Up to the present time, theories of sense perception have taken for granted that perception depends wholly on sensations that are specific to receptors. I have called these theories of sensation-based perception. The present theory asserts the possibility of perceptual experience without underlying sensory qualities that are specific to receptors and I have called this a theory of information-based perception. It is a new departure.
The various theories of perception constitute a main branch of the history of psychology, and they have been described by Boring (1942). They need not be reviewed here but it is worth noting the main issues over which they divided, since we should ask how the new proposal deals with them. The liveliest issued, now centuries old, was that between nativism, and empiricism. More recently, another issue has been raised by Gestalt theory in opposition to elementarism.
Consider first the debates between the nativists and the empiricists. One aspect of the controversy was a purely theoretical issue; whether the human being can be said to have a mind at birth -any sort of innate rational capacities or any basis for knowledge before the fact of actual perceiving- or whether, on the other hand, the infant starts life with nothing but a capacity for meaningless sensations and only learns to perceive the world by means of memory and association after an accumulation of past experience.
Another aspect of the controversy is a difference of emphasis not of theory: whether to stress the influence of heredity on the development of perception or the influence of learning. Since both kinds of influence are known to have some effect, the decision is not one between supposedly logical alternatives.
If the theory of information-based perception is accepted, the first controversy becomes meaningless and the logical issue can be thrown out of court. The second controversy is still meaningful, but it takes a new form. The perceptual capacities of the newborn animal or human, for getting information become a matter for investigation. The relative proportions [p.267] of the unlearned and the learned in perception might be expected to depend on the degree of maturity of the infant at birth, which in turn depends on his species and on the kind of environment the young of his species have been confronted with during evolution.
Consider next the question of whether perception was compounded of elements or organized into structures. The empiricists argued for learning or association as the only organizing principle in perception; the Gestalt theorists argued for autonomous field-forces in the brain as the organizing principle. The issue has not been resolved. According to the theory here proposed, this issue also disappears for the neural inputs of a perceptual system are already organized and therefore do not have to have an organization imposed upon them -either by the formation of connections in the brain or by the spontaneous self-distribution of brain processes.
The evidence of these chapters shows that the available stimulation surrounding an organism has structure, both simultaneous and successive and that this structure depends on sources in the outer environment. If the invariants of this structure can be registered by a perceptual system, the constants of neural input will correspond to the constants of stimulus energy, although the one will not copy the other. But then meaningful information can be said to exist inside the nervous system as well as outside. The brain is relieved of the necessity of constructing such information by any process -innate rational powers (theoretical nativism) the storehouse of memory (empiricism), or form-fields (Gestalt theory). The brain can be treated as the highest of several centers of the nervous system governing the perceptual systems. Instead of postulating that the brain constructs information from the input of a sensory nerve, we can suppose that the centers of the nervous system, including the brain, resonate to information.
With this formula, an old set of problems for the psychology of perception evaporates and a new set of problems emerges. We must now ask what kinds of information pickup are innate and what are acquired? What is the process of information pickup? How are the facts of association to be reconciled with the formula? The facts of so-called insight? What is the relation of perceiving to remembering in the new approach? The relation of perceiving to recognizing? To expecting? How is the detecting of information that has been coded into language related to the detecting of information that has not? These questions will be taken up in order.
What is innate and what acquired in perception?
The theoretical issue that divided nativism and empiricism was whether the interpretation of sensory signals did or did not presuppose inborn categories of understanding or "innate ideas." The empiricists wanted to [p.268] show that perceiving could be learned -all of it including the perception of depth. Neither camp ever doubted the assumption that sensations were innate, i.e., that the repertory appeared full blown at birth when the sense organs began to function. This abstract issue can now be disregarded, but a concrete question remains: Considering the infants of a given species, what mechanisms of detection appear at birth and what others depend on learning?
The theoretical assumption that sensations are innate, incidentally, can now be examined. It seems very dubious. Perhaps men learn to experience visual sensations, for example, to become aware of the field of view, or even sometimes to notice the excitation of receptors.
The concrete question of innate and acquired mechanisms in perception is not a two-way issue, for we now know there are intermediates between what is inherited and what is acquired. Pure genetics is one thing; pure learning is another thing; but in between there are types of development that we call growth or maturation. The anatomy and basic physiology of the organs of perception depend mainly on genetic factors as determined by evolution. The maturation of the perceptual systems depends on genetic and environmental determiners in concert. The education of the perceptual systems depends mainly on the individual's history of exposure to the environment. So there are really three questions: How much does perceiving depend on organs? How much does it depend on growth? How much does it depend on experience?
The answer to the first question has already been suggested in Chapters 4 through 8. The working anatomy of the vestibular organs, the ears, the ingestive equipment, the appendages, and the eyes has been described for man, and the evolution of these structures has been outlined so far as this is known. The organs with their receptors set limits on the kinds of stimulus information that can be registered. The five modes of attention, listening, smelling tasting, touching, and looking are specialized in one respect and unspecialized in another. They are specialized for vibration, odor chemical contact, mechanical contact and ambient light, respectively, but they are redundant for the information in these energies whenever it overlaps. Their ways of orienting, adjusting, and exploring are partly constrained by anatomy, but partly free. The basic neural circuitry for making such adjustments is built into the nervous system by the time of birth but it continues to develop in man for a long time after birth.
The answer to the second question has been suggested, but a fuller knowledge of how the perceptual systems develop in the child over time depends on evidence that has been accumulating only recently. We need to know more about overt attention, as in looking, listening, touching, and so on and more about the inner central nervous resonance to selected inputs that also occurs. In this country, experiments are beginning to [p.269] appear on the growing ability of infants to fix their eyes on certain kinds of visual structure (Fantz, 1961). Studies are being made on those features of an optic array that demand notice such as the information for a human face (Ambrose, 1961) or the information for a "visual cliff" (Walk and Gibson, 1961) or the information for "imminent collision" (Schiff, 1965). In Europe Jean Piaget has for many years pursued the study of perceptual development (e.g., Piaget and Inhelder, 1956), but his emphasis is on the inner intellectual aspect of perception. He inclines to the belief that the child constructs reality instead of detecting information. His experiments show, however, that the ability to attend to the higher-order features of objects and events develops in graded stages. At least the results can be interpreted in terms of information. In any case it seems to be true that the child cannot be expected to perceive certain facts about the world until he is ready to perceive them. He is not simply an adult without experience, or a sentient soul without memory. The ability to select and abstract information about the world grows as he does.
The answer to the third question, the extent to which perception depends on experience or learning in the theory of information pickup, is this: it does so to an unlimited extent when the information available to the perceiver is unlimited. The individual is ordinarily surrounded by it; he is immersed in it. The environment provides an inexhaustible reservoir of information. Some men spend most of their lives looking others listening, and a few connoisseurs spend their time in smelling, tasting, or touching. They never come to an end. The eyes and ears are not fixed-capacity instruments, like cameras and microphones, with which the brain can see and hear. Looking and listening continue to improve with experience. Higher-order variables can still be discovered, even in old age. Getting information to the receptors becomes troublesome when the lens of the eye and the bones of the ear lose their youthful flexibility, but higher-order variables in light and sound can still be discovered by the artist and musician.
However, this is not the kind of learning that the theory of association, or of conditioning, or of memorization, has been concerned with. It is not an accrual of associations, an attaching of responses, or an accumulation of memories. Perceptual learning has been conceived as a process of "enrichment," whereas it might better be conceived as one of "differentiation" (Gibson and Gibson, 1955). What can this differentiation consist of?
The Probable Mechanism of Learning to Perceive
Despite the ancient doctrine that sensations left behind ideas in the mind, or the modern version that they could become reconnected with [p.270] responses in the brain, there has always been plenty of experimental evidence to suggest a different sort of learning. This neglected evidence was surveyed and reinterpreted some years ago by Eleanor J. Gibson (1953). Even the supposedly sensory correspondence between physical intensities and phenomenal brightnesses and loudnesses has been shown to improve with practice in making comparisons. Similarly, the psychophysical correspondence between physical frequencies and phenomenal qualities of pure color and pitch improves with practice. Even the detection of physical separations of points on the retina and the skin, supposedly basic sensory acuities, can get better. When patterns of intensity, frequency, or separation are presented to an observer, learning is the rule, for patterns may carry information. A great number of psychophysical experiments have shown decreasing errors in discriminating, estimating, detecting, and recognizing even when the observer is kept in ignorance of his errors. The rule holds for every department of "sense." The author of this survey concluded that the observer learns to look for the critical features, to listen for the distinctive variations, to smell or taste the characteristics of substances (perfumes or wine) and to finger the textures of things (wool or silk). Both she and I now consider this an education of attention to the information in available stimulation.
This increase of discernment is not confined to the detection of finer and finer details. The span of attention is increased with practice. It can (within limits) be enlarged in scope. It can also be extended in time. A pilot, for example, can be trained to keep track of a whole array of aircraft instruments, and a production engineer can be trained to watch over a long sequence of mechanical operations if each episode is part of a whole. This increase of the span of apprehension over both space and time is very suggestive. It is probably a matter of detecting progressively larger forms composed of smaller ones, and progressively longer episodes composed of shorter ones. The spatial relations in an array, and the temporal relations in a sequence, permit the information to be taken in progressively larger and longer units or "chunks." One can finally grasp the simultaneous composition of a whole panel of instruments or a panorama, and apprehend the successive composition of a whole production line or a whole symphony. Note that this extension and protension of grasp is not inconsistent with the concentration of attention on smaller details of an array, or on briefer details of an episodic sequence.
The "differentiation theory" of perceptual learning proposed by Gibson and Gibson (1955) was programmatic at the outset, but the mechanisms for this learning are becoming clearer. The process is one of learning what to attend to, both overtly and covertly. For the perception of objects it is the detection of distinctive features and the abstraction of general properties. This almost always involves the detection of invariants under changing [p.271] stimulation. The dimensions of transformation are separated off, and those that are obtained by action get distinguished from those that are imposed by events (Chapter 2). The exploratory perceptual systems typically produce transformations so that the invariants can be isolated (Chapter 12). And the action of the nervous system is conceived as a resonating to the stimulus information, not a storing of images or a connecting up of nerve cells.
The "resonating" or "tuning" of a system suggests the analogy of a radio receiver. This model is inadequate because there would have to be a little man to twiddle the knobs. A perceiver is a self-tuning system. What makes it resonate to the interesting broadcasts that are available instead of to all the trash that fills the air? The answer might be that the pickup of information is reinforcing. This is essentially the answer that Woodworth suggested twenty years ago, in a paper on the "reinforcement of perception" (1947). Clarity in itself, he asserted, is good, is valued. A system "hunts" until it achieves clarity. The process can occur at more than one level. First, the pickup of information reinforces the exploratory adjustments of the organs that make it possible. And second, the registering of information reinforces whatever neural activity in the brain brings it about. We know something about the adjustments -for example, the accommodating of the eye where the clarity of detail is somehow "satisfying" to the ocular system. We do not know much yet about the neural action of resonance at higher centers, but it too may prove to be the reaching of some optimal state of equilibrium. If the neurophysiologists stopped looking for the storehouse of memory perhaps they would find it.
A perceptual system, to repeat, is not composed of an organ and a nerve. The nervous system is part and parcel of any perceptual system, and the centers of the nervous system, form lower to higher, participate in its activity. Organ adjustments are probably controlled by lower centers, selective attention by intermediate centers and conceptual attention by the highest centers.
The elaboration of this theory and the marshaling of the evidence for it is too much for this chapter, or for this book. Another book is needed. It will be published under the title, Perceptual Learning and Development, by Eleanor J. Gibson.
How are associations between events detected?
Psychologists have become accustomed to thinking of an association as something that is formed between two sensory impressions or between a [p.272] sensory impression and a response. They realized, of course, that there had to be a physical conjunction of events -fire and smoke, for example- before the psychological association could be formed, but this was not what they were interested in. Let us consider, however, the fact of ecological associations, as distinguished from the formation of associations. The result of this fact is an invariance of stimulus combinations. Brunswik considered the ecology of stimulus combinations (1956), but he treated them only as probabilities of sense data. To the extent that a fire always conjoins an optical flame with an acoustic sound, a cutaneous warmth, and a volatile odor, the combination is invariant and constitutes a stimulus of higher order; more exactly, each component contains the same stimulus information (Chapter 3, p. 54). If a peach always yields a certain color, form, odor, texture, and sour-sweet quality, the discriminated features are all characteristic of the same thing and constitute a single combination (Chapter 8, p. 137). The act of perceiving a fire or a peach then, might just as well be considered the pickup of the associated variables of information as the associating of sensory data. Two things are necessary: the dimensions of quality must have been differentiated, and the invariant combinations of quality must be detected. The formation of associations is not necessary.
Can the classical conditioning of responses be explained without resorting to the theory of association? Sign learning, at least, can be subsumed under the theory of information pickup. Consider Pavlov's dog isolated in a cubicle containing a food tray and a bell. The rule of this special environment was that whenever the bell sounded, food appeared. The dog in the cubicle soon began to salivate to the sound. The latter stimulus is then said to be conditioned (the sight and smell of food being associated with it) and the response of salivation is said to be conditioned to it. We say that a new stimulus-response connection has been formed -that the dog has a new stimulus for his old response of salivating or a new response for his old sensation of a bell. But we might as well say that the dog has learned to detect the bell-food invariant in the cubicle situation. As long as Pavlov chose to make this improbable sequence a law of the cubicle (and only so long as he did), the dog might be expected to detect it.
What about the instrumental conditioning of responses? We must now consider Skinner's rat isolated in a box containing a lever and a food cup (1938). Skinner had created this little world (perhaps in six days, resting on the seventh) so that depression of the lever caused delivery of a food pellet. In order to detect this strange invariant, the rat had to behave before he could perceive, but in the course of exploration the utility of the lever became evident: it afforded food. When Skinner made the law merely probable instead of certain, or willfully abolished it, the rat's [p.273] attention to the food affordance of the lever still persisted. The rat would continue to press the lever long after it had been disconnected from the food magazine, and a single success would send him off again.
What about the learning of nonsense syllables? Surely, you may say, this is a process of forming associations. Even here, it can be argued, Ebbinghaus required the learner (usually himself) to perceive the pairing of sequence YOK and LIF in an arbitrarily created list. This was not even an invariant that held for the laws of discourse, much less one holding for the world outside his laboratory, but it was an invariant of the task nevertheless.
Learning by association is defined in stimulus-response theory as an increase in the capacity of a certain stimulus to evoke a certain response the increase having been produced by associating the stimulus with another one that regularly evokes the response. This formula takes no account of stimulus information. In perception theory, at least in the kind being advocated, the response of interest is that to the association, not to either one of the stimuli alone. In short, learning by association becomes the learning of associations.
What is learning by insight?
Ebbinghaus, Pavlov, and Skinner have all given us experimental methods for studying learning by association. Köhler's (1925) observations on the learning of lifelike tasks by apes, however, did not fit into the theory of association. A famous example is the chimpanzee in a barred cage with food set outside his reach and a stick behind him. After many vain attempts, the animal suddenly turns, seizes the stick, and rakes in the banana. The animal is said to have perceived the relations between the elements of the situation and to have learned by insight.
The explanation offered for the chimp's perception was that a spontaneous reorganization of his phenomenal field had occurred which included the banana, the stick, the bars, and his body in one configuration. But again, a different interpretation is possible if the hypothesis of stimulus information is accepted, and this is foreshadowed in Köhler's description of the ape's behavior. Conceivably what he did was to perceive or notice the rake-character of the stick. This object, by virtue of a certain thickness and length, was graspable and reach-with-able. The information for its useability was available in the ambient light. There is no need to postulate reorganization in the brain -only perception of a fact.
This assertion about the useability of the stick does not imply that the chimp had any innate idea that a certain thickness was graspable or that a certain length was reach-with-able. The detection of these meanings [p.274] emerges, no doubt, from grasping (or having grasped) and from reaching (or having reached). The perceiving of rake-character may have developed slowly, after much primate manipulation. The suddenness of insight has been justly questioned (Thorpe, 1956, Ch. 6). The point is that these meanings do not consist of the memories of past manipulation, or of the acquired motor tendencies to manipulate. The acts of picking up and reaching with reveal certain facts about objects; they do not create them.
The hypothesis of the "invitation qualities" of objects, their valences or what they afford, was central to Gestalt theory, especially as developed by Lewin (1936), but the phenomenal field in which they appeared had an uncertain status, neither wholly internal nor wholly external. If these valences are taken to be invariants of stimulus information, the uncertainty disappears. The stick's invitation to be used as a rake does not emerge in the perception of a primate until he has differentiated the physical properties of a stick, but they exist independently of his perceiving them.
The invitations or demands of one animal to another, the affording of sexual partnership, for example, are usually specified by color and shape. But often, as if this were not enough information, the availability of a mate will be advertised by special movements called expressive. The optical transformations specify the fact, and seem to be registered with little previous experience. Displays of this sort are called "releasers" for instinctive behavior (Tinbergen, 1951), but it should be noted that they constitute stimulus information.
Insight vs. Association
The controversy over learning by insight as against learning by association is full of complications and is too big a subject for discussion here. We might, however, consider the physiology of the two processes. Insofar as the Gestalt theorists thought of insight as a neurological process of organization (e.g., Köhler, 1929), their theory was similar to that of the stimulus-response psychologists who thought of association as a neurological process of reinforcement (e.g., Hull, 1943). That is, both theories started from sense data, although they differed as to the kind of neural interaction ensuing. But insofar as the Gestalt theorists recognized the prior organization of stimuli, insofar as they acknowledged the "seeing" of structure (e.g., Wertheimer, 1945), their theory was similar to the present one. They did sometimes think of insight as detection. But they could never quite bring themselves to assume that environmental stimulation always has structure. The Gestalt theorists failed to realize that even dot patterns or inkblots cannot be wholly "unstructured." Hence their emphasis had to be on a hypothetical process that imposed structure on stimulus inputs.
What is the relation of perceiving to remembering?
All theories of learning by association presuppose some kind of central enrichment of an impoverished input to the nervous system. The supplementation, no matter how conceived, is supposed to depend on memory, that is, on some cumulative carryover of the past into the present. It may be conceived either as an accumulation of nervous bonds or connections, or of images or engrams, but at any rate an accumulation of traces in some sense of the term.
Lashley sought to discover the physiological basis of memory during a long career of investigation. But he had to conclude in the end that "it is not possible to demonstrate the localization of a memory trace anywhere within the nervous system" (1950, p. 477). Neither connections between neurons nor between images impressed on the tissue were consistent with the results of his experiments. The "search for the engram," as he put it, had failed. He could only suggest that "the learning process must consist of the attunement of the elements of a complex system in such a way that a particular combination or pattern of cells responds more readily than before the experience" (p. 479). This hypothesis of tuning or resonance implies something quite different from the accumulation of traces. When it is combined with the hypothesis of information pickup, it suggests a surprising possibility -that learning does not depend on memory at all, at least not on the re-arousal of traces or the remembering of the past. Let us follow up this possibility.
Hebb's Theory of Reverberation
Hebb, a student of Lashley, conceived of a way in which the brain might resonate or reverberate, described in a book called The Organization of Behavior (1949). But the reverberation was supposed to occur in the cortex, and the aim was to explain the awareness of a visual form, say a triangle, together with the engram of such an experience. Hebb was influenced by the theory of an isomorphism between visual form and cortical form, the notion that the firing of nerve cells must somehow be like consciousness. The resonance of a retino-neuro-muscular system at various levels to the information available in optical structure, to the variables of form but not the form as such, is quite different form Hebb's reverberating circuits. Only the concept of a circuit is the same. But both theories stem from Lashley.
The essence of memory as traditionally conceived is that it applies to the past, in contradistinction to sense perception, which applies to the [p.276] present. But this distinction is wholly introspective. It depends on the feelings of "now" and "then," not on the facts of life. The experience of "now" is the result of attention to the observer's own body and to the impressions made on it -to sensation, not perception. Information does not exist exclusively in the present as distinguished from either the past or the future. What is exclusively confined to the present is the momentary sensation. The stream of consciousness as described by William James (1890, Ch. 9, 15) exhibits the travelling moment of present time, with a past extending backward and a future extending forward, but this is the stream of self-consciousness, not the process of perception. Physical events conform to the relation of before and after, not to the contrast of past and future. Resonance to information, that is, contact with the environment, has nothing to do with the present.
The ordinary assumption that memory applies to the past, perception to the present, and expectation to the future is therefore based on analytic introspection. Actually, the three-way distinction could not even be confirmed, for the travelling moment of present time is certainly not a razor's edge, as James observed, and no one can say when perception leaves off and memory begins. The difficulty is an old one in psychology and Boring, (1942) has described the efforts to get around it in his chapter on the perception of time. The simple fact is that perceiving is not focused down to the present item in a temporal series. Animals and men perceive motions, events, episodes, and whole sequences. The doctrine of sensation-based perception requires the assumption that a succession of items can be grasped only if the earlier ones are held over so as to be combined with later ones in a single composite. From this comes the theory of traces, requiring that every percept lay down a trace, that they accumulate, and that every trace be theoretically able to reinstate its proper percept. This can be pushed into absurdity. It is better to assume that a succession of items can be grasped without having to convert all of them into a simultaneous composite.
The idea that "space" is perceived whereas "time" is remembered lurks at the back of our thinking. But these abstractions borrowed from physics are not appropriate for psychology. Adjacent order and successive order are better abstractions, and these are not found separate.
Even at its simplest, a stimulus has some successive order as well as adjacent order (Chapter 2, p. 40). This means that natural stimulation consists of successions as truly as it consists of adjacencies. The former are on the same footing as the latter. A visual transient between light and dark is no more complex than a visual margin between light and dark. The information in either case is in the direction of difference: on or off, skyward or earthward. The visual system in fact contains [p.277] receptive units for detecting both kinds of information. It is absurd to suppose that these sequence detectors have to make a comparison of intensity now with the memory of intensity then.
The improvement of information pickup with experience is thus not necessarily the dependence of perception on memory in the commonsense, introspective meaning of that term. The "attunement of a complex system," in Lashley's words, need not entail the reinstatement of earlier experiences, that is, recalling or recollecting. This proposal does not in the least deny that remembering can occur. It denies only that remembering is the basis of learning. Perhaps conscious remembering is an occasional and incidental symptom of learning in the same way the sensations are occasional and incidental symptoms of perceiving.
The ability of the human individual to contemplate parts of his past history is no mean achievement; the experimental psychologist as well as the psychotherapist and the novelist has reason to be fascinated by it; but there is some question whether it has to intervene in the simpler ability to perceive and learn.
The question of whether or not thinking always involves images was a controversy in psychology many years ago (Humphrey, 1951). The weight of the evidence indicated that problem-solving and reasoning could sometimes proceed with no awareness whatever of any copies of previous experience. If it is agreed that one can think without remembering, there is no great step to the conclusion that one can learn without remembering.
The "image" of memory and thought is derived by analogy to the image of art. The "trace" of a percept is analogous to the graphic act. The "storehouse" of memory is analogous to the museums and libraries of civilization. As we observed in Chapter 11, these inventions do make possible the preservation of human knowledge for subsequent generations. But to assume that experiences leave images or traces in the brain, that experience writes a record, and that the storage of memories explains learning, that, in short, the child accumulates knowledge as the race has accumulated it, is stultifying.
What is the relation of perceiving to recognizing?
It has often been pointed out that memory has quite different manifestations. To recognize is not the same as to recall. One can identify the same place, object, or person on another occasion without recalling it. "I recognize you," one says, "but I cannot recall your name, nor where we met." Often there is a mere "feeling of familiarity" or a bare judgment of "same [p.278] as before." Nevertheless, both are considered forms of memory and the theory of traces requires that, even for recognition, the present input must somehow retrieve the stored image of the earlier experience. If the input matches, recognition occurs; if not, recognition fails. This act of comparison is implied by the commonly accepted theory of recognition. There is, however, an alternative theory. It is to suppose that the judgment of "same" reflects the tuning of a perceptual system to the invariants of stimulus information that specify the same real place, the same real object, or the same real person. The judgment of "different" reflects the absence of invariants, or sometimes the failure of the system to pick up those that exist.
The "successions" of stimulation include both non-changes and changes, and therefore the detection of same is no less primary than the detection of different. One is the reciprocal of the other and neither requires an act of mental comparison. This is quite evident in the simplest possible case of recognition, in which one encounter with an object is followed immediately by another, as when one sees an object in two perspectives, or feels it on both sides. The invariants provide for the detection of same thing along with the detection of different aspect. In recognition over a long interval, when encounters with other objects, other places, or other persons have intervened, the attunement of the brain to the distinguishing features of the entity must be deeper and stronger than in recognition over a short interval but the principle need only be extended to cover it.
The same object is usually not encountered in wholly separate places; it is usually met with in the same place, to which one returns after having passed through other places. As we observed in Chapter 10 (Figure 10.10), places are linked by the transformations of vistas and the transitions between them. A vista, it will be remembered, is an array that "opens up" in front and "closes in" behind. Locomotion thus eventuates in a sort of cognitive map, consisting of the invariants common to all the perspectives. This helps to establish the recognition of the objects contained in the perspectives.
The problem of why phenomenal identity usually goes with the same physical thing and of why phenomenal distinctiveness usually goes with different physical things are actually two sides of the same problem. Identification and discrimination develop together in the child as reciprocals, and the experimental evidence shows it. Identifying reactions improve at the same time as discriminative reactions (Gibson and Gibson, 1955). Recognition does not have to be the successful matching of a new percept with the trace of an old one. If it did, novelty would have to be the failure to match a new percept with any trace of an old one after an exhaustive search of the memory store, and this is absurd. [p.279]
What is the relation of perceiving to expecting?
No one has ever been able to say exactly where perceiving ceases and remembering begins, either by introspection or by observation of behavior. Similarly, it is not possible to separate perceiving from expecting by any line of demarcation. Introspectively, the "conscious present," James observed, merges with both the past and the future. Behaviorally, the evidence we accept as showing that the subject of an experiment expects something, food or electric shock, is the same evidence we accept as showing that the animal remembers something or has learned something.
The theory of learning advocated by Tolman (1932) was characterized as a cognitive or perceptual theory. He argued that all kinds of learning consisted of expectations, the actual movements of behavior being secondary, and that the explanation of learning was to be found in the confirming or disconfirming of expectations, not in the reinforcing of responses by reward or punishment. The animal learned what led to what, not reactions. A conditioned stimulus, for example, came to arouse an expectancy of food or shock. The lever in a Skinner box came to induce an expectancy of food in the cup below. The successive alleys of a maze after running through them led to the anticipation of the goal box, which might or might not contain food. The marking on a door in a discrimination box or a jumping stand came to arouse an expectancy of food behind it. This emphasis on the animal's orientation to the future made it plausible to think of behavior in terms of "means-end readiness," and to conceive behavior as purposive.
It has already been suggested how these kinds of learning might be explained without any necessary reference to the future, namely as cases of perceiving or detecting an invariant. The causal connection in these experiments, the contingency, is one created by the experimenter. It was he that designed the conditioning experiment, the box with the lever the alleys of the maze, or the discrimination apparatus, and he that decided what the law of the experimental environment would be. The causal structure of this environment, its machinery, might not be very similar to that of the natural environment of a rat but it was predictable and controllable. The causal law of "what led to what" was present in the situation on repeated trials. If the animal could identify it over the series, he could be said to have learned, inasmuch as his behavior came to be determined by it. Whether or not the animal could fairly be said to expect, anticipate, or imagine the future, he could surly be said to have detected something.
Tolman's confirming of an expectation, it may be noted is similar in principal to what has here been called the discovering and clarifying of information as a consequence of exploratory search. To call the process [p.280] one of predicting an event and then verifying its occurrence makes it seem an intellectual accomplishment and dignifies the rat undeservedly. The rat's perception is more primitive than this.
The apprehension over time of the motion of an object, one might suppose, has nothing in common with the learning that may occur in the event sequences described above. The motion, we say, is simply perceived; remembering and expecting do not come into it. A kitten perceives the course of a rolling ball, an outfielder perceives the trajectory of a batted ball and that is all there is to it. Nevertheless, in a sense, the kitten and the ballplayer expect the ball to continue on a predictable path, and that is why they can both start out on a dead run to intercept it. This foreseeing is much like ordinary seeing, and not much like Tolman's expectancies for it depends on a continuous flow of stimulation. But the two kinds of situation do have something in common. The unbroken continuation of the optical motion is a consequence of the invariant laws of inertia and gravity in physics. The ball continues in a straight line or a trajectory, because of Newton's Laws. The invariant is implicit in the motion. Both the kitten and the ballplayer may have to practice and learn in order to detect it accurately but in a certain sense what they are learning is to perceive the laws of motion.
The experiments of Schiff, Caviness, and Gibson (1962) and Schiff (1965) on optical magnification of a silhouette in the field of view demonstrate that "looming," the visual information for imminent collision, is often detected by young animals who have never had painful encounters with an approaching object. They shrink away or blink their eyes, or otherwise make protective responses without having any reason to "expect" collision by reason of past experience. In this case the visual nervous system is presumably attuned to the information at birth. The behavior of human automobile drivers suggests that there are various degrees of attunement to the foreseeing of a collision when something starts to expand in the field of view.
What is the effect of language on perception?
Both men and animals perceived the environment, but the human perceiver has language while the animal does not. When the child begins to communicate by speech and to practice speaking, he starts on a line of development that makes his knowledge of the world forever different from what it would have been if he had remained a speechless animal. What are these consequences? We might suppose that the effect of language would be to make perceiving easier and better. But it has been argued [p.281] that there is an unfortunate and unavoidable effect which tends to make it distorted and stereotyped instead.
The argument seems to be based on a fact about language, but only one, and not necessarily the most important -the fact that it is a code. Language substitutes words for things. It depends on the lexicon, that is, on a sort of social agreement as to the signals that will stand for certain percepts. Every child must learn the code of his social group, and it is supposed that he learns it by forming associations between things and words, or by acquiring conditioned verbal responses to things. There are of course, many more things in the world than there are words in a language. Not everything can be coded. The verbal responses, it is argued, must therefore categorize or cut up the real world in conventional ways that are necessarily inadequate to its full complexity (Whorf, 1956). If, now, it is further assumed that perceptual identifying is not theoretically separable from verbal naming, then perceiving is perforce limited, as verbalizing is limited, and perception is to that extent distorted.
This line of reasoning presupposes an association theory of perception, assuming that words are utterances (or tendencies to utter, or auditory memories of utterances, or visual memories of writing) and that they have been attached to the stimuli from the world by association. The theory of information pickup, however, starts with a different assumption about words and ends with a different conclusion as to the effect of language on perception. Let us try to pursue the new line of reasoning.
For the child who is learning to use language and at the same time learning to perceive the world, words are not simply auditory stimuli or vocal responses. They embody stimulus information, especially invariant information about the regularities of the environment. They consolidate the growing ability of the child to detect and abstract the invariants. They cut across the perceptual systems or "sense modalities." The words are like the invariants in that they are capable of being auditory or visual or even tactual (as Braille writing is). They even cut across the stimulus-response dichotomy, for they can be vocal-motor or manual-motor. Hence, the learning of language by the child is not simply the associative naming or labeling of impressions from the world. It is also, and more importantly, an expression of the distinctions, abstractions, and recognitions that the child is coming to achieve in perceiving. Insofar as a code is a set of associations, the terms of the code have to be learned by association. But a language is more than a set of associations and the learning of language is therefore more than learning by association.
A language is more than a code because it permits predictions as well as labelings. It has a grammar as well as a vocabulary. So the child's discovery of facts about the world can be predicated in sentences, not simply [p.282] stereotyped in words. Predication can go to higher and higher levels, so the limitations of vocabulary do not set the same limits on the codifying of facts.
The learning of the language code as a vocabulary should be distinguished from the child's learning to consolidated his knowledge by predication. He gets information first by focusing, enhancing, detecting, and extracting it from nonverbal stimulation. Later, the extracting and consolidating go on together. Perceiving helps talking, and talking fixes the gains of perceiving. It is true that the adult who talks to a child can educate his attention to certain differences instead of others. It is true that when a child talks to himself he may enhance the tuning of his perception to certain differences rather than others. The range of possible discriminations is unlimited. Selection is inevitable. But this does not imply that the verbal fixing of information distorts the perception of the world.
In the theory of information pickup, the spontaneous activities of looking, listening, and touching, together with the satisfactions of noticing, can proceed with or without language. The curious observer can always observe more properties of the world than he can describe. Observing is thus not necessarily coerced by linguistic labeling and there is experimental evidence to support this conclusion.
Behavioral theories of perception get their force from the conviction that behavior is practically useful. In a behavioral theory of perception, however, exploratory activities are treated simply as responses. Perception must then be learned by the reinforcing of stimulus-response connections. The conclusion is unavoidable that perception is biased by the needs that motivate practical action, for discrimination serves only the interests of practical action. One should fail to see anything that leads to unpleasant consequences and should see anything that leads to satisfaction. Both psychic blindness and hallucination ought to be common occurrences. But in the theory being advocated, discrimination is itself a kind of useful action -an activity reinforced by clarity, not by punishments or rewards- and autism, or wishful perceiving, ought to be an uncommon occurrence.
The issue between the two kinds of theory can be illustrated by the following question. Does a child distinguish between two physically different things only after he has learned to make different responses to each, names, for example; or does he first learn to distinguish them and then (sometimes) attach names? On the former alternative he must learn to respond to the things; on the latter he must learn to respond to the difference.
From the first alternative it would be predicted that a child should be able to say names correctly before he can say "bigger" correctly; on the latter alternative the reverse would be predicted. The issue is deep and far-reaching. It cannot be compromised or avoided. [p.283]
The Probable Kinds of Development in Learning to Perceive
Associating, organizing, remembering, recognizing, expecting, and naming -all these are familiar psychological processes, and all of them have been appealed to in the effort to explain the growth of knowledge. But all these processes were first conceived as operations of the mind upon the diliverances of sense, and they still carry some of this implication. They have now been examined, one by one, and I have suggested that, as commonly understood, they are incidental, not essential, to the developing process of information pickup. They need to be reinterpreted. The deeper, underlying kinds of perceptual development seem to involve exploration and attention. What can be said by way of summary about the more fundamental types of development?
Differentiating the Range of Possible Inputs
Consider a very simple perceptual system -for example, that for detecting the direction of gravity (Chapter 4). The input of a statocyst is presumably different for every different position of the weight resting on its hair-cells, altering as the animal is tilted leftward, is upright, or is tilted rightward. But a given input of excited hairs constitutes information about the direction "down" only in relation to the other possible inputs of excited hairs. The range of inputs, from a horizontal posture through vertical to horizontal again, defines the meaning of any given input. Consequently the animal's nervous system must have differentiated this range if it is to detect "down" and make compensatory righting reactions. For this, the animal must have been subjected to the range of postures, or perhaps have explored the range of postures. The development might be prenatal, or innate, or even learned, but it must be a development.
The same differentiating of the range of inputs must occur for other perceptual systems as well as the vestibular. The dimensions of variation in the haptic and the visual system, for example, are much more elaborate than are the inputs of a statocyst. Discriminative learning may be required instead of neural growth or maturation. Active testing of the limits of the range may occur. Any perceptual system, however, has to have each of its inputs related to the other available inputs of the system.
Establishing the Covariation of Inputs between Different Systems
The "orienting system," it will be recalled, is actually a redundant combination of vestibular, tactual, articular, and visual information. The input of a statocyst is covariant with the input of the skin, the joints, and the eyes [p.284] whenever a young individual rolls, crawls, walks, or gets about. Consequently there must be another simple type of perceptual development, the registering of the concurrent covariation from different organs. The pull of gravity, the push of the ground, and the sky-earth difference are correlated. The vestibular, haptic and visual inputs are likewise correlated over time. Insofar as this linkage is invariant, the information is the same in all of them, that is, the systems are equivalent. Their inputs are associated, it is fair to say, but learning by association hardly need be assumed.
Covariation in time of differentiated inputs does not necessarily imply a one-to-one correspondence of sensory elements or qualities. Covariant but not coincident inputs from the statocyst and the skin will occur for an individual resting on a slope, as noted in Chapter 4 (Figure 4.3). The "calibration" of the ranges of inputs from different perceptual organs may well be a matter of learning, and it implies information of a higher order.
The learning of concurrent covariations in the external environment, of what goes with what, depends also on the pickup of concurrent covariation of neural input, but this requires that the exterospecific component of the input will have been isolated.
Isolating External Invariants
The perception of the color and layout of surfaces, of the distinctive features of objects, and of their real motions in space implies that the other-produced component of neural input is separated from the self-produced component. This separation is not difficult to explain if one supposes that relational inputs exist along with the anatomical inputs. The transformations of the anatomical pattern of excited receptors have subjective reference; the invariants of adjacent and successive order in the overall input specify the invariants of stimulation and thereby the invariants of the world.
This "constancy" of perception no doubt depends on development insofar as the invariants of input have to be differentiated from one another in the nervous system. But the registering of invariants is something that all nervous systems are geared to do, even those of the simplest animals. The visual perception of "depth," for example is surely not dependent on a gradual process by which the brain learns to interpret local sensations of color. Constancy is learnable in some degree, but not by a process of associating, organizing, or remembering.
Consider the origin of the child's perception of the permanence of objects. Does it have to depend on some kind of intellectual understanding of the causes of the child's impermanent sensations? Piaget (1954) and many others have assumed so. David Hume asserted (1739) that the senses "are incapable of giving rise to the notion of the continued existence of objects after they no longer appear to the senses. For that would be a [p.285] contradiction in terms" (Part IV, Sec. 2). Hume was quite right; the awareness of the continued existence of a thing after it has been hidden by the edge of something else cannot be derived from the visual sensation after it has been wiped out. But it can be explained by the detecting of stimulus information for occlusion, i.e., the property of the transformation that we call "wiping out," which is quite distinct from the transformation that we call "fading out" (Reynolds, 1966). This information was described in Chapter 10 (Figure 10.9). The child must distinguish, or learn to distinguish, between these two kinds of optical transformation in order to perceive when a thing merely goes out of sight and when it vanishes, but he does not have to "construct" reality out of impermanent sensations (Piaget, 1954). Nor does he have to associate tactual sensations with visual ones in order "to understand that the objects in his environment have a continuous and consistent identity entirely detached from himself" (Vernon, 1952, p. 10).
Learning the Affordances of Objects
When the constant properties of constant objects are perceived (the shape, size, color, texture, composition, motion, animation, and position relative to other objects), the observer can go on to detect their affordances. I have coined this word as a substitute for values, a term which carries an old burden of philosophical meaning. I mean simply what things furnish, for good or ill. What they afford the observer, after all, depends on their properties. The simplest affordances, as food for example, or as a predatory enemy, may well be detected without learning by the young of some animals, but in general learning is all-important for this kind of perception. The child learns what things are manipulable and how they can be manipulated, what things are hurtful, what things are edible, what things can be put together with other things or put inside other things -and so on without limit. He also learns what objects can be used as the means to obtain a goal, or to make other desirable objects, or to make people do what he wants them to do. In short, the human observer learns to detect what have been called the values or meanings of things, perceiving their distinctive features, putting them into categories and subcategories, noticing their similarities and differences and even studying them for their own sakes, apart from learning what to do about them. All this discrimination, wonderful to say, has to be based entirely on the education of his attention to the subtleties of invariant stimulus information.
Detecting the Invariants in Events
Along with the discrimination of objects goes the developing discrimination of events. The child learns how things work as well as how they differ. He begins to perceive falling, rolling, colliding, breaking, pouring, tracing, [p.286] and he ends by apprehending inertia, the lever, the train of gears, the chemical change, the electric current, and perhaps the concept of energy. The cause-and-effect relation in these observations becomes increasingly subtle. The simple perception of motion or of collision (Michotte 1963) gives way more and more to what we call inference. Nevertheless there remains an element of perception in the appreciation of even the most abstract law. The physical scientist visualizes atoms or particles; the savage or the child sees spirits or magical rules behind a complex sequence of events (Piaget, 1951), but everyone perceives some kind of invariant over time and change. The information for the understanding of the law in such a case may be of staggeringly high order, but it is theoretically open to observation.
The Development of Selective Attention
Still another probable kind of perceptual development is the acquiring of what might be called economical perception. It is the ability to avoid distraction -to concentrate on one thing at a time in the face of everything going on in the environment- and yet to accomplish as much knowing as possible. To accomplish this perceiving must be quick and efficient rather than slow and contemplative. As a result, the information registered about objects and events becomes only what is needed, not all that could be obtained. Those features of a thing are noticed which distinguish it from everything that is not -but not all the features that distinguish it from everything that it is not.
This has been called the schematic tendency in perception, and it has been much studied in the psychological laboratory. The rule is, I suggest, that only the information required to identify a thing economically tends to be picked up from a complex of stimulus information. All the other available information that would be required to specify its unique and complete identity in the whole universe of things is not attended to.
This rule emphasizes economy in detecting the diagnostic features of things in the structure of stimulation. It does not refer to economy in a process of organization that is supposed to produce structure where none existed. The "minimum principle" in the organization of perception is one of the tenets of Gestalt theory; this is also a minimum principle, but the economy is in a process of selection not one of organization.
"Introduction" (pp. 1-6).
"Chapter XIV: The Causes of Deficient Perception" (pp. 287-318).