Two Attentional Models of Classical Conditioning: Variations in CS Effectiveness Revisited

摘要

Attentional models offer alternatives for describing blocking, overshadowing, and many other features of classical conditioning. Two such models emphasize variations in the associability of CSs instead of variation in the effectiveness of the reinforcing event, the US. Early published variants do not always accurately portray the effects of nonreinforced CS presentations as represented in simulation experiments. In one case levels of conditioned responding under partial reinforcement are too low to reasonably approximate expectations based on the experimental literature, and extinction is too deep to produce the rapid reacquisition that typically follows extinction. These problems are corrected by changing the expressions in the model for decreasing associative strength. The revised model retains the positive features of the original, e.g., the ability to stimulate in real-time latent inhibition and compound CS effects such as blocking and conditioned inhibition. The other model is path dependent and highly nonlinear under partial reinforcement. The problem can be corrected either by modifying and restricting the rules for computing the associability of the CS, or by modifying the rules for computing associative strength. The revised model retains the original's ability to simulate latent inhibition, compound CS effects, and the transfer (positive or negative) from training with a weak US to training with stronger US.