Psychonomic Bulletin & Review 2000, 7 (2), 354-359

Direct versus indirect tests ofmemory: Directed forgetting meets the generation effect

COUN M. MAcLEOD and KAREN A. DANIELS University ofToronto, Scarborough, Ontario, Canada

Subjects read 20 words and generated 20 others from defInitions during a 40-item study phase. Pro­ duction ofeach word was followed by an instruction to remember or to forget that word. In free recall, a direct test ofmemory, words that had been generated were recalled much betterthan words that had been read. The remember-forget instructional manipulation affected read words but not generated words. In speeded word reading, an indirect test of memory, all studied words showed priming, but read words showed more priming than generated words. Here, the effect of remember versus forget instructions appeared only for generated words. These dissociations ofa direct and an indirect test in­ dicate that two powerful encoding manipulations affect separable processes to which these tests are differentially sensitive.

The prevalent lay view ofremembering rests on the con­ have come to be called priming effects. As reviewed by scious recollection ofpast episodes. Yet even over 100 years Richardson-Klavehn and Bjork, the combined evidence ago, in his ground-breaking empirical studies, Ebbinghaus from amnesic and normal subjects reveals a number of (1885/1964) distinguished two ways ofremembering: "in­ trends, of which we will highlight two. First, direct tests voluntary recollection" and "voluntary recollection." This generally are more influenced than are indirect tests by se­ distinction can be seen as a precursorto the current concepts mantic elaboration during encoding (Graf & Mandler, ofimplicit and explicit memory (Graf& Schacter, 1985), as 1984; Jacoby & Dallas, 1981), although exceptions do differentiated by two classes ofmemory tests. Indirect tests exist (Graf& Schacter, 1985). Second, a shift in modality reveal the influence ofmemory through facilitation ofper­ from study phase to test phase (e.g., from auditory to vi­ formance in the absence ofconscious recollection; in con­ sual) has little effect on direct measures yet sharply atten­ trast, direct tests require the conscious recollection ofpre­ uates priming on indirect measures (Bassili, Smith, & vious information. As illustrations, intentionally trying to MacLeod, 1989; Roediger & Blaxton, 1987), although ex­ recollect references that we wish to cite in an article consti­ ceptions exist here as well (Jacoby & Dallas, 1981). tutes a direct use of memory, whereas fluently using the It should be noted, however, that despite such strikingdis­ many features ofour word processors without realizing that sociations between direct and indirect tests, there also are memory is being tapped is an indirect use ofmemory. significant parallel effects. As illustrations, repetition of A variety ofdissociations in performance have been ob­ stimuli appears to facilitate performance on both types of served between direct and indirect tests, suggesting that test (Jacoby & Dallas, 1981; but see Challis & Sidhu, 1993), these two ways of using memory are quite different (see and same-context responses are significantly more effec­ Richardson-Klavehn & Bjork, 1988; Roediger & McDer­ tive than are different-context responses in aiding reten­ mott, 1993, for reviews). Studies of amnesia have made tion on both types oftest (Graf& Schacter, 1985). especially important contributions. Despite their frequently Roediger (1990) has offered a transfer-appropriate pro­ catastrophic failure on direct tests ofmemory, amnesics cessing account ofthe pattern ofresults seen on direct ver­ can show normal performance on indirect tests ofmemory sus indirect tests (see also MacLeod & Masson, 1997, in (Graf& Schacter, 1985; Graf, Shimamura,& Squire, 1985; press; Masson & MacLeod, 1992). Here, indirect tests are Warrington & Weiskrantz, 1970). These unconscious ef­ seen as relying on the fluent replaying ofthe encoding of fects in indirect tests-usually expressed as a benefit con­ stimuli, typically emphasizing perceptual components of ferred on subsequent processing by previous processing- their initial analysis. Direct tests typically rely more on any elaborative encoding done during study, emphasizing se­ mantic/conceptual processing that went beyond simply an­ alyzing the stimulus in context. Thus, a modality change This research was supported by Natural Sciences and Engineering Re­ search Council ofCanada Grant A7459, and formed the senior honors from study to test should disrupt recovery of the initial thesis ofK.A.D., who is now at the Georgia Institute ofTechnology. We perceptual analysis, affecting primarily indirect tests, thank Monica Ghosal and Shelley Hodder for their assistance in col­ whereas semantic elaboration at encoding should enhance lecting the data, and Barbara Basden, John Gardiner, Pierre Perruchet, conceptual processing, affecting primarily direct tests, as and Henry Roediger III, for their helpful comments during the review process. Correspondence should be addressed to C. M. MacLeod, Divi­ has been shown repeatedly.l sion of Life Sciences, University of Toronto, Scarborough, ON, MIC Our goal was to explore the direct-indirect test distinc­ IA4 Canada (e-mail: [email protected]). tion further by manipulating two powerful encoding vari-

Copyright 2000 Psychonomic Society, Inc. 354 DIRECTED FORGETTING AND GENERATION 355 abies: directed forgetting and generation versus reading. that required subjects to provide words in response to de­ These variables are particularly ofinterest because they per­ finitions-presumably a more conceptual indirect test­ mit us to investigate the separate effects ofinitial percep­ although only under the item method. Thus, the most gen­ tual/conceptual encoding and of subsequent rehearsal on eral conclusion emerging now from the directed forgetting direct versus indirect tests. In what follows, we will con­ literature is that there ordinarily is not an R-F difference on sider the literature relating to each ofthese issues in tum indirect tests ofmemory, although as yet there are only a before describing in detail the purpose ofour study. few relevant studies. It also appears to be the case that in­ direct tests can be very sensitive to what otherwise might Directed Forgetting seem to be minor changes over experiments. Over the past 30 years, it has been well established that Two other points regarding directed forgetting studies stimuli accompanied by an instruction to forget ordinarily are especially relevant. First, as Basden et al. (1993) have show reduced retention relative to those accompanied by argued (see also MacLeod, 1998, 1999), the operative mech­ an instruction to remember (for recent reviews, see John­ anism in directed forgetting appears to differ as a function son, 1994; MacLeod, 1998). Known as directed forget­ of the method of delivery of the instructions. In the list ting, this phenomenon can be seen as a proper subset of method, F words appear to be encoded as well as R words, the more general phenomenon of intentional forgetting, but inhibited from retrieval. In contrast, in the item method, which spans social psychology, psychology and the law, F words appear to be rehearsed less and hence are not as animal cognition, and a variety ofother domains (see the well encoded. In our study, we used the item procedure recent book edited by Golding & MacLeod, 1998). Yet by because we wished to focus on directed forgetting due to far the majority ofthe directed forgetting work in the mem­ selective rehearsal. ory literature has exclusively examined performance on Second, in a study by Golding, Long, and MacLeod direct tests of memory, primarily recall and recognition. (1994) that examined recall only, we found that the ability It is worth noting at the outset that directed forgetting to forget information on cue was a function of the relat­ can be realized in either of two ways. Under the item edness ofstudied items. Subjects studied a list containing method, each individual item presented for study is as­ some semantically related successive words (e.g., crab­ signed its own instruction, usually displayed just after the leg) and some unrelated successive words (e.g., cheese­ item. Under the list method, a single instruction is pre­ fire, constructed from the related pairs cheese-cake and sented after each halfofthe list ofitems, with one halfof camp-fire). Using the item method, successive words the list to be forgotten and the other half to be remem­ sometimes received the same instruction, sometimes dif­ bered. The item method produces a larger remember­ ferent instructions. Although recall was best when both forget (R-F) difference than does the list method in both words in a pair were given a remember cue (i.e., the R-R recall and recognition; indeed, the list method generally condition), we also found good recall ofsemantically re­ does not produce a reliable R-F difference in recognition lated word pairs in the R-F condition, much better than (see, e.g., MacLeod, 1999). when the first word ofthe pair was an F word. We inter­ Does directed forgetting affect performance on indirect preted this finding as suggesting that compliance with an tests as well? The picture is not clear: The answer seems instruction to forget a word is difficult when that word im­ to depend both on the particular indirect test employed mediately follows an R word to which it is semantically and the method ofdelivering the instructions to remember related. The elaborative, conceptual processing already versus forget, but only a very narrow set ofconditions has completed for the R word by the time the F word appears so far resulted in observation of the effect. Using word undermines the ability to differentially encode the F word fragment completion and repetition priming in lexical de­ and the R word. Perhaps, then, on a direct test, any study cision as the indirect tests, MacLeod's (1989) initial find­ task that encouraged elaborative encoding could defeat an ings with the item method indicated differential priming instruction to forget by making postencoding rehearsal favoring words to be remembered (R words) over words to differences inoperative. be forgotten (F words), the same pattern as on direct tests. Despite replicating MacLeod's (1989) finding with word Generation Versus Reading fragment completion, however, Paller (1990) found no re­ On a direct recognition test, Jacoby and Dallas (1981) liable R-F difference using word stem completion under found that words generated from anagrams during study the item method. Also using the item method, Basden, Bas­ were better recognized than words that were simply read. den, and Gargano (1993, Experiment 4) replicated Mac­ In contrast, on an indirect masked word identification test Leod's (1989) finding, but convincingly showed that even where subjects were to identify words appearing on the on a word fragment completion test, the effect was re­ screen for 30 msec prior to a mask, the opposite pattern stricted to a narrow range ofconditions. appeared: There was more priming for words that were Using the list method, neither E. L. Bjork and R. A. read than for words that were generated. Also using a Bjork (1996, Experiment 2) nor Basden et al. (1993) found masked word identification test, Weldon (1991, Experi­ differential priming for R words and F words on a word ment 1) supported this finding, aligning it with the well­ fragment completion test. Basden and Basden (1996, Ex­ established pattern in word fragment completion (see periment 2) did find an R-F difference on an indirect test Roediger & McDermott, 1993, for a review). 356 MACLEOD AND DANIELS

Since then, Masson and MacLeod (1992) and MacLeod might be possible to observe effects of differential re­ and Masson (1997) have confirmed the reliable priming for hearsal, so we anticipated that there could be an R-F dif­ generated words on the masked word identification test. ference in priming on the indirect speeded reading test for More recently, MacLeod and Masson (in press) have ex­ the generated words only. tended this to another indirect measure, speeded word reading. Here, subjects simply read aloud a fully exposed METHOD test word as quickly as they can into a microphone. Some­ times called "naming" (see, e.g., Balota & Chumbley, 1984), Subjects speeded word reading is an especially good candidate as The subjects were 89 undergraduates from introductory psychol­ an indirect measure of memory because the task is per­ ogy at the University of Toronto at Scarborough. They were re­ formed very rapidly-indeed virtually automatically. This cruited on a voluntary basis and received bonus points for their par­ ticipation. Fifty-four subjects took part in the speeded reading task. argument is buttressed by the well-known Stroop (1935) The data from 4 ofthese subjects were discarded due to microphone effect, where color words cannot be ignored when the task malfunction, leaving 50 in this condition.2 The remaining 35 sub­ is to name the ink colors in which they are printed. Word jects took part in the free-recall task only. reading should not suffer the intrusion of conscious rec­ ollection-the bane of indirect tests-because it is ex­ Materials tremely fluent and easy and does not require any of the The 60 stimuli were the definitions and target words used by Wel­ problem-solving types of activities required of indirect don (1991) and presented in the appendix to MacLeod and Masson (1997). An example would be "a large, triangular structure in tests such as word fragment completion, where the test Egypt-p?" and the corresponding target "pyramid." The program stimulus is quite degraded. randomly assigned the words to the generate, read, or new condition for each subject. An additional 12 words were chosen to serve as Rationale for the Present Experiment buffers in the study phase and the reading task. The buffer words This experiment brings together all of the factors that matched those in Weldon's list according to frequency ofusage and we have just set out, and addresses two primary questions. word length, and were selected from the Paivio, Yuille, and Madigan (\968) norms. The first relates to direct tests ofmemory. As already de­ scribed, the generation effect and directed forgetting Apparatus strongly affect recall and recognition. Yet there is evidence Subjects were tested using IBM-compatible 286 computers that, for words, directed forgetting cannot overcome an al­ equipped with Tatung CM-1496 14-in. VGA color monitors. When ready strong elaborative encoding, at least in the case of the subject spoke into a Realistic Highball-7 microphone, the signal preexisting semantic associations (Golding et a1., 1994; was amplified by a Realistic SA-ISO stereo amplifier and input to the computer via a modified keyboard that permitted millisecond but see Basden & Basden, 1996, Experiment 1, for a dif­ accuracy response timing using the routines taken from Graves and ferent result with pictures). Would this also be true for Bradley (\987, 1988). All programs were written in QuickBASIC new episodic elaborative encodings, as in the case ofgen­ 4.5. The screen background color was white (palate no. IS); the erated words (but not read words)? We expected recall stimuli were presented in black (palate no. 0). performance on R words to be better than that on F words when those words were read at study, consistent with the Study Procedure considerable evidence for selective rehearsal of the R Each subject studied 40 words, with the remaining 20 words in the set reserved to provide new, unstudied baseline items in the reading words (see MacLeod, 1998, for a review). However, ifour test. Halfofthe studied words were presented in isolation (e.g., cac­ logic is correct, words generated at study should not show tus); the subject read the word aloud and then the instruction for that an R-F difference on a recall test because the elaborative word appeared. For the other halfofthe studied words, a definition encoding already carried out on these words should be was presented (e.g., "this protects you from the rain-uT') and re­ able to withstand an instruction to forget; that is, subse­ mained on the screen until the subject responded aloud, at which quent differential rehearsal should not be able to override point its instruction appeared. Each instruction remained on the screen for a period ofJ sec: "RRRRR" indicated that the preceding such strong conceptual encoding. Ofcourse, recall ofgen­ word was to be remembered, and "FFFFF" indicated that it could be erated words should be much better than that ofread words forgotten. There were 10 remember and 10 forget instructions for (Slamecka & Graf, 1978). each of the 20 read and 20 generate words, presented in random Turning to indirect test performance, we chose to use the order to the subject. word reading task because ofthe virtues ofthis procedure Because only a subject's first response was accepted during study, as an indirect measure that we described earlier. On the accuracy was emphasized. The experimenter pressed one of two basis ofprior findings regarding directed forgetting (Bas­ keys to score accuracy. If the subject did not produce the intended answer, that word was eliminated from further analysis on the re­ den et a1., 1993; MacLeod, 1989; Paller, 1990) and re­ tention tests. (Note that leaving these incorrectly generated words in garding the generation effect (MacLeod & Masson, 1997, the data set for the analyses did not alter the pattern of results, just in press; Masson & MacLeod, 1992), we predicted that all as Masson and MacLeod, 1992, showed.) Stimuli were centered top studied words should show priming. We also anticipated to bottom and left justified on the computer screen, as were the in­ more priming for read words than for generated words, structions to remember or forget. given that the perceptual processing ofread words during Test Procedure initial encoding is probably optimal for the indirect On the speeded word reading task, the 50 subjects were required speeded reading test. However, when the weight of per­ to read aloud each test word that appeared on the screen as quickly ceptual encoding is reduced, as for the generated words, it as possible. Again, stimuli were centered top to bottom and left jus- DIRECTED FORGETTING AND GENERATION 357 tified and stayed on the screen until the subject responded. In addi­ ~el.1 as for t~e new condition. Note that trimming the in­ tion to the 40 studied words, 20 unstudied words were incorporated dIVIdual subject latencies to censor those shorter than 300 to provide a baseline measure, with the order of all 60 words ran­ or longer than 1,200 msec removed almost no data (only ~omized. Subjects spoke each test word into a microphone, allow­ five latencies in total). Response latency was faster for mg response latency to be recorded by the computer. All subjects were given a practice session using the microphone prior to the ex­ studied items than for new items [t(49) = 4.36, p < .001], periment. The experimenter again recorded accuracy by pressing demo~strating priming in the latency data. The principal one oftwo keys. analysIs, a 2 X 2 ANOVA, revealed reliable main effects The 35 subjects who did the free recall test were instructed to take ofencoding condition [F( 1,49) = 16.72, MSe = 348.76, p < as much time as they needed to write down, in any order, as many of .001] and of instruction [F(l,49) = 4.16, MSe = 413.27, the words as they could remember. They were explicitly told to dis­ p < .05]. Priming was greater for read items (548 msec) regard the study phase instructions to remember or forget. than for generated items (558 msec), consistent with typ­ ical findings on indirect measures (e.g., Jacoby Dallas, RESULTS & 1981). Also, R words (550 msec) showed more priming Study Phase than F words (556 msec), again consistent with prior re­ During the study phase, the mean proportions oferrors search (MacLeod, 1989). The interaction of these two were .000 for the words that were read and .055 for the variables, although not reliable, did point to a clear trend, words that were generated. evident in Table I [F(I,49) = 2.66, MSe = 439.73,p = .11]. Our primary concern was with the R-F difference as a All principal analyses were 2 X 2 analyses ofvariance (ANOyAs), w~t~ the two within-subjects variables being function ofencoding condition. Planned comparisons re­ encodmg condition (read vs. generate) and instruction (re­ vealed that this difference was not significant for read member vs. forget). Additional t tests were conducted on words (F < I), but that it was significant for generated the speeded word reading data for comparisons ofstudied words [F(l,49) = 4.94, MSe = 1,159.44,p < .05]. Put sim­ to new (unstudied) items; this was accomplished by aver­ ply, the R-F difference in speeded word reading was re­ aging over the four studied conditions and comparing this stricted to the words generated during study. mean to the single new condition. Free Recall Test Speeded Word Reading Test For free recall, the mean proportions of words accu­ Both accuracy and latency data were collected for the rately recalled following the study phase are displayed in speeded reading task. The subjects were highly accurate Table 1. Subjects recalled an average proportion of.314 of when reading the words from the computer screen. The the 40 study words during the recall task. The 2 X 2 mean accuracy rates for the five conditions were as fol­ ANOVA showed reliable main effects of both encoding lows: .998 for read-remember, .998 for read-forget, .998 condition [F(l,34) = 73.44, MSe = .022,p < .001] and in­ for generate-remember, 1.000 for generate-forget, and struction [F(l,34) = 5.16, MSe = .029,p < .05]. That more .993 for new. 3 Accuracy was greater for studied items than generated words (.420) were recalled than read words for new items [t(49) = 2.0I,p = .05], demonstrating prim­ (.208) concurs with prior findings for direct tests ofmem­ ory (MacLeod & Masson, 1997, in press; Masson & Mac­ ing in the accuracy data. Not surprisingly, though, a 2 X 2 ANOVA confirmed that there were no reliable differences Leod, 1992; Siamecka & Graf, 1978; Weldon, 1991). The in accuracy among the four studied conditions (all three overall better recall of R words (.346) than of F words Fs< I). (.281) is also in keeping with the directed forgetting liter­ The latency data were ofprimary interest. Table I dis­ ature (see MacLeod, 1998). The interaction was also sig­ plays the mean response latencies and the corresponding nificant [F(l,34) = 4.82, MSe = .024, p < .05]. standard errors for each ofthe four studied conditions, as With regard to the planned comparisons, whereas recall ofthe generated words showed no significant effect ofin­ struction (F< I), recall ofthe read words was significantly Table I better following remember instructions than following Mean Response Time (RT, in Milliseconds) and Standard Error forget instructions [F(I,34) = 9.97, MSe = .053,p < .01]. (SE) on the Indirect Speeded Word Reading Test and Mean Put simply, the R-F difference in recall was restricted to Proportion Correct (PC) and Standard Error (SE) on the the words that had been read during study. Direct Free Recall Test, as a Function of Encoding Condition Test DISCUSSION Speeded Reading Recall Condition RT SE PC SE We have observed three principal dissociations within Read a single experiment. First, we replicated the familiar read­ Remember 547 8.28 .269 .036 generate dissociation across tests: Our indirect speeded Forget 548 9.51 .146 .021 Generate reading test showed better memory (more priming) for Remember 553 8.88 .424 .030 read words, whereas our direct recall test showed better Forget 563 9.62 .416 .031 memory (greater recall) for generated words. This, of New 565 10.00 course, is the pattern reported by Jacoby (1983; Jacoby & 358 MAcLEOD AND DANIELS

Dallas, 1981) and fits nicely with transfer appropriate pro­ priming on speeded reading was optimal when words cessing accounts (e.g., Masson & MacLeod, 1992; Roedi­ were read during encoding, so the instructional manipula­ ger, 1990). Under such views, encoding at a more percep­ tion could not have any effect in this condition. For gen­ tuallevel is most beneficial to performance on the indirect erated words, however, where the benefit ofadditional re­ test, which is typically more perceptually driven, whereas hearsal for R words could be seen because priming was encoding at a more conceptual level is most beneficial to not optimal, the R-F difference did indeed appear. performance on the direct test, which is typically more con­ In summary, these three dissociations help to differen­ ceptually driven. tiate the processing that underlies performance on an in­ Second, there was a reliable effect ofinstruction for the direct test from that underlying performance on a direct read words but not for the generated words on the direct test. Under the transfer-appropriate processing frame­ recall task. The presence ofthe effect for read words is en­ work, it is the match between processing at study and pro­ tirely consistent with the extensive directed forgetting lit­ cessing at test that is crucial in determining the success of erature (see Johnson, 1994; MacLeod, 1998, for reviews remembering, and that explains the oft-observed between­ ofthe many relevant studies). Recall that under the item test dissociation ofthe read and generate conditions. But method, a remember instruction appears to motivate the what ofthe two within-test dissociations due to directed for­ subject to rehearse that item; by contrast, a forget instruc­ getting that we have reported here? Our argument is that tion functions to discourage rehearsal of that item the benefits of the additional rehearsal that accrues to R (see Basden et ai., 1993; R. A. Bjork, 1989). Given the items are visible only when performance on a test is not largely perceptual encoding for the words read at study, R optimized by the type of encoding. This occurs in direct words are better recalled because they have benefited tests where encoding was not elaborative, and in indirect from more rehearsal following initial encoding. Now con­ tests where encoding was not perceptual. sider generated words. Here, our claim is that the seman­ tic encoding ofboth the R words and the F words was al­ REFERENCES ready so rich that differential rehearsal was virtually BALOTA, D. A., & CHUMBLEY, J. I. (1984). Are lexical decisions a good irrelevant, and was essentially overridden. Consistent with measure oflexical access? The role ofword frequency in the neglected the Golding et al. (1994) finding that it is difficult to for­ decision stage. Journal ofExperimental Psychology: Human Percep­ get an F word that is semantically related to a prior R word, tion & Performance, 10,340-357. it appears that if every word is processed elaboratively BASDEN, B. H., & BASDEN, D. R. (1996). Directed forgetting: A further comparison ofthe list and item methods. Memory, 4, 633-653. prior to presentation ofits instruction, then the normal po­ BASDEN, B. H., BASDEN, D. R., & GARGANO, G. 1. (1993). Directed for­ tency ofthe forget instruction (which hinges on differen­ getting in implicit and explicit memory tests: A comparison ofmeth­ tial rehearsal) is completely eliminated on a direct test ods. Journal ofExperimental Psychology: Learning, Memory, & Cog­ such as free recall. nition, 19,603-616. The third dissociation occurred within the indirect BASSIL!, J. N., SMITH, M. c., & MACLEOD, C. M. (1989). Auditory and visual word-stem completion: Separating data-driven and conceptu­ speeded word reading task. Here, there was also a selec­ ally driven processes. QuarterlyJournal ofExperimental Psychology, tive effect ofthe directed forgetting manipulation, but now 41A, 439-453. only the generated words and not the read words showed BJORK, E. L., & BJORK, R. A. (1996). Continuing influences of to-be­ a reliable R-F difference. A critic might wish to suggest forgotten information. Consciousness & Cognition,S, 176-196. BJORK, R. A. (1989). Retrieval inhibition as an adaptive mechanism in that this was the result ofcontamination from explicit re­ human memory. In H. L. Roediger III & F. I. M. Craik (Eds.), Varieties trieval, given the better memory for generated words on ofmemory and consciousness: Essays in honour ofEndel Tulving direct tests. However, two arguments militate against such (pp. 309-330). Hillsdale, NJ: Erlbaum. a criticism. The first is empirical: There was no corre­ CHALLIS, B. H., & SIDHU, R. (1993). Dissociative effect ofmassed rep­ sponding R-F difference in recall for generated items to etition on implicit and explicit measures of memory. Journal ofEx­ perimental Psychology: Learning, Memory, & Cognition, 19, 115­ support such contamination. The second is more method­ 127. ological (see MacLeod, 1996; MacLeod & Masson, in press, EBBINGHAUS, H. (1885/1964). Memory: A contribution to experimental for relevant arguments): As an indirect measure ofmem­ psychology. New York: Dover. (Original work published 1885) ory, speeded word reading has the virtue that the test item GOLDING, J. M., LONG, D. L., & MACLEOD, C. M. (1994). You can't al­ ways forget what you want: Directed forgetting ofrelated words. Jour­ is presented without degradation and hence without the nal ofMemory & Language, 33, 493-510. need for problem solving, thereby obviating any need for GOLDING, J. M., & MACLEOD, C. M. (EDS.) (1998). Intentionalforget­ conscious recollection. ting: Interdisciplinary approaches. Mahwah, NJ: Erlbaum. So how can we explain this dissociation within the in­ GRAF, P., & MANDLER, G. (1984). Activation makes words more acces­ direct test? We rely again on the idea of differential re­ sible, but not necessarily more retrievable. Journal ofVerbal Learning & Verbal Behavior, 23, 553-568. hearsal. The logic is really the same as that used to explain GRAF, P., & SCHACTER, D. L. (1985). Implicit and explicit memory for the dissociation within the direct test: When initial encod­ new associations in normal and amnesic subjects. Journal ofExperi­ ing optimizes performance on any test, directed forgetting mental Psychology: Learning, Memory, & Cognition, 11,501-518. cannot exert its effect. In recall, performance was opti­ GRAF, P., SHIMAMURA, A. P., & SQUIRE, L. R. (1985). Priming across modalities and priming across category levels: Extending the domain mized by elaborative encoding of the generated words, ofpreserved function in amnesia. Journal ofExperimental Psychol­ making differential rehearsal redundant; thus, an R-F dif­ ogy: Learning, Memory. & Cognition, 11,385-395. ference could appear only for read words. In contrast, GRAVES, R., & BRADLEY, R. (1987). Millisecond interval timer and au- DIRECTED FORGETTING AND GENERATION 359

ditory reaction time programs for the IBM PC. Behavior Research ROEDIGER, H. L., III, & BLAXTON, T. A. (1987). Effects of varying Methods. Instruments. & Computers, 19,30-35. modality, surface features, and retention interval on priming in word­ GRAVES, R., & BRADLEY, R. (1988). More on millisecond timing and fragment completion. Memory & Cognition, 15,379-388. tachistoscopic applications for the IBM PC. Behavior Research Meth­ ROEDIGER, H. L., III, & McDERMOTT, K. B. (1993). Implicit memory in ods.Instruments. & Computers, 20, 408- 412. normal human subjects. In F. Boller & 1. Grafrnan (Eds.), Handbook JACOBY, L. L. (1983). Perceptual enhancement: Persistent effects ofan ofneuropsychology (Vol. 8, pp. 63-131). Amsterdam: Elsevier. experience. Journal ofExperimental Psychology: Learning, Memory. SLAMECKA, N. J., & GRAF, P. (1978). The generation effect: Delineation & Cognition, 9, 21-38. ofa phenomenon. Journal ofExperimental Psychology: Human Learn­ JACOBY, L. L., & DALLAS, M. (1981). On the relationship between auto­ ing & Memory, 4, 592-604. biographical memory and perceptual learning. Journal ofExperimen­ STROOP, J. R. (1935). Studies of interference in serial verbal reactions. tal Psychology: General, 1I0, 306-340. Journal ofExperimental Psychology, 18, 643-662. JOHNSON, H. M. (1994). Processes of successful intentional forgetting. WARRINGTON, E. K., & WEISKRANTZ, L. (1970). The amnesiac syn­ Psychological Bulletin, 1I6, 274-292. drome: Consolidation or retrieval? Nature, 228, 628-630. MACLEOD, C. M. (1989). Directed forgetting affects both direct and in­ WELDON, M. S. (1991). Mechanisms underlying priming on perceptual direct tests ofmemory. Journal ofExperimental Psychology: Learn­ tests. Journal ofExperimental Psychology: Learning, Memory, & ing, Memory. & Cognition, 15,13-21. Cognition, 17, 526-541. MACLEOD, C. M. (1996). How priming affects two speeded implicit tests ofremembering: Naming colors versus reading words. Consciousness NOTES & Cognition, 5, 73-90. MACLEOD, C. M. (1998). Directed forgetting. In 1. M. Golding & C. M. I. It is important to realize, though, that the mapping ofdirect test to MacLeod (Eds.), Intentionalforgetting: Interdisciplinary approaches conceptual processing and indirect test to perceptual processing is (pp. I-57). Mahwah, NJ: Erlbaum. clearly an oversimplification. Although tests tend to sort in this way, MACLEOD, C. M. (1999). The item and list methods ofdirected forget­ there certainly exist conceptual indirect tests and perceptual direct tests ting: Test differences and the role ofdemand characteristics. Psycho­ (see Roediger & McDermott, 1993). Because the speeded word reading nomic Bulletin & Review, 6,123-129. test would seem to be more perceptual and the recall test certainly is MACLEOD, C. M., & MASSON, M. E. J. (1997). Priming patterns are dif­ more conceptual, we have chosen for simplicity to emphasize the more ferent in masked word identification and word fragment completion. common mapping here. Consequently, our claims regarding indirect Journal ofMemory & Language, 36, 461-483. tests can be interpreted as referring only to perceptual indirect tests. MACLEOD, C. M., & MASSON, M. E. J. (in press). Repetition priming in 2. Actually, two sets ofdata-each made up of25 subjects-were col­ speeded word reading: Contributions of perceptual and conceptual lected about a year apart in the speeded word reading task. A preliminary processing episodes. Journal ofMemory & Language. analysis including these two subgroups as a factor revealed no reliable MASSON, M. E. 1., & MACLEOD, C. M. (1992). Re-enacting the route to main effect ofsubgroup and no reliable interactions involving subgroup, interpretation: Enhanced perceptual identification without prior per­ so this variable was deleted and the data for all 50 subjects were merged. ception. Journal ofExperimental Psychology: General, 121, 145-176. Note, too, that a recognition test was administered to all subjects after the PAIVIO, A., YUILLE, 1. c., & MADIGAN, S. A. (1968). Concreteness, im­ speeded word reading test. Because performance in the generate condi­ agery, and meaningfulness values for 925 nouns. Journal ofExperi­ tion was at ceiling on the recognition test, it will not be discussed further. mental Psychology Monographs, 76 (I, Pt. 2), 1-25. 3. Because the absence of variance in the generate-forget accuracy PALLER, K. A. (1990). Recall and stem-completion priming have differ­ data would have prevented analysis ofvariance, I subject's score in that ent electrophysiological correlates and are modified differently by di­ condition was changed from 1.000 to 0.999. rected forgetting. Journal ofExperimental Psychology: Learning. Memory. & Cognition, 16,1021-1032. RICHARDSON-KLAVEHN, A., & BJORK, R. A. (1988). Measures ofmem­ ory. Annual Review ofPsychology, 39, 673-690. ROEDIGER, H. L., III (1990). Implicit memory: Retention without re­ (Manuscript received November 24, 1997; membering. American Psychologist, 45, 1043-1056. revision accepted for publication June 30,1999.)