Memory & Cognition 1998.26 (3),435-443

Long-term positive and negative identity priming: Evidence for episodic retrieval

DOUG LOWE Trent University, Peterborough, Ontario, Canada

An episodic retrieval account of negative priming (Neill, 1997;Neill & Valdes, 1992)was evaluated in three experiments. Duringpractice, regular word pairs were presented to subjects differing num­ bers of times. The subjects named specific target words while they ignored specific distractor words. Following a 5-min retention interval, memory for practice was revealed: Test responses for target words exhibited positive priming that increased with increases in the number of times thatthe words had been attended. Test responses for distractor words exhibited either positive priming (Experi­ ment 1) or negative priming (Experiments 2-3) that also increased with increases in the number of times that the words had been ignored. The type of priming that distractors exhibited was deter­ mined by several contextual similarities between the practice environment, in which distractors were ignored initially, and the test environments, in which they were processed subsequently. Neg­ ative priming that spanned a 5-min interval, increased with increases in the number of times that a distractor was ignored, and was sensitive to contextual changes indicated that the direction of the effect was temporally backward because the test probe cued memory for earlier processing of the priming stimulus when the distractor had been ignored.

The deployment ofselective attention permits subjects (1985) termed transfer effects ofthis type negativepriming to respond optimally to target aspects ofstimulation while (NP). Subsequent research has shown the wide applica­ ignoring distractor aspects. There is now substantial evi­ bility ofNP to a broad range ofstimuli and tasks (see Fox, dence that selectingfor, or facilitation of, target information 1995; Houghton & Tipper, 1994; May, Kane, & Hasher, entails selecting against, or inhibition of, distractor infor­ 1995; Neill, Valdes, & Terry, 1995, for reviews). mation. Distractor inhibition has been inferred from the In considering the meaning ofNp' Neill (1977) specu­ findings ofpriming procedures during which the distractor lated that inhibition ofdistractor representations resulted item for a leading, or priming, stimulus becomes the tar­ in reduction oftheir activation below base levels. The "de­ get item for a trailing, or probe, stimulus. According to the activated" representations lingered into the probe trial, and logic ofthe procedure, ignoring an item by means ofits in­ performance was impaired because the appropriate re­ hibition on the prime trial will make the same item diffi­ sponses were temporarily unavailable. Neumann and De­ cult to process on the subsequent probe trial. Schepper (1991) proposed that inhibition resulted in a raised Dalrymple-Alford and Budayr (1966) first demonstrated threshold ofactivation for distractor representations. Tipper a distractor inhibition effect in a list ofStroop (1935) color and Cranston (1985) subsequently modified Neill's (1977) words. When each distractor color word in the list named proposal by arguing that inhibited representations re­ the next ink color, color-naming responses were slower and mained activated, but were "blocked" from access to re­ less accurate than when adjacent color wordswere unrelated. sponse mechanisms. However,to come full circle, Houghton Neill (1977) confirmed the finding in a discrete trials ver­ and Tipper (1994) proposed that, at prime offset, there are sion ofthe task, and made the important interpretation that negative levels ofactivation for distractor representations. the distractor representations for the prime had been se­ Clearly, there are shifting opinions on the meaning ofNP. lectively inhibited. To contrast the processing impairment Yet despite this absence ofa theoretical consensus, the for ignored stimuli with the facilitated processing, or pos­ different accounts agree that the direction ofNP effects is itive priming (PP), produced for attended stimuli, Tipper temporally "forward" from the prime, via lingering persis­ tence ofa weakened distractor representation, into the probe trial. Recent findings, however, suggest that NP may nei­ ther index selective processing ofthe prime, nor reflect ac­ This research was conducted from 1986 to 1989 while the author was funded by the National Science and Engineering Research Council of tive inhibition ofthe distractor. Canada (Grant AO 284), I am indebted to Mike Masson, Bruce Milliken, While investigating the duration of'Nl; Neill and his col­ Tramm Neill, and Bruce Whittlesea for many insightful comments on an leagues (Neill & Valdes, 1992; Neill, Valdes, Terry,& Gor­ earlier version ofthis manuscript. Correspondence should be addressed fein, 1992) discovered that although NP can last up to sev­ to D. Lowe, Department ofPsychology, Trent University, Peterborough, ON. Canada K9J 7B8 (e-mail: [email protected]). eral seconds, the dissipation ofNP was greatly reduced when the temporal interval between the response to the prime and -Accepted by previous editor, Geoffrey R. Loftus the onset ofthe probe matched the interval that preceded

435 Copyright 1998 Psychonomic Society, Inc. 436 LOWE the prime. In contrast, when the intervals surrounding the retrieved from memory before they can be provided by the prime were not constant, greater dissipation ofNP was ob­ algorithm. served (see also Hasher, Zacks, Stolzfus, Kane, & Con­ To apply the theory to NP, Neill and Valdes (1992) ar­ nelly, 1996; Neill, 1997). Since events preceding the gued that information about ignored events is also en­ prime affect the duration ofNP, it cannot be assumed that coded into memory; presumably, in the NP task, attributes probe performance is a direct index of selective process­ for the prime distractor are coded as "to-be-ignored" or ing ofthe prime or the activation state ofdistractor repre­ "nonreportable." Presentation of ignored items as probe sentations. Moreover, the observation that NP is affected targets cues retrieval ofthe priming episodes, and perfor­ by events that surround the prime and possibly by events mance is impaired because of interference between the that surround the probe suggests that NP may be deter­ available"response" and the retrieved"nonresponse" attrib­ mined by broader types ofprime-probe relationships be­ utes that are concurrently associated with the target. yond those involving prime distractors and probe targets. Alternatively, the distractor having been ignored, it may Toaccount for their findings, Neill (Neill & Valdes, 1992; be that no response information for the distractor gets Neill et aI., 1992) attributed NP to memory retrieval ofthe stored, and responses may be delayed because there is no priming episode, during which the probe target was ig­ appropriate response information available. By this ac­ nored. Baddeley's (1976) proposal that the success ofre­ count, NP depends on the differential processing of tar­ trieving an episode depends on its temporal discriminabil­ gets and distractors during selective attention to the prime. ity from other episodic memories explains the near identity However, in contrast to the logic ofthe priming procedure, ofNP at constant short and long intervals. When applied the direction of NP is temporally "backward" when the to the NP task, temporal discriminability of the priming probe cues retrieval ofmemory for past processing ofthe episodewould depend on the ratio ofthe most recent prime­ distractor. Accordingly, the processing impairments dur­ probe delay to the delay between the prime and previous ing NP trials do not reflect distractor inhibition during se­ trials. When the current prime-probe delay is short, and the lective attention to the prime. interval before the prime is long, the probability of re­ Despite its apparent promise, Logan's (1988) theory trieving the priming episode is high, yielding NP. Con­ may be problematic when applied to the NP task. Typical versely,when the current prime-probe interval is long, and NP studies use a few stimuli that are balanced to serve the delay before the prime is short, the probability ofre­ equally as targets and distractors and that are presented re­ trieving the priming episode is low, yielding reduced NP. peatedly throughout the experiment. Malley and Strayer When the intervals surrounding the prime are constant, (1995) have shown that NP is stimulus specific, restricted despite their durations, the ratio is the same. Consequently, to highly familiar stimuli experienced before. They found there is equal discriminability of the priming episodes, that NP occurred only when an ignored distractor was a and retrieval probability is unaffected, yielding equal de­ stimulus that had been repeated often throughout the ex­ grees of NP. More recently, Neill (1997) has directly periment. Pp, and not NP, occurred when a distractor was shown the importance oftemporal discriminability for re­ novel and had not been experienced before the prime. trieval ofthe priming episode, and for NP. By manipulat­ Thus, when the same stimuliare used repeatedly through­ ing the target-distractor onset interval for both prime and out the experiment, in principle, on any given NP probe probe trials, it was shown that NP was dependent on the trial, a retrieved episode could contain the "reportable" at­ identity ofdistractor onset conditions between primes and tribute as often as it contained the "nonreportable" attribute. probes. When primes and probes did not share the same In such an unstable environment, memory for past pro­ distractor onset information, retrieval did not occur, and NP cessing would provide little useful information (Logan, was not obtained. 1988). Therefore, how retrieval ofpast episodes would af­ The processes implicated by a retrieval account ofNP fect performance is not transparent, even if that retrieval are similar to those posited in Logan's (1988) instance the­ was certain to occur. Indeed, when "reportable" and "non­ ory ofautomaticity. According to the theory, performance reportable" attributes are equally retrievable, it could be ofany cognitive task is initially accomplished by an algo­ predicted that the attributes might cancel, and there would rithmic computation. Each time the task is performed, a sep­ be no NP at all. arate episodic trace, or instance, is stored in memory. The Experiment 1 was a direct test of a memory-based ac­ episode contains information about a particular stimulus count ofNP. To avoid problems inherent in the same stim­ event and all information that was associated with it in the uli's serving as both targetsand distractors, individualwords past, including the response that was made. Attributes im­ consistently served as either targets or distractors, but not portant for performing the task are attended and are in­ both. The words were painted in different ink colors, and, corporated into the episode; presumably, in the NP task, on each trial, subjects attended, and named aloud, the attributes for the prime target are coded as "to-be-attended," word in the target color while they ignored the word in the or "reportable." Subsequent encounters with the stimulus nontarget color. Following Logan (1988), during practice cue memory retrieval of the priming episode, providing specific word pairs were presented one, three, six, or nine the correct response. With repetition ofthe task, more in­ times. Ofnecessity, repetition ofthe items caused the final stances accumulate in memory, and correct responses are presentations ofthe most presented pairs to occur during LONG-TERM NEGATIVE PRIMING 437

the latter part of practice. To eliminate any "recency ad­ by unexpected newpairs duringtest. Experiment 2 was a test vantage" for these pairs, there was an unfilled 5-min re­ ofthis possibility; new items were eliminated, and the test tention interval before test. trials were composed entirelyofATTand IGN items. Under Following the retention interval, a test phase involved these circumstances, distractors exhibited progressive NP presentations ofprobe stimuli that were designed to yield with increases in the number of presentations. Experi­ single, one time, reaction time (RT) measures ofboth the ment 3 was a study ofa second contextual determinant of type and degree ofpriming for target and distractor words NP that emerged in Experiments 1 and 2. from practice. One third ofthe test trials assessed PP for targets, which had been attended one, three, six, or nine EXPERIMENTS 1 AND 2 times in an attended repetition (ATT) condition (Tipper, 1985) that involved a final presentation of prime word Method pairs from practice. Another third of the test trials as­ Design and Materials. One hundred ninety-two words, three to sessed NP for distractors, which had been ignored one, five letters in length, were selected from clusters seven and eight of three, six, or nine times in an ignored repetition (IGN) con­ the Battig and Togglia (1978) word norms. One hundred forty-four dition (Tipper, 1985). As in a typical IGN condition, dis­ ofthe words were randomly selected to create 72 pairs while ensur­ tractor words from practice were painted in the target color. ing that the items were not synonyms, did not rhyme, and had no ob­ vious association (e.g., BOX and CAR). For each pair, the words were However, unlike in a typical IGN condition, to ensure the printed above one another on white cards in uppercase 42-point Le­ consistency oftargets and distractors, test (i.e., probe) tar­ traset Instant Lettering (No. 723). For one halfofthe pairs, the upper gets were paired with novel words that were not part of word was painted in red ink and the lower word was presented in practice. The novel words served as the probe distractors blue. For the remaining pairs, the colors were reversed. and were painted in the nontarget color. On the remaining Forty-eight pairs were randomly selected to serve as practice stim­ test trials, which provided the baseline for assessing the uli; the remaining 24 pairs served as new test items. The 48 practice pairs were also altered to serve as IGN probes by replacing target degree of priming for ATT and IGN trials, novel (new) words with 1 ofthe remaining 48 words that had been selected from word pairs were presented that were unrelated to practice. the norms. The words that served as prime distractors remained in Memory-based theories (Logan, 1988; Logan & Ether­ their same upper or lower position but appeared in the target color. ton, 1994) propose that practice at attending and reporting The replacement (i.e., probe distractor) word was painted in the non­ target members ofthe pairs should create episodic mem­ target color. For example, the pair BOOK in red above FISH in blue ories that contain the attributes that specific words are was altered to DRUM in blue above FISH in red. For a subject naming red words, FISH would be the prime distractor and probe target, re­ consistently "reportable." With repetitions, more traces of spectively. A second set ofIGN stimuli was similarly constructed by the pairs and the appropriate responses should accumulate replacing the other word (i.e., FISH). Overall, there were 120 different in memory. Subsequent presentations ofthe pairs during pairs, and each was photographed separately onto 35-mm slides. ATT test trials should cue retrieval of that knowledge, Both experiments involved 300 trials composed of 228 practice yielding progressively improved performance, or Pp, with trials followed by 72 test trials, with a 5-min interval between. The increases in number ofrepetitions. According to Neill and practice phases ofall experiments were identical. Two different pre­ sentation orders of 144 trials each were used, with one half ofthe Valdes (1992), ignoring distractor members of the re­ subjects each receiving a different order. The subjects were pre­ peated pairs should create episodic memories that also in­ sented the 48 word pairs, which had been randomly separated into clude the knowledge that specific words are consistently four sets of 12 pairs each. Enough examples were created so that a "nonreportable," or memories that lack response informa­ particular set could be presented one, three, six, or nine times. The tion. Subsequent presentations ofthe distractors as targets stimuli for the four presentation conditions were fairly evenly dis­ during IGN test probes should cue retrieval of those tributed throughout practice, with the restriction that no identical episodes, yielding progressively impaired performance, or pair or position of the same colored item could appear more than twice in succession. One halfofthe subjects named red words, and Np, with increases in the number ofpresentations. "Long­ the remainder named blue words. For one half of the subjects, the term" NP that spans an interval of5 min and that increases practice pairs that served as ATT stimuli during test were used as with the number of times that an item has been ignored IGN probes for the other half ofthe subjects. For the remainder of would implicate memory retrieval as the cause ofthe pro­ the subjects, this relationship was reversed. cessing impairment. Neuman and DeSchepper (1992) Within the test phase ofeach experiment, all subjects received the have shown NP effects in a short-term memory task, and identical stimulus order. There were 24 new items and 6 pairs for each number of practice presentations for both ATT and IGN con­ there is evidence from a shape-matching task (DeSchep­ ditions. In Experiment I, the 24 pairs for each ATT, new, and IGN per & Treisman, 1991, 1996) that NP can be long term, and test condition were randomly intermixed with the same restrictions can increase with increases in the number oftimes that a used for practice. In Experiment 2, there were no new test items; the distractoris ignored(but see Malley & Strayer, 1995,p. 658). 24 stimuli within each ATT and IGN test condition appeared in the To anticipate the findings, in different experiments, test same order as in Experiment I. responses for distractors exhibited either PP or NP. In Ex­ Procedure. All stimuli were back projected onto an opaque screen via a Kodak carousel projector that was fitted with a Gerbrands shut­ periment 1, both targets and distractors exhibited progres­ ter. When the displays were viewed from a distance ofapproximately sive PP with increases in the number ofpresentations. It 66 ern, their subtended visual angles varied from approximately 3.3° appeared that between the practice and test phases, sub­ X 2.2° for three-letter words to 5.50 X 2.2° for five-letter words. The jects may havedetected a contextual change that was caused separation between upper and lower words was approximately 2.2°. 438 LOWE

Table 1 Mean Reaction Times (in Milliseconds), With Standard Deviations, for Practice Trials as a Function of Number ofPractice Presentations (Experiments 1-3) No. of PracticePresentations 2 3 4/5 6/7 8/9 Experiment M SD M SD M SD M SD M SD M SD 1 673 66.4 634 65.4 628 67.2 617 65.8 619 76.0 606 76.2 2 660 58.4 619 52.6 614 59.1 610 66.2 608 63.5 597 67.5 3 658 56.7 618 52.0 613 55.1 611 58.7 609 62.6 586 69.5

Presentation times were controlled by an Apple 11+ microprocessor, increased [Experiment 1, F(5,195) = 37.06, MSe = 563; which also measured and collected RTs. Experiment 2, F(5, 195) = 35.53, MSe = 544]. The great­ Each trial began with a central fixation dot on the blank screen, est improvements occurred between one and two presen­ above and below which the words appeared. The subjects initiated tations. Except for one minor reversal between four and stimulus presentation, which simultaneously started a software timer. The responses, registered with a voice key,simultaneously (I) stopped five presentations and six and seven presentations in Ex­ the timer, (2) removed the stimulus from view and (3) advanced the periment 1, in both experiments, there were small, pro­ projector, which signaled the onset ofthe next trial. The intertrial in­ gressive, RT improvements following two presentations. terval was approximately 3 sec. The subjects were instructed ( I) to Additional ANOVAs,that excluded one-presentation data, name the word that was painted in the target color as quickly and ac­ showedthat RTs decreased between two presentations and curately as possible, (2) that, although from trial to trial the location eightand nine presentations [Experiment 1,F(4,156) = 9.59, ofthe target word was unpredictable, the word would appear in both locations equally often, and (3) that some pairs would appear more MSe = 490; Experiment 2, F(4,156) = 5.58, MSe = 505]. than once. Between practice and test trials, the subjects chatted with Progressive RT improvements, or PP, with increasing the experimenter. numbers ofpresentations show that practice at reporting Subjects. The 40 subjects in each experiment were male and fe­ specific targets for the repeated pairs progressively aided male undergraduates at Trent University who participated as a subsequent performance. Presumably, subjects learned course requirement. and remembered which words were "reportable." Subse­ quent encounters with the pairs cued retrieval ofthat knowl­ Results and Discussion edge, facilitating performance, as instance theory (Logan, The rejection region for all statistical tests was p < .01, 1988) predicts. or better, unless stated otherwise. Data from trials on The greater improvement between one presentation and which subjects' responses were incorrect or on which sub­ two than between subsequent repetitions shows that there jects failed to stop the clock were excluded from analysis. were "warmup effects," so that performance was affected The error data were too few to allow meaningful analyses; also by general or nonspecific practice (Logan, 1988).Gen­ only 5 and 6 subjects in Experiments 1and 2, respectively, eral practice would increase with progressive trials, so the made any errors. RT improvements could have been due entirely to non­ Practice trials: Experiments 1 and 2. For each sub­ specific practice. To gauge the extent ofnonspecific prac­ ject, median RTs were calculated for each number ofpre­ tice, data for pairs that were presented only once, and for sentations. Since items that were repeated three, six, or which specific practice was impossible, were categorized nine times contributed practice data to all earlier presen­ according to whether they occurred during the first or sec­ tations, medians for one presentation were based on the ond half of practice. For each subject, median RTs were presentation ofall 48 items. For two and three presenta­ calculated. tions, medians were based on the 36 pairs that received In both experiments, mean RTs for nonrepeated items three, six, or nine presentations. For four through nine pre­ from the second half of practice were faster than during sentations, medians were based on the 24 pairs that re­ the first half, showing the presence ofnonspecific practice ceived either six or nine presentations. However, before [Experiment 1, first halfM = 683 msec vs. second half M = calculating the medians, data for four and five presenta­ 657 msec, t(39) = 3.88; Experiment 2, first half M = tions, six and seven presentations, and eight and nine pre­ 668 msec vs. second halfM= 646 msec, t(39) = 3.21]. If sentations were combined. Thus, medians for one, two, PP for repeated pairs was due entirely to nonspecific prac­ three, four and five, six and seven, and eight and nine pre­ tice, performance for repeated and nonrepeated items sentations were based on 48, 36, 36, 48, 36, and 24, or should have been nearly identical across practice. How­ fewer, correct RTs, respectively. The means of subjects' ever, during the second half, RTs for repeated pairs (i.e., median RTs, for each number ofpresentations, are shown four and five, six and seven, and eight nine presentations in Table 1. combined) were substantially faster than RTs for nonre­ Across experiments, the pattern ofresults was virtually peated items [Experiment 1, nonrepeated M = 657 msec identical. One-way repeated measures ANOVAs showed vs. repeated M= 615 msec, t(39) = 5.93; Experiment 2, that average RTs decreased as the number ofpresentations nonrepeated M = 646 msec vs. repeated M = 605 msec, LONG-TERM NEGATIVE PRIMING 439

Table 2 one/three presentations [M = 656 msec; t(39) = 3.24], Mean Reaction Times (in Milliseconds), With which in turn were faster than RTs for new items [M = Standard Deviations, for Test Trials as a Function 675 msec, t(39) =4.02]. ofNumber ofPractice Presentations and Test Response Condition (Experiments 1-3) Progressive PP with number ofrepetitions for both tar­ Response Condition get and distractor words indicates that test performance was memory based. The specific activities ofattending par­ Attended Ignored ticular target words while ignoring particular distractor Repetition Control Repetition words were not differentially influential, but frequency of 113 619 o l/3 619 exposure to both item types was. PP for IGN trials could Experiment M SD M SD M SD M SD M SD be an embarrassment for the inhibitory theory; apparently, I 654 84.7 636 78.0 675 72.5 658 81.9 641 80.6 2 624 69.5 610 55.3 639 71.2 656 92.9 the representations of target and distractor words were 3 636 68.3 621 70.1 641 67.9 642 79.1 664 69.0 equally available during test. PP for distractor processing Note-I!3, 619, 0, 113,619 = number of practice presentations. might also contradict Neill and Valdes's (1992) notion that information that distractors were "nonreportable" was in­ cluded, or that appropriate response information was lack­ t(39) = 8.09]. The advantage for repeated pairs is consis­ ing, in the retrieved episodes. tent with the idea that learning was item specific, and that However, these results were obtained with procedures performance was instance or item based. quite different from those usually used to test for NP (i.e., Test trials: Experiment 1. Since there were only six a large set of words as stimuli, consistency of stimuli as scores for each ATT and IGN condition for each number of targets and distractors, multiple practice on items, novel practice presentations, the data were quite variable. Con­ items as control stimuli, novel items as probe distractors, sequently,data for trials with one and three practice presen­ long-term test, etc.). Anyone ofthese changes might have tations and with six and nine practice presentations were eliminated NP. During practice, for example, the distractors combined before calculating the medians. Itshould be noted might not have been ignored; with repetitions, subjects that distractors for ATT probes were the same during prac­ might have deliberately attended to them. Perhaps IGN tice and test, whereas IGN test phase distractors were new trials failed to exhibit NP because distractor inhibition words, so that a direct comparison between these items never took place (i.e., ifNP had been measured during prac­ may not be meaningful. More interpretable results are re­ tice). Experiment 2 was performed to address these issues. vealed by the effects ofnumber ofpractice presentations on Tipper, Weaver, Cameron, Brehaut, and Bastedo (1991) ATT and IGN targets. Mean RTs are presented in Table 2. found that NP was reduced when novel, unexpected, events The design was not a complete factorial, so several intervened between primes and probes. Their test trials in­ analyses were used. To evaluate response conditions, data cluded unexpected presentations of new pairs that were for each subject were collapsed across number ofpractice unrelated to practice. Given that their subjects expected presentations (i.e., one, three, six, nine), and medians were repeated word pairings during practice, the introduction calculated for each ATT, new, and IGN test condition. A ofunfamiliar new items might have signaled that there had one-way repeated measures ANOVA revealed that RTs been a contextual change, thereby eliminating NP (Tipper differed as a function ofconditions [F(2,78)= 4.44, MSe = et aI., 1991). In the present Experiment 2, elimination of 437]. RTsfor ATT (M =645 msec) and IGN (M=649 msec) new word pairs might cause practice and test phases to ap­ trials did not differ (see also next analysis), but both were pear more alike, thereby yielding NP. faster than responses for new trials. Test trials: Experiment 2. Since there were no new A 2 X 2 repeated measures ANOVAwas used to evalu­ probes, a sole 2 X 2 repeated measures ANOVA was used ate number ofpractice presentations and ATT versus IGN to evaluate the effects ofnumber ofpractice presentations trials; RTs for ATT and IGN trials did not differ [F(l,39) on ATT versus IGN trials. Unlike in Experiment I, the test < I, MSe = 3,534], nor did response condition interact responses for distractors were slower than those for tar­ with number ofpresentations [F(l,39) < 1, MSe = 3,047]. gets [F(l,39) = 19.09, MSe = 1,949], and response condi­ Overall RTs decreased with number ofpractice presenta­ tion interacted with number of presentations [F(l,39) = tions, although the effect was marginally reliable [F( 1,39) = 11.14,MSe = 922]. Using theMSe from the ANOVA as the 2.92, MSe = 4,324, P < .10; see next analysis]. estimate of error, subsequent t tests showed that RTs for The effect ofnumber ofpractice presentations was ex­ ATT trials decreased with increases in the number ofpre­ amined in a third analysis by collapsing across ATTand IGN sentations [t(39) = 2.19,p < .05], while RTs for IGN trials trials. For each subject, median RTs for each number (i.e., increased [t(39) = 2.57,p < .05]. zero [new], one/three, six/nine) ofpresentations were cal­ The presence ofNP for IGN targets shows that distrac­ culated. A one-way repeated measures ANOVA showed tor words were ignored during practice, and that subjects that RTs decreased significantly with increases in number did not deliberately attend to them. Progressive PP and NP of presentations [F(2,78) = 7.30, MSe = 532]. Using the for targets and distractors, respectively, also suggests that MSe from the ANOVA as the estimate oferror, subsequent test performance was memory based. Knowledge that target t tests showed that RTs for trials with six/nine presenta­ words were "reportable" was stored in the practice episodes tions (M = 639 msec) were faster than those for trials with that were retrieved during ATTtrials (Logan, 1988). More 440 LOWE important, knowledge that distractors were "nonreportable" serving inconsistently as targets and distractors, control was also stored in the practice episodes that were retrieved items selected from the repeating stimulus set, short-term during IGN trials (Neill & Valdes, 1992). test, etc.). Lowe (1979; Experiment 4) documented that NP that spans an interval of 5 min clearly implicates whether particular probes exhibited PP or NP was con­ memory retrieval of past processing as the source of the trolled by the context that was set by the "mix" ofdiffer­ processing impairment. It may be that distractors were ent types ofprobes that were used within an experiment; "selected against" when they were ignored during practice similar findings have been reported by Moore (1994). As (but see Milliken & Joordens, 1996,and Milliken, Joordens, noted, Tipper et al. (1991) showed that NP is decreased Merikle, & Seiffert, 1995, for evidence that prime selec­ when the test context includes unfamiliar events. Lowe tion may not be necessary for NP), but it is not possible that (1979; Experiment 2) showed that NP was more likely to distractor representations were "deactivated" (Houghton occur when there were many repeated target (i.e., ATT) & Tipper, 1994; Neill, 1977; Neumann & DeSchepper, trials between primes and probes, so that NP can be con­ 1991) or "blocked" (Tipper, 1985). IfNP were caused by textually dependent on subjects' strategies to use informa­ persistence ofinhibited distractor representations into test, tion taken from primes to process probes. In Experi­ it is unlikely that they would be rendered unavailable for ment 1, ATT items made up one third ofthe test items, so such an extended duration. When considered in isolation the small proportion ofrepeated trials may have discouraged from Experiment 1, the findings provide considerable sup­ subjects' use ofmemory for practice pairs for responding port for Neill and Valdes's (1992) retrieval account ofNP. to test items, thereby eliminating NP. In Experiment 2, The pattern ofresults reflects context-specific learning ATT items made up one halfofthe test trials, thus increas­ during practice. NP is sensitive to broader types ofprime­ ing the likelihood that memory for practice items would probe relationships than those involving past distractors be useful for processing test stimuli, thereby yielding NP. and current targets. Targets and distractors for the repeated However, practice-test contextual dissimilarity in Ex­ pairs became connected to the environment or general periment 1 cannot explain why IGN test probes displayed background in which those items were processed. Appar­ PP.Ifcontextual changes prevented retrieval ofprocessing ently, general context was used along with specific test episodes from training, RTs for IGN and new trials should probes during memory access, for what was retrieved de­ have been similar. Typical studies ofNP have also shown pended on whether ornotthe "list-wide" context from prac­ that NP sometimes reverses to PP when probes consist of tice was reinstated during test. When the learning (i.e., nonconflict stimuli (e.g., Lowe, 1979; Tipper & Cranston, practice) and retrieval (i.e., test) contexts were consistent, 1985). NeiIl (1997) has suggested that contextual dissim­ distractor processing was impaired; when those contexts ilarity between conflicting prime stimuli followed by non­ were inconsistent, distractor processing was facilitated. conflicting probes prevents episodic retrieval, thereby The contexts would have been consistent when there were eliminating NP. To account for PP for nonconflict probes, no new items, because test responses consisted entirely of Neill (1997) proposed that there may be a second source of the contents ofpractice episodes. Contextual inconsistency PP that is independent of contextual similarity-that is, did not prevent the influence ofearlier episodes; rather, it activation may persist at some lower level of representa­ changed the way that processing was influenced. Memory tion. Thus, whereas NP depends on retrieval context, PP for the specific activities ofignoring particular distractor for distractor processing may result in part from a differ­ words was context dependent, whereas memory for fre­ ent mechanism that is insensitive to context, such as per­ quency of exposure to them was context independent. sistent activation in a semantic memory system. Overall, the results have something in common with other reports of "list-wide" context effects on PP during word EXPERIMENT 3 recognition tasks (e.g., Chawarski & Sternberg, 1993; Neely & Durgunoglu, 1985; Smith, Macleod, Bain, & Hoppe, When distractors became test targets, they were paired 1989), suggesting that a common mechanism may under­ with novel words, so they appeared in different contexts lie PP and NP. than when they were ignored. If contextual consistency is Context dependency ofNP strongly reinforces the claim critical for NP, it might have been the case that NP failed to that the direction ofthe different priming effects was tem­ occur at all. Thus, unlike in earlier studies (e.g., Lowe, porally "backward" when the test probes cued memory for 1979;Malley & Strayer, 1995; Moore, 1994), in the present past processing. According to "forward" accounts ofNP paradigm, composition ofthe specific test probes might be (e.g., MaIley & Strayer, 1995; NeiIl, 1977; Tipper & unimportant for NP.In Experiment 3, test responses for dis­ Cranston, 1985), impaired performance for ignored stimuli tractors were studied when the items appeared in the same should be "short term" and should be conditional only on specific context as when they were ignored during practice. how the distractors have been processed initially; it should In this experiment, IGN probes consisted ofthe same word not be "long term" and dependent on the context in which pairs that were used during practice. The words appeared in they are processed subsequently. the same upper and lower locations, but their colors were Context dependency of NP also indicates that the re­ "flipped" so that the distractors became targets and vice sults have several commonalities with more typical versa. "Flipped" probes have not been commonly used in demonstrations ofNP (e.g., a small set ofstimuli, stimuli NP experiments, presumably because ofthe complications LONG-TERM NEGATIVE PRIMING 441

of simultaneously reversing the roles of both targets and [F(1,39) = 47.34, MSe = 618]. From Table 2, it is evident distractors between primes and probes. However, using the that RTs for ATT trials decreased with increases in num­ Stroop task, Lowe (1979; Experiment 1) documented that ber ofpresentations [t(39) = 2.70], whereas those for IGN "flipped" probes do exhibit NP. As in earlier experiments, trials increased [t(39) = 3.60]. Ofless interest, overall RTs individual words consistently served as either targets or dis­ were unaffected by number ofpresentations [F(I,39) < I, tractors. Target-distractor consistency was preserved when MSe = 618]. The final ANOVAalso indicated that RTs for the pairs were repeated during ATTprobes; that consistency trials with zero/new (M = 641 msec) versus one/three (M = was reversed during "flipped" IGN probes. 638 msec) versus six/nine (M = 643 msec) prior presenta­ According to Tulving and Thomson's (1973) encoding tions did not differ [F(2,78) < 1, MSe = 623]. specificity principle, reinstated encoding contexts provide Since NP was obtained under conditions where PP was better retrieval cues because the practice episodes closely observed earlier, the opposite priming effects for distrac­ resemble the test items that have cued them. Therefore, tors in Experiments 1 and 3 reveal another contextual in­ when distractors appear at test in the same specific context fluence. Presumably, "flipped" items cued retrieval more in which they were ignored during practice, the more that effectively than the changed probes ofExperiment 1. Test responses can be made by retrieval alone, and the more responses for distractors were affected by memory that that performance should be affected by memory for ear­ they had been ignored, yielding NP. lier processing. Thus, "flipped" probes may reveal NP de­ The results bolster earlier claims of the importance of spite the contextual changes that accompany the intro­ contextual consistency for NP. When distractors became duction ofnew items. targets in the specific context in which they were ignored, their processing was impaired, despite changes in the gen­ Method eral context in which selective processing took place. In General. Except for the makeup of IGN probes, in all other as­ Experiment 2, when distractors became targets in a dif­ pects Experiment 3 was identical to Experiment I. ferent specific context than when they were ignored, their Subjects. The 40 new subjects were selected from the same pop­ processing was impaired, but only if the general context ulation as were the subjects in the earlier experiments. in which they had been ignored was maintained. In Ex­ periment 1, when both specific and general contexts Results and Discussion changed, distractor processing was facilitated. RT data were scored and analyzed as in Experiment 1. Since only 4 subjects made any errors, errors were too few to allow analyses. GENERAL DISCUSSION Practice trials. Table I shows that, as before, RTs de­ creased reliably with increases in number ofpresentations In three experiments performed to evaluate the claim [F(5,195) = 34.21, MSe = 458]. The greatest RT reductions that NP is a matter ofmemory retrieval (Neill & Valdes, were between one and two presentations, and there were 1992), it was demonstrated that NP can be both "long small, progressive, improvements between two presenta­ term" and dependent on the number oftimes that distractors tions and eight and nine presentations [F(4,156) = 14.93, have been ignored. Whether distractor processing exhibited MSe = 446]. NP or PP depended on several contextual relationships be­ RTs for nonrepeated items during the second half of tween the sequential environments in which distractors practice were reliably faster than those during the first half, were ignored initially and attended subsequently. NP that indicating nonspecific practice [first halfM = 664 msec is conditional on relationships between the initial process­ vs. second half M = 637 msec, t(39) = 3.52]. However, ing environment and later test circumstances is contrary to during the second half, RTs were faster for repeated pairs the idea that NP reflects temporary inhibition ofdistractor than for nonrepeated items, indicating that performance for information. Rather, NP is best explained as episodic re­ repeated pairs was memory based [nonrepeated M = trieval, wherein retrieval is governed by a host ofprime­ 637 msec vs. repeated M = 607 msec, t(39) = 7.65]. probe contextual relationships. Those contextual factors Test trials. As in Experiment 2, test responses for tar­ can involve memory for temporal information (Neill, gets exhibited Pp, whereas responses for distractors dis­ 1997; Neill et aI., 1992),the prime (Malley & Strayer, 1995) played NP. RTs for ATT (M = 628 msec), IGN (M = and probe components (Tipper et aI., 1991), the predictabil­ 658 msec), and new probes differed reliably [F(2,78) = ity ofprobes from primes (Lowe, 1979), and the blend of 11.96, MSe = 1,082]. Compared with performance for new materials composing the prime and probe sets (Lowe, trials, responses for ATT trials were marginally faster 1979; Moore, 1994). [t(39) = 1.85,p < .10], whereas RTs for IGN trials were re­ Context dependency of NP could be theoretically im­ liablyslower [t(39) = 2.31,p < .05; see also next analysis]. portant. That compatibility between sequential environ­ The 2 X 2 repeated measures ANOVA of number of ments governs the occurrence ofNP supports the claim that practice presentations and ATT versus IGN response con­ NP has adaptive consequences for future processing. Tipper ditions revealed that target responses were faster than dis­ et al. (1991) noted that, in most situations, particular stim­ tractor responses [F(1,39) = 12.63, MSe = 1,824], and that uli are consistently relevant and are selected for response, response condition interacted with number ofpresentations whereas other stimuli are consistently irrelevant and are 442 LOWE

ignored. Consequently, when the same environments are tiveness ofthe different probes in cuing earlier experiences encountered, the same stimuli are again categorized as rel­ of ignoring specific words. Those experiences were better evant or irrelevant and are attended or ignored, respec­ reinstated by the intact S-S associations provided by tively. "Therefore, if the prior experience ofignoring a stim­ "flipped" probes because the specific context in which the ulus [in the same environment] can be reinstated, then distractors were ignored was repeated (Tulving & Thomson, selective processing will be facilitated" (Tipper et aI., 1991, 1973). Performance was impaired because memory that p. 691). Facilitated processing for ATTitems, where across specific words had been consistently goal irrelevant and environments, targets are consistently relevant and dis­ should be ignored was retrieved, thereby interfering with tractors are consistently irrelevant, supports this claim. current responding. In contrast, those earlier experiences Similarly, when that prior experience of"ignoring" yields were not reinstated by the broken S-S connections provided performance impairments, as during inconsistent IGN by changed IGN probes in Experiment 1, and there was no trials, "the very mechanism by which attention to a partic­ effect ofmemory that specific words had been ignored. ular source of information can be maintained effectively Changes in general context also influenced memory ac­ overtime" (Neill et aI., 1992, p. 999) may be revealed. From cess. Targets and distractors became connected to the gen­ this perspective, sustained attention to some information eral background in which those items were differently is an effect (Johnson & Dark, 1986) when current process­ processed. General context affected memory access dur­ ing is guided by the constraints, or results, ofmemory for ing test, for what was retrieved depended on whether or past processing in that situation. not the background was reinstated. In both Experiments 1 When NP depends on reinstatement ofa prior experience and 3, unfamiliar, or new, items were part of test. None­ ofignoring a stimulus, how does context work to yield op­ theless, in Experiment 3, since two thirds ofall the trials posite PPand NP effects? The answer depends on achieving involved familiar S-S connections, practice and test lists an understanding of what has "gotten into" the represen­ were mostly related. Memory for co-occurrences during tations ofthe particular processing episodes stored during earlier processing (Logan & Etherton, 1994) was retrieved. practice, and what is later "taken out" ofthem during test Information that specific words had been ignored and (Logan & Etherton, 1994, p. 1022). A satisfactory expla­ should continue to be ignored was retrieved, yielding NP. nation also requires an appreciation ofwhat subjects learn In contrast, in Experiment 1, only one third oftest trials in­ and remember about the general environment (Chawarski volved familiar S-S connections, and practice and test & Sternberg, 1993), or "list-wide" context (Smith et al., lists were mostly unrelated. Thus, the context associated 1989) when those episodic memories are acquired. with ignoring items during practice was not reinstated Logan and Etherton (1994) consider that instances of during test. Consequently, there were no retrieval cues for processing occasions represent co-occurrences, or partic­ information about what was done with an item during ular combinations ofthe goal, stimuli encountered, inter­ practice. Memories ofpast processing during which spe­ pretations given to the stimuli, and response made. They cific words had been ignored were not retrieved, and NP found that when word pairs were consistently repeated, was not obtained. subjects learned and remembered which words went to­ Beyond implications for theories ofcognition that con­ gether, forming stimulus-stimulus (S-S) connections, and join processes ofattention and memory, the present results the responses made to the pairs, forming stimulus-response indicate that context, by itself, produces degrees ofreiat­ (S-R) connections. They showed that RTs improved with edness that are sufficient to produce qualitatively different practice, indicating that performance relied on memory for priming, or transfer, phenomena. These findings have im­ regular co-occurrences ofword pairings. When transferred portant implications for choosing suitable baselines against to a new environment in which word pairings changed, which transfer across processing tasks and environments subjects were forced to find new bases to support perfor­ should be assessed. That inclusion or exclusion ofpartic­ mance. The search for an alternate platform was revealed ular baseline events can determine type oftransfer is worthy by slowed RTs and quantitatively changed performances. offurther investigation. By applying these notions, one can characterize how the particular co-occurrences that were represented during REFERENCES practice changed during test: (1) Across all experiments, BADDELEY, A. D. (1976). The psychology ofmemory. New York:Basic during IGN trials, established S-Rconnections became un­ Books. reliable; (2) when distractors were paired with novel words BATTIG, W. E, & TOGGLIA, M. P. (1978). Handbook ofsemantic word during IGN probes, the S-S connections were broken; norms. Hillsdale, NJ: Erlbaum. (3) "flipped" IGN probes maintained the established S-S CHAWARSKI, M. c., & STERNBERG, R. J. (1993). Negative priming in word recognition: A context effect. Journal ofExperimental Psychol­ connections; (4) when test involved new items, correct re­ ogy: General, 122, 195-206. sponses were not available from memory and were provided DALRYMPLE-ALFORD, E. C, & BUDAYR, B. (1966). Examination of some by the algorithm; and (5) when there were no new pairs, all aspects of the Stroop color-word test. Perceptual & Motor Skills, 23, test responses had some type of episodic representation. 1211-1214. Within this framework, the opposite PPand NP effects DESCHEPPER, B., & TREISMAN, A. [M.] (1991, November). Novel visual shapes in negative priming. Paper presented at the meeting of the Psy­ for distractors in Experiments 1 and 3 are, perhaps, easi­ chonomic Society, San Francisco. est to understand. Retrieval was determined by the effec- DESCHEPPER, B., & TREISMAN, A. [M.l (1996). Visual memory fornove1 LONG-TERM NEGATIVE PRIMING 443

shapes: Implicit coding without attention. Journal ofExperimental tion priming.Journal ofExperimental Psychology: Learning, Memory, Psychology: Learning, Memory, & Cognition, 22, 27-47. & Cognition, 23,1291-1305. Fox, E. (1995). Negative priming from ignored distractors in visual se­ NEILL, W. T., & VALDES, L. A. (1992). Persistence ofnegative priming: lection: A review.Psychonomic Bulletin & Review, 2,145-173. Steady-stateor decay?Journal ofExperimental Psychology: Learning, HASHER, L., ZACKS, R. T., STOLZFUS, E. R., KANE, M. 1., & CONNELLY, Memory, & Cognition, 18, 565-576. S. L. (1996). On the time course of negative priming: Another look. NEILL, W.T., VALDES, L. A., & 'fERRY, K. M. (1995). Selective attention Psychonomic Bulletin & Review, 3, 231-237. and the inhibitory control of cognition. In F. N. Dempster & C. J. HOUGHTON, G., & TIpPER, S. P. (1994). A model of inhibitory mecha­ Brainerd (Eds.), Interference and inhibition in cognition (pp. 207-261). nisms in selective attention. In D. Dagenbach & T. H. Carr (Eds.), In­ San Diego: Academic Press. hibitory processes in attention, memory, and language (pp. 53-112). NEILL, W. T., VALDES, L. A., 'fERRY, K. M., & GORFEIN, D. S. (1992). San Diego: Academic Press. Persistence of negative priming: II. Evidence for episodic trace re­ JOHNSON, J. A., & DARK, V.J. (1986). Selective attention. Annual Re­ trieval. Journal ofExperimental Psychology: Learning, Memory, & view ofPsychology, 37, 43-75. Cognition, 18,993-1000. LOGAN, G. D. (1988). Towardan instance theory ofautomatization. Psy­ NEUMANN, E., & DESCHEPPER, B. G. (1991). Costs and benefits of tar­ chological Review, 95, 492-527. get activation and distractor inhibition in selective attention. Journal LOGAN, G. D., & ETHERTON, J. L. (1994). What is learned during au­ ofExperimental Psychology: Learning, Memory, & Cognition, 17, tomatization? The role ofattention in constructing an instance. Jour­ 1136-1145. nal ofExperimental Psychology: Learning, Memory, & Cognition, 20, NEUMANN, E., & DESCHEPPER, B. G. (1992). An inhibition-based fan 1022-1050. effect: Evidence for an active suppression mechanism in selective at­ LOWE, D. G. (1979). Strategies, context, and the mechanism ofresponse tention. Canadian Journal ofPsychology, 46, I I-50. inhibition. Memory & Cognition, 7, 382-389. SMITH, M. c., MACLEOD, C. M., BAIN, J. D., & HOPPE, R. B. (1989). MALLEY, G. B., & STRAYER, D. L. (1995). Effectofstimulus repetition on Lexicaldecision as an indirecttest of memory: Repetition and list-wide positiveand negativeidentitypriming.Perception & Psychophysics, 57, priming as a function of type ofencoding. Journal ofExperimental 657-667. Psychology: Learning, Memory, & Cognition, 15, 1109-1118. MAY, C. P., KANE, M. 1., & HASHER, L. (1995). Determinants ofnega­ STROOP, J. R. (1935). Studies ofinterference in serial verbal reactions. tive priming. Psychological Bulletin, 118, 35-54. Journal ofExperimental Psychology, 18, 643-662. MILLIKEN, B., & JooRDENS, S. (1996). Negative priming without overt TIPPER, S. P.(1985). The negative priming effect: Inhibitory priming by prime selection. Canadian Journal ofExperimental Psychology, 50, ignored objects. QuarterlyJournal ofExperimentalPsychology, 37A, 333-346. 571-590. MILLIKEN, B., JooRDENS, S., MERIKLE, P., & SEIFFERT, A. (1995). Se­ TIpPER, S. P.,& CRANSTON, M. (1985). Selective attention and priming: lective attention: A reevaluation ofthe implications ofnegative prim­ Inhibitoryand facilitatory effectsofignored primes. QuarterlyJournal ing. Manuscript submitted for publication. ofExperimental Psychology, 37A, 591-611. MOORE, C. M. (1994). Negative priming depends on probe-trial conflict: TIpPER, S. P.,WEAVER, B., CAMERON, S., BREHAUT, J. C., & BASTEDO, J. Where has all the inhibition gone? Perception & Psychophysics, 56, (199I). Inhibitory mechanisms of attention in identification and local­ 133-147. ization tasks: Time course and disruption. Journal ofExperimental NEELY, J. H., & DURGONO