Learning by Occasion Setting

Abstract

In a feature-positive discrimination, presentations of the ‘feature’ cue and ‘target’ cue are accompanied by the unconditioned stimulus, and presentations of the target alone are not reinforced, whereas in a feature-negative discrimination the target alone is reinforced and the compound non-reinforced. In these discriminations, the feature can either act as a ‘simple stimulus’—eliciting conditioned responses by signaling the occurrence of the unconditioned stimulus, or as an ‘occasion setter’—controlling the responses generated by the target by indicating its relation to the unconditioned stimulus. Three criteria can distinguish between simple conditioned stimulus properties (excitation or inhibition) and occasion setting attributes of the feature: (a) response form, (b) extinction on feature-positive discriminations and counter-conditioning on feature-negative discriminations; and (c) transfer during feature-positive and feature-negative discriminations. A neural network model of classical conditioning is used to (a) illustrate how the feature acts as a simple CS in simultaneous feature-positive discriminations, and (b) as an occasion setter in serial feature-positive discriminations, (c) explain why occasion setters are impervious to extinction, and (d) hypothesize about the conditions in which occasion-setter properties can be transferred to other targets.