Implicit and Explicit Memory for New Associations …...that implicit memory differs fundamentally...

Journal of Experimental Psychology:Learning. Memory, and Cognition1985. Vol. II. No. 3. 501-518

Copyright 1985 by the American Psychological Association, Inc.0278-7393/85/$00.75

Implicit and Explicit Memory for New Associationsin Normal and Amnesic Subjects

Peter Graf and Daniel L. SchacterUniversity of Toronto, Toronto, Ontario, Canada

Two experiments examined whether repetition priming effects on a word completiontask are influenced by new associations between unrelated word pairs that wereestablished during a single study trial. On the word completion task, subjectswere presented with the initial three letters of the response words from the studylist pairs and they completed these fragments with the first words that came tomind. The fragments were shown either with the paired words from the study list(same context) or with other words (different context). Both experiments showeda larger priming effect in the same-context condition than in the different-contextcondition, but only with a study task that required elaborative processing of theword pairs. This effect was observed with college students and amnesic patients,suggesting that word completion performance is mediated by implicit memoryfor new associations that is independent of explicit recollection.

Remembering is commonly assumed toinvolve the conscious or explicit retrieval ofinformation about particular experiences.This assumption derives largely from tradi-tional memory tests, such as free recall, cuedrecall, and recognition; a denning character-istic of these tests is the explicit instructionto retrieve information about a particularexperience. However, this assumption is notconsistent with the results of studies that haveassessed performance with tests that makeno reference to particular experiences. Onthese tests memory is expressed by a facili-tation in performance that need not be ac-companied by conscious recollection. For ex-ample, on a word completion test subjectsare presented word fragments (e.g., BREor B E D for BREAD) and are instructedto complete them with the first words thatcome to mind. When required to complete

This research was supported by Grant U0299 to PeterGraf and Grant UO361 to Daniel L. Schacter from theNatural Sciences and Engineering Research Council ofCanada and by a Special Research Program Grant fromthe Connaught Fund, University of Toronto.

We thank Reed Hunt, Larry Jacoby, Morris Moscov-itch, Norm Slamecka, and Endel Tulving for their com-ments on earlier versions of this paper, Karen Raaflaub-Walsh for testing the amnesic patients, and Carol Mac-donald for helping to prepare this manuscript.

Requests for reprints should be sent to Peter Graf,Department of Psychology, University of Toronto, Toronto,Ontario, Canada M5S 1A1.

fragments of recently presented words and ofnew words, subjects succeed more often withpresented words than with new words. Thisfacilitation in completion performance is re-ferred to as a repetition or direct primingeffect (cf. Cofer, 1967; Cramer, 1966).

Direct priming effects have also been dem-onstrated with other tasks, such as wordidentification, which requires subjects toidentify words that are presented very briefly,and lexical decision, which requires subjectsto decide whether or not a presented letterstring constitutes a word. A variety of termshas been used to distinguish between the typeof memory that is tapped by priming testson the one hand, and by recall and recognitiontests on the other hand (e.g., Baddeley, 1982;Bruner, 1969; Cohen & Squire, 1980; Graf,Squire, & Mandler, 1984; Jacoby & Wither-spoon, 1982; Mandler, 1980; Moscovitch,1982; O'Keefe & Nadel, 1978; Schacter, inpress-a; Schacter & Moscovitch, 1984; Tulv-ing, 1983; Warrington & Weiskrantz, 1982;Wickelgren, 1979). For descriptive purposes,we use the terms implicit memory and explicitmemory to distinguish between these formsof memory. Implicit memory is revealed whenperformance on a task is facilitated in theabsence of conscious recollection; explicitmemory is revealed when performance on atask requires conscious recollection of pre-vious experiences. The term priming effect is

501

502 PETER GRAF AND DANIEL L. SCHACTER

used in the present article to refer to thefaeilitative effects of a learning episode onperformance of a word completion or similartask.

Three converging lines of evidence suggestthat implicit memory differs fundamentallyfrom explicit memory. First, performance onimplicit and explicit memory tasks is affecteddifferentially by a variety of experimentalmanipulations (e.g., Graf & Mandler, 1984,1985; Graf, Mandler, & Haden, 1982; Graf,Squire, & Mandler, 1984; Jacoby & Dallas,1981; Roediger & Blaxton, 1983; Scarbor-ough, Gerard, & Cortese, 1979; Tulving,Schacter, & Stark, 1982). Second, perfor-mance on tests of implicit retention can bestatistically independent of recognition per-formance (Eich, 1984; Jacoby, & Wither-spoon, 1982; Schacter, McLachlan, Moscov-itch, & Tulving, 1984; Tulving et al., 1982).Third, patients with organic amnesia, whoare severely impaired on explicit memorytests, are less impaired, and frequently nor-mal, on various implicit tests (e.g., Brooks &Baddeley, 1976; Cohen, 1984; Cohen &Squire, 1980; Graf, Shimamura, & Squire,1984; Graf, Squire, & Mandler, 1984; Jacoby& Witherspoon, 1982; Moscovitch, 1982,1984; Schacter, in press-a; Schacter, Harbluk,& McLachlan, 1984; Warrington & Weis-krantz, 1968, 1970, 1974; Weiskrantz & War-rington, 1979).

A widespread interpretation of the forego-ing findings is that implicit memory is basedon the activation of preexisting memory rep-resentations (e.g., Graf et al., 1982; Graf,Squire, & Mandler, 1984; Mandler, 1980;Mortensen, 1980; Morton, 1969, 1979;Rozin,1976; Warrington & Weiskrantz, 1982; Wick-elgren, 1979). The study materials used inmost previous experiments were familiaritems, such as individual words, that arerepresented in long-term or semantic memoryprior to their appearance in a study list. Ithas been argued that these preexisting rep-resentations are activated as a result of pre-senting the study items, and that activationoccurs automatically and thus independentlyof processes that mediate explicit remem-bering.

An activation interpretation receives sup-port from several studies of amnesic patients.Amnesia occurs as a consequence of various

neurological dysfunctions. Amnesic patientsshow a pattern of cognitive functioning thatis typically characterized by poor memoryfor recent events along with relatively normalintellectual functions and premorbid knowl-edge and skills (e.g., Moscovitch, 1982;Schacter, 1983; Squire, 1982; Squire, Cohen,& Nadel, 1984; Warrington & Weiskrantz,1982). One of the hallmarks of anterogradeamnesia is the patients' inability to acquireand remember new associations; failure toremember a list of unrelated word pairs onan immediate recall test is diagnostic ofamnesia (e.g., Meyer & Yates, 1955; Rozin,1976; Scoville & Milner, 1957; Starr & Phil-lips, 1970). Nevertheless, amnesic patientsshow relatively normal priming or implicitretention, when the study materials are fa-miliar items that have a preexisting memoryrepresentation, such as words (e.g., Graf,Squire, & Mandler, 1984; Rozin, 1976; War-rington & Weiskrantz, 1970), highly relatedpaired associates (e.g., Shimamura & Squire,1984; Winocur & Weiskrantz, 1976), or lin-guistic idioms (e.g., Schacter, in press-a). Thefinding of normal priming effects, in con-junction with the observation that amnesicpatients seem unable to acquire new associ-ations, is consistent with the view that implicitmemory is mediated by activation of preex-isting representations. The reasoning here isstraightforward. A new association has nopreexisting memory representation that canbe activated and thus mediate implicitmemory.

It is important to observe, however, thatthe negative evidence for the retention of newassociations in amnesia has been obtainedon explicit memory tasks, such as pairedassociate learning; these findings thus do notpreclude the possibility that there is implicitmemory for new associations. Several recentstudies are relevant to this issue, but thefindings are inconsistent. On the one hand,there is some evidence for the implicit reten-tion of newly acquired associations in amnesicpatients who exhibit profound deficits ontests of explicit memory (e.g., Moscovitch,1984; Schacter, Harbluk, & McLachlan, 1984;Weiskrantz & Warrington, 1979). To illus-trate, Moscovitch required patients to mem-orize word pairs and later to read thesewords, either in the same or in different

RETENTION OF NEW ASSOCIATIONS 503

pairings, from a degraded display. He foundthat amnesic patients were faster at readingthe words in the same pairs than in differentpairs, suggesting that they had acquired newassociations between the paired words in thestudy list. On the other hand, several studieswith college students provide conflicting evi-dence concerning the implicit retention ofnew associations (Carroll & Kirsner, 1982;Franks, Plybon, & Auble, 1982; McKoon &Ratcliff, 1979). For example, McKoon andRatcliff had subjects study unrelated andrelated word pairs and then gave a lexicaldecision task in which some of the targetwords were preceded by the paired wordsfrom the study list. They found that latencyin making a lexical decision was reducedwhen a target was preceded by the pairedword from the study list, for both related andunrelated pairs. This finding suggests thatnew associations can influence performanceon a task that is sensitive to implicit memory.However, Carroll and Kirsner (1982) used asimilar design and they found no evidencefor associative effects on lexical decision per-formance. Thus, although it appears thatimplicit retention of new associations canoccur, the phenomenon remains to be ex-plored and established more conclusively.

The present study was designed to examinefurther the implicit retention of newly ac-quired associations. The general strategy wasto present unrelated and related word pairsand then test priming and explicit remem-bering of the response word from each pair.Explicit remembering was examined with atraditional cued-recall test in which the stim-ulus member of each list pair was presentedand subjects were instructed to rememberthe response word. Priming was assessed witha word completion test in which the firstthree letters of the response words were pre-sented and subjects were required to completethem with the first words that came in mind.To assess whether newly acquired associationsabout the specific pairing of words in thestudy list affect implicit retention, we com-pared completion of word fragments thatappeared on the test either together with thepaired words from the study list (same con-text) or together with other words (differentcontext).' On the view that the presentationof a word activates its preexisting represen-

tation, we expected an increase in completionperformance even in the different-contextcondition. The critical question is, however,whether completion performance in the same-context condition exceeds performance in thedifferent-context condition. To the extent thatinformation about the pairing of words inthe study list contributes to priming, testfragments should be completed with wordsfrom the study list more often in the same-context condition than in a different-contextcondition. By an activation interpretation,this outcome is expected for related wordpairs, because there is a previously establishedassociation between the two words that canbe activated as a result of presenting the pairin the study list. However, this outcome isnot expected for unrelated word pairs, becausethere is no preexisting representation thatcan be activated.

Experiment 1

This experiment examined whether eithernewly acquired associations or preexistingassociations affect completion test perfor-mance. We presented subjects with unrelatedword pairs and with related word pairs, understudy conditions that required either elabo-rating each list pair or comparing the numberof vowels in each pair of words. In the relatedpairs, the target words are linked by an oldand familiar association (e.g., BUTTONED-SHIRT). In the unrelated pairs, the targetwords have no preexisting relation, and hencemust be linked by a new association that isestablished for the first time during the studytrial (e.g., WINDOW-SHIRT). Subjects studiedword pairs either under elaborative or non-elaborative task conditions in order to ex-amine the effects of different study tasks onthe implicit retention of new and old associ-

1 It is important to distinguish between the localcontext provided by a paired word on the completiontest and the global context that is provided by the entirestudy list, the experimenter, the experimental room, andso forth (see Schacter, in press-a, for discussion). Althoughit is well known that amnesic patients have a severedeficit in memory for global context (e.g., Kinsbourne &Wood, 1975; Rozin, 1976; Schacter, Harbluk, & Mc-Lachlan, 1984; Winocur & Kinsbourne, 1978), when weuse the terms context or context effect in the presentarticle, we refer exclusively to the local context providedby a paired word on the completion test.


ations. Previous studies that tested memoryfor single words showed that similar taskmanipulations had no differential effect onthe magnitude of priming (e.g., Graf et al.,1982; Graf & Mandler, 1984; Jacoby & Dallas,1981).

MethodDesign. The main experimental conditions were

formed by the combination of two between-subjectsfactors: Type of Study List pairs (related vs. unrelated)and Study Task (elaborate vs. vowel comparison). Thedesign also included completion test context (same vs.different) as a within-subjects factor. In addition, weinvestigated explicit remembering with a standard cued-recall test.

Subjects. Sixty-four students participated in returnfor credits in an introductory psychology course. Theywere randomly arranged into four groups of 16 each.These groups were assigned to four conditions that weredenned by the orthogonal combination of two types ofstudy tasks (elaborate and vowel comparison) and Iwotypes of word pairs (related and unrelated): elaborate/related, elaborate/unrelated, vowel comparison/related,and vowel comparison/unrelated.

Materials. Forty-two moderately related word pairs(e.g., MOLD-BREAD, RIPE-APPLE) were selected from theShapiro and Palermo (1968) norms according to thefollowing criteria. First, pairs were chosen if the responsewas given as a primary associate to the stimulus with aprobability between .05 and .20; the selected pairs hadan average associative probability of .15. Second, theinitial three letters—the stem—of each response word(e.g., BRE, APP) had to be unique in the set of all wordsthat were included in the pairs, and for each stem apocket English dictionary had to list at least 10 commonwords with the same stem (e.g., BREAD, BREAK, BREAK-FAST, BREAST). Third, the response words had to bebetween 5 and 10 letters in length and of mediumfrequency; the selected words averaged 6 letters and theyhad a mean frequency of 48.4 occurrences per million(Kucera & Francis, 1967). The stimulus words werebetween 3 and 11 letters long (average 6.2) and they hada mean frequency of 45.1 occurrences per million.

Of these 42 related word pairs, a random set of 32was selected for use as critical pairs in the study list (Set1), and the remaining 10 pairs were used for practice (5pairs) and as fillers (5 pairs) at the beginning and end ofthe study list. A parallel set of unrelated word pairs wasalso required for the study; these were obtained by re-pairing the words from the related pairs. The constructionof the latter pairs was also constrained by the requirementto keep the same words as responses for the related pairsand for the unrelated pairs. Thirty-two unrelated pairs(Set 2) were formed by re-pairing words from Set I andwere used as critical items in the study list. The remainingunrelated pairs were used for practice and as fillers. Eachof the related and unrelated pairs was printed on anindex card, in lowercase letters.

Two additional sets of single words were selected foruse as distractors on the completion test. One of thesesets (Set 3) included eight words that were similar to thestimulus words used for the study list pairs in Sets 1 and

Table 1Examples of Study List Pairs and CompletionTest Items

Study list Completion test item

Related pairsMOLD-BREADRIPE-APPLEDEL1CATE-FRAGILECOVERING-BLANKET

Unrelated pairsKINDLY-STICKJAIL-STRANGEBALANCE-CHAIRDRYER-BLOCK

Test item withrelated context word

MOLD-BRERIPE-APP

Test item withunrelated context word

K1NDLY-STIJAIL-STR

Test item withstimulus distractorcontext word

DETAIL-FRABELIEF-CHA

Test item withno contextword—target alone

BLA

BLOTest item with

no contextword—distractor alone

ABSFLA

2; they had an average of 5.6 letters and a frequencycount of 43.6 occurrences per million. The other distractorset (Set 4) included 32 words that were similar to theresponse words from the study list pairs. The initial threeletters of each word from Set 4 was unique in the poolof all words required for the experiment, and a pocketdictionary listed at least 10 common words with thesame beginning. These response distractors had an averageof 5.8 letters and a frequency count of 47 occurrencesper million (Kucera & Francis, 1967).

Tests. The cued-recall test form consisted of a singlepage; it showed a randomly arranged list of the stimuluswords from the critical 32 study pairs- The word com-pletion test form also consisted of a single page. This testincluded 64 items, with each item defined either as asingle word fragment (40 items) or as a word fragmentprinted to the right of a context word (24 items), asillustrated in Table 1. The test fragments always consistedof the initial three letters of a word.

Table 1 illustrates the construction of completion testitems. Consider first the items that consisted of fragmentswith context words. Each of these items included afragment corresponding to a response word from thestudy list pairs. For 8 items, the context was a relatedword from Set I (e.g., MOLD-BRE ). These items wereused to assess completion performance in the same-context condition following study of related word pairs,and they were used to assess performance in a different-context condition following study of unrelated wordpairs. For an additional eight items, the context was anunrelated word from Set 2 (e.g., KINDLY-STI ).These items were used to assess completion performancein the same-context condition following study of unrelatedword pairs, and they were used to assess performance in


a different-context condition following study of relatedword pairs. For the remaining eight items, the contextwas a stimulus distractor word from Set 3 that neverappeared in the study list (e.g., DETAIL-CHA ); theseitems were used to assess completion performance in adifferent context. Of the single fragment test items, eightcorresponded to response words that were presented inthe study list pairs; these items were used to obtain afurther index of completion performance in a differentcontext. Finally, 32 single fragment test items corre-sponded to the words that had been selected as responsedistractors (Set 4) and they never appeared in the studylist. The distractor fragments were included in order todisguise the memory testing aspects of the completiontest by merging the target fragments among a longer listof lest cues. This disguise is critical because once thememory testing aspects of the completion test becomeapparent to subjects, the test can be transformed into acued-recall test. Previous studies have shown that thistransformation can be achieved by a simple change ininstructions (Graf & Mandler, 1984; Graf, Squire, &Mandler, 1984).

To ensure that each study list response word could bepresented in each condition of the completion test (i.e.,with a related, unrelated, or distractor context word, oralone), four alternative test forms were required. The useof these forms was counterbalanced within and acrossexperimental conditions.

Procedure. The same general procedure was used foreach subject group. It consisted of three parts: instructionand practice, study, and testing. Each subject was testedindividually. During instruction and practice, subjectswere instructed in a task that required either elaborativeprocessing or vowel comparison, and they practiced thesetasks either with related or unrelated pairs. For theelaborative task, subjects used a 5-point scale that hadthe labels "Easy to Relate" and "Difficult to Relate" atits ends; they had to "construct a sentence" that related"the two words in a meaningful manner," say thatsentence to the experimenter, and then record how easyor difficult it had been to produce that sentence. Subjectscreated a wide range of sentences, some short and simple,"He ate the RIPE APPLE," and some long and complex,"John found a quarter and went to the store to buy aRIPE APPLE with it." For the vowel comparison task,subjects were instructed to decide (Yes/No) whether thetwo words in each pair had the same number of vowelsand to record their decision. Practice continued untileach subject followed these instructions. The rate of pairpresentation was controlled by subjects' speed on eachprocessing task (approximately 8 vs. 5 seconds per pairfor the elaborate vs. vowel comparison task). Previouswork has shown that variations in the rate of presentationacross this range has no systematic effect on wordcompletion performance (Graf & Mandler, 1984).

Immediately after instruction and practice, each subjectwas presented with a study list consisting of either relatedwords or unrelated words. This list included 32 pairseither from Set I (related) or Set 2 (unrelated), and fivefiller pairs, three of which were at the list beginning andtwo at its end. The fillers were added to minimizeprimacy and recency effects on the subsequent tests. The32 critical pairs were arranged randomly within the studylist. The study list was presented once, self-paced by eachsubject's speed on the processing tasks described earlier.

During the third part of the experiment, each subjectreceived a sequence of three tests: name completion,word fragment completion, and cued recall. The namecompletion test was given immediately after study. Itsfunctions were to engage subjects in an unrelated activitybefore administering the word completion test, and moreimportantly, to induce an appropriate set for word com-pletion testing. The name completion test presentedsubjects with the fragmented names of 11 famous people(e.g., G e G d, L a

i e i), and 3 min were allowed for com-pleting as many names as possible.

The word fragment completion and cued-recall testswere described in the materials section. The completiontest instructions informed subjects that before they wouldreceive a memory test, they had to "complete each wordbeginning on the [completion test] form with the firstword that [came] to mind." Subjects were instructed thatthey could write any word except proper names, andwhen a proper name was given an alternative completionwas requested. Because some of the completion cues werepresented in the context of another word, several ofwhich were from the study list, the instructions empha-sized that the task was to complete each cue with thefirst word that came to mind, that the context wordwould sometimes help to think of a completion, but thatit was unimportant whether or not the completion wasrelated to the context word. Subjects who requested moreinformation about the purpose of this test, particularlythe role of the context words, were told that this fillertask had been borrowed from a different experiment. Weencouraged subjects to finish the completion test asquickly as possible (it required about 3 min). Then thecued-recall test was presented. The cued-recall instructionsemphasized explicit remembering. Subjects were remindedof the word pairs that they had studied, they wereinformed that the test cues were the stimulus words fromthe pairs that they had seen, and they were instructed torecall the response word that had been paired with eachstimulus in the study list. This test required about 5minutes and it was terminated when a subject had notwritten anything for about 1 minute.

In summary, each subject studied a 37-item list con-sisting of either related or unrelated word pairs, underconditions that required either elaborative processing orvowel comparison. The pairs were presented at a pacedetermined by each subject's speed on the study task.Immediately after study a distractor task was given (thename completion test) for 3 minutes, followed by a wordfragment completion test and then a cued-recall test.

Results

The dependent measures were the propor-tion of study list response words producedon the fragment completion test and theproportion of words remembered on the cued-recall test. In order to assess the completiontest effects attributable to different experi-mental conditions, we also required an esti-mate of baseline performance on this test.This estimate was obtained from a group of


I I ELABORATE

VOWEL COMPARISON

SAME DIFFERENTCONTEXT CONTEXT

UNRELATED PAIRS

SAME DIFFERENTCONTEXT CONTEXT

RELATED PAIRSFigure 1. Mean completion performance for related and unrelated word pairs that were presented in a

task that required generating a sentence for each pair (elaborate condition) or in a task that requiredcomparing the vowels in each pair {vowel comparison condition). [The completion test showed the initialthree letters of the response word from each pair, either with the paired stimulus word from the study list(same context).or with another word or alone (different context). A separate control group was used toobtain an estimate of baseline completion performance on the target response words, shown by the dashedlines. Vertical bars show the standard errors of the means.]

116 subjects who received the same comple-tion test without a presentation of the studylist. The performance of this control groupshowed that for fragments that were tested inthe context of a related word, baseline per-formance was 38.5%, whereas it was lowerfor fragments that were tested in the contextof an unrelated list word (12.2%), an unrelatednew word (11.9%), or alone (12.2%). Thelatter means did not differ significantly, F <1, and thus we used their overall mean (12.1 %)in subsequent analysis {p < .05 for this andall other statistical tests). These baseline levels,shown in Figure 1, provide the referencepoints for assessing the completion test effectsproduced by learning unrelated and relatedword pairs under elaborative and vowel com-parison study tasks and across different testingconditions.

Effects of context on completion of studiedwords. The primary goal of the experimentwas to compare completion performance onfragments tested in the same context as inthe study list (i.e., paired with the same wordas in the study list) and on fragments testedin different contexts (i.e., paired with another

word from the study list, paired with a newword, or tested alone). The relevant data arepresented in Figure 1. This figure shows thatfollowing elaborative processing of either typeof word pairs, completion performance onsame-context items exceeded performanceon different-context items. Following vowelcomparison, however, the level of completionperformance was similar across same- anddifferent-context test items for both types ofstudy list pairs.

This summary of the findings is supportedby a 2 X 2 X 2 analysis of variance (ANOVA)that treated materials (related and unrelatedpairs) and study task (elaborate and vowelcomparison) as between-subjects factors andtest context (same and different) as a within-subjects factors. For this analysis, the originaltest scores were adjusted by subtracting theappropriate baseline scores, thereby eliminat-ing effects attributable to the different baselinescores for unrelated and related pairs. Theanalysis showed significant main effects forstudy task, F(l, 60) = 17.5, MSt = 17.8, andfor test context, F(l, 60) = 16.6, MSC = 15.5.More importantly, however, there was a sig-


Table 2Cued Recall Performance (Means and Standard Errors) as a Function of Study Taskand Study List in Experiment I

Study task

Elaborate

Vowel comparison

Study list

Related pairsUnrelated pairsRelated pairsUnrelated pairs

Recall of targetwords {%)

M

66.835.48.8

.2

SE

5.54.52.9

.2

Recall test intrusions(no. of words)

1.563.061.311.63

nificant interaction between study task andtest context, F(\, 60) =12.9, MSC =^ \5.5,with no other effects approaching significance.This interaction indicates that for both un-related and related study list pairs, completionperformance was higher on same-contextitems than on different-context items afterthe elaborative study task, but not after thevowel comparison task.

An additional analysis showed that on thedifferent-context test items, performance wassystematically affected only by whether ornot the context word was related to the targetword. On the one hand, for items with un-related context words, overall completionperformance was 23.3 and 20.0% for theelaborative and vowel comparison task, re-spectively. Both of these means were signifi-cantly higher than the appropriate baselinelevel of 12,1%, *<31) = 6.5 and 5.1, respec-tively. On the other hand, for items withrelated context words, overall completionperformance was 53.9 and 50.0% for theelaborative and vowel comparison task, re-spectively. Both of these means also exceededthe appropriate baseline level of 38.5%,;(15) = 2.7 and 2.1, respectively.

Cued recall. Table 2 shows the overalllevels of cued-recall performance in eachexperimental condition. An analysis of thesedata showed significant main effects for ma-terials (related, unrelated), F(\f 60) = 27.6,MSe = 23.8, and for study task (elaborate,vowel comparison), F(l , 60) - 149.5, MSe -23.8, as well as a significant interaction be-tween these factors, F(l, 60) = 9.0, MSe -23.8. Overall recall was lower under vowelcomparison than elaborative task conditionsand with unrelated than related word pairs.

The interaction effect is probably an artifactthat stems from the near-zero level of recallunder vowel comparison task conditions. Thetable also shows the number of incorrectwords or intrusions that appeared on therecall test; these intrusions were infrequentin all experimental conditions.

Discussion

Experiment 1 yielded three main findings.First, following an elaborative study task,there was a higher level of completion per-formance when the study context was rein-stated at testing than when study and testcontexts were different, for both unrelatedand related study list word pairs. In contrast,following the vowel comparison task, therewas no same-different effect on either typeof word pairs. Second, across all types ofdifferent-context test items, there were similarand significant increases above baseline com-pletion performance under both elaborativeand vowel comparison task conditions. Thisfinding replicates and extends the results fromprevious studies that reported that primingof single words is not affected by similarstudy task manipulations (Graf & Mandler;1984; Graf et al., 1982; Graf, Squire, &Mandler, 1984; Jacoby & Dallas, 1981).Third, there was a higher level of recall forrelated pairs than for unrelated pairs as wellas a higher level of recall under elaborativethan vowel comparison task conditions.Overall, the pattern of findings suggests thatimplicit memory for new associations, asindexed by completion test performance, isaffected by manipulations of test context. Thesimilarity in the results for unrelated and


related word pairs indicates that the same-different completion test effect does not de-pend on preexisting associations betweenwords, because the normatively unrelatedword pairs were associated for the first timeduring study list presentation.

The finding of a higher level of completionperformance on same-context test items ver-sus different-context test items under elabo-rative study conditions, but not under vowelcomparison conditions, contrasts with theresults from previous studies that found com-parable levels of priming across similar taskmanipulations. This invariance of completionperformance across task manipulations (e.g.,Graf & Mandler, 1984; Graf et al., 1982;Jacoby & Dallas, 1981) was used to arguethat implicit memory is mediated by auto-matic processes. In contrast, explicit recalland recognition were assumed to be mediatedby strategic or controlled processes becauseof their dependence on study task manipu-lations. On this view, the observation that thesame-different effect depends on elaborativeprocessing suggests that the completion testused here measures explicit rather than im-plicit memory. That is, because performanceis affected by study task manipulations thatare assumed to operate independently of theautomatic processes that mediate implicitmemory, the completion test used here wasinfluenced by explicit remembering. Thus,the same-different context effect may notprovide evidence of implicit memory fornewly acquired associations.

One way to test this hypothesis is by ex-amining the relation between performanceon the word completion test and performanceon the cued-recall test. Previous studies haveshown that the likelihood of producing aword on a completion task does not predictits recognition; performance on a completiontask can be statistically independent of per-formance on a test of explicit remembering(see Tulving et al., 1982). This statisticalindependence was used to argue that perfor-mance across these tests was mediated bydifferent underlying memory processes.

In order to determine whether the same-different effects observed with the elaborativestudy tasks can be attributed to processesthat are also required for cued recall, weexamined the relation between completion

and recall performance for words that hadbeen tested in the same-context condition ofthe completion task. A finding of statisticalindependence would be difficult to reconcilewith the view that the same processes arerequired for completion and cued recall, be-cause common processes should have similareffects across tests and thus introduce perfor-mance dependence. However, the analysisrevealed signs of dependence: overall cued-recall performance was generally higher forwords that had been produced on the com-pletion test than for words that had not beengiven as completions (82.7 vs. 58.7% forrelated pairs and 51.8 vs. 25.2% for unrelatedpairs). A chi-square test confirmed this ob-servation by showing a significant amount ofdependence between completion and cued-recall test performance for related study listpairs, x 2 0) = 32.1, and for unrelated studylist pairs, x 2 (0 = 24.6.

One possible interpretation of this depen-dence is that both completion and cued recallare mediated by explicit remembering. How-ever, this dependence does not permit anunambiguous interpretation because it canalso occur for a variety of other reasons. Forexample, dependence can result becausewords that are produced on the completiontest receive an additional exposure betweenstudy and cued-recall testing and are thusbetter recalled than words that were notproduced on the completion test (a similarobservation was made by Tulving et al., 1982).The present results are consistent with thisview. We found that after the elaborativestudy task, recall of words that had beengiven as completions in the same-contextcondition was higher (80.7%) than recall ofwords that had been given as completions inthe different-context condition (56.4%). Incontrast, there was no comparable differencein recall for words that had not been givenon the completion test: recall of words fromthe same-context condition was 32.1% andrecall of words from the different-contextcondition was somewhat higher, 44.5%. Thesefindings indicate that completion testing hada large influence on recall performance, andthis influence may be sufficient to accountfor the observed dependence between com-pletion and cued-recall performance.

In summary, the same-different effect


found in Experiment 1 demonstrates thatcompletion test performance is affected bynewly acquired associations. The interpreta-tion of this finding must remain ambiguous,however, because completion test performancemay not provide a pure index of implicitmemory. Thus, the same-different effectsfound here may be attributed either to theimplicit or explicit memory for new associa-tions.

Experiment 2

The main goal of Experiment 2 was toexamine further whether an implicit or ex-plicit form of memory is responsible for thesame-different effect observed on completiontest performance in Experiment 1. For thispurpose, we compared cued recall and com-pletion test performances of amnesic patientsand of two groups of control subjects. In viewof the amnesic patients1 deficit on tests thatrequire explicit remembering and their intactperformance on tests that depend on implicitmemory, their performance on the wordcompletion test provides a critical test ofwhether the same-different effect is attribut-able to explicit remembering. If amnesic pa-tients and control subjects show a comparableperformance superiority on the completiontest in the same-context condition over thedifferent-context condition, there would bestrong support for the view that the same-different effect is mediated by implicit mem-ory for newly acquired associations. However,if amnesics showed a similar level of comple-tion performance on same- and different-context items, it would suggest that the same-different effect found in Experiment 1 wasmediated by explicit remembering.

Experiment 2 followed the strategy used inExperiment 1, with three important changes.First, all materials were presented with anelaborative study task because the same-different effect was observed only under elab-orative study conditions in Experiment 1.Second, the type of study material (relatedand normatively unrelated word pairs) wasincluded as a within-subjects factor. Third, asimplified word completion test was used,with fewer items for assessing performancein the different-context condition, becauseExperiment 1 showed that alternative types

of different-context items provide similar in-formation about completion performance.Each subject saw both related and unrelatedword pairs and then received a word comple-tion test followed by a cued-recall test. Onthe assumption that anterograde amnesia im-pairs only explicit remembering, and that thesame-different effect is attributable to implicitmemory, we expected the same pattern ofcompletion performance in each subjectgroup, together with different levels of cued-recall performance.

MethodSubjects. Three groups of subjects participated in the

experiment. The first group consisted of 12 amnesicpatients (8 men, 4 women) with diverse neurologicaldisorders that are commonly associated with memoryimpairments. These patients became amnesic as a resultof closed head injuries (6 patients), ruptured anteriorcommunicating artery aneurysms (3 patients), encephalitis(1 patient), and tumor of the third ventricle (I patient).No firm neurological diagnosis was available for onepatient. Eleven patients lived at home with their familiesand one lived in a chronic care facility in the Torontoarea. Their average age was 42.3 years (range 20-67),and they had an average of 14.0 years of formal education(range 10-22). Their average full scale IQ on the WechlserAdult Intelligence Scale-Revised (WAIS-R) was 93.3(range 82-117) and their average Wechsler Memory Scale(WMS) score was 80.8 (range 61-110). In normal subjectsthe WMS score is equivalent to IQ. None of the patientswere aphasic, anomic, apraxic, or agnosic, and none haddifficulty in following task instructions. Every patienthad difficulties on standard tests of explicit remembering,such as recall and recognition of recently presentedwords. None of them could recall part of a short paragraphafter a 30 min delay, and each showed a severe deficit inpaired-associate learning, recalling an average of .08,. 17,and .42 responses from six unrelated word pairs afterone, two, and three study trials, respectively.

The performance of the amnesic patients was comparedwith two control groups. The first control group included5 men and 7 women that were matched to the amnesicpatients in terms of age, IQ, and formal education. Thesesubjects lived with their families and volunteered for thestudy. Seven of them were chosen because they hadexperienced neurological traumas similar to those of theamnesic patients (mild head injuries for 5 subjects,ruptured anterior communicating artery aneurysms for2 subjects), but they had no measurable memory deficits.The remaining 5 subjects had not experienced anyneurological dysfunction, and were chosen because theirages and IQs were similar to those of the amnesicpatients. The group had an average age of 47.2 years(range 22-81) and 12.3 years of formal education (range8-17). Their average full scale WAIS-R IQ was 99.3(range 88-120). This matched group did not differ sig-nificantly from the amnesics in terms of age, tQ2) < 1,years of education, i{22) = 1.3, or full scale WAIS-R IQ,r(22) = 1.4. The experiment also included a control


group of 24 college students who participated in returnfor credits in an introductory psychology course. Thelatter group was included to provide a replication ofExperiment 1.

Design. The design included the three groups as abetween-subjects factor, and materials (related vs. unre-lated word pairs) and test type (completion vs. cuedrecall) as within-subjects factors.

Materials. Most of the materials from Experiment 1were used, including 30 word pairs each from the relatedand unrelated sets (Sets 1 and 2\ 30 of the responsedistractor words from Set 4, and a random selection of15 related and unrelated practice-filler pairs. In addition,40 words were required to serve as stimulus distractorsfor the completion test used in Experiment 2. This setof stimulus distractors consisted of words that weresimilar to the stimulus words used in the study lists, asdescribed in Experiment 1.

Tests. The cued-recall test form consisted of 2 pages,each of which showed the 15 stimulus words from onehalf of the study list pairs in a random arrangement. Thesame test form was used to assess recall for both relatedand unrelated word pairs, because the words from therelated pairs had simply been re-paired in order toconstruct the unrelated pairs.

The completion test was a simplified version of thatused in Experiment 1 because the results from thatexperiment showed that alternative types of differentcontext items provide similar measures of performance.A different test form was used to assess completion forrelated and unrelated pairs. Each test was constructed toobtain three measures of completion performance ontarget words from the study list and one measure ofbaseline completion performance on new words. Thethree measures of performance on target words wereobtained by presenting test fragments in three types ofcontexts. Each completion test consisted of 2 pages, andeach page showed a random arrangement of 30 test items(i.e., a fragment with a context word). Five items consistedof a fragment with a paired word from the study list(same-context items), 5 items consisted of a fragmentwith a presented word that had not been paired with thetarget in the study list (broken item), 5 items consistedof a fragment with a stimulus distractor word from Set3 (new-old item), and 15 items consisted of a responsedistractor fragment from Set 4 with a stimulus distractorword from Set 3 (new-new item). For each page of thecompletion test, three different forms were required inorder to present each study list target word in each testcontext.

Procedure. The general procedure was similar to thatof Experiment 1. Each subject was first instructed in theelaborative study task used in Experiment 1. This taskrequired subjects to generate a sentence for each studylist word pair and to evaluate on a 5-point scale howdifficult it had been to generate that sentence. Subjectspracticed this task on five word pairs. The rate of pairpresentation was self-paced by each subject's speed onthe study task.

Immediately after instruction and practice, each subjectreceived two study-test trials, with a short pause (about3 min) separating the two trials. Each trial included astudy list presentation and a sequence of three tests: namecompletion, word completion, and cued recall. The study

list consisted of either 15 related word pairs or 15unrelated word pairs, and 5 filler pairs—three at the listbeginning and two at its end. For one half of the subjectsin each group, the first list included the related pairs andthe second list included the unrelated pairs; the reverseorder was used for the remaining subjects. The first andsecond list had none of the same words, for any subject.The name completion test, which was described inExperiment 1, was given immediately after the study list;its functions were to engage subjects in an unrelatedactivity for 3 min before administering the word comple-tion test, and more importantly, to induce an appropriatetest set for word completion. The word completion andcued-recall tests were described in the materials section;they were administered as in Experiment 1.

Results

The dependent measures were the propor-tions of test fragments that were completedwith study list response words in the same-context condition and in the different-contextcondition by each subject group. The com-pletion test also yielded a measure of baselineperformance because it included some testitems corresponding to response distractorwords that were never presented in the studylist. All subject groups showed similar levelsof completion performance on these baselineitems (amnesic patients = 14%, matchedcontrols = 11.7%, and student controls =13.6%). An additional estimate of baselinecompletion performance was obtained froma group of 48 college students who receivedthe completion test without having seen thestudy lists. They completed 11.9% of thedistractor items. More importantly, as inExperiment 1, the results from the controlgroup showed that for target word fragmentsthat were tested in the context of a relatedword, baseline performance averaged 25.8%,whereas it was lower for fragments that werepresented in the context of an unrelated listword (12.9%) or an unrelated new word(13.5%). Because the latter two means didnot differ significantly (t = A), their overallmean (13.2%) was used as an estimate ofbaseline performance on test items with un-related context words. Figure 2 shows thebaseline level for related items and for unre-lated items; these levels provide referencepoints for assessing the completion test effectsproduced by learning unrelated or relatedword pairs.

Effects of context on completion of studiedwords. The primary goal of the experiment


1 \I I AMNESIC PATIENTS

1832 MATCHED CONTROLS

STUDENT CONTROLS

SAME CONTEXT

RELATEDPAIRS

SAME CONTEXT

UNRELATEDPAIRS

DIFFERENT CONTEXT

BOTHPAIRS

Figure 2. Mean word completion performance for related and unrelated pairs, studied under elaboratingtask conditions by three different subject groups: amnesic patients, matched controls, and student controls.[The completion test presented the initial three letters of the response word from each study list pair,either with the paired stimulus word from the study list (same context) or with an another word (differentcontext). A separate control group was used to obtain an estimate of baseline completion performance onthe target response words, shown by the dashed lines. Vertical bars show the standard errors of the means.]

was to compare completion performance ofamnesic patients and control subjects onfragments tested in the same context as inthe study list (i.e., paired with the same wordas in the study list) and on fragments testedin different contexts. The relevant data arepresented in Figure 2. The figure shows thatfor both types of word pairs and for allsubject groups, completion performance washigher in the same-context condition than inthe different-context condition. For the same-context condition, performance on items withunrelated context words averaged 31.7, 31.7,and 34.2% for amnesic patients, matchedcontrols, and student controls, respectively(F = . 1), and performance on items withrelated context words averaged 51.7, 60.0,and 55.0% for amnesic patients, matchedcontrols, and student controls, respectively(F = .3), This summary of the findings wassupported by an ANOVA that revealed a sig-nificant main effect for type of completiontest context (same related, same unrelated,different), F(2, 90) = 13.8, MSe = 1.0, withno other effects approaching significance. Forthis analysis the original test scores wereadjusted by subtracting the appropriate base-

line scores, thereby eliminating effects attrib-utable to the different baseline scores forunrelated and related pairs. Across subjectgroups, the overall increase in completionperformance was higher in the same-contextcondition (24.5%) than in the different-contextcondition (7.0%).

An additional analysis showed that in thedifferent-context condition, performance wassimilar across the three subject groups, aver-aging 17.9% (17.5 and 18.3% for the relatedand unrelated list, respectively) for the am-nesic patients, 20.4% (20.0 and 20.8% for therelated and unrelated list, respectively) forthe matched control group, and 21.9% (21.7and 22.1% for the related and unrelated list,respectively) for the student controls. Thesemeans did not differ significantly from eachother, F = .7. However, they reveal a signifi-cant increase over baseline completion per-formance; the smallest increase was shownby the amnesic patients (10 out of 12 showedan increase) but even this level of performancewas above baseline, t(\\) = 2.27.

Cued recall. Table 3 shows the level ofcued-recall performance in each experimentalcondition, as well as the number of incorrect


Table 3Cued-Recall Performance (Means and StandardErrors) as a Function of Study List andSubject Group in Experiment 2

Study list

Related pairsTarget words

MSE

Intrusions(no. ofwords)

Unrelated pairsTarget words

MSE

Intrusions(no. ofwords)

Amnesicpatients

33.36.6

5.0

2.21.3

7.4

Subject group

Matchedcontrols

72.85.3

1.6

35.0.7.7

3.0

Studentcontrols

92.22.1

.3

63.94.7

1.0

words or intrusions that appeared on therecall test. An analysis of the recall datashowed a significant main effect for group,F(2, 45) = 53.7, MSe = 11.1, attributable toa high level of recall in the student controlgroup (78.1%), a medium level of recall inthe matched group (53.9%), and a low levelof recall in the amnesic group (17.8%). Theanalysis also showed a main effect for studylist, F(\, 45) = 104.0, MSe = 4.9, with higherrecall of words from the related (66.1%) thanthe unrelated (33.7%) pairs. No other effectsreached significance.

Discussion

Experiment 2 yielded three new facts. First,for amnesic patients, matched controls, andstudent controls, there was a higher level ofcompletion performance when the study con-text was reinstated at testing (same context)than when study and test contexts were dif-ferent, for both unrelated and related wordpairs. Second, the level of completion perfor-mance on same-context test items did notdiffer among the three subject groups. Third,there was a higher level of recall for bothcontrol groups than for the amnesic group.The control groups' data replicate the findings

from Experiment 1. The amnesic group'sdata provide critical evidence in support ofthe view that the same-different effect on theword completion test is mediated by implicitmemory for newly acquired associations. Inspite of their severely impaired recall, theamnesic patients showed entirely normal lev-els of completion test performance afterstudying unrelated and related word pairs.This pattern of results suggests that amnesicpatients can have normal implicit memoryfor newly acquired associations, just as theycan show normal priming for items withpreexisting representations, such as words(e.g., Diamond & Rozin, 1984; Graf, Squire,& Mandler, 1984; Warrington & Weiskrantz,1970, 1974). Thus, the present findings sup-port the view that implicit and explicit mem-ory for new associations are mediated bydifferent underlying processes.

It is necessary, however, to consider also amore parsimonious account of the presentresults, one that does not require the as-sumption of two qualitatively different formsof memory. The present results might alsobe interpreted in terms of differences in testdifficulty; it could be that word completionis an easier test than cued recall because thecues provided on the completion test, partic-ularly in the same-context condition (e.g.,MOLD-BRE ), provide more informationthan do the cues provided on the recall test(e.g., MOLD- ). Thus, amnesic patientsmay show normal performance on the easytest and a deficit on the difficult test. Accord-ingly, there may be no need to assume thatan implicit form of memory mediates theretention of newly acquired associations inamnesic patients; it may be that the explicitform of memory is recruited whenever aretention test becomes sufficiently easy.

Three aspects of the results from the presentexperiment, however, argue against a testdifficulty explanation of the amnesic patients'performance on the word completion test.First, on the argument that word completionis an easier test than cued recall, it would beexpected that all subject groups would showa higher level of performance in the same-context condition of the completion test thanon the cued-recall test. However, this outcomewas not observed. The relevant data arepresented in Figure 2 and in Table 3, which


illustrate that (a) the student controls showeda higher level of cued-recall performancethan word completion performance in thesame-context condition, (b) the matched con-trols showed similar levels of cued recall andword completion performance, and (c) onlythe amnesic patients performed more poorlyon the recall test than on the completion test.This pattern of findings highlights the sub-stantial between-group differences on the re-call test, and the absence of such differenceson the completion test, thus arguing againsta simple quantitative difference interpretationfor the completion and recall test results.

Second, it also follows from the test diffi-culty argument that a word that was givenon cued recall—the hard test—should alsohave been given on completion—the easytest, and consequently, the probability offailing to complete a word that was given onthe recall test [i*(c|R)3 should be very low.To evaluate this possibility, we computedP(c|R) for words that had been tested forcompletion in the same-context condition.Contrary to the test difficulty argument, ahigh proportion of the recalled words hadnot been given on the word completion test.Specifically, P(c\R) was 44.1 and 62.1% forrelated and unrelated pairs, respectively, forstudent controls; 32.0 and 52.5% for relatedand unrelated pairs, respectively, for matchedcontrols; and 31.0% for related pairs foramnesic patients (recall of unrelated pairswas too low for a meaningful contingencyanalysis).

Third, the test difficulty argument wouldpredict statistical dependence between com-pletion and recall performance for words thathad been tested in the same-context conditionof the completion test. However, there weresigns of dependence only in the data for theage matched control group (x2C) = 3.8 and8.0 for unrelated and related word pairs); thedata from the amnesic patients and the stu-dent control group showed no evidence ofdependence (all x 2 0) < 2.5). In view of thefact that strong dependence between wordcompletion and cued recall was observed inExperiment 1, it is not entirely clear why wefound independence in some conditions ofthe present experiment. There are a numberof procedural differences between Experiment1 and 2 that could explain this pattern of

findings, such as length of study list and thecomposition of the word completion test.

Taken together, the foregoing findings arenot consistent with a test difficulty interpre-tation of the results from Experiment 2.Instead, they provide additional evidence forthe view that word completion performanceand cued-recall performance reveal qualita-tively different forms of memory. The presentresults also emphasize that the sensitivity toa particular form of memory is not deter-mined solely by the nominal amount ofinformation that is provided on the test.Instead, it appears that a test's sensitivity toa particular form of memory is heavily influ-enced by the instructions that are used toinduce a testing set for the implicit expressionof memory or for explicit remembering (cf.Graf & Mandler, 1984).

On the argument that instructions are crit-ical for defining the nature of a test, it followsthat if the completion test used here weregiven with instructions for explicit remem-bering, a different pattern of results wouldemerge. Specifically, if both the completiontest and the recall test were given with explicitremembering instructions, .P(c|R) should bevery low, because the two tests would thenmeasure the same form of memory. To assessthis possibility, a group of 6 students studiedthe list of unrelated pairs that was used inExperiment 2 and was then given the wordcompletion test followed by the cued-recalltest. In contrast to the previous experiments,however, these subjects were given instructionsthat emphasized that the cues on both testswere related to the words from the studiedpairs and that the task was to use these cuesas aids for recalling the words from the studylist. Consistent with expectations, the resultsshowed that for pairs tested with same-contextitems on the completion test, P(c\R) was only14% in contrast to 62.1% when completioninstructions were used in Experiment 2. Thisfinding reveals the critical nature of instruc-tion in defining what is tapped by a test, andit strengthens the view that word completionand cued recall are sensitive to qualitativelydifferent forms of memory.

General Discussion

The present study complements previousresearch on direct priming and explicit re-


membering in three ways. First, the resultsshow that word completion performance isaffected by newly acquired associations be-tween pairs of unrelated words. This dem-onstration extends the findings on priming ofnew associations shown previously with lexicaldecision, word identification, and listeningand reading tests (e.g., Carroll & Kirsner,1982; Franks et aL, 1982; McKoon & Ratcliff,1979; Moscovitch, 1984). Second, the resultsshow that implicit memory for new associa-tions can occur independently of the level ofexplicit remembering. It was found that am-nesic patients and control subjects showedsimilar levels of performance in all conditionsof the word completion test despite the am-nesics' severe deficit on the cued-recall test.This pattern of results indicates that implicitand explicit forms of memory can be disso-ciated for newly acquired associations as wellas for preexisting associations. Third, thisstudy identified a critical factor in implicitmemory for new associations by showing thatit depends on elaborative processing of to-be-remembered materials.

The present finding that implicit memoryfor new associations requires elaborative pro-cessing, differs from previous demonstrationsthat showed that implicit memory for itemswith preexisting representations, is indepen-dent of semantic-non-semantic task manip-ulations (e.g., Graf et al., 1982; Graf &Mandler, 1984; Graf, Squire, & Mandler,1984; Jacoby & Dallas, 1981). These con-trasting results raise questions about the rep-resentations that mediate implicit memoryfor new associations. One possibility is thatimplicit memory occurs independently of thespecific study task requirements with singlewords because they have highly integratedrepresentations that are completely activatedeven when only some of their componentsare processed deliberately (cf. Horowitz &Prytulak, 1969; Rumelhart & Ortony, 1978).In contrast, there is no preexisting unitizedrepresentation for a pair of unrelated words.Thus, the relation between these words mustfirst be constructed by engaging in elaborativeprocessing during the study trial, and onlythen can it support the redintegration of thepair subsequently.

The demonstration in Experiment Lthatthe same-different effect depended on elabo-

rative processing for both related and unre-lated pairs may seem inconsistent with thisinterpretation, because a pair of related words(e.g., RIPE-APPLE) is linked by a preexistingassociation, and thus even the nonelaborativevowel comparison task should have produceda same-different effect on the completiontest. However, these moderately related pairsmay not be represented in an integrated orunitized manner, and hence they still requireelaborative processing in order to affect per-formance on a test for implicit retention. Itis possible that highly related pairs (e.g.,TABLE-CHAIR), that are represented as inte-grated units (cf. Hayes-Roth, 1977), wouldhave produced a same-different effect on theword completion test even under nonelabor-ative task conditions.

Although our findings indicate that primingentails more than activation of preexistingrepresentations, implicit memory for a pairof newly related words may still depend crit-ically on the preexisting representations ofthe component words. Some degree of depen-dence on preexisting representations is sug-gested by the finding that in the presentexperiments, a same-different effect for un-related word pairs was observed when thecompletion test cues consisted of the stimulusword and part of the response word fromeach list pair (e.g., BALANCE-BRE ). How-ever, when the test presented only the stimulusword from each pair, and no part of theresponse word (e.g., BALANCE ), therewas no evidence of a priming effect. Thislatter finding was observed in a pilot studywith 4 amnesic patients. These patients wererequired to generate sentences for randomlypaired words as in the present study, andthen they were given a completion test forthe target words. In contrast to the completiontest used here, however, the test used in thepilot study presented only the stimulus wordfrom each pair and required subjects to re-spond with the first word that came to mind.The subjects never produced the responsewords, even after four study-test trials.

A recent study by Schacter (in press-a)provides a more systematic demonstration ofthe fact that priming of new associations isobserved only when the test cue consists ofparts of both words from an unrelated pair.Schacter presented amnesic patients and con-


trol subjects with unitized phrases that rep-resented linguistic idioms (e.g., SOUR GRAPES,SMALL POTATOES) and nonunitized phrasesthat were formed by re-pairing the compo-nents of the idioms (e.g., SMALL GRAPES,SOUR POTATOES), and then gave a completiontest followed by a free-recall test and a cued-recall test. The critical data come from thecompletion test, which showed the initialword from each phrase as a cue (e.g.,SOUR , SMALI ) and required sub-jects to write the first response word thatcame to mind. Schacter found that on thiscompletion test, subjects produced 26.3% ofthe correct response words for the unitizedphrases (27% for amnesics and 26% for con-trols), but they produced virtually none ofthe response words for the nonunitizedphrases (0% for amnesics and 1 % for controls).The absence of a priming effect on the non-unitized or unrelated items stands in contrastwith results from the present study, whichdiffered from Schacter's study primarily interms of the cues that were provided on thetest. When the completion test reinstated partor all of each preexisting component of anunrelated word pair, as in the present exper-iments, there was evidence of implicit mem-ory for new associations; but when the testprovided only one preexisting component ofa pair of unrelated words, as in Schacter'sstudy, there was no evidence for implicitmemory. Schacter's results show, however,that for a pair of words that forms a well-integrated, preexisting unit (e.g., SOURGRAPES), one word from the pair is sufficientfor cuing the entire unit on a completiontest. This pattern of results reveals a distinc-tion between the preexisting representationfor a pair of related words (e.g., a linguisticidiom) that can be redintegrated when oneof its components is provided as a cue andthe newly established representation for apair of previously unrelated words whoseredintegration requires that the completiontest cue provides access to each of its preex-isting components.

The present results, in conjunction withother studies, have implications for a numberof interpretive accounts that have been offeredfor implicit memory. As discussed earlier, ithas been argued that priming is mediated bythe automatic activation of preexisting rep-

resentations (e.g., Diamond & Rozin, 1984;Graf, Squire, & Mandler, 1984; Mandler,1980;Mortensen, 1980; Morton, 1969, 1979;Rozin, 1976; Warrington & Weiskrantz, 1982;Wickelgren, 1979). This view is difficult toreconcile with the finding of implicit memoryfor newly acquired associations that have nopreexisting representation, and with the factthat this type of memory occurs only undersemantic or elaborative study conditions andnot under nonsemantic conditions. A relatedaccount suggests that direct priming is me-diated by decontextualized semantic memoryrepresentations that can exist independentlyof context specific representations in episodicmemory (e.g., Kinsbourne & Wood, 1975,1982; Schacter & Tulving, 1982). This viewis not supported by the present finding of acontext effect on completion performancethat indicates that implicit memory can bemediated by representations established as aresult of relating two words on a single studytrial. A third alternative view suggests thatboth implicit and explicit memory for arecent experience are mediated by the sameepisodic memory representation (Jacoby,1983; Jacoby & Witherspoon, 1982). By thisview, an episode is retrieved with awarenessfor an explicit test, and the same episode isretrieved without awareness for an implicittest. The implicit memory for new associa-tions observed in the present study is consis-tent with this view, insofar as it indicates thatpriming can be affected by new representa-tions that were established during a singlestudy trial. If, however, implicit memory ismediated by the same episodic representationthat supports explicit remembering, primingshould be observed even when only one wordof a studied pair is presented as a cue on animplicit test. However, as discussed previously,this outcome is found only for highly related,unitized pairs and not for newly acquiredpairs. Priming of new pairs requires a testcue that includes a component correspondingto the preexisting representation of the targetresponse, whereas explicit remembering of anew pair can occur even when only one partof a new unit is presented as a test cue. Asimilar problem about underlying represen-tations is raised by the finding that amnesicpatients do not show priming for pseudowordswhich have no preexisting representations


(e.g., Cermak, 1984; Diamond & Rozin, 1984;see Schacter, in press-b, for discussion).

The present results also have implicationsfor the understanding of preserved and im-paired memory processes in amnesia. Mostaccounts of preserved memory processes inamnesia emphasize that amnesic patients canstrengthen existing representations, associa-tions, or procedures (e.g., Diamond & Rozin,1984; Mandler, 1980; Squire & Cohen, 1984;Warrington & Weiskrantz, 1982; Wickelgren,1979). It has been suggested that thisstrengthening or tuning occurs gradually as aresult of repetition (e.g., Mandler, 1980;Mishkin, Malamut, & Bachevalier, 1984;Squire, Cohen, & Nadel, 1984). The findingsfrom Experiment 2 do not support thesearguments because they show that at leastsome amnesic patients can retain a new as-sociation between normatively unrelatedwords that was established on a single studytrial. These results confirm previous reportsof memory for new associations in amnesicpatients (Moscovitch, 1984; Schacter, Har-bluk, & McLachlan, 1984; Weiskrantz &Warrington, 1979). Whether this aspect ofmemory is intact in all types and degrees ofamnesia remains to be determined.

The new associations retained by amnesicpatients, however, are only sufficient to sup-port implicit memory; they do not supportexplicit remembering. It could be argued thatthis finding was observed because amnesicpatients are capable of forming relativelyweak associations (cf. Milner, 1984) that re-quire the presence of both the stimulus andpart of the response in order to be expressed,whereas normal subjects form strong associ-ations that can be elicited by the stimulusword alone. This argument, however, wouldhave difficulty accommodating the resultsfrom the contingency analysis described ear-lier, which revealed that normals and evenamnesic patients often give a correct responseon the cued-recall test after having failed toproduce the same response on the completiontest. It seems implausible to suggest thatsubjects fail on a test that is sensitive to weakassociations, and then succeed on a test thatrequires stronger associations for successfulperformance. Instead, it seems preferable topostulate two qualitatively distinct represen-tational consequences of a learning episode,

only one of which occurs normally in amnesicpatients.

In conclusion, implicit memory for newassociations, independent of explicit remem-bering, is mediated by a representation thatis established as a result of semantic orelaborative processing on a single study trial.The representation that mediates implicitmemory for new associations depends onpreexisting representations, and its redinte-gration requires that part of each preexistingunit is presented on the memory test. Incontrast, the representation that mediates ex-plicit memory for new associations does notdepend on preexisting units in the same way;recall and recognition can occur even whenonly one part of a completely new unit isgiven as a test cue. These observations furtherstrengthen the suggestion that there are qual-itatively different representational conse-quences of an experience that mediate implicitand explicit memory for new associations.However, a detailed understanding of thenature of these differences will have to awaitthe results of further research.

References

Baddeley, A. D. (1982). Amnesia: A minimal model andan interpretation. In L. S. Cermak (Ed.), Humanmemory and amnesia (pp. 305-336). Hillsdale, NJ:Erlbaum.

Brooks, D. N., & Baddeley, A. D. (1976). What canamnesic patients learn? Neuropsychologia, 14, 111-122.

Bruner, J. S. (1969). Modalities of memory. In G. A.Talland & N. C. Waugh (Eds.), The pathology ofmemory (pp. 253-259). New \brk: Academic Press.

Carroll, M., & Kirsner, K. (1982). Context and repetitioneffects in lexical decision and recognition memory.Journal of Verbal Learning and Verbal Behavior, 21,55-69.

Cermak, L. (1984, June). Some attempts to improve thememory of amnesic patients. Paper presented to theInternational Neuropsychological Symposium, Beaune,France.

Cofer, C. C. (1967). Conditions for the use of verbalassociations. Psychological Bulletin, 68, 1-12,

Cohen, N. J. (1984). Preserved learning capacity inamnesia: Evidence for multiple memory systems. InL. R. Squire & N. Butters (Eds.), Neuropsychology ofmemory (pp. 83-103). New York: Guilford Press.

Cohen, N. J., & Squire, L. R. (1980). Preserved learningand retention of pattern-analyzing skill in amnesia:Dissociation of "knowing how" and "knowing that."Science, 210, 207-209.

Cramer, R (1966). Mediated priming of associative re-


sponses: The effect of time lapse and interpolatedactivity. Journal of Verbal Learning and Verbal Behavior,5, 163-166.

Diamond, R., & Rozin, P. (1984). Activation of existingmemories in the amnesic syndrome. Journal of Ab-normal Psychology, 93, 98-105.

Eich, J. E. (1984). Memory for unattended events: Re-membering with and without awareness. Memory &Cognition, 12, 105-111.

Franks, J. J., Ptybon, C. J., & Auble, P. M. (1982). Unitsof episodic memory in perceptual recognition. Memory& Cognition. 10, 62-68.

Graf, P., & Mandler, G. (1984). Activation makes wordsmore accessible but not necessarily more retrievable.Journal of Verbal Learning and Verbal Behavior, 23,553-568.

Graf, P., & Mandler, G. (1985). Direct priming acrosssensory modalities and word boundaries. Unpublishedmanuscript.

Graf, P., Mandler, G., & Haden, M. (1982). Simulatingamnesic symptoms in normal subjects. Science, 218,1243-1244.

Graf, P., Shimamura, A. P., & Squire, L. R. (1984).Priming across modalities and priming across categorylevels: Extending the domain of preserved function inamnesia. Journal of Experimental Psychology: Learning,Memory, and Cognition, 11, 385-395.

Graf, P., Squire, L. R., & Mandler, G. (1984). Theinformation that amnesic patients do not forget. Journalof Experimental Psychology: Learning, Memory, andCognition, 10, 164-178.

Hayes-Roth, B. (1977). Evolution of cognitive structuresand processes. Psychological Review, 84, 260-278.

Horowitz, L. M., & Prytulak, L. S. (1969). Redintegrativememory. Psychological Review, 76, 519-531.

Jacoby, L. L. (1983). Remembering the data: Analyzinginteractive processes in reading. Journal of VerbalLearning and Verbal Behavior, 22, 485-508.

Jacoby, L. L., & Dallas, M. (1981). On the relationshipbetween autobiographical memory and perceptuallearning. Journal of Experimental Psychology: General,110, 306-340.

Jacoby, L. L., & Witherspoon, D. (1982). Rememberingwithout awareness. Canadian Journal of Psychology,36, 300-324.

Kinsbourne, M., & Wood, F. (1975). Short-term memoryand the amnesic syndrome. In D. D. Deutsch & J. A.Deutsch (Eds.), Short-term memory. New York: Aca-demic Press.

Kinsbourne, M., & Wood, F. (1982). Theoretical consid-erations regarding the episodic-semantic memory dis-tinction. In L. S. Cermak (Ed.), Human memory andamnesia (pp. 195-217). Hillsdale, NJ: Erlbaum.

Kucera, M., & Francis, W. (1967). Computational analysisof present-day American English. Providence, RI: BrownUniversity Press.

Mandler, G. (1980). Recognizing: The judgment of pre-vious occurrence. Psychological Review, 87, 252-271.

McKoon, G , & Ratcliff, R. (1979). Priming in episodicand semantic memory. Journal of Verbal Learning andVerbal Behavior, 18, 463-480.

Meyer, V., & Yates, A. J. (1955). Intellectual changesfollowing temporal lobectomy for psychomotor epilepsy.Journal of Neurology, Neurosurgery and Psychiatry, 18,44-52.

Milner, P. (1984, June). Amnesia: Some neurophysiologicalconsiderations. Paper presented at the Canadian Psy-chological Association meeting, Ottawa, Ontario, Can-ada.

Mishkin, M., Malamut, B., & Bachevalier, J. (1984).Memories and habits: Two neural systems. In J. L.McGaugh, G. Lynch, & N. M. Weinberger (Eds.),Neuwbiology of learning and memory. New York:Guilford Press.

Mortensen, E. L. (1980). The effects of partial informationin amnesic and normal subjects. Scandinavian Journalof Psychology 21, 75-82.

Morton, J. (1969). The interaction of information inword recognition. Psychological Review, 76, 165-178.

Morton, J. (1979). Facilitation in word recognition: Ex-periments causing change in the logogen models. InP. A. Kolers, M. E. Wrolstad, & H. Bouma (Eds.),Processing of visible language (Vol. 1). New York:Plenum.

Moscovitch, M. (1982). Multiple dissociations of functionin amnesia. In L. S. Cermak (Ed.), Human memoryand amnesia (pp. 337-370). Hillsdale, NJ: Erlbaum.

Moscovitch, M. (1984). The sufficient conditions fordemonstrating preserved memory in amnesia. InL. R. Squire & N. Butters (Eds.), Neuropsychology ofmemory (pp. 104-114). New York: Guilford Press.

O'Keefe, J., & Nadel, L. (1978). The hippocampus as acognitive map. Oxford, England: Clarendon Press.

Roediger III, H., & Blaxton, X A. (1983). Priming inword fragment completion: Effects of modality andorthography. Paper presented at the 24th annual meetingof the Psychonomic Society, San Diego.

Rozin, P. (1976). The psychobiological approach to humanmemory. In M. R. Rosenzweig & E. L. Bennett (Eds.),Neural mechanisms of learning and memory (pp. 3 -48). Cambridge, MA: MIT Press,

Rumelhart, D. E., & Ortony, A. (1978). The representationof knowledge in memory. In R, C Anderson, R. J.Spiro, & W. E. Montague (Eds.), Schooling and theacquisition of knowledge. Hillsdale, NJ: Erlbaum.

Scarborough, D. L., Gerard, L., & Cortese, C. (1979).Accessing lexical memory: The transfer of word repe-tition effects across task and modality. Memory &Cognition. 7, 3-12.

Schacter. D. L. (1983). Amnesia observed: Rememberingand forgetting in a natural environment. Journal ofAbnormal Psychology, 92, 236-242.

Schacter, D. L. (in press-a). Priming of old and newknowledge in amnesic patients and normal subjects.Annals of the New York Academy of Sciences.

Schacter, D. L. (in press-b). Multiple forms of memoryin humans and animals. In N. Weinberger, J. McGaugh,& G. Lynch (Eds.), Memory systems of the brain. NewYork: Guilford Press.

Schacter, D. L., Harbluk, J. A., & McLachlan, D. R.(1984). Retrieval without recollection: An experimentalanalysis of source amnesia. Journal of Verbal Learningand Verbal Behavior. 23, 593-611.

Schacter, D. L., McLachlan, D. R., Moscovitch, M., &Tulving, E. {1984, August). Tracking of memory dis-orders over time. Paper presented to the AmericanPsychological Association, Toronto.

Schacter, D. L., & Moscovitch, M. (1984). Infants, am-nesics, and dissociable memory systems. In M. Mas-


covitch (Ed.), Infant memory (pp. 173-216). NewYork: Plenum.

Schacter, D. L., & Tulving, E. (1982). Memory, amnesia,and the episodic/semantic distinction. In R. L. Isaacson& N. E. Spear (Eds.), The expression of knowledge(pp. 33-65). New York: Plenum Press.

Scoville, W. B., & Milner, B. (1957). Loss of recentmemory after bilateral hippocampal lesions. Journalof Neurology, Neurosurgery, & Psychiatry, 20, 11-21.

Shapiro, S. L, & Palermo, D. S. (1968). An atlas ofnormative free association data. Psychonomic Mono-graph Supplement, 2(12, Whole No. 28), 219-250.

Shimamura, A. P., & Squire, L. R. (1984). Paired-associate learning and priming effects in amnesia: Aneuropsychological study. Journal of ExperimentalPsychology: General, 113, 556-570.

Squire, L. R. (1982). Comparisons between forms ofamnesia: Some deficits are unique to Korsakoff'ssyndrome. Journal of Experimental Psychology:Learning, Memory, and Cognition, 8, 560-571.

Squire, L. R., & Cohen, N. J. (1984). Human memoryand amnesia. In J. McGaugh, G. Lynch, & N. Wein-berger (Eds.), Proceedings of the Conference on theNeurobiology of Learning and Memory. New York:Guilford Press.

Squire, L. R., Cohen, N. J., &. Nadel, L. (1984). Themedial temporal region and memory consolidation: Anew hypothesis. In H. Weingartner & E. Parker (Eds.),Memory consolidation. Hillsdale, NJ: Erlbaum.

Starr, A., & Phillips, L. (1970). Verbal and motor memoryin the amnestic syndrome. Neuropsychologia, 8, 75 -88.

Tulving, E. (1983). Elements of episodic memory. Oxford,England: Clarendon Press.

Tulving, E., Schacter, D. L., & Stark, H. A. (1982).Priming effects in word-fragment completion are in-dependent of recognition memory. Journal of Experi-mental Psychology: Learning, Memory, and Cognition,8. 336-342.

Warrington, E. K., & Weiskrantz, L. (1968). New methodof testing long-term retention with special reference toamnesic patients. Nature, 217, 972-974.

Warrington, E. K., & Weiskrantz, L. (1970). Amnesia:Consolidation or retrieval? Nature, 228, 628-630.

Warrington, E. K., & Weiskrantz, L. (1974). The effectof prior learning on subsequent retention in amnesicpatients. Neuropsychologia, 12, 419-428.

Warrington, E. K., & Weiskrantz, L. (1982). Amnesia:A disconnection syndrome? Neuropsychologia, 20, 233-248.

Weiskrantz, L., & Warrington, E. K. (1979), Conditioningin amnesic patients. Neuropsychologia, 17, 187-194.

Wickelgren, W. A. (1979). Chunking and consolidation:A theoretical synthesis of semantic networks, configur-ing in conditioning, S-R versus cognitive learning,normal forgetting, the amnesic syndrome, and thehippocampal arousal system. Psychological Review, 86.44-60.

Winocur, G., & Kinsbourne, M. (1978). Contextualcueing as an aid to Korsakoff amnesics. Neuropsycho-logia, 16. 671-682.

Winocur, G., & Weiskrantz, L. (1976). An investigationof paired-associate learning in amnesic patients. Neu-ropsychologia, 14, 97-110.

Received June 13, 1984Revision received November 26, 1984

Date post:	04-Jun-2020
Category:	Documents
Upload:	others
View:	7 times
Download:	0 times

Implicit and Explicit Memory for New Associations …...that implicit memory differs fundamentally...

Documents