Stimulus selection at different stages of paired-associate learning.

Ss learned paired associates with compound stimuli (word + nonsense). Transfer was tested to individual components after varying amounts of training on the compounds. Ss who were trained to a single perfect trial showed no more transfer to nonsense components than Ss trained to 4 of 8 correct responses to the compounds; however, Ss receiving 10 overtraining trials on compounds showed substantially more transfer to the nonsense components. Then a group of Ss was given training after learning a list of compound stimuli to criterion, but during the extended training other items were added so that the list was not mastered while the initial items were receiving overtraining. This group showed no more transfer to nonsense components than a group trained just to criterion. The data are generally consistent with the hypothesis that Ss actively select among stimulus aspects until the list is mastered, and then relax the selection of attention during overtraining. It can be accepted now that when stimuli of paired associates are complex, 5"s often learn pairs by associating the response with a part of the stimulus. Underwood (1963) reviewed several items of evidence and concluded that £s carry on a process of stimulus selection, thereby producing a disparity between the nominal stimulus (presented by E} and the functional stimulus (with which 5 associates the response). Since Underwood's review, further studies by Newman and Taylor (1963), Jenkins and Bailey (1964), Cohen and Musgrave (1964), and Houston (1964) all have provided further demonstrations that after learning a list of pairs with complex stimuli, there are cases in which 5" can give the correct response to one element of a stimulus compound but not to another.