Semantic Similarity and Immediate Serial Recall: Is There an Effect on All Trials?

Article excerpt

In immediate serial recall, items are better recalled when they are all drawn from the same semantic category. This is usually accounted for by a two-stage retrieval-based framework, in which, at recall, long-term knowledge is used to reconstruct degraded phonological traces. The category shared by list items would serve as an additional retrieval cue restricting the number of recall candidates. Usually, the long-term search set is not defined, but some authors have suggested an extended search set and others a restricted set that is composed of the most recently presented items. This was tested in an experiment in which participants undertook an immediate serial recall task either alone or under articulatory suppression with either semantically similar or dissimilar lists. A trial-by-trial analysis revealed that, in both quiet and suppression conditions, items from similar lists were better recalled on all the trials, including the first one. In addition, there was no interaction between semantic similarity and trial, indicating that the effect of similarity was of similar size on all the trials. The results are best interpreted within a proposal suggesting an extended long-term search set.

In immediate serial recall, it is well established that long-term memory factors, such as lexicality, word frequency, and semantic similarity, contribute to performance (see, e.g., Hulme, Maughan, & Brown, 1991; Multhaup, Balota, & Cowan, 1996; Murdock, 1976; Poirier & Saint-Aubin, 1995; Stuart & Hulme, 2000). To account for these effects, a reconstruction hypothesis is often put forward (see, e.g., Hulme et al., 1991; Saint-Aubin & Poirier, 1999,2000; Schweickert, 1993). This hypothesis can account for the effects of most, if not all, long-term memory factors recently investigated in this field by suggesting that long-term knowledge of the to-be-remembered items is called upon at retrieval. However, as of yet, in the context of the reconstruction hypothesis, the nature of the set of items called upon at recall has not been clearly defined. The aim of the present study was to investigate this question by taking advantage of the semantic similarity effect. Because of the trial-by-trial analysis that we pursued, the data set is also of interest in terms of constraining models accounting for performance in short-term memory tasks.

The reconstruction hypothesis can be described as follows. At recall, phonological representations set up by list presentation are thought to be degraded and undergo a reconstruction process calling upon long-term knowledge of the to-be-remembered items. Broadly put, long-term knowledge of the language would be called upon to fill in the gaps left by degradation. Typical short-term memory factors, such as word length and articulatory suppression, are thought to influence the amount of degradation (Neath & Nairne, 1995; Saint-Aubin & Poirier, 2000). On the other hand, long-term memory factors are thought to influence the accessibility of the recall candidates (Saint-Aubin & Poirier, 1999, 2000; Schweickert, 1993; Stuart & Hulme, 2000).

Although the reconstruction hypothesis has been successful in accounting for the influence of long-term memory factors on immediate serial recall performance, in most versions of the hypothesis the memory search set called upon at retrieval has not been clearly defined. In fact, Brown and Hulme (1995) argued that within the reconstruction hypothesis there is an "absence of any specification of the 'vocabulary' [quotation marks in the original] used during the redintegration process. Are partially decayed traces compared to every possible vocabulary item? To every item that has been used in the experimental situation?" (p. 600). Multhaup et al. (1996) suggested that the reconstruction hypothesis "does assume that when subjects perform a word span task, they restrict their search of long-term memory information to the items on the to-be-remembered list" (p. …