Past experience, recency, and spontaneous recovery in choice behavior

Abstract

Pigeons' responses on two keys were recorded before and after the percentage of reinforcers delivered by each key was changed. In each condition of Experiment 1, the reinforcement percentage for one key was 50% for several sessions, then either 70% or 90%for one, two, or three sessions, and then 50% for another few sessions. At the start of the second and third sessions after a change in reinforcement percentages, choice percentages often exhibited spontaneous recovery-a reversion to the response percentages of earlier sessions. The spontaneous recovery consisted of a shift toward a more extreme response percentage in some cases and toward a less extreme response percentage in other cases, depending on what reinforcement percentages were previously in effect. In Experiment 2, some conditions included a 3-day rest period before a change in reinforcement percentages, and other conditions included no such rest days. Slightly less spontaneous recovery was observed in conditions with the rest periods, suggesting that the influence of prior sessions diminished with the passage of time. The results are consistent with the view that choice behavior at the start of a new session is based on a weighted average of the events of the past several sessions.Although the results of several experiments were well described by this model, it is too simple to accommodate where~~is the change in strength of~, and r is a reinforcement parameter that can range from 0 to 1. After each response that is not reinforced, the strength of the response decreases as follows:forcement, the size of a change in probabilities, and the overall rate of reinforcement.The results of these experiments contradicted the predictions of several models oftransitional choice behavior, including Myerson's kinetic model (Myerson & Miezin, 1980; see also Myerson & Hale, 1988), Staddon's ratioinvariance model (Staddon, 1988;Staddon & Horner, 1989), and the linear-operator model (Bush & Mosteller, 1955). However, the results could be described quite well by a simple mathematical model similar to one proposed by Couvillon and Bitterman (1985). The model states that each response alternative, i, has a separate strength, . Response strength increases each time the response is reinforced and decreases each time the response is not reinforced. Each time a response is reinforced,~increases as follows:where n is a nonreinforcement parameter that can range from 0 to 1. A simple matching rule is used to translate from the independent strengths of two responses (~and J-2) to the probability that one will occur:Vi PI = V; + v: .

Serial discrimination reversal learning in pigeons as a function of intertrial interval and delay of reinforcement

“…This effect was apparent for all four conditions. Such a pattern has been seen in previous studies of reversal learning, and it has been interpreted as the result of subjects' forgetting what was learned in Session 1 during the 24 h that separated the two training sessions (Woodward, Schoel, & Bitterman, 1971), or as the result of spontaneous recovery of the preferences that were established in previous training sessions (Mazur, 1995(Mazur, , 1996.…”

Extinction revisited: Similarities between extinction and reductions in US intensity in classical conditioning of the rabbit’s nictitating membrane response

“…To do so, unlearning theories have included additional mechanisms, the aim of which is to conserve associative strength while still yielding large reductions in responding. Within the unlearning tradition, the most detailed account of spontaneous recovery is stimulus-samplingtheory (Estes, 1955;Mazur, 1996). It assumes that a CS generates a large pool of hypothetical stimulus elements.…”

Choice in a Variable Environment: Effects of Unequal Reinforcer Distributions

“…Mazur and colleagues (Bailey & Mazur, 1990;Mazur, 1992Mazur, , 1995Mazur, , 1996Mazur, , 1997Mazur & Ratti, 1991) investigated the effects of a single within-session change in the reinforcer ratio available from two alternatives. Another approach used pseudorandom binary…”