In verbal paired-associate learning, both paradigms show facilitative effects; in instrumental generalization, VII shows generalization but not I; and in salivary conditioning, I shows generalization but not VII. Clearly, we must study paradigms in the context of situations and tasks.
A Hierarchy of Tasks. We may, I think, postulate a general ordering of tasks on the basis of their ordinary capacity to evoke associative arousal (realizing, of course, that in any given case this may be manipulated by instruction and reinforcement). I suspect this ordering ought to relate to the degree of incidental learning found in these situations if we accept Postman ( 1955) analysis of the process. A hypothetical dimension of arousal is given in Figure 6-4. No special brief is held for any given entry being at a particular value relative to some other given entry. The intent of the diagram is to suggest that there are graded series of tasks with respect to associative arousal, most probably ranging from irrelevant discrimination tasks ("Circle the word with the most letters.") to recall tasks ("Remember these words in any fashion and any order that you can."). I believe this is a very important aspect to which we must attend in designing further studies of the mediation processes.
Cofer's remarkable success with the Maier two-string problem and his findings with an ingenious "verbal chains" experiment ( Cofer, 1951) support the view that situations which require problem-solving and hypothesis formation are sensitive to associative arousal factors. In the first situation Cofer showed that learning an appropriate serial list, which was reinforced on several days before the experiment, was related to "pendulum" solutions (the "good-Gestalt solutions") of the two-string problem. In the second situation a chain of associates was imbedded in a set of irrelevant words in such a way that the first word of the chain could be repeatedly reinforced. Then the subject was presented with a new card with four words and asked to choose one. One of the words was an associate of the word on the first card. He was not informed whether his choice was correct or not but was presented with the next card to make another choice and so on. Given these circumstances, subjects overwhelmingly chose words associated with the word reinforced on the first card.
Such situations through their absence of specific instruction force the subject to structure the problem himself and the structure apparently conforms to what would be expected on the basis of normative data or learned associates.
This paper attempts to illustrate the shift in emphasis of one line of investigation in the pursuit of greater understanding of mediation phenomena. It was taken for granted that mediation phenomena may be