Acquisition of discriminations involving ambiguous or non-ambiguous features: an evaluation of two configural learning models

The present rat experiment evaluated the validity of two formal accounts of configural learning in the framework of discrimination tasks involving the serial presentation of feature and target stimuli: Rescorla's (1973) modification of the Rescorla-Wagner model (1972) and the Pearce model (1987). The first, ambiguous feature task was of the form X-->A+, Y-->A-, X-->B-, Y-->B+, in which X and Y represent visual features, '-->' signifies a serial arrangement, A and B are auditory target stimuli, and '+' and '-' symbolise food-reinforcement and non-reinforcement, respectively. The second, non-ambiguous feature task was of the form: X-->A+, Y-->A-, X-->B+, Y-->B-. The former task was much more difficult to solve than was the latter task. The Rescorla model is able to account for the observed differences between the two tasks in learning rates and in the associative strength of feature X with more plausible parameter values than is the Pearce model. It is suggested that models acknowledging a role for both elemental and configural learning can better account for discrimination learning in discrimination tasks of the sort presented in this study than do models that exclusively allow for configural learning.