| Inhibition of delay, defined as a progressive increase in the latency of initiation of the Pavlovian conditioned response over training, is predicted by real-time models that make use of a competitive Rescorla-Wagner (1972) learning rule (e.g., Buhusi, & Schmajuk, 1999) but not by models using a noncompetitive linear-operator (Hull, 1943) learning rule (e.g., Kirkpatrick & Church, 1998), which instead predict a progressive decrease in the latency of CR over training. Three experiments investigated inhibition of delay of the rabbit conditioned eye closure. In experiments 1 and 2, rabbits were trained with a tone CS of different durations that overlapped and terminated with a periorbital shock US (i.e., CS-US intervals of 250, 500, 1000 or 2000 ms in Experiment 1 and 250 or 1500 ms in Experiment 2). Inhibition of delay was apparent in animals trained with relatively long CS-US intervals (i.e., 500, 1000, 1500 and 2000 ms) but not in animals trained with a relatively short CS-US interval (250 ms), whose CR latency decreased, rather than increased, over training. In Experiment 3, after extensive training, the presentation of an extraneous stimulus prior to CS onset produced a reinstatement of short latency CRs in the group trained with a 1500 ms CS-US interval, consistent with the notion that the responses were otherwise "inhibited", but did not affect CR latency in the group trained with a 250 ms CS-US interval.; The finding that the latency of CR moves in opposite directions depending on the duration of the CS-US interval, poses problems for all existing models of Pavlovian conditioning, since their respective predictions for this phenomenon are independent of the CS-US interval. Since neither of the two theoretical approaches can account for this outcome, several potential modifications to each approach were examined. It was concluded that the apparently most powerful solution follows from assuming a constrained competitive Rescorla-Wagner (1972) rule and a representation of the CS comprised of two classes of elements, one class with a temporally distributed pattern of activity over the CS duration and the other class with a randomly distributed pattern of activity over the CS duration. |
