The influence of generalization decrement on the outcome of a feature-positive discrimination

In a feature-positive discrimination subjects receive training in which reinforced presentations of a compound, AB⁺, are intermixed among non-reinforced presentations of one element, A^o. The three reported experiments compared the results of this procedure when using a light or a tone as the feature-positive stimulus, B. In all three experiments, the tone elicited a stronger conditioned response than the light when it was presented alone. The main concern of Experiments 2 and 3 was to examine whether this effect was due to training being incomplete for subjects receiving the light as the B element. In Experiment 2 it was found that extended discrimination training did not diminish the difference between the properties of the light and the tone. Experiment 3 revealed that this effect was not due to differences in the asymptotic associative strength of either A or the compound. It is argued that these results are not readily compatible with contemporary theories of learning, and an alternative account is presented.