Stimulus predifferentiation
Educational films can be considered as everyday examples of stimulus predifferentiation, in which the individual gets preliminary information to be used in subsequent learning. The student who sees a film describing the various parts of a microscope is likely to be better prepared to learn the requisite skills when confronted with the instrument itself. In laboratory studies of stimulus predifferentiation, the subject is given experience with a particular stimulus situation ahead of time; later he is asked to learn new responses in the same situation. In one illustrative study, subjects first practiced labelling four different lights and then later were asked to learn to press selectively one of four switches, each connected to one light. The rate at which they learned the appropriate pressing reactions was related to how well they had learned to label the lights.
The results of a large number of experiments covering a variety of stimulus predifferentiation techniques suggest that when a learner has an opportunity to become generally acquainted with an environment, he retains some information about its different components that prepares him for learning to make new responses to them. Various explanations have been offered to account for this facilitation; some investigators suggest that the process of labelling enhances the distinctiveness of environmental stimuli for the labeller; others hold that perceptual acquaintance can more sharply differentiate an environment into its component parts for the perceiver or that it may encourage appropriate responses of observing or attending. Nevertheless, no single process has been identified as fundamental in stimulus predifferentiation. Perhaps a number of these processes operate in different combinations from one stimulus-predifferentiation transfer experiment to another, each process representing a different method by which a learner can become familiar with the details of his environment.
Transposition
Another phenomenon that has received considerable attention in theories of transfer of training is called transposition. An initial report of transposition came from a study in which chickens were trained by rewards to respond to the darker of two gray squares. After this discrimination task was learned, the chickens were shown the originally rewarded gray square along with one that was still darker. They seemed to prefer the darkest gray to the square that had been previously rewarded. This finding was interpreted to support the hypothesis that the birds had initially learned to respond to a relationship (what a human being would call the concept “darker”) and that this response to a relationship had been transposed or transferred to the new discrimination. This relational interpretation later was challenged by theorists who offered a formulation to show, on the basis of principles of stimulus generalization, how a response to a relational stimulus could be explained by assuming that organisms do indeed respond to the absolute properties of the stimuli. Both explanations were found to be too simple for the variety of findings obtained with transposition studies. As a result, the interest of many investigators shifted away from demonstrating the relative merits of absolute versus relational interpretations to identifying conditions that seem to influence transposition behaviour. Within this context, newer, more sophisticated formulations have been proposed that consider both the absolute and relational characteristics of the stimuli in transposition studies.
Learning to learn
When people are asked to learn successive lists of words, their performance tends to improve from one task to another so that much less time is commonly required to learn, say, the tenth list than was needed for mastering the first list. This improvement suggests that information beyond the specific content of lists of words is also learned. It would seem as if the subjects are learning how to learn; that is, they seem to be acquiring learning sets, or expectancies, that transfer from list to list to produce continually improving performance.
Some of the most intensive work on learning sets has been carried out with monkeys that were learning how to solve several hundred discrimination problems in succession. In each problem, the monkey learned which one of two objects (for example, a bottle cap and a cookie cutter) consistently contained a piece of food. Although the solution of each successive problem required the animals to discriminate between two previously unfamiliar objects, performance tended to improve on successive tasks; the monkeys made increasing numbers of correct choices on the second trial of each problem as the process continued. Manifestly there was no cue to indicate the correct choice on the first trial of any specific problem. If the animal responded correctly on the first trial, then on the second trial it would only have to choose the same object to be correct thereafter; if the monkey made an error on the first trial, then the other object would inexorably be the one that should be chosen next. During their efforts to solve the first few problems the monkeys were correct approximately half the time on the second attempt to solve each problem. This success increased to an average of 80 percent correct after each animal had solved 100 problems, to 88 percent after 200 correct solutions, and eventually to 95 percent after 300. Thus, after a long series of separate tasks, all of the same type, the monkey’s first response to the next problem usually provided sufficient information for the animal to make the correct choice.
Since each of the successive discrimination problems was different, what actually was being transferred from problem to problem? In these discrimination problems, the monkeys seemed to have several items of information to learn in addition to which one of the two objects contained the rewarding bit of food. The animals apparently had to learn to pay attention to that part of their environment where the objects were placed. To make the correct choice, it would seem that a monkey would have to learn to abandon any preference it might exhibit for objects on either the left or the right; indeed, the animals usually did show such preferences. (The correct object was shifted from side to side in a random sequence to control for these preferences.) Ostensibly, the monkeys also had to learn that one object consistently contained food while the other was always empty. Although these learning sets by themselves would not serve to identify the correct object in each new discrimination problem, it seems likely that they could help the animal locate the reward very rapidly by eliminating initially unprofitable responses.
Reversal learning
In reversal learning, the individual first learns to make a discrimination, such as choosing a black object in a black–white discrimination problem, and then is supposed to learn to reverse his choice—i.e., to choose the white object. Such reversals tend to be difficult for most learners since there are negative transfer effects; e.g., the individual tends to persist in responding to the black object that was originally correct. Eventually, however, one’s tendency to make the originally learned selection typically becomes weaker, and he makes the competing response (e.g., to white) more frequently until a point is reached where it is almost consistently evoked. Reversal learning can be accomplished very rapidly when a laboratory animal, such as a monkey, is presented with a series of reversal-learning problems in which the same sequence of shifts is repeated (as when black is initially correct, then white, then black, then white, and so on). After extended reversal training, some animals are able to make the next reversal in the sequence in one trial. They behave as if they have mastered the abstract concept of alternation or of regular sequence.
The speed with which representatives of a given species of animal, including human beings, can be taught to make a reversal of this kind seems to be related to the place biologists assign them in a hierarchy of evolutionary development. On first being exposed to a reversal-learning problem, normally competent adult humans who can use language are likely to achieve a solution with great rapidity. Monkeys can learn to perform equally well after a relatively longer series of reversal-learning tasks; but isopods such as pill bugs or sow bugs, small relatives of crabs and shrimp, have such primitive brains that they seem to be unable to improve their performance at all during a series of reversal-learning tasks.
Developmental processes and transfer
The manner in which a problem is learned seems to have an effect on what is transferred. This conclusion is supported by experiments in which comparisons are made of the relative ease with which children of different ages execute reversal and so-called extradimensional shifts. In performing both kinds of shift, experimental subjects learn two successive discriminations between two pairs of objects that vary simultaneously in two aspects or dimensions—e.g., white triangle versus black square, and black triangle versus white square. In training subjects initially, discrimination of only one dimension (for example, black–white) is made relevant, with the child’s selection of one of the cues (for example, white) being rewarded, while the other (black) is incorrect. After they have learned this, the children are shifted to the second discrimination. In the case of a reversal shift, the same stimulus dimension (black–white) remains relevant, but the child is now to learn to reverse his initial choice; black choices are now rewarded, and white selections become incorrect. For an extradimensional shift, the initially irrelevant dimension (square–triangle) is given relevance by rewarding selection of one of its alternatives and by failing to reward choices for the other.
The relative ease with which human beings learn to make extradimensional and reversal shifts is related to how old they are. Reversal shifts are relatively difficult for young children to learn and are relatively easy for adults to master. As people gain maturity, the relative ease with which they execute a reversal shift tends to increase in comparison with their ability to achieve an extradimensional shift.
Explanations for these developmental changes seem to be found in the manner in which the individual solves a discrimination problem. Very young children and laboratory animals tend to learn simple habits when faced with a discrimination problem for the first time; for example, they are most likely to learn simply to approach black objects and to avoid white. Reversal shift is often extremely difficult for them, and negative transfer effects are substantial. Subjects who primarily learn simple habits are faced with the task of eliminating one habit (e.g., to choose black) that has been rewarded and then of developing another habit (e.g., to choose white) that previously has not been rewarded.
Human adults, on the other hand, generally find a reversal shift relatively easy; they do not behave as if they simply associate their choices to the relevant stimuli (e.g., white and black) but instead appear symbolically (or conceptually) to react to both of them in terms of their common characteristic (brightness). A similar kind of symbolic or logical response is appropriate in solving reversal-shift problems; since the relevant dimension remains the same, this kind of shift tends to be easier to make than is one involving extradimensional shift, which requires the individual to switch to a new symbolic response (e.g., from brightness to size). In short, when they respond concretely, learners favour their potentials for achieving extradimensional transfer; those who tend to respond symbolically enhance the probability for reversal transfer.
Whatever the validity to be found in theoretical explanations of this sort, review of how transfer phenomena may be influenced suggests that no single principle or simple theory thus far put forward accounts for all of the observed data. Instead, the evidence is that several interacting processes underlie transfer of training and that their relative influence depends both on the nature of the tasks between which transfer takes place as well as on the characteristics of the learning organism. If one seeks to control the degree of transfer, as one does in educational settings, it seems useful to analyze transfer behaviour in terms of a number of component processes—e.g., stimulus and response similarity, stimulus predifferentiation and response learning, and the symbolic abilities of the learner.
The physiology of transfer of training
Although available evidence for a physiological basis of transfer of training is limited, some impressive data already are recorded. Some central (brain and spinal-cord) mechanisms seem to control transfer of training. A long-established transfer phenomenon is cross education, in which there is positive transfer of a skill learned with one part of the body to another, untrained part. For example, a person who learns to throw a dart with his preferred hand exhibits positive transfer to his non-preferred hand. Since different muscles are involved in the equivalent action of opposite limbs, positive transfer resulting from cross education cannot be attributed simply to common muscular movements; instead it would seem that cross education depends on central processes that control the actions of both limbs.
Among highly evolved animals, transfer of training between limbs from opposite sides of the body evidently is mediated through a massive system of neural fibres, known as the corpus callosum, that connects the two hemispheres of the brain. One of the many ways in which the validity of this principle may be demonstrated is first to train blindfolded cats to discriminate with one paw between two different pedals (by feeling raised horizontal lines on one pedal and by detecting raised vertical lines on the other). Since each eye sends some of its nerve impulses to both hemispheres of the cat’s brain while each paw only directs impulses to the hemisphere of the brain on the same side of the animal’s body, this procedure feeds the sensory information to just one hemisphere. After learning to make the discrimination with one paw (e.g., reward being given only for the pedal with the horizontal pattern), a cat that is confronted with making the same discrimination with the other front paw, which has its connections with the ostensibly “untrained” brain hemisphere, will nevertheless exhibit positive transfer. Indeed, even when the corpus callosum is surgically severed immediately after learning (to “disconnect” the two hemispheres), positive transfer will take place from one front paw to the other; manifestly, transfer of training takes place between connected hemispheres while the animal is learning. If the cat’s corpus callosum is severed before it initially learns to discriminate the two pedals, however, no transfer occurs between the animal’s limbs; the untrained paw fails to exhibit any benefit from what has been learned with the other paw. In other words, by severing the cat’s corpus callosum, the surgeon splits the brain into two independently functioning units. The same kinds of behaviour are observable among other split-brain animals, including chimpanzees and people.
The physiological foundations of transfer of training are not limited merely to the anatomical considerations of the central nervous system. To better understand how physiological processes mediate transfer of training means also to be able to specify more fully the anatomic, electrical, and chemical basis of learning in general, a goal that remains incompletely achieved. Many physiologists and psychologists hold that the search for the neurophysiological foundations of learning can be pursued most profitably by measuring physical and chemical changes that influence the transmission of nerve impulses. It has long been established that chemical changes are part of the process of neural transmission; and it is widely agreed that, in some way, biochemical activities also are responsible for all forms of learning, including transfer of training.
One popular theory in the 1960s was that learning and remembering depend on changes in the molecular structure of such chemicals as ribonucleic acid (RNA) and peptides that are incorporated in the cells of the body, including nerve cells. Some researchers have theorized that memory traces are physically coded within the molecules of cells.
Reports of experiments have been published offering evidence that skills have been transferred from one individual to another by injecting materials taken from the brains (or even other parts of the body) of trained animals into the bodies of untrained organisms (e.g., flatworms, rats, hamsters). These reports have encouraged many to hope that someday one might be able to learn a foreign language, for example, by simply taking a pill instead of through the usual time-consuming practice. Subsequent efforts to repeat such experiments sometimes have given positive results but more often have yielded no evidence of chemical transfer of training from one individual to the next. In view of such inconsistent findings, this question became a matter of considerable controversy. Many investigators seemed inclined to dismiss the notion that organisms can learn by swallowing chemicals or through injection as another of those oversimplified interpretations that continue to be offered in efforts to account for complex psychophysiological phenomena.
Howard H. Kendler