REMI and ROUSE: Quantitative Models for Long-Term and Short-Term Priming in Perceptual Identification

The REM model originally developed for recognition memory (Shiffrin & Steyvers, 1997) has recently been extended to implicit memory phenomena observed during threshold identification of words. We discuss two REM models based on Bayesian principles: a model for long-term priming (REMI; Schooler, Shiffrin, & Raaijmakers, 1999), and a model for short-term priming (ROUSE; Huber, Shiffrin, Lyle, & Ruys, in press). Although the identification tasks are the same, the basis for priming differs in the two models. In both paradigms we ask whether prior study merely reflects a bias to interpret ambiguous information in a certain manner, or instead leads to more efficient encoding. The observation of a ‘both-primed benefit’ in two-alternative forced-choice paradigms appears to show that both processes are present. However, the REMI model illustrates that the both-primed benefit is not necessarily indicative of an increase in perceptual sensitivity but might be generated by a criterion bias. The ROUSE model demonstrates how the amount of attention paid to the prime, and the consequent effect upon decision making, may lead to the reversal of the normal short-term priming effect that is observed in certain conditions.