حافظه کاذب آواشناسی در کودکان و بزرگسالان: مدارک و شواهد برای واژگونی توسعه

False memories created by the Deese/Roediger–McDermott (DRM) procedure typically show a developmental reversal whereby levels of false recall increase with age. In contrast, false memories produced by phonological lists have been shown to decrease as age increases. In the current study we show that phonological false memories, like semantic false memories produced by the DRM procedure, show a developmental reversal when list items converge on a single critical lure. In addition, effects of list length were observed in adults and older children but not in the younger children, again mirroring effects previously observed in semantic false memories. These findings suggest that differences in list structure underlie the divergent developmental trajectories previously reported in semantic and phonological false memories. The findings are discussed in relation to theories of false memory and theories of spoken word recognition.

Roediger and McDermott (1995) demonstrated a powerful and robust false memory illusion that does not depend on providing participants with suggestive materials at study or leading questions at test. Based on a procedure originally developed by Deese (1959), participants are presented with lists of semantic associates of a nonpresented critical lure (e.g., participants are presented with thread, pin, sewing, sharp, etc., which are associates of the critical lure needle). Deese found that the probability of extra-list intrusions at recall was proportional to the probability of the critical lure being generated as an associate in response to the list items; i.e., backward associative strength (BAS). Roediger and McDermott revived and extended what is now referred to as the Deese/Roediger–McDermott (DRM) paradigm and found robust false memory effects in tests of both recall and recognition, with levels of false memory matching or even exceeding levels of veridical memory. Two theoretical accounts of the DRM illusion have been proposed. According to activation-monitoring theory (Roediger, Watson, McDermott, & Gallo, 2001), critical lures are activated at encoding in response to the study lists. They are then subject to errors of source monitoring at test (see Johnson, Hashtroudi, & Lindsay, 1993) and falsely endorsed as having been studied. According to fuzzy trace theory (Reyna & Brainerd, 1998), participants encode two traces of study items; a verbatim trace that encodes specific details of an item and its encoding context, and a gist trace that preserves relational information about the meaning of an item or list of items. False memories occur when critical lures presented at test match the gist traces formed at study, but this effect can be suppressed via a process of recollection rejection whereby lures are rejected on the grounds that they do not contain the expected verbatim information. False memories have also been investigated using lists of words that are associated phonologically rather than semantically. For example, Sommers and Lewis (1999) presented participants with lists of the most confusable phonological neighbours of a single critical lure. They found high levels of false recall, with no difference between the mean probability of recalling the critical lure and that of recalling studied items (.54 and .58 respectively). These results parallel those found by Roediger and McDermott (1995) in semantic false memory. In recognition tests, participants responded old to critical lures 64% of the time, a rate similar to that at which participants falsely recognised semantic critical lures in Roediger and McDermott’s study. High levels of false memories produced by phonological lists have been reported in a number of subsequent studies (e.g., Ballardini et al., 2008, Budson et al., 2003, McDermott and Watson, 2001, Sommers and Huff, 2003, Watson et al., 2003 and Westbury et al., 2002). Sommers and Lewis (Experiment 3) found that the magnitude of the false memory effect was related to the confusability between critical lures and phonological neighbours. They used the Frequency Weighted Neighbourhood Probability (FWNP) calculation developed by Luce and Pisoni (1998) to calculate the confusability between a critical lure (e.g. hit) and one of its phonological neighbours (e.g. heat) whilst also controlling for word frequency. Significantly more critical lures were recalled in lists comprising most confusable neighbours and, at recognition more false alarms were made to critical lures of lists with most confusable neighbours. As discussed by Sommers and Lewis, these findings are consistent with the Neighbourhood Activation Model (NAM) of word recognition (Luce & Pisoni), in which words are organised into neighbourhoods based on phonological similarity. According to the NAM, words are similar to a stimulus word if they can be created by adding, subtracting or deleting a single phoneme. For example sadder (addition), add (deletion) and mad (substitution) all belong (amongst others) to the similarity neighbourhood of sad. Moreover, graded activation of phonemes occurs such that items that are more similar to stimulus input will receive greater activation than those that are less similar. This feature of the NAM can be compared with activation models of semantic false memories, in which levels of false memory are determined by the strength of association between studied items and critical lures (see Roediger et al., 2001). Despite the similarity between the phonological false memories reported by Sommers and Lewis (1999) and the semantic false memories produced by DRM lists, findings from subsequent studies suggest that they are supported by different underlying processes. Watson et al. (2003) found that hybrid lists (containing both semantic and phonological associates of the critical lure) produced levels of false recall and recognition that were greater than what would be expected from lists comprising only one type of associate. They suggested that this over-additive pattern was due to differences between the activation processes for the two list types (conceptually-based processes in semantic lists versus perceptually-based processes in phonological lists). In support of this, Ballardini et al. (2008) found that phonological false recall peaked after short (20 ms) presentation durations, whereas semantic false recall peaked after longer (250 ms) presentation durations. Ballardini et al. suggested that the 20 ms window was sufficient to activate the perceptual associations that give rise to phonological false memories, but not the conceptual associations that give rise to semantic false memories. Evidence for the independence of semantic and phonological false memories was also provided by Ballou and Sommers (2008). Using an individual differences approach, they found significant correlations between semantic and phonological lists for veridical memories but not for false memories. A further difference between semantic and phonological false memories, and the focus of the current study, is their developmental trajectory. A counterintuitive finding from the DRM literature is a developmental reversal, whereby young children produce lower levels of false memory than older children, adolescents, and young adults (for examples see Anastasi and Rhodes, 2008, Brainerd et al., 2002, Howe, 2005, Howe, 2006 and Sugrue and Hayne, 2006; for reviews see Brainerd et al., 2008 and Holliday et al., 2011). A similar pattern has been observed in studies using lists of category exemplars (Brainerd & Reyna, 2007). Developmental reversals in false memory have been interpreted in terms of activation models and fuzzy-trace theory. According to the associative activation theory proposed by Howe, 2005 and Howe, 2006; Howe, Wimmer, Gagnon, & Plumpton, 2009), developmental reversals reflect the increasing automaticity with which associates are activated at study. According to fuzzy-trace theory, developmental reversals reflect a trade-off between gist-extraction and the use of verbatim traces to reject critical lures. A key component of FTT is that developmental reversals are driven entirely by improvements in gist connection, rather than changes in the familiarity of the individual items (Brainerd et al., 2008). As discussed above, developmental reversals have been observed with lists of semantic associates (either associatively or categorically related). A similar reversal has also been reported in studies of memory suggestion (e.g., Pezdek and Roe, 1995, Pezdek and Roe, 1997 and Ross et al., 2006; see Brainerd et al., 2008). Developmental reversals thus appear to be common feature across a number of false memory paradigms. However, phonological false memories are an exception to this pattern as they have been shown to decrease as age increases. For example, Dewhurst and Robinson (2004) presented children aged 5, 8, and 11 years with shortened DRM lists in which each item had at least one possible rhyming word (although the studied words did not rhyme with each other). They observed a developmental shift from phonological to semantic false memories, whereby 5-year-olds falsely recalled more phonological associates of studied items, but fewer semantic associates of studied items, relative to 11-year-olds. This pattern suggests that young children make associations based primarily on phonological features of studied items, whereas older children (and adults) make associations based primarily on semantic features. Divergent developmental trajectories for semantic and phonological false memories were also observed by Holliday and Weekes in false recognition. Specifically, they found that false recognition of critical lures from semantically related lists increased with age, whereas false recognition of critical lures from phonological lists decreased with age (see Brainerd & Reyna, 2007, for a similar age-related decline in phonological false recognition). Holliday and Weekes (2006) discussed their findings in terms of the Cohort Model of word recognition (Marslen-Wilson & Tyler, 1980). Due to the fact that the initial phoneme of a word is necessarily perceived first during spoken word recognition, the Cohort Model stipulates that activation spreads primarily from this phoneme. Importantly, this implies that words sharing early phones with the stimulus input will receive greater levels of activation than words sharing mid or late phonemes (see Wallace, Stewart, Sherman, & Mellor, 1995, for an extension of the Cohort Model to false recognition). In line with the view that early phoneme overlap leads to higher levels of priming of associated but nonpresented words, Holliday and Weekes found higher levels of false memories when study items shared the head phoneme with the critical lure. Holliday and Weekes also discussed their findings in terms of fuzzy-trace theory. They suggested that the developmental reversal in semantic false memory occurs because improvements in recollection rejection are opposed by increases in the processing of semantic gist (which is infinite). In contrast, phonological relations do not show the same development as semantic relations because of the finite nature of phonological relations. Age-related improvements in recollection rejection are therefore unopposed by any developments in the extraction of phonological gist, hence the decrease in phonological false memory. The developmental decrease in phonological false memory has important implications for theories of false memory development, as it represents a departure from the developmental reversal typically observed in false memory paradigms (see Brainerd et al., 2008). However, it is possible that the divergent developmental trajectories between semantic and phonological lists may be due in part to inconsistencies in the construction of the lists used in previous studies. Semantic DRM lists typically converge on a single critical lure and are chosen on the basis of backward associative strength (the strength of association from the study item to the critical lure). In contrast, the lists used in previous developmental studies of phonological false memories have not always consisted of converging phonological associates of a single critical lure. As discussed above, the lists presented by Dewhurst and Robinson (2004) were simply shortened DRM lists (i.e., list of semantic associates). Although the lists used by Holliday and Weekes (2006) shared phonological features with critical lures, they did not converge on one critical lure, as indicated by the high numbers of false alarms to related distracters (i.e., items other than critical lures that shared phonological features with the study lists). A notable exception to this is the study by Brainerd and Reyna (2007) which used lists taken from Sommers and Lewis that converge on a single critical lure. However, Brainerd and Reyna presented participants with shortened versions of these lists consisting of either a single associate or eight associates, and it may be the case that different effects will emerge with longer lists (we will return to the issue of list length in the General Discussion). It is possible, therefore, that the divergent developmental trajectories of semantic and phonological false memories are due, at least in part, to differences in list structure. In the current study, we investigated the developmental trajectory of phonological false memories using lists that converge on a single critical lure. The primary aim of the current study was to determine whether phonological false memories follow the same developmental trajectory as semantic false memories when study lists consist of the closest phonological associates of the critical lures. A second aim was to investigate the degree to which phonological false memory is influenced by list length, as this factor has been found to have a significant influence on semantic false memories (e.g., Robinson and Roediger, 1997 and Sugrue and Hayne, 2006). Two alternative predictions may be made regarding the developmental trajectory of phonological false memories for short and long lists. Sugrue and Hayne found that effects of list length on false memories created by semantic DRM lists were age-dependent. Specifically, adults showed higher levels of false recall and recognition than children in response to long lists (14 items) but not in response to short lists (seven items). However, phonological false memory levels have been shown to be high in children even with short lists and to decrease across development (Dewhurst and Robinson, 2004 and Holliday and Weekes, 2006). On this basis one might predict an effect of length that is developmentally invariant. An alternative prediction can be based on the use of phonological lists in which associates uniquely constrain the critical lure. Children might be expected to make high numbers of false memories to lists that merely allow for phonological associations (c.f., Dewhurst & Robinson, 2004) or to lists of associates that all differ by the same phoneme from the critical lure (e.g., Holliday & Weekes, 2006) because these relationships between words reflect simple item-to-item associations. However, associates that converge on a single critical lure might require a more complex associative network in order to make the activation of the critical lure more likely. Since the complexity of similarity neighbourhoods has been shown to increase with age (in order to allow discrimination within an increasing large lexicon; see Charles-Luce and Luce, 1990 and Charles-Luce and Luce, 1995), one might predict that older children and adults will use extra associates to increase activation levels to critical lures, whereas younger children will not benefit from the additional associates in long lists due to the relative immaturity of their associative networks. In this case, phonological false memories will follow the same developmental pattern as semantic false memories and show effects of list length in adults but not in children.