Paired-associate models and the effects of list length

Abstract A two-process Markov model for paired-associate learning is presented in which stimulus-response associations may pass through an intermediate or short-term memory state before learning is complete. In the short-term state, forgetting may occur, and in the trial-dependent-forgetting (TDF) model, the likelihood that forgetting takes place on any trial is postulated to be a function of the number of S-R pairs remaining to be learned on that trial. To determine the quantitative accuracy of the model, a paired-associate experiment was conducted in which list length was varied. Specific response-sequence frequencies from experimental lists of 9, 15, and 21 items were reasonably well predicted by the TDF model. A much better account of the data was obtained by a revision of the model, in which it was assumed that the probability of learning on a given trial depended on whether an item was still in short-term memory or had been forgotten. Comparative predictions from the linear and all-or-none models, as well as an alternative two-process Markov model, are also presented.