To assess whether working memory contents can effectively bias visual selection even when they do not represent the current target in the attention task, we recorded the ERP activity from participants performing both a memory task and, in the retention period, a visual search task. In this task, a distracter matching the memory content could be presented on the same side (congruent trials) or on the opposite side (incongruent trials) relative to the target location (Experiment 1 and Experiment 2). On some trials, only the matching distracter (but no target) was presented (catch trials, Experiment 2). Results showed that the N2pc component was modulated by the presence and location of a matching distracter. We interpret these results as evidence that the involuntary control exerted by the irrelevant memory contents coexists with the strategic mechanism related to the search target, influencing attention selection with roughly equal power. In Experiment 3, we found that the modulation of the N2pc is not strictly related to the active maintenance of the memory-target features but can also be elicited by repetition priming. Overall, these findings suggest that, together with the physical properties of the stimuli presented in the visual field, irrelevant memory contents represent a powerful class of factors that lead to involuntary attentional control.
Most of the evidence concerning the interaction between memory and attention comes from studies in which the working memory contents define the current target (for a review, see Desimone & Duncan, 1995; Wolfe, 1994; Bundensen, 1990; Duncan & Humphreys, 1989). For this reason, the influence of memory on attention has exclusively been regarded as voluntary and goal-oriented.
In contrast with this assumption, recent behavioral evidence indicates that working memory representations can influence attentional selection even when they are irrelevant for the current task (Olivers, Meijer, & Theeuwes, 2006; Soto, Heinke, Humphreys, & Blanco, 2005; Downing, 2000; Pashler & Shiu, 1999; for a review, see Olivers, 2008; Soto, Hodsoll, Rotshtein, & Humphreys, 2008; Awh, Vogel, & Oh, 2006; but see Woodman & Luck, 2007; Houtkamp & Roelfsema, 2006; Downing & Dodds, 2004). Several visual search studies (e.g., Soto et al., 2005) have shown that a distracter element matching the irrelevant memory contents (i.e., the working memory representations that are related to a separate task, thus being irrelevant for the current visual search task) leads to faster response times when both the target and the matching distracter occupy the same location relative to when they occur at opposite locations, suggesting that attention is inadvertently guided by the irrelevant memory contents.
However, behavioral measures represent the result of several cognitive operations and could result from effects ensuing at different stages of processing. Hence, it is still unclear whether the behavioral effects found in previous studies truly reflect modulations occurring at the attentional selection stage or whether both costs and benefits might be due to facilitation/interference occurring during a subsequent stage, where the search target is further analyzed and identified. The present ERP study sought to provide evidence allowing a better understanding of the temporal brain dynamics underlying the above phenomenon.
A recent ERP study (Kumar, Soto, & Humphreys, 2009) provided the first evidence that irrelevant memory contents can bias attention selection, as reflected by the N2pc. The N2pc is an attention-related ERP component consisting of a lateralized activation with a latency of 180–300 msec poststimulus at posterior sites contralateral to the hemifield where a target is shown (e.g., Mazza, Turatto, & Caramazza, 2009a, 2009b; Mazza, Turatto, Umiltà, & Eimer, 2007; Hopf et al., 2000; Woodman & Luck, 1999; Luck, Girelli, McDermott, & Ford, 1997; Eimer, 1996; Luck & Hillyard, 1994). Because this component has mainly been found within the context of visual search tasks, it has been viewed as the ERP correlate of attentional focusing onto a relevant location or item in the visual field (for a discussion, see Woodman, Arita, & Luck, 2009).
In the Kumar et al. (2009) study, participants performed a search task that was preceded by a memory prime, on which they had to perform a match-to-sample task at the end of some trials. The search task required to indicate the orientation of a tilted line presented among three other vertical lines. Lines were shown within four distracter objects, and on some trials, one of these objects matched the memory item (matching distracter), being on the same (congruent condition) or opposite (incongruent condition) side with respect to the target location. Results showed that the N2pc was earlier and greater on congruent than neutral trials. An N2pc toward the target hemifield was recorded on incongruent trials as well, although it seemed to be smaller than the one elicited on neutral trials. These results provide the first evidence that irrelevant memory contents can effectively bias the selection stage reflected by the N2pc. However, it is unclear whether the modulation of the N2pc by the matching distracter in this study was mediated by a modulatory influence of this item on the attentional processing of the target or whether the matching distracter actually elicits an N2pc in its own right, reflecting automatic attentional deployment onto its location. The latter scenario, if confirmed, indicates that any N2pc recorded in the presence of both the target and the matching distracter would be the net result of (summing or subtracting) independent components (one related to the target and one related to the matching distracter). Our study aimed at resolving this ambiguity by measuring any N2pc elicited by the matching distracter alone in the absence of the target.
In a follow-up study, Telling, Kumar, Meyer, and Humphreys (2010) explored the extent to which attention tends to be deployed toward items that are semantically related to the search target. Consistent with prior work by Moores, Laiti, and Chelazzi (2003), the level of performance was influenced by the presence of a related distracter as well as by its spatial congruency relative to the target location. ERP measures showed that the related distracter affected the N2pc component generated by the target both in a synergistic and antagonistic fashion, respectively, enhancing the N2pc when the related distracter fell on the same side as the target and reducing it when it fell on the opposite side. No clear N2pc was elicited by the related distracter itself in the absence of the search target. This study provides converging evidence for the hypothesis that memory content biases attention selection, as reflected by the N2pc. However, the study concerns working memory contents that are not related to a separate task, thus posing a problem when one wishes to explore any involuntary effect of memory contents on attentional orienting. Indeed, unlike in other studies exploring the role of working memory in the inadvertent guidance of attention, in Telling et al., both the target and the related distracter were defined by the same content of memory directly relevant to the search task (search template). Thus, the interacting effects engendered by the coexistence of a target and a related distracter were not due to separate and independent sources of information.
The paradigm used in the present study, which was similar to the one used by Soto et al. (Soto, Humphreys, & Rotshtein, 2007; Soto et al., 2005), allowed us to address the unresolved issues that emerged from previous studies (Telling et al., 2010; Kumar et al., 2009). In Experiment 1, participants were engaged in two tasks. In the memory task, they had to memorize the initial object presented at the center of the screen to perform a delayed match-to-sample judgment at the end of the trial. In the retention period, participants performed a visual search task, judging the orientation of a lateralized target (i.e., the letter T) presented together with an irrelevant letter on the opposite side. Each letter appeared inside a colored shape (distracter). Crucially, whereas on half of the trials both distracters were different from the memory item (neutral condition), on the remaining trials one of the distracters matched the memorized item (matching distracter), and the target letter could appear inside the matching distracter (congruent condition) or at the opposite location (incongruent condition). Experiment 2 served the purpose of replicating the results of Experiment 1 and contained an additional condition—catch trials—in which no target but only a matching distracter was presented, and participants were asked to refrain from responding.
In terms of ERP measures, if the irrelevant memory contents affect the attention orienting stage, we expect that the N2pc should be modulated by the presence of a matching distracter and by its spatial congruency with respect to the target location (as in Kumar et al., 2009). To examine the alternative possibility that the irrelevant memory contents might instead affect later stages of processing, we additionally tested for any influence of these memory contents on the contralateral delay activity (CDA; also called sustained posterior contralateral negativity, see Jolicoeur, Sessa, Dell'Acqua, & Robitaille, 2006; Vogel & Machizawa, 2004; Wauschkuhn et al., 1998). This is a later, sustained ERP lateralized activity (starting at 300–350 msec poststimulus) that is especially observed in visual search tasks when the target needs to be encoded in greater detail (Mazza et al., 2007, 2009b; Jolicoeur et al., 2006). On the basis of our previous findings (e.g., Mazza et al., 2007, 2009b), we expected a CDA to be present in all conditions of both experiments, except for the catch trials, where no identification of the search target is required. If the involuntary mechanism associated with the irrelevant memory contents mainly operates at a stage of processing subsequent to attentional selection, the CDA should be modulated by the presence of a matching distracter and by its spatial congruency with the target location.
Two other issues were addressed in the present study. First, we assessed how the strategic and involuntary mechanisms of attentional control interact when they are pitted against each other during the same event. Strategic control refers to the fact that participants have to create a template of the search target (i.e., the letter T in the present experiments) to select it against the other (irrelevant) letter. In contrast, the involuntary mechanism of selection comes into play because of the fact that the object stored in memory for the purpose of the match-to-sample task may lead to automatic allocation of attention to any matching item in the visual search array. Since in our paradigm the representation of the memory item is irrelevant for the search task, any effect on attention due to this representation during the search task is likely to be involuntary in nature (for a discussion, see Olivers, 2008; Soto et al., 2008). The study by Kumar et al. (2009), which satisfies this criterion, found a clear increase of the N2pc on congruent relative to neutral and incongruent trials, suggesting a synergy of the strategic and involuntary mechanisms. However, no statistical assessment was provided concerning the incongruent and neutral conditions. Moreover, there were no trials where only a matching distracter was presented and, therefore, no opportunity to assess the effects of the irrelevant memory content in relative isolation. Most importantly, in that study, the target stimulus was conspicuous not only in strategic terms, being the relevant element to search for, but also because it was more salient than the other irrelevant lines, being the only tilted line presented among three homogenously vertical lines and having a unique color with respect to the other black lines (and to the outlined objects). Given this coincidence between strategic and exogenous factors, it is plausible to assume that the search target in the Kumar et al. study represents a privileged stimulus for attention orienting; hence, it may have more potently opposed any effect of the matching distracter, which did not possess salient, unique features. Therefore, this study did not allow an unambiguous assessment of the interplay between strategic and involuntary attentional control. By contrast, in the present study, both the use of catch trials and the fact that the target item was not particularly salient provide a cleaner assessment of the interaction between strategic and involuntary attentional control.
There are several predictions that can be made on how the strategic and involuntary mechanisms could affect attentional selection and its electrophysiological signature (i.e., the N2pc component) in the search task. One possibility is that the involuntary mechanism is suppressed, such that target selection is merely guided by the strategic mechanism. Thus, attention should always be focused onto the target location, irrespective of the presence of a matching distracter and of its spatial congruency with the target. In our study, we should, therefore, expect to find an N2pc in the neutral condition, namely the condition where only the search target is present, as seen in previous studies using similar paradigms (e.g., see Luck et al., 1997; Eimer, 1996). Furthermore, if attentional selection is controlled by the strategic mechanism only, no modulation of the N2pc should occur when a matching distracter is presented in the visual display. Thus, a target-related N2pc should be present in all three conditions, and it should not be modulated by the presence and/or location of a matching distracter, an unlikely outcome in the light of the results reported by Kumar et al. (2009). In addition, no N2pc should occur on catch trials, because on these trials, the search target is physically absent.
A second possibility is that the two mechanisms may coexist but may not be independent. For instance, attentional selection in the search task may be mainly guided by the strategic mechanism, but the involuntary mechanism might contribute to selection when it is not particularly detrimental for the task, namely when both the target and the matching distracter occur at the same location (congruent condition). According to this prediction, an enhanced N2pc might be found in the congruent condition relative to the neutral condition, for instance, because the component is now sustained by both items occurring at the same spatial location. In contrast, no difference should emerge between the neutral and the incongruent condition, in which orienting attention to the location of the matching distracter would be detrimental to task performance. Interestingly, according to the same logic, we may predict a clear N2pc elicited by the matching distracter on catch trials, because on these trials automatic orienting of attention to the location of the matching distracter is not particularly detrimental to task performance as no target is present.
A third possibility is that the strategic and involuntary mechanisms coexist in an independent fashion, influencing attentional selection with roughly equal power. Relative to the neutral condition and in addition to a modulation of the N2pc on congruent trials, this should translate into a modulation of the N2pc in the incongruent condition. A previous ERP study on exogenous attentional capture (Hickey, McDonald, & Theeuwes, 2006) indeed detected two consecutive N2pc components, one associated with the location of a salient but irrelevant distracter and a subsequent one associated with the target itself. In a similar fashion, we may predict the presence of two sequential N2pc components, one related to the matching distracter and one to the search target. As a slight variation of this prediction, the presence of the two stimuli at opposite locations may give rise to two simultaneous N2pc components of similar amplitudes but opposite in polarity, effectively eliminating any measurable N2pc component. Additionally, as mentioned above, if the content of working memory exerts an independent control on attention, an N2pc should also be observed on catch trials and of similar amplitude to that measured on neutral trials.
The second issue, addressed in Experiment 3 of the present study, concerns the nature of the effects of irrelevant memory contents on attention selection. Are these effects tightly related to the active maintenance of the target features in working memory or can other memory-related phenomena, such as repetition priming, lead to involuntary attentional control as well? Previous behavioral studies (Soto et al., 2005; Downing, 2000) have shown that the effects on the search task due to the presence of a matching distracter disappear when no memory task is required. Interestingly, however, some recent ERP studies using standard visual search tasks (e.g., Eimer, Kiss, & Cheung, 2010; Olivers & Hickey, 2010; Mazza et al., 2009b) have shown that repetition of both the relevant and irrelevant features in successive displays affects the N2pc latency, with shorter onset latencies for repeated than unrepeated trials, indicating that repetition priming can impact attentional orienting at a relatively early stage of processing. Thus, one may predict that at least some of the effects in the present paradigm may be attributed to forms of memory other than working memory. Kumar et al. (2009) found no N2pc modulation on those trials where no memory task was involved. However, the greater salience of the target relative to the other irrelevant lines might have played an important role in that case, potentially blocking any repetition effect that might have facilitated processing of the matching distracter.
To test for the impact of repetition priming, in Experiment 3 of the present study, participants did not have to memorize the initial object but instead performed an immediate task on it. All other aspects were identical to those in Experiment 1. We predicted that if the involuntary guidance of attention were effective but strictly associated with the active maintenance in working memory, then no modulation of the N2pc should be seen across the neutral, congruent, and incongruent conditions in this experiment. In contrast, a modulation of the N2pc as a function of congruency in this experiment would indicate a more general influence of memory on attention guidance.
A total of 50 healthy volunteers (Experiment 1: 20, ages 19–31 years, two are left-handed; Experiment 2: 18, ages 19–31 years, two are left-handed; Experiment 3: 12, ages 23–28 years, all are right-handed) participated in the experiments. Participants reported normal or corrected-to-normal vision and normal color vision and provided written informed consent. All the experiments were conducted following the guidelines laid down in the Helsinki declaration and were approved by the local ethics committee.
Stimuli and Procedure
Stimuli were presented on a dark gray background (6 cd/m2) and consisted of outlined geometrical shapes (circles, squares, triangles, or diamonds, each approximately covering an area of 2.6° × 2.6°) shown in different colors (green, red, blue or gray; all approximately 26 cd/m2). In Experiments 1 and 2, each trial began with the presentation of the Italian word for “Memorize” (“Memorizza”) above the fixation cross for 1000 msec. Subsequently, an object was presented around the fixation cross for 300 msec (Figure 1A), and participants were instructed to hold it in memory until the end of the trial. After 1000 msec, the search display was shown and consisted of two lateralized white letters (4° to the left and right of fixation, respectively) presented at the center of two surrounding distracter objects. The target stimulus was the letter T oriented leftward or rightward, whereas the other stimulus was the letter L oriented left, right, up, or down. The target appeared with equal probability and in random order within the left or right outlined object. On 50% of the total trials of Experiment 1 (neutral condition), the distracter objects surrounding the letters were different from the memorized object in both color and shape. On 25% of the trials (congruent condition), one distracter object (matching distracter) was identical to the memorized object in both color and shape and contained the target letter. On the remaining 25% of the trials (incongruent condition) the target letter and the matching distracter were presented on opposite sides (Figure 1B). The same conditions were used in Experiment 2 (neutral condition: 40% of the trials; congruent: 20% of the trials; incongruent: 20% of the trials). Additionally, on the remaining 20% of the total trials (catch trials), no target was presented (i.e., two letters L were displayed), and participants were instructed to refrain from responding (Figure 1B). On these trials, a matching distracter was always presented, on either side of fixation. In both experiments, the search display was presented for 150 msec to minimize the occurrence of eye movements to the target location. Participants had to indicate the orientation (left vs. right) of the target (search task). Half of the participants pressed one of two keys on a computer keyboard with the index or middle fingers of their left hand, whereas this assignment was reversed for the other participants. Speed and accuracy were emphasized equally. Maximum time for responding was 1500 msec, after which a test object was presented for 300 msec around the fixation cross. Three question marks prompted the participants to indicate without time pressure whether the test object was the same or different relative to the memorized object (memory task). Half of the participants pressed one of two keys on the computer keyboard with their index or middle finger of their right hand, with a reversed assignment for the other participants. Same and different trials were presented with equal probability and in random order. At the end of the trial, a visual feedback (either the words “Error” or “No response”) was presented for each task for 1000 msec if responses were incorrect or no response was executed. The intertrial interval was 1500 msec.
In Experiment 3, participants did not have to memorize the initial object. Instead, they had to judge its size by pressing the space bar only when the object had a specific size (either large or small, counterbalanced across participants; see Figure 1A), with the index and middle fingers of their left or right hand (counterbalanced across participants). All the other aspects were identical to those in Experiment 1. The size of the distracter objects containing the letters was always the opposite relative to that of the first object (i.e., if the first object was small, both objects in the search display were large, and vice versa). This was done to avoid any specific bias for those objects having the same size as the target of the first task (i.e., large objects when the relevant dimension was “large”).
Participants performed 10 experimental blocks in Experiment 1 and Experiment 2 (Experiment 1: 64 trials per block; Experiment 2: 80 trials per block) and 12 blocks of 64 trials per block in Experiment 3. One training block was delivered before the start of the experimental session of each experiment.
EEG Recording and Data Analysis
EEG was recorded with Ag–AgCl electrodes from FPz, F7, F3, Fz, F4, F8, FC5, FC6, T7, C3, Cz, C4, T8, CP5, CP6, P7, PO7, P3, O1, Pz, P4, PO8, P8, O2, and Oz. These electrodes and the left earlobe electrodes were recorded with a right earlobe reference and then re-referenced off-line to the average of the left and right earlobe sites. Horizontal EOG was recorded by means of two electrodes positioned on the outer canthii of both eyes. Impedance was kept below 6 KΩ for all electrodes. Amplifier bandpass was 0.01–40 Hz, and digitization rate was 500 Hz. Trials with horizontal eye movements (horizontal EOG exceeding ± 30 μV), eye blinks, head movements, and other artifacts (any electrode exceeding ± 80 μV) were excluded. The average number of trials retained for the participants was 80% (range, 71–88%) in Experiment 1, 84% (range, 70–92%) in Experiment 2, and 90% (range, 68–96%) in Experiment 3.
Averages for correct responses on both tasks (except for catch trials, where only the memory task was executed) were computed relative to the 100 msec interval preceding the search display onset, separately for each condition (neutral, congruent, incongruent, and, when present, catch trials) and target side (left vs. right). The following sets of statistical analyses were conducted for all the experiments of the present study. Additional analyses, aimed at examining in further detail the crucial findings of each experiment, are explained in the Results section.
The first analysis was conducted on mean difference amplitudes obtained by subtracting ERP waveforms at posterior electrodes ipsilateral to the target side (i.e., PO7 for left targets, PO8 for right targets for congruent, incongruent, and neutral trials; PO7 for left distracters, PO8 for right distracters for catch trials) from those recorded at electrodes contralateral to the target side (i.e., PO8 for left targets or left distracters on catch trials, PO7 for right targets or right distracters for catch trials), collapsed across target (or distracter) side, in the poststimulus interval 190–270 msec (N2pc range). These values were submitted to an ANOVA with congruency as factor (Experiments 1 and 3: congruent, incongruent, and neutral; Experiment 2: congruent, incongruent, neutral, and catch). When appropriate, Greenhouse–Geisser correction for sphericity violations was used, and only corrected p values are reported. Further analyses were conducted by means of pairwise comparisons (t tests). In addition, to assess the presence of a reliable N2pc, mean difference amplitudes for each condition were compared against the null mean (i.e., the absence of an N2pc) via one-sample t tests. To test for the presence of any effect of congruency on the latency of the N2pc onset, a second set of analyses was conducted on the basis of jackknifing (see Ulrich & Miller, 2001; Miller, Patterson, & Ulrich, 1998). For each condition of interest, this method computes each n subsample as the grand-averaged waveforms of N − 1 subjects. In this analysis, a criterion of 50% of the N2pc peak amplitude values was used. Because this procedure intrinsically reduces the variability in the data, the correction suggested in Ulrich and Miller (2001) and Miller et al. (1998) was subsequently applied. Finally, for each experiment, we conducted a median-split analysis on the basis of individual RTs, in which mean difference amplitudes in the 190–270 msec range for fast (i.e., below the median RT) and slow (i.e., above the median RT) trials were calculated for each condition (congruent, incongruent, and neutral) and submitted to an ANOVA with congruency (congruent, incongruent, and neutral) and speed (fast, slow) as factors.
The analysis on the CDA was conducted by means of an ANOVA on mean difference amplitudes (obtained in the same way as for the N2pc) in the poststimulus interval 300–600 msec, considering the factor congruency (Experiments 1 and 3: congruent, incongruent, and neutral; Experiment 2: congruent, incongruent, neutral, and catch).
In all the experiments of the present study, an ANOVA on RTs between 200 and 1500 msec for trials with correct responses in both tasks was conducted, which included the factor congruency (congruent, incongruent, neutral). For Experiment 1, the ANOVA revealed a significant main effect, F(2, 38) = 15.72, p < .001. Follow-up comparisons (t tests) showed that participants were significantly faster on congruent trials (M = 670 msec) than on both incongruent (M = 726 msec; t(19) = 4.037, p < .001), and neutral trials (M = 682 msec; t(19) = 2.66, p < .02). In addition, participants were faster on neutral than on incongruent trials, t(19) = 4.02, p < .001 (Figure 2). Thus, congruency led to both costs (incongruent vs. neutral trials) and benefits (congruent vs. neutral trials), as previously found in other studies (e.g., Soto et al., 2005).
The ANOVA on percentage of correct responses also revealed a significant effect of congruency, F(2, 38) = 9.72, p < .002, again indicating that participants were slightly more accurate on congruent (M = 98%) and neutral (M = 97%) trials than on incongruent trials (M = 96%), t(19) = 3.53, p < .002 and t(19) = 2.96, p < .008, respectively. No significant difference emerged between neutral and congruent trials.
From the ANOVA on percentage of correct responses a marginally significant effect of congruency emerged, F(2, 38) = 3.3, p = .07, indicating that participants were slightly more accurate when the test object followed a congruent trial (M = 95%) than a neutral trial (M = 94%). No difference involving incongruent trials (M = 95%) emerged, all p > .09.
Figure 3A depicts the activation at posterior electrodes ipsilateral and contralateral to the target side, separately for the congruent, incongruent and neutral condition. As can be seen, the N2pc amplitude was affected by congruency, being maximal on congruent trials, where the target was surrounded by a matching distracter, relative to both neutral trials, where no matching distracter was present, and incongruent trials, where target and the matching distracter occurred at opposite locations. In contrast, a CDA of similar amplitude was present in all three conditions.
The ANOVA on the mean difference amplitudes showed a significant effect of congruency, F(2, 38) = 22.4, p < .001, with follow-up comparisons (t tests) indicating that congruent trials elicited a more pronounced N2pc than neutral and incongruent trials, t(19) = 6.5, p < .001 and t(19) = 5.2, p < .001, respectively. To ascertain the presence of a reliable N2pc component for each condition, ERP mean difference values were separately compared against zero (i.e., absence of N2pc). A reliable N2pc component was observed in the congruent and neutral conditions, both t(19) > 2.2, both p < .04, but not in the incongruent condition, p = .3.
To further assess the contribution of the involuntary mechanism to attentional control, we estimated the net effect of the matching distracter by computing the ERPs based on the difference between congruent trials and incongruent trials and by calculating the N2pc as in the main analyses.1 To test whether the matching distracter per se could elicit an N2pc, we compared the mean difference values against the null mean (i.e., absence of N2pc) via a one-sample t test against zero. This analysis revealed the presence of an N2pc associated with the matching distracter, t(19) = 5.24, p < .001.
N2pc onset latency
The onset of the N2pc was 83 msec earlier for the congruent than for the neutral condition, t(19) = 21.3, p < .001. Incongruent trials could not be included in this analysis, because no clear N2pc emerged in this condition.
This analysis was conducted to better characterize the N2pc pattern found in the present experiment and to explore in more detail why no N2pc was detected for the incongruent condition. Specifically, we were interested in exploring whether by this procedure we might be able to reveal a reliable N2pc component even in the incongruent condition. We assume that relatively fast and slow trials reflect (at least partially) a different balance between voluntary deployment of attention to the search target and inadvertent deployment of attention toward the matching distracter. On the basis of this assumption, slow trials represent those trials on which attention deployment toward the matching distracter initially dominates processing, therefore causing a delay in target selection. We predicted that this could be indexed by an early N2pc component related to the matching distracter on slow trials.
The ANOVA on mean difference amplitudes with congruency (congruent, incongruent. and neutral) and speed (fast and slow) as factors revealed a significant main effect of speed, F(1, 19) = 15.4, p < .001, indicating an overall reduction of the N2pc amplitudes for slow relative to fast trials in all conditions. As can be seen in Figure 5A, this reduction led to a decrease of the N2pc on slow trials both for the congruent condition and for the neutral condition (in which in fact the N2pc seems to disappear) and, interestingly, to a reversal of the N2pc on slow incongruent trials, in which an N2pc toward the matching distracter was observed. These observations were confirmed by follow-up comparisons (one-sample t tests) in which for each condition (congruent, incongruent, and neutral) we compared the N2pc mean difference amplitudes on fast and slow trials (i.e., respectively below and above the median RT for each participant and for each condition) against the null mean (i.e., no N2pc). In the congruent condition, an N2pc was present for both fast and slow trials (both t > 4.4, both p < .001); an additional pairwise t test showed that the N2pc was greater on fast than on slow trials, t(19) = 3.0, p < .008 (Figure 5A, top). Whereas fast trials in the neutral condition also elicited a reliable N2pc (t(19) = 3.3, p < .004, no N2pc was found on slow trials for this condition, p = .36 (Figure 5A, middle). An inverse pattern was found for the incongruent condition, wherein slow trials elicited a marginally significant N2pc toward the location of the matching distracter, t(19) = 1.95, p = .06. On the contrary, no N2pc was visible on fast trials, p = .71 (Figure 5A, bottom).2
No significant effect emerged from the ANOVA on mean difference amplitudes, p = .25. However, a CDA was recorded in each condition, as confirmed by comparisons against the null mean, all t > 5.8, all p < .001 (Figure 3A).
Following the same approach as for Experiment 1, the ANOVA revealed a significant main effect of congruency, F(2, 34) = 29.4, p < .001. As in Experiment 1, participants were significantly faster on congruent trials (M = 649 msec) than on both incongruent (M = 693 msec) and neutral trials (M = 660 msec), t(17) = 5.7, p < .001 and t(17) = 3.22, p < .006, respectively. They were also faster on neutral than on incongruent trials, t(17) = 5.5, p < .001 (Figure 2).
Participants' accuracy was overall high (greater than 95% of correct responses). A significant main effect of congruency, F(2, 34) = 3.38, p < .05, revealed, as confirmed by planned comparisons, that participants were more accurate on neutral (M = 97%) than on incongruent trials (M = 95%), t(17) = 2.4, p < .04.
No significant effects emerged from the ANOVA (overall accuracy: 95%).
As in Experiment 1, the N2pc was affected by congruency, being greater on congruent than on both neutral and incongruent trials, with no reliable N2pc on incongruent trials (Figure 3B). The entirely novel result of the present experiment concerns catch trials, namely those trials where only a matching distracter was presented. As predicted by the hypothesis that the involuntary mechanism can guide attention as effectively as the strategic mechanism, these trials elicited a clear N2pc, which seems to be identical to the one evoked on neutral trials, where only the search target was presented (Figure 4A).
The ANOVA on the mean difference amplitudes showed a significant effect of congruency, F(3, 51) = 9.27, p < .001. As in Experiment 1, follow-up comparisons against the null mean indicated the presence of an N2pc for congruent and neutral trials, both t > 2.97, both p < .009, but not for incongruent trials, p = .83. Importantly, a reliable N2pc toward the matching distracter was observed on catch trials, t(17) = 3.4, p < .004. As in Experiment 1, congruent trials elicited a more pronounced N2pc relative to neutral trials, t(17) = 3.56, p < .003. In addition, the N2pc was also greater on congruent than catch trials, t(17) = 2.98, p < .009. As predicted, no difference in the N2pc amplitudes emerged between neutral and catch trials, p = .85.
To further investigate whether the involuntary mechanism of attentional control (as engaged by the mere presence of a matching distracter on catch trials) could add up algebraically to the strategic component (as engaged in pure form by the presence of the target of search on neutral trials), we computed for each subject the sum of the ERP averaged waveforms for catch trials and neutral trials. The resulting N2pc, computed from these “simulated” waveforms as in the main analyses, was compared with that elicited on congruent trials (Figure 4B). A t test confirmed that the simulated N2pc was not significantly different from the N2pc elicited on congruent trials, p = .41. In a similar fashion, we computed for each subject the difference of the ERP averaged waveforms between catch and neutral trials, and this simulated N2pc did not significantly differ from the N2pc measured on incongruent trials, p = .95.
N2pc onset latency
In this analysis, we used the same approach as in Experiment 1. Because three conditions were considered (i.e., congruent, neutral, and catch trials), an overall ANOVA was performed with the correction indicated by Ulrich and Miller (2001). The main effect was significant, Fc(2, 34) = 26.2, p < .001. Post-hoc comparisons confirmed that the N2pc onset on congruent trials was earlier than on neutral trials, t(17) = 2.1, p < .001, but did not differ from that on catch trials, p = .2. Catch trials also elicited significantly earlier N2pcs than neutral trials, t(17) = 2.1, p < .001.
As in Experiment 1, the ANOVA showed a significant effect of speed, F(1, 17) = 10.5, p < .006, replicating the overall decrease of the N2pc amplitudes on slow trials relative to fast trials (Figure 5B). As in Experiment 1, the congruent condition elicited an N2pc on both fast and slow trials (one-sample t tests against zero: both t > 3.9 and p < .001), but no reliable difference was found between the two (paired-sample t test: p = .27; Figure 5B, top). In the neutral condition fast trials also elicited a significant N2pc, t(17) = 3.8, p < .001, in contrast to slow trials, which did not elicit any N2pc, p = .23 (Figure 5B, middle). In analogy with the results of Experiment 1, the peculiar N2pc pattern for the incongruent condition was replicated, with an inverse N2pc (i.e., an N2pc toward the matching distracter) for slow trials, resulting in a clear significant difference from the null mean in this experiment, t(17) = 2.15, p = .05 (Figure 5B, bottom). In contrast, a marginally significant N2pc toward the target location was found on fast trials, p = .08.
The ANOVA indicated a significant effect of congruency, F(3, 51) = 35.3, p < .001, because of the fact that, as expected, no CDA was present on catch trials, as also confirmed by successive comparisons against the null mean, which showed the presence of a CDA for all conditions (all t > 7 and p < .001) except for the catch trials, t < 1.
To summarize, three main results emerged from both Experiment 1 and Experiment 2. First, in line with previous studies (e.g., Soto et al., 2005; Downing, 2000), the RT data showed both costs and benefits due to the presence of a matching distracter and to its spatial congruency with the target. Second, the ERP results showed that the presence of a matching distracter affected the N2pc, whereas no modulations of the CDA were recorded. Third, in both experiments the N2pc started earlier and was more pronounced on congruent trials than on neutral trials (see also Kumar et al., 2009). In contrast, no N2pc emerged for incongruent trials, with additional analyses showing that trials with slow RTs were associated with an N2pc toward the matching distracter, suggesting that RTs in these trials may have been especially slow for the very reason that attention was potently summoned by the incongruent, matching distracter before it could be deployed toward the target. Finally, Experiment 2 showed that catch trials also elicited an N2pc, whose amplitude was the same as the N2pc elicited on neutral trials.
Participants' accuracy in the size task was overall very high (greater than 96%).
The ANOVA on RTs showed a significant effect of congruency F(2, 22) = 7.4, p < .02, with participants being faster on congruent (M = 593 msec) or neutral (M = 598 msec) trials than on incongruent trials (M = 615 msec), t(11) = 2.94, p < .02 and t(11) = 2.65, p < .03, respectively. However, no difference emerged between neutral and congruent trials, p = .14 (Figure 2).
Participants' accuracy was overall high (greater than 97%), and no significant effect of congruency emerged from the ANOVA on percentage of correct responses, F < 1.
As in Experiments 1 and 2, the N2pc amplitude was larger on congruent than on incongruent trials (Figure 3C). However, no difference seemed to emerge between neutral and congruent trials.
The ANOVA showed a significant effect of congruency, F(2, 22) = 10.2, p < .001, because of the fact that both congruent and neutral trials elicited a more pronounced N2pc than incongruent trials, both t > 3.3, and p < .007. No difference emerged between congruent and neutral trials, p = .12. As in the previous experiments, an N2pc was present only for congruent and neutral trials, as confirmed by significant differences against the null mean, both t > 3.4, and p < .006. No N2pc was observed for the incongruent condition, p = .43.
N2pc onset latency
In line with the analysis on the N2pc amplitudes, no significant difference emerged in the N2pc onset latency between congruent and neutral trials, p = .28.
The results of the median-split analysis replicated those of Experiments 1 and 2 (see Figure 5C). Specifically, N2pc amplitudes were smaller on slow than on fast trials, as indicated by the significant effect of speed in the ANOVA, F(1, 17) = 21.3, p < .002. In the congruent condition, both fast and slow trials elicited an N2pc, as in the previous experiments, both t > 2.48, and p < .04. In the neutral condition, an N2pc was found for fast trials, t(11) = 4.3, p < .002, but not for slow trials, p = .68. In the incongruent condition, no N2pc was visible on fast trials, p = .29, but a marginally significant N2pc toward the matching distracter was seen on slow trials, t(11) = 1.8, p = .09.
As in Experiment 1, no significant effect emerged from the ANOVA, F < 1, but a CDA was recorded in each condition, all t > 4.9, and p < .001.
Overall, these results show a general reduction of the distracter effect, leading to the disappearance of benefits in RTs and of the difference in the N2pc parameters between congruent and neutral trials. By contrast, incongruent trials still led to slower RTs and to the disappearance of the N2pc.
The results of the present study shed new light on the role of working memory in guiding visual attention. First, we found that the N2pc, which is considered the indicator of attentional selection of a relevant location and/or item, was modulated in all the experiments by the presence of a matching distracter. In contrast, whereas a CDA toward the location of the search target was also recorded in all conditions where a target was presented, its amplitude was not modulated by the presence of a matching distracter, indicating that later stages of processing (i.e., stimulus identification) are not involved in the generation of the reported effects. Taken together, these findings provide converging evidence in favor of the notion that the locus of the effect due to irrelevant memory contents, as seen from previous studies (e.g., Kumar et al., 2009; Olivers et al., 2006; Soto et al., 2005; Downing, 2000), can be attributed to the attentional selection stage.
Second, the results on the N2pc reported here allow us to clarify the nature of the interaction between strategic and involuntary attentional control when both target and distracter are simultaneously present, providing converging evidence that the two mechanisms coexist in an independent fashion. Previous ERP studies on involuntary attentional capture have provided mixed results. On the one hand, Hickey et al. (2006) found that a salient but task-irrelevant distracter presented together with the search target elicits an N2pc that precedes the target-related N2pc, supporting the existence of strong attention capture. On the other hand, more recent studies (Wykowska & Schubö, 2010; Eimer & Kiss, 2008; Leblanc, Prime, & Jolicoeur, 2008; Lien, Ruthruff, Goodin, & Remington, 2008) challenge this claim by showing that salient distracters are able to orient attention (as inferred by the presence of an N2pc component) only when they share some of the target-defining features, supporting the view that attentional orienting is predominantly under strategic control. However, all these studies have concentrated on the issue of involuntary attentional capture associated with the presentation of salient distracter stimuli. By contrast, the current study addressed involuntary attentional guidance by means of stimuli that do not possess unique features and, therefore, do not capture attention in a stimulus-driven fashion, but they summon attention insofar as they match the object description held in working memory to perform a different, unrelated task.
In Experiments 1 and 2 the N2pc was modulated by the presence of a matching distracter and by the target–distracter spatial congruency, being larger on congruent trials (where both the search target and the matching distracter were at the same location) than on neutral trials. This result is in line with Kumar et al. (2009), which also showed an earlier and greater N2pc on congruent than neutral trials. It is worth noting that in their experiment the matching distracter never contained the target. In other words, the congruent condition referred to trials in which the target and the matching distracter were shown within the same hemifield, but not at the same location. Thus, the facilitation due to the spatial congruency between target and the matching distracter seems to apply to the entire hemifield rather than to the exact spatial location where the target is presented. In contrast to Kumar et al., no N2pc was observed in the present experiments on incongruent trials, where the search target and the matching distracter occupied opposite positions. As we discussed above, we attribute this difference mainly to the fact that, contrary to the items used in the present study, in Kumar et al. (2009) the target stimulus was conspicuous not only in strategic terms, being the relevant element to search for, but also because it was more salient than the other irrelevant lines, thus likely representing a privileged stimulus for attentional orienting.3
We interpret our findings in terms of the occurrence of two N2pcs of virtually identical amplitudes, one for the target and one for the matching distracter, which add up or cancel each other out on congruent and incongruent trials, respectively. This interpretation was supported by several additional observations. First, in Experiment 2, an N2pc was found on catch trials, where only a matching distracter was presented. Moreover, this N2pc did not differ from the one elicited on neutral trials, namely those trials where only the target was presented. Second, the analysis using simulated waveforms (Experiment 2) indicated that the N2pc obtained by the sum of neutral and catch trials equaled that for the congruent trials. In a similar way, the simulated N2pc obtained by subtracting the ERP waveforms on catch trials from those on neutral trials did not differ from that elicited by incongruent trials. Overall, of all the alternative scenarios discussed in the Introduction, the pattern of results presented here provides clear support to the hypothesis that the involuntary mechanism that is driven by the irrelevant memory contents coexists with the strategic one and that the two mechanisms are roughly equally effective in guiding attentional selection in a search task.
The additional analysis on the latency of the N2pc onset and the one on the basis of the median-split approach (also supported by the correlational analyses reported in footnote 2) provide further important observations. The earlier onset of the N2pc on congruent trials indicates that attentional orienting toward the target hemifield is speeded up when both the target and the matching distracter are presented on the same side, relative to when the search target is presented alone. The fact that in Experiment 2 the N2pc onset was earlier on catch and congruent trials than on neutral trials seems to suggest that the involuntary mechanism operates faster than the strategic one in orienting attention to a specific portion of the visual field. Both findings support the view that, when the strategic and involuntary mechanisms operate synergistically, target processing is facilitated, in line with the overall facilitation seen in the behavioral data.
The median-split analysis for the congruent condition in Experiment 1 indicated that the N2pc amplitude was greater on fast relative to slow trials, further suggesting that on trials where attention is directed toward both the target and the distracter (as inferred by the increase of the N2pc amplitude on fast trials), an overall facilitation of target processing occurs. However, because the difference in amplitude between fast and slow congruent trials was not replicated in Experiment 2 and given that the same trend was also found in the neutral condition (where in fact no N2pc was found for slow trials), we must be cautious in drawing any conclusion. Importantly, the median-split analysis for the incongruent condition in both Experiments 1 and 2 revealed the presence of an N2pc toward the matching distracter when slow trials were considered. This can be taken as an indication that on trials where the participants' attention is mainly summoned by the distracter, their processing of the target becomes particularly slow, as reflected in especially long RTs.
One may wonder whether the effects found in the present study can truly be considered as involuntary or whether the presence of congruent trials may have encouraged observers to strategically attend to the item that matches the memory item in Experiments 1 and 2 (see Woodman & Luck, 2007). There are two ways in which orienting of attention toward the matching distracter could potentially be seen as strategic. According to a first interpretation, the matching distracter could be useful to find the target location more efficiently and would, thus, be used as a strategic cue aiding search performance. However, the proportion of congruent trials, which might induce a strategic deployment toward the matching distracter as a way to localize the target, was clearly low (25% and 20% of the total trials in Experiment 1 and Experiment 2, respectively), likely discouraging a strategic use of the matching distracter for locating the target. In fact, the use of congruent trials in our displays (where only two pairs of elements were presented) allowed us to prevent participants from using the matching distracter strategically. Had we included incongruent and neutral trials only, participants could have realized that when a matching distracter was presented, the target would occur on the other side of the display, thus resorting on the matching distracter as a strategic cue for orienting attention toward the (opposite) target side.
According to a different interpretation, the matching distracter may strategically be used to refresh participants' memory to the benefit of the delayed match to sample task. However, the behavioral results of Experiment 1 (where only a marginally significant difference for congruent trials was found) and Experiment 2 seem to provide evidence against such a hypothesis. The fact that there was no clear difference in memory performance because of the presence and location of the matching distracter argues against the possibility that attention may have been strategically deployed to the matching distracter for the sake of the memory task. Moreover, a strategic account would predict that attentional orienting in the target-only (i.e., neutral) and matching distracter-only (i.e., catch) conditions should be similar. In contrast, the finding of an earlier N2pc latency for catch relative to neutral trials (Experiment 2) seems to suggest the involvement of two different kinds of attentional mechanisms, the one driven by the matching distracter being faster than the strategic one related to the search target.
In Experiment 3, we investigated whether explicit maintenance of the memory contents has a unique role in the involuntary control of attention or whether other (implicit) forms of memory associated with the reappearance of a previously seen stimulus may as well lead to involuntary attentional guidance (e.g., Eimer et al., 2010; Olivers & Hickey, 2010). In this experiment, the effects due to the presence of a matching distracter were overall reduced, and this was particularly evident for the congruent condition (which already showed a quantitatively smaller effect than the incongruent condition in Experiments 1 and 2). Indeed, the behavioral results of Experiment 3 showed that benefits, as seen from the comparison between congruent and neutral trials, disappeared. However, RTs still showed the presence of costs, albeit quantitatively reduced (i.e., 17 msec vs. 44 msec and 33 msec in Experiments 1 and 2, respectively). This result apparently conflicts with previous behavioral studies (e.g., Soto et al., 2005; Downing, 2000), which found no effects when no explicit maintenance was required. There are several differences between the paradigms used by previous studies and the present one that might explain this discrepancy. For instance, in the Soto et al. study (2005, Experiment 3) no task on the initial object was assigned to participants, thus making the two experimental conditions (the dual task condition used here and the single task condition used in their study) difficult to compare.
In the condition where no explicit memory task was required, Kumar et al. (2009) found no modulation of the N2pc because of the presence of a matching distracter. However, the relatively greater salience of the target might have counteracted any repetition effects that could facilitate processing of the matching distracter. A different pattern of results was found in our study. Specifically, the ERP results showed an N2pc for both the neutral and congruent conditions, but no variations in the N2pc amplitude or onset latency between the two conditions. On the basis of the interpretation given in relation to the previous experiments, this finding suggests that no N2pc for the matching distracter occurs on those trials, thus indicating a weakening of the involuntary mechanism in attention control. In contrast, as in Experiments 1 and 2, no N2pc was observed on incongruent trials, with further analyses replicating the finding of an N2pc toward the matching distracter on slow trials. Whereas future experiments will need to address the role of spatial congruency between target and matching distracters in the generation of memory-driven attention orienting, the present results indicate that the involuntary effect due to memory contents is not exclusively generated by the active maintenance of the information in working memory for an upcoming task. Results from Experiment 3 have an additional, important implication. Namely, given that no memory task was required in this experiment and to the extent that we obtained evidence of attentional deployment toward the matching distracter, we can rule out the possibility that effects due to the matching distracter in Experiments 1 and 2 were mediated by strategic deployment of attention to this item to the benefit of the memory task.
In conclusion, the current findings provide direct evidence that the activation of a memory representation effectively biases attentional selection even when this does not define the current target of the visual search task. Overall, these results complement previous research on attentional orienting by showing that, together with the physical properties of the stimuli presented in the visual field, memory representations make up a class of signals that sometimes exert powerful involuntary control of attention.
This research has been supported by a grant from the Fondazione Cassa di Risparmio di Trento e Rovereto (Italy) to V. M. and M. T. L. C. was supported by MIUR and Fondazione Cariverona.
Reprint requests should be sent to Veronica Mazza, Corso Bettini 31, 38068 Rovereto, Italy, or via e-mail: email@example.com.
This analysis was based on the assumption of additivity. On the basis of this assumption and given that the ipsilateral and contralateral labels to compute the N2pc were assigned with respect to the target location, the N2pc on congruent trials can be viewed as the sum of a target-related (T) N2pc and of a matching distracter-related (D) N2pc. In contrast, the N2pc in the incongruent condition can be viewed as the difference between TN2pc and DN2pc. Thus, by subtracting the incongruent condition (TN2pc − DN2pc) from the congruent condition (TN2pc + DN2pc), the target-related N2pcs cancel each other out, leaving the sum of the two distracter-related N2pcs. By averaging this waveform, we can obtain an indirect estimate of the N2pc related to the matching distracter.
Additional support to the results from the median-split analysis was provided by means of correlational analyses performed for each condition (congruent, incongruent, and neutral). For each participant, we computed the N2pc mean difference amplitudes from ERP epochs sorted out on the basis of a division in 10 bins of each individual's RT distribution. N2pc mean difference values were correlated with the mean RT values across bins. In all the conditions, the correlation between RTs and N2pc amplitudes was significant (p = .02 for the congruent condition; p = .001 for the neutral condition; p = .006 for the incongruent condition), thus confirming that progressively slower responses were associated with reduced (or inverted, for the incongruent condition) N2pcs. To anticipate, the same pattern of results emerged in Experiment 2 (congruent condition: p = .07; neutral condition: p < .001; incongruent condition: p < .001) and in Experiment 3 (congruent condition: p = .001; neutral condition: p = .001; incongruent condition: p = .02).
A reviewer suggested a different explanation for the difference between the Kumar et al. (2009) study and the present one by pointing to the different demands on filtering processes between the two studies (one item to be filtered in our study; three items to be filtered in Kumar et al.'s study). Whereas it is possible that the presence of fewer distracters in the present study may have led to an overall weakening of the N2pc, this interpretation cannot explain the stronger effects (in terms of the relative differences in N2pc magnitude across the various conditions) found in the present study relative to the Kumar et al. (2009) study. The notion of an overall weakening of the N2pc because of the small number of distracters used here should as well predict no increase of the N2pc in the congruent condition and no N2pc for catch trials, where the need for suppression should be minimal if not null (because only the matching distracter was presented), but neither result was observed.