Conditions of invariance and modifiability in simple reversal learning.

Abstract

Rats were run in a single -T unit with correction method to food reinforcement, different groups being given 1, 2 or 4 daily trials; after reaching a criterion on the original task, four successive reversals were made. Learning rate remained constant throughout the series for the 1-trial per day group, but showed significant increases for the multiple-trial groups. The increases were primarily due to significant reduction of errors on the second trial of a daily block, a result at least partially due to the multiple-trial animals learning to slow down at the choice point on trials following errors. The results also indicated that rate of learning was larger within than between daily blocks, that rate of approach to the asymptote of 100% correct was an increasing negatively accelerated function of the number of trials in a daily block, and that there was a regression in response probability from the end of one daily block to the beginning of the next, decreasing during each series and greater for reversal than for original learning. 16 references.