Selection history alters attentional filter settings persistently and beyond top-down control

Transcription

1 Psychophysiology, 54 (07), Wiley Periodicals, Inc. Printed in the USA. Copyright VC 07 Society for Psychophysiological Research DOI: 0./psyp.830 Selection history alters attentional filter settings persistently and beyond top-down control HANNA KADEL, a TOBIAS FELDMANN-W USTEFELD, a,b AND ANNA SCHUB O a a Cognitive Neuroscience of Perception and Action, Philipps-University of Marburg, Marburg, Germany b Institute for Mind and Biology, Department of Psychology, University of Chicago, Chicago, Illinois, USA Abstract Visual selective attention is known to be guided by stimulus-based (bottom-up) and goal-oriented (top-down) control mechanisms. Recent work has pointed out that selection history (i.e., the bias to prioritize items that have been previously attended) can result in a learning experience that also has a substantial impact on subsequent attention guidance. The present study examined to what extent goal-oriented top-down control mechanisms interact with an observer s individual selection history in guiding attention. Selection history was manipulated in a categorization task in a between-subjects design, where participants learned that either color or shape was the response-relevant dimension. The impact of this experience was assessed in a compound visual search task with an additional color distractor. Top-down preparation for each search trial was enabled by a pretrial task cue (Experiment ) or a fixed, predictable trial sequence (Experiment ). Reaction times and ERPs served as indicators of attention deployment. Results showed that attention was captured by the color distractor when participants had learned that color predicted the correct response in the categorization learning task, suggesting that a bias for predictive stimulus features had developed. The possibility to prepare for the search task reduced the bias, but could not entirely overrule this selection history effect. In Experiment 3, both tasks were performed in separate sessions, and the bias still persisted. These results indicate that selection history considerably shapes selective attention and continues to do so persistently even when the task allowed for high top-down control. Descriptors: Visual attention, Selection history, Npc, Attentional capture, Task preparation, Intertrial priming Visual selective attention allows for preferential processing of relevant and suppression of irrelevant, potentially distracting information. Several mechanisms work together in deciding where to attend: bottom-up mechanisms operating on visual features of stimuli and their salience (Itti & Koch, 00; Li, 00; Schoeberl, Fuchs, Theeuwes, & Ansorge, 04; Theeuwes, 00, 03) and goal-based top-down mechanisms, which allow humans to direct their attention to stimuli relevant to the current task or action goal (Ansorge, Kiss, Worschech, & Eimer, 00; Eimer & Kiss, 008, 00; Folk & Remington, 998; Kiss, Grubert, Petersen, & Eimer, 0; Leonard & Egeth, 008; Wolfe, 994; Wolfe, Butcher, Lee, & Hyle, 003; Wykowska & Schub o, 0). Recent results have suggested that prior knowledge (Wolfe et al., 003) or selection history (Awh, Belopolsky, & Theeuwes, 0) also affects priority in selection. For instance, in tasks in which observers respond to pop-out stimuli, responses are remarkably faster when the targetdefining pop-out feature repeats over trials (Fecteau, 007; Nakayama, Maljkovic, & Kristjansson, 004; Maljkovic & This research was supported by the Deutsche Forschungsgemeinschaft (German Research Foundation; SFB/TRR 35, TP B03). We thank John McDonald and an anonymous reviewer for insightful comments and suggestions. Address correspondence to: Hanna Kadel, Philipps-University Marburg, Experimental and Biological Psychology, Gutenbergstraße 8, 3503 Marburg, Germany. kadel@uni-marburg.de Nakayama, 994; Thomson & Milliken, 00). Longer-lasting selection history effects may originate from reward history, and it was shown that stimuli that were coupled with reward in the past can still capture attention long after they stopped being rewarded or task relevant (Anderson, Laurent, & Yantis, 0; Hickey, Chelazzi, & Theeuwes, 00; Hickey & van Zoest, 03; Kiss, Driver, & Eimer, 009). A recent study found that associative learning induced a selection history that biased attention deployment (Feldmann- W ustefeld, Uengoer, & Schub o, 05). In this study, a categorization learning task was combined with a visual search task in the course of an experiment. In the categorization learning task, participants were required to select and categorize singletons varying either in color or in shape. Half of the participants (color-predictive group) learned that stimulus color (blue vs. green) signaled category membership and was thus predictive for the correct response, while shapes were nonpredictive. For the other half (shape-predictive group), the categorization was based on the shapes (triangles vs. pentagons), and colors were nonpredictive. According to several attention theories of associative learning (George & Pearce, 0; Kruschke, 99; Logan, 00; Mackintosh, 975), organisms prioritize stimuli that enable accurate predictions of future events in a way such that attention is biased toward stimuli with a higher predictive value. One may therefore assume that participants who, for instance, had learned that color singletons were response 736

2 Selection history alters attentional filter settings 737 predictive would deploy their attention to the color singleton rather than to the shape singleton, which had no predictive value. ERPs showed that after a short acquisition phase attention indeed was deployed only to singletons of the predictive dimension, indicating that selection history had successfully been induced. In the further course of the experiment, the visual search task was added, in which a salient color singleton (red) served as distractor for all participants, and attention deployment was measured by means of behavior and the Npc, an ERP component first reported by Luck and Hillyard, 994, and since that time used in numerous paradigms as a neurophysiological marker of attention deployment to an object in the visual field (Brisson, Robitaille, & Jolicœur, 007; Eimer & Kiss, 00; Mazza, Turatto, & Caramazza, 009). It emerges approximately 00 ms after the onset of a visual search display as enhanced negativity observed at contralateral posterior electrodes compared to ipsilateral electrode sites for an attended item. In the study of Feldmann-W ustefeld et al. (05), the Npc served to determine how learning experience affected attention deployment in the search task. Results showed that the color distractor received more attention in participants from the colorpredictive group (who had attended to color singletons in the categorization learning task) than in participants from the shapepredictive group (who had attended to shape singletons in the categorization learning task). More precisely, when the color distractor was presented in the visual hemifields opposite to the search target, an inverse Npc (larger negativity at electrodes ipsilateral to the target and thus contralateral to the distractor) emerged prior to the target-elicited Npc, but only in participants for whom color was predictive in the categorization learning task. This early distractorelicited Npc signals that attention was captured by the irrelevant color singleton before it could be redirected to the target. The authors concluded that the individual selection history induced by preferential attention deployment to predictive stimuli in the categorization learning task had changed attentional filter settings, which were then transferred to the second, independent search task. Crucially, the same result pattern was also found when the search task was performed the next day, providing direct evidence that selection history can affect attention deployment also in the long run. The present study examined how such selection history effects relate to other attention control mechanisms. One can imagine many situations where prior experience biases attention in a way that conflicts with the organism s current goals (cf. Awh et al., 0, p. 437). The question of interest seems whether and how topdown mechanisms can counteract attention settings induced by selection history. Given the potency of top-down mechanisms (e.g., in avoiding or overcoming salience-driven attention capture; Ansorge et al., 00; Eimer & Kiss, 00; Kiss, Grubert, & Eimer, 03; Sawaki & Luck, 03; Wykowska & Schub o, 00, 0), one may assume that top-down adjustment of filter settings can modulate or even override selection history effects. However, an individual s history of selection and learning is a powerful factor in determining human behavior, and it seems hence conceivable that individual selection history may consistently bias attention beyond top-down control. To test whether top-down control can override individual selection history in attention deployment, we used the same categorization learning and search task as in Feldmann- W ustefeld and colleagues (05), and added three different manipulations to increase the amount of volitional control participants could exert in a single trial. In Experiment, a cue indicated the upcoming task for every trial. Results from task-switching studies investigating the flexible adjustment of attentional control settings between trials have shown reduced response time costs when pretrial cues allow observers to actively prepare for an upcoming task. When humans are performing multiple cognitive tasks at the same time, cue-induced task preparation can trigger cognitive control mechanisms that allow reconfiguring the task set (e.g., Rogers & Monsell, 995), thereby overcoming interference from leftover task set activations in previous trials. Such reconfiguration processes can be multifaceted, and may include a shift of action goal or goal state (Karayanidis et al., 00; Rubinstein, Meyer, & Evans, 00; Ruge, Jamadar, Zimmermann, & Karayanidis, 03), a shift in attentional set (i.e., the relative weighing of the relevant perceptual stimulus attributes; Longman, Lavric, & Monsell, 03; Meiran, 000; Meiran, Kessler, & Adi-Japha, 008; Ruge et al., 03; Rushworth, Passingham, & Nobre, 005), and activation of a response-related action set (Karayanidis et al., 00; Ruge et al., 03; Rushworth, Passingham, & Nobre, 00). All of the above are mediated by a frontoparietal control network that is activated upon cue presentation (Karayanidis et al., 00; Ruge et al., 03) and allow participants to optimally prepare for an upcoming task. In Experiment, task preparation was induced by applying an entirely predictable trial sequence in which categorization and search task alternated after every second trial. This pattern of two successive trials of each task type (C-C-S-S-C-C, etc.) allowed a high degree of top-down control implementation, as participants could prepare task repetitions or task switches on a larger time scale and systematically adjust their attentional control settings between changing and repeating trials. Both pretrial task cueing in Experiment and the fixed trial sequence in Experiment should enable participants to draw on more goal-oriented, top-down control strategies to counteract any detrimental attentional bias resulting from individual prior selection differences. Whereas previous studies that tackled the question of how attention can influence early sensory encoding have confounded goal-driven and historydriven sources of selection bias (Awh et al., 0), Experiment and of the present study will allow disentangling potential selection history effects on attention deployment from dynamic topdown control strategies, as any leftover effects of the categorization learning task on attention deployment in the search task will have resulted from selection history insufficiently counteracted by topdown task preparation. Experiment 3 investigated the persistence of potential attention bias resulting from selection history. The search task was performed in a different experimental session separate from the categorization task, allowing participants to reconfigure their attention settings in an optimal way for the search task without ever having to switch back to the learned categorization task set. If effects from the categorization on the search task still remain, they will signal continued effects of learned selection history on visual attention. Rationale of the Present Study The purpose of the present study was to investigate to what extent top-down induced task preparation can modify attentional changes induced by individual selection history. In three experiments, we combined a categorization task with a visual search task in the same (Experiment and ) or separated (Experiment 3) experimental block. In the categorization task, participants were required to categorize stimulus compounds varying in color and shape (Figure A). To induce differential category selection, half of the participants (color-predictive group) responded to color singletons (blue vs. green), while shapes were nonpredictive for categorization. For the other half (shape-predictive group), categorization was based

3 738 H. Kadel, T. Feldmann-W ustefeld, and A. Schub o A B Experiment : Categorization task Experiment : Search task Distractor-absent trial trial Figure. A: Exemplary displays in the categorization task in Experiment. Participants in the color-predictive group had to press one button for a green and another button for a blue circle. Participants in the shape-predictive group had to press one button for a pentagon and another button for a triangle. Participants were na ıve regarding their group assignment at the beginning of the experiment and had to learn it by pressing the buttons on a trial-and-error basis and receiving immediate auditory feedback on correctness of categorization. The central stimulus (empty circle) was already presented ms before onset of the other stimuli, announcing a categorization task trial. B: Exemplary displays in the search task in Experiment. Both the color- and shapepredictive group searched for the diamond-shaped target and reported the orientation (horizontal vs. vertical) of the embedded line. In half of the trials, an additional color distractor was presented (right panel), which had to be ignored. The central stimulus (circle with embedded cross) was already presented ms before onset of the other stimuli, announcing a search task trial. on the shapes (triangles vs. pentagons), and colors were nonpredictive. After feedback-guided learning of the categorization task, the visual search task (an additional singleton task) was added. The search task was identical for both groups: participants had to search for a shape target (diamond) and report the orientation of an embedded line while an additional color singleton distractor (red circle) was presented in some trials (Figure B). The Npc component elicited by the search target was compared for participants from the color-predictive group (who responded to color singletons in the categorization task), and participants from the shapepredictive group (who responded to shape singletons in the categorization task). If attention deployment is biased by individual selection history, participants in the color-predictive group should be more distracted by the color distractor than those in the shapepredictive group, which should result in a more pronounced response time (RT) cost and a smaller and later Npc toward the target. Possibly, the target Npc may be preceded by an inverse Npc (i.e., an Npc toward the salient distractor). The existence and functional role of such an inverse Npc has been investigated in several studies in the previous years. Woodman and Luck (999, 003) showed that the Npc can change polarity when attention is shifted between potential targets in opposite hemifields during serial search. More recent studies investigated whether salient distractors can elicit an Npc early in time. For example, Hickey and colleagues reported a polarity switch of the Npc when targets and salient distractors were presented in opposite hemifields; an initial Npc contralateral to the hemifield that contained the distractor was followed by an Npc contralateral to the hemifield that contained the target (Hickey, McDonald, & Theeuwes, 006). However, a follow-up study with increased sample size failed to replicate this result, suggesting that the earlier finding might have been attributable to a low signal-to-noise ratio (McDonald, Green, Jannati, & Di Lollo, 03). Thus, it is unclear whether to expect a robust Npc elicited by the salient distractor in the present search task. Assuming that accumulating selection history will strongly bias attention deployment toward a dimension selection was based upon distractor interference should increase and a distractorelicited Npc might be observed (see, for instance, also Feldmann- W ustefeld, Brandhofer, & Schub o, 06, where reward history enabled the emergence of a distractor-elicited Npc). To examine whether volitional task preparation could override individual selection history, a pretrial task cue announced which task participants had to perform in the upcoming trial in Experiment. Experiment used a fixed trial sequence in which categorization and search task were alternated in a predictable manner. Experiment 3 examined the persistence of selection history effects by informing participants that the categorization task would no longer appear in the second experimental session. If participants were fully able to use volitional, task-oriented control mechanisms to prepare for the upcoming task, attention deployment should no longer be affected by the individual selection history. Rather, participants from both groups should show similar search RT patterns and the Npc elicited by the target should be of similar onset and amplitude, regardless of individual prior experiences. Participants Experiment : Method Thirty volunteers (3 male) completed the experiment for course credit or monetary payment (8e/h). Written consent for participation was obtained before the experimental session. Twenty-six participants were right-handed, all had normal or corrected-to-normal vision. Each participant was randomly assigned to one of two equal-sized groups (color-predictive group, mean age 6 SD: years, and shape-predictive group, mean age 6 SD: years). Four participants were excluded from analysis due to excessive eye movements in over 30% of the trials in the EEG (see below for criteria). Stimuli and Apparatus Participants were seated in a comfortable chair in a dimly lit, electrically shielded, and sound attenuated room. For the manual responses, a customizable key pad with four response buttons was placed in the participant s lap (Ergodex DX). Two buttons were positioned on the left half of the pad for reactions to the categorization task, to be operated with the left hand ring finger and thumb. For search task trials, participants used two buttons on the right half of the response pad with their right index and middle finger.

4 Selection history alters attentional filter settings 739 Stimuli were controlled via E-Prime.0 routines (Psychology Software Tools, Inc.), running on a Windows XP computer and displayed on a LCD-TN screen (Samsung Syncmaster 33) placed 00 cm in front of the participant. Two stereo speakers were positioned behind the screen for auditory feedback (Logitech Z0.0). In both tasks, eight elements of.38 visual angle each were arranged on an imaginary circle on a dark gray background (RGB 60, 60, 60) equidistant to the screen center (distance screen center to stimulus center: 6.38; horizontal eccentricity: 5.78). In the categorization task, six neutral distractor stimuli (gray circles, RGB 0, 0, 0) and two singletons were presented (see Figure A): One stimulus was deviant in color, either green (RGB 48, 7, 48) or blue (RGB 48, 48, 7), and the other one was deviant in shape, either a triangle or pentagon. The color and shape singletons were always presented with exactly one neutral distractor in between them. All possible stimulus combinations (blue/triangle, blue/pentagon, green/triangle, green/pentagon) were presented equally often in all possible locations. In the search task, the target was a diamond-shaped stimulus, with a horizontally or vertically oriented line within. Neutral distractor stimuli were gray circles (RGB 0, 0, 0) with an embedded gray oblique line, tilted 458 to the left or right. In 50% of trials, only neutral distractors were displayed besides the target (distractor-absent trials, Figure B). In the remaining trials, an additional red singleton distractor (RGB 7, 48, 48, circular shape) with an embedded oblique line was presented in a position close to the target, with one neutral distractor in between (distractor-present trials, Figure B). The target and distractor appeared with equal frequency in each of the eight stimulus positions, equally often in the same side of the visual field (distractor-present, same side) and on opposite sides (distractor-present, opposite sides). Each task type was announced by a centrally presented cue ms before the stimulus display. An empty gray circle of 0.68 visual angle preceded the categorization task, while a circle with an embedded cross announced a search trial. The cues were chosen to resemble the type of stimuli in each task, and remained on the screen for fixation aid throughout the trial. Procedure General trial procedure. Categorization and search trials started with the respective task cue (empty circle for search task, circle with embedded cross for learning task) that was presented for ms, and which participants were instructed to fixate their eyes on. Afterward, the stimulus display was presented for 00 ms, followed by a blank screen for up to,000 ms after stimulus display onset, showing only the central task cue for aiding fixation. A correct response within that time interval triggered the beginning of the intertrial interval (,000 ms). An erroneous or missing response led to acoustic feedback in the form of a low buzzing tone previous to the intertrial interval. Feedback-guided categorization learning. The experiment started with a block of 64 categorization task trials. Participants were informed that, in each presented display layout, one stimulus would be different in color and another one in shape. They were asked to respond to each layout by pressing either the upper or lower response button with their left hand, and that errors were followed by a buzzing tone. However, they were not told which stimuli required which button press or that only one dimension would be response predictive. Thus, they were to find out about the correct stimulus-response mappings themselves with the help of the acoustic error feedback. For participants who were assigned to the color-predictive group, the response mapping and feedback depended on the color singletons only (blue or green) and were independent of the shape singletons, while the reverse contingency was applied in the shape-predictive group: here, only the shape singletons (triangle or pentagon) were associated with response choice while color singletons were irrelevant. By this means, participants had to learn whether the color or shape dimension was predictive for categorization response. The assignment of the respective singletons to the response buttons was counterbalanced across participants. Response accuracy and speed were emphasized equally. During the first 3 trials, presentation time of the stimulus display was prolonged to ms in order to facilitate acquisition to the stimulus material. If accuracy was below 75%, the block was repeated. Mixed practice phase. The next practice block consisted of 3 categorization trials and 3 search trials, intermingled in a random order. Categorization task trials remained unchanged. In search task trials, all participants responded to the orientation of the line embedded in the diamond-shaped target (horizontal or vertical) by pressing either the left or right button on the response board with their right hand. Stimulus-response mapping was again balanced over participants of both categorization groups. During the first 3 practice trials of the search task, presentation time was prolonged to,000 ms instead of 00 ms. If accuracy in this combined categorization and search task block was below 75%, the practice block was repeated. Main experiment. After both tasks had been successfully acquired by practice, 6 further blocks of 64 trials each (,04 trials in total) were administered, where trials of the categorization and search task were presented in random order with the limitation that maximal four trials of one task could follow each other. Half of the 5 search trials were distractor-absent trials, in the other half the distractor was presented either on the same side as the target (8 trials) or on the opposite side (8 trials). Immediate auditory feedback after each incorrect response was maintained. After each block, performance feedback (RT and accuracy for both tasks) was presented to participants on the screen, and a break of at least 0 s followed. Participants could decide when to continue with the task. EEG Recording Sixty-four Ag-AgCl active electrodes (acticap, Brain Products, Munich, Germany) were used for EEG recording, placed according to the International 0 system. Vertical and horizontal electrooculogram (EOG) were recorded from electrodes placed below and above the left eye, and from the outer canthi of both eyes, respectively. Impedances were kept below 5 kx. All electrodes were referenced to FCz and rereferenced offline to the average of all electrodes. The signal was recorded with a BrainAmp amplifier (Brain Products) at a sampling rate of 000 Hz with regard to highprecision latency analysis, and filtered with a low cutoff filter of 0.06 Hz and a high cutoff filter of 50 Hz (3 db cutoff, Butterworth filter, 30dB/oct rolloff). Data Analysis Behavioral data. In the categorization task, mean RT for correct trials and mean accuracy were calculated for each participant and then compared between the color-predictive and the shape-

5 740 H. Kadel, T. Feldmann-W ustefeld, and A. Schub o predictive group with a t test for independent measures. For the search task, mean RT for correct trials and mean accuracy were calculated for each participant, separately for trials with and trials without a color distractor. RT and accuracy data were then forwarded to a two-way analysis of variance (ANOVA) with the between-subjects factor categorization group membership (colorpredictive group vs. shape-predictive group) and the within-subject factor trial type (distractor-absent vs. distractor-present trials). In order to further analyze RT results with respect to the task performed in the preceding trial, and to increase comparability to Experiment, RT data were split into task switch trials (search trials preceded by a categorization task trial) and task repetition trials (search trials preceded by a search task trial) and analyzed with the additional factor trial sequence position. P values were adjusted with Bonferroni correction in case of additional pairwise comparisons. The first trial of each block and trials with exceedingly long RT (6 SD from mean RT calculated separately for each participant and separately for each task block and the categorization and search task) were removed from all RT and accuracy analyses. This led to exclusion of 6% of trials. EEG data. EEG was averaged offline over a 700-ms epoch including a 00-ms prestimulus baseline with epochs time-locked to display onset. Only trials with correct responses were analyzed. Vertical and horizontal EOG were calculated as lower veog minus Fp and F9 minus F0, respectively, and low-pass filtered (Butterworth, 3 db cutoff 35 Hz, 4 db/oct rolloff). Trials in which EOG revealed blinks (indicated by veog > 6 80 lv) or eye movements within the first 30 ms after stimulus onset (heog > 6 35 lv) were excluded from analysis. Additionally, segments were excluded from further analysis on an individual channel basis when the absolute voltage exceeded 00 lv. Only participants with at least 70% valid trials were included in the further analysis (6 participants). To examine the Npc in the categorization task, the EEG was averaged for each participant separately for electrode sites contralateral and ipsilateral to the predictive singleton. Mean contralateral and ipsilateral activity in the ERP was calculated for pooled electrodes O/, PO3/4, P7/8, and PO7/8 for an epoch of ms after categorization display presentation and forwarded to a twoway ANOVA with the between-subjects factor categorization group membership (color-predictive vs. shape-predictive group) and the within-subject factor electrode laterality (contra- vs. ipsilateral to the predictive singleton). To examine the Npc in the search task, the EEG was averaged for each participant separately for electrode sites contralateral and ipsilateral to the shape target. EEG was further averaged separately for distractor-absent trials, distractor-present (same side) trials, and distractor-present (opposite sides) trials. Thus, six waveforms were obtained per participant, which were analyzed for both groups separately. Mean activity at electrode sites contra- and ipsilateral to the target stimulus was analyzed for electrodes O/, PO3/4, P7/8, and PO7/8 from ms after stimulus onset. This exact time range was chosen based on average waveform polarity. A negative deflection in an early time range (65 30 ms) in distractorpresent target (T) and distractor (D) opposite-side trials in the color-predictive group indicated an Npc of inverse polarity (i.e., activity ipsilateral to the target and contralateral to the distractor was more negative), suggesting attentional capture by the opposite color distractor. In the later time range, a traditional, target-elicited Npc was visible for all trial types. To optimally comprehend and depict this pattern and to differentiate between eventual early and late effects in attention selection, mean amplitude was calculated for two epochs (65 30 ms, ms) one on either side of the polarity shift. A two-way ANOVA was calculated with the repeated measures factors laterality (contra vs. ipsilateral to target) and the between-subjects factor categorization group membership (color-predictive vs. shape-predictive group). ANOVAs were calculated separately for epoch (early vs. late) and trial type (distractor-absent, distractor-present same side, distractorpresent opposite sides). Furthermore, in order to thoroughly examine the early inverse Npc, observed only in the color-predictive group in distractor-present opposite-side trials, and to test whether it might have been due to low signal-to-noise ratio, we used a signed area approach to evaluate its magnitude (Gaspar et al., 06; Luck, 04; Sawaki, Geng, & Luck, 0). As we had no specific a priori assumptions regarding the exact timing of the inverse Npc, we had centered the epochs for mean amplitude comparison quite narrowly on the peak observed in the data (cf. above). The signed-area approach allowed us to use a wider time window that takes individual differences of Npc latency into account. For each participant, the magnitude of the area (Amplitude 3 Time Points) was determined for which activity was more positive at electrodes contralateral to the target 5 to 75 ms after search display onset (signed positive area). The same area was estimated for an equally long time window during the prestimulus baseline (50 to 0 ms), which served as an estimate of random noise in the data. A t test for dependent samples was used to compare signed positive area in the 5 75 ms and the 50 0 ms epoch. In addition to mean amplitude and signed positive area, we compared onset latency of the Npc component between categorization groups in distractor present T and D opposite-side trials. We chose a jackknife-based approach to comparing latency (Kiesel, Miller, Jolicœur, & Brisson, 008; Miller, Patterson, & Ulrich, 998). For each condition in the categorization and search task, 6 grand-averaged waveforms were calculated, each excluding one participant. The time point at which 50% of the maximal Npc amplitude was first reached (contra ipsilateral, 30 Hz low-pass filtered, half-power cutoff, 4 db) was determined separately for each grand average. Latency estimations were compared between the categorization groups with a t test for independent measures. To account for underestimations of variance resulting from the jackknifing procedure, t values were adjusted by dividing by (n ) before assessing significance (compare Luck, 04). Those corrected values are referred to as t c in the results section. Categorization Task Experiment : Results Participants in the color-predictive group (M ms) responded faster to the predictive singletons than observers in the shapepredictive group (M ms), t(4) 5.94, p 5.007, Figure A. Regarding accuracy, no group difference was observed: Participants were about as accurate in the color group (M 5 97.%) as in the shape group (M %), t(4) 5.47, p Electrophysiological results show that both predictors elicited an Npc in the ms time range, as the ERP was more negative for electrode sites contralateral (M 54. lv) compared to electrode sites ipsilateral (M lv) to the response predictor, see Figure B, F(,4) 5 4., p <.00, g p The Npc was of comparable size in the shape-predictive group and the colorpredictive group, p

6 Selection history alters attentional filter settings 74 A Experiment : Categorization task RTs B Experiment : Categorization task ERPs contralateral ipsilateral to color singleton contralateral ipsilateral to shape singleton Figure. A: Mean response times in the categorization task for the color-predictive group (left) and the shape-predictive group (right) in Experiment. Error bars represent standard errors of the mean. B: Grand-averaged ERPs in Experiment recorded at posterior-occipital electrode sites (pool of O/, PO3/4, P7/8, and PO7/8), elicited by singletons in the categorization task for the color-predictive group (upper panel) and the shape-predictive group (lower panel). Red lines represent the ERP activity elicited by the predictive singleton at contralateral electrode sites; black lines represent the ERP elicited by the predictive singleton at ipsilateral electrode sites. The difference between red and black lines in the marked epoch represents the Npc component. For illustration purposes, EEG waveforms were low-pass filtered at 30 Hz using digital filtering (half-power cutoff, 4 db). Search Task Behavioral results. (See Figure 3.) Observers responded to the shape target more rapidly in distractor-absent trials (M ms) than in distractor-present trials (M 5 67 ms), F(,4) , p <.00, g The RT advantage for distractor-absent trials was numerically more pronounced for the color-predictive group (DM 5 37 ms) compared to the shape-predictive group (DM 5 7 ms), but this difference did not reach significance, F(,4) 5.45, p 5.4. Splitting the data in task switch and task repetition trials (see Figure 3B) and adding the factor trial sequence position showed that observers responded slower in task switch trials (M 5 67 ms) than in task repetition trials (M 5 68 ms), F(,4) , p <.00, g This difference was as pronounced in distractor-absent as in distractor-present trials, F(,4) <, and did not differ between categorization groups, F(,4) <. Numerically, observers were more accurate in distractor-absent trials (M 5 94.%) than in distractor-present trials (93.%), F(,4) , p 5.07, g 5.3. ERP results. Distractor-absent trials. (See Figure 4A.) In the early epoch (65 30 ms), targets elicited an Npc, as the mean ERP amplitude was more negative on contralateral electrode sites (M 56. lv) than on ipsilateral electrodes (M 55.8 lv), F(,4) , p 5.005, g p 5.9. The Npc extended to the later epoch (30 95 ms; M contra 53.7 lv; M ipsi 5.5); F(,4) , p <.00, g p It was of comparable size in the shapepredictive and the color-predictive group in both epochs (both ps >.493, g p <.0). trials: Target and distractor in same hemifield. (See Figure 4B). A pronounced Npc was observed in the early epoch (M contra 56.0 lv; M ipsi 55.6), F(,4) , p 5.004, g p 5.3, and in the later epoch (M contra 53.5 lv; M ipsi 5.6), F(,4) 5 6.8, p <.00, g p 5.5. Again, the size of the Npc was not affected by group membership, both ps >.05. trials: Target and distractor in opposite hemifield. (See Figure 4C.) In the early epoch, laterality of the ERP depended on whether color or shape was predictive in the categorization task, F(,4) , p 5.04, g p 5.6. While targets elicited no Npc in the shape-predictive group in the first epoch (DM lv), p 5.68, targets elicited an inverse Npc (negativity contralateral to the color distractor) in the color-predictive group (DM lv), p The magnitude of this inverse Npc was reliably larger than fluctuations in prestimulus noise, t() , p 5 004, as determined with the signed area comparison. In the second epoch, targets always elicited an Npc (M contra 53.7 lv; M ipsi 5.5), F(,4) , p <.00, g p 5.68, which was equally pronounced in the shape-predictive

7 74 H. Kadel, T. Feldmann-W ustefeld, and A. Schub o and in the color-predictive group, p 5.8, g p <.0. The onset of this Npc component was 35 ms earlier in the shape-predictive (9. ms) than the color-predictive group (54.9 ms; t c (4) , p 5.00). A Experiment : Search task RTs Distractor-absent 750 Distractor absent Distractor present 700 Experiment : Discussion Results of Experiment showed that visual search was strongly biased by selection history, even though a task cue indicated the upcoming task, allowing participants to prepare for an upcoming trial. Whether a trial was a switch trial (i.e., was preceded by a categorization task trial) or a repetition trial (i.e., was preceded by another search trial) did not affect this bias. Npc results in trials in which search target and color distractor were presented in opposite hemifields revealed that, for participants in the shape-predictive group, the shape target was the first item to be selected in the visual field, while in the color-predictive group participants attention was first captured by the color distractor before being redirected to the target, as indicated by a reliable distractor-elicited Npc. As a result, the onset of the target-elicited Npc was strongly delayed. One may argue that modulations of the lateralized ERP may have been due to differential processing of both target and distractor, and that these processes may have overlapped in time, making an interpretation of the lateralized ERP components difficult. A recent study that has disentangled the attentional subprocesses of target prioritization and distractor suppression by systematically varying target and distractor lateralization showed more prioritization of the target (indicated by a larger target negativity [N T ] in the ERP) in the shape-predictive group and more distractor suppression (indicated by a more pronounced distractor positivity [P D ] in the ERP) in the color-predictive group (Feldmann-W ustefeld et al., 05). Although in the present study both singletons (target and colored distractor) were always lateralized (i.e., the contralateral-ipsilateral difference waveforms always reflected the cumulative contribution of both singletons), in the light of previous findings, our results show that the presence of a distractor affected attention deployment in the color-predictive group much more than in the shapepredictive group. This indicates that the presentation of a pretrial task cue cannot prevent that attention is biased toward a stimulus that observers have experienced as being predictive, and/or away from a task-relevant stimulus. Experiment To further examine the role of task preparation and selection history on selective attention, Experiment used a fixed trial sequence of two trials of each task: two categorization trials were always followed by two search trials, and so on (Figure 5A, 6A). This manipulation enabled participants to perfectly predict whether the next trial would be a task repetition or a task switch trial, and to prepare for an upcoming trial as early as possible. Results were analyzed separately for the respective first and second trials in each task, which allows examining between-trial effects arising from the trial sequence position. For the first trial of a task sequence, participants could prepare by deleting any leftover activation from the previous task and by retrieving the settings required for the upcoming task. For the second trial of a sequence, however, leftover activations may be helpful and may reduce the amount of effort needed to prepare for the upcoming trial. Participants Experiment : Method Thirty-three new volunteers ( male, 30 right-handed) completed the experiment for course credit or monetary payment. All had normal or corrected-to-normal vision. Participants randomly assigned to the color-predictive group had a mean age of years,intheshape group years. Written consent for participation was obtained before the experiment. Only participants with at least 70% valid trials were included in the analysis (8 included, 5 excluded participants). Stimuli, Apparatus, EEG Recording Apparatus, EEG recording, and stimuli were the same as in Experiment, except that a gray fixation cross of 0.68 visual angle was presented before and during each trial, regardless of task type, instead of pretrial task cue. Procedure B Experiment : Search task switch and repetition trials Distractor absent Distractor present Switch trial Repetition trial Switch trial Repetition trial Figure 3. A: Mean response times in the search task for the colorpredictive group (left) and the shape-predictive group (right) in Experiment. Black bars represent RTs for trials with an additional color distractor, white bars represent RTs for distractor-absent trials. Error bars represent standard errors of the mean. B: Depicts the same data as (A), split for task switch trials (preceded by a categorization task trial) and task repetition trials (preceded by a search task trial) Procedure was as in Experiment, except for changes in the trial sequence.

8 Selection history alters attentional filter settings 743 A Distractor-absent B T & D same side C Experiment : Search task ERPs T & D opposite side Difference Waves contralateral ipsilateral to target D Figure 4. Grand-averaged ERPs in Experiment in the search task recorded at posterior-occipital electrode sites (pool of O/, PO3/4, P7/8, and PO7/8), elicited by search displays for participants in the color-predictive group (left) and the shape-predictive group (middle). A C: Red lines represent the ERP elicited by targets at contralateral electrode sites; black lines represent the ERP elicited by targets at ipsilateral electrode sites. View (A) shows ERPs elicited by targets in distractor-absent trials in which only the target was presented. View (B) shows ERPs elicited by targets in distractor-present trials for trials with target and distractor on the same side and (C) with target and distractor on opposite sides. Activity in an early epoch (65 30 ms) and a later epoch (30 95 ms) was considered separately. A distractor negativity (negative deflection contralateral to the color distractor) was only found for the color group in the early marked epoch when target and distractor were shown on opposite sides (lower left). An Npc (negative deflection contralateral to the target) was found in all conditions. For illustration purposes, EEG waveforms were low-pass filtered at 30 Hz using digital filtering (half-power cutoff, 4 db). View (D) shows the same results plotted as Npc difference waves (contra- minus ipsilateral activity) for each display type. Blue lines represent the Npc of the color-predictive group, black lines of the shape-predictive group. Vertical dotted lines in the lower panel represent the time point at which 50% of the maximal amplitude was reached. Categorization learning and mixed practice phase. In a first experimental session, each participant completed a feedbackguided categorization learning and a mixed practice phase identical to those of Experiment. In the mixed practice phase, trials of both tasks were presented in a random order. Sequence practice and main phase. Participants returned for a second experimental session on the next day. At the beginning, they were explicitly informed that both tasks would alternate from now on in a predictable sequence of exactly two trials per task. One practice block of 64 trials was administered to allow familiarization with the consistent task sequence. After the sequence practice block, 4 further blocks of 64 trials each (,536 trials in total) were conducted, with trials of the categorization and search task presented in a fixed task sequence of two trials each. Blocks starting with the categorization and the search task were equally frequent and alternated regularly. Data Analysis Behavioral data. Analyses were the same as in Experiment, except that mean RT for correct trials and mean accuracy were also calculated separately for the first and second trial in the respective task. A total of 3.7% of all trials were excluded due to being the first trials in an experimental block, or exceedingly long RT. P values were adjusted with Bonferroni correction in case of additional pairwise comparisons. EEG data. Npc analysis paralleled Experiment except for the following changes: epochs were adjusted to ms poststimulus in the categorization task. ANOVAs were extended by the within-subject factor sequence position (C vs. C, and S vs. S, respectively). To incorporate this factor in Npc latency analysis, statistical comparisons were performed with two-way ANOVAs, F values were adjusted by dividing by (n ) before assessing significance (Luck, 04; Ulrich & Miller, 00). The corrected values are referred to as F c in the results section. Categorization Task Experiment : Results Participants in the color-predictive group (M 5 57 ms) were faster than participants in the shape-predictive group (M ms), F(,6) 5 6.6, p 5.06, g p 5., Figure 5B. Participants from both groups responded faster in second trials of the sequence (M 5 50

9 744 H. Kadel, T. Feldmann-W ustefeld, and A. Schub o A Experiment : Categorization trials in the task sequence Search (S) Search (S) Categorize (C) Categorize (C) Search (S) Search (S) Categorize (C) Categorize (C) B Categorization task RTs C Categorization task ERPs First Trial (C) contralateral ipsilateral to color singleton contralateral ipsilateral to shape singleton First Trial Second Trial (C) (C) First Trial (C) Second Trial (C) Second Trial (C) Figure 5. A: Depiction of the trial sequence in Experiment. Trials of the categorization and the search task alternated in fixed sequences of two trials per task: First and second categorization trials (C, C) are highlighted here. B: Mean response times in Experiment in the categorization task for the color-predictive group (left) and the shape-predictive group (right), separated according to sequence position: first (C) and second (C) categorization task trials. Error bars represent standard errors of the mean. C: Grand-averaged ERPs in Experiment recorded at posterior-occipital electrode sites (pool of PO3/4, P7/8, and PO7/8), elicited by singletons in the categorization task for the color-predictive group (left) and the shapepredictive group (right). Analysis was conducted separately for first (C, upper) and second (C, lower) trials in the sequence. Red lines represent the ERP activity elicited by the relevant singleton at the contralateral electrode sites; black lines represent the ERP at ipsilateral electrode sites. The difference between red and black lines in the marked epoch represents the Npc component. For illustration purposes, EEG waveforms were low-pass filtered at 30 Hz using digital filtering (half-power cutoff, 4 db). ms) than in the first trials (M 5 59 ms), F(,6) , p <.00, g p Regarding accuracy, no group difference was observed: Participants were about as accurate in the colorpredictive group (M 5 98%) as in the shape-predictive group (M %), F(,6) <, Figure 5B. However, accuracy was higher in C trials (98.5%) than in C trials (97%), F(,6) , p <.00, g p 5.6. There was no significant interaction of group membership and sequence position, either regarding reaction times, F(,6) 5.43, p 5.3, or accuracy, F(,6) <. Electrophysiological results (Figure 5C) showed that both predictors elicited an Npc, as the ERP was more negative at electrode sites contralateral (M lv) compared to electrode sites ipsilateral (M 53.8 lv) to the predictor, F(,6) 5 40., p <.00, g p 5.6. The Npc was of comparable size in both groups (p 5.494, g 5.0). Generally, Npc amplitude was larger in C (DM 5.08 lv) compared to C (DM 50.8 lv) trials, F(,6) 5 9.0, p <.00, g p 5.4. Search Task Behavioral results. (See Figure 6.) Participants responded faster to targets in distractor-absent trials (M ms) than in distractorpresent trials (M ms), F(,6) , p <.00, g p This RT cost for distractor-present trials was more pronounced in the color-predictive group (DM 5 43 ms) than for the shapepredictive group (DM 5 4 ms), F(,6) , p 5.045, g p 5.5. Observers responded faster in the second trials (S) of the sequence (M 5 65 ms) than in the first trials (S) (M ms), F(,6) , p <.00, g p Similar patterns were observed in accuracy: observers were more accurate in distractorabsent trials (M 5 96.%) than in distractor-present trials (95%), F(,6) 5 5., p 5.00, g p For the color-predictive group, this effect was slightly more pronounced (DM 5.9%) than for the shape-predictive group (DM 5 0.6%, F(,6) 5 4., p 5.05, g p 5.4). Furthermore, observers tended to be more correct in S trials (95.9%) than in S trials (95.3%), F(,6) 5 3, p 5.095, g p 5.. ERP results. Distractor-absent trials. (See Figure 7A C, upper panels.) In the early epoch (65 30 ms), targets elicited an Npc, as mean ERP amplitudes were more negative on contralateral electrode sites (M 57. lv) than on ipsilateral electrodes (M lv), F(,6) 5 3., p 5.00, g p Also, in the later epoch (30 95 ms) an Npc was observed (M contra 53.3 lv; M ipsi 5.4), F(,6) , p <.00, g p 5.66, which was larger in S trials (DM 5.09 mv) than in S (DM mv) trials. In both epochs, the Npc was of equal size for the colorpredictive and the shape-predictive group (both Fs <., ps >.9).