Slave systems in verbal short-term memory

Abstract

Background: The model of performance in short-term memory (STM) tasks that has been most influential in cognitive neuropsychological work on deficits of STM is the “working memory” model mainly associated with the work of Alan Baddeley and his colleagues.

Aim: This paper reviews the model. We examine the development of this theory in studies that account for STM performances in normal (non-brain-damaged) individuals, and then review the application of this theory to neuropsychological cases and specifications, modifications, and extensions of the theory that have been suggested on the basis of these cases. Our approach is to identify the major phenomena that have been discussed and to examine selected papers dealing with those phenomena in some detail.

Main Contribution: The main contribution is a review of the WM model that includes both normative and neuropsychological data.

Conclusions: We conclude that the WM model has many inconsistencies and empirical inadequacies, and that cognitive neuropsychologists might benefit from considering other models when they attempt to describe and explain patients' performances on STM tasks.

Keywords:

• Aphasia
• Aphasiology
• Neuropsychology
• Psychosocial
• Language
• Brain

Notes

¹The effect of concurrent tapping in Larsen and Baddeley (2003) is also relevant to their suggestion that the lack of an effect of irrelevant speech on phonologically similar lists is due to participants abandoning the PS. Concurrent syncopated tapping and articulation produced statistically indistinguishable lower spans in the phonologically similar condition. If the PS was not used when items are phonologically similar, then the reduction in span in the phonologically similar conditions due to concurrent syncopated tapping and concurrent articulation must be because of interference of these concurrent tasks with the CE, not the PS. The claim that concurrent syncopated tapping interferes with the CE is potentially consistent with the WM model (although it is not the view that Larsen & Baddeley, 2003, take), but the idea that concurrent articulation does so is obviously not. A second challenging result of this study is that the effect of concurrent regular single-item tapping was the same as that of irrelevant speech in not reducing the visual PSE. Larsen and Baddeley (2003) say that “the fact that tapping does not reduce the phonological similarity effect is consistent with the assumption that its impact is on the central executive, not on the phonological loop” (p. 1263). By this logic, irrelevant speech might also affect the CE.

The fact that Larson and Baddeley (2003) found no difference in the effect of concurrent tapping and concurrent articulation in Exps 2 and 3 suggests that these interference conditions operate at a common functional locus. Larson and Baddeley suggested that both concurrent syncopated tapping and concurrent articulation disrupted articulatory rehearsal by interfering with a timing process, a suggestion they attribute to Saito (1993, 1994) and that they consider to be a “modification” of the WM model. The idea that concurrent syncopated tapping selectively interferes with rehearsal by interrupting its timing is based on the view that production of articulatory timing is irregular. However, Guérard, Jalbert, Neath, Surprenant, and Bireta (2009) showed that regular tapping of spatially varied targets had the same effect as syncopated tapping, arguing that syncopation is not the critical factor.

²An exception is Watkins (1972), who found an equal length effect at all serial positions.

³Shallice (1975) also criticised Craik (1968a) on the grounds that phonological errors could have been perceptual errors and that a fast presentation rate was used leading to reliance on semantic processing, both of which would have underestimated phonological errors. The lesson to be drawn from Craik (1968a) is thus not clear.

⁴The double dissociation is unusual in that it involves normal participants under concurrent articulation and a patient, but its implications, such as they are (see Shallice, 1988, for reservations about the implications of double dissociations), are the same as those of the more usually cited double dissociations between patients or across tasks in normal participants (as in Shah & Miyake, 1996).

⁵Martin and Saffran also correlated the difference between the composite S and P indices (the “P-S Index”) with these measures, but these correlations with difference scores are hard to interpret (patients with equal P-S Index scores could have very different levels of performance). Correlations involving difference scores are also suspect because difference scores are less reliable than individual scores. The decrease in reliability increases with the correlation of the tests involved in the difference; Martin and Saffran's P and S indices were not correlated, making this issue less important but not eliminating the problem of reliability.

⁶Martin and Ayala's (2004) Table 7 (p. 475) appears to contain an error, indicating that there was a correlation between semantic abilities and repetition “modified spans” in patients with more severe phonological deficits (S > P patients).

⁷Nickels and Howard (1995) note that adapting the Dell and O'Seaghdha model to input (and thus also repetition) is problematic because the adaptation generates 37% phonological errors in comprehension (e.g., “cat” misunderstood as MAT).

⁸The data in the figures are the model's results for error types at the “lexical” level for semantic and formal errors and at the “phonological” level for neologisms. Martin et al.’s figure 7 depicts semantic errors at the lexical level because it reflects commitment to a lexical item (N. Martin, personal communication) but the use of lexical-level data is somewhat misleading, since NC's responses necessarily involve the phonological level.

Figure 2. Performance and simulation of NC's recall of low- and high-imageable words in first and second position in two word lists and after decay. Left panels: NC's performance on high-imageability (left) and low-imageability (right) words in first (top panel) and second (middle panel) position of two word lists, and after a filled delay (bottom panel). Right panels: Simulation. Top panel: Simulated performance on first word in two word lists with high-imageability (left) and low-imageability (right) words. Middle panel: Simulated performance on second word in two word lists with high-imageability (left) and low-imageability (right) words; Bottom panel: Martin et al.’s (1996) figure 7, showing effect of imageability on semantic errors collapsed over time. S = semantic error; F = formal paraphasia; N = neologism. Reprinted from Brain and Language, 52(1), Nadine Martin, Eleanor M. Saffran, & Gary S. Dell, Recovery in deep dysphasia: Evidence for a relation between auditory–verbal STM capacity and lexical errors in repetition, pages 83–113, Copyright 1996, with permission from Elsevier.

⁹An example of such a model is provided by R. Martin, Breedin, and Damian (1999), who simulated the performance of one patient, AP, in phoneme discrimination, lexical decision and repetition through rapid decay of phonemic representations in a modified version of McClelland and Rumelhart's (1981) “interactive activation” model of visual word perception. R. Martin, Breedin, et al. (1999) suggested that AP's STM limitations could have resulted from his lexical deficit, but this possibility was not explored computationally.