حافظه کاذب و اثرات آینه ای: نقش آشنایی و ارتباط عقب مانده در ایجاد خاطرات کاذب

The mirror effect refers to a phenomenon where the hit rate is higher for low frequency words while the false alarm rate is higher for high frequency distractors. Using a false memory paradigm (Roediger & McDermott, 1995), we examined whether false memory for non-presented lures would be influenced by the lure’s familiarity. The results revealed that false memory levels for low familiarity lures were higher than that for high familiarity lures, but only when the backward association strength between the presented list’s words and the lure was high. The veridical memory for the presented words also revealed greater accuracy for low familiarity words. In contrast, higher false alarms were observed for high frequency unrelated distractors. These results are discussed in light of current theories of the false memory effect, and it is suggested that they support an activation/monitoring account of the effect, according to which non-presented lures are activated during encoding.

Memory illusions, such as remembering events that never occurred, or remembering them differently from how they happened, are well known phenomena in the history of memory research (Roediger, 1996 and Schacter, 1995). One variant of these illusions is false memory for a non-presented word, which was first described by Deese (1959). Deese developed a paradigm in which participants were presented with lists of words, each list containing 12 primary associates of a critical not-presented word (lure). After the presentation of each list, participants were asked to recall the words from the previously exhibited list. Surprisingly, although the lure word was not presented in the list, it was falsely recalled at very high levels, reaching intrusion rates of 44% for some lures. In spite of the robust false recall elicited by the Deese (1959) paradigm, subsequent memory research generally ignored his findings (see Bruce & Winograd, 1998, for an analysis of the topic from philosophical and sociological perspectives). The paradigm was revived by Roediger and McDermott (1995), who replicated the original study with a new set of stimuli. In addition, they modified the paradigm by introducing a recognition test following the lists’ presentation. Their results also showed a high occurrence of false memory. Specifically, the non-presented associates were recalled 40–55% of the time, a rate similar to presented words appearing in the middle of the list. In the recognition test, the results were more dramatic: The false alarm rate for the critical lures was almost equivalent to the hit rate for the words appearing in the lists. Theoretical interpretations of false memory Two basic accounts of false memory have emerged in the literature: The first, which we refer to as memory-based approach (following Hirshman & Arndt, 1997), emphasizes encoding and representational factors as major determinants of the emergence of false memories. The second account, which we refer to as a decision-based account, claims that decision processes could influence the production of memories that did not occur. In the following section, we will elaborate on the former approach, while postponing the discussion on the latter until the General discussion. Within the broad category of memory-based theories, diverging accounts of false memory effects can be pinpointed. One line of theories emphasizes the role of associative processes in creating false memories (e.g., Roediger et al., 2001, Robinson and Roediger, 1998 and Roediger et al., 2001). According to this approach, the presentation of a word during the encoding phase activates its associates, including the critical lure, in the semantic memory network. The cumulative activation of the lure by the multiple words in the list causes participants, in a subsequent memory test, to experience source confusion and to misidentify the lure as a presented item (Johnson, Hashtroudi, & Lindsay, 1993). The activation of the lure during the encoding stages could be either conscious or unconscious. Underwood (1965), for example, argued that participants consciously generate the associated lure during the study phase. Others, however, have demonstrated that false memories can arise even when the list is presented at a rapid pace, preventing the conscious processing of both the words and the lure (Seamon, Luo, & Gallo, 1998; see also Thapar & McDermott, 2001 who showed false recall and recognition for lures from lists that were processed at a shallow level). A second line of theories focuses on faulty encoding of the presented words as a major determinant in the production of false memory (Dodson and Schacter, 2001 and Koutstaal and Schacter, 1997). According to fuzzy trace theory (Brainerd et al., 1995 and Reyna and Brainerd, 1995) memory is based on either a “gist” trace, which preserves the meaning and interpretations of the experience, or a “verbatim” trace, which preserves the specific features of the experience. False memories are attributed to remembering the gist but not the verbatim representation of the presented words, while veridical memories are driven by an item-specific verbatim trace. This reliance on gist traces may result from a pattern separation failure (McClelland, McNaughton, & O’Reilly, 1995), which is the inability to form non-overlapping unique representations of the items in the list. As a result, participants have good memory for the common aspects of the words, but poor memory for the items themselves. Thus, the lures will be identified as part of the presented list because of its common attributes (see also Hunt & McDaniel, 1993 for a similar account). The two approaches differ on the question of whether false memory for the non-presented lure is actually a genuine recollection of events occurring during study, or an illusory experience constructed during retrieval. According to the associative process account, the critical lure is activated (consciously or unconsciously) during the encoding stage and becomes an entity that is experienced during encoding and related to other events taking place at that time. In contrast, according to fuzzy trace theory and similar accounts, the failure to identify the critical lure as a non-presented item is not related to the activation status of the lure but to the encoding of the presented items. Thus, the activation of the lure is not a necessary condition for false memory. Support for the claim positing that activation processes underlie the creation of false memories is provided by participants’ phenomenological experiences of these non-studied lures. Using the remember-know judgement paradigm (Tulving, 1985) numerous studies have found that participants can report re-experiencing the lures’ presentation during study at levels equal to those of items that were actually studied (e.g., Gallo et al., 2001, Israel and Schacter, 1997 and Roediger and McDermott, 1995). The fact that participants claim conscious recollection of the lure suggests that it was indeed activated during study, thus sharing similar features with the list’s items (but see Brainerd, Wright, Reyna, & Mojardin, 2001 for an alternative account from a fuzzy-trace theory perspective). However, one drawback of this line of reasoning is that conclusions are based on subjective evaluation of previous memory judgements and not on more objective data (Brainerd et al., 2001). Extracting evidence directly from the data may be a more reliable technique for observing whether memory for non-presented lures resembles veridical memory. One possible approach would be to investigate how the lure’s frequency influences false memory, since past research has found diverging effects of frequency on memory for presented and non-presented items. The mirror effect in recognition memory A common practice in memory research is to manipulate word frequency to influence memory. Studies have found that in a typical old–new recognition paradigm, the hit rate is higher for low frequency than high frequency targets. The false alarm rate, however, is higher for high frequency than for low frequency distractors (Chalmers and Humphreys, 1998, Glanzer et al., 1993 and Glanzer et al., 1998). This phenomenon is known as the word frequency effect or as the mirror effect because of the opposite influence of word frequency on hits and false alarms. Several theoretical accounts have been offered to accommodate these findings. As in the field of false memory, they can be categorized as either decision-based or memory-based theories. Decision-based theories emphasize the role of decision processes in the emergence of the mirror effect (e.g., Glanzer et al., 1993, Greene, 1996 and Hirshman, 1995). Greene, for example, suggests that the mirror effect arises from participants’ inclinations to equate the pattern of their responses to the different types of stimuli. Since old low frequency words generate more yes responses than high frequency words in a yes–no recognition test, due to their being more memorable, fewer yes responses will be produced for new low frequency than for new high frequency words. Memory-based accounts, on the other hand, emphasize the unique encoding or representation of low frequency compared to high frequency words. The Source of Activation (SAC) model is one such account that was recently proposed to explain frequency effects in recognition memory ( Reder et al., 2000). According to the SAC, when a word is presented at study it activates its concept node and event node, the latter containing contextual information relating to the specific encoding episode. During retrieval, activation spreads from the concept node to its associated event nodes. Concept nodes of high frequency words are more activated, since they are more frequently and recently seen, but also more event nodes are attached to them, so each of their many event nodes receives little activation from the concept node. In contrast, concept nodes of low frequency words are less activated, but fewer event nodes are attached to them so each event node receives greater activation. The mirror effect stems from the reliance of participants on the activation of the event nodes, which leads to greater accuracy for low frequency words. In contrast, reliance on the activation of the concept nodes of new words will result in greater false alarms for high frequency words, which are characterized by higher activation of the concept node. Despite the large differences between decision-based and memory-based accounts of the mirror effect, they share a common notion that old and new words vary in their probability distributions across a memory strength dimension that represents their recent occurrence. Thus, the mirror effect can serve as a useful tool in comparing memory for presented and non-presented items (related as well as unrelated) and in delineating the processes underlying false memory. If indeed the critical lure is activated during the study phase, similar to the activation of a presented list word, then a word frequency effect should emerge: Greater old responses should be found for low frequency lures than for high frequency lures. However, if false memory arises as a result of pattern separation failure, originating at the encoding stage but manifesting itself during the retrieval stage when participant base their judgments on gist traces, there should be no difference in falsely remembering low versus high familiar lures. The predictions of this approach would be a lack of lure’s familiarity on false memory. Alternatively, high frequency lures would be falsely recognized more often than low frequency lures, just as more false alarms are found in recognition performance for new high frequency words. The effect of the familiarity of the lure on memory was first investigated by Deese (1959) but has not been pursued since then. The reason might be that Deese did not find a significant correlation between the lure’s frequency and the magnitude of its false recall (r = .06). However, two methodological drawbacks in his study cannot allow us to discount potential frequency effects on false memory: First, the word frequency counts of the lures (taken from Thorndike & Lorge, 1944) were quite low. Thus, the lack of variability in frequency might have obscured any potential effect, as Deese himself admitted (p. 20). Second, recall was the only measure assessing memory performance in his study. Thus, the question of whether the lure’s frequency can influence false recognition is yet to be answered. The reason to examine whether the lure’s frequency influences false recognition stems also from findings that frequency effects in recall are influenced by factors that are not crucial in recognition, such as list structure (i.e., pure versus mixed lists that contain both high and low frequency words; Ward, Woodward, Stevens, & Stinson, 2003). The primary aim of the present study was, therefore, to use the mirror effect as a tool to investigate the processes underlying false memory for non-presented lures. To this end, we manipulated the frequency scores of non-presented lures and probed their false memory with both recognition and recall. Studies in which the mirror effect was found used word frequency as the independent variable (e.g., Glanzer and Adams, 1990, Greene, 1996 and Kim and Glanzer, 1993). The frequency rating of a word is usually based on word frequency norms such as those of Kuĉera and Francis (1967) or Thorndike and Lorge (1944), which count the appearance of words in a large corpus sample of published literature. These norms are themselves an estimate of the number of times that a word appears in natural language. However, this reliance on word frequency norms is problematic for a number of reasons: First, occurrence counts of low frequency words may be unreliable since the majority of norms are based on published literature which is revised by editors inclined to use stylized low frequency words (Rudell, 1993). Second, since the norms themselves are samples, they may be subject to sampling errors. Some researchers have claimed that the size of the corpus may be a major determinant in the effective sampling of word use (Breland, 1996 and Burgess and Livesay, 1995). This problem is especially acute in the existing Hebrew norms (Balgur, 1968), which are based on a corpus of only 160,000 words sampled from published literature designed for elementary school pupils. In order to circumvent the potential problems that may arise with the use of normative frequency measures, we used a measure of experiential familiarity (Gernsbacher, 1984) in the present research, in addition to printed word frequency. This is a subjective rating in which participants were asked to rate how familiar they were with each word. The ability of this measure to approximate the frequency of occurrence of a word in natural language was found by Gernsbacher to reliably predict word recognition. In addition, the use of experiential familiarity measures instead of printed frequency measures has proved to be helpful in resolving inconsistencies in the word recognition literature where printed word frequency was manipulated along with a second variable. Furthermore, Gernsbacher found the converging validity of this measure to be quite high, yielding a correlation of .81 with printed frequency (see also Balota, Pilotti, & Cortese, 2001 for similar results). Recent studies by Chalmers and colleagues (Chalmers and Humphreys, 1998 and Chalmers et al., 1997) have also shown that the mirror effect can be obtained with measures other than normative word frequency. Finally, we also performed a pilot study, which will be described more elaborately in the Method section of Experiment 1, to demonstrate that the mirror effect is also obtained when experiential familiarity is used to define high versus low frequency items.