Encapsulation of Implicit and ExplicitMemory in Sequence LearningPaul - PDF document

Encapsulation of Implicit and ExplicitMemory in Sequence LearningPaul J. ReberUCSD School of MedicineLarry R. SquireDepartment of Veterans Affairs and UCSD School of MedicineAbstractn Contrasts between implicit and explicit knowledge in theserial reaction time (SRT) paradigm have been challenged be-cause they have depended on a single dissociation: intact im-plicit knowledge in the absence of corresponding explicitknowledge. In the SRT task, subjects respond with a corre-sponding keypress to a cue that appears in one of four loca-tions. The cue follows a repeating sequence of locations, andsubjects can exhibit knowledge of the repeating sequencethrough increasingly rapid performance (an implicit test) or bybeing able to recognize the sequence (an explicit test). In ourstudy, amnesic patients were given extensive SRT training. Theirimplicit and explicit test performance was compared to theperformance of control subjects who memorized the trainingsequence. Compared with control subjects, amnesic patientsexhibited superior performance on the implicit task and im-paired performance on the explicit task. This crossover inter-action suggests that implicit and explicit knowledge of theembedded sequence are separate and encapsulated and thatthey presumably depend on different brain systems. nINTRODUCTIONMemory is not a single faculty but is composed of mul-tiple separate abilities (Schacter, 1987; Squire, 1992; Tul-ving, 1985; Weiskrantz, 1990). One major distinctioncontrasts declarative (explicit) memory, which supportsconscious memory of facts and events, with nondeclara-tive (implicit) memory, which supports a range of phe-nomena including habit learning, simple conditioning,and priming. The study of high-level cognition, in whichtask decisions are based on complex rules and some-times detailed task information, has traditionally involvedthe assumption that task-relevant decision processes areguided mainly by declarative, conscious memory (Du-lany, Carlson, & Dewey, 1985; Perruchet & Gallego, 1993;Shanks & St. John, 1994a). However, a growing body ofevidence has identiŽed cognitive skill-learning tasks thatdepend primarily on nondeclarative memory (Squire &Zola, 1996).A powerful technique for examining the role of dec-larative and nondeclarative memory in cognitive skilllearning is to study amnesic patients. Amnesic patientshave damage to the medial temporal lobe or di-encephalic structures and have severely impaired decla-rative memory. Yet, for a number of rather complexcognitive tasks, amnesic patients have been found tolearn and retain information normally, despite being un-© 1998 Massachusetts Institute of TechnologyJournal of Cognitive Neuroscience 10:2, pp. 248–263able to recollect the task stimuli or the experimentalcontext. These tasks include sequence learning (Nissen& Bullemer, 1987; Reber & Squire, 1994), prototype learn-ing (Knowlton & Squire, 1994), artiŽcial grammar learn-ing (Knowlton, Ramus, & Squire, 1992), and probabilisticclassiŽcation learning (Knowlton, Squire, & Gluck, 1994).One challenge to this body of research in amnesic pa-tients is that the work has depended on a single disso-ciation: intact implicit knowledge in the absence ofcorresponding explicit knowledge. This line of argumentled to the proposal that there is a single, presumablyexplicit, source of knowledge and that implicit and ex-plicit tests are simply differentially sensitive to detectinggroup differences (Shanks & St. John, 1994a). This impor-tant, alternate idea has been termed the Sensitivity hy-pothesis.In one relevant study, Reber and Squire (1994) com-pared the performance of amnesic patients and controlsubjects on the SRT task. In this task, a cue could appearin any one of four locations, and subjects responded asquickly as possible to each appearance of the cue bypressing the key directly beneath the cue. Subjects werenot told that the cue followed a repeating sequence oflocations, but with practice they nevertheless exhibitedgradually decreasing reaction times to the cue. When therepeating sequence was replaced with a random se-quence of locations, subjects exhibited slower reaction times, indicating that some of their previous speed wasdue to knowledge of the repeating sequence. Theamount of sequence knowledge expressed in this waywas equivalent for amnesic patients and control subjects.In contrast, the amnesic patients were markedly im-paired relative to the control subjects at recognizing therepeating sequence when it was presented and, unlikethe control subjects, they were unable to report thesequence verbally.This Žnding demonstrates the standard single dissocia-t

ion: intact implicit knowledge for the sequence andimpaired explicit memory for the sequence. However,Shanks and St. John (1994a) pointed out that, if theimplicit memory test were less sensitive to group differ-ences than the explicit memory test, it could still be thecase that a single source of knowledge supports perfor-mance on both tests. That is, the amnesic patients mighthave actually learned less about the sequence than con-trol subjects, but only the explicit memory test is sensi-tive to this difference. The Sensitivity hypothesis wouldaccount for this pattern of results by supposing that theknowledge-performance relationship for the implicitmemory test is such that subjects can express signiŽcantSRT skill with a relatively small amount of knowledgeand that a small amount of knowledge is as effective asa large amount of knowledge at supporting performance.In addition, the explicit memory test may be relativelyinsensitive to a small amount of knowledge, but a largeamount of knowledge is easily detectable.By this hypothesis, amnesic patients should consis-tently perform slightly worse than control subjects onimplicit memory tests (although the difference betweengroups need not be statistically signiŽcant). Consistentwith this idea, in some of the studies carried out to date,overall performance of the amnesic patients was numeri-cally worse than that of control subjects (see Shanks &St. John, 1994b). Although such Žndings are consistentwith the predictions of the single source of knowledgeview and the Sensitivity hypothesis, they also do notcontradict the multiple memory systems view. Perhapscontrol subjects can sometimes engage explicit memorystrategies and thereby perform better than they couldby relying solely on implicit memory. For this reason, itis not surprising that control subjects sometimes per-form slightly, but not signiŽcantly, better than amnesicpatients.These alternate views cannot be easily distinguishedon the basis of the evidence available from a singledissociation. One approach has been to study additionalpatient groups to try to establish the opposite dissocia-tion (i.e., intact explicit memory in the presence ofimpaired implicit memory). Knowlton, Mangels, andSquire (1996) found that patients with Parkinson’s dis-ease were impaired at habit learning despite havingintact explicit memory for the testing session. Anotherapproach has been to examine the implicit learning ofamnesic patients who are so severely impaired that theyappear to have virtually no explicit memory at all. Forexample, a severely amnesic patient (EP) exhibited nor-mal prototype learning (Squire & Knowlton, 1995) andintact perceptual priming (Hamann & Squire, 1997) butperformed at chance on parallel tests of recognitionmemory.An approach that has not been explored is to try toobtain a crossover interaction in the performance ofamnesic patients and control subjects. If the multiplememory system view is correct, circumstances might befound such that one group exhibits signiŽcantly moreimplicit task knowledge than another group, but sig-niŽcantly less explicit task knowledge. Such a Žndingwould contradict the Sensitivity hypothesis by demon-strating that the implicit and explicit memory tests areboth sensitive enough to yield signiŽcant group differ-ences. The approach taken in the current study was totry to create a situation in which amnesic patients hadsigniŽcantly more implicit knowledge than control sub-jects but signiŽcantly less explicit knowledge. We usedthe SRT task, which has previously been shown to sup-port normal sequence learning in amnesic patients (Nis-sen & Bullemer, 1987; Nissen, Willingham, & Hartman,1989; Reber & Squire, 1994). The task has the additionaladvantage that tests of implicit sequence learning andexplicit sequence memory necessarily depend on infor-mation about the embedded, repeating sequence itself.That is, the implicit and explicit memory tests dependon similar information.Following training on the SRT task, knowledge of therepeating sequence was assessed with tests sensitive toeither implicit or explicit knowledge. Explicit sequenceknowledge was tested with a recognition memory test.Implicit sequence knowledge was tested by assessingSRT task performance. SpeciŽcally, learning the repeatingsequence makes the location of the cue predictable andleads to faster keypresses (shorter reaction times) inresponse to successive appearances of the cue. Reactiontimes can also decrease due to the learning of generalkeypressing skills (nonspeciŽc task learning). To assessthe extent to which faster keypressing speed dependson speciŽc knowledge of the tr

aining sequence, thetraining sequence was abruptly changed to a different,novel sequence without forewarning the subject. Theamount of reaction-time slowing that occurred when therepeating sequence was changed indexes the amount ofimplicit knowledge that was acquired about the trainingsequence.The important comparison was between amnesic pa-tients who learned a repeating sequence implicitly bypracticing it during extended training and four differentgroups of control subjects who learned the same se-quence explicitly by observing and memorizing butwithout actually practicing the sequence. We hypothe-sized that amnesic patients, who have impaired declara-tive memory but are capable of normal sequencelearning (Nissen & Bullemer, 1987; Nissen, Willingham, &Reber and Squire 249 Hartman, 1989; Reber & Squire, 1994), would acquiresigniŽcant implicit knowledge of the repeating sequencewithout acquiring much explicit knowledge about it. Wefurther hypothesized that control subjects, who at-tempted to memorize the sequence without practicingit, would acquire explicit knowledge of the repeatingsequence but would acquire little or no implicit knowl-edge about it.NonspeciŽc LearningAs described above, implicit knowledge of the sequencewas measured by noting the increase in reaction timesthat occurred between a block of trials containing thetraining sequence and an immediately following block oftrials containing a novel sequence. However, there is apossible complication in measuring implicit knowledgein this way. If sufŽcient nonspeciŽc learning occurredwhile practicing the training sequence, reaction timesmight be fast both during the Žnal trials with the trainingsequence and also during subsequent trials with thenovel sequence. In other words, evidence for speciŽcknowledge of the sequence (namely, reaction time slow-ing during the novel sequence) could be masked ifconsiderable nonspeciŽc learning continued to occurduring the test of implicit knowledge. Because of thispossibility, a failure to Žnd slower reaction times duringthe novel sequence could mean either that no implicitknowledge of the sequence had been acquired or thatimplicit knowledge had been acquired but was maskedby ongoing nonspeciŽc learning. We therefore gave pre-training on the SRT task to three groups of controlsubjects, who memorized the sequence. Pretraining per-mitted nonspeciŽc learning to occur prior to the test ofimplicit knowledge and provided a way to demonstratedirectly that nonspeciŽc learning was not preventing theexpression of speciŽc sequence knowledge.InterferenceSubjects who receive pretraining on the SRT task willhave acquired nonspeciŽc task knowledge and will alsohave acquired implicit knowledge of the pretrainingsequence itself. Another possible complication was thatimplicit knowledge of the pretraining sequence mightinterfere with learning and expressing implicit knowl-edge of the subsequent training sequence. To explorethis possibility, one group of control subjects was givenpretraining in which the cue appeared in a randomseries of locations. Another group received no pretrain-ing so that there could be no interference from a pre-viously learned sequence.RESULTSSRT performance was measured as mean reaction time(RT) for each 60-trial block of performance. In calculat-ing the mean RT for each 60-trial block, trials containingerrors (i.e., an incorrect keystroke response by the sub-ject) and trials containing RTs greater than 1000 msecwere eliminated. An average of 6% of the reaction timeswere eliminated.PretrainingMean RTs for each block of pretraining are shown inFigure 1 for the MemorizeOld, MemorizeYoung, Baseline,and MemorizeYoung Random Pretraining groups. For theMemorizeOld, MemorizeYoung, and Baseline groups, im-plicit sequence knowledge of the pretraining sequenceS2 was measured by the increase in reaction times onthe fourth block, which contained Žve repetitions of anovel sequence N2. Each group exhibited signiŽcantknowledge of the pretraining sequence S2. The meanincrease in RT on the fourth block for the MemorizeOldgroup was 33.4 ± 8.4 msec, t(14) = 3.97, p 0.01. Forthe MemorizeYoung group, the mean increase in RT was31.0 ± 5.8 msec, t(14) = 5.40, p 0.001. For the Baselinegroup, the mean increase in RT was 41.5 ± 11.5 msec,t(14) = 3.60, p 0.01.Evidence of nonspeciŽc SRT learning can also be seenin the pretraining performance. The Žnal pretrainingblock for the MemorizeOld, MemorizeYoung, and Baselinegroups, which consisted of repetitions of a novel se-quence, was generally performed faster than the Žrst SRTpretraining block. The decrease in RT was 35.2 ± 11.2msec for the Me

morizeOld group, t(14) = 3.15, p 0.01;54.3 ± 16.7 msec for the MemorizeYoung group, t(14) =3.25, p 0.01; and 17.9 ± 8.4 msec for the Baselinegroup, t(14) = 2.11, p 0.07. In addition, the Mem-Figure 1. Average reaction time for each 60-trial block during pre-training. For the MemorizeOld, MemorizeYoung, and Baseline groups,the open bars indicate the mean reaction time (RT) for the three 60-trial blocks containing repetitions of sequence S2, and the shadedbars indicate the mean RT for a 60-trial block containing repetitionsof a novel sequence N2. For the MemorizeYoung Random Pretraininggroup, each bar indicates the mean RT for 60-trial blocks in whichthe sequence of cue locations was pseudorandom. Standard errorsfor these scores ranged from 10 to 24 msec. 250 Journal of Cognitive NeuroscienceVolume 10, Number 2 orizeYoung Random Pretraining group also exhibited adecrease in mean RT of 48.4 ± 9.4 msec, F(1,9) = 32.4,p 0.001 that can only be attributed to nonspeciŽclearning because the sequence of locations was unpre-dictable.TrainingMean RTs for each block of SRT training for the amnesicpatients and the Control (CON) group are shown inFigure 2. Due to computer error, the training data fromthe second training session were lost for two subjects inthe CON group, and data were lost as well from the Žrsteight blocks of the second session for one amnesicpatient (EP). A 2 ´ 2 ´ 20 analysis of variance (ANOVA)comparing the performance of amnesic patients and theCON group across the two sessions (20 training blockseach) yielded a signiŽcant effect of training block, F(19,190) = 13.99, p 0.01, which reected the decreasingRTs during training. There was no effect of group or ses-sion and no interactions, Fs 1.10. A second 2 ´ 20 ANOVAfor the data from the Žrst session yielded the same results.Verbal ReportThe verbal report score is the length of the maximumsection of the repeating sequence contained in the sub-jects’ responses (maximum = 12, the length of the re-peating sequence). The amnesic patients reported anaverage of 5.0 (±3.5) elements of the repeating se-quence. The mean number of elements reported by theCON group was 4.2 (±0.3). For the groups who memo-rized the sequence, the mean number of elements con-tained in the responses was 4.7 (±0.5) for theMemorizeOld group, 8.8 (±0.8) for the MemorizeYounggroup, 9.0 (±0.9) for the MemorizeYoung Random Pre-training group, and 8.7 (±0.8) for the MemorizeYoung NoPretraining group. Performance of the three Mem-orizeYoung groups (MemorizeYoung, MemorizeYoung Ran-dom Pretraining, and MemorizeYoung No Pretraining) wassigniŽcantly better than the performance of the amnesicpatients (ts &#x 24;&#x 000; 2.71, ps 0.02) and better than the per-formance of the CON and MemorizeOld groups (ts &#x -1;! 0; 4.33,ps 0.001). The performance of the CON andMemorizeOld groups did not differ from that of the am-nesic patients (ts 0.50).The important Žnding here was that the threeMemorizeyoung subject groups did learn most of the se-quence (from 8.7 to 9.0 of the 12 sequence elements).The fact that the amnesic patients could report a fairpart of the sequence suggests the possibility that verbalreport performance was inuenced by implicit knowl-edge (see “Discussion” for consideration of these ideas).RecognitionThe mean recognition score for each group is shown inFigure 3. The score for the amnesic patients reects theaverage score for each of the two testing sessions. Table1 shows the performance of individual patients. TheCON, MemorizeOld, MemorizeYoung, MemorizeYoung Ran-dom Pretraining, and MemorizeYoung No Pretraininggroups all exhibited above-chance recognition of therepeating sequence (ts &#x 0; 4.27, ps 0.002). The amnesicpatients’ recognition performance was not differentfrom chance (t 0.88). The amnesic patients were im-Figure 2. Mean reaction timefor each 60-trial block of re-peating sequence S1 acrosstwo training sessions. Filledsquares = 5 amnesic patients;open squares are 10 controlsubjects. Blocks 1 through 20were administered on the Žrstsession, followed by the recog-nition memory test and thetest of implicit knowledge (anadditional 60-trial block of thetraining sequence S1 and a 60-trial block of a novel se-quence N1). Blocks 21through 40 were administered3 to 12 months later for theamnesic patients and 9 to 10months later for the controlsubjects. Blocks 21 through40 were also followed bytests of recognition memoryand implicit knowledge. Reber and Squire 251 paired at recognizing the sequence compared with eachof the control groups: CON group, t(13) = 2.58, p 0.03,MemorizeOld group, t(18) = 2.99, p 0.01, MemorizeYoung

group, t(18) = 8.12, p 0.001, MemorizeYoung RandomPretraining group, t(12) = 4.67, p 0.001, andMemorizeYoung No Pretraining group, t(12) = 4.91, p 0.01. The three MemorizeYoung groups all performed simi-larly, F(2,32) = 0.93. The CON and MemorizeOld groupsalso performed similarly, t(23) = 0.21. The MemorizeYounggroup remembered the sequence signiŽcantly betterthan the MemorizeOld group, t(28) = 3.47, p 0.01.Finally, the CON group exhibited less recognition mem-ory of the repeating sequence than each of the threeMemorizeYoung groups, ts &#x 0; 2.48, ps 0.03.Implicit Sequence KnowledgePerformance on the implicit sequence knowledge test isshown in Figure 4 (see also Table 1 for the performanceof individual amnesic patients in the two separate ses-sions). The amnesic patients exhibited signiŽcant knowl-edge of the repeating sequence as measured by asigniŽcant increase in reaction time on the second 60-trial block (which contained repetitions of sequence N1),Figure 3. Recognition perfor-mance. In the recognition test,subjects rated Žve sequences(the target sequence and fourfoil sequences) on a 0 to 100scale as to whether they hadseen the sequence duringtraining (either SRT practiceor Sequence Memorization).Each subject was given a rec-ognition score equal to the rat-ing assigned to the targetsequence minus the mean rat-ing assigned to the other foursequences. Bars show groupmean recognition scores.Brackets show the standard er-ror of the group mean. Table 1. Performance of Individual Amnesic Patients Session 1 Session 2 Both sessions Patient Recog. Verbalreport DRT Recog. Verbalreport DRT Recog. Verbalreport DRTPH31.28 7.3 12.54 48.2 21.96 27.8LJ 6.86 -4.3-36.83 112.2 15.04.5 54.0NF37.54 63.4 0 4 -32.0 18.84 15.7RC0 5 44.3-12.56 88.0 -6.25.5 66.2EP0 5 13.2 0 5 37.5 0 5 25.4Means15.15.6 24.8-7.44.4 50.8 9.95.0 37.8Control means(n = 10)46.8(9.1)3.8(0.3) 49.6(12.5)23.0(16.8)4.5(0.5) 63.4 (20.6) 34.9 (8.0)4.2(0.3) 58.4(15.0) Note: Recog. indicates the scores of individual patients on the recognition test of explicit memory for the repeating sequence (S1). Verbal re-port indicates the length of the longest section of the repeating sequence contained in each patient’s attempt to report the repeating se-quence. DRT indicates the difference in mean reaction time for the S1 and N1 (60-trial blocks) in the test of implicit knowledge for therepeating sequence. Each patient was tested twice in two separate sessions 3 to 5 months apart. Control subjects were tested twice in twoseparate sessions 9 to 10 months apart. For the control subjects, numbers in parentheses indicate the standard error.252 Journal of Cognitive NeuroscienceVolume 10, Number 2 t(4) = 3.97, p 0.02. The CON group also exhibitedsigniŽcant implicit sequence knowledge, t(9) = 3.90, p 0.01 and performed similarly to the amnesic patients,t(13) = 1.16. Note that the CON group did exhibitnumerically greater RT slowdown than the amnesic pa-tients as well as superior explicit knowledge (see Figure4 and Table 1). This pattern of Žndings has been reportedpreviously (Willingham, Nissen, & Bullemer, 1989) andprobably reects a contribution of explicit sequenceknowledge to SRT performance (see also “Discussion”).No other group exhibited signiŽcant knowledge of therepeating sequence (ts 1.31). In addition, the amnesicpatients expressed signiŽcantly more implicit knowl-edge of the sequence than the MemorizeOld group,t(18) = 3.49, p 0.01, the MemorizeYoung Random Pre-training group, t(12) = 2.99, p 0.02, and theMemorizeYoung No Pretraining group, t(12) = 3.43, p 0.01 and marginally more implicit knowledge of thesequence than the MemorizeYoung group, t(18) = 2.17, p0.055. Likewise, the CON group exhibited more im-plicit sequence knowledge than the other three groups,ts &#x -1;• 0; 3.04, ps 0.01. There was a trend for the Memor-izeYoung subjects to express more implicit sequenceknowledge than the MemorizeOld subjects, t(28) = 1.83,p 0.08. There was no difference in performance acrossthe three MemorizeYoung groups, F(2, 32) = 0.93.Finally, the Baseline group, which had not encoun-tered the training sequence (S1) prior to the test ofimplicit sequence knowledge, also did not exhibit im-plicit knowledge. There was virtually no increase in RTon the second block of the test, 2.4 ± 8.8 msec, t(14) =0.27. The amount of sequence knowledge expressed bythe Baseline group was less than that expressed by theamnesic patients, t(18) = 2.15, p 0.04, and the CONgroup, t(23) = 3.43, p 0.01. There was no difference inperformance between the Baseline group and any of thefour Memorize groups, ts 1.23.CrossoverThe perf

ormance scores of amnesic patients and controlsubjects on verbal report performance, the recognitionmemory test, and the implicit test of sequence knowl-edge were transformed to z scores (using the mean andstandard deviation for all subjects) in order to compareperformance across tasks (Figure 5). The two explicittests were averaged to produce a single z score. Becausethere was no difference in performance on either ofthese tests for the three groups of younger subjects(MemorizeYoung, MemorizeYoung Random Pretraining, andMemorizeYoung No Pretraining), these groups are com-bined for this analysis. A repeated measures 2 ´ 2 ANOVAshowed a signiŽcant interaction between group and taskF(1, 38) = 38.5, p 0.001 with no effect of either groupor task, Fs 1.20. An ANOVA comparing the z score dataof the amnesic patients and the MemorizeOld controlsubjects also shows a signiŽcant interaction betweengroup and task, F(1, 18) = 6.81, p 0.02 with no reliableFigure 4. Reaction times(RT) on the Implicit Knowl-edge test. (A) Open bars indi-cate the mean RT for one60-trial SRT block containingrepetitions of the training se-quence S1. Shaded bars indi-cate the mean RT for theimmediately following 60-trialSRT block, which containedrepetitions of a novel se-quence N1. (B) Implicit knowl-edge for the trainingsequence S1, as measured bythe difference in mean RT dur-ing 60 trials of sequence S1and the immediately follow-ing 60 trials of sequence N1.Thus the scores in panel (B)are subtraction scores basedon the RTs in panel (A). Brack-ets show the standard errorof the mean. Reber and Squire 253 effects of group, F(1, 18) = 2.99, p � 0.10 or task, F(1,18) = 4.13, p � 0.05. In addition, the CON group showsa similar crossover interaction when compared to thethree groups of younger subjects. A 2 ´ 2 ANOVA com-paring the CON group to the combined performance ofthe three younger groups found a signiŽcant interactionbetween group and task, F(1, 43) = 65.6, p 0.001, noeffect of group, F(1, 43) = 0.31, and a marginal effect oftask F(1, 43) = 3.29, p &#x -9; 00; 0.07. A similar ANOVA comparingthe amnesic patients to the CON group found no effectof group, F(1, 13) = 1.31, p &#x -9; 00; 0.20, an effect of task F(1,13) = 56.1, p 0.001, and no interaction, F(1, 13) = 0.38.(Note that differences in task indicated by the ANOVAscannot be readily interpreted because the scores foreach task have been normalized to z scores.)DISCUSSIONThe amnesic patients exhibited signiŽcantly more im-plicit sequence knowledge than any of the four controlgroups that memorized the sequence, and they simulta-neously exhibited signiŽcantly less explicit knowledge ofthe sequence than any of those four control groups. Thiscrossover indicates that performance on the tests ofimplicit and explicit knowledge cannot depend on thesame knowledge source. A crucial part of the crossoverinteraction is the fact that, after memorizing the se-quence, the control subjects did not exhibit any im-plicit sequence knowledge. This result was observed ineach of the four Memorize groups (MemorizeOld,MemorizeYoung, MemorizeYoung Random Pretraining, andMemorizeYoung No Pretraining), which makes it possibleto address several potentially complicating issues.NonspeciŽc LearningOne possibility is that during the test of implicit se-quence knowledge subjects acquired signiŽcant nonspe-ciŽc knowledge of the SRT task that was not relevant tothe repeating sequence. SigniŽcant nonspeciŽc learningfrequently occurs at the beginning of SRT practice. Forexample, it is apparent in Figure 1 that the RTs duringthe second 60-trial block of SRT training are faster thanthe RTs during the Žrst block for SRT pretraining (forboth repeating and random sequences). If signiŽcantnonspeciŽc learning occurred during the implicit test ofsequence knowledge, RTs for the N1 block might be asfast or faster than RTs for the S1 block (because the N1block followed the S1 block), giving the appearance thatno implicit knowledge had been acquired. The pretrain-ing given to the MemorizeOld and MemorizeYoung groupsaddressed this possibility. The fact that the MemorizeOldand MemorizeYoung groups exhibited signiŽcant implicitknowledge of the pretraining sequence S2 indicates thatafter three blocks of sequence learning, implicit se-quence knowledge could be detected and that nonspe-ciŽc SRT learning did not mask its expression.Accordingly, the implicit test of sequence knowledge forthe training sequence S1, which was administered shortlyafterward, could not have been contaminated by nonspe-ciŽc SRT learning for these groups.InterferenceIt seemed possible that, in th

e case of the MemorizeOldand MemorizeYoung groups, pretraining might interferewith the learning or expression of the second trainingsequence (S1). If so, the absence of measurable implicitknowledge in these groups might be due to interferencecaused by pretraining rather than the failure to acquireimplicit knowledge. However, this possibility can be dis-counted. The MemorizeYoung Random Pretraining groupexhibited no implicit knowledge of sequence S1 afterhaving received pretraining with a random sequence,and the MemorizeYoung No Pretraining group exhibitedno implicit sequence knowledge when no pretrainingoccurred.Figure 5. Crossover interaction between implicit and explicitknowledge. Filled squares indicate the performance of amnesic pa-tients (n = 5). Open squares indicate the combined performance ofthe three groups of young subjects who tried to memorize the se-quence (MemorizeYoung, MemorizeYoung Random Pretraining, andMemorizeYoung No Pretraining groups; n = 35). Open circles indicateperformance of MemorizeOld subjects (n = 15). Open triangles indi-cate the performance of the CON subjects who did not memorizebut received extensive SRT training (n = 10). The score for explicitknowledge was derived from the verbal report score and the scoreon the recognition memory test (Figure 3). The score for implicitknowledge was the increase in mean reaction time obtained on theŽnal 60-trial SRT block containing the novel sequence N1 comparedwith the immediately preceding 60-trial SRT block containing thetraining sequence S1. These scores were transformed to z scoresbased on the means and standard deviations obtained for these twotests by all subjects (n = 65). For the explicit tests, z scores for theverbal report and recognition memory test were averaged for eachsubject. Brackets indicate the standard error of the mean. 254 Journal of Cognitive NeuroscienceVolume 10, Number 2 Baseline Differences in RTA Žnal concern was that the younger control groupsexhibited much faster RTs overall than both the am-nesic patients and the older control groups (CON andMemorizeOld groups). Accordingly, it seemed possiblethat the younger control groups might not demonstrateimplicit sequence knowledge because they were fasterat responding and relatively insusceptible to RT slowingwhen the novel sequence was introduced. However, theMemorizeOld group performed similarly to the amnesicpatients in overall RT during the implicit sequenceknowledge test (see Figure 4A) but, like the threeyounger control groups, exhibited no implicit sequenceknowledge. In addition, the fact that the MemorizeYounggroup and the Baseline group exhibited signiŽcant se-quence knowledge during pretraining suggests that thesensitivity of the implicit sequence knowledge test is notbeing compromised by generally fast RTs in the case ofyounger control subjects.Key FindingsIn summary, although these potential complications allseemed plausible a priori, none appears to have been asigniŽcant factor. The four Memorize groups all behavedsimilarly to each other on the implicit knowledge testfor sequence S1. All four groups also performed similarlyto the subjects in the Baseline group, who had no expe-rience with the training sequence prior to the test ofimplicit sequence knowledge. Thus, neither pretraining,interference, or baseline differences in RT had a sig-niŽcant impact on performance on the implicit knowl-edge test. Together, these four control groups addressseveral possible concerns and strengthen the Žndingthat explicit memorization training does not improveperformance on the implicit sequence knowledge test.To establish a crossover interaction between explicitand implicit memory, it is important to ensure that ex-plicit memorization training did lead to signiŽcant ex-plicit knowledge of the repeating sequence. Althoughthe performance of the four Memorize groups on therecognition memory test could conceivably have arisenfrom explicit knowledge of only the Žrst few sequenceelements (just enough to distinguish the target sequencefrom the foils), the fact that the three MemorizeYounggroups reported the majority of the sequence (mean =8.7 to 9.0 elements out of 12 possible elements) in-dicates directly that these subjects did acquire signiŽcantexplicit knowledge. In spite of this explicit knowledge,these subjects exhibited little implicit knowledge of thesequence on the subsequent SRT test. Thus, as thecrossover interaction illustrates, explicit memorizationtraining of a sequence led to explicit knowledge of thesequence without corresponding implicit knowledge.It is interesting to note that, although explicit memo-rization

training appears to lead to signiŽcant explicitknowledge of the training sequence, the CON andMemorizeOld groups did not exhibit better performancein their verbal report than the amnesic patients did(although these two control groups did perform sig-niŽcantly better on the recognition test). It seems likelythat the verbal report scores obtained by the amnesicpatients were inuenced by their implicit sequenceknowledge. For example, in several instances after train-ing, some amnesic patients denied that there was anyembedded repeating sequence but after repeated en-couragement to guess, they reported (typically by point-ing) a fair number of sequence elements (e.g., PH,session 1, Verbal report score = 8; EP, for both sessions,Verbal report score = 5). Notably, patient EP is so se-verely amnesic that he has no detectable declarativememory capacity (Hamann & Squire, 1997).It might seem surprising that subjects can acquireexplicit sequence knowledge but then not express thisknowledge on an implicit memory test for the samesequence. Indeed, several studies have reported that ex-plicit knowledge of a repeating sequence can contributeto performance on the SRT task. In one study (Willing-ham, Nissen, & Bullemer, 1989), subjects who achievedfull explicit knowledge of a repeating sequence whileperforming the SRT task exhibited signiŽcantly fasterRTs than subjects who did not achieve full explicitknowledge of the sequence. This Žnding suggests thatexplicit sequence knowledge can contribute to perfor-mance on the SRT task. Similarly, Frensch and Miner(1994, Experiment 1) found that subjects who weregiven explicit instructions to attempt to discover theembedded repeating sequence while performing theSRT task showed better sequence learning than did sub-jects who were not so instructed. An important differ-ence between these studies and ours is that the subjectsin both the Willingham et al. and Frensch and Miner(1994) studies acquired their explicit knowledge whilepracticing the SRT task. In contrast, our control subjectshad no practice on the SRT task with the training se-quence before receiving the implicit knowledge test. Wesuggest that explicit sequence knowledge can improveSRT performance only after some implicit knowledge ofthat sequence has been acquired.In another relevant study, Curran and Keele (1993,Experiment 1) explicitly told subjects the sequence oflocations the cue would follow before beginning SRTtraining. These subjects appeared to exhibit faster reac-tion times on the Žrst block of 120 SRT trials thansubjects who had not been explicitly instructed (statis-tics for this comparison are not reported). However,there were several key differences between our proce-dure and the procedure used by Curran and Keele. Inthe Curran and Keele experiment, the embedded se-quence was only six locations in length, making it mucheasier to learn compared with our twelve-location se-quence. Also, the Žrst reported RT means are derivedfrom 120-trial blocks of SRT performance. In our test ofReber and Squire 255 implicit knowledge, subjects received only 60 trials ofpractice with the training sequence followed by 60 trialsof a novel sequence. This point deserves emphasis be-cause during 120 trials of SRT practice with a six-itemsequence, it is likely that signiŽcant implicit sequencelearning occurred. Thus, the apparently faster RTs in theŽrst SRT block may have been supported by the explicitsequence knowledge as well as by implicit sequenceknowledge that was acquired during the Žrst block ofSRT practice.Howard, Mutter, and Howard (1992) also report acondition under which subjects appeared to be able toexpress implicit knowledge of a repeating sequencewithout prior SRT training. In this study, subjects whosimply observed a repeating sequence subsequently ex-hibited as much sequence knowledge on an implicit testas subjects who had actually practiced the SRT task withthe training sequence. Although these subjects were notexplicitly told to memorize the sequence (as our sub-jects were), there is some reason to suspect that theirsubjects may have acquired explicit knowledge of therepeating sequence during their extended observationperiod. In the Howard et al. experiment, subjects ob-served 30 repetitions of the sequence before starting tomake keypress responses. In addition, the sequence usedin the Howard et al. experiment was only 10 locationsin length. By contrast, our subjects exhibited signiŽcantexplicit sequence knowledge after observing a 12-itemsequence only Žve times (although our subjects wereinstructed to attempt to memorize the sequence). Wesuggest that the subjects in

the Howard et al. experimenthad ample time to develop some explicit knowledge ofthe sequence. The test of implicit sequence knowledgeused by Howard et al. gave subjects 100 trials of practicewith the training sequence, followed by 100 trials of arandom sequence. SigniŽcant learning of the simpler10-location sequence may have occurred during the 100trials of practice with the training sequence. Figure 5 ofHoward et al. supports this idea, showing that during the100 trials of SRT practice that preceded the switch to arandom sequence, response times improved substan-tially. Thus, once again, good performance on their testof implicit sequence knowledge may have been sup-ported by implicit knowledge acquired during the 100trials of SRT practice, together with the explicit se-quence knowledge acquired during observation. In eachof these studies (Curran & Keele, 1993; Frensch & Miner,1994; Howard et al. 1992; Willingham et al. 1989), itappears to be the conjunction of explicit sequenceknowledge and some implicit sequence knowledge thatenabled subjects to exhibit very rapid response timesduring the SRT task.It is worth noting that the CON group in our study,who received extensive SRT practice and also acquiredsome explicit sequence knowledge, exhibited a numeri-cally larger RT slowdown on the implicit sequence testthan the amnesic patients (although this numerical dif-ference, 38 msec vs. 58 msec, was not signiŽcant). If thisnumerical difference were reliable, it could mean thatthe CON group was able to apply some explicit knowl-edge to the implicit memory test as the result of theirSRT practice, as in Curran and Keele (1993); Frensch andMiner (1994); Howard et al. (1992); and Willingham et al.(1989). In addition, if explicit knowledge can contributeto SRT performance when subjects have also acquiredsufŽcient implicit sequence knowledge, control subjectscould exhibit numerically (but not necessarily reliably)better SRT performance than amnesic patients (e.g.,Reber & Squire, 1994).Previous research indicated that implicit learning of arepeating sequence in the SRT paradigm is intact inamnesic patients (Nissen & Bullemer, 1987; Nissen, Will-ingham, & Hartman, 1989; Reber & Squire, 1994). Thissingle dissociation suggested that implicit sequencelearning does not depend on declarative memory (whichis impaired in amnesic patients), but it did not addressthe Sensitivity hypothesis that there is a single memorysystem and that the implicit and explicit tests are simplydifferentially sensitive to group differences (Shanks & St.John 1994a). The Sensitivity hypothesis predicts that con-trol subjects should typically perform numerically betterthan amnesic patients, and it therefore provides an alter-nate account of the single dissociations reported hereand elsewhere (Nissen & Bullemer, 1987; Reber & Squire,1994), as well as studies indicating that SRT performancecan depend on both implicit and explicit sequenceknowledge (Willingham et al. 1989; Howard et al. 1992).However, the Sensitivity hypothesis cannot account forthe crossover interaction reported in the current study.We suggest that the crossover interaction observed inour study is driven by the encapsulation of explicit andimplicit sequence knowledge in separate memory sys-tems.It is notable that the crossover interaction betweenimplicit and explicit knowledge can be described asresulting from transfer-appropriate processing (Morris,Bransford, & Franks, 1977; Roediger & Blaxton, 1987;Roediger, 1990). The best performance on the tests ofimplicit and explicit sequence knowledge followed train-ing (practice or memorization) that was appropriate foreach test. One could then suppose that transfer-appro-priate processing explains the results because the con-texts for the two tests are so different that transferbetween the two tasks is quite limited. However, a trans-fer-appropriate processing account of performance isalso consistent with the multiple memory systems view(Schacter, 1990). The idea is that the separate brain re-gions supporting parallel learning and memory systemsoperate according to the transfer-appropriate processingprinciple. We suppose that an explicit representationafforded by memorization training is supported by themedial temporal lobe structures that support declarativememory. An implicit representation of sequence knowl-edge afforded by SRT practice is supported by different256 Journal of Cognitive NeuroscienceVolume 10, Number 2 brain areas (e.g., the neostriatum, supplementary motorarea, and motor cortex, see below). The key point forunderstanding the crossover interaction is that it is moredifŽcult to a

pply explicit knowledge to an implicit test(or vice versa) than to perform in conditions where thetype of test matches the training that was given. In ourexperiment, control subjects who memorized the se-quence developed an explicit representation of the re-peating sequence that supported a high level ofperformance on the recognition memory test but didnot support expression of sequence knowledge whenthis information was transferred to the implicit test. Bycontrast, the amnesic patients developed an implicit rep-resentation of the repeating sequence through SRT prac-tice without developing explicit sequence knowledge(due to their declarative memory impairment). In thecase of the patients, implicit representation of sequenceknowledge supported superior performance on the im-plicit test but did not contribute to performance on therecognition memory test. The resulting crossover inter-action suggests that there are separate representationsof sequence knowledge (in a single memory system)that support performance on both implicit and explicittests.It should be clear that the conclusion that there areseparate underlying memory systems for implicit andexplicit sequence knowledge depends not only on theŽnding of a crossover dissociation presented here butalso on previous reports of preserved learning on thistask by amnesic patients (Nissen et al. 1989; Nissen &Bullemer, 1987; Reber & Squire, 1994). The fact thatamnesic patients are selectively impaired at explicit se-quence learning suggests that this type of learning de-pends on the medial temporal lobe and that implicitsequence knowledge is supported by brain areas outsidethe medial temporal lobe. As has been noted, evaluatingthe performance of amnesic patients in isolation has notruled out the possibility that observed single dissocia-tions arise from sensitivity differences between tests ofimplicit and explicit knowledge (e.g., Shanks & St. John1994a). The crossover interaction demonstrated hereindicates that differential performance on implicit andexplicit tests does not simply arise from sensitivitydifferences. The combination of two results (selectiveimpairment in amnesic patients and a crossover interac-tion) indicates that implicit and explicit sequence learn-ing must depend on separate brain systems supportingseparate representations of sequence knowledge.A number of reports have provided converging evi-dence that SRT performance and implicit sequencelearning are at least partially supported by the neostriatalhabit learning system. Patients with neostriatal damagedue to either Huntington’s disease (HD) or Parkinson’sdisease (PD) were impaired on learning the SRT task (forHD patients, Knopman & Nissen, 1991; Willingham &Koroshetz, 1993; for PD patients, Jackson, Jackson, Harri-son, Henderson & Kennard, 1995; Pascual-Leone et al.,1994). Neuroimaging with positron emission tomogra-phy indicated that the sensorimotor cortex andneostriatum were active during SRT learning in condi-tions in which little explicit knowledge was acquired(Grafton, Hazeltine, & Ivry, 1995; Rauch et al., 1995). Bothstudies also found evidence for the involvement of dif-ferent structures when the sequence was learned ex-plicitly. In a study of motor cortex mapping usingtranscranial magnetic stimulation (TMS) at several pointsduring extended SRT practice (Pascual-Leone, Grafman,& Hallett, 1994), the extent of the cortical map and themuscle response amplitude to TMS increased as subjectsdeveloped implicit knowledge of the repeating se-quence. Thus, the Žndings from all these studies areconsistent with the involvement of a corticostriatal sys-tem during implicit learning of the SRT task.In this view, amnesic patients can demonstrate normalsequence learning because their motor cortex andneostriatum are intact. CON subjects can also learn thesequence implicitly through SRT practice, and in addi-tion, by virtue of their intact medial temporal lobe anddiencephalic structures, they are able to memorize asequence of cue locations. However, explicit sequenceknowledge is encapsulated and does not contribute toSRT performance, at least not during the early stages ofpractice. With continuing practice, control subjects maybe able to improve their performance by engaging indeclarative memory strategies (Willingham et al., 1989).The encapsulation of declarative and nondeclarativememory is also suggested by our earlier report (Reber &Squire, 1994). In that study, the control subjects acquiredmore explicit knowledge of the sequence during SRTpractice than the amnesic patients. However, bothgroups performed similarly on the SRT task of implicitmemory. That

is, the explicit memory acquired by con-trol subjects did not contribute to performance on theimplicit test.It is important to note that the independence of im-plicit and explicit memory in the SRT task demonstratedhere does not imply that implicit and explicit sequenceknowledge cannot interact. The proposal that implicitand explicit memory in the SRT task are encapsulatedrefers to the operation of separate neural substrates thatsupport the two types of memory. Yet, in many para-digms, behavior can reect the operation of both sys-tems. Thus, behavior arises from the operation of twomemory systems, each of which provides a differentkind of information, and both sources of informationcontribute to overall performance.The idea that implicit sequence learning is supportedby a corticostriatal circuit that is functionally distinctfrom the brain system supporting declarative memory isnow based on a number of different studies involvingbehavioral data from normal subjects, behavioral datafrom several neurological patient groups, and functionalneuroimaging data. The preponderance of evidencepoints to the operation of functionally and neuroana-Reber and Squire 257 tomically distinct memory systems that can operate inparallel during high-level cognitive tasks.METHODSubjectsAmnesic PatientsFive amnesic patients (four men and one woman) par-ticipated in this study. Two of the patients have alcoholicKorsakoff’s syndrome. Both have participated in quanti-tative magnetic resonance imaging (MRI) studies thatdemonstrated reductions in the volume of the mammil-lary nuclei (for RC, Squire, Amaral, & Press 1990; for NF,unpublished observations). Patient NF also has bilateralreduction in the size of the hippocampal formation. Theremaining three patients have bilateral hippocampal for-mation damage (for PH, Polich & Squire, 1993; for EP,Squire & Knowlton, 1995; for LJ, unpublished observa-tions). Patient PH had a history of 1- to 2-min attacks (ofpossible epileptic origin) in association with gastricsymptoms and transient memory impairment. In 1989 hesuffered a series of small attacks that resulted in markedand persisting memory impairment. Patient EP devel-oped profound anterograde and retrograde amnesia in1992 after herpes simplex encephalitis. Patient LJ be-came amnesic gradually in 1988 to 1989 without anyknown precipitating event. Her memory impairment hasremained stable since that time.The patients averaged 69 years of age at the beginningof the study and had an average of 12.8 years of educa-tion. Immediate and delayed (12 min) prose recall aver-aged 3.2 and 0.0 segments, respectively (Gilbert, Levee,& Catalano, 1968; maximum score = 21). Scores on othermemory tests appear in Tables 2 and 3. The mean scoreon the Dementia Rating Scale (DRS) was 127.6 (Mattis,1976; maximum score = 144). Most of the points lost onthe DRS were from the memory subportion of the test(mean points lost = 10.6). The mean score for the BostonNaming Test was 53.4 (Kaplan, Goodglass, & Weintraub,1983; maximum score = 60). Scores for healthy subjectson these tests can be found elsewhere (Janowsky, Shima-mura, & Squire, 1989; Squire et al. 1990).CON SubjectsThe control subjects (26 men and 44 women) wereeither employees or volunteers at the San Diego VeteransAffairs Medical Center or were recruited from the retire-ment community of the University of California at SanDiego. The control subjects were assigned to six separategroups (Table 3). Two of the control subject groups(CON, n = 10; MemorizeOld, n = 15) were matched to theamnesic patients with respect to the mean and range oftheir ages, years of education, and scores on the Infor-mation and Vocabulary subtests of the Wechsler AdultIntelligence Scale-Revised (WAIS-R). These two groupsaveraged 66.6 years of age, 14.2 years of education, and20.5 and 52.8 on the Information and Vocabulary sub-tests, respectively (amnesic patients = 19.7 and 51.3,respectively). Immediate and delayed prose recall aver-aged 7.5 and 6.2 segments, respectively.Four groups of younger control subjects were alsotested (Table 4): MemorizeYoung, (n = 15) MemorizeYoungRandom Pretraining (n = 10), MemorizeYoung No Pretrain-ing (n = 10), and Baseline (n = 10). These four groupsaveraged 26.2 years of age (range of group means = 25.2to 27.1 years) and 16.8 years of education (range ofgroup means = 16.4 to 17.3 years). Table 2. Characteristics of Amnesic Patients WMS-R Patient Lesion Age(years) WAIS-RIQ Attention Verbal Visual General DelayPHHF74120 117 678370 57LJHF58 98 105 83606950RCaDien79106 115 769780 72NFaDien60 94 91 62735350EPHF74103 94 578261 56Means69104.2104.4697

9 66.7 57 Note: WAIS-R = Wechsler Adult Intelligence Scale-Revised; WMS-R = Wechsler Memory Scale-Revised. HF = Hippocampal formation; Dien =Diencephalon. The WAIS-R and the WMS-R indices yield a mean score of 100 in the normal population with a standard deviation of 15. TheWMS-R does not provide scores for subjects who score below 50. Therefore, the two scores below 50 were scored as 50 for calculating agroup mean.a These patients have alcoholic Korsakoff’s syndrome. NF has reduced volume of the hippocampal formation in addition to a diencephaliclesion.258 Journal of Cognitive NeuroscienceVolume 10, Number 2 MaterialsThe SRT TaskThe SRT task was presented on an EPSON 650 colorlaptop computer (10.4-in dual-scan LCD screen). Fourdashes (0.5 cm in width and 4.0 cm apart) appearedcontinuously 1.0 cm from the bottom of the screen todenote the four possible locations of the cue. The cuewas an asterisk 0.4 cm wide that could appear 5.3 cmabove any one of the four dashes. Subjects were in-structed, “When the asterisk appears, press the key un-derneath it as quickly as you can.” Responses were madeon the computer keyboard using the four keys directlybeneath the dashes: c, b, m, and the period key. Thesekeys were marked with white stickers to indicate wherethe subjects should place their Žngers during the task.Subjects used two Žngers from each hand, usually theŽrst and second digits, and were instructed to maintaincontact with the four marked keys throughout training.A correct keypress caused the asterisk to disappear and Table 3. Memory Test Performance Patient Diagramrecall Paired associates Wordrecall (%) Wordrecognition (%) 50words 50facesPH300127843634LJ300040933329RC300319853730NF400236762827EP000024652428Means 2.6001.229.280.631.629.6Control means(n = 8)20.66.07.68.971.097.041.138.1 Note: The diagram recall score is based on delayed (12-min) reproduction of the Rey-Osterrieth Žgure (Osterrieth, 1944; maximum score = 36).The average score for the amnesic patients for copying the Žgure was 27.6, a normal score (Kritchevsky et al. 1988). The paired associatescores are the number of word pairs recalled on three successive trials (maximum score = 10 per trial). The word recall score is the percent-age of words recalled across Žve successive study-test trials. The word recognition score is the percentage of words identiŽed correctly byyes/no recognition across Žve consecutive study-test trials. The score for words and faces is based on a 24-hr delayed recognition test of 50words or 50 faces (modiŽed from Warrington, 1984; maximum score = 50, chance = 25). The mean scores for control subjects shown for thesetests are from Squire and Shimamura (1986). Table 4. Experimental Procedure Group Pretraining Training Explicit test ImplicittestAmnesic patients (n = 5)—Practice 20 blocks S1Verbal report, recognize S1S1 N1Control Subjects (n = 10)—Practice 20 blocks S1Verbal report, recognize S1S1 N1MemorizeOld (n = 15)S2 S2 S2 N2Memorize 1 block S1Verbal report, recognize S1S1 N1MemorizeYoung (n = 15)S2 S2 S2 N2Memorize 1 block S1Verbal report, recognize S1S1 N1MemorizeYoungRandom Pretraining(n = 10)R1 R2 R3 R4Memorize 1 block S1Verbal report, recognize S1S1 N1MemorizeYoungNo Pretraining(n = 10)—Memorize 1 block S1Verbal report, recognize S1S1 N1Baseline (n = 15)S2 S2 S2 N2——S1 N1 Note: S1 represents one 60-trial block of the training sequence (124313214234). N1 represents one 60-trial block of a novel sequence(343213142412). S2 represents one 60-trial block of a second pretraining sequence (232413143421). N2 represents one 60-trial block of a sec-ond novel sequence (432423121413). Each 60-trial block contains Žve repetitions of the sequence. R1–4 represent four 60-trial blocks of a bal-anced, random sequence of cue locations (see text). The amnesic patients were tested twice with the procedure shown, with an interval of 3to 5 months between tests. The control subjects were also tested twice, with an interval of 9 to 10 months between tests. The four groups ofcontrol subjects who memorized the repeating sequence are grouped together by horizontal lines. These groups varied either in mean age orthe type of SRT pretraining they received before attempting to memorize the repeating sequence.Reber and Squire 259 then reappear in a new location after a 250-msec delay.The cue never appeared in the same location on succes-sive trials. An incorrect response resulted in the com-puter beeping once, and the next trial started only aftera correct response had been made. The task was admin-istered in 60-trial blocks with a 15-sec break followingeach block.Repeating Sequences and Random S

equencesEach 60-trial block of the SRT task contained Žve repe-titions of a 12-location sequence (except as noted belowfor pretraining of the MemorizeYoung Random Pretraininggroup). Each 12-location sequence contained three oc-currences of each of the 4 possible cue locations andone occurrence of each of the 12 possible transitionsbetween locations (e.g., 12, 13, 14, 21, 23, etc.). Thus,each sequence was balanced (or “ambiguous”; Curran &Keele, 1993). With a balanced sequence, subjects cannotpredict the next location of the cue simply by learninglocation or transition frequencies. They must learn sec-ond-order conditional (SOC) associations (Reed &Johnson, 1994). That is, in order to predict the next cuelocation, they must know the two immediately preced-ing cue locations. Reaction times typically decrease withpractice on the SRT task both because the repeatingsequence is being learned and because of nonspeciŽceffects of practicing the task. SpeciŽc knowledge of therepeating sequence was assessed by changing the se-quence of cue locations so that they followed a novel,balanced sequence. The increase in reaction time pro-duced by changing the sequence indicates the amountof sequence-speciŽc learning that has occurred. The se-quences were S1(124313214234), S2 (232413143421),N1(343213142412), and N2 (432423121413). Every sub-ject received the same training or pretraining sequence(S1, S2) to avoid introducing any additional variability inperformance that might have arisen from differences inthe rate at which the sequences could be learned. Forone group (MemorizeYoung Random Pretraining), pre-training on the SRT task followed a pseudorandom se-quence of 240 locations in four 60-trial blocks(R1R2R3R4). This sequence of locations did not repeat,but each location and transition occurred about equallyoften within each block.ProcedureAmnesic patients and six control groups receivedsome or all of the Žve tasks described below (also seeTable 3).TasksSRT Pretraining and Training. SRT pretraining con-sisted of four 60-trial SRT blocks in which subjectspressed the key beneath the cue as the cue movedamong the four possible locations. Pretraining was givento four control groups (MemorizeOld, MemorizeYoung,MemorizeYoung Random Pretraining, and Baseline) in or-der to familiarize them with the SRT task and to permitthe learning of nonspeciŽc task knowledge prior to thepresentation of the training sequence and prior to thetest of implicit knowledge for the training sequence. Forthree of these groups (MemorizeOld, MemorizeYoung, andBaseline), the Žrst three blocks of pretraining containeda repeating sequence (S2), and the Žnal block of pretrain-ing introduced a novel, repeating sequence (N2). A sig-niŽcant increase in RT for this Žnal block (N2) woulddemonstrate that subjects can express signiŽcant implicitknowledge of the pretraining sequence (S2). That is, anincrease in RT during block N2, compared with the pre-ceding block of S2, would show that nonspeciŽc SRTlearning (during the N2 block) was not preventing theexpression of implicit sequence knowledge. Moreover, inthe subsequent test of implicit sequence knowledge forthe training sequence (S1) the masking effect of nonspe-ciŽc SRT skill learning could not be an explanation for afailure to observe implicit sequence knowledge.For the fourth group (MemorizeYoung Random Pretrain-ing), pretraining consisted of four 60-trial blocks of apseudorandom sequence. This group could acquire non-speciŽc task knowledge but did not learn a speciŽcsequence during pretraining.Two groups received SRT training (amnesic patientsand CON group). SRT training consisted of practicingtwenty 60-trial blocks (1200 trials). Each training blockcontained Žve repetitions of the 12-location, repeatingsequence (S1).Explicit Sequence Memorization Training. Subjectswere instructed to watch the computer screen and at-tempt to memorize the sequence of locations of thetarget sequence (S1). Each cue appeared on the screenfor 750 msec with a 250-msec intertrial interval. No keyswere pressed by the subject during this task. After onepresentation of the 12-location sequence, there was ashort pause while subjects were instructed that theywould be shown the sequence again and they shouldagain attempt to memorize it. This procedure wasrepeated until subjects had observed the 12-location se-quence Žve times. It was hypothesized that memoriza-tion of the sequence, without any SRT experience withthe sequence, would lead to explicit knowledge of thetraining sequence without implicit sequence knowledge.Verbal Report. After completing SRT training, subjectswere asked to attempt

to report the repeating sequenceeither verbally or by pointing in turn to the appropriatecue locations on the computer screen. Subjects wereprompted to attempt to report the sequence even ifthey were not aware of its existence. Prompting contin-ued until subjects reported at least eight locations (evenif they had to guess). Responses were analyzed to deter-260 Journal of Cognitive NeuroscienceVolume 10, Number 2 mine the longest section of the repeating sequence thatwas contained in the verbal report.Recognition Memory Test. The recognition test waspresented in the same manner as the SRT task (i.e.,subjects pressed the key underneath the cue as soon asit appeared on the screen), but subjects were additionallyinstructed that after responding to 12 appearances of thecue, they would be asked to rate, on a 0 to 100 scale,whether they thought they had seen that same sequenceof cue locations in the immediately preceding trainingphase of the experiment. Five different 12-item se-quences were administered, and subjects made a 0 to100 rating after each sequence. The training sequence(S1) was always either the second or fourth of the Žvesequences. The recognition test score was calculated asthe rating given to the target sequence (S1) minus themean rating given to the other four sequences. The rec-ognition memory test assessed subjects explicit knowl-edge of the training sequence.Implicit Test of Sequence Knowledge. Two Žnal 60-trialblocks were given to each group in order to assessimplicit sequence knowledge for the training sequence(S1). In the Žrst of these blocks, a 60-trial block was giventhat contained Žve repetitions of the 12-location trainingsequence (S1). The starting position for the sequence wasthe same as the starting position used for SRT trainingand explicit sequence memorization training. The second60-trial block consisted of Žve repetitions of a 12-locationnovel sequence (N1). Subjects were instructed to re-spond to each appearance of the asterisk by pressing thekey underneath it as quickly as possible. No mention wasmade of any repeating sequence. Implicit knowledge ofthe training sequence was measured by the slower meanreaction time that occurred during the second SRT block(N1) compared with the Žrst SRT block (S1) (i.e., thedifference in mean RT between these two blocks).Subject GroupsAmnesic Patients. The Žve amnesic patients receivedtwenty 60-trial blocks of SRT training with the targetsequence S1. This training was followed by verbal report,the recognition test, and then by the implicit sequenceknowledge test. The amnesic patients repeated the sameprocedure (with the same training sequence) in a sepa-rate session 3 to 5 months later.CON Subjects. The CON group (n = 9) received thesame procedure as the amnesic patients, with the twotesting sessions separated by 9 to 10 months.MemorizeOld. The MemorizeOld group (n = 15) receivedfour 60-trial blocks of pretraining with the SRT task. TheŽrst three of these blocks each contained Žve repetitionsof the 12-location, pretraining sequence (S2). The fourthblock contained Žve repetitions of a novel sequence (N2).Pretraining was followed by explicit sequence memori-zation training with sequence S1, verbal report, the rec-ognition memory test, and the implicit sequenceknowledge test.MemorizeYoung. The MemorizeYoung group (n = 15) wastested in exactly the same way as the MemorizeOld group.These subjects were younger than both the amnesicpatients and the subjects in the CON and theMemorizeOld groups. For both the MemorizeOld andMemorizeYoung groups, the pretraining gave these sub-jects enough experience with the SRT task to acquirenonspeciŽc task knowledge and enough experience toexpress signiŽcant knowledge of sequence S2 (as dem-onstrated by slower RTs when the novel N2 sequencewas introduced). Because pretraining with repeating se-quence S2 could potentially have interfered with theability later to express implicit knowledge of the memo-rized sequence S1, two additional groups were givenpretraining either with random cue locations(MemorizeYoung Random Pretraining) or no pretraining atall (MemorizeYoung No Pretraining).MemorizeYoung Random Pretraining. The MemorizeYoungRandom Pretraining group (n = 10) was given SRT pre-training in which the cue followed a pseudorandomsequence of locations for four 60-trial blocks (R1R2R3R4).This sequence of locations did not contain an embeddedrepeating sequence, but each location and transition oc-curred about equally often. Pretraining was followed byexplicit sequence memorization training with sequenceS1, verbal report, the recognition memory test, andŽnally the implicit

sequence knowledge test. The proce-dure for this group differed from the MemorizeOld andMemorizeYoung procedure only with respect to pretrain-ing. Because the pretraining for the MemorizeYoung Ran-dom Pretraining group did not involve a repeatingsequence, the pretraining should not have interferedwith this group’s ability to express implicit sequenceknowledge of the training sequence after memorizing it.MemorizeYoung No Pretraining. The MemorizeYoung NoPretraining group (n = 10) was given explicit sequencememorization training with sequence S1, verbal report,the recognition memory test, and then the implicit se-quence knowledge test. The performance of this groupon the implicit sequence knowledge test, in comparisonwith the MemorizeOld, MemorizeYoung, and MemorizeYoungRandom Pretraining groups, provides a way to determinewhether pretraining had any effect on performing theimplicit sequence knowledge test.Baseline. The Baseline group (n = 15) was given thesame pretraining as the MemorizeOld and MemorizeYounggroups but was then given the implicit sequence knowl-edge test without any previous experience with theReber and Squire 261 training sequence (S1). This group provided an estimateof baseline performance on the implicit sequence knowl-edge test in the absence of any opportunity to acquireeither implicit or explicit knowledge of the training se-quence.AcknowledgmentsThis research was supported by the Medical Research Serviceof the Department of Veterans Affairs and by National Instituteof Mental Health Grants MH24600 (Larry R. Squire) and F32MH11150-01A1 (Paul J. Reber). We thank James Moore andJoyce Zouzounis for research assistance.Reprint requests should be sent to Larry Squire, Veterans AffairsMedical Center 116A, 3350 La Jolla Village Drive, San Diego, CA92161, or via e-mail: lsquire@ucsd.edu.REFERENCESCurran, T., & Keele, S. (1993). Attentional and nonattentionalforms of sequence learning. Journal of Experimental Psy- chology: Learning, Memory, and Cognition, 19, 189–202. Dulany, D. E., Carlson, R. A., & Dewey, G. I. (1985). A case ofsyntactical learning and judgment: How conscious andhow abstract? Journal of Experimental Psychology: Gen- eral, 113, 541–555. Frensch, P. A., & Miner, C. S. (1994). Effects of presentationrate and individual differences in short-term memory ca-pacity on an indirect measure of serial learning. Memory & Cognition, 22, 96–110. Gilbert, J., Levee, R., & Catalano, K. (1968). A preliminary re-port on a new memory scale. Perceptual and Motor Skills, 27, 277–278. Grafton, S. T., Hazeltine, E., & Ivry, R. (1995). Functional map-ping of sequence learning in normal humans. Journal of Cognitive Neuroscience, 7, 497–510. Hamann, S. B., & Squire, L. R. (1997). Intact perceptual mem-ory in the absence of conscious memory. Behavioral Neu- roscience, 111, 850–854. Howard, J. H., Mutter, S. A., & Howard, D. V. (1992). Serial pat-tern learning by event observation. Journal of Experimen- tal Psychology: Learning, Memory, and Cognition, 18, 1029–1039. Jackson, G. M., Jackson, S. R., Harrison, J., Henderson, L., &Kennard, C. (1995). Serial reaction time learning and Park-inson’s disease: Evidence for a procedural learning deŽcit. Neuropsychologia, 33, 577–593. Janowsky, J. S., Shimamura, A. P., & Squire, L. R. (1989). Sourcememory, impairment in patients with frontal lobe lesions. Neuropsychologia, 27, 1043–1056. Kaplan, E. F., Goodglass, H., & Weintraub, S. (1983). The Bos-ton Naming Test. Philadelphia: Lea Febiger.Knopman, D. S., & Nissen, M. J. (1991). Procedural learning isimpaired in Huntington’s disease: Evidence from the serialreaction time task. Neuropsychologia, 29, 245–254. Knowlton, B. J., Mangels, J. A., & Squire, L. R. (1996). Aneostriatal habit learning system in humans. Science, 273, 1399–1402. Knowlton, B. J., Ramus, S. J., & Squire, L. R. (1992). Intact ar-tiŽcial grammar learning in amnesia: Dissociation of clas-siŽcation learning and explicit memory for speciŽcinstances. Psychological Science, 3, 172–179.Knowlton, B. J., & Squire, L. R. (1994). The information ac-quired during artiŽcial grammar learning. Journal of Ex- perimental Psychology: Learning, Memory, and Cogni- tion, 20, 79–91. Knowlton, B. J., Squire, L. R., & Gluck, M. (1994). ProbabilisticclassiŽcation learning in amnesia. Learning and Memory,1, 106–120.Kritchevsky, M., Squire, L. R., & Zouzounis, J. A. (1988). Tran-sient global amnesia: Characterization of anterograde andretrograde amnesia. Neurology, 38, 213–219. Mattis, S. (1976). Dementia Rating Scale. In R. Bellack &B. Keraso (Eds.), Geriatric psychiatry (pp.

77–121). NewYork: Grune and Stratton.Morris, C. D., Bransford, J. D., & Franks, J. J. (1977). Levels ofprocessing versus transfer appropriate processing. Journal of Verbal Learning and Verbal Behavior, 16, 519–533. Nissen, M. J., & Bullemer, P. (1987). Attentional requirementsof learning: Evidence from performance measures. Cogni- tive Psychology, 19, 1–32. Nissen, M. J., Willingham, D., & Hartman, M. (1989). Explicitand implicit remembering: When is learning preserved inamnesia? Neuropsychologia, 27, 341–352. Osterrieth, P. A. (1944). Le test de copie d’une Žgure com-plexe [The test of copying a complex Žgure]. Archives dePsychologie, 30, 206–356.Pascual-Leone, A., Grafman, J., Clark, K., Stewart, M., Mas-saquoi, S., Lou, J., & Hallett, M. (1994). Procedural learningin Parkinson’s disease and cerebellar degeneration. Annals of Neurology, 34, 594–602. Pascual-Leone, A., Grafman, J., & Hallett, M. (1994). Modula-tion of cortical motor output maps during developmentof implicit and explicit knowledge. Science, 263, 1287– 1289. Perruchet, P., & Gallego, J. (1993). Association between con-scious knowledge and performance in normal subjects: Re-ply to Cohen and Curran (1993) and Willingham, Greely,and Bardone (1993). Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 1438–1444. Polich, J., & Squire, L. R. (1993). P300 from amnesic patientswith bilateral hippocampal lesions. EEG Clinical Neuropsy-chology, 86, 408–417.Rauch, S. L., Savage, C. R., Brown, H. D., Curran, T., Alpert,N. M., Kendrick, A., Fischman, A. J., & Kosslyn, S. M. (1995).A PET investigation of implicit and explicit sequence learn-ing. Human Brain Mapping, 3, 271–286. Reber, P. J., & Squire, L. R. (1994). Parallel brain systems forlearning with and without awareness. Learning and Mem-ory, 2, 1–13.Reed, J., & Johnson, P. (1994). Assessing implicit learningwith indirect tests: Determining what is learned about se-quence structure. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 585–594. Roediger, H. L., III (1990). Implicit memory: Retention with-out remembering. American Psychologist, 45, 1043–1056. Roediger, H. L., III, & Blaxton, T. A. (1987). Effects of varyingmodality, surface features and retention interval on prim-ing in word-fragment completion. Memory and Cogni- tion, 15, 379–388. Schacter, D. L. (1987). Implicit memory: History and currentstatus. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 501–518. Schacter, D. L. (1990). Toward a cognitive neuropsychologyof awareness: Implicit knowledge and anosognosia. Jour-nal of Clinical and Experimental Neuropsychology, 12,155–178.Shanks, D. R., & St. John, M. F. (1994a). Characteristics of dis-sociable human learning systems. Behavioral and BrainSciences, 17, 367–447.262 Journal of Cognitive NeuroscienceVolume 10, Number 2 Shanks, D. R., & St. John, M. F. (1994b). How should implicitlearning be characterized? Behavioral and Brain Sci-ences, 17, 427–447.Squire, L. R. (1992). Memory and the hippocampus: A synthe-sis from Žndings with rats, monkeys, and humans. Psycho- logical Review, 99, 195–231. Squire, L. R., Amaral, D. G., & Press, G. A. (1990). Magneticresonance measurements of hippocampal formation andmammillary nuclei distinguish medial temporal lobe and di-encephalic amnesia. Journal of Neuroscience, 10, 3106– 3117. Squire, L. R., & Knowlton, B. J. (1995). Learning about catego-ries in the absence of memory. Proceedings of the Na- tional Academy of Sciences, 92, 12470–12474. Squire, L. R., & Shimamura, A. P. (1986). Characterizing amne-sic patients for neurobehavioral study. Behavioral Neuro- science, 100, 866–877. Squire, L. R., & Zola, S. M. (1996). Structure and function ofdeclarative and nondeclarative memory systems. Proceed- ings of the National Academy of Sciences, USA, 93, 13515–13522. Tulving, E. (1985). How many memory systems are there? American Psychologist, 40, 385–398. Warrington, E. K. (1984). Recognition Memory Test. Windsor,England: FER-Nelson.Willingham, D. B., & Koroshetz, W. J. (1993). Evidence for dis-sociable motor skills in Huntington’s disease patients. Psy-chobiology, 21, 173–182.Willingham, D. B., Nissen, M. J., & Bullemer, P. (1989). On thedevelopment of procedural knowledge. Journal of Experi- mental Psychology: Learning, Memory, and Cognition, 15, 1047–1060. Weiskrantz, L. (1991). Problems of learning and memory. Oneor multiple memory systems? In J. R. Krebs & G. Horn(Eds.), Behavioral and neural aspects of learning andmemory (pp. 1–10). Oxford: Clarendon.Reber and Squire