Journal of Experimental Psychology: Copyright 1991 by the American Psychological Association, Inc. Learning, Memory, and Cognition 0278-7393/91/S3.00 1991, Vol. I7,NO. 6,1105-1123 Implicit Memory for New Associations: An Interactive Process Approach Angela Micco and Michael E. J. Masson University of Victoria Victoria, British Columbia, Canada The contributions of data-driven and conceptually driven processes to implicit memory for new associations were examined using a word stem completion task. Targets encoded in the context of an unrelated word were more likely to be produced on the completion task if tested in the presence of the original, rather than a different, cue word. This context effect was obtained using a semantic elaboration encoding task and a copying task that required no elaboration, but not when orthographically or semantically similar cue words were used at test. With homograph cue words the context effect depended on reinstating the appropriate interpretation of the cue word. Implicit memory for new associations is based on episodic memory for integrated data-driven and conceptually driven encoding operations that have a converging influence on the initial interpretation of a stimulus configuration. The transfer-appropriate processing framework developed memory tests, particularly when these tests have been classi- by Roediger and his colleagues (e.g., Roediger & Blaxton, fied as primarily conceptually driven and data driven, respec- 1987; Roediger, Weldon, & Challis, 1989; Srinivas & Roedi- tively. For example, performance on some explicit but not ger, 1990) has provided an effective characterization of the implicit memory tests is strongly influenced by the degree of distinction between explicit and implicit tests of memory. semantic elaboration during encoding (e.g., Graf, Mandler, & Explicit memory tests are those that require a subject to Haden, 1982; Jacoby & Dallas, 1981). In contrast, perform- consciously recollect a prior encoding episode, whereas im- ance on some implicit memory tests is more sensitive than plicit memory tests carry no such requirement and instead performance on explicit memory tests to changes in an item's express memory for prior occurrence through facilitation in surface structure between study and test (Bassili, Smith, & tasks such as word identification. Within the transfer-appro- MacLeod, 1989; Kirsner & Dunn, 1985; Weldon & Roediger, priate processing framework, memory for prior occurrence 1987). Within this framework, however, there is no necessary results from an overlap between the retrieval processes in- relationship between the explicit and implicit nature of a duced by a memory test and the encoding operations applied memory test and the mode of processing it requires. It is during learning. Memory tests can be classified as primarily possible to develop explicit memory tests that emphasize data- conceptually driven, in the sense that performance relies on driven processes and implicit tests that depend on concep- the recapitulation of the elaborative semantic processes that tually driven processes (e.g., Blaxton, 1989; Srinivas & Roe- were applied during the encoding of the stimulus or primarily diger, 1990). data driven, in which case performance depends on the match The transfer-appropriate processing framework runs into between the sensory perceptual analysis applied to the stim- some difficulty, however, in attempting to account for dem- ulus during encoding and at test. Within this framework, onstrations of memory for new associations on implicit tests explicit and implicit memory test performance can rely on (Graf & Schacter, 1985, 1989; Schacter & Graf, 1989). In different retrieval processes and therefore benefit from differ- these experiments unrelated cue-target word pairs (e.g., ent types of processing at study. MOTHER-WINDOW) were studied and at test the first three For experiments in which items consist of familiar words, letters of the target were presented in the context of either the the transfer-appropriate processing framework has been able word with which it had been paired at study (e.g., MOTHER- to account for dissociations between explicit and implicit WIN ) or with a different word (e.g., REASON- WIN ), Subjects were given one of two memory tests. For the explicit memory test, they were instructed to recall This research is based on a master of arts thesis submitted by the study list word that fit the stem (letter-cued recall), and Angela Micco to the University of Victoria and was supported by for the implicit test, they were to complete the stem with the Operating Grant A7910 from the Natural Sciences and Engineering first word that came to mind (word stem completion). Con- Research Council of Canada to Michael E. J. Masson. text-dependent memory was demonstrated on both tasks, in We are grateful to Peter Graf, Colin MacLeod, Henry Roediger, that the proportion of target items used to complete the stems and Marilyn Smith for very helpful comments on earlier versions of was higher when the stems were presented in the context of this article and to Fergus Craik for discussion regarding some of the ideas expressed here. the original cue word. The cue-target pairs were unrelated Correspondence concerning this article should be addressed to prior to the experiment, so context sensitivity was attributed Michael E. J. Masson, Department of Psychology, University of to the development of a new association between the members Victoria, P.O. Box 3050, Victoria, British Columbia, Canada V8W of a pair. This result might be explained within the transfer- 3P5. Electronic mail may be sent to [email protected]. appropriate processing framework by proposing that the pres- 1105 1106 ANGELA MICCO AND MICHAEL E. J. MASSON ence of a context word at study and test invoked conceptually arise because redintegration of unitized representations (Ho- driven processing. As we shall see, however, this is not an rowitz & Prytulak, 1969) is the basis for implicit memory for entirely satisfactory account of the phenomenon. new associations, and redintegration is driven by modality- Memory for new associations as demonstrated in letter- specific test cues. In effect, modality-specific information cued recall and word stem completion appears to be contin- provides an access key to unitized representations. gent on processing a semantic relationship between members The emphasis on elaboration, unitization, and redintegra- of a pair (Graf & Schacter, 1985; Schacter & Graf, 1986; tion in the Graf and Schacter (1989; Schacter & Graf, 1989) Schacter & McGlynn, 1989). For both recall and stem com- explanation of implicit memory for new associations is con- pletion tests, the context effect was obtained when subjects sistent with views that promote the coordination of data- were required to encode a meaningful association between driven and conceptually driven pattern-analyzing operations the cue-target pairs, but not when the encoding task empha- (Kolers, 1975, 1979; Kolers & Roediger, 1984; Masson & sized the individual meaning of the words (e.g., pleasantness Sala, 1978). This view is somewhat different from that of ratings), nor when a nonsemantic vowel comparison task was Roediger and his colleagues (e.g., Blaxton, 1989; Roediger et used (Graf & Schacter, 1985). al., 1989), in which it is assumed that the degree of data- Recall and stem completion tests were, however, differen- driven and conceptually driven processing varies across mem- tially affected by type of relational processing. For example, ory tasks. Rather than emphasizing the relative amount of generating as opposed to reading a sentence that relates the contribution made by these two types of processing to per- members of a word pair produced a larger associative effect formance on a memory test, we are most interested in the only for the letter-cued recall test (Schacter & Graf, 1986). integration or coordination of the two types of processes. Other ways in which explicit and implicit tests of memory for There is some difficulty, however, in developing a clear picture new associations have been dissociated are: (a) Proactive and of how coordinated processing goes forward. In particular, we retroactive interference manipulations detrimentally affected are concerned with the issues of how unitized representations performance on explicit memory tests but not on implicit are formed and how they are redintegrated during a test. tests (Graf & Schacter, 1987); (b) explicit but not implicit tests The formation of unitized representations that influence yielded improved performance when the encoding task in- implicit memory for new associations appears to depend on volved forming connections among word pairs (Graf & Schac- semantic elaboration. In all of the experiments described in ter, 1989); and (c) some types of amnesics who showed deficits the line of research by Graf and Schacter, implicit memory on the explicit memory test demonstrated normal associative for new associations has been obtained only when subjects memory on the implicit test (Cermak, Blackford, & O'Con- engaged in some form of elaboration of the meaning of each nor, 1988; Graf & Schacter, 1985). word pair.1 We use the term elaboration to refer to the mental A particularly important dissociation between explicit and manipulation in working memory of a symbol's form or its implicit tests of memory for new associations is that implicit meaning in relation to other knowledge (Anderson & Reder, but not explicit tests yield a modality-specific effect.