Discrimination and reversal in capuchin monkeys as a function of irrelevant cue salience

Abstract

Eight capuchin monkeys were trained on a simultaneous shape discrimination and a subsequent reversal with size, brightness, and position cues irrelevant. One group (HS) was trained with highly salient size and brightness cues, and the other group (LS) was trained with less salient size and brightness cues. The two groups did not differ significantly in learning either the original discrimination or its reversal. However, during reversal, Group HS responded more to its preferred size than did Group LS. This result indicates that the emergence of previously absent preferences along an irrelevant dimension is a function of the salience of the dimension. Responding to irrelevant dimensions was also found to increase from acquisition to reversal for the HS group, but not for the LS group. This increase is not consistent with Mackintosh's recent theory of discrimination learning.A variety of animal learning experiments have amply demonstrated that responses to irrelevant dimensions do not necessarily decrease monotonically as learning proceeds. For example, Turner (1968) reported that position habits increased during acquisition of brightness discrimination by rats in a Lashley jumping stand , and Shepp and Schrier (1969) obtained comparable results during learning set acquisition by macaque monkeys . Increased position responding was also previously observed by the authors (Scanlon & King, 1976) in sameness-difference learning by capuchin monkeys. In addition, previously suppressed position habits and other systematic modes of responding correlated with irrelevant dimensions are potentially prone to elicitation during reversal of a previously learned discrimination, a phenomenon that Harlow (1959) has referred to as "regression." An example of regression was the appearance during reversal of stimulus preference and differential cue error factors on a series of multiple discrimination reversal problems (Harlow, 1950).Increased responding to irrelevant dimensions during acquisition and regression during reversal are inconsistent with a recent theory of discrimination learning proposed by Mackintosh (1975). According to this theory , a dimension-specific learning rate parameter, a, changes for each stimulus dimension according to that dimension's correlation with reinforcement. Increases in aA occur only when Dimension A is correlated with reinforcement and, as in overshadowing, is more salient than another stimulus dimension also correlated with reinforcement, or, as in blocking, has been previously established as the signal for the reinforcer . Conversely, aB decreases only when Dimension B is not correlated