Journal of Experimental Psychology: Learning, Memory, and Cognition 1997, Vol. 23, No. 5,12L7-1232 Copyright 1997 by the American Psychological Association, Inc. O278-7393/97/S3.00 Word-Frequency Effects on Short-Term Memory Tasks: Evidence for a Redintegration Process in Immediate Serial Recall Charles Hulme and Steven Roodenrys University of York Gordon D. A. Brown University of Warwick Richard Schweickert Purdue University Sarah Martin and George Stuart University of York Four experiments investigated the mechanisms responsible for the advantage enjoyed by high-frequency words in short-term memory tasks. Experiment 1 demonstrated effects of word frequency on memory span that were independent of differences in speech rate. Experiments 2 and 3 showed that word frequency has an increasing effect on serial recall across serial positions, but Experiment 4 showed that this effect was abolished for backward recall. A model that includes a redintegration process that operates to "clean up" decayed short-term memory traces is proposed, and the multinomial processing tree model described by R- Schweickert (1993) is used to provide a quantitative fit to data from Experiments 2, 3, and 4. It is now clearly established that verbal short-term memory makes heavy use of some form of phonological or articula- tory coding (see Baddeley, 1986; Penney, 1989; for re- views). According to trace decay models, verbal short-term memory is limited in capacity, with items being represented by traces that decay within a short period of time (e.g., Baddeley, 1986; Schweickert & Boruff, 1986). In such trace decay models, decay can be overcome by rehearsal that involves some form of subvocal articulation. A simple and influential model of this type is embodied in the notion of an articulatory loop (e.g., Baddeley & Hitch, 1974; Baddeley, Lewis, & Vallar, 1984). This model pro- vides a parsimonious explanation for many short-term memory effects. So, for example, the word-length effect (Baddeley, Thomson, & Buchanan, 1975), the fact that people can recall more short than long words in order, has been explained in terms of long words being spoken and so rehearsed more slowly than short words. Phonologically similar words are recalled less well than dissimilar words, Charles Hulme, Steven Roodenrys, Sarah Martin, and George Stuart, Department of Psychology, University of York, York, United Kingdom; Richard Schweickert, Department of Psychol- ogy, Purdue University; Gordon D. A. Brown, Department of Psychology, University of Warwick, Warwick, United Kingdom. Steven Roodenrys is now at the Department of Psychology, Uni- versity of Wollongong, Wollongong, New South Wales, Australia. This research was supported by Grants ROOO 232 567 and R000 236 216 from the Economic and Social Research Council of the United Kingdom and also in part by National Science Foundation Grant 9123865-DBS. We thank Dave Balota, Richard Chechile, Ian Neath, and Philip Quinlan for helpful discussions, and Xiangen Hu for help with his parameter estimation program. We also thank John Wixted for his constructive comments on the article. Correspondence concerning this article should be addressed to Charles Hulme, Department of Psychology, University of York, York YO1 5DD, United Kingdom. Electronic mail may be sent via Internet to CHI @york.ac.uk. and this has been attributed to the detrimental effect of similarity between traces on retrieval from a phonological store (for reviews of these effects, see Baddeley, 1986; Hulme & Mackenzie, 1992). One variable that cannot be accommodated in a straight- forward way by such trace decay models is word frequency. Word frequency has been shown to affect both long- and short-term memory tasks. In long-term memory tasks, performance is generally better on high-frequency words than on low-frequency words; the notable exception to this pattern being the episodic recognition task (e.g., Mandler, Goodman, & Wilkes-Gibbs, 1982). Such episodic recogni- tion tasks require access to knowledge about a particular encounter with a given word; this, however, is a very different task from the short-term immediate serial-recall tasks we are concerned with in the present article. In these tasks participants appear to maintain an ordered representa- tion of the phonological forms of words that have been presented on a given trial. We argue below that the processes involved in short-term memory tasks are sensitive to the properties of the representations of items held in lexical memory and that frequency has effects on recall through its effects on these lexical representations. High-frequency words are recognized more easily in speech than are low-frequency words (e.g., Howes, 1957), and our arugment is that processes analogous to speech perception are in- volved in short-term recall tasks. In short-term memory tasks the effects of frequency are complex and not well understood. Watkins (1977) found that memory span scores (the mean number of words that could be correctly recalled in order) were higher when the first half of the list comprised high-frequency words and the second half comprised low-frequency, than when this arrangement was reversed. He argued from this that there was a long-term memory component in the span task. The greater effect of frequency in the initial part of the list was interpreted in terms of items early in the sequence being recalled from 1217 This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.