Unifying models of paired associates and serial learning: insights from simulating a chaining model

Abstract

Recent evidence suggests a basic dissociation between paired associates (PAL) and serial learning (SL). Forward and backward probes of PAL are nearly perfectly correlated; for probed recall of a learned serial list, the correlation is moderate. We suggest that SL and PAL are ends of a continuum within a single theoretical framework. A single parameter controls the degree to which pairs of list items are isolated from the rest of the list. This Isolation Principle may be flexibly implemented into many models, and we demonstrate, in simulations, how Isolation can account for the distinction between memory for pairs and triples.