اندازه گیری بازیابی خودکار: یک مقایسه از حافظه ضمنی، فرآیند گسست و روش پاسخ شتاب گرفته

Using the stem completion task, we compared estimates of automatic retrieval from an implicit memory task, the process dissociation procedure, and the speeded response procedure. Two standard manipulations were employed. In Experiment 1, a depth of processing effect was found on automatic retrieval using the speeded response procedure although this effect was substantially reduced in Experiment 2 when lexical processing was required of all words. In Experiment 3, the speeded response procedure showed an advantage of full versus divided attention at study on automatic retrieval. An implicit condition showed parallel effects in each study, suggesting that implicit stem completion may normally provide a good estimate of automatic retrieval. Also, we replicated earlier findings from the process dissociation procedure, but estimates of automatic retrieval from this procedure were consistently lower than those from the speeded response procedure, except when conscious retrieval was relatively low. We discuss several factors that may contribute to the conflicting outcomes, including the evidence for theoretical assumptions and criterial task differences between implicit and explicit tests.

Several procedures have been developed in an attempt to identify the contributions to memory performance of automatic retrieval processes—retrieval of previously studied information with no intent to do so. These include implicit memory tasks (Graf and Schacter, 1985 and Schacter, 1987), the retrieval intentionality criterion (Schacter, Bowers, & Booker, 1989), and the process dissociation procedure (Jacoby, 1991 and Jacoby, 1998). The utility of each has been challenged on several grounds. For example, Jacoby, 1991 and Richardson-Klavehn and Bjork, 1988 noted that parallel effects of variables on implicit and explicit tests may indicate contamination of implicit performance with conscious retrieval. In terms of the retrieval intentionality criterion, Richardson-Klavehn, Gardiner, and Java (1996) argued that conscious retrieval is not a necessary consequence of awareness of the episodic history of an item (Graf and Komatsu, 1994 and Roediger and McDermott, 1993). Finally, some of the theoretical assumptions of the process dissociation procedure have been questioned (e.g., Bodner et al., 2000, Horton et al., 2001, Joordens and Merikle, 1993 and Richardson-Klavehn et al., 1996). Horton et al., 2001, Vonk and Horton, in press and Wilson and Horton, 2002 described an alternative procedure that uses RTs to identify retrieval strategy. The speeded response procedure is based on the assumption that automatic retrieval executes faster than conscious retrieval (de Houwer, 1997, Reingold and Toth, 1996, Richardson-Klavehn and Gardiner, 1995, Richardson-Klavehn and Gardiner, 1996, Richardson-Klavehn and Gardiner, 1998, Toth, 1996, Vaterrodt-Plünnecke et al., 2002, Weldon and Jackson-Barrett, 1993 and Yonelinas and Jacoby, 1994), and we have reported data to support this assumption (Horton et al., 2001 and Wilson and Horton, 2002). Following a study task, subjects in a speeded response group first received practice stem completion tests in which none of the stems corresponded to studied items. Their instructions were to respond as quickly as possible with the first word that came to mind. To increase response speed, average RTs were presented to subjects at the end of each test and faster responding was encouraged on the subsequent test. Because subjects were encouraged to respond quickly and no stems corresponded to the studied items, subjects had no basis for adopting conscious retrieval on the practice tests. These design features were implemented to maximize the likelihood that conscious retrieval strategies would be excluded. The critical stem completion tests immediately followed the practice tests with the only difference being that, on the critical test, 50% of the stems corresponded to studied items. Although it would be possible to switch to conscious retrieval on the critical tests, a comparison of speeded response group RTs with those from a baseline group and an explicit group argued against that conclusion. Subjects in the baseline group performed exactly the same tasks as subjects in the speeded response group except that none of the stems on the critical test corresponded to previously studied items. The explicit group was treated the same as the speeded response group except that, immediately prior to the critical test, they were instructed to switch to a conscious retrieval strategy. That is, they were to use the stems to retrieve previously studied items while still responding as quickly as possible. The RTs on the critical test for the explicit group were longer than those of either the speeded response or the baseline group, which did not differ. Longer RTs for the explicit group compared to the baseline group are consistent with the assumption that conscious retrieval is slower than automatic retrieval (Richardson-Klavehn and Gardiner, 1995, Richardson-Klavehn and Gardiner, 1996, Richardson-Klavehn and Gardiner, 1998, Toth, 1996 and Vaterrodt-Plünnecke et al., 2002). Critically, these findings also indicate that the speeded response group did not switch to a conscious retrieval strategy on the critical test. In the present research, we extended this work to include a direct comparison of automatic estimates from the speeded response group with those from standard implicit and PDP groups. Depth of processing was manipulated in Experiments 1 and 2 and full/divided attention was manipulated in Experiment 3. A comparison of performance in the implicit condition with that of the speeded response group would indicate whether the standard implicit stem completion task is routinely contaminated with explicit retrieval. Similar automatic estimates in the speeded response and PDP conditions would provide converging evidence for the critical assumptions of these two measures of automatic retrieval.

The purpose of our research was to compare automatic estimates derived from our speeded response group with those from an implicit memory group and from PDP groups. Given the RT evidence supporting the assumption that subjects in the speeded response group adopted automatic retrieval throughout the test phase, the finding of identical automatic estimates for the speeded response and implicit groups suggests that, under our experimental conditions, the implicit group also employed automatic retrieval. By contrast, the PDP groups consistently revealed underestimates of automatic retrieval when conscious retrieval was high but not when conscious retrieval was comparatively low, suggesting that automatic and conscious retrieval were positively correlated rather than independent. As a consequence of this underestimation, results from the PDP groups suggested that neither depth of processing nor the attention at study manipulation affected automatic retrieval, although a small depth effect was found on automatic estimates in Experiment 2. In contrast, automatic estimates derived from the standard and speeded response tasks showed that both of these manipulations affected automatic retrieval. These findings suggest caution in interpreting the lack of effect of several variables on estimates automatic retrieval derived from PDP.