Version 3 of the retrieving effectively from memory model: extensions and limitations

Abstract

Shiffrin and Steyvers (1997) presented a computational model of human memory they called Retrieving Effectively from Memory (REM). In their original report of the model, they described several REM variants. To date, the majority of papers using REM have employed the simplest version of the model, known as REM.1. Although it does not matter for most applications, REM.1 makes an important, simplifying assumption: namely, that item strengthening always accumulates in a single mnemonic trace. In other words, there is one trace for each study item, regardless of the time for which items were presented or the number of times items were repeated. In the case of spaced repetitions, this assumption of automatic, single-trace storage is untenable. Here, I use a version of REM Shiffrin and Steyvers termed REM.3. In REM.3, spaced repetitions are stored in one trace if subjects identify the repetition as previously studied, and in separate traces otherwise. I show that this model can account for two findings previously believed to be inconsistent with the REM framework: the spacing effect (Delaney, Verkoeijen, & Spirgel, 2010) and positive list-strength effects observed with spaced strengthening (Ratcliff, Clark, & Shiffrin, 1990, Experiment 5) and the strong-interference paradigm (Norman, 1999). I then test a novel prediction from REM.3 concerning the list-strength effect. Three experiments were unable to find support for this prediction. Instead, I present an explanation of the list-strength effect based on strategy disruption.