Illusory Recollection 1 Running Head: ILLUSORY RECOLLECTION Illusory Recollection

Higham and Vokey (1999, Experiment 1) demonstrated that identifying a briefly displayed word prior to presenting that item in the clear for recognition, enhanced the bias to respond “old” relative to misidentification and that this effect was independent of item selection artifacts. Three experiments investigated the phenomenology associated with this illusion of memory using the standard R/K procedure and a new independent scales technique. Contrary to expectations based on the fluency heuristic, which predicts effects of prime identification on familiarity only, all experiments revealed that the illusion was experienced as both familiarity and recollection. Furthermore, prime identification produced false recollection. The results, in particular the implications of false recollection, are discussed in terms of dual-process, dual-trace and two-criteria signal-detection models of R/K judgments. Illusory Recollection 3 Illusory Recollection Research in recognition memory has increasingly focussed on the subjective experience of remembering. One approach toward this goal that has received a great deal of attention recently is the remember/know procedure, first introduced by Tulving (1985). With this procedure, participants are instructed to indicate, for each test item rated “old” on a recognition memory test, whether it was “remembered” or “known” (R and K judgments, respectively). If the item is given an R rating, that is an indication that some aspect or aspects of the encoding context was remembered at the time the item was rated “old.” K ratings, on the other hand, indicate that the person believes that the test item was in the previous study list, but that the details regarding the previous encoding of the item are not available at the time the “old” response was made. Remember/Know and Dual-Process Models In Tulving’s (1985) original proposal, R and K judgments were associated with retrieval from the episodic and semantic memory systems, respectively. However, Gardiner and colleagues (e.g., Gardiner, 1988; Gardiner & Java, 1990, 1993; Gardiner and Parkin, 1990) extended the distinction so that it applied to dual-process models of recognition memory (e.g., Jacoby & Dallas, 1981; Mandler, 1980). According to this new distinction, when participants give an R rating, they are indicating that the test item was consciously recollected (Jacoby & Dallas, 1981) or that the item underwent inter-item elaboration (Mandler, 1979, 1980) at encoding. Conversely, K judgments, either directly (e.g., Gardiner & Java, 1993), or with a correction factor (Yonelinas & Jacoby, 1995), are thought to indicate feelings of familiarity (in the absence of recollection) caused from perceptual fluency (Jacoby & Dallas, 1981) or integration of the perceptual features of the item (Mandler, 1979, 1980). It appears to be generally accepted among researchers in this literature that R judgments are a good measure of conscious recollection (e.g., Gardiner, Gawlik & Richardson-Klavehn, 1994; Jacoby, Yonelinas & Jennings, 1997; Knowlton & Squire, 1995; Yonelinas & Jacoby, 1995). On the other hand, a great deal of controversy has centered around K judgments and whether or not a correction factor should be applied to the K rates before they can be considered accurate estimates of familiarity (see Jacoby, 1998; Jacoby et al.,1997; Yonelinas & Jacoby, 1995). Illusory Recollection 4 Familiarity has traditionally been associated with perceptual processing in dual-process models, whereas conscious recollection has been associated with conceptual or more elaborative processing (e.g., see Jacoby & Dallas, 1981, although see also Jacoby 1991, Rajaram, 1996, 1998 and Whittlesea, 1993 for a more recent discussion of conceptual effects on familiarity). If, indeed, R and K judgments are measuring conscious recollection and familiarity, respectively, then manipulating conceptual versus perceptual processing differentially should produce dissociations of R and K responding. Many studies have produced results consistent with this hypothesis. For example, semantic encoding of study items, which should enhance conceptual processing, leads to more R judgments, but not more K judgments, than nonsemantic encoding (Gardiner, 1988; Rajaram, 1993). In contrast, increasing perceptual fluency, by preceding a recognition stimulus with a brief presentation of itself rather than a different word, results in more K judgments, but not more R judgments (Rajaram, 1993). Such differential sensitivity of R and K judgments to conceptual and perceptual manipulations would be expected by early dual-process models of recognition. More recently, Rajaram (1998; see also Rajaram, 1996; Rajaram & Roediger, 1997) has suggested an alternative to the mapping of R and K judgments to conceptual and perceptual processing. She notes that there are a number of experiments in which perceptual, rather than conceptual, processing has an effect on R judgments. For example, the reinstatement of both the size (Rajaram, 1996, Experiment 2) and orientation (Rajaram, 1996, Experiment 3) of picture stimuli between study and test significantly enhanced R judgments. Because reinstatement of these perceptual characteristics should have no effect on conceptual processing, these results directly counter the view that R judgments are selectively sensitive to variations in conceptual processing. To account for these results, as well as all other published findings using the remember/know procedure, Rajaram (1998) proposed that R judgments are sensitive to the distinctiveness or salience of encoding of the recognition stimulus, whereas K judgments are sensitive to variations in processing fluency. The distinction between distinctiveness/salience of the item versus its fluency is orthogonal to the distinction between conceptual versus perceptual processing. For example, it is possible to enhance both the perceptual and conceptual fluency of a recognition stimulus, and that will result in more K judgments. Similarly, it is Illusory Recollection 5 possible to enhance the distinctiveness of an item with either conceptual or perceptual variables, and that will result in more R judgments. Regardless of the particular dual-process distinction that is most useful for characterizing R and K judgments, at least one statement can be made that is true for all of them: R judgments are characterized as reflecting memory for encoding details of old items, whether that availability is produced by conceptual encoding processes or by the distinctiveness/salience of the old item. Stated simply, if R is thought to be measuring the degree to which the details of the initial encoding of old items become available at retrieval, it is necessarily measuring a dimension that only varies among previously encountered stimuli. An analogous item-based conceptualization of recollection is present in “fuzzy trace theory” (e.g., Brainerd, Reyna & Brandse, 1995; Reyna & Brainerd, 1995), although this theory is probably best described as a dual-trace theory, as opposed to a dual-process theory. According to this theory, encountering a stimulus serves to lay down two representations: a verbatim representation, which contains literal details of the original encoding incident, and a gist representation, which contains semantic content. At test, old items retrieve both verbatim and gist representations, whereas new item retrieve only the gist representation. Old/new recognition is based on the extent to which the retrieved representation(s) is similar to the test item’s surface form and/or gist. This model has been used to explain some false memory data (e.g., Brainerd et al., 1995; Payne, Elie, Blackwell & Neuschatz, 1996) and we will return to it in the General Discussion after describing our experiments. For present purposes, we need only mention that, under the assumptions of fuzzy trace theory, it is clear that the detailed, contextual, episodic content necessary to support R judgments is contained only in verbatim representations (representations derived from encountering old items). Gist representations (the only basis for responses to new items), by definition, do not contain this information. False recollection and Dual-Process Models What about false recollection and what role does it play in dual-process models? On the one hand, there is a growing body of research that has used the R/K procedure or some variant to examine the nature of false memories (e.g., Brainerd et al., 1995; Dalla Barba, 1993; Higham, 1998; Holmes, Illusory Recollection 6 Waters & Rajaram, 1998; Lane & Zaragoza, 1995; Norman & Schacter, 1997; Payne et al., 1996; Read, 1996; Roediger & McDermott, 1995; Roediger, Jacoby & McDermott, 1996; Schacter, Verfaellie & Anes, 1997). This research has found that participants quite often rate their false memories as “remembered” and a number of explanations have been advanced to explain such illusory recollection, including inappropriate binding of episodic content to false memories (Holmes et al.,1998) and incorrect source monitoring (Higham, 1998; Johnson, Hashtroudi & Lindsay, 1993; Lane & Zaragoza, 1995). Regardless of the particulars, any suitable explanation of false recollection must specify how a person can “reexperience” a memory that is rich in encoding detail, but which never happened. Although some R/K research has been directed at the nature of false memories and illusory recollection, there is also clear evidence that other R/K research, motivated from a dual-process theoretical viewpoint, is not even attempting to explain such effects. Consider first the fact that analyses of R data in published research have focussed principally on the R hit rate (although see Donaldson, 1996; Gardiner, Richardson-Klavehn & Ramponi, 1997; Hicks & Marsh, 1999; Hirshman & Henzler, 1998; Hirshman & Master, 1997; Strack & Förster, 1995 and a few others for exceptions). The R false alarm rate is assumed (and often shown) to be minimal and have a negligible effect on performance. As a result, analyses of false alarms or even discriminability measures (d’ or A’) are rare. As Donaldson has stated, “Recently, there has been an increased tendency [in recognition memory research] to use the hit rate as if it accurately measures memory” (p. 523). Second, most researchers attempting to manipulate R judgments have done so by varying the kind of encoding that old item undergo. These variations include the level of processing of the study items (Gardiner, 1988; Rajaram, 1993), reading versus generating study words (Gardiner, 1988; Java, 1994), the attentional resources available during encoding (Gardiner & Parkin, 1990), the orthographic distinctiveness of the study words (Rajaram, 1998), the conceptual salience of the study words (Rajaram, 1998), the perceptual form of the study items (picture versus word: Dewhurst & Conway, 1994; Rajaram, 1993, 1996; Wagner, Gabrieli & Verfaellie, 1997), and the number of study time presentations (Jacoby, Jones & Dolan, 1998), to name a few. The focus on old item encoding when it comes to manipulating R judgments indicates that investigating false recollections is of lower priority in these lines of research. Illusory Recollection 7 It is also clear that some dual-process models of recognition memory do not even have the provisions, in their current form, to explain false recollection data. For example, in a recent formulation of the process dissociation procedure, a procedure that is based on a dual-process model, the likelihood that a new item will be rated “old” (false alarm) is assumed to be equal to the probability of the new item’s familiarity exceeding some response criterion (e.g., P[“old”|new]=Fnew; Yonelinas, Kroll, Dobbins, Lazzara & Knight, 1998, p. 329). In other words, recognition false alarms, according to this dual-process model, are entirely a function of response bias and the familiarity of the item, not recollection; therefore, conscious recollection, as measured by R responses, is necessarily detectable only in hit rate performance. False recollection and Signal-Detection Models In contrast to dual-process models, Donaldson (1996) has recently proposed a signal-detection model of R/K performance (see also Donaldson, MacKenzie & Underhill, 1996; Hirshman and Master, 1997). In this model, participants asked to make R/K judgments set two response criteria along a single dimension of familiarity. The first, more liberal criterion corresponds to yes/no recognition, whereas the second, more conservative criterion corresponds to yes/no remember. Thus, “new” judgments are made to items with familiarity below the first, recognition criterion, K judgments are made to items with familiarity between the criteria, and R judgments are made to items with familiarity above the second R criterion. In contrast to dual-process models, the signal-detection model can easily account for false recollection: it reflects the area of the new item distribution that is to the right of (above) the R criterion (i.e., high false-familiarity judgment). A signal detection model of this sort makes several testable predictions and has the potential to explain new findings from research on false recollections in terms of R criterion placement and old/new discriminability. Indeed, new research has been investigating how well such a model accounts for new and extant R/K data (e.g., see Donaldson,1996; Donaldson et al., 1996; Gardiner & Gregg, 1997; Hicks & Marsh, 1999; Hirshman & Henzler, 1998; Hirshman & Master, 1997; Hockley & Consoli, in press; Inoue and Bellezza, 1998). In summary, research using the R/K procedure has branched into investigations examining the nature of false memory and investigations testing the predictions and assumptions of dual-process Illusory Recollection 8 models of recognition. Whereas false recollection is the main focus in the first line of research, many dual-process models of recognition memory are unable to account for false recollection data at all because they assume that recollection is a process applied only to old items. In contrast, a two-criteria signal-detection model of recognition memory can predict false recollection results via criterion and sensitivity mechanisms. Overview of the Experiments There were two purposes to the research presented in this paper. First, given that a signal detection model can account for false recollection whereas some dual-process models cannot, we sought to further investigate false recollection in our experiments. Our experiments differed from other studies that have demonstrated illusory recollection in that our manipulation was not likely to affect the conceptual or semantic aspects of encoding. Many previous researchers have induced false recollections with material that is conceptually and semantically rich, such as lists of semantic associates (Deese, 1959; Israel & Schacter, 1997; Read, 1996; Roediger & McDermott, 1995) or complex sentences containing gist (Holmes et al., 1998). In contrast, our experiments reveal false recollection by varying a factor that many would describe as “perceptual”: prime display duration. Following a procedure we developed elsewhere (Higham & Vokey, 1999; see also Watkins & Gibson, 1988), participants first studied a long list of words, and later were administered a recognition memory test. However, prior to the presentation of each test item (target) for recognition, the same item was presented very briefly (prime) and then masked with ampersands (e.g., the target table was primed with a brief presentation of table). Participants were asked to try to identify the prime and then make a recognition response to the target presented in the clear. Unbeknownst to participants, we varied the duration of the prime presentation; primes in the long duration were presented for 30 ms longer than primes in the short duration condition.1 For the brief durations we used, we found that targets following long duration primes were more likely to be rated “old” under these testing conditions than were targets following short duration primes and that this effect of duration was mediated completely by identification performance. That is, prime display duration affected recognition to the extent that it varied the likelihood that the prime was identified. Consequently, we expected a similar effect of prime Illusory Recollection 9 identification on recognition in the experiments reported here. However, our main concern here was whether or not this illusion would be manifested in false alarms, and if so, whether it would be manifested as K judgments (familiarity) and/or as R judgments (recollection). The second purpose of the current research was to introduce a new methodology for investigating participants’ phenomenology during recognition. In Experiments 2 and 3, instead of using the standard R/K procedure, we asked participants to make two independent memory judgments for each item. After being taught the distinction between familiarity and recollection by adapting instructions that are now standard in the R/K literature, participants were asked to indicate how familiar the item was and, separately, to indicate how much the item was recollected. A different 4-point scale was used to make each judgment (1=low; 4=high for both). As we discuss below, this methodology avoids a number of undesirable side-effects of the R/K procedure, such as a necessary arithmetic dependence between the the R and K rates. Experiment 1 Other research has been conducted that uses prime manipulations similar to ours. For example, Rajaram (1993) replicated an illusion of memory first demonstrated by Jacoby and Whitehouse (1989). The illusion occurs when participants increase their bias to respond “old” to a recognition stimulus that follows a brief presentation of itself versus a brief presentation of some other, unrelated word. Rajaram found that when the remember/know procedure was used to measure participants’ phenomenology, this illusion of memory was manifested in K judgments. Rajaram maintained that her results supported Jacoby and Whitehouse’s claim that the perceptual fluency of the target was enhanced by the matching prime compared to the mismatching prime. This enhanced fluency resulted in a feeling of familiarity which, in turn, enhanced the likelihood of a K judgment. These results suggest that test-time perceptual manipulations of fluency are manifested as “feelings of familiarity” (e.g., Jacoby & Whitehouse, 1989; Whittlesea, Jacoby & Girard, 1990; Whittlesea, 1993). To explain the effect of prime identification on recognition described above, it could be argued that prime identification enhances the fluency with which the subsequent matching target is processed and that participants experience this fluency as familiarity which increases their bias to Illusory Recollection 10 respond “old.” From a fluency perspective, then, any effect of prime identification should be manifested in familiarity judgments (K). However, we argued (Higham & Vokey, 1999) that the memory illusion we observed was due to use of an identification heuristic, not target processing fluency. In short, under the testing circumstances of our experiments, participants used their prime identification performance as a basis of their recognition judgments. Participants may rely on identification performance because old items are identified better than new items (perceptual priming), so such performance is predictive of prior presentation. However, the memory illusion arises when participants over-rely on identification as a basis of performance and attribute too much variability in identification performance to prior presentation. This over-reliance results in prime identification having unwanted residual effects on recognition performance. Nonetheless, if this memory illusion is, indeed, based on prime identification success and not target processing fluency, it is not clear what the phenomenology associated with the memory illusion should be. To find out, in Experiment 1, we induced the memory illusion using a prime display duration manipulation and investigated participants’ phenomenology using standard R/K methodology. Method