بررسی فعال سازی بدون نظارت منبع در پارادایم DRM حافظه کاذب

Four experiments examined whether studying a single Deese/Roediger-McDermott (DRM) list produces semantic priming for nonstudied critical items (CIs) and semantic + repetition priming for studied associates. After 30 s of mental arithmetic that followed the study of a DRM list, priming was assessed in a lexical decision task when the nonwords were either pronounceable (Experiment 1) or pseudohomophones (Experiments 2–4). Priming was measured relative to a baseline containing exactly the same CIs and associates that had not been primed by their related DRM lists. Significant CI semantic priming effects occurred in all four experiments, whether or not there was within-test priming from a related associate preceding the CI by 3–7 items. To our knowledge, these are the first experiments using standard DRM study procedures to provide a convincing demonstration of a genuine CI semantic priming effect in a delayed indirect memory test that should be free of intentional retrieval strategies. Discussion focuses on measuring long-term semantic activation effects without the influence of source monitoring in a lexical decision task.

In the Deese/Roediger-McDermott (DRM) false memory paradigm (Deese, 1959a, Deese, 1959b and Roediger and McDermott, 1995), people study a DRM list in which words (e.g., bed) are all associates of a single nonstudied critical item (CI, e.g., sleep). The probability of falsely recalling or recognizing nonstudied CIs as having been studied is often similar to (or sometimes higher than) the probability of correctly recalling or recognizing associates that were actually presented in the middle of the studied DRM list (e.g., Roediger and McDermott, 1995 and Gallo et al., 1997). This false memory effect occurs under a wide variety of conditions (see Roediger et al., 2001 and Roediger et al., 1998, for reviews). Roediger et al.’s two-process Activation/Monitoring theory assumes that the presentation of the associate blanket indirectly activates the representations in memory corresponding to related words such as the CI sleep (either automatically, cf. Neely, 1977, or strategically, through implicit associative responses, IARs— Underwood, 1965). This indirect activation of the CI accumulates as each of its associates is studied. In the recall or recognition test, false memories occur for the highly activated CI when source monitoring fails, that is, when people misattribute the source of the indirect activation of the nonstudied CI to the direct activation that would have been produced if it had been studied. In short, Activation/Monitoring theory suggests that false memories are produced by the conjunction of a heightened indirect activation of nonstudied CIs and a failure of source monitoring. The Activation/Monitoring theory encourages one to attempt to isolate activation effects from the influence of source monitoring.1 One approach for doing this in the standard explicit recognition memory test is to reduce source monitoring by having a short response deadline that does not allow enough time for source monitoring to operate (e.g., Benjamin, 2001) or by testing populations with breakdowns in source monitoring (e.g., the early-stage Alzheimer’s Disease patients tested in Balota et al., 1999). Another approach, which is the focus of the current research, is to use indirect memory tests that presumably do not engage source monitoring. We now turn to a review of experiments that have used this approach. Indirect memory tests in the DRM paradigm Indirect memory tests presumably eliminate source monitoring because people are not explicitly asked to decide if a word was studied, but instead are asked to do a task apparently unrelated to the study phase, such as completing, with the first word that comes to mind, word stems or fragments in which some of the answers are previously studied words (Graf and Schacter, 1985 and Roediger and McDermott, 1993). Because people are unaware their memories are being assessed, they presumably do not attempt to intentionally retrieve the items from the prior study list or to monitor the source of the activation for these retrieved items. Thus, performance in these tests should not be influenced by source monitoring. To measure activation, a priming effect is computed that compares performance for test items actually presented in the study list (or nonstudied test items related to the study items) with performance on nonstudied baseline test items totally unrelated to the study items (e.g., McDermott, 1997). When the item in the indirect memory test has actually been studied, one is measuring semantic plus repetition (SEM + REP) priming because the test item is semantically related to the other studied associates and is also a repetition of an item that itself was studied. (This SEM + REP priming effect differs from the pure repetition priming effect measured for a word studied in a study list that does not contain other semantically related items.) By similar reasoning, for a test item, typically the CI, that was itself not studied but is related to the studied DRM items, one is measuring semantic (SEM) priming. 2 In both cases, priming is said to have occurred if the probability (or speed) of responding with a CI or associate is higher (or faster) when its related DRM list was studied compared to when it was not. Word-production tasks as indirect memory tests for assessing activation from DRM lists To our knowledge, seven published reports have used word-production tasks in indirect memory tests to measure activation from intentionally studied DRM lists without the influence of source monitoring. Word production was cued by either (a) a semantic cue [e.g., drowsy as a cue for sleep in a word association test ( McDermott, 1997)], (b) an orthographic cue [e.g., sl___, s_e_ _ or the anagram eslpe as a cue for sleep ( Hicks and Starns, 2005, Løvdén and Johansson, 2003, McDermott, 1997, McKone and Murphy, 2000 and Smith et al., 2002)], or (c) a briefly presented (50 ms) backward-masked presentation of the word itself as a cue (e.g., the perceptual identification tests in Cleary and Greene, 2004 and Hicks and Starns, 2005). To perform these tasks, people presumably generate possible candidate words to each cue and then select one that is an appropriate response. The studies using word-production tasks focused primarily on CI SEM priming (which presumably can serve as a measure of the activation underlying false alarms for nonstudied CIs in direct memory tests) and associate SEM + REP priming (which presumably can serve as a measure of the activation underlying hits for studied associate in direct memory tests). In these investigations, perceptual identification tests consistently yielded only SEM + REP priming effects, but the other indirect memory tests yielded both SEM and SEM + REP priming effects. 3 This is congruent with the claim (e.g., Thapar and Greene, 1994 and Weldon, 1991) that a perceptual identification test is more data-driven than other kinds of indirect memory tests, perhaps due to the stimuli in this test being intact words and/or being presented briefly and masked. Of special relevance to the present research is the concern that source monitoring or other deliberate intentional (i.e., direct-memory) retrieval strategies could have, along with activation, contributed to these significant CI SEM priming effects in the prior word-production studies. Such a concern seems justified on two grounds. First, the long response deadlines employed may have encouraged intentional retrieval strategies. Second, for 48% or more of the test cues, one of the potential correct responses was related to or presented in the studied DRM lists. (Exceptions are Hicks & Starns, 2005 and Smith et al., 2002, in which fewer than 26% of the test cues were for items related to or presented in the studied DRM lists.) If people became aware that many of the to-be-produced words were related to or were actually presented in studied lists, this would have encouraged them to use intentional retrieval strategies (see Kinoshita, 2001), for which the long response deadlines would have allowed time.4 We therefore believe that a better indirect memory test for assessing activation in the absence of source monitoring and other intentional retrieval strategies is the lexical decision task (LDT), to which we now turn. The LDT as an indirect memory test for assessing activation from DRM lists In the LDT, people press one of two keys to indicate whether a letter string is a word or a nonword. Because a lexical decision can be made by referring only to one’s knowledge of orthography, phonology, or meaning, there is no need to (i) retrieve previous study episodes, (ii) generate potential response candidates and verify their appropriateness for output, or (iii) monitor the source of the generated test item’s activation. Hence, the LDT seems to be a good candidate for providing a relatively pure measure of how much activation the study of a DRM list produces for studied associates and nonstudied CIs. That is, the greater the activation that the prior study of a DRM list produces for a word, the faster or more accurate lexical decisions to that word should be. However, there are reasons to question whether the LDT is well suited for assessing SEM priming from a DRM list. On the one hand, although repetition priming in the LDT can last for minutes (e.g., Dannenbring and Briand, 1982, Duchek and Neely, 1989 and Forster and Davis, 1984) or even hours (e.g., Scarborough, Cortese, & Scarborough, 1977), SEM priming from a single semantically related prime is greatly reduced or even eliminated if one or more unrelated items or a few seconds intervene between the prime and target (e.g., Dannenbring and Briand, 1982, Joordens and Besner, 1992 and Masson, 1991). On the other hand, more recent papers have reported long-term semantic priming both from multiple related primes (Becker, Moscovitch, Behrmann, & Joordens, 1997) and even from single related primes (Hughes and Whittlesea, 2003 and Joordens and Becker, 1997) under conditions in which both the primes (and targets) were processed to a deep semantic level. Thus, it seems reasonable that multiple related primes in a DRM list studied in anticipation of a subsequent memory test could produce CI SEM priming in the LDT. To our knowledge, five different laboratories have conducted 13 LDT experiments all of which used visually presented study lists to examine CI SEM priming effects, and these experiments have yielded mixed results (see Table 1). Specifically, whereas associate SEM + REP priming has been reliably found in all five laboratories, CI SEM priming has been found in only two of them—Hancock, Hicks, Marsh, and Ritschel (2003) and Whittlesea (2002). We now discuss the procedural differences that we believe underlie these conflicting findings.

The present results are the first we know of that convincingly demonstrate that a genuine CI SEM priming effect is produced by a single DRM list studied under standard DRM conditions in a delayed indirect memory test that likely eliminates (or at least greatly minimizes) intentional retrieval strategies and source monitoring processes. More specifically, the present results firmly establish three new findings that we believe are important: (1) The procedures developed by Hancock et al. (2003) do indeed produce a true CI SEM priming effect from a single DRM list that lasts for about 35 s, although their use of frequency/length-matched control items rather than unprimed CIs did not allow them to conclude with certainty that their CI SEM priming effect was genuine. (2) This CI SEM priming effect occurs whether the DRM study list is presented at a 2 or 5 s rate but dissipates rather rapidly after 35 s unless it is bolstered by within-test priming from a related associate that was studied in the related DRM study list that produced the SEM priming for that CI. (3) A long-term semantic priming effect, at least from multiple primes, is replicable across a number of different procedural variations and labs and does not depend on exactly the same processing task being performed on the primes and target. Hence, long-term semantic priming from multiple primes is a phenomenon that theories of semantic priming can no longer ignore.