SEMANTIC PRIMING FROM FLANKER WORDS: SOME LIMITATION TO AUTOMATICITY

We explore under which conditions words flanking a centrally presented digit in the pr ime display can c l ic i t semant ic pr iming on the lexical decis ion to a subsequcnt letter string appearing at fixation about I sec later. No significant priming is found when the prirlc display requires an immediate odd/even class i l i ca t i ono f ad ig i t (Expe r imen t l ) . ade layed reca l l o f ad ig i t (Expe r imen t3 ; . or the detection of an infiequent change fiom thc digit 4 to the lctter A (Exper iment. l ) . I t is only in Ëxper imenr 2. in which norhing is presented at f ixation during the pr-imc display in positive lexical dccision trials, that a positive sen'lantic priming elÈct is f ound. These results arc discussed in the fiamework of quant i tat ive and qual i tat ive l i rn i tat ions to proccssing automut ic i ry.

The literature provides ample evidence that words presented visually within a few degrees from fixation are semantically processed. This is demonstrated with both Stroop-like interference tasks and with priming tasks. A flanker compatibility effect was observed in a modified version of the task devised by Eriksen and Eriksen (1914), in which participants classrfied a target word presented at fixation into one of four semantic categories, while ignoring simultaneously (Shaffer & LaBerge, 1979) or nearly simultaneously (Broadbent & Gathercole, 1990) presented flanker words belonging to the same or a different semantic category. Semantic priming effects were obtained from flanker words presented near hxation (instead ofat fixation as in the standard procedure) in prime displays occurring well before the probe This study was complctcd while Katia f)uscherer was a rcsearch assuciatc of the National Fund ot Scicntific Research (FNRS), Belgium. She is now supported by a post-doctoral grant ofthe Fyssen Foundation. Part ofthis work was financed by an Action de Recherche Concertée (ARC) grant front the Direction Générale de la Recherche Scientiliquc-Communauté Françarse de Belgiquc, and by the Fund fbr Joint Basic Research (FRFC. Convenrion 2.4513.9-5). Experinrcnts I and 2 were prcsented at the 38th Annual Meeting of the Psychonomic Society, Philadelphia. Pennsylvania Novcmber I 997. Wc want to thank Elvira van Eyll for hcr assisrance in collccting data.
The present study is a follow up of our previous work (Duscherer & Holender,2002) based on a procedure set up by Fox (1996). On each trial of Experiment I of Fox (1996) and of our replication thereof, a prime display consisting of a central target digit flanked by a distractor word appearing twice, once above and once below the digit, was followed after a stimulus onset asynchrony (SOA) of about 950 ms by a probe display consisring of a letter string appearing at fixation. The flanker words in the prime display, which were preceded and followed by plus-sign masks, were presented for 150 ms at a distance of 2.4" from fixation. When the probe letter string was a word, it could be semantically related or unrelated to the flanker word. Participants made two consecutive responses. The first response was based on the outcome of a binary odd/even classification of the target digit in the prime display while ignoring the flanker words; the second response was based on the outcome of a word/pseudoword classification of the probe letter string (i.e., a lexical decision). Fox's (1996) predictions were that most participants would be unaware of the identity and meaning of the flanker words and, because these words were presented outside the focus of attention, they would exert a negative semantic priming effect on the probe words. Both predictions were borne out by her data. However, the observation of strong negative semantic priming effects in three out of four conditionsr in Fox's Experiment 1 appeared problematic and surprising to us, because the procedure did not fulfil the conditions we thought necessary for negative semantic priming from unconscious distractor words to occur (Duscherer & Holender, 2002). Here follows a brief summary of our argumentation.
Although doubts about the mere existence of negative semantic priming r ln the studies concerned with negative prirning. at least one. but generally both successivc displays are two-component stirnuli in which one component is a relevani tur.qetihat has to bc attended to and thc other cor.nponent is an irrelevant rli.ttro(br thal has to be ignorcd. We fbllou' the standard terntinology of the ncgative priming literature in calling the lirst display a 7rrùirr, and the second a prabr,. irrespcctive of whethcr both displays, onlv one displa.v'.. or eveu no dis play at all contain two componcnts (or attributes). The standard ternrinokruy of the posirtrc primrng litcrature, which is esscntiallv based on single-conrponcnt stirnuli. is to call rhe llrst stimulus a prirnc and the second a target (see Neely. l99l lbr a rcview). I Strong ncgative semantic prirning eflècts wcrc lbund in the tùo within-language conditiuns and in one ol'the bctween Ianguage condition ol 'Fox's (1996) Expcrinre.nt l. Our rcplicarion involved only a within language condition.
were cast by Fox (1995) and Damian (2000), our appraisal including later studies (Duscherer & Holender, 2002) led us to conclude that negative semantic priming should provisionally be taken as a genuine phenomenon.
Moreover, it appears that a sufficient condition for negative semantic pnming to occur is that a conflicting prime display is followed by a nonconflicting probe display. Therefore, both the persisting inhibition model (c.g., Tipper, 1985;Tipper & Cranston, 1985) and the episodic retrieval model (e.g., Neill, Valdes, Terry, & Gorfein, 1992) can account for negative semantic priming with conscious prime distractors:. However, even a potentially competing distractor would loose its potency if it were presented in perceptually impoverished conditions preventing awareness of its identity and meaning, thereby removing any need for selection. It follows that the only way negative priming from unconscious prime distractors could occur is from a synergy between an act of selection taking place during prime processing and diversion of attention taking place immediately afterwards. As the retrieval process would be disabled in such a case, only the persisting inhibition model can account for negative semantic priming with unconscious distractors. This interpretation is germane to that offered by Tipper (1985) to account for the initial demonstration of negative semantic priming with unconscious prime distractors by Allport, Tipper, and Chmiel ( 1985). With respect to the procedure implemented in Experiment I of Fox (1996), the crucial question is whether an act of selection is needed during prime processing. We surmised that covert naming is the most probable incipient response that would be automatically activated by the flanker words and that covert naming is competing neither with the binary odd-even classification task performed on the prime target digit, nor with the lexical decision performed on the probe letter string. Theretbre, this procedure is inappropriate to generate negative semantic priming in any circumstances. Only positive priming is expected, the magnitude of the effect depending on the degree to which the flanker words are available to consciousness.
The frrst step in our investigation (Duscherer & Holender, 2002) was to replicate Fox's Experiment 1, while correcting for two methodological inadequacies consisting in an imperfect matching of the small subsets of related and unrelated words used to measure pnming and in imperfect balancing of response transitions between the prime and the probe trials. We confirmed ' ,\ thiftl mitittr nodcl ol negiitivc pr-irning. the temporal discrirnination model ol'Millikcn.
S IIM A N7-I C P R I M I NC F' R() M I.' IA N K F,R IY('R1-}.5 that prime words presented 2.4" above and below fixation are mostly unavailable to consciousness, and we failed to get any priming eflèct, as expected. The second step in our investigation, which is reported in Experiment I of the present paper, was to move the prime words closer to fixation, at a distance of 0.8', thus expecting a substantial increase in the availability to consciousness of these prime words, and the emergence of a significant positive priming effect. The first prediction was borne out by the results of Experiment 1 reported below, but the second was not. This absence of pnming is in apparent contradiction with the positive priming effect fbund by Den Heyer ( , 1986 with a procedure somewhat srmilar to ours. The aim of the present study is to investigate the reason why reportable prime words may fail to induce reliable positive semantic priming effects. Before proceeding, a few remarks are in order about the assessment of prirne distractor availability to aweueness. Fox (1996) relied on a surprise question replacing the probe of the last trial of the experiment. The question was whether participants could remember what was presented in the last display. Then, participants were interviewed about whether they had noticed anything above or below the digit in any trial. None of the 19 participants of Fox's Experiment I could name the flanker words of the last trial: only one participant reported awareness of words being presented during digit classification. In both the present and the previous study (Duscherer & Holender, 2002), we relied only on a postexperimental interview to assess participants' awareness of the flanker words.
The major advantage of the postexperimental assessûlent of awareness resides in the fact that it does not modify the economy of the task. The major drawbacks of this method are that it almost certainly underestimates the fiequency with which participants iue aware of the flanker words (see Holender, 1986), and that it does not allow to estimate accurately the proportion of rrials in which participants are aware of the meaning of the flanker words at the time of their presentation. However, it should be noted that assessing awareness on a trial-by-trial basis also poses inextricable problems, especially in situations in which distractor word processing is resource lirnited (see Holender, 1986, especially Section 3.3 about parafbveal processing). We submit that the postexperimental interview is sufficient to distinguish between cases in which there is negligible awareness of the prime words, like in Fox ( 1996) and Duscherer & Holender (2002), and cases in which the proporrion of trials in which participants are aware of the prirne words is substantial enough to wonder why no priming occurs, like in the present study.
In fact, we will argue that flanker words presented 0.8" irbove and below fixation can loose their prirning potency fbr reasons less trivial than rnerc perceptual degradation making most of thern unavailable to awareness. In addition to sensory limitation due to flanker distance from fixationr, there is now mounting evidence that both the capacity demands and the nature of a task performed concurrently with the display of the distractor words can affect the semantic processing of these words, or at least their potency to elicit semantic priming. With respect to quantitative limitation in processing, Lavie (1995;Lavie & Tsal, 1994) suggested that distractor processing is automatic only to the extent to which the relevant task performed on the target does not exhaust all the available capacity, a point that will be dealt with in the general discussion. With respect to qualitative limitation in processing due to the nature of the concurrent task, a long series of studies inaugurated by Smith, Theodor, and Franklin (1983) and Henik, Friedrich, and Kellog ( 1983) shows that positive semantic priming is reduced or eliminated if participants have to search the prime word for a letter (see also Hoffmann & MacMillan, 1985;Stolz & Besner, 1996and see Maxfield, 1997;Neely & Kahan, 2001, for reviews). Similarly, colouring only a single letter instead of the whole word can reduce the Stroop effect (e.g., Besner & Stolz, 1999a, 1999bBesner, Stolz, & Boutilier, 1997). Even more intriguingly, although abolishing semantic priming, the letter search task does not necessarily block conscious access to the meaning of the searched words. Maxfield (1991 . p. 215) reported that "217 of 244 (approximately 89Vo) letter search subjects participating in our experiments have reported an awareness of the word relationships... . Some subjects continued to argue that they must have shown priming effects as they were sure that they were aware of the relationships as they were participating in the experiment" .
The existence of such a dissociation not only challenges the use of the priming paradigm to probe the semantic processing of unattended stimuli, but also the explanations of semantic priming based on a spreading activation process occurring prior to the conscious identification of the prime stimulus. Consequently, it is important to examine in which circumstances consciously identified words would induce either no priming at all, or only small, depleted effects. In Experiments 2 to 4, we investigated the possibility that the absence of semantic priming effect may be due to limitations in the automatic processing of the prirne words generated by the task implemented on the target of the prirne display. rEriksenandEriksen(197.1:sccalsoDriver&Ba1,lis. l99l:N4iller. l99l)havcshounrhat distractor lcttL-rs prcsenled ltvond I fronr a letter targct induccd liltle ur no intcrl'crcncteflects ()n a calcg()nratiort ttsk on thrl tarsct. On the other hand. Broadtrent and Gathcrcolc (1990) ttbserred sctnantic intcrll'rcnce el'lects lirnr di\trrctor nortls prcscnted as lar:rs 2.6' lrorn tlxa t ion. r58 SI'MAN llC PRIMIN(; fRoll4 F l,\NKI'R WORDS Experirnent I ln our previous study based on Fox's procedure ( 1996), we (Duscherer & Holender, 2002) used a prime word distance of 2.4o fiom fixation, creating conditions in which participants werc unaware of the flanker wortls. By using a prime word distance of only 0.8" from fixation in the prescnt experiment. we enhance the sensory quality of the stimulus input. thereby increasing the probability that the prime words could be processed up to a semantic level and access awareness. If this manipulation succeeds, we expect positive semantic priming to occur.

Method
Participonts. 32 undergraduate students at Université Libre de Bruxelles participated as part of a course requirement. Three extra participants were replaced: Two because more than l}Va of their data were eliminated through the cut-off procedure (see below) and one because his error rate on the lexical decision task exceeded 10Vo. All participants had nonnal or corrected-tonormal vision, and for all participants French was their first language. Most of them were in their late teens or early twcnties.
Stimuli. The material used in this and all the next experiments consisted of 64 French semantically related word pairs. All words were one to two syllables long, contained between three and seven letters, and were of a relatrvely high fiequency (occurrences per rnillion words) in French according to the BRULEX lexical database (Content, Mousty, & Radeau, 1990). Sixty-four unrelated word pairs were created by repairing randomly the first and the second member of the 64 related word pairs, and by coffecting fbr any remaining association. Two additional sets of 64 French words were selected, one being matched in fiequency, number of letters, and number of syllables with the 64 probes, and the other being matched with the 64 prirres. The first set was used to generate 64 pronounceable pseudowords by changing one or two letters in each word. The words of the second set were used as primes for the pseudoword probes.
We checked that this material yieldcd a substantial priming efïèct in a preliminary experiment that followed the design and procedure described bckrw, except that the prirne word was presented only once at llxation instead of twice (once above and once below ïlxation); and the SOA betr.veen thc prime and the probe was constant at 1,000 ms, as no response was made to the prime display. We fbund a positive semantic priming effect of 47 rns (SD = 27). This difference between the mean RTs fbr the related (M = 555 rns. 51) = 62) and unrelated (M = 602 ms, SD = 58) word pairs was significant, F( l, 151= 47.97, MSE = 363, p <.001. The priming effect was slightly larger in the second (52 ms) than in the first (40 ms) half of the experiment, but this difference was not significants.
Apparatus. The experiments were designed using Micro Experimental Laboratory (MEL; Version 2.01) software (for a descriptive article, see Schneider, 1988). Stimuli were presented on a NEC Multisync XElT colour monitor controlled by a Pentium lBM-compatible computer, which also recorded the RTs in milliseconds via an MEL manual response box.
Design and procedure. Participants perforrned two successive tasks on each experimental trial: During the prime display they categorized a singledigit target as odd or even; during the probe display they performed a lexical decision task. The same responses (Buttons I and 5 of the MEL response box) were used for both the prime digit classification and the probe word or pseudoword classification. The response mapping for the word or pseudoword classification was the same for all participants: pseudoword-left, wordright, while the response mapping for the digit classification (odd/even) was counterbalanced between participants. The prime display was composed of a single-digit |arget (4, 5, 6, or 7) presented at fixation and of a distractor word presented twice, once above and once below fixation. At an average viewing distance of 60 cm, the center-to-center distance between the central digit and either distractor word was 0.8". The probe display consisted of a centrally presented letter string (a word or a pseudoword). When the probe was a word, it could be either semantically related or semantically unrelated to the parafoveal prime distractor. AII stimuli were presented in light grey on a black background, using the uppercase standard font of the computer. At a viewing distance of 60 cm, the visual angles subtended by the entire prime display were 2.2" in height, with each character subtending 0.52" in height in both the prime and the probe displays.
One experimental trial comprised the following consecutive events: (a) A 'The motc contmon proccdure in thc scrrantic priming literature is to avoid within-participant repctition of anv stimulus. Hou'eler. if no repetition of the probc words is alloucd. the set of words to bc uscd in the relatcd trials has to bc rnatched vcry carefully with the set of u,ords to be used in thc unrelated trials. a nrethodolo-rrical difTicLrlty which we identifled (Duschcrcr & Holerrder. 2002) as a potential artctàct rn For's (1996) procedure. Allouing the same words to scrve in hoth thc rclated;rnd unrclated conditions overconlcs this dil-ficulty. The reason fbr arrlrding stimulus repetition in prinring crpcrinrents is prrtly stemnring from the lear that the long livecl repctition prrming clflct shor,"n in scrial lerical dccision tasks (c.g.. Scarborgucn. Cortese.&Scarborough. 1977)coultl erentuallvhamperthcshort-liredscmanticprimingctÈct. Subsequent stutlics shou ecl this lcar to be littlc justitied becausc associativc prirnine and rcpcti lion prirnint harc adtlitire cl'lects on 1.rc-r'lirrmarrcc. at lelst as krng as thcre is onlv a sinulc rcp ctition ot tltc' prinre and thc-probc nortls (Derr Hcycr ct al. 500-ms flxation display, consisting of a central plus (+) sign flanked above and below by a row of seven plus signs that occupied the locations where the prime words would be presented; (b) a black screen for 100 ms: (c) a 150-rns prime display; (d) a 100-ms masking pattern, identical to rhe initial fixation display; (e) a black screen lasting until the digit classification response, or lasting for 2,000 ms if no response was detected; (f) another black screen for 300 ms after the response to the digit; (g) the probe letter string lasting until the lexical decision, or lasting fbr a maximurn of 2,000 ms; and, linally, (h) a 2,000-ms black screen until the fixation display of the next trial. The set of 64 related word pairs was split into two subsets of 32 pairs, each subset being matched as closely as possible in terms of letter length, syllable length, and frequency. The split of the set of 64 unrelated word pairs was fïlly determined by that of the related word pairs, because the probes had to be the same in the corresponding subsets of related word pairs and unrelated word pairs. Four lists of 128 trials were built according ro the following rules. List I contained one subset of 32 related word pairs, the subset of unrelated word pairs containing the rernaining probe words, and the full set of 64 word-pseudoword pairs. List 2 had the other subset of 32 related word pairs, the other subset of unrelated word pairs, and the same full set of 64 word-pseudoword pairs. In each list, half the probes of each type-related, unrelated, and pseudoword-were preceded by an odd digit (equally often 5 and 7), and the other half were preceded by an even digit (equally often 4 and 6). An imporlant constraint was that any specific probe word was preceded by the same digit in its two presentations. The salne constraint was applied to the two presentations of the pseudoword probes. List I' and List 2' were derived from List I and List 2 by crossing the digit-probe pairing. The last stage in list construction was the pseudorandomization of the 128 stimuli in each list with the constraint that there were never more than three consecutive trials of the same kir.rd in tenns of the outcome of either the odd or even digit or the word or pseudoword classification. The resulting sequence of trials in each list was the same for all participants. A practice block of 32 trials containing no related word pairs was also constructed. The words and pseudowords used in this practice block were different from those used in the experirnent.
Each participant was tested individually in one session of about 40 rnin, consisting of one practice block of 32 trials followed by two lists of 128 trials. Each list was divided into two blocks of 64 trials with a rest period between. Two warm-up trials were added at the beginning of each block. The order of the two lists, the order of the two blocks within the lists. and the response mapping fbr the digit classification (i.e., odd-left. even-right, or vice versa) were counterbalanced between 8 participants. A total of 16 participants were needed to fully balance the design; 8 receil,ing Lists I and 2 and 8 receiving Lists I'and 2'. The response rnapping for the word or pseu-doword classification was the same for all participants: pseudoword-left, word-right. Participants were instructed to respond quickly and accurately on both the prime digit and on the probe letter string. After the experiment, participants completed a questionnaire in which they were asked if they had noticed the presentation of the flanker prime words and the existence of any associations between the prime and the probe words.
Data analysis. For each participant, we flrst computed the mean RT and the standard deviation for all the responses tàlling in the 1-to 2000-ms time window for both the digit classification and the lexical decision task. Then, for each task, RTs exceeding three standard deviations above and below the mean RT were eliminated from further analysis. Error rates for both the digit classification and the lexical decision tasks were computed on the remaining trials in each condition. Mean RTs for each task and each condition were computed only for trials in which participants made an error neither in the digit classification nor in the lexical decision. Participants having more than 107o of their data eliminated through this cut-off procedure were replaced. Remaining participants having more than l07o of their data unavailable because oferrors in either the digit classification, the lexical decision, or both, were also replaced. The data of main interest is the priming effect computed by subtracting the mean RT for related from the mean RT fbr unrelated trials. If anythrng, we expect only a positive semantic priming effect to occur. However, in order not to bias the issue, we performed a paired sample /-test between primrng conditions in a bilateral way. Although the usual p < .05 probability of making the Type I error is used as a criterion of significance, the exact values of p are also reported. In addition, we report the magnitude of the priming effect needed to reach a power I -Ji =.80. No analyses were conducted on the error rates as those were generally very low (cf.

Results and Discussion
The cut-ofï procedure entailed an overall elimination rate of 3.07o of the trials. Table I shows the average and the standard deviation of the individual mean RTs, and the average of the individual error rates fbr each type of probe display-semantically related, semantically unrelated, and pseudow6ld5-45 well as the overall results for the lexical decision task. In the digit classification task, the mean RTs for the odd and the even digits were 593 ms (SD = 722, enor rcte = 2.0Vo), and 601 ms (SD = 123, error rate = 2.2c/o), respectively. The 4-ms (SD = 19.9) positive priming effect computed by subtracting the mean RT for the related (M = 615 ms) from that of the unrelated trials (M = 619 ms) was not significant, (31) = 1.20, p =.24. Note that this failure to reach significance was not due to the repetition of the word material, as no significant priming effèct was observed either in the first (-2 ms; (31) = .03,p = .98) or in the second (8 ms; r(31) = 1.52, p = .14) half of the experirnent. A priming eft'ect of 10 ms in absolute value was needed to reach a power of .80. Out of 32, 19 participants reported they could ofïen read the prime words; 13 of thern even noticed that sometimes the prime and the probe words were semantically related.
As expected, the reduction of flanker word distance from fixation increased the availability of the prirne words to awareness: While not a single participant had been aware of the flanker words presen Ied 2.4' from fixation in Experiment I of our previous study (Duscherer & Holender, 2002), more than half of the participants in this experiment were aware of the fact that flanker words were presented 0.8' fiom fixationo. However. while we succeeded in increasing the availability of the prime words to awareness, we fàiled to get the expected concomitant positive priming effect.
Why did the reportable flanker words fail to induce a signif'rcant positive priming el1èct? One trivial possibility is that only a very srnall proportion of these prime words was actually available to awareness. However, thc l9 participants who claimed awareness of the flanker words were confldent about having read many of thern. Yeq neither the subgroup of 13 parlicipants who were also aware of the possible semantic relation between the flanker words and the probe words, nor the subgroup of 13 participants not repofiing any awareness of the flanker words, did show significant prining effects (nonsignificant effects of 6 and 8 ms, respectively). Hence, it does not appear that the small, I We uant to strcss that whilc cxtretne caution should prcvail lirr assessing palticipants' ?//?awareness through a posterpcrinrcntal report (cl'. Holenclcr. l9tl6). r,"e 10 t[l\1 our participaltt\ whcn thcy state to bc uîurc ol llankcr u'ords. lrtd evcn nlore so. uhcn the.l arc ablc trt reD91 a lar-gc number ol'the m. nonsignificant priming effects of Experiment I can simply be explained by insufficient processing of perceptually impoverished flanker words.
As noted in the introduction, positive semantic priming was obtained by Den Heyer ( , 1986 with prime displays similar to ours in terms of the distance liom fixation of the flanker words. However. there are two major differences between the two experimental set-ups. In Den Heyer's situation, the flanker words remained visible for more than 1000 ms, first with the central position left vacant for the initial 550 ms, and then concurrently with the central target letter string staying on until the response. In our Experiment 1, the masked flanker words were displayed for only 150 ms, and followed by an unflanked probe letter-string after an average SOA of about 950 ms. In addition, the central position of the prime display was occupied by a digit calling for an overt classification response. As both the nature and the capacity demands of the task perfbrmed during the prime display can eventually lead to a dissociation between the identification of the meaning of the distractor words and their potency to induce semantic priming efïects (Maxfield, 1997), the following three experiments were aimed at disentangling the reasons why positive semantic priming was obtained by Den Heyer, but not by us in Experiment 1.

Experiment 2
In Experimenl 2, we simply suppressed the target digit from most prime displays, which makes our situation more similar to that of Den Heyer, while keeping all the other procedural details identical to those of Experiment L However, in order to encourage participants to maintain fixation, we presented a central digit in one third of the pseudoword trials (i.e., in one sixth of the trials, overall) and asked participants to keep fixation in order to be able to report any occurrence of such a digit. The digit recall-prompt always followed the lexical decision on the probe letter-string.

Method
Participants, stimuli, and appuratus. Thirty-two students from the same pool as in the other experiments were selected. None of them were involved in any other experirnent. Two extra pafticipants, having more than 10o/c of their data eliminated through the cut-off procedure were replaced. The same stimuli and apparatus were used as in Experiment 1.
Dcsigrt tutd procedutr. The only difïerence with the previous experiment is that no central digit was presented in the prirne display except in one sixth of the trials. Digits were never presented in the critical trials leading to a positive lexical decision, but only in one third of the non-critical trials in which the probe was a pseudoword. We used a 1000-ms constant SOA between the prime and the probe because there was never any immediate response required to the prime display. Participants were asked to keep fixation rn order not to miss any digtt that might appear in the prirne display. They had always to report the digit after the lexical decision rvas completed. They were asked not to slow down in their lexical decision and to wait until prompted by the experimenter fbr making their digit report.
Data analysis. The main data analysis was the same as in Experiment l, except that the pseudoword data conesponding to trials in which a digit was presented in the prime display were not taken into account. Hence, the mean RT for the pseudowords was estimated from two thirds of the trials calling for a negative lexical decision.

Results and Discussion
The cut-off procedure entailed an overall elimination rate of l.4Vo. Digit report was nearly perfect, being of 98.9Vo. The results of the lexical decision task are shown in Table 1. A significant positive semantic priming effect of 24 ms (SD =29.1), /(31) = 4.66,p =.0001 was found, the mean RTs being 583 ms and 607 ms for the related and unrelated trials, respectively, with a significant effect emerging both in the first (18 ms; (31) = 2.8,p =.008) and in the second (31 ms; (3 I ) = 4.9, p =.0001) half of the experiment. A priming effect of 15 ms in absolute value was needed to reach a power of .80. All participants reported that they could have read at least some of the flanker primes and all but three had noticed that sometirnes the prime distractor and the probe were semantically related.
We observed a positive semantic priming from flanker words close enough to fixation to be available to awareness. However, this was achieved only by making radical changes in the prime display. In Experiment l, the prime flanker words were always accompanied by a central digit that had to be attended to. This digit had to be identified and mapped onto a response according to an unfamiliar rule, and the classification response had to be executed. By contrast, in Experiment 2, the fixation position of the prime display was leff vacant on all the critical trials in which the probe was a word. Hence, no infbrmation had to be processed and no response had to be executed during the prime presentation of these trials. To determine which of these two differences between the critical trials of Experiments I and Experirnent 2 are responsible for the presence or absence of a significant positive priming effect, we carried out two additional experiments. \l:,11,\i\'7 l( l'RliVIlNG I"ROAI f l,\NKt.R lt?)RD"S Experiment 3 In Experiment 3, a single digit (out of a set of four) was always presented at fixation in the prime display. Participants had to identify and to memorize this digit, as they could be asked to report it later in the trial, after the lexical decision was completed. Hence, in Experiment 3, the prime display was identical to the one used in Experiment l; still, like in Experiment 2, no overt response had to be executed during the prime display. With respect to the overall incentive of keeping fixation as requested, Experiment 3 is similar to Experiment 2 in that participants had to report the prime digit after completion of one sixth of the trials (i.e., one third of the negative lexical decisions). However, in Experirnent 2, the uncertainty was about whether a digit would appear at all, whereas in the present experiment, the uncertainty was about whether recall of the digit would be requested at all.

Method
Participants, stimuli, and apparatas. Thirty-two students were drawn from the same pool as in Experiment l. None of these students participated in any other experiment based on the same material and procedure. Two extra participants, exceeding 10Vc of errors in the lexical decision task were replaced. The same stimuli and apparatus were used as in Experiment l.
Design, procedure, und data analysis. The design and procedure were sirnilar to those of Experirnent 1, except that no immediate classiflcation response was required to the central digit. Participants were asked to hold the digit into memory because they could be asked to report it afïer the lexical decision was completed. This happened in one third of the trials in which the probe was a pseudoword. On those trials, a prompt consisting of the word rttppel (English 'recall') was shown immediately after the probe response was made. The data analysis was the same as in Experiment I (i.e., all the pseudowords were included).

Result.y und Disc'ussion
The cut-ofï procedures entailed an overall elimination rate of |.6Vc of the trials. Digit recall was nearly perf'ect, being of 9'7.1%,. The results of the lexical decision task are shown in Table l. The 5-ms (SD = 16.4) difference bctween the related (M = 610 ms) and the unrelated trials (M = 615 ms) was not significant, (31) = 1.70, p = .10. No significant priming was observed eithcr in the flrst (5 ms; r(31)= .95. p = .34) or in the second (6 ms; r(31) = 1.38,p =.17) half of the experiment. A priming effect of 8 ms in absolute value was needed to reach a power of .S0.Twenty-two participants out of'32 reported that they were sometimes able to read the flanker words in the prime display; l6 of thern noticed the semantic relatedness existing between some of the prime and probe words.
The results of Experiment 3 are similar to those of Experirnent I both in terms of the proportion of participants aware of the sernantic prirning procedure and in showing a small, nonsignificant positive semantic prirning effect.
Thus, it appears that semantic priming is equally depleted by the execution of a digit categorization response irnmediately on presentation, like tn Experiment l, and by memorization of this digit fbr later recall, like in the present Experiment 3.

Experiment 4
Experiment 4 is an attempt to render the prirne target task even less demanding. Instead of one out of four possible digits, the same digit 4 was presented at fixation in all the prime displays, except for one third of those preceding a pseudoword probe, in which the letter A was presented instead. Participants had to detect these occurrences and to name the letter A immediately on presentation while trying not to slow down in their lexical decision to the probe. Like Experiments 1 and 3, Experiment 4 involved a visual event consisting of a single character presented at lixation during all prime stimuli. However the discrimination of A from .l probably requires less processing resources than the identification and memorization of one among four digits. Like in Experiments 2 and 3, no response was required to the prime targets of the critical trials of Experiment 4. Overall incentive to keep fixation was also the same as in Experiments 2 and 3 because the letter A occurred in one sixth of the trials overall, that is, in one third of the noncritical trials in which the probe was a pseudoword.

Method
Particilttrnts, stimuli, and dpporatus. Thirty-two students were drawn from the same pool as in Experiment l. None of these students participated in any of the other experiments. The same stimuli and apparatus were used as in Experiment 1.
Design and procedure. Instead of one out of fbur possible digits as in Experirnents I and 3, only the digit 4 was presented at fixation during the prime display except for one third of the pseudoword trials in which the let-ter A was presented. Participants were instructed to name this letter as fast as possible as soon as it was presented. They were asked to try not to miss any letter and not to slow down on the following lexical decision. No response, either immediate or delayed, was required to the prime displays in which the digil I was presented.
Datu analysis. The main data analysis was the same as in the other experiments except that, Iike in Experiment 2, the pseudoword data corresponding to the noncritical trials in which the letter A was presented in the prime display were not taken into account. Hence, the mean RT for the pseudowords was estimated from two thirds of trials in which the probe called for a negative lexical decision.

Results and Discussion
The cut-off procedures entailed an overall elirnination rate of 1.4Vo of the Irials.99.2Vc of the occurrences of letter A were detected. The results fbr the lexical decision are shown inTable 1. The 3-ms (SD = 16.6) priming effect computed by subtracting the mean RT for the related (M = 517 ms) from that of the unrelated trials (M = 580 ms) was not significant, (31) = 1.18, p = .25. The f,rrst hall of the experiment produced a significant priming effect of 8 ms, l(31) = 2.52, p = .017, whereas the -1 rns-efIèct of the second half was not significant: (31) = .16, p =.873. A priming efÏect of 8 ms in absolute value was needed to reach a power of .80. Eighteen participants out of 32 reported that they were sometirnes able to read the flanker primes; fifteen of thern noticed thc semantic relatedness existing between some of the prirre and probe words.
Again no significant positive semantic priming fiom flanker words close to fixation is obtained, although. like in Experiment 1 and 3, nearly half of the participants were aware of the semantic relations sometimes existing between the prime and the probe words.

Comparison Between Experirnents
Thc ibur experiments reported in this article all have a common underlying structure, which corresponds to the stanclard procedure used in the investigation ol sernantic priming in the lexical dccision task, except that the prime words were presented twice as flanker distractors instead of once under the fbcus of attention (i.e., at frxation). They difTèr in terms of the supplementary task performed on a prirne target presented at fixation. Table 2  .1. .j. 6. 0r 7 1 I 00ci ) Idenrillcalion + immediare clasificarion I i00.,i I Blmk{l00crt lv'lonitoring(100.Çitioradilit .,l. .i. 6. or / ( 100? r ldentillcittion and memoflzalron r l00q ) N'lonitoring t l00i' r for rn,1 li1001r Nolc. The percentagcs in parcntheses indicatc the frcqucncy with which each typc ol printe targct is prescntcd tnd the liequency with uhich cach type of process and action are called for in dcaling wrtn thc pnme targct. of (a) which possible stimuli are presented at fixation in the prime display, and (b) which processes and which immediate or differed actions are called fbr in dealing with the prime target according to the specific instructions given in each experiment. The left part of Table 2 provides this information fbr the task as a whole; the right part of the table restricts this information to the trials in which the probe is a word, that is, the trials from which the semantic priming effect is computed.
The upper part of Figure I shows the average RT obtained by collapsing over the related and unrelated probe words in the four experiments, disregarding the pseuclowol'd trialsT: the lower part of Figure I shows the corresponding mean priming effects. In both graphs, the error bars indicate the overall 957c conhdence intervals, computed following Equation I of Loftus and Masson (1994) for between-subject designs, which is based on the pooled estirnate of the within-condition variance (with 124 dfl. Horv do the difïerent prime target tasks afïect the absolute level of performance in the subsequent lexical decision on the probe? Examination of the upper part of Figure I shows that there is no statistical difference between the mean RTs for the positive lexical decision in Experiments I and 3, both experiments requiring the identification of one out of four possible prime target digits, whereas the mean RT is significantly shorter in Experiment 4 'We prclèr 1o evaluatc thc absolutc' levcl ol'pcrlornrance orr trials in uhich thc targcr is l word instt'ird ol on all trials bccausc the negatile lcxical dccision is based on dilfcrcnt propor, tions ol pse'udowords in dil'lircnt cxperitnents t100? in Experintents I and 3 vs.67% in Erpcritttcnts I ancl -1). and bccausc prirrring is assessctl on uord probes onl1,. In anv case. an anal\':is pcrlirrrued on thc overall RTs (including the pscuclouortls.; shoun in Tablc l. vicltlctl thc sanrc rcsLrlts. requiring a simpler discrimination between one possible digit and one possible letter. Experiment 2, requiring to monitor the empty prirne target position for identifying infrequent (ll7o of the trials) target digits, occupies an intermediate, ambiguous, position in having a mean RT statistically different neither from the slower mean RTs in Experiments I and 3, nor fiom the faster mean RT in Experiment 4.
The grouping of the experiments in terms of priming is, however, difïèrent from that based on overall mean RT for the positive lexical decision. As can be seen in the lower part of Figure l, significant semantic priming is found only in Experiment 2, whereas there is no significant priming in Experiments 1,3, and 4. It has to be noted that the very same grouping of experiments is also found with respect to awareness of the prime words and of the prime-probe contingency. In Experiment 2, all the 32 participants were aware of the prime flanker words, 29 of them also being aware of the primeprobe contingency. By contrast, only 19, 22, and 18 participants were aware of the prime flanker words in Experiments l, 3, and 4, in that order, with, respectively, 13, 16, and l5 ofthem also being aware ofthe prime-probe contingency. It thus appears that the reduction in processing of the prime distractor is observed in both the slower (Experiments I and 3) and the faster (Experiment 4) positive lexical decision, whereas Experiment 2 occupying an intermediate position shows both positive semantic priming and a higher level of awareness of the prime distractor words and of the prime-probe contingency. Hence, we can rule out that the presence or absence of significant priming eff'ects stem from differences in the absolute level of performance for carrying out the lexical decision on the probes.
Before interpreting these results, two remarks are in order about the failure to get signihcant priming effects in Experiments 1, 3, and 4. First, this is not due to a lack of power. Priming effects of l0 ms in Experiment 1 and of 8 ms in Experiments 3 and 4 are needed to reach a power of .80 (compared to 15 ms in Experiment 2). Second, we do not take this lack of statistical significance as evidence that there is no priming at all. In the complete absence of priming, small effects of both signs would have been found. Rather, we think that the positive effects of 4, 5, and 3 ms found respectively in Experiments l, 3, and 4 reflect genuine positive priming, the effects being simply too small to reach significance. Actually, in averaging over the 96 participants of Experiments 1, 3, and 4, the 4-ms overall effect was significant, (95)=2.35,p<.05.

General Discussion
We investigated under which conditions semantic priming could be obtained from prime words presented at a distance of 0.8" from hxation. We observed no significant priming eflèct in Experiment l, in which participants had to classify a target digit appearing at fixation in all prirne displays. Similarly, no significant priming efïects were fbund in Experiments 3 and 4, which cliflèred rnainly liorn Experirnent I by requiring no immediate response to the prirne target when the probe was a word, but in which a target character (generally a digit, but sometimes a letter in Experiment 4) was always presentecl at llxation. It was only in Experiment 2. with no target at all presented at flxation in the prime display of the critical trials in which the probe was a word, that we u'ere able to get the expected positive semantic priming effect.
The positive semantic priming effect found in Experimenl 2 conhrms the results obtained by Den Heyer ( , 1986. This result is little unexpected in view of the close sirnilarity between the procedure used in Experiment 2 and that used by Den Heyer and collaborators. Experiment 2 can thus be considered as a baseline showing that our procedure is appropriate fbr generating positive semantic priming, at least in these specilic conditions. In all other experiments, a supplementary task perfbrmed on a prime target presented at fixation seems to impede, or at least to reduce, this priming effect. One tentative explanation of this observation-stemming from the studies centred on qualitative limitations to automatic processing-is that no semantic priming occurs because participants' attention was not directed to the meaning of the prime words at the moment of their encoding (e.g., Srnith et al., 1983). An alternate point of view we want to consider is that there may also be quantitative lirnitations to automaticity, that is, the absence of semantic priming efïects from reportable distractor words may be explained by a lack of processing resources during the encoding of the prime words.
In fact, Lavie (1995, Lavie & Tsal, 1994 suggested that disrracror processing is automatic only to the extent to which the relevant task performed on the target does not exhaust all the available capacity. As long as target processing leaves spare capacity, processing of the irrelevant inforrnation is automatic in the sense of being irrepressible, that is, not under voluntary control. In contrast, iI insufficient l'esources are available to guarantee both the processing of task relevanl and task irrelevant information, resources are ailocatecl in priority to thc relevant target processing. Until now. Lavie has been mairrly interested in tcsting the irnpact of pen,eptual lottd on perceptual processing cicllned as proccssing leading to stimulus identification. Her research strategy consists in rnirnipulating thc perceptuai load of the targct in Ill t72 S E M A N T I C P R I M I N G I.' RO h,I F' LA N K I.:R IYORI).S modified versions of the letter flanker task of Eriksen and Eriksen (19j4), while keeping all the other task requirerrents constant. Distractor processing is thus considered more unlikely under conditions of high perceptual load (e.g., with complex stimulus displays or with resource-demanding target tasks) than under conditions of low perceptual load (e.g., with simple stimulus displays or with target tasks requiring only shallow perceptual processing). Indeed, Lavie succeeded in eliminating distractor interfèrence effects by increasing the number of nontarget elements in the attended region of the display, or by varying the processing requirement of the attended target-nontarget discrimination task (Lavie, 1995(Lavie, , 2000Lavie & Cox, 1997;Lavie & Robertson, 2001;Rees, Frith, & Lavie, 1991: see also Handy, Soltani, & Mangun, 2001;Kumada & Humphreys, 2002).
Both the reportability and the prirning potency of the distractor words is increased when less-with no prime target in Experiment 2-rather than more-with digit classification in Experiment l-processing capacity is diverted by the relevant task in the prime display. More specilically, the comparison between Experiments 2 and 4 provides some evidence for the role of perceptual load in prime processing. As can be seen in the right part of Table  2, both experiments require monitoring the prime display for detecting a target, but differ for the relevant experimental trials (i.e., the word probe trials) in terms of the perceptual load of the prime display: While no stimulus is presented at fixation in Experiment 2, in Experiment 4 the digit 4 is always presented at fixation. Thus, it is ternpting to attribute both the presence of positive priming and of a high level of report of the prime to the low perceptual load of the prime display in Experiment 2, whereas both the absence of priming and the depleted level of report of the prime flanker words would be attributed to the higher perceptual load of the prime display in Experiment 48. This tentative account of the results is consonant with other studies tiom the literature showing that primes that are reportable at the tirne of their presentation may cause no immediate positive semantic prirning (Maxfield, 1997), no delayed repetition priming (Stone, Ladd, Vaidya, & Gabrieli, 1998, Experiment l), and leave poor episodic memory traces (Stone et al., 1998, Experiments 2 and 3).
By contrast, in our previous work (Duscherer & Holender, 2002) with flanker words farther away from fixation, the manipulation of the capacity x Besides the requirement to proccss an additional stimulus at llxation. hoth cxperintents di1'fèr in the spccilic nature ol'the prirne task. Hoscvcr. it has to bc noted that nronitoring a blank fleld for the appcarance ol'a digit (Experinrent 2) rs probably less capacitl' demanding thiin cliscriminating the presentation ol'thc lctter ,1 from that of the digit J (Expc.rinrenr .l) and thar no signilrcant differencc in the ovelall lcvcl of pc-rlirrrlance tbr the positive lerical dccisiun is observed betrveen thcse experiments (scc upper part of Figure I ). that has to be invested on the prime target had no influence on priming, which was null in Experiments I and 2, and had little influence on flanker word reportabilityu. Hence, in keeping the other parameters of visual presentation constant (i.e., flanker duration, preand postmasking), increasing the distance from fixation of the flanker words from 0.8" (Experiments I and 2, present study) to 2.4" (Duscherer & Holender. 2002) suffices to shifï the processing of these words from the resource-limited to the datalimited region of the perfbnnance-resource function (see Norman & Bobrow,. Before concluding, we have to discuss two apparent discrepancres between some of our results and some of the results obtained by Mari-Beffà, Fuentes, Catena. and Houghton (2000) in their Experiment 1. Their procedure was similar to that of our Experirnent 1, except that both the prime and the probe display consisted of a centrally presented letter string (instead of a digit here) flanked above and below by a distractor word repeated twice, at a distance of 0.95" frorn fixation (comparable to the 0.8' used here). When the probe target was a word, it could be semantically related or unrelated to either the prime target or the prime distractor. The task on the probe target was a lexical decision; the task on the prime target was a lexical decision in one group of participants and a letter search in the other group. There was a strong 35rns positive semantic effect stemming fiorn the prime target requiring a lexical decision that was reduced to a non-signiircant l5-ms eflèct in the letter search condition. This is just another confirmation of the fàct that semantic priming stemming fiom an attendecl prime word can be reduced or abolished if this word is searched fbr a specific letter (see also Hoffman & MacMillan,198,5,Experirnents ? and 3.). Turning now to the results we want to discuss, there was a significant 17-rns negotivc semantic priming eff'ect stemming fiorn the prime clistractor word when a lexical decision was perfbrmed on the prirne target, that reversecl into a strong positive semantic priming effèct of 39 ms when a letter search was perfbrmed on the prime target.
llouerer. all thcsc participant\ attrihutccl thcrl ahilitl" to rcad sontc plilne uorcis to thcir occa sional checlins lirl thc paralirreul contcnt thlough sazt-shil'ting. All ol these participants \\'crc cortrinccd thal \\ithout shiliing gazc thcr could not hare hcconrc anare ol'thc prcsortution of l'lankcr uortls. \&treit not lhrrr thcsc parlicipar)ts'conrnrcntarics gathered in the postcxpr-li nrerttal itttcrricw. rntl considcrins thal onl-\' lTrrr ol thc trials inrolrc.d I to-be-rcpoflcd targct digit. Lhc utkrplion ol srrclt a \trttes\ \\()ul(l protrablr hare pas:cd ultnoticr.d br us. In thc plc \e rt \tuLl\. u ith I'lrtnkct uords close r to llrltion. no participants in Erpc-r'irlcnt l. or in artv ol' thc irtltt'r.1 crpctinrcnls. crcr nuttlc \Lrclr conmrcnls lirr.iustil'ring thcil abilitr'to reud sonlc llrllkL'r \\ofds anel t0 hccotnc r$arc ol'thc printtt Drohe c0ntiltentr. effect stemming from the flanker words when the prime target was searched fbr a letter in Experiment 1 of Marf-Befïa et al. (2000) and the absence of such priming when the prime target digit was classified as odd or even in our Experiment L One possible explanation for this discrepancy lies in the f'act that our pairs of words were related only by association whereas those of Marf-Beffà et al. were related both by association and by their semantic category membership. Maxfield (1991 , see also Smith, Bentin, & Spalek, 2001 ) mentioned unpublished data from her own research showing that searching the attended prime for a letter abolishes semantic prirning completely when the prime-target relation is exclusively associative, whereas it only reduces semantic priming when this relation is semantic/categorical. Another possible explanation is in terms of the capacity demands of flanker words processing. It is likely that less resources are required for word processing in Experiment I of Mari-Beffa et al., in which all the words belonged to four semantic categories used repeatedly, with each individual words repeated 5 or 6 times, whereas disparate words were repeated only twice in our Experiment I (see Broadbent & Gathercole, 1990). Moreover, the prime display was easier to process in the experiment of Mari-Beffa et al., because it was presented unmasked until the participants responded to the prime target, than in our experiment, in which it was it was presented for only 150-ms and masked.
The second apparent discrepancy is between the negative semantic prilning effect obtained when a lexical decision was performed on the prime target in Experiment I of Mari-Beffa et al. (2000) and the positive semanric priming effect found both when the prime target was searched for a letter in Experiment I of Marf-Beffa et al. and when the prime target position was monitored for the presence of a digit in our Experiment 2.
To explain this discrepancy, we shall rely on our previous analysis (Duscherer & Holender, 2002) showing that a necessary condition for ncgative semantic priming to occur is that an act of selection takes place dunng prime processing, in order to prevent the prime distractor from controlling action. This is clearly demonstrated by the fàct that when a prime target word has to be selected fiom an adjacent distractor, in order to be recalled later, either no priming or negative semantic priming is found (Ortells, Abad, Noguera, &Lupiâfre2,2001;Ortells & Tudela, 1996, Experiment 2), whereas either no priming or positive semantic priming is found when participants simply have to pay attention to the prime word, without being instructed to perfonn any task on it (Fuentes, Carmona, Agis, & Catena. 1994;Fuentes & Tudela, 1992;Ortells & Tudela, 1996, Experiment 1).
Thus, for explaining negative semantic priming in Experiment I of Mari-Beffa et al. (2000) with a lexical decision perforrned on the prime target, we have to assume that sometimes participants could not help performing the lexical decision on the prime distractor 3s vygllttt, which entails that they had to prevent this irrelevant covert "word" decision fiom controlling the response to the prirne target. In contrast, when the prime target had to be searched for a predesignated letter in Experiment I of Mari-Beffa et al. or when the prime target position was monitored for the presence of a digit in our Experiment 2, the most probable covert response elicited by the distractor word, if any, would be a naming response. As this incipient naming response is not assumed to compete with the binary decision on the prime target, no negative semantic priming is expected in such conditions. ln these conditions, only positive semantic priming is expected, and actually found, provided the perceptual load of either the prime distractor words (in Experiment I of Mari-Beffa et a1.) or of the prime target position (in our Experiment 2) is not too high. lf, however, the task on the prime target is more capacity dernanding, while still eliciting no compering co\,ert response from the distractor words, like in our Experiments 1,3, and 4, only srnall, nonsignificant, positive priming effects are found.
To conclude, our frnding that semantic activation up to conscious identification of a substantial proportion of the prime words (Experiments 1, 3, and 4) does not necessarily cause significant semantic priming effècts clearly questions the existence of strong automatic priming processes. Moreover, these results refute an eventual interpretation of the lack of semantic priming with parafoveal prime words or with an additional prime task in terms of an impairment of the visual feature integration process (cf. Neely & Kahan, 2001), as both in the present study and in the experiments discussed by Maxfield (1997), no significant priming effects were obtained with reportable prime words.