Two States of Paired-Associate Learning

Rock (1957) presenzed evidence which he interpreted to suggest that learning to give one nonsense syllable as a response to another always occurs in a single trial. If, as McGuire ( 1961) has shown, such learning involves several independent steps, e.g., learning to put together the letters of the response term as well as associating it with the stimulus, it seems implausible that all the learning should take place on a single trial. To show that learning is not strictly all or none, it would be sufficient to demonstrate the existence of an intermediate state between not-learned and learned. Both Rock and Estes ( 1960) claimed to have demonstrated the a5sence of such an intermediate state by showing that pairs which an S did not recall on a given trial were no more likely to be recalled on the next trial than were pairs which he had never seen before. This method is subject to considerable bias since pairs learned on one trial versus pairs nor so learned are selected by S rather than imposed by E. This bias has been experimentally demonstrated (Williams, 1961). An unbiased and simpler merhod is to use a second and more sensitive index of learn~ng in addition to recall. When a recognition test is employed in this manner, an intermediate state seems to emerge quite clearly, as the following experiment indicates.