Re-Examining the Phonological Similarity Effect in Immediate Serial Recall: The Roles of Type of Similarity, Category Cueing, and Item Recall

Study of the phonological similarity effect (PSE) in immediate serial recall (ISR) has produced a conflicting body of results. Five experiments tested various theoretical ideas that together may help integrate these results. Experiments 1 and 2 tested alternative accounts that explain the effect of phonological similarity on item recall in terms of either feature overlap, linguistic structure, or serial order. In each experiment, participants’ ISR was assessed for rhyming, alliterative, and similar nonrhyming/nonalliterative lists. The results were consistent with the predictions of the serial order account, with item recall being higher for rhyming than alliterative lists, and higher for alliterative than similar nonrhyming/nonalliterative lists. Experiments 3 and 4 showed that these item recall differences are reduced when list items repeat across lists. Experiment 5 employed rhyming and dissimilar one-syllable and two-syllable lists to demonstrate that recall for similar (rhyming) lists can be better than for dissimilar lists even in a typical ISR task using words, providing a direct reversal of the classic PSE. These and other previously published results are interpreted and integrated within a proposed theoretical framework that offers an account of the PSE. Phonological Similarity Effect Reexamined 3 Re-Examining the Phonological Similarity Effect in Immediate Serial Recall: The Roles of Type of Similarity, Category Cueing, and Item Recall A classic finding in the study of immediate serial recall of lists of verbal materials is that recall is poorer for lists consisting of phonologically similar items such as {cad, map, man, cap, mad} than for lists consisting of phonologically dissimilar items such as {pit, day, pen, bar, few} (e.g., Baddeley, 1966; Conrad, 1964). This is known as the phonological similarity effect (PSE); it has come to be viewed as one of the key phenomena characterizing immediate serial recall of verbal lists (e.g., Baddeley, 1992; Baddeley & Hitch, 1994; Baddeley, Gathercole, & Papagno, 1998). It also formed the basis of the theoretical proposal that the verbal short-term memory component of working memory employs a phonological code (Baddeley & Hitch, 1974; Baddeley, 1986); that is, the PSE constitutes key evidence for positing the very existence of what has come to be known as the phonological loop. It is thus a phenomenon of central importance to the influential working memory model of short-term memory. The standard interpretation of the PSE is that it arises as a result of interference between similar phonological memory traces in the phonological store (Baddeley, 1986). The classic PSE is a robust effect, and has been replicated numerous times (e.g., Baddeley, Lewis, & Vallar, 1984; Poirier & Saint-Aubin, 1996; Watkins, Watkins, & Crowder, 1974; Wickelgren, 1965). Other investigations have indicated, however, that phonological similarity may have no detrimental effect on immediate serial recall (e.g., Fallon, Groves, & Tehan, 1999, Experiment 1), and that it may even facilitate memory for item identity (as opposed to order) in immediate serial recall (e.g., Gathercole, Gardiner, & Gregg, 1982), and memory for order (as opposed to item identity) in order reconstruction tasks (Nairne & Kelley, 1999). This research has highlighted the importance of distinguishing between the typical serial, item-in-position criterion of correctness, whereby a list item is scored as correctly recalled only if it is recalled in the correct serial position (“strict serial recall”), and an order-free item criterion of correctness (“item recall”), whereby a list item is scored as correctly recalled if it is produced during recall of a list, whether or not it Phonological Similarity Effect Reexamined 4 was produced in the correct serial position. For example, Watkins et al. (1974) compared serial recall of phonologically similar and phonologically dissimilar lists. When performance was assessed using the strict serial recall measure, it was better for the phonologically distinct lists, demonstrating the classic PSE. However, when performance was assessed using the item recall measure, it was no different for the phonologically similar versus dissimilar lists. Similarly, Gathercole et al. (1982) compared serial recall of phonologically similar and phonologically dissimilar lists. Using the strict serial recall measure, performance was better for the phonologically distinct lists; however, item recall was actually better for the phonologically similar lists than for the phonologically dissmilar lists (Gathercole et al., 1982, p.180). Along similar lines, a study by Poirier and Saint-Aubin (1996) examined serial recall of lists of 2-syllable words. Strict serial recall was better for the phonologically distinct lists, but item recall was no different for the phonologically similar versus dissimilar lists. However, there have also been studies that obtained the classic PSE (better recall for dissimilar than similar lists) using both the strict serial recall and item recall scoring criteria. For example, Coltheart (1993) found that recall was better for phonologically dissimilar than for phonologically similar lists in terms of both the strict serial recall and item recall measures (Coltheart, 1993, Experiment 1). Similarly, Drewnowski (1980) found that recall was better for the dissimilar than for the similar lists in terms of both strict serial recall and item recall (Drewnowski, 1980, Experiment 3). In an insightful analysis, Fallon et al. (1999) noted that the differing effects of phonological similarity obtained in different investigations appear to be related to how the notion of phonological similarity had been operationalized in the studies. The studies that obtained a facilitatory effect of phonological similarity at the item level employed rhyming list items in their phonologically similar lists (Gathercole et al., 1982; Wickelgren, 1965). The studies that obtained a detrimental effect or no effect of phonological similarity at the item level operationalized phonological similarity in terms of list items that shared high phonemic overlap but that did not Phonological Similarity Effect Reexamined 5 all rhyme within a list (Coltheart, 1993; Drewnowski, 1980). Fallon et al. (1999) suggested that the former type of phonological similarity (rhyming similarity) can act as an effective category cue, and therefore facilitates item recall; but that the latter type of phonological similarity (phonological overlap without rhyme) does not provide an effective category cue, and therefore does not facilitate item recall. Fallon et al. (1999) tested these hypotheses by examining immediate serial recall of rhyming lists, phonologically similar but nonrhyming lists, and phonologically dissimilar lists. In their first experiment, they found that item recall was indeed greater for rhyming than for phonologically dissimilar lists, consistent with the hypothesis that rhyme similarity can produce a category cuing effect. Item recall for phonologically dissimilar lists was greater than for the similar nonrhyming lists, consistent with the hypothesis that a category cuing effect is obtained only with rhyme similarity. Stric serial recall was equivalent for the similar rhyming lists and the dissimilar lists. However, strict serial recall was higher for the dissimilar lists than for the similar nonrhyming lists, a replication of the classic PSE. These results were consistent with the hypothesis that for similar rhyming lists, a detrimental effect of phonological similarity on order recall is offset by the beneficial category cuing effect of rhyme on item recall, leading to strict serial recall that is equivalent to that for dissimilar lists. For similar nonrhyming lists, however, there is no beneficial category cuing effect to facilitate item recall, but there still is a detrimental effect of phonological similarity on order information, so that strict serial recall is worse than for dissimilar lists, yielding the classic PSE. Fallon et al. (1999) also tested the hypothesis that the effectiveness of category cuing for rhyming lists is a function of the uniqueness of the cue. The results just described were from their first experiment, in which list items did not recur across lists – that is, list items were drawn from what may be termed an open set. In a second experiment, Fallon et al. (1999) again examined immediate serial recall of dissimilar, similar rhyming, and similar nonrhyming lists, but the lists in each condition were now drawn from a closed set. That is, in each condition, lists were drawn from a small set of items, so that list items did recur across lists within a condition. Under these Phonological Similarity Effect Reexamined 6 circumstances, item recall for the rhyming lists was equivalent to that for dissimilar lists (rather than greater, as in their first experiment), consistent with the hypothesis that the effectiveness of rhyme as a category cue is affected by the uniqueness of the rhyme. Strict serial recall was greater for dissimilar than for rhyming lists, consistent with the hypothesis that the detrimental effect of phonological similarity on order information was not offset by a beneficial category cuing effect on item recall. There is at least one important issue that remains unresolved. Fallon et al. (1999) showed that item recall in phonologically similar rhyming lists was greater than in phonologically similar nonrhyming lists, and hypothesized that it is the presence of a rhyme category that is critical. However, the rhyming-nonrhyming manipulation in their experiments was confounded with a difference in the degree of within-list phonological overlap. Each member of similar rhyming lists such as {mat, fat, sat, rat, hat, bat} shared two phonemes, whereas each member of similar nonrhyming lists such as {rat, map, tab, fad, can, gag} shared on average only one phoneme. The difference in item recall for these two types of lists could therefore have been due to cuing by the degree of phonemic overlap rather than by rhyme category, a possibility that the authors acknowledged in discussing transposition errors (Fallon et al., 1999, p. 303). The question therefore is whether the effects obtained by Fallon et al. (1999) were due to cuing by rhyme category or cuing by the degree of phonemic overlap. An obvious way to address this would be to compare item recall for lists of rhyming versus non-rhyming stimuli with a controlled degree of phonemic overlap. However, the question makes contact with issues of considerably broader significance than simply controlling for a confound, because different theoretical accounts of the phonological similarity effect make different predictions about the relative effects of rhyme versus phonemic overlap. Let us consider these alternative accounts. In doing so, we will consider their predictions with regard to three types of lists. One type of list is comprised of phonologically similar rhyming stimuli that share a certain degree of phonemic overlap, say two phonemes, and this overlap is in the vowel and final consonant (e.g., {mat, fat, sat, rat, hat, bat}). Let us refer to these as rhyming lists. A second type Phonological Similarity Effect Reexamined 7 of list is comprised of phonologically similar non-rhyming stimuli that share as much phonemic overlap as the rhyming lists, in this case two phonemes, but overlap in the first consonant and phoneme (e.g., {cat, cab, cad, can, cap}). Let us refer to these as alliterative lists. Comparison of item recall for lists of these two kinds would address the question of whether cuing in the Fallon et al. (1999) study was a result of the degree of phonemic overlap or of the rhyme category. Let us also, however, consider a third type of phonologically similar list, in which the list items share some overlap, but the overlap is not consistently in the same place, and the total overlap within a list is not as much as in the other types of similar lists (e.g., {cad, cat, map, can, man}). Let us refer to these as canonically similar lists, as acknowledgment that this is the type of similarity utilized in some of the seminal studies that originally demonstrated the classic PSE (e.g., Baddeley, 1966). Let us now consider the predictions that various theoretical accounts would make for item recall of such lists. One kind of theoretical account that has been proposed is a feature model. In feature models in general, memory traces are represented as vectors of features. The feature model of Nairne (1990; Nairne & Kelley, 1999; Nairne & Neumann, 1993) incorporates representations of this type. In this model, the effect of phonological similarity in serial memory arises from overlap of the feature vectors that represent the phonologically similar list items. Phonological similarity makes it difficult to recover an item’s correct position within a list because there are overlapping features; however, common phonological features among list items can be used to discriminate the list as a whole from other lists, thus aiding item recall (e.g., Nairne & Kelley, 1999, p. 49). This latter aspect constitutes a means for phonological commonality to act as what Fallon et al. (1999) termed a category cue. Implicit in the feature account is the notion that it is the degree of overlap that matters for these effects. The feature model would therefore predict that item recall for lists comprised of phonologically similar rhyming stimuli should be equivalent to that for lists comprised of phonologically similar non-rhyming stimuli such as alliterative lists, if the degree of phonological overlap is controlled. In addition, it would predict that, based on the degree of phonological overlap, item recall should be greater for alliterative than for canonically similar Phonological Similarity Effect Reexamined 8 lists. We will refer to this as the feature account. An alternative account is derived from linguistic theory. According to linguistic analysis, a syllable has two constituents, an onset and a rime. The onset contains any consonants that precede the vowel, e.g., c in cat. The rime contains the vowel and any following consonants, e.g., at in cat. Thus, according to linguistic theory, for a consonant-vowel-consonant (CVC) syllable, the combination of the second and third segments corresponds to a theoretically defined linguistic category (the rime), whereas the combination of the first and second segments does not correspond to a linguistic category. An account of phonological similarity effects based on this linguistic analysis would view a two-phoneme overlap between the rimes of words in a list as being quite different in nature from a two-phoneme overlap between the first two segments of the words; and would predict that the potential for a category cuing effect should only arise in the case of rime overlap, which corresponds to linguistic category overlap (e.g., Nimmo & Roodenrys, 2004). This predicts that item recall for lists comprised of phonologically similar rhyming stimuli should be better than for lists comprised of phonologically similar non-rhyming stimuli such as alliterative lists, even if the degree of phonological overlap is controlled. Further, it predicts no difference in item recall between alliterative and canonically similar lists, because in neither case is there systematic linguistic category overlap between list items. We will refer to this as the linguistic structure account. A third account is derived from consideration of the computational requirements of a processing system that performs list recall in the verbal domain (e.g., Gupta, 1996; Gupta & MacWhinney, 1997). One point highlighted by computational analysis is that list recall requires the processing of serial order at both the level of lists (i.e., the serial order of items within lists), and at the level of words (i.e., the serial order of the constituents of individual words, which are, after all, phonological sequences). This raises the question of what effect, if any, the serial ordering within words might be expected to have on serial recall of the lists containing them. In a theory in which the serial order of phonemes within words is explicitly represented (as in the Gupta & MacWhinney, 1997 account), there is a basis for considering how serial order at this Phonological Similarity Effect Reexamined 9 level might affect serial ordering at the next level up. How might it play a role? One suggestion comes from work by Gupta and Dell (1999), who noted that a sequence of words such as {cat, cab} that share overlap at the beginning is more difficult to produce than a sequence of words such as{cat, mat} that share overlap at the end, as shown by Sevald and Dell (1994). Following this earlier work, Gupta and Dell (1999) suggested that this is because of the serial order of phonemes within word forms. The idea is that words can be thought of as dynamic trajectories over time, in phonological space. Trajectories that start similarly but end differently (i.e., words that share overlap at the beginning but not at the end) are more difficult to discriminate than trajectories that start differently but end similarly (i.e., words that share overlap at the end but not at the beginning). Applying these ideas to the current issue, this suggests that cuing effects in list recall should be sensitive, not merely to the degree of overlap between list items, but to the serial position of the overlap. This predicts that there will be a difference between recall of lists whose items share overlap at their beginings versus at their ends. Specifically, it predicts that item recall for lists comprised of phonologically similar rhyming stimuli (in which the overlap is at the ends of the words) should be better than that for lists comprised of phonologically similar non-rhyming stimuli such as alliterative lists in which the overlap is at the beginnings of the words, even if the degree of phonological overlap is controlled. It also predicts that item recall will be greater for alliterative than for canonically similar lists, because of greater phonemic overlap in the former. We will refer to this as the serial order account. It is worth noting that the serial order account incorporates important elements of the feature account. In particular, the degree of feature overlap does form the basis of cuing effects, just as in the feature account. However, the serial order account additionally posits that the location of the overlap matters, thus leading to different predictions regarding item recall for rhyming versus alliterative lists. Thus the serial order account can be seen as building on and extending the feature account; the proposal that the serial order within list items is relevant is, nevertheless, an important difference. Thus the three accounts differ in their overall sets of predictions. The item recall predictions of the feature account are Rhyming = Alliterative > Canonical. The item recall predictions Phonological Similarity Effect Reexamined 10 of the linguistic structure account are Rhyming > Alliterative = Canonical. The item recall predictions of the serial order account are Rhyming > Alliterative > Canonical. The first goal of the present work was to test these predictions and thus discriminate between these theoretical accounts, by comparing item recall in immediate serial recall of rhyming and alliterative lists with the same degree of overlap, as well as immediate serial recall of canonically similar lists, with lesser overlap; Experiments 1 and 2 addressed this goal. A second goal was to test the hypothesis proposed by Fallon et al. (1999) and Nairne and Kelley (1999) that a category cue will be more effective when using an open rather than a closed set; this aim was addressed by Experiments 3 and 4. A third goal, addressed in Experiment 5, was to examine whether item recall could be boosted sufficiently to lead to greater strict serial recall for similar than dissimilar lists. A fourth overarching goal was to articulate a theoretical framework that can serve to interpret and integrate the varied findings regarding the PSE in immediate serial recall into a coherent and unified account; in the General Discussion, we offer such a framework, and attempt to show that it can relate various findings to each other.