A forgetting model for paired-associate learning

This paper presents a Markov model for paired-associate learning which is based on conclusion drawn from experimental studies of short-term memory. Learning of paired-associates is considered as a decrease in the probability of forgetting an association between trials. The ability of the model to account for experimental results is demonstrated by considering data from two experiments involving paired-associate lists. Observed values of the mean learning curve and response probabilities conditional on the outcome of the preceding trial are compared with the values predicted by the forgetting model and the one-element model. Comparison of parameter values for several experimental situations provides interpretations for the forgetting model's parameters. Finally, the model is applied to an experiment using an RRTTTT design to demonstrate its potential generality.