Individual Differences in Memory Span

Rehearsal speed has traditionally been seen to be the prime determinant of individual differences in memory span. Recent studies, in the main using young children as the participant population, have suggested other contributors to span performance, notably contributions from long-term memory and forgetting and retrieval processes occurring during recall. In the current research we used structural equation modelling to explore at the construct level individual differences in immediate serial recall with respect to rehearsal, search, and speed of access to lexical memory. We replicate standard short-term phenomena; we show that the variables that influence children's span performance influence adult performance in the same way; and we show that lexical memory access appears to be a more potent source of individual differences in immediate memory than either rehearsal speed or search factors. Individual Differences in Memory Span 3 The Contribution to Word Memory Span of Rehearsal, Search Speed, Access to Lexical Memory, and Phonological Coding: An Investigation at the Construct Level. Most current models of immediate serial recall have three basic assumptions. Firstly, that short-term information is lost very rapidly and as a result serial recall is often based upon a degraded memory trace. Some models attribute forgetting to decay (Baddeley, 1986; Burgess & Hitch, 1999; Cowan, 1999; Page & Norris, 1999), others see interference as the cause of forgetting (Brown & Hulme, 1995; Nairne, 1990). The second assumption is that in order to recall a list item, a degraded memory trace must be reconstructed or redintegrated in some manner (Brown & Hulme, 1995; Farrell & Lewandowsky, 2002; Schweickert, 1993). Thirdly, the assumption is made that the trace that supports recall is speech-based in nature. That is, immediate recall relies upon phonological codes. While these models assert that these features are important determinants of immediate recall, they say little about the relative importance of each factor or what contribution to span each factor makes. One way of addressing this problem is to take an individual differences approach and to assess the contribution of each of these factors to immediate serial recall (ISR). This approach is adopted in the current study. The standard account of immediate memory (Nairne, 2002) sees performance to be the outcome of a balancing act between rapid de-activation of the memory trace and rehearsal, where rehearsal is assumed to refresh and re-activate the fading trace. Many of the variables that are believed to influence memory span are thought to have their influence via the ease with which rehearsal can be carried out and the memory trace refreshed. The concept of rehearsal has been notoriously difficult to measure and most researchers have opted for indirect measures, the most widely adopted being speed of overt articulation. In probably the first study to use this indirect measure of rehearsal, Baddeley, Thomson and Buchanan (1975) had their participants repeat three words as quickly as possible ten times and, as a second measure, read a list of 50 words as quickly as possible. Using regression procedures they found a strong relationship between immediate memory span and the speeded articulation measures; span could be predicted on the basis of the number of items a participant could articulate in approximately 2 seconds. The correspondence between articulation and span was taken as prima facie evidence that rehearsal speed was a prime determinant of span. Span-rehearsal rate correspondences have subsequently been used to account for many of the structural (Schweickert & Boruff, 1986; Tehan & Humphreys, 1988), developmental (Hulme, Thomson, Muir & Lawrence, 1984), cross-cultural (Chen & Stevenson, 1988), and individual differences that are observed in span performance (Baddeley et al. 1975). Despite the initial appeal of the rehearsal speed explanation, recent evidence suggests that its role is not as dominant as first thought. For a start, there are many instances where rehearsal rate and span dissociate. It is possible to match items for spoken duration and still see differences in span (Hulme, Maughan & Brown, 1991). Individual Differences in Memory Span 4 Likewise, it is possible to see differences in spoken duration without accompanying changes in span (Caplan, Rochon & Waters, 1992). In the light of these and similar findings, it has been suggested that rehearsal may not actually be all that important (Brown & Hulme, 1995; Nairne 2002). Cowan and his colleagues (Cowan, Day, Saults, Keller, Johnson, & Flores, 1992) presented an alternative account of the rapid forgetting in immediate memory that is based upon decay during the recall process. They demonstrated that longer words at the start of the list caused more forgetting for the words at the end of a list than if short words were presented at the start of the list. Cowan argued that the trace that supported recall decayed during the recall process and that more decay occurred during the articulation of long words during output than short words. From this perspective, articulation speed was an indicator not of rehearsal speed but of the amount of rapid decay that would occur during output. If a person can rapidly articulate the remembered item, decay of the non-recalled items would be minimised. Furthermore, through their exploration of the timing characteristics of verbal output of subspan and span length lists, Cowan et al. (1994) noted that as lists increase in difficulty, participants produce longer and longer pauses between list items. Given a memory architecture where a limited number of items are rapidly returning to threshold levels of activation, they proposed that during pauses participants reactivated the memory trace and searched through these activated traces for the next response. Cowan, Wood, Wood, et al. (1998) tested these ideas using an individual differences approach and were able to show via structural equation modelling (SEM) that articulation and search measures both made significant and independent contributions to span. In a similar study, Hulme, Newton, Cowan, Stuart and Brown (1999) also explored the relationship between span and memory search rate. They found a significant correlation between search rate and span for one syllable words, but this relationship did not hold for 5 syllable words, nor for one and five syllable non-words. As such, the role of memory search rate is worthy of further exploration. It is now commonly accepted that a critical process in immediate serial recall is the ability to reconstruct or redintegrate a degraded short-term trace. While current models of immediate memory incorporate such a stage (Burgess & Hitch, 1999; Farrell & Lewandowsky, 2002; Page & Norris, 1999), these models are not overly specific about the memory system that is accessed during this process. It is becoming increasingly clear that redintegration involves accessing lexical and/or semantic memory (Hulme et al., 1991; Schweickert, 1993). Lexical features of verbal materials often have a similar impact upon immediate recall. Most obviously, words are better recalled than non-words (Hulme et al., 1991). Word-frequency likewise has an impact upon span performance that is not mediated by rehearsal (Tehan & Humphreys, 1988). Content words (e.g., nouns, verbs, and adjectives) are recalled better than function words (e.g., prepositions, pronouns, articles) under both rehearsal and suppression conditions (Tehan & Humphreys, 1988). Results such as these have led to the proposition that at least two components underlie span performance, a rapidly decaying phonemic trace which is augmented by long-term knowledge about the Individual Differences in Memory Span 5 lexical attributes of words (Schweickert, 1993; Brown & Hulme, 1995; Hulme et al., 1991). The reason for assuming that lexical access involves phonemic traces is based on the finding that immediate recall seems to be based primarily upon phonological representations. It has long been known that participants have difficulty remembering lists of items that rhyme or have substantial phonological overlap (Baddeley, 1966). Given the ubiquitous role of phonological codes in immediate recall, Hulme et al. assumed that long-term phonological information would be accessed to facilitate the reconstruction of degraded phonological representations. Moreover, at least in the developmental literature, there is substantial evidence that children’s ability to process phonological information is a substantial contributor to span (Kail, 1997). In summary, the notion that rehearsal is the sole or even major contributor to span has to be questioned. Decay during recall, speed of search, the speed of access to lexical memory, and facility with phonological coding have all been proposed as alternative or additional mechanisms. Tehan and Lalor (2000) used an individual differences approach to explore the relationships between some of these measures, namely rehearsal, speed of lexical access, and output time. With multiple measures of each construct, a factor analysis supported the three latent constructs. Tehan and Lalor attempted to gauge the relative contributions of the three factors to memory span by conducting a series of hierarchical regression analyses and varying the order of entry of the factor scores. In their first study, they found that all three factors made significant contributions to memory span but that Lexical Access contributed most to digit span scores when it was the first factor entered in the equation (29%). Rehearsal and output measures contributed equivalent amounts to span (15%) when they were each entered first in the regression equation. In a second study, they found that Lexical Access accounted for 19% and Rehearsal for a mere 4% of the variance in span. Although they were able to produce strong evidence for the importance of speed of lexical access in memory span, the Tehan and Lalor study was confined to the comparison of speed of lexical access, rehearsal, and output speed (Study 1), and to lexical access and rehearsal (Study 2). No measures of search speed or facility with phonological codes were included. As pointed out earlier in the introduction to this paper, both of these variables have been linked with memory span. Cowan et al. (1998) argued for the importance of search processes and Hulme et al. (1991) emphasised the role of phonological coding. While some of these variables have been considered in combination, no single study has combined all four factors. In the following study, the rehearsal and speed of lexical access measures used by Tehan and Lalor were included along with measures of speed of search, and measures of phonological coding in an attempt to determine the relative contribution of these factors to immediate recall. Issues with Measures The above aims can be adequately evaluated only if valid markers of the Individual Differences in Memory Span 6 variables are selected. We have indicated what some of those markers might be but it is important to justify our selection because, in many instances, prior research has raised significant doubts about some of the measures used. Rehearsal: In the original work exploring the relationship between span and rehearsal rate (Baddeley et al., 1975), two indirect measures of rehearsal were utilised: reading a list of 50 words as quickly as possible and repeating three words as quickly as possible ten times. The reading time estimate of rehearsal speed has faded from use due to the complexity of the reading task. Moreover, in the Tehan and Lalor (2000) data the two tasks loaded upon different factors. So the two tasks are not equivalent. The repetition task has been widely adopted as a measure of maximal articulation rate/rehearsal speed. This measure is not free of problems either. For example, Chase, Lyon and Ericsson (1981) reported two studies in which rehearsal of 3, 4 or 5 digit lists did not correlate with span whereas rehearsing a list of six digits did correlate with span. Because the correlation emerged only as span was approached, they argued that the correlation of span with articulation rate is an artefact of the involvement of memory for order in both tasks. Hulme et al. (1984) explored these issues by measuring articulation rates for single words and groups of three words in a sample of children. They found that both measures correlated equally well with memory performance. Likewise, Ferguson, Bowey & Tilley (2002) explored speech rate in children for one word and three word lists. Like Hulme et al, they found that both measures correlated with span with the correlation higher for the three word lists. Using regression techniques they found that if speech rate for single item words was partialled out of speech rate for the word triples, the residuals were still correlated with span. They concluded, as did Chase et al., that the speeded articulation of word triples involved a memory component that inflated the correlation between span and articulation rate. The above research suggests that one needs to be careful in the selection of speeded articulation measures so as to ensure that they have minimal memory load. Consequently, we have adopted those tasks used by Lalor and Tehan (2002). They used the triple word repetition task but supplemented these with two tasks that involved the speeded articulation of overlearned sequences that should not involve a memory load component. Thus participants were required to recite the letters of the alphabet as quickly as possible and to count backwards from 20 to 1 as quickly as possible. These latter two measures are assumed to be free of a memory load