the examples previously discussed. Broadly speaking, the pupil must turn the handle-bar in the direction he is falling, which will make him tend to fall in the opposite direction, and so forth, until he gradually 'gets the feel' of the amount of correction required. This is certainly trial-and-error learning in the sense that errors are punished by a fall; but the trials are by no means random, and the errors are all in the right direction -- they merely over- or under-shoot she mark. The code which is formed by successive adjusments of the neural 'servo-mechanism' is presumably of the analogue-computer type -- and the same applies probably to dancing, skating, or tennis-playing.
/> But once the skill has been mastered and formed into a habit, its integrated pattern is represented as a unit on the next-higher level in the hierarchy, and can be triggered off by a single (verbal or nonverbal) command. To take a more complicated example: the soccer-player must acquire a variety of basic routines of taking command of the ball -- 'stopping' it with foot, thigh, chest, or head; volleying it in flight without stopping; kicking it with the instep, the inner or outer side-wall of the boot; dribbling, passing, and shooting at the goal, etc. When these elementary, yet very complex, techniques have been mastered, each of them will become a self-contained sub-skill in his repertory, and he will be able to decide, in a split second, which of them to employ according to the layout of the field. The decision whether to shoot or pass is based on discrete yes-no alternatives of the digital type; but the execution of the actual move -- shooting, passing, etc. -- seems to require an analogue-computer type of code. A further step down the 'analogue' process of flexing the leg-muscles for a pass of appropriate length is again converted into the digital on-off processes in individual motor units; while on the top level a fluid strategy is converted into discrete tactical decisions.
It would seem that behavioural matrices on every level of a given hierarchy are triggered off by digital-type all-or-nothing impulses; if the matrix is flexible it will function as a digital-to-analogue converter; and will in its turn trigger off sub-codes at certain critical limits as analogue-to-digital converters. But theorizing about the nervous system in terms of computer models is a risky affair, and may yet lure psychology into a cul-de-sac -- as the telephone-exchange model did half a century ago.*